{"id":"bf421f2e-8149-49fb-8512-cac6a4ce8720","arxiv_id":"2504.19190","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"For two of the three fractional Brownian motion definitions, a fluctuating diffusivity only rescales standard results by the mean diffusivity, while the Langevin-equation version predicts a crossover from anomalous to normal diffusion.","lead":"This paper compares three ways of generalizing fractional Brownian motion to include a randomly fluctuating diffusion coefficient, finding that the choice of representation changes the predicted motion. The results give experimentalists a guide for which model to use when analyzing non-Gaussian anomalous diffusion in messy biological and soft-matter environments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Independence of the diffusivity noise from the Brownian driver is implicit and load-bearing; Eqs. 43–45 and 38 only hold under that hypothesis.","rationale":"The reader's weakest-assumption analysis identifies exactly the point on which the central claim rests. The MN-FBM-DD and RL-FBM-DD calculations are mathematically sound conditional on independence of the diffusivity process and the Brownian motion, and the LE-FBM-DD expression from prior work uses the same independence in factorizing the correlation function. The simulations in Figs. 1–4 provide independent support for the analytic expressions under the standard independent-noise construction, so I do not see an internal inconsistency or a numerical error. The remaining issue is that the manuscript never explicitly states independence of η and B, nor the full negative-time equilibrium requirement for the MN representation. Because the claimed MN-vs-LE distinction would fail under correlated noises, this omission is load-bearing for the central claim. It is a presentation and hypothesis-specification gap rather than a demonstrated flaw, so the CONDITIONAL verdict stands unchanged.","tokens_in":19418,"tokens_out":29154,"duration_ms":306637,"concrete_test":"Run the MN-FBM-DD generator (Eq. 37) twice with identical parameters, once with η and B drawn from independent Gaussian streams and once with η(t)=ρḂ(t)+√(1−ρ²)ζ(t) at ρ=0.5; compare the MSD to Eq. 44. If the second MSD deviates from 2⟨D⟩t^{2H}, independence is a necessary hypothesis and should be stated explicitly; if it coincides, the concern would not be load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III.B defines D(t)=Y²(t) with OU noise η(t), but never states the joint law of η and the Brownian motion B that drives the FBM integrals. The headline MN-vs-LE distinction depends on this silence: Eq. 43 factors E[D(s)] out of Itô integrals; Eq. 38 factors ⟨√D(u)√D(v)⟩=K(u−v) out of the fGn double integral; Eq. 55 factorizes the ACVF as a product of K(Δ) and ⟨ξ_H²⟩_Δ. All three steps are valid only if D (equivalently η) is independent of B. If η and B are correlated, additional cross terms appear in the second moment and MN-FBM-DD need not equal 2⟨D⟩t^{2H}; the claimed difference between MN-FBM-DD and LE-FBM-DD would not be the robust model prediction stated. The same applies to the auxiliary equilibrium assumption for D(s) for all s≤0 in Eq. 28: it is needed for ⟨D(s)⟩=⟨D⟩ on the negative-time integral in Eq. 43. This is a standard model choice, not an error, but it is the load-bearing condition and should be stated as an explicit hypothesis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript generalizes three equivalent representations of fractional Brownian motion (Langevin equation driven by fractional Gaussian noise, LE-FBM; Mandelbrot–van Ness integral, MN-FBM; and Riemann–Liouville integral, RL-FBM) by replacing the constant diffusivity with a random diffusivity D(t) that evolves as the square of an Ornstein–Uhlenbeck process. The authors derive the mean-squared displacement (MSD), mean-squared increment (MSI), autocovariance function of increments (ACVF), and probability density function for the resulting FBM-DD models. The central claim is that although MN-FBM and LE-FBM are equivalent for constant diffusivity, their diffusing-diffusivity counterparts behave differently: MN-FBM-DD and RL-FBM-DD both give MSD = 