Computational Physics
See recent articles
Showing new listings for Tuesday, 17 November 2026
- [1] arXiv:2611.18863 [pdf, html, other]
-
Title: An Analytic Two-Dimensional Benchmark for Integrated Autocorrelation Times in Branching Monte Carlo Part I: Transport Spectra and Branching TheoryComments: 11 pages, 1 figure, 0 tables. Submitted to M&C 2027, the International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering (American Nuclear Society). Part I of two; the companion paper is Part II, submitted separatelySubjects: Computational Physics (physics.comp-ph); Computation (stat.CO)
Predicting correlations between source generations can support uncertainty estimation and variance reduction in Monte Carlo transport. We derive a two-dimensional benchmark for planar isotropic exponential flights in a reflecting square. An image construction gives the exact collision and absorption spectra. Critical branching preserves the mean transport operator but changes mode covariances through the offspring second moment. These covariances give tally autocorrelations and integrated autocorrelation times, with a large-population, finite-window approximation for normalized source tallies. Direct Monte Carlo measurements test the predictions for absorption/rebirth and two-offspring fission. The benchmark provides a reference for assessing reduced correlation models, for correcting reported tally uncertainties, and for checking the tallies of a production code for correlation bias.
- [2] arXiv:2611.18870 [pdf, html, other]
-
Title: An Analytic Two-Dimensional Benchmark for Integrated Autocorrelation Times in Branching Monte Carlo Part II: Operator Estimation from Functional Expansion TalliesComments: 11 pages, 2 figures, 2 tables. Submitted to M&C 2027, the International Conference on Mathematics and Computational Methods Applied to Nuclear Science and Engineering (American Nuclear Society). Part II of two; the companion paper is Part I, submitted separatelySubjects: Computational Physics (physics.comp-ph); Numerical Analysis (math.NA)
Functional expansion tallies can represent transport operators for Monte Carlo correlation prediction without first constructing a fine cell transition matrix. We tally Galerkin operators directly from birth-to-absorption histories, retaining the sampled departure measure through an empirical mass matrix. Operator moments and an estimated offspring second moment give branching autocorrelations and integrated autocorrelation times without an eigendecomposition. The analytic benchmark from Part I supplies continuum and noise-free finite-model references. At the compared sizes, cosine and Legendre models have smaller IACT errors and require less operator storage than the cell model, and the branching reweighting reduces their representation error by a further factor near five while leaving the cell error unchanged. The measured runtime benefit is basis dependent. Profiling separates shared transport from the additional costs of operator assembly and prediction.
- [3] arXiv:2611.18961 [pdf, html, other]
-
Title: Accelerating Atomistic-Continuum Simulations of Multilayer 2D Materials with Functional Principal Component Analysis-Based Machine LearningSubjects: Computational Physics (physics.comp-ph); Machine Learning (cs.LG); Mathematical Physics (math-ph)
Atomistic-continuum simulations of complex wrinkling deformations in large-area multilayer two-dimensional (2D) materials are computationally expensive. To accelerate these simulations, this work presents a machine-learning (ML) framework that first constructs a low-dimensional representation of the deformation field, then learns the deformation in this reduced space. Dimensionality reduction of deformed 2D membranes is challenging because of the extremely high dimensionality of discretized deformation fields, the need to preserve surface continuity and prescribed geometric constraints, the stochastic nature of wrinkling patterns, and the limited availability of training data. We address these challenges using two-dimensional functional principal component analysis (FPCA). The proposed FPCA representation reduces deformation fields containing millions of degrees of freedom to latent spaces with only tens of dimensions while preserving the continuity of the deformed surface. In addition, prescribed homogeneous boundary constraints are satisfied exactly through an appropriate choice of basis functions in the FPCA representation. A neural network is then trained to predict the principal components from the system and loading parameters. The resulting ML-predicted deformation fields are used as high-quality initial configurations for the atomistic-continuum solver, which subsequently performs energy minimization to obtain equilibrium solutions. The framework is demonstrated for multilayer graphene subjected to uniaxial compression, biaxial compression, and combined biaxial compression and simple shear. For the most computationally intensive large-strain cases, the proposed approach achieves speedups of at least 6.8 times while converging to final energies comparable to, and in some cases lower than, those obtained using conventional atomistic-continuum initialization.
- [4] arXiv:2611.19410 [pdf, html, other]
-
Title: An implicit high-order flux reconstruction method with logarithmic variables for hypersonic flows in thermochemical nonequilibriumRayan Dhib, Vatsalya Sharma, Vincent F. Giangaspero, Firas Ben Ameur, José Miguel Luzia Murteira, Stefaan Poedts, Andrea LaniSubjects: Computational Physics (physics.comp-ph); Numerical Analysis (math.NA); Fluid Dynamics (physics.flu-dyn)
Simulating hypersonic flows in thermochemical nonequilibrium requires solving the Navier-Stokes equations for reacting, possibly ionised mixtures. Compact high-order computations of viscous two-temperature air flows on unstructured grids have not gone beyond P = 2. We present an implicit Flux Reconstruction (FR) method for such flows, built on three ingredients. First, the solver stores and updates the logarithms of the partial densities and temperatures, making every state admissible without a positivity limiter and, for ionised mixtures, imposing charge neutrality consistently with the implicit update. Second, shocks are captured by subcell order blending, extended to steady implicit computations on curved Gauss-Legendre meshes, with second-order subcell blending in subsonic regions and an automatic freeze of the blending near convergence. Third, a pseudo-transient Newton method with a physicality-limited update and automatic CFL control reaches steady state from the free stream with the same settings at every order. The design order is verified up to sixth order on a manufactured solution for five-species air with finite-rate chemistry and vibrational relaxation. Steady viscous solutions up to sixth order are obtained for five- and eleven-species air on the High Enthalpy shock tunnel Göttingen (HEG) cylinder and for eleven-species air on the RAM-C II vehicle. To our knowledge, this work gives the first design-order verification of a high-order method with finite-rate air chemistry and vibrational relaxation, and the first high-order finite-element-type computations of viscous eleven-species, two-temperature air. On the HEG cylinder, the stagnation-point heat flux is within 2% of finite-volume references and within 6% of the measurements, with a comparable number of degrees of freedom to second-order finite-volume computations but on meshes with 20 to 40 times fewer elements.
- [5] arXiv:2611.20035 [pdf, html, other]
-
Title: U-HamGNN: a universal model for magnetic Hamiltonians conditioned on spin and Hubbard UComments: 25 pages, 5 figuresSubjects: Computational Physics (physics.comp-ph); Materials Science (cond-mat.mtrl-sci); Machine Learning (cs.LG)
The electronic structure of a magnetic crystal depends jointly on its spin configuration and the on-site Hubbard correction U, requiring a separate DFT+U calculation for each combination. We introduce U-HamGNN, an equivariant network that takes these two inputs directly and predicts spin-resolved Hamiltonians across diverse crystal chemistries. A spectral-hybrid optimizer that weights each update direction by how consistently the gradient points along it reaches 8% lower Hamiltonian error than Muon and 26% lower than AdamW in a ten-epoch benchmark. The predicted bands reproduce U-dependent gap opening and altermagnetic spin splitting within the trained window and beyond it; exchange couplings and magnetic-order calls derived from the Hamiltonian agree with DFT+U on a calibration set. This universal model provides a new tool for magnetic research.
- [6] arXiv:2611.20054 [pdf, html, other]
-
Title: Physics-Spawned Neural Integration of Rotating Shallow-Water DynamicsSubjects: Computational Physics (physics.comp-ph); Numerical Analysis (math.NA); Chaotic Dynamics (nlin.CD); Atmospheric and Oceanic Physics (physics.ao-ph); Fluid Dynamics (physics.flu-dyn)
Equation-derived neural integration can execute prescribed dynamics without training, while informative ensemble learning can estimate uncertain parameters and identify governing interactions without requiring differentiability. Together, these approaches motivate Physics-Spawned Informative Networks ($\Phi$SpIN), a framework for prescribed and adaptive hybrid dynamical systems. As a first step toward flood modeling applications, we present $\Phi$SpIN-SW, its prescribed-physics realization for the rotating shallow-water equations. Compatible C-grid operators on spherical Voronoi meshes define a computational graph that retains non-polynomial operations and advances the state through recurrent neural integration compiled with JAX. Analytical benchmarks characterize numerical errors, and global and coastal experiments closely reproduce reference solutions generated using the Model Prediction Across Scales (MPAS) model. At approximately 14~km global resolution, A100, H200 and B200 GPUs attain 9-, 19- and 34-fold higher timestep throughput than a 96-core MPAS-SW CPU reference under matched workloads. These results extend training-free neural integration beyond polynomial dynamics to complex geophysical simulation, establishing a computational foundation for subsequent adjoint-free ensemble estimation of physical and neural parameters and informative adaptation of operator structure.
- [7] arXiv:2611.20316 [pdf, html, other]
-
Title: Solving the Discretised Shallow Water Equations on Distorted Structured Grids with Neural PhysicsComments: 43 pages, 12 figures, 2 tablesSubjects: Computational Physics (physics.comp-ph); Computational Engineering, Finance, and Science (cs.CE); Fluid Dynamics (physics.flu-dyn)
This paper presents a method for solving the shallow water equations on highly distorted structured grids using convolution-based operators following the Neural Physics approach. By mapping distorted physical grids to a uniform computational domain with Jacobian transformations, the governing advection and diffusion operators are reformulated in a structured convolutional form. This enables the use of efficient stencil-based operations and, following the Neural Physics concept, we do this through convolutional layers whose weights are determined by the chosen numerical methods rather than by training.
The method introduces consistent mass-lumped formulations, conservative and non-conservative advection forms, and a diffusion operator, all expressed as convolutions. A nonlinear Petrov-Galerkin stabilisation is incorporated to control numerical oscillations on distorted meshes. Two approximations for the Laplacian operator are also proposed to accelerate pressure and diffusion solves. The framework is validated using the Carlisle 2005 flood event, demonstrating accurate predictions of water depth and velocity on distorted meshes for both linear and quadratic discretisations.
The results show that distorted structured grids can retain conservation, stability and accuracy, whilst adapting the mesh statically. The convolutional formulation makes the method well suited to Neural Physics implementations and GPU acceleration, offering a scalable approach for flood modelling and large-scale geophysical simulations. - [8] arXiv:2611.20426 [pdf, html, other]
-
Title: Efficient scheduling of particle annihilation for long-time Wigner simulationsComments: 20 pages, 3 figuresSubjects: Computational Physics (physics.comp-ph); Probability (math.PR)
Particle annihilation (PA) plays an indispensable role in alleviating the numerical sign problem in long-time simulations with the Wigner branching random walk (WBRW) method. However, the effect of PA frequency on long-time accuracy remains poorly understood, leaving its choice largely empirical. To provide systematic guidance for PA scheduling, we establish an overall error estimate for PA through a uniform mesh, separating WBRW variance, resampling variance and squared bias. The analysis shows that, beyond merely controlling particle growth, PA replaces the exponential-in-time variance bound with a linear-in-time bound, thereby alleviating the numerical sign problem. It also reveals bias-variance trade-off induced by PA frequency: frequent PA limits variance but can increase accumulated bias. More importantly, building on the newfound error structure, we propose a variance-constrained cost optimization strategy for adaptive PA scheduling across several PA methods. An intrinsic-time reparameterization provides a common clock for variance control, PA scheduling, and branching dynamics under time-dependent potentials. Numerical experiments validate the theoretical variance behavior and bias-variance trade-off, and results of our schedule populate the low-cost regime of the empirical Pareto frontier objecting to balance accuracy and efficiency.
