betaFeedback

WorksMathematics

Uniquely balanced polyhedral complexes are highly connected

Abstract

We establish higher connectivity of polyhedral complexes admitting a unique positive balanced top cycle. Our main result shows that, after quotienting by common lineality, every skeleton satisfies the corresponding higher connectivity property. The proof bounds the number of connected components surviving facet deletions by the dimension of the top-cycle space using homological and linear-algebraic arguments. As applications, we prove higher connectivity for all skeleta of arbitrary pointed polyhedral structures on tropical linear spaces of valuated matroids and Bergman fans including nonrealizable cases. These results answer questions raised by Maclagan and Yu.

Introduction

The tropicalization of an irreducible algebraic variety is connected through codimension one [9]: any two maximal cells can be joined by a sequence of maximal cells meeting along codimension-one faces. This connectivity underlies algorithms for computing tropical varieties [BJS+07]. [15] strengthens this property by proving higher connectivity: for a dd-dimensional irreducible variety over a field of characteristic zero, its tropicalization is (d−ℓ)(d-\ell)-connected through codimension one where ℓ\ell is the dimension of the lineality space of the chosen polyhedral complex (see [GHM+21] for positive characteristic). This means that connectivity persists after deleting any d−ℓ−1d-\ell-1 closed maximal cells. Its significance lies in the use of algebraic geometry methods, notably a tropical Bertini theorem, to establish strong combinatorial consequences from algebraic irreducibility.

[15 Section 5] asks whether tropical irreducibility suffices for higher connectivity. We approach this question through the space of balanced top-dimensional cycles. For a pure dd-dimensional polyhedral complex Δ\Delta, let ker⁡∂dΔ\ker\partial_d^\Delta denote its top-cycle space, defined using the cellular boundary map with coefficients in the top exterior powers of the direction spaces of its facets. We say that Δ\Delta is uniquely balanced if it admits a positive top cycle and dim⁡ker⁡∂dΔ=1\dim\ker\partial_d^\Delta=1, as its positive top cycle is unique up to scaling. Our first main result shows that this condition forces higher connectivity, without any rationality assumption.

Theorem 1. Let Δ\Delta be a finite pure dd-dimensional real polyhedral complex admitting a positive top cycle, and suppose dim⁡ker⁡∂dΔ=1\dim\ker\partial_d^\Delta=1. Then, the lineality spaces of all cells of Δ\Delta coincide, defining a subspace UU such that Δ/U\Delta/U is pointed. For every 1≤k≤d−dim⁡U1 \le k \le d-\dim U, its kk-skeleton is kk-connected through codimension one. Equivalently, the (dim⁡U+k)(\dim U+k)-skeleton of Δ\Delta is kk-connected through codimension one.

In the usual rational setting of tropical geometry, the uniquely balanced condition is equivalent to tropical irreducibility. Consequently, our theorem establishes higher connectivity for tropically irreducible balanced complexes. The distinction from algebraic irreducibility is essential because an algebraically irreducible variety may not have a tropically irreducible tropicalization. For example, the irreducible curve

X=V((x+1)(y+1)+t)⊂(C((t))−{0})2X = V((x+1)(y+1)+t) \subset(\mathbb{C}((t))-\{0\})^2

has tropicalization Trop⁡(X)={u=0}∪{v=0}\operatorname{Trop}(X)=\{u=0\}\cup\{v=0\} where (u,v)(u,v) are the tropical coordinates. This tropicalization has two independent balanced top cycles. Conversely, the Bergman fan of the Vámos matroid is uniquely balanced but cannot arise as the tropicalization of an algebraic variety. It illustrates that our work applies beyond the scope of the algebraic irreducibility result (see Example 21).

