{"id":"0e6aa566-eb6a-438c-a663-1ec90075ee2a","arxiv_id":"2506.09921","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Sign-thresholding or magnitude-rounding of fractional Brownian motion increments yields lattice random walks with the same power-law correlations as super-diffusive fBm.","lead":"A new recipe turns a super-diffusive fractional Brownian motion into a random walk on a lattice with integer time steps. It gives scientists a simple tool to simulate memory-driven diffusion in discrete settings.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Table I's fitted TLD/FOD exponents deviate from 2−2H by many times the quoted errors, contradicting the 'perfect agreement' that supports the central analogue claim.","rationale":"The reader's weakest assumption pointed to the unproven preservation of the power-law exponent based on finite-time fits. My stress-test confirms this is the load-bearing issue, but sharpens it: the paper's own Table I shows fitted exponents that are incompatible with the theoretical value when the quoted errors are taken at face value. This is not a matter of extrapolation outside the tested range; the tested range already fails the claimed test. The TLD scheme admits a closed-form covariance via the arcsin transform of the Gaussian correlation, so the claim 'exactly the same covariance function' is analytically false; the apparent power-law is only asymptotic. Because the central conclusion is stated in absolute terms ('perfect agreement', 'exactly the same'), the paper should be accepted only conditionally, after correcting the overstated claims and either deriving or explicitly acknowledging the finite-lag deviation. The concrete analytic test above would settle whether the numerical bias is inherent or a simulation artifact.","tokens_in":9889,"tokens_out":9169,"duration_ms":90883,"concrete_test":"Compute the exact TLD increment covariance C_TLD(dτ)=(2/π)arcsin(dτ^{-(2-2H)}) for H=0.6,0.7,0.8,0.9 and fit a power law C_TLD(dτ)∝dτ^{-ε} over the same lag range as Table I/Fig. 2 (e.g., dτ∈[1,10^3]). Compare the fitted ε with ε_TLD in Table I. If ε matches ε_TLD (≈0.83 for H=0.6, ≈0.63 for H=0.7, ≈0.225 for H=0.9), the deviation from 2−2H is real and the exact-covariance claim fails; if ε matches ε_th (0.8, 0.6, 0.2), the simulation bias is a numerical artifact. Repeating the fit on simulated long trajectories (T=2^22) with more realizations would distinguish finite-size effects from systematic bias.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (Sec. IV) that TLD and FOD are discrete counterparts of fBm rests on the fits in Table I purportedly showing that the increment covariance decays as (dτ)^{-(2-2H)}. But the quoted errors do not support this. For H=0.7, ε_th=0.6 while ε_TLD=0.628±0.004 and ε_FOD=0.626±0.004; for H=0.9, ε_th=0.2 while ε_TLD=0.225±0.002 and ε_FOD=0.211±0.001. Each difference is many times the reported standard error, so the numerical evidence actually contradicts 'perfect agreement'. Moreover, for TLD the covariance is exactly computable: the continuous-time increments (δ_t, δ_{t+dτ}) are bivariate Gaussian with correlation ρ=|dτ|^{-(2-2H)}, so the sign covariance is E[sign(δ_t)sign(δ_{t+dτ})]=(2/π)arcsin(ρ), which is not a constant multiple of a pure power law. Fitting this exact expression over a finite lag window yields an apparent exponent larger than 2-2H, matching the bias in Table I. The FOD rescaling (Eq. 6) is applied per realization and introduces an additional uncontrolled time reparametrization. Thus the exact-covariance goal stated in the introduction is not met; the schemes at best reproduce the asymptotic tail exponent, and even that is not established by the reported fits.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes two constructions of a discrete-space, integer-time random walk intended to serve as a lattice analogue of super-diffusive fractional Brownian motion (fBm) with Hurst exponent H>1/2. In the TLD scheme, increments are binarized by their sign; in the FOD scheme, increments are rounded in magnitude and the time axis is rescaled per trajectory via Eq. (6). The authors validate the constructions numerically using Davies-Harte generated fBm trajectories (2000 realizations, T=2^22, H=0.6-0.9): they report power-law fits for the increment covariance in Table I, show good collapse of the averaged absolute displacement in Fig. 1 and of the reduced covariance in Fig. 