Living systems exhibit various robust dynamics during system regulation, growth, and motility. However, how robustness emerges from stochastic components remains unclear. Towards understanding this, I develop topological theories that support robust edge currents and localization, effectively reducing the system function to a lower-dimensional subspace. I will introduce stochastic networks in molecular reaction space that model long and stable time scales, such as the circadian rhythm. More generally, we prove that unlike their quantum counterparts, stochastic topological systems require non-Hermiticity for edge states and strong localization. I will close by discussing experimental platforms for the detection and use of edge currents for self-assembly and replication in living systems.

Neural responses in association brain areas during cognitive tasks are heterogeneous, and the widespread assumption is that this heterogeneity reflects complex dynamics involved in cognition. However, the complexity may arise from a fundamentally different coding principle: the collective dynamics of a neural population encode simple cognitive variables, while individual neurons have diverse tuning to the cognitive variable, similar to tuning curves of sensory neurons to external stimuli. We developed an approach to simultaneously infer neural population dynamics and tuning functions of single neurons to the latent population state. Applied to spike data recorded from primate premotor cortex during decision-making, our model revealed that populations of neurons encoded the same dynamic variable predicting choices, and heterogeneous firing rates resulted from the diverse tuning of single neurons to this decision variable. The inferred dynamics indicated an attractor mechanism for decision computation. Our results reveal a unifying geometric principle for neural encoding of sensory and dynamic cognitive variables.

Connections between neurons can be mapped by acquiring and analyzing electron microscopic (EM) brain images. In recent years, this approach has been applied to chunks of brains to reconstruct local connectivity maps that are highly informative yet inadequate for understanding brain function more globally. We reconstructed the first neuronal wiring diagram of a whole adult brain, containing >50 million chemical synapses between 139,255 neurons reconstructed from a female Drosophila melanogaster. In this talk, I will highlight the technological progress leading up to the creation of this resource. Further, I will discuss how the connectome can be used to study the global organization of the brain and facilitate the tracing of synaptic pathways from the inputs (e.g., sensory neurons) to outputs (e.g., descending neurons). The technologies and open ecosystem developed for the fly brain connectome set the stage for future large-scale connectome projects in other species. I will give an outlook on connectomes for large mammalian brains.

Shear thickening in concentrated suspensions occurs when the imposed stress drives particles together to form a frictional contact network. This implies a balancing resisting force, whose magnitude sets the threshold for the imposed stress to induce contacts. Using an established simulation technique [1] and motivated by recent work showing the rapid onset of system-spanning rigid structures (identified by a pebble game algorithm) with increase of volume fraction [2], we show for a two-dimensional (monolayer) suspension that the number of contacts in the network is an increasing function of stress, but it also fluctuates during flow. We go on to explore the development of minimally rigid structures in the shear-thickened suspension as it approaches jamming at high stress, and show that the onset of large rigid clusters exhibits critical behavior. The possible relationship of this behavior to other work showing critical scaling in this system will be discussed.

[1] R. Mari, R. Seto, J. F. Morris & M. M. Denn 2014 Shear thickening, frictionless and frictional rheologies. J. Rheol. 58, 1693.

[2] M. van der Naald, A. Singh, T.T. Eid, K. Tang, J. J. de Pablo & H. M. Jaeger, H.M. 2024. Minimally rigid clusters in dense suspension flow. Nature Physics 20, 653–659.

In this talk, I will share two stories. First, I will share our discoveries on why aquatic worms braid, tangle, and knot with their neighbors to form extraordinary living “blobs”. These soft, squishy “living polymers” shape-morph, crawl, float, and untangle – the stuff of science fiction. Through mathematical models and robotic analogs, I will explain how these worms solve Gordian knot problems, inspiring active, topological tunable robotic swarms.

Second, I will explore the century-old practice of sheepdog trials to uncover strategies for controlling noisy, indecisive herds of sheep. In these competitions, sheep stochastically switch between following and ignoring control signals, yet skilled dog-handler teams excel at herding and splitting them on demand. Using simulations and temporal network theory, I will demonstrate how stochastic indecisiveness—a behavior often seen as a challenge—can instead be harnessed to efficiently control noisy swarms, with applications in swarm robotics, opinion dynamics, and adaptive networks.

Mass loss from the Antarctic Ice Sheet occurs due to the viscous ice flow of glaciers from the interior of ice sheets towards the ocean, along with brittle fracture of ice into icebergs. Viscous flow and brittle fracture of ice are governed by microphysical processes that occur on scales of ice crystals or smaller. However, incorporating physics on these small spatial and temporal scales into ice sheet models is made difficult by gaps in our physical understanding of these microphysical processes and by the computational cost of models that resolve them. Here, we propose a framework for incorporating small-scale processes governing ice deformation into large-scale ice sheet models. In particular, we present a bulk parameterization method for incorporating ice crystal-scale processes into models of viscous ice flow, and we show that this parameterization provides a way of representing the effect of varying flow conditions on future ice sheet behavior. Our results help to explain the range of ice deformation processes previously inferred in laboratory and field studies and allow for the future development of a microphysics-driven ice flow model.

