Conference

The newest large-language reasoning models are for the first time powerful enough to perform mathematical reasoning in theoretical physics at graduate level. In the mathematics community, data sets such as FrontierMath are being used to drive progress and evaluate models, but theoretical physics has so far received less attention. In this talk I will present our dataset TPBench (arxiv:2502.15815, tpbench.org), which was constructed to benchmark and improve AI models specifically for theoretical physics. We find extremely rapid progress of models over the last months, but also significant challenges at research level difficulty. I will also briefly outline strategies to improve these models for theoretical physics.

Recent advancements have positioned Large Language Models (LLMs) as transformative tools for scientific research, capable of addressing complex tasks that require reasoning, problem-solving, and decision-making. Their exceptional capabilities suggest their potential as scientific research assistants, but also highlight the need for holistic, rigorous, and domain-specific evaluation to assess effectiveness in real-world scientific applications. This talk describes a multifaceted methodology for Evaluating AI models as scientific Research Assistants (EAIRA) developed at Argonne National Laboratory. This methodology incorporates four primary classes of evaluations. 1) Multiple Choice Questions to assess factual recall; 2) Open Response to evaluate advanced reasoning and problem-solving skills; 3) Lab-Style Experiments involving detailed analysis of capabilities as research assistants in controlled environments; and 4) Field-Style Experiments to capture researcher-LLM interactions at scale in a wide range of scientific domains and applications. These complementary methods enable a comprehensive analysis of LLM strengths and weaknesses with respect to their scientific knowledge, reasoning abilities, and adaptability. Recognizing the rapid pace of LLM advancements, we designed the methodology to evolve and adapt so as to ensure its continued relevance and applicability. This talk describes the current methodology's state. Although developed within a subset of scientific domains, the methodology is designed to be generalizable to a wide range of scientific domains.

The standard way to do this is to use the chain rule to backpropagate gradients through layers of neurons. I shall briefly review a few of the engineering successes of backpropagation and then describe a very different way of getting the gradients that, for a while, seemed a lot more plausible as a model of how the brain gets gradients. Consider a system composed of binary neurons that can be active or inactive with weighted pairwise couplings between pairs of neurons, including long range couplings. If the neurons represent pixels in a binary image, we can store a set of binary training images by adjusting the coupling weights so that the images are local minima of a Hopfield energy function which is minus the sum over all pairs of active neurons of their coupling weights. But this energy function can only capture pairwise correlations. It cannot represent the kinds of complicated higher-order correlations that occur in images. Now suppose that in addition to the "visible" neurons that represent the pixel intensities, we also have a large set of hidden neurons that have weighted couplings with each other and with the visible neurons. Suppose also that all of the neurons are asynchronous and stochastic: They adopt the active state with a log odds that is equal to the difference in the energy function when the neuron is inactive versus active. Given a set of training images, is there a simple way to set the weights on all of the couplings so that the training images are local minima of the free energy function obtained by integrating out the states of the hidden neurons? The Boltzmann machine learning algorithm solved this problem in an elegant way. It was proof of principle that learning in neural networks with hidden neurons was possible using only locally available information, contrary to what was generally believed at the time.

Building on recent advancements in understanding particle transport in magnetized media, we present a first-principles scaling law for the formation of non-thermal tails in particle spectra within mildly and strongly magnetized turbulent plasmas. This scaling is validated using results from kinetic Particle-In-Cell simulations, which show excellent agreement with our theoretical predictions. Finally, we discuss the astrophysical implications of these findings, particularly for the proton spectra in the coronae of supermassive black holes.

It has been proposed recently that the breaking of MHD waves in the inner magnetosphere of strongly magnetized neutron stars can power different types of high-energy transients. We have studied the steepening and dissipation of a strongly magnetized fast magnetosonic wave propagating in a declining background magnetic field, by means of particle-in-cell simulations that encompass MHD scales. Our analysis confirms the formation of a monster shock as $B^2-E^2$ approaches zero, that dissipates about half of the fast magnetosonic wave energy, and reveals, for the first time, the generation of a high-frequency precursor wave at the monster shock, carrying a fraction of 0.001 of the total energy dissipated at the shock. The spectrum of the precursor wave exhibits several sharp harmonic peaks, with frequencies in the GHz band under conditions anticipated in magnetars. Such signals may appear as fast radio bursts.

Neutron star magnetospheres are a source of abundant X-ray activity. They have transients observed in different bands, like the fast radio burst (FRB) and associated hard X-ray flare from the Galactic magnetar SGR 1935+2154. We present global models for magnetar X-ray emission, including a landmark first-principle radiative particle-in-cell simulation of the twisted magnetar magnetosphere with the GPU-PIC code Entity. In one scenario, plasma particles accelerated by surface-motion-induced discharges interact resonantly with thermal background photons. Our GPU-accelerated particle-in-cell simulations track up-scattered high-energy photons that drive secondary pair production and ignite a magnetospheric circuit that persistently generates X-rays. We divulge the plasma properties of such a magnetospheric circuit, including densities and velocities, and give an outlook on alternative ignition scenarios for persistent magnetar X-ray emission.

Black hole X-ray binaries (BHXRBs) and Active Galactic Nuclei (AGN) transition through a series of accretion states in a well-defined order. The accretion states, each associated with different luminosities, spectral and variability characteristics, quasi periodic oscillations (QPOs) and outflow properties, are thought to be triggered by physical changes in the accretion disk around the central black hole.The mechanisms behind state transitions, the geometry of transitional disks and the physical mechanisms driving the emission characteristics we observe remain highly debated. General relativistic magneto-hydrodynamic simulations (GRMHD) are increasingly providing crucial insights into the accretion process, the launch of outflows and the physical processes driving state transitions in BHXRBs and AGN. Using GRMHD simulations conducted with the H-AMR code I: 1) Discuss how high and low-frequency QPOs can be produced by a highly tilted, geometrically thin accretion disk. 2) Present the first GRMHD simulation showing the self-consistent formation of a truncated accretion disk– a proposed disk model for the hard intermediate accretion state, in which the accretion flow is thick and hot close to the black hole, while the outer regions of the flow are thin and cool. 3) Describe how QPOs can be generated at the truncation radius (the radius at which the disk transitions from thick to thin) in a truncated accretion disk.

Fast magnetic dissipation powers the bursting activity of isolated neutron stars, merging neutron stars, and flares from black holes. Four mechanisms of fast dissipation can operate in the magnetospheres of compact objects: Alfvenic turbulent cascade, magnetic reconnection, collision of Alfvén waves, and monsters shocks. Radiation produced by these dissipation modes is controlled by the compactness parameter. The main radiative output is usually in X-rays; in some cases, radio bursts are produced.

Radio pulsars display a range of puzzling long-term variability in their magnetospheric emission and spin-down. In this talk I will discuss the mechanisms for internal magnetic field evolution of pulsars, how it differs from the MHD of classical conducting fluids, and how the neutron star interior couples to the magnetosphere and its radiation.