REVIEW 3 major objections 9 minor 2 references
Engineered Complete Intersections: Algorithmic Aspects
T0 review · 3 major / 9 minor · reviewed 2026-07-30 · grok-4.5
Pith's one-line read Mixed subdivisions of engineered complete intersections give their root counts and eliminant Newton polytopes, and a tropical homotopy computes them.
desk verdict Solid algorithmic companion to Esterov’s ECI/MCI theory: mixed cells, a workable tropical homotopy, and an eliminant vertex oracle, with code and timings that beat some prior tropical pipelines on the reported examples. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
Mixed cell cones: the closed set of height vectors for which a given mixed cell remains a dual tropical root is a polyhedral cone cut out by explicit affine-circuit inequalities; the WallWalk step of the tropical homotopy updates the mixed subdivision precisely when a straight-line path crosses a facet of one of these cones.
What would settle it
On any small engineered complete intersection whose generic root count is already known by Gröbner bases or certified numerical solving, run the mixed-subdivision algorithm and check whether the sum of cell volumes equals that known count and whether the eliminant Newton polytope matches an independently computed one.
Extended reading notes
Core claim
For a zero-dimensional matroid complete intersection and a sufficiently generic height vector, the mixed volume equals the sum of the volumes of the dual tropical roots (mixed cells) at that height; those cells are exactly the data needed both to count roots of a generic engineered complete intersection and to read off vertices of the Newton polytope of any of its eliminants.
Load-bearing premise
Height vectors and the straight-line homotopy path must avoid a finite arrangement of bad linear subspaces and must cross only the relative interiors of mixed-cell-cone facets; otherwise the cell list and the volume sum can be wrong.
Editorial extensions
If this is right
- Generic root counts of square engineered systems, including many chemical-reaction steady-state systems, are obtained by summing mixed-cell volumes without solving the system.
- Coupling the computed mixed subdivision with existing tropical start-system techniques yields an optimized numerical homotopy for the same systems.
- Vertices of Newton polytopes of A-discriminants and other ECI eliminants are returned by a single linear-form evaluation on the mixed cells, enabling evaluation-interpolation recovery of the eliminant itself.
- Real-patchworking for engineered systems reduces real-root counting for large parameter values to linear algebra over the field with two elements on the simplicial cancellations of each mixed cell.
Reading between the lines
- The same cancellation-and-mixed-cell formalism should extend, with only notational changes, to the larger class of systems that are nondegenerate upon monomial cancellation.
- Once mixed-cell cones are available, certified path-tracking or interval methods could replace the probabilistic dual-number perturbation used to guarantee genericity.
- The vertex-oracle speed-up observed for A-discriminants suggests that other classical resultant and discriminant polytopes admitting an engineered presentation become practical targets for the same pipeline.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops algorithmic tools for engineered complete intersections (ECIs) and their combinatorial abstraction, matroid complete intersections (MCIs), building on the first author's prior theory [Est24; Est25]. The main contributions are: (1) a generalization of Huber–Sturmfels mixed subdivisions to MCIs, with Theorem 2 expressing the mixed volume MV(r,A) as the sum of volumes of dual tropical roots at a generic height vector; (2) a tropical homotopy continuation algorithm (Algorithms 1–2) that tracks mixed subdivisions along a path in height space, with an explicit polyhedral description of mixed cell cones (Lemma 2) and a wall-crossing update rule (Corollary 1, WallWalk); (3) an effective real-patchworking pipeline, including a reduction of real root counting for systems with mutually transversal simplicial supports to linear algebra over F₂ (Theorem 4), applied to prove the existence of a degree-4 hypersurface in 3 variables whose discriminant curve has all 24 cusp singularities real (Theorem 6); and (4) a vertex oracle for the Newton polytope of an ECI eliminant (Theorem 5, Corollary 3, Algorithm 3), built on a support-function formula from [Est25] and combined with Huggins' method, applied to A-discriminants and ED-type eliminations. All algorithms are implemented in a publicly available Julia package (MCISubdivisions.jl, built on OSCAR), and the paper reports experiments on steady-state systems from chemical reaction networks, A-discriminants, and euclidean-dista
