REVIEW 2 major objections 8 minor 30 references
Beyond Self-Repellent Kernels: History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs
T0 review · 2 major / 8 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read The paper proposes a history-driven target framework: replace the MCMC target $\boldsymbol{\mu}$ with $\pi_i[\mathbf{x}] \propto \mu_i (x_i/\mu_i)^{-\alpha}$, and proves the empirical measure still converges almost surely to…
desk verdict A real upgrade to self-repellent graph sampling that moves history dependence into the target and gets a clean V_base/(2alpha+1) CLT; the one genuine gap is an unproven SA regularity condition for non-reversible augmented samplers. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the history-driven target $\pi_i[\mathbf{x}] \propto \mu_i (x_i/\mu_i)^{-\alpha}$, whose unnormalized form $\tilde{\pi}_i = \tilde{\mu}_i (\tilde{x}_i/\tilde{\mu}_i)^{-\alpha}$ depends only on the current state's visit count and target weight, not on neighbors. Lemma 3.1 proves Conditions C1–C4 (scale invariance, local dependence, fixed point at $\boldsymbol{\mu}$, history dependence) force exactly this power-law form. The analysis treats the empirical measure recursion as stochastic approximation with mean field $h(\mathbf{x}) = \pi[\mathbf{x}] - \mathbf{x}$ and ODE $\dot{\mathbf{x}} = \pi[\mathbf{x}] - \mathbf{x}$; the Lyapunov function $V(\mathbf{x}) = \sum_i \mu_i (x_i/\mu_i)^{-\alpha}$ establishes global asymptotic stability of $\boldsymbol{\mu}$, and the Jacobian $\nabla h(\boldsymbol{\mu}) = \alpha \boldsymbol{\mu}\mathbf{1}^T - (\alpha+1)I$ produces the factor $1/(2\alpha+1)$ in the covariance.
What would settle it
Run HDT-MCMC on a graph with a known spectral gap, such as a barbell graph with Metropolis-Hastings, and compare the empirical covariance of $\sqrt{n}(\mathbf{x}_n - \boldsymbol{\mu})$ at large $\alpha$ and large $n$ against $V^{\mathrm{base}}/(2\alpha+1)$; if the ratio drifts with $n$ or the CLT fails to stabilize for $\alpha = 10$ with $n = 10^8$, the continuity or Poisson-equation assumption is violated.
Extended reading notes
Core claim
The central claim is Theorem 3.3: for any ergodic base MCMC sampler whose kernel $P[\mathbf{x}]$ targets $\pi[\mathbf{x}] \propto \mu_i (x_i/\mu_i)^{-\alpha}$, the HDT update $\mathbf{x}_{n+1} = \mathbf{x}_n + \frac{1}{n+1}(\delta_{X_{n+1}} - \mathbf{x}_n)$ satisfies $\mathbf{x}_n \to \boldsymbol{\mu}$ almost surely and $\sqrt{n}(\mathbf{x}_n - \boldsymbol{\mu}) \rightsquigarrow \mathcal{N}(0, V^{\mathrm{base}}/(2\alpha+1))$, where $V^{\mathrm{base}}$ is the base sampler's limiting covariance when it targets $\boldsymbol{\mu}$ directly. The proof works for non-reversible samplers on augmented state spaces by defining the empirical measure only on the original space and using the marginal stationary distribution, something the SRRW construction cannot do because its kernel modification requires detailed balance. A companion cost-based CLT (Lemma 3.6) gives $C_{\mathrm{HDT}} V_{\mathrm{HDT}}(\alpha) \preceq (2/\mathbb{E}_{i\sim\mu}[|N(i)|]) \, C_{\mathrm{SRRW}} V_{\mathrm{SRRW}}(\alpha)$, so the advantage grows with average degree.
Load-bearing premise
The result depends on the base sampler's kernel $P[\mathbf{x}]$ being continuous in $\mathbf{x}$ and satisfying uniform Poisson-equation bounds for the stochastic-approximation CLT; for non-reversible or augmented-state samplers this is asserted rather than verified, and if the chain's mixing slows dramatically as $\pi[\mathbf{x}]$ concentrates, the variance reduction would require impractically many samples.
