During the Fall 2025 semester, seminars will take place either primarily in-person or virtually on Tuesdays at 1:15 p.m. The list below will detail which seminars are virtual and which ones are in-person. An informal pre-seminar chat with the speaker, to which all are invited, will be held at at 1:00 p.m. When in-person, we also have bring-your-own-lunch with the speaker at 12:30 p.m. in the seminar room.
Location for in-person seminars (unless exception noted): IPST Building (8108 Regents Drive, College Park), Room 1116. Please note that these will not be broadcast through zoom.
Location for virtual seminars: The zoom link for the chat and seminar is https://go.umd.edu/statphys_zoom
This seminar series is organized by Chris Jarzynski and Pratyush Tiwary. Please send suggestions for speakers to Pratyush Tiwary, ptiwary at umd.edu
Fall 2025 seminars
9/16/25 (in-person) Prof. Dio Margetis, University of Maryland
Title: Optical conductivity and edge plasmons in the twisted bilayer graphene
Abstract: This talk focuses on recent progress in understanding implications of the optical conductivity tensor of the twisted bilayer graphene (TBG). This material system has attracted much attention, mainly because of its novel electronic phases. First, I will discuss features of the optical conductivity via the Kubo formulation of linear response theory. This theory, together with symmetry considerations, yields a spatially homogeneous conductivity tensor of a certain form. Second, I will use this tensor form in order to analytically derive the dispersion relation of non-retarded edge plasmons, which are electromagnetic modes confined near edges, for a TBG consisting of two semi-infinite flat conducting sheets. This relation explicitly depends on the chirality of the system. Third, by invoking this result, I will describe a correspondence of the chiral optical plasmon in the TBG to the magnetoplasmon in the single-layer graphene, by introducing an effective magnetic field.
9/30/25 (in-person) Twesh Upadhyaya, University of Maryland
Title: Majorization theory for quasiprobabilities
Abstract: Majorization theory is a powerful mathematical tool to compare the disorder in distributions, with wide-ranging applications in many fields including mathematics, physics, information theory, and economics. While majorization theory typically focuses on probability distributions, quasiprobability distributions provide a pivotal framework for advancing our understanding of quantum mechanics, quantum information, and signal processing. Here, we introduce a notion of majorization for continuous quasiprobability distributions over infinite measure spaces. Generalizing a seminal theorem by Hardy, Littlewood, and Pólya, we prove the equivalence of four definitions for both majorization and relative majorization in this setting. We give several applications of our results in the context of quantum resource theories, obtaining new families of resource monotones and no-goes for quantum state conversions. A prominent example we explore is the Wigner function in quantum optics. More generally, our results provide an extensive majorization framework for assessing the disorder of integrable functions over infinite measure spaces.
10/7/25 (in-person) Prof. Mihail Anisimov, University of Maryland
Title: DEGENERATE FLUID POLYAMORPHISM: IS THERE ANYTHING LEFT TO DISCOVER IN CLASSICAL THERMODYNAMICS?
Abstract: TBA
10/28/25 (virtual) Dr. Yuanqi Du, Cornell University
Title: TBA
Abstract: TBA
11/4/25 (in-person) Dr. Weishun Zhong, Institute for Advanced Study at Princeton
Title: Statistical mechanics of real memories in natural language
Abstract: There has long been a divide in the study of memory: on one side, memories are complex and subjective, shaped by our individual experiences; on the other, traditional theories rely on oversimplified assumptions in artificial settings, rendering them largely irrelevant to real memories. We attempt to bridge this gap by introducing a statistical mechanical model that captures real memories acquired from and recalled in natural language. Analytical solutions of the model align with observations from large-scale narrative recall experiments, accounting for key features of memory such as summarization and abstraction. Grounded in the hierarchical organization of language itself, our model enables a unified framework for memory and language. Surprisingly, when applied directly to language, the model yields a prediction for the entropy of natural language, which we confirm through experiments with modern large language models.
11/11/25
11/18/25 (in-person) Prof. Will Jacobs, Princeton University.
Title: TBA
Abstract: TBA
12/2/25
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Previous Seminars
Seminars for Fall 2025
9/2/25 (in-person) Prof. Emanuela Del Gado, Georgetown University
Title: Fluctuations, rheology, and memory in self-assembled gel networks
Abstract: Many materials we eat, spread, squeeze, or 3D print are gels, soft amorphous solids whose solid component comprises self-assembled networks of particles, fibers, or agglomerates of proteins, polymers, and colloids. The space between and within human cells is permeated by self-assembled gel networks, the extra-cellular matrix and the cytoskeleton, whose self- organization and heterogeneity is central to biological functions. Self-assembled gels have adaptive, tunable, and nonlinear rheology determined by a complex interplay between the molecular cohesion and surface interactions, the aggregation kinetics that drive for- mation of various types of structures, and the effect of external forces that can promote breaking or reforming of the load-bearing backbone. Solidification processes are typically sources of frozen-in stresses and help build a memory of the processing history in these amorphous solids. Disorder and self-organization determine stress localization under load and the feedback between stress heterogeneities, structural disorder, and nonequilibrium conditions is therefore key to the mechanical response of these fascinating and ubiquitous materials.
Seminars for Spring 2025
2/11/25 (in-person) Prof. Haitao Quan, Peking University
Title: Hierarchical structure of fluctuation theorems for a driven system in contact with multiple heat reservoirs
Abstract: For driven open systems in contact with multiple heat reservoirs, we find the marginal distributions of work or heat do not satisfy any fluctuation theorem, but only the joint distribution of work and heat satisfies a family of fluctuation theorems. A hierarchical structure of these fluctuation theorems is discovered from microreversibility of the dynamics by adopting a step-by-step coarse-graining procedure in both classical and quantum regimes. Thus, we put all fluctuation theorems concerning work and heat into a unified framework. We also propose a general method to calculate the joint statistics of work and heat in the situation of multiple heat reservoirs via the Feynman-Kac equation. For a classical Brownian particle in contact with multiple heat reservoirs, we verify the validity of the fluctuation theorems for the joint distribution of work and heat. If time allows, I will also discuss the extension of these fluctuation theorems into relativistic covariant forms.
2/18/25 (virtual) Prof. Kranthi Mandadapu, University of California Berkeley
Title: Odd transport phenomena in classical fluids: Breaking parity and time reversal symmetry
2/25/25 (virtual) Prof. Eric Vanden-Eijnden, New York University
Title: Enhancing Monte Carlo Sampling Methods with Non-Equilibrium Transport
Abstract: Sampling high-dimensional probability distributions is a common task in Science, Engineering, and Statistics. Monte Carlo (MC) samplers are the methods of choice to perform these calculations, but they are often plagued by slow convergence properties. I will discuss recent advances in generative modeling that can be used to assist nonequilibrium MC methods and improve their performance. These approaches are based on using MC data to learn how to dynamically transport a measure towards the target of interest, then use use this transport to generate more data, in a positive feedback loop. As a specific illustration, I will present a variant of Neals’ annealed importance sampling (AIS) based on Jarzinsky equality, in which the stochastic differential equation used to generate the data in AIS is augmented with an additional drift term that enhance the method’s capabilities. This drift can be learned variational via minimization of a tractable objective function that can be shown to control the Kullback-Leibler divergence of the estimated distribution from its target. I will also illustrate the method on standard benchmarks, such as high-dimensional Gaussian mixture distributions, and a model from statistical lattice field theory.
This is joint work with Michael Albergo.
3/4/25 (virtual) Dr. Britta Johnson, Pacific Northwest National Laboratory (PNNL)
Title: Including Nuclear Quantum Effects in Condensed Phase Systems through Path-Integral Methods
Abstract: Nuclear quantum effects such as tunneling and nuclear zero-point energy are important components in obtaining accurate structural and dynamic properties for a host of chemical processes including charge and energy transfer reactions, the behavior of molecules under confinement, and systems with hydrogen bonding networks. However, most traditional computational methods, like ab-initio molecular dynamics, do not include quantum effects in the nuclear degrees of freedom. Path-integral based methods, like path-integral molecular dynamics, are a class of methods that capture nuclear quantum effects through replacing a quantum particle with an n-bead classical ring polymer. Here, I present our work on expanding
path-integral methods into a new atomistic path integral molecular dynamics software program where each atom in a system can be treated with a different quantization (n-)level. This method can drastically improve the computational scalability of path integral methods and allows for the simulation of previously inaccessible condensed phase systems. I will explore the success of this method using a series of model systems. This work was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Chemical Sciences, Geosciences, and Biosciences Division, Condensed Phase and Interfacial Molecular Science program, FWP 16249.
