{"id":"3d9e754b-4265-4353-9c4d-61f148e3a2a3","arxiv_id":"2502.10005","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of polynomialization and quadratization of ODEs, covering existence theorems, algorithms, software tools, and worked examples from neural networks and cell signaling.","lead":"This chapter reviews how to rewrite nonlinear differential equations into polynomial or quadratic form by introducing new variables. It is a practical survey for scientists who want simpler, more computationally tractable models of complex dynamics.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No load-bearing flaw found in the central existence theorems; the chain-rule error in Example 4 is real but localized.","rationale":"The reader's conditional verdict is reasonable because Example 4 contains a genuine mathematical error, even though it is localized. I do not, however, regard the abstract's 'majority' wording or the differentiability/differential-algebraicity scope as a load-bearing objection: the paper states its assumptions, and the central theorems do not overreach them. The strongest support for the central claim, Theorem 2, is correctly proved in the text, and Theorem 1 is a standard cited result. The Example 4 error should be corrected before publication, but it does not change the validity of the main existence and algorithm-survey conclusions. Thus I leave the reader's conditional verdict unchanged.","tokens_in":15246,"tokens_out":13099,"duration_ms":135777,"concrete_test":"Correct Example 4 by setting w1 = 1/x, w2 = J_alpha(x), w3 = J_alpha'(x) and recomputing the derivatives with the chain rule. The corrected system is ˙x = w2, ˙w1 = -w1^2 w2, ˙w2 = w2 w3, ˙w3 = -w1 w2 w3 - (1 - alpha w1^2) w2^2. If this system is polynomial, the example's conclusion is repaired and the central theorem is unaffected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I could not find a load-bearing flaw in the central claims. Theorem 2's construction with the monomial set M is sound: every monomial in p(x) lies in M, and the derivative of any m in M is a sum of products of two elements of M, so the resulting system is quadratic. Theorem 1 is a standard result cited to [33]; although no proof is reproduced in the chapter, I see no internal inconsistency or missing hypothesis that would make the statement false. The concrete defect is in Example 4: the variables are written as functions of t rather than of the state x, and the displayed time derivatives omit the chain-rule factor. Replacing t by x and including the chain rule gives a still-polynomial system, so the intended point of the example survives. This is an editing defect in an illustrative example, not a failure of the main existence results.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This chapter surveys the theory and algorithms for converting finite-dimensional ODE systems into polynomial or quadratic form by adjoining new variables. It defines polynomialization and quadratization, states existence results for differentially-algebraic right-hand sides and for polynomial systems, proves a bound for monomial quadratization (Theorem 2), describes the BioCham and QBee software, and illustrates the methods on a sigmoid perceptron model and a MAPK signaling model. The chapter closes with a list of open problems.","tokens_in":15375,"tokens_out":12857,"duration_ms":114842,"significance":"The central existence guarantees are useful and, as far as the surveyed derivations go, sound. The self-contained proof of Theorem 2 is elementary and correct, and most examples verify their constructions explicitly. The chapter is valuable in bringing together results from symbolic computation, control, and systems biology and in pointing to concrete software implementations. The reduction of a sigmoid perceptron model to a two-dimensional quadratic system is a particularly effective demonstration. However, the Bessel-function example contains a mathematical error that must be corrected before the chapter can serve as a reliable reference, and the MAPK claim would benefit from a reproducible output.","major_comments":[{"comment":"The displayed polynomial system in Example 4 is not a polynomialization of dot x = J_alpha(x) in the sense of Definition 1. If the new variables are w1 = 1/x, w2 = J_alpha(x), w3 = J'_alpha(x), then along solutions the correct derivatives are dot w1 = -w1^2 w2, dot w2 = w2 w3, and dot w3 = (-w1 w3 + (alpha w1^2 - 1)w2) w2, because each derivative must carry the chain-rule factor dot x = w2. The system printed in the manuscript (with dot w1 = -w1^2, dot w2 = w3, dot w3 = -w1 w3 + (alpha w1^2 - 1)w2) is instead valid for an auxiliary variable t satisfying dot t = 1, after conflating the state x with the independent variable of Bessel's equation. Please correct the example; the intended point survives, since the corrected system is still polynomial.","section":"Section 2.3, Example 4"}],"minor_comments":[{"comment":"The symbol for the nonnegative integers is printed as \"Z/greaterorequalslant0\"; this is a corrupted encoding of \\mathbb{Z}_{\\ge 0} and