Article
Chemistry, Physical
D. Kumar, S. Patinet, C. E. Maloney, I. Regev, D. Vandembroucq, M. Mungan
Summary: We developed a mesoscopic model to study the plastic behavior of an amorphous material under cyclic loading. By tuning the aging duration, different levels of glass stabilities can be achieved and the disorder-landscape can be characterized through the analysis of transition graphs.
JOURNAL OF CHEMICAL PHYSICS
(2022)
Article
Physics, Multidisciplinary
Samuel Albert, Giulio Biroli, Francois Ladieu, Roland Tourbot, Pierfrancesco Urbani
Summary: The study revealed the absence of a Gardner phase transition in glassy glycerol, suggesting that standard molecular glasses may suppress its existence. Instead, a specific localized excitation pattern was observed at low temperatures.
PHYSICAL REVIEW LETTERS
(2021)
Article
Physics, Multidisciplinary
Siheon Ryee, Myung Joon Han, Sangkook Choi
Summary: Motivated by the discovery of superconductivity in infinite-layer nickelates, this study investigates the role of Hund coupling J in a quarter-filled two-orbital Hubbard model. The results reveal distinctive regimes of correlated metals, with one related to Mott insulator proximity. Defines criteria for characterizing these metals, establishing the existence of Hund metallicity in two-orbital systems.
PHYSICAL REVIEW LETTERS
(2021)
Article
Chemistry, Physical
Swagata Acharya, Dimitar Pashov, Alexander N. Rudenko, Malte Rosner, Mark van Schilfgaarde, Mikhail Katsnelson
Summary: The passage discusses the use of embedding methods to handle strong electronic correlations in materials, measuring success by the quality of the self-energy sigma. It also highlights that factors such as choice of parameters, double-counting corrections, and the adequacy of the low-level Hamiltonian can hinder a clear understanding of these effects in some cases.
NPJ COMPUTATIONAL MATERIALS
(2021)
Article
Multidisciplinary Sciences
Kumpei Shiraishi, Hideyuki Mizuno, Atsushi Ikeda
Summary: Supercooled liquids with complicated structural relaxation processes have been a long-standing problem in condensed matter physics. Previous experiments observed that relaxation dynamics in many molecular liquids separate into two distinct processes at low temperatures. This study uses molecular dynamics simulations to investigate the potential energy landscape and provides the first direct evidence of the topographic hierarchy that induces relaxation. The results contribute to a fundamental and comprehensive understanding of relaxation dynamics in supercooled liquids.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2023)
Article
Chemistry, Physical
Boris N. Solomonov, Mikhail I. Yagofarov
Summary: The study examines the relationship between Gibbs energies and enthalpies of solvation and vaporization in various classes of organic non-electrolytes, and finds that most organic non-electrolytes follow a linear correlation, while long-chain aliphatics have a lower slope.
JOURNAL OF MOLECULAR LIQUIDS
(2022)
Article
Quantum Science & Technology
Hamza Fawzi, Omar Fawzi, Samuel O. Scalet
Summary: We present a classical algorithm for approximating the free energy of one-dimensional quantum systems with local, translation-invariant properties in the thermodynamic limit. While solving the ground state problem for these systems is computationally difficult, our algorithm provides a subpolynomial time solution for fixed temperatures above absolute zero. The algorithm computes the spectral radius of a linear map, which is interpreted as a noncommutative transfer matrix and has been studied in relation to the analyticity of the free energy and correlation decay. The corresponding eigenvector of this map allows for the computation of various thermodynamic properties of the quantum system.
Article
Quantum Science & Technology
Raffaele Salvia, Vittorio Giovannetti
Summary: This research proves that, under certain conditions, the energy distribution of states reached through cyclic processes closely resembles a Gaussian distribution with respect to the Haar measure. Limits are derived for the average energy of the state, indicating that the deviation from the normal distribution diminishes as the dimension of the system's Hilbert space increases.
Article
Multidisciplinary Sciences
Peter K. Morse, Sudeshna Roy, Elisabeth Agoritsas, Ethan Stanifer, Eric Corwin, M. Lisa Manning
Summary: The similarities in mechanical properties between dense active matter and sheared amorphous solids have been investigated through a mean-field model, showing equivalent critical behavior in infinite dimensions. Numerical tests in two dimensions confirm the accuracy of these predictions, suggesting a universal framework for predicting flow, deformation, and failure in active and sheared disordered materials.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2021)
Article
Materials Science, Multidisciplinary
K. G. Wang
Summary: The kinetics of phase coarsening in a dense binary, two-phase system were theoretically studied and existing relations were recovered and generalized. Equations for particle size distribution and coarsening were rigorously derived. An interesting finding is that the scaling exponent, m, for the kinetics of phase coarsening at ultra high volume fractions takes values in the range 2 < m < 3, depending on the precise volume fraction of the dispersed phase, when varied over the narrow range 0.9 < V-V < 1. The particle size distributions derived in this study depend on volume fractions, which is different from Wagner's particle size distribution for interface-reaction-controlled phase coarsening. The current work substantiates that the kinetics of phase coarsening at ultra high volume fractions exhibits a blend of both interface-reaction-controlled and volume diffusion-controlled phase coarsening.