Significance. If the results hold, this is a useful and concrete advance in computational tropical geometry. It provides the first implementation of tropical homotopy continuation that manipulates mixed cell cones of ECI/MCIs directly (rather than via chains of flats of matroids as in [DR24]), with reported speedups of one to two orders of magnitude over [HHR24] and [Fel+26] on chemical-reaction-network examples. The eliminant vertex oracle yields A-discriminant Newton polytopes in times comparable to or better than dedicated tropical implicitization [RST25]. Strengths that deserve explicit credit: the key formulas are parameter-free and proved in the text; the software is public with reproducible example scripts; the F₂-linear-algebra reduction for real root counting (Theorem 4) is elegant and effective; and Theorem 6 is a concrete, machine-verified existence result (a degree-4 surface with all 24 discriminant cusps real) that showcases the whole pipeline. Proposition 1 (multiplicities of positive-dimensional MCI tropicalizations as mixed volumes of cancellations) generalizes a result of Sturmfels–Tevelev and is of independent interest.
major comments (3)
- [§3.3, Corollary 1] Corollary 1 (the wall-crossing update on which the correctness of Algorithm 1 rests, cf. Remark 8) is internally inconsistent in orientation. The statement concludes 'for every M ∈ M+ there is M′ ∈ M− with supp(M) ⊆ supp(M′) ∪ supp(c)', but the proof takes M ∈ M− with M ∉ M+ ('Since we have M∉M+, but M ∈ M−') and constructs M′ ∈ M+ — which is the direction WallWalk actually implements (input M−, output M+), and Example 6 also follows the proof's direction. As written, the statement's quantifiers and the proof's labels are swapped. Please make statement, proof, and Algorithm 1 mutually consistent, and state explicitly in which direction the support inclusion holds, since a reader verifying the homotopy's correctness needs the orientation (and the sign of ∑_{s∈S′_i} c_s < 0 relative to the crossing direction) pinned down.
- [§3.3, Notation 6 / Lemma 3] Notation 6 defines r″(S) := max{r(S∩A) + r′(S∩B), n}. This cannot be a matroid rank function (e.g. r″(∅) = n, and ranks of small sets exceed their cardinality). The proof of Lemma 3 uses r″(S) = r′(S) for S ⊆ B, which holds for min{r(S∩A) + r′(S∩B), n} — the truncation at rank n of the direct sum — so min is evidently intended. Since Lemma 3 is the initialization step of Algorithm 2, please correct the definition and re-check the statements of Lemma 3 and Algorithm 2 (lines 6–9) accordingly.
- [§5, Table 3 (Euclidean Distance Degree)] Three of the seven ED-degree instances in Table 3 report 'rounding error' for Algorithm 3: facet inequalities of mixed cell cones are computed in floating point and become ill-conditioned when one support entry dominates. Since a misclassified cone facet silently invalidates the WallWalk update, this bears on the practical-reliability claim of §5. The authors acknowledge the issue and sketch an exact fallback; I ask that the paper (a) state explicitly which parts of the pipeline are exact (dual-number bookkeeping, circuit arithmetic) and which are floating-point, and (b) discuss cheap a posteriori certification — e.g. verifying the output subdivision's cone inequalities in exact arithmetic, or certifying the computed vertex/Σsvol against a known degree or an independent evaluation — so that a user can detect rather than inherit such failures.
minor comments (9)
- Lemma 2, proof: the second family of inequalities is written as π_A(c(i,i+1))·d ≥ 0, but the defining condition in the same proof is the strict ordering (ω,1)(S_{1,d}) > (ω,1)(S_{2,d}) > …, which yields strict inequalities for points of C°_M(r,A). Please reconcile strict vs. non-strict inequalities with the open cone / euclidean closure distinction in the statement.
