arXiv++ Combinatorics

Let $G$ be a cubic graph admitting a $2$-factor consisting of exactly two odd circuits, and let the complementary $1$-factor contain precisely three spokes (along with an arbitrary number of chords). We show that four perfect matchings can cover $G$. As a consequence, $G$ fulfils the $7/5$-Conjecture of Alon and Tarsi.

We give a matroid-theoretic analogue of Gaussian elimination and use it to construct, for every matroid M, the syzygy matroid Syz(M) on the collection of circuits of M. Dualizing the construction yields the combinatorial adjoint Adj(M) on the collection of maximal proper flats of M, and the two are related by Adj(M) \simeq Syz(M*). Neither construction depends on a realization. We determine Syz(M) when M is U_{k,k+3} or the Vamos matroid V_{8}. We also show that Adj(M) is an adjoint of M when M has rank 3. When M has rank 4 and V_{8} is a restriction of M, Adj(M) has rank 3 and is not an adjoint of M.

Long weighted walks seem to sample mod-$t$ flows uniformly, but at even moduli a persistent parity bias survives. Writing $C_G(t)$ for the flow polynomial of a connected graph $G$ of genus $g$, the proportion of nowhere-zero flows under uniform sampling is $C_G(t)/t^g$. For arbitrary positive directed weights and all moduli, we determine the exact leading corrections for both closed walks and nonbacktracking closed walks, which we call circuits. Only the circuit result requires $g\ge2$. For closed walks, the canonical character governs the leading correction through the canonical modulating flow polynomial $D_G$; for circuits, the period character, when it exists, governs the leading correction through the period modulating flow polynomial $D_{G,η}$. At even moduli these give parity-dependent corrections: the two mechanisms coincide on nonbipartite graphs and separate only in the bipartite circuit problem. More generally, each integral-homology character $ζ$ determines a balanced-dichromatic specialization $D_{G,ζ}$. These polynomials form a character-indexed family whose finite-order members give Fourier coefficients of the nowhere-zero flow set and whose retained edge sets are the central sets of a gain semimatroid. We further show that $D_G$ is a cycle-matroid invariant and a signed-Tutte specialization satisfying signed deletion-contraction. Finally, an integral nowhere-zero approximation theorem implies that asymptotically almost every weighted circuit traverses every edge.

We give a short-description decomposition of bounded-width CNFs. For every fixed $\varepsilon>0$, every width-$b$ CNF on $m$ variables is an exact disjoint union of at most $2^{\varepsilon m}$ width-$b$ CNFs, each determined by $2^{O_\varepsilon(b)}\log(m+1)$ satisfying assignments. For every fixed $d\ge3$, we prove a worst-case size lower bound of $2^{Ω_d((n\log n)^{1/(d-1)})}$ against depth-at-most-$d$ unbounded-fan-in AND/OR circuits with input negations for three polynomial-time Boolean families: inverse trace over binary fields, the Goldreich--Tal trilinear function, and OpenAI's polynomial evaluator with its original parameters. All parameter bits count as inputs. For inverse trace, every such circuit of size at most $S$, where $n\le S\le2^{c_dn^{1/(d-1)}}$, has uniform sign correlation at most $2^{-η_dn\log\log S/(\log S)^{d-1}}$. Here $c_d,η_d>0$ depend only on $d$. This improves on the classical parity correlation bound. The lower bounds combine the decomposition with restriction arguments; the correlation bound additionally uses common partial decision trees. We also construct exact nonuniform AND/OR circuits for inverse trace of size $2^{O(n^{2/3})}$ at depth eight and $2^{O(\sqrt n\log n)}$ at depth nine. For every fixed integer $h\ge2$, the construction gives depth-$(4h+1)$ circuits of size $2^{O_h(n^{1/h}\log n)}$. The upper bounds implement finite-field evaluation and interpolation using circuits that enumerate intermediate states.

