Following its inception in the mid-19th century, our understanding of thermodynamic entropy has undergone many revisions, most notably through the development of microscopic descriptions by Boltzmann and Gibbs, which led to a deep understanding of equilibrium thermodynamics. The role of entropy has since moved beyond the traditional boundaries of equilibrium thermodynamics, towards problems for which the development of a statistical mechanical theory seems plausible but the a-priori probabilities of states are not known, making the definition and calculation of entropy-like quantities challenging. In this talk, we will discuss two new classes of methods that enable these computations: one based on pattern matching ideas from information theory, and the other based on basin volume calculations. These approaches provide us with very general frameworks for computing entropy, density of states, and entropy production in systems far from equilibrium. We will discuss applications of these ideas to a variety of contexts: from granular systems, to absorbing-state models, to active matter, in simulations and in experiments. Throughout the talk, I will highlight challenges and promising future directions for these measurements.

The earliest stages of planetary assembly occur in a protoplanetary disk, the cloud of dust and gas that orbits a young star. The building blocks of planets are assembled in this disk, as dust grains collide and grow, and concentrate in regions to form planetesimals. Throughout this process, the disk itself evolves dynamically, driving mass onto the star as part of the final stages of pre-main sequence growth, while also expanding in size due to angular momentum conservation. All of these processes combine to change the physical environments present in the disk, from temperature and density to the flux of ultraviolet photons, which in turn leads to constantly changing chemistry. In this talk, I will discuss how we are investigating the complex feedbacks that exist between these processes, and what they are telling us about how planetary properties are set, both in our Solar System and around other stars.

Lyotropic chromonic liquid crystal (LCLC) solutions in the nematic phase have peculiar properties. They are tumbling materials, which means that flows can easily destabilize the director alignment, and they possess a large elastic anisotropy where twist elastic deformations are energetically much cheaper than splay or bend deformations. We show how these characteristics can be exploited to induce controlled growth morphology transitions from the generic dense-branching growth to dendritic growth in the viscous- fingering instability, and how they lead to unique structure formation as the LCLC solutions are driven out-of-equilibrium by a pressure-driven flow in a microfluidic channel. In particular, we report the surprising emergence of chiral domains despite the achiral nature of the material. The chirality results from a periodic double-twist deformation of the liquid crystal and leads to striking stripe patterns vertical to the flow direction. We discuss the mechanism of this unique pathway to spontaneous mirror symmetry breaking and rationalize the selection of a well-defined period of the chiral domains.

A beautiful example of spontaneous pattern formation occurs in certain dryland environments around the globe. Stripes of vegetation alternate with stripes of bare soil, with striking regularity and on a scale readily monitored via satellites. Positive feedbacks, between infiltration of water into the soil and the vegetation itself, help concentrate this essential, but limited, resource into the vegetated zones. These feedbacks play out on the short timescales of the rare and unpredictable rainstorms that sustain life in these dry regions. In contrast, the vegetation may change very little over decades, aside from a gradual upslope colonization. In this talk I will tell you a little bit about the empirical data, the models, their analysis and the results of numerical simulations, focusing on the basic question of what sets the spacing of the vegetation bands. Our work suggests some new questions about how these fragile ecosystems might respond to changes in precipitation characteristics, such as storm strength, frequency, and seasonality, all of which are occurring as a consequence of climate change.

When we attend a demanding task, our performance is poor at low arousal (drowsy) or high arousal (anxious), but we achieve optimal performance at intermediate arousal, yielding the celebrated Yerkes-Dodson inverted-U law. Despite decades of investigations, the neural mechanisms underlying this inverted-U law are unknown. In this talk, I will elucidate the behavioral and neural mechanisms linking arousal and performance. I will show that mice during auditory and visual decision-making express an array of discrete strategies, including optimal, suboptimal and disengaged, as revealed by an HMM analysis. The optimal strategy occurs at intermediate arousal levels, measured by pupil size, consistent with the inverted-U law. Using neuropixels recordings from neural populations in the auditory cortex, we show that sound encoding is optimal at intermediate arousal level, suggesting that performance modulations occur as early as primary sensory cortex. To explain the computational principle underlying this inverted-U law, we show that in a recurrent network with E/I populations arousal induces a phase transition from a multi-attractor to a single attractor phase, and performance is optimized near the critical region. The model further predicts a monotonic decrease in neural variability induced by arousal, which we confirmed in the empirical data. Our results establish a biologically plausible theory of optimal performance based on phase transitions in attractor networks with E/I balance, whose implications for brain-inspired AI models will be briefly outlined.

