{"id":"ed89a87b-544f-4e76-883d-fd8af88195cf","arxiv_id":"2507.15988","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"The authors claim that hypercubes, cycles, tori, and lattices can be convolved into smaller weighted graphs while preserving continuous-time quantum walk dynamics.","lead":"This paper proposes a method for compressing symmetric graphs (hypercubes, cycles, tori, lattices) into smaller weighted graphs that are claimed to preserve quantum walk dynamics. If the method worked, it could save qubits when simulating quantum walks on photonic and NISQ hardware.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Sec. 3.2's lattice-convolution rule is an unproved classical-probability heuristic; the paper's broad claim that quantum walk dynamics are preserved is therefore not established, and the two numerical examples cannot carry the claimed generality.","rationale":"The reader's weakest-assumption analysis identifies the same load-bearing step: the lattice-convolution heuristic in Sec. 3.2 is the only constructive procedure connecting the original lattices to the 'ultimate graphs' whose dynamics the paper claims are identical. My stress-test agrees that this heuristic is not derived from quantum mechanics, that the paper explicitly labels it as heuristic, and that the numerical experiments are too limited to establish a general theorem. The hypercube and cycle reductions are genuinely exact quotients, and the torus-to-lattice step can be repaired using the Cartesian-product formula, so the central problem is not the existence of exact reductions but the unjustified lattice step. Because the claimed generality of 'preserves QWs dynamics' depends on that step, the reader's REJECT verdict is supported. I found no additional concern that would change the verdict; the concern is the same one, and it is load-bearing enough on its own.","tokens_in":10206,"tokens_out":4356,"duration_ms":53466,"concrete_test":"Implement exact CTQW evolution with a sink via Eqs. (2) and (14) for: (i) a uniform n x n lattice (n = 4, 5, 6); (ii) the 'ultimate graph' produced by the Sec. 3.2 algorithm; and (iii) the correct symmetric-subspace quotient obtained by grouping vertices equivalent under graph automorphisms fixing the chosen start and target, with edge weights read off from the Hamiltonian's block structure. Compare sink hitting probability p_sink(t) for the same start/target pair, Gamma = 1, and threshold (15). If the Sec. 3.2 curve deviates from (i) by more than the integrator tolerance, the central claim fails. As an analytical check, derive the quotient weights for the n = 4 lattice: they are determined by counts of edges between distance shells, not by sqrt(p1^2 + p2^2), so exact equality would require an additional identity that the paper neither states nor proves.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in Sec. 4.2 is that graph convolution preserves continuous-time quantum walk dynamics for both lattices and toruses. For the hypercube and cycle, the reductions are exact symmetric-subspace quotients (the hypercube weights sqrt(i(D+1-i)) correspond to the correct quotient Hamiltonian). But the lattice reduction that produces the 'ultimate graph' is introduced in Sec. 3.2 as a heuristic: discard the lower triangular, merge diagonal rectangles into lines, and assign edge values by sqrt(p_side1^2 + p_side2^2), where p_side1 and p_side2 are transition probabilities. This rule is not derived from Eq. (2), from the Hamiltonian, or from the GKSL equation (14). In a CTQW, edge weights are Hamiltonian couplings (amplitudes), and an exact reduction must come from restricting the Hamiltonian to a symmetric subspace or from an equitable partition; a Euclidean norm of classical probabilities has no such derivation. The paper itself says in Sec. 4.3 that the equiprobability grouping can be used 'for verification, but not for derivation,' and the eigenvalue observations in Sec. 5 are explicitly unproved. The numerical support in Fig. 6 consists of one 6x6 torus and one 4x4 lattice, each with a single start/target pair, with no code, data, error bars, or larger tests. Thus the claimed general preservation of QW dynamics rests on an unproved postulate plus two numerical coincidences.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes 'graph convolution' reductions for continuous-time quantum walks on hypercubes, cycles, toruses, and lattices. The stated central claim is that the reduced (smaller) graphs preserve the quantum walk dynamics, so that a larger graph can be replaced by a smaller weighted graph for photonic waveguide simulation, saving qubits. The exact reductions for hypercubes and cycles are based on symmetric-subspace quotients, while the lattice reduction relies on a heuristic algorithm in Sec. 3.2. Numerical simulations of hitting probabilities are presented in Sec. 4, and Sec. 5 proposes an eigenvalue criterion for minimal graph size.","tokens_in":10443,"tokens_out":6626,"duration_ms":68109,"significance":"If the central claim were established, the