Blog

'Better Call Saul' Season 6 Final Episodes Directors Went to 'Master Class' in Black & White – Showbiz Cheat Sheet


Binge-worthy entertainment news and celebrity interviews
Better Call Saul Season 6 had the show’s first all black and white episode with “Nippy.” The subsequent episode, “Breaking Bad,” had significant black and white segments too. As established by previous season premieres, Jimmy McGill (Bob Odenkirk)’s future life as Gene Takovic is in black and white. Now that it’s become a focal point of the Better Call Saul final episodes, the directors went to black and white school
Director Michelle MacLaren and cinematographer Paul Donachie appeared on the Better Call Saul Insider podcast on July 26 after “Nippy” aired. They described the black and white process of Better Call Saul’s final episodes. Better Call Saul Season 6 airs Mondays at 9 p.m. on AMC.
MacLaren was a regular director on Breaking Bad too. She credited co-creator Peter Gould with schooling the directors of the Better Call Saul final episodes how he wanted the black and white to look.
RELATED: ‘Better Call Saul’ ‘Breaking Bad’ Episode Changed Saul Goodman’s Introduction to Walter White
“Peter, you did something really cool at the very beginning of this when you knew that you were going into black and white,” MacLaren said on Better Call Saul Insider. “You got all the directors who were shooting the last few episodes together. And you gave us a master class on shooting black and white film, and it was phenomenal.”
MacLaren didn’t specify the movies Gould showed. But they were enough to convey what he wanted from the Better Call Saul final episodes. MacLaren explained how they used the black and white to convey Gene’s journey. 
Stay on your toes, the penultimate episode of #BetterCallSaul premieres tomorrow. pic.twitter.com/IOLgx7PR8f
RELATED: ‘Better Call Saul’ ‘Breaking Bad’ Episode Draws Subtle Parallels Between Walter White and Saul Goodman
“We took a deep dive that Peter had put together from all these clips of different movies that were decades old,” MacLaren said. “And talked about what you liked about them and what you were hoping for this. The cinematographers were there too. And Paul Donachie our DP really did a phenomenal job with playing with the light and the dark, and having Jimmy/Gene walk from the darkness into the light. And vice versa, depending where we were in the story.”
Donachie discovered filming Better Call Saul in black and white required different skills. To say filming is a bit of a misnomer too. Donachie worked with the digital cameras to create black and white for the final episodes of Better Call Saul.
I found some technical differences. What I found was black and white, funnily enough, you need more light. I have no idea [why] to be honest. I’m not a technical person at all but I just needed more light in there to see stuff. The rating of the camera’s the same. You’re actually recording it raw which is not in color, it’s not in black and white, you’re just recording ones and zeros. It’s fundamentally having a color picture which is turned into a black and white one. Other than that really it was just interesting to try to create more shadow and actually trying to create darker and lighter areas, things that were a bit more contrasty, having deep blacks.
RELATED:

source

Fractional neural sampling as a theory of spatiotemporal probabilistic computations in neural circuits – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 13, Article number: 4572 (2022)
460 Accesses
3 Altmetric
Metrics details
A range of perceptual and cognitive processes have been characterized from the perspective of probabilistic representations and inference. To understand the neural circuit mechanism underlying these probabilistic computations, we develop a theory based on complex spatiotemporal dynamics of neural population activity. We first implement and explore this theory in a biophysically realistic, spiking neural circuit. Population activity patterns emerging from the circuit capture realistic variability or fluctuations of neural dynamics both in time and in space. These activity patterns implement a type of probabilistic computations that we name fractional neural sampling (FNS). We further develop a mathematical model to reveal the algorithmic nature of FNS and its computational advantages for representing multimodal distributions, a major challenge faced by existing theories. We demonstrate that FNS provides a unified account of a diversity of experimental observations of neural spatiotemporal dynamics and perceptual processes such as visual perception inference, and that FNS makes experimentally testable predictions.
Humans and other animals operate in a world that is noisy and ambiguous. Moreover, it has been widely observed that the firing activity of neural systems is inherently stochastic and neural responses to stimuli exhibit large variability within and across trials1,2,3. Such exogenous and endogenous randomness naturally leads to the view that neural computations are carried out in a probabilistic way4,5,6,7. Indeed, a host of brain functions from sensory processing8,9 and cognitive tasks10,11 to motor behaviors12 have been successfully characterized from the perspective of probabilistic inference. The success of probabilistic accounts of these brain functions raises the fundamental question of how neural activity represents probability distributions and how neural circuits implement probabilistic inference based on that representation.
Two prominent types of models have been proposed to understand the neural basis of probabilistic representations and computations. One type is based on neural population activity that encodes the parameters of probability distributions (e.g., probabilistic population codes) such as the variance of Gaussian distributions4,13. The other type of model employs sampling-based probabilistic representations14,15,16,17,18,19. Probabilistic representations in these models are mainly based on the variability or fluctuations of neural activity with Poisson or Gaussian statistics. However, empirical evidence has increasingly demonstrated that neural fluctuations occur at multiple scales with heavy-tailed, non-Gaussian statistics20,21, and that such fluctuations unfold in both time and space to give rise to rich, complex spatiotemporal dynamics of neural circuits22,23.
Here we argue that these rich spatiotemporal dynamics enable a new type of sampling-based probabilistic neural computation. This probabilistic computation offers a unique perspective on the functional role of spatiotemporal neural dynamics and provides a solution to a long-standing challenge of reliably sampling complex probability distributions such as multimodal distributions. Sampling multimodal distributions is fundamentally important for processing natural environments replete with multiple, distributed salient patches (i.e., modes)24,25, and for learning and inference26. This form of processing remains a major challenge in previous studies, because Gaussian fluctuations or noise used in existing models essentially implement Brownian-motion-based Markov chain Monte Carlo (MCMC) sampling and its variants14,15,17. Due to the lack of ‘large jumps’ in Brownian motion (Fig. 1a), when faced with multimodal distributions with far-apart modes, neural samplers are unable to traverse through low-probability regions; such samplers are thus prone to be trapped in one local mode and lack the capacity of freely switching from one mode to another27.
a Typical sample path of a Brownian motion and that of a Lévy motion. The latter consists of clusters of short step sizes that are intermittently interspersed by long jumps. In contrast, the Brownian motion lacks such long jumps. Color indicates time. b Schematic diagram of the two-dimensional E-I spiking neural circuit. The circuit consists of recurrently connected excitatory and inhibitory neurons spanning a two-dimensional feature space (top layer), receiving multiple feedforward inputs (bottom layer). The spiking activity pattern (dots) emerging from the circuit exhibits rich, complex spatiotemporal dynamics. As indicated by the red arrow, the localized activity pattern intermittently switches between different parts of the neural circuit (black circles) and can implement sampling-based probabilistic computations. c Schematic diagram of a generative process for sampling-based probabilistic inference in our model.
We illustrate our probabilistic computation theory through a biophysically realistic, spatially-extended spiking neural circuit (Fig. 1b), demonstrating that population activity patterns (i.e., neural ensembles) emerging from the circuit possess large intrinsic fluctuations at multiple spatial and temporal scales as empirically observed. Rather than Brownian motion, the activity patterns propagating across the neural circuit with such large fluctuations exhibit clusters of short step-sizes that are intermittently interspersed by long jumps (Fig. 1a), with the movement step sizes showing heavy-tailed, non-Gaussian (Lévy) statistics28. These intermittent, long jumps inherent in Lévy motion enable the activity patterns to adaptively and freely switch between different modes of multimodal distributions, thus sampling these distributions with great efficiency. Besides these heavy-tailed Lévy motions in space, our sampling approach exploits certain temporal oscillatory components that play a fundamental role in speeding up the sampling process. We demonstrate that our sampling-based representation accounts for key response properties of neural circuits such as the reduction of neural variability after stimulus onset3,29 and theta oscillations (3–8 Hz) accompanied by 1/f activity as widely observed during environmental sampling tasks30, and that the sampling dynamics of the neural circuit are the underlying mechanism for perceptual switching31.
We further elucidate that the key spatial and temporal properties of the sampling-based representation implemented in the circuit model can be characterized by a stochastic differential equation with a fractional order derivative, which generalizes the notion of differentiation to fractional orders and captures heavy-tailed Lévy motions occurring at multiple scales28. We thus develop a mathematical model of our sampling approach; due to the fractional nature revealed by this mathematical model, we term our sampling approach fractional neural sampling (FNS). Based on this mathematical model, we illustrate the essential algorithmic properties of FNS for efficiently sampling multimodal distributions and further validate predictions from the mathematical model in our spiking neural circuit, thus revealing how and why FNS works at both the circuit implementation and algorithmic levels.
FNS-based probabilistic inference produces estimates about the mean and variance of probability distributions that match those of optimal Bayesian inference. However, due to the presence of small movements occasionally interrupted by large ones in FNS, it naturally generates switching-like behaviors and the resultant multi-modes of the estimate probability distributions. Our FNS-based inference thus provides a mechanistic account of why there exists bimodality in estimate distributions, as observed in a recent psychophysics study of visual perception inference32; this experimental result is not explained by conventional models of probabilistic inference. Our FNS-based inference model further makes quantitative predictions about how statistics of perceptual estimate distributions are related to stimulus contrast; these predictions are consistent with reanalysis of the existing experimental data. Dynamical switching between different activity patterns representing either multiple external sensory inputs33 or multiple choices in decision making34,35 have been widely observed, suggesting that FNS-based probabilistic computations could be of general applicability to understanding brain functions ranging from sensory processing to decision making.
We first illustrate how FNS works based on a biophysically realistic circuit model of spiking neurons and then reveal the circuit mechanism underlying the emergence of key spatial and temporal properties of FNS. Second, we formulate FNS by using a mathematical model derived from fractional diffusion formalisms, based on which we elucidate the computational properties of FNS for sampling multimodal distributions. We then validate these properties in the spiking neural circuit model. Finally, we illustrate how FNS-based probability inference can be implemented in the spiking neural circuit with the prior embedded in recurrent synaptic weights, and demonstrate that our model provides a novel account of visual perceptual inference as observed in experimental studies.
We consider a biophysically realistic spiking neural circuit model of excitatory and inhibitory neurons that incorporates experimentally established properties of the cortex, such as distance-dependent synaptic connectivity36,37 and correlated excitatory and inhibitory synaptic inputs38 (Fig. 1b, see “Methods”). The neural circuit spans a two-dimensional feature space39 with its x and y coordinates representing feature values such as orientation (angle) and color (hue), respectively, both ranging from −π to π. The circuit model exhibits a rich repertoire of dynamical activity states including asynchronous activity (i.e., disordered state) and localized propagating waves (i.e., ordered state)40, depending on the relative strengths of synaptic inhibition and excitation (characterized by an I–E ratio ξ; see “Methods”). Near the phase transition (ξ = ξc = 3.4) between these two states, localized activity patterns with rich, complex spatiotemporal dynamics emerge and behave like random walkers, exhibiting clusters of small movement step-sizes that are intermittently interspersed by long jumps (Fig. 2a). Such intermittent motion of the localized patterns can be characterized by Lévy motion (Fig. 1a), a type of non-equilibrium motion that has been shown to be essential for animals to optimally search for spatially distributed food41,42, for T-cells to efficiently find target pathogens in brain explants43, and for optimally transporting energy in turbulent fluids44. Notably, it has been shown that Lévy motion underlies the propagation of gamma (30–100 Hz) burst patterns in the MT area of marmoset monkeys45, and that hippocampal sharp wave ripples exhibit random movements with occasional long jumps46.
a Snapshot of the localized activity pattern when the I–E ratio ξ = ξc = 3.4. Dots denote spikes during a 5 ms period. The small circle marks the center-of-mass (CoM) of the pattern and the large circle represents one standard deviation of the spike locations. The line shows the CoM trajectory of the pattern over the previous 80 ms, with color indicating time. b Mean squared displacement (MSD) of the trajectory of the activity pattern as a function of time lag. The red line indicates a power-law fit, MSD(Δt) ∝ Δtη, with the diffusion exponent η = 1.18. c Distribution of trajectory increments. The red line indicates a fitted symmetric α-stable distribution, with the tail index α = 1.28. Inset: distribution of the trajectory increment on a log–log scale. d A typical sample trajectory overlaid on top of the sampled probability density. Large jumps with lengths greater than 0.25 are highlighted with connected green dots. e Autocorrelation of the sample trajectories for different stimulus contrast levels. Averaged across 100 trails of 10 s duration. f Mean squared error of the sample mean decreases with time lag, with a small baseline due to quenched noise (randomness in synaptic connectivity). g Dependence of the sample variance on input contrast; shades indicate the standard deviation of the sample variance across 20 trials.
To quantify the Lévy motion of the localized activity pattern (Fig. 2a), we calculate its mean-squared displacement and the distribution of the increment of the pattern’s movement as in ref. 43. We track the pattern over time and calculate its center of mass (CoM), ({hat{{{{{{{{bf{s}}}}}}}}}}_{t}=({hat{x}}_{t},{hat{y}}_{t})) (see “Methods”). As shown in Fig. 2a, the CoM trajectories exhibit random diffusive properties with variable step sizes, resulting in small-movement clusters occasionally interspersed by long jumps to new locations. The mean-squared displacement of the pattern can be calculated based on the CoM as, ({{{{{{{rm{MSD}}}}}}}}({{Delta }}t)=langle parallel {hat{{{{{{{{bf{s}}}}}}}}}}_{t+{{Delta }}t}-{hat{{{{{{{{bf{s}}}}}}}}}}_{t}{parallel }^{2}rangle), where Δt is the time lag. As shown in Fig. 2b, the mean-squared displacement is linear on a log–log scale, indicating that it is a power function of Δt, such that MSD(Δt) ∝ Δtη. The diffusion exponent η determines the type of the random motion: a value of η = 1 indicates Brownian motion, commonly used for implementing MCMC sampling47; a value of η > 1 indicates a superdiffusive process and η < 1 indicates a subdiffusive process. We find that the localized activity pattern emerging in our network has η = 1.18 (Fig. 2b), indicating that its movement is superdiffusive. We further examine the distribution of the increment ({{Delta }}{hat{x}}_{t}={hat{x}}_{t+{{Delta }}t}-{hat{x}}_{t}) of the pattern’s CoM trajectory by fixing the time interval Δt = 15 ms and find that it exhibits a heavy tail. The increments can be fitted to a symmetric Lévy stable distribution with index 1 < α ≤ 2 (see “Methods”); using maximum likelihood, we find that α = 1.28 (Fig. 2c). The tail of this distribution asymptotically follows a power law (p({{Delta }}hat{x}) sim|{{Delta }}hat{x}{|}^{-1-alpha }), where α is often referred to as the tail index. The heavy-tailed, power-law distribution of the increment and the corresponding superdiffusion are the characteristic features of Lévy motions28,48.
We next illustrate how the localized activity pattern (i.e., assemblies of neurons) with Lévy motion is able to implement FNS. To provide a conceptual understanding of the computational goal of FNS, consider the probabilistic generative process shown in Fig. 1c. A stimulus x (such as an image) is defined in terms of its latent feature s (such as its orientation and color) and a global contrast level c. For clarity, we denote the true value of the stimulus feature in a particular trial as s* to distinguish it from the generic latent variable s. The stimulus x evokes a sensory response r (spike counts). We assume that the sensory response depends on the latent feature in the form of a probabilistic population code; that is, the sensory response ri of each neuron i follows an independent Poisson distribution with a bell-shaped firing rate profile centered at s* with height proportional to the stimulus contrast c (see Eq. (10) in “Methods”). Thus, the sensory response r conveys both information about the latent feature s* as well as the strength of the sensory evidence (proportional to stimulus contrast). The sensory response is used as the feedforward input to the recurrent circuit, where probabilistic inference is performed by combining the sensory evidence conveyed by the feedforward input and the prior embedded in the recurrent circuit. Mathematically, this is expressed with Bayes’ rule
where p(s∣r) and p(s) represent the posterior and the prior distributions, respectively. In our FNS approach, a sample (hat{{{{{{{{bf{s}}}}}}}}}) from the feature space is represented by the instantaneous CoM of the localized spiking pattern in the recurrent circuit. The distribution resulting from the random motion of the localized activity pattern then approximates the posterior distribution.
In this section, we first consider the simple case of sampling a unimodal distribution corresponding to sensory evidence with a flat prior to illustrate the key spatial and temporal properties of FNS. The neural circuit receives a unimodal feedforward input centered at the true value of the stimulus feature s*, with its strength proportional to the stimulus contrast level c, as described by Eq. (10) (“Methods”). Figure 2d shows a typical sample path traced by the CoM of the localized spiking pattern, which exhibits complex spatial and temporal dynamics. Spatially, the sampling path exhibits the signature of Lévy motion, i.e., the presence of occasional large jumps in space (Fig. 2d). Similar to the case of spontaneous activity, the distribution of the sample path increment is characterized by a heavy tail. Temporally, the sample path displays an oscillatory component as indicated by its autocorrelation (Fig. 2e), which can be fitted as a function of the form (exp (-{{Delta }}t/tau )cos (2pi f{{Delta }}t)) with the decay constant τ ≈ 36 ms and the oscillation frequency f ≈ 5.5 Hz. The frequency of the oscillation is higher for the stronger contrast level (Fig. 2e). It is interesting to note that this frequency is in the range of theta oscillations (3–8 Hz), an elementary oscillatory component widely observed during environmental sampling tasks such as spatial attention sampling30 and whisking in rodents49. We will come back to this point in “Discussion”.
We next calculate the convergence speed of statistical estimates based on these samples toward the true value of the stimulus feature; in particular the convergence of the sample mean, ({bar{{{{{{{{bf{s}}}}}}}}}}_{T}=frac{1}{T}intnolimits_{0}^{T}{hat{{{{{{{{bf{s}}}}}}}}}}_{t}dt). As shown in Fig. 2, the mean-squared error of the sample mean, ({mathbb{E}}[parallel {bar{{{{{{{{bf{s}}}}}}}}}}_{T}-{{{{{{{{bf{s}}}}}}}}}{*}{parallel }^{2}]), with s* denoting the true value of the stimulus feature, decreases to half-maximum at THM = 128 ms for contrast c = 1 and THM = 81 ms for c = 2, which are around the same order as the adaptation time constant τK = 80 ms and are several times the membrane time constant (τm = 15 ms). The rate of convergence is faster for the higher contrast level, as consistent with the observation that the frequency of oscillation increases with contrast (Fig. 2e). This result indicates that the activity pattern emerging from the circuit implements sampling-based representation of the stimulus features, and that the sampling process is quite efficient.
As the stimulus contrast level represents the strength of sensory evidence, the increase in contrast should lead to a reduction of the sample variance, as suggested by the generative model in Eq. (1) as well as by existing studies13. We find that this property is indeed satisfied in FNS: as shown in Fig. 2g, the sample variance (frac{1}{T}intnolimits_{0}^{T}{({hat{x}}_{t}-{bar{x}}_{T})}^{2}dt) (calculated over a time period T = 10 s) decreases with stimulus contrast. From the dynamics aspect, the reduction of sample variance with contrast is primarily due to the modulation of the random motion of the activity pattern by the feedforward input; that is, as the contrast level of the input increases, the CoM of the pattern becomes more concentrated around the center of the input.
We now demonstrate that the sampling dynamics of FNS capture key response properties of the cortex including the reduction of neural firing variability3,29. To this end, we calculate the spike-count Fano factor for individual neurons by using a fixed time window of Δt = 100 ms. The Fano factor following stimulus onset has a pronounced drop compared to the Fano factor during spontaneous activity3. We find that this reduction of Fano factor displays a U-shaped dependence on the stimulus feature (orientation) as found in the middle temporal area of monkeys29, with the largest reduction occurring when the stimulus orientation is equal to the neuron’s preferred orientation (Fig. 3a). As shown in Fig. 3b, the Fano factor is greater than one for the spontaneous activity (c = 0) and is reduced by an amount largely proportional to stimulus contrast, with the exception of low contrast levels (c ≤ 0.3).
a Fano factor tuning of neural responses for varying stimulus feature values s relative to the neuron’s preferred stimulus spref for contrast level c = 1 (dots). Dashed line corresponds to spontaneous activity when c = 0. Fano factor is calculated using spike counts over 100 ms time windows. b Trough of the Fano factor in (a) as a function of the contrast level c of the feedforward input. Shades indicate the standard deviation across different neurons. c Theta oscillations revealed by the power spectral density (PSD) of the population firing rate with a peak frequency at 6.2 Hz. Solid line and shade indicate the average and standard deviation across 10 trials, respectively, each with a duration of 10 s. Inset: the PSD on a log–log scale reveals a power-law tail with an exponent of −2.28 (dashed line, offset for visibility).
We also investigate how the Fano factor in the recurrent circuit depends on the width of the feedforward input. As shown in Fig. S1a, when the width of the input is smaller than that of the receptive field, the Fano factor decreases with the input width; the converse is true when the input width is larger than the receptive field. It has been shown that in the primary visual cortex of monkeys, Fano factor exhibits a similar trend as the stimulus size varies (see Fig. 2f of ref. 50). In addition, we find that as the input width increases, the uncertainty of the sampled distribution changes in the similar way as the Fano factor (Fig. S1b). This correspondence between the uncertainty of neural responses and that of the sampled distribution is due to the reduction (or increase) in the variability of the instantaneous firing rates of neurons as the variance of the samples (represented by the CoM of the spiking pattern) decreases (or increases).
Another key feature of neural responses arising from the sampling dynamics is that neural population firing rate possesses a theta oscillatory component (3–8 Hz), as indicated in its power spectral density (PSD) (Fig. 3c). Interestingly, these oscillations ride on top of a 1/f arrhythmic component. Certain oscillatory components such as theta accompanied by 1/f-like activity have been widely observed in the cortex20,21; they are particularly relevant for cognitive functions such as visual-spatial attention51,52. These results provide further neurophysiological validity of our circuit model of FNS.
We next elucidate the circuit mechanism underlying the emergence of the sampling dynamics in our circuit model. More specifically, we aim to pinpoint the origin of the two key features of FNS: the fractional Lévy motion and the theta oscillatory component.
The circuit model exhibits a rich repertoire of dynamical activity states, ranging from the asynchronous to propagating wave states40. Particularly, by varying the I–E ratio (ξ), it has shown that around the transition state between the asynchronous (i.e., disordered) and localized propagating wave (i.e., ordered or coherent) states, the circuit model explains a range of nonlinear neural response properties; this state transition has been characterized by calculating the susceptibility and the branching indices of neural spikes40. Figure 4a shows that when the I–E ratio ξ is large, the network exhibits an asynchronous state without any structured patterns (State III, Fig. 4b), in which neural correlations (see “Methods”) are low. On the other hand, when the I–E ratio is small, coherent patterns emerge from the circuit in the form of a localized propagating wave; this wave pattern propagates across the neural circuit with a relatively smooth and regular trajectory (Fig. 4c). In this state (State I), neural correlations are much greater than those in the asynchronous state. In the transition regime (State II) between these two states (State I and State III), neural correlation changes rapidly as ξ varies. Only in State II, the CoM of the pattern exhibits fractional Lévy motion with small steps occasionally interrupted by long jumps (Fig. 1a)51, as indicated by a tail index of α < 2 with the trough of α coinciding with the transition point at ξ = ξc = 3.4. This value of I–E ratio (ξc = 3.4) is quantitatively consistent with that measured in the visual cortex of awake mice53. It is interesting to note that this dynamical mechanism underlying Lévy motion in our circuit model is similar to other complex physical systems whose critical phase transitions are essential for the emergence of Lévy motion54.
a Phase diagram of the emergent activity states in the spiking neural circuit as the inhibition-to-excitation ratio ξ varies. Neural correlations (blue line) of spike counts reveal a state transition from the asynchronous state (State III, weak correlation) to propagating wave state (State I, strong correlation). Only within the transition regime (State II), the increment distribution of the trajectory of the localized spiking pattern follows the α-stable distribution with its power-law tail index α smaller than 2, indicating the presence of fractional Lévy motion. b, c Snapshots of the spiking pattern for State I and State II featuring asynchronous activity (ξ = 4.25, 15 ms time window) and a localized propagating wave (ξ = 2.55, 2 ms time window), respectively. Red curve indicates the CoM trajectory of the propagating wave. d When adaptation is removed, the histogram of the sample increments (solid line) still exhibits a power-law tail (dashed line). e Sample path autocorrelation shows the absence of anticorrelation as a result of removing neural adaptation. f Mean squared error of the sample mean decreases slowly with the time lag, with a baseline due to quenched noise (randomness in synaptic connectivity).
Previous studies have shown that negative feedback mechanisms such as spike-frequency adaptation or depression are essential for the emergence of oscillations55,56. Our realistic spiking network model also incorporates spike-frequency adaptation in the form of slow potassium currents, which could be the origin of oscillation in our model. To test this, we remove the adaptation completely from the network by setting the potassium current in Eq. (5) to zero. We use the spontaneous activity to re-calibrate the inhibitory synaptic strength by a factor of 1.2 so that the neural dynamics is restored to the transition regime between the asynchronous and localized pattern states. Within this regime, the increments of the CoM of the localized activity pattern still follow a symmetric Lévy stable distribution, with the tail index α = 1.19 (Fig. 4d), indicating that the pattern moves in a fractional Lévy manner. However, we find that in the circuit without adaptation, the sample autocorrelation function of the sample path decays exponentially as (exp (-{{Delta }}t/tau )) with τ = 13–19 ms when the pattern samples the unimodal distribution; it does not show any feature of oscillation or anticorrelation (Fig. 4e). This lack of theta oscillations in the circuit model without adaptation results in a slower decay (time-at-half-maximum ΔtHM = 317 ms for c = 1 and ΔtHM = 220 ms for c = 2) of the mean squared error of the mean estimate (Fig. 4f) than for the case with oscillations (Fig. 2f), indicating that theta oscillations play a role of speeding up sampling of FNS. These results indicate that the oscillatory component of FNS originates from neural adaptation in our model.
We next develop a mathematical model to gain further theoretical insights into the probabilistic sampling processes implemented in our spiking circuit model. Based on this mathematical model, we reveal the unique computational properties of FNS through highlighting the functional roles of temporal oscillations and Lévy motion.
As demonstrated above, the spiking pattern behaves like a random walker that exhibits far richer spatiotemporal dynamics than Brownian motion: the pattern exhibits occasional long jumps in space, a characteristic feature of Lévy motion, and an oscillatory component in its CoM. To model a random walk with these features, we use a stochastic differential equation (SDE) driven by Lévy motion with an auxiliary momentum term
where ({hat{x}}_{t}) is the CoM of the activity pattern, vt is an auxiliary variable representing momentum, β is the damping coefficient, b(x) is a drift term related to the probability landscape, γ is the strength of the noise, and ({L}_{t}^{alpha }) is the Lévy motion whose step sizes over a time period Δt follow a symmetric Lévy stable distribution ({{{{{{{mathcal{S}}}}}}}}alpha {{{{{{{mathcal{S}}}}}}}}(alpha,{{{Delta }}t}^{frac{1}{alpha }})) possessing a power-law tail with a tail index 1 < α ≤ 2 (see “Methods”)28. It has been shown that the momentum term vt is responsible for generating temporal oscillations in the trajectory of the random walker57, with the frequency of the oscillations controlled by the damping coefficient β. Thus, the mathematical model (Eq. (2)) is able to capture the essential dynamical features of the localized pattern, i.e., the Lévy motion and the temporal oscillations. For analytical tractability we consider the case with one spatial dimension.
As the trajectory of a random walker evolves over time, its probability density converges towards a stationary distribution (target distribution) π(x). Therefore, the trajectory of the random walker, after a sufficient burn-in time, can be considered as approximate samples from π(x). In the case of Eq. (2), the stationary distribution π(x) is the stationary solution (marginalized by treating momentum v as a nuisance variable) to the corresponding fractional Fokker-Planck equation, and it is related to the drift term b(x) by (see Method for mathematical derivations of this result)
where ({{{{{{{{mathcal{D}}}}}}}}}_{x}^{alpha -2}) is the partial Riesz fractional derivative. Using Eq. (3), we can determine the drift term b(x) required for producing samples from any desired target distribution π(x).
This mathematical model provides a simple yet effective theoretical framework for revealing the key computational properties of FNS. Without loss of generality, we set the tail index in Eq. (2) to be α = 1.2, similar to the tail index characterizing the Lévy motion of the spiking patterns emerging in the neural circuit model, and set β = 1 to capture the oscillatory aspect of FNS; other values of 1 < α < 2 and β > 0 would generate qualitatively similar results. We then compare the computational performance of this default model with three other sampling processes: (1) Sampling without oscillation (β = 0) but with fractional Lévy motion (α = 1.2). (2) Sampling with oscillation (β = 1) but without fractional Lévy motion (α = 2), which is the standard Hamiltonian Monte Carlo (HMC) sampling57. Such Hamiltonian dynamics have been previously mapped to a neural network with excitatory and inhibitory populations58 for performing efficient sampling. (3) Sampling with neither oscillation nor fractional Lévy motion (α = 2 and β = 0). For this case, the sampling process reduces to the standard MCMC driven by Brownian motion47, called Langevin sampling, which is used as a general-purpose algorithm in machine learning; it has been proposed that neural networks may implement such sampling14,15,16,59.
Typical sample paths of the four cases for sampling a standard normal distribution with zero mean and unit variance are shown in Fig. 5a, each exhibiting distinct features. Immediately noticeable are the jumps in the sample paths of the two sampling processes powered by the fractional Lévy motions, but are absent in their non-fractional counterparts. Despite the drastic differences in the sample path structure, all four cases are able to produce samples from the target distribution, as seen through the agreement between the sample histogram and the target distribution (right panels in Fig. 5a). The finer temporal structures of these sampling processes are further revealed through calculating the autocorrelation function of the sample path, (langle {hat{x}}_{t+{{Delta }}t}{hat{x}}_{t}rangle), as we have calculated for the spiking pattern in the circuit model. As shown in Fig. 5b, the autocorrelation functions of the two cases without the oscillatory component (β = 0) decay exponentially to zero. However, those with the oscillatory component (β = 1) drop to zero quickly and exhibit negative-going lobes. This indicates that successive samples rapidly decorrelate, speeding up the sampling processes and thus resulting in a faster convergence to the target distribution. As shown in Fig. 5c, the mean-squared error decays significantly faster for the two cases with the oscillatory component (β = 1), with a time-at-half-maximum of 1.42 for α = 1.2 and 1.70 for α = 2, than that without oscillation (β = 0), with a time-at-half-maximum of 1.95 for α = 1.2 and 2.51 for α = 2. These results are consistent to the observations in the spiking neural circuit model in Fig. 2, and show that the temporal oscillatory property of FNS plays an essential role in improving sampling speed. Note that the cases with the momentum (β > 0) and without the momentum (β = 0) in the mathematical model correspond to neural sampling in the circuit model with and without adaptation, respectively. In the mathematical model, the momentum pushes successive samples further apart, giving rise to the oscillatory behavior and accelerating the sampling speed. In the neural circuit model, adaptation plays a similar role by pushing the localized activity pattern away from its present location, thus speeding up sampling.
a Sampling processes driven by fractional Lévy motion (α = 1.2, top two panels) exhibit long jumps, in contrast to their non-fractional counterparts (α = 2, bottom two panels). Right panels: sample histograms agree with the target distributions (dashed line). b Autocorrelation functions of different sampling processes. The processes with momentum (β = 1) exhibit an oscillatory component. c Mean squared error as a function of Δt. The two cases with an oscillatory component (β = 1) have significantly faster convergence to the true mean (gray line) than those without (β = 0). Fractional Lévy dynamics do not affect the convergence speed in a significant way.
We next elucidate that FNS possesses the powerful computational property of sampling multimodal distributions with far-away modes. For this purpose, we set the target distribution as in Eq. (3) to be a bimodal Gaussian mixture
Without loss of generality, we set s* = 2.5 and σ = 0.32. For clarity and simplicity, here we consider a bimodal distribution, but the results can be generalized to any multimodal distributions. Note that the modal separation Δs* is significantly larger than the modal width σ, implying vanishingly small probability density between the modes; in other words, the modes are separated by a high potential barrier, (U(x)=-log pi (x)), which is extremely difficult to penetrate. For the two fractional cases (α = 1.2), despite the impenetrable energy barrier between the two modes located at ± s*, the sample path is able to switch intermittently between them due to the inherent long jumps of Lévy motion (Fig. 6a). This property enables efficient sampling of the bimodal distribution, with the sample histogram matching the target distribution. In contrast, in both sampling processes driven by conventional Brownian motion (α = 2), i.e., the Langevin Monte Carlo sampling and the Hamiltonian Monte Carlo (HMC) sampling, the sample paths are completely trapped in one of the modes and are unable to explore the entire state space, thus failing to sample bimodal distributions. Particularly, we stress that while HMC has been previously implemented in neural networks for speeding up sampling processes, it cannot on its own resolve the issue of sampling multimodal distributions with far-apart modes because the sampler has a negligible chance to gather a sufficiently large momentum to overcome high potential barriers60.
a Sampling processes driven by fractional Lévy motion (α = 1.2, top two panels) are able to alternate between two modes rapidly (modal separation Δs* = 5), rather than being stuck at one mode as their non-fractional counterparts (α = 2.0, bottom two panels). Right panels: sample histograms and target distribution (solid line). b Mean exit time of the sampling processes driven by fractional Lévy motion grows linearly with modal separation, whereas those driven by Brownian motion grow exponentially; averaged across 24 trials of duration T = 104. Circles and crosses represent results from the mathematical model; solid lines denote the fitted exponential and linear curves.
The Lévy motion can fundamentally improve the mixing ability of the sampling process, that is, the ability to traverse through the probability landscape including the low-probability regions between the two modes. To quantify this mixing property, we calculate the exit time τexit (see “Methods”), which is the duration the random walker remains in one of the modes, and examine how the mean exit time changes with the modal separation Δs*. As shown in Fig. 6b, the dependence of the mean exit time on the modal separation has striking differences between the cases with and without fractional Lévy dynamics. For the two cases driven by fractional Lévy motions, the mean exit time increases linearly with the modal separation; in contrast, for the conventional cases driven by Brownian motion (Langevin sampling and Hamiltonian sampling), either with or without the oscillatory component, the mean exit time blows up exponentially as Δs* increases. Note that these results are achieved with all the other parameters such as the strength of the noise and the shape of the probability landscape fixed. Therefore, the mixing ability can be fundamentally attributed to the fractional Lévy motion ({L}_{t}^{alpha }). The linear dependence of the mean exit time on modal separation is the hallmark of FNS, indicating that it can effectively sample multimodal distributions regardless of variations in modal separation.
As illustrated in the mathematical model of FNS, the drift term b(x) determines the sampled distribution according to Eq. (3). To demonstrate that the spiking neural circuit performs non-trivial probabilistic computations as in the mathematical model, it is necessary to show how the drift term is related to the feedforward input (for encoding sensory evidence) and recurrent synaptic connections (for encoding prior). Due to the mathematical intractability of the spiking neural circuit model, we employ a neural field model (Eq. (27)) to derive the relationships between b(x) and parameters of the feedforward input and the recurrent synaptic weights (see “Methods” for the neural field model and Supplementary Information for details of the mathematical derivations); we then numerically validate these relationships in our spiking neural circuit model, thus providing insights into the probabilistic computations implemented in the spiking neural circuit.
Figure 7a (upper panel) shows b(x) when the neural field model receives a bell-shaped feedforward input (Eq. S5) centered at s1 = 0 with varying contrast levels c1. The drift is positive when x < s1 = 0 and is negative when x > s1 = 0, indicating that the localized pattern experiences a drift toward s1 = 0 and subsequently samples from a probability distribution centered at s1. Further analysis shows that the drift magnitude is proportional to stimulus contrast c1 (Eq. (28)). Using b(x), we then analytically obtain an explicit solution to the resulting sampled distribution (Fig. 7a, lower panel), whose variance is inversely proportional to c1 (Eq. S20). In addition, our analysis shows that a bell-shaped perturbation centered at s0 = 0 to the recurrent synaptic weights (Eq. S6) can induce a drift b(x) with a similar shape around s0 (Fig. 7b, upper panel); the magnitude of the drift is proportional to the strength of synaptic perturbation c0 (Eq. (28)). For this case, we also derive the corresponding sampled probability distribution (Fig. 7b, lower panel) and find that its variance is inversely proportional to c0 (Eq. S21); note that this probability distribution caused by changes in synaptic weights is the prior distribution.
a Drift term (upper panel) and sampled distribution (lower panel) when the neural field model receives a feedforward input with varying contrast levels c1. b Drift term (upper panel) and sampled distribution (lower panel) when the neural field model receives synaptic weight perturbation with varying strengths c0.
We next calculate the drift b(x) in the spiking neural circuit model from the CoM trajectory of the spiking pattern. Since the increments of the CoM are equal to the sum of a deterministic drift and random noise according to the mathematical model (Eq. (2)), we can obtain b(x) by averaging the increments of the CoM trajectory over time and across trials to remove the noise. We then divide the two-dimensional space into bins and calculate the drift as a function of the feature coordinate x. We find that when the neural circuit receives a feedforward input as in Eq. (10), the drift b(x) is positive for x < s1 = 0 and negative for x > s1 = 0, resulting in a sampled distribution centered at s1 (Fig. S2a). A bell-shaped perturbation to the recurrent synaptic weights (Eq. (12)) also yields a similar outcome (Fig. S2b), thus properly embedding the prior to the circuit. These results are qualitatively consistent with the analytical results based on the neural field model, indicating that the drift term is determined by the feedforward input and recurrent synaptic connectivity.
We next validate the powerful mixing property for sampling multimodal distributions in our spiking circuit implementation of FNS. For this purpose, we change the spatial profile of the feedforward input to be the superposition of two Gaussian functions centered at s1 and s2 with contrast levels c1 and c2, respectively (Eq. (11) in “Methods”); for simplicity, here we use a bimodal input as an example, but multimodal distributions can be similarly sampled in the circuit. Such multimodal inputs can be thought as representing sensory responses from multiple sensory channels, as in multisensory perception tasks61. It may also arise from a single stimulus with alternative interpretations such as the spatial orientation of a Necker cube. In both cases, the true state s* of the latent stimulus feature could be either s1 or s2.
Figure 8a shows that the localized pattern wanders around the location of one of the modes for a short while and then jumps to the other one, thus alternately sampling these modes in a similar fashion as in the mathematical model. Consequently, the sampling process leads to bimodal sample histograms whose modes are separated by low-probability regions (right panel in Fig. 8a). To calculate the exit time for the localized activity pattern performing sampling, we assign the spatial coordinates of the CoM trajectory at each time moment to one of the stimulus peaks depending on which peak location it is closer to. The exit time is then defined as the duration the trajectory spends in one region before it switches to the other. We investigate how the exit time changes as the modal separation Δs increases. As shown in Fig. 8b, the mean exit time ranges from 30 to 60 ms, indicating that the alternating sampling between the two modes occurs rapidly. The mean exit time increases linearly for ({{Delta }}s , < , frac{3}{4}pi), before it saturates due to the periodic boundary condition of the circuit model. This result indicates that FNS implemented in the neural circuit exhibits the salient feature as predicted in the mathematical model; that is, the mean exit time of the activity pattern increases linearly with the modal separation.
a Time series of the x– and y-coordinates of the sample path when c1 = c2 = 1. Samples (blue dots) are concentrated around the stimuli (red solid line) and switch rapidly between them without being trapped. Right panel: contour plot of the log sample probability density with a segment of the sample path overlaid on top. b Mean exit time (circle) grows linearly with the modal separation (with saturation due to periodic boundary conditions). c Mean exit time for varying c1 with c2 fixed (c2 = 1). d Mean exit time decreases as both contrast levels are kept the same and increased simultaneously. Error bars in b–d show the standard deviation of the exit time within trials. e Exit time (perceptual dominance duration) follows a right-skewed distribution better fitted with a Burr distribution (red line) with a more slowly decaying tail than a gamma distribution (blue line).
We next demonstrate the effectiveness of our circuit implementation of FNS for different contrast levels by systematically varying c1 while keeping c2 = 1 fixed. Rather than being trapped in the spot with a higher contrast level, the activity pattern is still able to dynamically switch between the two modes intermittently. By calculating the exit time, we find that when the contrast level of one of the stimuli becomes stronger, the mean exit time from this stimulus location increases monotonically while that from the other stimulus location decreases (Fig. 8c).
In summary, these results indicate that FNS can sample multimodal probability distributions in such a flexible and effective way that it is insensitive to their modal separations and contrast differences, thus retaining the excellent mixing ability. These dependencies of the exit time on modal separations and contrast levels form testable predictions of our FNS theory.
When presented with a stimulus with alternative interpretations such as the Necker cube or conflicting stimuli such as in binocular rivalry, humans may experience two alternating percepts31. Such phenomenon known as perceptual switching can be naturally explained in the framework of FNS. To model a simple case of binocular rivalry in which two conflicting stimuli with opponent colors are presented to two eyes, we add inputs to two neural groups whose preferred hues differ by π, as in existing modeling studies62. Alternations between the responses of the two neural groups (left panel, Fig. 8a) are interpreted as switching between different percepts. The exit time from each neural group calculated above is equivalent to the dominance duration of each percept in the context of perceptual switching. The result with varying contrast level c1 and fixed c2 (c2 = 1), as shown in Fig. 8c, is thus consistent with one of the key properties of perceptual switching of binocular rivalry31. When the contrast level of one stimulus is increased from zero with the other fixed, it primarily decreases the mean dominance duration of the stronger percept while increasing that of the weaker one to a lesser extent and equi-dominance is reached when the contrast levels of both stimuli are equal (c1 = c2). Further increasing the contrast level of the stimulus primarily increases the dominance duration of the stronger percept while decreasing that of the weaker one to a lesser extent31. We also increase the contrast levels of the two stimuli simultaneously and find that the mean dominance duration of both percepts decreases (Fig. 8d), which is another key property of perceptual switching31.
In addition, dominance durations in our model follow a right-skewed distribution (Fig. 8e), consistent with psychophysical studies of perceptual switching. Although the specific forms of this type of right-skewed distribution have been found to vary across different experimental and modeling studies, which are still a matter of debate31, we find that the dominance durations in our model are best fitted by a right-skewed distribution called Burr distribution rather than a gamma distribution (see “Methods” for the fitted parameters of these distributions). The key difference is that the former features a slowly decaying power-law tail whereas the latter has a faster decaying exponential tail. It is interesting to note that such a slowly decaying tail in the distribution of perceptual dominance duration has been previously found in experiments63, albeit not explicitly stated. Our results thus propose another right-skewed distribution for capturing dominance duration, which can be tested in future experimental studies.
As in our model, previous noise-driven attractor models have successfully reproduced the dependence of perceptual switching on stimulus contrast and the distribution of dominance duration through a combination of adaptation and externally added noise64. In our model, however, intrinsically emergent fluctuations in the form of fractional Lévy motion are the dynamical origin of perceptual switching. To demonstrate this, in the transition regime of the circuit model without adaption as described above (see Fig. 4), we apply the same feedforward input described by Eq. (11) and find that the localized activity pattern exhibits rapid switching between the two neural groups (Fig. S3). This result thus indicates that the fractional Lévy motion underlies perceptual switching. It is important to note that in our model, the movement of the activity pattern in space has large fluctuation across multiple scales, as evidenced by the heavy-tailed, power-law distribution of its movement step sizes; however, switching behaviors between different attractors possess characteristic scales in existing modeling studies. In ref. 15, it has been shown that noise-driven activity states in a neural network model can be used to perform MCMC sampling and to model perceptual switching. In this study, the effect of modal separation on perceptual switching was not investigated; however, as demonstrated in the mathematical model (Eq. (2)), at the algorithmic level, the MCMC sampler with Langevin dynamics is unable to traverse through low-probability regions, particularly when multimodal distributions have far-apart modes. In contrast, the FNS theory suggests that Lévy motion with large fluctuations emerging from the spiking neural circuit can serve as a robust mechanism for explaining perceptual switching, as the FNS-based sampling dynamics are insensitive to modal separation.
We next illustrate the properties of FNS-based probabilistic inference particularly through accounting for the key features of human perceptual inference in a motion direction task32. In this experiment, subjects were presented with visual stimuli consisting of moving dots, some of which move in the same direction while the rest move in random directions; the fraction of dots moving in the same direction is called coherence. It has been shown that the mean and variance of perceptual estimates match those predicted by the basic Bayesian observer model. However, the distribution of the perceptual estimates was found to be bimodal32; this observation was at odds with the basic Bayesian observer model which gave rise to a unimodal posterior. We now demonstrate that FNS-based inference provides a neural mechanism for resolving this puzzle and further makes new predictions about perceptual inference.
To proceed, we first establish a connection between the experimental setup and our model by interpreting the x-component of the 2D feature space as motion direction. The x-component of the CoM of the localized pattern at a particular time instant thus represents the perceived motion direction, whereas the y-component is considered as representing a nuisance parameter. We assume that the visual stimulus is represented through a population code as a bell-shaped feedforward input to the spiking neural circuit model (Eq. (10)), such that its center s1 represents the motion direction and its contrast c1 is proportional to the motion coherence, similar to that used in ref. 32. With this understanding, we compare the FNS-based inference results with the experimental results by carrying out the similar procedures as in the original experiment. In the experiment, the subjects were first trained with a block of visual stimuli to learn a prior distribution, and then were presented with a visual stimulus for a brief period of 300 ms before the subject was required to report the motion direction. Correspondingly, in our spiking neural circuit model we first embed a prior by applying a bell-shaped perturbation to the synaptic weights centered at s0 with amplitude c0 according to Eqs. (12) and (13) (see “Methods”). As illustrated above, these synaptic changes can be used to properly embed a prior to the recurrent circuit (Fig. 7). We then add a unimodal feedforward input with contrast c1 centered at s1 (Eq. (10)), which encodes the sensory evidence about the motion direction of the visual stimulus. The sampler (i.e., the localized activity pattern) is able to switch freely between the prior and the newly added sensory evidence within the short stimulus presentation period. The instantaneous position of the activity pattern at the end of the stimulus presentation is then interpreted as the perceptual estimate.
We then use our model to explain the three sets of experiments in ref. 32. The first set of experiments involves fixing the coherence of the random moving dots and the prior while varying the motion direction relative to the prior mean. Correspondingly, in our neural circuit model, we fix the contrast level c1 = 0.5 in Eq. (10) and the strength of synaptic perturbation c0 = 0.002. Without loss of generality, we fix the prior mean s0 = (0, 0) and systematically vary the center s1 = (s, 0) of the feedforward input with s varying from −π to π. As shown in Fig. 9a (left panel), the estimate distribution is characterized by two peaks: one corresponding to the prior centered at the origin and another corresponding to the sensory evidence with varying motion directions. Notably, the similar feature of bimodality versus modal separation has been observed in the experiments32 (see Fig. 4 of their paper). As illustrated above, the bimodality in our circuit model arises as the spiking activity pattern intermittently alternates between the two modes (Fig. 8a). We further quantify the FNS-based probabilistic inference by calculating the mean and variance of perceptual estimates across trials. As the center of the feedforward input (thus the modal separation relative to the prior at the origin) increases from −π to π, the mean of the bimodal perceptual estimate distribution also increases from negative to positive values in an approximately linear fashion (Fig. 9a, top right panel); this relationship is consistent with the experimental results32 (see Fig. 3a of their paper). The variance of the perceptual estimate is approximately a quadratic function of modal separation (Fig. 9a, bottom right panel). We find that the linear and quadratic dependences of the mean and variance of perceptual estimates on the modal separation are also consistent with the analytical results obtained from the neural field model (see Eqs. S23–S24 in Supplementary Information).
Perceptual estimate histogram, mean and standard deviation for: a varying modal separations between the stimulus (red bar) and the prior (black bar); b for varying stimulus contrast and fixed prior; c for varying prior strength (synaptic perturbation) and fixed stimulus contrast. Shades indicate the standard deviation across 20 trials. d, e Statistics of estimate distribution based on reanalysis of published experimental data. d Dependence of perceptual estimate variance (dots) on the modal separation can be approximately fitted by quadratic functions (solid curves), excluding outliers due to periodic boundary conditions. Stimulus coherence is fixed at 6%. e Standard deviation of the perceptual estimates decreases with stimulus coherence (solid lines), except for low coherence levels. Dashed lines denote extrapolations based on the theoretical argument that the perceptual estimate distribution should be equal to the prior when the coherence is zero. Modal separation is fixed at 20°.
To test the predictions from our inference results, we re-analyze the published data32 and find that the variance of the perceptual estimates also matches a quadratic curve when the prior standard deviation is large (80 degrees), as shown in Fig. 9d (red dots). However, for smaller prior standard deviations (Fig. 9d, green and orange dots), whether the experimental data matches our prediction remains inconclusive due to the limited data range32. Specifically, since the perceptual stimuli used in the experiments were pooled from the same dataset with which the prior was trained, its range was limited by the prior standard deviation (e.g., for a prior std of 10 degrees, the range of the presented stimulus is only about ±20 degrees). Future experiments could test our predictions by extending the range of the perceptual stimuli even when the prior standard deviation is small.
The second set of experiments involves varying the motion coherence while fixing the prior uncertainty and the distance of motion direction relative to the prior mean32. In our neural circuit model, this corresponds to varying the contrast level of the feedforward input c1 while fixing the strength of the synaptic perturbation c0 = 0.002 and the modal separation between them. The sample distributions for this case are shown in Fig. 9b (left panel), likewise exhibiting bimodality. As the contrast level increases, the height of the mode corresponding to sensory evidence increases whereas the height of the prior decreases. The mean and variance of perceptual estimates are also largely consistent with the experimental results: as the contrast level of the stimulus increases from zero, the mean of the bimodal perceptual estimate distribution gradually shifts from the prior at s0 toward the sensory evidence at s1 (Fig. 9b, top right panel). Meanwhile, the perceptual estimate variance decreases with the contrast level, but with the exception at low contrast levels (Fig. 9b, bottom right panel).
This finding that the variance of perceptual estimates varies non-monotonically with the stimulus coherence therefore provides a testable prediction for future experimental studies. By re-analyzing the published data32, we find that the perceptual estimate standard deviation decreases with the stimulus coherence for a number of coherence values (Fig. 9e). Experimental data for coherence smaller than 5% is unavailable so whether monotonicity is violated cannot be directly verified. However, if we impose the theoretical argument that the perceptual estimates should simply be samples from the prior distribution when the coherence is zero, we can then extrapolate the estimate standard deviation with the prior standard deviation. By presenting all data points together, we find that the monotonicity is indeed violated (Fig. 9e). To directly test our prediction, future experiments should more closely examine changes in the variance of perceptual estimates when the stimulus contrast levels are low.
The third set of experiments involves varying the uncertainty of the prior while fixing the coherence of the stimulus and the motion direction32. In the neural circuit model, this corresponds to varying the strength of the heterogeneous synaptic perturbation c0 while fixing the contrast of the feedforward input c1 = 0.5 and the modal separation. As the synaptic perturbation becomes stronger, the height of the prior increases and the height of the mode corresponding to sensory evidence decreases (Fig. 9c, left panel). The perceptual estimate mean gradually shifts from the sensory evidence at s1 to the prior s0 (Fig. 9c, top right panel). The perceptual estimate variance also depends non-monotonically on the strength of synaptic perturbation (Fig. 9c, bottom right panel). This observation cannot be verified in the experimental data due to the limited data size and thus forms a testable prediction for future experiments.
In this study, we have presented a theory (i.e., FNS) of probabilistic neural computations through the illustration of its neural circuit implementation as well as the normative formulation of its computational properties. By extending existing models that are mainly based on temporally variable or fluctuating dynamics13,14, FNS exploits rich, complex fluctuations of neural population activity both in time and in space for efficiently performing sampling-based probabilistic computations; FNS thus offers an approach to addressing the long-standing challenge of sampling and representing multimodal distributions. Our probabilistic neural computation theory provides a unified account of a variety of findings on neural dynamics at the individual neuron and circuit levels as well as on perceptual phenomena such as perceptual switching and visual perception inference, thus establishing a framework for understanding neurophysiological and computational mechanisms of brain function.
Our FNS theory provides a new perspective on the role of complex spatiotemporal neural dynamics that exhibit large fluctuations across multiple scales, which are of a non-Gaussian (i.e., heavy-tailed) nature. Particularly, FNS-based probabilistic computations harness the power of the fractional Lévy motion of population activity patterns (i.e., neural ensembles); these patterns hover around one location for a while and then move or switch to another location in an intermittent manner, with their movement step sizes following heavy-tailed, power-law distributions. Such fractional motions in space give rise to irregular propagation trajectories and speeds with large variability. Propagating activity patterns have been widely observed at the circuit and the whole brain levels22,23; notably, localized gamma activity patterns with Lévy motion have been found in the MT area of marmoset monkeys45. We have repeated the analysis on the broadband LFP from ref. 45 (see Supplementary Information) and found that localized activity patterns of such broadband activity also show Lévy motion (Fig. S4). It has been shown that hippocampal sharp wave ripples exhibit movements with clusters of short step sizes that are intermittently interspersed by long jumps46. Nevertheless, to directly test our modeling prediction of propagating spiking patterns with Lévy motion, future studies need to focus on massive individual-neuron recordings and to analyze spiking patterns in the same way as done in our modeling study.
Another key property of sampling dynamics of FNS is that the autocorrelation function of the sample path exhibits an oscillatory component. Such an oscillatory component induces negative-going lobes in the autocorrelation; consequently, this would result in independent samples occurring in a much shorter time, thus playing an essential role in increasing the effective sampling speed as in the Hamiltonian Monte Carlo method57,58. In a previous modeling study58, such oscillatory activity arises from mapping Hamiltonian dynamics to the activity of excitatory and inhibitory neurons and its peak frequency is around 40 Hz (i.e., gamma oscillation). Importantly, it has been demonstrated that such gamma oscillations can naturally emerge in a circuit model trained for optimally performing sampling-based probabilistic inference65. In our model, as illustrated in the autocorrelation analysis of the sampling process, the temporal oscillation is in the theta range (3–8 Hz). We have found that the spike frequency adaptation is essential for the genesis of the theta oscillatory component, in accordance with other modeling studies56. However, different from oscillatory activity with clock-like periodicity in these models, neural oscillations in our model exhibit great variability and are accompanied by heavy-tailed, 1/f-like activity, as shown in the power spectrum of the firing rates of neural population activity. Notably, such temporal fluctuations with a 1/f component have been widely observed in neural population activity as recorded by LFP, EEG, and MEG during both spontaneous activity and task conditions20. Traditionally, such 1/f-like activity has been deemed unimportant and often removed from analyses to emphasize oscillatory components. In our theory, however, it is an integral part of our circuit implementation of FNS.
It has been found that theta oscillations are a hallmark feature of environmental sampling having now been linked to spatial attention sampling 30,52, eye movements in primates66, and whisking in rodents49. In some of these environmental sampling tasks, such as eye movement-based sampling of natural scenes, the existence of Lévy-like motion has been reported67. These observations lead us to predict that our FNS-based probabilistic computation might be the underlying computational mechanism of these cognitive functions; indeed, recently we have successfully applied FNS dynamics to explain key neural and behavioral effects of visual attention51. Furthermore, it is interesting to note that the phase of theta oscillations often modulates the amplitude of gamma fluctuations68. Such theta-gamma coupling suggests that it is important to explore whether and how each theta sampling cycle can be implemented through gamma activity, potentially unifying our FNS approach with the gamma dynamics-based sampling approaches58,65 and gaining a comprehensive understanding of probabilistic computations in neural circuits.
As formulated in the mathematical model with fractional order derivative and further illustrated in the spiking neural circuit, FNS exploiting the complex spatiotemporal dynamics mentioned above possesses some profound computational advantages. The sampling processes of FNS are quicker than those of the other methods such as the classical Langevin sampling method that has been implemented in stochastic recurrent neural networks. Importantly, FNS enables the efficient sampling of multimodal distributions even with far-away modes, due to the long jumps inherent to fractional movements of neural activity patterns. Sampling multimodal distributions is crucial for probabilistic computations, which, however, would not be possible in existing models, because they rely on Gaussian dynamics (i.e., Brownian motion) for implementing classical Langevin Monte Carlo sampling methods used in machine learning69. These methods are notorious for becoming trapped in a single mode, and lack the capacity to efficiently traverse through complex probability landscapes. Based on both the circuit and mathematical models, our theory of FNS makes a core prediction; that is, the mean exit time (i.e., the average time the sample path takes to leave one of the modes of a bimodal distribution) depends linearly on the modal separation. This prediction is falsifiable and hence represents a strong test of our FNS theory.
Our FNS can be implemented in biophysically realistic, two-dimensional spiking neural circuits incorporating two well-established properties of the cortex. One property is the dependence of synaptic connection probability on distance36,37 and another is the balance between excitatory and inhibitory inputs, with the former closely tracked by the latter38. Most previous models of sampling-based representations14,15,58,59, however, have focused on either normative probabilistic models or theories without specifying their underlying neural circuit mechanisms14,15, or on abstract neural circuits models 58, some of which have no separation of excitatory and inhibitory neurons thus violating Dale’s principle59. Networks with one-dimensional structure (i.e., a ring structure) have been trained to produce a fast sampling-based inference exploiting cortical-like temporal dynamics65. Since these networks do not intrinsically produce variable neural dynamics, an external source of fluctuations is required. In contrast, in our circuit model, spatiotemporal activity patterns with large fluctuations are intrinsically generated and the 2D spatial structure supports the emergence of such fluctuations. This scenario is similar to other complex physical systems where such a 2D spatial extension is important for the emergence of spatiotemporal patterns with complex dynamics70. In our model, the complex spatiotemporal patterns with fractional motions only emerge when the circuit works near the transition between different activity states; these findings are consistent with recent studies proposing that complex cortical dynamics could be better understood in the dynamical regime close to the transition between different cortical states (i.e., asynchronous and synchronous/coherent states)71,72,73, but go beyond them by revealing the fundamental functional role of complex spatiotemporal cortical dynamics in probabilistic neural computations.
The circuit mechanism of FNS also has important implications for constructing efficient artificial neural networks in machine learning. Large-scale artificial neural networks such as the Boltzmann machine74 and deep belief networks75 often perform complex computations through employing probabilistic sampling. Most of these artificial neural networks perform probabilistic sampling by using the classical MCMC driven by Gaussian noise, so they suffer from the similar problem as faced by the existing models of neural probabilistic sampling, in that these methods lack the ability to jump across low-probability regions and to traverse through the probability landscape (also known as the ‘mixing’ ability). The problem of mixing becomes even more pronounced when dealing with larger complex datasets. The common remedy for these methods relies on some variants of simulated tempering76, which change the temperature parameter in order to globally flatten the solution landscapes during the sampling process. These tempering methods, however, come with a cost of their own, because they require extra computations and parameter-tuning that assume knowledge about the global state of the artificial neural networks. However, our circuit-based FNS mechanism indicates that powerful mixing abilities for representing multimodal distributions can emerge from the circuits that are essentially locally coupled, and importantly this happens in a fundamentally autonomous manner without tuning any global parameters during the sampling process. Motivated by these properties of the circuit-implementation of FNS, we thus suggest that future large-scale, information processing neural network models may benefit from our circuit mechanism of FNS, if they are designed to exploit fractional Lévy-like diffusion and oscillations for sampling complex, high dimensional probability landscapes.
As we have demonstrated, the FNS-based probabilistic sampling can implement perceptual inference that explains both perceptual estimate statistics and the bimodality of perceptual estimate distributions as found in a recent study of motion direction estimation task32. Regarding probabilistic inference, our model makes two main predictions: first, the perceptual estimate variance is approximately a quadratic function of the modal separation between the sensory evidence and the prior; second, the perceptual estimate variance decreases with stimulus contrast but this trend is reversed when the contrast level is small. These model predictions are consistent with our reanalysis of the experimental data32, except for the cases when the stimulus contrast level is low; a conclusive test of these predictions will require more specific manipulation of stimuli to cover the low contrast range. Previously, a binary switching model with Gaussian noise was used to explain the bimodal property of perceptual estimates32. In our model, due to the non-equilibrium, fractional nature of Lévy motion, the activity patterns exhibit short-step clusters that are interspersed by long jumps, thus giving rise to clustering and switching-like behaviors and the resulting bimodality of the estimate probability distributions. Rather than being simplistic and binary, the switching-like behavior in our FNS model happens across multiple scales with large fluctuations, as evidenced by step sizes with heavy-tailed, power-law distributions. Switching-like dynamics with large fluctuations have indeed been found in the inferotemporal area of monkey during sensory processing33, in the orbitofrontal cortex34, and in the lateral intraparietal area35 during decision making. Our FNS mechanism of probability inference might therefore be generally applicable to understanding these key brain functions.
The spatially extended spiking neural circuit model consists of NE excitatory neurons and NI inhibitory neurons embedded in a two-dimensional feature space with periodic boundary conditions40. Each neuron i is assigned to spatial coordinates si = (xi, yi) representing the preferred stimulus features (orientation and color) of that neuron, with both xi and yi ranging from −π to π. We consider NE = 63 × 63 = 3969 excitatory neurons on a square grid and NI = 1000 inhibitory neurons with uniformly random locations. The ratio between NE and NI is around 4. The Euclidean distance is calculated between each pair of neurons i from population α and j from population β, where α and β are either excitatory (E) or inhibitory (I). The connection probability between neurons is proportional to the distance-dependent factor ({{{Omega }}}_{ij}^{alpha beta }={e}^{-{D}_{ij}^{alpha beta }/{tau }_{D}^{alpha beta }}). For the excitatory connections the spatial scales are ({tau }_{D}^{{{{{{{{rm{EE}}}}}}}}}=8) and ({tau }_{D}^{{{{{{{{rm{IE}}}}}}}}}=10) grid units, whereas for the inhibitory connections the spatial scales are ({tau }_{D}^{{{{{{{{rm{II}}}}}}}}}={tau }_{D}^{{{{{{{{rm{EI}}}}}}}}}=20) grid units.
The subthreshold membrane potential ({V}_{i}^{alpha }) of neuron i in population α follows
where the membrane capacitance C = 0.25 nF, the leak conductance gL = 16.7 nS, and VL = −70 mV is the reversal potential for the leak current. ({I}_{i,{{{{{{{rm{K}}}}}}}}}^{alpha }(t)) is the potassium current, ({I}_{i,{{{{{{{rm{rec}}}}}}}}}^{alpha }(t)) is the recurrent synaptic current received by the neuron and ({I}_{i,{{{{{{{rm{ext}}}}}}}}}^{alpha }(t)) is the external current. When the membrane potential reaches the threshold Vth = − 50 mV, a spike is emitted and the membrane potential is reset to the potential Vrt = −60 mV for an absolute refractory period τf = 4 ms. The potassium current is given by ({I}_{i,{{{{{{{rm{K}}}}}}}}}^{alpha }(t)=-{g}_{i,{{{{{{{rm{K}}}}}}}}}^{alpha }(t)({V}_{i}^{alpha }(t)-{V}_{{{{{{{{rm{K}}}}}}}}})), where ({g}_{i,{{{{{{{rm{K}}}}}}}}}^{alpha }(t)) is the active potassium conductance and VK = −85 mV. The dynamics of the potassium conductance are described by
where ({t}_{j,k}^{alpha }) is the time of the kth spike emitted by neuron i from population α, ΔgK = 10 nS and τK = 80 ms. Because spike frequency adaptation has been primarily observed in cortical pyramidal neurons, we only include such adaptation for excitatory neurons in our model. The recurrent synaptic current ({I}_{i,{{{{{{{rm{rec}}}}}}}}}^{alpha }(t)) in Eq. (5) is
where ({g}_{i}^{alpha beta }(t)) is the conductance of the recurrent current from the presynaptic population β. The excitatory and inhibitory reversal potentials are ({V}_{{{{{{{{rm{rev}}}}}}}}}^{{{{{{{{rm{E}}}}}}}}}=0) mV and ({V}_{{{{{{{{rm{rev}}}}}}}}}^{{{{{{{{rm{I}}}}}}}}}=-80) mV, respectively. The conductance ({g}_{i}^{alpha beta }(t)) is given by
where ({a}_{ij}^{alpha beta }) and ({J}_{ij}^{alpha beta }) represent the coupling topology and the connection strength, respectively, as detailed in the original paper40. Specifically, the synaptic connection probability between neurons decreases in an exponential manner as distance between neurons increases, consistent with experimental findings36,37. The non-dimensional gating ({s}_{ij}^{alpha beta }(t)) describes the synaptic dynamics
where ({tau }_{{{{{{{{rm{d}}}}}}}}}^{beta }) and ({tau }_{{{{{{{{rm{r}}}}}}}}}^{beta }) are the decay and rise time constants, respectively, ({t}_{j,k}^{beta }) is the time point of the kth spike of neuron j from population β, and ({d}_{ij}^{alpha beta }) is the conduction delay drawn from a uniform distribution between 0 and 4 ms.
In our model, we consider an essential neurophysiological feature of local cortical circuits, that is, the excitatory post-synaptic currents are proportional to the inhibitory ones, with a homogeneous ratio across all excitatory neurons, as found in mouse primary visual cortex layer 2/353. To model this in our heterogeneous circuit, we consider the I–E ratio ({xi }_{i}=mathop{sum }nolimits_{k=1}^{{K}_{i,{{{{{{{rm{in}}}}}}}}}^{{{{{{{{rm{EI}}}}}}}}}}{J}_{jk}^{{{{{{{{rm{EI}}}}}}}}}/{sum }_{j}{J}_{ij}^{{{{{{{{rm{EE}}}}}}}}}), where ({K}_{i,{{{{{{{rm{in}}}}}}}}}^{{{{{{{{rm{EI}}}}}}}}}) denotes the number of connections (in-degree) received by excitatory neuron i from the inhibitory population and the connection strength values ({J}_{ij}^{{{{{{{{rm{EE}}}}}}}}}) are determined by the reverse pooling method40. To equalize the I–E ratio ξi across the neurons to a desired network-wide ratio, that is, 〈ξi〉 = ξ, the ({J}_{ij}^{{{{{{{{rm{EI}}}}}}}}}) for neuron i are sampled from a Gaussian distribution with a mean equal to (xi {sum }_{j}{J}_{ij}^{{{{{{{{rm{EE}}}}}}}}}/{K}_{i,{{{{{{{rm{in}}}}}}}}}^{{{{{{{{rm{EI}}}}}}}}}) and a standard deviation that is 25% of the mean. The I–E ratio ξ is varied as a system parameter for demonstrating the emergence of localized activity patterns with Lévy motion (Fig. 4) but is otherwise fixed at the value ξc = 3.4.
It has been shown previously that the circuit model exhibits a rich repertoire of dynamical activity states, ranging from asynchronous to localized and global propagating wave states40. Particularly, it has shown that around the phase transition between the asynchronous and localized propagating wave states, our circuit model exhibits nonlinear response properties different from the classical balanced state and can quantitatively reproduce a variety of major empirical findings regarding neural spatiotemporal dynamics. The phase transition was characterized by calculating the susceptibility and the branching parameter of spikes40. In the present study, we only change one parameter, the I–E ratio, to study the spatiotemporal properties (i.e., Lévy motion in space and theta oscillations in time) of the dynamical activity pattern emerging in this transition regime and their fundamental roles in probabilistic computations.
To study how the activity pattern samples non-trivial distributions, external stimuli are applied in the form of independent Poisson spike trains with specific firing rates. First, for sampling unimodal distributions, we apply an external feedforward input to each neuron i in the recurrent population as independent Poisson spikes whose firing rate is equal to
where σ = 0.6 is the width of the feedforward input, and c ≥ 0 is the contrast. When c = 0, the network receives the baseline background input with a uniform rate r0 = 0.85 kHz. The true value of the stimulus feature is fixed at s* = (0, 0) without loss of generality. Second, for sampling bimodal distribution, we apply an external feedforward input with firing rate equal to
The baseline firing rate is reduced to r0 = 0.5 kHz so that spontaneous activity (corresponding to when c1 = c2 = 0) is suppressed, ensuring a low-probability region between the modes of the sampled bimodal distribution. The true value of the stimulus feature s* is unknown by the observer and could be either of the alternative interpretations s1 or s2. We investigate the performance of our network in sampling bimodal distributions by varying the stimulus contrast levels c1 and c2 and the modal separation Δs = ∥s1 − s2∥.
To embed prior in the recurrent synaptic connectivity, we apply a spatially heterogeneous perturbation to the excitatory-to-excitatory synaptic weight
where si = (xi, yi) is the coordinates of neuron i in the feature space, c0 is the magnitude of the weight perturbation, and ϕ is a two-dimensional bell-shaped function
centered at s0 = (x0, y0) with the width parameter d. In this study, we fix s0 = (0, 0) and d = 1.3.
Our spatially extended network exhibits localized spiking activity patterns with complex spatiotemporal dynamics. To characterize the dynamics of the pattern, we first track its center of mass (CoM), ({hat{{{{{{{{bf{s}}}}}}}}}}_{t}=({hat{x}}_{t},{hat{y}}_{t})) based on the population vector of excitatory neurons77
where xj ∈ [−π, π) is the x-coordinate of the jth excitatory neuron, nj is the spike count over a small time window [t − τ, t) (here we set τ = 15 ms), i is the imaginary unit, and Arg is the principal complex argument in [−π, π). The y-coordinate of the CoM ({hat{y}}_{t}) is defined similarly. From a population coding point of view, Eq. (14) is also known as a complex estimator, which is equivalent to fitting a cosine function to the spike counts in the least-square sense78. In our study, the trajectory of the CoM of the activity pattern is the sample path of FNS.
Based on the trajectories of CoM, we next calculate the increments and the MSD. Specifically, the raw coordinates ({hat{{{{{{{{bf{s}}}}}}}}}}_{t}) over the periodic space are first unwrapped to linear coordinates in order to accommodate distance calculations across the periodic boundary. The mean-squared displacement of the CoM is then ({{{{{{{rm{MSD}}}}}}}}({{Delta }}t)=langle parallel {hat{{{{{{{{bf{s}}}}}}}}}}_{t+{{Delta }}t}-{hat{{{{{{{{bf{s}}}}}}}}}}_{t}{parallel }^{2}rangle), where Δt is the time lag. For fitting the increment distribution, we set ΔT = 15 ms to ensure no temporal correlation is artificially introduced due to overlaps in the spike count time window. The increments are then fitted to a symmetric Lévy stable distribution ({{{{{{{mathcal{S}}}}}}}}alpha {{{{{{{mathcal{S}}}}}}}}(alpha,gamma )) with the tail exponent α ∈ (0, 2] and the scale parameter γ > 0, as defined by Eq. (15). The parameters of the distribution are estimated with maximum likelihood. The MSD is averaged across 100 independent realizations of networks, each collecting 50 s of samples, whereas the increment histogram is produced with samples collected over 10 s, repeated across 5 random realizations of the network.
To characterize the state transition from the asynchronous to propagating wave state, we calculate local pairwise correlations of spike counts. We first divide the two-dimensional network into non-overlapping local patches of size 9 × 9 grid units and then calculate the pairwise spike count correlation within each patch. Finally, the mean local pairwise correlation is obtained by averaging the correlation coefficients of individual patches (49 patches in total). A weak local pairwise correlation indicates asynchronous activity without any structured patterns, whereas a strong local pairwise correlation indicates a spatially coherent wave pattern. The spike count time window is set to 100 ms with a total simulation time of 10 s repeated over 5 independent trials.
To understand the algorithmic nature of FNS implemented by our spiking circuit model, we develop a mathematical model based on stochastic dynamics driven by Lévy motion, rather than Brownian motion as in the classical Langevin MCMC. We consider a class of Lévy motion ({L}_{t}^{alpha }) whose increments follow symmetric Lévy stable distributions, denoted as ({{{{{{{mathcal{S}}}}}}}}alpha {{{{{{{mathcal{S}}}}}}}}(alpha,gamma )), with a tail index α and a scale parameter γ. For clarity and analytical tractability, we restrict our discussion to the one-dimensional case. The probability density of ({{{{{{{mathcal{S}}}}}}}}alpha {{{{{{{mathcal{S}}}}}}}}) can be expressed as28
The probability density exhibits a heavy tail, with power-law asymptote of the form79
for all 0 < α < 2. Note that α = 1 corresponds to the special case of a Cauchy distribution and α = 2 corresponds to a Gaussian distribution. The increment ({{Delta }}{L}_{t}^{alpha }={L}_{t}^{alpha }-{L}_{t^{prime} }^{alpha }) with ({{Delta }}t=t-t^{prime} > , 0) follows ({{{{{{{mathcal{S}}}}}}}}alpha {{{{{{{mathcal{S}}}}}}}}(alpha,{{{Delta }}t}^{frac{1}{alpha }})).
To derive the mathematical model described by Eq. (2), we start with the general formula presented in ref. 80
where ({{{{{{{{bf{z}}}}}}}}}_{{{{{{{{bf{t}}}}}}}}}={({x}_{t},{v}_{t})}^{T}), D is a positive semi-definite matrix describing the fractional diffusion, and Q is a skew-symmetric matrix describing the interplay between position xt and momentum vt. The temporal evolution of the probability density p(z, t) is given by the fractional Fokker-Planck equation,
where the fractional Riesz derivative ({{{{{{{{mathcal{D}}}}}}}}}^{alpha }) is defined as
through the Fourier transform ({{{{{{{mathcal{F}}}}}}}}[f]=int d{{{{{{{bf{z}}}}}}}}{e}^{-i{{{{{{{bf{k}}}}}}}}cdot {{{{{{{bf{z}}}}}}}}}f({{{{{{{bf{z}}}}}}}})). Fractional order derivatives generalize the notion of differentiation to fractional orders and are powerful mathematical tools for describing complex dynamics81; recently, fractional Fokker-Planck equations have been used to explain how non-Gaussian neural dynamics emerge from biologically realistic neural circuits82. It can be shown that the temporal evolution of p(z, t) converges to a stationary distribution (pi ({{{{{{{bf{z}}}}}}}})propto phi ({{{{{{{bf{z}}}}}}}})=exp (-H({{{{{{{bf{z}}}}}}}}))=exp (-U(x)-T(v))), related to the drift term by
To retain both the efficient sampling provided by the Hamiltonian dynamics as well as the powerful mixing ability provided by the fractional dynamics, we set ({{{{{{{bf{D}}}}}}}}=big[begin{array}{ll}gamma &0\ 0&0end{array}big]) and ({{{{{{{bf{Q}}}}}}}}=big[begin{array}{ll}0&-beta \ beta &0end{array}big]) and obtain
Although convergence to the stationary distribution is guaranteed mathematically, numerically simulating this set of equations is challenging since there are no straightforward ways for evaluating the Riesz fractional derivative with two dimensions. To overcome this numerical challenge, we exploit the fact that we are only interested in sampling the target distribution of π(x) instead of the joint stationary distribution of π(x, v) and make two simplifications as follows. First, we approximate the Riesz fractional derivative ({{{{{{{{mathcal{D}}}}}}}}}^{alpha }) with its partial versions ({{{{{{{{mathcal{D}}}}}}}}}_{x}^{alpha }) and ({{{{{{{{mathcal{D}}}}}}}}}_{v}^{alpha }), such that
The partial Riesz fractional derivative can then be effectively evaluated using a fractional centered difference scheme83,84. We find that this approximation in practice provides the correct sample distribution through numerical simulation. Second, by setting (psi (v)=exp (-T(v))) to be a symmetric α-stable distribution SαS(1/α), it holds that
By applying these two simplifications, we obtain the fractional Hamiltonian dynamics for modeling fractional neural sampling (Eq. (2)). As we show numerically in Results, the distribution of the samples generated by Eq. (2) converges to the target distribution π(x). For all simulations we use the Euler-Maruyama scheme with a step size of Δt = 0.001 with γ = 1, and apply a clipping (tilde{b}(x)approx min (b(x),{b}_{max })) with ({b}_{max }=500) to avoid numeric overflow of the Riesz derivative near low-probability regions.
A number of properties about the Riesz derivative are worth noting. First, when α = 2, the Riesz fractional derivative coincides with the ordinary second-order derivative with a change of sign, that is, ({{{{{{{{mathcal{D}}}}}}}}}^{2}=-{partial }_{xx}-{partial }_{vv}). Second, unlike integer-order derivatives, the Riesz derivative cannot be broken down into the sum of the partial derivatives in each of its dimension, that is, ({{{{{{{{mathcal{D}}}}}}}}}^{alpha }, ne , {{{{{{{{mathcal{D}}}}}}}}}_{x}^{alpha }+{{{{{{{{mathcal{D}}}}}}}}}_{v}^{alpha }), as the right-hand side of this equation cannot preserve isotropy in the space spanned by (x, v).
To characterize the ability of FNS for sampling bimodal distributions, we calculate how the mean exit time, which measures the average duration the sampler spends near one of the modes, changes as a function of modal separation. Suppose that the two modes of a bimodal distribution are centered at s1 and s2, then the exit time τexit from the first mode to the second mode can be defined as the duration the sample trajectory remains closer to s1 than to s2
where the ({hat{{{{{{{{bf{s}}}}}}}}}}_{0}) is any point satisfying (parallel {hat{{{{{{{{bf{s}}}}}}}}}}_{{0}_{+}}-{{{{{{{{bf{s}}}}}}}}}_{1}parallel le parallel {hat{{{{{{{{bf{s}}}}}}}}}}_{{0}_{-}}-{{{{{{{{bf{s}}}}}}}}}_{2}parallel) with ({hat{{{{{{{{bf{s}}}}}}}}}}_{{0}_{pm }}) denoting the one-sided limits (mathop{lim }nolimits_{tto pm 0}{hat{{{{{{{{bf{s}}}}}}}}}}_{t}). The mean exit time is then calculated as the average of all exit times in a sample trajectory. This definition applies to both the 1D mathematical model and the 2D neural circuit implementation. For the mathematical model, we use an open boundary condition rather than a periodic boundary condition in order to highlight the impact of modal separation on sampling. For the case of periodic boundary condition, the exit time increases similarly for small modal separations but saturates for large modal separations.
In the context of perceptual bistability, the exit time can be interpreted as perceptual dominance duration which is known to follow a right-skewed distribution. We fit the exit time calculated from the neural circuit model to two candidate distributions using the method of maximum likelihood. The first one is the Burr distribution with probability density function
with parameters found to be λ = 65.0 ms (95% CI 62.7–65.3), c = 8.05 (95% CI 7.58–8.55), and k = 0.528 (95% CI 0.476–0.586). The second one is the gamma distribution with probability density
with parameters found to be k = 10.4 (95% CI 9.9–10.9) and θ = 7.51 ms (95% CI 7.14–7.88). Note that the former features a power-law tail (with an exponent equal to −1 − ck) whereas the latter features a faster decaying exponential tail. Outliers (dominance duration <29 ms) are occasionally produced during the asynchronous phase of the spiking pattern and thus are omitted for this analysis.
To gain theoretical insights into how the drift term and sampled distribution are related to sensory input and synaptic weights, we consider a continuous neural field model85, which captures some features of the spiking circuit model, including the distance-dependent synaptic coupling and a localized activity pattern (bump activity) that moves randomly in the presence of noise. The continuous neural field model with one spatial dimension is described as
where u(x, t) represents the synaptic current, (J(x !,, x^{prime} )) is the synaptic coupling function, ρ is the neural density, I(x) is the feedforward input current, and F is a quadratic neural activation with divisive normalization. We consider the following form of synaptic coupling (J(x !,, x^{prime} )=bar{J}(x-x^{prime} )+{c}_{0}tilde{J}(x !,, x^{prime} ;{s}_{0})), where (bar{J}) is a translation-invariant synaptic coupling function and (tilde{J}) is a heterogeneous synaptic perturbation centered around s0, and c0 is the strength of the perturbation. Similarly, the feedforward input is (I(x)=bar{I}+{c}_{1}tilde{I}(x;{s}_{1})+xi (x),) where (bar{I}) is a constant background input, (tilde{I}(x;{s}_{1})) is a stimulus-dependent feedforward input centered at s1, c1 represents stimulus contrast, and ξ(x) is an additive noise; see Supplementary Information for the detailed definitions of these terms. By applying a projection method, we obtain an explicit analytical expression for the drift term
where a is the width of the homogeneous synaptic coupling, d1 is the width of the feedforward input, d0 is the width of the synaptic perturbation, and r0 is the height of the localized activity. The corresponding sampled distribution is bimodal when s1 and s0 are far apart, in which case we can apply Laplace approximation to cast it into the form of a Gaussian mixture
where g(x; μ, σ2) denotes a Gaussian distribution with mean μ and variance σ2. The variance of each of the modes is inversely proportional to the strength of perturbation c1 or c0, with the constant of proportionality equal to ({kappa }_{1}=frac{gamma {c}_{alpha }tau {({a}^{2}+{d}_{1}^{2})}^{3/2}}{2sqrt{2}{a}^{2}{d}_{1}}), or ({kappa }_{0}=frac{gamma {c}_{alpha }tau {({a}^{2}+{d}_{0}^{2})}^{2}}{4sqrt{pi }{a}^{3}{d}_{0}^{2}{r}_{0}}), respectively. The mixture proportion is determined by ({w}_{1}=sqrt{2pi {kappa }_{1}/{c}_{1}}exp Big(frac{{c}_{1}}{gamma {c}_{alpha }tau }frac{4sqrt{2}{a}^{2}{d}_{1}}{{({a}^{2}+{d}_{1}^{2})}^{1/2}}Big)) and ({w}_{0}=sqrt{2pi {kappa }_{0}/{c}_{0}}exp big(frac{{c}_{0}}{gamma {c}_{alpha }tau }frac{8sqrt{pi }{a}^{3}{d}_{0}^{2}{r}_{0}}{3({a}^{2}+{d}_{0}^{2})}big)). See Supplementary Information for the details of the mathematical derivation of Eq. (28) and Eq. (29).
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The local field potential data used in this study are available on Zenodo86 (https://doi.org/10.5281/zenodo.6806648). Source data are provided with this paper.
The code for simulations and analyses of the spiking neural circuit model87 (https://doi.org/10.5281/zenodo.6806437) and the code for fractional Hamiltonian Monte Carlo sampling88 (https://doi.org/10.5281/zenodo.6799461) are available without restrictions on Github.
Softky, W. R. & Koch, C. The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs. J. Neurosci. 13, 334–350 (1993).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Tomko, G. J. & Crapper, D. R. Neuronal variability: non-stationary responses to identical visual stimuli. Brain Res. 79, 405–418 (1974).
CAS  PubMed  Article  Google Scholar 
Churchland, M. M. et al. Stimulus onset quenches neural variability: a widespread cortical phenomenon. Nat. Neurosci. 13, 369–378 (2010).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Pouget, A., Beck, J. M., Ma, W. J. & Latham, P. E. Probabilistic brains: knowns and unknowns. Nat. Neurosci. 16, 1170–1178 (2013).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Fiser, J., Berkes, P., Orbán, G. & Lengyel, M. Statistically optimal perception and learning: from behavior to neural representations. Trends Cogn. Sci. 14, 119–130 (2010).
PubMed  PubMed Central  Article  Google Scholar 
Mach, E. Contributions to the Analysis of the Sensations (ed. Williams, C. M.) (Open Court Publishing Co., 1980).
Von Helmholtz, H. Physiological Optics, Vol. III: The Perceptions of Vision (ed. Southall, J. P.) Vol. 3 (Optical Society of America, 1925).
Knill, D. C. & Richards, W. (eds.) Perception as Bayesian Inference (Cambridge University Press, 1996).
Yuille, A. & Kersten, D. Vision as Bayesian inference: analysis by synthesis? Trends Cogn. Sci. 10, 301–308 (2006).
PubMed  Article  Google Scholar 
Griffiths, T. L., Steyvers, M. & Tenenbaum, J. B. Topics in semantic representation. Psychol. Rev. 114, 211 (2007).
PubMed  Article  Google Scholar 
Vul, E., Goodman, N., Griffiths, T. L. & Tenenbaum, J. B. One and done? Optimal decisions from very few samples. Cogn. Sci. 38, 599–637 (2014).
PubMed  Article  Google Scholar 
Wolpert, D. M. Probabilistic models in human sensorimotor control. Hum Movement Sci 26, 511–524 (2007).
Article  Google Scholar 
Ma, W. J., Beck, J. M., Latham, P. E. & Pouget, A. Bayesian inference with probabilistic population codes. Nat. Neurosci. 9, 1432–1438 (2006).
CAS  PubMed  Article  Google Scholar 
Orbán, G., Berkes, P., Fiser, J. & Lengyel, M. Neural variability and sampling-based probabilistic representations in the visual cortex. Neuron 92, 530–543 (2016).
PubMed  PubMed Central  Article  CAS  Google Scholar 
Buesing, L., Bill, J., Nessler, B. & Maass, W. Neural dynamics as sampling: a model for stochastic computation in recurrent networks of spiking neurons. PLOS Comput. Biol. 7, e1002211 (2011).
ADS  MathSciNet  CAS  PubMed  PubMed Central  Article  Google Scholar 
Moreno-Bote, R., Knill, D. C. & Pouget, A. Bayesian sampling in visual perception. Proc. Natl. Acad. Sci. USA 108, 12491–12496 (2011).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 
Hoyer, P. & Hyvärinen, A. In Advances in Neural Information Processing Systems (eds. Becker, S., Thrun, S. & Obermayer, K.) Vol. 15 (MIT Press, 2003).
Haefner, R. M., Berkes, P. & Fiser, J. Perceptual decision-making as probabilistic inference by neural sampling. Neuron 90, 649–660 (2016).
CAS  PubMed  Article  Google Scholar 
Zhang, W.-H., Lee, T. S., Doiron, B. & Wu, S. Distributed sampling-based Bayesian inference in coupled neural circuits. Preprint at bioRxiv https://doi.org/10.1101/2020.07.20.212126 (2020).
He, B. J. Scale-free brain activity: past, present, and future. Trends Cogn. Sci. 18, 480–487 (2014).
PubMed  PubMed Central  Article  Google Scholar 
Donoghue, T. et al. Parameterizing neural power spectra into periodic and aperiodic components. Nat. Neurosci. 23, 1655–1665 (2020).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Townsend, R. G. & Gong, P. Detection and analysis of spatiotemporal patterns in brain activity. PLOS Comput. Biol. 14, e1006643 (2018).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 
Muller, L., Chavane, F., Reynolds, J. & Sejnowski, T. J. Cortical travelling waves: mechanisms and computational principles. Nat. Rev. Neurosci. 19, 255 (2018).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Ruderman, D. L. & Bialek, W. Statistics of natural images: scaling in the woods. Phys. Rev. Lett. 73, 814 (1994).
ADS  CAS  PubMed  Article  Google Scholar 
Munn, B. & Gong, P. Critical dynamics of natural time-varying images. Phys. Rev. Lett. 121, 058101 (2018).
ADS  CAS  PubMed  Article  Google Scholar 
MacKay, D. J. Information Theory, Inference and Learning Algorithms (Cambridge University Press, 2003).
Sanborn, A. N. & Chater, N. Bayesian brains without probabilities. Trends Cogn. Sci. 20, 883–893 (2016).
PubMed  Article  Google Scholar 
Klages, R., Radons, G. & Sokolov, I. M. Anomalous Transport: Foundations and Applications (John Wiley & Sons, 2008).
Ponce-Alvarez, A., Thiele, A., Albright, T. D., Stoner, G. R. & Deco, G. Stimulus-dependent variability and noise correlations in cortical MT neurons. Proc. Natl. Acad. Sci. USA 110, 13162–13167 (2013).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 
Fiebelkorn, I. C. & Kastner, S. A rhythmic theory of attention. Trends Cogn. Sci. 23, 87–101 (2019).
PubMed  Article  Google Scholar 
Brascamp, J. W., Klink, P. C. & Levelt, W. J. The ‘laws’ of binocular rivalry: 50 years of Levelt’s propositions. Vision Res. 109, 20–37 (2015).
CAS  PubMed  Article  Google Scholar 
Laquitaine, S. & Gardner, J. L. A switching observer for human perceptual estimation. Neuron 97, 462–474 (2018).
CAS  PubMed  Article  Google Scholar 
Caruso, V. C. et al. Single neurons may encode simultaneous stimuli by switching between activity patterns. Nat. Commun. 9, 1–16 (2018).
ADS  CAS  Article  Google Scholar 
Rich, E. L. & Wallis, J. D. Decoding subjective decisions from orbitofrontal cortex. Nat. Neurosci. 19, 973–980 (2016).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Meister, M. L. R., Hennig, J. A. & Huk, A. C. Signal multiplexing and single-neuron computations in lateral intraparietal area during decision-making. J. Neurosci. 33, 2254–2267 (2013).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Horvát, S. et al. Spatial embedding and wiring cost constrain the functional layout of the cortical network of rodents and primates. PLOS Biology 14, e1002512 (2016).
PubMed  PubMed Central  Article  CAS  Google Scholar 
Levy, R. B. & Reyes, A. D. Spatial profile of excitatory and inhibitory synaptic connectivity in mouse primary auditory cortex. J. Neurosci. 32, 5609–5619 (2012).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Okun, M. & Lampl, I. Instantaneous correlation of excitation and inhibition during ongoing and sensory-evoked activities. Nat. Neurosci. 11, 535–537 (2008).
CAS  PubMed  Article  Google Scholar 
Garg, A. K., Li, P., Rashid, M. S. & Callaway, E. M. Color and orientation are jointly coded and spatially organized in primate primary visual cortex. Science 364, 1275–1279 (2019).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 
Gu, Y., Qi, Y. & Gong, P. Rich-club connectivity, diverse population coupling, and dynamical activity patterns emerging from local cortical circuits. PLOS Comput. Biol. 15, 1–34 (2019).
Article  CAS  Google Scholar 
Viswanathan, G. M. et al. Lévy flight search patterns of wandering albatrosses. Nature 381, 413–415 (1996).
ADS  CAS  Article  Google Scholar 
Viswanathan, G. et al. Optimizing the success of random searches. Nature 401, 911–914 (1999).
ADS  CAS  PubMed  Article  Google Scholar 
Harris, T. H. et al. Generalized Lévy walks and the role of chemokines in migration of effector CD8+ T cells. Nature 486, 545–548 (2012).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 
Shlesinger, M. F., West, B. J. & Klafter, J. Lévy dynamics of enhanced diffusion: application to turbulence. Phys. Rev. Lett. 58, 1100–1103 (1987).
ADS  MathSciNet  CAS  PubMed  Article  Google Scholar 
Liu, Y., Long, X., Martin, P. R., Solomon, S. G. & Gong, P. Lévy walk dynamics explain gamma burst patterns in primate cerebral cortex. Commun. Biol. 4, 739 (2021).
PubMed  PubMed Central  Article  Google Scholar 
Pfeiffer, B. E. & Foster, D. J. Autoassociative dynamics in the generation of sequences of hippocampal place cells. Science 349, 180–183 (2015).
ADS  CAS  PubMed  Article  Google Scholar 
Andrieu, C., De Freitas, N., Doucet, A. & Jordan, M. I. An introduction to MCMC for machine learning. Mach. Learn. 50, 5–43 (2003).
MATH  Article  Google Scholar 
Zaburdaev, V., Denisov, S. & Klafter, J. Lévy walks. Rev. Modern Phys. 87, 483 (2015).
ADS  MathSciNet  CAS  Article  Google Scholar 
Berg, R. W. & Kleinfeld, D. Rhythmic whisking by rat: retraction as well as protraction of the vibrissae is under active muscular control. J. Neurophysiol. 89, 104–117 (2003).
PubMed  Article  Google Scholar 
Festa, D., Aschner, A., Davila, A., Kohn, A. & Coen-Cagli, R. Neuronal variability reflects probabilistic inference tuned to natural image statistics. Nat. Commun. 12, 1–11 (2021).
Article  CAS  Google Scholar 
Chen, G. & Gong, P. A spatiotemporal mechanism of visual attention: superdiffusive motion and theta oscillations of neural population activity patterns. Sci. Adv. 8, eabl4995 (2022).
ADS  PubMed  PubMed Central  Article  Google Scholar 
Helfrich, R. F. et al. Neural mechanisms of sustained attention are rhythmic. Neuron 99, 854–865 (2018).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Xue, M., Atallah, B. V. & Scanziani, M. Equalizing excitation–inhibition ratios across visual cortical neurons. Nature 511, 596–600 (2014).
ADS  CAS  PubMed  PubMed Central  Article  Google Scholar 
Paczuski, M., Maslov, S. & Bak, P. Avalanche dynamics in evolution, growth, and depinning models. Phys. Rev. E 53, 414 (1996).
ADS  CAS  Article  Google Scholar 
Matsuoka, K. Sustained oscillations generated by mutually inhibiting neurons with adaptation. Biol. Cybernet. 52, 367–376 (1985).
MathSciNet  CAS  MATH  Article  Google Scholar 
Moldakarimov, S., Rollenhagen, J. E., Olson, C. R. & Chow, C. C. Competitive dynamics in cortical responses to visual stimuli. J. Neurophysiol. 94, 3388–3396 (2005).
PubMed  Article  Google Scholar 
Neal, R. M. In Handbook of Markov Chain Monte Carlo (2011).
Aitchison, L. & Lengyel, M. The Hamiltonian brain: efficient probabilistic inference with excitatory-inhibitory neural circuit dynamics. PLOS Comput. Biol. 12, e1005186 (2016).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 
Savin, C. & Denève, S. In Advances in Neural Information Processing Systems (2014).
Girolami, M. & Calderhead, B. Riemann manifold Langevin and Hamiltonian Monte Carlo methods. J. R. Stat. Soc. B 73, 123–214 (2011).
MathSciNet  MATH  Article  Google Scholar 
Wozny, D. R., Beierholm, U. R. & Shams, L. Probability matching as a computational strategy used in perception. PLOS Comput. Biol. 6, 1–7 (2010).
MathSciNet  Article  CAS  Google Scholar 
Laing, C. R. & Chow, C. C. A spiking neuron model for binocular rivalry. J. Comput. Neurosci. 12, 39–53 (2002).
PubMed  Article  Google Scholar 
Brascamp, J. W., van Ee, R., Pestman, W. R. & van den Berg, A. V. Distributions of alternation rates in various forms of bistable perception. J. Vis. 5, 1–1 (2005).
Article  Google Scholar 
Moreno-Bote, R., Rinzel, J. & Rubin, N. Noise-induced alternations in an attractor network model of perceptual bistability. J. Neurophysiol. 98, 1125–1139 (2007).
PubMed  Article  Google Scholar 
Echeveste, R., Aitchison, L., Hennequin, G. & Lengyel, M. Cortical-like dynamics in recurrent circuits optimized for sampling-based probabilistic inference. Nat. Neurosci. 23, 1138–1149 (2020).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Otero-Millan, J., Troncoso, X. G., Macknik, S. L., Serrano-Pedraza, I. & Martinez-Conde, S. Saccades and microsaccades during visual fixation, exploration, and search: foundations for a common saccadic generator. J. Vis. 8, 21–21 (2008).
PubMed  Article  Google Scholar 
Brockmann, D. & Geisel, T. The ecology of gaze shifts. Neurocomputing 32, 643–650 (2000).
Article  Google Scholar 
Lisman, J. & Jensen, O. The theta-gamma neural code. Neuron 77, 1002–1016 (2013).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Gelfand, A. E. & Smith, A. F. Sampling-based approaches to calculating marginal densities. J. Amer. Statist. Assoc. 85, 398–409 (1990).
MathSciNet  MATH  Article  Google Scholar 
Bak, P., Tang, C. & Wiesenfeld, K. Self-organized criticality. Phys. Rev. A 38, 364 (1988).
ADS  MathSciNet  CAS  MATH  Article  Google Scholar 
di Santo, S., Villegas, P., Burioni, R. & Muñoz, M. A. Landau–Ginzburg theory of cortex dynamics: scale-free avalanches emerge at the edge of synchronization. Proc. Natl. Acad. Sci. USA 115, E1356–E1365 (2018).
PubMed  PubMed Central  Article  CAS  Google Scholar 
Fontenele, A. J. et al. Criticality between cortical states. Phys. Rev. Lett. 122, 208101 (2019).
ADS  CAS  PubMed  Article  Google Scholar 
Chen, G. & Gong, P. Computing by modulating spontaneous cortical activity patterns as a mechanism of active visual processing. Nat. Commun. 10, 4915 (2019).
ADS  PubMed  PubMed Central  Article  CAS  Google Scholar 
Ackley, D. H., Hinton, G. E. & Sejnowski, T. J. A learning algorithm for Boltzmann machines. Cogn. Sci. 9, 147–169 (1985).
Article  Google Scholar 
Hinton, G. E., Osindero, S. & Teh, Y.-W. A fast learning algorithm for deep belief nets. Neural Comput. 18, 1527–1554 (2006).
MathSciNet  PubMed  MATH  Article  Google Scholar 
Neal, R. M. Sampling from multimodal distributions using tempered transitions. Stat. Comput. 6, 353–366 (1996).
Article  Google Scholar 
Georgopoulos, A., Kalaska, J., Caminiti, R. & Massey, J. On the relations between the direction of two-dimensional arm movements and cell discharge in primate motor cortex. J. Neurosci. 2, 1527–1537 (1982).
CAS  PubMed  PubMed Central  Article  Google Scholar 
Pouget, A., Zhang, K., Deneve, S. & Latham, P. E. Statistically efficient estimation using population coding. Neural Comput. 10, 373–401 (1998).
CAS  PubMed  Article  Google Scholar 
Nolan, J. P. Univariate Stable Distributions: Models for Heavy-Tailed Data (Springer International Publishing, 2020).
Ye, N. & Zhu, Z. Stochastic fractional Hamiltonian Monte Carlo. In International Joint Conference on Artificial Intelligence (2018).
West, B. J. Colloquium: Fractional calculus view of complexity: a tutorial. Rev. Mod. Phys. 86, 1169 (2014).
ADS  Article  Google Scholar 
Wardak, A. & Gong, P. Fractional diffusion theory of balanced heterogeneous neural networks. Phys. Rev. Res. 3, 013083 (2021).
CAS  Article  Google Scholar 
Şimşekli, U. Fractional Langevin Monte Carlo: exploring Lévy driven stochastic differential equations for Markov chain Monte Carlo. Proc. Mach. Learn. Res. 70, 3200–3209 (2017).
Google Scholar 
Li, C. & Zeng, F. Finite difference methods for fractional differential equations. Int. J. Bifurcation Chaos 22, 1230014 (2012).
ADS  MathSciNet  MATH  Article  Google Scholar 
Fung, C. C. A., Wong, K. Y. M. & Wu, S. A moving bump in a continuous manifold: a comprehensive study of the tracking dynamics of continuous attractor neural networks. Neural Comput. 22, 752–792 (2010).
MathSciNet  PubMed  MATH  Article  Google Scholar 
Qi, Y. & Gong, P. Fractional neural sampling as a theory of spatiotemporal probabilistic computation in neural circuits, local field potentials. Zenodo, https://doi.org/10.5281/zenodo.6806648 (2022).
Qi, Y. & Gong, P. Fractional neural sampling as a theory of spatiotemporal probabilistic computation in neural circuits, fractional Hamiltonian Monte Carlo sampler. Github, https://doi.org/10.5281/zenodo.6806437 (2022).
Qi, Y. & Gong, P. Fractional neural sampling as a theory of spatiotemporal probabilistic computation in neural circuits, SpikeNet. Github, https://doi.org/10.5281/zenodo.6799461 (2022).
Download references
This work was supported by Australian Research Council Grants DP160104316 (P.G.).
School of Physics, University of Sydney, Sydney, NSW, 2006, Australia
Yang Qi & Pulin Gong
Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, 200433, China
Yang Qi
Key Laboratory of Computational Neuroscience and Brain-Inspired Intelligence (Fudan University), Ministry of Education, Shanghai, China
Yang Qi
MOE Frontiers Center for Brain Science, Fudan University, Shanghai, 200433, China
Yang Qi
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Conceptualization: Y.Q. and P.G.; methodology: Y.Q. and P.G.; investigation: Y.Q. and P.G.; writing—original draft: Y.Q. and P.G.; writing—review & editing: Y.Q. and P.G.; funding acquisition: P.G.; supervision: P.G.
Correspondence to Pulin Gong.
The authors declare no competing interests.
Nature Communications thanks Biyu He and the other anonymous reviewer(s) for their contribution to the peer review of this work
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Qi, Y., Gong, P. Fractional neural sampling as a theory of spatiotemporal probabilistic computations in neural circuits. Nat Commun 13, 4572 (2022). https://doi.org/10.1038/s41467-022-32279-z
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-022-32279-z
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Focus
Advertisement
Advanced search
Nature Communications (Nat Commun) ISSN 2041-1723 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