- [9] arXiv:2611.20429 [pdf, html, other]
-
Title: FiuMD: Heterogeneous Batching for Coarse-Grained Molecular Dynamics with Neural Network PotentialsHugo D. Hernández R. (1 and 2), Paola B. Torres (3), Lara Negri (1 and 2), Rocío L. Cecchini (4), Nicolás A. García (1 and 2) ((1) Instituto de Física del Sur (IFISUR), CONICET, Bahía Blanca, Argentina, (2) Departamento de Física, Universidad Nacional del Sur (UNS), Bahía Blanca, Argentina, (3) Grupo Vinculado Bionanotecnología y Sistemas Complejos, INFAP (CONICET-UNSL-UTN), San Rafael, Mendoza, Argentina, (4) Instituto de Ciencias e Ingeniería de la Computación (ICIC), UNS-CONICET, Bahía Blanca, Argentina)Comments: 13 pages, 7 figures, 4 tablesSubjects: Computational Physics (physics.comp-ph); Soft Condensed Matter (cond-mat.soft)
FiuMD is a GPU molecular dynamics engine that advances many independent biomolecular simulations as a single batch, even when the systems differ in size, topology, force-field parameters, periodic box, timestep, or thermostat parameters. All systems are concatenated into global tensors, so the classical interaction terms and a shared TorchMD-Net neural network potential are evaluated for the whole batch at once, while each system keeps its own integration and thermostat. FiuMD started as a fork of TorchMD and reworks its system representation, force evaluation, and integration to handle such heterogeneous batches. We test it with coarse-grained C$\alpha$ models of chignolin, Trp-cage, BBL, and $\lambda$-repressor. In double precision, forces within a heterogeneous batch agree with isolated FiuMD and TorchMD evaluations to within the precision of repeated evaluations, with a maximum relative difference of $7\times10^{-16}$ across 60 systems. Over 1000 steps without a thermostat, batched trajectories deviate by at most $2.5\times10^{-11}$ Angstrom from isolated ones, even when periodic images lie within the neural network cutoff. With a different target temperature for each system of the ensemble, the mean and width of each system's kinetic temperature match the values expected from its own thermostat. We show that when individual simulations are too small to fill the GPU, batching can improve throughput several-fold over TorchMD. On a single GPU, batching four different protein systems achieves a 4.6-7.8-fold speedup per step compared with sequential TorchMD simulations. Compared with sequential homogeneous FiuMD batches, heterogeneous batching provides up to a 4-fold speedup for smaller batches while maintaining comparable performance for larger ones.
- [10] arXiv:2611.20551 [pdf, other]
-
Title: GranularGPT: Natural-Language-Driven End-to-End Discrete Element Method SimulationsComments: 21 pages, 9 figuresSubjects: Computational Physics (physics.comp-ph); Artificial Intelligence (cs.AI)
Discrete element method (DEM) simulations are widely used to investigate granular and particulate systems, but their application often requires specialized expertise in particle physics, numerical modelling, and simulation post-processing. Large language models (LLMs) offer new opportunities to simplify these workflows by enabling users to interact with simulation tools through natural language. This study offers GranularGPT, an LLM-assisted agentic framework for end-to-end DEM simulation workflows, from natural-language problem descriptions to simulation execution and automated post-processing. The framework enables users to generate DEM simulation cases, execute and monitor simulations, and automatically extract and report physical and process-level metrics. Evaluation across 60 normal-condition cases covering 12 DEM scenarios and parameter variations achieve a simulation success rate of 91.7%. In the error-handling evaluation, all 10 repairable errors are detected and 9 are successfully resolved, while all 5 non-repairable cases are correctly identified and reported. The framework automatically extracts a range of DEM-relevant metrics, including particle size distributions, packing and density measures, velocity profiles, discharge rates, mixing indices, contact force characteristics, and thermal quantities. GranularGPT provides a foundation for improving the accessibility and efficiency of DEM simulation workflows while reducing the specialized expertise required for their operation.
- [11] arXiv:2611.20802 [pdf, html, other]
-
Title: Parametrised-Quadratic Interface Reconstruction and Geometric Velocity Extension in the Volume-of-Fluid FrameworkSubjects: Computational Physics (physics.comp-ph)
A conservative Volume-of-Fluid (VOF) framework using local parametric-curve interface reconstruction (PI-VOF) is developed for geometric interface propagation on fixed Cartesian grids. While VOF methods achieve conservative and second-order accurate interface advection under prescribed velocity fields, normal-velocity-driven interface propagation requires the interfacial velocity to be extended to neighbouring staggered-grid locations.
Existing studies on interface evolution employ auxiliary distance functions, kernel smoothers, or high-resolution VOF in multiple grids to compute propagation velocities and address first-order-accuracy limitations. In contrast, PI-VOF directly employs higher-order parametric interface geometry for local normal-velocity extension, aiming to achieve second-order accurate interface evolution on fixed Cartesian grids. The resulting velocity field is employed in conservative geometric advection using a directionally split Eulerian-implicit and Lagrangian-explicit (EI-LE) scheme, while the reconstructed interface provides the geometric information required for curvature evaluation and the ghost-fluid finite-difference discretisation of sharp interfacial Dirichlet conditions. Static reconstruction and reversible-vortex advection tests demonstrate higher-order interface geometry and approximately second-order transport convergence, with mass conservation to the chosen tolerance on refined grids. Prescribed normal-velocity and curvature-dependent propagation tests establish near second-order convergence. Coupled Stefan benchmarks involving capillarity and attachment kinetics demonstrate approximately second-order thermal convergence and decreasing interface errors under mesh refinement. The results establish a geometric formulation connecting conservative VOF transport, local normal-velocity extension, and sharp Stefan conditions for evolving interfaces. - [12] arXiv:2611.21489 [pdf, html, other]
-
Title: Local topology hallucinations in universal machine-learned interatomic potentialsComments: 23 pages, 5 figures, 3 tables (includes Supplementary Information)Subjects: Computational Physics (physics.comp-ph)
Predicting metastable states with universal machine-learned interatomic potentials requires that they preserve the minima and saddles of the reference potential-energy surface. Here we identify local topology hallucination, in which distinct model minima relax into the same reference basin. This failure arises from mode-selective curvature-sign errors shared by MACE, CHGNet and M3GNet. Curvature signs predicted by two of the three models agreed with density-functional calculations in 55 of 57 positive but only 5 of 56 negative cases. In LiGe$_2$S, two MACE minima connected by a verified saddle relaxed into one density-functional basin. Reference relaxation likewise returned nine of ten selected model minima to their reference basins. A directional test with two displaced reference calculations rejected 9 of 12 prospectively tested model instabilities. Model agreement alone therefore cannot establish that predicted minima are distinct metastable states.
New submissions (showing 12 of 12 entries)
- [13] arXiv:2608.01360 (cross-list from math.NA) [pdf, html, other]
-
Title: Control variates with neural surrogates for uncertainty quantification in kinetic equationsSubjects: Numerical Analysis (math.NA); Computational Physics (physics.comp-ph); Plasma Physics (physics.plasm-ph)
Efficient uncertainty quantification for kinetic equations with random inputs is challenging because it requires repeated simulations of high-dimensional models, such as the Boltzmann, Landau, and related collisional equations, whose computational cost can quickly become prohibitive. Multifidelity control variates address this difficulty by coupling a small number of high-fidelity simulations with many evaluations of lower-complexity reduced models. In this work, we analyze the case in which the reduced model is replaced by a neural surrogate rather than evaluated through a classical numerical scheme. We show that the resulting estimator remains unbiased and that the change in the optimal variance induced by the neural approximation is controlled by the error between the exact low-fidelity observable and its neural approximation. This estimate is then combined with residual stability estimates for inhomogeneous Fokker--Planck and Bhatnagar--Gross--Krook surrogates. We also extend the analysis to several control variates and to an asymptotic-preserving (AP) hierarchy containing the limiting Euler observable. In the fluid limit, the optimal hierarchical variance converges to the variance associated with the limiting Euler control, while the contribution of the intermediate kinetic correction vanishes. Numerical tests based on micro--macro neural surrogates illustrate the predicted variance stability and the behavior of the AP hierarchy.
- [14] arXiv:2611.18748 (cross-list from physics.chem-ph) [pdf, html, other]
-
Title: Dissipative many-body theory of electron-phonon-photon systemsComments: 18 pages, 1 figureSubjects: Chemical Physics (physics.chem-ph); Materials Science (cond-mat.mtrl-sci); Strongly Correlated Electrons (cond-mat.str-el); Computational Physics (physics.comp-ph); Optics (physics.optics)
Optical cavities offer opportunities to control quantum materials, but radiative leakage and vibrational relaxation call for microscopic theories that treat interactions and dissipation on equal footing. We develop a Keldysh-Lindblad nonequilibrium Green's function framework for interacting electrons coupled to quantized photon and phonon fields. The formulation accommodates boson-number-nonconserving Hamiltonians and dissipative channels that combine electronic transitions with bosonic processes. We derive the mixed electron-boson diagrammatic expansion and the corresponding Kadanoff-Baym equations, retaining environmental noise, initial correlations, and interaction-induced memory. The equal-time equations govern the electronic density matrix and the bosonic covariance. As an illustration, we construct jump operators that preserve a semiconductor with filled valence bands and an empty photon field. Within a mean-field treatment, this state remains stationary although its two-time propagators exhibit dissipative broadening. The framework opens new avenues for studying correlated dynamics in lossy cavities and vibrational environments.
- [15] arXiv:2611.18969 (cross-list from nlin.CD) [pdf, html, other]
-
Title: Intrinsic memory limits resonant control of complex cardiac spiral wavesSubjects: Chaotic Dynamics (nlin.CD); Adaptation and Self-Organizing Systems (nlin.AO); Pattern Formation and Solitons (nlin.PS); Biological Physics (physics.bio-ph); Computational Physics (physics.comp-ph)
Resonant forcing is known to induce controlled drift of phase-locked spiral waves in excitable media. Here, we show that intrinsic memory fundamentally alters this paradigm in complex meandering spirals typical of cardiac tissue, where phase-dependent responses to successive perturbations over the meander cycle interfere destructively to produce metastable quasi-stationarity or constructively to induce stable confinement. These dynamical insights motivate an improved low-energy multi-pulse feedback protocol that robustly drives spiral waves to termination through sustained directed drift.
- [16] arXiv:2611.19014 (cross-list from cond-mat.stat-mech) [pdf, html, other]
-
Title: Beyond global density in a stochastic sandpile: the role of spatial signatures in the predictability of extremesSubjects: Statistical Mechanics (cond-mat.stat-mech); Computational Physics (physics.comp-ph)
In self-organized critical systems the global load is a strong precursor of extreme events. In the two-dimensional Manna sandpile, grain density alone predicts extreme avalanches with an AUC of about 0.96. We ask whether density fully determines the system's readiness for an extreme event, that is, whether microstates with the same number of grains are equivalent. Because density is such a strong predictor, a method that sees the full microstate can score well just by counting grains. We therefore compare only states with an identical number of grains. Even then, both a parameter-free statistic and a convolutional network distinguish states that precede an extreme avalanche from typical ones, while shuffled controls do not. The signal is a load contrast between the center and the dissipative boundaries, carried by the two lowest symmetric cosine modes. Their normalized energy is a single variable that needs no training. Together with density it recovers most of the gain that a network trained on the full microstate achieves over density alone. Readiness for an extreme avalanche therefore depends both on the total load and on how that load is distributed relative to the boundaries.
- [17] arXiv:2611.19128 (cross-list from quant-ph) [pdf, html, other]
-
Title: Efficient simulation of the generalized Langevin equation using a basis of dynamically relevant histories for memory and noiseSubjects: Quantum Physics (quant-ph); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
The Generalized Langevin Equation (GLE) is a cornerstone for modeling non-Markovian dynamics, but its numerical solution remains challenging due to the history-dependent convolution and the need to generate colored noise with a prescribed correlation function. We introduce a forward construction that addresses both challenges within a single framework. We represent an arbitrary trajectory as sum of relevant and irrelevant parts, where relevant part can be decomposed into a small number of characteristic history patterns that make the main contribution to future dynamics. It allows to reduce numerical cost to \(O(d^2)\) for $d$ chosen patterns instead of full history. The same construction applies to colored noise generation, where the causal filter plays the role of the memory kernel. We demonstrate the accuracy and efficiency of the method for several representative kernels, including long-range kernels with power-law tails.
