arXiv subjects

Peter Wirnsberger

Publications and source records attributed to Peter Wirnsberger.

13 recordsLinked to original sources

arXiv2024

Target Score Matching

Denoising Score Matching estimates the score of a noised version of a target distribution by minimizing a regression loss and is widely used to train the popular class of Denoising Diffusion Models. A well known limitation of Denoising Score Matching, however, is that it yields poor estimates of the score at low noise levels. This issue is particularly unfavourable for problems in the physical sciences and for Monte Carlo sampling tasks for which the score of the clean original target is known. Intuitively, estimating the score of a slightly noised version of the target should be a simple task in such cases. In this paper, we address this shortcoming and show that it is indeed possible to leverage knowledge of the target score. We present a Target Score Identity and corresponding Target Score Matching regression loss which allows us to obtain score estimates admitting favourable properties at low noise levels.

cs.LG↗

arXiv2023

Estimating Gibbs free energies via isobaric-isothermal flows

We present a machine-learning model based on normalizing flows that is trained to sample from the isobaric-isothermal ensemble. In our approach, we approximate the joint distribution of a fully-flexible triclinic simulation box and particle coordinates to achieve a desired internal pressure. This novel extension of flow-based sampling to the isobaric-isothermal ensemble yields direct estimates of Gibbs free energies. We test our NPT-flow on monatomic water in the cubic and hexagonal ice phases and find excellent agreement of Gibbs free energies and other observables compared with established baselines.

physics.comp-ph↗

arXiv2023

GraphCast: Learning skillful medium-range global weather forecasting

Global medium-range weather forecasting is critical to decision-making across many social and economic domains. Traditional numerical weather prediction uses increased compute resources to improve forecast accuracy, but cannot directly use historical weather data to improve the underlying model. We introduce a machine learning-based method called "GraphCast", which can be trained directly from reanalysis data. It predicts hundreds of weather variables, over 10 days at 0.25 degree resolution globally, in under one minute. We show that GraphCast significantly outperforms the most accurate operational deterministic systems on 90% of 1380 verification targets, and its forecasts support better severe event prediction, including tropical cyclones, atmospheric rivers, and extreme temperatures. GraphCast is a key advance in accurate and efficient weather forecasting, and helps realize the promise of machine learning for modeling complex dynamical systems.

cs.LG↗

arXiv2022

MultiScale MeshGraphNets

In recent years, there has been a growing interest in using machine learning to overcome the high cost of numerical simulation, with some learned models achieving impressive speed-ups over classical solvers whilst maintaining accuracy. However, these methods are usually tested at low-resolution settings, and it remains to be seen whether they can scale to the costly high-resolution simulations that we ultimately want to tackle. In this work, we propose two complementary approaches to improve the framework from MeshGraphNets, which demonstrated accurate predictions in a broad range of physical systems. MeshGraphNets relies on a message passing graph neural network to propagate information, and this structure becomes a limiting factor for high-resolution simulations, as equally distant points in space become further apart in graph space. First, we demonstrate that it is possible to learn accurate surrogate dynamics of a high-resolution system on a much coarser mesh, both removing the message passing bottleneck and improving performance; and second, we introduce a hierarchical approach (MultiScale MeshGraphNets) which passes messages on two different resolutions (fine and coarse), significantly improving the accuracy of MeshGraphNets while requiring less computational resources.

cs.LG↗

arXiv2022

Normalizing flows for atomic solids

We present a machine-learning approach, based on normalizing flows, for modelling atomic solids. Our model transforms an analytically tractable base distribution into the target solid without requiring ground-truth samples for training. We report Helmholtz free energy estimates for cubic and hexagonal ice modelled as monatomic water as well as for a truncated and shifted Lennard-Jones system, and find them to be in excellent agreement with literature values and with estimates from established baseline methods. We further investigate structural properties and show that the model samples are nearly indistinguishable from the ones obtained with molecular dynamics. Our results thus demonstrate that normalizing flows can provide high-quality samples and free energy estimates without the need for multi-staging.

physics.comp-ph↗

arXiv2021

SyMetric: Measuring the Quality of Learnt Hamiltonian Dynamics Inferred from Vision

