--- title: Memory Multi-Fractional Brownian Motion url: https://www.emergentmind.com/topics/memory-multi-fractional-brownian-motion-mmfbm-9f21d431-d5a7-4481-a478-c84f98da59e1 type: topic --- # Memory Multi-Fractional Brownian Motion Memory multi-fractional Brownian motion (mmfBm) denotes a family of Gaussian, generally non-stationary extensions of fractional Brownian motion in which a time-dependent Hurst or memory exponent is combined with an explicit mechanism that preserves dependence on past increments. In the current literature the designation is not fully standardized: it is used for continuous-memory Volterra processes with exponent $\alpha(t)$, for kernel-based models of the form $\mathcal M(t)=\int_0^t K(t,s)\,dB^{H(s)}(s)$, for switching constructions with piecewise-constant $H(t)$ and diffusivity $D(t)$, and, under the same acronym, for “multi-mixed fractional Brownian motion,” a superposition of independent fBms with different Hurst indices [2303.01551], [2507.20097], [2307.12919], [2103.02978]. What unifies these constructions is the attempt to go beyond constant-Hurst fBm and beyond multifractional schemes in which changing the local exponent effectively resets the memory structure. ## 1. Terminological scope and canonical constructions The literature uses closely related but non-identical definitions of mmfBm. The common theme is that the roughness or memory index varies with time while long-memory effects are retained rather than reinitialized. | Formulation in the literature | Representative definition | Distinguishing feature | |---|---|---| | Continuous-memory MMFBM | $X(t)=\int_0^t \sqrt{\alpha(s)}\,(t-s)^{(\alpha(s)-1)/2}\,dB(s)$ | Entire history of $\alpha(s)$ enters the kernel | | Volterra-kernel mmfBm | $\mathcal M(t)=\int_0^t K(t,s)\,dB^{H(s)}(s)$ | Separate causal memory kernel and local Hurst field | | Switching-FBM / mmfBm | $X(t)=\int_0^t \sqrt{2D(s)H(s)}\,(t-s)^{H(s)-1/2}\,dB(s)$ | Piecewise-constant or stochastic regime changes | | Multi-mixed fBm | $M_t=\sum_{k=1}^\infty \sigma_k B_t^{H_k}$ | Superposition over a spectrum of Hurst indices | In the continuous-memory formulation of Pagnini, Sposini, and Metzler, the exponent is written as $\alpha(t)\in(0,2]$ and the process is defined by a Riemann–Liouville-type Volterra–Itô integral; the defining feature is that the kernel uses the past values $\alpha(s)$ rather than the current value $\alpha(t)$ alone [2303.01551]. In the Volterra-kernel formulation used for superconducting-qubit charge noise, the process is written as \[ \mathcal M(t)=\int_0^t K(t,s)\,dB^{H(s)}(s), \] where $H:[0,T]\to(0,1)$ is smooth and $K(t,s)$ is a causal kernel of Volterra type [2507.20097]. In the switching-FBM model, both the local Hurst exponent $H(t)$ and the generalized diffusivity $D(t)$ jump between states, but the kernel still reaches over the full past interval $[0,t]$, so the process does not forget its pre-switch history [2307.12919]. By contrast, the “multi-mixed fractional Brownian motion” of Almani and Sottinen is not a time-varying-exponent Volterra process at all; it is \[ M_t=\sum_{k=1}^\infty \sigma_k B_t^{H_k}, \qquad \sum_{k=1}^\infty \sigma_k^2<\infty, \] with distinct Hurst indices $\{H_k\}_{k\ge1}\subset(0,1)$ [2103.02978]. A related neighboring object is the multifractional bifractional Brownian motion $B^{H(\cdot),K}$, introduced by replacing the constant Hurst parameter in bifractional Brownian motion by a Hölder function $H(\cdot)$ while keeping $K\in(0,1]$ fixed. It is not usually labeled mmfBm, but it belongs to the same broader program of time-varying Gaussian self-similar models [2004.03999]. ## 2. Volterra kernels, covariance structure, and memory encoding Most memory-preserving multifractional models are naturally expressed as Gaussian Volterra processes. In this representation the entire dependence structure is encoded in the kernel. For the continuous-memory MMFBM, \[ X(t)=\int_0^t G(t,s)\,dB(s), \qquad G(t,s)=\sqrt{\alpha(s)}\,(t-s)^{(\alpha(s)-1)/2}, \] so the response function is simply \[ R(t,s)\equiv \frac{\delta X(t)}{\delta B(s)}=G(t,s),\qquad 0\le s\le t. \] The use of $\alpha(s)$ rather than $\alpha(t)$ is the mathematical origin of the “continuous correlations” emphasized in that work [2303.01551]. In the qubit-oriented Volterra model, \[ \mathcal M(t)=\int_0^t K(t,s)\,dB^{H(s)}(s), \] the covariance is given formally by \[ \mathbb E\big[\mathcal M(t)\mathcal M(s)\big] = \int_0^{t\wedge s} K(t,u)K(s,u) \Big[ 2H(u)u^{2H(u)-1}+u^{2H(u)}H'(u)\ln u \Big]\,du. \] When the term $H'(u)\ln u$ is small, the integrand reduces to $2H(u)u^{2H(u)-1}$, making the covariance resemble a locally weighted fBm covariance with an additional memory kernel [2507.20097]. A second qubit formulation uses the Riemann–Liouville representation \[ x(t)=M(t)=\frac{1}{\Gamma\!\bigl(H(t)+\tfrac12\bigr)} \int_0^t (t-s)^{H(t)-1/2}\,dW(s), \] with adaptive memory kernel \[ K(t,s)= \begin{cases} (t-s)^{H(t)-1/2}, & 0\le s0$ and $p\to0^+$, \[ \left\{ \frac{B^{H(\cdot),K}_{t+pu_i}-B^{H(\cdot),K}_t}{p^{H(t)K}} \right\}_{i=1}^n \Rightarrow \{B^{H(t)K}_{u_i}\}_{i=1}^n, \] in finite-dimensional distributions, where the limit is a standard fBm of Hurst index $H(t)K$. At large lags the same model exhibits long-range dependence: the correlation obeys a power law with exponent $\delta(s)\in(0,1)$, hence $\sum_t \mathrm{Corr}(B_t,B_s)=+\infty$ [2004.03999]. For mBm in the White Noise framework, increments near a fixed time $t_0$ behave like those of an fBm with Hurst parameter $h(t_0)$, so the local Hölder exponent is $H(t_0)$ and the local memory index is $2H(t)-1$ [1305.0342]. In the qubit mmfBm model this same exponent appears in the non-stationary noise spectrum: \[ S_M(\omega;t)\propto |\omega|^{-(2H(t)-1)} \equiv |\omega|^{-\beta(t)}, \qquad \beta(t)=2H(t)-1, \] which gives a slowly drifting $1/f^{\beta(t)}$ law [2605.18914]. The continuous-memory MMFBM of [2303.01551] exhibits a stronger history effect than memory-reset multifractional models. Its mean-squared displacement is \[ \langle X^2(t)\rangle = \int_0^t \alpha(s)(t-s)^{\alpha(s)-1}\,ds. \] For a step protocol \[ \alpha(t)= \begin{cases} \alpha_1, & t\le \tau,\\ \alpha_2, & t>\tau, \end{cases} \] one obtains \[ \langle X^2(t)\rangle = \begin{cases} t^{\alpha_1}, & t\le\tau,\\ t^{\alpha_1}-(t-\tau)^{\alpha_1}+(t-\tau)^{\alpha_2}, & t>\tau, \end{cases} \] and for $t\gg\tau$, \[ \langle X^2(t)\rangle\sim (\alpha_1\tau)t^{\alpha_1-1}+t^{\alpha_2}. \] Thus, if $\alpha_1>\alpha_2+1$, the long-time scaling can be governed by the pre-step exponent rather than by the post-step exponent. This is a defining memory effect, not a perturbative correction [2303.01551]. ## 4. Switching and superposed formulations A major branch of mmfBm research studies regime changes rather than smoothly varying exponents. In the switching-FBM model, $H(t)\in(0,1]$ and $D(t)>0$ jump between finitely many states and the process is defined by \[ X(t)=\int_0^t \sqrt{2D(s)H(s)}\,(t-s)^{H(s)-1/2}\,dB(s). \] Two switching laws are analyzed. In the Markovian two-state case, the active state $\sigma(t)\in\{1,2\}$ has exponential dwell times with means $\tau_1,\tau_2$. In the scale-free intermittent case, one state has heavy-tailed waiting-time density \[ \psi_1(t)=\frac{\beta t_0^\beta}{t^{1+\beta}},\qquad t>t_0,\quad 0<\beta<1, \] while the other remains exponential; this produces ageing and ergodicity breaking [2307.12919]. For Markovian switching, the long-time occupation fractions are \[ A_i=\frac{\tau_i}{\tau_1+\tau_2},\qquad i=1,2, \] and the ensemble-averaged time-averaged MSD is \[ \langle \overline{\delta^2(\Delta)}\rangle = A_1\,2D_1\Delta^{2H_1} + A_2\,2D_2\Delta^{2H_2}. \] The single-trajectory PSD satisfies the analogous weighted-sum rule \[ \langle S(\omega)\rangle = A_1\langle S_1(\omega)\rangle + A_2\langle S_2(\omega)\rangle. \] At short lag $\Delta t\ll\tau$, the increment distribution is approximately a bimodal Gaussian mixture, \[ P(\Delta x;\Delta t)\approx A_1\,\mathcal N(0,\sigma_1^2) + A_2\,\mathcal N(0,\sigma_2^2), \qquad \sigma_i^2\propto \Delta t^{2H_i}, \] whereas for $\Delta t\gg\tau$ it converges to a single Gaussian by the Central Limit Theorem [2307.12919]. The multi-mixed fractional Brownian motion of [2103.02978] is structurally different. Here \[ M_t=\sum_{k=1}^\infty \sigma_k B_t^{H_k}, \] and \[ \mathrm{Cov}[M_t,M_s] = \frac12\sum_{k=1}^\infty \sigma_k^2 \bigl(t^{2H_k}+s^{2H_k}-|t-s|^{2H_k}\bigr). \] Its increment autocovariance behaves as \[ \varrho(\delta;t) \sim \delta^2\sum_{k=1}^\infty \sigma_k^2 H_k(2H_k-1)t^{2H_k-2} = O\bigl(t^{2H_{\sup}-2}\bigr). \] Hence the increments are long-range dependent if $\max_k H_k>1/2$ and short-range dependent if $\max_k H_k<1/2$. The almost-sure Hölder exponent is $H_{\inf}$, and the $p$-variation index is $1/H_{\inf}$ [2103.02978]. ## 5. Stochastic integration and driven dynamics Because these processes are typically not semimartingales, stochastic integration requires dedicated constructions. A key route is the White Noise approach for mBm. If $h$ is $\beta$-Hölder continuous and $h_n$ is the associated piecewise-constant approximation on a partition of $[0,1]$, then the patchwork process $B^{h_n}$ converges in law to $B^h$ in $C([0,1])$. This makes it possible to define stochastic integration with respect to mBm as a limit of stochastic integrals with respect to tangent fBms [1305.0342]. In Hida–White Noise calculus, if $Y_t\odot W_t^{H}$ is Bochner-integrable, the Wick–Itô integral with respect to fBm is \[ \int_0^1 Y_t\,d^\diamond B_t^H = \int_0^1 Y_t\odot W_t^H\,dt. \] For mBm, the Hida derivative is \[ W_t^h=\frac{dB_t^h}{dt} = W_t^{h(t)}+h'(t)\,\partial_H B(t,h(t)), \] and the limiting fractional