{"id":"f670f97a-8bcc-4f00-8970-d9f2df0c9fe7","arxiv_id":"2504.12429","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For uniform and cyclic Markov chains, the optimal packing of neural attractor states is solved under a symmetry ansatz, yielding a simplex for uniform statistics and a generally non-square quadrilateral for four-state cycles.","lead":"This paper formulates how neural representations should pack attractor states to balance transition speed against noise robustness, and solves the resulting optimization for two toy transition structures: uniform and cyclic. A symmetry argument reduces the problem to a few distances, giving a simplex for the uniform case and a generally distorted quadrilateral for the four-state cyclic case.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The optimality claim rests on an unproven Purkiss-style symmetry assumption; the paper optimizes only symmetric configurations, so the reported packings are not established as global optima.","rationale":"The reader's weakest_assumption identifies exactly the gap I consider most load-bearing: the paper solves a symmetry-restricted problem and calls it the optimal packing. The Purkiss principle is cited but its hypotheses are not verified, and the authors explicitly disclaim the analysis. Since J is nonconvex, the symmetric solution could be a saddle or a locally optimal but globally suboptimal configuration. My proposed multistart full-coordinate optimization would provide direct evidence: if asymmetric configurations with lower J are found, the central optimality claim fails; if not, the symmetry assumption is supported. I additionally note a concrete feasibility risk in Section 4.2: the claimed large-bias optimum d=0, L>0 violates the Euclidean distance-matrix condition L <= sqrt(2)d for four points with all cycle edges d, so even granting the symmetry ansatz the reported solution may not correspond to any neural state configuration. This reinforces the need for the empirical check but does not change the appropriate disposition: the paper is transparent about its main gap and the toy results are plausible, so CONDITIONAL (the reader's verdict) remains appropriate until the optimality is either proved or the claims are softened.","tokens_in":12754,"tokens_out":18814,"duration_ms":197732,"concrete_test":"Use the released code or an independent optimizer to minimize Eq. 8 in full coordinates for M=4 uniform and M=4 cyclic, without imposing the symmetry ansatz: for a grid of (alpha,b) values and many random symmetry-breaking initializations, compare the lowest J found with the J of the reported all-equal-distance and D4 solutions (using the corrected Appendix B coefficient). Also verify that the reported (d,L) optima satisfy the Euclidean embedding condition L <= sqrt(2) d; if the reported large-bias solution d=0, L>0 appears, it is infeasible as a packing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3 states that for convex problems a unique global minimum inherits symmetries, but then says the present problem is 'probably not convex' and that a rigorous Purkiss-principle analysis is 'not pursued.' Despite this, Sections 4.1 and 4.2 optimize only the symmetry-constrained ansatz (all distances equal, or two distances d and L under D4), and the abstract and title present these as the optimal packings. Because Eq. 8 is nonconvex, a symmetric stationary point need not be the global minimum; symmetry-broken configurations in R^D could give lower J. The appendix also contains an arithmetic slip (Appendix B's e^{-d^2} coefficient should be M^2-3M+3, not M^2-3M+5), so even the symmetric solution is not numerically reliable as written. The availability of code makes the proposed numerical check straightforward.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript formulates a normative toy model for arranging M internal attractor states in a neural representation. Environment statistics are modeled by a Markov chain, internal transitions by distance-dependent rates, and encoding/decoding by a noisy symmetric channel; the objective J (Eq. 8) trades off the KL divergence between the desired and actual internal transition distributions against a firing-rate regularization. After a change of variables, J depends only on the mean location and pairwise distances (Eqs. 10–11), and the mean is set to zero. Relying on a Purkiss-style symmetry argument, the paper restricts to symmetric configurations and claims to solve two cases: uniform statistics, for which all distances are equal (an (M−1)-simplex), and M=4 cyclic statistics, for which the optimum is a D4-symmetric quadrilateral with two distances d and L. A large-b approximation for general cyclic topologies is also sketched. Code is provided.","tokens_in":12857,"tokens_out":17308,"duration_ms":165038,"significance":"The paper's modeling pipeline is clear and the reduction to pairwise distances is a useful simplification; the uniform-prior decoding step and the symmetry of the objective under Markov-chain permutations are correctly derived. The two-parameter model yields concrete, falsifiable geometric predictions, and the availability of code supports reproducibility. However, the central optimality claim is not established: the symmetry restriction is an assumption about the global minimizer, not a proof, and there are concrete arithmetic and geometric errors in key results. In its current form the contribution is best read as a symmetry-constrained analysis of a plausible objective rather than as a proof of optimal packing.","major_comments":[{"comment":"The paper's central claim that it has found the optimal packing is not supported. The text explicitly says Eq. (8) is 'probably not convex', that the authors 'settle for something weaker' in the spirit of the Purkiss principle, and that a rigorous analysis of Eq. (11) 'is not pursued'. Sections 4.1 and 4.2 then optimize only the symmetry-constrained ansatz (all distances equal, or two distances d and L under D4). Because a symmetric stationary point of a nonconvex function need not be a global minimum, and because the objective is defined over all configurations in R^D, the results as stated ('optimal packing', 'optimal solution') go beyond what is