4/1/25 (in-person) Prof. Venkat Kapil, University College London
Title: Machine Learning for Full-Quantum Simulations of Condensed Phases and Interfaces: application to the first-principles phase diagram of nanoconfined water
Abstract: Understanding the microscopic structure, dynamics, and reactivity of condensed-phase and interfacial systems is essential for fields ranging from biochemistry to battery science and catalysis. While quantum mechanical simulations, in theory, provide predictive accuracy, incorporating all relevant quantum effects in condensed-phase systems at finite temperatures has remained computationally prohibitive. In this talk, I will present our recent progress in developing machine-learning (ML) approaches that efficiently and accurately address these challenges. These include ML interatomic potentials approaching the accuracy of correlated electronic structure theory at a fraction of their cost [1], physics-based ML models of electronic properties such as dielectric response tensors for vibrational spectroscopy [2], and ML quantum effective corrections to Born-Oppenheimer potentials for describing quantum statistics and approximate dynamics at a classical cost [3]. As an application, I will demonstrate a fully quantum description of the phase behaviour of nanoconfined water [4]. This system, relevant to water treatment, catalysis, and battery sciences, exhibits se veral anomalous properties and hitherto has been poorly understood due to inaccuracies in simulations. Our findings reveal that monolayer water displays rich phase behaviour and is highly sensitive to the van der Waals pressure exerted by the confining material. We observe ice phases that break conventional ice rules [5] and predict the existence of a superionic phase under significantly milder conditions than in bulk water. This phase exhibits ionic conductivity comparable to or exceeding that of many battery materials, with quantum nuclear motion playing a crucial role [6]. Our work shows that large-scale, fully quantum modelling of condensed-phase and interfacial systems is now feasible, paving the way for quantitative operando studies of processes relevant to clean energy and catalysis that are key to sustainability.
References
[1] Kaur, Della Pia, Batatia, Advincula, Shi, Lan, Csányi, Michaelides, & Kapil (2024). https://doi.org/10.48550/ARXIV.2405.20217
[2] Kapil, Kovács, Csányi, & Michaelides (2023). Faraday Discussions. https://doi.org/10.1039/D3FD00113J
[3] Musil, Zaporozhets, Noé, Clementi, & Kapil (2022). The Journal of Chemical Physics. https://doi.org/10.1063/5.0120386
[4] Kapil, Schran, Zen, Chen, Pickard, & Michaelides (2022). Nature. https://doi.org/10.1038/s41586-022- 05036-x
[5] Ravindra, Advincula, Schran, Michaelides, & Kapil (2024). Nature Communications. https://doi.org/10.1038/s41467-024-51124-z
[6] Ravindra, Advincula, Shi, Coles, Michaelides, & Kapil (2024). arXiv.
4/8/25 (in-person) Prof. Zhiyue Lu, University of North Carolina, Chapel Hill
Title: Far-from-Equilibrium Statistical Physics: Information and Response in Living Matter
Abstract: Statistical mechanics, first developed for equilibrium systems, stands as one of physics' greatest triumphs. Living systems, however, operate far from equilibrium, processing information and responding to signals in ways that challenge traditional equilibrium frameworks. We demonstrate that by shifting focus from state distributions to complete stochastic trajectories, we can extend the success of statistical physics to non-equilibrium biological systems. Our trajectory-based approach reveals that these systems encode more information than previously recognized, enabling phenomena like multiplexed sensing through single molecular pathways. We develop universal theoretical tools - integrating information theory and response theory - that hold arbitrarily far from equilibrium, uncovering fundamental principles governing how biological systems sense and respond to their environment. These insights not only explain existing biological designs but also provide quantitative guidelines for engineering synthetic molecular systems. Our work bridges the gap between equilibrium statistical physics and the dynamic complexity of living matter, establishing design principles for biological information processing.
4/15/25 (in-person) Prof. Jan Sengers, University of Maryland, College Park
Title: Nonequilibrium Fluctuating Hydrodynamics
Abstract: Fluctuating hydrodynamics is based on Onsager’s hypothesis that the regression of thermal fluctuations in fluids is governed by the hydrodynamic equations. Fluctuating hydrodynamics has been primarily used to describe thermal fluctuations in fluids in thermodynamic equilibrium states. Extension of fluctuating hydrodynamics to fluids in nonequilibrium states has revealed that, in contrast to equilibrium fluctuations,
nonequilibrium fluctuations become gigantic and extend over the entire system, even far away from any hydrodynamic instability. This discovery has led to a variety of studies of nonequilibrium fluctuations that will be reviewed. Most recently, it has been predicted that these fluctuations will cause giant fluctuation-induced Casimir forces in confined liquid layers, possibly opening a new field of nonequilibrium micro-mechanics.
5/6/25 (in-person) Prof. Fabio Anza, University of Maryland, Baltimore County
Title: Kinetic theory of quantum information transport: a geometric approach to the statistical physics of open quantum systems
Abstract: Geometric quantum mechanics (GQM) is a differential-geometry-based approach to quantum mechanics that brings it closer to classical mechanics. In this talk, I will discuss an ongoing research effort to exploit the tools of GQM to advance our understanding of open quantum systems. In particular, I will show how GQM provides a powerful approach to studying the physics of quantum ensembles underlying density matrices. After a brief introduction to GQM and a walkthrough of recent studies that motivate the larger theoretical approach I am pursuing, I will present the Kinetic theory of quantum information transport. This novel theoretical framework can be used to understand quantum ensembles' equilibrium and out-of-equilibrium physics and, in particular, improve our understanding of the interplay between energetics and information-theoretic features of open quantum systems. This is an ongoing research effort, so I will also mention several directions of interest, which I hope could serve as a fruitful ground for future collaborations.
5/12/25 Postponed to fall (in-person) Prof. Emanuela Del Gado, Georgetown University.
Title: TBA
Abstract: TBA
Seminars for Fall 2024
12/10/24 (in-person - 12PM-1PM. Note special time) Hoony Kang, University of Maryland
Title: Rhythmic sharing, a bio-inspired algorithm for zero-shot adaptation and learning in neural networks
Abstract: Artificial neural networks learn by finding fixed optimal weights, which are the strengths of connections between neuronal nodes. These optimal values are found by minimizing the error between the neural network’s output and what we know should be the true output for given inputs used during training. However, much of the real world is governed by non-stationary dynamics, wherein the underlying statistics of the dynamics drift in time. When an artificial neural network (ANN) is trained with a loss function that is uninformed of the different probability spaces present in the training data (e.g., through contextual tokens), the ANN will treat the entire data as if it had come from one stationary probability distribution—in effect, as one ‘event’—and cannot differentiate between the different dynamics without instruction.
Yet the living neural networks of brains can rapidly infer contextual changes in real-time, adapt their behavior under this new environment, and even induce how the organism should act in an unencountered environment. In this talk, I introduce a paradigm that associates learning with the coordination of link strength rhythms, inspired by mutimodal interactions in the brain, and implement it into an ANN. We term this new paradigm "rhythmic sharing". When nonstationary time series that consist of mixed events are sent into our modified ANN, the links coordinate their rhythms in unique patterns for different dynamics. In other words, the link dynamics can be interpreted as classification tokens that the network generates to separate the different events in the data without prior training. Furthermore, the oscillations themselves allow the network to extrapolate dynamics to previously unseen contexts. Because it is agnostic to the specific details of the neural network architecture, our study also opens the door for introducing rapid adaptation and learning capabilities into leading AI models.
10/1/24 (virtual) Dr. Cal Floyd, University of Chicago
Title: Limits on the computational expressivity of non-equilibrium biophysical processes
Abstract: Many biological decision-making processes can be viewed as performing a classification task over a set of inputs, using various chemical and physical processes as "biological hardware." In this context, it is important to understand the inherent limitations on the computational expressivity of classification functions instantiated in biophysical media. Here, we model biochemical networks as Markov jump processes and train them to perform classification tasks, allowing us to investigate their computational expressivity. We reveal several unanticipated limitations on the input-output functions of these systems, which we further show can be lifted using biochemical mechanisms like promiscuous binding. We analyze the flexibility and sharpness of decision boundaries as well as the classification capacity of these networks. Additionally, we identify distinctive signatures of networks trained for classification, including the emergence of correlated subsets of spanning trees and a creased "energy landscape" with multiple basins. Our findings have implications for understanding and designing physical computing systems in both biological and synthetic chemical settings.
