{"id":"00bc3e0d-2d91-4aa2-8d81-69ed4a451e60","arxiv_id":"2505.04215","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Random walks with resetting on hypergraphs can be analyzed through the spectral decomposition of the reset-free walk, and resetting improves search when the coefficient of variation of first passage times is large.","lead":"This paper derives spectral formulas for random walks with resetting on hypergraphs, including occupation probability, stationary distribution, and mean first passage time. It also claims that these walks give different node rankings than the usual clique-expanded graph and that resetting can shorten cover time.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Optimal-reset condition is proven only for r=i; the abstract's 'general condition' for arbitrary reset nodes or subsets is unsupported.","rationale":"The reader's weakest_assumption identifies the same gap: the optimal-resetting condition is derived only for reset node equal to the starting node, yet the abstract claims generality. My own check of the spectral derivations found no algebraic error: the eigenvalue relation zeta_l = (1-gamma)lambda_l, the stationary distribution (54), and the MFPT formula (60) are consistent with direct Markov-chain calculations, and I verified numerically for a nontrivial first-passage distribution that Eq. (83) indeed holds at the interior optimum when r = i. The overgeneralization is therefore not a sign of internal inconsistency but of an unsupported scope claim. Because the authors explicitly state the restriction in Section V but do not qualify the abstract, and because the subset-reset case is mentioned but never treated, the paper needs a correction or proof extension. The reader's CONDITIONAL verdict already reflects this; my stress-test does not move it, so I recommend UNCHANGED. A concrete numerical test with r != i would settle whether the condition actually extends; if it fails, the authors must restrict the claim to r = i or supply a proof for the general case.","tokens_in":20464,"tokens_out":28380,"duration_ms":255913,"concrete_test":"On a small hypergraph (or a graph, which is a hypergraph with 2-edges), take a 5-node path with start i = 1, target j = 5, and reset node r = 3. Compute <T(gamma)> and <T^2(gamma)> exactly by solving the linear equations of the Markov chain with transition matrix Pi(r,gamma) = (1-gamma)W + gamma Theta(r) on a fine grid of gamma in [0,1]. Find gamma* that minimizes <T(gamma)>, then evaluate z^2(gamma*) = (<T^2> - <T>^2)/<T>^2 and compare with 1 + 1/<T(gamma*)>. Repeat with r = i = 1 as a control. If the equality holds for r = i but fails for r = 3, the paper's 'general condition' is false outside the proven special case.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section V begins 'For simplicity, we assume r = i and i != j' and derives the optimal-reset criterion z^2(gamma*) = 1 + 1/<T(gamma*)> (Eq. 83) and the sufficient condition (Eq. 87) only under this assumption. The derivation hinges on Eq. (79), which simplifies the eigenvector product <i|psi_l(i;gamma)><bar_psi_l(i;gamma)|j> using r = i so that the reset-dependent term proportional to <r|phi_l> cancels. For r != i, this cancellation does not occur, and the algebra leading to Eqs. (81)-(83) does not carry over. The abstract and introduction present the result as a 'general condition' without this restriction, and Section II mentions a subset of reset candidates but the associated treatment is 'omitted' and never actually provided. Since the optimal resetting condition is one of the headline theoretical contributions, the unsupported generality is a load-bearing gap. The core spectral formulas (eigenvalues, stationary distribution, MFPT) appear internally consistent and are not the source of the concern; the problem is the scope of the reset-optimality claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies random walks with resetting on hypergraphs using the spectral decomposition of the transition matrix. It derives exact formulas for the occupation probability, stationary distribution, and mean first passage time (Eqs. (54), (55), and (60)), establishes a spectral relation between the reset and reset-free transition matrices (Eqs. (38)--(40)), proposes a generalized hypergraph transition matrix, and gives conditions for optimal resetting in terms of the coefficient of variation (Eqs. (83) and (87)). It also presents applications to node ranking and cover time. The core spectral algebra is self-contained and internally consistent; the main concern is that the optimal-resetting condition is derived only for the special case r = i, i ≠ j, although the abstract and introduction present it as a general condition.","tokens_in":20691,"tokens_out":14828,"duration_ms":142684,"significance":"If