The Right Measure for Physics-Constrained Generation: A Co-Area Correction for Posterior-Consistent PDE Inverse Problems

2026-06-03 · Source: Takara TLDR - Daily AI Papers · Field: Science & Research — Mathematics & Computational Sciences, Engineering & Applied Sciences · Depth: Expert, quick

Summary

Generative models, including diffusion and flow matching, are increasingly applied to partial differential equation (PDE) inverse problems, enforcing physics as a hard constraint. However, this common approach, using projection or guidance, samples the wrong distribution. Conditioning on a measure-zero manifold omits a crucial co-area (Fixman) Jacobian factor, $[det(JJ^{\top})]^{-1/2}$, leading to significant bias. This bias escalates with constraint sensitivity heterogeneity, inflating posterior error up to $20\times$ the sampling-noise floor, while minimal-displacement projection shows $9\times$ bias. A naive scalar reweighting fails to correct this. Researchers introduce CoCoS, a measure-aware constrained sampler that correctly incorporates this factor, matching the gold-standard posterior within sampling noise. This highlights that "satisfying the physics" is distinct from "sampling the posterior."

Key takeaway

For AI Scientists and Research Scientists calibrating uncertainty in PDE inverse problems using generative models, you must recognize that current projection and guidance methods yield biased posteriors. Your models will produce inaccurate uncertainty estimates if they omit the co-area Jacobian factor. Adopt measure-aware sampling techniques like CoCoS to ensure your scientific inference is principled and your uncertainty quantification is accurate, avoiding significant posterior error.

Key insights

Hard PDE constraints in generative models require a co-area correction for accurate Bayesian posterior sampling.

Principles

• Hard constraints on measure-zero manifolds are intrinsically ambiguous.
• Omitted co-area Jacobian factors bias posterior distributions.
• Physics satisfaction does not equate to posterior sampling.

Method

CoCoS introduces a measure-aware sampling method that incorporates the co-area (Fixman) Jacobian factor $[det(JJ^{\top})]^{-1/2}$ to correctly target the posterior distribution in PDE inverse problems.

In practice

• Implement CoCoS for accurate uncertainty quantification.
• Avoid projection/guidance for hard PDE constraints.
• Account for constraint sensitivity heterogeneity.

Topics

• PDE Inverse Problems
• Generative Models
• Diffusion Models
• Flow Matching
• Bayesian Posterior
• Co-Area Correction
• Uncertainty Quantification

Best for: AI Scientist, Research Scientist

Related on AIssential

• Separating Intrinsic Ambiguity from Estimation Uncertainty in Deep Generative Models for Linear Inverse Problems — A new method separates intrinsic ambiguity from estimation uncertainty in deep generative models for inverse problems.
• MoSA: Motion-constrained Stress Adaptation for Mitigating Real-to-Sim Gap in Continuum Dynamics via Learning Residual Anisotropy — MoSA learns residual anisotropy and heterogeneity using physics priors to close the real-to-sim gap in continuum dynamics.
• Phantom: Physics-Infused Video Generation via Joint Modeling of Visual and Latent Physical Dynamics — Integrating latent physical property inference into video generation improves physical plausibility and visual realism.
• On the Regularity and Generalization of One-Step Wasserstein-guided Generative Models for PDE-Induced Measures — The optimal transport map for PDE-induced measures is Hölder continuous, justifying one-step generative models.
• Error Bounds for a Diffusion Model-Based Drift Estimator — Theoretical guarantees for a diffusion model-based drift estimator in SDEs are now established through an explicit risk bound.
• Structure-Aware Epistemic Uncertainty Quantification for Neural Operator PDE Surrogates — Restricting uncertainty sampling to the lifting module improves neural operator uncertainty quantification efficiency and accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.