Search

We report on a comprehensive thermodynamic study of a quasi-two-dimensional (quasi-2D) quantum magnet Cu$_2$(OH)$_3$Br which in the 2D layer can be viewed as strongly coupled alternating antiferromagnetic and ferromagnetic chains. In an applied magnetic field transverse to the ordered spins below $T_N=9.3$ K, a field-induced phase transition from the 3D ordered to a disordered phase occurs at $B_c=16.3$ T for the lowest temperature, which is featured by an onset of a one-half plateau-like magnetization. By performing quantum Monte Carlo simulations of the relevant 2D model, we find that the plateau-like magnetization corresponds to a partial symmetry restoration and the full polarization in the ferromagnetic chains. Our numerical simulations also show that the magnetization saturation occurs with full symmetry restoration at a much higher field of $B_s \simeq 95$ T, corresponding to a 1D quantum phase transition in the antiferromagnetic chains. We argue that the experimentally observed field-induced phase transition at $B_c$ follows from the partial symmetry restoration and the concomitant dimensional reduction., Comment: 6 pages, 4 figures

We study domain walls (DWs) arising in field theories where $Z_2$-symmetry is spontaneously broken by a scalar expectation value decreasing proportionally to the Universe temperature. The energy density of such melting DWs redshifts sufficiently fast not to overclose the Universe. For the first time, evolution of melting DWs and the resulting gravitational waves (GWs) is investigated numerically using lattice simulations. We show that formation of closed melting DWs during radiation domination is much more efficient compared to the scenario with constant tension DWs. This suggests that it can be the main mechanism responsible for reaching the scaling regime similarly to the case of cosmic strings. However, the scaling behaviour of melting DWs is observed, provided only that the initial scalar field fluctuations are not very large. Otherwise, simulations reveal violation of the scaling law, potentially of the non-physical origin. The spectrum of GWs emitted by melting DWs is also significantly different from that of constant tension DWs. Whether the system has reached scaling or not, the numerical study reveals a GW spectrum described in the infrared by the spectral index $n \approx 1.6$ followed by the causality tail. We attribute the difference from the value $n=2$ predicted in our previous studies to a finite lifetime of the DW network. Notably, the updated index is still in excellent agreement with the recent findings by pulsar timing arrays, which confirms that melting DWs can be responsible for the observed (GW) signal. We also point out that results for evolution of melting DWs in the radiation-dominated Universe are applicable to constant tension DW evolution in the flat spacetime., Comment: 38 pages, 9 figures

We provide the first proof of convergence for normalized error feedback algorithms across a wide range of machine learning problems. Despite their popularity and efficiency in training deep neural networks, traditional analyses of error feedback algorithms rely on the smoothness assumption that does not capture the properties of objective functions in these problems. Rather, these problems have recently been shown to satisfy generalized smoothness assumptions, and the theoretical understanding of error feedback algorithms under these assumptions remains largely unexplored. Moreover, to the best of our knowledge, all existing analyses under generalized smoothness either i) focus on single-node settings or ii) make unrealistically strong assumptions for distributed settings, such as requiring data heterogeneity, and almost surely bounded stochastic gradient noise variance. In this paper, we propose distributed error feedback algorithms that utilize normalization to achieve the $O(1/\sqrt{K})$ convergence rate for nonconvex problems under generalized smoothness. Our analyses apply for distributed settings without data heterogeneity conditions, and enable stepsize tuning that is independent of problem parameters. Additionally, we provide strong convergence guarantees of normalized error feedback algorithms for stochastic settings. Finally, we show that due to their larger allowable stepsizes, our new normalized error feedback algorithms outperform their non-normalized counterparts on various tasks, including the minimization of polynomial functions, logistic regression, and ResNet-20 training.

We present the largest 3D Particle-in-Cell shearing-box simulations of turbulence driven by the magnetorotational instability, for the first time employing the realistic proton-to-electron mass ratio. We investigate the energy partition between relativistically hot electrons and subrelativistic ions in turbulent accretion flows, a regime relevant to collisionless, radiatively inefficient accretion flows of supermassive black holes, such as those targeted by the Event Horizon Telescope. We provide a simple empirical formula to describe the measured heating ratio between ions and electrons, which can be used for more accurate global modeling of accretion flows with standard fluid approaches such as general-relativistic magnetohydrodynamics.