Whereas [15] uses a tropical Bertini theorem, we prove by filling boundaries on deleted facets and recovering the balancing condition through linear algebra. The key ingredient is Proposition 7 which bounds the number of connected components surviving the deletion of at most d−1d-1 facets by the dimension of the top-cycle space. The idea is to construct an independent top cycle from each surviving component by filling the boundary of the restricted positive cycle on the deleted facets. We apply Theorem 1 to tropical linear spaces of valuated matroids and Bergman fans. In particular, we establish higher connectivity for all skeleta of arbitrary pointed polyhedral structures on these spaces, including nonrealizable cases (see Theorem 18 and corollary 19). These results answer the corresponding questions raised in [15 Section 5].

The paper is organized as follows. Section 2 introduces balanced top cycles and their behavior under subdivisions and quotients. Section 3 establishes the bound on surviving connected components using filling arguments. Section 4 proves the connectivity of lower-dimensional skeleta and Theorem 1. Section 5 applies the results to tropical linear spaces and Bergman fans. Section 6 examines coordinate-projection matroids and related examples.

Balanced cycles and polyhedral structures

Throughout, VV is a finite-dimensional real vector space. A polyhedral complex is a finite collection of closed convex polyhedra satisfying the usual face and intersection conditions. We make no rationality assumption; this allows nonrational subdivisions and quotients by arbitrary real subspaces. For a polyhedral complex Δ\Delta, its support is ∣Δ∣=⋃σ∈Δσ|\Delta|=\bigcup_{\sigma\in\Delta}\sigma, and Δq\Delta_q denotes the set of its qq-dimensional cells. If Δ\Delta is pure of dimension dd, its facets and ridges are its cells of dimension dd and d−1d-1, respectively. Its kk-skeleton consists of the cells of dimension at most kk; if Δ\Delta is pointed, it is pure of dimension kk. The lineality space of a polyhedron σ\sigma is the largest linear subspace UU with σ+U=σ\sigma+U=\sigma. A polyhedron is pointed if its lineality space is zero, equivalently if it contains no affine line, and a complex is pointed if all its cells are. If a linear subspace UU lies in the lineality space of every cell of Δ\Delta, the images of the cells in V/UV/U form a polyhedral complex Δ/U\Delta/U with the same face poset.

Definition 1. A hypergraph is connected if any two vertices are joined by a sequence of vertices in which consecutive vertices lie in a common hyperedge; we regard the empty hypergraph as connected. A hypergraph is strongly pp-vertex-connected if it remains connected after deleting any at most p−1p-1 vertices together with every hyperedge containing one of them. The facet-ridge hypergraph of a pure polyhedral complex Δ\Delta has the facets of Δ\Delta as vertices and, for each ridge τ\tau, the hyperedge consisting of the facets containing τ\tau. We say that Δ\Delta is kk-connected through codimension one if its facet-ridge hypergraph is strongly kk-vertex-connected.

Thus, deleting a facet removes every ridge it contains: a ridge of a deleted facet cannot be used to connect surviving facets, even if two other facets containing it survive. We call this strong deletion.

Let Δ\Delta be a pure dd-dimensional polyhedral complex in VV. For each cell σ∈Δ\sigma\in\Delta, let Tσ=span⁡(σ−σ)T_\sigma= \operatorname{span}(\sigma-\sigma) denote its direction space. Fix an orientation for each cell of Δ\Delta. When V=N⊗RV=N\otimes\mathbb{R} for a lattice NN, we call TσT_\sigma rational if it is spanned by vectors in NN. For a fixed real vector space AA, the cellular Borel-Moore chains of Δ\Delta with coefficients in AA are

CqBM(Δ;A)=⨁σ∈ΔqA[σ]C^{\mathrm{BM}}_q(\Delta; A) = \bigoplus_{\sigma\in\Delta_q} A[\sigma]

with boundary operator

∂(a[σ])=∑τ<σϵστa[τ]\partial(a[\sigma])=\sum_{\tau<\sigma}\epsilon_{\sigma\tau}a[\tau]

where the sum is over codimension-one faces τ\tau of σ\sigma and ϵστ\epsilon_{\sigma\tau} is the incidence sign.