resource reduction would be practically relevant: a D-dimensional hypercube would require only D+1 rather than 2^D waveguide modes, and analogous reductions for cycles and lattices would simplify continuous-time quantum walk simulations. The hypercube and single-cycle mappings are indeed exact quotient constructions; the coupling coefficients in Eq. (3) coincide with the known equitable partition of the hypercube, and the line graph obtained from C_8 has exactly the unique eigenvalues of C_8 as shown in Sec. 5. However, the manuscript's extension to toruses and lattices is not established: the adjacency-matrix relation Eq. (8) is incorrect, the Sec. 3.2 lattice rule is an unproved postulate, and the numerical support consists of two examples with a single start-target pair each. The manuscript also provides no code or data files. As written, the paper is a useful observation about exact quotient reductions for hypercubes and cycles combined with unsupported generalizations.","major_comments":[{"comment":"Equation (8) writes the adjacency matrix of the D=2 hypercycle as A_Circle^k ⊗ A_Circle^k, but the torus C_k × C_k is the Cartesian product, whose adjacency is A(G×H) = A(C)⊗I + I⊗A(C), as correctly stated in Eq. (10). Since the manuscript uses Eq. (10) two lines later to derive the mapped Hamiltonian (12), Eq. (8) is internally inconsistent: the graph whose adjacency is a Kronecker product is not the graph whose dynamics are simulated. This invalidates the derivation of the lattice mapping as a representation of the torus.","section":"Sec. 3.1.1, Eq. (8)"},{"comment":"The list 'eigenvalues original = [2, -2, 1.4142, 1.4142, 0, 0, -1.4142, -1.4142]' is the spectrum of the cycle C_8, not of the torus C_8 × C_8. The torus has 64 vertices and its eigenvalues are λ_i + λ_j, where λ_i and λ_j are the eigenvalues of C_8; it has many more distinct values (e.g., 4, 2±√2, 2√2, ...). Consequently, the observation that 'the eigenvalues of the convoluted graph are only unique eigenvalues of the original one' is false for the torus as stated, and the claimed criterion for minimal graph size is not supported.","section":"Sec. 5, Eq. (20)"},{"comment":"The lattice convolution rule is introduced as a heuristic with no derivation. Step 3 assigns merged edge weights by sqrt(p_side1^2 + p_side2^2), where p_side1 and p_side2 are classical transition probabilities. In the continuous-time quantum walk, the Hamiltonian edge weights are amplitudes/couplings, not probabilities, and the classical transition matrix does not share the spectral structure of the Hamiltonian in the way required by the paper; a Euclidean norm over classical probabilities has no known quantum-mechanical justification. Since the 'ultimate graph' produced by this rule is used in Figs. 5 and 6 to support the central claim that quantum walk dynamics are preserved, this is a load-bearing unproved postulate rather than a derived reduction.","section":"Sec. 3.2, lattice convolution algorithm"},{"comment":"The numerical evidence for the preservation claim consists of one torus with D=2, k=6 and one 4×4 lattice, each tested with a single starting and target pair. No error bars, statistics over random pairs, code, or data files are provided, and the hitting-probability curves are compared only visually. Two examples do not establish the general statement that 'the proposed method of graph convolution preserves QWs dynamics; it is valid for both lattice and torus.'","section":"Sec. 4.2, Fig. 6"},{"comment":"The criterion for the 'smallest graph' is nearly tautological: the convoluted graph is said to be the smallest when its node count equals the number of equiprobable groups, and the existence of such groups is inferred from the outcome probabilities of the quantum walk on the original graph. Because the authors state that this technique can be used 'for verification, but not for derivation of the convoluted graph,' it does not provide a constructive or falsifiable characterization of minimality.","section":"Sec. 4.3, Fig. 8(b)"}],"minor_comments":[{"comment":"Equation (6) contains a mismatched parenthesis: the line reads '+ √6 (|3⟩ ⟨2| + (|4⟩ ⟨3| + H.C.', which should be closed and grouped consistently.","section":"Eq. (6)"},{"comment":"The symbol Hhc→line is used for both the hypercube Hamiltonian in Eq. (3) and the hypercycle Hamiltonian in Eq. (7), which is confusing because the graphs and couplings are different.","section":"Secs. 2.2 and 3.1"},{"comment":"The text says 300 simulations were performed on both the original and convoluted graphs, but Fig. 7 plots only the fastest result of each walk; the relationship between the 300 runs and the plotted points, including the treatment of failed walks, should be specified.","section":"Sec. 4.2, Fig. 7"},{"comment":"The two observations at the end of Sec. 5 are presented without proof, and the authors acknowledge this; they should be explicitly labeled as conjectures, since they are used as a criterion in Sec. 4.3.","section":"Sec. 5"},{"comment":"The Data