2, and report very high Pearson correlations between individual original and discretized trajectories. They conclude that TLD and FOD are discrete counterparts of super-diffusive fBm and argue that super-diffusion is universal while sub-diffusion is not.","tokens_in":10234,"tokens_out":4639,"duration_ms":55535,"significance":"If the central claim were established, the paper would provide simple, practical algorithms (O(T log T) using circulant embedding) for generating lattice walks with fBm-like increment correlations, and it would sharpen the distinction between super-diffusive persistence and sub-diffusive anti-persistence. The numerical effort is substantial and the algorithms are transparent, with TLD being essentially parameter-free. However, the central claim is currently overstated: the exact covariance equality asserted in the text is false for TLD, the numerical fits in Table I deviate from the theoretical exponent by many quoted error bars, and the FOD per-trajectory time rescaling undermines the integer-time interpretation. These issues are load-bearing because the paper's conclusion rests on the claim that the increment covariance 'obeys Eq. (4)' and that the fits show 'perfect agreement'. With a reformulation in terms of asymptotic tail equivalence and a more careful statistical analysis, the contribution could still be useful.","major_comments":[{"comment":"The fitted exponents do not agree with the theoretical exponent ε_th=2−2H within the quoted errors. For H=0.7, ε_TLD=0.628±0.004 versus ε_th=0.6, a 7σ deviation; for H=0.9, ε_TLD=0.225±0.002 versus 0.2 (12.5σ) and ε_FOD=0.211±0.001 (11σ). Even the parent fBm fit shows deviations (ε_X=0.838±0.008 versus 0.8 for H=0.6, and 0.206±0.001 versus 0.2 for H=0.9). Thus the statement that results agree 'in the limits of the statistical error' is not supported by the reported numbers. The authors should report the fitting range, the method used to estimate errors (e.g., bootstrap over realizations), and should either correct for the apparent finite-window bias or explicitly weaken the claim to asymptotic agreement.","section":"Sec. IV, Table I"},{"comment":"For the TLD scheme, the increment covariance is not the same power law as Eq. (4). Since the continuous-time increments (δ_t, δ_{t+dτ}) are jointly Gaussian with correlation ρ(dτ)=|dτ|^{-(2-2H)} (up to normalization), the sign covariance is exactly E[sign(δ_t)sign(δ_{t+dτ})]=(2/π)arcsin(ρ(dτ)). This equals a constant times (dτ)^{-(2-2H)} only in the asymptotic limit ρ→0; over any finite lag window a power-law fit will produce an effective exponent larger than 2−2H, which is consistent with the positive deviations seen in Table I. Therefore the claim that TLD increments have 'exactly the same power-law covariance function' as fBm is incorrect. The authors should either replace this claim with the explicit arcsin formula and an asymptotic statement, or restrict the claim to asymptotic tail equivalence.","section":"Sec. III, TLD rule and Eq. (4)"},{"comment":"The FOD time rescaling is applied per realization: the factor T/T_FOD depends on the random variable T_FOD, the number of nonzero steps in that particular trajectory. This means each trajectory is reparametrized differently, so the ensemble-averaged covariance C̃(dτ) computed in Eq. (7) mixes the increment correlations with the distribution of T_FOD. Moreover, the rescaled time step τ'_FOD = τ × T/T_FOD is generally not an integer, so the FOD process is not an integer-time process in the sense claimed in the abstract and introduction. The authors should specify the ensemble procedure unambiguously (e.g., a fixed global rescaling or a conditional analysis) and clarify in what precise sense FOD is a discrete-time analogue.","section":"Sec. III, Eq. (6)"}],"minor_comments":[{"comment":"The text refers to 'FOB schemes' in the paragraph after Fig. 2; this should read 'FOD schemes'.","section":"Sec. IV"},{"comment":"Table I lists H=0.6, 0.7, and 0.9 but omits H=0.8, which is the representative case used in Figs. 1-3; either include H=0.8 in the table or explain the omission.","section":"Table I"},{"comment":"The caption says 'ensemble averaged absolute value of X(t)', but the text does not distinguish between ⟨|X(t)|⟩ and √⟨X^2(t)⟩; please state explicitly which quantity is plotted and why the mean squared displacement itself is not shown.","section":"Fig. 1"},{"comment":"The Pearson values, e.g., '.999937', should