This is joint work with Suri Vaikuntanathan, Arvind Murugan, and Aaron Dinner (https://arxiv.org/abs/2409.05827).
10/15/24 (virtual) Dr. Emmanuel Bengio, Recursion Pharmaceuticals
Title: GFlowNets for scientific discovery and beyond
Abstract: In this talk I will share some recent progress on training GFlowNets and applying them to scientific problems, as well as speculate on the bigger picture; using GFNs to be good Bayesians, and create models that reason about their environment. I will start the talk with a quick tutorial/refresher on GFlowNets to set the stage, and hope to finish with an open discussion of next steps and open problems.
10/22/24 (in-person) Prof. Purushottam Dixit, Yale University
Title: Emergent low dimensionality of microbiomes: causes and application
Abstract: Microbiomes are complex ecosystems wherein thousands of microorganisms spanning several kingdoms of life can coexist. Observed species covariation in microbiomes suggests that the number of degrees of freedom that govern these ecosystems (a.k.a. their dimensionality) may be substantially lower than the number of organisms. To identify the dimensionality of microbiomes, we developed a latent variable model derived from the consumer/resource (C/R) framework. Fitting this model to publicly available microbiomes showed that their dimensionality varied substantially across communities. We used the C/R framework to investigate the observed heterogeneity. Low dimensionality of microbiomes can arise either from correlations in their nutrient preferences or correlations in inflows of nutrients. Our simulations showed that dimensionality could be modulated by low dimensional structures in resource inflows but not by low dimensional structures in resource preferences. Notably, the model predicted that low dimensionality of microbiomes correlated with increased competition amongst species. Analysis of multiple microbiomes and metabolic modeling confirmed these predictions in real microbiomes. Additionally, low dimensional microbiomes accurately predicted the state of their environment. We used this to develop a generative model that allowed rational microbiome engineering. In the model, variation in ecosystem composition arose due to differences in the availability of effective resources (latent variables) while species’ resource preferences remained conserved. The same latent variables were used to model phenotypic states of hosts. In silico microbiomes generated by our model accurately reproduced universal and dataset-specific statistics of bacterial communities. The model allowed us to address three salient questions in host-associated microbial ecologies: (1) which host phenotypes maximally constrain the composition of the host-associated microbiomes? (2) how context-specific are phenotype/microbiome associations, and (3) what are plausible microbiome compositions that correspond to desired host phenotypes? Therefore, mechanistic models are not only able to shed qualitative light on organization of complex microbial ecosystems but when appropriately modified can also allow us to design these ecosystems.
10/29/24 (virtual) Prof. Matteo Salvalaglio, University College London
Title: Avoiding rare events: Quantitative insight into the thermodynamics of polymorphism without sampling phase transitions
Abstract: Free energy calculations enable the quantitative understanding of physicochemical phenomena in material science, chemistry, and physics. Nevertheless, free energy methods are typically faced with computational efficiency issues, which limit their applicability in large-scale, high-throughput applications. One such application is the computational prediction of polymorphism, where the relative stabilities of tens to hundreds of putative polymorphs need to be evaluated to provide rational, physics-based prediction. A source of such limitations is that interesting metastable states, representing i.e. putative polymorphs, are usually characterized by nonoverlapping configurational Boltzmann distributions, and thus, computing free energy differences between them requires sampling intermediate states characterized by high free energies and low probabilities. In this seminar, I will discuss how theoretical, or machine learning (ML) models of the free energy landscape informed by locally ergodic molecular dynamics simulations can provide a blueprint to boost large-scale studies of the relative thermodynamic stability of polymorphs both in the crystalline bulk and in multicomponent solutions. On the one end of the model complexity spectrum, physically interpretable theoretical models built on classical thermodynamics can be used to interpret the equilibrium configuration of canonical simulations to extract physically relevant parameters such as polymorph specific solubility and surface tension. [1,2] Within the confines of the theoretical assumptions embedded in the models used to interpret simulations and of the empirical molecular models necessary to sample explicitly solvated nuclei, these approaches can be used provide estimates of the relative polymorphic stability and of relative nucleation rates from equilibrium MD trajectories. On the opposite end of the model complexity spectrum, ML models enable an efficient and accurate estimate of relative free energy differences between polymorphs without sampling any physically relevant intermediate state. In this seminar, I will discuss the combination of targeted free energy perturbation, flow-based ML models and low variance free energy estimators to obtain free energy estimates without sampling intermediate states.[3,4] In this context, ML techniques are essential to provide trainable model architectures to map between configurations belonging to different macrostates, i.e. crystal polymorphs.[5] ML-based approaches at the high-end of the model complexity spectrum are agnostic with respect to theoretical interpretations and promise to enable significant gains in sampling efficiency, thus providing an actionable strategy to implement large-scale thermodynamic rankings of putative polymorphs at finite temperatures.
References
Finney, A. R., & Salvalaglio, M. (2024). Wiley Interdisciplinary Reviews: Computational Molecular Science, 14(1), e1697.
Bachtiger, F., Rahimee, A., Li, L., & Salvalaglio, M., chemRxiv (2024).
Jarzynski, C., (2002). Physical Review E, 65(4), p.046122.
Olehnovics, E., Liu, Y.M., Mehio, N., Sheikh, A.Y., Shirts, M.R. and Salvalaglio, M., (2024). Journal of Chemical Theory and Computation, 20(14), pp. 5913-5922.
Olehnovics, E., Liu, Y.M., Mehio, N., Sheikh, A.Y., Shirts, M.R. and Salvalaglio, M., (2024). in preparation.
11/5/24 (virtual) Prof. Michael Chertkov, University of Arizona
Title: Harmonic Path Integral Diffusion
Abstract: In this talk, I present a novel approach for sampling from a continuous multivariate probability distribution, which may either be explicitly known (up to a normalization factor) or represented via empirical samples. The approach consists in constructing a time-dependent bridge from a delta function centered at the origin of the state space at t=0, optimally transforming it into the target distribution at t=1. We formulate this as a Stochastic Optimal Control problem of the Path Integral Control type, with a cost function comprising (in its basic form) a quadratic control term, a quadratic state term, and a terminal constraint. This framework, which we refer to as Harmonic Path Integral Diffusion (H-PID), leverages an analytical solution through a mapping to an auxiliary quantum harmonic oscillator in imaginary time. The H-PID framework results in a set of efficient sampling algorithms, without the incorporation of Neural Networks. The algorithms are validated on two standard use cases: a mixture of Gaussians over a grid and images from CIFAR-10. We contrast these algorithms with other sampling methods, particularly simulated annealing and path integral sampling, highlighting their advantages in terms of analytical control, accuracy, and computational efficiency on benchmark problems. Additionally, we extend the methodology to more general cases where the underlying stochastic differential equation includes an external deterministic, possibly non-conservative force, and where the cost function incorporates a gauge potential term. These extensions open up new possibilities for applying our framework to a broader range of statistics specific to applications. The talk is based on collaboration with Hamidreza Behjoo reported in https://arxiv.org/abs/2409.15166 .
11/12/24 (virtual) Dr. Lingle Wang, Schrodinger
Title: Advancing Drug Discovery via Physics Based Free Energy Calculations
Abstract: Protein-ligand binding free energy calculation has become increasingly important in modern computationally driven drug discovery. This presentation will provide a concise overview of the history of free energy method development, highlight key advances in modern implementations that enhance accuracy and broaden application domains, and showcase example applications of how these methods are driving modern drug discovery projects. Additionally, integration of physics based modeling with machine learning to efficiently explore diverse chemical spaces will also be discussed.
11/19/24 (virtual) Prof. Christoph Dellago, University of Vienna
Title: Reducing statistical correlations in rare event simulations, with and without machine learning
Abstract: The microscopic dynamics of many condensed matter systems occurring in nature and technology is dominated by rare yet critical barrier crossing events. Examples of such processes include nucleation at first order phase transitions, chemical reactions and the folding of biopolymers. Computational methods developed to address this problem typically suffer from statistical correlations, particularly for systems with multiple reaction channels. In my talk, I will discuss various approaches to reduce such correlations and enhance the efficiency of rare event simulations.