proven. This is load-bearing for the abstract and title. The authors should either prove global optimality for the two solved cases (e.g., by a convexity or monotonicity argument on the reduced objectives, combined with explicit feasible regions) or explicitly reframe the contribution as the optimal symmetric packing.","section":"Section 3, Sections 4.1–4.2"},{"comment":"The coefficient of e^{-d^2} in the uniform-case p_int and in the displayed J is wrong. For x≠y, the number of ordered pairs (a,b) with a≠x, b≠y, and a≠b is (M−1)^2 − (M−2) = M^2 − 3M + 3, not M(M−1) − 2(M−2) + 1 = M^2 − 3M + 5. Direct counting for M=2 gives one cross term, whereas the printed formula gives three. Consequently the objective plotted in Fig. 2 and the associated phase-transition statements must be recomputed; this is an arithmetic error in a central equation, not a typo.","section":"Appendix B, Section 4.1"},{"comment":"The claimed large-b degenerate solution 'd = 0 and L ≠ 0' is not realizable as a Euclidean configuration. If the four side distances d12=d23=d34=d41 are zero, then all four vertices coincide and the diagonals L=d13=d24 must also be zero. The feasible line-degenerate configuration for the four-cycle is the opposite limiting case, L=0 with d>0 (two pairs of states coalesce). The text and Fig. 3 panels that describe this phase therefore need correction, and the optimization should explicitly enforce the Euclidean embedding constraints on (d,L), at minimum the triangle inequalities for the face triangles.","section":"Section 4.2"}],"minor_comments":[{"comment":"The section is titled 'Optimal packing for cyclic topology' but, for M>4, it only derives an approximate large-b system of equations and states that it cannot be solved analytically. The section should be retitled or clearly scoped as an approximate analysis, since the abstract's 'cyclic transition statistics' case is only exactly solved for M=4.","section":"Section 4.3"},{"comment":"The statement that setting p(x|x)=0 is 'without loss of generality' needs a sentence of justification, since deleting self-loops generally changes transition probabilities on a fixed time scale.","section":"Section 2"},{"comment":"The exponents in the sums after the derivative appear to mix d_k^2 and d_k^2/2; please recheck these equations against the bracket in Eq. (32) and the derivative of Z.","section":"Appendix D, Eqs. (33)–(35)"},{"comment":"The caption says the objective is plotted versus d and L, but the main text describes a two-variable landscape; please specify whether the figure shows a contour or a surface and state the fixed parameter values.","section":"Figure 3b"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is transparent about the Purkiss gap and the difficulty of the general problem, which I appreciate. My main concern is that the published version, if it keeps the current title and abstract, will be read as having proven global optimality when it has not. I recommend requiring either a proof of global optimality or a consistent reframing to symmetry-constrained optimality, plus correction of the Appendix B coefficient and the degenerate-solution interpretation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"John,\n\nThe paper sets up a normative optimization problem for where to place attractor states in a neural representation, balancing transition speed against noise robustness, and solves two toy cases exactly under a symmetry assumption. The genuinely new piece is the M=4 cyclic case: the optimal symmetric configuration is generally not a square, with a non-trivial ratio L/d. That result, plus the general framework, makes the paper worth a careful read.\n\nWhat's done well: the objective combining Kramers-type transition probabilities, noisy encoding/decoding, KL divergence to the environmental Markov chain, and a firing-rate penalty is clearly motivated. The algebra in the appendices is mostly explicit, and the code is available for numerical checks. The paper is honest in the main text that the Purkiss-principle step is a hope rather than a proof, and that the problem is probably non-convex.\n\nThe soft spots, in order of importance. First, Appendix B has a concrete arithmetic error: the coefficient of e^{-d^2} in the uniform-case pint should be M^2-3M+3, not M^2-3M+5. My own count for M=2 gives 1, for M=3 gives 3; the printed formula gives 3 and 5. The mistake propagates to the uniform-case objective and any quantitative statements about the simplex solution. It is localized and easily fixed, but as written the uniform calculations are unreliable. Second, the abstract says \"we show that symmetries in environmental transition statistics imply certain symmetries of the optimal neural representations.\" That overstates what is proven: the paper optimizes only the symmetry-constrained ansatz and does not establish that the global minimum inherits the symmetry. The text is transparent about this, but the abstract and title are not, and the claim \"optimal packing\" in the title is stronger than the established result. If the true global optimum breaks symmetry, the reported packings are only stationary points within a restricted family. Third, the objective has two free parameters (b and alpha) and no data are fit, so it is a toy model; that's fine, but it means the paper's contribution is conceptual rather than empirical.\n\nOverall, the central idea is new and the M=4 observation is likely to be useful. The paper deserves a serious referee, not a desk reject. I would ask the authors to fix the coefficient, soften the abstract/title to \"symmetry-constrained optima\" unless they can prove the Purkiss step, and maybe add a numerical check of global optimality using their own code. If those are done, I'd happily cite it.\n\nRecommendation: send to peer review, conditional on revision.","headline":"A fresh normative framework for neural geometry with a genuinely non-square M=4 result, but the symmetry-inheritance claim is over-sold and an appendix arithmetic slip needs fixing.","tokens_in":13436,"tokens_out":4746,"would_cite":false,"duration_ms":42284,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-16T12:32:56.753296+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}