arXiv subjects

Franco Pellegrini

Publications and source records attributed to Franco Pellegrini.

At least 19 recordsLinked to original sources

arXiv2026

A fast summation method for the DFT-D3 dispersion correction

The DFT-D3 dispersion correction is routinely added to machine learning force fields (MLFFs) trained on dispersion-deficient functionals such as PBE. Its environment-dependent pair coefficients, however, break the atom-centered separability that fast summation methods require, forcing practitioners either to truncate D3 or to accept a substantial slowdown. We introduce FourierD3, a method that uses a functional low-rank decomposition to restore this separability and enable particle-mesh evaluation in $O(N\log N)$ time without a real-space cutoff on the dispersion sum.

physics.comp-ph↗

arXiv2025

Molecular simulations of Perovskites CsXI3 (X = Pb,Sn) Using Machine-Learning Interatomic Potentials

Cesium based halide perovskites, such as CsPbI3 and CsSnI3, have emerged as exceptional candidates for next generation photovoltaic and optoelectronic technologies, but their practical application is limited by temperature dependent phase transitions and structural instabilities. Here, we develop machine learning interatomic potentials within the LATTE framework to simulate these materials with near experimental accuracy at a fraction of the computational cost compared to previous computational studies. Our molecular dynamics simulations based on the trained MLIPs reproduce energies and forces across multiple phases, enabling large scale simulations that capture cubic tetragonal orthorhombic transitions, lattice parameters, and octahedral tilting with unprecedented resolution. We find that Pb based perovskites exhibit larger octahedral tilts and higher phase transition temperatures than Sn based analogues, reflecting stronger bonding and enhanced structural stability, whereas Sn based perovskites display reduced tilts and lower barriers, suggesting tunability through compositional or interface engineering. Beyond these systems, our work demonstrates that MLIPs can bridge first principles accuracy with simulation efficiency, providing a robust framework for exploring phase stability, anharmonicity, and rational design in next generation halide perovskites.

cond-mat.mtrl-sci↗

arXiv2025

Convergence of Body-Orders in Linear Atomic Cluster Expansions

We study the convergence of a linear atomic cluster expansion (ACE) potential with respect to its basis functions, in terms of the effective two-body interactions of elemental Carbon and Silicon systems. We build ACE potentials with descriptor sets truncated at body-orders $K=2$ to $K=5$ trained on a diverse Carbon dataset and on Silicon dimers to pentamers. The potentials trained on diverse structures with standard ACE bases are not able to recover the correct dimer curves much less produce stable curves or solutions. While employing ACE bases removed of self-interactions still does not generalize to the DFT-expected results, properly tailored datasets and basis sets are able to show signs of convergence and stability in the curves and expansions, suggesting a method to build potentials with interpretable bases with respect to the cluster expansion.

physics.comp-ph↗

arXiv2025

Efficient Molecular Conformer Generation with SO(3)-Averaged Flow Matching and Reflow

Fast and accurate generation of molecular conformers is desired for downstream computational chemistry and drug discovery tasks. Currently, training and sampling state-of-the-art diffusion or flow-based models for conformer generation require significant computational resources. In this work, we build upon flow-matching and propose two mechanisms for accelerating training and inference of generative models for 3D molecular conformer generation. For fast training, we introduce the SO(3)-Averaged Flow training objective, which leads to faster convergence to better generation quality compared to conditional optimal transport flow or Kabsch-aligned flow. We demonstrate that models trained using SO(3)-Averaged Flow can reach state-of-the-art conformer generation quality. For fast inference, we show that the reflow and distillation methods of flow-based models enable few-steps or even one-step molecular conformer generation with high quality. The training techniques proposed in this work show a path towards highly efficient molecular conformer generation with flow-based models.

cs.LG↗

arXiv2025

Optimizing Data Distribution and Kernel Performance for Efficient Training of Chemistry Foundation Models: A Case Study with MACE

Chemistry Foundation Models (CFMs) that leverage Graph Neural Networks (GNNs) operating on 3D molecular graph structures are becoming indispensable tools for computational chemists and materials scientists. These models facilitate the understanding of matter and the discovery of new molecules and materials. In contrast to GNNs operating on a large homogeneous graphs, GNNs used by CFMs process a large number of geometric graphs of varying sizes, requiring different optimization strategies than those developed for large homogeneous GNNs. This paper presents optimizations for two critical phases of CFM training: data distribution and model training, targeting MACE - a state-of-the-art CFM. We address the challenge of load balancing in data distribution by formulating it as a multi-objective bin packing problem. We propose an iterative algorithm that provides a highly effective, fast, and practical solution, ensuring efficient data distribution. For the training phase, we identify symmetric tensor contraction as the key computational kernel in MACE and optimize this kernel to improve the overall performance. Our combined approach of balanced data distribution and kernel optimization significantly enhances the training process of MACE. Experimental results demonstrate a substantial speedup, reducing per-epoch execution time for training from 12 to 2 minutes on 740 GPUs with a 2.6M sample dataset.

cs.DC↗

arXiv2024

Can Neural Networks Learn Nanoscale Friction?