Article
Materials Science, Multidisciplinary
Tommaso Rizzo
Summary: The article discusses computating the exponentially small probability of a system jumping from one metastable state to another, focusing on the evaluation of path integrals using mean-field models and saddle-point methods, and solving the resulting dynamical equations with numerical algorithms.
Article
Optics
Yue-Xun Huang, Ming Li, Zi-Jie Chen, Yan-Lei Zhang, Xu-Bo Zou, Guang-Can Guo, Chang-Ling Zou
Summary: Mean-field treatment (MFT) is commonly used for approximating the dynamics of quantum optics systems. However, neglecting quantum correlations between modes can lead to unexpected quantum effects. This study presents a theoretical framework based on perturbation theory and MFT to capture these effects and predicts the form and relationship of nonlinear dissipation, parasitic Hamiltonian, and nonlinear coupling rate. The framework is applied to quantum frequency conversion and shows excellent agreement with numerical simulations, revealing the neglected quantum effects by MFT and providing a more precise framework for nonlinear and quantum optics.
LASER & PHOTONICS REVIEWS
(2023)
Article
Materials Science, Multidisciplinary
Julian Thoenniss, Michael Sonner, Alessio Lerose, Dmitry A. Abanin
Summary: We propose an efficient method for simulating the dynamics of an interacting quantum impurity coupled to noninteracting fermionic reservoirs. By treating the impurity as an open quantum system, we describe the reservoirs using Feynman-Vernon influence functionals (IFs) represented as matrix-product states in the temporal domain. The method demonstrates favorable performance in studying quantum quenches and transport in an Anderson impurity model, including highly nonequilibrium setups, compared to existing methods. The computational resources needed to accurately compute the dynamics scale polynomially with evolution time, indicating efficient solvability of a broad range of out-of-equilibrium quantum impurity problems. This approach will offer additional insights into the dynamical properties of mesoscopic devices and correlated materials.
Article
Chemistry, Multidisciplinary
Guangming Wang, Xuefeng Chen, Xun Li, Ying Zeng, Kaka Zhang
Summary: This study reports the emergence of a high-performance organic afterglow in pyrylium induced photopolymerization systems, as well as the mechanism landscape of the afterglow systems as a function of monomer types. Methyl methacrylate exhibits a TADF-type organic afterglow with an afterglow efficiency of 70.4% after pyrylium-catalyzed photopolymerization. Using heavy-atom-containing methacrylate, the external heavy atom effect speeds up phosphorescence decay and enables room-temperature phosphorescence in pyrylium-polymer systems.
Article
Physics, Multidisciplinary
M. Schmidt, P. F. Dias
Summary: The correlated cluster mean-field (CCMF) theory is an approximate method that has been applied to study spin-1/2 Hamiltonians and extended to Ising-like systems with spin S > 1/2. Research shows that the CCMF method results on honeycomb, square, and simple cubic lattices can be compared to state-of-the-art methods, and applications for higher spin and mixed-spin systems on the honeycomb lattice have been compared with other techniques. Results indicate that the reduced critical temperature obtained within the CCMF theory overestimates by only 5% the exact result for the mixed spin-(1, 1/2) system.
PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS
(2021)
Article
Physics, Multidisciplinary
Marylou Gabrie, Jean Barbier, Florent Krzakala, Lenka Zdeborova
JOURNAL OF PHYSICS A-MATHEMATICAL AND THEORETICAL
(2020)
Article
Physics, Multidisciplinary
Stefano Sarao Mannelli, Giulio Biroli, Chiara Cammarota, Florent Krzakala, Pierfrancesco Urbani, Lenka Zdeborova
Article
Computer Science, Information Systems
Jean Barbier, Nicolas Macris, Mohamad Dia, Florent Krzakala
IEEE TRANSACTIONS ON INFORMATION THEORY
(2020)
Article
Computer Science, Information Systems
Benjamin Aubin, Bruno Loureiro, Antoine Maillard, Florent Krzakala, Lenka Zdeborova
Summary: Investigated the statistical and algorithmic properties of random neural-network generative priors in spiked-matrix estimation, establishing the performance of Bayesian optimal estimator and identifying statistical threshold for weak-recovery of spike; derived a message-passing algorithm considering latent structure of spike, showing asymptotically optimal performance for natural generative network choices, highlighting absence of algorithmic gap compared to sparse spikes.
IEEE TRANSACTIONS ON INFORMATION THEORY
(2021)
Article
Multidisciplinary Sciences
Antoine Baker, Indaco Biazzo, Alfredo Braunstein, Giovanni Catania, Luca Dall'Asta, Alessandro Ingrosso, Florent Krzakala, Fabio Mazza, Marc Mezard, Anna Paola Muntoni, Maria Refinetti, Stefano Sarao Mannelli, Lenka Zdeborova
Summary: Research suggests that probabilistic risk estimation can enhance the performance of digital contact tracing, aiding in mitigating the impact of epidemics.
PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA
(2021)
Article
Mechanics
Francesca Mignacco, Florent Krzakala, Pierfrancesco Urbani, And Lenka Zdeborova
Summary: This study analyzes the learning dynamics of stochastic gradient descent in a high-dimensional Gaussian mixture classification problem, revealing how the algorithm's performance varies with changes in control parameters in the loss landscape.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2021)
Article
Mechanics
Federica Gerace, Bruno Loureiro, Florent Krzakala, Marc Mezard, Lenka Zdeborova
Summary: In this study, we focus on generalised linear regression and classification for a synthetically generated dataset, presenting closed-form expressions for asymptotic generalisation performance using the replica method from statistical physics. We highlight the double descent behavior in logistic regression and the superiority of orthogonal projections in learning with random features, while considering the role of correlations in data generated by the hidden manifold model. This theoretical formalism not only addresses specific problems but also opens a pathway for extending to more complex tasks.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2021)
Article
Mechanics
Antoine Maillard, Florent Krzakala, Marc Mezard, Lenka Zdeborova
Summary: The paper discusses matrix factorization and extensive-rank matrix denoising problems using high-temperature expansions to find more accurate solutions. It provides a systematic approach to derive corrections to existing approximations, taking into account the specific structure of correlations in the problems.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2022)
Article
Mechanics
Hugo Cui, Bruno Loureiro, Florent Krzakala, Lenka Zdeborova
Summary: This manuscript investigates kernel ridge regression (KRR) under the Gaussian design and explores the impact of the interplay between noise and regularization on the decay rates of excess generalization error. By studying different settings, we provide a characterization of all observed regimes and demonstrate the existence of a transition phenomenon in the noisy setting.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2022)
Article
Mechanics
Bruno Loureiro, Cedric Gerbelot, Hugo Cui, Sebastian Goldt, Florent Krzakala, Marc Mezard, Lenka Zdeborova
Summary: Teacher-student models provide a framework for describing the performance of high-dimensional supervised learning. This paper introduces a Gaussian covariate generalisation of the model that captures learning curves for a broad range of realistic data sets. The study also discusses the power and limitations of the framework.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2022)
Article
Biochemical Research Methods
Sebastian Goldt, Florent Krzakala, Lenka Zdeborova, Nicolas Brunel
Summary: The advent of comprehensive synaptic wiring diagrams of large neural circuits has created the field of connectomics. This study addresses the question of whether it is possible to reconstruct the information stored in a recurrent network of neurons given its synaptic connectivity matrix. It provides a practical algorithm based on statistical physics for approximate Bayesian inference to solve this inference problem.
PLOS COMPUTATIONAL BIOLOGY
(2023)
Article
Computer Science, Information Systems
Cedric Gerbelot, Alia Abbara, Florent Krzakala
Summary: There has been a recent surge of interest in studying the asymptotic reconstruction performance of generalized linear estimation problems, especially for the case of i.i.d standard normal matrices in the teacher-student setting. In this study, an analytical formula for the reconstruction performance of convex generalized linear models with rotationally-invariant data matrices is proven, confirming a conjecture derived using the replica method. The proof leverages on message passing algorithms and statistical properties of their iterates, characterizing the asymptotic empirical distribution of the estimator.
IEEE TRANSACTIONS ON INFORMATION THEORY
(2023)
Article
Computer Science, Artificial Intelligence
Lucas Clarte, Bruno Loureiro, Florent Krzakala, Lenka Zdeborova
Summary: Being able to assess the accuracy and uncertainty of models' predictions is important in machine learning. Computational challenges arise in high-dimensional problems when sampling the posterior probability measure. This manuscript characterizes uncertainty for learning from limited samples and provides a formula for investigating the calibration of the logistic classifier.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2023)
Proceedings Paper
Acoustics
Alessandro Cappelli, Ruben Ohana, Julien Launay, Laurent Meunier, Iacopo Poli, Florent Krzakala
Summary: We propose a new defense mechanism inspired by an optical co-processor that provides robustness against adversarial attacks without compromising natural accuracy in both whitebox and black-box settings. This hardware co-processor performs a nonlinear fixed random transformation with unknown parameters that cannot be retrieved with sufficient precision. In the whitebox setting, our defense works by obfuscating the parameters of the random projection, but we find it challenging to build a reliable backward differentiable approximation for obfuscated parameters.
2022 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Maria Refinetti, Sebastian Goldt, Florent Krzakala, Lenka Zdeborova
Summary: Theoretical works indicate that two-layer neural networks with few neurons can outperform kernel learning on simple classification tasks, especially in high-dimensional limits. Small neural networks can achieve near-optimal performance, while lazy training methods like random features and kernel methods do not. Over-parameterizing neural networks can lead to faster convergence but does not necessarily improve final performance.
INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 139
(2021)