Flocks of birds and schools of fish are delightful and awe-inspiring examples of collective motion that we see in nature, where groups of individuals, each possessing only limited, local information, nevertheless come together and display coordinated motion. This phenomenon also extends to much smaller scales, as in migrating clusters of cells that mediate physiological processes such as embryonic development, wound healing, and cancer metastasis. The collective, co-ordinated motion of cells allows for emergent behaviors unavailable to single cells that are critical for proper function. In this talk, I shall describe our work on modeling such phenomena in cancer cell clusters, highlighting how frustration can arise at the group level because of heterogeneity in behavior among individual cells in the cluster. I shall show how this frustration can be resolved leading to new collective phases of motion that are experimentally observed in malignant lymphocyte clusters and functionally important – enabling robust chemotaxis and “load sharing” among cells.

With conventional digital computing technology reaching its limits, there has been a renaissance in analog computing across a wide range of physical substrates. In this talk, I will introduce the concept of Physical Neural Networks [1] and describe a method my group has developed to train any complex physical system to perform as a neural network for machine-learning tasks. We have tested our method experimentally with three different systems – one mechanical, one electronic, and one photonic – and have been able to show MNIST handwritten-digit classification using each of these systems, despite the fact that none of the systems were initially designed to do machine learning.

I will describe several possible future research directions on Physical Neural Networks, including the potential to create large-scale photonic accelerators for server-side machine learning [2], smart sensors that pre-process acoustic, microwave or optical [3] signals in their native domain before digitization, new kinds of quantum neural network that don't require a carefully engineered quantum computer, and generally the prospect to endow analog physical systems with new, unexpected functionality.

[1] L.G. Wright*, T. Onodera* et al. Nature 601, 549-555 (2022)

[2] T. Wang et al. Nature Communications 13, 123 (2022)

[3] T. Wang*, M. Sohoni* et al. arXiv:2207.14293, to appear in Nature Photonics (2023)

I will review the complex phenomena displayed by one of the simplest physical systems one can think of: Three identical masses connected to each other by three harmonic springs on a frictionless plane. This system conserves angular momentum, yet can persistently rotate with zero total angular momentum. It has a well-understood regular harmonic oscillatory limit but can also show fully chaotic behavior. When combined together these two phenomena give rise to a rotational random walk with fractional exponents determined by a single parameter; the total energy of the system. We will show how the notions of Arnold diffusion, geometric phases and classical gauge field naturally arise, and how they can be intuitively understood in this simple system. No prior knowledge will be assumed.

Sequencing-based technologies allow resolving the composition of microbial ecosystems to strain-level detail; however, coarser representations are often found to be more reproducible and more predictive of community-level properties. The general principles for selecting an appropriate level of description for modeling remain elusive. I will describe a framework for systematically comparing all possible coarse-grained descriptions by explicitly quantifying their prediction power and information content, allowing us to define the Pareto front of optimal descriptions for a given property of interest. Crucially, this Pareto front depends on ecological context; in particular, a high diversity of strains (while nominally more complex) may, in fact, facilitate coarse-grainability. I'll discuss an empirical example of diversity-enhanced coarse-grainability, and show how our framework nuances the notion of "emergent simplicity" in microbial ecology.

Despite the great success of deep learning, it remains largely a black box. In this seminar, we will describe our recent work in understanding learning dynamics and generalization of deep neural networks based on concepts and tools from statistical physics.

SGD Learning dynamics: The main search engine in deep neural networks is the Stochastic Gradient Descent (SGD) algorithm. However, despite its effectiveness, little is known about how SGD finds ``good" solutions (low generalization error) in the high-dimensional weight space. By studying weight fluctuations in SGD, we find a robust inverse relation between the weight variance in SGD and the landscape flatness, which is the opposite to the fluctuation-dissipation(response) relation in equilibrium statistical physics. We show that the noise strength in SGD depends inversely on the landscape flatness, which explains the inverse variance-flatness relation. Our study suggests that SGD serves as an ``intelligent" annealing strategy where the effective temperature self-adjusts according to the loss landscape, which allows it to find the flat minimum regions that contain generalizable solutions. Time permit, we discuss some application of these insights for solving machine learning problems [1,2].