WHAT DOES FAMILY MEAN? EIGHT LOCAL FILMMAKERS OFFER UNIQUE PERSPECTIVE DURING FREE SCREENING EVENT AT MDC'S TOWER THEATER MIAMI AUGUST 25 – Miami Dade College

New “Family Commissions” program invested a total of $120,000 into local filmmakers to develop original new works
MIAMI (August 9, 2022) – Using vastly different storylines, eight Miami filmmakers will showcase their own perspective on what “family” means during a free screening at Miami Dade College’s (MDC) Tower Theater Miami from 7-9 p.m. on Thursday, August 25. The films are part of “Family Commissions,” a program of eight short films presented by Oolite Arts and MDC’s Miami Film Festival. Co-funded by the Lynn and Louis Wolfson II Family Foundation, each filmmaker received a share of $120,000 to produce original works using the theme of family.
“Alongside our friends at Oolite Arts, we are thrilled to be supporting the work of these immensely talented local filmmakers and helping them take this next step in advancing their promising careers,” said Nicolas Calzada, interim executive director of MDC’s Miami Film Festival and Tower Theater Miami. “These films are both entertaining and impressive. We can’t encourage the community enough to come out and experience the unique stories that these emerging filmmakers have to share.”
The stories include two teenagers’ quest to find an ancient good luck charm in hopes of saving their family booths and a flea market from closure; a first-person account of a filmmaker’s quarter-life crisis brought on by his ever-changing family dynamics, and the escapades of a young man who will stop at nothing to recover a stolen family heirloom.
“Our talented community of indie filmmakers has done it again, turning a topic we all embrace  – family – into an on-screen investigation of the human condition,” said Dennis Scholl, Oolite Arts’ president & CEO. “We’re delighted to support our local filmmakers by bringing their stories to the big screen in partnership with the Miami Film Festival and the Lynn and Louis Wolfson II Family Foundation.”
At the August 25 screening event, attendees will also have an opportunity to meet the eight filmmakers who had up to six months to work on their respective films. They are:
To RSVP to the FREE screening on August 25, visit https://www.towertheatermiami.com/event/mff-oolite-arts-family-commissions.
CALENDAR NOTICE
“Family Commissions” Works-in-Progress Screening
Eight short films by local filmmakers centered on the theme family
Presented by Oolite Arts and MDC’s Miami Film Festival
7-9 p.m. Thursday, August 25 at Tower Theater Miami
1508 SW 8th Street, Miami, FL 33135
RSVP: https://www.towertheatermiami.com/event/mff-oolite-arts-family-commissions
About Oolite Arts
Oolite Arts helps Miami-based artists advance their careers and inspires the cultural community to engage with their work. Established in 1984, the nonprofit is both a community and a resource, providing visual artists with the studio space, exhibition opportunities and financial support they need to experiment, grow and enrich the city. Through its educational programming, Oolite Arts helps Miamians learn about contemporary art and develop their own artistic skills. Coming in 2024—in time for the organization’s 40th anniversary—Oolite Arts will open a new campus in the City of Miami that will help build Miami’s next creative chapter. For more, visit oolitearts.org.
Exhibitions and programs at Oolite Arts are made possible with the support of the Miami-Dade County Department of Cultural Affairs and the Cultural Affairs Council; the Miami-Dade County Mayor and Board of County Commissioners; the Miami Beach Mayor and City Commissioners; the State of Florida, Department of State, Division of Arts and Culture and the Florida Council on Arts and Culture; the National Endowment for the Arts; the Jorge M. Perez Family Foundation at the Miami Foundation, the Lynn and Louis Wolfson II Family Foundation, and the John S. and James L. Knight Foundation. For more information, visit oolitearts.org. Follow @oolitearts on social media.
About Miami Dade College’s Miami Film Festival   
Miami Film Festival is an international film event that serves as a major film showcase for world cinema. Considered the preeminent film festival for highlighting Ibero-American cinema in the U.S, Miami Film Festival has become renowned for championing films made by the South Florida community, first-time feature filmmakers, and International Feature Film submissions to the Academy Awards. The annual Festival, produced and presented by Miami Dade College, attracts more than 60,000 audience members and more than 400 filmmakers, producers, talent, and industry professionals. Since 1984, the Festival has screened over 2,500 films from more than 75 countries and given out more than $2 million in cash awards to filmmakers and distributors. Miami Film Festival is the oldest major festival housed in a university or college and also operates a year-round art cinema, Tower Theater Miami and Miami Film Festival GEMS in November. For more information, visit miamifilmfestival.com or call 305-237-FILM (3456).
Oolite Arts + Miami Film Festival Media Relations Contacts:
Andrea Salazar, Pinzur Communications | 954-756-0652 | [email protected]
Rachel Pinzur, Pinzur Communications | 305-725-2875 | [email protected]

source

The 10 Best Restaurants In Gainesville, FL – Tasting Table

It’s great to be a Florida Gator. There’s the school spirit, warm weather, and the prestigious university accolades (one of the top public universities in the country). The University of Florida (UF) is in the small city of Gainesville, Florida, and the local food scene is thriving. Charming brunch spots perfect for your Instagram pictures, casual dinner places ideal for students balling on a budget, and cuisines from many different cultures are around this town. Sadly, some Gainesville classics like The Swamp Restaurant, Leonardo’s By The Slice, Larry’s Giant Subs, and Felipe’s Mexican Taqueria have closed due to the pandemic, reports The Gainesville Sun (though The Swamp is luckily being rebuilt in a different part of town). These places and others will all be missed. If you graduated from UF a few years ago and are wondering what the best spots in town are (that are still open or new spots) and want to avoid rolling up to Felipe’s only to find it closed, this article is for you.
If you’re going to be a new student at UF this fall and are looking to try some spots not in the Reitz Union or in Midtown, here are the best local food spots you need to check out with your parents during a move-in weekend or with friends. Exploring the city’s best places will make you feel more at home, and you can try some amazing food. Here are the 10 best restaurants in Gainesville, Florida.

If you walk through Downtown Gainesville at 11 a.m. and notice people with mimosas and plates full of delectable brunch tacos, eggs, and avocado toast, you’re looking at Original American Kitchen (OAK). This place has delicious brunch and dinner with elevated classics like wild mushroom ravioli, the best grilled cheese and tomato soup ever (no, seriously), fried green tomato sliders, and bottomless mimosas. Satisfy your sweet tooth with desserts like homemade bread pudding, pecan pie, and fried Oreos. You can sit either inside, outside, or at the bar if you choose.
Wherever you sit, you can bet the decor will be rustic and sweet. Tons of unique mirrors create a gallery wall at the back of the restaurant, a wall on the right features mason jars filled with flowers, and a large chalkboard wall has doodles and menu details. OAK is an absolute delight that will leave you coming back for more delicious meals and photos with friends. It’s also conveniently located Downtown, so once those mimosas and brunch food have filled you up, take a stroll around town and appreciate historic Gainesville. Pop in a record store, get an iced coffee from Wyatt’s Coffee, and scope out where you want to go for dinner later (or where to order from). What a perfect Saturday morning.

This restaurant is for all the folks who look forward to Publix’s $5 Wednesday sushi rolls (yum). One of the fancier dinner spots in town, this place has some of the highest quality sushi you’ll ever have (it also features a vegetarian menu, which provides options for a wide range of people). Dragonfly Sushi & Sake Company (commonly called Dragonfly Sushi) is not budget-friendly like most other places on this list, but it’s perfect for your special occasions. For example, when your family pops into town for parents’ weekend, graduation, or a football game, go to Dragonfly Sushi with them. It’ll be nice to get an elevated meal after months of Panda Express, Chipotle, and Wawa (absolutely no shade to any of those places, though). If you go with family, taking them to a local spot will make you feel like a local and probably impress them.
Set in Downtown, Dragonfly Sushi sits on an ideal corner right in the middle of everything. After getting absolutely stuffed from sushi, seafood dishes, and a few drinks, pop next door to The Hyppo Gourmet Ice Pops for some light and refreshing fruit pops. They come in unique flavors like blueberry lavender lemonade and mango cheesecake. Or, if you’re more of a dinner-and-a-show type, check out what’s playing at the Hippodrome Theatre one block away from the restaurant, as there is always a new show to see there.

If you’re looking for authentic Mexican food, La Tienda has the best in Gainesville. If you’re a new student looking to impress the friends you just made by taking them to a great local Mexican restaurant that’s not Chuy’s (sorry Chuy’s, we really do love you), go to La Tienda. You won’t have to stress about navigating the mildly stressful drive-through situation at El Indio (it has delicious Mexican food, but with only a few tables outside, it’s not a sit-down place). The food at La Tienda is affordable and offers high value. Though usually packed with hungry people ready for food, it is worth scoping out a table here because the tacos are unmatched.
This spot is on SW 13th Street past Shands Hospital, so it’s not super walkable from campus, but it’s a short two-minute drive (or hop on a scooter to get there quickly). We recommend getting either the enchiladas, soft tacos, or a burrito. La Tienda also has a vegetarian menu, so bring your veggie friends! Get an order of chips, salsa, and queso to share. You won’t regret it. La Tienda is open from 11 a.m. to 9:30 p.m. Monday to Thursday, 11 a.m. to 10 p.m. Friday and Saturday, and 11 a.m. to 9 p.m. on Sunday. If you don’t have time to stop here, you can always order the food to-go or opt for delivery!

Satchel’s Pizza is one of those places that you almost don’t believe exists until you see it for yourself. Mosaic tiling covers every inch of this pizza joint, from the floor to the ceiling. Tiles line the inside spaces, the exterior, the patio area, all the outdoor tables, and even the converted van. Getting a van seat is a coveted experience and a great photo opportunity. However, Satchel’s isn’t just about the quirky aesthetics — the food is incredible. If you’re a New York or New Jersey native who has found themselves in Florida, Satchel’s Pizza might have the best pizza in the state. Not only is the pie high-quality, but the sodas are homemade, and the desserts are decadent. Even if you’re not hungry for dessert, you have to ask for the dessert menu. It actually comes in a View-Master toy that you hold up to your eyes and use to click through images of the dessert options.
This restaurant is almost on the edge of town, near the airport, but it is worth the extra ten minutes of driving. Satchel’s also frequently hosts live music, so if you want the best pizza in town for an affordable price, a unique and fun atmosphere that will heal your inner child, and an experience you’ll be talking about for weeks to come, head to Satchel’s Pizza.

Just because Gainesville is two hours from the coast doesn’t mean you can’t find great seafood here. Harry’s Seafood Bar and Grille is a New Orleans-inspired spot with delicious seafood and steaks perfect for a celebratory dinner with family or friends. This restaurant is one of the more upscale ones in town, so it’s ideal for a birthday dinner, graduation celebration, or when family comes into town to visit. Located Downtown (can you tell how many good spots are Downtown?), Harry’s has signature crab cakes that are authentic and to-die-for. The New Orleans ribeye is tender and juicy, and the French-baked scallops are delicious.
If you don’t mind eating dinner a little earlier than the average person or are a big “linner” enthusiast, take advantage of Harry’s happy hour from 2 p.m. to 7 p.m. All draft beer is $3, premium cocktails and glasses of wine are $4, and many more discounts await. From the bar menu, we recommend the Bayou Bloody Mary for those who love a little kick or The Front Porch (made with Southern iced tea, bourbon, orange liqueur, and muddled mint) for people craving a sweeter beverage (it’s a spiked sweet tea). For dessert, sink into the mouthwatering bananas foster, key lime pie, or Oreo beignets. No matter what you order from the menu, it will be delicious at Harry’s Seafood Bar and Grille.

Don’t be fooled by the name. Flaco’s Cuban Bakery is serving up top-notch lunches and dinners. The Cuban sandwiches are the perfect simple lunch, and the arepas are a surefire hangover cure. Flaco’s, as it’s lovingly referred to, is the ideal drunk food if you’re out Downtown. After a Downtown bar crawl (Midtown is a different story), going from UC (University Club) to Arcade Bar to Cry Baby’s to Downtown Fats, end your night on a high note at Flaco’s. With a closing time of 2:30 a.m., this establishment knows that people need some delicious arepas, Cuban sandwiches, tacos, burritos, burrito bowls, or empanadas to settle their stomachs and end the night right. If you’re choosing Taco Bell instead of Flaco’s at 2 a.m., sorry, we’re judging you.
If you’re not a night owl, don’t worry because Flaco’s opens at 11 a.m., the perfect time for lunch on a sunny Gainesville day. Another great thing to note about Flaco’s is that it’s an ideal spot for students, with menu items averaging around $10. Not bad at all! If you’re out studying at a coffee shop Downtown and start to want something a bit more substantial than coffee, go to Flaco’s. You won’t be disappointed.

If the line wrapped around Luke’s New York Bagel Shop every morning is any indication, this small Gainesville bagel shop serves the most authentic New York bagels and breakfast sandwiches around. The shop exterior recently got a fire engine red facelift, so if you’re driving around trying to find the shop, you can’t miss it. Be sure to park in the neighboring lot, as the parking lot at Luke’s will feature a line of people waiting for their coffee and bagels. If you want the authentic Gainesville experience, park your car for free in the Depot Park parking lot right across the street. Then, walk over to Luke’s, grab your food, and set up a blanket in idyllic Depot Park to sit and make a picnic out of breakfast (don’t worry if you forgot a blanket; there are plenty of picnic tables in the park).
If you want Luke’s full range of bagel flavors, cream cheeses, and breakfast sandwiches, make a point to get there early, before 9 a.m. The breakfast spot opens at 7 a.m. and closes at 1 p.m., and if you’re an early riser anyway, the line isn’t too long at opening time. Around 8 a.m. is when it is the longest. But no matter the length of the line, the food will be worth it. Authentic, huge New York bagels with generous amounts of cream cheese await.

Mi Apá Latin Café (casually referred to as Mi Apá) is a sensation. Located right off NW 34th Street, Mi Apá is perfect for an after-class pick-me-up. If you’re going there for breakfast, get the cafe con leche and stuffed arepa (with scrambled eggs, ham, and cheese inside). If you’re popping by for lunch or dinner, we recommend ordering one of the entrees and definitely getting rice and beans on the side (if they don’t come with your meal already). The menu is vast but not overwhelming, meaning everyone will find something they like. Mi Apá’s prices are very student-friendly, and the drive-through line is efficient and easy, so if you’re a busy student on the go, this restaurant is perfect.
If you’re looking for a place to rejuvenate after some shopping on 34th Street, Mi Apá is right near all the shops and is the perfect place to stop and get a bite to eat. If you want to do something fun with your friends after lunch at Mi Apá, stop by Corks & Colors Studio to paint some pottery (located right in the same plaza). All in all, Mi Apá is a great local spot to check out if you’re on 34th Street or want to find good Cuban food in Gainesville.

Weekend brunch in Gainesville can mean you’re fighting everyone else to get a reservation somewhere, but the short wait at Afternoon is always worth it. This brunch spot is hidden away on NW 10th Avenue, but the quality is well-known to most people in town. Afternoon is a Pinterest girl’s dream. Mint green accents in the decor, rose gold fixtures, a record player churning out tunes, and a beautiful feature wall are just a few things that make the vibe in this restaurant excellent. However, the good vibes don’t stop at the decor; the food here is also phenomenal. The coffee options are top-tier, and the unique brunch options are incredible. For drinks apart from coffee, we recommend the matcha tea latte, a spritz (an alcoholic fizzy cocktail), or a butterfly flower lemonade. For your main course, some tried-and-true items are the Dutch baby, mushroom tartine, poached eggs, shakshuka, or the classic breakfast.
After your brunch at Afternoon, keep your car parked and walk over to the other side of the plaza into Serpentine Plants + Provisions. Owned by the same folks as Afternoon, this plant and gift shop is just as charming as the restaurant, and you’ll find some great coffee table books, wine, earrings, candle holders, and plants here.