A recently proposed class of models attempts to learn latent dynamics from high-dimensional observations, like images, using priors informed by Hamiltonian mechanics. While these models have important potential applications in areas like robotics or autonomous driving, there is currently no good way to evaluate their performance: existing methods primarily rely on image reconstruction quality, which does not always reflect the quality of the learnt latent dynamics. In this work, we empirically highlight the problems with the existing measures and develop a set of new measures, including a binary indicator of whether the underlying Hamiltonian dynamics have been faithfully captured, which we call Symplecticity Metric or SyMetric. Our measures take advantage of the known properties of Hamiltonian dynamics and are more discriminative of the model's ability to capture the underlying dynamics than reconstruction error. Using SyMetric, we identify a set of architectural choices that significantly improve the performance of a previously proposed model for inferring latent dynamics from pixels, the Hamiltonian Generative Network (HGN). Unlike the original HGN, the new HGN++ is able to discover an interpretable phase space with physically meaningful latents on some datasets. Furthermore, it is stable for significantly longer rollouts on a diverse range of 13 datasets, producing rollouts of essentially infinite length both forward and backwards in time with no degradation in quality on a subset of the datasets.

stat.ML↗

arXiv2021

Which priors matter? Benchmarking models for learning latent dynamics

Learning dynamics is at the heart of many important applications of machine learning (ML), such as robotics and autonomous driving. In these settings, ML algorithms typically need to reason about a physical system using high dimensional observations, such as images, without access to the underlying state. Recently, several methods have proposed to integrate priors from classical mechanics into ML models to address the challenge of physical reasoning from images. In this work, we take a sober look at the current capabilities of these models. To this end, we introduce a suite consisting of 17 datasets with visual observations based on physical systems exhibiting a wide range of dynamics. We conduct a thorough and detailed comparison of the major classes of physically inspired methods alongside several strong baselines. While models that incorporate physical priors can often learn latent spaces with desirable properties, our results demonstrate that these methods fail to significantly improve upon standard techniques. Nonetheless, we find that the use of continuous and time-reversible dynamics benefits models of all classes.

stat.ML↗

arXiv2020

Targeted free energy estimation via learned mappings

Free energy perturbation (FEP) was proposed by Zwanzig more than six decades ago as a method to estimate free energy differences, and has since inspired a huge body of related methods that use it as an integral building block. Being an importance sampling based estimator, however, FEP suffers from a severe limitation: the requirement of sufficient overlap between distributions. One strategy to mitigate this problem, called Targeted Free Energy Perturbation, uses a high-dimensional mapping in configuration space to increase overlap of the underlying distributions. Despite its potential, this method has attracted only limited attention due to the formidable challenge of formulating a tractable mapping. Here, we cast Targeted FEP as a machine learning problem in which the mapping is parameterized as a neural network that is optimized so as to increase overlap. We develop a new model architecture that respects permutational and periodic symmetries often encountered in atomistic simulations and test our method on a fully-periodic solvation system. We demonstrate that our method leads to a substantial variance reduction in free energy estimates when compared against baselines, without requiring any additional data.

physics.comp-ph↗

arXiv2019

Microscopic analysis of thermo-orientation in systems of off-centre Lennard-Jones particles

When fluids of anisotropic molecules are placed in temperature gradients, the molecules may align themselves along the gradient: this is called thermo-orientation. We discuss the theory of this effect in a fluid of particles that interact by a spherically symmetric potential, where the particles' centres of mass do not coincide with their interaction centres. Starting from the equations of motion of the molecules, we show how a simple assumption of local equipartition of energy can be used to predict the thermo-orientation effect, recovering the result of Wirnsberger et al. [Phys. Rev. Lett. 120, 226001 (2018)]. Within this approach, we show that for particles with a single interaction centre, the thermal centre of the molecule must coincide with the interaction centre. The theory also explains the coupling between orientation and kinetic energy that is associated with this non-Boltzmann distribution. We discuss deviations from this local equipartition assumption, showing that these can occur in linear response to a temperature gradient. We also present numerical simulations showing significant deviations from the local equipartition predictions, which increase as the centre of mass of the molecule is displaced further from its interaction centre.