We study tropical circuits with scalar multiplication gates, that is, algebraic circuits whose gates implement $\max$, $+$, or multiplication with a positive constant. For such circuits, we prove exponential size lower bounds for computing maximum weight directed spanning trees and maximum weight bipartite perfect matchings. As a corollary, we obtain an exponential size separation between monotone and non-monotone maxout neural networks, which generalize the popularly used ReLU neural networks. One conclusion from this is that neural network models with enforced convexity constraints, such as input-convex neural networks (ICNNs), sometimes need to be exponentially larger than their unrestricted counterparts in order to express the same functions.

For a partition $\boldsymbolλ = (λ_1,\ldots,λ_r)$ and $n\ge r$, the monomial symmetric polynomial is defined as $$m_{\boldsymbolλ}(x_1,\ldots,x_n) := \sum x_{i_1}^{λ_1} x_{i_2}^{λ_2} \cdots x_{i_r}^{λ_r},$$ where the sum ranges over all distinct monomials obtained by assigning the parts of $\boldsymbolλ$ to pairwise distinct variables. We establish an almost-complete dichotomy for the complexity of the monomial symmetric polynomials $m_{\boldsymbolλ}$: If a family of partitions contains polynomially many distinct parts, then the corresponding monomial symmetric polynomials (on sufficiently many variables) are VNP-complete. If the number of distinct parts is constant, the family admits polynomial-size circuits. Thus the only remaining gap concerns the intermediate regime where the number of distinct parts grows subpolynomially. As part of our proof, we use combinatorial properties of matching polynomials to establish VNP-hardness of the perfect matching polynomial for blow-up graphs; these are graphs with non-generic edge-weights that are obtained by "blowing up" a smaller graph. We reduce these matching polynomials to monomial-symmetric polynomials by combining classical techniques from algebraic complexity theory with structural properties of Sidon sets.

Let \(b(M)\), \(d(M)\), and \(r(M)\) denote the number of bases, the number of circuits, and the rank of a matroid \(M\), respectively. We prove that every nonempty simple binary matroid with no coloops satisfies \[ 2b(M)\ge(r(M)+1)d(M), \] with equality if and only if \(M\) is isomorphic to the Fano matroid. This confirms a conjecture recorded by Oxley in 1983. For the same class, we prove that deleting any element leaves at least as many bases as there are circuits in the original matroid: \(b(M\backslash e)\ge d(M)\) for every \(e\in E(M)\). We determine all equality cases and deduce a sharp linear lower bound for basis growth under successive series extensions. We also give a circuit-size exclusion condition guaranteeing distinct bases containing \(C-e\) for circuit-element pairs \((C,e)\) with prescribed circuit sizes.

We prove that there exist uniform $(+,\times,/)$-circuits of size $O(n^3)$ to compute the basis generating polynomial of regular matroids on $n$ elements. By tropicalization, this implies that there exist uniform $(\max,+,-)$-circuits and ReLU neural networks of the same size for weighted basis maximization of regular matroids. As a consequence in linear programming theory, we obtain a first example where taking the difference of two extended formulations can be more efficient than the best known individual extended formulation for the base polytope of regular matroids of size $O(n^6)$ by Aprile and Fiorini. Such differences have recently been introduced as virtual extended formulations. The proof of our main result relies on a fine-tuned version of Seymour's decomposition of regular matroids which allows us to identify and maintain graphic substructures to which we can apply a local version of the star-mesh transformation.

A biclique decomposition of a graph is a partition of its edges into complete bipartite subgraphs. We consider graphs whose vertices can be ordered such that the neighborhood of every vertex is the union of a sublinear number of intervals. We observe that these graphs admit compact representations in the form of biclique decompositions of small size. Here, the size of a decomposition is measured as the sum of the number of vertices of its bicliques. Combining this result with the existence of suitable vertex orderings for graphs of low neighborhood complexity, as proven by Welzl in 1988, we recover and extend several known results up to logarithmic factors. These results include upper bounds on the Zarankiewicz problem, matrix multiplication, quantum circuit complexity, and shortest path algorithms in "well-structured" instances.

