Föllmer Measures in Stochastic Analysis

• Föllmer measures are probability measures on path space defined via quadratic variation, forming a cornerstone in pathwise Ito calculus and entropy minimization.
• They arise from minimizing relative entropy under marginal constraints, linking concepts such as Schrödinger bridges and minimal martingale measures.
• Applications extend to optimal transport, financial risk management, and rough path theory, offering robust tools for model uncertainty and pathwise integration.

A Föllmer measure is a probability measure on path space that arises from the minimization of relative entropy (Kullback–Leibler divergence) with prescribed marginal constraints or as the canonical pathwise quadratic variation measure attached to a deterministic path. The notion permeates mathematical finance, optimal transport, stochastic analysis, and rough path theory, serving as the foundational object in pathwise Itô calculus, Schrödinger bridges, and the minimal martingale measure construction in incomplete markets. The key role of Föllmer measures is to provide rigorous, often canonical, ways to deal with model uncertainty, pathwise integration, and entropy minimization under marginal constraints.

1. Classical Definition and Pathwise Construction

On a filtered probability space $(\Omega, \mathcal{F}, (\mathcal{F}_t)_{t\ge0}, P)$, Föllmer's original pathwise construction considers a (possibly deterministic) càdlàg path $X$ in a Banach space $E$ and attaches to $X$ a canonical measure $\mu^X$ on $[0, \infty)$ defined via the quadratic variation process $Q(X)$. Given a sequence of nested partitions $\Pi = (\pi_n)$, the scalar quadratic variation along $\Pi$ is

$$Q(X)_t := \lim_{n\to\infty}\!\sum_{I\in\pi_n} \|X_{v\wedge t} - X_{u\wedge t}\|^2,$$

for intervals $X$0; for one-dimensional paths, this reduces to the classical quadratic variation. The associated Föllmer measure is

$X$1

This measure captures both the continuous and jump parts of $X$2, with purely atomic support at jump times and continuous support elsewhere (Hirai, 2021).

Föllmer integration is then defined via deterministic Riemann sums along $X$3; for sufficiently regular functions $X$4,

$X$5

is the pathwise limit of sums $X$6 (Hirai, 2021). In higher dimensions or Banach spaces, arbitrary bounded bilinear maps $X$7 can be used to define quadratic covariation, and a tensor-valued Föllmer measure encodes joint variability.

2. Variational Characterization and Entropy-Minimizing Paths

The variational notion of a Föllmer measure formalizes the solution to the Schrödinger bridge problem with degenerate initial law: given a target measure $X$8 on $X$9 and Brownian motion as the reference, the Föllmer measure is the unique minimizer of

$E$0

over all probability measures $E$1 on path space such that $E$2, $E$3. In the formulation via stochastic interpolants, a family of diffusions

$E$4

can be constructed such that, among all diffusions with prescribed time-marginals, the Föllmer process is the unique minimizer of path-space KL divergence with respect to a given reference process (Chen et al., 11 Feb 2026), generalizing classical Schrödinger-bridge and stochastic control treatments.

The drift $E$5 can be expressed in terms of the conditional expectation or as an explicit function of the marginals' score $E$6, and the Föllmer process results from optimizing over possible diffusion coefficients to minimize sensitivity to estimation error in the induced pathwise KL (Chen et al., 11 Feb 2026).

3. Applications: Minimal Martingale and Schrödinger-Föllmer Measures

Minimal Martingale Measure in Incomplete Markets

In mathematical finance, the Föllmer–Schweizer minimal martingale measure $E$7 is the unique equivalent local martingale measure under which the local martingale part of the asset-price process remains a local martingale. Its density is given by

$E$8

where $E$9 is the market-price-of-risk process. $X$0 minimizes the $X$1-variance among all equivalent local martingale measure densities and yields crucial stability properties for factor processes, ensuring mean-reversion under Hurwitz conditions on the factor-dynamics matrix. In risk-sensitive portfolio optimization with model uncertainty, adversarial market selection of equivalent measures leads to the worst-case being attained by the minimal martingale measure, enforcing the "no-regret" (i.e., cash-only) strategy under certain constraints (Deshpande, 2014).