should be fixed.","section":"Section 1.4"},{"comment":"The sentence \"there exists and integer h\" should read \"there exists an integer h\", and it may be worth specifying that h is nonnegative.","section":"Section 2.2, Definition 2"},{"comment":"The bound written as \"\\Pi^{n+r}_{i=1}(d_i+1)\" should be typeset as a product over i from 1 to n+r, and the definition of d_{n+i} should clarify that these are the degrees in the input variables u_i.","section":"Section 3.2, Theorem 4"},{"comment":"The reconstruction formula x = -1/a ln(w4/w1^2) - b/a assumes a is nonzero and that the logarithm argument is positive; these assumptions should be stated explicitly.","section":"Section 4.1"},{"comment":"The 15-variable quadratization of the MAPK model is reported without the resulting quadratic system, so the computation cannot be checked from the chapter; please include the full system in an appendix or provide the QBee output and version in the text.","section":"Section 4.2"},{"comment":"The phrase \"These/emdash.cyrpossibly incomplete/emdash.cyrresearch questions\" appears to be a corrupted encoding of \"These—possibly incomplete—research questions\" and should be corrected.","section":"Section 5"}],"recommendation":"minor_revision","confidential_remarks":"The mathematical error in Example 4 is the only substantive technical defect I found; the central theorems appear sound and the fix is local. The MAPK section is the weakest part from a reproducibility standpoint because the output is not shown. The chapter cites the authors' own QBee package and related papers heavily; this is understandable for a survey of the authors' tools, but the editors may wish the balance to be checked. With the Example 4 correction and the minor fixes, I would support publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The one thing to know: this is a review chapter, not a new-results paper. It collects the known existence theory for polynomialization and quadratization of ODEs, describes the two main software tools (BioCham and QBee), and works through a few instructive examples. The advertised value is real: if you work in model reduction, chemical reaction networks, or control, this is a usable entry point into that literature, and Section 5's open problems are honestly framed.\n\nWhat's genuinely good: the proof of Theorem 2 (monomial quadratization always exists) is clear and self-contained, and the examples that I checked—Exponent, the sigmoid perceptron, the suboptimality example—are algebraically correct. The authors are also honest about what is and isn't algorithmic, and they flag the manual step in the perceptron reduction rather than hiding it. That's decent scholarship.\n\nNow the soft spots, in proportion. First, Example 4, the Bessel function polynomialization, has a genuine mathematical error. The new variables are written as functions of t, and their derivatives omit the chain-rule factor from dx/dt. It's a localized defect in an illustrative example, and the intended point—that Bessel functions can be handled if you know their differential equation—survives if the variables are defined as functions of x and the chain rule is applied. But as printed, the ODE system for w is wrong, and a reader who uses it as a template will get confused. The stress-test note agrees this is an editing defect, not a load-bearing flaw, and I concur.\n\nSecond, the MAPK quadratization (15 new variables) is reported as a QBee computation, but the quadratic system itself is omitted and no reproducible artifacts are provided. It's not central to the paper, but for a survey that wants to showcase software, this is a missed opportunity—the reader has to take the claim on faith. Third, the paper leans heavily on the authors' own prior work (QBee, the non-autonomous quadratization results). That's not a flaw by itself, but it means the survey's selection of algorithms is to some extent a map of their own research program.\n\nNet verdict: the central existence claims are sound, the writing is clear, and the errors are localized. This deserves a serious referee, but only after the Bessel example is corrected and ideally with the MAPK computation made reproducible. I'd send it to review, but with a clear request for revision, not as a straight accept.\n\nFor your reading group: maybe. It's a useful survey to skim, but there's no new theorem or algorithm to sink your teeth into.","headline":"Solid survey, not a research paper: the central existence theorems hold up, but Example 4 has a real chain-rule error that needs fixing before this is citable as a reference.","tokens_in":15906,"tokens_out":1443,"would_cite":false,"duration_ms":16093,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["34-02","34C20","68W30"],"pacs":[],"model":"deepseek-v4-flash","headline":"Most nonlinear ODE models can be recast as quadratic systems by adding new variables, this chapter argues.","keywords":["polynomialization","quadratization","lifting transformations","nonlinear ODEs","differentially-algebraic functions","monomial quadratization","symbolic computation","chemical