We report an extensive high-field study of atacamite Cu$_2$Cl(OH)$_3$, a material realization of quantum sawtooth chains with weak interchain couplings, in continuous and pulsed magnetic fields up to 58 T. In particular, we have mapped the entropy landscape for fields as high as 35 T and have identified a field-induced quantum critical point at 21.9(1) T for $\mathbf{H} \parallel c$ axis. The quantum critical point separates field regions with and without magnetic order, evidenced by our thermodynamic study and $^1$H nuclear magnetic resonance spectroscopy, but lies far below full saturation of the magnetization. Corroborated by numerical results using density-matrix renormalization group (DMRG) calculations, we find this behavior associated with a decoupling of the sawtooth chain into two magnetic subsystems consisting of a spin-$1/2$ antiferromagnetic Heisenberg chain (backbone of the sawtooth chain) in the presence of an exchange field produced by the remaining field-polarized spins.

In this study, we probe the coupling between magnetism and lattice dynamics in UO2, a 3k antiferromagnet that undergoes magnetic ordering below its Neel temperature (TN) of 30.8 K. Ultrasound measurements provide insights into the interplay between the material's magnetic properties and lattice vibrations in response to the applied high magnetic field. A model analysis based on ab initio calculated superexchange interactions predicts a change in the magnetic structure from 3k to 2k at around 50 T. Although this change is not evident in the magnetization curve, we observe a crossover of ultrasound velocity Dv/v using a phase-sensitive pulse-echo technique in pulsed magnetic fields of up to 65 T. We show that at Hz > 50 T, the structure remains antiferromagnetic in the (x, y) plane and becomes ferromagnetic in the z direction. A further transition into the 1k structure is theoretically predicted to take place at a magnetic field of ~104 tesla.

Due to the non-smoothness of optimization problems in Machine Learning, generalized smoothness assumptions have been gaining a lot of attention in recent years. One of the most popular assumptions of this type is $(L_0,L_1)$-smoothness (Zhang et al., 2020). In this paper, we focus on the class of (strongly) convex $(L_0,L_1)$-smooth functions and derive new convergence guarantees for several existing methods. In particular, we derive improved convergence rates for Gradient Descent with (Smoothed) Gradient Clipping and for Gradient Descent with Polyak Stepsizes. In contrast to the existing results, our rates do not rely on the standard smoothness assumption and do not suffer from the exponential dependency from the initial distance to the solution. We also extend these results to the stochastic case under the over-parameterization assumption, propose a new accelerated method for convex $(L_0,L_1)$-smooth optimization, and derive new convergence rates for Adaptive Gradient Descent (Malitsky and Mishchenko, 2020)., Comment: 51 pages, 1 figure

Controlling the spectral norm of the Jacobian matrix, which is related to the convolution operation, has been shown to improve generalization, training stability and robustness in CNNs. Existing methods for computing the norm either tend to overestimate it or their performance may deteriorate quickly with increasing the input and kernel sizes. In this paper, we demonstrate that the tensor version of the spectral norm of a four-dimensional convolution kernel, up to a constant factor, serves as an upper bound for the spectral norm of the Jacobian matrix associated with the convolution operation. This new upper bound is independent of the input image resolution, differentiable and can be efficiently calculated during training. Through experiments, we demonstrate how this new bound can be used to improve the performance of convolutional architectures., Comment: ECCV 2024

We study the production of hypothetical vector particles, dark photons $\gamma^\prime$, with masses in the the range 0.4--1.8 GeV via inelastic proton bremsstrahlung. We further develop the approach of Refs. [1-3], where for the first time we considered the contributions to the cross section that are associated with the Pauli form factor in $pp\gamma^\prime$ vertex and obtained new splitting functions. We demonstrate numerically the importance of these corrections to full inelastic proton bremsstrahlung cross section and refine the sensitivity of the ongoing and future fixed target experiments T2K, DUNE and SHiP to the parameters of dark photon model. A dedicated experiment on measurements of the proton electromagnetic form factors in the time-like region below the proton-antiproton threshold, like those suggested in PANDA at FAIR, would help to obtain the robust predictions for the dark photon production by a proton beam at fixed target., Comment: 26 pages, 4 figures, 2 tables