the results are correct, the paper provides a useful spectral toolkit for reset processes on hypergraphs: all key quantities are expressed directly in terms of the eigenvalues and eigenvectors of the reset-free transition matrix, with no fitted parameters. The relationship between the spectra of the reset and reset-free matrices is cleanly derived and checkable. The proposed generalized transition matrix and the node-ranking experiments illustrate a genuinely hypergraph-native alternative to clique expansion. The optimal-resetting conditions, once properly scoped to the r = i setting, connect to the known restart literature and are a useful discrete-time refinement. The paper is largely a theoretical contribution; the experimental sections are illustrative rather than exhaustive.","major_comments":[{"comment":"The optimal-resetting criterion is derived only under the restriction r = i and i ≠ j, stated at the beginning of Section V. The key simplification in Eq. (79) relies on r = i so that ⟨r|φ_l⟩/⟨r|φ_1⟩ cancels or reduces to a simple factor; for arbitrary reset node this cancellation does not occur, and for i = j the term δ_ij in Eq. (57) changes the derivation. The abstract and Contribution (3) nevertheless call Eq. (83) a 'general condition' for the optimal reset probability. This is a load-bearing overclaim because the optimal-resetting condition is one of the headline contributions. Please either prove the general case or explicitly state the theorem as applying to r = i, i ≠ j, and adjust the abstract and introduction accordingly.","section":"Section V, Eqs. (61)--(83)"},{"comment":"The paper twice announces a treatment of resetting to a subset of candidate nodes. Section II states that 'the associated transition matrix is obtained in a similar manner, which is omitted here,' and Section IV says 'As mentioned above, a subset of nodes may be selected as reset candidates. This case is also discussed in the following way.' However, no subset version is actually provided anywhere in the manuscript. Either include the subset-resetting formulation or remove these promises; as written, the reader is left with an unfulfilled claim.","section":"Section II and Section IV"},{"comment":"As printed, Eq. (77) does not follow from Eq. (73): the second term in Eq. (77) should contain an additional factor ⟨T_ij(γ)⟩. In addition, Eq. (79) writes the eigenvector product without the overline on the left eigenvector; the correct factor should be ⟨i|ψ_l(i;γ)⟩⟨ψ̄_l(i;γ)|j⟩, and the middle expression should involve ⟨φ̄_l|j⟩ rather than ⟨φ_l|j⟩ in both terms. These are typos in the derivation of the headline condition, but they make the displayed algebraic chain incorrect as written and must be fixed before the result can be verified.","section":"Section V, Eqs. (77) and (79)"}],"minor_comments":[{"comment":"The notation 'N∑_{l=2}' is ambiguous: it appears to mean ∑_{l=2}^N, as in standard LaTeX rendering, but it can be read as a multiplicative factor N. Please use an explicit ∑_{l=2}^{N} or ∑_{l=2}^{n} throughout Section V to avoid confusion.","section":"Section V, Eqs. (61)--(63)"},{"comment":"There is a typo in the sentence 'Revisiting the first term of the occupation probability Pij(t), whcih is...': 'whcih' should be 'which'.","section":"Section III-B"},{"comment":"The phrase 'revisiting Eq.(21) and q.(27)' should read 'Eq. (27)' rather than 'q. (27)'.","section":"Section V, text after Eq. (73)"},{"comment":"The relationship between Eq. (83) and Reuveni's known universal condition CV = 1 at optimal restart should be discussed explicitly. In the continuous-time or large-mean limit the extra term 1/⟨T⟩ becomes negligible, but the discrete-time correction should be acknowledged.","section":"Section V, Eq. (83)"},{"comment":"The claim that the hypergraph-based ranking is 'more reliable' is not supported by any quantitative measure such as a rank correlation or an error bar. Consider adding a quantitative comparison or softening the wording.","section":"Section VI-A"}],"recommendation":"major_revision","confidential_remarks":"The core spectral framework is sound and publishable once the scope of the optimal-resetting theorem is stated correctly. I recommend requiring the authors to either prove the optimal-resetting condition for arbitrary reset nodes or clearly restrict the claim to r = i, i ≠ j, and to fix the typographical errors in Eqs. (77) and (79). The experimental sections are secondary and should not be the basis for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the paper does what it claims on the spectral side: it writes the resetting transition matrix as (1-gamma)W + gamma*Theta(r), derives its eigenvalues and eigenvectors in Theorems 4-7, and gets clean formulas for occupation probability, stationary distribution, and MFPT. The derivations are self-contained and, as far as I can tell, algebraically consistent. Second, the optimal-reset part is narrower than advertised. Section V states \"For simplicity, we assume r = i and i != j\" and then derives the coefficient-of-variation condition z^2(gamma*) = 1 + 1/<T(gamma*)> only under that assumption. The abstract presents this as a general condition, and the subset-of-reset-candidates case is mentioned in Section II but explicitly omitted. That is a load-bearing gap.