- [18] arXiv:2611.19400 (cross-list from astro-ph.HE) [pdf, html, other]
-
Title: Neural Surrogates for Tabulated Nuclear Equations of State in Core-Collapse Supernova SimulationsComments: 15 pages, 14 figures, accepted for publication in PASJSubjects: High Energy Astrophysical Phenomena (astro-ph.HE); Instrumentation and Methods for Astrophysics (astro-ph.IM); Computational Physics (physics.comp-ph)
Large-scale core-collapse supernova simulations rely on tabulated nuclear equations of state (EOSs), whose memory requirements can become a practical constraint in massively parallel calculations. We explore a compact alternative by replacing the tabulated thermodynamic mapping with a neural-network surrogate, using the Lattimer--Swesty EOS as a representative reference model. The surrogate directly predicts pressure, sound speed, and effective adiabatic index from $(\log \rho, \log e, Y_e)$ and accurately reconstructs the nonlinear thermodynamic structure of the EOS, including its behavior across phase-transition regions. It reproduces one-dimensional shock-tube solutions with hydrodynamic errors at the $\sim 10^{-3}$ level under different thermodynamic conditions and yields hydrostatic equilibrium configurations closely matching those obtained with the reference EOS. We find that smaller regression errors do not necessarily yield better hydrodynamic performance, demonstrating the importance of hydrodynamic validation in surrogate model selection. A compact network with only $\sim 10^4$ parameters retains this accuracy while reducing the memory required to store the EOS from $\sim 160$ MB to $\sim 100$ KB, corresponding to a reduction by a factor of $\sim 1500$. The surrogate also remains stable through collapse, bounce, shock formation, and post-bounce evolution in a one-dimensional adiabatic core-collapse calculation. These results demonstrate the potential of compact neural EOS surrogates for reducing memory requirements in large-scale core-collapse supernova simulations.
- [19] arXiv:2611.19494 (cross-list from physics.flu-dyn) [pdf, html, other]
-
Title: A quantitative measure of predictability for surface gravity wavesSubjects: Fluid Dynamics (physics.flu-dyn); Computational Physics (physics.comp-ph); Data Analysis, Statistics and Probability (physics.data-an)
We introduce a Bayesian inference framework to characterize the predictability of a one dimensional ocean surface wave field given surface elevation measurements from a sensor. Unlike the classical theories that only sketch the boundary of a predictable zone based on group velocity, we use the Bayesian posterior uncertainty of surface elevation to quantify the local predictability at every spatiotemporal location. Our approach first constructs a prior distribution to encode existing knowledge of the wave field such as the spectrum and propagation direction. By conditioning on the (possibly noisy) measurement data, we infer the posterior distribution of the wave field, from which a spatiotemporal uncertainty map is extracted. This map agrees well with the group-velocity-based classical predictable zone theory when the frequency spectrum has bounded support. Meanwhile, it also demonstrates predictability even for a spectrum with unbounded frequency support where the classical predictable zone degenerates. A refined analysis of posterior uncertainty using observability Gramian matrix reveals a `data-informed' part and an `un-informed' part of the wave field, with the latter being fundamentally un-predictable. Furthermore, nonlinear effects can be naturally introduced into the Bayesian framework by incorporating the nonlinear ocean surface wave equations. This is the first time that nonlinearity is rigorously treated in the predictable zone problem. We perform numerical tests using both bounded, piecewise constant spectrum and unbounded JONSWAP spectrum. The numerical results confirm the proposed advantages of our method for linear waves and reveal how wave steepness affects the predictability of nonlinear waves. Nonlinear modal interactions redistribute uncertainty between the inside and the outside of the predictable zone with an amount proportional to the wave steepness.
- [20] arXiv:2611.19520 (cross-list from physics.plasm-ph) [pdf, html, other]
-
Title: Simulations of turbulent heat and particle transport in stellarator plasmasComments: PhD thesis, Universidad Carlos III de Madrid (2025). Advisors: J. M. García-Regaña and I. Calvo. 152 pages, 59 figures, 11 tablesSubjects: Plasma Physics (physics.plasm-ph); Computational Physics (physics.comp-ph)
Achieving sustained energy production in magnetic confinement fusion reactors requires a deep understanding of turbulent transport of heat and particles. This thesis investigates turbulent transport with a particular emphasis on stellarators, using simulations with the $\delta f$ gyrokinetic code stella for the W7-X, LHD, TJ-II and NCSX stellarators, the ASDEX Upgrade (AUG) tokamak and the Cyclone Base Case (CBC). It addresses two questions: how do density gradients affect turbulent heat transport, and can turbulence help retain particles in the plasma core?
In the first part, an inter-machine study shows a strong reduction of the ion heat flux in NCSX and W7-X with increasing density gradients, attributed to the stabilization of ion-temperature-gradient-driven modes without significantly destabilizing trapped-electron modes or universal instabilities. This beneficial effect is absent in tokamaks, where the heat flux increases with the density gradient, highlighting the role of magnetic geometry. The ion heat flux in stellarators does not follow the growth rate of the fastest-growing linear mode, underscoring the importance of nonlinear effects. Trapped-electron modes dominate in CBC, AUG, TJ-II, LHD and NCSX, while passing-particle-driven universal instabilities contribute significantly in W7-X.
The second part addresses the long-standing discrepancy between the strongly hollow density profiles predicted by neoclassical theory in large stellarators and the flat or weakly peaked profiles measured in W7-X. Nonlinear simulations and a particle balance analysis show that turbulence provides a robust inward particle flux, driven by finite ion and electron temperature gradients, that can account for the observed profiles: the sign of the turbulent particle flux matches the sign of the discrepancy between the experimental and neoclassical fluxes across the entire plasma radius of W7-X. - [21] arXiv:2611.19547 (cross-list from physics.chem-ph) [pdf, html, other]
-
Title: Affordable Analytic Gradients for G0W0-BSE Excited States for Multiple GW Variants via Back-PropagationSubjects: Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
We derive, implement and validate analytic nuclear gradients for three variants of $GW$ self-energies---the contour-deformation (CD), the analytic continuation (AC), and the contour deformation with $W$-analytic-continuation (CD-WAC)---and for the Bethe--Salpeter equation (BSE) built on top of them. All gradients are expressed as contractions of a set of covectors connected to derivatives of primitive quantities in SCF calculations. The derivation of these covectors is a back-propagation procedure through the $GW$ and BSE equations. Central finite differences confirm the correctness of all implemented gradients. With the AC-$GW$-BSE gradient we optimize the excited-state geometries of carbon monoxide, a donor-acceptor dye labelled as DA1, and DMABN, a typical model system for twisted intramolecular charge-transfer phenomena (TICT), and have obtained results that successfully reproduce observations from experiments or high-level calculations. We also performed trajectory surface hopping (TSH) simulations of nonadiabatic molecular dynamics (NAMD) of DMABN using the SHARC NAMD package, showing the capacity of our method to simulate NAMD problems. We performed timing and memory scaling tests, demonstrating that the gradient calculations of both $GW$ and BSE scale asymptotically as $N^4$ for timing and $N^3$ for memory. All source codes used in the present work, including the REST program package and the modified SHARC NAMD package that supports REST as the quantum chemistry backend, are available on Gitee this https URL and this https URL
- [22] arXiv:2611.19645 (cross-list from physics.chem-ph) [pdf, html, other]
-
Title: Formulation of Brownian chain molecular dynamics and the path-resolved quasi-density operatorComments: 20 pages, 5 figuresSubjects: Chemical Physics (physics.chem-ph); Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
Brownian chain molecular dynamics (BCMD) [M. Shiga, J. Comput. Chem. 43, 1864 (2022)] was introduced as a practical path-integral-based approach to approximate quantum vibrational dynamics in molecular and condensed-phase systems. Its equations of motion combine Newtonian dynamics of the centroid variables with overdamped Brownian dynamics of the non-centroid path variables. Here, we introduce a path-resolved quasi-density-operator (path-QDO) representation by extending the QDO construction underlying centroid molecular dynamics to the full discretized imaginary-time path. For two position-diagonal observables, this representation gives an exact expression for the Kubo-transformed correlation function in the fine-discretization limit, without requiring either observable to be linear. BCMD can be viewed as a phenomenological, time-homogeneous Markov approximation to the exact reduced propagation in these path variables, although its generator is not derived from the path-QDO formulation. We also formulate and validate a multiple-time-step algorithm that allows pseudo-Hamiltonian errors in the deterministic propagation to be controlled by separating the Brownian refresh interval from the inner integration time step. We test the path-QDO representation on one-dimensional models and present BCMD infrared spectra of q-TIP4P/F ice, ambient water, and hot water, comparing them with published results from other approximate quantum-dynamics methods.
- [23] arXiv:2611.19676 (cross-list from cs.LG) [pdf, html, other]
-
Title: Score-Based Stochastic Closure Modeling via Offline Indirect Learning and Online Data AssimilationSubjects: Machine Learning (cs.LG); Computational Physics (physics.comp-ph); Fluid Dynamics (physics.flu-dyn)
Stochastic closure modeling is an emerging area in reduced-order modeling of large and complex (e.g., multiscale, multi-physics) systems, for which the computationally affordable reduced-order physical descriptions (e.g., digital twins for additive manufacturing, world models for AI agents) in numerical simulations inevitably lead to information losses that cannot be well addressed by classical deterministic closures. However, learning stochastic closures typically requires closure data that are difficult to obtain from measurements or often prohibitively expensive to compute using first-principles simulations. In this work, we develop a stochastic closure modeling framework that builds on score-based diffusion models, leveraging indirect offline training and online data assimilation without requiring the direct data of closures. Numerical experiments on two-dimensional turbulent flows demonstrate that indirect offline learning captures key closure statistics and improves predictions of system state. In addition, combining the learned closure with known physical dynamics leads to better generalization capability across different regimes than purely data-driven models that directly predict system state. Compared to standard data assimilation methods with a Gaussian assumption, the posterior-score assimilation better represents non-Gaussian features of the learned closure, and its combination with state-level data assimilation further improves prediction of the system state. The proposed framework enables stochastic closure modeling in realistic and challenging scenarios of complex dynamical systems in mechanics and engineering problems, where non-Gaussian statistics of model-form uncertainties are non-negligible (e.g., due to the nonlinearities of the system dynamics and/or the observations) and true closure data are not available, for both offline learning and online assimilation.
- [24] arXiv:2611.19793 (cross-list from cond-mat.mtrl-sci) [pdf, html, other]
-
Title: IANN: A Unified Framework for Equivariant InterAtomic Neural Networks with User-Friendly ImplementationComments: 31 pages, 7 figures, 4 tables; Supplementary Information included (49 figures, 11 tables). IANN is MIT-licensed. Code: this https URL - Documentation: this https URL - Curated DFT databases: this https URL - Foundation models: this https URLSubjects: Materials Science (cond-mat.mtrl-sci); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
Machine-learning interatomic potentials (MLIPs) now approach density functional theory (DFT) accuracy at a fraction of its cost, but the practical benefit is limited by software fragmentation: each state-of-the-art architecture is released as a separate codebase, with its own data format, configuration system, training loop and molecular-dynamics interface. Comparing two architectures on the same dataset, or moving a trained model into a production simulation, therefore requires substantial re-engineering. We introduce IANN (InterAtomic Neural Network), an open-source PyTorch framework that reimplements equivariant graph neural networks - PaiNN, NequIP, MACE, Allegro, EquiformerV2, EquiformerV3 and UMA - behind a unified data object, a unified trainer and a single deployment route, so that the architecture is a one-word change in a configuration dictionary. Forces, stresses and virials are obtained for every architecture from one analytic differentiation of the energy with respect to the edge vectors of the atomic graph, exact under periodic boundary conditions. Distributed training is configured automatically from the SLURM, PBS or MPI environment, so the same script runs unchanged from a workstation to a multi-node GPU allocation. Self-describing checkpoints serve directly as ASE calculators, as ensembles reporting uncertainties, or as TorchScript models run inside LAMMPS through bundled C++ pair styles with single- and multi-GPU inference. Trained on a common dataset of millions of structures, the architectures can be compared under identical conditions. We further release PaiNN foundation models trained on 41.9 million curated DFT structures from MPtrj, Alexandria, OMat24, MatPES, OC20, OC22, OC25, MPr2SCAN and MatPESr2SCAN, spanning the PBE, RPBE and r2SCAN functionals. Additionally, IANN provides an agent interface to facilitate integration with autonomous agents.