Wick–Itô integral becomes \[ \int_0^1 Y_t\,d^\diamond B_t^h = \int_0^1 Y_t\odot W_t^{h(t)}\,dt + \int_0^1 h'(t)\,Y_t\odot \partial_H B(t,h(t))\,dt. \] The additional term is the explicit contribution of the time variation of the Hurst function [1305.0342]. In the qubit mmfBm model of [2605.18914], the noise enters a Young-integral SDE, \[ \chi(t)=\chi(0)+\int_0^t F(s,\chi(s))\,ds +\sigma\int_0^t G(s,\chi(s))\,dM(s), \] and when $G$ is effectively constant, \[ \langle \chi(t)^2\rangle \approx \frac{\sigma^2}{2H(t)\Gamma^2\!\bigl(H(t)+\tfrac12\bigr)}\,t^{2H(t)}. \] A related qubit charge-offset model embeds the Volterra mmfBm directly into \[ d\chi(t)=-\lambda\,\chi(t)\,dt+\sigma_\phi\,\mathcal M(t), \] with $\lambda$, $\sigma_\phi$, $\beta$, and the target Hurst function calibrated from charge-noise data [2507.20097]. ## 6. Physical applications The most developed application domain in the supplied literature is superconducting-qubit decoherence. In the unified stochastic drift model of [2605.18914], the classical sector uses a time-dependent Hurst exponent $H(t)\in(1/2,1)$ and adaptive memory kernel $K(t,s)$ to represent non-stationary $1/f^{\beta(t)}$ noise, while the quantum extension is a time-dependent Caldeira–Leggett bath with spectral density \[ J(\omega;t)=\eta(t)\,\omega_c^{1-s(t)}\,\omega^{s(t)}e^{-\omega/\omega_c}, \qquad s(t)=2H(t)-1. \] For a pure-dephasing qubit with \[ H_S=\frac{\hbar\omega_0}{2}\sigma_z, \qquad H_{SB}=\sigma_z\otimes B, \] the reduced dynamics take the exact time-local Lindblad form \[ \frac{d\rho}{dt} = -\frac{i\omega_0}{2}[\sigma_z,\rho] -\Gamma_\phi(t)[\sigma_z,[\sigma_z,\rho]], \] with \[ \chi(t)\propto t^{2H(t)}, \qquad \Gamma_\phi(t)\propto t^{2H(t)-1} \] in the low-frequency limit. The corresponding coherence envelopes are stretched exponentials, \[ C_{\rm Ramsey}(t)\approx e^{-A_R t^{2H(t)}}, \qquad C_{\rm Echo}(t)\approx e^{-A_E t^{2H(t)+2}}. \] The paper reports four central results: relaxation and noise amplitudes act independently on energy decay; time-varying $H(t)$ matches experimental $1/f$ spectra more accurately than any constant exponent; adaptive kernel dynamics preserve correlations without artificial damping; and simulations predict coherence times $T_1\approx5.0\times10^6\,\mathrm{ns}$ and $T_2\approx4.18\times10^5\,\mathrm{ns}$ when charge noise dominates. It also identifies a thermal crossover time \[ t_{\rm cross}\sim \frac{\hbar}{k_B T}, \] with temperature-independent dephasing below $\sim50$ mK and $\Gamma_\phi\propto T$ above $\sim100$ mK [2605.18914]. The charge-noise model of [2507.20097] calibrates the mmfBm parameters to qubit data by fitting the relaxation rate to known $T_1$ times $(\sim50\,\mu\mathrm s^{-1})$, setting the noise amplitude by the low-frequency charge-noise level $(\sim3\,\mu\mathrm V/\sqrt{\mathrm{Hz}})$, using a spectral exponent such as $\beta\approx1.4$, and converting it to a target Hurst function \[ H(t)\approx 1-\frac{\beta}{2}+0.1\sin\!\Bigl(\frac{2\pi t}{T}\Bigr). \] Simulation uses the causal kernel \[ K(t_k,t_j) = \frac{(t_k-t_j)^{H(t_k)-1/2}}{\Gamma\!\bigl(H(t_k)+\tfrac12\bigr)} \exp\!