\n\nThe main soft spot beyond the scope restriction is a concrete error in Section V. In Eqs. (62)-(63), C_ij and S_ij are defined with an extra factor N (total hyperdegree sum). Eq. (60) has no such factor, and Eq. (80), which identifies (1-gamma)[gamma*S'+S] with R^(0)_ij, would be false with that N present. It looks like a typo left over from writing the sums, and removing N restores consistency, but as printed the displayed algebra does not parse.\n\nThe empirical section is weak. The ranking experiment is one DBLP component, a scatter plot, and a qualitative claim of \"more reliable\" with no metric; the cover-time experiment uses 50 runs and no error bars. That part does not justify the \"extensive experiments\" language, but it is not central to the math.\n\nOn novelty: the weighting K_ij is from Carletti et al. and the paper says so, but the introduction still calls it \"we propose a research framework.\" The spectral relation between resetting and non-resetting walks is the standard result from Riascos et al. applied to a hypergraph transition matrix. So the contribution is specialization, not a new mechanism. That is fine for a niche paper, but the framing should be honest about it.\n\nWho gets value: anyone doing spectral random-walk theory on hypergraphs, especially if they want MFPT formulas under resetting. It is a reasonable target for a specialized journal after revision. I would send it to peer review, with a referee asked to check the Section V algebra and demand a proof or explicit scope for the optimality condition.","headline":"Useful spectral formulas for resetting walks on hypergraphs, but the headline optimal-reset condition is proved only for r=i and there is an extra factor N in the Section V algebra; both need fixing before the paper can be trusted as written.","tokens_in":21205,"tokens_out":3811,"would_cite":false,"duration_ms":38441,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C65","05C82","60J10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Resetting a hypergraph random walk is solved in closed form from the spectrum of the reset-free walk.","keywords":["hypergraph random walks","stochastic resetting","spectral theory","occupation probability","stationary distribution","mean first passage time","node ranking","cover time"],"falsifier":"Take a small connected hypergraph (for instance three nodes in a single three-node hyperedge), set the reset node $r$ different from the starting node $i$, and compute $\\langle T_{ij}(r,\\gamma)\\rangle$ exactly by solving the $n\\times n$ linear system for many $\\gamma$. If the minimizer does not satisfy $z^2(\\gamma^*)=1+1/\\langle T_{ij}(r,\\gamma^*)\\rangle$, the condition as stated is false; if it holds across all such small cases, that supports extending the proof to $r\\neq i$.","tokens_in":20272,"feed_emoji":"🎲","tokens_out":12839,"duration_ms":120119,"temperature":0.7,"pith_summary":"The paper considers a random walk on a hypergraph that, at each step, moves to a neighbour with probability $1-\\gamma$ or jumps back to a fixed reset node with probability $\\gamma$, and it asks what this resetting does to standard search statistics. Its answer is that the reset process is a rank-one perturbation of the ordinary hypergraph walk: the reset transition matrix $\\Pi(r,\\gamma)=(1-\\gamma)W+\\gamma\\Theta(r)$ has eigenvalues $1$ and $(1-\\gamma)\\lambda_l$, and its eigenvectors are explicit linear corrections to those of $W$. From this, the paper obtains closed-form expressions for the occupation probability, the stationary distribution, and the mean first passage time in terms of the reset-free eigenvalues and eigenvectors. It also derives a condition for the optimal reset probability, expressed through the coefficient of variation of the first-passage time, and a sufficient condition for a small reset rate to help. Because the framework keeps the hypergraph intact instead of converting it to a clique graph, node rankings differ from the standard expansion, and low reset rates can shorten cover time.","feed_headline":"One spectral formula solves resetting on hypergraphs","feed_subtitle":"The reset walk's eigenvalues are scaled copies of the