{"id":"7c737d8c-fd6d-47f3-bd2a-68041d7e4605","arxiv_id":"2412.14390","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"SimTeEx simplifies tensor polynomials with arbitrary symmetry relations by combining graph-based handling of dummy indices with reduced row echelon form linear algebra.","lead":"This paper presents SimTeEx, a Mathematica package that simplifies polynomial expressions with tensors by treating dummy-index contractions as graphs and tensor symmetries as linear equations. The package targets physicists computing in general relativity and effective field theories, where such expressions quickly become unwieldy.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Canonical-form claim is not established: default output is a normal form, and the true-canonical mode lacks a specified graph ordering.","rationale":"I examined the reader's weakest assumption, the Section 4 claim that putting input-present columns last prevents term-count growth. This is actually correct: any pivot row subtracted from the first row has its pivot in a column where c_i is nonzero (i.e. in the suffix), and therefore has zeros in all prefix columns; subtracting it cannot create nonzeros where c was zero. So I do not regard this as the load-bearing issue. The load-bearing gap is the mismatch between the advertised 'canonical form' and the algorithm actually specified. The default mode is explicitly a normal form (eqs. 31-32), and the true-canonical flag is not described at the level needed to verify canonicity: the paper never defines the total order on graphs that determines the RREF column order, nor the procedure for closing the graph set under the symmetry relations. Since the RREF result depends on column ordering, canonicity is contingent on an ordering rule that is absent. This supports the reader's CONDITIONAL verdict: the package may well work, but the central claim is under-specified. A direct equivalence test with $TrueCanonicalForm=True would settle whether the implementation delivers the claimed invariance.","tokens_in":21345,"tokens_out":30422,"duration_ms":247341,"concrete_test":"Download SimTeEx; with $TrueCanonicalForm=True, generate 100 pairs of equivalent expressions involving a multi-term symmetry (e.g., permutations of the Riemann-tensor example or the kappa relation of eq. (5)) by relabelling dummy indices and reordering terms; run CanonicalForm on each pair and require byte-identical outputs. If any true-canonical pair disagrees, the canonical-form claim fails as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim (abstract, Section 1) is that SimTeEx can handle any tensor symmetry and produce a canonical form. Section 5, eqs. (31)-(32), explicitly concedes that the default CanonicalForm gives different outputs for equivalent inputs (e.g. TiTi vs TjTj), so it is a normal form, not a canonical form. The offered remedy, setting $TrueCanonicalForm=True, is asserted to produce a canonical form, but the paper never specifies the total ordering of graphs that fixes the column order in the RREF of Section 4. Since the RREF first row depends on the column ordering (the paper itself shows different orderings giving different simplifications in (27)-(29)), a canonical form is only defined once this ordering is fixed and shown to be input-independent. The paper also omits a closure step: to apply a multi-term symmetry, the set of graphs must be expanded to include all graphs generated by permuting tensor slots of the input monomials; Section 4 assumes the set {g_i} is already closed without describing how this orbit is constructed or bounded. Without these specifications, the 'any symmetry / canonical form' claim is under-determined. This gap is real but fixable, so it does not refute the underlying algorithm.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents SimTeEx, a Mathematica package for simplifying polynomial expressions in tensors with dummy-index contractions and user-supplied permutation symmetries. Tensor monomials are encoded as labeled directed multigraphs, with equality of monomials reduced to graph isomorphism through canonicalized generalized adjacency matrices. Polynomials are treated as vectors over a basis of graph representatives, and tensor symmetries, including multi-term ones, are encoded as linear null relations among these representatives. The simplification step is phrased as reduced row echelon form (RREF) of a matrix whose first row holds the input coefficients and whose remaining rows hold the symmetry relations; columns corresponding to terms present in the input are placed last to control term growth. The paper documents the main CanonicalForm function, the TrueCanonicalForm option that renames dummy indices to a fixed list, handling of anti-commuting tensors, and several auxiliary group-theoretic functions. It validates the approach on the Riemann tensor identity (2) and on SMEFT-type examples.","tokens_in":21603,"tokens_out":6294,"duration_ms":59807,"significance":"If the advertised properties hold, SimTeEx fills a practical niche: it accepts arbitrary multi-term tensor symmetries directly as null equations, without requiring the user to provide Young symmetrizers, and the underlying linear-algebra-plus-graph-isomorphism strategy is simple and coherent. The paper has concrete strengths: the examples are reproducible and include a Riemann-tensor identity check with a symbolic parameter x, making the vanishing genuinely dependent on x = 1/4; the core construction is self-contained and