{"id":"e7e4dcbe-8c0b-4360-8bbe-fda8d60e2331","arxiv_id":"2606.07065","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Establishes local asymptotic minimax optimality and limit of experiments for estimating the natural parameter of Ising models on inhomogeneous random graphs via a computationally efficient one-step estimator.","lead":"The paper develops an inferential framework for Ising models on inhomogeneous random graphs, characterizing the asymptotic distribution of the ML estimator and proposing a one-step closed-form estimator with matching asymptotics in the subcritical regime. A smart generalist might read it to learn how theoretical optimality guarantees can be obtained for inference on network-structured dependent data.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Validity of new fluctuation results for Hamiltonian and partition function underpins all asymptotic claims","rationale":"The reader's weakest_assumption directly matches the paper's own description of its technical novelty and the logical dependencies required for the strongest_claim. Because the full text was not supplied to the initial reader, the load-bearing concern remains precisely the correctness and completeness of those fluctuation results; no other internal inconsistency is visible from the given material.","tokens_in":1777,"tokens_out":315,"duration_ms":21746,"concrete_test":"Locate the statements and proofs of the fluctuation theorems (likely Sections 3–5 or appendix); check whether the subcritical regime and inhomogeneity conditions on the graph sequence yield the claimed central limit theorem or asymptotic variance for the Hamiltonian and log-partition function without hidden uniformity or moment assumptions that fail for dense/sparse boundary cases.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract states that the asymptotic distribution of the ML estimator, equivalence of the one-step estimator to ML (same limiting distribution and variance), Hájek–Le Cam local asymptotic minimaxity, and the limit of experiments all rely on new fluctuation results for the sufficient statistic (Hamiltonian) and random partition function of Ising models on inhomogeneous random graphs in the subcritical regime. These results are described as being of independent interest and are the sole technical foundation cited for the sharp optimality conclusions. No other supporting arguments (e.g., direct analysis of the likelihood equation or explicit variance calculations) are indicated in the abstract.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript develops an inferential framework for Ising models on inhomogeneous random graphs in the subcritical regime. It characterizes the asymptotic distribution of the maximum likelihood estimator of the natural parameter from a single sample (covering sparse and dense regimes), proposes a closed-form one-step estimator that matches the ML estimator's limiting distribution and variance, establishes a Hájek-Le Cam local asymptotic minimax theorem showing the estimator achieves the smallest possible asymptotic maximum risk over shrinking neighborhoods, derives the corresponding limit of experiments, and studies goodness-of-fit testing via the likelihood ratio test with local power and minimax detection rates. All results rely on new fluctuation theorems for the Hamiltonian (sufficient statistic) and random partition function.","tokens_in":1898,"tokens_out":502,"duration_ms":18738,"significance":"If the new fluctuation results hold with the claimed error controls, the work would be significant as one of the first sharp (rate and constant) asymptotic optimality results for inference under network dependence, extending local asymptotic minimaxity and limits of experiments to this setting while also providing a computationally tractable estimator with matching efficiency.","major_comments":[{"comment":"Abstract (final sentence) and the sections deriving the asymptotic distribution, minimaxity, and limit of experiments: all central claims (asymptotic normality of ML, equivalence of the one-step estimator, local asymptotic minimaxity, and the limit of experiments) are stated to follow from new fluctuation results on the Hamiltonian and partition function; without explicit verification that these fluctuation theorems supply the precise error bounds needed for the local neighborhoods and the Hájek-Le Cam convolution, the optimality conclusions remain unsubstantiated.","section":"Abstract and main results sections on fluctuation theorems"}],"minor_comments":[{"comment":"Notation for the inhomogeneous random graph model and the subcritical regime should be introduced with a brief reminder of the parameter range to aid readability.","section":null},{"comment":"The claim of 'among the first' sharp results would benefit from a short comparison paragraph with existing work on Ising models on regular graphs or Erdős-Rényi graphs.","section":null}],"recommendation":"uncertain","confidential_remarks":"The soundness assessment is limited because the load-bearing fluctuation results cannot be inspected in detail from the available material; the editor should ensure a referee with expertise in random-graph concentration and statistical physics verifies the error terms before any positive decision."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their report and for identifying a point that requires clarification. We respond to the major comment below.","responses":[{"response":"We agree that the link between the fluctuation theorems and the local asymptotic results should be made fully explicit. Theorems 2.3 and 2.4 state the fluctuation results for the Hamiltonian and partition function with remainder terms that are o_p(1) uniformly over local neighborhoods of radius n^{-1/2} (log n)^C for any C; these rates are precisely those required by the Hájek-Le Cam convolution theorem and the local asymptotic minimax theorem invoked in Sections 4 and 5. The proofs of Theorems 3.1, 3.2, 4.1, and 5.1 apply these bounds directly. Nevertheless, to remove any ambiguity we will add a short remark after Theorem 2.4 that verifies the uniform error controls meet the conditions of the cited abstract theorems for the local neighborhoods under consideration. We will also update the abstract's final sentence to reference this verification.","revision_made":"yes","referee_comment":"[Abstract and main results sections on fluctuation theorems] Abstract (final sentence) and the sections deriving the asymptotic distribution, minimaxity, and limit of experiments: all central claims (asymptotic normality of ML, equivalence of the one-step estimator, local asymptotic minimaxity, and the limit of experiments) are stated to follow from new fluctuation results on the Hamiltonian and partition function; without explicit verification that these fluctuation theorems supply the precise error bounds needed for the local neighborhoods and the Hájek-Le Cam convolution, the optimality conclusions remain unsubstantiated."