- [25] arXiv:2611.19801 (cross-list from cond-mat.str-el) [pdf, html, other]
-
Title: Exact environment checkpointing for memory-limited DMRG and TDVPComments: 13 pages, 6 figures. Code and data: this https URL (doi:https://doi.org/10.5281/zenodo.24193745)Subjects: Strongly Correlated Electrons (cond-mat.str-el); Computational Physics (physics.comp-ph); Quantum Physics (quant-ph)
Finite matrix-product-state sweep algorithms, such as the density-matrix renormalization group (DMRG) and the time-dependent variational principle (TDVP), store one environment tensor per site. At the bond dimensions needed for frustrated and two-dimensional systems this stack, rather than the state itself, exhausts the memory of the device. We adapt checkpointing to these sweeps: every $k$-th environment is kept, with $k\approx\sqrt{N}$ for $N$ sites, and the blocks in between are rebuilt when the sweep returns to them. The rebuild repeats the original contractions, so energies and states are reproduced bit for bit when the contractions are deterministic. Environment memory falls by 4.6-10x on cylinders of 96-384 sites. Bond-dimension searches with an implementation that rebuilds checkpoints at each half-sweep give gains of 1.77 for one level and 2.07 for two levels with eight contraction chunks on a single 141 GB GPU, using a 144-site triangular $J_1$-$J_2$ cylinder and stored baselines with matching chunking. Separate timings with checkpoints retained across half-sweeps give a one-level overhead of 6-15%, depending on the local eigensolver. At $D=1536$ the two-level schedule with eight chunks takes 1.42 times the sweep time of the equally chunked stored baseline, or 1.79 times that of the unchunked stored baseline. Simple models of the memory and the run time reproduce the measurements to within a few per cent. Unlike offloading, the method needs no host memory or disk, and the two can be combined. Implementations are provided for this http URL, TeNPy, this http URL and quimb.
- [26] arXiv:2611.19888 (cross-list from cond-mat.mtrl-sci) [pdf, other]
-
Title: Metallic or insulating magnetic phases of transition-metal oxides in light of the Slater vs Mott controversyComments: 12 pages, 5 figuresSubjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph); Quantum Physics (quant-ph)
Magnetic d-electron oxides usually have a long-range-ordered ground state and an above-ground-state disordered para phase. The historical Slater-Mott controversy still reverberating to date pertains to understanding the origin of the insulating vs metallic nature of these two phases. Slater suggested that the insulating character evolved from the long-range order (LRO) antiferromagnetic (AFM) cell-doubling, leading to energy band repulsion and thus to forming a band gap. The reliance on LRO to explain gapping suggests that the absence of LRO in the spin-disordered para-phase would deprive the system from band repulsion, thus predicting a metallic paramagnetic (PM) phase. This, however, conflicts with numerous current observations. The alternative Mott-Peierls-Hubbard view abandons the mean-field-like band structure view emphasizing instead the local interelectronic repulsion U as a strong correlation signature of gapping both at the presence or absence of LRO. However, U is not an easily measured or calculated quantity in periodic systems, a situation often leading in practice to treating U as an adjustable parameter. The current work uses a parameter-free mean-field-like approach (Density Functional Theory, DFT), allowing energy-lowering symmetry breaking. This is done by considering local magnetic motifs as building blocks of the macroscopic AFM or PM phases. It is found that even when such motifs are isolated electronically from each other (i.e. vanishing LRO) by embedding them in quantum barriers (such as magnetic NiO6 octahedra surrounded locally by wide-gap MgO) they exhibit an intrinsic band gap. Thus, the origin of the band gaps in AFM or PM phases need not be LRO or strong correlation and can be understood within the mean-field-like band structure approach with symmetry breaking.
- [27] arXiv:2611.19941 (cross-list from physics.flu-dyn) [pdf, html, other]
-
Title: Magnetic control of hydrodynamic spin coupling between rotors: Ferrohydrodynamic effects in the two-rotor problemEfstratios E. Tzirtzilakis (University of the Peloponnese, Patras, Greece)Comments: 51 pages, 18 figures, 8 tables. The following article has been submitted to Physics of FluidsSubjects: Fluid Dynamics (physics.flu-dyn); Computational Physics (physics.comp-ph)
The hydrodynamic spin coupling between two rotors ("fluid gears") is investigated numerically for a magnetic fluid in non-uniform magnetic fields. An active cylinder drives a torque-free passive one inside a circular corral, with one rotor heated. The model combines magnetohydrodynamics (MHD) and ferrohydrodynamics (FHD), with the full Kelvin force and temperature-dependent Langevin magnetisation. For the oil-based ferrofluid EFH1 the Lorentz force is negligible, and the Kelvin force drives thermomagnetic convection from the heated rotor. Steady torque-free states are computed with Taylor-Hood finite elements, the gear ratio $\Gamma$ being an additional unknown. Without a field the results agree within 1-8% with published planar simulations up to $\mathrm{Re}\approx 50$. About 17,000 steady states cover 34 arrangements of wires, NdFeB rods and magnet cores, gaps $G=0.1$-$0.7$, $\mathrm{Re}=10$-$40$ and heating of 5-20 K. Because the coupling is weak, fields of 1-2 mT change $\Gamma$ by an amount comparable to $\Gamma$ itself. At 2 mT it is reversed, strengthened or weakened in nine cases out of ten. The outcome is set by the source position: at the experimental gap and $\mathrm{Re}=20$ a source below the corral or behind the active rotor strengthens the gear-like counter-rotation up to 2.7-fold, and a source above the corral reverses it into belt-like corotation. Several sources do not act additively, and a magnet core inside the active rotor raises $|\Gamma|$ at small gaps to seven times the largest hydrodynamic value. At fixed geometry and Reynolds number, field strength and heating act approximately through a single thermomagnetic source parameter.
- [28] arXiv:2611.20002 (cross-list from cond-mat.soft) [pdf, html, other]
-
Title: Self-assembly of dipolar nanoclusters on an attractive substrate: A coarse-grained simulation approachDenis Gessert, Harshita Chauhan, Thi Ngoc Ha Nguyen, Christoph Tegenkamp, Michael Mehring, Varsha Banerjee, Martin WeigelComments: 16 pages, 12 figures, AIP styleSubjects: Soft Condensed Matter (cond-mat.soft); Materials Science (cond-mat.mtrl-sci); Statistical Mechanics (cond-mat.stat-mech); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
Recently, Thomas et al. [Nanoscale 17, 18291 (2025)] showed that bismuth oxido nanoclusters (BiO-NCs) of the type [Bi$_{38}$ O$_{45}$ (Boc-L-Phe-O)$_{24}$] form chains when deposited onto gold or pyrolitic graphite substrates. This chain formation is significantly enhanced when the bismuth in the BiO-NCs is partially replaced by cerium through doping. In this work, we study a simplified, coarse-grained model of this process, namely (magnetic) Stockmayer particles in the presence of an attractive substrate. The model is investigated by means of molecular dynamics simulations. The deposition experiment is mimicked by a rapid quench from high temperature to a low temperature below the bulk coexistence curve. The results are found to be qualitatively similar to the experimental observations reported earlier. In our simulations, we find significant chain formation in the bulk even at rather low concentrations, which is in contrast to the BiO-NCs in ethanol for which dynamic light scattering reveals negligible aggregation in the absence of the substrate. This lends support to a scenario, conjectured in the work of Thomas at al., where the dipole moments are induced by the particle-substrate interaction and hence there is little dipole moment in the bulk. In a second part, we study the system in equilibrium. We consider the scaling of chains, the number of layers populated as a function of two-dimensional density, and we draw analogies to polymers on interfaces, which have been extensively studied over the past decades. Most notably, it is found that the chains behave as self-avoiding walks at low densities, which gives way to rod-like scaling at higher densities.
- [29] arXiv:2611.20175 (cross-list from physics.flu-dyn) [pdf, html, other]
-
Title: Solver-consistent learned curvature for algebraic volume-of-fluid simulations of rising bubbles: symmetry-preserving networks, closed-loop data aggregation and a posteriori benchmarksSubjects: Fluid Dynamics (physics.flu-dyn); Computational Physics (physics.comp-ph)
Curvature reconstructed from compressed volume fractions exhibits a persistent force-weighted bias in the tested algebraic volume-of-fluid formulation. In OpenFOAM interFoam, face curvature is underestimated by 11-22% for radii of 2.5-40 cells, explaining the 14-17% static Laplace-pressure deficit. This bias is consistent with approximately 2% rise-velocity underprediction in the first Hysing benchmark. We estimate curvature from 5 x 5 volume-fraction stencils using a two-hidden-layer perceptron with exact phase-exchange antisymmetry and invariance under all eight square-grid symmetries. Training uses solver-generated profiles of periodic translating bubbles, droplets, and flat films with exact analytical curvature labels. Iterative data aggregation incorporates model-solver feedback and resolves deployment instabilities found in 15 of 76 periodic cases with the initial model. Band-weighted face interpolation supplies the same capillary force to momentum prediction and pressure correction. Independent tests show static pressure-jump errors of 2-6% and reductions in time-averaged maximum capillary numbers by factors of 2.8-16. Across three grids, time-integrated rise-velocity errors decrease from 2.06-2.76% to 0.65-1.04% in the first Hysing benchmark and from 5.64-8.72% to 4.72-7.61% in the second. Tuned curvature smoothing gives smaller static pressure-jump errors, but its first-benchmark accuracy deteriorates with refinement and second-benchmark gains remain marginal. A separate interface-gated dissipative momentum closure achieves a held-out coefficient of determination of 0.47 on Favre-filtered targets but changes rise velocity by less than 0.03% of its peak. Capillary-force discretisation is therefore the more influential correction target for the tested laminar bubbles.
- [30] arXiv:2611.20247 (cross-list from cond-mat.mtrl-sci) [pdf, html, other]
-
Title: Machine Learning the $GW$ Self-Energy Operator for Real MaterialsSubjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
Machine learning of $GW$ in solids has largely focused on prediction of scalar quasiparticle energy corrections. Here, we introduce a phase-aware many-body transformer that predicts the full $\mathbf{k}$-resolved complex self-energy matrix. As a proof of principle, we apply the model to predict quasiparticle energies before and after diagonalization of the quasiparticle Hamiltonian, obtaining a test set mean absolute error (MAE) of 29 meV for a dataset of 1745 2D materials in the static COHSEX approximation and 60 meV for a more limited dataset of 433 materials calculated within a plasmon-pole model. COHSEX pretraining followed by finetuning reduces this MAE to 42 meV. The predicted matrices also accurately reproduce QP wavefunctions.
- [31] arXiv:2611.20329 (cross-list from physics.optics) [pdf, html, other]
-
Title: Ternary-Valued FDTD Method for Nonlinear Optics: Third-Harmonic Generation on a Multiplier-Free SolverSubjects: Optics (physics.optics); Mathematical Physics (math-ph); Numerical Analysis (math.NA); Computational Physics (physics.comp-ph)
I show that a finite-difference time-domain (FDTD) solver with field variables restricted to the ternary alphabet $\{-1,0,+1\}$ can reproduce nonlinear optical physics quantitatively on both CPUs and field-programmable gate arrays (FPGAs). Building on the general ternary-valued FDTD algorithm, in which each Yee cell stores an integer state that accumulates the discrete curl of neighbouring symbols without being drained, I extend the scheme to inhomogeneous and Kerr media and to Mur absorbing boundaries, without performing any multiplication operations. Dielectric contrast is implemented as a multiplexer, since the difference between ternary symbols takes only a few values, the instantaneous Kerr nonlinearity as a coefficient look-up table and the Mur boundary coefficients as canonical signed-digit shift--add chains. The scheme reproduces the third-harmonic conversion efficiency of a floating-point reference to within $1.4\%$. A symmetry of the governing physics, expressed through the even harmonics forbidden by centrosymmetric $\chi^{(3)}$, provides a reference-free self-check for both ternary and floating-point models. The complete solver fits on a small, low-cost FPGA and contains no multiplier anywhere in its datapath, placing nonlinear optical modelling within reach of embedded and autonomous platforms rather than of high-performance computers.
