Canonical Rough Path over Tempered Fractional Brownian Motion: Existence, Construction, and Applications

Atef Lechiheb

arxiv: 2512.04646 · v2 · submitted 2025-12-04 · 🧮 math.PR

Canonical Rough Path over Tempered Fractional Brownian Motion: Existence, Construction, and Applications

Atef Lechiheb This is my paper

Pith reviewed 2026-05-17 01:43 UTC · model grok-4.3

classification 🧮 math.PR

keywords rough pathtempered fractional Brownian motiongeometric rough path2D rho-variationrough differential equationsMilstein schemesignature method

0 comments

The pith

Tempered fractional Brownian motion supports a canonical geometric rough path lift for any Hurst index above one quarter.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proves that d-dimensional tempered fractional Brownian motion admits a canonical geometric rough path for every Hurst parameter greater than 1/4 and every positive tempering parameter. The key step is showing that the covariance function has finite two-dimensional rho-variation with rho equal to one over twice the Hurst index. This property satisfies the Friz-Victoir criterion and therefore guarantees the existence of a rough path lift that can be constructed explicitly as an L2 limit. The resulting framework distinguishes integration regimes, establishes well-posedness for rough differential equations, supplies a Milstein scheme of optimal rate, and yields the signature with factorial decay.

Core claim

The central claim is the existence of the canonical geometric rough path B_{H,λ} = (B_{H,λ}, BB_{H,λ}) over tempered fractional Brownian motion. This is obtained by establishing that the non-homogeneous covariance possesses finite 2D ρ-variation for ρ = 1/(2H), which directly invokes the Friz-Victoir theorem, followed by an L2-limit construction that supplies explicit bounds depending on H, λ and T.

What carries the argument

The finite 2D ρ-variation of the tempered fractional Brownian motion covariance with ρ = 1/(2H), which meets the Friz-Victoir criterion and thereby produces the geometric rough path lift.

If this is right

Young integration works against tfBm when H > 1/2, while rough path theory is required and sufficient when 1/4 < H ≤ 1/2.
Rough differential equations driven by tfBm are well-posed and admit a Milstein scheme with strong convergence rate of order n to the power of minus H.
The signature of tfBm exists and decays factorially.
The boundary case H = 1/2 recovers the Stratonovich lift of the Ornstein-Uhlenbeck process, and the limit λ to zero recovers classical Itô calculus.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same covariance-variation argument may apply to other Gaussian processes whose covariance mixes power-law and exponential regimes.
Pathwise methods developed here could support numerical simulation of mean-reverting processes with memory in applications such as finance or turbulence modeling.
Higher-order rough path lifts or alternative approximations to the Lévy area might be derived from the same explicit L2 construction.

Load-bearing premise

The covariance of tempered fractional Brownian motion has finite two-dimensional rho-variation with rho equal to one over twice the Hurst parameter.

What would settle it

A direct calculation showing that the two-dimensional rho-variation of the covariance is infinite for some choice of Hurst index H greater than 1/4 and tempering parameter λ greater than 0.

Figures

Figures reproduced from arXiv: 2512.04646 by Atef Lechiheb.

**Figure 1.** Figure 1: Convergence of the Lévy area approximation. (Left) L 2 error e(N) versus the number of intervals N for H ∈ {0.3, 0.4, 0.6} and λ = 1. Dashed lines show the theoretical slope −2H. (Right) Error for fixed H = 0.4 and varying λ ∈ {0.1, 1, 10}; the rate remains −0.8 while the constant prefactor decreases with larger λ (stronger tempering). The shaded regions indicate ±1 standard error from M = 1000 samples. Th… view at source ↗

**Figure 2.** Figure 2: Strong convergence of the Milstein scheme for the linear RDE (14). (Left) Strong error Estrong(n) versus the number of steps n for H ∈ {0.3, 0.4, 0.6, 0.7}, λ = 1. Dashed lines indicate the theoretical slope −H. (Right) A single sample path of the solution Yt for H = 0.4, λ = 1, together with its Milstein approximation (n = 100). 5.4 Signature-Based Feature Extraction As an illustration of the signature ca… view at source ↗