Football isn’t for everybody, and that’s okay. If you’re not a football fan, you might not know the name Steve Spurrier. Spurrier is one of the most famous University of Florida football players in history, and he even won the legendary Heisman Trophy in 1966, per Heisman Trophy. After his playing days, he went on to be a college football coach. He eventually came back and coached at his alma mater and guided the Gators to win the SEC championship in 1990, 1991, 1993, 1994, 1995, 1996, and 2000. Spurrier is basically a Gainesville legend.
Steve Spurrier has since retired from coaching, but he wasn’t done with Gainesville just yet. In 2021, he opened Spurrier’s Gridiron Grille, a classy restaurant in the Celebration Point plaza. He does food as well as football. This upscale place offers to-die-for cocktails and appetizers (the orange and blue shrimp are great), and the entrees are just as unforgettable. This eatery is a bit pricey for students, but it’s the perfect place to bring your family members who were there to witness Steve Spurrier’s career. The food is excellent, and we recommend the Emory & Henry Seafood Pasta, Spurrier’s bacon cheeseburger, or a steak cooked to your liking. Desserts include cheesecake, lemon tarts, and chocolate chip cookies. The bar menu is extensive, even featuring signature cocktails and mocktails. We recommend the raspberry lemon drop martini or the spicy mango mocktail. Spurrier’s is also an ideal spot for those who are gluten-free.

source

Cutting-edge American technology given away to Chinese company : Consider This from NPR – NPR

MARY LOUISE KELLY, HOST:
Just north of Seattle near the shores of Puget Sound sits an empty warehouse in Mukilteo, Wash.
CHRIS HOWARD: We used to have 10 shipping containers here. There were empty containers back here, customers and clients coming for visits.
KELLY: Chris Howard used to work in that warehouse with more than a dozen other engineers and researchers for an American company called UniEnergy. The name is still on the sign out front. They were building a battery – not just any battery, something called a vanadium redox flow battery. It was about the size of a refrigerator. And Howard and the rest of the employees thought it was going to change the world.
HOWARD: It was more than a job. It was a lot of blood, sweat and tears into developing a product that we were really excited about and really proud of.
KELLY: Unlike batteries in cell phones, even cars, these batteries could charge and discharge energy for as long as 30 years. And this particular design seemed to hold enough energy to power a house. Researchers pictured people plunking them down next to their air conditioners, attaching solar panels to them, and everyone living happily ever after off the grid.
HOWARD: It was beyond promise. We were seeing it functioning as designed, as expected.
KELLY: They thought the batteries would be the next great American success story. That is not what happened. Today, that warehouse sits, shuttered, empty. All the employees who worked there were laid off. And those batteries, they are now being made at a factory in northeastern China. CONSIDER THIS – cutting edge technology developed in an American national lab, paid for with American tax dollars, was given away to a Chinese company. Using internal government documents, an NPR investigation tells the story for the very first time and shows how the U.S. may have lost the next big thing to China yet again.
(SOUNDBITE OF MUSIC)
KELLY: From NPR, I’m Mary Louise Kelly. It’s Tuesday, August 9.
It’s CONSIDER THIS from NPR. It’s a favorite promise of politicians – keep manufacturing jobs and technology in America.
(SOUNDBITE OF ARCHIVED RECORDING)
BARACK OBAMA: Thank you. Please, everybody, have a seat.
KELLY: Let’s go back to Feb. 22, 2011. It was a Tuesday, and then-President Barack Obama was in Cleveland, Ohio.
(SOUNDBITE OF ARCHIVED RECORDING)
OBAMA: The world wants American goods, and my administration is going to go to bat for America’s businesses around the world. You should know that.
(APPLAUSE)
KELLY: After shouting out LCD screens and fiber optic cables, Obama turned to an up and coming technology that seemed to show promise.
(SOUNDBITE OF ARCHIVED RECORDING)
OBAMA: And I have no idea what this is – vanadium redox fuel cells.
(LAUGHTER)
OBAMA: That’s one of the coolest things I’ve ever said out loud.
(LAUGHTER)
KELLY: Researchers at the federal government’s Pacific Northwest National Lab surely would have agreed. For the past five years, they had been developing vanadium redox technology, getting it ready for commercial use. And this came at a time when the Obama administration was making a major push to reinvigorate American manufacturing.
(SOUNDBITE OF ARCHIVED RECORDING)
OBAMA: So we have a huge opportunity at this moment to bring manufacturing back, but we have to seize it.
KELLY: That’s Obama at his State of the Union address the following year, 2012.
(SOUNDBITE OF ARCHIVED RECORDING)
OBAMA: Tonight, my message to business leaders is simple. Ask yourselves what you can do to bring jobs back to your country, and your country will do everything we can to help you succeed.
(APPLAUSE)
KELLY: That same year, UniEnergy Technologies was founded with a license to make the vanadium redox flow batteries developed at the National Lab. And as the company grew, claims about supporting American technology and manufacturing kept coming from the White House, regardless of which president lived there.
(SOUNDBITE OF ARCHIVED RECORDING)
DONALD TRUMP: We’re bringing manufacturing back to the United States big league. We’re reducing taxes.
KELLY: It was a particular fixation for Donald Trump during his time in office.
(SOUNDBITE OF ARCHIVED RECORDING)
TRUMP: I remember when I was growing up, I’d see made in America all over the place – a little bit made in the USA – but made in America. Everything had made in America. And we’re starting that again. And it’s happening again, as you know. We’re here today…
KELLY: And Trump’s successor would pick up that baton, too.
(SOUNDBITE OF ARCHIVED RECORDING)
PRESIDENT JOE BIDEN: And if we think we’re not in a race, well, guess what? Take a look at China. Take a look at China.
KELLY: President Biden last summer at a Mack Trucks facility in Pennsylvania.
(SOUNDBITE OF ARCHIVED RECORDING)
BIDEN: It’s a straightforward solution – support and grow more American-based companies.
(SOUNDBITE OF MUSIC)
KELLY: And yet, it was the federal government itself that let those cutting edge batteries slip away. NPR’s Laura Sullivan and Courtney Flatt from Public Radio’s Northwest News Network will pick up the story from here with their investigation into how vanadium redox flow batteries ended up being made in a factory in China.
LAURA SULLIVAN, BYLINE: The Chinese company didn’t steal this technology. It was given to them by the U.S. Department of Energy. An NPR investigation found the department allowed the technology and jobs to move overseas, violating its own licensing rules while failing to intervene on behalf of U.S. workers in multiple instances according to internal department emails. Now, China is forging ahead, investing millions into this cutting edge green technology that was supposed to help keep the U.S. and its economy out front.
JOANNE SKIEVASKI: It just is mind-boggling.
SULLIVAN: Joanne Skievaski is the vice president of finance for a U.S. company called Forever Energy that has been trying to get a license from the department to make the batteries here for more than a year.
SKIEVASKI: This is technology made from taxpayer dollars. It was invented by a national lab. And it’s deployed in China, and it’s held in China. To say it’s frustrating is an understatement.
SULLIVAN: Department of Energy officials declined NPR’s request for an interview and wouldn’t explain how technology that cost U.S. taxpayers $15 million ended up in China. But after NPR sent department officials detailed questions laying out the timeline of events, officials terminated the license it gave to the Chinese company. In a statement, officials said the department, quote, “takes American manufacturing obligations extremely seriously” and is now, quote, “undertaking an internal review of the licensing of vanadium battery technology.” The story of how this happened begins where the battery was born, three hours southwest of Seattle, in the basement of a government lab called the Pacific Northwest National Laboratory, where Courtney went to visit.
COURTNEY FLATT, BYLINE: Safety goggles on.
VINCE SPRENKLE: Yep.
FLATT: Vince Sprenkle works with energy storage here at the lab.
SPRENKLE: We’re going to go down into the redox flow battery lab.
FLATT: It was down here in 2006 that more than two dozen scientists began to suspect that a special mix of acid and electrolyte could hold unusual amounts of energy without degrading. They turned out to be right.
Do you feel kind of, like, on the cutting edge of learning about these batteries?
SPRENKLE: Yeah, we are. I mean, I think we’ve got one of the leading research groups in the country and probably the world in this technology.
FLATT: It’s because of this leading edge that when a success happens, the lab encourages scientists to go out and see if they can make and sell the inventions in the real world. The lab and the U.S. government still hold the patents because American taxpayers paid for the research. But the Department of Energy licenses the patents to scientists and companies willing to take a shot.
In this case, it took six years and millions of taxpayer dollars to discover the perfect battery recipe. Gary Yang was the lead scientist, and he was excited to see if he could make them. In 2012, he left the lab with the license in hand and started UniEnergy Technologies at the warehouse in Mukilteo, Washington.
GARY YANG: I left the lab – total legal process – and start UniEnergy Technology (ph) in Washington state.
FLATT: He hired engineers and researchers, but then he ran into trouble. He says he couldn’t find any U.S. investors.
YANG: I talked to almost all major investment bank. None of them invest in battery.
FLATT: So he turned to a Chinese businessman and a company called Dalian Rongke Power and its parent company, which agreed to invest and even help manufacture the batteries. And so began a slow shift. First, Chris Howard said, it was just some parts. Ultimately, it was the whole process.
HOWARD: Manufacturing was subsequently shifted to our sister company in China, and they would take on that role.
FLATT: In 2017, Yang and UniEnergy formalized the situation and gave Rongke Power an official sublicense, allowing the company to make the batteries.
SULLIVAN: So here’s the thing – companies can choose to manufacture in China, but in this case, Yang’s original license clearly says on page six, he has to sell batteries in the United States, and those batteries have to be, quote, “substantially manufactured” here. Yang acknowledges he didn’t do that. He was mostly selling batteries in China, and the batteries he did sell here were largely made in China. But he says in all those years, the department never raised any concerns or intervened.
Then, in 2019, Chris Howard said he and the other engineers were called to a conference room. Supervisors told them they, too, would have to go to China to work there for four months at a time at Rongke Power.
HOWARD: And that was set to be increased on the premise that there were certain government programs, Chinese government programs, that would support funding efforts. So it was unclear, certainly to myself and some of the other engineers, what the plan was.
SULLIVAN: In a statement, the department said that license monitoring is a priority and that a review of this case is underway – all of which brings us to Yang. Yang was born in China, but he is a U.S. citizen and got his Ph.D. at the University of Connecticut. He says he wanted to manufacture here, but at the time, China was doing more to encourage battery production. And he told us China could do it better.
YANG: In this field – manufacturing, engineering – China ahead of U.S.
SULLIVAN: China is ahead of the U.S.
YANG: They’re ahead of U.S. – many wouldn’t believe, engineering wise, ahead.
SULLIVAN: He says, far from helping China, the Chinese engineers were helping his U.S. employees. But you can see in several news reports at the time, it was helping China. The Chinese government launched several large demonstration projects and announced millions in funding.
FLATT: As things began to take off in China, here in the States, Yang was once again in financial trouble. So he made a decision that would again keep the technology from staying in the U.S. He transferred the license from UniEnergy to a company called Vanadis.
ROELOF PLATENKAMP: Vanadis is based in the Netherlands, and we will set up the holding company in Switzerland.
FLATT: Roelof Platenkamp is Vanadis’s founding partner. Platenkamp told NPR the company’s plan was to continue making the batteries in China and then set up a factory in Germany and eventually maybe the U.S. He says he has to manufacture in Europe because the European Union has strict rules about these things.
PLATENKAMP: I have to be a European company – yeah – or certainly a non-Chinese company in Europe.
FLATT: But the United States has these rules, too. And any transfer of a U.S. government license needs U.S. government approval.
SULLIVAN: Which Yang apparently had no trouble getting. We looked at department emails and found that last summer, on July 7, one of the top officials at UniEnergy wrote to a government manager at the Department of Energy Lab in Washington, saying they were going to make a deal with Vanadis. We believe they have the right blend of technical expertise, the official wrote. The manager wrote back that he would need confirmation. A second employee sent confirmation an hour and a half later, and the license was transferred.
Now, if anyone from the lab or the Department of Energy during that hour and a half thought to check whether Vanadis was an American company or whether it intended to manufacture in the United States is unclear. Even Vanadis’s website says it plans to make the batteries in China. Department of Energy officials told us they take all license transfer seriously and have recently closed significant loopholes. But they acknowledge their efforts rely to some extent on, quote, “good faith disclosures” by the companies, which means if companies like UniEnergy don’t say anything, the U.S. government may never know. It’s a problem government investigators found has been going on for years. In 2018, the Government Accountability Office found the department lacked resources to properly monitor its licenses, was relying on antiquated computer systems and didn’t have consistent policies across its labs.
FLATT: It was an American company, Forever Energy, that actually read the vanadium battery license and raised a red flag more than a year ago. Joanne Skievaski and others there say they repeatedly warned department officials that the UniEnergy license was not in compliance. Officials repeatedly told them it was.
SKIEVASKI: How is it that the national lab did not require U.S. manufacturing? Not only is it a violation of the license, it’s a violation to our country.
FLATT: Skievaski hopes that now that the department has revoked the license, Forever Energy will get a chance. They’re hoping to open a factory in Louisiana.
SKIEVASKI: We are ready to go with this technology.
SULLIVAN: Skievaski told us it will be hard at this point for any American company to catch up. Industry trade reports list Dalian Rongke Power as the No. 1 manufacturer of vanadium flow batteries worldwide. And the bigger question looming over all of this is whether China will stop making the batteries once an American company is granted the right to start making them.
FLATT: That may be unlikely. Chinese news reports announced this summer that China is about to bring online one of the largest battery farms in the world, hoping to set new records for energy output. The reports say the entire battery farm is built out of vanadium redux flow batteries.
(SOUNDBITE OF MUSIC)
KELLY: That was Courtney Flatt from public radio’s Northwest News Network along with NPR’s Laura Sullivan. You can read their entire investigation at npr.org.
(SOUNDBITE OF MUSIC)
KELLY: It’s CONSIDER THIS from NPR. I’m Mary Louise Kelly.
Copyright © 2022 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
Sponsor Message
Become an NPR sponsor

source

Music Industry Moves: Steven Rowen Named SVP of Island Records, Luke Armitage Upped to SVP of Astralwerks – Variety

Variety Print Plus Subscriber
Variety Print Plus Subscriber
Variety Print Plus Subscriber
Variety Print Plus Subscriber
Variety Print Plus Subscriber

By Thania Garcia
Island Records has promoted Steven Rowen to senior vice president, head of international marketing. In his new role, Rowen will continue to oversee global marketing across the Island roster which includes Shawn Mendes, Demi Lovato, the Killers, Lauren Spencer Smith and more.
“Steven is one of the most knowledgeable and experienced executives in the global marketing field,” said Mike Alexander, general manager of Island Records. “His creativity and strategy have been at the forefront of Island’s biggest breakthrough stories over the last decade and we are confident he will continue to make enormous strides as Island moves into its next historic chapter.”
Rowen began his career in music at Universal Music Group in 2008, as a coordinator in the artist develop­ment department of the former Island Def Jam Music Group (IDJ). He then joined the international department in 2012 and was eventually promoted to VP of international marketing. In 2021, he officially started working solely for Island as head of international marketing.

“Thank you to Justin, Imran, and Mike for the opportunity to lead our global marketing efforts for the new era of Island Records.” Rowen said. “I’m excited to continue building our incredible roster around the world.”
Rowen will be based in New York and report directly to Alexander.
+ Luke Armitage has been promoted to senior VP of Astralwerks where he will oversee global marketing and daily operations for the label and its overseas brands, including Astralwerks Asia.
Astralwerks President Toby Andrews, to whom Armitage reports, said: “Luke has been integral to the growth and success of Astralwerks since he joined in 2018, leading numerous game-changing campaigns and expertly guiding our team. His forward-thinking approach to international and domestic marketing as a combined discipline has really taken the label to the next level, and I’m so thankful to have him growing into this new role as we look into the years ahead.”
Armitage assumes his new role after four years as VP of marketing for Astralwerks, during which he oversaw the label’s domestic and international marketing efforts. During this period, Armitage worked on a variety of projects, including Marshmello and Bastille’s “Happier” and the Grammy-nominated “Piece Of Your Heart,” from Meduza.
Other artists Armitage has worked closely with during his time at Astralwerks include Alesso, Martin Garrix, the Avalanches, Jax Jones, Jonas Blue and Alison Wonderland. He also worked in partnership with Netflix on the soundtrack to the film, “He’s All That,” which marked the acting debut of Addison Rae.
Armitage is based in Hollywood within Capitol Music Group’s Capitol Tower.
+ ASCAP has upped Deepak Mohite to chief technology officer (CTO), reporting jointly to ASCAP chief executive officer Elizabeth Matthews and to ASCAP executive VP and chief operating officer Brian Roberts.
Mohite joined the company in 2018 as VP of infrastructure and platform services engineering, and in 2019, he was promoted to SVP, software engineering, architecture and quality assurance. During his time in that role, Mohite led ASCAP’s engineering teams in product delivery and strategy planning for software that supported ASCAP’s web and mobile applications.

“Deepak has been responsible for major tech initiatives at ASCAP that have ensured we are the best-in-class music rights management platform for music creators and licensees. The innovations he has implemented have helped us build a fast, flexible and efficient technology infrastructure that has streamlined our operations, made us more agile and set us up to continue to innovate for the future,” said Matthews.
Roberts added: “Deepak’s expertise in emerging technologies will ensure that ASCAP’s technology infrastructure is always evolving along with the businesses we serve. His strategic vision and guidance have helped to drive key technology enhancements at ASCAP, from the migration of our data platforms to the cloud, to critical enhancements to our cybersecurity and data protection, to building software that supports our web and mobile applications.”
Mohite previously held senior technology roles at Tower Research Capital and The New York Times. He holds two master’s degrees and multiple product and technology certifications from MIT.
+ Columbia Records has promoted Betsy Whitney to VP of marketing, as announced by Erika Alfredson and Jay Schumer, both EVP, marketing at the company.
Most recently, Whitney held the position of senior director of marketing where she led campaigns for artists such as Bruce Springsteen, Hozier, Leon Bridges, Haim, Arcade Fire, Barbra Streisand and John Mayer.
She will continue to report to Alfredson and Schumer, who in a joint statement said, “In the many roles that Betsy has held at Columbia, she has consistently been an incredible resource, leading successful campaigns for some of the label’s most storied artists. Betsy brings passion, drive, and creativity to everything she does. We are proud of her growth and leadership, and we are excited to see her continue to bring exciting campaigns to life across the roster.”
Whitney began her career as an intern at NME and in 2008, started at Columbia in New York City as a coordinator in the marketing division. She spent eight years in digital marketing before re-joining the marketing team in Los Angeles.
+ Gilad Woltsovitch has joined Revelator in the company’s climb to elevate its Web3 development team. Revelator powers the organization, tracking, and promotion of music in both Web2 and Web3.
“By making the ecosystem more decentralized and open, we will benefit the entire supply chain. We are thrilled that Gilad will join us in this endeavor,” said Revelator founder and CEO Bruno Guez.
+ The Beggars Group of Labels has promoted Andy Larsen to head of sales and Steve Glauber moves into the newly created role of head of U.S. operations.
Glauber, who joined the company in 2009 as a warehouse worker, will oversee several departments including sales and production, in addition to several aspects of U.S. facilities. He previously served as the U.S. director of digital operations.
Larsen will now oversee the sales team and work closely with distributor Redeye. She has been with the company since 2016 when she joined from the Rough Trade shop. She moves from her previous post as director of sales.
+ Big Machine Music (BMM), a subsidiary of HYBE America, has signed singer-songwriter Anna Vaus to an exclusive songwriting agreement.
A graduate of Belmont University, Vaus was the first-ever recipient of the Miranda Lambert Women’s Creator Fund. She joins BMM’s growing roster of artists which includes Brett Young, Ryan Hurd, Laura Veltz, Matt Dragstrem, Jessie Jo Dillon, Brandy Clark and more.
“We have admired Anna’s unique songwriting from afar for years and are so excited for the opportunity to work together directly,” said BMM general manager Mike Molinar.
“Songwriting is such a foundational part of my artistry and I’m thrilled to get to work with the Big Machine Music team,” Vaus shared. “I’m incredibly excited to have Mike, Tim [Hunze] and Lizzy [Rector] and rest of the team alongside me as I continue to grow as a songwriter.”
A Variety and iHeartRadio Podcast
The Business of Entertainment

source

Watch Chelsea train live at Stamford Bridge! | News | Official Site | Chelsea Football Club – Chelsea FC

You have the opportunity to see Thomas Tuchel and his players up close and personal at an exclusive open training session being held at Stamford Bridge later this month…
The event will be held on Tuesday 23 August but be warned, you’ll have to be quick to secure your seat for this popular event!  
This will be the perfect family event and attendees at the event will also have the chance to feature in the squad photo with the Chelsea players. The Blues squad will line up in front of the stand full of fans for a big group photo (including the fans) which will be shared on our digital channels. 
There are a limited number of tickets available for the open training session, which are available for fans to come and attend for FREE. 
Starting tomorrow, (Wednesday 10 August), there will be a 24-hour early access period only available to Chelsea members and season ticket holders. After this, tickets will be available to all Chelsea fans until they’re sold out. Get yours at the links below.
Stamford Bridge Open Day – Tuesday 23 August – Doors open 2.30pm
Members and season ticket holders book your early-access place here! (Opens 10/08)
Club Chelsea and Westview members book your early-access place here! (Opens 10/08)
All other fans can book your place here! (Opens 11/08)
We look forward to welcoming you to the Bridge for what promises to be a special day! 

source

TG Therapeutics Provides Business Update and Reports Second Quarter 2022 Financial ResultsConference call to be held today, August 8, 2022 at 8:30 AM ET – BioSpace

Skip to main content
Published: Aug 08, 2022
NEW YORK, Aug. 08, 2022 (GLOBE NEWSWIRE) —TG Therapeutics, Inc. (NASDAQ: TGTX) today announced its financial results for the second quarter ended June 30, 2022 and recent company developments, along with a business outlook for the remainder of 2022.
Michael S. Weiss, the Company’s Chairman and Chief Executive Officer, stated, “Our primary focus for the remainder of this year is working toward the approval of ublituximab, which has a PDUFA goal date of December 28, 2022, and preparing for its potential launch in early 2023. If approved, we believe ublituximab has the potential to be a meaningful treatment option for patients with relapsing forms of multiple sclerosis.” Mr. Weiss continued, “During the first half of the year we implemented a number of cost-saving measures, and we are pleased to report those efforts have resulted in a lower than expected 2Q burn, which we believe puts us in a good financial position as we approach the potential launch of ublituximab.”

Recent Highlights
Ublituximab in Multiple Sclerosis
Key Objectives for 2022
Financial Results for the Three and Six Months Ended June 30, 2022
CONFERENCE CALL INFORMATION
The Company will host a conference call today, August 8, 2022, at 8:30 AM ET, to discuss the Company’s second quarter 2022 financial results.
In order to participate in the conference call, please call 1-877-407-8029 (U.S.), 1-201-689-8029 (outside the U.S.), Conference Title: TG Therapeutics. A live audio webcast will be available on the Events page, located within the Investors & Media section, of the Company’s website at http://ir.tgtherapeutics.com/events. An audio recording of the conference call will also be available for replay at www.tgtherapeutics.com, for a period of 30 days after the call.

ABOUT TG THERAPEUTICS
TG Therapeutics is a biopharmaceutical company focused on the acquisition, development and commercialization of novel treatments for B-cell diseases. In addition to a research pipeline including several investigational medicines, TG has completed a Phase 3 program for ublituximab, an investigational glycoengineered monoclonal antibody that targets a unique epitope on CD20-expressing B-cells, to treat patients with relapsing forms of multiple sclerosis (RMS). For more information, visit www.tgtherapeutics.com, and follow us on Twitter @TGTherapeutics and LinkedIn.

Cautionary Statement
This press release contains forward-looking statements that involve a number of risks and uncertainties. For those statements, we claim the protection of the safe harbor for forward-looking statements contained in the Private Securities Litigation Reform Act of 1995. Such forward looking statements include but are not limited to statements regarding the Company’s plans, goals, strategies, timelines, anticipated milestones, and expectations for our current or future approved drugs and drug candidates, including; plans and timelines for potential approval of ublituximab monotherapy in RMS and, if approved, plans and timelines for commercializing ublituximab in RMS; the timing of initiation of clinical trials or the results of ongoing and planned clinical trials; the potential benefits of ublituximab in RMS or any of the Company’s drug candidates in treating patients; and financial guidance regarding the period in which we will have sufficient capital resources to fund our operations.
All forward-looking statements in this press release are based on management’s current expectations and beliefs and are subject to a number of risks and uncertainties. In addition to the risk factors identified from time to time in our reports filed with the Securities and Exchange Commission, factors that could cause our actual results to differ materially from those expressed or implied by any forward-looking statements contained in this press release include the following: our ability to, if approved, establish, maintain and enhance our commercial infrastructure, to market and sell ublituximab; the potential for variation from the Company’s projections and estimates about the potential market for ublituximab due to a number of factors, including for example, limitations that regulators may impose on the required labeling for the product; our ability to reach certain regulatory milestones at all or within the timelines projected; our ability to obtain, or to obtain within the timeline projected or for the indications sought, marketing authorization for our product candidates, including ublituximab monotherapy in RMS; our ability to successfully complete analyses of our clinical study results and present data within the timeframes projected; the risk that the interim, top-line and preliminary data from our clinical trials that we announce or publish may change, or the perceived product profile may be impacted, as more patient data or additional endpoints are analyzed; the risk that regulatory authorities disagree with the conclusions we have reached or data we have publicly disclosed and we are unable to obtain approval for, or successfully commercialize, our product candidates; the risk that preclinical and clinical results for the Company’s drug candidates may not support further development of such drug candidates; actions of regulatory agencies, which may affect the initiation, timing and progress of clinical trials; the Company’s reliance on third parties to perform manufacturing, distribution and supply services, and a range of other support functions for its clinical products; the risk that the ongoing COVID-19 pandemic and associated government control measures or other global issues such as the ongoing conflict in the Ukraine have an adverse impact on our clinical trials and other research and development plans or our regulatory filings and commercialization efforts; the accuracy of our estimates regarding expenses, future revenue, capital requirements and needs for additional financing; our financial performance; and the sufficiency of our existing capital resources to fund our future operating expenses for the timelines projected. Further discussion about these and other risks and uncertainties can be found in our Annual Report on Form 10-K for the fiscal year ended December 31, 2021 and in our other filings with the U.S. Securities and Exchange Commission, including the most recent quarterly report on Form 10-Q for the second quarter ended June 30, 2022.
Any forward-looking statements set forth in this press release speak only as of the date of this press release. We do not undertake to update any of these forward-looking statements to reflect events or circumstances that occur after the date hereof. This press release and prior releases are available at www.tgtherapeutics.com. The information found on our website is not incorporated by reference into this press release and is included for reference purposes only.

CONTACT:
Investor Relations
Email: [email protected]
Telephone: 1.877.575.TGTX (8489), Option 4
Media Relations:
Email: [email protected]
Telephone: 1.877.575.TGTX (8489), Option 6
TG Therapeutics, Inc.
Selected Condensed Consolidated Financial Data
Statements of Operations Information (in thousands, except share and per share amounts; unaudited):
 
Condensed Balance Sheet Information (in thousands):
* Condensed from audited financial statements


Primary Logo
Back to news
Back to top
© 1985 – 2022 BioSpace.com. All rights reserved. Powered by Madgex Job Board Software