Editorial extensions
If this is right
- Every base sampler, reversible or not, inherits a $1/(2\alpha+1)$ contraction of its asymptotic covariance while remaining unbiased for $\boldsymbol{\mu}$.
- Variance orderings between samplers are preserved: if sampler A beats sampler B in covariance, HDT-A beats HDT-B at every $\alpha$.
- Under a fixed compute budget, HDT's cost-scaled covariance is at most $2/\mathbb{E}[|N(i)|]$ times SRRW's, so the advantage widens as average degree grows.
- The framework admits non-reversible samplers such as MHDA and 2-cycle chains, which SRRW cannot use.
- With an LRU cache holding about 10% of the visit counts, HDT-MHRW still beats the base sampler in total variation distance.
Reading between the lines
- Beyond the paper, the same target-replacement trick could apply to continuous or very high-dimensional state spaces where enumerating neighbors is impossible, because the acceptance ratio only needs the two local pairs $(x_i,\mu_i)$ and $(x_j,\mu_j)$.
- The LRU experiments suggest the empirical measure can be approximated by temporal locality; a testable question is whether a streaming frequency sketch preserves the $1/(2\alpha+1)$ covariance constant or degrades it by a known factor.
- The cost-based comparison implies HDT's value is largest when neighbor enumeration is expensive, such as dense graphs, high-degree nodes, or rate-limited graph APIs, and minimal on low-degree structures like paths where $\mathbb{E}[|N(i)|]$ is near 2.
- If the same Lyapunov structure transfers to continuous target densities, HDT could reduce variance in high-dimensional MCMC without gradient information.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a history-driven target (HDT) framework for MCMC on finite discrete state spaces (graphs). Instead of modifying the transition kernel as in the self-repellent random walk (SRRW), HDT replaces the target distribution μ by a history-dependent target π[x] ∝ μ_i (x_i/μ_i)^{-α}, where x is the empirical visit measure. Any base reversible or non-reversible MCMC sampler can be run with this adaptive target, and the empirical measure is updated after each draw. The main theoretical claims are: a.s. convergence of x_n to μ (Theorem 3.3(a)), a CLT with covariance V_HDT(α)=V_base/(2α+1) (Theorem 3.3(b)), and a cost-based CLT/Lemma 3.6 showing an advantage over SRRW under a fixed compute budget. Experiments on several real-world graphs with MH, MTM, MHDA, and 2-cycle samplers show consistent TVD/NRMSE improvements, and an LRU cache is introduced for memory reduction in large graphs.
Significance. If the theoretical claims hold, this is a significant and useful contribution: it offers a universal, cheap variance-reduction wrapper that extends the self-repellent mechanism to non-reversible and augmented-state MCMC samplers, with an elegantly derived 1/(2α+1) covariance reduction. The cost-based comparison with SRRW under a fixed budget is practically important, and the experimental validation is extensive. However, the proof of the central CLT for non-reversible augmented-state samplers has a gap in the verification of the stochastic-approximation conditions; the universality claim is therefore not yet fully established. The linear-algebra derivation of the covariance itself is correct and the ODE stability proof is sound.
major comments (2)
- [Appendix E.1] The verification of condition (B2) of Theorem E.1 is incomplete. Equation (30) requires sup_{x∈C,i} ||m_x(i)||^2 < ∞ for every compact C, but the proof only shows pointwise finiteness of the fundamental matrix for each x∈Int(Σ). Assumption 1 only guarantees x_n∈Int(Σ), not that the trajectory lies in a compact subset of Int(Σ) bounded away from the boundary; the update (2) lets components decay as O(1/n). For x approaching ∂Σ, π[x] becomes extremely concentrated and the spectral gap of P[x] may shrink, making (I−P[x]+1π[x]^T)^{-1} unbounded. The reference to truncation devices in Doshi et al. (2023, Remark 4.5 and Appendix E) is not implemented or analyzed for Algorithm 1, so the CLT formula (10) is not justified for the claimed class of non-reversible samplers.