Current nanofriction experiments on crystals, both tip-on-surface and surface-on-surface, provide force traces as their sole output, typically exhibiting atomic size stick-slip oscillations. Physically interpreting these traces is a task left to the researcher. Historically done by hand, it generally consists in identifying the parameters of a Prandtl-Tomlinson (PT) model that best reproduces these traces. This procedure is both work-intensive and quite uncertain. We explore in this work how machine learning (ML) could be harnessed to do that job with optimal results, and minimal human work. A set of synthetic force traces is produced by PT model simulations covering a large span of parameters, and a simple neural network (NN) perceptron is trained with it. Once this trained NN is fed with experimental force traces, it will ideally output the PT parameters that best approximate them. By following this route step by step, we encountered and solved a variety of problems which proved most instructive and revealing. In particular, and very importantly, we met unexpected inaccuracies with which one or another parameter was learned by the NN. The problem, we then show, could be eliminated by proper manipulations and augmentations operated on the training force traces, and that without extra efforts and without injecting experimental informations. Direct application to the sliding of a graphene coated AFM tip on a variety of 2D materials substrates validates and encourages use of this ML method as a ready tool to rationalise and interpret future stick-slip nanofriction data.

cond-mat.mes-hall↗

arXiv2024

EquiJump: Protein Dynamics Simulation via SO(3)-Equivariant Stochastic Interpolants

Mapping the conformational dynamics of proteins is crucial for elucidating their functional mechanisms. While Molecular Dynamics (MD) simulation enables detailed time evolution of protein motion, its computational toll hinders its use in practice. To address this challenge, multiple deep learning models for reproducing and accelerating MD have been proposed drawing on transport-based generative methods. However, existing work focuses on generation through transport of samples from prior distributions, that can often be distant from the data manifold. The recently proposed framework of stochastic interpolants, instead, enables transport between arbitrary distribution endpoints. Building upon this work, we introduce EquiJump, a transferable SO(3)-equivariant model that bridges all-atom protein dynamics simulation time steps directly. Our approach unifies diverse sampling methods and is benchmarked against existing models on trajectory data of fast folding proteins. EquiJump achieves state-of-the-art results on dynamics simulation with a transferable model on all of the fast folding proteins.

cs.LG↗

arXiv2024

LATTE: an atomic environment descriptor based on Cartesian tensor contractions

We propose a new descriptor for local atomic environments, to be used in combination with machine learning models for the construction of interatomic potentials. The Local Atomic Tensors Trainable Expansion (LATTE) allows for the efficient construction of a variable number of many-body terms with learnable parameters, resulting in a descriptor that is efficient, expressive, and can be scaled to suit different accuracy and computational cost requirements. We compare this new descriptor to existing ones on several systems, showing it to be competitive with very fast potentials at one end of the spectrum, and extensible to an accuracy close to the state of the art.

physics.comp-ph↗

arXiv2023

Neural Network Interatomic Potentials For Open Surface Nano-mechanics Applications

Material characterization in nano-mechanical tests requires precise interatomic potentials for the computation of atomic energies and forces with near-quantum accuracy. For such purposes, we develop a robust neural-network interatomic potential (NNIP), and we provide a test for the example of molecular dynamics (MD) nanoindentation, and the case of body-centered cubic crystalline molybdenum (Mo). We employ a similarity measurement protocol, using standard local environment descriptors, to select ab initio configurations for the training dataset that capture the behavior of the indented sample. We find that it is critical to include generalized stacking fault (GSF) configurations, featuring a dumbbell interstitial on the surface, to capture dislocation cores, and also high-temperature configurations with frozen atom layers for the indenter tip contact. We develop a NNIP with distinct dislocation nucleation mechanisms, realistic generalized stacking fault energy (GSFE) curves, and an informative energy landscape for the atoms on the sample surface during nanoindentation. We compare our NNIP results with nanoindentation simulations, performed with three existing potentials -- an embedded atom method (EAM) potential, a gaussian approximation potential (GAP), and a tabulated GAP (tabGAP) potential -- that predict different dislocation nucleation mechanisms, and display the absence of essential information on the shear stress at the sample surface in the elastic region. We believe that these features render specialized NNIPs essential for simulations of nanoindentation and nano-mechanics with near-quantum accuracy.