source

Systems of axon-like circuits for self-assembled and self-controlled growth of bioelectric networks | Scientific Reports – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 13371 (2022)
158 Accesses
1 Altmetric
Metrics details
By guiding cell and chemical migration and coupling with genetic mechanisms, bioelectric networks of potentials influence biological pattern formation and are known to have profound effects on growth processes. An abstract model that is amenable to exact analysis has been proposed in the circuit tile assembly model (cTAM) to understand self-assembled and self-controlled growth as an emergent phenomenon that is capable of complex behaviors, like self-replication. In the cTAM, a voltage source represents a finite supply of energy that drives growth until it is unable to overcome randomizing factors in the environment, represented by a threshold. Here, the cTAM is extended to the axon or alternating cTAM model (acTAM) to include a circuit similar to signal propagation in axons, exhibiting time-varying electric signals and a dependence on frequency of the input voltage. The acTAM produces systems of circuits whose electrical properties are coupled to their length as growth proceeds through self-assembly. The exact response is derived for increasingly complex circuit systems as the assembly proceeds. The model exhibits complicated behaviors that elucidate the interactive role of energy, environment, and noise with electric signals in axon-like circuits during biological growth of complex patterns and function.
Electric phenomena are essential in the development of complex biological structures and their function. By guiding cell migration1, electric fields and potentials influence wound healing and tissue regeneration2, and direct pattern formation (like left or right, up or down3) in processes such as early neuronal development4 or growth of plant roots5 and pollen tubes6. The aggregate effect of potential differences across membranes, gap junctions, action potentials in axons, and other bioelectric phenomena forms networks of electric potentials that communicate among cells to influence gene expression and thus, formation of biological structure and function in embryogenesis and morphogenesis7,8,9. Changes in physical structure are communicated electrically throughout the network, resulting in coordination of distant growth processes to produce spatially differentiated target structures10. The flow of information from the environment to biological structure and function through electric potentials represents computation without neurons and is postulated to be a primitive and ancient mechanism11.
Self-assembly models are inspired by autonomous interactions among component parts that build complex structures, including examples such as biomolecules (DNA and proteins), living organisms, social networks, and even galaxies. Theoretically, self-assembly is an algorithmic process, resulting in complex and powerful behavior that is capable of Turing universal computation12,13, and has produced new methods for building nanostructures14.
Originally motivated by DNA-based self-assembly14, the circuit tile assembly model (cTAM) was introduced to demonstrate how self-controlled growth15 and self-replication16 can be achieved as emergent properties without explicit programming. Rather, they are made possible by a finite resource (e.g., an electric potential (nu _0)) that is consumed as tiles with simple circuits bind to a growing circuit assembly15, as illustrated in Fig. 2. In the cTAM, tiles attach if a constant voltage (Direct Current or DC) at the terminus is greater than or equal to a threshold (tau). This assembly process produces a family of circuits whose electrical properties change as the circuit grows, which in turn, modulates electric signals that are communicated throughout, as illustrated in Fig. 2. In contrast to models based on differential equations, the cTAM is abstract and discrete, with the potential to reveal the essence of electrical effects in biological phenomena. As an abstract model, the cTAM is amenable to analysis for exact prediction of its maximum circuit size through self-controlled growth and thus, of its electrical properties, as a function of the voltage source (nu _0) or threshold (tau)15,17,18, and their effects on growth of the assembled circuit (such as self-replication16). Figure 1 shows examples of phenomena to which this model could be applied to gain quantitative insights into their dynamical properties at the macrolevel.
Examples of biological phenomena that could potentially be analyzed by the acTAM models and its variants to gain quantitative insights into their dynamical properties at the macrolevel (Galvanic Phenomena19, DNA–Protein Interactions20, Cellular Membranes21, Axonal Networks22) .
The goal of this paper is to further elucidate the role of electric signals in the growth of biological structures by adding a capacitor to the attaching tile in the cTAM model and extending the analysis to the frequency and time domains. The specific example biosystem that the cTAM represents can be regarded as an axon, and its electrically-driven growth as thresholded signal communication with its environment. Thus, this cTAM is termed the axonal cTAM (or just acTAM), which also denotes the time-varying nature of the model (‘ac’ for alternating current, or AC). The acTAM circuits approximate equivalent circuits for the propagation of action potentials in cable theory23 and the Hodgkin-Huxley model24. Propagation of electric signals along axons are the physical basis for communication from sensory to information-processing neurons and are fundamental to neural information processing. Moreover, axonal networks are capable of information processing25,26 and computation27 without neurons.
Capable of modeling both spiked and graded signals, acTAMs inherit a unique property of the cTAM models in that they produce a dynamically changing family of circuits in which both structure and electrical properties vary as growth proceeds. The acTAM captures the coupled effects of growth and electric potential in an axon-like circuit. In early neural circuit assembly, spontaneous electrical activity is important in development, which is followed by network refinement by signals evoked from sensory input4,28. Electrical activity also promotes neural repair29. Thus, the cTAM could also serve as a tool to understand the effect of electrical activity on the growth of axonal networks, as well as its role in the electric potential distributions that arise in bioelectric networks8 that lead to observable properties of form and function in biological organisms.
In what follows, the acTAM is defined, and the expressions for the node potentials as a function of the parameters and size of the circuit are derived Nodal analysis of acTAM systems and applied to the analysis of steady-state behavior. The transient and complete solutions are derived. The product of the acTAM is a system of ladder circuits, and their mature size is characterized. Finally, a discussion of the significance of the acTAM for development and function of structures in biological systems is presented.
An cTAM ladder circuit is assembled as ladder tiles attach to a seed with a DC voltage source (nu _0) and subsequently, each other as long as the voltage across the resistor R at the terminus is at least (tau). In addition, Watson–Crick complementary DNA glues at the top (a–a(^{prime })) and bottom (b–b(^{prime })) bind the assembly together.
This section presents definitions of the cTAM (to make the paper self-contained) and its extension to the acTAM.
(acTAM Circuit) An axonal circuit is a tuple
on a graph (N, E), where N denotes the set of nodes corresponding to electrical nodes in the circuit, E denotes the set of edges, C is a set of circuit components (chosen from resistors, capacitors, and voltage sources) assigned to edges (e_{(i,j)} in E) where ({i, j} in N), and g maps some subset of nodes (partial N) to some subset of glues labeled from a finite alphabet (Sigma), i.e. (g: partial N rightarrow Sigma). (partial N = N_{in} cup N_{out}) consists of two finite subsets of nodes, input nodes (N_{in}) and output nodes (N_{out}) of the circuit, and are the points at which glues bind tiles together on the boundary of the circuit. The size of the circuit is the number of tiles in it. (A cTAM is an acTAM with only resistor tiles, without capacitors.)
(acTAM) An axonal tile assembly system (acTAM) is a tuple (mathscr {C} = (Gamma , S, nu _0, zeta , tau )), where (Gamma) is a finite nonempty alphabet of tiles, (S subset Gamma) is a subset of seed tiles, (nu _0) is the potential at the power source, (zeta : Gamma (N_{in}) times Gamma (N_{out}) rightarrow {0,1}) is a glue indicator function that determines whether glues on input nodes of a tile match or bind to glues on output nodes of acTAM circuits, and (tau in {mathbb {R}}_{+}) is the threshold voltage that sets one of the criteria for further tile attachments.
The simplest acTAM consists of two circuit tiles, a seed tile (Fig. 3) and an unlimited number of copies of one ladder tile (Fig. 4). The circuit in the tiles consists of a resistor (alpha R) in series with a parallel combination of a capacitor C and a resistor R. The ladder circuit is equivalent to that used in cable theory23 to model signal propagation down axons, with (alpha R) being the longitudinal resistance, R the membrane resistance, and C the membrane capacitance. Growth of the ladder is determined by the electric potential difference across the RC pair, or between the nodes joining the (alpha R) resistors and the RC pair and the common ground (bottom of the circuit). Ladder tiles attach to the seed and subsequently, to other ladder tiles, if and only if the node potential at the output nodes of last tile in the assembly is at least (tau) (Fig. 5). The electric potential at node k in a given acTAM circuit of size n tiles at time t will be denoted (nu ^n_k(t)) (Or just (nu _k(t)) if a certain size n is assumed), where (k=0,ldots ,n), including the source potential (nu _0^n(t)) (or just (nu _0)) at the seed.
In the simplest model, ladder tiles bind to form assemblies in the shape of growing ladders. In a reaction-rate limited regime, ladder tiles are present in saturation and always bind whenever the tip potential is at least (tau), whereas in a diffusion-limited regime, ladder tiles only arrive and attach at set time intervals. The specific mechanism by which tiles attach can be left unspecified under either assumption. One specific implementation would follow the well known aTAM model of DNA self-assembly12,13, in which a tile has a pair of oligonucleotides a, b on the output nodes of a DNA molecular tile that may bind to their corresponding Watson-Crick complements (a^{prime }, b^{prime }) on input nodes of the attaching ladder tile. Other models can use protein-protein interactions resulting from electric potentials forming across ion channels11. In this paper, assembly processes are monotonic, i.e., once a tile is attached, the attachment will never dissolve. The behavior of the family of acTAM circuits is characterized below in general, for both DC and AC.
The seed tile for the acTAM consists of a voltage source (nu _0(t)), two resistors, R and (alpha R), and a capacitor C. The top node has glue a, and bottom node has glue b.
The ladder tile for the acTAM consists of resistors R and (alpha R), and a capacitor C. The input nodes have glues (a^{prime }) and (b^{prime }) on top and bottom and the output nodes have glues a and b on top and bottom, both respectively.
An acTAM ladder circuit is assembled as ladder tiles attach to a seed with a voltage source (nu _0(t)) and subsequently, each other as long as the voltage across the RC pair at the terminus is at least (tau). In addition, the hybridized glues at the top (a–a(^{prime })) and bottom (b–b(^{prime })) bind the assembly together. Nodes are labeled k and the terminal (last) node n.
Kirchoff’s Current Law (KCL) is about conservation of charge and states that the sum of the currents entering and exiting any node in a circuit must be 0. In particular, the seed (Fig. 3) has a distribution of potentials at its three nodes (source (nu _0), ground (nu _{-1}), and node potential (nu _1) at the tip of the tile (between (alpha R) and the parallel C–R pair). Attachment of successive ladder tiles causes a (speed of light, nearly instantaneous for relatively small circuits) propagation of the signals to the other tiles, which reconfigures the node potentials at the previous nodes into a new steady state after a brief transient, as illustrated in Fig. 6. Over time, the self-assembly process in an acTAM model generates a family of circuits of increasing size (number of tiles) with a dynamic potential distribution (nu ^n_k) at nodes k in a ladder of size n ((kin {1,ldots ,n})). Because of the series-parallel resistance of the circuits in the family, the tip potentials (nu ^n_n) are a decreasing function of time or size, so they eventually become unable to support new attachments, as shown in the DC case in15,17. Thus an acTAM really defines a dynamical system of growing circuits of increasing size that exhibit emergent characteristic behavior.
Response of first node potential in the seed tile for (alpha = 1, R=1 Omega , C=10^{-3} F), and (|nu _0| = 50 text {V}). The second tile attaches to the seed at t = 0.003 s, causing the node potential at node 1 ((nu _1)) to undergo a transient. The steady-state (ss) DC value 25 V (top line) for the seed alone is shown for reference.
For the analysis of the acTAM, the differential equations for a system of circuits are derived using nodal analysis (KCL). First, the system is solved for a non-time-varying (DC) case, which makes the capacitor an open circuit and the circuit purely resistive (“Nodal analysis of acTAM systems”). Then, in “Steady-state phasor analysis”, the steady-state response in the frequency domain is derived from the DC equations with complex impedances. Thus, the steady-state time response is derived for a sinusoidal input voltage. The transient response (“Transient response”) was determined from results in30 and initial conditions, which are set by the node potentials when a new ladder tile attaches. The complete response is steady-state plus transient (“Complete solution”). The dynamic behavior is shown in Fig. 7.
Response of node potentials for parameters (alpha = 1, R=1 Omega , C = 10^{-3} F), and (|nu _0| = 50 text {V}). Tiles attach every t = 0.003 s, causing the node potentials to undergo transients before returning toward steady-state values. With a threshold of (tau = 3 text {V}), no further attachments occur past node (k=4).
The temporal dynamics of the circuit system assembled in the acTAM model can be characterized in general using the equations for the node voltages (nu ^n_k) obtained from Kirchoff’s Current Law (KCL). (We will drop the superindex n when the context is not ambiguous.) Given the circuit in Fig. 5, for an arbitrary intermediate node k, where the current in the capacitor is (C d nu _k / d t), the node equation is
where (nu _k) is the electric potential at node k. Rearranging and using the shorthand (nu _k^{prime } = d nu _k/d t), the differential equation for (nu _k) is
For the first node (k=1), the equation is
and for the last node (k=n,)
In matrix form, the system of differential equations thus becomes
Under direct current (DC, time invariant supply), the capacitor C is an open circuit and can be ignored, and thus, the general form of the node equations for node k of n is
The set of equations then becomes
where the matrix of coefficients is
and the source vector is
with (nu _0) the DC source potential. According to Cramer’s rule, the node potentials are then given by
where ({mathbf {A}}^{(k)}_n) is the matrix with the k-th column of ({mathbf {A}}_n) (Eq. 8) replaced with the vector ({mathbf {b}}_n) from Eq. (9), and (|cdot |) or a corresponding capital letter denotes the determinant of the corresponding matrix. Since ({mathbf {A}}_n) is a tridiagonal matrix for (n ge 2), the determinants (A_n = |{mathbf {A}}_n|) satisfy a recurrence relation
with (A_1 = 1+alpha), and (A_{0} = 0). Equation (11) is a linear recurrence, and has a solution similar to that derived in31 for equivalent resistance. Therefore, the general solution for a circuit ladder of size n is a linear combination
where (a_1) and (a_2) are constants, and (rho _1) and (rho _2) are the roots of the characteristic equation of the recursion,
Solving Eq. (13) yields two solutions,
Using the the initial conditions from Eq. (11) with (k=1) and (k=2),
yields the following two equations,
Solving for (a_1) and (a_2) gives
The recurrence for (A_n^{(k)}=|{mathbf {A}}_{n}^{k}|) is
where
with (a_1) and (a_2) given by Eq. (15). Therefore, according to Cramer’s rule (Eq. 10), the node potentials at node k are
and for the last (terminal) node ((k=n)) in the ladder, the node potential is
where (nu _0) is the source voltage.
Since (|rho _2| < 1), as n becomes large, the node potential at the terminus of the ladder in the last tile is bounded by
With a sinusoidal time-varying source (nu _0(t)), phasor analysis can be used to derive the steady state response of the circuit in Fig. 5. Phasor analysis represents circuit components with complex impedances, allowing the steady state behavior of the circuit to be solved using nodal analysis, as in the DC case. The impedances are (Z_1 = R/(1 + j omega R C)) for the parallel combination of R and C and (Z_2 = alpha R) for the series resistor. For an intermediate node k other than 1 or n, the node equation is
with (omega) the angular frequency. If we let (alpha ^{prime } = alpha (1 + j omega R C)), then, the system of equations from nodal analysis is identical to those for DC (Eq. 7), except the node potentials are now in the phasor or frequency domain. Therefore, with that substitution, the solution for node potentials is the same as Eq. (17),
with complex
and
in which (alpha ^{prime }) has been substituted for (alpha). The node potential at the terminus of the ladder is bounded by Eq. (20) with (alpha ^{prime }) substituted for (alpha),
where the source voltage is now a function of frequency (omega). Converting from frequency back to the time domain, the steady state node voltages are
where (|nu _k|) is the magnitude and (angle nu _k) is the phase of the complex node potentials (nu _k(omega )). Like axons, the acTAM ladder is a low-pass filter. The magnitude of the frequency response (Eq. 22) and phase are shown in Figs. 8 and 9, respectively.
Magnitude of the frequency response for parameters that approximate an axon ((alpha = 10^{-6}, R = 10^{12} Omega , C = 10^{-8} F), and (|nu _0| = 100 text {mV})).
Phase of the frequency response for parameters that approximate an axon ((alpha = 10^{-6}, R = 10^{12} Omega , C = 10^{-8} F), and (|nu _0| = 100 text {mV})).
According to30, the eigenvalue for node k of n for the matrix in Eq. (8) with (alpha ^{prime }) substituted for (alpha) is
and the ith value in the kth eigenvector (chi _{i,k}) for (lambda _k) is
The transient response for node k of n is thus
where the (b_i)’s are determined from initial conditions.
With a sinusoidal time-varying source (nu _0(t)), the complete solution for a node (k in {1,ldots ,n}) is the sum of the transient (Eq. 29) and steady-state (Eq. 26) responses,
for a ladder of length n. The constants (b_i) in Eq. (29) are determined by initial conditions, which are the node potentials when a new tile attaches or are the values of the potentials (nu ^{n-1}_k(t)) for nodes (k in {1,ldots ,n-1}) in a ladder of length (n-1), i.e. the node potentials ((nu ^{n-1}_k(t))) in the ladder of length (n-1) are the initial conditions for the node potentials ((nu ^n_k(t))) in a ladder of length n. For an attaching tile, since it is disconnected from a source, the initial condition is (nu ^{n}_n(0)=0). If attachments to a growing ladder only occur after it has practically reached steady state, then, the initial conditions for attachment to a ladder of size n are the DC values for the node potentials, or (nu ^n_k) (Eq. 18).
Equation (30) for different nodes k is plotted for a couple of binding regimes. The first binding regime is an approximation for diffusion-limited growth in which tiles are available for binding at a fixed time interval T. Every T seconds, a new tile arrives at a growing ladder and attaches if the terminal potential is at least (tau). In Fig. 10, with circuit parameters that approximate those for axons, a DC voltage source is applied at (t=0) and tiles arrive every (T = 0.3) s. There is relatively little attenuation as growth proceeds because the value of (alpha) is small compared to R. The same circuit parameters and (T = 0.2) s are used in Fig. 11 with (nu _0(t) = 0.1 sin (omega t + phi )) and (omega = 20 pi) (rad/s). An exponential decrease in amplitude is observed, as well as shifts in the phase of the sinusoid. The other regime approximates reaction-rate limited growth in which tiles are available in saturation and is shown in Fig. 12. As soon as the terminal potential (nu _n ge tau), a new tile will attach. Because the time to reach (tau) is shorter than the time to reach steady-state, the response at each node is attenuated to a transient value at the attachment and changes to new time behavior with (nu ^{n-1}_{k}(t)) as the initial condition, where for the last tile (nu ^{n}_n(0) = 0) and for the next to last tile (nu ^{n}_{n-1}(0)=tau).
Complete response for parameters that approximate an axon ((alpha = 10^{-6}, R=10^{12} Omega , C=10^{-8} F), and (|nu _0| = 100 text {mV})) under a binding regime that approximates diffusion-limited growth with a time between attachments of (T = 0.3) s.
Complete response for parameters that approximate an axon ((alpha = 10^{-6}, R = 10^{12} Omega , C=10^{-8} F), and (nu _0(t) = 100 sin (omega t + phi ) text {mV}) with (omega = 20 pi) (rad/s)) under binding regime that approximates diffusion-limited growth with a time between attachments of (T = 0.2) s.
Complete response for parameters that approximate an axon ((alpha = 10^{-6}, R=10^{12} Omega , C=10^{-8} F), and (nu _0(t) = 100 text {mV})) under binding regime that approximates reaction-rate limited growth with tiles in saturation and a threshold of (tau = 50 text {mV}).
Using the expressions derived in previous sections, bounds on the length of the ladder can be derived. Given the biological connection of the acTAM ladder circuit and growth/signal propagation in axons, the length provides a measure of both the driving potential (nu _0) and the threshold (tau). In our model, (nu _0) represents the source of energy for the growth or the strength of the signal from a sensor, and the threshold (tau) represents those randomizing forces in the environment that oppose either growth or signal propagation. Thus, the length of the ladder is a measure of both.
To determine bounds on the length n of the ladder, the terminal voltage has to be less than the threshold, i.e.,
where the expression for (nu _n) is from Eq. (19). Rearranging, it is found that (A_n > nu _0/tau), which has no ready solution for n. Nevertheless, by using Eq. (20) for n large, it is found that the condition for growth to cease is bounded by
Thus, the growth is bounded and the length given (nu _0) and (tau) can be estimated. Moreover, for complex (alpha ^{prime }), length is not only modulated by input potential and threshold, but also by the frequency of the input. The dependence of length n on frequency is shown in Fig. 13.
Dependence of length n on (nu _0/tau) at different frequencies for parameters that approximate an axon ((alpha = 10^{-6}, R = 10^{12} Omega , C = 10^{-8} F), and (nu _0(t) = 100 text {mV})).
Likewise, the input potential (nu _0) and threshold (tau) that generate a given length can be determined from Eq. (31) as
For (alpha ^{prime } = 1) in Eq. (18), (A_n) becomes
By using relations for the golden ratio (phi) in Eq. (34), Eq. (33) becomes
where (F_{2n+1}) are the odd Fibonacci numbers, which is similar to a result for the equivalent resistance of long resistive ladders32. The golden ratio and Fibonacci numbers appear frequently in nature, mathematics, and human designs. Thus, the relationship of driving force (nu _0) and environment (tau) that produce a given length ladder to the golden ratio and the Fibonacci sequence provide further evidence of the connection of the acTAM to biological systems.
In addition, using the equation for the node potential (Eq. 18) and recurrence for the determinant (Eq. 11),
it is seen that this ratio is bound by a Lucas sequence U(P, Q) with (P=2+alpha ^{prime }), (Q=1), (U_1(P,Q) = 1), and (U_0(P,Q) = 0). Comparing (nu _0/tau) to generate lengths n and (n-1) as n becomes large gives
which is one of the solutions, (rho _1), to the characteristic (Eq. 14), the limit of the ratio of two numbers in the Lucas sequence, and for (alpha ^{prime } = 1) is equal to (1 + phi), where (phi) is the well-known golden ratio.
As mentioned earlier, (nu _0) represents a signal or energetic driving force for ladder elongation, (tau) represents randomizing forces or noise in the environment, and the length of the ladder for which the signal is (ge tau) is an artifact produced by the ratio (nu _0/tau), a kind of signal-to-noise ratio for the acTAM. Thus, the ladder is produced by an interaction between its own internal assembly process and its environment. Therefore, it is natural to ask, what is the sensitivity of the structure of the ladder, i.e. length, to changes in the environment, i.e (nu _0/tau). For any given (nu _0/tau), a unique ladder length is produced. Nevertheless, there are an infinite number of (nu _0/tau) that can produce a ladder of a given length n. Thus, the system itself, the ladders, can only tell us so much about the environment, (nu _0/tau). This is a consequence of the discreteness of the ladder growth in a continuous environment. Using Eq. (19), for a ladder of length (n-1) to add one additional tile, (nu _{n-1} = nu _0/A_{n-1} ge tau), and for growth to terminate at length n, (nu _n = nu _0/A_{n} < tau). Solving both for (nu _0/tau)
which represents the continuous interval that will produce a ladder of length n, spanning a voltage range
where (rho _1) and (rho _2) are given in Eq. (14), the solutions to the characteristic equation. For (alpha ^{prime }=1), Eq. (39) is equal to
which are the even Fibonacci numbers (F_{2n}). The size of this interval grows unboundedly as n becomes large. Thus, larger and larger (nu _0/tau) are required to produce longer ladders, and there is a larger spread in its values between adjacent lengths.
There are two interpretations of this remark. First, the length of the ladder, which is also a measure of the spatial extent over which a signal exceeds the threshold, is the artifact by which the acTAM assembly senses its environment. As the length of the ladder increases, the uncertainty about the (nu _0/tau) that produced the ladder increases, and alternatively, the length of the ladder becomes less sensitive to changes in (nu _0/tau). Autopoiesis33 is a theory that describes the ability of a system to maintain itself through a constructive relationship with an environment. In an abstract way, the acTAM captures some of the tradeoffs between a system being able to “know” its environment and also, remaining stable in the face of environmental changes. Living organisms have evolved to find a middle ground of sensing and responding to changes while maintaining those characteristics that define it as a species and have enabled it to survive in a given environment.
Living systems transduce flows of energy and matter from the environment to grow, produce complex biological patterns, respond to sensory input, and reproduce. Because of their complexity, it is a challenge to model the input-output relationships among the various flows into the system and predict the response, be that a structure or a behavior.
In the original cTAM (Fig. 1), simple circuit tiles attach to a growing ladder circuit if the voltage at the terminus is greater than or equal to a threshold. The cTAM enables self-controlled growth driven by a finite source of energy (voltage source) in a background environment. This complex and characteristic behavior comes as a direct consequence of the interaction among components of the model that are proxies for components in a biological network, namely matter (represented by circuit tiles), energy (input voltages), communication (through signal propagation and dynamic interaction among components representing somatic computation10,11 without a central information processor) and interaction with the environment (dynamic growth to homeostasis) at various levels of abstraction.
In this paper, the model has been extended to the acTAM to incorporate other components that are considered important for biological circuit models, such as capacitors and time-varying bioelectric signals, while keeping the model amenable to quantitative and predictive analysis.
Figures 8, 9, 10, 11, 12 and 13 represent results for the acTAM with circuit parameters approximating an axon. After all, the acTAM circuit is highly similar to that of axonal cable theory, which at this stage, is its most appropriate benchmark. In these figures, the acTAM produces a response that approximates signaling in axonal networks, namely it acts as a low-pass filter.
Levin and collaborators7,8,9 describe how complicated distributions of electric potentials might influence biological pattern formation through interactions with genetic mechanisms. The acTAM is a tool that enables exploration of the interaction between electric potentials that depend on the growth of a biological structure, and vice versa. Growth itself introduces more complicated time-varying behavior, as seen in Figs. 10, 11 and 12, and additional frequency components in the signal. Interestingly, in Figs. 10 and 12, transients from growth by attachment of circuit tiles results in waveforms that are starting to look like action potentials. It has been suggested11 that bioelectric phenomena are an ancient mechanism to control growth and form, possibly predating cells themselves. Transduction of energy to overcome disorder through growth of organized structures would seem to be a prerequisite for early life, and at least in its electric signals, the acTAM starts to demonstrate how those early growth mechanisms may have contributed to the development of complex biological shape and function.
In this paper, the acTAM is monotonic, i.e. once a tile attaches, it cannot detach. Nevertheless, extensions to the acTAM are possible that allow detachment. In order for growth to proceed, the model would have to incorporate forward and reverse reaction rates that would determine the extent of the growth, with this cTAM representing a kind of abstract non-equilibrium thermodynamics that might lend itself to quantitative analysis. Certainly, a series of attachment and detachment events would produce even more complex electric signals and growth behavior.
In conclusion, the acTAM abstracts and is equivalent to other other models of bioelectric networks7,8,9 and offers the potential to uncover deep clues on the etiology of biological structure formation (shape and dynamics). For example, what really controls and limits growth is access to environmental resources and the coupling between components and the environment. It also captures mechanisms that couple growth and electrical signaling in axon-like networks with predictive capability for both length and electrical response. These models also provide full and realistic examples of the well known property of autopoiesis33, in which the structure of an organism is determined and preserved by, not in spite of, interaction between components and the environment. Furthermore, the model itself (although not the analysis) is relatively simple and so affords the possibility of models for more complex and realistic phenomena (proteins and neuronal assemblies) in which the representation of matter is less abstract than in conventional models and closer to direct electrical phenomena. These properties would also allow the possibility of providing a more objective and systematic framework to address issues such as homeostasis, self-healing, and other epigenetic biological phenomena as emergent properties34,35 rather than being explicitly programmed.
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. These include Mathematica programs that were used to generate simulation data for figures.
Funk, R. H. Endogenous electric fields as guiding cue for cell migration. Front. Physiol. 6, 143 (2015).
Article  Google Scholar 
Tyler, S. E. B. Nature’s electric potential: A systematic review of the role of bioelectricity in wound healing and regenerative processes in animals, humans, and plants. Front. Physiol. 8, 627 (2017).
Article  Google Scholar 
Robinson, K. R. & Messerli, M. A. Left/right, up/down: The role of endogenous electrical fields as directional signals in development, repair and invasion. BioEssays News Rev. Mol. Cell. Developmental Biol. 25, 759–766 (2003).
Article  Google Scholar 
Spitzer, N. C. Electrical activity in early neuronal development. Nature 444, 707–712 (2006).
ADS  CAS  Article  Google Scholar 
Canales, J., Henriquez-Valencia, C. & Brauchi, S. The integration of electrical signals originating in the root of vascular plants. Front. Plant Sci. 8, 2173 (2018).
Article  Google Scholar 
Agudelo, C., Packirisamy, M. & Geitmann, A. Influence of electric fields and conductivity on pollen tube growth assessed via electrical lab-on-chip. Sci. Rep. 6, 19812 (2016).
ADS  CAS  Article  Google Scholar 
Levin, M. Endogenous bioelectrical networks store non-genetic patterning information during development and regeneration. J. Physiol. 592, 2295–2305 (2014).
CAS  Article  Google Scholar 
Pietak, A. & Levin, M. Bioelectric gene and reaction networks: Computational modelling of genetic, biochemical and bioelectrical dynamics in pattern regulation. J. R. Soc. Interface 14(134), 20170425 (2017).
Article  Google Scholar 
Srivastava, P., Kane, A., Harrison, C. & Levin, M. A meta-analysis of bioelectric data in cancer, embryogenesis, and regeneration. Bioelectricity 3(1), 42–67 (2021).
Article  Google Scholar 
Manicka, S. & Levin, M. Modeling somatic computation with non-neural bioelectric networks. Sci. Rep. 9, 18612 (2019).
ADS  CAS  Article  Google Scholar 
Levin, M. & Martyniuk, C. J. The bioelectric code: An ancient computational medium for dynamic control of growth and form. Bio Syst. 164, 76–93 (2018).
Google Scholar 
E. Winfree, Algorithmic self-assembly of DNA. PhD thesis, California Institute of Technology, June 1998.
E. Winfree and P. Rothemund, The program-size complexity of self-assembled squares, in STOC ’00: Proceedings of the thirty-second anual ACM symposium on Theory of Computing, (New York), pp. 459–468, ACM, ACM Press, 2000.
Winfree, E., Liu, F., Wenzler, L. A. & Seeman, N. C. Design and self-assembly of two-dimensional DNA crystals. Nature 394, 539–544 (1998).
ADS  CAS  Article  Google Scholar 
Deaton, R., Garzon, M., Yasmin, R. & Moore, T. A model for self-assembling circuits with voltage-controlled growth. Int. J. Circ. Theory Appl. 48(7), 1017–1031 (2020).
Article  Google Scholar 
Yasmin, R., Garzon, M. & Deaton, R. Model for self-replicating, self-assembling electric circuits with self-controlled growth. Phys. Rev. Res. 2, 033165 (2020).
CAS  Article  Google Scholar 
Yan, Y., Garzon, M. & Deaton, R. Harmonic circuit self-assembly in cTAM models. IEEE Trans. Nanotechnol. 18, 195–199 (2019).
ADS  CAS  Article  Google Scholar 
Yan, Y., Garzon, M. & Deaton, R. Self-assembly of 2-d resistive electric grids. IEEE Trans. Nanotechnol. 18, 562–566 (2019).
ADS  CAS  Article  Google Scholar 
Wells, D. A. The Science of Common Things: A Familiar Explanation of the First Principles of Physical Science. For Schools, Families, and Young Students, Ivison, Phinney, Blakeman, 1859, p. 290. This work is in the public domain in its country of origin and other countries and areas where the copyright term is the author’s life plus 70 years or fewer.https://commons.wikimedia.org/wiki/File:Galvani-frogs-legs-electricity.jpg. Accessed 5 July 2020.
Wheeler, R. (Zephyris). 2007. B-/Z-DNA junction bound to a Z-DNA binding domain. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License. https://commons.wikimedia.org/wiki/File:B-,_Z-DNA_junction_2ACJ.png. Accessed 5 July 2020.
Nrets. (2008). This file is licensed under the Creative Commons Attribution-Share Alike 3.0 Unported license. https://commons.wikimedia.org/wiki/File:MembraneCircuit.jpg. Accessed 5 July 2020.
B. Blausen. (2013). This file is licensed under the Creative Commons Attribution-Share Alike 3.0 Unported license. https://commons.wikimedia.org/wiki/File:Blausen_0657_MultipolarNeuron.png. Accessed 5 July 2020.
Scott, A. C. The electrophysics of a nerve fiber. Rev. Mod. Phys. 47, 487–533 (1975).
ADS  Article  Google Scholar 
Hodgkin, A. L. & Huxley, A. F. A quantitative description of membrane current and its application to conduction and excitation in nerve. J. Physiol. 117, 500–544 (1952).
CAS  Article  Google Scholar 
Debanne, D. Information processing in the axon. Nat. Rev. Neurosci. 5, 304–316 (2004).
CAS  Article  Google Scholar 
Alcami, P. & El Hady, A. Axonal computations. Front. Cell. Neurosci. 13, 413 (2019).
CAS  Article  Google Scholar 
Zhang, X., Pan, L. & Paun, A. On the universality of axon p systems. IEEE Trans. Neural Netw. Learn. Syst. 26, 2816–2829 (2015).
MathSciNet  Article  Google Scholar 
Ibarretxe, G., Perrais, D., Jaskolski, F., Vimeney, A. & Mulle, C. Fast regulation of axonal growth cone motility by electrical activity. J. Neurosci. 27(29), 7684–7695 (2007).
CAS  Article  Google Scholar 
Goldberg, J. L. Role of electrical activity in promoting neural repair. Neurosci. Lett. 519, 134–137 (2012).
CAS  Article  Google Scholar 
Yueh, W.-C. Eigenvalues of several tridiagonal matrices. Appl. Math. E-Notes 5, 66–74 (2005).
MathSciNet  MATH  Google Scholar 
Kagan, M. On equivalent resistance of electrical circuits. Am. J. Phys. 83(1), 53–63 (2015).
ADS  Article  Google Scholar 
M. Kagan & X. Wang. Infinite circuits are easy. How about long ones? arXiv:1507.08221 [physics.gen-ph] https://doi.org/10.48550/arXiv.1507.08221(2015).
H. R. Maturana & F. J. Varela. Autopoiesis and Cognition: The Realization of the Living, vol. 42 of Boston Studies in Philosophy of Science. (D. Reidel Pub. Co., Dordrecht, 1972).
Sosik, P., Garzon, M., Smolka, V. & Drastik, J. Morphogenetic systems for resource bounded computation and modeling. Inform. Sci. 547, 814–827 (2021).
MathSciNet  Article  Google Scholar 
Garzon, M., Sosik, P., Drastik, J. & Skalli, O. A self-controlled and self-healing model of bacterial cells. Membranes 12, 678–694 (2022).
CAS  Article  Google Scholar 
Download references
Department of Electrical and Computer Engineering, University of Memphis, Memphis, TN, 38152, USA
Russell Deaton & Rojoba Yasmin
Department of Computer Science, University of Memphis, Memphis, TN, 38152, USA
Max Garzon
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
R.D., M.G., and R.Y. made substantial contributions to the conception of the acTAM model. R.D. drafted the work and M.G. substantively revised it. R.Y. confirmed the model with simulations.
Correspondence to Russell Deaton.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Deaton, R., Garzon, M. & Yasmin, R. Systems of axon-like circuits for self-assembled and self-controlled growth of bioelectric networks. Sci Rep 12, 13371 (2022). https://doi.org/10.1038/s41598-022-17103-4
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-17103-4
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
Scientific Reports (Sci Rep) ISSN 2045-2322 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Thermodynamics of organic electrochemical transistors – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 13, Article number: 4514 (2022)
480 Accesses
1 Altmetric
Metrics details
Despite their increasing usefulness in a wide variety of applications, organic electrochemical transistors still lack a comprehensive and unifying physical framework able to describe the current-voltage characteristics and the polymer/electrolyte interactions simultaneously. Building upon thermodynamic axioms, we present a quantitative analysis of the operation of organic electrochemical transistors. We reveal that the entropy of mixing is the main driving force behind the redox mechanism that rules the transfer properties of such devices in electrolytic environments. In the light of these findings, we show that traditional models used for organic electrochemical transistors, based on the theory of field-effect transistors, fall short as they treat the active material as a simple capacitor while ignoring the material properties and energetic interactions. Finally, by analyzing a large spectrum of solvents and device regimes, we quantify the entropic and enthalpic contributions and put forward an approach for targeted material design and device applications.
Organic mixed ion-electron conductors (OMIECs)1,2 emerged as an ideal class of materials for bioelectronic applications, both in-vivo and in-vitro. The attention to OMIECs stems from a reversible and energy-efficient redox mechanism that allows controlling the charge carrier density of the material. These appealing features, absent in traditional organic and inorganic electronic devices, have been harnessed to design human-compatible electronics, e.g., circuits that can be implanted within an organism and interact with tissues, neurons, blood, etc.3,4,5. The most prominent device based on OMIECs is the organic electrochemical transistor (OECT)6, whereby a gate electrode is employed to flood or deplete the OMIEC with ionic species, therefore switching on and off its conductivity. The traditional OECT architecture, sketched in Fig. 1a, has been employed for a broad spectrum of applications of technological interest as proven by the number of successful demonstrations of OECT-based devices for biosensors7, artificial synapses8,9, hardware capable of computing10,11,12, and wearable electronics for biofluid and biosignal monitoring3,13,14. More recently, the application scope of OECTs has been further expanded, entering dry electronics: using solid electrolytes, OECT-based devices were shown to be an excellent candidate for non-volatile memories15 as well as a good alternative to traditional low-cost organic and inorganic digital circuitry16,17. This multifaceted range of applications demands a detailed understanding of the physical and chemical processes underlying the OECT operation, as well as comprehensive electronic modeling. In their seminal paper, Bernards et al. borrowed the mathematical framework of thin-film field-effect transistors (FETs)18 to describe the channel current Id as a function of the gate voltage Vgs and drain voltage Vds of poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS)-based OECTs, according to
where Vp is the pinch-off voltage defined as Vp = eρ0/C*, Λh is the hole mobility (we chose this uncommon notation to avoid confusion with the chemical potential μ extensively used below), C* is the volumetric capacitance, ρ0 is the intrinsic charge carrier density, e is the elementary charge, and t, W, and L are the geometrical channel thickness, width, and length. Bernards model is an elegant and often used framework to describe the operation OECTs. However, it does not allow for an accurate quantitative analysis of the important figures of merit of OECT operation (e.g., transconductance, saturation in on- and off-state, subthreshold behavior, etc.), nor for qualitative description of the OECTs IV curves, such as the peculiar S-shaped transfer curve, especially for high-performance devices. The latter, is the result of an electrochemical, rather than electrostatic, charging mechanism mediated by the redox reaction
Figure 1 shows how such a reaction is reflected in the electrical properties of the PEDOT when a gate voltage is applied.
a Typical setup for the IV measurement of an OECT using a non-polarizable Ag/AgCl pellet as a gate electrode. Such setup is used throughout this work too. b Redox reaction driving the operation of OECTs. The amount of oxidized and reduced PEDOT leads to the transfer curve sketched below. c Key difference between the FET-based model for OECTs and our thermodynamic model: the nonlinear charge accumulation resulting from the Gibbs free energy, in strong contrast with a linear accumulation typical of a capacitor.
In order to address these discrepancies, several approaches have been proposed to extend Bernards model, usually adding parameters that capture OECT-specific properties not seen in conventional FETs, e.g., ion concentration-dependent mobility19, disorder20, or ion diffusion21. Bernards model and its extensions rely on the assumption that the accumulated ions in the channel can be calculated as a simple parallel-plate capacitor, similar to a metal-oxide-semiconductor FET, but experimental evidence shows a marked nonlinear potential-dependent capacitance, as shown in Supplementary Fig. 1 and ref. 22. Moreover, the chemistry behind the redox process leading to the OECT switching is not taken into account: electrostatic interactions alone cannot comprehensively describe a redox reaction and therefore cannot explain the basic mechanism that underlies the OECT operation. A few studies put forth the idea, or the necessity, to investigate the metal/electrolyte/polyelectrolyte system on a thermodynamic basis, therefore rendering the purely electrostatic picture insufficient23. For example, thermodynamic notions (Nernst law and Boltzmann statistics) need to be integrated into the model in order to describe the concentration-dependent threshold voltage exploited in OECTs for ion sensing7,24. Romele et al.25 investigated the transport properties of OMIECs and observed that the channel resistance and capacitance are independent of the electrolyte concentration, in contrast to the double-layer theory. This ionic buffering is explained in terms of Donnan’s equilibrium. Tybrandt et al.26 reasoned in terms of the chemical potential of the charged species within the polymeric channel: their study results in accurate numerical drift-diffusion simulations and in the key observation that polyelectrolyte/conductive polymer blends should be viewed as two-phase materials. Rebetez et al. investigated the temperature-dependence of the reaction equilibrium that drives the polaron and bipolaron formation and revealed the importance of entropy in the establishment of the redox equilibrium27. These approaches highlight physical and chemical mechanisms behind the charging dynamics of OECTs and provide good tools to understand the interaction at the gate/electrolyte and electrolyte/polymer interface. However, a fully-developed mathematical framework capable of reproducing experimental data while considering the material properties was never proposed. A number of empirical models and numerical simulations were also proposed23,28,29. A detailed overview of the mathematical models of OECTs and their evolution over time was recently reported by Colucci et al.30.
Here, we propose a universal approach derived from thermodynamic axioms capable of explaining a plethora of OECT and OMIEC properties in a variety of applications. We argue that, although FET models can provide a useful and convenient tool to investigate the properties of OECTs, they fail in giving a realistic physical and chemical picture of the system. Instead, we reason in terms of chemical equilibrium and electrochemical potential. Key in our approach is the fact that we do not rely on the direct proportionality between applied voltage and accumulated charge typical of a parallel-plate capacitor. Rather, a nonlinear voltage-dependent charge accumulation (Fig. 1c) naturally derives from the treatment of the Gibbs energy. We analyze the equilibrium that establishes between the conductive polymer and electrolyte, and how the gate-source voltage perturbs it. In doing so, key features of OECTs naturally emerge from the math without artificial parametrization. For example, previously overlooked peculiarities of OECTs, such as saturation in the on-state and the strong voltage-dependent capacitance are intrinsically captured by our model, and so is the effect on the device performance based on the doping level.
We support our findings with experimental data employing PEDOT:PSS (the most widely employed OMIEC) OECTs as a reference system, and with numerical simulations. We capitalize on the investigation and quantification of intrinsic material properties such as the interactions occurring between the polymer and ions mediated by the solvent.
When a PEDOT:PSS film is immersed in an aqueous electrolytic solution (e.g., NaCl in H2O), the polymeric matrix swells, and ions can penetrate. Cations act as dedopants, reducing a fraction of the existing polarons PEDOT+ into PEDOT0 according to Eq. (2). This process happens spontaneously, without the need for external forces (e.g., a gate potential), and it is proven by the conductance lowering of PEDOT:PSS when going from salt-free solvents to the electrolytic environment. Interestingly, the resulting amount of Na:PSS is independent of the concentration of salt in the bulk electrolyte, as shown experimentally by Romele et al.25. This appears evident looking at the equilibrium constant keq
where the activity of the electrons provided by the electrodes is considered unitary. Here, [PEDOT 0] is the percentage of PEDOT that reduces after the PSS moiety pairs with a cation, and [PEDOT +] is the percentage of PEDOT in its oxidized state. Since the sum [PEDOT 0]+[PEDOT +] = 1 is constant, and so are the temperature and the pressure, we treat the system as a canonical ensemble with two phases31. Eq. (3) provides a bridge between a chemical picture of the system, since it describes the chemical reaction, and an electronic picture, because it quantifies the doping level at a certain temperature. Hence, we use it as starting point for our analysis. We set out this investigation with the definition of Gibbs free energy G, of a chemical reaction, and by treating the channel as a binary mixture of the two species. The definition of Gibbs energy of a binary mixture is
where H is the enthalpy, T is the temperature, and Smix is the entropy of mixing. In the simplest approximation, the enthalpy of mixing Hmix is neglected and the only contribution to the enthalpy is H0, namely the sum of the enthalpies of the pure PEDOT 0 and of pure PEDOT + with the corresponding chemical potentials μp0 and μp+, respectively31. By defining ϕ as the relative concentration of reduced PEDOT, defined as
which spans from 0 (purely oxidized phase) to 1 (fully reduced phase), we can describe the entropy of mixing31. Accordingly, Eq. (4) becomes
Where kB is the Boltzmann constant. Eq. (6) is plotted in Fig. 2a for different temperatures. It is important to point out that the free energy of the whole system (electrolyte + OMIEC) should also take into account the free energy of the electrolyte Ge, and not only the one of the OMIEC. When ions migrate from the electrolyte into the polymer, a ΔGe may be expected. However, such change is negligible as calculated in Supplementary Note 1. Therefore, Ge is a constant, allowing us to calculate the variation of G only on the OMIEC, as well as calculate the derivative (see next section) without incurring an error.
a Temperature and ϕ dependence of the Gibbs free energy (with Hmix = 0, using ({mu }^{{p}^{+}}=-60 ;{{{{{rm{meV/mol}}}}}}) and ({mu }^{{p}^{0}}=0)). Higher temperatures shift the minimum towards ϕ0 = 0.5 and to lower energies. b Measurement of the current flowing in a PEDOT:PSS thin-film at a different temperatures, in deionized water (DIW) and saltwater DIW:NaCl 100 mM (circles). Curves are fitted with a linear regression (solid lines). The experiment in DIW (see Methods for details) is used to estimate the contribution of the temperature-dependent mobility, which is then subtracted, leading to the blue dashed line. The same contribution is subtracted from the current measured in DIW:NaCl. The negative slope shows that, besides the increased mobility, the doping density is lowered. Note that at room temperature, the current drops by 8% with respect to when it is in distilled water (DIW), hence ϕ0(T = 300 K) = 0.08. c Calculated concentration of reduced PEDOT0 and doped/oxidized PEDOT+ in the film at different temperatures.
Since the pure phases are entropically unfavorable, the systems’ equilibrium at a certain temperature lies in its energetic minimum characterized by the coordinates ϕ0 and G0. ϕ0 can be found by measuring the conductance of a PEDOT:PSS film in deionized water (DIW) and in saltwater. The current flowing through a film of PEDOT:PSS is given by
where [PSS−] is the concentration of PSS in PEDOT:PSS (in m−3), and it is the maximum charge carrier density that the film can reach (when ϕ = 0). This happens in DIW, where in absence of a reducing agent (ϕ = 0) the charge density is [PSS−]. Thus, by varying the temperature of the solution, the temperature dependence of the hole mobility Λh(T) is found. This is shown in Fig. 2b, where a linear increase of the conductance is obtained, with a slope of 1.9 μA/K We ascribe it to a temperature-activated hopping mechanism, typical in conductive polymers. The situation is different in NaCl(aq) where both ϕ and Λh are expected to vary with temperature. Again, a linear increase is observed. However, the slope is 1.4 μA/K, 26% lower than in DIW. By assuming Λh(T) is the same for PEDOT:PSS in DIW and NaCl(aq), the difference must lay in the different charge carrier concentrations. We can therefore calculate ϕ(T), reported in Fig. 2c. In doing so, we are considering the hole mobility to be independent of the ion concentration. Although the mobility may depend on the doping concentration, Friedlein et al. showed that its effect is weak and should not compromise our results in the small range of dedoping level i.e., from 8% to 11.5%19. Accordingly, the doping level (defined as 1–ϕ) decreases, revealing that the cation concentration within PEDOT:PSS increases as a consequence of increased entropy of mixing.
A temperature-dependent doping concentration, linked to a higher entropic contribution (ϕ0 → 0.5 for T → ∞) is compatible with a thermodynamic picture. We now aim to generalize the mathematical framework in order to describe the behavior of PEDOT:PSS in the presence of an external gate electrode, hence, an OECT.
Our mathematical analysis is outlined in Fig. 3a and proceeds as follow: firstly, the equilibrium state of the reaction is analyzed (blue panel); eventually, by applying an external voltage, the equilibrium is perturbed and the electrochemical potential must be introduced (yellow panel). the Gibbs energy can be employed, by deriving with respect to ϕ, to obtain the chemical potential μ
μ is plotted in Fig. 3a for different temperatures. The equilibrium state, characterized by μ0 = μ(ϕ0) = 0, can be altered by an external stimulus e.g., an electrical (gate) potential Vgs, hence introducing the electrochemical potential in the channel (bar{mu })
with Vch being the potential in the channel, which depends on the drain-source voltage Vds according to Eq. Supplementary Eq. 4. By plugging Eq. (9) into Eq. (8), and integrating Eq. (7) over the channel length (refer to Supplementary Note 2 for the full procedure), we obtain the following expression for the current Id flowing in the OECT channel for a given T, Vds, and Vgs
where β = e/kBT is the ratio of the elementary charge to thermal energy and (Z=exp (frac{{mu }^{p+}-{mu }^{p0}}{{k}_{B}T})) is a material-dependent constant that includes the chemical potential of the unmixed species. In doing so, the crucial point distinguishing this model from FET-based models is the calculation of the accumulated charge in the channel: the charge accumulation is not directly proportional to the gate-source voltage as in a plate capacitor model. Rather, it is a nonlinear function of the gate potential (see Fig. 1c) that is derived directly from the equilibrium constant in Eq. (3). Under the condition that Hmix = 0, i.e., particle-particle interactions are neglected. While this seems a far-reaching simplification, it shines a light on some basic features of OECTs as can be seen in the output and transfer characteristics plotted in Fig. 4. Firstly, Eq. (10) includes the linear proportionality with the geometric parameters of the channel and with the hole mobility, as in the Bernards model. In contrast to it, it does not feature capacitance, underscoring the fact that we did not consider the system as a plate capacitor. Additionally, unlike traditional FETs, the equation does not need to be split between linear and saturation regimes. Rather, the dependence of Vd becomes weaker as Vg becomes more negative, and the saturation at large drain voltages is inherently captured. Moreover, it reproduces the saturation in the off-state (due to the complete PEDOT reduction (ϕ = 1) as well as the saturation in the on-state (complete oxidation (ϕ = 0)) as shown in Fig. 3a. The latter was never described in previous theories for OECTs. From a thermodynamic standpoint, μ(ϕ = 0) → − ∞ and μ(ϕ = 1) → + ∞. This means that huge changes in chemical potential i.e., large variation of Vgs, bring about small changes in cation concentration, hence saturation of the drain current. Finally, this model confirms a shift of the threshold voltage at different temperatures32 and foresees that, at room temperature, a gate voltage of <1 V is needed to drive a PEDOT:PSS-based OECT from the on-state to the off-state, in good agreement with state-of-the-art OECTs. In Fig. 3b, c, the transfer characteristics are fitted with the thermodynamic model and compared to the traditional FET-based model (Bernards model), revealing a much better agreement with experimental data even for the simplistic model without particle-particle interactions. In Fig. 2 the same curves are reported on a logarithmic scale. Nevertheless, we stress that this model does not aim to describe the properties of the channel for currents below the threshold voltage, where the leakage current becomes non-negligible and diffusion currents may play an important role. Moreover, decoupling thermodynamic and kinetic effects is a challenging task. To this end, each voltage sweep was performed very slowly (1 V/min) to ensure that the system is always at equilibrium. This is a “static” model, and the description of transient effects may go beyond the thermodynamic approach.
a Step-by-step workflow of our thermodynamic analysis. Starting with the Gibbs energy G, the chemical potential μ can be obtained. These properties are material-specific and by measuring the resistance of the PEDOT/electrolyte system only the equilibrium can be probed. By applying an external field, the equilibrium is perturbed (yellow box), and the current-voltage characteristics can be obtained as explained in the main text and in Supplementary Note 2. The pink and blue spheres indicated a fully oxidized and fully reduced PEDOT phase, respectively. b Transfer characteristics of a PEDOT:PSS OECT with a channel of 100 mM aqueous NaCl solution at Vds = −0.6 V and −0.3 V. The experimental curves (red circles) have been fitted (solid lines) with b the FET-based model (Bernards model), c with the thermodynamic model in the assumption of Hmix = 0, and d with the thermodynamic model including the enthalpy of mixing. The device features W = 50 μm, L = 200 μm, t = 200 nm. The parameters for the fits are: Vp = 0.57 V and ΛhC = 40 F/cm/s/V for the FET model. For the scenario with Hmix = 0: ({mu }^{{p}^{+}}+{mu }^{{p}^{0}}=1cdot 1{0}^{-21}J), Λh[PSS−] = 1.06 × 1022 C/cm2/s/V, T = 300K, and α = 0.2 (see Eq. Supplementary Eq. 14). When adding the enthalpy: h1 = 0.20, h2 = 0.18, h3 = 0.20.
Investigation of the chemical potential and free energy underlying the transfer characteristics of an OECT immersed in a 100 mM NaCl aqueous solution, b 100 mM KCl aqueous solution, c ionic liquid, d 100 mM LiClO4 in butanol, and e 100 mM LiClO4 in water. Experimentally determined curves with Vds = 5 mV (circles) are fitted with the blue solid curves. The histograms on the right display the parameters used for the fits. In Supplementary Table 1 the values and the errors of the parameters used for the fits are reported, together with the fitting procedure. All energies are in kBT units.
In order to achieve better quantitative agreement with experimental data, we must go beyond the interaction-free approximation and allow for enthalpic contributions Hmix arising from the interaction between reduced PEDOT pairs (interaction parameter h1), between oxidized PEDOT pairs (interaction parameter h2), or between each other (interaction parameter h3) using the following term describing interactions as
Differently from the previous scenario, ϕ(μ) cannot be explicated if Hmix ≠ 0. We need to resort to numerical solutions for the integration in Eq. Supplementary Eq. 7 plotting and data fitting (see Supplementary Note 3 for the full code). With this method, we fit the transfer characteristics as shown in Fig. 3d: the agreement with experimental data is striking and decisively improving with respect to the scenario with Hmix = 0.
The presence of underlying S-shaped chemical potential can be directly probed by measuring a transfer curve for Vds → 0 (See Eq. Supplementary Eq. 10). Lowering the drain-source voltage to such a low value (5 mV in our experiments) makes the channel potential homogeneous and, mathematically speaking, compresses the integral along the channel length to a point (See Supplementary Fig. 3). We do this for a number of material combinations, fit the data, and analyze the parameters used for the modeling.
For example, using larger cations (K instead of Na as in Fig. 4a, b), the charges are forced to stay farther away, lowering the Coulomb interaction energy (which we believe is mainly reflected in the decrease of the interaction parameters h1, h2, and h3). Moreover, different solvents can be used to validate the model. If a water-free solvent is used, the enthalpy of mixing is larger than in the case of water. This is visible when employing an ionic liquid (Fig. 4c) or butanol (Fig. 4d) as electrolyte. Figure 4d, e compare OECTs immersed in a solution of 100 mM LiClO4, first in butanol and then in water. Water, with its higher dielectric constant, significantly lowers the contribution of Hmix, again reflected in the decrease of the interaction parameters.
Besides the trend, it is worth discussing the absolute values of the Hmix contribution: its magnitude is a small correction to the interaction-free scenario with Hmix = 0. This may be surprising as the electrostatic interaction that arises between holes, cations, and polyelectrolyte may make a larger effect be expected. This apparent paradox is readily resolved by reasoning in terms of energy variation ΔG: in the fully oxidized phase, the PSS dopants are paired with positively charged polarons. When cations are injected, they replace the polaronic charge without altering dramatically the energetic landscape. This process is sketched in Supplementary Fig. 5. Therefore, the driving force behind the operation of an OECT is entropy, with small perturbations coming from the enthalpic terms. Therefore, for targeted material design, the entropic term must be addressed in order to change the device’s figures of merits decisively.
The thermodynamic model offers a way to understand OMIECs and OMIEC-based devices, as well as a pathway for intelligent device engineering and material design. The very fact that chemical equilibrium is at the core of the operation of an OECT (i.e., the system lies at the minimum of the Gibbs energy function), leads to a transconductance that is highest for Vgs = 0, which is an excellent feature for possible self-powered devices. Note that this is true only if the drain voltage approaches zero, and there may be an additional shift due to the reaction at the gate/electrolyte interface. Therefore, minimizing the channel length ensures a high on-state current despite the low Vds used. Moreover, the same approach (low Vds, small L), allows for an easier mathematical analysis of the entropic e enthalpic analysis and fitting and should be used to compare different OECTs.
As discussed above, the entropic term dominates the enthalpic, giving the OECTs a dissimilar behavior when compared to traditional FETs and OFETs. Unfortunately, the entropic term is also very challenging to tune, posing the difficult question of how to improve the device’s performance. On the contrary, one can change the enthalpic term more easily by employing different ions and solvents, although these may be dictated by the application. In general, by decreasing the enthalpic contribution h1 and h2, as well as increasing h3, the resulting transconductance of the device can be increased. This can be seen by solving numerically Eq. (4).
Another point to pay attention to is the type of material used. Our measurements have been carried out using PEDOT:PSS channels. We believe, however, that the model it can be extended to any type of depletion mode OECTs in which a scheme analogous to the one in Supplementary Fig. 5 applies. However, novel OMIECs allow for accumulation mode OECTs, where the species in the off-state may be fully neutral (i.e., only the neutral polymer without any dopant), and only in the on-state ions generate charge carriers. On the one hand, the model cannot be readily interpreted for accumulation mode OECTs. On the other hand, the theory is not material-specific but rather universal. The basic features deriving from the thermodynamic model e.g. S-shaped transfer curves, bell-shaped transconductance, and transition from on- to off-state in the range of 1 V33,34 still apply to such devices, suggesting that the model may be applied to them as well. In doing so, one must adjust the initial equations and take into account the fact that the device may be off-state when is doped (See Supplementary Eq. 11).
A thermodynamic analysis of the OECTs is presented. Our approach stands on general thermodynamic axioms and leads to a comprehensive, compact, and quantitative framework capable of describing the current-voltage characteristics of such devices and how these arise from the interactions of the material. We reveal that the main driver in the switching of the OECT is the entropy of mixing, in strong contrast with traditional models based on purely electrostatic effects. By measuring the channel current at very low drain voltages, we are able to quantify the enthalpic term of the total free energy, and conclude that it only represents a minor contribution when compared to the entropy (1–10% depending on the system). We also highlight that some important advantages of OECTs, e.g., high transconductance and low voltage operation, are purely a consequence of such thermodynamic features. Our model allows excellent fitting of experimental transfer curve and leads to the conclusion that a model based purely on electrostatic interactions, like the ones typically used, is insufficient to interpret experimental data correctly and unravel chemical and physical interaction of the material with the ions. In fact, the electrostatic (enthalpic) contribution is minor and we quantified it across a large range of materials. Our approach can be used for a better understanding of the physical and chemical interactions that drive the operation of OECTs, as well as a precise guideline for targeted material modeling and engineering.
OECTs were fabricated on 1 × 1 inch glass substrates covered with Cr (3 nm) and Au (50 nm). The photoresist AZ 1518 (MicroChemicals GmbH) was spin-coated (3000 rpm for 60 s; SAWATEC AG), followed by baking at 110 °C for 60 s. Gold traces were shaped by illuminating in a mask aligner photolithography system (I-line 365 nm, lamp power 167 W, SUESS Microtec AG) for 10 s, photoresist developing (AZ 726 MIF, MicroChemicals GmbH), and Au- and Cr-etching for 60 s and 20 s (10% diluted aqueous solutions of Standard Gold/Chromium etchants, Merck KGaA). After O2-plasma cleaning, PEDOT:PSS (Clevios PH1000, Heraeus Deutschland GmbH & Co. KG) with 5% v/v ethylene glycol (Merck KGaA) was spin-coated (2000 rpm for 60 s) to yield an ~200 nm thick layer. This step was followed by drying at 120 °C for 20 min. Orthogonal photoresist OSCoR 5001 (Orthogonal Inc.) was spin-coated (3000 rpm for 60 s), baked for 60 s at 100 °C, and exposed for 12 s to structure channel and gate. After post-baking (60 s at 100 °C), development followed by covering the sample with Orthogonal Developer 103a (Orthogonal Inc.) and removing it by spinning after 60 s (twice). Excess PEDOT:PSS was removed by O2-plasma etching (5 min, 0.3 mbar, Diener electronic GmbH & Co. KG), after which the sample was placed in Orthogonal Stripper 900 (Orthogonal Inc.) overnight at room temperature. Electrical characterizations were performed using Keithley 2600 SMUs controlled by the software SweepMe! (sweep-me.net) with the PEDOT:PSS film immersed in 100 mM water:NaCl. The transfer curves were measured slowly (1 V/min) using an Ag/AgCl pellet as a gate in order to ensure the system operated at equilibrium and so minimized transient and hysteretic behaviors. After a preliminary test to establish where the saturation in the on and off states occurred, the measurement was carried out from the on-state to the off-state. The temperature-resolved measurement was carried out with the device immersed in a beaker filled with water and NaCl 100 mM, with the beaker on top of a hotplate, and a thermocouple attached to the substrate. A voltage of 10 mV was applied constantly, while temperature and current were controlled with a parameter analyzer Keythley interfaced with the software SweepMe! (sweep-me.net). Numerical simulations were coded in Matlab, code in Supplementary Note 3. In Fig. 4, The experimental I–V transfer curves are transformed into a μ − ϕ as explained in Supplementary Note 2. This is possible only because the curves are measured at vanishing drain voltage. By integrating the chemical potential, the Gibbs free energy is obtained, which is the curve on which the fit is carried out. In order to measure a significant current even with very small Vd (5 mV), the channel was patterned to be wide and short: L = 50 μm, W = 500 μm.
Data available on request from the authors or downloadable from https://osf.io/dufsz/?view_only=e884dff34dc646599f95ae02a44d42ce.
Codes are included in the Supplementary Information (Note S3). For more information please contact the corresponding author.
Paulsen, B. D., Tybrandt, K., Stavrinidou, E. & Rivnay, J. Organic mixed ionic–electronic conductors. Nat. Mater. 19, 13–26 (2020).
ADS  CAS  PubMed  Article  Google Scholar 
Rivnay, J., Owens, R. M. & Malliaras, G. G. The rise of organic bioelectronics. Chem. Mater. 26, 679–685 (2014).
CAS  Article  Google Scholar 
Khodagholy, D. et al. In vivo recordings of brain activity using organic transistors. Nat. Commun. 4, 1–7 (2013).
Google Scholar 
Strakosas, X., Bongo, M. & Owens, R. M. The organic electrochemical transistor for biological applications. J. Appl. Polym. Sci. 132, 41735 (2015).
Leleux, P. et al. Organic electrochemical transistors for clinical applications. Adv. Healthc. Mater. 4, 142–147 (2015).
CAS  PubMed  Article  Google Scholar 
Rivnay, J. et al. Organic electrochemical transistors. Nat. Rev. Mater. 3, 1–14 (2018).
Article  Google Scholar 
Lin, P., Yan, F. & Chan, H. L. Ion-sensitive properties of organic electrochemical transistors. ACS Appl. Mater. Interfaces 2, 1637–1641 (2010).
CAS  PubMed  Article  Google Scholar 
Cucchi, M. et al. Directed growth of dendritic polymer networks for organic electrochemical transistors and artificial synapses. Adv. Electron. Mater. 2100586 (2021).
Ji, X. et al. Mimicking associative learning using an ion-trapping non-volatile synaptic organic electrochemical transistor. Nat. Commun. 12, 1–12 (2021).
ADS  Article  Google Scholar 
Van De Burgt, Y. & Gkoupidenis, P. Organic materials and devices for brain-inspired computing: from artificial implementation to biophysical realism. MRS Bull. 45, 631–640 (2020).
ADS  Article  Google Scholar 
Cucchi, M. et al. Reservoir computing with biocompatible organic electrochemical networks for brain-inspired biosignal classification. Sc. Adv. 7, eabh0693 (2021).
ADS  Article  Google Scholar 
Petrauskas, L. et al. Nonlinear behavior of dendritic polymer networks for reservoir computing. Adv. Electron. Mater. 8, 2100330 (2021).
Keene, S. T. et al. Wearable organic electrochemical transistor patch for multiplexed sensing of calcium and ammonium ions from human perspiration. Adv. Healthc. Mater. 8, 1901321 (2019).
CAS  Article  Google Scholar 
Campana, A., Cramer, T., Simon, D. T., Berggren, M. & Biscarini, F. Electrocardiographic recording with conformable organic electrochemical transistor fabricated on resorbable bioscaffold. Adv. Mater. 26, 3874–3878 (2014).
CAS  PubMed  Article  Google Scholar 
van de Burgt, Y. et al. A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing. Nat. Mater. 16, 414–418 (2017).
ADS  PubMed  Article  Google Scholar 
Ersman, P. A. et al. All-printed large-scale integrated circuits based on organic electrochemical transistors. Nat. Commun. 10, 1–9 (2019).
Article  Google Scholar 
Weissbach, A. et al. Photopatternable solid electrolyte for integrable organic electrochemical transistors: operation and hysteresis. J. Mater. Chem. C 10, 2656–2662 (2022).
CAS  Article  Google Scholar 
Bernards, D. A. & Malliaras, G. G. Steady-state and transient behavior of organic electrochemical transistors. Adv. Funct. Mater. 17, 3538–3544 (2007).
CAS  Article  Google Scholar 
Friedlein, J. T., McLeod, R. R. & Rivnay, J. Device physics of organic electrochemical transistors. Org. Electron. 63, 398–414 (2018).
CAS  Article  Google Scholar 
Friedlein, J. T. et al. Influence of disorder on transfer characteristics of organic electrochemical transistors. Appl. Phys. Lett. 111, 023301 (2017).
ADS  Article  Google Scholar 
Gentile, F. et al. A theoretical model for the time varying current in organic electrochemical transistors in a dynamic regime. Org. Electron. 35, 59–64 (2016).
CAS  Article  Google Scholar 
Inal, S., Malliaras, G. G. & Rivnay, J. Benchmarking organic mixed conductors for transistors. Nat. Commun. 8, 1–7 (2017).
CAS  Article  Google Scholar 
Kaphle, V., Paudel, P. R., Dahal, D., Krishnan, R. K. R. & Lüssem, B. Finding the equilibrium of organic electrochemical transistors. Nat. Commun. 11, 1–11 (2020).
Article  Google Scholar 
Tseng, H., Cucchi, M., Weissbach, A., Leo, K. & Kleemann, H. Membrane-free, selective ion sensing by combining organic electrochemical transistors and impedance analysis of ionic diffusion. ACS Appl. Electron. Mater. 3, 3898–3903 (2021).
CAS  Article  Google Scholar 
Romele, P., Ghittorelli, M., Kovács-Vajna, Z. M. & Torricelli, F. Ion buffering and interface charge enable high performance electronics with organic electrochemical transistors. Nat. Commun. 10, 1–11 (2019).
CAS  Article  Google Scholar 
Tybrandt, K., Zozoulenko, I. V. & Berggren, M. Chemical potential–electric double layer coupling in conjugated polymer–polyelectrolyte blends. Sci. Adv. 3, eaao3659 (2017).
PubMed  PubMed Central  Article  Google Scholar 
Rebetez, G., Bardagot, O., Affolter, J., Réhault, J. & Banerji, N.What drives the kinetics and doping level in the electrochemical reactions of pedot: Pss? Adv. Funct. Mater. 32, 2105821 (2021).
Paudel, P. R., Kaphle, V., Dahal, D., Radha Krishnan, R. K. & Lüssem, B. Tuning the transconductance of organic electrochemical transistors. Adv. Funct. Mater. 31, 2004939 (2021).
CAS  Article  Google Scholar 
Szymański, M. Z., Tu, D. & Forchheimer, R. 2-d drift-diffusion simulation of organic electrochemical transistors. IEEE Trans. Electron. Devices 64, 5114–5120 (2017).
ADS  Article  Google Scholar 
Colucci, R., de Paula Barbosa, H. F., Günther, F., Cavassin, P. & Faria, G. C. Recent advances in modeling organic electrochemical transistors. Flex. Print. Electron. 5, 013001 (2020).
CAS  Article  Google Scholar 
Dill, K. A., Bromberg, S. & Stigter, D. Molecular driving forces: statistical thermodynamics in biology, chemistry, physics, and nanoscience (Garland Science, 2010).
Bongartz, L. M., Weissbach, A., Cucchi, M., Leo, K. & Kleemann, H. Temperature-dependence of all-solid-state organic electrochemical transistors. In 2022 IEEE International Conference on Flexible and Printable Sensors and Systems (FLEPS), 1–4 (IEEE, 2022).
Pappa, A. M. et al. Direct metabolite detection with an n-type accumulation mode organic electrochemical transistor. Sci. Adv. 4, eaat0911 (2018).
ADS  PubMed  PubMed Central  Article  Google Scholar 
Inal, S. et al. A high transconductance accumulation mode electrochemical transistor. Adv. Mater. 26, 7450–7455 (2014).
CAS  PubMed  Article  Google Scholar 
Download references
We thank Dr. Ron Dockhorn for the fruitful conversations about the thermodynamics of conductive polymers. H.T. thanks the financial support from the European Social Fund (ESF) and the Free State of Saxony in the framework of OrgNanoMorph (project number: 100382168). L.M.B. and H.K. are grateful for funding from the German Research Foundation (DFG, KL 2961/5-1). A.W. thanks the project ESF Re-Learning (cfaed), ESF Number: 100382146.
Open Access funding enabled and organized by Projekt DEAL.
Laboratory for Soft Bioelectronic Interfaces, Neuro-X Institute, École Polytechnique Fédérale de Lausanne (EPFL), Geneva, Switzerland
Matteo Cucchi
Technische Universität Dresden, Dresden, Germany
Matteo Cucchi, Anton Weissbach, Lukas M. Bongartz, Richard Kantelberg, Hsin Tseng, Hans Kleemann & Karl Leo
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
M.C. originated the idea, developed the math, and designed the experiments. A.W., L.M.B., H.T., H.K., and K.L. equally contributed to the discussion. R.K. generated the code for the fits for the enthalpy of mixing.
Correspondence to Matteo Cucchi.
The authors declare no competing interests.
Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Cucchi, M., Weissbach, A., Bongartz, L.M. et al. Thermodynamics of organic electrochemical transistors. Nat Commun 13, 4514 (2022). https://doi.org/10.1038/s41467-022-32182-7
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41467-022-32182-7
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
Nature Communications (Nat Commun) ISSN 2041-1723 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