parameter-free; and Appendix B makes a useful pedagogical point about the insufficiency of Young symmetrizers for general tensor symmetries. The significance is nevertheless conditional. The manuscript does not yet specify the total ordering needed for a true canonical form, does not describe the closure/orbit step needed to apply multi-term symmetries to arbitrary monomials, and asserts without proof the no-term-growth property. These are fixable gaps rather than refutations of the underlying approach.","major_comments":[{"comment":"The assertion that placing columns for terms present in the input last guarantees that the number of non-zero coefficients never increases is stated without proof. This property is load-bearing for the claim that CanonicalForm is a simplifier rather than just a normal-form map; indeed Eq. (29) shows that without the reordering the number of non-zero terms can increase. Please supply a proof, or state precisely under which conditions the property holds and qualify the simplification guarantee accordingly.","section":"Section 4, after Eq. (30)"},{"comment":"The true-canonical claim is under-specified. Footnote 2 defines the canonical representative of a monomial by picking 'some sorting criteria' for generalized adjacency matrices, and Section 4 shows that the RREF result depends on the column order. For CanonicalForm(expr1) = CanonicalForm(expr2) to hold for all equivalent expr1 and expr2 in TrueCanonicalForm mode, the package must fix a concrete total order on the graph representatives that is independent of the input expression and specify how it is computed. The paper does not do this, so the advertised canonical form is not well defined. Please specify the ordering used in the implementation and explain why it yields a unique, input-independent representative.","section":"Footnote 2 and Section 5 (TrueCanonicalForm mode)"},{"comment":"The application of multi-term symmetries is missing a closure step. In Eq. (24) the relations are written as linear combinations of the same basis graphs g_i that appear in the input, but a symmetry such as Eq. (21) or the Riemann relations (1) does not by itself specify which of the many slot-permuted monomials, with dummy indices relabeled, should be included in the matrix. The algorithm must generate the orbit of each input monomial under all permutations of tensor slots and all dummy-index renamings, canonicalize those graphs, and assemble the corresponding columns. The paper never states this construction or a bound on its size. Without it, Section 4 is not directly executable on the examples of Section 5.","section":"Section 4, Eq. (24)"},{"comment":"The advertised claim of producing a canonical form and handling any tensor symmetry is stronger than what is demonstrated. Equations (31)-(32) explicitly concede that the default output is only a normal form, and the true-canonical mode depends on the unspecified ordering and the unproved no-growth step noted above. Please either supply the missing specifications and proofs, or revise the abstract and introduction to describe the default behavior as normal-form simplification with an optional canonical mode whose properties are fully stated.","section":"Abstract, Section 1 and Section 5"}],"minor_comments":[{"comment":"Several displayed equations are garbled in the text: the irreducible representations in Eqs. (7)-(8) and in Appendix B appear as empty or incomplete diagrams, which makes the group-theoretic discussion hard to follow.","section":"Introduction and Appendix B"},{"comment":"There are numerous typos and awkward phrasings, e.g. 'not does on need to know' in Section 1 and 'it is always be true' after Eq. (32). A careful proofread is needed.","section":"Throughout"},{"comment":"The phrase 'some sorting criteria' is too vague even for a footnote; if the full ordering cannot be described in the main text, the author should at least identify where in the code or an appendix the concrete criterion is defined.","section":"Footnote 2"},{"comment":"In the anti-commuting tensor example, the notation mixes ordinary multiplication and NonCommutativeMultiply; a brief explanation of why this is necessary would help readers unfamiliar with Mathematica's operator conventions.","section":"Section 5, fermionic example"}],"recommendation":"major_revision","confidential_remarks":"This is a software-and-algorithm paper rather than a formal-proof paper, and the gaps identified are of the kind that can be fixed by adding an appendix with the concrete graph-ordering rule, the orbit-generation procedure, and a proof or precise statement for the no-term-growth property. I therefore recommend major revision rather than rejection. The editor may also wish to encourage the author to align the abstract and title with the actual default behavior, since the current wording overstates the canonical-form guarantee."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a genuinely useful tensor simplifier with a real new idea, and it deserves review, but the paper's canonical-form claim is not backed by a full specification. The new input format—symmetries given directly as linear null equations instead of Young tableaux—is a genuine improvement; it covers multi-term symmetries and cases Young symmetrizers cannot express (Appendix B makes that point cleanly). The graph-based dummy index handling is standard but solid, the Riemann identity check in Section 5 is convincing evidence the code actually works, and the auxiliary functions (SameEquationsQ, SingleProjector, etc.) are useful add-ons. Credit where due: the package is functional and the examples reproduce.