- [32] arXiv:2611.20591 (cross-list from physics.chem-ph) [pdf, html, other]
-
Title: GPRidge: Gaussian-process Accelerated Calculations of Minimum Energy Paths and First-order Saddle PointsRohit Goswami (1 and 2), Maxim Masterov (1), Satish Kamath (1), Alejandro Pena-Torres (3 and 2), Andreas Lynge Vishart (4), Thomas Bligaard (4), Hannes Jonsson (5 and 2) ((1) SURF B.V., Amsterdam, The Netherlands, (2) Science Institute, University of Iceland, VR-III, 107 Reykjavik, Iceland, (3) Icelandic Research e-Infrastructure (IREI), University of Iceland, 107 Reykjavik, Iceland, (4) Department of Energy Conversion and Storage, Technical University of Denmark (DTU), 2800 Kgs. Lyngby, Denmark, (5) Faculty of Physical Sciences, University of Iceland, 107 Reykjavik, Iceland)Comments: Supporting Information included as an appendix. Prepared for submission to the Journal of Chemical Information and Modeling. Reproducibility archive, including the GPRidge 0.5.0 source snapshot and run records: this https URLSubjects: Chemical Physics (physics.chem-ph); Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
GPRidge, a Gaussian-process (GP) library for accelerating calculations of minimum energy paths and first-order saddle points with compiled-language and Python bindings, is presented. An evaluation controller uses GP predictions to iteratively adjust the Cartesian coordinates of the atoms. The fitted GP energy surface and measured energies and forces pass from the path calculation to local refinement of the saddle point. New evaluations improve the surrogate and check the candidate geometry. For the molecular calculations, acceptance requires measured-force convergence and an independent Hessian with only one negative eigenvalue. A second negative eigenvalue triggers an escape and renewed refinement. The controller supports a joint inverse-distance GP for molecules and local GP residuals on periodic systems. Applications to the 24 Baker-Chan and 20 Birkholz-Schlegel molecular reactions using the Universal Model for Atoms (UMA-S-1.1), and to 72 Pt heptamer and island rearrangements on Pt(111) using a Morse potential, are presented. On matched Baker-Chan reactions, GPRidge uses 7.8- and 7.0-fold fewer energy-and-force evaluations than climbing-image nudged elastic band (CI-NEB) and off-path climbing-image nudged elastic band (OCI-NEB), respectively, reducing validation-inclusive CI-NEB time by 41%. All 20 Birkholz-Schlegel saddle points pass the independent index-one check, compared with 17 for the classical dimer method started from the same path maxima; on those 17 reactions, the GP dimer stage uses 6.4-fold fewer evaluations. All 72 Pt(111) paths meet the OptBench force criterion, with median evaluation counts of 30 and 23 for the heptamer and island problems, against 778 and 443 for CI-NEB.
- [33] arXiv:2611.20609 (cross-list from cs.LG) [pdf, html, other]
-
Title: Newton-PINet: A fast physics-informed neural network with Newton linearization for meta-learning nonlinear PDEsComments: Accepted at NeurIPS 2026Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computational Engineering, Finance, and Science (cs.CE); Computational Physics (physics.comp-ph)
Scientific machine learning has opened new avenues for solving parameterized partial differential equations (PDEs), enabling models to learn a family of PDEs and generalize to unseen instances. In this context, data-driven operator learning methods typically require large training datasets, while physics-informed neural networks (PINNs) suffer from difficult optimization and limited generalization, especially for nonlinear PDEs. We propose Newton-PINet, a physics-informed network enhanced by Newton linearization, offering an effective meta-learning framework for nonlinear PDEs. Newton-PINet (i) employs a physics-informed multilayer network with skip connections, where the output-layer weights are solved by least squares; (ii) adopts a two-stage learning strategy that first leverages gradient-based training to learn robust representations from the available training tasks, and then performs gradient-free fine-tuning on the output layer for fast task-specific generalization; and (iii) incorporates a Newton linearization method to speed up the least-squares iteration for nonlinear PDE problems. On a challenging nonlinear reaction-diffusion benchmark, Newton-PINet achieves up to three orders of magnitude lower relative error than recent neural solvers, while using 16x fewer training tasks and over an order of magnitude less training time (about 5 minutes versus several hours). This work advances the meta-learning of PINNs toward data-efficient, fast, and generalizable physics solvers. The code and datasets are publicly available at: this https URL.
- [34] arXiv:2611.20789 (cross-list from cond-mat.mtrl-sci) [pdf, html, other]
-
Title: Detecting False Self-Consistency in Density-Functional CalculationsComments: 31 pages main text + 65 pages Supplementary Information, 3 figuresSubjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
Density-functional calculations feed materials databases and machine-learning models, and are declared converged when energy changes and residuals fall below thresholds. We show that these criteria, and even the agreement of independent optimizers, can accept a state from which the energy still descends or whose Hamiltonian is inconsistent with the energy. A variational formulation of self-consistency in the space of Kohn-Sham Hamiltonians yields a descent probe that exposes lower states accessible from a returned solution, and a frozen-orbital Janak test detects such an inconsistency. Both apply to any code without modification. In VASP, a UO$_2$ PBE+$U$ calculation converges 1.46 eV above a lower state, and CoO and FeO converge as metals 0.8-0.9 eV above insulating states. In a census of 339 converged calculations on tungsten-free Materials Project structures, recomputed with a modified static protocol, 57 (16.8%) show energy decreases above the threshold, 49 without an initial energy rise exceeding the tolerance.
- [35] arXiv:2611.20990 (cross-list from hep-lat) [pdf, html, other]
-
Title: Transferable neural-network preconditioners for Wilson--Clover propagator inversionsComments: 15 pages, 6 figuresSubjects: High Energy Physics - Lattice (hep-lat); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
We investigate two single--layer neural-network architectures for accelerating four-dimensional Wilson--Clover propagator inversions on lattice ensembles in practical use; both transfer directly without retraining. The first is a gauge-covariant parallel transport convolution (PTC) network that directly reduces the number of generalized conjugate residual (GCR) solver iterations. The second is a CNN-like network used together with a multigrid-inspired algorithm. The two approaches can be combined to further reduce GCR iterations. PTC transfers without retraining across lattice volumes, lattice spacings, and pion masses while maintaining comparable acceleration, whereas the CNN-like network transfers directly from small- to large-volume ensembles with strong acceleration. A cost comparison accounting for both basis construction and solves shows that the proposed scheme is more advantageous for large-volume ensembles than for small-volume ones.
- [36] arXiv:2611.21063 (cross-list from cs.CE) [pdf, html, other]
-
Title: A Hierarchically Compressed Immittance Operator Formulation for Fast Statics-to-Full-Wave Electromagnetic Simulation with the Boundary Element MethodComments: 15 pages, 12 figures. Submitted to the IEEE Transactions on Antennas and PropagationSubjects: Computational Engineering, Finance, and Science (cs.CE); Computational Physics (physics.comp-ph)
The boundary element method allows the electromagnetic simulation of three-dimensional problems with a two-dimensional discretization, using a surface integral representation of Maxwell's equations. For applications involving broadband simulations or multiscale geometries, the well-known low-frequency breakdown of conventional integral equation formulations must be addressed, while ensuring that dielectric, semiconducting, and lossy metallic regions are modeled rigorously, including the skin effect. In this work, a new boundary element formulation based on an immittance operator is proposed, where the underlying integral equations are inherently valid from statics to the full-wave regime. To enable efficient matrix assembly and solution, a hierarchical matrix compression strategy is devised, leveraging fast direct factorization as an intermediate step, and fast matrix-vector products for the iterative solution of the final system of equations. The proposed formulation is capable of electrostatic, magnetostatic, quasistatic, and full-wave simulations of dielectrics, semiconductors, and lossy metals. Representative numerical examples drawn from a variety of applications demonstrate the accuracy and efficiency of the proposed method.
- [37] arXiv:2611.21147 (cross-list from cond-mat.mtrl-sci) [pdf, html, other]
-
Title: Large-scale Monte Carlo/Molecular Dynamics simulations with Atomic Cluster Expansion potentials in LAMMPSSubjects: Materials Science (cond-mat.mtrl-sci); Mesoscale and Nanoscale Physics (cond-mat.mes-hall); Computational Physics (physics.comp-ph)
Monte Carlo/molecular dynamics (MCMD) simulations are widely used to study the properties of materials at the mesoscale. Simulations with hundreds of thousands of atoms usually rely on empirical potentials because MCMD with machine-learning interatomic potentials is impractical at this scale in existing implementations. Here, we introduce a hybrid MPI/OpenMP scheme for MCMD with Atomic Cluster Expansion (ACE) potentials in LAMMPS. Monte Carlo moves are evaluated from local energy changes and carried out in parallel over MPI ranks, while molecular dynamics is parallelized over OpenMP threads. At a fixed number of CPUs, the wall time of the Monte Carlo moves decreases by a factor equal to the number of MPI ranks. Larger simulation cells also reduce the statistical errors of free energy calculations because thermal fluctuations decrease with system size. We demonstrate the scheme by calculating free energies of the HCP and BCC phases in the strongly anharmonic Hf--Ti system and by simulating zirconium segregation in a nanocrystalline copper sample of 670,000 atoms. For Cu--Zr, an ACE potential and an empirical potential predict similar grain boundary compositions but different amounts of zirconium dissolved in the grains. The scheme makes MCMD simulations with machine-learning interatomic potentials practical at the length scales of microstructure.
- [38] arXiv:2611.21194 (cross-list from cond-mat.mes-hall) [pdf, html, other]
-
Title: Interaction Landscapes of Structurally Distorted Toroidal Classical Spin TexturesComments: to appear in Adv. Theory SimulSubjects: Mesoscale and Nanoscale Physics (cond-mat.mes-hall); Materials Science (cond-mat.mtrl-sci); Other Condensed Matter (cond-mat.other); Computational Physics (physics.comp-ph); Quantum Physics (quant-ph)
We develop a classical discrete magnetostatic model of localized magnetic moments arranged in a toroidal texture, where the moments are represented by static current loops and study the behavior of the magnetic dipole-dipole interaction under several types of structural deformations. The purpose is to disentangle effects due to geometric and magnetic deformations, including elliptical, lopsided, alternating, and tilted configurations. In particular, we investigate how distortions modify the toroidal moment and the interaction energy between pairs of toroidal structures. The resulting interaction landscapes exhibit characteristic symmetry patterns and anisotropies that are highly sensitive to the underlying deformation. As a separate proof-of-concept comparison, first-principles calculations on a molecular spin-toroidal system reveal qualitatively similar orientation-dependent behavior. The calculated interaction is microscopically the conventional pairwise magnetic dipole-dipole interaction between localized moments arranged in a spin-toroidal texture; the toroidal moment characterizes the collective texture and should not be interpreted as introducing an additional elementary interaction term. Our results provide a platform for exploring orientation-dependent interactions of spin-toroidal textures in molecular magnets and related nanoscale magnetic systems that are essential to understand the role of toroidal order.
- [39] arXiv:2611.21295 (cross-list from cs.LG) [pdf, other]
-
Title: Deep Learning for Physicists: From Linear Regression to Reasoning ModelsComments: 115 pages, many figures. Based on lectures given at Queen Mary University of LondonSubjects: Machine Learning (cs.LG); High Energy Physics - Lattice (hep-lat); High Energy Physics - Theory (hep-th); Computational Physics (physics.comp-ph)
These lecture notes are based on an introductory course on neural networks and deep learning that the author taught in the School of Mathematical Sciences at Queen Mary University of London in 2023/24 and 2024/25. The author, however, is a physicist, and it shows: wherever possible, ideas are explained with physical intuition rather than heavy formalism. Readers with a background in physics or mathematics who want a self-contained, introductory route into modern machine learning may therefore find these notes particularly friendly.
