Pathwise Stochastic Integration Frameworks

• Pathwise stochastic integration frameworks are deterministic methods that define stochastic integrals directly from individual sample paths.
• They encompass diverse approaches—Föllmer-type, rough paths, fractional calculus, and outer measure—each imposing unique regularity and stability conditions.
• These frameworks enable robust financial modeling, adaptive numerical schemes, and model-free analysis where classical Itô calculus is insufficient.

Pathwise stochastic integration frameworks provide methodologies for defining and analyzing stochastic integrals directly from individual realizations of processes (“paths”), without using probability-theoretic or expectation-based constructions. Such frameworks underpin probabilistic numerics, model-free financial mathematics, and rough differential equations, particularly when classical Itô calculus is inapplicable or too restrictive. Approaches in this area include pathwise quadratic variation integrals à la Föllmer, rough paths theory, functional calculus extensions, fractional calculus, and frameworks specifically designed for model-free and robust finance. Each framework imposes distinct regularity or structural conditions on the integrator, the integrand, or both, and yields different analytic, stability, or isometry properties.

1. Core Frameworks: Föllmer-Type and Functional Itô Calculus

A central paradigm in pathwise integration is the Föllmer approach, which constructs integrals for continuous (or càdlàg) paths possessing quadratic variation along a sequence of vanishing-mesh partitions $\{\pi^n\}$. For a path $X:[0,T]\to\mathbb{R}^d$, quadratic variation along $\pi^n$ is defined as

$$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$

when the limit exists and is nontrivial. The pathwise Itô integral of an adapted process $h$ against $X$ is constructed as the uniform limit (if it exists) of discrete sums

$\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$

with the key requirement that $h$ arises as the vertical derivative of a non-anticipative functional $F\in C^{1,2}$ in the sense of the Dupire functional calculus (Ananova et al., 2016).

A powerful extension is provided by Ananova–Cont’s functional calculus (Ananova et al., 2016), which yields:

• Precise definition: Integrands are vertical derivatives $\nabla F(t, X_{\cdot\wedge t})$; integrator $X:[0,T]\to\mathbb{R}^d$0 must have finite quadratic variation along the partition sequence.
• Pathwise isometry: The quadratic variation of the integral $X:[0,T]\to\mathbb{R}^d$1 is $X:[0,T]\to\mathbb{R}^d$2. This is a deterministic, pathwise analog of the Itô isometry.
• Signal-plus-noise decomposition: Any sufficiently regular process can be decomposed as $X:[0,T]\to\mathbb{R}^d$3, with $X:[0,T]\to\mathbb{R}^d$4, uniquely specifying $X:[0,T]\to\mathbb{R}^d$5 up to $X:[0,T]\to\mathbb{R}^d$6-null sets.
• Hilbert space structure: The integrand space, completed under $X:[0,T]\to\mathbb{R}^d$7, admits an isometric integral operator.

This approach builds on, but is more robust than, Föllmer’s original gradient-only construction, and notably remains deterministic once the data path $X:[0,T]\to\mathbb{R}^d$8 is given (Ananova et al., 2016).

2. Rough Path Theory and Pathwise Lévy Areas

Pathwise integration via rough path theory, originating from Lyons and expanded in subsequent work (Imkeller et al., 2014, Gubinelli et al., 2014, Das et al., 23 Jul 2025), addresses driving signals of low regularity (e.g., Hölder $X:[0,T]\to\mathbb{R}^d$9, or $\pi^n$0-variation with $\pi^n$1) by constructing an enhanced path $\pi^n$2, where $\pi^n$3 encodes “Lévy areas”:

$\pi^n$4

and must satisfy algebraic (Chen's relation) and analytic ($\pi^n$5-variation) conditions.

• Controlled paths: Integrands $\pi^n$6 satisfy $\pi^n$7, with the remainder $\pi^n$8 of higher regularity.
• Sewing lemma and pathwise area: Existence of the Lévy area is a genuinely pathwise, analytic matter, constructed via the sewing lemma (Feyel–de la Pradelle). As soon as $\pi^n$9 is controlled by $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$0, one may construct

$$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$1

• Stability: The integral map is locally Lipschitz continuous in the rough path topology, essential for robust numerics and sensitivity analysis (Allan et al., 2021, Das et al., 23 Jul 2025).
• Partition invariance: Invariant limits with respect to the partition sequence are tied to regularity properties known as “quadratic and Lévy roughness” (Das et al., 23 Jul 2025).
• Multiplicity: This apparatus yields unified access to Itô, Stratonovich, and backward integrals as different Riemann sum limits (Das et al., 23 Jul 2025).

