Published November 16, 2026 | Version 0.2.0-candidate

Maximal correlation of Lévy bridges

Authors/Creators

Description

Unrefereed research candidate. A reduction to jump configurations and a one-cut principle.

Context. The maximal correlation of two random objects is the largest correlation between any function of the first and any function of the second: 0 means independence and 1 means the two share a nonconstant function exactly. Chang and Chen (The Annals of Applied Probability 36 (2026), doi:10.1214/25-AAP2293) computed it for the two coordinate paths of a two-dimensional Lévy process, as max{|ρ|, Op(ν)}, and asked in their Section 6, Problem 1, for the answer when the process is pinned at a terminal point, that is, for a Lévy bridge. Pinning destroys independent increments and can create dependence that the Lévy measure does not show.

Main results. (1) Bridge formula (Theorem 2.1): the bridge coefficient is the largest of three numbers: the absolute correlation of the Gaussian part; the maximal correlation c0 of the two unordered lists of jump sizes; and the norm c1 of an operator that counts common jumps. Jump times matter only through which jumps are shared. It holds for almost every endpoint, for every endpoint when the Gaussian covariance is positive definite (standard bridge), and at every atom of the terminal law; infinite activity is allowed. (2) One-cut principle (Corollary 2.2): with the jump sizes retained, knowing which jumps occur before one fixed time preserves the coefficient; two different cut times shrink the common-jump and Gaussian terms by an explicit factor. (3) Computable cases: without common jumps the size term reduces to a two-dimensional problem for the jump totals, and pure-jump coordinates become independent; for finitely many lattice jump types both terms are norms of finite matrices; Chang and Chen's Examples 2-4, pinned, have coefficient one at every endpoint at which the second coordinate jumps; without conditioning the same machinery gives an independent proof of Chang and Chen's formula on every finite horizon. (4) The engine is a correlation identity for independent random marks attached to the classes of a hidden random partial matching; Chang and Chen's labelled random-subset theorem (their Theorem 1.5) is essentially its visible-matching special case.

Examples. Pinning can raise the coefficient from 0.2 to above 0.997 without common jumps (certified enclosure 0.997457967473837283247 < R < 0.997457967473837283251 by interval arithmetic), and can also lower it. In a four-type pure-jump bridge the dominant term switches at c = 1/4. A hidden swap has coefficient 1/√2 while the revealed matching has coefficient 1.

Prior work credited. The fixed-overlap spectrum and the random-elements-in-common construction behind the shared-shock example are due to Diaconis and Griffiths (2012); the unpinned shared-shock class of bivariate Poisson processes is due to Griffiths and Milne (1978), with correlation ranges studied by Griffiths, Milne and Wood (1979). The clock representation is Kallenberg's.

Limits. The answer is a reduction, not a closed formula: the two jump terms are operator norms with no general elementary formula; they are computed explicitly only in the cases above. No claim is made at exceptional endpoints: every-endpoint statements for pure-jump bridges such as gamma or stable bridges remain open. Pinning can raise or lower the coefficient, and no general comparison is proved.

Contents. The paper (LaTeX source and PDF); exact-rational finite checks of the matching identity and the Poisson spectra (324 one-cut spectra, 81 matching laws and 100 random matching laws, residuals below 1e-15); an interval-arithmetic certificate for the Gaussian-Poisson example; checks of the extensions and examples; the figure script and data; the bibliography; a machine-readable index; a research-gate record; a review summary; a replay receipt and SHA-256 sums. Checks exit with status 1 on failure, also under python -O. The computations check formulas on finite cases; the proofs carry the mathematical claims.

Assurance. Written proofs with exact and interval computational checks. Internal AI-assisted review only: five reviewer roles run as Claude subagents and an adversarial review by a second model (Astra), a clean replay, one revision round, then a re-review and a Claude proof audit of the new proofs. The planned Astra audit of the new proofs did not run because of a usage limit. Not peer review, independent reproduction or formal verification. Creator: Anonymous. Text CC0-1.0; code MIT.

Notes

Subtitle: A reduction to jump configurations and a one-cut principle. Code (*.py and the CI workflow) is MIT; prose, manuscript, figures and recorded outputs are CC0-1.0.

Files

levy-bridge-maximal-correlation-0.2.0-candidate.pdf

Files (1.2 MB)

Name Size
md5:fb2672e0b8a4775d82ed50a96ea9e486
464.4 kB Preview Download
md5:7f1ba7164a9e50cda68d3c94579364ad
738.7 kB Preview Download
md5:2cc7cb9c191b89cbdc6a1ca5c82f51ae
236 Bytes Download

Additional details