Seminars for Spring 2024
2/9/24 (Special Friday 3PM seminar, in-person) Matthew Gerry, University of Toronto
Title: Two stories of nonequilibrium transport: dynamics of modular chains and specificity of efflux pumps
Abstract: Systems away from equilibrium at the micro- and nano-scales are often modelled as stochastic processes, wherein a master equation governs the evolution of a probability distribution over a network of states. I will discuss two projects which make use of this general approach, but with different objectives. The first considers continuous-time random walks on modular chains consisting of alternating domains. We carry out full counting statistics for a particular example of such a chain, along with numerical simulations for a wider variety of systems, and show that underlying modularity is not reflected in the mean veclocity in any way. Higher order cumulants can, however, show signatures of the local structure, with the kurtosis being the lowest-order cumulant to do so when the transport is unbiased, while the diffusion coefficient is sufficient in the presence of bias. These results may help inform strategies for determining architectures underlying various stochastic processes. The second project focuses on a specific class of systems, namely, ion-powered efflux pumps on bacterical cell membranes. We find that far from equilibrium, these pumps' throughput exhibits complex and nontrivial dependence on various environmental parameters, including the concentration gradient of ions across the membrane. In particular, in addition to powering the pump, this ion gradient seems to tune the range of drug binding affinities at which the pump is most effective, hinting at a strategy by which bacterial populations may be able to respond to drugs used in treatment and providing insight into the role these pumps play in broad antibiotic resistance.
2/13/24 (Physics Colloquium special time and location, in-person) Gavin Crooks, Normal Computing
Special time and location: https://umdphysics.umd.edu/events/physicscolloquia.html#thermodynamic-linear-algebra
Title: Thermodynamic Linear Algebra
Abstract: Linear algebraic primitives are at the core of many modern algorithms in engineering, science, and machine learning. Hence, accelerating these primitives with novel computing hardware would have tremendous economic impact. I'll discuss how a variety of linear algebra problems can be solved by sampling from the thermodynamic equilibrium distribution of a collection of coupled harmonic oscillators.
3/5/24 (virtual) Dr. Jascha Sohl-Dickstein, Google DeepMind --> Anthropic AI
Title: Understanding infinite width neural networks from the perspective of statistical mechanics
Abstract: As neural networks become wider their accuracy improves, and their behavior becomes easier to analyze theoretically. I will give an introduction to a growing body of work which examines the learning dynamics and distribution over functions induced by infinitely wide, randomly initialized, neural networks. Core results that I will discuss include: that the distribution over functions computed by a wide neural network often corresponds to a Gaussian process with a particular compositional kernel, both before and after training; that the predictions of a class of wide neural networks are linear in their parameters throughout training; that the posterior distribution over parameters also takes on a simple form in wide Bayesian networks. These results provide for surprising capabilities -- for instance, the evaluation of test set predictions which would come from an infinitely wide trained neural network without ever instantiating a neural network, or the rapid training of 10,000+ layer convolutional networks. I will argue that this growing understanding of neural networks in the limit of infinite width is foundational for theoretical and practical understanding of deep learning.
Neural Tangents: https://github.com/google/neural-tangents
4/9/24 (virtual) Prof. Alessandro Laio, SISSA, Trieste, Italy
Title: Identifying informative distance measures in high-dimensional feature spaces
Abstract: Real-world data typically contain a large number of features that are often heterogeneous in nature, relevance, and also units of measure. When assessing the similarity between data points, one can build various distance measures using subsets of these features. Finding a small set of features that still retains sufficient information about the dataset is important for the successful application of many statistical learning approaches. We introduce an approach that can assess the relative information retained when using two different distance measures, and determine if they are equivalent, independent, or if one is more informative than the other. This test can be used to identify the most informative distance measure out of a pool of candidates (https://academic.oup.com/pnasnexus/article/1/2/pgac039/6568571). The approach can be used to identify the most appropriate set of collective variables in molecular systems and to infer causality in high-dimensional dynamic processes and time series (https://arxiv.org/abs/2305.10817).
4/23/24 (virtual) Prof. Arnaud Doucet, Oxford and Google DeepMind, UK
Title: Diffusion Schrödinger Bridge Matching
Abstract: Solving mass transport problems, i.e. finding a map transporting one given probability distribution to another, has numerous applications in machine learning and related areas. Novel mass transport methods motivated by generative modeling have recently been proposed, e.g. Denoising Diffusion Models (DDMs) and Flow Matching Models (FMMs) implement such a transport through a Stochastic Differential Equation (SDE) or an Ordinary Differential Equation (ODE). However, while it is desirable in many applications to approximate the deterministic dynamic Optimal Transport (OT) map which admits attractive properties, DDMs and FMMs are not guaranteed to provide transports close to the OT map. In contrast, Schrödinger bridges (SBs) compute stochastic dynamic mappings which recover entropy-regularized versions of OT. Unfortunately, existing numerical methods approximating SBs either scale poorly with dimension or accumulate errors across iterations. In this work, we introduce Iterative Markovian Fitting (IMF), a new methodology for solving SB problems, and Diffusion Schrödinger Bridge Matching (DSBM), a novel numerical algorithm for computing IMF iterates. DSBM significantly improves over previous SB numerics and recovers as special/limiting cases various recent transport methods. We demonstrate the performance of DSBM on a variety of problems.
This is joint work with Yuyang Shi, Valentin De Bortoli and Andrew Campbell (https://arxiv.org/abs/2303.16852).
Seminars for Fall 2023
9/26/23 (in-person) Dr. Eric Beyerle, University of Maryland
Title: Modeling the slow dynamics of biological systems over a range of time- and lengthscales
Abstract: Essential biological processes occur over a range of timescales, lengthscales, and complexity. To explicate such heterogeneous dynamics, models that effectively describe the relevant dynamics at each time-, length-, and complexity scale must be developed. In this talk I will discuss two such models, one for describing near-equilibrium protein dynamics and a more generic, machine-learning based model for describing rare events, including those in biological systems. When combined with sampling from either equilibrium or biased molecular dynamics simulations, these models give fine spatial and temporal detail regarding biological processes. The first model is an analytical, linear model that elucidates the conformational dynamics of the small protein ubiquitin from the picosecond to microsecond timescale and gives information regarding how free ubiquitin selects its binding partners to perform its biological activity. The second model is a generally nonlinear neural network that is parameterized separately to describe the mechanisms of small molecule permeation through a lipid bilayer, hydrophobic ligand unbinding, and the formation of glycine crystals from aqueous solution. The results of these models offer insights regarding human health and disease, drug efficacy, and fundamental physics regarding the specificity of biological processes.
10/10/23 (in-person) Annual Shih-I Pai Lecture. See lecture website for details.
10/17/23 (in-person) Darya Kisuryna (Dodson group, University of Maryland. Chemical Physics Candidacy exam)
10/24/23 (virtual) Rianne van den Berg, Microsoft Research
Title: AI4Science at Microsoft Research: Diffusion Models and Force Fields for Coarse-Grained Molecular Dynamics
Abstract: In July 2022 Microsoft announced a new global team in Microsoft Research, spanning the UK, China and the Netherlands, to focus on AI for science. In September 2022 we announced that we have also opened a new lab in Berlin, Germany, and recently another team in Redmond (USA) joined our initiative. In this talk I will first briefly discuss some of the research areas that we are currently exploring in AI4Science at Microsoft Research, covering topics such as drug discovery, material generation and neural PDE solvers. Then I will dive a little deeper into recent work that was done at AI4Science. I will cover work on the use of score-based generative modeling for coarse-graining (CG) molecular dynamics simulations. By training a diffusion model on protein structures from molecular dynamics simulations we show that its score function approximates a force field that can directly be used to simulate CG molecular dynamics. While having a vastly simplified training setup compared to previous work, we demonstrate that our approach leads to improved performance across several small- to medium-sized protein simulations, reproducing the CG equilibrium distribution, and preserving dynamics of all-atom simulations such as protein folding events.
10/31/23 (virtual) Hannes Stärk, Gabriele Corso and Bowen Jing, MIT
Title: Diffusion on Manifolds for Generative Modeling of Molecules
Abstract: We will talk about two of our works where we employ diffusion generative models for the task of protein-ligand docking and molecular conformer generation. We will highlight the suitability and flexibility of diffusion for generative modeling on manifolds.
1) Torsional Diffusion defines a generative process over the torsion angles of a molecule to sample its 3D conformations.