Geometric determinants of generalization: We first report the discovery of an exact duality relation between changes in activities in a densely connected layer of neurons and the changes in their weights connecting to the next layer. The activity-weight duality leads to an explicit expression for the generalization loss, which can be decomposed into contributions from different directions in weight space. We find that the generalization loss from each direction is the product of two geometric factors (determinants): sharpness of the loss landscape at the solution and the standard deviation of the dual weights, which scales as an activity-weighted norm of the solution. By using the generalization loss decomposition, we uncover how hyperparameters in SGD, different regularization schemes (e.g., weight decay and dropout), training data size, and labeling noise affect generalization by controlling one or both factors [3].

[1] “The inverse variance-flatness relation in Stochastic-Gradient-Descent is critical for finding flat minima”, Y. Feng and Y. Tu, PNAS, 118 (9), 2021.

[2] “Phases of learning dynamics in artificial neural networks: in the absence and presence of mislabeled data”, Y. Feng and Y. Tu, Machine Learning: Science and Technology (MLST), July 19, 2021. https://iopscience.iop.org/article/10.1088/2632-2153/abf5b9/pdf

[3] “The activity-weight duality in feed forward neural networks: The geometric determinants of generalization”, Y. Feng and Y. Tu, https://arxiv.org/abs/2203.10736

The survival of many microorganisms depends on their ability to navigate through complex biological environments. Complexity may come in many forms, from circuitous channels to non-Newtonian bulk fluid rheology. We will discuss analytical and numerical insights into swimming through model viscoelastic and anisotropic fluids, with a special focus on the outsized roles played by the presence of nearby boundaries. And we will propose a classification which seeks to organize a multitude of systems based on the relative sizes and timescales associated with the active bodies and any surrounding obstacles, broadly interpreted.

The emergence and subsequent evolution of ever-increasing complexity is one of the key signatures of life. It manifests itself in a dramatic reduction of the information entropy accompanied by the emergence of functional activity. Understanding the onset of such behavior in the early prebiotic world is essential for solving the problem of the origin of life. We studied a general problem of heteropolymers capable of nonenzymatic template-assisted replication and driven out of thermodynamic equilibrium by cyclic changes in the environment [1-4]. One of our central results is a dramatic reduction of sequence entropy [2] that was recently tested experimentally [3]. Another finding is the pathway to the emergence of an early catalytic function [4]. Specifically, we show how our system could acquire sequence-specific cleavage activity through functional differentiation of polymers into catalysts and substrates. This mechanism provides an escape route from a relatively simple pairwise replication of mutually complementary chains to a more complex behavior involving both information transfer and enzymatic activity.

[1] Tkachenko AV, Maslov S. (2015) Spontaneous emergence of autocatalytic information-coding polymers. J Chem Phys. 143(4):045102. [2] Tkachenko AV, Maslov S. Onset of natural selection in auto-catalytic heteropolymers (2018) J Chem Phys. 149, 134901 [3] Kudella PW, Tkachenko AV, Salditt A, Maslov S, Braun D. Structured sequences emerge from random pool when replicated by templated ligation. PNAS (2021);118(8). [4] Tkachenko AV, Maslov S. (2023) Emergence of catalytic function in prebiotic information-coding polymers. BioRxiv (submitted for publication).

Holes in elastic metamaterials, defects in 2D curved crystals, localized plastic deformations in amorphous matter and T1 transitions in epithelial tissue, are typical realizations of stress-relaxation mechanisms in different solid-like structures, interpreted as mechanical screening.

While screening theories are well established in other fields of physics, e.g. electrostatics, a unifying theory of mechanical screening applicable to crystalline, amorphous, and living-cellular matter, is still lacking. In this talk I will present a general mechanical screening theory that generalizes classical theories of solids, and introduces new moduli that are missing from the classical theories. Contrary to its electrostatic analog, the screening theory in solids is richer even in the linear case, with multiple screening regimes, predicting qualitatively new mechanical responses. Specifically, we predict a regime of screening that is mechanically similar to the celebrated Hexatic phase, in disordered matter.

The theory is tested in different physical systems, among which are disordered granular solids and models of epithelial tissue. Experiments and numeric simulations in granular, glass, and tissue models uncover a mechanical response that strictly deviate from classical elasticity, and is in full agreement with the theory. Finally I will discuss the relevance of the theory to 3D granular solids and a new Hexatic-like state in three-dimensional matter.

