Physics Colloquia

The beauty of physics lies in its ability to describe nature’s complex phenomena through a few fundamental principles and equations. Beneath this complexity, universal patterns emerge that transcend the microscopic details of individual systems. Such patterns are especially striking near continuous phase transitions, where systems with vastly different constituents and interactions can exhibit the same collective behavior. In this colloquium, I will present recent results that decipher such patterns within the framework of quantum field theory, drawing on the principles of unitary quantum mechanics, local interactions, and symmetry. We will see how these principles constrain physical observables and yield quantitative predictions even when a direct solution of the underlying dynamics is out of reach, echoing Einstein’s belief that “pure thought can grasp reality.”

Generative AI represents a groundbreaking development within the broader “Machine Learning Revolution,” significantly influencing technology, science, and society. In this talk, I will focus on remarkable connections between generative AI, in particular diffusion models, and statistical physics. Diffusion models are the state of the art to generate images, videos, and sounds. They are very fascinating algorithms for physicists, as they are very much connected to concepts from stochastic thermodynamics, particularly time-reversed Langevin dynamics. These diffusion models start from a simple white noise input and make it evolve through a Langevin process to generate complex outputs such as images, videos, and sounds. I will show that statistical physics provides principles and methods to characterise this generation process. The emergence of features can be understood through the lens of symmetry breaking and dynamical phase transitions, whereas the emergence of creativity - or its absence (memorisation) - can be explained using theories of disordered systems, and dynamics in complex energy landscapes.
I will conclude by discussing the fundamental reasons behind the success of diffusion models, highlighting a deep connection between their multi-scale generation mechanism and the renormalisation group analysis of physical systems.

Physical understanding benefits from simplicity and interpretability, yet quantitative prediction in complex systems often demands much richer descriptions. In this talk, I will show how compression can be used to extract physical insights from systems in which the microscopic details still matter quantitatively. Using examples drawn from heteropolymers, biomolecular systems, and programmable soft matter, I will demonstrate that the thermodynamic structure of multicomponent systems can itself be compressed, so that molecules acquire low-dimensional representations through their interactions with other components. These representations retain the relevant physical information while enabling quantitatively accurate predictions of thermodynamic properties across a wide range of mixtures and conditions. By revealing which microscopic details control collective behavior, this approach provides a systematic route to infer effective variables, extrapolate to new systems and conditions, and rationally design multicomponent matter.

One of the largest uncertainties for future predictions of sea level rise is the dynamic mass loss of ice into the ocean. Icebergs break off from ocean-terminating glaciers and form granular piles that jam up future calving events. In this iceberg purgatory, a floating granular mélange, forces are transmitted through contacts between individual icebergs 10-100 meters in size. Further upstream, before reaching the ocean, bulk glacial ice creeps slowly over time, behaving like a non-Newtonian liquid where the relationship between stress and deformation rate is a key factor for making predictions about the stability of ice shelves and ice sheets. Glacial ice is composed of polycrystalline grains that can be centimeters in size and form a disordered network of grain boundaries separating crystal orientation. Within each grain are disordered arrays of dislocations whose dynamics determine molecular-scale creep. Within this multiscale problem, I will primarily focus on the slow creep of glaciers, from molecules to grains. Recently we developed a physics-based rheological model that captures all three regimes of transient creep in polycrystalline ice under constant stress. The model replicates the minimum strain rate near 1% strain, which results from a competition of two timescales that depend on the driving stress. The model also exhibits apparent power-law exponents in the stress dependence of the strain-rate minimum, which is a quantity often reported in experimental data and directly used in large-scale climate models.

In this talk, we study the propagation, entanglement, and entropy of quantum information using 3x3 and 4x4 arrays of superconducting qubits that emulate the two-dimensional (2D) hard-core Bose-Hubbard model. The 2D arrays feature site-selective, simultaneous control and readout of all qubits. We highlight several recent experimental demonstrations using these arrays, including quantum random walks, Anderson and Wannier-Stark localization [1], the crossover from area-law to volume-law entanglement [2], the realization of synthetic electromagnetic fields throughout the array [3], and studies of flat-band physics in rhombic lattices [4].
[1] npj Quantum Information 8, 35 (2022) | arXiv:2107.05035 (2021)
[2] Nature 629, 561-566 (2024) | arXiv:2306.02571 (2023)
[3] Nature Physics 20, 1881-1887 (2024) | arXiv:2405.00873 (2024)
[4] Physical Review X 15, 021091 (2025) | arXiv:2410.07878 (2024)