2) DiffDock builds on this by including translations and rotations in the diffusion process to generate the 3D structures in which a small molecule binds to a protein.
11/7/23 (virtual) Andrei Klishin, University of Washington
Title: Beyond Optimization: Statistical Mechanics of Design and Data
Abstract: Many challenges in engineering and natural science, from industrial design to machine learning, are commonly formalized as optimization problems. Many numerical methods can find the exact or approximate point in the solution space and thus answer what solution should be chosen. However, such quantitative approaches are limited in addressing the problems that are ill-defined or have combinatorially large solution spaces. In this talk, I argue that statistical mechanics can provide a more nuanced, probabilistic, uncertain perspective than optimization alone and thus answer a wider range of scientific questions. Why do functional units tend to cluster in naval ship design? How much does an optimal configuration of sparse sensors degrade if perturbed? What if there is too much noise in the trajectory used for system identification? Combined with efficient numerical algorithms, addressing these questions enables a more principled selection of both design solutions and data-driven models.
11/14/23 (in-person) Jeremy Owen, Princeton University
Title: "How biology breaks the rules: the nonequilibrium laws of sensitivity"
Abstract: Living things enjoy exquisite molecular sensitivity in their key processes, including DNA replication, transcription and translation, sensing, and morphogenesis. Understanding the diverse chemical mechanism responsible for this sensitivity poses a basic challenge for nonequilibrium physics. At equilibrium, strict constraints relate sensitivity to system structure and spontaneous fluctuations. But away from equilibrium, we lack knowledge of even the most basic limits to sensitivity set by coarse properties like system size or dissipation. In this talk, I'll tell you about the simple laws that constrain sensitivity in small, nonequilibrium systems. These new results extend and unify a patchwork of prior biophysics results on the energetic and structural costs of sharp biochemical switches, accurate sensors, and molecular discrimination—revealing their common origin in the perturbation theory of Markov chains. In pursuit of mechanism that saturate these bounds, we discover a nonequilibrium binding mechanism with remarkable sensitivity, exponential in system size, with implications for our understanding of models of gene regulation.
11/28/23 (in-person) Richard Remsing, Rutgers University
Title: Modeling quantum effects with path integrals: From nanomaterials to planetary science
Abstract: The quantum mechanical behavior of particles can give rise to important processes in materials science and chemistry, including effects like quantum confinement and tunneling through reaction barriers. However, quantum mechanics-based computer simulations of molecular systems are often expensive and limited to small systems and short timescales. To overcome these limitations, path integral-based approaches can enable the simulation of quantum systems using classical simulation methods applied to an extended system. In this talk, I will discuss the development and application of path integral simulation techniques in two areas. I will focus mostly on an approach for modeling excess electrons and holes in materials that incorporates aspects of electronic structure theory into the path integral framework. I will demonstrate the ability of this approach to model complex systems, such as excess electrons in two-dimensional semiconductors at interfaces with liquids that are relevant to nanoelectronics, sensing, and catalysis. Such systems require a description of the quantum behavior of excess electrons or holes in the semiconductor, including quantum confinement, as well as adequate sampling of the configuration space of the liquid, both of which can be achieved using our approach. I will show that we can efficiently model the interactions of charge carriers with defects in two-dimensional semiconductors, screening of these interactions by interfacial liquids, effects of liquid screening on transport, and the effects of defect-induced trap states on interfacial liquid properties relevant to catalysis. I will also discuss the use path integral methods for modeling nuclear quantum effects in confined liquids and in chemical reactions at low temperatures relevant to chemistry on the surface of cold planetary bodies.
Seminars for Spring 2023
2/14/23 (in-person) Prof. John Biddle, Holy Cross College at Notre Dame
Title: Reversal symmetries for cyclic paths in Markovian systems far from thermodynamic equilibrium
Abstract: If a system is at thermodynamic equilibrium, an observer cannot tell whether a film of it is being played forward or in reverse: any transition between states, and any sequence of transitions between states, will occur with the same frequency in the forward as in the reverse direction. However, if expenditure of energy maintaining the system away from equilibrium changes the rate of even a single transition, such time-reversal symmetry undergoes a widespread breakdown reaching far beyond the point at which the energy is expended. System properties that are locally determined at equilibrium then come to depend in a complicated manner on the rate of every transition in the system. Cyclic paths, however, have reversibility properties that remain locally determined, and which can exhibit reversal symmetry, no matter how far the system is from thermodynamic equilibrium. Specifically, given any cycle of reversible transitions, the ratio of the frequencies with which the cycle occurs in one direction versus the other is determined, in the long-time limit, only by the thermodynamic force on the cycle itself. In particular, if there is no net energy expenditure on the cycle, then the cycle occurrence frequencies are the same in either direction in the long-time limit.
3/7/23 (in-person) Prof. Martin Gruebele, University of Illinois
Title: Quantum information scrambling in molecules
Abstract: Black holes may be the most efficient quantum information scramblers, but chemical reactions obey surprisingly similar upper limits on the information scrambling rate. I will discuss some simple models and computational results that show how the quantum analog of weak classical chaos can promote rapid steering of quantum systems to targets in state space, how out-of-time-order correlators (OTOCs) describe scrambling of vibrational information in molecules and isomerization reactions, with a bound that was originally discovered by Herzfeld using old quantum theory in 1919 similar to the one applied by Maldacena to quantum scrambling. These models may also shed light on how state collapse during measurement could be simply an approximation to quantum localization induced by statistical variations between many-body detectors.
3/13/23 (in-person) Prof. Steve Campbell, University College Dublin
Title: Characterising the quantum work distribution
Abstract: The steady interest in understanding the thermodynamics of quantum systems has led to several approaches to defining work in a quantum mechanically consistent way (at least almost consistent). The two-point measurement protocol is one such approach that, despite some limitations, has provided a wealth of insight. While many analyses have focussed on the first and/or second moment (relating to the average and fluctuations) of the distribution arising from this protocol, interesting physics can be captured by going beyond and examining properties of the full probability distribution. In this talk I will present a snapshot of this for several systems, in particular for many-body systems host to quantum and/or disorder-driven phase transitions for both sudden quenches and finite time ramps. Useful summary tools to characterise the distribution, chiefly the Shannon entropy and measures of Gaussianity, will be discussed.
This talk will touch-on/draw-from results in the following papers:
- Entropy of the quantum work distribution, Anthony Kiely, Eoin O'Connor, Thomás Fogarty, Gabriel T. Landi, and Steve Campbell, arXiv:2210.07896;
- Non-Gaussian work statistics at finite time driving, Krissia Zawadzki, Anthony Kiely, Gabriel T. Landi, and Steve Campbell, Phys. Rev. A 107, 012209 (2023);
- Work statistics and symmetry breaking in an excited state quantum phase transition, Zakaria Mzaouali, Ricardo Puebla, John Goold, Morad El Baz, and Steve Campbell, Phys. Rev. E 103, 032145 (2021);
- Criticality revealed through quench dynamics in the Lipkin-Meshkov-Glick model, Steve Campbell, Phys. Rev. B 94, 184403 (2016)
3/17/23 (in-person) Dr. Yasaman Bahri, Google Research
When: Friday, March 17 at 2pm
Where: 2400 ATL
Title: Understanding deep learning through the lens of solvable models, dynamical systems, and statistical mechanics
Abstract: Deep neural networks – the backbone of modern machine learning – are prevalent across and impacting many areas of science and engineering. However, the mechanisms behind how they learn from data are not fully understood. Will deep neural networks always remain a black-box, or can we understand some aspects of how they work? I will describe my research of the past few years towards building scientific foundations for deep learning, developing approaches based on solvable models, dynamical systems, and statistical physics. I’ll present three threads from my past work. First, I will describe exact connections between large-width deep neural networks with Gaussian field theories and new classes of kernel methods. Second, I will discuss an equivalence between deep neural networks and linear models and then characterize a nonlinear regime where the equivalence breaks down, which we study through a simple dynamical system. Finally, I will describe insights we have gained into the power-law performance of deep neural networks, as a function of the number of parameters and the amount of data, by leveraging prior results on solvable models. I’ll close by discussing directions for future work both in deep learning theory and at the intersection of physics and machine learning.