Schrödinger–Föllmer Bridges and Generative Diffusions

In the context of entropic optimal transport, the Föllmer measure is the path-space law that minimizes relative entropy to Brownian motion subject to marginals constraints. The associated SDE, typically with drift

$X$2

leads to the so-called Schrödinger–Föllmer diffusion. The Schrödinger-Föllmer sampling framework enables efficient sampling from complex, especially multimodal, distributions without requiring ergodicity or gradient evaluation. The key numerical result is an $X$3 convergence rate in $X$4-Wasserstein distance for Euler discretization, under appropriate regularity assumptions (Wang et al., 30 Dec 2025).

The high-temperature variant of SFS diffusions can escape local modes by amplifying both drift and noise, a property critical for successful sampling of multimodal targets (Wang et al., 30 Dec 2025).

4. Martingale Structures, Duality, and Weak Transport

Föllmer measures admit a Doob–martingale transform ("Föllmer martingale"), which is the continuous-time martingale minimizing a weighted energy functional among all martingales with prescribed initial and terminal laws. In the irreducible case, the continuous martingale Schrödinger bridge coincides with the Föllmer martingale. Dualizing, one recovers a weak optimal transport problem in which the static optimizer has a Gibbs form. In explicit cases (Gaussian marginals, Bernoulli two-point laws), Föllmer drifts, diffusions, and associated martingales have closed-form representations, tightly linking entropic optimal transport, dynamic (Schrödinger) bridges, and weak martingale transport (Backhoff et al., 1 Apr 2026).

5. Föllmer Measure as a Tool in Filtration Shrinkage and Mathematical Finance

When projecting a strict local martingale onto a subfiltration, the local martingale property may be lost, and a unique Föllmer measure can be constructed up to the explosion time by patching together stopped Girsanov changes of measure. The compensator $X$5 of the (possibly explosive) stopping time under the extended measure quantifies the loss of the martingale property in the smaller filtration. In a topological path setting, the support of the measure's finite-variation component is contained in the closure of the zero set of the harmonic function defining the reciprocal local martingale; in many cases, it is purely singular with respect to Lebesgue measure (Larsson, 2011).

This framework rigorously explains how "information shrinkage" transforms strict local martingale phenomena into the perception of arbitrage under restricted filtrations, shaping the analysis of bubbles and asset-price collapse in finance (Larsson, 2011).

6. Pathwise and Rough Path Extensions

Föllmer's approach to stochastic integration establishes a fully pathwise (nonprobabilistic) calculus based on quadratic variation along partitions. This construction has been extended to Banach spaces using scalar and bilinear quadratic variation, and to rough paths by enhancing the driving signal to a $X$6-rough path whenever suitable "quadratic-roughness" or "Lévy-roughness" properties hold. In such cases, the partition-invariance of both the quadratic variation and Lévy area is characterized precisely, and deterministic Riemann sum integrals along partitions coincide with rough path integrals (Das et al., 23 Jul 2025). This unifies classical stochastic calculus, pathwise methods, and rough path theory, providing a robust platform for infinite-dimensional and irregular pathwise integration.

7. Summary Table: Interpretations and Domains

Context	Föllmer Measure Characterization	Reference
Pathwise Itô Calculus	Measure encodes quadratic variation of $X$7	(Hirai, 2021)
Schrödinger Bridge/Optimal Transport	KL minimizer w/prescribed marginals	(Chen et al., 11 Feb 2026)
Minimal Martingale Measure	EMM minimizing $X$8-variance for assets	(Deshpande, 2014)
Weak Martingale Transport	Dynamic Doob–martingale minimizing energy	(Backhoff et al., 1 Apr 2026)
Filtration Shrinkage	Measure extension for loss of martingality	(Larsson, 2011)
Pathwise and Rough Integration	Limiting measure for robust pathwise calculus	(Das et al., 23 Jul 2025)

Föllmer measures unify and rigorously underpin a wide array of disciplines by encoding minimality, entropy, and pathwise quadratic variation, and they play a canonical role in mathematical finance, stochastic analysis, and modern generative modeling.