\Bigl[-\frac{t_k-t_j}{\tau_c}\Bigr], \qquad \tau_c\approx1\,\mu\mathrm s, \] with Euler–Maruyama integration and an ensemble of $100$ realizations. Coupled to a transmon Hamiltonian, the model yields fidelity $\mathcal F(t)\sim \exp[-(t/T_2^*)^2]$ with $T_2^*\approx3\,\mu\mathrm s$, coherence envelopes with non-Markovian revivals, and excited-state population below $1\%$ [2507.20097]. Outside quantum hardware, the switching-FBM framework has been validated against single-particle tracking of quantum dots in the cytoplasm of live mammalian cells, where intermittent changes in transport parameters coexist with persistent correlations along trajectories [2307.12919]. ## 7. Conceptual distinctions, misconceptions, and related multifractional models A recurrent misconception is that any time-dependent Hurst exponent automatically produces a memory-preserving process. The literature explicitly distinguishes between memory-reset multifractional Brownian motion and continuous-memory MMFBM. In the memory-reset model, \[ Y(t)=\sqrt{\alpha(t)}\int_0^t (t-s)^{(\alpha(t)-1)/2}\,dB(s), \] the current value $\alpha(t)$ reweights the entire past kernel. As a result, $\langle Y^2(t)\rangle=t^{\alpha(t)}$, a step in $\alpha$ produces an abrupt change in the MSD slope, the linear response to a short perturbation is identically zero for $T>0$, and the autocovariance immediately switches to the new scaling. In MMFBM, by contrast, the kernel keeps the whole history of $\alpha(s)$, so pre-step exponents remain visible in the MSD, autocovariance, and response function [2303.01551]. A second source of confusion is terminological. The acronym “mmfBm” is used both for memory multi-fractional Brownian motion and for multi-mixed fractional Brownian motion. The former varies the exponent in time and encodes memory through a Volterra kernel or related construction; the latter mixes independent fBms with fixed but different Hurst indices through a superposition measure $\mu=\sum_k \sigma_k^2\delta_{H_k}$ [2103.02978]. A third distinction concerns nearby but non-equivalent generalizations. Multifractional bifractional Brownian motion $B^{H(\cdot),K}$ preserves the bifractional parameter $K$ and reduces to mBm when $K=1$; its local Hölder exponent is $H(t)K$, not $H(t)$, and its long-range dependence follows from the asymptotic power-law behavior of its correlation function [2004.03999]. Related multifractional, rather than explicitly memory-kernel, models also appear in finance: a multifractional Black–Scholes calibration on SPX ATM calls reported mean squared errors $456.8$ for the multifractional model, $493.7$ for the fractional model with fixed $H$, and $555.5$ for classical Black–Scholes, illustrating the empirical value of time-varying roughness even outside the mmfBm terminology [2303.16314]. Taken together, the arXiv literature presents mmfBm not as a single canonical process but as a research program: Gaussian models with time-varying roughness, explicit history dependence, and analytically tractable covariance structure. The central technical question across these variants is how to allow $H(t)$ or $\alpha(t)$ to change without destroying long-memory effects already accumulated in the past. Source: https://www.emergentmind.com/topics/memory-multi-fractional-brownian-motion-mmfbm-9f21d431-d5a7-4481-a478-c84f98da59e1