plain walk's, giving exact search statistics","key_machinery":"The engine of the paper is the rank-one reset matrix $\\Theta(r)$, the matrix whose $r$-th column is all ones and whose other entries are zero, added to the plain hypergraph transition matrix as $\\Pi(r,\\gamma)=(1-\\gamma)W+\\gamma\\Theta(r)$. Because this is a rank-one perturbation, the spectrum of $\\Pi$ can be written directly from the spectrum of $W$: non-unit eigenvalues are scaled by $(1-\\gamma)$, and each eigenvector is shifted along the all-ones direction $|\\phi_1\\rangle$ by a coefficient $\\gamma\\langle r|\\phi_l\\rangle/\\big((1-(1-\\gamma)\\lambda_l)\\langle r|\\phi_1\\rangle\\big)$. That relation is what turns occupation probability, stationary distribution, and mean first passage time into finite sums over the reset-free eigen data, and it is also what converts the optimal-reset problem into the coefficient-of-variation condition $z^2(\\gamma^*)=1+1/\\langle T(\\gamma^*)\\rangle$.","core_discovery":"The central claim is a spectral dictionary between the hypergraph walk with resetting and the one without. Given the spectral decomposition $W=\\sum_{l=1}^n \\lambda_l|\\phi_l\\rangle\\langle\\bar\\phi_l|$, with $\\lambda_1=1$, the reset matrix $\\Pi(r,\\gamma)=(1-\\gamma)W+\\gamma\\Theta(r)$ has eigenvalue $1$ with right eigenvector $|\\phi_1\\rangle$, eigenvalues $(1-\\gamma)\\lambda_l$ for $l\\ge 2$, right eigenvectors $|\\phi_l\\rangle-\\frac{\\gamma}{1-(1-\\gamma)\\lambda_l}\\frac{\\langle r|\\phi_l\\rangle}{\\langle r|\\phi_1\\rangle}|\\phi_1\\rangle$, and left eigenvectors $\\langle\\bar\\phi_l|$ except for the stationary one, which picks up a sum of corrections. The paper then writes the stationary distribution under resetting as $\\frac{d_j}{N}+\\gamma\\sum_{l=2}^n\\frac{\\langle r|\\phi_l\\rangle\\langle\\bar\\phi_l|j\\rangle}{1-(1-\\gamma)\\lambda_l}$, the occupation probability as a one-line spectral sum, and the mean first passage time as Eq. (60). For the optimal reset rate, Section V derives, under the simplification $r=i$ and $i\\neq j$, the condition $z^2(\\gamma^*)=1+1/\\langle T_{ij}(\\gamma^*)\\rangle$, where $z$ is the coefficient of variation of the first-passage time, together with the sufficient condition $z^2(0)>1+1/\\langle T_{ij}(0)\\rangle$ for resetting to be beneficial at small $\\gamma$.","pith_inferences":["The coefficient-of-variation optimality condition likely applies beyond the proven $r=i$ case, because the proof uses only the first two moments of the first-passage distribution; a brute-force sweep on small hypergraphs with $r\\neq i$ would test this before a general proof is attempted.","Since $\\Pi$'s non-unit eigenvalues are $(1-\\gamma)\\lambda_l$, resetting strictly increases the spectral gap, so the reset walk is expected to mix faster in the spectral-gap sense; the paper does not discuss this corollary.","The exact mean-first-passage formula allows cover time to be optimized as a one-dimensional search over $\\gamma$ of the maximum target MFPT, replacing the paper's simulation-based sweep with a deterministic computation.","The weighting $K_{ij}=\\sum_\\alpha(C_{\\alpha\\alpha}-1)e_{i\\alpha}e_{j\\alpha}$ is a specific choice; substituting any edge-dependent vertex weighting gives a different $W$ but the same spectral machinery, so the framework can compare higher-order transition rules without changing the resetting analysis."],"forward_implications":["For any fixed hypergraph, one diagonalization of the reset-free walk $W$ is enough to evaluate every reset statistic for all $\\gamma$; no matrix inverse or new spectral calculation is needed per reset rate.","The optimal reset rate is determined by the first-passage distribution: the best $\\gamma^*$ is exactly where the relative fluctuation satisfies $z^2=1+1/\\langle T\\rangle$.","Resetting is guaranteed to improve the mean first passage time whenever the original process has enough fluctuation, $z^2(0)>1+1/\\langle T(0)\\rangle$, which is the signature of heavy-tailed searches.","Node importance in a collaboration hypergraph is not the same as in its clique expansion: the hypergraph-preserving stationary distribution rewards authors with fewer, larger collaborations relative to authors with many small collaborations.","On the tested 40-hyperedge network, small reset rates create a dip in cover time below the no-reset value, so resetting can be tuned to speed up full coverage even though large reset rates worsen it."],"supporting_citations":[{"why":"Defines the generalized adjacency and transition matrix on hypergraphs ($K_{ij}$ and $W$) whose spectral decomposition the paper uses as the reset-free base.","marker":"[23]"},{"why":"Introduces stochastic resetting on network random walks via $\\Pi(r,\\gamma)=(1-\\gamma)W+\\gamma\\Theta(r)$, the