cond-mat.mtrl-sci↗

arXiv2023

PANNA 2.0: Efficient neural network interatomic potentials and new architectures

We present the latest release of PANNA 2.0 (Properties from Artificial Neural Network Architectures), a code for the generation of neural network interatomic potentials based on local atomic descriptors and multilayer perceptrons. Built on a new back end, this new release of PANNA features improved tools for customizing and monitoring network training, better GPU support including a fast descriptor calculator, new plugins for external codes and a new architecture for the inclusion of long-range electrostatic interactions through a variational charge equilibration scheme. We present an overview of the main features of the new code, and several benchmarks comparing the accuracy of PANNA models to the state of the art, on commonly used benchmarks as well as richer datasets.

physics.comp-ph↗

arXiv2021

Sifting out the features by pruning: Are convolutional networks the winning lottery ticket of fully connected ones?

Pruning methods can considerably reduce the size of artificial neural networks without harming their performance. In some cases, they can even uncover sub-networks that, when trained in isolation, match or surpass the test accuracy of their dense counterparts. Here we study the inductive bias that pruning imprints in such "winning lottery tickets". Focusing on visual tasks, we analyze the architecture resulting from iterative magnitude pruning of a simple fully connected network (FCN). We show that the surviving node connectivity is local in input space, and organized in patterns reminiscent of the ones found in convolutional networks (CNN). We investigate the role played by data and tasks in shaping the architecture of pruned sub-networks. Our results show that the winning lottery tickets of FCNs display the key features of CNNs. The ability of such automatic network-simplifying procedure to recover the key features "hand-crafted" in the design of CNNs suggests interesting applications to other datasets and tasks, in order to discover new and efficient architectural inductive biases.

cs.LG↗

arXiv2020

A Systematic Approach to Generating Accurate Neural Network Potentials: the Case of Carbon

Availability of affordable and widely applicable interatomic potentials is the key needed to unlock the riches of modern materials modelling. Artificial neural network based approaches for generating potentials are promising; however neural network training requires large amounts of data, sampled adequately from an often unknown potential energy surface. Here we propose a self-consistent approach that is based on crystal structure prediction formalism and is guided by unsupervised data analysis, to construct an accurate, inexpensive and transferable artificial neural network potential. Using this approach, we construct an interatomic potential for Carbon and demonstrate its ability to reproduce first principles results on elastic and vibrational properties for diamond, graphite and graphene, as well as energy ordering and structural properties of a wide range of crystalline and amorphous phases.

cond-mat.mtrl-sci↗

arXiv2020

An analytic theory of shallow networks dynamics for hinge loss classification

Neural networks have been shown to perform incredibly well in classification tasks over structured high-dimensional datasets. However, the learning dynamics of such networks is still poorly understood. In this paper we study in detail the training dynamics of a simple type of neural network: a single hidden layer trained to perform a classification task. We show that in a suitable mean-field limit this case maps to a single-node learning problem with a time-dependent dataset determined self-consistently from the average nodes population. We specialize our theory to the prototypical case of a linearly separable dataset and a linear hinge loss, for which the dynamics can be explicitly solved. This allow us to address in a simple setting several phenomena appearing in modern networks such as slowing down of training dynamics, crossover between rich and lazy learning, and overfitting. Finally, we asses the limitations of mean-field theory by studying the case of large but finite number of nodes and of training samples.

stat.ML↗

arXiv2019

PANNA: Properties from Artificial Neural Network Architectures

Prediction of material properties from first principles is often a computationally expensive task. Recently, artificial neural networks and other machine learning approaches have been successfully employed to obtain accurate models at a low computational cost by leveraging existing example data. Here, we present a software package "Properties from Artificial Neural Network Architectures" (PANNA) that provides a comprehensive toolkit for creating neural network models for atomistic systems. Besides the core routines for neural network training, it includes data parser, descriptor builder and force-field generator suitable for integration within molecular dynamics packages. PANNA offers a variety of activation and cost functions, regularization methods, as well as the possibility of using fully-connected networks with custom size for each atomic species. PANNA benefits from the optimization and hardware-flexibility of the underlying TensorFlow engine which allows it to be used on multiple CPU/GPU/TPU systems, making it possible to develop and optimize neural network models based on large datasets.