reaction networks"],"falsifier":"Find or construct a polynomial ODE system in $n$ variables whose optimal monomial quadratization requires a number of new variables that grows faster than any polynomial in $n$; the paper notes that Conjecture 1 says such systems exist, so exhibiting one with a certified exponential lower bound would show that low-order quadratization cannot be expected for all polynomial systems.","tokens_in":15023,"feed_emoji":"🧮","tokens_out":6211,"duration_ms":56621,"temperature":0.7,"pith_summary":"This chapter argues that polynomialization and quadratization—adding auxiliary variables so that an ODE system's right-hand side becomes polynomial, then quadratic—are broadly available and practically useful transformations for nonlinear models. The paper's central existence results state that any system whose right-hand side is made of differentially-algebraic functions (functions that satisfy an algebraic differential equation, such as exponentials, sines, and logarithms) can be polynomialized, and that every polynomial system admits a monomial quadratization whose order is bounded by a product of one plus the per-variable degrees. On the algorithmic side, the paper reviews two software tools that can discover these representations automatically, with the caveat that current polynomialization tools offer no optimality guarantee. If these claims hold, many difficult nonlinear models become accessible to the standard quadratic-system toolbox used in simulation, model reduction, reachability analysis, and chemical-reaction-network computing.","feed_headline":"Most nonlinear ODE models can be recast as quadratic systems","feed_subtitle":"Adding variables reveals hidden structure, making simulation, control, and learning easier.","key_machinery":"The central objects are the lifting transformation and the differential-algebraicity condition. A polynomialization is a map $x \\mapsto [x, w(x)]$ with $w$ a vector of new functions chosen so that the augmented system has polynomial right-hand sides; a quadratization is the same idea with degree at most two, and in the monomial case the new entries are monomials. The machine that makes existence work is the class of differentially-algebraic functions, which is closed under arithmetic and composition; Theorem 1's proof is constructive, and for quadratization the construction is explicit: take the finite set $M$ of all monomials $x_1^{e_1}\\cdots x_n^{e_n}$ with $0 \\le e_i \\le d_i$, and the derivative of any $m \\in M$ is quadratic in $M$. Algorithmically, the search is carried out by either SAT-solving over this large monomial set or tree search over candidate variable sets.","core_discovery":"On the paper's own terms, the core discovery is a pair of existence theorems with practical reach. Theorem 1 says that if the functions $f(x)$ are analytic and differentially-algebraic with respect to $x$ on a domain $D$, then system $\\dot{x}=f(x)$ has a polynomialization: there are new variables $w(x)$ and polynomials $q_1,q_2$ such that trajectories of the original system are mapped to trajectories of the polynomial system $\\dot{x}=q_1(x,w)$, $\\dot{w}=q_2(x,w)$. Theorem 2 says that every polynomial ODE system $\\dot{x}=p(x)$ has a monomial quadratization, with an optimal monomial quadratization of order at most $\\prod_{i=1}^n(d_i+1)$, where $d_i$ is the degree of $p$ with respect to $x_i$; the proof takes all monomials with exponents bounded by $d_i$ as new variables. Together these theorems turn 'look for hidden polynomial or quadratic structure' from an art into a search problem with existence guarantees, and the chapter documents that these guarantees are already implemented in practice for a useful class of elementary functions.","pith_inferences":["A straightforward extension would be to let users of polynomialization software supply a special function by its defining differential equation, such as the Bessel function $J_\\alpha$; the paper points to this idea but does not implement it.","If optimal quadratization is NP-hard in general, then in practice one should expect a trade-off between expressive new-variable languages (Laurent monomials, arbitrary polynomials) and tractability; the paper notes these can reduce dimension but no general algorithm exploits them.","For data-driven model reduction, the existence of low-order quadratizations suggests that learned latent coordinates might be constrained to monomial or polynomial functions rather than free-form neural-network coordinates; that connection is implicit but not developed.","The same lifting idea could be applied to partial differential equations, but the chapter only cites preliminary ODE-based results; a testable next step is to quadratize spatially discretized PDEs and compare the resulting reduced-order models with direct discretization."],"forward_implications":["Every polynomial ODE has a monomial quadratization of finite order, so the question of whether a degree reduction is possible is settled; the remaining question is how small the order can be.","Polynomialization turns models