- Proposition 3: item 2 concludes 'MV(r,A) = …' but the quantity on the left should presumably be the mixed volume of the sequence (r_i,S_i)_i in the sense of Definition 8(4); also item 1 writes A_i where the hypothesis names the sets S_i. Example 3(2) refers to 'item 3 in Definition 4', but Definition 4 has no items — presumably Definition 8(3) is meant.
- Definition 11: 'if only if' should be 'if and only if'. Remark 6: 'appears more than once than we consider' should be 'then'. Theorem 4's statement has a stray unbalanced parenthesis ('not satisfying the condition of Lemma 1)'), and Lemma 1 and the proof of Theorem 4 use A_i where S_i is meant.
- Algorithm 1, Homotopy, line 4: the loop condition quantifies over 'M ∈ M′' but should range over the current subdivision M; moreover, after setting t_curr to the minimal crossing time, the search should be over (t_curr, 1] to avoid re-detecting the same wall at t_curr. WallWalk, line 11: 'for any circuit S′_i …' reads existentially; presumably all such circuits are enumerated ('for every circuit').
- Algorithm 2, line 3: 'the regular subdivision of A at height d′' should be the set of maximal cells of the regular triangulation of A at d′, since the circuits of the uniform matroid u_{A,n} are the (n+1)-element subsets.
- §5, Tables 1–2: the speedups over [HHR24] (60s → 0.7s) and [Fel+26] (262s → 3.1s) quote timings reported in those papers, presumably on different hardware. Please add a one-line caveat that the comparison is against published timings, or ideally rerun one baseline on the same machine.
- Theorem 6 and its proof: Theorem 3 only guarantees equality of real root counts for t0 ≫ 0, with no effective bound, and the proof does not exhibit or certify a specific t0. Since the theorem claims only existence this is logically sufficient, but the manuscript currently leaves unclear whether the witnessing polynomial is ˜f of Appendix A itself or an engineered f_{i,d}(t0); please clarify, and note that the msolve certification was applied to the cancellation systems, not to the final polynomial.
- A brief complexity discussion would strengthen §3.3: even an empirical statement about the number of wall crossings and the per-WallWalk cost (facet-inequality computation, circuit enumeration in matroid quotients) would help readers gauge where the bottleneck lies, particularly in light of the 86-variable odebase example where the initial regular triangulation already fails.
- Figure 1's caption/introduction cross-reference ('see Section 1') is self-referential; please point to the relevant part of §5 instead. In the abstract and §1, 'chemical reaction networks' examples would benefit from a forward pointer to Table 1.
Circularity Check
No significant circularity: combinatorial algorithms and proofs are self-contained; Est24/Est25 supply the algebraic identification in the usual theory-to-algorithms way.
full rationale
The paper’s load-bearing equalities (Theorem 2: mixed volume equals sum of dual-tropical-root volumes; Corollary 3 / Algorithm 3: vertex oracle from svol of mixed cells; Theorem 4: real orthant count via F2 linear algebra) are proved inside the manuscript from the definitions of mixed cells, cancellations, and mixed-cell cones, or are standard polyhedral identities. Algorithms 1–3 compute those combinatorial objects; they do not fit parameters to data and then re-label the fit as a prediction. Dependence on Esterov’s prior theory [Est24; Est25] is explicit and ordinary: Theorem 1 (trop(I_p)=trop(r,A) for generic p) and the support-function formula restated as Theorem 5 are cited as the bridge from algebra to the MCI combinatorics the algorithms manipulate. That bridge is not re-derived here, but neither is it smuggled in as an unacknowledged ansatz that forces the output by construction—the algorithms remain correct as combinatorial procedures on MCIs even if one only cares about mixed volumes of conewise-linear functions. Experimental tables compare runtimes and root counts against external software and databases, not against quantities fitted from the same runs. No self-definitional loop, no fitted-input-as-prediction, and no uniqueness theorem used to forbid alternatives. Score 1 reflects only the expected first-author theory citation for the ECI interpretation, not circular derivation.