We study the production of vector portal mediators, dark photons, with masses in the range 0.4--1.8\,GeV in proton-proton collisions via the process of inelastic proton bremsstrahlung. In contrast to previous studies, we take into account the contribution of Pauli electromagnetic form factor to differential cross section and introduce two new splitting functions of the proton. We show that their contributions can become leading in the mass region 0.9--1.8\,GeV and present the updated estimate for the sensitivity of the future SHiP experiment to visible decays of dark photons., Comment: 10 pages, 3 figures, contribution to the Scientific Session of the Physical Sciences Division Nuclear Physics Section of the Russian Academy of Sciences, 1-5 April 2024, Dubna, Russia, accepted for publication in Phys.Part.Nucl. 56 (2025)

In this paper we present new constraints on velocity-independent cross section of keV-scale mass annihilating Dark Matter particles obtained with SRG/ART-XC after 4 full-sky surveys. These constraints are derived from observations of the Milky Way Halo, 33 Local Group spheroidal dwarf (dSph) galaxies and separately for the dSph galaxy Ursa Major III/UNIONS 1. The constraints from the Milky Way Halo are the strongest among others and among all available in literature for this class of Dark Matter models with particle masses from 4 to 15 keV., Comment: 7 pages, 8 figures

Employing the publicly available CosmoLattice code, we conduct numerical simulations of a domain wall network and the resulting gravitational waves (GWs) in a radiation-dominated Universe in the $Z_2$-symmetric scalar field model. In particular, the domain wall evolution is investigated in detail both before and after reaching the scaling regime, using the combination of numerical and theoretical methods. We demonstrate that the total area of closed walls is negligible compared to that of a single long wall stretching throughout the simulation box. Therefore, the closed walls are unlikely to have a significant impact on the overall network evolution. This is in contrast with the case of cosmic strings, where formation of loops is crucial for maintaining the system in the scaling regime. To obtain the GW spectrum, we develop a technique that separates physical effects from numerical artefacts arising due to finite box size and non-zero lattice spacing. Our results on the GW spectrum agree well with Refs. [29, 30], which use different codes. Notably, we observe a peak at the Hubble scale, an exponential falloff at scales shorter than the wall width, and a plateau/bump at intermediate scales. We also study sensitivity of obtained results on the choice of initial conditions. We find that different types of initial conditions lead to qualitatively similar domain wall evolution in the scaling regime, but with important variations translating into different intensities of GWs., Comment: 32 pages, 9 figures; typos fixed, gravitational wave spectral indices are slightly updated; matches version accepted in JCAP

We present a new approach based on the Personalized Federated Learning algorithm MeritFed that can be applied to Natural Language Tasks with heterogeneous data. We evaluate it on the Low-Resource Machine Translation task, using the dataset from the Large-Scale Multilingual Machine Translation Shared Task (Small Track #2) and the subset of Sami languages from the multilingual benchmark for Finno-Ugric languages. In addition to its effectiveness, MeritFed is also highly interpretable, as it can be applied to track the impact of each language used for training. Our analysis reveals that target dataset size affects weight distribution across auxiliary languages, that unrelated languages do not interfere with the training, and auxiliary optimizer parameters have minimal impact. Our approach is easy to apply with a few lines of code, and we provide scripts for reproducing the experiments at https://github.com/VityaVitalich/MeritFed, Comment: 18 pages, 7 figures

The increasing size of neural networks has led to a growing demand for methods of efficient fine-tuning. Recently, an orthogonal fine-tuning paradigm was introduced that uses orthogonal matrices for adapting the weights of a pretrained model. In this paper, we introduce a new class of structured matrices, which unifies and generalizes structured classes from previous works. We examine properties of this class and build a structured orthogonal parametrization upon it. We then use this parametrization to modify the orthogonal fine-tuning framework, improving parameter and computational efficiency. We empirically validate our method on different domains, including adapting of text-to-image diffusion models and downstream task fine-tuning in language modeling. Additionally, we adapt our construction for orthogonal convolutions and conduct experiments with 1-Lipschitz neural networks.