- [Appendix E.3] The extension to non-reversible augmented-state samplers (MHDA, 2-cycle) replaces H by Φ on X×Y, but the required uniform bounds on the augmented fundamental matrix are exactly the missing control. The proof asserts continuity of P[x] in x and claims this suffices for Fort's condition (31); however, continuity on the non-compact domain Int(Σ) does not yield a uniform modulus over the set actually visited by the SA recursion, which may approach the boundary. For MHDA and the 2-cycle chain, the kernel is piecewise defined or depends on the auxiliary state, and the spectral gap may vanish near ∂Σ. Thus the proof does not establish Theorem 3.3(b) for these samplers, which is the paper's headline claim of compatibility with non-reversible MCMC.
minor comments (8)
- [Appendix B] In the final step of the proof of Lemma 3.1, the condition '1 − ρ2 ≤ 1' appears to be a typo; to derive α ≥ 0 from C4 the intended condition is 1−ρ2 < 0. As written, the derivation is logically incomplete.
- [Lemma 3.1] The 'if and only if' statement assumes differentiability of f in the proof, while C2 only requires continuity. Either C2 should be strengthened to 'continuously differentiable' or the proof should be adapted to the weaker condition.
- [Section 2 / Appendix E] The step size in the empirical-measure update is 1/(n+1) in Eq. (2) but 1/(n+2) in Eq. (35). The indexing should be made consistent.
- [Algorithm 2] Step 5 has a typo: 'Stpe' should be 'Step'.
- [Abstract] The phrase 'unbiased samples' is imprecise for MCMC; use 'asymptotically unbiased' or 'consistent' to avoid confusion with finite-sample unbiasedness.
- [Appendix H.1] The list T = {3000, 15000, 15000, 3000} is not explicitly matched to the four graphs; please specify the correspondence.
- [Section 4.5] Equation (15) for the LRU cache is heuristic; a brief explanation of why the neighbor-average approximates the missing frequency would improve readability.
- [Section 2] The assumption that the base kernel P is full-rank is not needed for formula (4) and excludes some ergodic chains with zero eigenvalues; consider restating the condition as 'ergodic and aperiodic'.
Circularity Check
No significant circularity: the central CLT is derived from standard stochastic-approximation theorems and an explicit target design; self-citations serve as external baselines, not as load-bearing inputs.
full rationale
The load-bearing derivation, Theorem 3.3, is not circular. The empirical-measure recursion is written as a stochastic approximation with H(x, X_{n+1}) = delta_{X_{n+1}} - x and mean field h(x) = pi[x] - x (Appendix E, Eq. (35)); the asymptotic covariance is then computed from the Fort/Delyon CLT, with U identified as V_base through the identity pi[mu] = mu (condition C3). The factor 1/(2alpha+1) arises from the Jacobian grad h(mu) = alpha mu 1^T - (alpha+1)I, not from any fitted parameter. Lemma 3.1 is a characterization, not a tautology: conditions C1-C4 are explicit design axioms, and Appendix B proves via scale-invariance and local-dependence arguments that any such pi must be the power law mu_i (x_i/mu_i)^{-alpha}; the qualitative condition C4 only fixes the sign of the exponent, so the derivation has independent content. The self-citations are not load-bearing: Doshi et al. (2023) supplies the external SRRW baseline used only in the comparison Lemma 3.6, and Hu et al. (2024b) is cited only for the standard Poisson-equation solution form, which is also supported by Bremaud (2013). Footnote 6 in Appendix B admits a missing literature reference but provides a self-contained proof, making it a completeness caveat rather than a circular step. The genuine weakness is in Appendix E.1: condition (B2) for non-reversible augmented-state chains is asserted via continuity of P[x] rather than a uniform spectral-gap bound, so the CLT for MHDA and 2-cycle chains rests on an unverified technical hypothesis; this is a correctness risk, not a circular reduction. Finally, the cost-based comparison in Lemma 3.6 is elementary algebra from the two covariance formulas and the stated per-sample cost model, with no fitted quantity renamed as a prediction.