Living matter functions conceptually differently from non-living matter. It is active and is organized in space and time through the interaction of five major types of processes: biochemical reactions, diffusion, noncovalent self-assembly, phase separation, and mechanical motion. This design provides adaptivity, evolvability, and the ability to self-replicate, which are unique for life. In contrast, the chemists’ ability to build dynamically organized systems (e.g., chemical oscillators) is limited. Interconnections and feedback loops between different processes make them non-modular (holistic) and, consequently, hard to understand and rationally construct. Nevertheless, the ability to construct dynamically organized systems opens possibilities (i) to obtain materials with life-like properties and (ii) to probe the role of dynamic self-assembly in the origin of Life.

In this talk, I propose using the chemists’ ability to design and synthesize molecules for the rational construction of dynamic systems and materials. I will illustrate this strategy with the rational design of chemical oscillators, waves, patterns, actuators, and microstructures. In perspective, this work opens a path toward constructing life-like dynamic materials and observing emergent phenomena in prebiotically relevant chemistry.

As atmospheric carbon dioxide levels continue to rise, there has been increasing interest from both the public and private sectors in methods and technologies for directly removing carbon dioxide from the atmosphere (CDR). Large-scale CDR may be a useful complement to emissions reductions in achieving specific carbon or climate goals. The American Physical Society's Panel on Public Affairs (POPA) has recently released a report that focuses on the physical constraints and requirements of large-scale CDR efforts. This talk will present the main aspects of the science analysis and policy conclusions of the report. In particular, the talk will describe the energy, land area, and materials requirements for a variety of CDR approaches, as well as uncertainties in effectiveness and impact. These basic scientific observations can help inform sensible carbon management policies.

In classical mechanics we typically study systems that can be represented by point-like particles and are subject to external forcing. In contrast, in living systems we encounter extended objects which undergo internal forcing to generate cyclic changes in shape or configuration. Such dynamics occur across scales, from proteins undergoing conformational changes to cells crawling on surfaces to centipedes wiggling through soil and debris. The physics of such “self-deforming” systems is remarkably rich, especially when coupled to non-trivial environments. In this talk I will narrate a few examples of surprising emergent dynamics in living and nonliving (robot) systems arising from the interplay of traveling waves of body bending and environmental heterogeneities. I will describe how undulatory limbless robophysical models that mix wave and particle-like properties can mechanically “diffract” in regular arrays of obstacles and exponentially localize in disordered environments. Such insights have proven useful in highlighting the importance of mechanics in living systems, e.g. the role of passive dynamics in the locomotion of snakes and nematode worms in heterogeneous environments. Discovery of principles of self-deforming systems is informing and guiding the commercialization of elongate robots (e.g. in a startup company I co-founded) that can self-propel effectively in dirty, dull, and dangerous situations.

Biology provides numerous examples of phase-separated protein and nucleic acid condensates, which establish distinct compartments for spatially organizing biomolecules within living cells. This mechanism of spatial organization relies on the ability of biomolecular systems to navigate high-dimensional phase diagrams by tuning the interactions among proteins and nucleic acids in a multicomponent mixture. In this talk, I will discuss optimization and machine learning approaches that can be used to design the compositions of multicomponent biomolecular condensates. These approaches reveal how compositional specificity is encoded in complex mixtures and how intrinsically disordered protein sequences can be designed to stabilize coexisting multicomponent condensates. These results shed light on the physicochemical limits of phase-separation-mediated spatial organization in biological systems and establish practical strategies for engineering fully programmable biomolecular condensates.

It is possible to reduce some of the climate risks of accumulated CO2 by deliberately altering the Earth's albedo using Sunlight Reflection Methods (SRM) also called solar geoengineering. It is possible to remove carbon from the atmosphere at large scale using various methods for Carbon Dioxide Removal (CDR). Estimates of the cost, risks, and efficacy these tools will remain uncertain but it is now possible to make some policy-relevant quantitative comparisons between risks and benefits, and to speculate about the appropriate use of energy-system decarbonization, CDR, and SRM.

Whereas the governing equations of fluids and solids as single phases were derived over a century ago, the physics at the interface between these continua can be surprisingly rich and complex – instabilities emerge, interfacial forces become dominant, and mechanical fields vary across scales, from the molecular- to the sample-scale. Our lab probes these phenomena with impacting droplets and cracks in hydrogels, where we directly image the kinematics of obscured interfaces using 3D microscopy. In this talk, I will discuss two vignettes: first, the emergence of an interfacial instability beneath an impacting droplet of alcohol on an atomically smooth mica substrate, and second, the geometry and stability of complex cracks. These seemingly disparate systems are connected on a variety of levels, from their sensitivity to defects, to the duality of contact formation and bond rupture, which implicate mechanical fields across scales. A discussion of some open questions and future perspectives will conclude the talk.