**Figure 3.** Figure 3: Signature features of tfBm. Scatter plot of the first two signature levels (S 1 , S2 ) for 500 sample paths (T = 1, λ = 1). Colors indicate the true Hurst parameter H. The ellipses show the theoretical 95% confidence regions derived from Theorem 4.6. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗

read the original abstract

We construct a canonical geometric rough path over $d$-dimensional tempered fractional Brownian motion (tfBm) for any Hurst parameter $H > 1/4$ and tempering parameter $\lambda > 0$. The main challenge stems from the non-homogeneous nature of the tfBm covariance, which exhibits a power-law structure at small scales and exponential decay at large scales. Our primary contribution is a detailed analysis of this covariance, proving it has finite 2D $\rho$-variation for $\rho = 1/(2H)$. This verifies the criterion of Friz and Victoir, guaranteeing the existence of a rough path lift. We provide an explicit construction of the rough path $\mathbf{B}_{H,\lambda} = (B_{H,\lambda}, \mathbb{B}_{H,\lambda})$ via $L^2$-limits, establishing its basic properties with explicit constants $C(H,\lambda,T)$. As direct consequences, we obtain: (i)~a complete characterisation of integration regimes, with Young integration applicable for $H > 1/2$ and rough path theory necessary and sufficient for $H \in (1/4, 1/2]$; (ii)~the well-posedness of rough differential equations driven by tfBm, together with a Milstein-type numerical scheme of optimal strong convergence rate $\bigO(n^{-H})$; and (iii)~the foundation for signature calculus for tfBm, including the existence and factorial decay of the signature. The boundary case $H = 1/2$ is treated explicitly, recovering the Stratonovich lift of the Ornstein--Uhlenbeck process and, as $\lambda \to 0^+$, classical It\^o calculus. Numerical experiments confirm the theoretical convergence rates $\bigO(N^{-2H})$ for the L\'evy area approximation and $\bigO(n^{-H})$ for the Milstein scheme. This work provides the first comprehensive pathwise framework for stochastic calculus with tfBm.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

They verify finite 2D rho-variation on the tempered fBm covariance to get a canonical rough path lift via Friz-Victoir, then build explicit L2 construction and Milstein scheme.

read the letter

The main point is that this paper gives the first canonical geometric rough path over tempered fractional Brownian motion for H > 1/4. They do it by showing the covariance has finite 2D rho-variation with rho = 1/(2H), which triggers the Friz-Victoir existence theorem, followed by an L2-limit construction of the lift with explicit constants depending on H, lambda, and T. From there they recover the usual consequences: Young integration above 1/2, rough-path integration below, well-posed RDEs, a Milstein scheme with strong rate O(n^{-H}), and signature existence with factorial decay. The H = 1/2 case reduces to the Stratonovich lift of the Ornstein-Uhlenbeck process, and the lambda to 0 limit recovers classical fBm, which are useful sanity checks. Numerical tests confirm the predicted Levy-area and scheme rates. What is actually new is the variation estimate for this particular non-homogeneous covariance that mixes power-law small-scale behavior with exponential large-scale decay; prior fBm work does not cover the tempered case directly. The construction itself follows standard lines once the variation bound is in hand. The load-bearing step is that 2D-variation proof. Because the covariance is neither stationary nor translation-invariant, the argument has to control the double sums uniformly over all partitions of [0,T] x [0,T] by splitting small-scale increments from the tempered tails. If that splitting works cleanly without extra regularity assumptions, the rest is routine. The abstract claims a detailed analysis, so the paper presumably supplies the estimates, but that is the section worth checking line by line. This is useful for readers who already work with rough paths on Gaussian processes and want to extend the toolkit to models with tempered correlations. It is not a foundational rewrite of the theory, but it fills a concrete gap with explicit rates and numerics. I would send it to a serious referee rather than desk-reject; the existence claim and the numerical validation are concrete enough to merit review even if the variation estimates need tightening.