Let $G=(V(G),E(G))$ be a finite connected simple cubic graph. A total coloring $μ:V(G)\cup E(G)\rightarrow\{0,1,2,3\} $ is efficient if each color yields an efficient dominating set, where the efficient dominating condition applies to the restriction of each color class to $V(G)$. If the girth of $G$ is 4 and $μ$ uses 4 colors both on vertices and edges of each 4-cycle, then $μ$ is an ETGC. Maps $M(G)$ of such $G$ are considered on the orientable surfaces. We show that a necessary condition for ETGCs to exist on $M(G)$ is that they have only face lengths divisible by 4. When such ETGCs exist, for which 7 constructive tools are provided, the faces of the maps $M(G)$ possessing them are shown to be colorable by the group $S_3$ of 3-permutations whenever the ETGCs are periodic. For the vertex coloring of each ETGC, there are two orthogonal edge colorings: directed and reversed. There are two injections into the set of face $S_3$-colorings: one from the set of periodic directed ETGCs, the other from that of periodic reversed ETGCs. We take each edge of $G$ as a pair of opposite arcs. Let $A^\leftrightarrow$ be the set of such arcs and let $G^\leftrightarrow$ be the digraph with vertex set $V(G)$ and arc set $A^\leftrightarrow$. A ditotal coloring $μ^\leftrightarrow$ of $G^\leftrightarrow$ is given by having bijections via $μ^\leftrightarrow$ from the sets formed by: {\bf(a)} each vertex and the arcs arriving to it; {\bf(b)} each vertex and the arcs departing from it; {\bf(c)} each vertex and its neighbors; {\bf(d)} each pair of oppositely oriented arcs of an edge and its end-vertices. Then, each ETGC of an $M(G)$ has a ditotal coloring. The Petrie dual of the resulting map $M(G^\leftrightarrow)$ is formed by monochromatic directed cycles.

We extend combinatorial summation (CoS) of connected Feynman diagrams to Hubbard mod- els with flavor-dependent propagators and counterterm shifts. The construction combines an interaction-occupancy mask with labels for the open fermion cycle, allowing equivalent endpoint his- tories to be accumulated while retaining the flavor and sublattice information needed by subsequent contractions. In the flavor-dependent version, denoted symmetry-broken CoS (SB-CoS), normal heads carry the cycle flavor and counterterm heads inherit the flavor at their interaction partner. We prove the correctness of the resulting recurrence and introduce terminal readouts that return all nonempty interaction-subset values from a shared circuit, as required by a many-configuration Markov-chain sampler. For three flavors, a second formulation uses determinant polynomials and a local interaction tensor, with connected coefficients extracted by subset recursion or Fourier interpo- lation and a support sieve. Applications to the two-spin honeycomb and three-flavor square-lattice Hubbard models illustrate the treatment of sublattice structure and non-uniform reference shifts.

We construct a fixed triangular family of noncommutative polynomials that spans the joint cyclic module of a real symmetric pair from any prescribed vector. In dimension at most $N$, when the second matrix has spectral diameter at most $N-1$, the family has $N(N+1)/2$ members, degree at most $N-1$, and a shared arithmetic circuit of size $O(N^2)$. No rank-dependent choice of coefficients is made. The proof uses a rational differential connection with a small self-adjoint residue: a primitive polynomial exterior vector for an invariant subspace must have an integer degree smaller than one. For connected trees, the probes and the pointed adjacency--degree reconstruction theorem give one explicit self-adjoint polynomial $p_N\in\mathbb{Z}\langle a,d\rangle$ whose trace separates all trees of order at most $N$. Its circuit has $O(N^2)$ gates, formal degree at most $3N-2$, and polynomial bit complexity. Only the original adjacency and degree matrices, the identity, and ordinary matrix arithmetic are used. A separate integer sum-of-squares construction gives a fixed positive semidefinite representative on which the all-one vector is cyclic; its degree is $O(N^2)$. A linear trace-degree lower bound shows that the first construction has the correct asymptotic degree scale.

