Are neural scaling laws leading quantum chemistry astray?

2025-09-30

AI Summary

This work challenges the applicability of neural scaling laws to quantum chemical modeling: does simply increasing model capacity and training dataset size—comprising quantum-chemical calculations on stable (equilibrium) molecular configurations—sufficiently ensure physical consistency and extrapolative reliability? Using the H₂ bond dissociation curve as a benchmark, we systematically evaluate the generalization performance of neural networks of varying scales, trained either with or without stretched/compressed configurations. Results demonstrate that models trained exclusively on equilibrium geometries fail—even at extreme scale—to reproduce Coulombic repulsion at short interatomic distances or accurately predict the dissociation-limit energy, revealing a fundamental failure to learn core physical principles such as Coulomb’s law. This study provides the first empirical evidence that data distribution bias is more critical than model scale; physically robust extrapolation necessitates explicit inclusion of key dynamical regimes—not mere architectural or data scaling.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Game Theory and Economic Paradigms: Adversarial Learning
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Responsible Web: Ethical and legal aspects of web-scale data analysis, uses and collection practices

Abstract

Neural scaling laws are driving the machine learning community toward training ever-larger foundation models across domains, assuring high accuracy and transferable representations for extrapolative tasks. We test this promise in quantum chemistry by scaling model capacity and training data from quantum chemical calculations. As a generalization task, we evaluate the resulting models' predictions of the bond dissociation energy of neutral H$_2$, the simplest possible molecule. We find that, regardless of dataset size or model capacity, models trained only on stable structures fail dramatically to even qualitatively reproduce the H$_2$ energy curve. Only when compressed and stretched geometries are explicitly included in training do the predictions roughly resemble the correct shape. Nonetheless, the largest foundation models trained on the largest and most diverse datasets containing dissociating diatomics exhibit serious failures on simple diatomic molecules. Most strikingly, they cannot reproduce the trivial repulsive energy curve of two bare protons, revealing their failure to learn the basic Coulomb's law involved in electronic structure theory. These results suggest that scaling alone is insufficient for building reliable quantum chemical models.

Problem

• Testing neural scaling laws in quantum chemistry for predicting molecular properties
• Evaluating model performance on bond dissociation energy of H2 molecule
• Assessing failure to learn basic Coulomb's law in electronic structure

Innovation

• Models trained only on stable structures fail
• Including compressed and stretched geometries improves predictions
• Scaling alone is insufficient for reliable quantum chemistry