be written as '0.999937'; also report the number of trajectories used to compute the coefficient and the statistical uncertainty, since the reported values differ only in the fourth decimal.","section":"Sec. IV, Pearson coefficient"},{"comment":"The reduced covariance is normalized by ⟨δ_t δ_{t+1}⟩; for the TLD and FOD processes this denominator is not equal to H(2H−1), so the comparison in Fig. 2 should state how each curve is normalized before dividing by the common factor.","section":"Eq. (7)"},{"comment":"The sentence 'The method can be easily applied to D dimensional systems with D>1' is not accompanied by any test; either provide a D>1 demonstration or label this as a conjecture.","section":"Sec. IV"}],"recommendation":"major_revision","confidential_remarks":"The paper's central claim is fixable by reframing the result as asymptotic equivalence rather than exact equality of covariance functions, and by redoing the fitting analysis with a proper treatment of finite-window bias and error bars. The current version overstates the numerical agreement, and the FOD rescaling needs a cleaner definition. The authors' data/code availability statement says 'available from the authors under reasonable request'; given the quantitative nature of the claims, I would encourage depositing the code and data in a permanent repository."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a useful methods paper with one new ingredient and one load-bearing overstatement. The FOD scheme—rounding fBm increments to integers and rescaling time per trajectory—is new as far as I can tell. The TLD scheme is sign binarization, essentially Ref. 30, and the authors say so. The numerical validation is serious: 2000 realizations, T=2^22, several H values, and they check both ensemble covariance and individual trajectory correlation.\n\nWhere it wobbles: the paper claims 'perfect agreement' between the fitted decay exponents and 2−2H, but Table I doesn't support that. For H=0.7, theory says 0.6 and they fit 0.628±0.004 for TLD and 0.626±0.004 for FOD; that's many times the quoted error. For H=0.9, theory 0.2, fits 0.225±0.002 and 0.211±0.001. The stress-test note has the right explanation: for TLD, the covariance of the sign of a bivariate Gaussian with correlation ρ is (2/π) arcsin ρ, which is not a pure power law. Fitting that over a finite window yields an apparent exponent systematically above 2−2H, matching the observed bias. So the schemes reproduce the asymptotic tail, not the exact covariance function. That distinction matters for the title claim.\n\nThe FOD time rescaling, Eq. (6), is also ad hoc and applied per realization. That can bias ensemble averages, though the reported moments look stable. And the broader statement about sub-diffusion being 'non-universal' is speculative; the paper doesn't have a construction for sub-diffusive fBm, so the comparison is not supported.\n\nWhat's genuinely good: the FOD recipe is simple and seems to track individual trajectories remarkably well (Pearson coefficients >0.999). If you need correlated binary sequences or a lattice walk with long-range persistence, this is a practical starting point. The paper is honest about not solving the sub-diffusive case.\n\nBottom line: worth a serious referee, but the authors need to either derive the covariance for the discretized processes or explicitly claim only asymptotic equivalence, report the fits with honest error bars, and release code/data. The central 'counterpart' claim is too strong as written.","headline":"A useful practical recipe for lattice super-diffusive fBm, but the numerical evidence undercuts the paper's central claim of exact covariance preservation.","tokens_in":10726,"tokens_out":2776,"would_cite":true,"duration_ms":31205,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G22","60G15","82C31"],"pacs":[],"model":"deepseek-v4-flash","headline":"Two simple rounding rules turn a super-diffusive fractional Brownian motion into an integer-time lattice walk, preserving its power-law memory.","keywords":["fractional Brownian motion","fractional Gaussian noise","super-diffusion","lattice random walk","power-law correlations","Hurst exponent","binary sequence discretization","anomalous diffusion"],"falsifier":"Simulate the TLD and FOD walks for $H=0.55$ and $H=0.95$ with $T=2^{22}$ and roughly 2000 realizations, fit the reduced increment covariance $\\widetilde{C}(d\\tau)$ over lags from 1 to 1000, and check whether the fitted exponents agree with $2-2H$ within error; a mismatch, or a strong dependence on the per-trajectory rescaling rule in FOD, would falsify the claimed universality.","tokens_in":9686,"feed_emoji":"🎲","tokens_out":7145,"duration_ms":71968,"temperature":0.7,"pith_summary":"Fractional Brownian motion (fBm) is the canonical Gaussian process with power-law memory, but its trajectories live in continuous space and time, which is awkward for lattice simulations and for any setting where steps must be integers. This paper proposes two elementary discretizations — taking only the sign of each increment (TLD), or rounding its magnitude to a positive integer while keeping the sign (FOD) — and argues numerically that both produce an integer-time random walk on the integers with the same power-law increment covariance as the parent fBm for Hurst exponents $H>1/2$. The claim matters because a faithful lattice analogue makes the super-diffusive regime accessible to exact numerics and to models built from correlated binary or integer sequences, while also exposing a structural difference from sub-diffusive fBm, for which no such analogue is achieved.","feed_headline":"A lattice walk that keeps fractional Brownian motion's power-law memory","feed_subtitle":"Thresholding or rounding each increment preserves the $2-2H$ covariance exponent and the shape of individual trajectories.","key_machinery":"The load-bearing identity is the increment covariance of fBm, $C(|t-t'|)=\\langle\\delta_t\\delta_{t'}\\rangle=H(2H-1)/|t-t'|^{2-2H}$, whose power-law exponent $\\varepsilon=2-2H$ is the quantity that must survive discretization. The two algorithms are the machinery: TLD maps each increment $\\delta_t$ to $\\operatorname{sign}(\\delta_t)$, while FOD maps it to $\\pm\\lfloor|\\delta_t|+1/2\\rfloor$ and rescales the effective time step by the realized walk length so that comparison with the continuous path is meaningful. Because the paper works with centered Gaussian processes, matching this second-order statistic fixes the process; the numerics check the reduced covariance at lags spanning three decades.","core_discovery":"The central discovery is that a super-diffusive fBm with Hurst index $H>1/2$ survives a drastic discretization of its increments. If $\\delta_t$ is an increment of a continuous fBm trajectory, the TLD rule replaces it by $+1$ or $-1$ according to its sign, and the FOD rule replaces it by $\\pm\\lfloor|\\delta_t|+1/2\\rfloor$ with a per-trajectory rescaling of the time step. Numerical fits for $H=0.6$, $0.7$, $0.8$, and $0.9$ show that the reduced increment covariance $\\widetilde{C}(d\\tau)=\\langle\\delta_t\\delta_{t+d\\tau}\\rangle/\\langle\\delta_t\\delta_{t+1}\\rangle$ decays as $(d\\tau)^{-\\varepsilon}$ with $\\varepsilon$ matching $2-2H$ for both schemes, and individual discretized trajectories follow the parent trajectory with Pearson coefficients above $0.999$. The FOD scheme is slightly closer to the parent process. The same construction fails for sub-diffusion, where correlations are integrable and short-time sign structure matters.","pith_inferences":["If the exponent preservation holds for all $H\\in(1/2,1)$, one can view sign-thresholded fractional Gaussian noise as a constructive source of binary sequences with arbitrary tunable power-law correlations, potentially useful in coding and memory models; the paper only demonstrates this numerically for four values of $H$.","The per-realization time rescaling in FOD may bias ensemble covariance estimates; testing a global, ensemble-level rescaling would clarify whether the apparent advantage of FOD over TLD is intrinsic or an artifact of the normalization.","The near-perfect Pearson correlation between continuous and discretized individual trajectories suggests that, for $H>1/2$, the sign process carries almost all the macroscopic shape of the path; a formal bound on this linear dependence would turn the numerical observation into a theorem."],"forward_implications":["For any $H>1/2$, one can generate an integer-time walk on $\\mathbb{Z}$ whose increment correlations decay as $|t-t'|^{-(2-2H)}$, giving a ready-made lattice model for super-diffusive transport.","The TLD scheme produces a binary $\\pm1$ sequence with power-law correlations, useful wherever correlated binary inputs are needed.","The FOD scheme is the more faithful of the two, nearly