Availability statement says that all data are available within the article, but no dataset, simulation code, or numerical parameters are provided; this makes the numerical results irreproducible as presented.","section":"Data Availability"}],"recommendation":"reject","confidential_remarks":"The paper's stated scope exceeds what is demonstrated. The exact hypercube and cycle reductions are sound and could form the core of a shorter paper, but the torus/lattice claims rest on a false adjacency formula, a false eigenvalue list for the torus, and an unproved heuristic. These are not presentation-level issues; they concern the main claim. I would advise the editor that a rejection is appropriate, though the authors might be encouraged to resubmit a revised manuscript restricted to the exact quotients after correcting Eqs. (8) and (20)."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful core here is small: the hypercube-to-weighted-line quotient (couplings sqrt(i(D+1-i))) and the even-cycle-to-weighted-line quotient (endpoint coupling sqrt(2)) are correct. They reproduce known symmetric-subspace reductions, and the numerical curves for the tested cases do coincide. The authors are also honest that the lattice rule is a heuristic and that their eigenvalue observations are unproved. That honesty is real credit.\n\nBut the paper's central claim goes much further. The torus section contains two outright errors: Eq. (8) writes the torus adjacency as a Kronecker product A_C ⊗ A_C, which contradicts the Cartesian-product formula Eq. (10) they use two lines later; and Eq. (20) lists eight eigenvalues as the spectrum of the k=8 torus, which has 64 vertices. These are not typos in the margins—they sit in the derivation of the claimed mapping.\n\nThe lattice-convolution rule is the load-bearing piece for the claimed generality, and it is not derived. Merging probabilities via sqrt(p_side1^2 + p_side2^2) has no basis in the Schrödinger equation or in the classical transition-matrix equation. In a CTQW the edge weights are Hamiltonian couplings, and an exact reduction should come from an equitable partition or a symmetric-subspace restriction. The authors themselves say in Sec. 4.3 that the equiprobability grouping is for verification, not derivation, which undercuts their own conclusion in Sec. 4.2 that the method preserves QW dynamics for both lattices and tori. The numerical support is one 6x6 torus and one 4x4 lattice, each with a single start/target pair, with no code, data, or error bars. That cannot carry the claimed generality.\n\nThe eigenvalue-minimality observations in Sec. 5 are explicitly unproved and are false in general, so they should not be stated as observations. The correct statement is the familiar one: an equitable partition can reduce the graph, and the reduced graph's spectrum is a subset of the original's—but that is standard spectral graph theory, not a new criterion.\n\nWho gets value here? Someone wanting a compact review of hypercube and cycle symmetric-subspace reductions might find the early sections useful. The lattice convolution, as presented, is not reliable enough to use as a method.\n\nMy recommendation: send it to peer review, but the referee should require the Eq. (8) and Eq. (20) errors to be fixed, and the lattice heuristic to be either derived or explicitly downgraded to a conjecture with a much narrower claim. If the authors can do that, the paper could become a modest but useful note. As written, it overreaches.","headline":"The known exact reductions are correct but standard, the torus section has two clear errors, and the lattice-convolution claim rests on an unproved heuristic; the paper is worth a referee's time but not acceptance as written.","tokens_in":11074,"tokens_out":2388,"would_cite":false,"duration_ms":27645,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["81P68","05C50","05C81"],"pacs":[],"model":"deepseek-v4-flash","headline":"High-dimensional graphs such as hypercubes, tori, and lattices can be replaced by much smaller weighted graphs—weighted lines or small lattices—without changing the continuous-time quantum walk dynamics at a chosen start and target node.","keywords":["quantum walks","graph convolution","hypercubes","hypercycles","torus","lattice graphs","photonic waveguide arrays","eigenvalue degeneracy"],"falsifier":"Apply the three-step lattice convolution to a $5\\times 5$ or $6\\times 6$ rectangular lattice (or to a torus with $k=8$) using the same sink rate $\\Gamma=1$ and threshold (15), and compare the full time trace of target-node hitting probability with the original graph beyond the time window shown in Fig. 6; any persistent divergence before the threshold falsifies the preservation claim. A sharper test is to check whether the convoluted graph's set of unique eigenvalues always equals the original's—a counterexample lattice whose convoluted graph has an eigenvalue outside the original spectrum would break the stated criterion.","tokens_in":1909,"feed_emoji":"🧩","tokens_out":2213,"duration_ms":71075,"temperature":0.7,"pith_summary":"The