Learning the Fermion sign structure in path-integral Monte Carlo
arXiv:2607.15060
2026
Training
2 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper provides a transferable probabilistic-numerics mechanism for learning a numerically unstable signed observable without directly estimating a near-zero ratio. Its key construction is to quotient configurations into permutation families, impose asymptotically correct physical priors on family probabilities and energies, and let an LSTM learn only the many-body residual. A neural-network analogue is a symmetry-aware family-conditioned predictor whose signed aggregate is evaluated analytically, together with uncertainty-driven sampling that preferentially visits families contributing large residual variance. The strongest falsifiable signatures are recovery of known asymptotic scaling, reduced variance near cancellation, and a sharp reduction in failure once direct signed-denominator estimation becomes noise dominated.
Ideas from this paper
✗ Mechanism failed
2026
Replace direct learning of a highly cancelling signed observable with a quotient-space model over symmetry orbits of inputs. Predict a physically constrained baseline for each family and use an LSTM or set/graph encoder only for the residual many-body correlation, then aggregate family predictions with known signed weights instead of forming a noisy sample-level ratio.
Useful8/10
Difficulty5/10
Novelty7/10
△ Mechanism confirmed, baseline not beaten
2026
Use the family predictor not only as a post-processing estimator but also as a feedback controller for data collection. Reweight Monte Carlo proposals or minibatch selection toward under-sampled families whose signed contribution and predictive uncertainty are large, rather than spending samples on already well-known positive families.
Useful7/10
Difficulty6/10
Novelty5/10