3/28/23 (in-person) Prof. Zhiyue Lu, University of North Carolina
Title: Energy Landscape Design Principles of Life-like Responsiveness to Non-equilibrium Driving Forces
Abstract: Living organisms can respond to changes in their environment and harvest energy from non-equilibrium systems. This talk explores the design principles that allow chemical systems to exhibit life-like responsiveness to non-equilibrium driving forces. The signal responsiveness and energy harvesting ability are demonstrated by two toy models. The first model demonstrates that given a rough and frustrated energy landscape, a single polymer can discern and respond to temporal patterns of external stimuli. The polymer behaves as a deterministic finite automata. The second model demonstrates a geometric approach to derive the optimal energy landscape for catalysts and enzymes with the ability to harvest environmental energy from oscillatory environments. The theoretical framework implies the existence of a new type of catalyst that is enhanced by non-equilibrium oscillations.
4/11/23 (in-person) Prof. Todd Gingrich, Northwestern University
Title: Nonequilibrium steady states: principles, simulations, and rates
Abstract: Chemical systems can deviate from equilibrium for a variety of reasons: because they are kinetically trapped, because they are subject to a time-varying drive, or because they are simultaneously in contact with multiple incommensurate reservoirs. This last scenario, which generates a nonequilibrium steady state (NESS), yields a stationary distribution over microstates that is not Boltzmann and that sustains currents. I will discuss three parallel efforts to better understand the chemical dynamics of such steady states. First, I will introduce a thermodynamic uncertainty relationship constraining the magnitude of fluctuations in currents. Next, I will discuss my group’s efforts to elucidate structure-function relationships in model molecular motors by simulating the NESS with a mixture of Langevin dynamics and Grand Canonical Monte Carlo chemostats. Time permitting, I will show how tensor network methods, popular in physics literature for quantum dynamics of spin chains, can be repurposed to compute rates for nonequilibrium reaction-diffusion processes.
4/25/23 (in-person) Prof. Glen Hocky, New York University
Title: Modeling assembly of colloids with charges and with mobile binders
Abstract: In this talk, I will present our recent efforts in probing the physical processes underlying self-assembly of colloidal gels and crystals. Nano-meter to micron sized particles in suspension can be a powerful platform for assembly novel functional materials, but the challenge is program in their interactions such that desired functionality is achieved. Moreover, for practical purposes this must be done on a large scale. First, I will discuss our work on using particles with many mobile binding sites, where particles can 'choose' their number of neighbors by assembling adhesion patches between particles. Second, I will discuss nucleation and growth of crystals formed from pairs of charged colloidal particles in suspension.
Refs:
1) Ionic Solids from Common Colloids. Theodore Hueckel, Glen M. Hocky, Jeremie Palacci, and Stefano Sacanna. Nature, 580, 487-490 (2020)
2) A Coarse-Grained Simulation Model for Colloidal SelfAssembly via Explicit Mobile Binders. Gaurav Mitra, Chuan Chang, Angus McMullen, Daniela Puchall, Jasna Brujic, and Glen M. Hocky. arXiv:2212.11946, In revision (2023)
5/2/23 (in-person) Dr. Guillaume Stirnemann, Institut de Biologie Physico-Chimie, Paris
Title: When the study of the origins of life meets AI and non-equilibrium thermodynamics
Abstract: The emergence of life is one of the most fascinating and yet largely unsolved questions in the natural sciences, and thus a significant challenge for scientists from many disciplines. I will present some of our recent results that specifically address two major issues: the accumulation of biological precursors under temperature gradients and the formation of phosphoester bonds (linking successive nucleotides) in abiotic condition. In the first part, I will discuss our recent investigation of the poorly-understood thermally-driven process of thermophoresis, which was shown to be an efficient process for the accumulation of dilute precursors in the absence of biological compartments. I will show how our non-equilibrium molecular simulations and calculations [1,2] can shed light on the molecular mechanisms associated with thermodiffusion, but also how important open questions remain. For the phosphoester bond formation, I will show how deep-learning strategies used for the design of reactive forcefields [3] enable to reach unprecedented simulation timescales with quantum accuracy and therefore, enable to use unprecedented statistical sampling strategies that could not have been considered even a few years ago. They overpower both traditional ab-initio molecular dynamics approaches, which suffer either from very poor sampling and/or poor quantum accuracy [4], as well as static energy map calculations often done in reactivity studies that oversimplify environment and entropy effects.
[1] Diaz-Marquez & Stirnemann, J. Chem. Phys. 155, 174503 (2021)
[2] Diaz-Marquez & Stirnemann, Eur. Phys. J. E 45, 37 (2022)
[3] Wang, Zhang, Han & E, Comp. Phys. Commun. 228, 178 (2018)
[4] Benayad, Saint-André & Stirnemann, J. Phys. Chem. B 126, 8251 (2022)
Seminars for Fall 2022
12/6/22 (virtual) Prof. Peter Bolhuis, University of Amsterdam
Title: A maximum caliber approach for tuning molecular kinetics
Abstract: Empirical force fields used in Molecular Dynamics simulations of complex biomolecules and materials are often approximate. Statistical methods based on the maximum-entropy principle are able to increase the accuracy of “structural and thermodynamical ensembles” obtained by such molecular dynamics simulations through incorporating experimental information. One would like to go even further, by obtaining ‘kinetic ensembles’, comprising the structures of the different states of a molecular system, their populations and their interconversion rates. However, up to recently there was no systematic way to incorporate experimental dynamical and kinetic information into molecular dynamics simulations. We developed a method of imposing known rate constants (and other dynamical observables) as constraints in molecular dynamics simulations, based on a combination of the maximum-entropy (MaxEnt) and maximum-caliber principles (MaxCal) [1]. Starting from an existing ensemble of (rare event) dynamical trajectories or paths, e.g. obtained from transition path sampling, each path is reweighted in order to match the calculated and experimental interconversion rates of a molecular transition of interest, while minimally perturbing the prior path distribution. This kinetically corrected ensemble of trajectories leads to improved structure, kinetics, thermodynamics, free energy landscapes, as well as mechanistic insights that may not be readily evident directly from the experiments. Finally, the MaxCal approach can also be used to fine-tune the molecular force fields themselves, and might provide an efficient means for designing kinetic behavior [2].
[1] Z. F. Brotzakis, M. Vendruscolo, and P. G. Bolhuis, Proc. Natl. Acad. Sci. 118, (2020).
[2] P. G. Bolhuis, Z. F. Brotzakis, and B. G. Keller, arXiv:2207.04558 (2022).
11/29/22 (in-person) Prof. Jordan Horowitz, University of Michigan
Title: Bounding nonequilibrium response: from biochemical sensitivity to transport in active fluids
Abstract: Diverse physical systems are characterized by their response to small perturbations. Near thermodynamic equilibrium, the fluctuation-dissipation theorem provides a powerful theoretical and experimental tool. Its great utility near equilibrium has led to significant interest in expanding its validity and developing generalizations for nonequilibrium situations. In this talk, I will introduce a novel parameterization of response that allows us to derive equalities and inequalities that quantitatively capture the limits to far-from-equilibrium response. As illustrations, I will show how these predictions rationalize known energetic requirements for some biochemical motifs and provide new limits to others. I will also demonstrate how these predictions can be used to derive Green-Kubo relations for the transport coefficients that enter hydrodynamic descriptions of active fluids.
11/8/22 (virtual) Prof. Kateri DuBay, University of Virginia
Title: Modeling nanoscale self-assembly within spatially complex and temporally variant environments
Abstract: Nanoscale self-assembly arises from and is highly sensitive to interactions among the assembling components as well as interactions between them and their environment. Environmental complexities, such as spatial and temporal heterogeneities, are ubiquitous in self-assembling biological systems and materials processing protocols, yet our understanding of how they influence the assembly process remains limited. Our group uses numerical simulations to investigate how particles self-organize within these complex environments. Specifically, this talk will describe our investigations into the emergence of sequence-influencing assemblies of nascent oligomers during step-growth copolymerizations and the self-assembly of viral capsid-like structures within oscillatory environments.
11/1/22 (in-person) Prof. Abraham Nitzan, University of Pennsylvania
Title: Collective response in light-matter interactions: The interplay between strong coupling, local dynamics, and disorder
Abstract: Because molecular size is much smaller than optical wavelengths molecules can interact with light collectively. Manifestations of such collective response depend on other factors such as dephasing and disorder. Under strong (molecules-light) coupling conditions such collective response can be more pronounced because collective modes are energy-shifted from the spectrum of individual emitters and appear as polaritonic spectral features in linear and non-linear optical signals. In this talk I will describe recent studies of several phenomena that reflect this interplay between collective response, local dynamics and (static and dynamic) disorder.