2<D> t^{2H} with the increment-stationarity properties of the underlying FBM, whereas LE-FBM-DD exhibits an unexpected crossover to Brownian diffusion for H<1/2 and to a rescaled anomalous scaling for H>1/2, driven by correlations of the random diffusivity. Simulations are presented for the MSD, MSI, ACVF, and PDF and are reported to agree with the analytical predictions.","tokens_in":19673,"tokens_out":10221,"duration_ms":92618,"significance":"If the results hold, the paper is significant because it demonstrates that the three standard representations of fractional Brownian motion, equivalent for constant diffusivity, are not equivalent when coupled to a random diffusivity. It provides a systematic comparison, summarized in Table I, and makes falsifiable predictions, such as the subdiffusive crossover to normal diffusion in LE-FBM-DD, that could be used to discriminate between competing FBM-DD models in single-particle-tracking experiments on viscoelastic, heterogeneous media. The derivations for the MN- and RL-FBM-DD models are exact and transparent, relying only on the Itô isometry and the stationarity of the squared-OU diffusivity, and the simulation methodology is described in enough detail to be reproduced.","major_comments":[{"comment":"The model definitions do not specify the joint law of the OU noise η(t) and the Brownian motion or fractional Gaussian noise that drives the particle. The derivation of the LE-FBM-DD MSD in Eq. (38) and the ACVF in Eq. (55) requires that D(t), equivalently η(t), be independent of ξ_H(t); if this independence fails, cross-correlation terms appear and the factorizations in Eqs. (38) and (55) are invalid. Since the predicted crossover in Eq. (41) is the paper's headline result, the independence hypothesis must be stated explicitly as part of the model definition. Note that the MN- and RL-FBM-DD results do not require this independence, only the Itô isometry and stationarity of D(t), so the missing hypothesis is specific to the LE-FBM-DD claims.","section":"Section III.B, Eqs. (38) and (55)"}],"minor_comments":[{"comment":"The sentence 'MN-FBM and LE-FBM are equivalent in the sense that they exhibiting the same MSD...' contains a grammatical error; 'they exhibiting' should be 'they exhibit'.","section":"Section III.A.2, text near Eq. (9)"},{"comment":"The large-Δ expansion of K(Δ) from Eq. (25) gives σ²τ/π [1 + (1/2)e^{-2Δ/τ}] to leading order, so the exponential term in Eq. (56) should carry a prefactor 1/(2π); as written, the coefficient of the exponential term is too large by a factor of 2π.","section":"Eq. (56)"},{"comment":"A brief derivation of the LE-FBM-DD ACVF would be helpful, as this result is used in interpreting the MSD crossover and is not explicitly derived in the manuscript.","section":"Section V.C, Eq. (55)"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the journal's scope and builds on the authors' earlier work (ref. [72]). The novelty is the systematic comparison across the three FBM representations under diffusing diffusivity. The missing independence assumption is standard in the diffusing-diffusivity literature and is easily fixable by an explicit statement; with that addition the central claim should be sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nHere is my take on Wang et al. (arXiv:2504.19190). The paper asks a good question: when you add diffusing diffusivity (DD) to fractional Brownian motion, does it matter which representation of FBM you start from? The answer is yes, and the paper shows it carefully. The MN-FBM-DD and RL-FBM-DD results are new: their MSDs, MSIs, and ACVFs just pick up an effective diffusivity equal to the mean <D>. In contrast, LE-FBM-DD shows a crossover in the MSD/MSI for H<1/2, which was already reported in the authors' earlier work (ref 72). The systematic comparison and the summary table are the main contributions.\n\nThe derivations for MN- and RL-FBM-DD are straightforward and exact, relying only on Itô isometry and stationarity of the OU-squared diffusivity. The simulations agree with the analytic expressions. The paper is clearly written and honest about what is taken from prior work.