The Critical 2D KPZ Process

Quentin Berger, Francesco Caravenna, Jeremy Clark, Alexander Dunlap, Yu Gu, Ziyang Liu, Shuta Nakajima, Rongfeng Sun, Li-Cheng Tsai, Zuodi Xie, Nikos Zygouras

For directed polymers on Z² in a Gaussian environment in the specified critical window with fixed parameter ϑ, centred log-partition fields converge to a uniquely constructed critical 2D KPZ random distribution, jointly for finitely many times and initial data in the stated local negative Sobolev topology. For flat initial data the limit is non-Gaussian, despite Gaussian small-scale averages.

Sources for this record

What was already known

Caravenna, Sun and Zygouras constructed the critical 2D stochastic heat flow in 2023. The new work constructs and characterizes the centred logarithmic field at criticality. It does not claim the same result for every microscopic growth model or all forms of the two-dimensional KPZ equation.

AI involvement AI-centered

Appendix C identifies AI-derived covariance/tightness bounds, a new simpler conditional-Gaussian-convexity route completing convergence, and a Gaussian-rigidity argument for non-Gaussianity. The human authors supplied structural inputs, rejected a flawed switch-off construction and revised the retained approach with AI. Inclusion concerns the surviving arguments, not the abandoned proposal.

Importance Notable

Provides a precise critical logarithmic scaling limit and identifies non-Gaussian structure in a central stochastic-growth model.

Limitations

Gaussian directed-polymer disorder, the critical-window scaling and the stated initial-data class are essential. Armstrong’s separate formalization records corrections and deliberate changes of conventions, including precise SHF predicates and topology. Its identification with the earlier literature’s SHF is a literature comparison, not established by Comparator alone. His four additional results are separately credited and are outside this contribution.

Technical verification Partial

Armstrong reports Lean kernel checking of the principal formal counterparts using only the three standard Lean axioms. The source documents 126 deliberate deviations and corrections for the main paper. Exact identification of the artifact’s SHF predicate and topology with the original literature has not been fully reconciled in this review, so the original mathematical claim remains Partial rather than inferring full checking from registration.

Human understanding Full

The named authors explain the convergence decomposition and non-Gaussianity mechanism in Section 2 and Section 9 and document the strategy changes in Appendix C. Armstrong’s formalization and later four-results note retain separate credit.

Correction and publication history

2026-10-28 · amendment

Lists the original paper once while retaining the specific checking and provenance references. Research claims and rating levels are unchanged.

Read this published revision
2026-10-28 · amendment

Clarifies that the source reports completed checking but does not supply its exact completion date. Research claims and rating levels are unchanged.

Read this published revision
2026-10-28 · publication

Newly reviewed source-backed research contribution with contribution-specific AI attribution and checking limits.

Read this published revision