10/11/22 (virtual) Prof. Cristina Marchetti, University of California, Santa Barbara
28th Annual Shih-I Pai Distinguished Lecture, link for details
9/27/22 (virtual) Dhruv Bansal, Co-Founder and CSO of Unchained Capital
Title: Why physicists should care about bitcoin (no I'm not selling anything)
Host: Prof. Victor Yakovenko
Abstract: Bitcoin is a new kind of digital money that's often in the news these days, usually for the wrong reasons. But bitcoin is also a real-world network that uses energy and transports information. This means we can use the tools of physics to analyze and constrain bitcoin's behavior. I propose some simple models for bitcoin from the perspectives of dynamical systems, thermodynamics, and statistical mechanics. I hope these lead to some interesting questions and speculations.
Biography: Graduate student at University of Texas Austin, Center for Nonlinear Dynamics (CNLD) in 2008-2011, studying networks, fluids, school systems, computational physics, and data science. Started Infochimps company in 2008 with partners, dropped out of graduate school in 2011, sold Infochimps in 2013. Discovered bitcoin in 2011, ignored it. Bought "too late" in 2013. By 2016, working in bitcoin full-time with Unchained.
9/13/22 (virtual) Prof. Juan Garrahan, University of Nottingham
Title: Tensor network methods and large deviations
Abstract: I will describe the application of tensor networks for the study of dynamical fluctuations in systems with stochastic dynamics. I will discuss several related questions: (i) how to obtain the long-time statistics, i.e. the large deviations, of trajectory observables from tilted Markov generators (in analogy with finding quantum ground states); (ii) how to exploit these methods to efficiently sample rare events; (iii) how to extend this approach to finite time trajectories (in analogy with calculating quantum partition sums); (iv) the emergence of symmetry protected topological phases in atypical dynamics.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Seminars for Spring 2022
5/3/22 (in-person) Prof. Monica Olvera de la Cruz, Northwestern University
Title: Metallization of colloids
Abstract: Numerous colloidal crystals have been designed and devised following design rules akin to atomic crystals. I will review techniques using functionalization and/or electrostatic interactions in mediating interactions among nanoparticles. I will discuss the resulting properties of the crystals including transport mediated by a localization to delocalization transition (LDT) in mixtures of asymmetric in size nanoparticles where the small nanoparticles roam the crystal holding the large nanoparticles in specific lattices sites, akin to electron clouds in atomic metals. The LDT, when discontinuous, is driven by lattice vibrations. It is observed in many systems, including atomic system, such as superionics, where it is known as sublattice melting.
4/26/22 (in-person) Prof. Silvina Matysiak, University of Maryland
Title: Effect of Physiological Environment on Neurodegenerative Peptide Aggregation
Abstract: Molecular-level self-assembly/aggregation processes are common in biomolecular systems. Specifically, peptide aggregation results in the formation of biomolecular deposits (amyloid), which has been associated with neuronal dysfunction leading up to neurodegeneration. Peptide aggregation is often influenced by several environment-derived entities, which can modulate this pathological process in a specific or non-specific manner. In this talk, I will present on the molecular mechanism of how the environment modulates amyloid aggregation with biomolecular simulations. To study this process at long spatial and temporal scales, we created and applied computational models at coarse-grained (CG) resolution. Our CG models can capture the biophysics of environment-stimulated conformational transitions that are common with neurodegenerative peptides. Experimentally it has been observed that lipid membrane surfaces, hyperglycemic conditions and components of the extracellular matrix can modulate in different ways amyloid formation and the aggregate morphology. In this presentation, I will describe the molecular mechanism of how crowding, membrane curvature, peptide-lipid and peptide-polysaccharide/monosaccharides interactions tune the emergence of different amyloid peptide aggregation kinetics and morphologies.
4/19/22 (virtual) Prof. Jindal Shah, Oklahoma State University
Title: What Does Non-Ideality Mean for Binary Ionic Liquid Mixtures?
Abstract: Ionic liquids are substances that are composed entirely of ions. Negligible vapor pressures and the availability of a large number of cations and anions to tune physicochemical and biological properties for a given chemical process have been the primary drivers for research in this field over the last two and half decades. A major thrust of these investigations is on elucidating changes in the properties of pure ionic liquids by altering the cation, anion or substituents on the ions. Another approach to expand the range of available ionic liquids is to form ionic liquid-ionic liquid mixtures. From a thermodynamic point of view, the knowledge of the extent of non-ideality in these binary ionic liquid mixtures and the molecular level details enable a priori prediction of thermophysical properties of ionic liquid mixture. In this presentation, we will demonstrate that the difference in the molar volume of the ionic liquids forming the mixture and the difference in the hydrogen bonding ability of the anions can serve as metrics for the prediction of non-ideality in the binary ionic liquid systems. Such non-idealities appear in the form local structural organization of anions around the cation in ionic liquid mixtures bearing a common cation and two different anions. We will further highlight that these non-native structures lead to a different dissolution mechanism for CO2 in mixtures in comparison to that for pure ionic liquids although the CO2 solubilities obey apparent ideal mixing rule.
4/5/22 (in-person) Prof. Joel Yuen-Zhou, University of California San Diego
Title: Unusual light-molecule interactions: Polariton condensates and other nonlinear phenomena
Abstract: Confinement of molecules in optical microcavities gives rise to collective strong light- matter coupling and the formation of hybrid light-matter modes known as polaritons. While these modes potentially offer unusual chemical reactivity, their use is limited by their very short lifetimes owing to an entropic flow of population into the so-called dark modes. However, past a certain threshold of optical pumping, the population of polariton modes can become self-catalytic, giving rise to a nonequilibrium version of Bose-Einstein condensation. Polariton condensation has been routinely demonstrated with organic dyes at room temperature since the last decade. I will provide a chemical outlook for these condensates, demonstrating that their photochemical reactivity can be rather rich and different from plain laser-driven molecules. In the second part of my talk, I will discuss interesting connections between topological phases of matter and chiroptical spectroscopy of molecules, describing a scheme to distinguish molecular enantiomers using the recently introduced ideas of topological frequency conversion.
REFERENCES
1. S. Pannir-sivajothi, J. A. Campos-Gonzalez-Angulo, L. A. Martínez-Martínez, S. Sinha, and J. Yuen-Zhou, Driving chemical reactions with polariton condensates, arXiV:2105.10449, Nat. Commun., in press (2022).
2. K. Schwennicke and J. Yuen-Zhou, Enantioselective topological frequency conversion, J. Phys. Chem. Lett. 13, 10, 2434 (2022).
3/29/22 (virtual) Prof. Valeria Molinero, University of Utah
Title: Towards the elucidation of the mechanisms of synthesis of zeolites
Abstract: Zeolites are porous silicates that constitute the main solid catalysts used by the chemical industry. These structurally complex solids are synthesized from aqueous solutions through a multi-stage process that involves multiple phase transformations mediated by the chemistry of polymerization of silica. Organic cations, typically tetraalkylammonium ions, are used to direct the synthesis towards specific zeolite polymorphs. Nevertheless, the molecular mechanisms by which the cations and silicates form the zeolites are not well understood. This presentation will discuss our current work using molecular simulations and nucleation theory to elucidate at which stage zeolitic order emerges from the synthesis mixture, the roles of nucleation and growth in the selection of zeolite polymorphs, and what is the smallest size of nanozeolite that can be synthesized.
3/8/22 (virtual) Dr. Peter Battaglia, DeepMind
Title: Physical inductive biases for learned simulation and scientific discovery
Abstract: This talk will explore both how our knowledge of physics can improve our machine learning approaches, and how our machine learning tools can be used to improve our knowledge of physics. I'll describe work my collaborators and I have done using particle- and mesh-based approaches for learning simulation, how we leverage inductive biases about ODEs, Hamiltonian and Lagrangian mechanics in learned simulators, and how we can use neural networks with symbolic regression to discover physical governing equations from simulated and real data.
