Fractional Brownian Motion, Non-Semimartingales, and the Lift

Job: develop the fBM (Hurst \(H\), covariance, \(p\)-variation), the canonical rough path lift (\(H>1/4\)), the non-semimartingale, and the RDE driven by fBM — with the fBM \(H>1/4\) canonical lift / RDE as the worked example.

Core

The fractional Brownian motion (fBM)

fBM / \(H\) / \(p\)-variation / rough Description
fBM \(B^H\) the Hurst \(H\), the covariance $r(t,s)=\frac12(
\(p\)-variation \(p>1/H\) (the \(H\) the Hurst, the \(p\) the roughness)
Rough path \(H>1/4\) the canonical rough path (the \(H>1/3\) the level-2)

The lift and the canonical

Lift / canonical / \(H>1/4\) / level-2 Description
Canonical lift \(H>1/4\) the canonical rough path lift (the \(\mathbb{B}^H\) the iterated)
Level-2 \(H>1/3\) the level-2 suffices (the \(H\in(1/4,1/3]\) the level-3)
Itô / Stratonovich the BM \(H=1/2\), the Itô / Stratonovich the \(\mathbb{B}\) (the stochastic processes)

The non-semimartingale and the rough path

Non-semimartingale / fBM / rough Description
Non-semimartingale \(H\ne1/2\) the fBM not the semimartingale (the stochastic processes)
Rough path the fBM the rough path (the \(H>1/4\) the canonical, the non-semimartingale)

The RDE driven by fBM

RDE fBM / \(H>1/4\) / Gaussian / SPDE Description
RDE \(dY=f(Y)\,dB^H\) the \(H>1/4\) the well-posed (the SPDE)
Gaussian the Gaussian the rough (the \(H\) the Hurst, the Gaussian the driver)

See it

fBM \(B^H\) for \(H>1/4\) the canonical rough path and the RDE. The fBM \(B^H\) the Hurst \(H\), the covariance, the \(p\)-variation \(p>1/H\). The canonical lift: the \(H>1/4\) the canonical rough path (the \(\mathbb{B}^H\) the iterated, the \(H>1/3\) the level-2). The RDE: the \(dY=f(Y)\,dB^H\) the \(H>1/4\) the well-posed (the SPDE). The key point: the fBM the canonical rough path (\(H>1/4\)), the RDE the well-posed — the non-semimartingale the rough path.

Check

Show that the fBM \(B^H\) for \(H>1/4\) admits a canonical rough path, and explain the non-semimartingale.

The fBM \(B^H\): the Hurst \(H\), the covariance \(r(t,s)=\frac12(|t|^{2H}+|s|^{2H}-|t-s|^{2H})\), the \(p\)-variation \(p>1/H\). The canonical lift: the \(H>1/4\) the \(\mathbb{B}^H\) the iterated integral (the Gaussian, the \(H>1/4\) the convergence, the \(H>1/3\) the level-2 the \(H\in(1/4,1/3]\) the level-3). The non-semimartingale: the \(H\ne1/2\) the fBM not the semimartingale (the stochastic processes, the \(H<1/2\) the not-martingale, the \(H>1/2\) the not-semimartingale). So the fBM the canonical (\(H>1/4\)), the non-semimartingale.

Explain the RDE driven by fBM and the well-posedness.

The RDE: \(dY=f(Y)\,dB^H\) (the fBM the driver, the \(f\in C^{[p]+1}\)). The well-posedness: the \(H>1/4\) the canonical rough path \(\mathbf{B}^H\) (the \(p>1/H<4\) the rough path), the Itô–Lyons the \(\mathbf{B}^H\mapsto Y\) the continuous (the controlled-and-sewing), the existence / uniqueness. The \(H\le1/4\) the open (the canonical the not-converge, the SPDE the renormalization). So the fBM the RDE (\(H>1/4\)), the well-posed.

Pitfalls

  • The fBM the \(H\) (the Hurst, the covariance, the \(p>1/H\)) — the \(H\) the roughness, the \(p\) the variation.
  • The canonical \(H>1/4\) (the \(\mathbb{B}^H\) the iterated, the \(H>1/3\) the level-2) — the \(H>1/4\) the canonical, the \(H\le1/4\) the open.
  • The non-semimartingale (\(H\ne1/2\), the stochastic processes) — the fBM the not-semimartingale, the rough path the pathwise.
  • The RDE the \(H>1/4\) (the well-posed, the Itô–Lyons) — the \(H>1/4\) the well-posed, the \(H\le1/4\) the open / the SPDE.

Refs

  • Mandelbrot–Van Ness, "Fractional Brownian motions, fractional noises" (SIAM 1968); the fBM
  • Coutin–Qian, "Stochastic differential equations driven by fBM" (2002); the \(H>1/4\) canonical
  • Friz–Victoir, Multidimensional Stochastic Processes as Rough Paths (2010); the fBM the lift

Next

RP index · Foundations · Signature & Features · Controlled & Sewing · Branched & Hopf · Stochastic processes · SPDE