\n\nThe main soft spot is a missing hypothesis. The derivations factor E[D(s)] out of Itô integrals and factor the correlation of sqrt(D) out of the double integral. This only holds if the diffusivity process D(t) is independent of the Brownian motion B(s) driving the FBM integrals. That independence is standard in the field, but it is load-bearing and never stated. The same applies to the equilibrium assumption for D(s) on negative times in the MN representation. This is not an error, but it should be an explicit model hypothesis. A referee should ask for that.\n\nMinor issues: Eq. (55) for the LE-FBM-DD ACVF appears without derivation, though the form is plausible. The crossover results for LE-FBM-DD are quoted from ref 72, so the genuinely new material is narrower than the title suggests. Also, the paper does not discuss what happens if D(t) and B(s) are correlated, which would change all three predictions, but that is beyond the intended model.\n\nWho is this for? Experimentalists analyzing single-particle tracking in viscoelastic, heterogeneous media, and theorists building model libraries for anomalous diffusion. It deserves a serious referee. I would accept it for peer review with a request to make the independence assumption explicit and to add a sentence on its physical meaning.\n\nBest.","headline":"Solid model comparison with clean derivations for two of the three FBM-DD variants; the load-bearing independence assumption should be stated explicitly.","tokens_in":20238,"tokens_out":2650,"would_cite":true,"duration_ms":25164,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G22","82C31"],"pacs":["05.40.Fb","02.50.Ey"],"model":"deepseek-v4-flash","headline":"Adding a fluctuating diffusivity to fractional Brownian motion yields three distinct sets of predictions: the Mandelbrot–van Ness and Riemann–Liouville forms keep an effective diffusivity equal to the mean $\\langle D\\rangle$, while the…","keywords":["fractional Brownian motion","diffusing diffusivity","anomalous diffusion","fractional Gaussian noise","mean-squared displacement","non-Gaussian diffusion","Ornstein-Uhlenbeck process","Brownian yet non-Gaussian"],"falsifier":"Simulate the LE-FBM-DD equation with diffusivity noise that is partially correlated with the fractional Gaussian noise increments $\\xi_H(t)$ and compare the MSD: the exact factorized form (38) would acquire an extra correlation term, and the clean crossover from $2\\langle D\\rangle t^{2H}$ to $2D_{\\mathrm{eff}} t$ would shift or vanish. A cleaner laboratory test is to measure the MSI at fixed lag as a function of absolute time in a viscoelastic fluid: RL-FBM-DD predicts an aging MSI, while LE-FBM-DD and MN-FBM-DD predict a stationary one.","tokens_in":19187,"feed_emoji":"🎲","tokens_out":9080,"duration_ms":79710,"temperature":0.7,"pith_summary":"This paper generalizes fractional Brownian motion (FBM) to heterogeneous environments by letting the diffusion coefficient itself become a random process, the square of an Ornstein–Uhlenbeck process, in each of FBM's three standard representations. It argues that the three resulting models are not equivalent: the Mandelbrot–van Ness and Riemann–Liouville versions behave as FBM with an effective diffusivity equal to the mean value $\\langle D\\rangle$, while the Langevin version feels the diffusivity's own correlations, producing a crossover in the mean-squared displacement (MSD) and mean-squared increment (MSI) from anomalous scaling $2\\langle D\\rangle t^{2H}$ to normal diffusion $2D_{\\mathrm{eff}}t$ when $H<1/2$. The Riemann–Liouville version additionally has nonstationary increments, unlike the other two. These distinctions matter for experimentalists fitting single-particle tracks in viscoelastic, heterogeneous media, where the same data could be read as different transport physics depending on which FBM representation underlies the model.","feed_headline":"Fluctuating diffusivity splits fractional Brownian motion models","feed_subtitle":"For subdiffusive Langevin FBM, the long-time MSD turns linear; the other two representations keep their anomalous scaling.","key_machinery":"The central object is the squared Ornstein–Uhlenbeck process $D(t)=Y^2(t)$, with mean $\\langle