Assumptions & free parameters
free parameters (3)
- α (self-repellency strength) =
None (user-chosen; experiments use α=5, grid α∈{0,0.5,1,2,5,10})
- LRU cache capacity ratio r =
r∈{0.01,...,0.2} in experiments
- Fake visit count initialization x0 =
Uniform/Degree/Dirichlet(0.5)
assumptions (5)
- domain assumption Base MCMC kernel P is full-rank and continuous in its target distribution (Section 2, Preliminaries)
- domain assumption Assumption 1: x_n ∈ Int(Σ) almost surely for all n
- ad hoc to paper C2: π_i[x] depends only on (x_i, μ_i), not on the neighborhood
- ad hoc to paper C4: under-sampled states get higher target probability (formalized as α ≥ 0)
- standard math Stochastic approximation theory (Fort 2015, Delyon 2000) applies to the empirical measure recursion (35) with controlled Markovian noise
Cite this review
Pith. "Pith review of Beyond Self-Repellent Kernels: History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs." pith.science (2026). https://pith.science/paper/3U4LMGK7
@misc{pith2026250518300,
author = {Pith},
title = {Pith review of: Beyond Self-Repellent Kernels: History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs},
year = {2026},
howpublished = {\url{https://pith.science/paper/3U4LMGK7}},
note = {Machine review of arXiv:2505.18300}
}
abstract
We propose a history-driven target (HDT) framework in Markov Chain Monte Carlo (MCMC) to improve any random walk algorithm on discrete state spaces, such as general undirected graphs, for efficient sampling from target distribution $\boldsymbol{\mu}$. With broad applications in network science and distributed optimization, recent innovations like the self-repellent random walk (SRRW) achieve near-zero variance by prioritizing under-sampled states through transition kernel modifications based on past visit frequencies. However, SRRW's reliance on explicit computation of transition probabilities for all neighbors at each step introduces substantial computational overhead, while its strict dependence on time-reversible Markov chains excludes advanced non-reversible MCMC methods. To overcome these limitations, instead of direct modification of transition kernel, HDT introduces a history-dependent target distribution $\boldsymbol{\pi}[\mathbf{x}]$ to replace the original target $\boldsymbol{\mu}$ in any graph sampler, where $\mathbf{x}$ represents the empirical measure of past visits. This design preserves lightweight implementation by requiring only local information between the current and proposed states and achieves compatibility with both reversible and non-reversible MCMC samplers, while retaining unbiased samples with target distribution $\boldsymbol{\mu}$ and near-zero variance performance. Extensive experiments in graph sampling demonstrate consistent performance gains, and a memory-efficient Least Recently Used (LRU) cache ensures scalability to large general graphs.
Figures
Figures from the paper (12 more)
Reference graph
Works this paper leans on
-
[1]
Thus, α ≜ ρ2 − 1 ≥ 0, and πi[x] ∝ µi(xi/µi)−α, which completes the proof. C. Implementation of Algorithm 1 For Advanced MCMC Samplers C.1. Multiple-try Metropolis algorithm with locally balanced weighs function (MTM) We followed the MTM algorithm in (Chang et al., 2022), where weights function incorporated locally balanced function h introduced by (Zanell...
work page 2022
-
[2]
Summary of graph datasets. Name # of nodes. # of edges Average degree WikiV ote 889 2,914 6.55 Facebook 4039 88,234 43.69 p2p-Gnutella08 6301 20,777 6.59 p2p-Gnutella04 10876 39,994 7.35 Table 2 shows the detailed summary of the graph used in our experiments. We used different numbers of iterations for each graph T = {3000, 15000, 15000, 3000}, respective...
