Information on Trajectories: Martingales and Random Times — Paper Overview

Forum topic · 小凯 · 2026-08-24

Summary

A paper by Akshay Balsubramani (arXiv:2608.20337) models information flow on the path space of nonnegative martingale trajectories, deriving exact variational identities that hold even at arbitrary random times. The framework recovers widely used classical concentration inequalities — from Ville's inequality to PAC-Bayes bounds — and quantifies precisely what information each inequality discards. The controlled tail is itself a relative entropy that decomposes via the chain rule into per-step conditional divergences. The discarded slack takes exact forms in three geometries: Gibbs tilting for Azuma-Hoeffding and PAC-Bayes bounds, crossing itself for Ville's inequality and merging tests, and control certificates for L^p maximal inequalities. The optional-stopping deficit of a certificate decomposes stepwise into a Bregman divergence of the running maximum. On path-time space, the identity yields a pricing factor: any random time carries an e-process 'peeking penalty'. The partition function can be interpreted as a coalescent process — the prefix-sharing probability of independent copies — and geometric mixtures of test martingales yield merging benefits for multi-model safe testing.

Paper Overview

Field: Machine Learning
Author: Akshay Balsubramani
Published: 2026-08-22
arXiv: 2608.20337

Key Points

Accounting for information flow on the path space of trajectories of a nonnegative martingale yields exact variational identities for it, even at arbitrary random times. This recovers widely used classical concentration inequalities — from Ville to PAC-Bayes — and measures what each one discards.

The controlled tail itself is a relative entropy, decomposable via the chain rule into per-step conditional divergences.
The discarded slack has exact forms in three geometries:
Gibbs tilting for Azuma-Hoeffding and PAC-Bayes bounds
The crossing itself for Ville's inequality and merging tests
Control certificates for L^p maximal inequalities
The optional-stopping deficit of the certificate decomposes stepwise into a Bregman divergence of the running maximum.
On path-time space, the same identity acquires a factor pricing expectations: any random time carries an e-process "peeking penalty".
The partition function admits an interpretation as a coalescent process — the probability that independent copies share a prefix.
Geometric mixtures of test martingales yield merging benefits for multi-model safe testing.

Abstract (Original)

> Accounting for information flow on the path space of trajectories of a nonnegative martingale yields exact variational identities for it, even at arbitrary random times. This recovers the widely used classical concentration inequalities, from Ville to PAC-Bayes, and measures what each one discards.

---

*Auto-collected on 2026-08-24.*