Referee Report

2 major / 2 minor

Summary. The manuscript constructs a canonical geometric rough path over d-dimensional tempered fractional Brownian motion (tfBm) for Hurst indices H > 1/4 and tempering parameters λ > 0. The key step is proving that the covariance function has finite two-dimensional ρ-variation with ρ = 1/(2H), which allows application of the Friz-Victoir theorem to guarantee existence of the lift. An explicit L²-limit construction is given, with basic properties stated using explicit constants depending on H, λ, and T. Applications include characterization of integration regimes, well-posedness of rough differential equations, a Milstein-type scheme with rate O(n^{-H}), and foundations for signature calculus. The boundary cases H = 1/2 and λ → 0 are treated, recovering Stratonovich and Itô calculus respectively. Numerical experiments confirm the predicted convergence rates.

Significance. If the central analytic estimate on the covariance holds, the work establishes the first comprehensive pathwise rough-path framework for tfBm, extending the theory beyond stationary fractional Brownian motion to processes with tempered correlations. The explicit constants, L² construction, and numerical validation of rates O(N^{-2H}) for Lévy area and O(n^{-H}) for the scheme are notable strengths. This provides a foundation for stochastic calculus with tfBm that is both theoretically rigorous and practically applicable, with potential impact on modeling in finance and physics where tempered long-memory processes arise.

major comments (2)

[Abstract (covariance variation analysis)] The claim that the covariance R_{H,λ}(s,t) possesses finite 2D ρ-variation for ρ = 1/(2H) is the load-bearing analytic input (see abstract and the detailed analysis referenced therein). While the argument exploits power-law small-scale behavior together with exponential large-scale decay, the non-stationary and non-translation-invariant nature of the covariance requires that the 2D-variation sums be controlled uniformly over all partitions of [0,T]×[0,T]. Any gap in the splitting argument between small-scale increments and tempered tails would invalidate the application of the Friz-Victoir criterion and therefore the existence of the canonical lift.
[Construction section (L²-limits)] The L²-limit construction of the rough path B_{H,λ} = (B_{H,λ}, ℬ_{H,λ}) is stated to satisfy basic properties with explicit constants C(H,λ,T). It would strengthen the result to verify explicitly that these constants remain finite and independent of the approximating sequence uniformly for the full range H > 1/4 (including the boundary behavior as H ↓ 1/4).

minor comments (2)

[Notation] Ensure consistent notation for the rough path lift (boldface vs. non-bold) throughout the manuscript.
[References] The citation to the Friz-Victoir theorem should include the precise statement or theorem number invoked for the existence criterion.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. The two major comments concern the uniformity of the 2D ρ-variation estimate and the explicit control of constants in the L² construction. We address each point below and will revise the manuscript to incorporate clarifications and additional verifications.

read point-by-point responses

Referee: [Abstract (covariance variation analysis)] The claim that the covariance R_{H,λ}(s,t) possesses finite 2D ρ-variation for ρ = 1/(2H) is the load-bearing analytic input (see abstract and the detailed analysis referenced therein). While the argument exploits power-law small-scale behavior together with exponential large-scale decay, the non-stationary and non-translation-invariant nature of the covariance requires that the 2D-variation sums be controlled uniformly over all partitions of [0,T]×[0,T]. Any gap in the splitting argument between small-scale increments and tempered tails would invalidate the application of the Friz-Victoir criterion and therefore the existence of the canonical lift.

Authors: We thank the referee for highlighting this crucial point. In Section 3, the finite 2D ρ-variation is established by splitting the double sum over an arbitrary partition of [0,T]×[0,T] into small-scale increments (where the time separation is less than a fixed δ) and large-scale increments. The small-scale contribution is controlled uniformly by the local power-law singularity of the covariance, which is position-independent due to the explicit form of the tempered kernel. The large-scale contribution is bounded using the exponential decay e^{-λ|s-t|}, which produces a geometric factor independent of the partition and uniform in T. The resulting bound depends only on H, λ, and T. While we believe the argument is complete, we agree that the uniformity deserves a more explicit statement. We will revise the manuscript by adding a remark after the main variation estimate that isolates the splitting constants and confirms their independence from the choice of partition. revision: yes
Referee: [Construction section (L²-limits)] The L²-limit construction of the rough path B_{H,λ} = (B_{H,λ}, ℬ_{H,λ}) is stated to satisfy basic properties with explicit constants C(H,λ,T). It would strengthen the result to verify explicitly that these constants remain finite and independent of the approximating sequence uniformly for the full range H > 1/4 (including the boundary behavior as H ↓ 1/4).