Notably, pathwise rough integration includes non-gradient integrands, accommodates path-dependent functionals, and underpins deterministic PDE and ODE solution theory for systems driven by rough signals.

3. Fractional Pathwise Calculus and Volterra Processes

Fractional calculus supplies a flexible framework for integrating with respect to Volterra processes (including fractional Brownian motions, Lévy-driven fields, and general ambit fields) (Nunno et al., 2016). Here, the integral is defined via:

• Fractional derivatives & integrals: Riemann–Liouville or Weyl fractional operators, $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$2, $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$3, are applied to both the integrand $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$4 and the integrator $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$5:

$$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$6

• Regularity conditions: Integrability is determined by matching fractional Besov/Slobodetski regularities (e.g., $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$7, $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$8, $$[X]_t = \lim_{n\to\infty} \sum_{t_i\in\pi^n,\,t_i\leq t} (X_{t_{i+1}} - X_{t_i})^2,$$9).
• Unification: This framework interpolates between deterministic Riemann–Stieltjes integrals (for sufficiently smooth paths), Young’s theory, and classical stochastic integration for Volterra-type noise, connecting to Malliavin, S-transform, and rough paths approaches.

The pathwise fractional calculus achieves model-independent existence and stability results, provided the kernel and driving process meet well-specified regularity criteria (Nunno et al., 2016).

4. Pathwise Stochastic Integration for Discontinuous Integrands

Several frameworks extend pathwise integration to discontinuous evaluation functions and processes of unbounded $h$0-variation, relevant for Gaussian, fractional, and multifractional processes (Chen et al., 2014, Chen et al., 2016). The key aspects are:

• Composite integrands: For $h$1 of locally bounded variation and $h$2 a continuous process, the path $h$3 typically has infinite $h$4-variation, violating Young or rough path conditions.
• Density and variability requirements: For $h$5 to be Riemann–Stieltjes integrable pathwise, $h$6 must spread sufficiently through its range, quantified via density bounds and level-crossing estimates (Chen et al., 2016).
• Fractional Sobolev embedding: Integrands $h$7 and integrators $h$8 are handled in fractional Besov or Sobolev–Slobodeckij spaces $h$9 and $X$0, with the integral defined as a fractional Zähle–Stieltjes integral, agreeing with Riemann sum limits.
• Change of variables: With additional smoothness, classical Itô or change-of-variables formulas persist in a pathwise form (Chen et al., 2016, Chen et al., 2014).

This approach strictly extends Young integration to classes of discontinuous functionals with pathwise convergence of Riemann sums and error analysis.

5. Model-Free and Robust Finance: Outer Measure and Typical Paths

Model-free frameworks construct integrals for “typical” price paths based on (super-)hedging dualities and outer measures, without reference to any probability. Key features are (Perkowski et al., 2013, Bartl et al., 2018, Allan et al., 2021):

• Minimal superhedging outer measure $X$1: The least initial capital $X$2 so that gains from simple trading strategies and their limits dominate a contingent claim. A property holds for typical paths if the set where it fails is $X$3-null.
• Itô-style pathwise extension: Integrals of adapted càdlàg integrands against continuous paths $X$4 are extended via completeness in the $X$5-type outer-measure metric, with explicit tail bounds via pathwise Hoeffding inequalities.
• Controlled rough paths for finance: Every typical price path has a canonical Itô rough path lift $X$6; the rough integral of a controlled path against $X$7 is again a pathwise limit of non-anticipating Riemann sums, admitting robust analytic continuity and supporting robust SDE and PDE solution theory (Perkowski et al., 2013, Allan et al., 2021).
• Analytic stability: The rough-path topology on the space of “typical” price paths enables robust pricing, hedging, and sensitivity analyses under Knightian uncertainty, integrating statistical and adversarial (worst-case) financial models (Allan et al., 2021, Bartl et al., 2018).
• Infinite-dimensional and Hilbert space extensions: Outer measure and pathwise constructions extend naturally to $X$8 for infinite-dimensional $X$9, though certain continuity and BDG-type estimates are established in finite dimensions (Bartl et al., 2018).