paper tries to establish that certain high-dimensional graphs can be replaced by much smaller weighted graphs without changing the continuous-time quantum walk dynamics observed from a chosen start node to a chosen target node. The graphs in question are hypercubes, hypercycles (tori), and 2D lattices; the replacements are weighted line graphs, weighted lattices, and \"ultimate\" graphs obtained by a simple convolution recipe. If this is true, quantum-walk simulations on photonic waveguide arrays would need far fewer waveguides and, in circuit implementations, far fewer qubits, because only the distinct probability groups of the original graph need to be represented. The paper supports the claim with numerical hitting-probability curves that coincide for original and convoluted graphs, and with an eigenvalue observation: the convoluted graph's spectrum is the set of unique eigenvalues of the original.","feed_headline":"Graph shrinking preserves quantum-walk dynamics, simulations show","feed_subtitle":"Replacing tori and lattices with weighted mini-graphs keeps target-hit probabilities identical, cutting qubits for photonic quantum walks.","key_machinery":"The central object is the weighted line or lattice obtained by lumping vertices that always carry equal probability—vertices with the same Hamming distance in a hypercube, or symmetric positions in a cycle or torus. The mechanism is the Cartesian-product identity $A(G\\times H)=A(G)\\otimes I + I\\otimes A(H)$ for adjacency matrices, which lets a torus (Cartesian product of two cycles) be replaced by the Cartesian product of two weighted lines. For lattices, the load-bearing step is the heuristic rule that merges diagonal rectangles into lines and sets the new edge probability to $\\sqrt{p_{\\mathrm{side1}}^2+p_{\\mathrm{side2}}^2}$. The supporting spectral claim is that duplicated eigenvalues in the original adjacency matrix signal equiprobable vertex groups, so the convoluted graph's eigenvalues are the unique eigenvalues of the original graph.","core_discovery":"On the paper's own terms, the central discovery is a graph-convolution method that preserves the dynamics of continuous-time quantum walks. For hypercubes, grouping vertices by Hamming distance gives a weighted line with couplings $\\beta_{i,i+1}=\\sqrt{i(D+1-i)}$; for cycles, a similar weighted line; for tori, a Cartesian product of these lines yields a weighted lattice; and a further three-step heuristic—dropping the lower triangle, merging diagonal rectangles into lines, and reassigning edge probabilities with $\\sqrt{p_{\\mathrm{side1}}^2+p_{\\mathrm{side2}}^2}$—produces the smallest \"ultimate\" graph. Numerical solutions of the GKSL master equation with a sink node show that the target-node hitting probability as a function of time is the same on the original torus and on its two-step convolution, and on a $4\\times 4$ lattice and its convoluted graph. The paper concludes that the proposed method preserves quantum-walk dynamics for both lattices and tori, and it also observes, without proving, that the eigenvalues of the convoluted graph are exactly the unique eigenvalues of the original graph, with the minimal number of nodes equal to the number of unique eigenvalues.","pith_inferences":["If the eigenvalue criterion generalizes, it offers a purely spectral test for when a graph can be replaced by a smaller one: the number of distinct eigenvalues bounds the minimal representation size, independent of the walk's starting point; this is a stronger statement than the paper's numerical examples show.","The Euclidean-norm merging rule resembles the composition rule for independent paths in a classical diffusion picture, but it has no obvious quantum derivation; a natural test is to derive it from a requirement that the lumped graph's generator has the same eigenvalues on the relevant subspace, which would turn the heuristic into a theorem.","Because the method works by identifying vertex groups that remain equiprobable for all times, it is closely related to Markov-chain lumpability and could be tested on other symmetric graph families such as complete bipartite graphs or Johnson graphs, where the paper does not apply it.","The sink-based detection used here means the preserved quantity is the hitting probability at a single target, not the full unitary evolution; a reader should not infer that convoluted and original graphs are unitarily equivalent, only that the chosen detection statistics match."],"forward_implications":["Quantum-walk circuits for hypercubes and tori can use only the number of qubits needed for the much smaller convoluted graph, because representing equiprobable vertex groups is sufficient.","For photonic waveguide implementations, a torus or lattice can be realized as a one-dimensional weighted chain or small lattice with tailored coupling coefficients, avoiding the need for high-dimensional physical geometry.","Classical and quantum hitting-time comparisons on the convoluted lattice