Authors: We agree that an explicit verification of uniformity with respect to the approximating sequence would strengthen the presentation. The constants C(H,λ,T) appearing in the basic properties are derived directly from the ρ-variation norm of the covariance and are therefore independent of any particular mollifier or dyadic approximation. The L²-Cauchy criterion for the iterated integrals likewise depends only on these a priori bounds. For each fixed H > 1/4 the constants are finite; they diverge as H ↓ 1/4, consistent with the loss of regularity. To make this transparent, we will add a short paragraph (or short appendix) in the construction section that records the dependence of the estimates on the approximation parameter and confirms that the passage to the limit preserves the same constants for every fixed H > 1/4. revision: yes

Circularity Check

0 steps flagged

No circularity: independent analytic estimate of covariance variation feeds external Friz-Victoir theorem

full rationale

The paper's derivation chain begins with a primary contribution consisting of a detailed analysis proving that the tfBm covariance possesses finite 2D ρ-variation for ρ = 1/(2H). This estimate is then used to verify the hypothesis of the external Friz-Victoir theorem, which directly yields existence of the canonical geometric rough path lift. The subsequent explicit L²-limit construction, basic properties, integration regimes, RDE well-posedness, and signature results are all logical consequences of this existence result. No step reduces by definition to its own output, no fitted parameter is relabeled as a prediction, and no load-bearing premise rests on a self-citation chain. The argument is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper invokes one standard external theorem and performs a direct analytic estimate on the given covariance; no free parameters are fitted and no new entities are postulated.

axioms (1)

standard math Friz-Victoir criterion: finite 2D ρ-variation of the covariance implies existence of a geometric rough-path lift
Invoked after the authors establish the variation bound for the tfBm covariance.

pith-pipeline@v0.9.0 · 5667 in / 1437 out tokens · 53183 ms · 2026-05-17T01:43:01.575668+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Our primary contribution is a detailed analysis of this covariance, proving it has finite 2D ρ-variation for ρ = 1/(2H). This verifies the criterion of Friz and Victoir, guaranteeing the existence of a rough path lift.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

14 extracted references · 14 canonical work pages

[1]

Meerschaert, M. M. and Sabzikar, F. (2013). Tempered fractional Brownian motion. Statistics & Probability Letters, 83(10):2269–2275

work page 2013
[2]

Meerschaert, M. M. and Sabzikar, F. (2014). Stochastic integration for tempered fractional Brownian motion.Stochastic Processes and their Applications, 124(7):2363–2387

work page 2014
[3]

Young, L. C. (1936). An inequality of the Hölder type, connected with Stieltjes integration. Acta Mathematica, 67(1):251–282

work page 1936
[4]

Lyons, T. J. (1998). Differential equations driven by rough signals.Revista Matemática Iberoamericana, 14(2):215–310

work page 1998
[5]

Friz, P. K. and Victoir, N. B. (2010).Multidimensional Stochastic Processes as Rough Paths: Theory and Applications. Cambridge University Press. 19

work page 2010
[6]

and Qian, Z

Coutin, L. and Qian, Z. (2002). Stochastic analysis, rough path analysis and fractional Brownian motions.Probability Theory and Related Fields, 122(1):108–140

work page 2002
[7]

Azmoodeh, E., Mishura, Y., and Sabzikar, F. (2020). How does tempering affect the local and global properties of fractional Brownian motion?Journal of Theoretical Probability, 33(2):789–811

work page 2020
[8]