2/22/22 (virtual) Dr. Christopher Moakler, University of Maryland
Title: The Hydra String Method and its Application to High Dimensional Potential Energy Surfaces Arising from Granular Systems
Abstract: Granular materials are a ubiquitous yet ill-understood class of media. Different approaches and techniques have been developed to understand the many complex behaviors they exhibit, but none have been completely successful. I have instituted a novel means to understand granular materials. This novel method, the Hydra String Method (HSM), is an efficient and autonomous way to trawl the potential energy surfaces (PESs) to enumerate the saddles, minima, and connections between them. I have applied the Hydra String Method to bi-disperse configurations of soft spheres to map out ensembles of pathways between stable packings of the system. These transition pathways are a low-dimensional projection of the larger PES. By understanding these pathways and how they connect to one another may allow for the prediction of the dynamics of a granular system as it moves between stable packings.
2/8/22 (in-person) Prof. Ichiro Takeuchi, University of Maryland
Title: Closed-loop autonomous combinatorial experimentation for streamlined materials discovery
Abstract: Machine learning has become an integral part of many aspects of fundamental research. It is particularly useful in high-throughput materials exploration where it can be used to predict and navigate a series of experiments, as well as perform rapid data analysis. In this talk, I will discuss how we are incorporating active learning in combinatorial screening and discovery of functional materials. The array format with which a large number of different composition samples are laid out on combinatorial libraries is particularly conducive to active learning driven autonomous experimentation. We have previously demonstrated discovery of a new phase change memory (PCM) material using the closed-loop autonomous materials exploration and optimization (CAMEO) strategy. The discovered PCM material has been tested in various scaled-up device formats and continues to exhibit superior performance to industrial standards. Recent efforts in developing live autonomous synthesis–measurement as well as experiment-theory closed loops will be discussed. This work is performed in collaboration with A. Gilad Kusne, V. Stanev, H. Yu, H. Liang, M. Li, E. Pop, and A. Mehta. This work is funded by SRC, ONR, AFOSR, and NIST.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Seminars for Fall 2021
9/14/21 (virtual) Joan-Emma Shea, University of California, Santa Barbara
Title: Self-Assembly of the Tau Protein: Liquid-Liquid Phase Separation and Fibril Formation
Abstract: Intrinsically disordered peptides (IDP) are a special class of proteins that do not fold to a unique three-dimensional shape. These proteins play important roles in the cell, from signaling to serving as structural scaffolds. Under pathological conditions, they can self-assemble into structures that are toxic to the cell, and a number of neurodegenerative diseases are associated with this self-assembly process. My talk will focus on the Tau protein, an IDP that binds to microtubules and can form fibrillar aggregates, a process that has been linked with Alzheimer’s disease. In addition to forming fibrils, the Tau protein can also phase separate into a protein rich and a protein depleted phase, a process known as liquid-liquid phase separation (LLPS). This process may play a protective role in the cell against pathological fibrillization. I will present molecular dynamics and field theoretic simulations that map out the phase diagram for Tau LLPS, and use this phase diagram to predict the conditions under which Tau can be driven towards LLPS under live cell coculturing conditions.
9/21/21 (virtual) David Sivak, Simon Fraser University
Title: Information thermodynamics of the transition-path ensemble
Abstract: The reaction coordinate describing a transition between reactant and product is a fundamental concept in the theory of chemical reactions. Within transition-path theory, a quantitative definition of the reaction coordinate is found in the committor, which is the probability that a trajectory initiated from a given microstate first reaches the product before the reactant. Here we develop an information-theoretic origin for the committor and show how selecting transition paths from the equilibrium ensemble induces entropy production which exactly equals the information that system dynamics provide about the reactivity of trajectories. This equality of entropy production and dynamical information generation also holds at the level of arbitrary individual coordinates, providing parallel measures of the coordinate's relevance to the reaction, each of which is maximized by the committor.
9/28/21 (virtual) Gerhard Stock, University of Freiburg
Title: Learning biomolecular collective variables
Abstract: The statistical analysis of molecular dynamics simulations requires dimensionality reduction techniques, which yield a low-dimensional set of collective variables x_i = x that in some sense describe the essential dynamics of the system. Considering the distribution P(x) of the collective variables, the primal goal of a statistical analysis is to detect characteristic features of P(x), in particular, its maxima and their connection paths. This is because these features characterize the low-energy regions and the energy barriers of the corresponding free energy landscape, and therefore amount to the metastable states and transition regions of the system. In this seminar, we outline a systematic strategy to identify collective variables and metastable states, which subsequently can be employed to construct a Langevin or a Markov state model of the dynamics. In particular, we account for the still limited sampling typically achieved by molecular dynamics simulations, which in practice seriously limits the applicability of theories (e.g., assuming ergodicity) and black-box software tools (e.g., using redundant input coordinates). We show that it is essential to use internal (rather than Cartesian) input coordinates, employ dimensionality reduction methods that avoid rescaling errors (such as principal component analysis), and perform density based (rather than k-means-type) clustering. Finally we discuss a machine learning approach to dimensionality reduction, that highlights the essential internal coordinates of a system and may reveal hidden reaction mechanisms.
F. Sittel and G. Stock, Perspective Article in J. Chem. Phys 149, 150901 (2018)
S.Brandt, F. Sittel, M. Ernst, and G. Stock, Machine Learning of Biomolecular Reaction Coordinates, J. Phys. Chem. Lett. 9, 2144 (2018)
10/5/21 (virtual) Chris Mundy, PNNL
Title: Top-down and bottom-up modeling of aggregation and assembly in complex material systems
Abstract: Herein, I will discuss progress in defining a theoretical framework to describe the assembly and aggregation processes in complex systems. It is well established that reduced models provide a useful path forward in describing nucleation and assembly. The experimental systems that we will consider will have different assembly outcomes that depend on the underlying solution conditions (pH, ionic strength, electrolyte). We will discuss our approach to modeling the effects of the solution conditions using the tools of colloidal theory in conjunction with molecular simulation. We will establish the important role of the short-range molecular detail in conjunction with long-range mean field phenomena to establish quantitative reduced models of assembly. Both bulk and interfacial systems will be considered. This work is supported by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences (BES), Division of Material Sciences and Engineering and supported by the DOE, Office of Science, BES, Division of Chemical Sciences, Geosciences, and Biosciences and DOE, Office of Science, Office of BES as part of the Energy Frontier Research Centers program: CSSAS, The Center for the Science of Synthesis Across Scales under [award number DESC0019288]. PNNL is a multiprogram national laboratory operated for the DOE by Battelle under contract no. DE-AC05-76RL01830
10/12/21 (virtual) Jonathan Weare, New York University
Title: A fresh look at learning long-timescale phenomena from trajectory data
Abstract: Events that occur on very long timescales are often the most interesting features of complicated dynamical systems. Even when we are able to reach the required timescales with a sufficiently accurate computer simulation, the resulting high dimensional data is difficult to mine for useful information about the event of interest. Markov state modeling (MSM) in particular has proven a powerful tool for turning trajectory data into useful understanding of long-timescale processes. Taking a new perspective on trajectory data analysis methods I will describe a family of methods that generalize MSMs with the aim of computing predictions of specific long timescale phenomena using only relatively short trajectory data (e.g. much shorter than the return time of the event). This new perspective points in exciting new directions for both rare event analysis algorithms as well mathematical analysis. In particular, I will explain the remarkable error properties in approximations of specific rare event forecasts achievable using a data set of short trajectories alone.
10/19/21 (virtual) Annual shih-i Pai lecture
10/26/21 (in-person) Jan Sengers, University of Maryland
Title: Mass and Thermodiffusion in Multicomponent Fluid Mixtures
Abstract: PDF file
11/2/21 (virtual) Armita Nourmohammad, University of Washington
Title: Adaptive immunity in light of host-pathogen coevolution
Abstract: The adaptive immune system consists of highly diverse B and T cells whose unique surfacereceptors enables them to mount specific responses against a multitude of pathogens. Adaptive immunity incorporates all aspects of life, from molecular signaling to cellular evolution. The result is an information processing molecular organization with many interacting components, which can reliably sense and adaptively respond to diverse and evolving pathogens. With the growing amount of molecular data, we can now quantify the sequence diversity generated in immune repertoires. However, we still lack an understanding of how such diversity translates to immune function. In this talk, I will introduce a principled statistical framework to integrate interpretable biophysical models of immune receptor generation with flexible and powerful deep learning approaches to characterize sequence determinants of immune receptor function. Apart from these data-driven approaches, I will establish a theoretical framework to characterize the organization and encoding of information in the adaptive immune system to counter the out of equilibrium evolutionary drive that a host experiences from pathogens. These approaches will shed light on how the diversity of immune repertoires shape functional responses to ever- changing pathogenic environments.