{"id":"3ff6ffca-e590-4fbd-a358-6db90ab420a2","arxiv_id":"2605.13690","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":8.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Hypergraph neural networks obey a strict expressivity hierarchy indexed by hypertree width, creating a Width Wall that no fixed-depth model, hidden dimension, or training procedure can cross for wider patterns.","lead":"The paper shows that hypergraph neural networks are limited by a Width Wall based on the hypertree width of patterns they can detect and count via homomorphism densities. This limit explains failures of current models on complex data and points toward density-aware architectures that could go beyond it.","discovery_kind":"paradigm_shift","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"Reader correctly flags that abstract-only review leaves soundness dependent on the detailed proofs. No concrete flaw in the high-level argument is visible from the supplied material, so the UNVERDICTED status is appropriate pending full verification of the hierarchy strictness.","tokens_in":1712,"tokens_out":223,"duration_ms":21061,"concrete_test":"Re-derive the main theorem (homomorphism densities generate all continuous invariants) from the homomorphism-count completeness result without using the invariant-approximation step; if the generation property still holds, the Width Wall construction is robust.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on homomorphism densities generating all continuous hypergraph invariants via classical completeness plus invariant approximation, then inducing a strict hypertree-width hierarchy. This is a standard route in expressivity theory; the abstract and described framework show no internal gap in the logic that would invalidate the Width Wall for fixed-depth HGNNs. The experimental suite is presented only as validation, not as the proof.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims that hypergraph expressivity is governed by homomorphism densities, which measure motif occurrences. Combining classical homomorphism-count completeness with invariant approximation, it shows these densities generate all continuous hypergraph invariants and induce a strict hierarchy indexed by hypertree width. This establishes the Width Wall: no hidden dimension, training procedure, or fixed-depth HGNN can represent invariants requiring wider patterns. The framework unifies 15 HGNN architectures, identifies information lost by clique expansion, and is validated experimentally on a node classification suite of real-world hypergraphs where the wall predicts failure of graph-reduction baselines.","tokens_in":1814,"tokens_out":401,"duration_ms":23323,"significance":"If the result holds, the work provides a fundamental, parameter-free characterization of HGNN limits via homomorphism densities and hypertree width. The unification of 15 architectures and precise identification of clique-expansion losses are notable strengths, as is the motivation for density-aware models. The experimental validation on real hypergraphs adds practical relevance by showing when the theoretical wall manifests in application node classification.","major_comments":[{"comment":"The central derivation (combining classical homomorphism-count completeness with the invariant-approximation step) lacks explicit error bounds or full derivation details for the approximation. This is load-bearing for the claim that homomorphism densities generate all continuous invariants and that the resulting hypertree-width hierarchy is strict for every fixed-depth HGNN architecture.","section":"Main theorem on invariant generation and Width Wall"}],"minor_comments":[{"comment":"The abstract capitalizes 'APPLICATION NODE CLASSIFICATION SUITE' without defining it; clarify whether this is a named benchmark or rephrase for readability.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The low soundness score in the provided reader assessment stems directly from the missing approximation-error handling noted in the major comment; this is a fixable presentation gap rather than an internal inconsistency."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their positive assessment of the paper's contributions, including the unification of 15 HGNN architectures, the identification of clique-expansion losses, and the experimental validation on real-world hypergraphs. We address the major comment below and will revise the manuscript accordingly to strengthen the central derivation.","responses":[{"response":"We agree that the proof of the main theorem would benefit from expanded details. In the revised manuscript we will provide a complete, self-contained derivation that first recalls the classical homomorphism-count completeness theorem and then explicitly constructs the invariant-approximation step. We will derive explicit error bounds using the uniform continuity of continuous invariants on the compact space of hypergraphs with bounded degree and size; these bounds will quantify the approximation error in terms of the modulus of continuity and the sampling density of the homomorphism densities. The revised proof will thereby rigorously establish both that homomorphism densities generate all continuous hypergraph invariants and that the induced hypertree-width hierarchy is strict for any fixed-depth HGNN, independent of hidden dimension or training procedure.","revision_made":"yes","referee_comment":"[Main theorem on invariant generation and Width Wall] The central derivation (combining classical homomorphism-count completeness with the invariant-approximation step) lacks explicit error bounds or full derivation details for the approximation. This is load-bearing for the claim that homomorphism densities generate all continuous invariants and that the resulting hypertree-width hierarchy is strict for every fixed-depth HGNN architecture."