Remark 2. Unlike ordinary cellular chains, Borel-Moore chains allow unbounded cells. We use the same coefficient space AA for every cell, so the boundary maps involve only incidence signs, unlike tropical chains whose coefficient spaces depend on the cells [2].

Taking A=⋀dVA=\bigwedge^d V, we restrict the boundary map to the subspace ⨁σ∈Δd⋀dTσ[σ]\bigoplus_{\sigma\in\Delta_d}\bigwedge^d T_\sigma[\sigma] and write

∂dΔ:⨁σ∈Δd⋀dTσ[σ]⟶Cd−1BM(Δ;A).(1)\partial_d^\Delta:\bigoplus_{\sigma\in\Delta_d}\bigwedge^d T_\sigma[\sigma]\longrightarrow C^{\mathrm{BM}}_{d-1}(\Delta; A). \tag*{(1)}

When it is clear from the context, we suppress the Δ\Delta and denote it by ∂d\partial_d. We call ker⁡∂dΔ\ker\partial_d^\Delta the top-cycle space of Δ\Delta. For each σ∈Δd\sigma\in\Delta_d, choose a generator ξσ∈⋀dTσ\xi_\sigma\in\bigwedge^d T_\sigma compatible with the chosen orientation of σ\sigma. A cycle z=∑σ∈Δdwσξσ[σ]∈ker⁡∂dz=\sum_{\sigma\in\Delta_d}w_\sigma\xi_\sigma[\sigma]\in\ker\partial_d is called positive if wσ>0w_\sigma>0 for every σ∈Δd\sigma\in\Delta_d. The positivity of a cycle is independent of the chosen orientations and positive rescalings of ξσ\xi_\sigma. We call Δ\Delta uniquely balanced if it admits a positive top cycle and dim⁡ker⁡∂dΔ=1\dim\ker\partial_d^\Delta=1. For rational fans, this agrees with [2 Definition 3.12] since in top degree the tropical Borel-Moore homology there coincides with ker⁡∂d\ker\partial_d.

Suppose that V=N⊗RV=N\otimes\mathbb{R} for a lattice NN and that all direction spaces are rational. Choose primitive lattice generators ξσ\xi_\sigma of ⋀d(Tσ∩N)\bigwedge^d(T_\sigma\cap N) compatible with the orientations. Then, ∂z=0\partial z=0 is the usual balancing condition ∑τ⊂σwσuσ/τ=0\sum_{\tau\subset\sigma}w_\sigma u_{\sigma/\tau}=0 in V/TτV/T_\tau at each ridge τ\tau where uσ/τu_{\sigma/\tau} is the primitive normal vector of τ\tau in σ\sigma [2 Definition 3.12 and Example 3.14].

Definition 2. A positive balanced polyhedral support with rational direction spaces is tropically irreducible if it cannot be expressed as the union of two proper positive balanced supports of the same dimension.

To relate tropical irreducibility to the uniquely balanced condition, we show that the top-cycle space is preserved under subdivision and quotient.

Lemma 3. Let Δ\Delta be a pure dd-dimensional polyhedral complex and Δ′\Delta' a subdivision of Δ\Delta. Then, there is a natural isomorphism ker⁡∂dΔ≅ker⁡∂dΔ′\ker\partial_d^\Delta\cong\ker\partial_d^{\Delta'} preserving positivity. Moreover, if a real subspace UU lies in the linearity space of every cell, quotienting gives an isomorphism ker⁡∂dΔ≅ker⁡∂d−dim⁡UΔ/U\ker\partial_d^\Delta\cong\ker\partial_{d-\dim U}^{\Delta/U} preserving positivity.