indistinguishable from the parent fBm covariance over three decades of lag.","The construction extends to $D$ dimensions by running independent copies per axis.","A lattice analogue of sub-diffusive fBm remains open, and the authors argue sub-diffusion is governed by non-universal short-time features rather than by long-range integrated correlations."],"supporting_citations":[{"why":"Supplies the foundational definition of fBm and the increment covariance $H(2H-1)/|t-t'|^{2-2H}$ that the lattice analogue must reproduce.","marker":"[25]"},{"why":"Provides an earlier use of sign-thresholding of correlated sequences that the TLD scheme extends with a detailed covariance and trajectory analysis.","marker":"[30]"},{"why":"Supplies the circulant (Davies-Harte) simulation algorithm used here to generate the parent continuous fBm trajectories.","marker":"[36]"},{"why":"Introduces the circulant embedding method that makes fast exact simulation of stationary Gaussian series possible.","marker":"[38]"},{"why":"Extends the circulant simulation approach to stationary Gaussian vector fields, underpinning the numerical generation of fBm trajectories at large times.","marker":"[41]"},{"why":"Provides the spectral simulation method for fBm that the numerical analysis relies on for efficiency at large $T$.","marker":"[37]"}],"fun_headline_variants":["Super-diffusive fBm keeps memory under lattice discretization","Sign-based lattice walk reproduces fBm covariance exponent","Discrete-time analogue of super-diffusive fractional Brownian motion","Integer-time fBm analogue with power-law memory intact"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that replacing an fBm increment by its sign, or by its rounded magnitude, preserves the power-law decay exponent $\\varepsilon=2-2H$ of the increment covariance for every $H>1/2$; the paper checks this only by finite-time fits at $H=0.6$, $0.7$, $0.8$, and $0.9$, and the FOD rescaling is applied separately to each trajectory, which could bias ensemble averages.","fun_headline_variants_meta":{"raw":{"variants":["Super-diffusive fBm keeps memory under lattice discretization","Sign-based lattice walk reproduces fBm covariance exponent","Discrete-time analogue of super-diffusive fractional Brownian motion","Integer-time fBm analogue with power-law memory intact"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000872,"raw_usage":{"total_tokens":3779,"prompt_tokens":955,"completion_tokens":2824,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":2753}},"tokens_in":571,"tokens_out":2824,"duration_ms":21574,"temperature":1.0,"reasoning_tokens":2753,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:37:37.983827+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the TLD and FOD walks for $H=0.55$ and $H=0.95$ with $T=2^{22}$ and roughly 2000 realizations, fit the reduced increment covariance $\\widetilde{C}(d\\tau)$ over lags from 1 to 1000, and check whether the fitted exponents agree with $2-2H$ within error; a mismatch, or a strong dependence on the per-trajectory rescaling rule in FOD, would falsify the claimed universality.","supporting_citations":[{"cited_title":"Mandelbrot and J","cited_arxiv_id":null,"evidence_quote":"Supplies the foundational definition of fBm and the increment covariance $H(2H-1)/|t-t'|^{2-2H}$ that the lattice analogue must reproduce."},{"cited_title":"Czirok, R","cited_arxiv_id":null,"evidence_quote":"Provides an earlier use of sign-thresholding of correlated sequences that the TLD scheme extends with a detailed covariance and trajectory analysis."},{"cited_title":"Dieker, Simulation of fractional Brownian motion , PhD Thesis, (University of Twente, Enschede, 2004)","cited_arxiv_id":null,"evidence_quote":"Supplies the circulant (Davies-Harte) simulation algorithm used here to generate the parent continuous fBm trajectories."},{"cited_title":"Davies and D.S","cited_arxiv_id":null,"evidence_quote":"Introduces the circulant embedding method that makes fast exact simulation of stationary Gaussian series possible."},{"cited_title":"Chan and A","cited_arxiv_id":null,"evidence_quote":"Extends the circulant simulation approach to stationary Gaussian vector fields, underpinning the numerical generation of fBm trajectories at large times."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the spectral simulation method for fBm that the numerical analysis relies on for efficiency at large $T$."}],"review_version":1}