D\\rangle=\\sigma^2\\tau/2$ and square-root correlation $K(\\Delta)=\\langle \\sqrt{D(t)}\\sqrt{D(t+\\Delta)}\\rangle$ given by Eq. (25). When inserted into the Langevin representation, $K(\\Delta)$ is convolved with the fractional Gaussian noise autocovariance $\\langle \\xi_H^2\\rangle_\\Delta$; for $H<1/2$ the long-time integral converges to a linear-in-$t$ term, producing the crossover to normal diffusion. In the Mandelbrot–van Ness and Riemann–Liouville representations the diffusivity appears linearly inside the Itô integral, so under the independence assumption the average factors into $\\langle D\\rangle$ times the deterministic FBM kernel, and the standard $t^{2H}$ scaling survives unchanged. The large-lag expansion $K(\\Delta)\\sim\\sigma^2\\tau(\\pi^{-1}+e^{-2\\Delta/\\tau})$ reveals the competing persistent-noise and truncated-power-law contributions that shape the LE-FBM-DD ACVF and explain the crossover.","core_discovery":"On the paper's own terms, for $D(t)=Y^2(t)$ with $Y$ an Ornstein–Uhlenbeck process, the three generalized models give distinct statistical signatures. The Langevin form (LE-FBM-DD) has $\\langle x^2(t)\\rangle_{\\mathrm{LE}} = 4\\int_0^t (t-s)K(s)\\langle \\xi_H^2\\rangle_s\\,ds$, which at short times is $2\\langle D\\rangle t^{2H}$ and at long times $2K_{\\mathrm{eff}} t^{2H}$ for $H>1/2$ but $2D_{\\mathrm{eff}} t$ for $H<1/2$, with $K_{\\mathrm{eff}}=\\lim_{\\Delta\\to\\infty}K(\\Delta)$ and $D_{\\mathrm{eff}}=2\\int_0^\\infty K(s)\\langle \\xi_H^2\\rangle_s\\,ds$. The Mandelbrot–van Ness form (MN-FBM-DD) gives $\\langle x^2(t)\\rangle_{\\mathrm{MN}}=2\\langle D\\rangle t^{2H}$ exactly, the same as the Riemann–Liouville form (RL-FBM-DD), because the diffusivity factors out of the Itô integrals under the equilibrium and independence assumptions. The increments of MN-FBM-DD and LE-FBM-DD are stationary; those of RL-FBM-DD are not, with MSI $4H\\langle D\\rangle[I_H(t/\\Delta)+1/(2H)]\\Delta^{2H}$. The autocovariance function (ACVF) of LE-FBM-DD carries the diffusivity correlation $K(\\Delta)$ multiplied by the fractional-noise autocovariance, while the other two carry only $\\langle D\\rangle$ times their usual noise autocovariance. All three displacement PDFs cross over from a short-time non-Gaussian (Bessel-function) form to a long-time Gaussian form.","pith_inferences":["Because the factorization to $\\langle D\\rangle$ relies on $D(t)$ being independent of the driving noise, a testable extension is to couple $D(t)$ weakly to the particle position: the clean $t^{2H}$ scaling should break first in the MN- and RL-forms, while the LE-form's crossover would shift.","The same distinction should apply to multifractional Brownian motion with a time-dependent Hurst exponent: representations based on the Langevin form should show diffusivity-correlation crossovers in addition to Hurst-exponent aging effects.","Single-particle tracking pipelines that treat the three FBM representations as interchangeable could misclassify subdiffusive trajectories, because the fitted long-time exponent would be $1/2$ under LE-FBM-DD but $2H$ under MN-FBM-DD for the same underlying physics."],"forward_implications":["For $H<1/2$, the LE-FBM-DD model predicts that a tracer's long-time MSD becomes linear in time, so fitting only long-time trajectories would make the motion look Brownian even though the process is fractional at short times.","The MSI and ACVF are stationary for LE-FBM-DD and MN-FBM-DD but nonstationary for RL-FBM-DD, providing experimental signatures in increment statistics that do not require measuring the MSD.","All three DD-generalized models predict a crossover of the displacement PDF from a short-time non-Gaussian shape to a long-time Gaussian, with the crossover scale set by the diffusivity correlation time $\\tau$.","MN-FBM-DD and RL-FBM-DD both yield $\\langle x^2(t)\\rangle=2\\langle D\\rangle t^{2H}$ for all $H$ in $(0,1)$, meaning the long-time MSD alone cannot distinguish those two representations."],"supporting_citations":[{"why":"Supplies the Mandelbrot–van Ness and Riemann–Liouville integral representations of fractional Brownian motion that the paper generalizes.","marker":"[37]"},{"why":"Introduces