‘The time is right’: reggae colossus Burning Spear on racism, rebellion and returning to Britain – The Guardian

He is the Jamaican legend who liberated reggae, taking it out of Kingston, drenching it in horns – and giving it a joyous, spiritual kick. As Burning Spear hits the road, he looks back on his astonishing life
It’s been two decades since Burning Spear set foot on a UK stage and almost as long since he took on anything that could qualify as a tour: a remarkable state of affairs for the man who has been Jamaica’s most exhilarating performer for half a century.
With an agenda that was equal parts celebration, Black history lesson and pulpit preaching – immersed in horn-drenched music that was at once modern and ancient – Burning Spear entranced and energised audiences across the world. Subdued lighting, smoke (much of which came from the crowd) and live dub versioning built the atmosphere around roots reggae’s greatest voice as he offered hope for the future while addressing the injustices of the past.
Throughout Spear’s career his music has been motivated by his devout Rastafari faith, looking towards African culture as the roots of Black consciousness, and sowing cerebral salvation throughout the diaspora: roots reggae was a vital, joyous contemporary expression of what was forcibly stolen 400 years ago. Burning Spear and his fellow Rastas celebrated the Black separatist activist Marcus Garvey and railed against the evils of slavery and colonialism, warning that a slavery of the mind can be more dangerous than physical subjugation. His goal has been to uplift.
“The time was right,” he says of his return to the UK this week at the age of 77, speaking on the phone from his New York home. “Look at the last two years – nobody going no place, nobody doing things together, nobody even shaking hands! If somebody stand next to you in a line and that man sneeze, everybody look at him with a bad eye! Now, as things start to come around again, the time is right to be playing my music live in a way that will bring people together.
“It’s not like I retired, but it’s best sometime to hold back. These shows are going to be amazing because people haven’t seen me for a while and that means we’re going to have a stronger force of energy in the place.”
Doing what he believes to be the right thing, as opposed to the most lucrative, has been a Burning Spear characteristic since he was still known as Winston Rodney. It’s a motivation that took him in and out of Jamaica’s legendary Studio One record company at the start of his career in the late 1960s.
A tiler by trade, Rodney arrived in Kingston from St Ann parish on Jamaica’s rural north coast in his early 20s, armed with an acoustic guitar and some self-penned hymns to Rastafari. Although the faith had been present on the island since the 1930s, it was still widely despised in the broadly Christian, British-ruled country – particularly in the embryonic music business. Coxsone Dodd of Studio One (“the best musicians, singers, engineers were in there,” Spear says) was the only producer in town that encouraged Rastas.
Spear has mixed memories of his time there. “Remember your first day going to school? You’re shy, you don’t know how to communicate, you don’t know who is a friend – that was what going to Studio One was like. I knew if I don’t say something that makes sense they would have kicked my arse out and I wouldn’t be here now. I stand up strong, and when I sing my first song to Mr Dodd of course he like the sound of it and he was like: ‘Damn! We’ll have to keep this one!’”
Sympathetic as Dodd was to Rastafari as a faith, he didn’t get it as a music. He recorded the Abyssinians’ Satta Massagana then refused to release it because he thought it too confrontational to have much appeal. Likewise, with Spear: Dodd cut two albums with him at the start of the 1970s but the recording was too restrained to capture the essence of Spear’s dread delivery.
This wasn’t the owner’s only failing, Spear recalls. “I could hang out with some of the best musicians and studio engineers and watch them work, so I learn a lot. Then I realised that you’re not getting paid. People write Mr Dodd was a good man. No! Mr Dodd would be taking credit for all of we songs. When it come to fixing up on the music, he can grow the music, he was good at that, but getting some money? No, that’s not Mr Dodd; he wasn’t an honest man.”
After several years, Burning Spear returned to St Ann, understandably disillusioned – he had failed at the island’s number-one hit factory, and his brand of roots music hadn’t found a place on the all-important Kingston sound systems. Back then reggae was an almost exclusively urban affair, the industry in the capital dictating musical development with a conceit that assumed anyone else had straw in their ears. Upon returning to life away from music, Spear fell in with local sound system owner and nascent record producer Lawrence Lindo, AKA Jack Ruby, who talked him into recording again. It was a revelation.
The producer’s lethargic time signatures, arrangements that left big spaces, and unique Addis Ababa-slanted horn phasing, gave Spear the ideal platform to properly explore his compelling vocal range. Most importantly he had hijacked Kingston’s edgier take on roots music and shifted it to the hills. This had the pace and tinge of melancholy that evoked the weariness brought on by life in rural Jamaica, where hard work from sun up to sun down – “from nah see to nah see” – was the norm, and carrying things along seemingly endless dusty roads was what people did.
On top of that, Spear’s lyrics had the spirit of rural resistance: these folk might be worked to exhaustion, but they would never be put down. This essence dates back to Queen Nanny of the Maroons, whose guerrilla army of rebel slaves held off British redcoats on the island’s mountainous interior in the 18th century until they were granted their own settlement.
Since the majority of Jamaica’s population were rural and a significant proportion of Kingstonians were recent transplants, hearing themselves in this music gave Burning Spear the success that had evaded him at Studio One. “Jamaican people always expected from their music that some consciousness and culture get mixed in – not all of it, but there must be some,” he says. “Music that carry a lot of strong vibes is a force to uplift the thoughts and minds of the people, so you would have more consciousness, less violence and people wouldn’t be so ignorant and so stupid towards each other.”
Four mid-1970s albums – Marcus Garvey, Man in the Hills, Dry & Heavy and Social Living – cemented Spear alongside Bob Marley as the most significant voice in reggae at the time. The songs weren’t so much an incitement to fight Babylon, but rather an explanation of why it was such a good idea; and although anger and indignation had their place, they set out the motivation for Black pride and self-celebration.
“When I put my lyrics together I know I have to tell a story, with a beginning, a middle and an end that people can follow and it will educate them,” he says. “If it’s about Christopher Columbus or the end of war or living good, the music needs to carry some form of education within it – if people know these stories, they can think properly.”
Songs such as Civilised Reggae and Slavery Days are Spear giving advice or providing information, rather than targeting individuals or situations. Even through Jamaica’s sustained political violence of the 1970s, he believed that people should and could govern themselves if given the knowledge and self-awareness to do so; for Burning Spear, “the oppressor” is a nebulous force rather than a particular regime, and as likely to come from within a person as from the outside.
In recent years, the issue that has most stirred Burning Spear is to see Garvey, who so powerfully influenced the Rastafari faith, recognised as the Black Jamaican hero he was. “The time is right for Marcus Garvey to have a public holiday in his honour and be studied in school. The time is right for the people of Jamaica to come together and make sure that take place.”
These albums pretty much established Burning Spear’s future career path – atmospheric, politicised roots music of the highest calibre, best experienced in a live environment. After 50-odd years in the business, he knows what works in a concert. “People knows what to expect, which is me, Burning Spear, and they are ready to accept it. They know. I will just be who I is – I can’t be less or more.”
But he hasn’t given up on recording, actually enjoying lockdown as a chance to “gather” ideas to take to the studio later this year. Judging by his 2021 single Mommy, which brought all the vintage Spear resonance and musical depth to bear but with modernised instrumentation and sounds, this new music will not disappoint.
“You can’t just come up with a melody, or you can’t just create lyrics – you have to let them come to you,” he says. “When everything closed, I could relax and enjoy the melodies and lyrics coming to me. I have to prepare myself to store them, then when I need them in the studio I can take whatsoever I need and put it where it belong. I’m filled with lyrics and melody ideas. That’s me, that’s what I is – music.”
Burning Spear plays the Birmingham Forum on 13 August and O2 Academy Brixton, London, on 14 August.

source

Get to know CCM's incoming faculty members and new unit heads – University of Cincinnati

UC’s College-Conservatory of Music will welcome six new full-time faculty members to its roster of distinguished performing and media arts experts this fall. In addition, the college is also welcoming five current faculty members into new leadership roles for 2022-23.
The six newest members of CCM’s faculty will begin their appointments on Aug. 15, 2022. Those faculty members are:
513-556-5803
[email protected]
513-556-2696
[email protected]
513-556-3442
[email protected]
513-556-6017
[email protected]
513-556-3442
[email protected]
513-556-3442
[email protected]
The following current CCM faculty members have recently assumed new roles:
513-556-3737
[email protected]
513-556-5803
[email protected]
513-556-9530
[email protected]
513-556-6046
[email protected]
312-882-5647
[email protected]
Learn more about CCM’s divisional and departmental leadership here.
At CCM, tomorrow’s arts luminaries work with today’s industry leaders. More than 150 internationally recognized faculty members work with students from around the world. With a 7:1 student-to-faculty ratio, CCM students receive personalized mentorship from professors who guide their individual scholarly, creative, musical, performing and teaching growth.
Who are our faculty members? CCM professors are Grammy, Emmy and Tony Award-winning performers and designers, ASCAP and BMI Award-winning songwriters and composers, Naumburg Gold Medalists and Fulbright Scholars. They are members of major American orchestras, as well as directors and conductors for the world’s leading opera houses and summer festivals. They are researchers, educators, choreographers, filmmakers and playwrights, as well as experts on health and wellness for artists. They have published widely-used textbooks, performed at the White House and produced multimedia for US presidents. They have been featured on major broadcasts like Good Morning America, the Today Show, the Tonight Show, the BBC Proms and Sesame Street, to mention just a few of their accomplishments.

At the University of Cincinnati, we realize the impact our teaching, research, artistry and service can have on our community and the world. So, we don’t wait for change to happen. We break boundaries, boldly imagine and create what’s Next. To us, today’s possibilities spark tomorrow’s reality. That’s why we are leading urban public universities into a new era of innovation and impact, and that’s how we are defining Next for the performing and media arts.
We’re about engaging people and ideas – and transforming the world.
We are UC. Welcome to what’s Next.
Featured image at top: The exterior of CCM’s Patricia Corbett Theater. Photo/UC Creative + Brand
Rebecca Butts | Assistant Public Information Officer | UC College-Conservatory of Music
| 513-556-2675

August 2, 2022
The Cincinnati Symphony Orchestra (CSO) and the UC College-Conservatory of Music (CCM) are proud to announce recent successes of their CSO/CCM Diversity Fellowship program. Established in 2015 and launched with a grant from The Mellon Foundation, the nationally-recognized program amplifies and supports extraordinary young string players from populations historically underrepresented in American orchestras through a specialized, two-year graduate degree-level education, professional development and mainstage performance opportunities with the Cincinnati Symphony Orchestra. Four CSO/CCM Diversity Fellows have obtained full-time positions with major American orchestras for the 2022-23 season; additional Diversity Fellows successes include careers in music education, social change and orchestral administration.

August 1, 2022
UC College-Conservatory of Music Costume Design and Technology student Brittanie McKenna Travis shares her experience as an intern at the Television Academy Foundation

August 1, 2022
CCM Prep at the University of Cincinnati College-Conservatory of Music offers a wide variety of programs in music dance, and theatre arts for arts enthusiasts of all ages and abilities — including classes for children, teens, adults and students preparing for a future in the performing arts.

University of Cincinnati
2600 Clifton Ave.
Cincinnati, OH 45220
513-556-0000
UC Tools
About Us
University of Cincinnati | 2600 Clifton Ave. | Cincinnati, OH 45221 | ph: 513-556-6000
Alerts | Clery and HEOA Notice | Notice of Non-Discrimination | eAccessibility Concern | Privacy Statement | Free Speech | Copyright Information
© 2020 University of Cincinnati
University of Cincinnati
2600 Clifton Ave.
Cincinnati, OH 45220
513-556-0000
© 2020 University of Cincinnati