At ambient conditions, SiC is known to be resistant to irradiation with swift heavy ions (SHI) decelerating in the electronic stopping regime. However, there is no experimental data on the SiC irradiation at elevated temperatures. To investigate this problem, we evaluate the stability of SiC to SHI impacts at high temperatures up to 2200 K. We apply the combination of the Monte-Carlo code TREKIS-3, describing excitation of the electronic and atomic systems using temperature-dependent scattering cross-sections, with molecular-dynamic modeling of the lattice response to the excitation. We demonstrate that increasing irradiation temperature increases the energy transferred to the atomic lattice from the excited electronic system. This material heating leads to formation of a stable nanometric damaged core along the trajectory of 710 MeV Bi ion when the irradiation temperature overcomes the threshold of ~1800 K. In this case, a chain of nanometric voids along the ion trajectory forms due to the mass transport from the track core by edge dislocations. Voids of larger sizes appear at higher irradiation temperatures. At lower irradiation temperatures, the damaged regions recrystallize completely within ~100 ps after the ion passage.

Methods with adaptive stepsizes, such as AdaGrad and Adam, are essential for training modern Deep Learning models, especially Large Language Models. Typically, the noise in the stochastic gradients is heavy-tailed for the later ones. Gradient clipping provably helps to achieve good high-probability convergence for such noises. However, despite the similarity between AdaGrad/Adam and Clip-SGD, the high-probability convergence of AdaGrad/Adam has not been studied in this case. In this work, we prove that AdaGrad (and its delayed version) can have provably bad high-probability convergence if the noise is heavy-tailed. To fix this issue, we propose a new version of AdaGrad called Clip-RAdaGradD (Clipped Reweighted AdaGrad with Delay) and prove its high-probability convergence bounds with polylogarithmic dependence on the confidence level for smooth convex/non-convex stochastic optimization with heavy-tailed noise. Our empirical evaluations, including NLP model fine-tuning, highlight the superiority of clipped versions of AdaGrad/Adam in handling the heavy-tailed noise., Comment: 37 pages, 8 figures

Using tremendous photon statistics gained with the stray light aperture of the NuSTAR telescope over 11 years of operation, we set strong limits on the emission of close to monochromatic photons from the radiative decays of putative dark matter sterile neutrinos in the Milky Way. In the energy range of 3-20 keV covered by the NuSTAR, the obtained limits reach the edge of theoretical predictions of realistic models leaving only a small room left to explore., Comment: 11 pages, 11 figures, submitted to Phys. Rev. Lett.; Comments and suggestions are welcome

Variational inequalities represent a broad class of problems, including minimization and min-max problems, commonly found in machine learning. Existing second-order and high-order methods for variational inequalities require precise computation of derivatives, often resulting in prohibitively high iteration costs. In this work, we study the impact of Jacobian inaccuracy on second-order methods. For the smooth and monotone case, we establish a lower bound with explicit dependence on the level of Jacobian inaccuracy and propose an optimal algorithm for this key setting. When derivatives are exact, our method converges at the same rate as exact optimal second-order methods. To reduce the cost of solving the auxiliary problem, which arises in all high-order methods with global convergence, we introduce several Quasi-Newton approximations. Our method with Quasi-Newton updates achieves a global sublinear convergence rate. We extend our approach with a tensor generalization for inexact high-order derivatives and support the theory with experiments.

We consider a family of linear operators, diagonalized by the Hankel transform. The Fredholm determinants of these operators, restricted to $L_2[0, R]$, are expressed in a convenient form for asymptotic analysis as $R\to\infty$. The result is an identity, in which the determinant is equal to the leading asymptotic multiplied by an asymptotically small factor, for which an explicit formula is derived. We apply the result to the determinantal point process with the Bessel kernel, calculating the speed of the convergence of additive functionals with respect to the Kolmogorov-Smirnov metric., Comment: 24 pages

Adaptive methods are extremely popular in machine learning as they make learning rate tuning less expensive. This paper introduces a novel optimization algorithm named KATE, which presents a scale-invariant adaptation of the well-known AdaGrad algorithm. We prove the scale-invariance of KATE for the case of Generalized Linear Models. Moreover, for general smooth non-convex problems, we establish a convergence rate of $O \left(\frac{\log T}{\sqrt{T}} \right)$ for KATE, matching the best-known ones for AdaGrad and Adam. We also compare KATE to other state-of-the-art adaptive algorithms Adam and AdaGrad in numerical experiments with different problems, including complex machine learning tasks like image classification and text classification on real data. The results indicate that KATE consistently outperforms AdaGrad and matches/surpasses the performance of Adam in all considered scenarios., Comment: 27 pages, 12 figures