We characterize the generic consistency of vertical systems in terms of the iterated circuits of the column matroid of the coefficient matrix. More precisely, we prove that such a system is generically consistent if and only if there exists an iterated circuit such that the dimension of the Minkowski sum of the associated Newton polytopes is equal to the rank of the coefficient matrix. We also give an equivalent characterization in terms of $G$-spanning flags of the dual matroid. With this, we derive a combinatorial formula for the codimension of the locus of parameter values for which the system has a solution, extending known results on dual defects of toric varieties. When this locus has codimension one, its defining equation yields a generalization of the sparse resultant to vertical systems, which applies also to $\mathcal{A}$-discriminants. Finally, we characterize when the solution set is contained, for generic parameter values, in a finite union of hyperplanes parallel to a coordinate hyperplane.

Exact conditional tests for contingency tables require sampling from fibers with fixed margins. Classical Markov basis MCMC is general but often impractical: computing full Markov bases that connect all fibers of a given constraint matrix can be infeasible and the resulting chains may converge slowly, especially in sparse settings or in presence of structural zeros. We introduce a SAT-based alternative that encodes fibers as Boolean circuits which allows modern SAT samplers to generate tables randomly. We analyze the sampling bias that SAT samplers may introduce, provide diagnostics, and propose practical mitigation. We propose hybrid MCMC schemes that combine SAT proposals with local moves to ensure correct stationary distributions which do not necessarily require connectivity via local moves which is particularly beneficial in presence of structural zeros. Across benchmarks, including small and involved tables with many structural zeros where pure Markov-basis methods underperform, our methods deliver reliable conditional p-values and often outperform samplers that rely on precomputed Markov bases.

The classical Descartes' rule of signs bounds the number of positive roots of a real polynomial in one variable by the number of sign changes of its coefficients. This bound only depends on the order of the exponents and on the signs of the coefficients, that is, on the oriented matroids of the exponent and coefficient matrices. We give a multivariate generalization of this rule to arbitrary fewnomial systems: for systems of $n$ real polynomial equations in $n$ variables with $n+k+1$ monomials, we obtain upper bounds for the number of nondegenerate positive solutions which only depend on sign conditions on the maximal minors of the coefficient matrix and of the exponent matrix and its liftings. For $k=1$ they are the Descartes rules of signs for circuits of Bihan-Dickenstein and Bihan-Dickenstein-Forsgård. Our approach starts from the Gale duality and the Khovanskii-Rolle argument of Bihan and Sottile, which reduce the problem to counting common zeros of iterated Jacobians in a polyhedron $Δ$. This polyhedron is a chamber of a hyperplane arrangement whose combinatorics is determined by the oriented matroid of the coefficient matrix. We bound the number of these zeros by the Bézout number minus the zeros forced in the other chambers by a topological degree, which we compute by a facet-count formula involving the signs of minors of a Gale dual matrix of the exponent matrix. This leads to a recursion over the flats of the arrangement, which ends on lines with the Descartes rules for circuits. In this way the coefficients enter the bounds through the combinatorics of the arrangement, and the exponents through sign conditions on minors.

Oxley conjectured that a matroid with a circuit-cocircuit intersection of size $k\geq 4$ has one of size $k-2$. Using the standard polynomial representation of the classical extended binary Golay code, we construct an identically self-dual binary matroid of rank $12$ on $24$ elements. Its circuit-cocircuit intersection sizes are exactly $0,2,4,6,8$, and $12$, disproving the conjecture for $k=12$. The proof is self-contained and does not rely on the full weight distribution of the code or on an enumeration of all circuits.

A square is a word $xx$, where $x$ is nonempty. We show that a circular word of length $n$ contains at most $\lfloor 3n/2 \rfloor$ distinct squares of length at most $n$. The proof combines known results relating squares to circuits in Rauzy graphs. The coefficient $3/2$ agrees with the known lower bound.