source

A heat-flux upper boundary for modeling temperature of soils under an embankment in permafrost region | Scientific Reports – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 13295 (2022)
86 Accesses
Metrics details
Building roads in permafrost region is challenged because permafrost is sensitive to temperature increase. As an embankment gains/drains heat mostly at the upper surface, accurately modeling the heat transfer in the upper surface is crucial to understand the thermal stability of the road. Popular methods treat the upper boundary as a temperature-controlled model (TCM), where temperature of the upper surface is set as a sinusoidal function. This simple function, however, fails to identify the influences of solar irradiance, heat convection, and thermal irradiance on the heat transfer on the ground surface. Here we introduce a heat-flux model (HFM) to calculate the heat fluxes at the embankment upper surface and at the adjacent ground surface. HFM-predicted temperature under an embankment is compared against the observed temperature to validate the model, and is compared to the TCM-predicted temperature. While TCM-predicted temperatures and HFM-predicted ones are similar in trend and in pattern, the HFM-predicted temperatures are far more coincident with the observed ones. The pros and cons of both HFM and TCM are discussed. Further studies are expected to use HFM to understand the heat flux components such as solar absorption, heat convection, and thermal irradiance on the temperature of permafrost under embankments.
Permafrost regions cover approximately 22.8 million km2 earth surface, of which permafrost in China makes up 1.59 million km21. Roadways across permafrost regions are usually layered upon a built-up embankment to keep the underlying soil intact2. The construction of the embankment varies the permafrost stratum undesirably and causes differential settlement at the ground surface3,4,5,6. Accurately simulating the heat transfer at the ground surface is critical to precisely predict the temperature of the permafrost stratum after the construction of the embankment. When simulating the temperature of soils under a permafrost embankment, popular methods usually treat this boundary as a temperature-control model (TCM)7,8,9, in which the temperature at the ground surface is assumed a sinusoidal wave. This sinusoidal wave is used to bulkily represent the influences of solar radiation, heat convection, thermal irradiance, and evaporation (if any) on the ground surface10. While this model is simple, the parameters of the sinusoidal wave are estimated by rule of thumb, but not by art (Table 3) and is thus hard to be estimated accurately.
If the ground-surface temperature can be guessed reliably, the temperatures of the deeper soil vary less and thus can be estimated with greater confidence. Qin et al. has proposed a heat-flux upper boundary (HFUB) for the natural ground surface to predict the temperature of permafrost stratum10,11. The model is suitable to predict the temperature of permafrost under a flat, open area only. Liu et al. then proposed an earth-atmosphere coupled model to predict the temperature of soil under an embankment12. This model, however, does not consider the variation of daily solar irradiance. Without a HFM for predicting the temperature of the embankment and its underlying soils, it is unscientific to guide the selection of performance grading of asphalt binder for asphalt pavement under the specific climate condition; it is difficult to understand the partition of radiant heat flux at the ground and to find practical solutions that cut the heat gain of the embankment to preserve the underlying permafrost; it is unknown how to eliminate the sunny-shady effect of embankment in permafrost region by equalizing the conductive heat at both embankment side slopes. In addition, as the global warming heats up the underlying permafrost and causes deformation to the embankment, the HFM is indispensable to predict the settlement of the embankment precisely.
Here we propose a comprehensive heat-flux model (HFM) for the heat transfer at the embankment upper surface and at the natural ground adjacent to the embankment. The proposed model considers the heat flux from solar irradiance, heat convection, and thermal irradiance, and counters the shading effect of the embankment on the adjacent ground. Temperatures predicted by the HFM model are compared against the temperature profiles that are observed from a testing section at the Beiluhe Permafrost Station in the Qinghai-Tibet Plateau. The TCM is also used to predict the temperature at the same site to demonstrate the difference between the TCM and the HFM. The pros and cons of both the HFM and the TCM are explained. The application of the HFM is briefly discussed.
The heat transfer in the embankment and in the soils below an embankment is conduction-dominated, which obeys
where ρ (kg/m3) is the density of the soils; T (°C) is the temperature of the soils; ceq (J/(kg K)) is the equilibrium heat capacity of the soils; and λ (W/(m⋅K)) is the heat conductivity of the soil. It is noted that ceq represents the latent heat of the permafrost when the phase change of water in the permafrost occurs. Methods to calculate ceq and λ have been documented well and can be referred in Appendix A of Supplementary Information.
To solve Eq. (1), one must specify a proper boundary condition to the computational domain, which should be sufficiently large so that a further extent of the domain does not influence the temperature simulation of the embankment and of the underlying soils. As an embankment is usually an isosceles trapezoid with the short base at the top, the computational domain is selected as the trapezoid centered over a rectangle envelop that represents the natural ground (Fig. 1). The heat flux from right and left sides of the rectangle envelop is adiabatic, and the heat flux from the bottom of the base is the product of the local geothermal gradient and the heat conductivity of the soil at the base (Fig. 1). Most of heat fluxes that change the temperature of the embankment and of the underlying soils come from the upper boundary, which consist of the embankment upper surface, the side slope surface, and the adjacent natural ground surface (Fig. 1). In published articles10,13, the upper boundary can be treated as either a temperature-controlled boundary (Fig. 1a) or a heat-flux boundary (Fig. 1b).
The computational domain of an embankment underlain by permafrost. (a) temperature-controlled upper boundary, (b) heat-flux upper boundary. Ti (i = 1, 2, …, 5) stands for a specific sinusoidal temperature function.
Popular models to predict the temperature of the permafrost under a roadway embankment prefer using a temperature boundary, which is termed TCM. The TCM specifies to the upper boundary as a sinusoidal temperature function
where Ts (°C) is the ground-surface temperature, T0 (°C) represents the mean annual ground temperature, and A (°C) is the amplitude of the ground-surface temperature, ω (rad/s) represents the angular frequency, ω = 2π/(24 × 3600 × 365); and ϕ stands for the phases (in radians) of annual ground surface temperature.
Equation (2) considers the daily temperature only, although some models occasionally take the daily temperature pattern into account14. As the ground-surface temperature is difficult to be measured directly, T0 and A are coarsely estimated from the annual air temperature (data from the nearly weather station) according to the adhesive layer theory15. Parameters in Eq. (2) bulkily represent the complex heat transfer at the ground surface, without considering the solar irradiance, convection, thermal irradiance, and evaporation.
The HFM focuses on formulating the upper boundary to solve the temperature of soils in the computational domain using Eq. (1). The most variable heat-flux boundaries are the ground surfaces and the embankment’s upper surface. Both surfaces absorb irradiance from the sun and downward thermal irradiances from the outer space Ld (W/m2). Both surfaces release a portion of the absorption as heat convection H (W/m2), upward long-wave radiation to the outer space Lu (W/m2), and evaporation E (W/m2) (if any) simultaneously. The surplus heat flux, G (W/m2), stores at the ground surface to vary the skin temperature and conducts it to the deeper soils. The heat balance of the upper surface exposed to the air obeys,
where I (W/m2) is the incoming solar irradiance to the surface, R (–) is the reflectance (albedo) of the surface.
In Eq. (3), Lu and Ld are proportional to the ground-surface thermal emissivity, ε (–); in practice, it is more convenient to compute the net long-wave radiation, which is
where σ is the Stefan-Boltzmann constant, σ = 5.67 × 10−8 W ·m−2 ·K−4; the subscripts “s” and “y” represent the ground surface and the outer sky, respectively.
In Eq. (4), the ground-surface temperature, Ts, can be calculated from Eq. (1) once the heat flux is specified to the boundary of the computational domain. The sky temperature, Ty (°C), can be estimated:
where Ta (°C) is the air temperature; εy (–) is the sky emissivity, which is
where Td (°C) is dew point:
where a0 = 17.3, b0 = 237.7, and r0 = a0Ta/(b0 + Ta) + ln(χ/100)12, with χ is the relative humidity of the surrounding air.
The heat convection, H, can be calculated by
where hc (W/(m2·°C)) is the heat convection coefficient. The computation of hc has been well documented in many models, among which the following simple one can provide reliable accuracy16
where v (m/s) is the wind speed recorded at a height of 9.0 m. For a wind recorded from a weather station that is not set at 9.0 m height, the wind speed must be converted according to
In Eq. (3), the evaporation E is difficult to calculate because it is hard to know both the water availability at the ground surface and the evaporation resistance of the ground surface. When water is available at the ground surface, the evaporation is proportional to the incoming solar radiation, with the peak evaporation occuring around solar noon and with a negligible evaporation at nighttime. According to Qin et al.10, the evaporation term can be reasonably neglected by calibrating the solar reflectance at the ground surface. In this model, the evaporation is neglected, and the ground-surface reflectance is adjusted unless the observed ground temperature is coincident with the predicted one. As a result, the heat flux to the upper boundary of the computational domain is
In Eq. (11), formulas to calculate the components H and L have been shown. The most complex component in Eq. (11) is the solar irradiance, I, on the embankment surface and on the ground surface adjacent to the embankment.
Solar irradiance on a global horizontal surface is the global horizontal solar irradiance logged from the local weather station. Solar irradiance is partitioned to beam radiation and diffuse radiation. The side slope of an embankment can be either shaded or sunlit, depending on the time of the day, on the day number of the year, on the orientation of the embankment, and on the angle of the side slope. The solar irradiance on the side slope is thus more complicate. To find the solar irradiance on a tilt surface, one must find the solar azimuth angle, which varies with day and time. On a specific day, the solar declination angle, δ (rad), can be estimated from
where N is the day number of the year, with N = 1 on January 1.
On a specific time of the day, the solar zenith angle can be estimated from
where β (rad) is the latitude of the observer; α (rad) is the solar hour angle, which is 0 at solar noon. The time span from sunrise to sunset can be derived from:
where α at the sunset time is positive, and at the sunrise time is negative.
With the solar declination angle, the solar zenith angle and the solar hour angle, one can find the solar azimuth angle, (gamma) (rad), is given by
If (left|mathrm{sin }gamma right|>1) or if (left|mathrm{sin}theta right|) in Eq. (15) is infinitesimal, the solar azimuth angle has to be estimated by:
It is noted that at noon θ = 0, the denominator in Eqs. (15) and (16) is equal to zero. To circumvent this problem, one can discrete the time sequence between sunrise to sunset in an even number so that the sequence does not have θ = 0.
With the solar azimuth angle, one can determine if a side slope of an embankment is sunlit or shaded. At low-sun case, beam radiation hits one side slope only, while the other is shaded. Assuming the height of the embankment is h (m), beam radiation creates a shaded belt parallel with the embankment orientation. The width of the belt is
where γe (rad) is the orientation of the embankment.
Whether a side slope of an embankment is sunlit or shaded can be estimated by
where η (rad) is the angle of the side slope of the embankment.
Global solar horizontal solar irradiance, I (W/m2), is divided into beam irradiance Ib (W/m2) and diffuse irradiance Id (W/m2), that is
The division has been documented in many models, in which one simple, well-cited model can provide reasonable accuracy17:
where ky is the sky cleanness factor with 0 representing a fully cloudy day and 1.0, a fully sunny day.
The global horizontal solar irradiance, I, can be directly specified to the upper surface of the embankment. However, the solar irradiance on the side slope and on the natural ground surface adjacent to the embankment is more complicate. According to Eq. (18), if a surface is sunlit, the solar irradiance is Id + Ib falling on that surface; otherwise, is the Id only. Diffuse irradiance on the side slope is
The beam radiation on the side slope is the dot product of the beam irradiance Ib and the normal of the side slope.
where the azimuth of the side slope, ψ (rad), can be estimated by
In Eq. (22), if Ibs < 0, which means the slope is shaded, then Ibs = 0. The solar irradiance on the side slope is
The solar irradiance on the natural ground surface, Ig, is also the addition of diffuse irradiance Idg and beam irradiance Ibg. The construction of the embankment blocks the sky view of the natural ground surface, especially the surface that is close to the side slope toe of the embankment. Setting the side slope toe is the origin, the x axis perpendicular to the embankment orientation, and the y axis normal to the ground surface, one gets the sky view factor of any point (x,0) at the natural ground surface as
The diffuse radiation on the natural ground surface adjacent to the side slope is
The beam radiation on the natural ground surface is
The solar irradiance on the natural ground surface adjacent to the side slope is
It is noted that a factor of sunlight reflecting from the side slope can be captured by the adjacent natural ground surface, and vice versa. This solar trapping effect increases the solar absorption of the side slope and of the adjacent ground, especially for the surface near the side slope toe. This topic has been documented and it is found the additional solar absorption caused by the solar trapping effect is negligibly small compared to the solar absorption18.
This study uses both the TCM and HFM to simulate the temperature of underlying permafrost under an embankment in the Beiluhe Permafrost Station (34.8° N, 92.9° E). The simulated results are compared against the logged temperature. The embankment is isosceles with a height of 3 m, a top (short) base of 8.0 m, and a lower (long) base of 17 m. The computational domain is extended horizontally 20 m from the side-slope toe of the embankment, and vertically 22 m from the natural ground surface. The orientation of the embankment is set as east–west orientation, because the thermal asymmetry of this embankment is the most-deleterious case. The embankment filler consists of coarse grained soil. The computational domain in the natural ground, from top to down, consists of a 2 m-thick loose silty clay, a 2 m-thick silty clay, a 6 m-thick layer with ground ice and clay, and finally a 12 m-thick mudstone silty clay (Fig. 1). Soils’ thermal parameters used for the simulation are listed in Table 1. Both left and right sides of the computational domain are set as adiabatic, and the thermal gradient at the bottom of the domain is 0.03 °C/m.
Firstly, the heat-flux boundary is specified to the upper surface of the computational domain. The air temperature Ta, the global horizontal solar radiation I, and the wind speed v, the relative humidity χ, and the sky cleanness factor ky at the embankment site are logged from typical annual weather data. The solar irradiance on the top of the embankment is Eq. (19); that on the side slope, Eq. (24); and that on the adjacent ground surface, Eq. (28). The solar reflectance of the embankment top, side slope, and adjacent ground is set as 0.22 according to Qin et al.19. Considering that water evaporates when it is available at the topsoil and that the evaporation pattern is coincident with solar radiation pattern, the simulation adjusts the reflectance some degree unless the predicted temperatures are best matched with the observed ones.
Then, the TCM is used to simulate the permafrost temperature at the same site. As a guess of the parameters (T0, A, and ϕ) in Eq. (2) could cause substantial errors, this study does not use the adhesive layer theory to parameterize the average, amplitude, and phase of annual ground-surface temperature. Instead, this study uses the least square fitting method to fit the ground-surface temperatures that are predicted via the HFM. The fitting T0, A, and ϕ in Eq. (2) are used as the parameters for the TCM to simulate the soil temperature under the embankment, while the soils’ thermal properties are kept unchanged.
In numerical simulation, either the HFM or the TCM needs specifying an initial temperature to the embankment and to the soil under the embankment. Initial temperatures of the natural ground soils are obtained by repeatedly using the heat-flux boundary to simulate the ground temperature (without embankment) until the differences of the temperature profile at a specific time of two sequential years are less year 0.01 °C. The temperature profile at 00:00 (hh:mm) on September 30 is treated as the initial temperature because the construction of the embankment is completed around this date usually. The initial temperature of the embankment is set as 2 °C uniformly, as usually used in published articles on this topic. This initial temperature (for the soil and the embankment) is used to simulate the temperature of the soils for 50 years after the construction of the embankment.
The soil temperature predicted by the HFM is compared against the temperature logged an earthen embankment in the Beiluhe Permafrost Station from September 2001 to October 2006. The embankment, DK1136 + 400, is an earthen roadbed of the Qinghai-Tibet Railway. Thermistors were buried inside the embankment central borehole and inside the boreholes rightly below the side slope toes to log the soil temperature. In addition, at the natural ground that was 20 m away from the embankment side slope toe, a borehole was also drilled to log temperature at the natural ground. For convenience, the temperature reported in the following section are the daily mean temperature, unless otherwise noted.
Since it was assumed that the ground thermal regime has reached an equilibrium state before the construction of the roadbed. Therefore, to ascertain the ground temperature error caused by the iteration, field-observed temperature from September 7, 2001 to September 7, 2006 was compared with that of simulation. In addition, we adopted the coefficient of determination (R2), mean value (Mean), and standard deviation (Std) to examine the accuracy of the simulation results, where all three indicators were calculated using MATLAB software. The specific calculation methods are as follows:
where To, Ts, (overline{{T }_{o}}) are the observed temperature, the simulated temperature and the average value of the observed temperature, respectively; n represents the total number of values (n = 1, 2, 3, …), and the subscript i stands for the serial number of the values.
Figure 2a,b show the temperature of the soils at 0.5 m depth under the natural ground that is 20 m away from the embankment side slope. It is found that the predicted temperatures (straight line) are well coincident with the simulated ones (dotted data), with R2 = 0.975, a mean error of 0.134 °C and a standard deviation of 0.702 °C (Fig. 2a,b). This statistical data substantiates that the HFM can well predict the temperature of the permafrost under the embankment. The good agreement between the simulation and the observation can be further substantiated from the temperature profiles in the natural borehole. As shown in Fig. 2c, the predicted temperature profile on April 20, 2002 is well agreed with the observed one. Similarly, in Fig. 2d, the predicted temperature profile is highly coincident with the observed one, with R2 = 0.963, a mean error of 0.029 °C and a standard deviation of 0.056 °C.
Soil temperature predicted by HFM. Location: at the natural borehole that is 20 m away from the side-slope toe (a) temperature serial at 0.5 m depth, with R2 = 0.975, Mean = 0.134 °C and Std = 0.702 °C, (b) temperature serial at 0.5 m depth, (c) on April 20, 2002, and (d) on October 19, 2002.
We further compare the predicted temperature profiles at the central borehole of the embankment to the observed ones. It is found that the predicted temperature near the embankment upper surface is greatly deviated from the observed one (Fig. 3). This great deviation is because the embankment is assumed to be completed as September 30 and the embankment temperature is assumed as 2 °C uniformly. These assumptions are somewhat unrealistic because the real construction of the embankment takes months and the temperature of the earthen filler is unknown. Despite this great deviation at near the ground surface, the predicted temperatures at deeper layer are well agreed with the observed ones (Fig. 3). As the heat of the embankment drains overtime and the influence of the initial temperature on the simulation vanishes, it is believed that the HFM can well predict the temperature of the soil under the embankment.
Soil temperature predicted by HFM. Location: at the central borehole. (a) on April 20, 2002, and (b) on October 19, 2002.
Here we compare the soil temperatures predicted by the TCM against the observed ones. To impose the TCM to the embankment surfaces, one must know T0, A, and ϕ in Eq. (2). In this study, the least square fitting method is used to fit the HFM-predicted ground-surface temperature to get T0, A, and ϕ in Eq. (2). The values of T0, A, and ϕ tabulated in Table 2 are used, while the others are the same as those used in the HFM.
It is found that the TCM can also predict the temperature of the soil under the natural ground. As shown in Fig. 4a,b, the temperature serial at 0.5 m depth is well coincident with the observed one, with a R square of 0.925, a mean error of 0.17, and a standard deviation of 1.213 °C. A close comparison between the predicted and observed temperatures reveals that when the temperature is close to the freezing/ thawing point (− 1 to 0 °C), the predicted temperature serial increases or decreases faster than the observed one (Fig. 4b). In reality, when the temperature of a soil is close to the freezing/thawing point, the soil needs more heat (cold) to increase (decrease) its temperature (Fig. 5a). But the TCM unrealistically supplies the sufficient heat (cold) to the soil such that the temperature at the ground surface follows a sinusoidal function (Fig. 5b,c). As a result, the TCM-predicted temperatures deviate somewhat from the observed ones, especially for temperature of the soils at shallow ground where they weather cyclic thawing and freezing annually (Fig. 4c,d). At deeper ground where the soil remains frozen, the predicted temperatures are well agreed with the predicted ones (Fig. 4c,d).
Soil temperatures predicted by TCM. Location: at the natural borehole that is 20 m away from the side-slope toe. (a) Temperature serial at 0.5 m depth, with R2 = 0.925, Mean = 0.17 °C and Std = 1.213 °C, (b) temperature serial at 0.5 m depth, (c) on April 20, 2002, and (d) on October 19, 2002.
During the freezing/thawing period, the TCM imposes more energy to the natural ground surface than the HFM does. (a) The entire year, (b) the thawing period, and (c) the freezing period.
As the temperature profiles and temperature serial in the natural boreholes and central borehole cannot represent the temperature of the entire embankment. Here we show the embankment’s temperature contours that are predicted by the HFM and the TCM, respectively. In case that the embankment has been built for five years, the daily mean temperature predicted by the HFM, in pattern and in magnitude, are highly similar with those predicted by the TCM (Fig. 6a,b). Using the HFM-predicted temperatures to minus the TCM-predicted ones, one can find the difference between the two modelings is about 0.1–0.2 °C (Fig. 6c), which is negligibly small and further implies that both models can predict the temperature of the soil under the embankment. Similar differences and similar patterns are found when the predicted embankment’s temperature contours in 10th year and 50th year are compared (Appendix B in Supplementary Information).
Comparison between the HFM-predicted temperatures and TCM-predicted ones after the embankment is constructed for 5 years. Date: On October 19. (a) HFM-predicted temperature, (b) TCM-predicted temperature, and (c) the HFM-predicted temperatures to minus the TCM-predicted ones.
To further understand the difference between the two models, we analyze the temperature profiles at the embankment central borehole and at the southern and northern side-slope toes. It is found that at the central borehole, the temperature profile predicted by the HFM is closely matched with that predicted by the TCM, especially within 5 year after the construction of the embankment (Fig. 7a). Almost the same difference is observed when compared is the temperature profiled predicted by the two models at 10 year after the construction of the embankment (Fig. 7b). As time elapses, the TCM-predicted temperature is about 0.1–0.2 °C higher than the HFM-predicted model, especially at the deeper layers (Fig. 7c). Similar differences and similar deviations between the two models are observed at the temperatures under the side slopes toes (Figs. 8 and 9). We do not know the reason why the TCM-predicted temperature is about 0.1–0.2 °C higher than the HFM-predicted model, a phenomenon that deserves further investigation.
The central-borehole temperature profile predicted by HFM is compared with that by TCM. Date: on October 27. (a) 5th year, (b) 10th year, and (c) 50th year.
The southern side-slope borehole temperature profile predicted by HFM is compared with that by TCM. Date: on October 27. (a) 5th year, (b) 10th year, and (c) 50th year.
The northern side-slope borehole temperature profile predicted by HFM is compared with that by TCM. Date: on October 27. (a) 5th year, (b) 10th year, and (c) 50th year.
This study is not intended to referee that the HFM is superior to the TCM, or vice versa. Each model has its advantage and disadvantage. The HFM mimics the natural heat transfer at the ground surface and thus can predict the temperature of the soil under the embankment more precisely. The HFM includes the influences of embankment thermal properties and weather data on the temperature of a permafrost stratum. It is thus capable to use the HFM to perform a sensitivity study about the influence of each parameter on the soil temperature. The HFM can help identify if variation of a specific parameter, such as solar reflectance, is favorite to preserve the permafrost under the embankment. If the deformation of the permafrost soil is concerned, it is preferred to use the HFM model to precisely predict the permafrost temperature so that the temperature-related deformation of the soil is estimated properly. In comparison, the TCM model is not a realistic boundary because the ground-surface temperature is not a rigidly sinusoidal wave. The TCM-predicted temperatures could deviate more from the observed ones. But the TCM can predict the temperature of the permafrost in a pattern that is similar to the observed one. As a result, if the trend of the temperature of permafrost under an embankment is interesting, the TCM is superior to the HFM because it is simple and it costs very little simulation time.
Currently, simulation of the temperature of permafrost soils under an embankment needs to guess some parameters. The TCM needs to reasonably guess the mean annual ground-surface temperature, T0, and the amplitude of this temperature, A. In this study, T0, A, and ϕ are regressed from the HFM-predicted temperature. As a result, the TCM-predicted temperature is coincident with the observed ones. However, in most cases, T0, A, and ϕ are unknown. Many documented articles states that T0, A, and ϕ are estimated on the basis of the adhesive layer theory, in which both T0 and A are some cesium degree greater than the local air temperature. But how to estimate both T0 and A is seldom explained. Table 3 lists T0, A, and ϕ that have been used to predict the temperature of permafrost under embankments in the Qinghai-Tibet Plateau. It can be seen that the T0, A, and ϕ vary greatly and that their values, sometime, are specified empirically and randomly.
While the HFM does not need T0, A, and ϕ, it needs to guess some parameters like sky clearness factor, solar reflectance and others. As the heat in the ground comes primarily from the sunlight, the most challenge of the HFM is to guess a reliable solar reflectance R on the upper surface. Although the R value of a surface can be measured, the true R value varies during the courses of day and year, with the lowest reflectance in summer and greatest value in winter12. It has been advocated that the use of energy-based R value to replace the true R so that the seasonal and daily variations of the solar reflectance at the embankment surface are weighed. The data for the energy-based R value, however, is very limited currently28. Specifying a reliable R value for the upper ground surface is thus challenged. Further studies are thus expected to understand the solar reflectance of ground surface in permafrost regions.
The HFM is helpful to select the right performance grading of an asphalt binder according to the climatic condition in the area of uses. In permafrost regions, asphalt pavement is preferred because its flexible properties are advantageously adapted to differential settlement that is caused by the warming of the underlying frozen soils29. Asphalt pavement is typically placed on the top of an embankment. It is critical to select the right performance grading of the asphalt binders to the right climate condition in the area of uses. Performance grading is a function of annual maximum and minimum pavement surface temperature, with one being the average seven-day maximum pavement temperature and the other being the minimum30,31. With the HFM, these temperature extremes can be precisely estimated according to the texture and color of the asphalt pavement and to the specific climatic conditions.
The HFM is powerful to estimate the heat gain and loss at the embankment surface precisely. The most concern heat flux component at the ground surface is the conductive heat flux, which mainly determines the temperature of the underlying frozen soils32,33,34. The HFM can reliably estimate the conductive heat flux to an embankment in a specific climate condition. Under the context of global warming, the HFM can be a useful tool to envision if the underlying permafrost can be cooled by varying embankment surface properties like albedo and emissivity. In addition to the conductive heat flux, how the radiant heat flux partitions at the embankment surface can be calculated. Understanding the heat flux balance at the ground surface is crucial to develop cool embankments and cool pavements for preserving the underlying permafrost under global warming.
The HFM is robust to find practical solutions to eliminate the thermal asymmetry of a roadway embankment in permafrost regions. After an embankment is built, the southern side slope absorbs more solar irradiance than the northern one does. As a result, the permafrost under the southern side slope is typically some Celsius degrees warmer than that under the northern one, a phenomenon that is termed as the sunny-shady effect and has been blamed as the culprit to the longitudinal cracks of the embankment35. Since this effect is caused by differential solar irradiance on both embankment side slopes, it can be mitigated by increasing the albedo of the southern side slope. But it remains unknown that for a specific embankment, increasing the albedo of the southern side slope to which degree can equalize the heat gain of both side slopes and thus eliminate the sunny-shady effect. The HFM can fill this knowledge gap and serves as a powerful tool to estimate the sunny-shady effect.
The HFM is indispensable to precisely calculate the deformation of warming permafrost under the embankment. Building embankment alters the thermal regimes such that the embankment absorbs more sunlight than the adjacent natural ground surface even if the surface materials are the same36. Plus the construction thermal disturbance, the permafrost underlain a roadway embankment has been found 0–1 °C warmer than the adjacent natural ground, especially in warm permafrost regions37. It has been observed that this temperature increment has caused 10–20 cm settlement to the embankment in warm permafrost region, and that the warming of permafrost still continues due to the global warming38. Correctly simulating the temperature increment caused by global warming and construction-induced warming is crucial to predict the embankment’s settlement, which is directly correlated to the damage of the embankment. As the HFM is far more reliable to predict the temperature of the underlying permafrost (Figs. 2 and 3), it is thus indispensable to foresee the serviceability of embankments in permafrost regions.
Other applications of the HFM may be vast but cannot be exhausted herein. For instance, the use of the HFM is not limited to the embankments in permafrost regions but also to those in seasonal frozen regions. It can be used to study on the influences of parameters like surface albedo, surface emissivity, surface roughness, pavement thermal inertia, and other controllable factors on the temperature of the underlying soils and of the pavement structures. It can be also adopted to estimate the impacts the uncontrollable factors like solar irradiance, wind speed, air temperature, and others on the soil temperature. Furthermore, it can be employed to estimate the temperature under embankments that is proactively cooled by inserting the embankment with thermosyphons, covering the side slope with crushed rock layer, and embedding the embankment with air convective duct, etc. While other applications of the HFM is unlimited and await further studies, this study is just a starting of the use of HFM to predict the temperature of permafrost under a roadway embankment.
This study proposed a heat-flux model (HFM) for the heat transfer at the upper surface of an embankment in permafrost regions. Different from traditional temperature-controlled model (TCM) that considers the annual ground-surface temperature varied as a sinusoidal wave, the HFM jointly considers the solar irradiance, heat convection, and thermal irradiance on the ground surface, as well as the shading effect of the embankment on the adjacent ground. The HFM model thus can be used to predict the influence of ground-surface thermal properties and local weather on the temperature of the permafrost under an embankment. The permafrost temperature predicted by HFM is compared against that observed ones. It is found that the HFM model can predict the permafrost soils’ temperature precisely, for instance, with R2 = 0.975, a mean error of 0.134 °C and a standard deviation of 0.702 °C. In comparison, the TCM-predicted temperature is also compared against the observed one. It is found that TCM predicts the permafrost soil temperature in a lower accuracy.
Temperatures predicted by the HFM are 0.1–0.2 °C lower than those predicted by the TCM, but the temperature-varying trend and temperature contour are similarly. It is concluded that if the temperature trend of the soil under an embankment is of concern, the TCM can be recommended if the mean annual ground-surface temperature and its amplitude can be guessed precisely. If the sensitivity of the weather data and ground-surface thermal properties on the permafrost temperature is of concern, the HFM is recommended but the solar reflectance of the ground surface must be weighted wisely. Further studies are expected to understand the energy-based solar reflectance of the natural ground surface and of the embankment surface so that the solar absorption in the HFM is weighted precisely. The HFM can be applied to guide the selection of the performance grading of asphalt pavement in permafrost regions, to understand the heat partition on the embankment surface, to understand the sunny-shady effect of embankments, to calculate the deformation of embankment, and explore other applications related to the HFM.
Some or all data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request.
Temperature (°C)
Ground-surface temperature (°C)
Air temperature (°C)
Dew point temperature (°C)
Bottom limit temperature (°C)
Upper limit temperature (°C)
Mean annual temperature (°C)
Heat capacity (J/(kg·K))
Equilibrium heat capacity (J/(kg·K))
Time (s)
Amplitude of the temperature (°C)
Solar irradiance (W/m2)
Beam irradiance (W/m2)
Beam irradiance on the side slope (W/m2)
Beam irradiance on the ground (W/m2)
Diffuse irradiance (W/m2)
Diffuse irradiance on the side slope (W/m2)
Diffuse irradiance on the ground (W/m2)
Solar irradiance on the side slope (W/m2)
Reflectance (–)
The net long-wave radiation (W/m2)
Downward thermal irradiance (W/m2)
Upward long-wave radiation (W/m2)
Heat convection (W/m2)
Evaporation (W/m2)
Conduction (W/m2)
Heat convection coefficient (W/(m·°C))
Wind speed (m/s)
Height (m)
Day number of the year (–)
Sky cleanness factor (–)
Height of the embankment (m)
Width of the shaded area (m)
Sky view factor (–)
Water content (–)
Latent heat of water (kJ/kg)
Density (kg/m3)
Heat conductivity (W/(m⋅K))
Thermal emissivity (–)
Stefan–Boltzmann constant (–)
Relative humidity (–)
Latitude (rad)
Solar declination angle (rad)
Solar hour angle (rad)
Solar zenith angle (rad)
Solar azimuth angle (rad)
Orientation of the embankment (rad)
Angle of the side slope (rad)
Azimuth of the side slope (rad)
Angular frequency (rad/s)
Phases (rad)
Unfrozen soil
Frozen soil
Water
Unfrozen water
Ice
Sky
Natural ground surface
Side slope
South-facing side slope
North-facing side slope
Qinghai-Tibet Highway
Qinghai-Tibet Railway
Qinghai-Tibet Plateau
Ran, Y. et al. Distribution of permafrost in China: An overview of existing permafrost maps. Permafrost Periglac. 23, 322–333. https://doi.org/10.1002/ppp.1756 (2012).
Article  Google Scholar 
Ma, W. et al. Remedying embankment thaw settlement in a warm permafrost region with thermosyphons and crushed rock revetment. Can. Geotech. J. 49, 1005–1014. https://doi.org/10.1139/t2012-058 (2012).
Article  Google Scholar 
Wagner, A. M. et al. Permafrost degradation and subsidence observations during a controlled warming experiment. Sci. Rep. 8, 10908. https://doi.org/10.1038/s41598-018-29292-y (2018).
ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 
Pei, W., Zhang, M., Li, S., Lai, Y. & Jin, L. Enhancement of convective cooling of the porous crushed-rock layer in cold regions based on experimental investigations. Int. Commun. Heat Mass 87, 14–21. https://doi.org/10.1016/j.icheatmasstransfer.2017.06.019 (2017).
Article  Google Scholar 
Shang, Y., Niu, F., Lin, Z. & Sun, T. Analysis of the cooling effect of a concrete thermal pile in permafrost regions. Appl. Therm. Eng. 173, 115254. https://doi.org/10.1016/j.applthermaleng.2020.115254 (2020).
Article  Google Scholar 
Lu, Q., Zhao, D. & Wu, S. Simulated responses of permafrost distribution to climate change on the Qinghai-Tibet Plateau. Sci. Rep. 7, 3845. https://doi.org/10.1038/s41598-017-04140-7 (2017).
ADS  CAS  Article  PubMed  PubMed Central  Google Scholar 
Sun, B., Yang, L., Liu, Q. & Xu, X. Numerical modelling for crushed rock layer thickness of highway embankments in permafrost regions of the Qinghai-Tibet Plateau. Eng. Geol. 114, 181–190. https://doi.org/10.1016/j.enggeo.2010.04.014 (2010).
Article  Google Scholar 
Sun, H., Ge, X., Zhu, D., Niu, F. & Chen, J. Numerical investigation of the temperature field of a new convection-intensifying composite embankment in permafrost regions. J. Cold Reg. Eng. 33, 06018001. https://doi.org/10.1061/(ASCE)CR.1943-5495.0000174 (2019).
Article  Google Scholar 
Han, F., Yu, W., Yi, X., Hu, D. & Lu, Y. Thermal regime of paved embankment in permafrost regions along the Qinghai-Tibet Engineering Corridor. Appl. Therm. Eng. 108, 330–338. https://doi.org/10.1016/j.applthermaleng.2016.07.126 (2016).
Article  Google Scholar 
Qin, Y., Hiller, J. E. & Bao, T. Modeling cold region ground temperatures with a heat flux upper boundary model. J. Cold Reg. Eng. 27, 29–43. https://doi.org/10.1061/(ASCE)CR.1943-5495.0000051 (2013).
Article  Google Scholar 
Qin, Y. & Bao, T. Modeling geothermal regime in the Qinghai-Tibet Plateau: An examination of the upper-boundary condition. Arab. J. Geosci. 7, 855–864. https://doi.org/10.1007/s12517-013-0889-2 (2013).
Article  Google Scholar 
Liu, Z., Chen, J., Jin, L., Zhang, Y. & Lei, C. Roadbed temperature study based on earth-atmosphere coupled system in permafrost regions of the Qinghai-Tibet Plateau. Cold Reg. Sci. Technol. 86, 167–176. https://doi.org/10.1016/j.coldregions.2012.10.005 (2013).
Article  Google Scholar 
Zhang, M., Lai, Y., Gao, Z. & Yu, W. Influence of boundary conditions on the cooling effect of crushed-rock embankment in permafrost regions of Qinghai-Tibetan Plateau. Cold Reg. Sci. Technol. 44, 225–239. https://doi.org/10.1016/j.coldregions.2005.12.002 (2006).
Article  Google Scholar 
Qin, Y. & Tang, P. Effects of diurnal temperature rhythm on the geothermal regime under the embankment in Qinghai-Tibet plateau. Arab. J. Geosci. 4, 259–267. https://doi.org/10.1007/s12517-010-0166-6 (2011).
CAS  Article  Google Scholar 
Zhu, L. Study of the adherent layer on different types of ground in permafrost regions on the Qinghai-Xizang plateau. J. Glaciol. Geocryol. 10, 8–14 (In Chinese with English Abstract) (1988).
Qin, Y. & Hiller, J. E. Ways of formulating wind speed in heat convection significantly influencing pavement temperature prediction. Heat Mass Transfer 49, 745–752. https://doi.org/10.1007/s00231-013-1116-0 (2013).
ADS  Article  Google Scholar 
Orgill, J. F. & Hollands, K. G. T. Correlation equation for hourly diffuse radiation on a horizontal surface. Sol. Energy 19, 357–359. https://doi.org/10.1016/0038-092X(77)90006-8 (1977).
ADS  Article  Google Scholar 
Zhang, X., Yan, L.-E. & Qin, Y. Regulating solar absorptance for remedying thermal asymmetry of a roadway embankment. Int. J. Heat Mass Tran. 121, 64–71. https://doi.org/10.1016/j.ijheatmasstransfer.2017.12.151 (2018).
Article  Google Scholar 
Qin, Y., Tan, K., Yang, H. & Li, F. The albedo of crushed-rock layers and its implication to cool roadbeds in permafrost regions. Cold Reg. Sci. Technol. 128, 32–37. https://doi.org/10.1016/j.coldregions.2016.05.004 (2016).
Article  Google Scholar 
Lai, Y., Wang, Q., Niu, F. & Zhang, K. Three-dimensional nonlinear analysis for temperature characteristic of ventilated embankment in permafrost regions. Cold Reg. Sci. Technol. 38, 165–184. https://doi.org/10.1016/j.coldregions.2003.10.006 (2004).
Article  Google Scholar 
Lai, Y., Zhang, S., Zhang, L. & Xiao, J. Adjusting temperature distribution under the south and north slopes of embankment in permafrost regions by the ripped-rock revetment. Cold Reg. Sci. Technol. 39, 67–79. https://doi.org/10.1016/j.coldregions.2004.04.003 (2004).
Article  Google Scholar 
Liu, Z., Lai, Y., Zhang, M. & Zhang, X. Numerical analysis for random temperature fields of embankment in cold regions. Sci. China Ser. D 50, 404–410. https://doi.org/10.1007/s11430-007-2043-z (2007).
Article  Google Scholar 
Zheng, B., Zhang, J. & Qin, Y. Investigation for the deformation of embankment underlain by warm and ice-rich permafrost. Cold Reg. Sci. Technol. 60, 161–168. https://doi.org/10.1016/j.coldregions.2009.08.012 (2010).
Article  Google Scholar 
Chen, L., Yu, W., Yi, X., Hu, D. & Liu, W. Numerical simulation of heat transfer of the crushed-rock interlayer embankment of Qinghai-Tibet Railway affected by aeolian sand clogging and climate change. Cold Reg. Sci. Technol. 155, 1–10. https://doi.org/10.1016/j.coldregions.2018.07.009 (2018).
CAS  Article  Google Scholar 
Su, B., Li, N. & Quan, X. The numerical study on the ventilated embankment in permafrost regions in Qinghai-Tibet railway. Cold Reg. Sci. Technol. 38, 229–238. https://doi.org/10.1016/j.coldregions.2003.11.003 (2004).
Article  Google Scholar 
Qian, J., Yu, Q., You, Y., Hu, J. & Guo, L. Analysis on cooling effect of new hollow concrete brick revetment under gravel pavement in permafrost regions. Cold Reg. Sci. Technol. 85, 15–21. https://doi.org/10.1016/j.coldregions.2012.06.003 (2013).
Article  Google Scholar 
Yuan, C., Yu, Q., You, Y. & Guo, L. Formation mechanism of longitudinal cracks in expressway embankments with inclined thermosyphons in warm and ice-rich permafrost regions. Appl. Therm. Eng. 133, 21–32. https://doi.org/10.1016/j.applthermaleng.2017.12.118 (2018).
Article  Google Scholar 
Sailor, D. J. & Fan, H. Modeling the diurnal variability of effective albedo for cities. Atmos. Environ. 36, 713–725. https://doi.org/10.1016/S1352-2310(01)00452-6 (2002).
ADS  CAS  Article  Google Scholar 
Duojie, C. et al. Assessment of freeze-thaw cycles impact on flexural tensile characteristics of asphalt mixture in cold regions. Math. Probl. Eng. 2021, 6697693. https://doi.org/10.1155/2021/6697693 (2021).
CAS  Article  Google Scholar 
Ramadhan, R. H. & Al-Abdul Wahhab, H. I. Temperature variation of flexible and rigid pavements in Eastern Saudi Arabia. Build. Environ. 32, 367–373. https://doi.org/10.1016/S0360-1323(96)00072-8 (1997).
Hao, J., Cao, P., Liu, Z., Wang, Z. & Xia, S. Developing of a SBS polymer modified bitumen to avoid low temperature cracks in the asphalt facing of a reservoir in a harsh climate region. Constr. Build. Mater. 150, 105–113. https://doi.org/10.1016/j.conbuildmat.2017.05.200 (2017).
CAS  Article  Google Scholar 
Qin, Y., Liang, J., Huang, Z. & Tan, K. Painting the roadway embankment with non-white high reflective pigments to raise the albedo. Environ. Earth Sci. 75, 359. https://doi.org/10.1007/s12665-016-5273-6 (2016).
Article  Google Scholar 
Wang, Q., Fang, J., Zhao, X. & Hu, K. The influence of pavement type on the thermal stability of block-stone embankments in the warm permafrost region. Transp. Geotech. 23, 100334. https://doi.org/10.1016/j.trgeo.2020.100334 (2020).
Article  Google Scholar 
Liu, J., Tai, B. & Fang, J. Ground temperature and deformation analysis for an expressway embankment in warm permafrost regions of the Tibet plateau. Permafrost Periglac. 30, 208–221. https://doi.org/10.1002/ppp.2007 (2019).
Article  Google Scholar 
Wang, S. et al. The temporal effect of distress developments of frozen embankments in the permafrost regions along the Qinghai-Tibet Highway. J. Test. Eval. 47, 3059–3079. https://doi.org/10.1520/JTE20170487 (2018).
Article  Google Scholar 
Qin, Y., Huang, H., Ying, J. & Yang, T. Simulation of short-wave solar radiative transfer across a roadway embankment. Int. J. Heat Mass Tran. 108, 332–340. https://doi.org/10.1016/j.ijheatmasstransfer.2016.11.093 (2017).
Article  Google Scholar 
Qin, Y. & Li, G. Permafrost warming under the earthen roadbed of the Qinghai-Tibet Railway. Environ. Earth Sci. 64, 1975–1983. https://doi.org/10.1007/s12665-011-1014-z (2011).
Article  Google Scholar 
Qin, Y., Zhang, J., Li, G. & Qu, G. Settlement characteristics of unprotected embankment along the Qinghai-Tibet Railway. Cold Reg. Sci. Technol. 60, 84–91. https://doi.org/10.1016/j.coldregions.2009.08.002 (2010).
Article  Google Scholar 
Anderson, D. M. & Tice, A. R. Predicting unfrozen water contents in frozen soils from surface area measurements. High. Res. Rec. 393, 12–18 (1972).
Google Scholar 
Download references
This work is jointly supported by the high-level innovation team and outstanding scholar program in Guangxi colleges and the Science Foundation of Guangxi (No. 2018GXNSFAA294070, 2018GXNSFDA138009).
College of Civil Engineering and Architecture, Guangxi University, 100 University Road, Nanning, 530004, Guangxi, China
Tianyu Wang & Li-E. Yan
The Key Laboratory of Disaster Prevention and Structural Safety of Ministry of Education, Guangxi University, Nanning, 530004, China
Tianyu Wang & Li-E. Yan
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
T.W. and L.Y. wrote the manuscript text together and T.W. prepared the numerical simulation and all figures. All authors reviewed the manuscript.
Correspondence to Tianyu Wang.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Wang, T., Yan, LE. A heat-flux upper boundary for modeling temperature of soils under an embankment in permafrost region. Sci Rep 12, 13295 (2022). https://doi.org/10.1038/s41598-022-17529-w
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-17529-w
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
Scientific Reports (Sci Rep) ISSN 2045-2322 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

Taiwan holds live-fire drills as China extends war games – Al Jazeera English

Taiwan’s top diplomat accuses China of simulating an invasion as Taiwanese and Chinese militaries conduct drills in Taiwan Strait.
Taiwan’s military has held live-fire artillery drills simulating a defence of the island as China continued the air and naval exercises it had launched in protest against US House Speaker Nancy Pelosi’s visit to the self-ruled territory.
Taiwanese officials say Tuesday’s drills, which involved soldiers firing howitzer artillery out to sea, had long been scheduled and were not a reaction to China’s unprecedented war games.
Shortly after the hour-long Taiwanese drills ended, China’s Eastern Theater Command announced that it was continuing its exercises in the seas and airspace around Taiwan.
The Reuters news agency, citing an unnamed source, also reported that about 20 Chinese and Taiwanese navy boats were locked in a “standoff” in the Taiwan Strait, with some of the Chinese boats attempting to “press” into the waterway’s median line, the de facto sea border between China and Taiwan.
China claims democratically governed Taiwan, an island of 23 million people, as its own and has called Pelosi’s visit there last week a violation of its sovereignty and a provocation by the United States.
The joint air and sea exercises it launched in response have been its largest in the Taiwan Strait and involved test launches of ballistic missiles over Taipei for the first time. They took place in six zones around the island and were scheduled to run for four days and end on Sunday.
But China’s Eastern Theater Command announced on Monday that it would extend the drills, focusing on anti-submarine and sea assault operations. It said maritime patrol aircraft, fighter jets, helicopters and a destroyer practised locating and attacking targets, including submarines, in the waters off Taiwan.
Tuesday’s drills, it said in a statement, focus on joint blockades and resupply logistics. It did not say when the extended exercises will end.
PLA Eastern Theater Command continued its joint combat training exercises on Tuesday in the waters and airspace around #Taiwan Island, focusing on joint blockade and support. (file pic) pic.twitter.com/o4bCiIdmSv
— People’s Daily, China (@PDChina) August 9, 2022

In the Taiwanese capital, Taipei, foreign minister Joseph Wu accused China of using Pelosi’s visit as an excuse to prepare for an invasion of the island.
“China has used the drills and its military playbook to prepare for the invasion of Taiwan,” he told reporters, calling the war games a “gross violation” of the territory’s rights.
“After the drills conclude, China may try to routinise its action in an attempt to wreck the long-term status quo across the Taiwan Strait,” he said.
Such moves threatened regional security and provided “a clear image of China’s geostrategic ambitions beyond Taiwan”, Wu said, urging greater international support to stop China effectively controlling the strait.
The scale and intensity of the Chinese exercises – as well as its decision to cut some communication with the US – has prompted concern in Washington, but US President Joe Biden has said he does not expect an escalation.
“I’m not worried, but I’m concerned they’re moving as much as they are. But I don’t think they’re going to do anything more than they are,” Biden told reporters at Dover Air Force Base on Monday.
A senior Pentagon official also told Reuters that Washington was sticking to an earlier assessment that Beijing would not try to invade Taiwan in the next two years. Under Secretary of Defense for Policy Colin Kahl said the US military would continue to carry out voyages through the Taiwan Strait in the coming weeks.
China, meanwhile, has defended its behaviour as “firm, forceful and appropriate”.
“[We] are only issuing a warning to the perpetrators,” foreign ministry spokesman Wang Wenbin told a regular briefing, promising China would “firmly smash the Taiwan authorities’ illusion of gaining independence through the US”.
“We urge the US to do some earnest reflection, and immediately correct its mistakes.”
Follow Al Jazeera English:

source

Assessment of thermal distribution through an inclined radiative-convective porous fin of concave profile using generalized residual power series method (GRPSM) | Scientific Reports – Nature.com