The notes start from maximum likelihood, least squares, linear regression and (stochastic) gradient descent. They then cover neural networks, backpropagation, the vanishing gradient problem, batch and layer normalization, overfitting and regularization, adversarial examples, and the kernel method. The second half turns to natural language processing: word embeddings, recurrent networks and LSTMs, the attention mechanism and Transformers, and large language models, including scaling laws, the history of GPT, instruction tuning, reinforcement learning from human feedback, hallucination, alignment faking, chain-of-thought reasoning and test-time scaling. The emphasis on large language models reflects requests from the students.
The notes are aimed at third- and fourth-year students in the UK system. Because the course was also taken by students from other departments, very little is assumed: even basic material such as matrix multiplication is explained from scratch, so readers familiar with it can safely skip such parts. No originality is claimed; the aim is a coherent, intuitive path through the material. Developments after spring 2025 are not covered, but the basic ideas emphasized here are unlikely to go out of date soon. - [40] arXiv:2611.21434 (cross-list from cs.LG) [pdf, html, other]
-
Title: Variational Monte Carlo for Quantum Excited States via Nested Low-Rank ApproximationComments: Accepted for NeurIPS 2026. The first two authors contributed equally. 41 pages, 6 figures, 9 tables. Code is available at this https URLSubjects: Machine Learning (cs.LG); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
The electronic Schrödinger equation encodes a molecule's electronic structure and quantum properties, but solving it scales exponentially with the number of electrons. Solving this equation with neural-network variational Monte Carlo (VMC) has reached chemical accuracy on the ground state of small molecules; extending this success to excited states is constrained by two existing paradigms. Penalty-based methods need $K$ Laplacians per step but require tuning the overlap-penalty weights against the energy gaps they aim to compute. Natural excited states VMC (NES-VMC) is penalty-free, but it combines the $K$ ansatzes into one determinant, whose local energy requires $K^2$ Laplacians per step. We resolve this trade-off via low-rank approximation (LoRA) reparameterization of the rank-$K$ variational objective. The reparameterized objective admits a VMC-compatible stochastic gradient estimator via MCMC sampling from the per-state Born densities, and a sequential nesting scheme recovers the lowest $K$ eigenstates in eigenvalue order. The resulting algorithm, NestedLoRA-VMC, inherits a global optimality guarantee from the LoRA principle. On the first-row atoms Li through Ne at $K=10$, its accuracy is comparable to that of NES-VMC and of a state-of-the-art penalty-based method.
- [41] arXiv:2611.21466 (cross-list from physics.plasm-ph) [pdf, other]
-
Title: A spherical body in plasma simulated with adaptive mesh and algorithm refinementComments: Companion to arXiv:2611.14820Subjects: Plasma Physics (physics.plasm-ph); Computational Physics (physics.comp-ph)
An absorbing sphere in a weakly ionized plasma is the oldest kinetic problem of probe theory. It presents key difficulties in plasma simulation: disparate velocity scales, orbital motion, phase-space regions that no characteristic of the plasma reaches, a boundary set by a kinetic flux balance, and collisions that qualitatively change the answer. We use this problem to demonstrate the Plasma Adaptive Mesh and Algorithm Refinement (PAMAR) framework introduced in a companion paper. Electrons and ions have separate solvers on their own adaptive $(r,v,\mu)$ octrees, or on reduced phase spaces at the angular-moment level (Fokker-Planck and two-stream models), coupled through a radial Poisson field solver. Two exact references, a method of characteristics for the collisionless problem and integration along the orbits for elastic scattering, audit the solvers. We compute the current-voltage characteristic of the probe on both electron and ion branches, and the profiles of the potential and of the moments of both species agree with the exact solution. Far from the sphere, the potential follows the Coulomb law in neither limit: it falls as $r^{-2}$ without collisions and as $r^{-1}$ only beyond the ion mean free path. With elastic electron scattering, the kinetic solver follows the exact dependence of the collected current on the mean free path, from free flight toward diffusion, to 1 percent. Finally, we demonstrate the ion model selection cell by cell, using a fluid model coupled to the kinetic solver, with the interface set by the pressure anisotropy; with charge exchange, a sensor of ion trapping is also needed. According to exact solutions, the ion anisotropy extends to several probe radii, much farther than the electron loss cone.
- [42] arXiv:2611.21487 (cross-list from physics.flu-dyn) [pdf, html, other]
-
Title: A conditional quadrature method of moments for populations of oblate spheroidal particlesSubjects: Fluid Dynamics (physics.flu-dyn); Computational Physics (physics.comp-ph)
Non-spherical particle populations are encountered in numerous industrial processes, where particle size, shape, and velocity jointly influence transport, breakage, and classification dynamics. The resulting coupled distributions of geometric and kinematic particle properties evolve within a high-dimensional phase space whose accurate representation remains a significant challenge. We present a conditional quadrature method of moments (CQMOM) for oblate spheroidal particles described by a five-dimensional phase space that includes their size, aspect ratio, and velocity. The dimensionality of the problem is reduced by assuming that the particles continuously adopt their drag-maximizing orientation. In the solution algorithm, the resulting conditional quadrature representation of the particle phase space is combined with a Strang splitting of the continuous and discontinuous mathematical operators governing the moment transport. We verify this CQMOM framework against Monte Carlo simulations of the same particle model for representative test cases of increasing complexity, including drag, breakage, recoil, and attrition, at different Stokes numbers. With two nodes per coordinate, the particle number, the mean size, and the mean aspect ratio agree with the Monte Carlo solution within its statistical uncertainty, and the velocity statistics to within 7 %. Time-step and quadrature resolution studies identify the conditional velocity quadrature as the main source of the remaining deviations, for which we discuss potential remedies. Overall, the results demonstrate that the conditional quadrature representation preserves the geometric-kinematic correlations of the particle population in a five-dimensional phase space, at a cost more than 60 times lower than that of a single Monte Carlo realization.
- [43] arXiv:2611.21541 (cross-list from cs.AI) [pdf, html, other]
-
Title: SciResearch-Eval: Evaluating Scientific Judgment in Autonomous Experimental AgentsComments: 28 pages, 7 figures. Code and leaderboard: this https URLSubjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
We introduce SciResearch-Eval, an interactive benchmark of scientific judgment built on a standard laboratory problem: measuring a material's thermal properties by heating it and fitting a heat-conduction simulator. An agent designs its own experiments, uses a supplied simulator that may be under-resolved or physically wrong, and reports each property with a 95% interval. To our knowledge it is the first benchmark to combine agent-designed experiments, a deliberately flawed simulator, and proper scoring of calibrated intervals against exact ground truth. On a held-out split (93 tasks, 9 models, 2,511 episodes), models handle procedural traps. When the simulator is missing physics, most report wrong values while stating they are 95% sure: their ranges contain the truth in 20-62% of cases, against 94% for GPT-6 Astra and 99% for Claude Fable 5.2. At default effort the failing Claude 5.5 models rarely diagnose the missing physics (Opus in 35% of episodes, against 66% for Astra), and even when they do, their ranges often miss; Astra and Fable design experiments in which the flawed simulator is accurate, such as gentle heating, which keeps Astra calibrated even when it does not name the cause. Asking for wider intervals does not help; stating the strategy improves interval scores 1.6-2.5-fold, replicated on a pre-registered fresh split, but coverage barely moves. A pre-registered effort ablation shows that the correction is latent in the Claude 5.5 models: at maximum effort Opus improves more than tenfold and is no longer detectably behind Astra at default, but only with 13-31 times more reasoning; older and open-weight models gain nothing. Fable 5.2 is the most accurate model at default effort (interval score 1.9 times better than Astra's) but writes 6 times as many tokens, and at Astra's token count it falls short. We release the environment, graders and harness.
Cross submissions (showing 31 of 31 entries)
- [44] arXiv:2605.17243 (replaced) [pdf, other]
-
Title: Phase-Sum Coupling for Gradient-Flow Dynamics in Oscillator Ising MachinesComments: 34 pages, 11 FiguresSubjects: Computational Physics (physics.comp-ph)
Oscillator Ising machines are commonly viewed as dynamical systems that minimize an energy function, yet this interpretation is not generally valid for nonlinear oscillator hardware. Here, we show that conventional phase-difference coupling requires a stringent harmonic-by-harmonic quadrature condition between the oscillator waveform and its phase response. In contrast, symmetric phase-sum interactions intrinsically admit gradient-flow dynamics without this constraint. When combined with second-harmonic injection, the canonical sine phase-sum interaction recovers the Ising Hamiltonian on the binary phase manifold while defining a distinct continuous energy landscape. However, the physical realization of phase-sum coupling can introduce unwanted cross-harmonic interactions that compromise the gradient-flow structure. We derive a Nyquist-like phase-domain sampling criterion that specifies the number of coupling paths required to eliminate these interactions within a prescribed harmonic bandwidth. Guided by this result, we develop a CMOS architecture combining phase-resolved oscillator taps with second-harmonic-modulated coupling. Transistor-level simulations demonstrate an approximately fifteenfold reduction in non-gradient curl compared with conventional phase-difference coupling, while simulations of oscillator networks validate the resulting phase-sum dynamics. These results establish interaction symmetry as an additional design degree of freedom for engineering gradient-flow dynamics in nonlinear oscillator Ising machines.
- [45] arXiv:2607.06421 (replaced) [pdf, html, other]
-
Title: Gradient-Based Inverse Design of Free-Energy Landscapes with Diffusion ModelsSubjects: Computational Physics (physics.comp-ph)
Free-energy surfaces govern the populations of metastable states and the barriers that control transitions between them, making their direct optimization a central challenge in molecular and materials design. In this work, we introduce Gradient-Based Free Energy Surface Optimization (GB-FESO), an inverse design framework that uses a trained conditional diffusion model as a differentiable surrogate for the ensemble distribution. After training, the diffusion model is frozen, and the conditioning variables defining the system are optimized so that the generated ensemble reproduces a prescribed target free-energy surface. The optimization is carried out by backpropagating a distribution-level loss, based on kernel density estimates of the Kullback-Leibler divergence, through a deterministic diffusion sampling trajectory. We first validate GB-FESO on one-dimensional Gaussian ensembles, demonstrating that both continuous and relaxed discrete conditioning variables can be optimized to recover target distributions, including those outside the training domain. We then apply the method to a four-particle Lennard-Jones toy peptide exhibiting multiple metastable conformational states. In this more physically motivated setting, GB-FESO successfully optimizes the interaction parameters to reproduce target free-energy landscapes in the majority of test cases, with optimization performed either in the full internal-coordinate space or in a reduced collective-variable representation. These results establish GB-FESO as a promising first step toward an ensemble-level inverse design framework for molecular systems with prescribed thermodynamic and kinetic behavior.
- [46] arXiv:2607.24574 (replaced) [pdf, html, other]
-
Title: Reframing the coupling force of adaptive resolution simulation in terms of the Liouville-type hierarchy for open systemsComments: To be published in Commun. Appl. Math. Comput. Sci., see this https URLSubjects: Computational Physics (physics.comp-ph); Statistical Mechanics (cond-mat.stat-mech); Chemical Physics (physics.chem-ph)
In this work, we present a novel perspective on the coupling force employed to compensate the interface artifacts prevalent in adaptive resolution simulations (AdResS) of open many-particle systems. We show that a substantial part of this "thermodynamic force" can be framed in terms of the theoretical model of the Liouville-type hierarchy for open systems. The correspondence is made explicit for the case of a simple atomistic fluid, for which a one-dimensional integral expression is derived. This enables the analysis for dependencies of the thermodynamic force on important simulation parameters, which is taken to inspire both simplifications for the numerical calculation of the thermodynamic force and new criteria for its validation that are adequate to the interfacial nature of the problem. The theoretical claims are then verified in a simulation study of the atomistic supercritical Lennard-Jones fluid at different thermodynamic states.