}],"tokens_in":1307,"tokens_out":315,"duration_ms":20717,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing to know is that this work claims hypergraph neural networks are bounded by a Width Wall indexed by hypertree width: no fixed-depth model, regardless of hidden dimension or training, can capture invariants that require wider patterns. The argument rests on homomorphism densities generating all continuous hypergraph invariants and organizing them strictly by width, using classical completeness plus an invariant-approximation step.","headline":"The paper defines a Width Wall for HGNNs via a strict hypertree-width hierarchy on homomorphism densities, which unifies 15 architectures and flags what clique expansions lose.","tokens_in":2296,"tokens_out":156,"would_cite":true,"duration_ms":23105,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[],"headline":"Hypergraph expressivity hierarchy via homomorphism densities and hypertree width is orthogonal to RS forcing chain","alignment":"orthogonal","rationale":"The paper's core machinery (homomorphism densities t(F,A), pattern alignment scores ΘP, generalized hypertree width Wgh(F), strict hierarchy Hr ⊊ Hr+1, and Width Wall) operates entirely in combinatorial expressivity theory for HGNNs. It relies on classical homomorphism-count completeness and invariant approximation on symmetric tensors, with no reference to J-cost functions, ratio symmetry, φ-ladder spacings, 8-tick periodicity, or parameter-free derivations of physical constants. RS theorems such as reality_from_one_distinction, absolute_floor_closure, and those deriving J(x) = ½(x + x⁻¹) − 1 or φ from a single distinction have no bearing on or contradiction with this domain.","tokens_in":59242,"confidence":"high","tokens_out":196,"duration_ms":12687,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Hypergraph neural networks cannot represent invariants beyond a fixed hypertree width.","keywords":["hypergraph neural networks","expressivity hierarchy","homomorphism densities","hypertree width","width wall","higher-order interactions","motif counting","graph invariants"],"falsifier":"A fixed-depth HGNN that accurately distinguishes or classifies hypergraphs differing only by an invariant whose minimal representation requires a hypertree width strictly larger than the model's message-passing width.","tokens_in":2631,"feed_emoji":"","tokens_out":562,"duration_ms":29476,"temperature":0.7,"pith_summary":"The paper establishes that homomorphism densities, which count how often small structural motifs appear, generate all continuous hypergraph invariants. These densities sort the invariants into a strict hierarchy indexed by hypertree width. The resulting Width Wall sets a hard bound: no fixed-depth HGNN, regardless of hidden dimension or training procedure, can capture invariants that require wider patterns. The framework unifies analysis across fifteen existing architectures and shows exactly what information is lost when hypergraphs are reduced to ordinary graphs by clique expansion.","feed_headline":"Hypergraph networks hit a width wall on expressivity","feed_subtitle":"Homomorphism densities organize invariants into a strict hypertree-width hierarchy that bounds all fixed-depth HGNNs","key_machinery":"Homomorphism densities that quantify motif frequencies, arranged into a hierarchy by hypertree width.","core_discovery":"Homomorphism densities generate all continuous hypergraph invariants and organize them into a strict hierarchy indexed by hypertree width. This produces a Width Wall that no fixed-depth hypergraph neural network can cross, even with arbitrary hidden dimensions or training.","pith_inferences":["Node-classification performance on real hypergraphs should degrade precisely when the task requires patterns above the model's width.","Explicit computation of homomorphism densities offers one route to bypass the wall without increasing depth.","The same width-based separation may appear in other relational models that rely on local aggregation over higher-order relations."],"forward_implications":["Clique expansion necessarily discards information carried by patterns wider than the chosen clique size.","Fifteen distinct HGNN architectures receive a uniform characterization by the maximum hypertree width they can access.","Density-aware architectures can represent invariants lying above the width bound of ordinary message passing.","No increase in model size or training data overcomes the limit for invariants requiring wider hypertrees."],"fun_headline_variants":["HGNNs bounded by hypertree width hierarchy","Width Wall limits hypergraph network expressivity","Hypergraph invariants tied to hypertree width","Homomorphism densities define HGNN width limit"],"cache_read_input_tokens":64,"weakest_assumption_plain":"Homomorphism densities together with invariant approximation fully capture the continuous invariants relevant to HGNN expressivity and the resulting hierarchy is strict for every architecture considered.","fun_headline_variants_meta":{"raw":{"variants":["HGNNs bounded by hypertree width hierarchy","Width Wall limits hypergraph network expressivity","Hypergraph invariants tied to hypertree width","Homomorphism densities define HGNN width limit"]},"model":"grok-4.3","cost_usd":0.006222,"raw_usage":{"total_tokens":2824,"prompt_tokens":617,"num_sources_used":0,"completion_tokens":54,"cost_in_usd_ticks":62215500,"prompt_tokens_details":{"text_tokens":617,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2153,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":617,"tokens_out":54,"duration_ms":17481,"temperature":1.0,"reasoning_tokens":2153,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-14T20:22:38.410398+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A fixed-depth HGNN that accurately distinguishes or classifies hypergraphs differing only by an invariant whose minimal representation requires a hypertree width strictly larger than the model's message-passing width.","supporting_citations":[],"review_version":1}