reproduce the known pattern: classical walks win at short distances (d < 3), while quantum walks win for distances d at least 3.","Any graph that is a Cartesian or Kronecker product of line-like graphs can in principle be reduced by the same two-step convolution procedure.","The minimal faithful convolution size of a graph is the number of unique eigenvalues of its adjacency matrix, offering a spectral bound on how small the replacement graph can be."],"supporting_citations":[{"why":"Supplies the sink-node detection scheme, the detection threshold $p_{\\mathrm{th}}=1/\\log(n)$, and the classical-quantum random-walk comparison method used in all numerical experiments.","marker":"[22]"},{"why":"Defines hitting time for quantum walks on the hypercube, the prototype structure whose convolution the paper generalizes.","marker":"[16]"},{"why":"Provides the continuous-time quantum walk analysis on a cycle graph, the basis for the hypercycle convolution step.","marker":"[15]"},{"why":"States the qubit cost of quantum-walk simulation (2^n vertices for n qubits), which motivates the graph-reduction goal.","marker":"[25]"},{"why":"Demonstrates mapping high-dimensional network dynamics onto one-dimensional waveguide chains, the photonic implementation context the paper aims to serve.","marker":"[32]"},{"why":"Reports experimental realization of quantum walks in waveguide lattices, supporting the physical feasibility of the proposed weighted chains.","marker":"[29]"}],"fun_headline_variants":["Mini-graphs replicate quantum-walk hitting probabilities","Graph convolution preserves quantum walks, cuts qubits","Shrunk graphs keep quantum-walk dynamics intact","Quantum walk simulation on compressed graphs saves qubits","Tori and lattices compressed without altering quantum walks"],"cache_read_input_tokens":13056,"weakest_assumption_plain":"The load-bearing premise is the lattice-convolution rule—discard the lower triangle, merge diagonal rectangles into lines, and set new edge probabilities by $\\sqrt{p_{\\mathrm{side1}}^2+p_{\\mathrm{side2}}^2}$—which is asserted as a heuristic rather than derived from the Schrödinger equation or the transition-matrix equation; if this rule is not generally valid, the claimed dynamics preservation for lattices and tori does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Mini-graphs replicate quantum-walk hitting probabilities","Graph convolution preserves quantum walks, cuts qubits","Shrunk graphs keep quantum-walk dynamics intact","Quantum walk simulation on compressed graphs saves qubits","Tori and lattices compressed without altering quantum walks"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000387,"raw_usage":{"total_tokens":2016,"prompt_tokens":888,"completion_tokens":1128,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":504,"completion_tokens_details":{"reasoning_tokens":1054}},"tokens_in":504,"tokens_out":1128,"duration_ms":9662,"temperature":1.0,"reasoning_tokens":1054,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:22:29.509459+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply the three-step lattice convolution to a $5\\times 5$ or $6\\times 6$ rectangular lattice (or to a torus with $k=8$) using the same sink rate $\\Gamma=1$ and threshold (15), and compare the full time trace of target-node hitting probability with the original graph beyond the time window shown in Fig. 6; any persistent divergence before the threshold falsifies the preservation claim. A sharper test is to check whether the convoluted graph's set of unique eigenvalues always equals the original's—a counterexample lattice whose convoluted graph has an eigenvalue outside the original spectrum would break the stated criterion.","supporting_citations":[{"cited_title":"A., Fedichkin, L","cited_arxiv_id":null,"evidence_quote":"Supplies the sink-node detection scheme, the detection threshold $p_{\\mathrm{th}}=1/\\log(n)$, and the classical-quantum random-walk comparison method used in all numerical experiments."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines hitting time for quantum walks on the hypercube, the prototype structure whose convolution the paper generalizes."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the continuous-time quantum walk analysis on a cycle graph, the basis for the hypercycle convolution step."},{"cited_title":"L., Wang, J","cited_arxiv_id":null,"evidence_quote":"States the qubit cost of quantum-walk simulation (2^n vertices for n qubits), which motivates the graph-reduction goal."},{"cited_title":"J., Wang, K., Dovgiy, A","cited_arxiv_id":null,"evidence_quote":"Demonstrates mapping high-dimensional network dynamics onto one-dimensional waveguide chains, the photonic implementation context the paper aims to serve."},{"cited_title":"B., Lahini, Y., Pozzi, F., Sorel, M., Morandotti, R., Silberberg, Y","cited_arxiv_id":null,"evidence_quote":"Reports experimental realization of quantum walks in waveguide lattices, supporting the physical feasibility of the proposed weighted chains."}],"review_version":1}