- [47] arXiv:2608.17032 (replaced) [pdf, html, other]
-
Title: Validating direct solvers for Newton's gravitational N-body problem, and the systematic comparison between IEEE floating point and PositsSimon Portegies Zwart (Leiden Observatory)Comments: Computer Physics Communication (accepted for publication)Subjects: Computational Physics (physics.comp-ph); Instrumentation and Methods for Astrophysics (astro-ph.IM); Programming Languages (cs.PL); Chaotic Dynamics (nlin.CD)
We present a systematic comparison between arbitrary precise arithmetic and integration, IEEE-754 compliant floating point arithmetic (fp16, bfp16, fp32, double precision fp64, and quadruple precision fp128), and two implementations of Posits (type III unum) for solving Newton's chaotic N-body problem. Each implementation is benchmarked with arbitrary precise calculations to objectively evaluate their performance in precision as well as speed. We rely on hardware and compiler implementations for fp64, and software implementations for arbitrary-precision arithmetic and Posits. Half precision arithmetic (fp16, bfp16, and Posits$<16,1>$) are insufficiently precise for solving Newton's equations of motion. Single precision (fp32, and Posits$<32,2>$) could be used for statistical ensemble calculations, but lead to relatively large errors in any individual strong encounter. All 64-bit implementations fp64 as well as Posits (Posits$<64,3>$) experience difficulty in our tests. One of the implementations of Posits (Universal) gives precision comparable to fp64 but is slow (by at least an orders of magnitude compared to fp64 after correcting for the more efficient hardware support for the latter). The other (CPPPosits) has a speed comparable to fp64 but has systematically larger errors (by about an order of magnitude compared to fp64 with excesses exceeding two orders of magnitude). As a consequence, this implementation leads to a systematic drift in the result space and has difficulty resolving close encounters. Posits and fp64 have difficulty when integrating a dynamical system in a moving reference frame; testing Galileo invariancy. In their current implementation, Posits do not seem to be the ideal alternative for fp64 when integrating chaotic or stiff ordinary differential equations, such as Newton's equations of motion.
- [48] arXiv:2601.16528 (replaced) [pdf, html, other]
-
Title: Electronic structure, phase stability, and transport properties of the AlTiVCr lightweight high-entropy alloy: A computational studyComments: v1: 14 pages, 9 figures, 1 table. v2: 16 pages, 9 figures, 1 tableJournal-ref: Journal of Materials Research (2026)Subjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
We investigate the thermodynamics and phase stability of the AlTiVCr lightweight high-entropy alloy using a combination of ab initio electronic structure calculations, a concentration wave analysis, and atomistic Monte Carlo simulations. In alignment both with experimental data and with results obtained using other computational approaches, we predict a $\textrm{B2}$ (CsCl) chemical ordering emerging in this alloy at comparatively high temperatures, which is driven by Al and Ti moving to separate sublattices, while V and Cr express weaker site preferences. The impact of this $\textrm{B2}$ chemical ordering on the electronic transport properties of the alloy is investigated within a Kubo--Greenwood linear response framework and it is found that, counter-intuitively, the alloy's residual resistivity increases as the material transitions from the $\textrm{A2}$ (disordered bcc) phase to our predicted $\textrm{B2}$ (partially) ordered structure. Our results give fresh insight into the atomic-scale structure and consequent physical properties of this well-studied, technologically relevant material.
- [49] arXiv:2602.13438 (replaced) [pdf, html, other]
-
Title: Quantum Algorithm Framework for Phase-Contrast Transmission Electron Microscopy Image SimulationComments: 28 pages, 9 figures (including appendices); supplementary code available at this https URLSubjects: Quantum Physics (quant-ph); Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
We present a quantum algorithmic framework for simulating phase-contrast transmission electron microscopy (CTEM) image formation using a fault-tolerant, gate-based quantum circuit model. The electron wavefield on an $N\times N$ grid is amplitude-encoded into a $2\log_2 N$-qubit register. Free-space propagation and objective-lens aberrations are implemented via two-dimensional quantum Fourier transforms (QFTs) and diagonal phase operators in reciprocal space, while specimen interaction is modeled under the phase-object approximation (POA) in a single-slice multislice formulation as a position-dependent phase grating. We validate projected potentials, contrast transfer function (CTF) behavior, and image contrast trends against classical multislice simulations for MoS$_2$ over experimentally relevant parameters, and provide resource estimates and key assumptions that determine end-to-end runtime. Extracting a complete $N\times N$ intensity image by direct sampling requires $O(N^2/\epsilon^2)$ repetitions, which rules out a generic polylogarithmic-cost image-generation algorithm. We therefore analyze end-to-end cost, including state preparation, repeated execution, measurement precision, and the best classical calculation of the same output, and identify the conditions under which limited-output tasks could be candidates for quantum advantage; the present benchmarks establish implementation agreement and do not demonstrate such an advantage. This framework provides a physics-grounded mapping from CTEM theory to quantum circuits and establishes a baseline for extending toward full multislice and inelastic scattering models.
- [50] arXiv:2602.14667 (replaced) [pdf, html, other]
-
Title: Recursive regularisation in lattice Boltzmann magnetohydrodynamicsSubjects: Fluid Dynamics (physics.flu-dyn); Computational Physics (physics.comp-ph)
Recursive regularisation (RR) improves the stability of lattice Boltzmann schemes, but its structure and limitations in magnetohydrodynamics (MHD) remain insufficiently understood. This work connects RR to central-Hermite-moment (CHM) collisions on D2Q9 and D3Q27 through a closed-form reconstruction valid for arbitrary second-order relaxation rates. The central moments of the Maxwell stress coincide, up to sign, with the recursion coefficients, providing a systematic basis for MHD collision operators. Common two-dimensional raw- and central-moment schemes thus correspond to regularisation with a unit bulk relaxation rate. In three dimensions, a CHM implementation reproduces RR to round-off, while a monomial central-moment collision differs at fourth order and produces a four- to twelve-fold larger kinetic-energy error in the tested flow. Chapman--Enskog analysis shows that, with a magnetic field, the hydrodynamic recursion reconstructs the third-order non-equilibrium with a relative error of $78$--$96\%$ ($4\%$ without a field). A local correction largely removes it, yet changes the solutions by less than their discretisation error, consistent with its Burnett-order character. Three-dimensional vortex simulations show approximately second-order convergence at $\mathrm{Re}=100$ and agree with pseudo-spectral solutions through the peak current at $\mathrm{Re}=1000$ and, with a unit bulk rate, at $\mathrm{Re}=3000$. There, single-rate RR develops a compressive instability whose onset depends on the grid, lattice Mach number and initial magnetic configuration. For the standard initial field, a unit bulk rate suppresses it at $\mathrm{Ma}=0.034$ for $N\ge128$. This improvement does not extend to BGK collisions or prevent failure at $\mathrm{Re}=10,000$. These results show that robust three-dimensional MHD simulations require regularisation and bulk dissipation to be considered together.
- [51] arXiv:2604.20384 (replaced) [pdf, html, other]
-
Title: Scalable Hessian-vector products for tensor networks with application to quantum circuit compressionComments: 15 pages + 10 pages appendix, 7 figures + 2 figures in the appendixSubjects: Quantum Physics (quant-ph); Computational Physics (physics.comp-ph)
Optimizing tensor networks with standard first-order methods often leads to slow convergence and getting trapped in local minima. Although second-order optimization offers enhanced robustness, explicitly constructing the full Hessian matrix is computationally prohibitive for large-scale systems. In this work, we bypass this bottleneck by introducing an analytical Hessian-vector product kernel designed for arbitrary compositions of linear maps. This two-pass algorithm leverages a recursive tangent-state propagation with a bounded virtual bond dimension to guarantee scalability. We demonstrate its use within a Riemannian trust-region method for quantum circuit compression. For time-evolution circuits with up to 17 layers, our method reproduces previous results in substantially fewer iterations, reducing wall-time by up to an order of magnitude. For deeper quantum circuits, incorporating second-order derivative information yields a better solution, reducing the final error to less than half that of the first-order result.
- [52] arXiv:2606.14590 (replaced) [pdf, html, other]
-
Title: Lattice dynamics and the spectroscopic signatures of H-bond disorder in $δ$-AlOOHComments: 10 pages, 6 figures; Supplemental Material: 5 pages, 4 figures. Accepted for publication in Physical Review BSubjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph); Geophysics (physics.geo-ph)
Compression of $\delta$-AlOOH produces a multistage evolution of H-bond configurations, average symmetry, and proton centering. Raman and infrared anomalies at 5-10 GPa mark early changes in phonon frequencies and linewidths within this sequence, while static harmonic phonons soften only at much higher pressure. We combine harmonic phonon calculations with strongly constrained and appropriately normed (SCAN)-based deep-potential molecular dynamics and phonon quasiparticle analysis at 300 K. Vibrational spectra show branch reorganization and linewidth growth during H-bond disordering. Loss of coherence in a coupled proton-lattice mode develops before the bulk compression anomaly and the change in average symmetry near 9-10 GPa; full proton centering occurs later, at 16-18 GPa. Quasiparticle renormalization and directional longitudinal-optical--transverse-optical (LO-TO) splitting reproduce the three higher-frequency components of the ambient-pressure OH-stretching Raman multiplet; the lowest-frequency component remains unresolved. Finite-temperature H-bond dynamics and loss of spectral coherence are necessary to explain the early spectroscopic anomalies.
- [53] arXiv:2607.05298 (replaced) [pdf, other]
-
Title: Phase-field modeling of elastically driven abnormal grain growthYazhuo Liu, Yin Zhang, Kunqing Ding, Yichen Yang, Alejandro Barrios, Xavier Maeder, Olivier Pierron, Xing Liu, Ting ZhuComments: Accepted ManuscriptJournal-ref: J. Mech. Phys. Solids 219 (2027) 106913Subjects: Materials Science (cond-mat.mtrl-sci); Applied Physics (physics.app-ph); Computational Physics (physics.comp-ph)
Grain-refined metals offer high strength, yet their engineering applications are often constrained by grain coarsening under mechanical loading. Recent experiments have revealed room-temperature abnormal grain growth (AGG) in ultrafine-grained Ni thin films subjected to cyclic loading. This phenomenon occurs in the regime of macroscopic elastic deformation, unlike conventional AGG, which generally requires substantial plastic deformation or high temperatures. The AGG process is characterized by the preferential growth of grains whose in-plane <100> orientations are aligned with the loading direction. Here, we investigate the underlying physical mechanisms by combining phase-field simulations with micromechanical analysis. The results indicate that a reduction in elastic energy provides a thermodynamic driving force for orientation-selective grain growth. Phase-field simulations reveal the kinetic evolution of AGG and demonstrate that local grain geometry and stress state influence grain growth pathways. By applying this framework to materials with varying degrees of elastic anisotropy, we establish a general approach for investigating elastically driven AGG in polycrystalline systems.
- [54] arXiv:2608.14455 (replaced) [pdf, html, other]
-
Title: Quantifying internal variability in large-ensemble climate data with the Wasserstein distanceSubjects: Atmospheric and Oceanic Physics (physics.ao-ph); Computational Physics (physics.comp-ph)
Estimating forced and internal variability in climate data is fundamental to detecting the climate change signal and assessing uncertainty in climate projections. We propose a metric that quantifies the relative magnitude of internal variability in single-model initial-condition large ensembles (SMILEs). We measure forced and internal variability by two 1-Wasserstein distances, a form of optimal transport cost. The proposed metric is the distance for internal variability divided by the sum of the two. Computing this ratio requires only sorting the samples and differencing the resulting quantiles, with no additional parameters. The metric reflects the entire distribution shape and applies to non-Gaussian variables, because the 1-Wasserstein distance measures the difference between any two probability distributions. We validate the metric with synthetic climate data from Gaussian, uniform, and lognormal distributions. Unlike the two existing metrics, the proposed metric gives stable estimates in the presence of outliers irrespective of the distribution shape, provided that the ensemble has about 40 members or more. We then apply the proposed and existing metrics to the 2 m air temperature and total precipitation of the Community Earth System Model Large Ensemble (CESM-LE) under two different forcing scenarios. All three metrics agree in the spatial pattern of the difference between the scenarios but not in amplitude. The amplitude of the difference is largest for the proposed metric, consistent with the synthetic-data result that this metric changes most rapidly when internal variability dominates. The 1-Wasserstein distance thus provides a simple and useful tool for analyzing large-ensemble datasets.