These frameworks underpin model-free hedging, superhedging duality, and capital process definitions for a broad class of pathwise strategies and price dynamics.

6. Càdlàg Path Integration and Extended Föllmer Calculus

Pathwise Itô calculus has been generalized to càdlàg paths via partition-based quadratic variation (matching jumps squared) and extends Itô’s formula, integration by parts, and associativity to jump processes (Hirai, 2017):

• Definition via Riemann sums: Integrals $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$0 are defined as limits of sums $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$1 for admissible integrands (vertical derivatives of $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$2-functionals of $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$3 and auxiliary finite-variation paths).
• Fundamental properties: Associativity, extended Itô formula with drift, continuous and jump covariation terms, and explicit solutions for linear and nonlinear SDEs or portfolio insurance strategies persist pathwise (Hirai, 2017).
• Generalized drift and dispersion: Using scale transforms, one handles highly non-smooth dispersions and generalized drift terms (e.g., as in Karatzas-Ruf) entirely pathwise, via functional equations indexed by fixed paths (Karatzas et al., 2013).

This framework is particularly well adapted to financial modeling with jumps, transaction costs, and various exotic derivatives.

7. Numerical Pathwise Approximation and SPDEs

Pathwise integration underpins pathwise convergence analysis for stochastic numerical schemes and SPDE discretizations (Shardlow et al., 2014, Auestad, 22 Aug 2025):

• Truncation-error pathwise bounds: Almost-sure error bounds are derived via endpoint-local errors and cumulative truncation processes, enabling direct pathwise Grönwall-type estimates.
• Adaptive step-schemes: By tying time-step selection to local pathwise behavior, robust adaptive numerical schemes (e.g., for Euler–Maruyama) can be analyzed and shown to achieve optimal pathwise rates (Shardlow et al., 2014).
• Burkholder–Davis–Gundy inequalities in $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$4: Pathwise versions extend BDG-type estimates to stochastically integrable integrands, permitting pathwise analysis of convergence in probability and time-regularity for SPDE solutions (Auestad, 22 Aug 2025).
• Pathwise Kolmogorov continuity: Fine probabilistic and analytic criteria yield almost sure Hölder continuity of stochastic integrals in both finite and infinite dimensions.

This analytic architecture supports finite element and adaptive approximations of stochastic PDE solutions, even for irregular noise and minimal integrability assumptions.

---

Table: Summary of Distinct Pathwise Integration Paradigms

Framework/Method	Integrator Regularity	Integrands Allowed
Föllmer/Functional Itô	Finite quadratic variation	Gradients (vertical deriv.)
Rough paths	Finite $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$5-variation ($\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$6)	Controlled paths, path-dependent
Fractional calculus	Kernel-Fractional regularity	$\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$7
Discontinuous/Unbounded $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$8-var (Chen et al., 2014, Chen et al., 2016)	Hölder + density/variability	Locally bounded variation $\sum_{i=0}^{N(n)-1} h_{t_i} (X_{t_{i+1}}-X_{t_i}),$9
Outer measure (Finance)	Any continuous (typical) path	Càdlàg, step-approximable
Càdlàg Föllmer calculus	Càdlàg + jump-matching	Admissible (gradient of $h$0)
Pathwise numerics/SPDE	$h$1-or pathwise a.s. regular	Stochastically integrable

Conclusion

Pathwise stochastic integration frameworks have matured from simple Riemann sum approximations to rich, robust analytic theories encompassing roughness, discontinuities, non-semimartingale dynamics, high or infinite dimension, and robust financial applications. Their interplay with classical Itô theory is mediated via revisited notions of quadratic variation, Lévy area, and stability estimates, enabling functional extensions, product integrals, and solutions to stochastic differential and partial differential equations in purely pathwise settings. Model-free finance and probabilistically robust numerics crucially depend on these developments, making pathwise stochastic integration a cornerstone of modern stochastic analysis.