work page 2000
-
[3]
0 500 1000 1500 2000 2500 3000 Steps 0.4 0.5 0.6 0.7 0.8 0.9 1.0TVD WikiVote Graph MHRW HDT-MHRW ( =
work page 2000
-
[5]
0 2000 4000 6000 8000 10000 12000 14000 Steps 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200NRMSE facebook Graph MHRW HDT-MHRW ( =
work page 2000
-
[7]
Li, Y ., Wu, Z., Lin, S., Xie, H., Lv, M., Xu, Y ., and Lui, J. C. Walking with perception: Efficient random walk sampling via common neighbor awareness. In 2019 IEEE 35th International Conference on Data Engineering (ICDE), pp. 962–973. IEEE,
work page 2019
-
[9]
Neal, R. M. Improving asymptotic variance of mcmc esti- mators: Non-reversible chains are better. arXiv preprint math/0407281,
-
[10]
Pynadath, P., Bhattacharya, R., HARIHARAN, A. N., and Zhang, R. Gradient-based discrete sampling with auto- matic cyclical scheduling. In ICML 2024 Workshop on Structured Probabilistic Inference & Generative Model- ing,
work page 2024
-
[13]
Challenging the limits: Sampling online social networks with cost constraints
Xu, X., Lee, C.-H., et al. Challenging the limits: Sampling online social networks with cost constraints. In IEEE Infocom 2017-IEEE Conference on Computer Communi- cations, pp. 1–9. IEEE,
work page 2017
Show all 30 references
-
[16]
Proof of Lemma 3.2 The proof consists of three parts: We first show that V (x) = X i∈X µi(xi/µi)−α (26) is the Lyapunov function of the ODE ˙x = π[x] − x
end for 16 History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs D. Proof of Lemma 3.2 The proof consists of three parts: We first show that V (x) = X i∈X µi(xi/µi)−α (26) is the Lyapunov function of the ODE ˙x = π[x] − x. (27) Then, we prove that µ is the u...
2000
-
[17]
In our Algorithm 1, replacing the target of the base MCMC sampler from µ to π[x] results in an ergodic transition matrix P [x] parameterized by x, which is π[x]-invariant
Now, we focus on (B2). In our Algorithm 1, replacing the target of the base MCMC sampler from µ to π[x] results in an ergodic transition matrix P [x] parameterized by x, which is π[x]-invariant. For a function m : X →Rd, we define the operator (P [x]m)(i) ≜ X j∈X Pij[x]m(j). F...
2013
-
[18]
tX s=1 Φ(θ∗, Zs) #
The third last equality comes from the fact that I = P|X | i=1 uivT i = µ1T + P|X | i=2 uivT i and I T = 1µT + P|X | i=2 viuT i = I. E.3. Adaptation to Non-Reversible Markov Chains For non-reversible Markov chains that directly modify the transition probabilities such as Suwa ...
2010
-
[20]
For simulation setting, We run 50 independent trials with 300,000 steps and set one-third of the number of steps as burn-in period before collecting samples
Furthermore, we test our design using the ego-Gplus graph (Leskovec & Krevl, 2014), which contains about 100K nodes, and implemented r = 0.01 and r = 0.1 in LRU. For simulation setting, We run 50 independent trials with 300,000 steps and set one-third of the number of steps as...
2014
-
[22]
More results on HDT-MCMC in graph sampling In this section, we present more simulation results on the algorithm that utilizes HDT-MCMC framework for graph sampling in other graph
H.2. More results on HDT-MCMC in graph sampling In this section, we present more simulation results on the algorithm that utilizes HDT-MCMC framework for graph sampling in other graph. To demonstrate the impact of HDT-MCMC, we implement HDT-version of both reversible and nonre...
2000
-
[24]
0 2000 4000 6000 8000 10000 12000 14000 Steps 0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200NRMSE p2p-Gnutella08 Graph MHRW HDT-MHRW ( =
2000
-
[25]
Moreover, in p2p-Gnutella08 graph, we notice that the performance of LRU design is robust to the choice of the memory capacity
We also observed LRU with 80% memory reduction can still outperforms standard MH. Moreover, in p2p-Gnutella08 graph, we notice that the performance of LRU design is robust to the choice of the memory capacity. A comparable performance 26 History-Driven Target Towards Efficient...
