Vertical Derivative in Modern Analysis

• Vertical derivative is a concept describing differentiation in a direction orthogonal to horizontal or temporal derivatives, used in non-commutative and path-dependent analyses.
• It underpins the construction of vertical Sobolev spaces and fractional operators, playing a key role in proving embedding theorems and fractional Poincaré inequalities.
• In functional Itô calculus, the vertical derivative quantifies the sensitivity of path-dependent functionals, which is crucial for hedging strategies and non-Markovian financial models.

The vertical derivative is a concept that arises in multiple advanced mathematical fields, notably the analysis on the Heisenberg group and the functional Itô calculus for path-dependent problems. In both contexts, it describes differentiation in a direction orthogonal to conventional "horizontal" or temporal derivatives: for the Heisenberg group, vertical refers to the central direction in the group's stratification; for path-space calculus, vertical refers to changes in the trajectory of a stochastic process at a fixed time. The vertical derivative is crucial for expressing regularity and constructing differential or integral representations in settings where classical calculus is insufficient.

1. Vertical Derivative on the Heisenberg Group

The Heisenberg group $\mathbb{H}^n$ is identified with $\mathbb{R}^{2n+1}$ having coordinates $(z,t)$, with $z=(x,y)\in\mathbb{R}^{2n}$ and $t\in\mathbb{R}$. The group operation encodes non-commutative geometry, and the vertical direction in this setting corresponds to the center $Z=\{ (0,t): t\in\mathbb{R} \}$. For a real-valued function $f(z,t)$, its vertical difference is defined as $\Delta_r f(z,t) := f(z,t+r) - f(z,t)$ for $r \in\mathbb{R}$. The vertical fractional derivative of order $\alpha>0$ is implemented as a principal value operator:

$\mathbb{R}^{2n+1}$0

This operator generalizes basic differentiation to non-integer orders and leverages the vertical structure of $\mathbb{R}^{2n+1}$1 (Fässler et al., 2019).

2. Vertical Sobolev Spaces and Regularity

The vertical Sobolev space $\mathbb{R}^{2n+1}$2, for $\mathbb{R}^{2n+1}$3 and $\mathbb{R}^{2n+1}$4, consists of functions $\mathbb{R}^{2n+1}$5 such that, for almost every $\mathbb{R}^{2n+1}$6, the slice $\mathbb{R}^{2n+1}$7 belongs to the one-dimensional Bessel potential Sobolev space $\mathbb{R}^{2n+1}$8. The $\mathbb{R}^{2n+1}$9-norm is given by

$(z,t)$0

where $(z,t)$1. A seminorm employing difference quotients is equivalent:

$(z,t)$2

The main theorem is a continuous embedding: if $(z,t)$3 (the horizontal Sobolev space), then $(z,t)$4 and $(z,t)$5 (Fässler et al., 2019).

3. Consequences for Bounded Lipschitz Functions

For bounded Lipschitz functions on $(z,t)$6—specifically, $(z,t)$7 where the horizontal gradient is in $(z,t)$8—there is a significant result: the half-order vertical derivative $(z,t)$9 exists almost everywhere (both in $z=(x,y)\in\mathbb{R}^{2n}$0 and distributionally), and

$z=(x,y)\in\mathbb{R}^{2n}$1

This establishes the $z=(x,y)\in\mathbb{R}^{2n}$2-regularity for half-order vertical derivatives of horizontally regular functions, an endpoint result that does not hold in $z=(x,y)\in\mathbb{R}^{2n}$3 but does in $z=(x,y)\in\mathbb{R}^{2n}$4 (Fässler et al., 2019). The proof uses decomposition over Korányi balls, analysis of compactly-supported factors, and adaptation of local oscillation estimates.

4. Fractional Vertical–Horizontal Poincaré Inequalities

Advanced mixed-norm inequalities connect vertical and horizontal Sobolev/Besov regularity. For $z=(x,y)\in\mathbb{R}^{2n}$5, $z=(x,y)\in\mathbb{R}^{2n}$6, $z=(x,y)\in\mathbb{R}^{2n}$7, the vertical Besov-type norm is defined:

$z=(x,y)\in\mathbb{R}^{2n}$8

For $z=(x,y)\in\mathbb{R}^{2n}$9, $t\in\mathbb{R}$0,

$t\in\mathbb{R}$1

This generalizes the Poincaré inequalities of Lafforgue–Naor and establishes mixed regularity estimates, linking horizontal and vertical fractional orders via functional analytic methods (Fässler et al., 2019).

5. Vertical Derivative in Functional Itô Calculus

In the theory of path-dependent stochastic calculus, Dupire's vertical derivative is defined for non-anticipative functionals $t\in\mathbb{R}$2, $t\in\mathbb{R}$3. The vertical derivative at $t\in\mathbb{R}$4 is

$t\in\mathbb{R}$5

where the "bumped path" $t\in\mathbb{R}$6 is $t\in\mathbb{R}$7 for $t\in\mathbb{R}$8 and $t\in\mathbb{R}$9 for $Z=\{ (0,t): t\in\mathbb{R} \}$0 (Bouchard et al., 2021). The vertical derivative detects the sensitivity of $Z=\{ (0,t): t\in\mathbb{R} \}$1 to infinitesimal modifications in the path's future at $Z=\{ (0,t): t\in\mathbb{R} \}$2, providing a pathwise gradient in non-Markovian settings.

For functionals $Z=\{ (0,t): t\in\mathbb{R} \}$3 (continuous and vertically differentiable), the process $Z=\{ (0,t): t\in\mathbb{R} \}$4 has left-continuous paths, and modulus continuity bounds hold under local uniform boundedness (Bouchard et al., 2021).

6. Functional Itô Formula and Applications to Mathematical Finance

The $Z=\{ (0,t): t\in\mathbb{R} \}$5-functional Itô formula decomposes a path-dependent functional of a weak Dirichlet process $Z=\{ (0,t): t\in\mathbb{R} \}$6 as

$Z=\{ (0,t): t\in\mathbb{R} \}$7

with $Z=\{ (0,t): t\in\mathbb{R} \}$8 the local martingale component and $Z=\{ (0,t): t\in\mathbb{R} \}$9 continuous and orthogonal to any local martingale. For pricing in mathematical finance, given a payoff $f(z,t)$0, the pathwise replication formula is

$f(z,t)$1

where $f(z,t)$2 is the pricing functional and $f(z,t)$3 prescribes the hedging strategy (Bouchard et al., 2021).

In robust hedging under volatility uncertainty, the value function $f(z,t)$4 and the associated optimal super-hedge $f(z,t)$5 also depend fundamentally on the vertical derivative, enabling universal integrand extraction for all martingale measures (Bouchard et al., 2021).

7. Significance in Analysis and Probability

The vertical derivative distinguishes itself by its capacity to capture fine regularity orthogonal to conventional gradients or time derivatives and to facilitate functional analytic and stochastic representations where classical notions fail. On the Heisenberg group, it enables sharp embedding and $f(z,t)$6-type regularity results, as well as mixed-norm Poincaré inequalities linking directions. In functional Itô calculus, it is pivotal for extending Itô's formula to non-Markovian and path-dependent problems, directly encoding the sensitivity required for path-dependent hedging and super-hedging in finance. Its operationalization in both pure analysis (Fässler et al., 2019) and stochastic calculus (Bouchard et al., 2021) demonstrates its versatility and centrality to modern mathematical frameworks.