physics.comp-ph↗

arXiv2018

Thermolubricity and the Jarzynski equality

We discuss and qualify a previously unnoticed connection between two different phenomena in the physics of nanoscale friction, general in nature and also met in experiments including sliding emu- lations in optical lattices, and protein force spectroscopy. The first is thermolubricity, designating the condition in which a dry nanosized slider can at sufficiently high temperature and low velocity exhibit very small viscous friction f ~ v despite strong corrugations that would commonly imply hard mechanical stick-slip f ~ log(v). The second, apparently unrelated phenomenon present in externally forced nanosystems, is the occurrence of negative work tails ("free lunches") in the work probabilty distribution, tails whose presence is necessary to fulfil the celebrated Jarzynski equality of non-equilibrium statistical mechanics. Here we prove analytically and demonstrate numerically in the prototypical classical overdamped one-dimensional point slider (Prandtl-Tomlinson) model that the presence or absence of thermolubricity is exactly equivalent to satisfaction or violation of the Jarzynski equality. The divide between the two regimes, satisfaction of Jarzynski with ther- molubricity, and violation of both, simply coincides with the total frictional work per cycle falling below or above kT respectively. This concept can, with due caution, be extended to more complex sliders, thus inviting crosscheck experiments, such as searching for free lunches in cold ion sliding as well as in forced protein unwinding, and alternatively checking for a thermolubric regime in dragged colloid monolayers. As an important byproduct, we derive a parameter-free formula expressing the linear velocity coefficient of frictional dissipated power in the thermolubric viscous regime, correcting previous empirically parametrized expressions.

cond-mat.stat-mech↗

arXiv2017

Friction Anomalies at First-Order Transition Spinodals: 1T-TaS$_2$

Revealing phase transitions of solids through mechanical anomalies in the friction of nanotips sliding on their surfaces is an unconventional and instructive tool for continuous transitions, unexplored for first-order ones. Owing to slow nucleation, first-order structural transformations generally do not occur at the precise crossing of free energies, but hysteretically, near the spinodal temperatures where, below and above the thermodynamic transition temperature, one or the other metastable free energy branches terminates. The spinodal transformation, a collective one-shot event with no heat capacity anomaly, is easy to trigger by a weak external perturbations. Here we propose that even the gossamer mechanical action of an AFM tip may locally act as a surface trigger, narrowly preempting the spontaneous spinodal transformation, and making it observable as a nanofrictional anomaly. Confirming this expectation, the CCDW-NCCDW first-order transition of the important layer compound 1T-TaS$_2$ is shown to provide a demonstration of this effect.

cond-mat.mes-hall↗

arXiv2017

Quantum Lubricity

The quantum motion of nuclei, generally ignored in sliding friction, can become important for an atom, ion, or light molecule sliding in an optical lattice. The density-matrix-calculated evolution of a quantum Prandtl-Tomlinson model, describing the frictional dragging by an external force of a quantum particle, shows that classical predictions can be very wrong. The strongest quantum effect occurs not for weak, but for strong periodic potentials, where barriers are high but energy levels in each well are discrete, and resonant tunnelling to excited states in the nearest well can preempt classical stick-slip with great efficiency. The resulting permeation of otherwise impassable barriers is predicted to cause quantum lubricity

cond-mat.mes-hall↗

arXiv2016

CDW slips and giant frictional dissipation peaks at the NbSe$_2$ surface

Accessing, controlling and understanding nanoscale friction and dissipation is a crucial issue in nanotechnology, where moving elements are central. Recently, ultra-sensitive noncontact pendulum Atomic Force Microscope (AFM) succeeded in detecting the electronic friction drop caused by the onset of superconductivity in Nb, raising hopes that a wider variety of mechanisms of mechanical dissipation arising from electron organization into different collective phenomena will become accessible through this unconventional surface probe. Among them, the driven phase dynamics of charge-density-waves (CDWs) represents an outstanding challenge as a source of dissipation. Here we report a striking multiplet of AFM dissipation peaks arising at nanometer distances above the surface of NbSe$_2$ - a layered compound exhibiting an incommensurate CDW. Each peak appears at a well defined tip-surface interaction force of the order of a nN, and persists until T=70K where CDW short-range order is known to disappear. A theoretical model is presented showing that the peaks are connected to tip-induced local 2$π$ CDW phase slips. Under the attractive potential of the approaching tip, the local CDW surface phase landscape deforms continuously until a series of 2$π$ jumps occur between different values of the local phase. As the tip oscillates to and fro, each slip gives rise to a hysteresis cycle, appearing at a selected distance, the dissipation corresponding to "pumping" in and out a local slip in the surface CDW phase of NbSe$_2$.

cond-mat.mes-hall↗