\n\nSoft spots are about claims versus specs. The default CanonicalForm is explicitly a normal form, not a canonical form—the paper admits this in (31)–(32), which is honest. But the $TrueCanonicalForm mode is under-defined: the graph ordering that fixes the RREF column order is left as 'some sorting criteria' (footnote 2), and without a specified total order there is no canonical form, just another normal form. That is fixable, but the abstract's claim is stronger than the description. Second, Section 4 assumes the set of graphs is already closed under slot permutations but never says how that orbit is generated or bounded; users and correctness arguments need that. Third, the no-term-growth property is asserted without proof. I checked a few examples and it is actually true—present columns last keeps the RREF from touching the first row unless the rank of the symmetry system exceeds the number of absent columns, in which case the output has fewer terms—so this is a missing proof, not a false claim.\n\nFor whom? Anyone doing SMEFT or GR with multi-term symmetries, and people building tensor tools. I would send it to a serious referee, asking for the canonical ordering to be specified, the closure step described, and the no-growth property justified.","headline":"Useful new tensor simplifier with a genuinely better input format, but the canonical-form claim is under-specified and needs a concrete graph ordering.","tokens_in":22065,"tokens_out":24360,"would_cite":true,"duration_ms":193310,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper presents an algorithm, implemented in the SimTeEx package, that puts any tensor polynomial into a simplified form under arbitrary symmetries, including multi-term ones.","keywords":["tensor simplification","dummy index","multi-term symmetries","graph isomorphism","reduced row echelon form","SimTeEx","normal form","Mathematica package"],"falsifier":"Run a brute-force scan over small symmetry systems (for example rank-3 and rank-4 tensors with a few multi-term relations) and over all inputs with few monomials, and check whether CanonicalForm with input-present columns placed last ever returns more nonzero terms than the input; the paper's own equation (29) shows the naive ordering fails, and the unproven assertion is precisely that the special reordering never does. One concrete counterexample would refute the simplification guarantee.","tokens_in":21138,"feed_emoji":"🧮","tokens_out":7029,"duration_ms":62144,"temperature":0.7,"pith_summary":"The paper claims that simplifying polynomial expressions in tensors, including the dummy-index contractions that make $T_{ia}U_{aj}$ and $T_{ib}U_{bj}$ look different to a computer, can be reduced to two well-understood tasks: graph isomorphism for index contractions and linear algebra for symmetries. The algorithm represents each tensor monomial as a labeled graph, canonicalizes that graph so dummy indices become irrelevant, and encodes all tensor symmetries as linear null equations in a matrix; putting that matrix in reduced row echelon form yields the simplified polynomial. A careful ordering of the matrix columns is meant to keep the number of output terms from exceeding the number of input terms. The method is implemented in the Mathematica package SimTeEx, whose main function is CanonicalForm. A sympathetic reader would care because such simplifications are a daily chore in particle physics and general relativity, and the input format asks only for symmetry equations, not for representation-theoretic data like Young tableaux.","feed_headline":"One command simplifies tensor polynomials with any symmetry","feed_subtitle":"Multi-term symmetries like the Riemann Bianchi identity are handled by linear algebra, no Young tableaux needed.","key_machinery":"The central object is a labeled multigraph, encoded as a generalized adjacency matrix, that represents a tensor monomial; canonicalizing it by row and column permutations turns the question of whether two contractions are the same up to dummy-index relabeling into the question of whether two labelings are the same graph, which is decided by comparing canonical representatives. The second load-bearing mechanism is a matrix in reduced row echelon form: the first row holds the coefficients of the input polynomial, the remaining rows encode the symmetry null equations, and the simplified coefficients are read off the first row after reduction. A column ordering that places the columns belonging to monomials actually present in the input last is what the paper claims guarantees that the number of nonzero output coefficients never grows, mirroring the effect of the meld algorithm in earlier tensor computer algebra.","core_discovery":"At its core, the paper identifies a reduction: tensor-polynomial simplification with arbitrary permutation symmetries is equivalent to linear algebra over a vector space whose basis elements are canonical graphs. Dummy indices are handled exactly once by building a generalized adjacency matrix for each monomial, with vertices representing tensors and external indices and edges recording which slot of which tensor contracts with which slot of another, and then choosing a canonical representative among all isomorphic relabelings. Symmetries are supplied as null equations, each a sum of monomial graphs equal to zero; these become extra rows in a matrix whose first