Pathwise Stochastic Integration à la Föllmer

• Pathwise stochastic integration à la Föllmer is a model-free method that constructs stochastic integrals from path properties like finite quadratic variation and convergent Riemann sums.
• It extends classic integration theory by incorporating càdlàg paths, Banach space trajectories, and rough path techniques, enhancing robustness in financial applications.
• The framework underpins universal function approximation via extended signatures and ensures numerical stability through partition invariance across varied path classes.

Pathwise stochastic integration à la Föllmer is a model-free, analytic construction of stochastic integrals that eschews reliance on probabilistic or martingale properties, instead anchoring the theory in pointwise properties of paths such as finite quadratic variation and the convergence of discrete Riemann sums. Initially developed for continuous paths, the theory now encompasses càdlàg and Banach space-valued trajectories, leverages rough paths and controlled paths, and underpins universal approximation results via signatures of extended rough paths. Föllmer’s construction has catalyzed advances in robust mathematical finance, rough path theory, and model-free stochastic analysis.

1. Quadratic Variation and Lévy’s Area: Definition and Role

The analytic foundation of pathwise stochastic integration is the notion of quadratic variation along a fixed sequence of partitions. For a path $X:[0,T]\rightarrow\mathbb{R}^d$ and a sequence of refining partitions $\pi^n = \{ t^n_0 < t^n_1 < \cdots < t^n_{N_n}=T \}$, $X$ is said to have quadratic variation along $\pi$ if

$$[X]^{\pi}_t := \lim_{n \to \infty} \sum_{k=0}^{N_n-1} (X_{t^n_k \wedge t,\, t^n_{k+1} \wedge t}) \otimes (X_{t^n_k \wedge t,\, t^n_{k+1} \wedge t})$$

exists uniformly in $t$ (Das et al., 23 Jul 2025). In multi-dimensional settings, the existence of the Lévy area—formally,

$$L_t^{\pi}(X) := \lim_{n\to\infty}\sum_{k=0}^{N_n-1}(X_{t^n_k \wedge t}+X_{t^n_{k+1}\wedge t}) \otimes X_{t^n_k \wedge t,\, t^n_{k+1}\wedge t}$$

—is essential for describing the antisymmetric part of the second-order structure. For vector-valued signals, these higher-order increments are crucial for any pathwise extension beyond classical integration, especially Stratonovich-type and rough-integral regimes (Das et al., 23 Jul 2025, Imkeller et al., 2014).

2. Föllmer’s Pathwise Integral: Construction and Fundamental Properties

The Föllmer integral is defined as the uniform limit of Riemann sums along the fixed partition sequence, typically taking the form

$$\int_0^t H_{s^-}\,dX_s := \lim_{n\to\infty} \sum_{k=0}^{N_n-1} H_{t^n_k} (X_{t^n_{k+1}\wedge t}-X_{t^n_k\wedge t})$$

for a càdlàg integrand $H$ and quadratic variation path $X$ (Hirai, 2017).

The integral is linear in the integrand, exhibits associativity (a chain-rule for nested integrals of admissible, typically gradient-type, integrands), and satisfies the integration by parts formula: $$X_t Y_t - X_0 Y_0 = \int_0^t X_{s^-}\, dY_s + \int_0^t Y_{s^-}\, dX_s + [X,Y]_t.$$ Key algebraic and analytic properties, such as the pathwise Itô formula, hold for $C^2$ functions $f$: $$f(X_t) - f(X_0) = \int_0^t f'(X_{s^-})\,dX_s + \frac{1}{2} \int_0^t f''(X_{s^-})\,d[X]_s + \Sigma_{jumps},$$ with jump and second-order corrections captured explicitly (Hirai, 2017, Hirai, 2021).

3. Extension to Non-Geometric and Rough Paths

Recent developments generalize Föllmer’s framework using rough path theory. For a fixed $\gamma\in[0,1]$, Föllmer-type integrals interpolate between Itô ($\gamma=0$), Stratonovich ($\gamma=\frac12$), and backward Itô ($\gamma=1$), through generalized Riemann sums: $$S_n(t;Y,X;\gamma) := \sum_{k=0}^{N_n-1} (Y_{t^n_k} + \gamma (Y_{t^n_{k+1}}-Y_{t^n_k}))(X_{t^n_k\wedge t, t^n_{k+1}\wedge t}).$$ The existence of both quadratic variation and Lévy area ensures that a controlled path $(Y, Y')$ ($Y_t \sim Y'_t X_{t,t+h}$) allows for well-posed rough integration, with the rough path object $(X, \mathbb{X}^2)$ and a suitable second-order lift (Das et al., 23 Jul 2025, Imkeller et al., 2014).