Friz, P. K. and Hairer, M. (2014).A Course on Rough Paths: With an Introduction to Regularity Structures. Springer

work page 2014
[9]

and Oberhauser, H

Chevyrev, I. and Oberhauser, H. (2018). Signature moments to characterize laws of stochastic processes.Journal of Machine Learning Research, 19(1):3655–3691

work page 2018
[10]

J., Caruana, M., and Lévy, T

Lyons, T. J., Caruana, M., and Lévy, T. (2014).Differential Equations Driven by Rough Paths. Springer

work page 2014
[11]

Salvi, C., Lemercier, M., Liu, C., et al. (2021). The signature kernel is the solution of a Goursat PDE.SIAM Journal on Mathematics of Data Science, 3(3):873–899

work page 2021
[12]

Dietrich, C. R. and Newsam, G. N. (1997). Fast and exact simulation of stationary Gaussian processes through circulant embedding of the covariance matrix.SIAM Journal on Scientific Computing, 18(4):1088–1107

work page 1997
[13]

(2014).A theory of regularity structures

Hairer, M. (2014).A theory of regularity structures. Inventiones Mathematicae, 198(2):269– 504

work page 2014
[14]

Gatheral, J., Jaisson, T., and Rosenbaum, M. (2018). Volatility is rough. Quantitative Finance, 18(6):933–949. 20

work page 2018

[1] [1]

Meerschaert, M. M. and Sabzikar, F. (2013). Tempered fractional Brownian motion. Statistics & Probability Letters, 83(10):2269–2275

work page 2013

[2] [2]

Meerschaert, M. M. and Sabzikar, F. (2014). Stochastic integration for tempered fractional Brownian motion.Stochastic Processes and their Applications, 124(7):2363–2387

work page 2014

[3] [3]

Young, L. C. (1936). An inequality of the Hölder type, connected with Stieltjes integration. Acta Mathematica, 67(1):251–282

work page 1936

[4] [4]

Lyons, T. J. (1998). Differential equations driven by rough signals.Revista Matemática Iberoamericana, 14(2):215–310

work page 1998

[5] [5]

Friz, P. K. and Victoir, N. B. (2010).Multidimensional Stochastic Processes as Rough Paths: Theory and Applications. Cambridge University Press. 19

work page 2010

[6] [6]

and Qian, Z

Coutin, L. and Qian, Z. (2002). Stochastic analysis, rough path analysis and fractional Brownian motions.Probability Theory and Related Fields, 122(1):108–140

work page 2002

[7] [7]

Azmoodeh, E., Mishura, Y., and Sabzikar, F. (2020). How does tempering affect the local and global properties of fractional Brownian motion?Journal of Theoretical Probability, 33(2):789–811

work page 2020

[8] [8]

Friz, P. K. and Hairer, M. (2014).A Course on Rough Paths: With an Introduction to Regularity Structures. Springer

work page 2014

[9] [9]

and Oberhauser, H

Chevyrev, I. and Oberhauser, H. (2018). Signature moments to characterize laws of stochastic processes.Journal of Machine Learning Research, 19(1):3655–3691

work page 2018

[10] [10]

J., Caruana, M., and Lévy, T

Lyons, T. J., Caruana, M., and Lévy, T. (2014).Differential Equations Driven by Rough Paths. Springer

work page 2014

[11] [11]

Salvi, C., Lemercier, M., Liu, C., et al. (2021). The signature kernel is the solution of a Goursat PDE.SIAM Journal on Mathematics of Data Science, 3(3):873–899

work page 2021

[12] [12]

Dietrich, C. R. and Newsam, G. N. (1997). Fast and exact simulation of stationary Gaussian processes through circulant embedding of the covariance matrix.SIAM Journal on Scientific Computing, 18(4):1088–1107

work page 1997

[13] [13]

(2014).A theory of regularity structures

Hairer, M. (2014).A theory of regularity structures. Inventiones Mathematicae, 198(2):269– 504

work page 2014

[14] [14]

Gatheral, J., Jaisson, T., and Rosenbaum, M. (2018). Volatility is rough. Quantitative Finance, 18(6):933–949. 20

work page 2018