High-energy neutrinos could be produced in the interaction of charged cosmic rays with matter or radiation surrounding astrophysical sources. To look for transient sources associated with neutrino emission, a follow-up program of neutrino alerts has been operating within the ANTARES Collaboration since 2009. This program, named TAToO, has triggered robotic optical telescopes (MASTER, TAROT, ROTSE and the SVOM ground based telescopes) immediately after the detection of any relevant neutrino candidate and scheduled several observations in the weeks following the detection. A subset of ANTARES events with highest probabilities of being of cosmic origin has also been followed by the Swift and the INTEGRAL satellites, the Murchison Widefield Array radio telescope and the H.E.S.S. high-energy gamma-ray telescope. The results of twelve years of observations are reported. No optical counterpart has been significantly associated with an ANTARES candidate neutrino signal during image analysis. Constraints on transient neutrino emission have been set. In September 2015, ANTARES issued a neutrino alert and during the follow-up, a potential transient counterpart was identified by Swift and MASTER. A multi-wavelength follow-up campaign has allowed to identify the nature of this source and has proven its fortuitous association with the neutrino. The return of experience is particularly important for the design of the alert system of KM3NeT, the next generation neutrino telescope in the Mediterranean Sea., Comment: 27 pages, 14 figures, submitted to JCAP

Oscillations of conductance observed in strong magnetic fields are a striking manifestation of the quantum dynamics of charge carriers in solids. The large charge carrier density in typical metals sets the scale of oscillations in both electrical and thermal conductivity, which characterize the Fermi surface. In semimetals, thermal transport at low-charge carrier density is expected to be phonon dominated, yet several experiments observe giant quantum oscillations in thermal transport. This raises the question of whether there is an overarching mechanism leading to sizable oscillations that survives in phonon-dominated semimetals. In this work, we show that such a mechanism exists. It relies on the peculiar phase-space allowed for phonon scattering by electrons when only a few Landau levels are filled. Our measurements on the Dirac semimetal ZrTe5 support this counter-intuitive mechanism through observation of pronounced thermal quantum oscillations, since they occur in similar magnitude and phase in directions parallel and transverse to the magnetic field. Our phase-space argument applies to all low-density semimetals, topological or not, including graphene and bismuth. Our work illustrates that phonon absorption can be leveraged to reveal degrees of freedom through their imprint on longitudinal thermal transport.

In Federated Learning (FL), the distributed nature and heterogeneity of client data present both opportunities and challenges. While collaboration among clients can significantly enhance the learning process, not all collaborations are beneficial; some may even be detrimental. In this study, we introduce a novel algorithm that assigns adaptive aggregation weights to clients participating in FL training, identifying those with data distributions most conducive to a specific learning objective. We demonstrate that our aggregation method converges no worse than the method that aggregates only the updates received from clients with the same data distribution. Furthermore, empirical evaluations consistently reveal that collaborations guided by our algorithm outperform traditional FL approaches. This underscores the critical role of judicious client selection and lays the foundation for more streamlined and effective FL implementations in the coming years.

In this paper, we consider non-smooth convex optimization with a zeroth-order oracle corrupted by symmetric stochastic noise. Unlike the existing high-probability results requiring the noise to have bounded $\kappa$-th moment with $\kappa \in (1,2]$, our results allow even heavier noise with any $\kappa > 0$, e.g., the noise distribution can have unbounded expectation. Our convergence rates match the best-known ones for the case of the bounded variance. To achieve this, we build the median gradient estimate with bounded second moment as the mini-batched median of the sampled gradient differences. We apply this technique to the stochastic multi-armed bandit problem with heavy-tailed distribution of rewards and achieve $\tilde{O}(\sqrt{dT})$ regret. We demonstrate the performance of our zeroth-order and MAB algorithms for different $\kappa$ on synthetic and real-world data. Our methods do not lose to SOTA approaches, moreover, they dramatically outperform SOTA for $\kappa \leq 1$.