cond-mat.stat-mech↗

arXiv2018

Theoretical prediction of thermal polarisation

We present a mean-field theory to explain the thermo-orientation effect in an off-centre Stockmayer fluid. This effect is the underlying cause of thermally induced polarisation and thermally induced monopoles, which have recently been predicted theoretically. Unlike previous theories that are based either on phenomenological equations or on scaling arguments, our approach does not require any fitting parameters. Given an equation of state and assuming local equilibrium, we construct an effective Hamiltonian for the computation of local Boltzmann averages. This simple theoretical treatment predicts molecular orientations that are in very good agreement with simulation results for the range of dipole strengths investigated. By decomposing the overall alignment into contributions from the temperature and density gradients, we shed further light on how the non-equilibrium result arises from the competition between the two gradients.

cond-mat.stat-mech↗

arXiv2018

Controlling cargo trafficking in multicomponent membranes

Biological membranes typically contain a large number of different components dispersed in small concentrations in the main membrane phase, including proteins, sugars, and lipids of varying geometrical properties. Most of these components do not bind the cargo. Here, we show that such `inert' components can be crucial for precise control of cross-membrane trafficking. Using a statistical mechanics model and molecular dynamics simulations, we demonstrate that the presence of inert membrane components of small isotropic curvatures dramatically influences cargo endocytosis, even if the total spontaneous curvature of such a membrane remains unchanged. Curved lipids, such as cholesterol, as well as asymmetrically included proteins and tethered sugars can hence all be actively participating in controlling membrane trafficking of nanoscopic cargo. We find that even a low-level expression of curved inert membrane components can determine the membrane selectivity towards the cargo size, and can be used to selectively target membranes of certain compositions. Our results suggest a robust and general way to control cargo trafficking by adjusting the membrane composition without needing to alter the concentration of receptors nor the average membrane curvature. This study indicates that cells can prepare for any trafficking event by incorporating curved inert components in either of the membrane leaflets.

cond-mat.soft↗

arXiv2016

Numerical Evidence for Thermally Induced Monopoles

Electrical charges are conserved. The same would be expected to hold for magnetic charges, yet magnetic monopoles have never been observed. It is therefore surprising that the laws of non-equilibrium thermodynamics, combined with Maxwell's equations, suggest that colloidal particles heated or cooled in certain polar or paramagnetic solvents may behave as if they carry an electrical/magnetic charge [J. Phys. Chem. B $\textbf{120}$, 5987 (2016)]. Here we present numerical simulations that show that the field distribution around a pair of such heated/cooled colloidal particles agrees quantitatively with the theoretical predictions for a pair of oppositely charged electrical or magnetic monopoles. However, in other respects, the non-equilibrium colloids do not behave as monopoles: they cannot be moved by a homogeneous applied field. The numerical evidence for the monopole-like fields around heated/cooled colloids is crucial because the experimental and numerical determination of forces between such colloids would be complicated by the presence of other effects, such as thermophoresis.

physics.comp-ph↗

arXiv2016

Non-equilibrium simulations of thermally induced electric fields in water

Using non-equilibrium molecular dynamics simulations, it has been recently demonstrated that water molecules align in response to an imposed temperature gradient, resulting in an effective electric field. Here, we investigate how thermally induced fields depend on the underlying treatment of long-ranged interactions. For the short-ranged Wolf method and Ewald summation, we find the peak strength of the field to range between $2 \times 10^7$ and $5 \times 10^7~\text{V/m}$ for a temperature gradient of $5.2~\text{K/}\unicode{x212B}$. Our value for the Wolf method is therefore an order of magnitude lower than the literature value [J. Chem. Phys. 139, 014504 (2013) and 143, 036101 (2015)]. We show that this discrepancy can be traced back to the use of an incorrect kernel in the calculation of the electrostatic field. More seriously, we find that the Wolf method fails to predict correct molecular orientations, resulting in dipole densities with opposite sign to those computed using Ewald summation. By considering two different multipole expansions, we show that, for inhomogeneous polarisations, the quadrupole contribution can be significant and even outweigh the dipole contribution to the field. Finally, we propose a more accurate way of calculating the electrostatic potential and the field. In particular, we show that averaging the microscopic field analytically to obtain the macroscopic Maxwell field reduces the error bars by up to an order of magnitude. As a consequence, the simulation times required to reach a given statistical accuracy decrease by up to two orders of magnitude.

physics.comp-ph↗