We prove Sabidussi's compatibility conjecture. Let $G$ be a finite connected multigraph in which every vertex has even degree and the minimum degree is at least four, and let $T$ be an Euler tour of $G$. The edges of $G$ can be partitioned into circuits (connected $2$-regular subgraphs) so that no circuit contains two edges used consecutively anywhere in $T$. In fact, the edges can be four-coloured so that every such pair receives different colours and every colour class has even degree at every vertex. We use a counting argument based on the Chevalley-Warning theorem to show that a four-colouring with the required properties exists. Splitting each colour class into circuits then gives the desired compatible decomposition. Formalization in Lean 4 is also available in the author's github.

The recently established duality between tensor, symmetric and exterior products of uniform matroids on the one hand, and abstract (bi-)rigidity matroids on the other hand, connects theorems and open questions from both areas, previously thought unrelated. In particular, a 1981 question of Mason asks if any two uniform matroids admit a freest product. Via duality, its reformulation due to Cruickshank, Jackson, Jordán and Tanigawa asks if the generic birigidity matroid is the freest abstract birigidity matroid for its parameters. A related conjecture of Jackson and Tanigawa asks if the generic $2$-hyperconnectivity matroid is the freest $\{K_4,K_{3,3}\}$-matroid. We show that these problems can be addressed effectively, and often solved completely for the class of linearly representable matroids. To this end, we introduce star-basis normal forms that allow to compare representations of abstract (bi-)rigidity matroids over the same field, resulting in a refinement of the weak order relation on the underlying matroids. Utilizing it, we prove that the generic birigidity matroid is the freest linearly representable abstract $(a,b)$-birigidity matroid. This confirms Mason's conjecture for linear matroids. We prove that every representable abstract $2$-rigidity matroid admits a rigidity matrix representation. This is a strengthening of the maximality property of the generic $2$-rigidity matroid. We also prove that every representable abstract $2$-rigidity matroid in which every copy of $K_{3,3}$ is a circuit admits a hyperconnectivity matrix representation. It follows that the generic rigidity and hyperconnectivity families $\mathcal{R}_2$ and $\mathcal{H}_2$ are the only linearly representable $2$-rigidity families.

In 1993, Ruzsa showed that for every $k \geq 2$, there exists a constant $C$ such that every subset $A \subseteq [N]$ of size at least $C N^{1/k}$ contains $2k$ distinct elements $a_1, \ldots, a_k, b_1, \ldots, b_k \in A$ such that $a_1 + \cdots + a_k = b_1 + \cdots + b_k$. We strengthen this result by proving that the elements $a_1, \ldots, a_k, b_1, \ldots, b_k$ can be chosen to have the additional property that $\{a_1, \ldots, a_k, b_1, \ldots, b_k\}$ has $2^{2k}-1$ distinct subset sums, with the only coincidence being that $\{a_1, \ldots, a_k\}$ and $\{b_1, \ldots, b_k\}$ have the same sum. Our proof also applies to any finite Abelian group of odd order $N$, and it provides a corresponding supersaturation result: that whenever $|A| \geq CN^{1/k}$, there are at least $Ω(|A|^{2k}/N)$ choices for $a_1, \ldots, a_k, b_1, \ldots, b_k \in A$ satisfying these properties. We prove a slightly weaker statement for Abelian groups of even order. We also apply our methods to the vector space setting and prove the following $\mathbb F_q$-analogue of the Bondy-Simonovits Theorem on the extremal number of even cycles in graphs: Any rank-$n$, simple, $\mathbb F_q$-representable matroid with no circuit of size exactly $2k$ has size at most $C q^{n/k}$ for some constant $C$ depending only on $q$ and $k$. When $q=2$, this is best possible up to the constant $C$ for all $k \geq 2$. Our methods also apply to the original graph setting and give a new proof of the Bondy-Simonovits Theorem and its supersaturation version.