built from exponentials, sines, logarithms, and compositions of them into polynomial systems, opening them to standard quadratic-system analysis.","Quadratized systems can be interpreted as chemical reaction networks with bimolecular reactions, which is a step toward analog and Turing-complete chemical computing.","A non-autonomous polynomial ODE with differentiable inputs admits a monomial quadratization; if the input is not differentiable, an input-free quadratization may fail, so differentiability is a real boundary.","For applications like a sigmoid perceptron and the MAPK cell-signaling model, automatic quadratization produces compact systems and even reveals closed subsystems that are not obvious from the original equations."],"supporting_citations":[{"why":"Supplies Theorem 1's existence result that differentially-algebraic right-hand sides can be polynomialized.","marker":"[33]"},{"why":"Source of Theorem 2's monomial quadratization existence and the order bound $\\prod(d_i+1)$.","marker":"[18]"},{"why":"Establishes NP-hardness of optimal quadratization and provides the SAT-based approach implemented in one of the reviewed tools.","marker":"[31]"},{"why":"Introduces the tree-search algorithm for optimal monomial quadratization used by the QBee package.","marker":"[14]"},{"why":"Extends quadratization to non-autonomous polynomial ODEs with inputs and gives the differentiability conditions.","marker":"[13]"},{"why":"Presents the polynomialization algorithm with quadratic-time behavior used in the BioCham tool.","marker":"[32]"},{"why":"Shows polynomial differential equations can compute all real computable functions, linking quadratization to Turing-complete chemical reaction networks.","marker":"[11]"}],"fun_headline_variants":["Existence proofs turn quadratization from art to search","Adding variables unlocks polynomial and quadratic ODE forms","Two theorems guarantee polynomial and quadratic representations","Nonlinear ODEs: hidden structure made explicit via new variables"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the system's right-hand side is smooth and differentially-algebraic on the domain of interest; if a model contains a non-differentially-algebraic or non-smooth term, such as a piecewise or data-defined function, the existence of a polynomialization and the software guarantees do not apply.","fun_headline_variants_meta":{"raw":{"variants":["Existence proofs turn quadratization from art to search","Adding variables unlocks polynomial and quadratic ODE forms","Two theorems guarantee polynomial and quadratic representations","Nonlinear ODEs: hidden structure made explicit via new variables"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000237,"raw_usage":{"total_tokens":1500,"prompt_tokens":930,"completion_tokens":570,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":546,"completion_tokens_details":{"reasoning_tokens":507}},"tokens_in":546,"tokens_out":570,"duration_ms":5477,"temperature":1.0,"reasoning_tokens":507,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T19:43:53.434381+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find or construct a polynomial ODE system in $n$ variables whose optimal monomial quadratization requires a number of new variables that grows faster than any polynomial in $n$; the paper notes that Conjecture 1 says such systems exist, so exhibiting one with a certified exponential lower bound would show that low-order quadratization cannot be expected for all polynomial systems.","supporting_citations":[{"cited_title":"In: Cinquemani, E., Paulev´ e, L","cited_arxiv_id":null,"evidence_quote":"Supplies Theorem 1's existence result that differentially-algebraic right-hand sides can be polynomialized."},{"cited_title":"Electron","cited_arxiv_id":null,"evidence_quote":"Source of Theorem 2's monomial quadratization existence and the order bound $\\prod(d_i+1)$."},{"cited_title":"Computer Methods in Applied Mechanics and Engineering 369, 113171 (2020)","cited_arxiv_id":null,"evidence_quote":"Establishes NP-hardness of optimal quadratization and provides the SAT-based approach implemented in one of the reviewed tools."},{"cited_title":"In: Flocchini, P., Moura, L","cited_arxiv_id":null,"evidence_quote":"Introduces the tree-search algorithm for optimal monomial quadratization used by the QBee package."},{"cited_title":"SIAM Journal on Applied Dynamical Systems 23(1), 982–1016 (2024)","cited_arxiv_id":null,"evidence_quote":"Extends quadratization to non-autonomous polynomial ODEs with inputs and gives the differentiability conditions."},{"cited_title":"In: Abate, A., Petrov, T., Wo lf, V","cited_arxiv_id":null,"evidence_quote":"Presents the polynomialization algorithm with quadratic-time behavior used in the BioCham tool."},{"cited_title":": Polynomial differential equations compute all real computable functions on computa ble compact intervals","cited_arxiv_id":null,"evidence_quote":"Shows polynomial differential equations can compute all real computable functions, linking quadratization to Turing-complete chemical reaction networks."}],"review_version":1}