Biological systems must process large amounts of information with limited resources: neurons generate only a limited number of action potentials, genetic control signals are carried by molecules at low concentration. Is it possible that biologically relevant information is maximized given the physical constraints? This old idea has languished in part because theory has not been powerful enough to solve the relevant optimization problems in realistic settings. I’ll describe progress on this class of problems, using a genetic network in the fly embryo as inspiration. After more than a decade, we now have methods to search the 50+ dimensional parameter space of (moderately) realistic networks to find the ones that transmit as much positional information as possible with a limited number of molecules. Remarkably, these optimal networks look very much like the real network, and this agreement happens with all parameters determined by optimization.

There comes a time in each of our lives where we grab a thick section of the morning paper, roll it up and set off to do battle with one of nature’s most accomplished aviators - the fly. If, however, instead of swatting we could magnify our view and experience the world in slow motion we would be privy to a world-class ballet full of graceful figure-eight wing strokes, effortless pirouettes, and astonishing acrobatics. After watching such a magnificent display, who among us could destroy this virtuoso? How do flies produce acrobatic maneuvers with such precision? What control mechanisms do they need to maneuver? How is this control implemented at the neuromuscular scale? In this talk I will discuss how we are combining newly-emerging tools for neural manipulation with quantitative behavioral modeling to gain a deeper understanding of how neural circuits are organized to control such complex motor behaviors.

Many sensory systems act primarily to amplify a signal which is not statistically significant at the level of a single receptor. For example, bacteria must sense and respond to minute (~1%) changes in ligand concentrations to navigate shallow chemical gradients. To achieve amplification, receptors and signaling components are organized in a lattice. This structure consumes chemical energy and transmits a highly amplified read out of the time-derivative of concentration. At much larger scales certain snakes can sense the blackbody radiation from small mammals a meter away- by the heating of their pit organ by ~1mK, 1000 times less than the sensitivity of thermal-TRP channels that are their molecular thermometers.

In this talk I will focus on chemosensing; I will provide a physicist's overview of the remarkable experiments conducted on chemosensing arrays, along with a review of previous theoretical work. I will then present our model for the amplification mechanism, which proposes that amplification arises from proximity to an active critical point. In our models, amplification stems from the diverging susceptibility observed near critical points, and proximity is maintained through a mechanism inspired by self-organized criticality. I will also connect to recent work of ours on the snake pit organ, which while molecularly very different, uses a distinct critical point in an analogous way.

Active particles, such as swimming bacteria or self-propelled colloids, have the propensity to spontaneously organize into large-scale dynamic structures. Geometric constraints, however, often enforce different spatio-temporal patterns compared to unconfined environment and thus may serve to control active matter behavior. In this talk, I will discuss two findings from our Lab: (1) Quincke rollers in soft confinement (a droplet) drive droplet deformation and amoeba-like motility, and (2) Quincke rotors, hovering between two parallel planar boundaries, assemble into crystal-like monolayer that with increasing activity (particle rotation rate) “melts” and breaks into clusters and snaking chains.

My talk will have two parts on two different active biological systems. First, I will tell you how a single cell produces light to survive; then, I will explain how a huddle of chloroplasts in cells reacts to light to optimize plant life. Part I: Bioluminescence (light generation in living organisms) has mesmerized humans since thousands of years ago. I will first go over the recent progress in the physics of single-cell bioluminescence (PRL 125 (2), 028102, 2020) and then will go beyond and present a lab-scale model of bioluminescent breaking waves. Part II: To remain efficient during photosynthesis, plants can re-arrange the internal structure of cells by the active motion of chloroplasts. I will show that the chloroplasts can behave like a densely packed light-sensitive active matter, whose non-gaussian athermal fluctuations can lead to various self-organization scenarios, including active glassy dynamics under dim lights (PNAS 120 (3), 2216497120, 2023).