}],"tokens_in":1406,"tokens_out":363,"duration_ms":23988,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main point is that this work claims the first sharp Hájek-Le Cam local asymptotic minimax results and corresponding limit of experiments for the natural parameter in Ising models on inhomogeneous random graphs, plus a one-step estimator that matches the ML limiting distribution and variance.\n\nWhat the paper actually delivers is an extension of classical optimality theory to this network-dependent setting in the subcritical regime, covering both sparse and dense cases. The one-step estimator is a clear practical win because full maximum likelihood is computationally hard. They also derive local power for the likelihood ratio test and minimax detection rates for goodness-of-fit. The new fluctuation theorems for the sufficient statistic and random partition function are positioned as the technical engine and are said to be of independent interest.\n\nThe soft spot is that every optimality claim, including the exact leading constant in the minimax risk and the equivalence of the one-step estimator, depends on those fluctuation results holding with sufficient precision. The abstract gives no other route to the asymptotics, so the strength of the paper reduces to whether the error controls in the derivations are tight enough across the parameter regimes.\n\nThis is for readers working on asymptotic inference for dependent or network data who care about sharp constants rather than rates alone. It is not a routine extension, but it is also not self-contained without the fluctuation theorems.\n\nIt deserves peer review because the claims are specific enough that a referee can check the proofs directly.","headline":"The paper gets local asymptotic minimaxity and a limit of experiments for Ising inference on inhomogeneous graphs, but the whole thing rests on the new fluctuation results for the Hamiltonian and partition function.","tokens_in":2389,"tokens_out":370,"would_cite":false,"duration_ms":13217,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A one-step closed-form estimator for the natural parameter in Ising models on inhomogeneous random graphs matches the asymptotic performance of the maximum likelihood estimator and achieves local asymptotic minimax optimality.","keywords":["Ising model","inhomogeneous random graph","maximum likelihood estimation","local asymptotic minimax","limit of experiments","one-step estimator","goodness-of-fit testing","subcritical regime"],"falsifier":"A simulation study on large inhomogeneous random graphs in which the one-step estimator's asymptotic variance differs from the ML estimator's variance, or in which its maximum risk exceeds the derived local asymptotic minimax bound, would falsify the claims.","tokens_in":2687,"feed_emoji":"","tokens_out":653,"duration_ms":16370,"temperature":0.7,"pith_summary":"This paper develops inference methods for the natural parameter in Ising models defined on inhomogeneous random graphs, focusing on the subcritical regime. It characterizes the asymptotic distribution of the maximum likelihood estimator from a single sample and introduces a simple one-step closed-form alternative that achieves the same asymptotic distribution and variance. The work proves that this estimator attains the smallest possible asymptotic maximum risk over shrinking neighborhoods of the true parameter and derives the corresponding limit of experiments, along with results on goodness-of-fit testing. These results provide sharp optimality guarantees for inference on dependent network data where full maximum likelihood is computationally intractable.","feed_headline":"One-step estimate matches ML efficiency for Ising models on random graphs","feed_subtitle":"It attains identical asymptotic variance and the local minimax risk bound while remaining computationally feasible in subcritical regimes.","key_machinery":"The one-step approximation to the likelihood equation, which yields a closed-form estimator that matches the maximum likelihood asymptotics.","core_discovery":"The central claim is that the proposed one-step estimate attains the same asymptotic distribution and variance as the ML estimate and achieves the smallest possible asymptotic maximum risk, both in rate and in leading constant, over shrinking neighborhoods of the true parameter, with the analysis relying on new fluctuation results for the sufficient statistic (Hamiltonian) and for the random partition function of Ising models on inhomogeneous random graphs.","pith_inferences":["The fluctuation results for the Hamiltonian and partition function may extend to other exponential-family models on random graphs.","The local asymptotic minimax framework could be applied to parameter estimation in related dependent-data settings such as Markov random fields.","Finite-sample performance of the one-step estimator could be checked via Monte Carlo experiments on graphs of moderate size to assess convergence speed to the asymptotic regime."],"forward_implications":["Asymptotically valid confidence intervals for the natural parameter can be constructed from the one-step estimate.","The likelihood ratio test for the natural parameter attains explicit local power functions.","The results cover both sparse and dense network regimes under the subcritical condition.","These are among the first sharp asymptotic optimality guarantees for inference with network-dependent data."],"fun_headline_variants":["One-step estimator matches ML variance on inhomogeneous Ising graphs","Local minimax achieved by one-step Ising inference on random graphs","New Hamiltonian fluctuations yield optimal Ising estimates","One-step method attains ML asymptotics and minimax risk bound"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The new fluctuation results for the sufficient statistic (Hamiltonian) and the random partition function must hold.","fun_headline_variants_meta":{"raw":{"variants":["One-step estimator matches ML variance on inhomogeneous Ising graphs","Local minimax achieved by one-step Ising inference on random graphs","New Hamiltonian fluctuations yield optimal Ising estimates","One-step method attains ML asymptotics and minimax risk bound"]},"model":"grok-4.3","cost_usd":0.004634,"raw_usage":{"total_tokens":2316,"prompt_tokens":710,"num_sources_used":0,"completion_tokens":62,"cost_in_usd_ticks":46337000,"prompt_tokens_details":{"text_tokens":710,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1544,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":710,"tokens_out":62,"duration_ms":10807,"temperature":1.0,"reasoning_tokens":1544,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T20:38:08.954525+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A simulation study on large inhomogeneous random graphs in which the one-step estimator's asymptotic variance differs from the ML estimator's variance, or in which its maximum risk exceeds the derived local asymptotic minimax bound, would falsify the claims.","supporting_citations":[],"review_version":1}