Lipid membranes are nanometer-thin soap-like films composed of molecules with hydrophilic and hydrophobic segments. To eliminate their edge energy lipid membranes form closed spherical shells or vesicles, structures that are ubiquitous and versatile, with applications in encapsulation, molecular transport, and drug delivery. Controlling vesicle topology is essential in these processes. The rapid dynamics and small scales make it challenging to study topological transitions of lipid vesicles. We develop and study fluid colloidosomes, which are micron-sized analogs of lipid vesicles assembled from rod-like particles. Their unique features enable real-time visualization of colloidosome assembly and disassembly pathways. Increasing the lateral size of a fluid two-dimensional disk-like sheet induces mechanical instability generating monodisperse permeable capsules. Closed vesicles disintegrate into open disks via an intermediate state that is topologically distinct from both the initial and final state.

Like silicon for electronics, DNA is the revolution-enabling material for biotechnology. It is chemically stable, readily produced with high purity, and, when spiked with well-chosen defects, its predictable structure supports a range of devices. Also like silicon, the DNA-based devices of today are being engineered on the nanoscale, where technologically relevant material properties are often dominated by aspects that are negligible in bulk. As a prime example, short (~100 bp) dsDNA are able to cyclize much more readily than the well-studied stiffness of long (~10 kbp) dsDNA would seem to allow. Years of debate and study suggest kinking of the double helix is the explanation. We have built a DNA device that allows these such bent states to reconfigure a micron-scale structure, so that fluorescence videomicroscopy can reveal the bend angles and their stiffness. We find the kinked states are meta-stable, with intermediate states that require twisting. Prospects for using kinks as the conformational change that animates a DNA-machine will be discussed.

Ultracold molecules are an emerging platform for quantum science that combines the techniques of atomic physics pioneered over the last half century, including quantum-state control and single particle detection/manipulation, with molecules' inherently rich internal structure.  I will present new efforts at UChicago toward building novel quantum phases of matter using the emerging technology of highly polar molecules cooled to nanokelvin temperatures. Specifically, we hope to realize exotic topological superfluids built from interacting gases of KAg molecules, which could feature extraordinary characteristics such as resistance to disorder, frictionless flow, and the emergence of Majorana particles. Another complementary goal is to leverage the strong dipole-dipole interactions to pioneer novel ways to load molecules into defect-free, low-entropy arrays for realizations of lattice spin models.

Complex systems composed of many interacting parts can display striking emergent behaviors as their size increases. In this talk, I will discuss recent work exploring such phases in two seemingly unrelated domains: ecological communities and neural networks. In laboratory microcosms, we demonstrate experimentally that as the number of species or their interaction strength increases, communities lose stability and display alternative attractors and persistent fluctuations. In parallel, we propose an explanation for the predictable improvements in performance of neural networks with model size. Because large language models represent more concepts than they have representational dimensions, concepts are represented through superposition. Using Anthropic’s toy model, we show that reconstruction error associated with superposition naturally gives rise to a power-law scaling of error with model width. Together, these results suggest that emergent behaviors due to interactions in ecosystems and in neural networks may reflect generic transitions resulting from system size and complexity.

Human gut commensal bacteria are routinely exposed to various stresses, and collateral effects are difficult to predict. I will discuss our recent efforts to build predictive models of microbiota responses to drugs, phage, and coalescence with other communities. As one example: to systematically interrogate community-level effects of drug perturbations, we screened stool-derived in vitro communities with 707 clinically relevant small molecules. Across ~5,000 community–drug interaction conditions, compositional and metabolomic responses were predictably impacted by nutrient competition, with certain species exhibiting improved growth due to adverse impacts on competitors. Changes to community composition were generally reversed by reseeding with the original community, although occasionally species promotion was long-lasting, due to higher-order interactions, even when the competitor was reseeded. Despite strong selection pressures, emergence of resistance within communities was infrequent. Finally, while qualitative species responses to drug perturbations were conserved across community contexts, nutrient competition quantitatively affected their abundances, consistent with predictions of consumer-resource models. Our study reveals that quantitative understanding of the interaction landscape, particularly nutrient competition, can be used to anticipate and potentially mitigate side effects of drug treatment on the gut microbiota.

Dense Associative Memories are recurrent neural networks with fixed point attractor states that are described by an energy function. In contrast to conventional Hopfield Networks, which were popular in the 1980s, their modern versions have a very large memory storage capacity, which makes them appealing tools for many problems in AI and neuroscience. In this talk, I will provide an intuitive understanding and a mathematical framework for this class of models, and will give examples of problems in AI that can be tackled using these new ideas. Specifically, I will explore the relationship between Dense Associative Memories and two prominent generative AI models: transformers and diffusion models. I will present a neural network, called the Energy Transformer, which unifies energy-based modeling, associative memories, and transformers in a single architecture. Furthermore, I will discuss an emerging perspective that views diffusion models as Dense Associative Memories operating above the critical memory storage capacity. This insight opens up interesting avenues for leveraging associative memory theory to analyze the memorization-generalization transition in diffusion models.