Away from equilibrium, energy can serve as a resource that opens the door to what seems like limitless potential for novel phenomena. To date, however, our understanding of energy’s role has been largely gleaned through individual case studies, leaving it an open question to identify universal principles. In this talk, I will discuss how advances in the field of nonequilibrium thermodynamics allow us to make such general quantitative statements. I will introduce a series of equalities and inequalities---akin to the Fluctuation-Dissipation theorem but valid arbitrarily far from equilibrium---that constrain a system’s sensitivity by its structure and how strongly it is driven away from equilibrium. To illustrate these results, I will draw on examples from biophysics, where the effectiveness of numerous biochemical systems depends on being exquisitely sensitive to changes in chemical inputs. We will see how these predictions rationalize known energetic requirements of some common biochemical motifs and provide new limits to others.

Complex systems theory has taught us that simple, higher-level laws with few effective parameters can emerge from the interaction of small-scale components. As biology is becoming more and more quantitative, one can use a combination of first-principle theoretical modeling with machine learning techniques to build accurate and tractable theories of biological dynamics. Those dynamics can often be best understood in (abstract) latent spaces, giving « physics-like » intuition, interpretability and eventually allowing for new predictions and applications. I will illustrate the power of such approaches on the dynamics of the adaptive immune system, in particular T cell response. We used a robotic platform combined with machine learning to uncover a 'Universal encoding' from cytokine dynamics, and I will show how this response structure fits parsimonious models of immune recognition. Our approach suggests a new 'antagonistic' strategy for cancer immunotherapy that we validated both in vitro and in vivo.

Immune repertoires provide a unique fingerprint reflecting the immune history of individuals, with potential applications in precision medicine. Can this information be used to identify a person uniquely? The immune system is also dynamic, constantly changing in response to pathogen stimulation. What is the optimal update to best anticipate the next viral strain? If they are the result of coevolution with viral environments, can we identify signatures of these interactions? Lastly, I will show how we can map out likely evolutionary paths immune systems take in response to pathogens.

Living cells need to organize chemical reactions. In school, we learned that cells compartmentalize reactions using lipid bilayers. However, many functional domains don’t have a membrane and appear to be held together by liquid-like cohesion — i.e. capillarity. While we know a lot about the biochemistry of these condensates, we know relatively little about how their emergent properties affect cellular physiology. In this talk, I will briefly highlight how living cells can exploit condensate capillarity to sort molecules and do mechanical work. Then, we’ll take a closer look at how the coupling of chemical reactions and phase behavior. I’ll show that condensates can quite effectively compartmentalize chemical reactions. This localized activity can lead to some stunning ‘behavior’ which can be rationalized with some simple physical principles.

Our ability to collect data is rapidly increasing thanks to computational power and the unprecedented diversity of sensors. But how good are we at extracting, reconstructing, and understanding information from them? We present a short overview of some recent advancements for data-assimilation and modelling of turbulent multi-scale flows using both data-driven and equations-informed tools, starting from sparse and heterogeneous observations of complex fluid systems. Issues connected to validations and benchmarks in the presence of full or partial observability will be discussed. A few examples of data-generation and data- augmentation based on Generative Adversarial Learning, Diffusion Models and Nudging, for Eulerian and Lagrangian turbulence will be quantitatively discussed.

Related papers:

arXiv:2307.08529 (2023)

Journal of Fluid Mechanics 971, A3 (2023)

Physical Review X 10 (1), 011023 (2020)

Living matter, such as biological tissue, can be seen as a nonequilibrium hierarchical assembly of assemblies of smaller and smaller active components, where energy is consumed at many scales. The remarkable properties of such living or “active-matter” systems make them promising candidates to study and synthetically design. While many active-matter systems reside in fluids (solution, blood, ocean, air), so far, studies that include hydrodynamic interactions have focused on microscopic scales in Stokes flows. At those microscopic scales viscosity dominates, and inertia can be neglected. What happens as swimmers slightly increase in size (say ~0.1mm-100cm) or as they form larger aggregates and swarms? The system then enters the intermediate Reynolds regime where both inertia and viscosity play a role, and where nonlinearities are introduced in the fluid. In this talk, I will present a simple model swimmer used to understand the transition from Stokes to intermediate Reynolds numbers, first for a single swimmer, then for pairwise interactions and finally for collective behavior. We show that, even for a simple model, inertia can induce hydrodynamic interactions that generate novel phase behavior, steady states and transitions.

We'll provide an overview of the Bulletin of the Atomic Scientists, and the organization's connection to the University of Chicago. We'll discuss some of the issues behind the setting of the Doomsday Clock, as well as discuss some of the activities of the UChicago Existential Risk Laboratory (XLab).