The Geometry of Rough Path Space

Abstract: We describe $H^p(V)$, a subset of $p$-rough path space $Ω_p(V)$ which is a vector space under an addition operation $\boxplus$ and a scalar multiplication $\odot$. We show that the domain of $\boxplus$ can be extended to $Ω_p(V)\times H^p(V)$, allowing any $p$-rough path $X$ to be additively perturbed by an $H\in H^p(V)$. We prove associativity $(X\boxplus H)\boxplus \tilde H = X\boxplus (H\boxplus \tilde H)$ and trivial kernel $X\boxplus H = X \Leftrightarrow H = 1$, where $1$ is the additive zero in $(H^p(V),\boxplus,\odot)$. Finally, we show that enlarging $H^p(V)$ to almost rough paths $H^{am,p}(V)$ does not enlarge the set of displacements of a given $X$, i.e. ${X\boxplus H: H\in H^p(V)}={X\boxplus H: H\in H^{am,p}(V)}$.

• The paper introduces a canonical linear structure by defining an addition (boxplus) and scalar multiplication on a subspace, transforming rough path perturbations.
• It establishes the vector space structure of H^p(V) through a bijective lift map, linking genuine rough increments with additive Young regular paths.
• It extends the framework to almost rough paths, ensuring stability in perturbations and providing insights for numerical methods and potential manifold generalizations.

The Geometry of Rough Path Space: An Authoritative Summary

Introduction and Motivation

This paper introduces a novel linear structure within the well-established but notoriously non-linear framework of rough path theory. Building on Lyons' rough path space $\Omega^p(V)$—traditionally lacking a canonical addition operation—the authors construct a subspace $\mathscr{H}^p(V)$ and define on it an intrinsically motivated addition $\boxplus$ and scalar multiplication $\odot$. These operations render $\mathscr{H}^p(V)$ a vector space and enable canonically defined additive perturbations of rough paths, extending classical results to genuinely rough increments rather than only perturbations of bounded variation or Young type.

The motivation is both geometric and analytic: while rough path theory gives a robust framework for solving controlled differential equations and stochastic analysis, the absence of linear or tangent space structure for $\Omega^p(V)$ has remained a structural limitation. Previous approaches (e.g., Friz–Victoir's translation operators, Qian–Tudor's tangent bundles, and Bellingeri et al.'s smooth rough models) handle particular classes of perturbations or impose additional smoothness or geometric constraints, often failing to give an internal, genuinely linear displacement structure across the entire rough path space.

Main Results and Theoretical Contributions

H-space: Construction and Structure

The primary innovation is the identification of a distinguished subset $\mathscr{H}^p(V) \subset \Omega^p(V)$ whose elements act as "rough displacements." Each $H \in \mathscr{H}^p(V)$ is a $p$-rough path with level-wise regularity

$\|H^j_{s,t}\|\leq K \omega(s,t)^\phi$

for some $\phi \in (1-1/p,1]$ and control $\omega$. This condition enables a controlled notion of additive perturbation: given any rough path $X$, the perturbed path $X \boxplus H$ is defined via the rough sewing lemma applied to the almost rough path $X \oplus H$ (unit-preserving pointwise addition in the truncated tensor algebra).

Key results include:

• Vector space structure: $\mathscr{H}^p(V)$ admits canonical addition $\boxplus$ and induced scalar multiplication $\odot$ (pulled back from increments of degree-wise Young paths), making it a genuine vector space. The zero element is the trivial rough path $\mathbb{1}$.
• Associativity: The extension $\boxplus$ is associative, i.e., $$(X\boxplus H)\boxplus \widetilde H = X\boxplus(H\boxplus \widetilde H)$$ for $X\in\Omega^p(V)$ and $H,\widetilde{H}\in\mathscr{H}^p(V)$.
• Trivial kernel: The action is free; $X\boxplus H = X$ implies $H = \mathbb{1}$.