row is the coefficient vector of the input polynomial. Reduced row echelon form on this matrix yields simplified coefficients, and any set of linear null equations can be fed in, including the multi-term symmetries of the Riemann tensor or the SMEFT Wilson coefficient $\\kappa$, including cases that cannot be described by Young symmetrizers. The paper is careful to state that the default output is a normal form, not a canonical form: two equivalent expressions may print differently, but their difference always simplifies to zero.","pith_inferences":["A natural test would be to search small symmetry sets exhaustively for an input where every column order gives more nonzero terms than the input; such a case would pin down exactly when the simplification guarantee can fail to shrink an expression.","Because the dummy-index part is graph isomorphism, the worst case is exponential, so a practical benchmark suite of Riemann-type quartics would clarify where the package becomes slow.","The linear-algebra formulation looks portable: any algebraic structure whose monomials can be canonicalized and whose relations are linear could use the same RREF-based simplification, not just tensors."],"forward_implications":["Riemann-tensor identities such as equation (2) reduce automatically from the three standard symmetry equations, with no special-case code.","Wilson-coefficient relations of the SMEFT, like equation (6), follow from a linear null equation such as equation (5), without computing Young tableaux.","Tensor symmetries that cannot be expressed by Young symmetrizers, such as arbitrary linear relations between mixed-symmetry components, can still be simplified because the input format is general linear equations.","Anti-commuting fermion fields are supported: declaring them as such and using noncommutative multiplication makes Grassmann parity part of the simplification.","With default settings the result is a normal form (equivalent expressions may look different), while turning on the TrueCanonicalForm flag produces a true canonical form with canonical dummy indices."],"supporting_citations":[{"why":"Defines the only other package with multi-term symmetry support, used as the main comparison for the algorithm's capabilities.","marker":"[2]"},{"why":"Represents tensor monomials with graphs to canonicalize dummy indices, the approach that the present graph representation generalizes.","marker":"[11]"},{"why":"Uses reduced echelon form for multi-term symmetries, the same linear-algebra idea that drives the central RREF step.","marker":"[12]"},{"why":"Supplies the meld algorithm and the canonical-form versus normal-form distinction invoked in Section 4.","marker":"[13]"},{"why":"Provides the standard definition of normal form used to justify the default behaviour of CanonicalForm.","marker":"[17]"}],"fun_headline_variants":["SimTeEx: Tensor simplification for any symmetry","Any tensor symmetry? SimTeEx reduces it to linear algebra","Simplify tensor polynomials: SimTeEx handles all symmetries","From dummy indices to any symmetry: SimTeEx simplifies tensors","One command, any tensor symmetry: SimTeEx linear algebra simplification"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the unproven claim in Section 4 that putting the columns of the terms that already appear in the input last always keeps the number of nonzero output coefficients from increasing; unless that holds for every symmetry set, the algorithm's output can be longer than the input, contradicting its name.","fun_headline_variants_meta":{"raw":{"variants":["SimTeEx: Tensor simplification for any symmetry","Any tensor symmetry? SimTeEx reduces it to linear algebra","Simplify tensor polynomials: SimTeEx handles all symmetries","From dummy indices to any symmetry: SimTeEx simplifies tensors","One command, any tensor symmetry: SimTeEx linear algebra simplification"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000639,"raw_usage":{"total_tokens":2937,"prompt_tokens":933,"completion_tokens":2004,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":549,"completion_tokens_details":{"reasoning_tokens":1920}},"tokens_in":549,"tokens_out":2004,"duration_ms":13129,"temperature":1.0,"reasoning_tokens":1920,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T12:16:47.552638+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a brute-force scan over small symmetry systems (for example rank-3 and rank-4 tensors with a few multi-term relations) and over all inputs with few monomials, and check whether CanonicalForm with input-present columns placed last ever returns more nonzero terms than the input; the paper's own equation (29) shows the naive ordering fails, and the unproven assertion is precisely that the special reordering never does. One concrete counterexample would refute the simplification guarantee.","supporting_citations":[{"cited_title":"Peeters,Cadabra2: computer algebra for field theory revisited, J","cited_arxiv_id":null,"evidence_quote":"Defines the only other package with multi-term symmetry support, used as the main comparison for the algorithm's capabilities."},{"cited_title":"ATENSOR - REDUCE program for tensor simplification","cited_arxiv_id":"1811.05409","evidence_quote":"Uses reduced echelon form for multi-term symmetries, the same linear-algebra idea that drives the central RREF step."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the standard definition of normal form used to justify the default behaviour of CanonicalForm."}],"review_version":1}