Assumptions & free parameters
free parameters (2)
- Homotopy path perturbation (dual-number ε or generic v) =
symbolic dual ε or random generic vector
- Random height vectors and random signs in patchworking search =
witness d_r, d_ε and sign pattern in Appendix A
assumptions (5)
- domain assumption For generic p, trop(I_p)=trop(r,A) for the MCI built from ECI (V,A) (Theorem 1 / Est24).
- standard math Stable intersection, balancing, and corner-locus formalism for conewise-linear functions as in MS21/Est12.
- domain assumption Generic height vectors avoid finitely many linear subspaces so dual tropical roots are affinely independent mixed cells (Theorem 2).
- domain assumption Real patchworking equality of Theorem 3 (Est25) relating engineered systems at large t to real roots of cancellations.
- standard math Vertex enumeration from a vertex oracle via Huggins iB4e-style method [Hug06].
invented entities (3)
-
Mixed cell / dual tropical root of an MCI
independent evidence
-
Mixed cell cone C_M(r,A)
independent evidence
-
Moving-set MCI and svol linear forms for eliminant vertices
independent evidence
Cite this review
Pith. "Pith review of Engineered Complete Intersections: Algorithmic Aspects." pith.science (2026). https://pith.science/paper/BPQXTNGX
@misc{pith2026260723622,
author = {Pith},
title = {Pith review of: Engineered Complete Intersections: Algorithmic Aspects},
year = {2026},
howpublished = {\url{https://pith.science/paper/BPQXTNGX}},
note = {Machine review of arXiv:2607.23622}
}
abstract
Engineered Complete Intersections (ECI's) are a class of sparse polynomial systems frequently arising in a number of contexts, both in pure mathematics (e.g. enumerative geometry) and applications (e.g. chemical reaction networks). Based on theoretical results given by the first author, we give several contributions. First we give a new effective technique to tropicalize such systems by generalizing the classical notion of mixed subdivisions introduced by Huber and Sturmfels (1995) to ECI's with the particular goal to efficiently count solutions of square systems of equations in ECI form. We further design a tropical homotopy continuation algorithm for computing such mixed subdivisions, inspired by Jensen (2016), Malajovich (2017) and Daisey and Ren (2024). Our techniques can be used to numerically solve such systems by coupling them with the algorithms introduced by Helminck, Henriksson and Ren (2024). Finally, we give an algorithm to compute Newton polytopes of eliminants of ECI's. This gives a new way to compute, for example, Newton polytopes of so-called $A$-discriminants. Coupled with evaluation-interpolation paradigms our algorithm gives an efficient approach to compute such eliminants. We implemented our algorithms in the form of a software package which we use to demonstrate their practical feasibility on a range of examples.
Reference graph
Works this paper leans on
-
[1]
The Number of Roots of a System of Equations
[Ber75] D. N. Bernshtein. “The Number of Roots of a System of Equations”. In:Func- tional Analysis and Its Applications9.3(1975), pp.183–185.doi: 10 . 1007 / BF01075595. [BES21] Jérémy Berthomieu, Christian Eder, and Mohab Safey El Din. “Msolve: A Library for Solving Polynomial Systems”. In:ISSAC’21. Saint Petersburg, Russia, 2021.doi:10.1145/3452143.3465...
arXiv 1975
-
[10]
A Polyhedral Method for Solving Sparse Polynomial Systems
1007/s11538-015-0125-1. 32 [HHR24] Paul Alexander Helminck, Oskar Henriksson, and Yue Ren.A Tropical Method for Solving Parametrized Polynomial Systems.2024.doi: 10.48550/arXiv.2409. 13288. [HS95] Birkett Huber and Bernd Sturmfels. “A Polyhedral Method for Solving Sparse Polynomial Systems”. In:Mathematics of Computation64.212(1995), pp.1541– 1555.doi:10....
Reviewed July 30, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.