We present first estimates of NICA sensitivity to Standard Model extensions with light hypothetical particles singlet under the known gauge transformations. Our analysis reveals that NICA can explore new regions in the parameter spaces of models with a hidden vector and models with an axion-like particle of masses about 30-500\,MeV. Some of these regions seem unreachable by other ongoing and approved future projects. NICA has good prospects in discovery ($5\sigma$) of the new physics after 1 year of data taking., Comment: 12 pages, 7 figures; replaced with the journal version

We present 3D fully kinetic shearing-box simulations of pair-plasma magnetorotational turbulence with unprecedented macro-to-microscopic scale separation. While retrieving the expected fluid behavior of the plasma at large scales, we observe a steepening of turbulent spectra at kinetic scales and substantial angular-momentum transport linked with kinetic processes. For the first time, we provide a definitive demonstration of nonthermal particle acceleration in kinetic magnetorotational turbulence agnostically of initial conditions, by means of a novel strategy exploiting synchrotron cooling., Comment: Accepted in PRL

Distributed learning has emerged as a leading paradigm for training large machine learning models. However, in real-world scenarios, participants may be unreliable or malicious, posing a significant challenge to the integrity and accuracy of the trained models. Byzantine fault tolerance mechanisms have been proposed to address these issues, but they often assume full participation from all clients, which is not always practical due to the unavailability of some clients or communication constraints. In our work, we propose the first distributed method with client sampling and provable tolerance to Byzantine workers. The key idea behind the developed method is the use of gradient clipping to control stochastic gradient differences in recursive variance reduction. This allows us to bound the potential harm caused by Byzantine workers, even during iterations when all sampled clients are Byzantine. Furthermore, we incorporate communication compression into the method to enhance communication efficiency. Under general assumptions, we prove convergence rates for the proposed method that match the existing state-of-the-art (SOTA) theoretical results. We also propose a heuristic on adjusting any Byzantine-robust method to a partial participation scenario via clipping., Comment: 52 pages; 4 figures. Changes in v2: a heuristic extension of the proposed method, new numerical results, a simpler presentation of the main results, and corrections of small typos

Robustness to Byzantine attacks is a necessity for various distributed training scenarios. When the training reduces to the process of solving a minimization problem, Byzantine robustness is relatively well-understood. However, other problem formulations, such as min-max problems or, more generally, variational inequalities, arise in many modern machine learning and, in particular, distributed learning tasks. These problems significantly differ from the standard minimization ones and, therefore, require separate consideration. Nevertheless, only one work (Adibi et al., 2022) addresses this important question in the context of Byzantine robustness. Our work makes a further step in this direction by providing several (provably) Byzantine-robust methods for distributed variational inequality, thoroughly studying their theoretical convergence, removing the limitations of the previous work, and providing numerical comparisons supporting the theoretical findings., Comment: NeurIPS 2023; 69 pages, 12 figures

We consider stochastic optimization problems with heavy-tailed noise with structured density. For such problems, we show that it is possible to get faster rates of convergence than $\mathcal{O}(K^{-2(\alpha - 1)/\alpha})$, when the stochastic gradients have finite moments of order $\alpha \in (1, 2]$. In particular, our analysis allows the noise norm to have an unbounded expectation. To achieve these results, we stabilize stochastic gradients, using smoothed medians of means. We prove that the resulting estimates have negligible bias and controllable variance. This allows us to carefully incorporate them into clipped-SGD and clipped-SSTM and derive new high-probability complexity bounds in the considered setup., Comment: AISTATS 2024. 60 pages, 3 figures. Changes in V2: small typos were fixed, extra experiments and discussion were added. Code: https://github.com/Kutuz4/AISTATS2024_SMoM

In this paper, we consider non-smooth stochastic convex optimization with two function evaluations per round under infinite noise variance. In the classical setting when noise has finite variance, an optimal algorithm, built upon the batched accelerated gradient method, was proposed in (Gasnikov et. al., 2022). This optimality is defined in terms of iteration and oracle complexity, as well as the maximal admissible level of adversarial noise. However, the assumption of finite variance is burdensome and it might not hold in many practical scenarios. To address this, we demonstrate how to adapt a refined clipped version of the accelerated gradient (Stochastic Similar Triangles) method from (Sadiev et al., 2023) for a two-point zero-order oracle. This adaptation entails extending the batching technique to accommodate infinite variance -- a non-trivial task that stands as a distinct contribution of this paper.