Lift Map and Equivalence with Additive Increments

A canonical bijection is established between $\mathscr{H}^p(V)$ and a space of additive, degree-wise Young regularity path increments $\mathfrak{I}^p(V)$, implemented via the iterated extension ("lift") map $\mathbb{1}^{(\cdot)}$ and its inverse, the development map $\operatorname{dev}$. This correspondence is robust at the level of topology and operations:

• $\mathscr{H}^p(V) \simeq \mathfrak{I}^p(V)$ as vector spaces.
• The operation $\boxplus$ on lifts is compatible with addition in increment space: $$\mathbb{1}^I \boxplus \mathbb{1}^{\widetilde{I}} = \mathbb{1}^{I+\widetilde{I}}$$ for $I, \widetilde{I} \in \mathfrak{I}^p(V)$.
• Scalar multiplication is defined by transport of structure: $a\odot \mathbb{1}^I = \mathbb{1}^{aI}$.

Robustness to Almost Rough Perturbations

The framework is further generalized to include "almost rough paths" ($\mathscr H^{\text{am},p}(V)$), where the strict multiplicativity is relaxed to a quasi-multiplicativity bound. The main result is that the set of attainable perturbations from any base path $X$ is unchanged when extending from $\mathscr{H}^p(V)$ to $\mathscr{H}^{\text{am},p}(V)$: $$\{X\boxplus H : H\in\mathscr{H}^p(V)\} = \{X\boxplus H : H\in\mathscr{H}^{\text{am},p}(V)\}.$$ Thus, the space of true displacements is exhausted already by genuine rough increments in $\mathscr{H}^p(V)$.

Comparison with Existing Literature

This construction generalizes and unifies various attempts at introducing linear structure and perturbation theory in rough path spaces:

• For smooth paths and Young regular increments, classical translation and signature operations are recovered as a special case of $\boxplus$.
• Unlike the translation operators of Friz–Victoir, which act only on weakly geometric rough paths (and involve group or Young regularity constraints), the $\boxplus$ operation is defined for all rough paths in $\Omega^p(V)$ and applies to genuinely non-geometric objects.
• In contrast to the approach by Qian–Tudor, which provides a tangent bundle structure dependent on ambient Banach spaces and non-canonical extensions, the present formalism yields a globally trivialized vector bundle of displacements, internal to $\Omega^p(V)$, and valid for all $p \geq 1$.
• While the work of Bellingeri–Friz–Paycha–Preiss constructs a canonical vector space only on smooth geometric rough models, the authors' $\mathscr{H}^p(V)$-space acts on (possibly non-smooth, non-geometric) genuine rough paths.

Numerical and Structural Implications

The precise identification of $\mathscr{H}^p(V)$ as vector space, along with constructive formulas for perturbation ($\boxplus$), supports not only new perspectives in the geometric analysis of $\Omega^p(V)$ but also tangible algorithmic consequences:

• The explicit induction and formulaic characterization of $(X\boxplus H)^j$ in terms of lower level (tensor) increments, canonical extension maps, and additive innovations enable direct computation of perturbed signatures.
• The vector space properties facilitate the analysis of variations, sensitivity, and derivatives of path-dependent functionals in rough path analysis, potentially impacting the study of flows, stability, and Malliavin calculus for rough differential equations.
• The closure under almost rough perturbations supports the stability of solution maps in the presence of non-ideal (e.g., non-multiplicative) discretizations or noise, relevant for both theoretical analysis and numerical approximation.

Broader Implications and Future Directions

This treatment closes a conceptual gap in rough path theory, endowing the path space $\Omega^p(V)$ with a distinguished displacement space and canonical action. Further directions that naturally arise include:

• Extending these constructions to infinite-dimensional Banach (or even Fréchet) spaces $V$ and links to signature-based machine learning or regularity structures.
• Investigating the manifold structure or generalized geometry (e.g., developing local linearizations and a differential geometry of rough path space) grounded in this internal vector space.
• Applying the $\mathscr{H}^p(V)$-displacement formalism in stochastic analysis for the rigorous treatment of pathwise differentiability, stochastic flows, and renormalization in singular SPDEs.

Conclusion

By identifying a canonical vector space of increments inside rough path space, providing a robust, associative addition compatible with Lyons' sewing lemma, and establishing a bijective correspondence with degree-wise Young regular additive paths, the authors give a comprehensive and flexible linearization theory for rough paths. This advances the structural understanding of $\Omega^p(V)$ and furnishes powerful tools for further analysis, both theoretical and computational, across stochastic analysis and applied probability.