Pith. sign in

physics.comp-ph

Computational Physics

All aspects of computational science applied to physics.

Papers reviewed in the last 7 days lead, then the papers readers actually read. Ranking is not a quality score.

sort pith recommended most recent

One training loop prunes neural nets and quantifies their uncertainty

On a mechanics benchmark, active parameters drop from 1020 to as few as 17 while accuracy holds.

· “Condensed Stein Variational Gradient Descent for Uncertainty Quantification of Neural Networks”

open re-runnable review →
Figure from the paper

Computing nonlocal kinetic energy once speeds orbital-free DFT by 100x

Accurate nonlocal functionals become practical for large-scale simulations with accuracy nearly unchanged.

· “Fast and stable tight-binding framework for nonlocal kinetic energy density functional reconstruction in orbital-free density functional calculations”

open re-runnable review →
Figure from the paper

Sharp-interface FSI solver matches fine-grid results on coarse grids

A geometric volume-of-fluid method keeps interfaces sharp and stable, so 128x128 runs rival 1024x1024 diffusive solvers.

· “Monolithic framework to simulate fluid-structure interaction problems using geometric volume-of-fluid method”

open re-runnable review →

TeNeS-v2 adds real-time and thermal simulation to 2D tensor-network solver

One software package now handles ground states, quench dynamics, and finite-temperature benchmarks against quantum Monte Carlo.

· “TeNeS-v2: Enhancement for Real-Time and Finite Temperature Simulations of Quantum Many-Body Systems”

open re-runnable review →
Figure from the paper

Coupling two particle methods predicts surface temperature within 4.4%

Swapping DSMC's simplified surface model for atomistic molecular dynamics in chosen cells cuts mean error from 730% to 4.4%.

· “A Concurrent Multiscale Framework Coupling Direct Simulation Monte Carlo and Molecular Dynamics”

open re-runnable review →
Figure from the paper

Diffusion-generated weights speed up seismic wavefield PINNs

A latent diffusion model pretrained on 2600 velocity models produces starting parameters that beat random and meta-learned initializations.

· “DiffPINN: Generative diffusion-initialized physics-informed neural networks for accelerating seismic wavefield representation”

open re-runnable review →
Figure from the paper

Onshore oil terminal spreads simulated spills farther into Texas bays

Harbor Island spills reach seashore and inner bays; offshore releases mostly stay offshore in every scenario modeled.

· “Potential Effects of Loading Terminal Locations on Surface Trajectories of Oil Spill Transport”

open re-runnable review →

Jacobian-free multigrid makes implicit DG weather solvers competitive

On fine grids it runs about twice as fast as explicit time stepping while taking steps roughly 500 times larger.

· “Jacobian-free Multigrid Preconditioner for Discontinuous Galerkin Methods applied to Numerical Weather Prediction”

open re-runnable review →

Hybrid model predicts hypersonic heat shield temperatures 100x faster

Trained on thermocouple readings alone, it extrapolates to new heat loads and materials with under 1% error.

· “Physics-Infused Reduced-Order Modeling for Analysis of Multi-Layered Hypersonic Thermal Protection Systems”

open re-runnable review →

VLE errors, not heat data, decide dividing wall column purity loss

Worst-case losses grow with stage count, and the medium boiler's vapor pressure is the decisive property, across three ideal mixtures.

· “Influences of Uncertainties in Thermodynamic Models on Pareto-optimized Dividing Wall Columns for Ideal Mixtures”

open re-runnable review →
Figure from the paper

Spline networks fit rubber data and predict untested biaxial response

Sparsified splines become compact symbolic strain-energy laws without losing accuracy or extrapolation.

· “Constitutive Kolmogorov-Arnold Networks (CKANs): Combining Accuracy and Interpretability in Data-Driven Material Modeling”

open re-runnable review →
Figure from the paper

QuantumDNA predicts how DNA mutations shift exciton lifetimes

Open-source package combines LCAO couplings, tight-binding models, and Lindblad noise for fast DNA screening.

· “QuantumDNA: A Python Package for Analyzing Quantum Charge Dynamics in DNA and Exploring Its Biological Relevance”

open re-runnable review →
Figure from the paper

Bias flips oxygen reduction pathway on doped graphene

Solvent-mediated vs adsorbed electron transfer in explicit atomistic simulations of the ORR.

· “Modeling Reactions on the Solid-Liquid Interface With Next Generation Extended Lagrangian Quantum-Based Molecular Dynamics”

open re-runnable review →
Figure from the paper

Aging detected in velocity correlations of heated granular gas

Slower decay with longer waiting time shows a 3D dissipative gas remembers its past velocities.

· “Aging Dynamics and Velocity Field Correlations in Three-Dimensional Uniformly Heated Granular Gases: A Molecular Dynamics Study”

open re-runnable review →
Figure from the paper

In-situ GPU workflow captures Rayleigh-Bénard convection at Ra 10^12

A 46.7-billion-point simulation on 3360 GPUs renders boundary-layer turbulence every 100 steps at 42 percent overhead.

· “Deciphering boundary layer dynamics in high-Rayleigh-number convection using 3360 GPUs and a high-scaling in-situ workflow”

open re-runnable review →
Figure from the paper

Yes, fixed-node DMC is reproducible across 11 codes

With T-move, DLA, or DTM schemes, eleven codes agree to ~3 meV; the old locality approximation spreads ~25 meV.

· “Reproducibility of fixed-node diffusion Monte Carlo across diverse community codes: The case of water-methane dimer”

open re-runnable review →
Figure from the paper

Crystal diffusion lands 0.005 Å from DFT ground states

It also steers crystal symmetry, so users can request a target point group for inverse materials design.

· “CrystalGRW: Generative Modeling of Crystal Structures with Targeted Properties via Geodesic Random Walks”

open re-runnable review →
Figure from the paper

WeTICA recovers unfolding kinetics at a tenth of the event time

The binless method matches reported microsecond unfolding rates using only nanosecond-scale cumulative simulation time.

· “WeTICA: A directed search weighted ensemble based enhanced sampling method to estimate rare event kinetics in a reduced dimensional space”

open re-runnable review →
Figure from the paper

browse all of physics.comp-ph → full archive · search · sub-categories