- [55] arXiv:2609.24785 (replaced) [pdf, html, other]
-
Title: A high-order ghost-point immersed boundary method with auxiliary differential equations for time-domain impedance wallsComments: 19 pages, 9 figuresSubjects: Fluid Dynamics (physics.flu-dyn); Numerical Analysis (math.NA); Computational Physics (physics.comp-ph)
This paper presents a method for performing time-domain simulations of a sound field interacting with a wall having a locally-reacting impedance. The linearized Euler equations are solved using a high-order finite difference method. To impose the impedance condition on a wall that is not aligned with the Cartesian grid, an immersed boundary method (IBM) is employed. It relies on a ghost-point approach with a normal-stencil-based reconstruction, where the impedance is accounted for via auxiliary differential equations (ADEs), which are integrated using the same Runge-Kutta scheme as the governing equations. The method is validated against semi-analytical solutions for a pressure pulse reflecting off a flat or cylindrical wall with an impedance corresponding to a mass-spring-damper system. The cylindrical case solution is derived in the appendix. The measured convergence order is approximately 4 for a flat wall and just slighlty smaller for a curved wall. Long-term stability is verified by the diffraction of a monopole by a cylinder.
- [56] arXiv:2609.28782 (replaced) [pdf, html, other]
-
Title: When a better baseline makes delta learning worse: separating residual scale from learnability to guide target designComments: 16 pages, 4 main-text figures, and 22 pages of Supplementary Information provided as an ancillary PDF. Revised title and exposition, expanded Methods and Supplementary Information, and updated referencesSubjects: Machine Learning (cs.LG); Chemical Physics (physics.chem-ph); Computational Physics (physics.comp-ph)
In scientific machine learning, $\Delta$-learning trains models on residual errors relative to approximate baselines, assuming that smaller residual scales inherently yield better downstream accuracy. Here, we demonstrate that residual scale alone is an insufficient heuristic for learnability. For molecular total energies, replacing a linear baseline with a neural network fitted to the same elemental counts reduced the residual scale, yet more than doubled GemNet's prediction error. Across local structural descriptors, more flexible baselines produced smaller residuals that varied sharply between structurally similar molecules. By contrast, semi-empirical baselines left larger residuals, yet yielded lower test error and improved out-of-domain transferability. Across the targets studied, scale-normalized graph Dirichlet roughness tracked scale-normalized prediction error, suggesting its use as a pre-training diagnostic for target selection. These results motivate baseline complementarity as a target-design principle in scientific machine learning: a useful baseline captures contributions the downstream architecture struggles to predict.
- [57] arXiv:2611.02403 (replaced) [pdf, other]
-
Title: Machine-Learned Interatomic Potential for the MoS$_2$/Sapphire SystemComments: 31 pages, 9 figures, 4 tables; supplementary material (starts on pg. 32): 38 pages, 23 figures, 11 tablesSubjects: Materials Science (cond-mat.mtrl-sci); Computational Physics (physics.comp-ph)
We develop a machine-learned neuroevolution potential (NEP) for the MoS$_2$/ $\alpha$-Al$_2$O$_3$(0001) system. The potential is trained on 63,611 PBE+D3 structures distilled from ~1.1 million curated density-functional-theory (DFT) calculations, using an energy-difference loss function that directly targets interfacial and orientation-dependent energies. On 2,820 held-out test structures from the DFT pool, the energy and force errors are 29 meV/atom and 0.23 eV/angstrom. Validated against DFT on a suite of physical properties, the NEP reproduces the lattice constants within 0.3 %, the elastic constants within 17 %, the phonon dispersions with mean deviations below 2.5 %, and the surface energies within 1.3 %. At the MoS$_2$/sapphire interface, it reproduces the adhesion energy minimum within 0.8 %, the interfacial energy within 2.2 % and the most stable orientation of the monolayer. It preserves the DFT-predicted energetic ordering of point defects (Spearman $\rho$ = 0.995) and mirror-twin boundaries ($\rho$ = 1.0), and reproduces tilt grain-boundary energies within 8 % ($\rho$ = 0.98) and stacking-fault energy barriers within 14 %.
- [58] arXiv:2611.14041 (replaced) [pdf, html, other]
-
Title: Accelerating matrix-free fast diagonalization and preconditioned linear solvers for high-order finite element discretized elliptic problems on distributed GPUsComments: 69 pages (26 pages main manuscript, 43 pages Supplementary information)Subjects: Numerical Analysis (math.NA); Computational Physics (physics.comp-ph)
The time to solution of an iterative solver depends on the number of iterations and the cost per iteration. This paper develops algorithms that reduce both for elliptic problems discretized with high-order continuous finite elements on distributed GPUs, implemented in a unified matrix-free code supporting CUDA, HIP and SYCL. A distributed direct fast-diagonalization (FD) Poisson solver provides exact solutions on tensor-product meshes with direction-wise grading and coefficients varying along one coordinate direction. Its tensor data remain distributed, moved between slab layouts using collective transposes or kept in place by a Schur-complement approach. A two-level hybrid Schwarz preconditioner with FD cell solves (HS-FD) is extended to 2:1 hanging-node octrees with exact local cell inverses, while a p-multigrid preconditioner with Chebyshev--Jacobi smoothing (CJ-pMG) resolves hanging-node constraints at every level within fused kernels. Both approaches exploit FP32, where cell shapes and coefficients allow, within an FP64 conjugate-gradient iteration, adaptively select replicated or distributed coarse levels, support block solves for multiple right-hand sides and determine all parameters from the operator without case-specific tuning. Comparisons with Jacobi-preconditioned conjugate gradients (Jacobi PCG), MFEM, nekRS and this http URL span uniform, graded and octree meshes, layered coefficients, and four applications across NVIDIA, AMD and Intel GPUs. The direct solver is two to three orders of magnitude faster than Jacobi PCG on uniform meshes. HS-FD converges in 11 iterations for the generalised Poisson problem on octree meshes across 8 to 1024 MI250X graphics compute dies. These results provide practical guidelines for solver selection based on mesh structure and coefficient characteristics.
- [59] arXiv:2611.14562 (replaced) [pdf, html, other]
-
Title: A Neural Physics Solver for Richards EquationKatherine MacKenzie, James Rand, Ben Ashby, Tristan Pryer, Claire E. Heaney, Christopher C. Pain, Thomas Rodding KjeldsenComments: 36 pages (including appendix), 7 figures, 3 tablesSubjects: Computational Engineering, Finance, and Science (cs.CE); Numerical Analysis (math.NA); Computational Physics (physics.comp-ph)
A new approach to discretise and solve Richards equation based on the finite element method (FEM) is presented. The novelty of the approach, called the Neural Physics method, is that it expresses the discretisation as a convolution neural network (CNN) whose weights are predetermined by the integrals of the basis functions in the FEM. The Neural Physics method is physically consistent as it represents a discretisation of the governing equation and can run on CPUs, GPUs and the latest AI accelerators, resulting in an increase in computational speed. The Neural Physics method is demonstrated for two different discretisations of Richards equation, including a multigrid method that is realised through a U-Net architecture. The accuracy and speed of these methods is tested on a problem with a known analytical solution and compared to a standard FEM, and shows that the Neural Physics method is almost an order of magnitude faster than a comparable FEM. Finally, the Neural Physics method is applied to a real-world test case with integrated experimental data and simulates the infiltration of water through a rain garden. This new approach therefore provides a route to efficiently and accurately solve Richards equation for a wide array of hydrological applications.
- [60] arXiv:2611.15759 (replaced) [pdf, html, other]
-
Title: CCGeant4: charge-state-resolved transport and energy-loss fluctuations of relativistic heavy ions in Geant4Subjects: Instrumentation and Detectors (physics.ins-det); Instrumentation and Methods for Astrophysics (astro-ph.IM); Computational Physics (physics.comp-ph)
CCGeant4 models energy-loss fluctuations of heavy ions in Geant4, accounting for both charge exchange and relativistic corrections to electron scattering beyond the first Born approximation. We use GLOBAL cross sections for electron capture and loss, ATIMA equilibrium stopping powers and charge-dependent delta-electron spectra calculated from the Dirac equation in a frozen-shell potential. The package also supports tabulated elastic ion-nucleus scattering. The silicon example includes interference between Coulomb and nuclear scattering amplitudes. We compare charge evolution with GLOBAL and energy-loss spectra with MOCADI for matched beams and targets. Experimental comparisons cover published energy-loss measurements and several measurements of energy deposited by xenon ions in silicon. For uranium at 800 MeV/u, the energy-loss variances from our Dirac configuration agree with published screened-ion calculations to within 4% in Be, Al and Pb. Both calculations lie within the reported experimental uncertainties for all three targets, while calculated bismuth spectra in silver are narrower than measured. Calculations in thin silicon layers show that delta-electron escape can make deposited-energy distributions substantially narrower than primary-ion energy-loss distributions, highlighting the importance of secondary-electron transport for detector-response modelling. The distribution includes the Geant4 source patch, the transport tables and their converters, and a runnable example.
- [61] arXiv:2611.17037 (replaced) [pdf, other]
-
Title: UV and deep-UV ultrafast harmonic generation enhanced by resonant gold nano-pillar metasurfaceShroddha Mukhopadhyay, Yuxuan Chen, Hooman Barati Sedeh, Natalia M. Litchinister, Michael Scalora, Jose Trull, Maria Antonietta Vincenti, Crina CojocaruComments: 14 pages, 6 figuresSubjects: Optics (physics.optics); Computational Physics (physics.comp-ph)
We report experimental observations and full-vector numerical simulations of second and third-harmonic generation from a two-dimensional periodic gold nano-pillar metasurface, for efficient ultrafast light conversion from near-infrared to the UV and deep-UV spectral ranges. The metasurface is designed to support an angle-tunable plasmonic resonance in the near-infrared spectral range, with its resonance wavelength shifting as a function of the pump angle of incidence. Under resonant conditions, strong electromagnetic-field confinement near the nano-pillar corners enhances the nonlinear light-matter interactions. Compared with a flat gold mirror, the nanostructured metal exhibits enhanced nonlinear frequency conversion, with conversion efficiencies of second and third-harmonic generation increased by two and three orders of magnitude, respectively. Full-wave theoretical modeling of linear and nonlinear light-matter interaction in nanoscale structures validates the experimental results. We demonstrate that periodically arranged square gold pillars can be engineered to achieve enhanced second and third-harmonic generation at wavelengths that can be tailored through the structural design, without requiring complex geometries or hybrid structures that pose fabrication challenges. The close agreement between theoretical predictions and experimental measurements further advances our understanding of surface-mediated nonlinear optical processes in metals and highlights the potential of metallic nanostructures as ultrafast light sources in the UV and deep-UV.
- [62] arXiv:2611.17655 (replaced) [pdf, html, other]
-
Title: Foundation Tensor Networks Generalize Across Lattice HamiltoniansComments: 7 Pages, 4 FiguresSubjects: Statistical Mechanics (cond-mat.stat-mech); Computational Physics (physics.comp-ph)
Neural quantum states and tensor networks have emerged as leading techniques for simulating quantum many-body systems. However, disordered systems remain extremely challenging in both frameworks as they require computation of the ground state for every realization of the disorder. Recently, neural quantum states have been used to construct foundation models that generate the ground state directly from the Hamiltonian. Still, they are hindered by the need to perform extensive sampling to estimate observables. To address these difficulties, we introduce the foundation tensor network (FTN), which uses a neural network to generate the tensor network representation of the ground-state wavefunction from the Hamiltonian. To accelerate training, we develop a technique, called \texttt{GridBP}, based on iterated one-dimensional belief propagation that reduces the cost of computing the energy expectation value to just two tensor contractions. We use FTNs to calculate the critical parameters of the glass transition in the quantum Edwards-Anderson model via finite-size scaling analysis. To facilitate rapid averaging, we use a moment-generating function to evaluate the order parameter, lowering the cost to a single tensor contraction. We obtain high-fidelity estimates for the critical parameters by averaging over 500,000 disorder realizations.