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Scientific Reports volume 12, Article number: 13275 (2022)
187 Accesses
Metrics details
The thermal distribution in a convective-radiative concave porous fin appended to an inclined surface has been examined in this research. The equation governing the temperature and heat variation in fin with internal heat generation is transformed using non-dimensional variables, and the resulting partial differential equation (PDE) is tackled using an analytical scheme, generalized residual power series method (GRPSM). Moreover, a graphical discussion is provided to examine the consequence of diverse non-dimensional variables including the parameters of convection-conduction, ambient temperature, radiation, heat generation, and porosity effect on the thermal field of the fin. Also, a graph is plotted to analyze the variations in unsteady temperature gradient using the finite difference method (FDM) and generalized residual power series method (GRPSM). The major result of this investigation unveils that as the convection-conduction parameter scale upsurges, the distribution of temperature in the fin diminishes. For the heat-generating parameter, the thermal distribution inside the fin increases.
Heat transfer is the transmission of energy induced by temperature variations and if two systems in contact have different temperatures, heat transfer occurs until thermal equilibrium is reached. The innovation of effective heat transfer liquids with elevated thermal conductivity and heat transfer coefficient is required to improve the efficiency of the heat transmission process and to reduce the cost and size of the relevant modules and devices. The suspension of tiny solid particulate in liquids is an effective method of increasing the thermal conductivity of liquids and thereby enhances the heat transference phenomenon. Using these kinds of liquids, several investigators explored the features of heat transferal1,2,3,4,5,6,7. On the other side, heat transfer is enhanced using the extended surface. Excessive heat is produced in machine parts in several industrial applications, which can lead to a variety of material flaws. Heat transfer through the extended surface of the apparatus is one strategy for avoiding material damage. A fin is an extended surface that is utilized to augment the rate of heat transference from the primary surface to the environment. It has extensive technological applications, namely air-cooled craft engines, compressors, nuclear reactors, heat exchangers, refrigeration, electrical and electronic apparatuses, and so on. Meanwhile, porous material fins have considerable advantages over conventional fins, and their research is one of the most comprehensive topics in the field of mass and energy transference. In the heat transferal scrutiny of permeable material fins, energy and mass transference of both solid and fluid media must be taken into account. Several analyses have been performed to explore efficient and productive methods of heat transferring through permeable finned surfaces. Ndlovu and Moitsheki8 discussed the one-dimensional heat transport and thermal aspects in a moveable porous straight fin of the uniform area of cross-section. With the impact of radiative, magnetic, and convective mechanisms, Madhura et al.9 depicted the features of the thermal field of a permeable longitudinal fin. The sinc collocation method was executed to study the thermal behavior of permeable fin by Nabati et al.10 under the influence of magnetic force. With the implementation of analytical procedures, Kundu and Yook11 determined the analytical approximation of the porous fin and thus investigated the heat transfer features of the considered fin. Considering the local thermal non-equilibrium model, Buonomo et al.12 researched the energy transfer aspects of a permeable rectangular extended surface. Implementing the spectral collocation method, Kumar et al.13 described the temperature and energy variation in a permeable trapezoidal extended surface with radiation phenomenon.
Fins with a non-uniform cross-sectional area, which contribute to a lighter structure, are recommended in airborne and space application fields over heavier rectangular-shaped fins, despite the fact that such lighter fin constructions are more complicated and costlier to produce. Aziz and Fang14 expounded on the thermal variations inside the straight fin of variable thickness. Further heat transfer aspects are discussed by considering various fin profiles namely trapezoidal, rectangular and concave. Using the DTM approach, Torabi et al.15 debriefed the thermal performance of radiative-convective concave profiled fin. The heat transference features of the concave parabolic extended surface were discussed by Kang16. Recently, Wang et al.17 employed the technique of DTM to examine the aspects of energy dissipation through a permeable fin of the inclined surface. The thermal performance and heat flow in a dovetail fin with internal heat production were probed by Goud et al.18. In the presence of convection and radiation, Jagadeesha et al.19 expounded on the thermal performance of a fully wetted semi-spherical fin using a non-Fourier heat conduction model. Several researchers have worked on solving ordinary differential equations using various techniques, including the differential transform method20,21,22,23,24, sinc collocation approach10, finite difference method25, spectral collocation method26,27, least square method28 and among others. Several of these methodologies are computationally complex because they are trial-and-error in nature or involve complicated symbolic calculations. The residual power series method (RPSM) is one of the analytical techniques which is broadly utilized to achieve an approximate solution because it does not necessitate any restrictive assumptions or linearization. This method can be employed effectively to the given problems and it is easier to achieve accurate approximate solutions without more complications. The RPSM is an innovative approach for procuring analytical Taylor series solutions for linear and nonlinear differential equations. As a result of applying the residual error notion, a series solution, as well as a truncated series solution, can be obtained. Arqub et al.29,30 applied the RPSM for solving the initial value problems. Using RPSM, Syam31 discussed the solution of the fractional-order Fredholm integrodifferential equation. Az-Zo’bi32 applied RPSM to analyze the numerical solution of the time-dependent motion of the van der Waals gas model. Refs.33,34,35,36,37,38 highlights the importance of fluid flow regarding various assumption via different geometries.
The majority of the studies were focused on analyzing the one-dimensional thermal distribution of the porous straight fin or fin with a tapered profile, as evidenced by the above investigations. Further numerical and analytical solutions are provided in other cases. Also, there are no detailed analyses with analytical solutions on the unsteady thermal distribution through a tapered inclined fin. Thus, the principal objective of this analysis is to scrutinize the unsteady temperature variance through an inclined concave porous fin with internal heating. Moreover, the temperature profile of the fin has been solved analytically using the generalized residual power series method (GRPSM).
The unsteady thermal performance of permeable concave fin with convection and radiation is studied. (L) and (W) are the length and width of the permeable concave fin with base thickness (t_{b}). The dissipation of heat to the ambient environment occurs due to the effect of radiation and convection at (T_{a}). The fin which is attached to an inclined surface at an angle (alpha) is considered in this analysis, as demonstrated in Fig. 1. It is assumed that the porous medium is saturated with a single phase, homogeneous, and isotropic fluid. Darcy’s model characterizes the interaction of a porous medium and fluid media. Another implication involves that the fin’s thickness is small in comparison to its length. The temperature inside the fin is also presumed to vary only in the x-direction, with the temperature variation in the y-direction being small enough to ignore.
Physical depiction of an inclined concave parabolic fin.
The transient energy transmission with the above-mentioned assumption is stated by the following governing equation (Torabi et al.15, and Ma et al.39):
where (h^{*}) is the convective heat transfer coefficient, (q^{*} (T)) signifies internal heat generation, (rho) is the density, (sigma) indicates Stefan–Boltzmann constant, (g) symbolizes acceleration due to gravity, (K) is the permeability, (c_{p}) represents specific heat, and (t(x) = left[ {Lambda left( {left( frac{x}{L} right)^{2} – 1} right) + t_{b} } right]) is the local semi-fin thicknesses.
The internal heat generation is presumed to be a function of temperature:
where, (q_{a}) denotes heat generation at ambient temperature, and (xi) is the heat generation parameter. As this analysis considers a finite-length fin with an insulated tip, there will be no heat flow through the tip of the fin. Thus, the relevant boundary conditions (BCs) for Eq. (1) are:
The appropriate non-dimensional terms involved in this study are:
Equation (1) and BCs are transformed to a non-dimensional form using Eq. (2) and Eq. (4) to yield,
The above equation includes radiation-conduction parameter (Nr) which is directly associated with the surface emissivity, fin taper ratio (C) that describes the concave tapered profile, convection-conduction parameter (Nc) is the ratio of convention to conduction, internal heat production parameter (Q), and porosity parameter (S_{H}).
Simultaneously with the help of Eq. (4), Eq. (3) yields,
Let (F) be a function of two variables (x) and (t) i.e.(F(x,t)) and consider the following PDE,
with initial and boundary conditions,
where, (Omega) is a differential operator and (g) denotes the source term.
Assume that the solution of Eq. (7) has the power series in the form shown below
Rewriting the Eq. (7) yields
To evaluate the coefficient functions, (Y_{i} (x),,,,i = 0 ldots k), kth residual function is defined as
Reiterating this operation to find the nth-truncated solution yields
Rearrange the Eq. (5) as:
Let the series solution be in the form
Applying the GRPSM to Eq. (13) gives
To obtain the coefficients (Psi_{k} left( X right),,,k = 1,2,3, ldots ,m), replace (k)th a truncated series of (Theta left( {X,tau^{*} } right)) in Eq. (15), and apply the below derivative formula on ({text{Re}} s^{k} left( {X,tau^{*} } right)) (Modanli et al.40),
For simplificantion, the values for the corresponding parameters are taken as (Nc = 2), (S_{H} = 0.5), (Nr = 3), (Theta_{a} = 0.2), (k_{r} = 0.1), (C = 0.1), (phi = 0.1), (gamma = 0.2), (Q = 0.8) and (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}). Upon substituting these values and with the use of BCs the coefficients of (Psi_{k} left( X right),;k = 1,2,3, ldots ,8) are determined,
Express the above coefficients in a truncation series and ({text{M}} = 0.5937942627) is the value obtained with the help of BC. Using the achieved (k)th truncated series in Eq. (14), the final series solution of the fin problem is represented as
An internal heat generation, natural convection, and radiation impact are taken into account in formulating the unsteady thermal model of an inclined concave porous fin. Equation (1) expresses the corresponding balanced heat equation and is converted to a PDE using dimensionless terms along with BCs. The obtained equation discloses that dimensionless parameters effect the thermal distribution of a porous fin. Subsequently, Eq. (5) is derived analytically using the elementary properties of the proposed technique. For the analysis of the inclination effect, Table 1 is provided to witness the variation in the transient thermal profile (Theta left( {X,tau^{*} } right)) of the inclined porous fin with respect to different angles of inclination (alpha). It is detected from this table, that thermal distribution diminishes from base to tip of the porous fin for all non-dimensional variables considered in the study (Nc = 1), (S_{H} = 10), (Nr = 1), (Theta_{a} = 0.1), (k_{r} = 0.1), (C = 0.1), (phi = 0.1), (gamma = 0.1), (Q = 0.8) at different values of inclination angle. The thermal profile signifies greater thermal distribution at (alpha = 0), resulting in a lower rate of heat transfer. Meanwhile, as the (alpha) value is modified ((alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6},,{pi mathord{left/ {vphantom {pi 3}} right. kern-nulldelimiterspace} 3},{pi mathord{left/ {vphantom {pi 2}} right. kern-nulldelimiterspace} 2})), the thermal profile of the fin reveals decreasing characteristics at all locations of fin length. The consequence of the aforementioned dimensionless parameters on the thermal gradient (Theta left( {tau^{*} ,X} right)) of the fin is graphically examined in this section. The nonlinear PDE (Eq. (5)) is solved using the finite difference method (FDM) technique in the domain (0 le X le L) and (0 le tau^{*} le T). Along with uniform mesh, the finite-difference approximation is applied in the direction of (X) and the step size of time and space domains are chosen as (Delta tau^{*} = Delta X = 0.001). Further, the present analysis (GRPSM) is compared with the numerical result (FDM) as displayed in Fig. 2 and they are found to be in excellent agreement. The significance of dimensionless parameters on the temperature deviance of the fin has been portrayed in Figs. 3, 4, 5, 6 and 7.
Validation of the present result with the numerical method.
(a) Nature of (Theta left( {X,tau^{*} } right)) for various (Nc) values (b) Nature of (Theta left( {X,tau^{*} } right)) for various (Nr) values.
(a) Nature of (Theta left( {X,tau^{*} } right)) for various (Theta_{a}) values (b) Nature of (Theta left( {X,tau^{*} } right)) for various (S_{H}) values.
(a) Nature of (Theta left( {X,tau^{*} } right)) for various (Q) values (b) Nature of (Theta left( {X,tau^{*} } right)) for various τ* values.
(a) Nature of (Theta left( {X,tau^{*} } right)) for (tau^{*} = 0.1) (b) Nature of (Theta left( {X,tau^{*} } right)) for various (tau^{*} = 0.3).
(a) Nature of (Theta left( {X,tau^{*} } right)) of solid non-porous fin (b) Nature of (Theta left( {X,tau^{*} } right)) of porous fin.
The impact of convective parameter (Nc) on the temperature distribution of the concave fin is seen in Fig. 3a for (S_{H} = 0.5), (Nr = 1), (Theta_{a} = 0.1), (k_{r} = 0.1), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.8), and (tau^{*} = 0.5) by considering different (C) values ((C = 0.1,{text{and}},C = 0.3)). It is detected that with a rise in the convective parameter ((Nc = 1,,3,,5)) thermal profile decreases. This is due to the effect of the natural convection on the concave surface of the fin. The convection will carry the heat on fin surface and hence helps in decreasing the heat and aids in fin cooling. With all the considered parameter (Nc = 1), (S_{H} = 0.6), (Theta_{a} = 0.2), (k_{r} = 0.1), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.8), Fig. 3b displays the impact of radiative parameter (Nr) on the thermal performance of the concave fin. The radiative parameter exhibits the same nature as that of the convective parameter, i.e., a decrease in temperature profile with increase in (Nr) (2, 4, 6). The radiation effect increases the transfer of heat from fin surface to the surrounding. Therefore, the reduction in temperature has been observed with rising in radiation parameter. The significance of ambient temperature (Theta_{a} left( {0,,0.3,,0.5} right)) on the fin thermal field of a concave inclined porous fin with (Nc = 1), (S_{H} = 0.2), (Nr = 1), (k_{r} = 0.1), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.8), and (tau^{*} = 0.5) is exhibited in Fig. 4a. The enhancement in ambient temperature results in an increase in thermal profile. This is because, (Theta_{a}) is the ratio of surrounding temperature to base temperature. Hence, with augment in (Theta_{a}) there is a decrease in the transmission of heat from fin surface to surrounding which leads to the increase in thermal profile. By setting (Nc = 2), (Nr = 1), (Theta_{a} = 0.3), (k_{r} = 0.1), (phi = 0.1),(gamma = 0.1) , (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.7), and (tau^{*} = 0.5), the effect of (S_{H}) on (Theta) of the fin has been revealed in Fig. 4b. The enhancement in (S_{H}) ((1,,3,,5)) leads to the reduction in the temperature profile. This is due to the fact that the porosity parameter helps in better interaction of surrounding air with the pores of the fin. Hence the porosity parameter helps in the fin cooling effect. The consequences of heat generation parameter (Q) on the (Theta) of the fin have been depicted in Fig. 5a with (Nc = 1), (S_{H} = 0.3), (Nr = 1), (Theta_{a} = 0.2), (k_{r} = 0.1), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), and (tau^{*} = 0.5). Here the augment in (Q) ((0,,0.4,,0.8)) enlarges the thermal value along the axial length of the fin because of the presence of internal heat within the fin. The internal heat enhances the fin surface temperature therefore it decreases the rate of cooling from the surface of the fin. Figure 5b explicates the impact of (tau^{*}) on (Theta) of the concave porous fin. As the (tau^{*}) ((0.2,,0.3,,0.4)) values increase, the temperature improves remarkably under the consideration of (Nc = 1), (S_{H} = 0.1), (Nr = 1), (Theta_{a} = 0.1), (k_{r} = 0.1), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.8). Figure 6a and b demonstrate the variation in the unsteady thermal distribution of the concave inclined porous fin as a function of time in a three-dimensional (3D) plot for (Nc = 1), (S_{H} = 0.1), (Nr = 1), (Theta_{a} = 0.1), (k_{r} = 0.1), (C = 0.3), (phi = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), and (Q = 0.8). In particular, in Fig. 6a, the value of (tau^{*}) is picked as 0.1, and for this value, a decrement in thermal variation has been discovered, whereas an increment in thermal variation is recognized for (tau^{*} = 0.3), as exhibited in Fig. 6b. A comparison of the thermal profile values of the concave inclined solid and porous fin has been performed to uphold the significance of modeled fin problem. For a better interpretation of the thermal variation of the concave inclined porous and solid fin, Fig. 7a and b are plotted when the parameter values are set to (Nc = 1), (Nr = 1), (Theta_{a} = 0.1), (k_{r} = 0.1), (C = 0.1), (gamma = 0.1), (alpha = {pi mathord{left/ {vphantom {pi 6}} right. kern-nulldelimiterspace} 6}), (Q = 0.8), and (tau^{*} = 0.5). At (X = 0), the thermal field value of the solid fin (left( {S_{H} = phi = 0} right)) is comparatively higher than the porous fin (left( {S_{H} = 10,,phi = 0.1} right)) and the same behavior is observed at all the considered values of (X) (0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6). Thus, it is deduced that the porous fin helps in thermal dissipation and thereby provides a higher heat transfer rate. The effect of the fin taper ratio is recognizable in all of the aforementioned thermal investigation cases, and it is determined that the temperature inside the fin decreases as the scale of (C) (0.1, 0.3) increases.
The current study explains transient thermal dispersion in a concave permeable fin exposed to convective-radiative heat transfer. The governing equation is nondimensionalized by employing non-dimensional terms, and the arising PDE is solved analytically using the GRPSM. The consequences of significant non-dimensional factors on the temperature gradient are also depicted utilizing graphical portrayal. The following are some of the most significant study results of the present research:
Unlike prior investigation, the heat transfer performance of the fin is effected not only by the radiation and convection mechanisms but also by the fin taper ratio, and inclination of the primary surface.
The thermal dispersal in the porous fin is elevated in increasing time.
With a greater level of fin taper ratio, the fin undergoes diminishing thermal dispersal.
As the convection-conduction and porosity parameter values rise, the thermal dispersal in the fin declines.
The temperature profile of the fin exhibits an enhanced nature for the raised heat-generating parameter.
The thermal dispersal in the fin drops as the magnitude of the radiation-conduction attribute rises.
It may be highly beneficial to a designer to select the tapered fin design with an inclined vertical surface for thermal effectiveness in a practical application such as latent heat thermal energy storage systems.
GRPSM provides an analytical solution for nonlinear differential equations and the obtained outcomes in this research work are in close agreement with numerical ones. The exactness of the proposed method signifies that GRPSM is an alternative to other techniques for solving nonlinear PDEs.
All the data are clearly available in the manuscript.
Thermal conductivity
Convection–conduction parameter
Angle of inclination
Time
Emissivity
Stefan–Boltzmann constant
Acceleration due to gravity
Fin taper
Permeability
Specific heat
Radiation–conduction parameter
Convective heat transfer coefficient
Density
Width of the fin
Non-dimensional temperature
Local semi-fin thicknesses
Porosity
Fin taper ratio
Semi-base thickness
Fin’s length (dimensionless)
Internal heat generation parameter (dimensionless)
Porosity parameter
Length of the fin
Time(dimensionless)
Internal heat generation
Volumetric expansion index
Fin axial distance
Temperature
Ambient
Base
Solid
Relative quantity
Aly, E. H. & Pop, I. MHD flow and heat transfer near stagnation point over a stretching/shrinking surface with partial slip and viscous dissipation: Hybrid nanofluid versus nanofluid. Powder Technol. 367, 192–205 (2020).
CAS  Article  Google Scholar 
Song, Y.-Q. et al. Bioconvection analysis for Sutterby nanofluid over an axially stretched cylinder with melting heat transfer and variable thermal features: A Marangoni and solutal model. Alex. Eng. J. 60, 4663–4675 (2021).
Article  Google Scholar 
Bilal, M., Arshad, H., Ramzan, M., Shah, Z. & Kumam, P. Unsteady hybrid-nanofluid flow comprising ferrousoxide and CNTs through porous horizontal channel with dilating/squeezing walls. Sci. Rep. 11, 12637 (2021).
ADS  CAS  Article  Google Scholar 
Madhukesh, J. et al. Physical insights into the heat and mass transfer in Casson hybrid nanofluid flow induced by a Riga plate with thermophoretic particle deposition. Proc. Inst. Mech. Eng. Part E J. Process Mech. Eng. https://doi.org/10.1177/09544089211039305 (2021).
Article  Google Scholar 
Waini, I., Ishak, A., Pop, I. & Nazar, R. Dusty hybrid nanofluid flow over a shrinking sheet with magnetic field effects. Int. J. Numer. Methods Heat Fluid Flow 32, 1067–1091 (2021).
Article  Google Scholar 
Nadeem, M., Siddique, I., Awrejcewicz, J. & Bilal, M. Numerical analysis of a second-grade fuzzy hybrid nanofluid flow and heat transfer over a permeable stretching/shrinking sheet. Sci. Rep. 12, 1631 (2022).
ADS  CAS  Article  Google Scholar 
Algehyne, E. A. et al. Investigation of thermal performance of Maxwell hybrid nanofluid boundary value problem in vertical porous surface via finite element approach. Sci. Rep. 12, 2335 (2022).
ADS  CAS  Article  Google Scholar 
Ndlovu, P. L. & Moitsheki, R. J. Steady state heat transfer analysis in a rectangular moving porous fin. Propuls. Power Res. 9, 188–196 (2020).
Article  Google Scholar 
Madhura, K. R., Kalpana, G. & Makinde, O. D. Thermal performance of straight porous fin with variable thermal conductivity under magnetic field and radiation effects. Heat Transf. 49, 5002–5019 (2020).
Article  Google Scholar 
Nabati, M., Jalalvand, M. & Taherifar, S. Sinc collocation approach through thermal analysis of porous fin with magnetic field. J. Therm. Anal. Calorim. 144, 2145–2158 (2021).
CAS  Article  Google Scholar 
Kundu, B. & Yook, S.-J. An accurate approach for thermal analysis of porous longitudinal, spine and radial fins with all nonlinearity effects: Analytical and unified assessment. Appl. Math. Comput. 402, 126124 (2021).
MathSciNet  MATH  Google Scholar 
Buonomo, B., Cascetta, F., Manca, O. & Sheremet, M. Heat transfer analysis of rectangular porous fins in local thermal non-equilibrium model. Appl. Therm. Eng. 195, 117237 (2021).
Article  Google Scholar 
Varun Kumar, R. S., Sarris, I. E., Sowmya, G., Madhukesh, J. K. & Prasannakumara, B. C. Effect of electromagnetic field on the thermal performance of longitudinal trapezoidal porous fin using DTM-Pade approximant. Heat Transf. 51, 3313–3333 (2022).
Article  Google Scholar 
Aziz, A. & Fang, T. Alternative solutions for longitudinal fins of rectangular, trapezoidal, and concave parabolic profiles. Energy Convers. Manag. 51, 2188–2194 (2010).
Article  Google Scholar 
Torabi, M., Aziz, A. & Zhang, K. A comparative study of longitudinal fins of rectangular, trapezoidal and concave parabolic profiles with multiple nonlinearities. Energy 51, 243–256 (2013).
Article  Google Scholar 
Kang, H. S. Analysis of a concave parabolic fin with vertically cutting fin tip and variable fin base thickness. Int. J. Heat Mass Transf. 173, 121258 (2021).
Article  Google Scholar 
Wang, F. et al. LSM and DTM-Pade approximation for the combined impacts of convective and radiative heat transfer on an inclined porous longitudinal fin. Case Stud. Therm. Eng. 35, 101846. https://doi.org/10.1016/j.csite.2022.101846 (2022).
Article  Google Scholar 
Goud, J. S. et al. Role of ternary hybrid nanofluid in the thermal distribution of a dovetail fin with the internal generation of heat. Case Stud. Therm. Eng. 35, 102113 (2022).
Article  Google Scholar 
Jagadeesha, K. C. et al. A physical depiction of a semi-spherical fin unsteady heat transfer and thermal analysis of a fully wetted convective-radiative semi-spherical fin. J. Indian Chem. Soc. https://doi.org/10.1016/j.jics.2022.100457 (2022).
Article  Google Scholar 
Das, R. & Kundu, B. New forward and inverse solutions for wet fins generalized profiles with all nonlinear phenomena. J. Heat Transf. 143 (2020).
Zeeshan, A., Arain, M. B., Bhatti, M. M., Alzahrani, F. & Beg, O. A. Radiative bioconvection nanofluid squeezing flow between rotating circular plates: Semi-numerical study with the DTM-Padé approach. Mod. Phys. Lett. B (2021).
John Christopher, A., Magesh, N., Punith Gowda, R. J., Naveen Kumar, R. & Varun Kumar, R. S. Hybrid nanofluid flow over a stretched cylinder with the impact of homogeneous–heterogeneous reactions and Cattaneo-Christov heat flux: Series solution and numerical simulation. Heat Transf. 50, 3800–3821 (2021).
Article  Google Scholar 
El Ibrahimi, M. & Samaouali, A. Closed-form approximate solution for heat transfer analysis within functionally graded plate with temperature-dependent thermal conductivity. Compos. Struct. 271, 114140 (2021).
Article  Google Scholar 
Sowmya, G., Varun Kumar, R. S., Alsulami, M. D. & Prasannakumara, B. C. Thermal stress and temperature distribution of an annular fin with variable temperature-dependent thermal properties and magnetic field using DTM-Pade approximant. Waves Random Complex Media 0, 1–29 (2022).
Varun Kumar, R. S. et al. Exploration of transient heat transfer through a moving plate with exponentially temperature-dependent thermal properties. Waves Random Complex Media 0, 1–19 (2022).
Sun, Y. et al. Investigation of transient coupled conduction and radiation heat transfer in the linearly anisotropic scattering cylindrical medium by spectral collocation method. Int. J. Therm. Sci. 172, 107308 (2022).
Article  Google Scholar 
Weera, W. et al. Convective-radiative thermal investigation of a porous dovetail fin using spectral collocation method. Ain Shams Eng. J. 7, 101811. https://doi.org/10.1016/j.asej.2022.101811 (2022).
MathSciNet  Article  Google Scholar 
Biswal, U., Chakraverty, S., Ojha, B. K. & Hussein, A. K. Numerical simulation of magnetohydrodynamics nanofluid flow in a semi-porous channel with a new approach in the least square method. Int. Commun. Heat Mass Transf. 121, 105085 (2021).
CAS  Article  Google Scholar 
Abu Arqub, O., Abo-Hammour, Z., Al-Badarneh, R. & Momani, S. A reliable analytical method for solving higher-order initial value problems. Discrete Dyn. Nat. Soc. 2013, e673829 (2013).
MathSciNet  Article  Google Scholar 
Abu Arqub, O., El-Ajou, A., Bataineh, A. S. & Hashim, I. A representation of the exact solution of generalized lane-emden equations using a new analytical method. Abstr. Appl. Anal. 2013, e378593 (2013).
MathSciNet  Article  Google Scholar 
Syam, M. I. Analytical Solution of the Fractional Fredholm Integrodifferential Equation Using the Fractional Residual Power Series Method. Complexity 2017, e4573589 (2017).
MathSciNet  Article  Google Scholar 
Az-Zo’bi, E. A., Yıldırım, A. & AlZoubi, W. A. The residual power series method for the one-dimensional unsteady flow of a van der Waals gas. Phys. Stat. Mech. Appl. 517, 188–196 (2019).
MathSciNet  Article  Google Scholar 
Chu, Y. M., Nazir, U., Sohail, M., Selim, M. M. & Lee, J. R. Enhancement in thermal energy and solute particles using hybrid nanoparticles by engaging activation energy and chemical reaction over a parabolic surface via finite element approach. Fractal Fract. 5, 17 (2021).
Article  Google Scholar 
Zhao, T. H. et al. A fuzzy-based strategy to suppress the novel coronavirus (2019-NCOV) massive outbreak. Appl. Comput. Math 20, 160–176 (2021).
MathSciNet  MATH  Google Scholar 
Nazeer, M. et al. Theoretical study of MHD electro-osmotically flow of third-grade fluid in micro channel. Appl. Math. Comput. 420, 126868 (2022).
MathSciNet  MATH  Google Scholar 
Chu, Y. M. et al. Combined impact of cattaneo-christov double diffusion and radiative heat flux on bio-convective flow of maxwell liquid configured by a stretched nano-material surface. Appl. Math. Comput. 419, 126883 (2022).
MathSciNet  MATH  Google Scholar 
Zhao, T. H., Khan, M. I. & Chu, Y. M. Artificial neural networking (ANN) analysis for heat and entropy generation in flow of non-Newtonian fluid between two rotating disks. Math. Methods Appl. Sci. https://doi.org/10.1002/mma.7310 (2021).
Article  Google Scholar 
Khan, N. M., Chu, Y. M., Khan, M. I., Kadry, S. & Qayyum, S. Modeling and dual solutions for magnetized mixed convective stagnation point flow of upper convected Maxwell fluid model with second-order velocity slip. Math. Methods Appl. Sci. https://doi.org/10.1002/mma.6824 (2020).
Article  Google Scholar 
Ma, J., Sun, Y. & Li, B. Simulation of combined conductive, convective and radiative heat transfer in moving irregular porous fins by spectral element method. Int. J. Therm. Sci. 118, 475–487 (2017).
CAS  Article  Google Scholar 
Modanli, M., Abdulazeez, S. T. & Husien, A. M. A residual power series method for solving pseudo hyperbolic partial differential equations with nonlocal conditions. Numer. Methods Partial Differ. Equ. 37, 2235–2243 (2021).
MathSciNet  Article  Google Scholar 
Download references
The authors acknowledge the financial support provided by the Center of Excellence in Theoretical and Computational Science (TaCS-CoE), KMUTT. This research was funded by National Science, Research and Innovation Fund (NSRF), and King Mongkut’s University of Technology North Bangkok with Contract no. KMUTNB-FF-65-24. The authors would like to thank the Deanship of Scientific Research at Umm Al-Qura University for supporting this work by Grant Code: 22UQU4331317DSR18.
Department of Mathematics, Davangere University, Davangere, Karnataka, 577002, India
R. S. Varun Kumar & B. C. Prasannakumara
Department of Mathematics, M S Ramaiah Institute of Technology, Bangalore, Karnataka, 560054, India
G. Sowmya
Department of Information Technology, University of Technology and Applied Sciences, Al Mussanah, Sultanate of Oman
M. C. Jayaprakash
Department of Mathematics and Statistics, Riphah International University, I-14, Islamabad, 44000, Pakistan
M. Ijaz Khan
Nonlinear Analysis and Applied Mathematics (NAAM)-Research Group, Department of Mathematics, Faculty of Sciences, King Abdulaziz University, P.O. Box 80203, Jeddah, 21589, Saudi Arabia
M. Ijaz Khan
Mechanical Engineering Department, College of Engineering and Islamic Architecture, Umm Al-Qura University, P.O. Box 5555, Makkah, 21955, Saudi Arabia
Kamel Guedri
Center of Excellence in Theoretical and Computational Science (TaCS-CoE), KMUTT Fixed Point Research Laboratory, Room SCL 802 Fixed Point Laboratory, Science Laboratory Building, Departments of Mathematics, Faculty of Science, King Mongkut’s University of Technology Thonburi (KMUTT), 126 Pracha-Uthit Road, Bang Mod, Thung Khru, Bangkok, 10140, Thailand
Poom Kumam
Department of Medical Research, China Medical University Hospital, China Medical University, Taichung, 40402, Taiwan
Poom Kumam
Intelligent and Nonlinear Dynamic Innovations Research Center, Department of Mathematics, Faculty of Applied Science, King Mongkut’s University of Technology North Bangkok (KMUTNB), 1518, Wongsawang, Bangsue, Bangkok, 10800, Thailand
Kanokwan Sitthithakerngkiet
Mechanical Engineering Department, College of Engineering, Prince Sattam Bin Abdulaziz University, Wadi ad-Dawasir, 11991, Saudi Arabia
Ahmed M. Galal
Production Engineering and Mechanical Design Department, Faculty of Engineering, Mansoura University, P.O. 35516, Mansoura, Egypt
Ahmed M. Galal
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
Conceptualization: R.S.V.K. and G.S., methodology: M.I.K. and M.C.J.; writing-original draft preparation K.G. and B.C.P., writing-review and editing: K.S. and A.M.G.; funding acquisition: P.K., investigation, supervision: B.C.P. and M.I.K. All authors have read and agreed to the published version of the manuscript.
Correspondence to Poom Kumam or Ahmed M. Galal.
The authors declare no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Varun Kumar, R.S., Sowmya, G., Jayaprakash, M.C. et al. Assessment of thermal distribution through an inclined radiative-convective porous fin of concave profile using generalized residual power series method (GRPSM). Sci Rep 12, 13275 (2022). https://doi.org/10.1038/s41598-022-15396-z
Download citation
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41598-022-15396-z
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Advanced search
Scientific Reports (Sci Rep) ISSN 2045-2322 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

StarX MusicX Lab Enters the era of Digital Content Creation with the Release of Its First AI-composed Songs – Yahoo Finance

BEIJING, July 19, 2022 /PRNewswire/ — Kunlun Tech’s social entertainment platform StarX MusicX Lab, specializing in AI composition, recently released five songs, all of which were composed by AI software. The five AI-generated content (AIGC) compositions are now available on more than 180 music platforms worldwide including Spotify, SoundClould, QQ Music and NetEase Cloud Music.
Music labels, singers and music producing organizations have started to use the AIGC songs StarX MusicX Lab produced. The songs, many of which have been recorded by mainstream artists in Indonesia and Malaysia, are already proving popular on StarX’s entertainment social platform StarMaker.
In the Web3.0 era, a new generation of platforms is on the rise. “AIGC is emerging as a key player in this new form of content creation,” says Kunlun Tech CEO Han Fang. “To this end, StarX MusicX Lab has started rolling out AIGC songs, providing a new kind of AI application scenario and giving us the freedom to let our imagination fly in terms of what is possible when it comes to content creation.”
The efforts by StarX MusicX Lab marked a material change in how AI-composed audio is produced. With the latest technological advances now having been converted into applications, AI can do what is truly imaginative, making a contribution to the harmony between the virtual and real worlds. Over time, more AI-originated creations are bound to go mainstream.
Based on AI application scenarios, the Lab is a part of Kunlun Tech dedicated to developing AI-composed musical tracks. With sophisticated and professional full-chain music production and worldwide distribution capabilities, the platform is able to deliver high-quality singles, soundtracks and ambient music to the four corners of the globe.
The Lab is equipped with a highly-efficient and low-cost AI composer, which provides creators with the capability of producing multiple forms of content, no matter whether what is created is destined for an application in social media, gaming and entertainment, film and television production, or a marketing campaign.
With the development and improvement of AI technology, AIGC will be more helpful than ever in terms of content creation by improving the efficiency and diversity of content production, while offering more dynamic and interactive content. The massive virtual world that is the Metaverse spans a wide range of sectors including AI, gaming and social media, among many others. In the medium and long term, AIGC can be expected to bring innovation to the virtual world and enhance the co-prosperity of all links in the industry chain by identifying and leveraging new growth points.
The next stages of development in the Metaverse require not only technological progress, but also content-side support. In 2021, Kunlun Tech completed its Metaverse portfolio and the build out if its AI-driven applications for social entertainment and information distribution. As AIGC evolves, the Metaverse business will serve as a new contributor to Kunlun Tech’s exponential growth.
China’s top Internet firms, including Byte and Baidu, have already entered and are quickly making progress in the realm of AIGC, while StarX MusicX Lab has already taken the lead in related technologies.
The digital human ecology built around the virtual world and the Metaverse infrastructure has begun to take shape, and so did the music ecology built around the new media of the Internet. The wide application of AI technologies will dramatically change how content is created. As AIGC evolves, StarX MusicX Lab will continue to expand its business footprint, as the firm further improves its creation capabilities, while producing exclusive songs for the world’s influencers.
Supported by increasingly stronger computing power, AI that features self-computing and greater adaptability will be a perfect match for content production needs by unleashing the vitality of content creation and by making the creative space more diverse and dynamic. With the popularity of AI products and the growing demands of fans, StarX’s creative content destined for application in the entertainment, audio, and social media sectors is bound to have even greater potential and room for development.
SOURCE Kunlun Tech
Tesla CEO Elon Musk raised the stakes in his court battle with Twitter on Friday, filing a counter lawsuit that accuses the social media company of fraud.
Apple and Amazon have thrown billions of dollars at sports media rights over the last few months, outbidding traditional competitors and spiraling the streaming wars into its next stage of battle.
Warner Bros. Discovery's recent HBO Max retreat may signal good things to come for competitor Netflix.
Box office receipts hit a nearly three-month low this weekend, but there's still time for a Hollywood ending.
AMC CEO Adam Aron has become a hero to retail investors amid the meme stock era, but one analyst argues that a new offering of AMC Preferred Equity (APE) shares is disingenuous and ultimately bad for investors.
After the company's sale setback, Kohl's is pivoting toward inclusion, premier denim, and a whole new fashion attitude.
Singer Nolan Neal, who was a contestant on both America’s Got Talent and The Voice has died. He was 41. Neal was found dead in his Nashville apartment on Monday, July 18, with no cause of death yet determined. That said, he did have a self-admitted history of drug abuse. His passing was announced …
So many things I saw on Ugly Betty were a lie.View Entire Post ›
Is "The Miz" in a new documentary series about Woodstock '99? Eagle-eyed viewers of Netflix’s recent three-part documentary series “Trainwreck: Woodstock ‘99” raised their eyebrows during the first few moments of Episode 1 when the camera passes over the crowd and a young, shirtless man who looks an awful lot like the Miz shouts “Woodstock ‘99 baby, yeah!” directly into the camera. The documentary series is the latest to explore the disastrous music festival that took place in 1999, three years
The star of Marvel's "She-Hulk" isn't afraid to use her clout to support trans rights, VFX artists facing "incredibly difficult" conditions and more.
News Corporation's (NWSA) better-than-expected fourth-quarter fiscal 2022 results are driven by gains from revenue growth across most segments and contributions from the recent buyouts.
When looking for the best artificial intelligence stocks to buy, identify companies using AI technology to improve products or gain a strategic edge, such as Google, Microsoft and Nvidia.
Are BTS's J-Hope and model Irene Kim more than just friends?
'Jeopardy!' is airing reruns while the quiz show is currently on hiatus before new episodes. 'Jeopardy!' fans had strong reactions to seeing Matt Amodio again before the 2022 Tournament of Champions.
The Tasty singer replied to Bey Hive members on social media.
The changes will let users silently leave group chats and control who sees their online status.
After reports that the Queen had invited Meghan Markle and Prince Harry to Balmoral, sources claim the Sussexes haven't been invited after all.
Pitt apparently has a "good list" and a "s**t list" for actors he's worked with over his career, according to his co-star.
Thought it’s only been a little over a week since Beyoncé dropped her seventh studio album Renaissance, the long-awaited project has already further cemented Queen Bey’s status in history as it’s landed at the number one spot on the Billboard 200 album chart.
'Grease' actor Eddie Deezen cannot face trial for his charges of assault because he is not mentally fit, according to a state order. Deezen was arrested in April.

source

Live news updates from August 8: China continues military drills around Taiwan, Falling tech stocks send SoftBank to $23.3bn loss – Financial Times

Expert insights, analysis and smart data help you cut through the noise to spot trends, risks and opportunities.
Join over 300,000 Finance professionals who already subscribe to the FT.
Then $69 per month
New customers only
Cancel anytime during your trial
OR
BEST VALUE – SAVE 20%
Then $74.75 every 3 months
Sign in
Check if your university has an FT membership to read for free.
We use cookies for a number of reasons, such as keeping FT Sites reliable and secure, personalising content and ads, providing social media features and to analyse how our Sites are used.
International Edition

source

The best places to live in North America – The Economist

For a look behind the scenes of our data journalism, sign up to Off the Charts, our weekly newsletter
OVER 630,000 people moved to North America from other parts of the world in the first half of 2022, a rise of 51% from the same period a year earlier. It is easy to see why. The region is home to some of the world’s most “liveable” cities, according to an index by EIU, our sister company. It ranks living conditions in 172 cities around the world, taking into account more than 30 factors related to education, culture, the environment, health care, infrastructure and stability. This year, every city in North America included in the index scored above 80 (out of 100), with an average of 88. That makes North America the second-most “liveable” region in the world, narrowly behind Western Europe. Which of its cities stood out?
Those looking for a new home may want to consider Calgary. The Canadian city scored 96 points, the most of any North American city in the region. It is not its first time at the top: Calgary has come first in the region twice before in the past decade. Vancouver, Toronto and Montreal filled the next three spots. Residents of all four Canadian cities enjoy top scores for health care and education. Canada’s rollout of covid-19 vaccine boosters and the quick reopening of schools during the pandemic aided its high marks.
The top ranked city in the United States was Atlanta, which came fifth overall. Los Angeles, the region’s biggest climber, rose to ninth place from 15th, thanks to big gains in the culture and environment category after it eased capacity restrictions for venues in February. New York, meanwhile, remained close to the bottom in 21st place—two below its pre-pandemic rank. The city scored poorly on stability and infrastructure (as New Yorkers worried about high crime rates or stuck on an ageing subway will understand). Mexico is not included in EIU’s North American index, but its cities were among the bottom 50 of the global survey. When it comes to liveability, it seems the further north you go the better. ■
A look at excess deaths suggest summers will become more deadly as temperatures rise
The country wants to resume its pre-war export levels by the autumn
The overturning of Roe v Wade has also pushed Democrats and women to the polls
Published since September 1843 to take part in “a severe contest between intelligence, which presses forward, and an unworthy, timid ignorance obstructing our progress.”
Copyright © The Economist Newspaper Limited 2022. All rights reserved.

source

Shambhala Music Festival Was an Extraordinary Experience – EDM Identity

Shambhala Music Festival this year was an incredible experience and worth the three years it took to come back to fruition. To give an accurate review of this year’s Shambhala, you must know what Shambhala encompasses. Shambhala music festival is a 3-day festival at the Salmo River Ranch in the gorgeous forests of British Columbia.
Though it’s advertised as a three-day festival, attendees can arrive as early as Tuesday, leave as late as the following Tuesday, and most do. This is an excellent way for attendees to beat the line traffic getting in, meet their neighbors, and decompress and sleep after the festival to prepare to drive home. Music starts Thursday, but sound camps throughout the campgrounds go all hours. Having been unable to attend Shambhala since 2019, I was excited to make my way back to the festival – and I was impressed by the experience.
Even with all the additional days to enter, it’s been anywhere from three to upwards of 16 hours with only one available entrance. In 2019, arriving at the early hour of 7am still had us waiting six hours to enter. This year, an additional entry was added for staff which maybe cut the line down but not by as much as you would think. Though many Shambhala veterans have mastered the best times to try and enter, other first-timers this year mentioned wait times ranging from two to nine hours. Issues were credited to new software and being short-staffed.
That being said, although Shambhala was understaffed this year, they held it together incredibly well. The staff was very attentive, and even though medical crews were picking up extra shifts, every call was handled with care and alertness. I truly commend the Shambhala staff and volunteers for holding together what could have been a problematic disaster. The shortage was barely noticeable while being in attendance, and all in all, the event was handled smoothly.
From the incredible lasers at the Pagoda stage to the intense bass at the AMP stage, absolutely nothing was lackluster. Everywhere you turned, there was some sort of fascinating piece to look at or explore. For example, inside the festival grounds, there was a treasure hunt that I followed. We would stumble upon a box and a QR code next to it.
That QR code led to a personality quiz that ended with a riddle. The answer to the puzzle would unlock the box, which had little trinkets to take, and then the next riddle would lead you to the next box. Ultimately, the final box contained Shambhala merchandise! This treasure hunt was a great way for us to interact and meet other attendees also trying to figure out the riddle.
At each stage, there was an art exhibit walk-through that contained some of the most remarkable pieces of art I’ve ever seen. One notable piece interacted with an app on your phone, which, when held up to the painting, made it slightly 3-D and made the painting move and transform like a trippy scenic clip.
Shambhala is superb in creating areas around the event grounds and near stages that ideally offer places to sit or lay down and still enjoy the music and production. One of my favorite hangout spots was sitting in the bench areas out by the Grove stage and hanging by the beachside of the Salmo River, listening to the Living Room stage.
I made it my mission to try every single vendor at least once, and some of the standouts for me were the warm ricotta balls, a great snack at the end of the night when it was colder, and the cold salad wraps, perfect for the sweltering sunny day. One of the better parts of the food vendors here compared to other festivals is that the portion sizes are larger; therefore, you get more bang for your buck.
The only concern I had with the meal amenities was that the food was only sold within the festival grounds, which was a bit of a walk for attendees out in further camping areas. In the future, it would be nice to see more of a camper’s “lounge/vendor” area, where attendees could grab a coffee, food, or merchandise closer to the festival grounds.
While some don’t enjoy this rule, I find that it creates the best atmosphere. Being someone that attends many festivals which sell and allow alcohol, I’ve seen that it can create disruptive and sloppy attendees and, in some cases, dangerous; when people mix alcohol with other substances. Rarely did I see any attendees being rude or unable to function. It was so refreshing to have a high-spirited and loving crowd around at Shambhala instead of seeing people passed out everywhere.
It was so fun seeing the buzzing excitement pulsing every time someone mentioned the return of Shambhala after three years. Additionally, A Hundred Drums gave a standout performance on Thursday, flowing through a tantalizing groove of heavy, deep bass that entranced the crowd and connected to the heart of Shambhala.
Trapeze artists, flow artists, and an incredible vocalist came together to put on a 45-minute complete show unveiling the new, upgraded stage. The artists of Le Cirque de la Nuit gave one of the best performances I have seen, adding elements of fire, heights, swords, and heavy structural objects while performing to a custom score by Hollywood composer Zain Effendi. The unveil showed the replacement of the old geodesic structure with a new updated dojo-inspired stage design.
Having become renowned for his past performances at the festival, it made sense for Excision to christen the newly developed stage with a performance. He delivered one of the best classic sets I have seen from him in a while, reminiscent of his earlier years. The energy level from him and the crowd was unmatched as he cycled through multiple genres and had every single person jumping and dancing.
This included the spectacular visual array CloZee brought to the Pagoda while blessing us with a new unreleased track she had with Ganja White Night. The Sponges and Defunk brought funky vibes to the Fractal Forest, while the DNB Girls Takeover threw down in an epic three-hour showdown. Additionally, Deathpact unveiled on Sunday that the mysterious artist is actually two artists, leaving attendees with more questions than answers on who they could be.
This year, the Shambhala antics were a highlight for me. Some of these moments include seeing a sporadic, fun lightsaber fight ensue in the food market, volunteering to teach a silly water aerobics class in the river, and seeing random dance circles pop up with someone towing a prominent speaker around.
Starting from the first day of attendance, I befriended neighbors around my campsite and created friends I will continue to hold in my heart. A crew of Canadian Shambhala veterans welcomed us into their inner circle and were adamant about showing us around new artists and spending time including us in their ventures.
From hanging out on a giant ducky in the middle of the river to watching one of their friends, Metafloor, absolutely crush it with some deep, bass-heavy beats at the Cedar Lounge, complete with their own dancing pole and gumball machine, this crew truly embodies the spirit of Shambhala. We even participated in their group’s themed outfits created by their merchandise line Swerve Supply Co.
A notable moment for many attendees was the number of animals in attendance. I’m not talking about service animals; this one is aimed at the farm animals. Starting Friday night, the cows in a nearby fenced-off pen managed to escape and roam around the campground tents. Though the cows were harmless, animals such as wild turkeys and bears, which are known to be more aggressive, were also roaming the campgrounds.
As the ‘farmily’ expands, a few logistical items can be reviewed to improve future experiences. Water stations this year were few and far between, as many of them were rarely full over the weekend. Many attendees in the ShambaLodging had to go inside the festival grounds even to get water. Additionally, the water stations were not distinctively marked within the grounds, and for newcomers, they proved challenging to find, especially as the heat levels rose.
Shambhala has proven time and time again that it truly is a safe and loving place that brings in not just an array of exquisite music but also the community festivals first began with. If you haven’t been yet, I strongly recommend it; you won’t regret the experience and the friends you make.
Website | Facebook | Twitter | Instagram | SoundCloud | YouTube

source

Digital Audio Workstations (DAWs) Market with Tremendous growth by 2029 – Ableton Live, Acoustica, Adobe – openPR

Permanent link to this press release:

All 5 Releases

source

‹ Previous Next ›