Comparison with classical rough paths: In classical rough path theory, Stratonovich integrals yield weakly geometric rough paths (group-like signatures). In contrast, Föllmer integration in the Itô setting, combined with extension of quadratic variation as a path-level object, produces non-geometric rough paths, but appropriately extended signatures regain the algebraic universal approximation property (Ceylan et al., 5 Feb 2026).

4. Pathwise Universal Approximation and Signature Methods

The extension of rough path signatures by quadratic variation coordinates under Föllmer integration yields strong universal function approximation properties. For continuous paths with finite $p$-variation ($2 < p < 3$), the extended signature

$\widetilde X_t := (t, X_t, \langle X \rangle_t),$

where $\langle X \rangle_t$ collects all quadratic variations, forms a non-geometric rough path whose signatures are dense in the space of continuous functionals on compact rough path sets (Ceylan et al., 5 Feb 2026). This enables linear functionals of signatures to approximate arbitrary continuous path functionals, both in a purely analytic and probabilistic (semimartingale) context.

Applications encompass model calibration and financial derivatives pricing where incorporating realized quadratic variation as an explicit feature yields lower mean squared error and more accurate pricing for path-dependent claims (Ceylan et al., 5 Feb 2026).

5. Invariance and Stability: Partition Independence

Quadratic variation and Lévy area, as computed along partition sequences, can potentially depend on the choice of partition. Recent analysis introduces criteria for "quadratic roughness" and "Lévy roughness," ensuring, respectively, that quadratic variations and areas coincide across classes of partitions—chiefly, balanced dyadic and adaptive sequences. When these roughness criteria are satisfied, the pathwise integral and the associated rough path object are invariant to the partition choice, strengthening both the conceptual foundation and numerical stability of the theory (Das et al., 23 Jul 2025).

6. Banach Space-Valued Integrators and Functional Calculus

Generalizations to Banach space-valued paths have been established through the introduction of tensor or $B$-covariation. Letting $X:[0,T]\to E$, $E$ a Banach space, quadratic variation is defined as convergence in the projective tensor topology: $${}^\alpha [X,X]_t := \lim_{n\to\infty} \sum_{[r,s]\in\pi_n} (X_{s\wedge t}-X_{r\wedge t})^{\otimes2}$$ in norm (or weak tensor topology). The pathwise Itô–Föllmer formula then extends to vector-valued, and even Hilbert-space-valued, paths (Hirai, 2021, Bartl et al., 2018), with the central result being: $$f(A_t,X_t)-f(A_0,X_0) = \int_0^t D_a f\, dA + \int_0^t D_x f\, dX + \frac{1}{2}\int_0^t D^2_B f\, dQ_B(X,X) + \sum_{0<s\le t} [\Delta f - D_x f \Delta X].$$ This formula forms the basis for functional Itô calculus and facilitates unique pathwise solutions for SDEs driven by such integrators on typical paths (Hirai, 2021, Bartl et al., 2018).

7. Applications, Numerical Aspects, and Algorithmic Developments

Pathwise stochastic integration à la Föllmer underpins model-free pricing and hedging frameworks in mathematical finance, particularly those seeking robustness to probabilistic model instability. In this context, tools like Vovk’s outer measure (superhedging price) ensure that quadratic variation and pathwise integrals exist quasi-surely under broad model classes, and can be calculated as limits of Riemann–Stieltjes sums computed along adaptive or dyadic partitions (Łochowski et al., 2016, Perkowski et al., 2013).

Algorithmic computation of Föllmer integrals, especially via Schauder function expansions and recursive blockwise approximations, results in explicit convergent schemes with quantifiable error rates in Hölder norms. This framework handles both Stratonovich and Itô types, together with rough path-based and controlled path methodologies, in a unified numerical fashion (Gubinelli et al., 2014).

8. Limitations, Extensions, and Open Problems

While Föllmer’s approach rests on minimal (analytic) assumptions and is fully model-free, several limitations persist. For full generality and “universality,” existence of quadratic variation (and sometimes Lévy area) must be verified along the specified partition family—a feature not inherent to all rough or irregular paths without augmentation. Admissibility of integrands is restricted relative to the Bichteler–Dellacherie–Kunita–Watanabe Itô theory; only gradient-types or controlled paths guarantee existence of limits unless additional structure (e.g., superhedging topology, density constraints, or pathwise functional analytic arguments) is invoked (Hirai, 2017, Chen et al., 2016). Further scope lies in integration with respect to non-semimartingale, highly singular, or discontinuous sample paths (Łochowski, 2012).

Nevertheless, Föllmer’s pathwise paradigm remains the analytic and conceptual prototype for model-free stochastic calculus, universal rough representations, and robust quantitative finance.