Functions of Bounded Variation on the Classical Wiener Space and an Extended Ocone-Karatzas Formula

Maurizio Pratelli, Dario Trevisan

This paper, published in 2011 by Pratelli and Trevisan, bridges a gap between Malliavin calculus and geometric measure theory. It extends the Ocone-Karatzas formula—a fundamental tool for representing random variables as stochastic integrals—to the broader class of functions of Bounded Variation (BV) on the classical Wiener space. This allows researchers to rigorously handle non-differentiable random variables, such as indicator functions and the maximum of a process, using the language of stochastic analysis.

TL;DR

The authors extend the Ocone-Karatzas integral representation to non-smooth "Bounded Variation" functions on Wiener space. By developing a new chain rule and leveraging measure-theoretic tools, they provide explicit stochastic integral representations for complex, non-differentiable functionals like the maximum of a Wiener process.

What Problem Does This Paper Address?

In stochastic calculus, we often want to represent a random variable (a "functional" of a path) as an initial value plus a running integral. The Ocone-Karatzas formula is the standard way to do this, but it traditionally requires the random variable to be "smooth" in the sense of Malliavin differentiability.

However, many real-world applications—especially in mathematical finance (e.g., digital options) and physics—involve "rough" functions, such as indicator functions ($I_{x>g}$) or the maximum value a process reaches. These functions aren't differentiable in the classical Malliavin sense. The paper addresses this limitation by moving into the realm of BV functions, which allow for "jumps" or singular behavior by treating their derivatives as measures rather than simple functions.

Methods

The authors employ several sophisticated mathematical techniques to bridge this gap:

  1. Measure-Valued Derivatives: Instead of a standard Malliavin derivative (which is a stochastic process), the authors define the derivative of a $BV$ function as a vector-valued measure ($Df$).
  2. Predictable Projections: A major challenge with $BV$ functions is that their derivatives can be "singular" (concentrated on tiny sets). The authors prove that when you project this measure onto the strictly predictable $\sigma$-algebra, it becomes "absolutely continuous." This allows them to extract a well-behaved density ($H$) that serves as the integrand in the stochastic integral.
  3. Approximation: They use a sequence of smooth functions to approximate $BV$ functions (Theorem 6), proving that the properties of the smooth versions carry over to the $BV$ limit.
  4. Chain Rule for BV Functions: They establish a new chain rule for composite functions (e.g., $f = \phi \circ g$), which is essential for calculating the representations of functions like $I_{\{g > x\}}$.

Datasets

Not discussed in the paper. This is a theoretical research paper in pure mathematics and stochastic analysis; it does not use empirical datasets or benchmarks.

Key Insights and Ideas

  • The "Aha!" Moment on Predictability: The most striking insight is that while the derivative of a $BV$ function might be a messy, singular measure on the whole Wiener space, it becomes a standard, integrable process once you restrict your view to "predictable" information. This is what makes the stochastic integral representation possible.
  • Simplifying the Technical Burden: The authors argue that $BV$ theory provides a "clear proof without advanced technical results," such as the theory of distributions (S-transforms) often used in white noise analysis.
  • Geometric Measure Theory meets Stochastic Paths: By adapting tools from Euclidean geometric measure theory to the infinite-dimensional Wiener space, the authors provide a more robust framework for "level sets"—essentially the boundaries where a process hits a certain value.

Key Results

  • Theorem 16 (The Extended Ocone-Karatzas Formula): This is the centerpiece. It proves that for any $BV$ function $f$, there exists a predictable process $H$ such that $f = E[f] + \int H_s dW_s$.
  • Explicit Representations: The paper provides concrete formulas for:
  • Indicator Functions: The representation for $I_{\{W_t \ge 0\}}$, which is a classic result in finance, is derived here as a simple application of the new theory.
  • The Maximum Process: They derive an explicit representation for $\phi(M)$, where $M$ is the maximum of a Wiener process over time $T$. This involves the density of the running maximum, showing the theory's power in handling path-dependent functionals.

Limitations and Future Work

  • Chain Rule Constraints: The authors admit that their chain rule is "still very distant from the deep results" available in finite-dimensional Euclidean settings. It works for specific classes of $BV$ functions but isn't yet a universal "master" chain rule for all infinite-dimensional cases.
  • Abstract Wiener Spaces: While the results are formulated for the classical Wiener space (trajectories of Brownian motion), the authors suggest that many of their findings are likely valid in more general "Abstract Wiener Spaces," though they leave the formal extension for future research.
  • Application to "Greeks": The paper hints that this framework is a useful tool for dealing with classical problems, likely referring to the calculation of "Greeks" (sensitivities) for non-smooth payoffs in finance.

Summary updated

Abstract

We prove an extension of the Ocone-Karatzas integral representation, valid for all $BV$ functions on the classical Wiener space. We establish also an elementary chain rule formula and combine the two results to compute explicit integral representations for some classes of $BV$ composite random variables.
BibTeX
@inproceedings{pratelli2012functions,
  author = {Maurizio Pratelli and Dario Trevisan},
  title = {{Functions of Bounded Variation on the Classical Wiener Space and an Extended Ocone-Karatzas Formula}},
  booktitle = {Stochastic},
  year = {2012},
  doi = {10.1016/j.spa.2012.03.010},
}
Figure