2000
-
[26]
500 1000 1500 2000 2500 3000 Steps 0.4 0.5 0.6 0.7 0.8 0.9 1.0TVD WikiVote Graph MHRW MHRW-HDT ( = 0.5) MHRW-HDT ( =
TVD Comparison for HDT-2cyc. 500 1000 1500 2000 2500 3000 Steps 0.4 0.5 0.6 0.7 0.8 0.9 1.0TVD WikiVote Graph MHRW MHRW-HDT ( = 0.5) MHRW-HDT ( =
2000
-
[28]
with r=1 2000 4000 6000 8000 100001200014000 Steps 0.5 0.6 0.7 0.8 0.9 1.0TVD p2p-Gnutella08 Graph MHRW HDT-MHRW ( =
2000
-
[29]
Simulation of incorporating LRU in HDT-MHRW framework to reduce memory issues. 27 History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs 0 500 1000 1500 2000 2500 3000 Steps 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40NRMSE WikiVote Graph MHRW HDT-MHRW ( =
2000
-
[30]
µi ∝ di with α = 1 in HDT
and Whois graph (with 7476 nodes and 56.9K edges) (Mahadevan et al., 2006), and set the target distribution proportional to the degree distribution, i.e. µi ∝ di with α = 1 in HDT. Similar to the simulation settings in our manuscript, we compare HDT-MCMC to baseline algorithms...
2006
-
[1999]
and Hwang, C.-R
Chen, T.-L. and Hwang, C.-R. Accelerating reversible markov chains. Statistics & Probability Letters, 83(9): 1956–1962,
1956
-
[2000]
Ma, Y .-A., Chen, T., Wu, L., and Fox, E. B. A unifying framework for devising efficient and irreversible mcmc samplers. arXiv preprint arXiv:1608.05973,
-
[2002]
Revisiting random walks for learning on graphs
Kim, J., Zaghen, O., Suleymanzade, A., Ryou, Y ., and Hong, S. Revisiting random walks for learning on graphs. arXiv preprint arXiv:2407.01214,
-
[2005]
Online statistical inference for nonlinear stochastic approximation with markovian data
Li, X., Liang, J., and Zhang, Z. Online statistical inference for nonlinear stochastic approximation with markovian data. arXiv preprint arXiv:2302.07690,
-
[2007]
and Livingstone, S
Andrieu, C. and Livingstone, S. Peskun–tierney ordering for markovian monte carlo: beyond the reversible scenario. The Annals of Statistics, 49(4):1958–1981,
1958
-
[2010]
Nonreversible mcmc from conditional invertible transforms: a complete recipe with convergence guarantees
Thin, A., Kotelevskii, N., Andrieu, C., Durmus, A., Moulines, E., and Panov, M. Nonreversible mcmc from conditional invertible transforms: a complete recipe with convergence guarantees. arXiv preprint arXiv:2012.15550,
2012 arXiv
-
[2013]
Given a sequence of random variable θ1, θ2, · · ·with partial sum Sn ≜ Pn s=1 θs such that 1√n Sn n→∞ − − − − → dist
Theorem 14.4) . Given a sequence of random variable θ1, θ2, · · ·with partial sum Sn ≜ Pn s=1 θs such that 1√n Sn n→∞ − − − − → dist. N (0, V ). Let nl be some positive random variable taking integer value such that θnl is on the same space as θn. In addition, for some determi...
2000
-
[2020]
Decentralized learning with random walks and communication-efficient adaptive optimization
Triastcyn, A., Reisser, M., and Louizos, C. Decentralized learning with random walks and communication-efficient adaptive optimization. In Workshop on Federated Learn- ing: Recent Advances and New Challenges (in Conjunc- tion with NeurIPS 2022),
2022
-
[2021]
Lifting markov chains to mix faster: limits and opportunities
Apers, S., Ticozzi, F., and Sarlette, A. Lifting markov chains to mix faster: limits and opportunities. arXiv preprint arXiv:1705.08253,
-
[2022]
Matrix Form of Covariance Matrix (4) The multivariate version of Br´emaud (2013, eq
12 History-Driven Target Towards Efficient Nonlinear MCMC on General Graphs A. Matrix Form of Covariance Matrix (4) The multivariate version of Br´emaud (2013, eq. (6.33)) shows that lim t→∞ 1 t E tX s=1 f (Xs) − ¯f ! tX s=1 f (Xs) − ¯f !T = Varµ(f (X0)) + 2 |X |X i=2 ...
2013
-
[2023]
Efficient unbiased sparsification
Barnes, L., Chow, T., Cohen, E., Frankston, K., Howard, B., Kochman, F., Scheinerman, D., and VanderKam, J. Efficient unbiased sparsification. arXiv preprint arXiv:2402.14925,
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.