construction this paper transplants to hypergraphs.","marker":"[27]"},{"why":"Supplies the spectral and Laplace-transform derivation of occupation probability, stationary distribution, and mean first passage time that Theorems 1–3 adapt.","marker":"[16]"},{"why":"Establishes that hypergraph random walks need edge-dependent vertex weights, motivating the paper's choice of $K_{ij}$ rather than simple adjacency weights.","marker":"[22]"},{"why":"Provides the optimal-restart result that ties optimality to the relative fluctuation of first passage times, the origin of the coefficient-of-variation condition.","marker":"[54]"},{"why":"Shows that stochastic resetting can optimize hitting and search, the background for the paper's optimal-reset and cover-time applications.","marker":"[29]"},{"why":"Supplies the DBLP collaboration dataset used in the node-ranking comparison between hypergraph-preserving and clique-expansion walks.","marker":"[36]"}],"fun_headline_variants":["Resetting hypergraph walks: exact spectral dictionary","Eigenvalues of reset walk: scaled plain walk's","Optimal reset probability from first-passage variance","Hypergraph reset walks without graph conversion","Exact resetting statistics on hypergraphs via spectra"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The optimal-reset condition $z^2(\\gamma^*)=1+1/\\langle T_{ij}(\\gamma^*)\\rangle$ is derived only when the reset node equals the starting node and $i\\neq j$, but the paper states it as a general condition for arbitrary reset nodes or reset-candidate subsets; that unproved generality is the load-bearing premise.","fun_headline_variants_meta":{"raw":{"variants":["Resetting hypergraph walks: exact spectral dictionary","Eigenvalues of reset walk: scaled plain walk's","Optimal reset probability from first-passage variance","Hypergraph reset walks without graph conversion","Exact resetting statistics on hypergraphs via spectra"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000674,"raw_usage":{"total_tokens":3151,"prompt_tokens":1112,"completion_tokens":2039,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":728,"completion_tokens_details":{"reasoning_tokens":1967}},"tokens_in":728,"tokens_out":2039,"duration_ms":15992,"temperature":1.0,"reasoning_tokens":1967,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T23:35:53.529092+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a small connected hypergraph (for instance three nodes in a single three-node hyperedge), set the reset node $r$ different from the starting node $i$, and compute $\\langle T_{ij}(r,\\gamma)\\rangle$ exactly by solving the $n\\times n$ linear system for many $\\gamma$. If the minimizer does not satisfy $z^2(\\gamma^*)=1+1/\\langle T_{ij}(r,\\gamma^*)\\rangle$, the condition as stated is false; if it holds across all such small cases, that supports extending the proof to $r\\neq i$.","supporting_citations":[{"cited_title":"Random walks on hypergraphs,","cited_arxiv_id":null,"evidence_quote":"Defines the generalized adjacency and transition matrix on hypergraphs ($K_{ij}$ and $W$) whose spectral decomposition the paper uses as the reset-free base."},{"cited_title":"Random walks on networks with stochastic resetting,","cited_arxiv_id":null,"evidence_quote":"Introduces stochastic resetting on network random walks via $\\Pi(r,\\gamma)=(1-\\gamma)W+\\gamma\\Theta(r)$, the construction this paper transplants to hypergraphs."},{"cited_title":"Random walks on complex networks,","cited_arxiv_id":null,"evidence_quote":"Supplies the spectral and Laplace-transform derivation of occupation probability, stationary distribution, and mean first passage time that Theorems 1–3 adapt."},{"cited_title":"Random walks on hypergraphs with edge-dependent vertex weights,","cited_arxiv_id":null,"evidence_quote":"Establishes that hypergraph random walks need edge-dependent vertex weights, motivating the paper's choice of $K_{ij}$ rather than simple adjacency weights."},{"cited_title":"Optimal stochastic restart renders fluctuations in first passage times universal,","cited_arxiv_id":null,"evidence_quote":"Provides the optimal-restart result that ties optimality to the relative fluctuation of first passage times, the origin of the coefficient-of-variation condition."},{"cited_title":"Diffusion with stochastic resetting,","cited_arxiv_id":null,"evidence_quote":"Shows that stochastic resetting can optimize hitting and search, the background for the paper's optimal-reset and cover-time applications."},{"cited_title":"Dblp: some lessons learned,","cited_arxiv_id":null,"evidence_quote":"Supplies the DBLP collaboration dataset used in the node-ranking comparison between hypergraph-preserving and clique-expansion walks."}],"review_version":1}