the squared Ornstein–Uhlenbeck diffusing-diffusivity process and its equilibrium statistics $\\langle D\\rangle=\\sigma^2\\tau/2$ used throughout.","marker":"[61]"},{"why":"Prior LE-FBM-DD analysis giving the MSD integral (38) and the effective diffusion coefficient $D_{\\mathrm{eff}}$ whose crossover is the paper's central contrast.","marker":"[72]"},{"why":"Treats Riemann–Liouville FBM with fluctuating diffusivity and reports no MSD crossover, the comparison result the paper explains.","marker":"[48]"},{"why":"Provides the nonstationary MSI and ACVF formulas for Riemann–Liouville FBM that the RL-FBM-DD results inherit.","marker":"[77]"},{"why":"Shows how truncated power-law noise correlations produce a linear long-time MSD component, the mechanism behind the LE-FBM-DD crossover for $H<1/2$.","marker":"[84]"}],"fun_headline_variants":["Three fractional Brownian definitions diverge under fluctuating diffusivity","Diffusing diffusivity reveals hidden split in fractional Brownian motion","Langevin FBM shows crossover, Mandelbrot and Riemann forms don't","Nonstationary increments emerge in one fractional Brownian diffusivity model","Which fractional Brownian model survives diffusing diffusivity?"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivations assume the random diffusivity $D(t)$ is statistically independent of the Brownian motion $B(s)$ driving the fractional Gaussian noise (and, for the Mandelbrot–van Ness form, that the diffusivity has already reached equilibrium for all times $s\\le 0$), because the clean factorizations into $\\langle D\\rangle$ and $K(\\Delta)$ require the averages to separate; if the diffusivity were correlated with the noise, every predicted scaling and crossover would change.","fun_headline_variants_meta":{"raw":{"variants":["Three fractional Brownian definitions diverge under fluctuating diffusivity","Diffusing diffusivity reveals hidden split in fractional Brownian motion","Langevin FBM shows crossover, Mandelbrot and Riemann forms don't","Nonstationary increments emerge in one fractional Brownian diffusivity model","Which fractional Brownian model survives diffusing diffusivity?"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000393,"raw_usage":{"total_tokens":2219,"prompt_tokens":1257,"completion_tokens":962,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":873,"completion_tokens_details":{"reasoning_tokens":873}},"tokens_in":873,"tokens_out":962,"duration_ms":6825,"temperature":1.0,"reasoning_tokens":873,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T06:00:58.766089+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the LE-FBM-DD equation with diffusivity noise that is partially correlated with the fractional Gaussian noise increments $\\xi_H(t)$ and compare the MSD: the exact factorized form (38) would acquire an extra correlation term, and the clean crossover from $2\\langle D\\rangle t^{2H}$ to $2D_{\\mathrm{eff}} t$ would shift or vanish. A cleaner laboratory test is to measure the MSI at fixed lag as a function of absolute time in a viscoelastic fluid: RL-FBM-DD predicts an aging MSI, while LE-FBM-DD and MN-FBM-DD predict a stationary one.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the squared Ornstein–Uhlenbeck diffusing-diffusivity process and its equilibrium statistics $\\langle D\\rangle=\\sigma^2\\tau/2$ used throughout."},{"cited_title":"Sposini, D","cited_arxiv_id":null,"evidence_quote":"Prior LE-FBM-DD analysis giving the MSD integral (38) and the effective diffusion coefficient $D_{\\mathrm{eff}}$ whose crossover is the paper's central contrast."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Treats Riemann–Liouville FBM with fluctuating diffusivity and reports no MSD crossover, the comparison result the paper explains."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the nonstationary MSI and ACVF formulas for Riemann–Liouville FBM that the RL-FBM-DD results inherit."},{"cited_title":"Luo and M","cited_arxiv_id":null,"evidence_quote":"Shows how truncated power-law noise correlations produce a linear long-time MSD component, the mechanism behind the LE-FBM-DD crossover for $H<1/2$."}],"review_version":1}