Fractional Itô Motion (FIM) Overview

• Fractional Itô Motion (FIM) is a self-similar, non-Gaussian process exhibiting anomalous diffusion with Markov and martingale properties.
• It employs a unique volatility structure |x|^(1-1/(2H)) that ensures selfsimilarity and analytical tractability with explicit closed-form densities.
• FIM accurately models sub-, normal, and super-diffusive regimes, facilitating simulation, inference, and application in fields like biology and climate.

Fractional Itô Motion (FIM) is a family of stochastic processes constructed to model anomalous diffusion phenomena, generalizing the scaling behavior of fractional Brownian motion (FBM) while restoring key probabilistic properties—namely, the Markov and martingale structures—that FBM lacks. In contrast to FBM, FIM is a non-Gaussian, Markovian, and martingale process with uncorrelated, nonstationary increments. This makes FIM analytically tractable and straightforward to simulate, while preserving selfsimilar scaling with mean-square displacement (MSD) exponent $2H$, $0Eliazar et al., 2021).

1. Definition and Construction

A one-dimensional fractional Itô motion $I_H(t)$ is defined as the unique solution to the stochastic differential equation

$$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$

with standard Brownian motion $B(t)$ and volatility

$\sigma(x) = |x|^{1-\frac{1}{2H}}, \qquad 0<H<1,$

interpreted in the Itô sense. This process is a zero-drift Itô diffusion with state-dependent multiplicative noise. The construction is dictated by the requirement that $I_H$ be selfsimilar of order $H$, i.e., for all $s>0$,

$$\{I_H(st)\}_{t\ge0} \overset{\rm law}{=} \{s^H I_H(t)\}_{t\ge0}.$$

The volatility structure $0Eliazar et al., 2021).

2. Scaling Laws and Selfsimilarity

FIM satisfies the scaling law

$0

which determines the anomalous diffusion regime:

• Subdiffusive: $0
• Normal diffusion: $0
• Superdiffusive: $0

Moments scale as

$0

so the mean-square displacement is proportional to $I_H(t)$0, mirroring FBM but with fundamentally different increment structure (Eliazar et al., 2021).

3. Probabilistic Properties

Markov and Martingale Structure

FIM is a (strong) Markov process as a consequence of being an Itô diffusion. Furthermore, with zero drift, $I_H(t)$1 is a martingale: $I_H(t)$2 for all $I_H(t)$3.

Non-Gaussianity

FIM is not Gaussian (unless $I_H(t)$4). The one-point marginal density is given by

$I_H(t)$5

which exhibits a power-law singularity at the origin for $I_H(t)$6, a zero at the origin for $I_H(t)$7, and exponential tails. Higher cumulants $I_H(t)$8 are nonzero for all $I_H(t)$9, and characteristic functions are not Gaussian.

Increment Properties

The velocity process $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$0 exists in the generalized sense, with $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$1 for $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$2, i.e., increments are uncorrelated. However, increment variances depend on the "age" $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$3: $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$4 Therefore, increments are nonstationary for $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$5 (Eliazar et al., 2021).

4. Analytical Tractability and Simulation

FIM is analytically tractable due to its Itô diffusion form. The one-point density is explicit, and the transition density admits representation via modified Bessel functions by a mapping to diffusion in a logarithmic potential. Euler–Maruyama discretization suffices for simulation: $B(t)$2 In contrast, FBM simulation requires non-local (Cholesky or circulant matrix) methods due to its correlated increments (Eliazar et al., 2021).

5. Connection to Diffusion in Logarithmic Potential

A monotone transformation $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$6 with $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$7 maps FIM to a process satisfying the Langevin SDE

$$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$8

representing diffusion in the logarithmic potential $$dI_H(t) = \sigma\bigl(I_H(t)\bigr)\,dB(t), \qquad I_H(0)=0,$$9. This mapping establishes a one-to-one correspondence between FIM and diffusion in a log-potential for $B(t)$0, enabling further analytic results (Eliazar et al., 2021).

6. Comparison with Fractional Brownian Motion

The contrasts between FIM and FBM are summarized in the following table (for $B(t)$1):

Property	FBM	FIM
Gaussianity	Gaussian	Non-Gaussian
Markov property	Non-Markov	Markov
Martingale	Not a martingale	Martingale
Increment nature	Correlated, stationary	Uncorrelated, nonstationary
Analytical tract.	Limited	Closed-form densities

Both models share selfsimilarity, continuity, and symmetric scaling of the MSD, but differ fundamentally in Markovianity, increment correlation, and tractability (Eliazar et al., 2021).

7. Applications and Modeling Implications

FIM’s non-Gaussian density accommodates observed non-Gaussian dissipation patterns in complex fluids, biological media, and climate data—phenomena such as central cusps or bimodal peaks in empirical distributions. Its Markov and martingale properties permit use of standard prediction, filtering, control, and inference tools. Closed-form densities enable likelihood-based estimation and rapid simulation, expanding usefulness in machine learning and data-driven stochastic modeling of anomalous diffusion (Eliazar et al., 2021).