PaperScope
LIVE · 2026-09-29 05:40 UTC

Apparent Compression, Real Stability: The Intrinsic Dimension of Learning a Quantum Wavefunction

Lu Wei, Yufeng Wang, Chenfeng Cao, Haibin Ling

Latestcs.CLcs.LGcs.AIcs.CV
arXiv ID
2609.33193 v1
Category
Submitted
2026-09-27

Abstract

How many directions in weight space does training need? The intrinsic dimension answers this with the smallest number of random directions in which training still reaches a target accuracy, and small values have motivated parameter-efficient methods such as LoRA. We measure it for variational Monte Carlo (VMC), which trains a neural network to represent the ground state of a quantum many-body system. VMC is a demanding test, because the network generates its own training samples and every gradient is noisy, and a revealing one, because the exact answer is known and every run can be scored. We train only a small latent vector that a frozen random map turns into the network's weights, with no change to the standard natural-gradient optimizer. We find that a small dimension can be misleading, while the stability it brings is real. On a magnet with a hard sign pattern, a network that cannot represent signs reaches its best energy in 8 of 28,642 directions, but only because no such network can go lower; once signs are learnable, neither the signs nor the magnitudes are cheap. The dimension rises across a quantum phase transition, so it tracks how difficult a state is at far less compute than fitting a scaling law, yet it never falls below a floor set by the random subspace itself, even where the ground state is nearly trivial. Training in the subspace, in contrast, never diverged in our experiments, whereas full-parameter training with the same settings did, and a control with matched solvers attributes the difference to the reduced dimension.

Comment: 18 pages, 9 figures, 6 tables

arXiv abs page · PDF · same-day batch