Proof. Suppose that a maximal cell σ\sigma is subdivided into σ1,…,σm\sigma_1,\ldots,\sigma_m with inherited orientations and the same exterior generator ξσi=ξσ\xi_{\sigma_i}=\xi_\sigma. This is possible since Tσi⊂TσT_{\sigma_i}\subset T_\sigma for all ii. Let wσξσw_\sigma\xi_\sigma be the coefficient of σ\sigma in the cycle. For each σi\sigma_i, we impose the same exterior coefficients wσiξσi=wσξσw_{\sigma_i}\xi_{\sigma_i}=w_\sigma\xi_\sigma for all i=1,…,mi=1,\ldots,m. Let τ\tau be a new interior ridge shared by two subfacets σi\sigma_i and σj\sigma_j. Then, we have ϵσiτ=−ϵσjτ\epsilon_{\sigma_i\tau}=-\epsilon_{\sigma_j\tau}, and hence ϵσiτwσξσ+ϵσjτwσξσ=0\epsilon_{\sigma_i\tau}w_\sigma\xi_\sigma+\epsilon_{\sigma_j\tau}w_\sigma\xi_\sigma=0. Along the subdivided original ridges, the boundary equations agree with the original ones and hence vanish.

Conversely, let z′=w1ξσ[σ1]+w2ξσ[σ2]+⋯z'=w_1\xi_\sigma[\sigma_1]+w_2\xi_\sigma[\sigma_2]+\cdots be a cycle on Δ′\Delta' with coefficient wiξσw_i\xi_\sigma on the subfacets of σ\sigma. At any new interior ridge τ\tau shared by σi\sigma_i and σj\sigma_j, the cycle condition gives

0=(∂z′)∣τ=ϵσiτ(wi−wj)ξσ,0=(\partial z')|_\tau=\epsilon_{\sigma_i\tau}(w_i-w_j)\xi_\sigma,

and hence wi=wjw_i=w_j. Since the subfacets of σ\sigma are connected through interior ridges, we have w1=⋯=wmw_1=\cdots=w_m. Thus, a subdivision does not induce an independent weight, and the cycle spaces are naturally isomorphic. Since the weights are unchanged, positivity is also preserved.

For the quotient, fix an oriented volume element ηU∈⋀dim⁡UU\eta_U\in\bigwedge^{\dim U}U. The map

vˉ1∧⋯∧vˉd−dim⁡U⟼ηU∧v1∧⋯∧vd−dim⁡U\bar v_1\wedge\cdots\wedge\bar v_{d-\dim U}\longmapsto\eta_U\wedge v_1\wedge\cdots\wedge v_{d-\dim U}

identifies the top exterior lines. Note that it is independent of the lift. With product orientations, the boundary equations correspond up to a common sign, so zero boundaries and positivity are preserved. Since every cell is invariant under UU, quotienting preserves the face and incidence relations.

Lemma 4. For a complex Δ\Delta with rational direction spaces admitting a positive balanced rational weighting, tropical irreducibility is equivalent to the space of top-dimensional cycles being one-dimensional, that is, dim⁡ker⁡∂d=1\dim\ker\partial_d=1.

Proof. Since the boundary map ∂d\partial_d is represented by a rational matrix, its kernel has a basis over Q\mathbb{Q}. Let ww be a positive rational weighting and aa a rational balanced weighting not proportional to ww. Let s=min⁡σaσwσs=\min_\sigma\frac{a_\sigma}{w_\sigma} and t=max⁡σaσwσt=\max_\sigma\frac{a_\sigma}{w_\sigma}. Then, we have s<ts<t and a−swa-sw and tw−atw-a are nonnegative balanced weightings. Each has nonempty proper support and their supports cover ∣Δ∣|\Delta|. A contradiction to tropical irreducibility.

Conversely, a proper balanced sub-support gives a nonproportional weighting after a common refinement; the subdivision assertion in Lemma 3 identifies it with a weighting on Δ\Delta, and hence dim⁡ker⁡∂d≥2\dim\ker\partial_d\ge2. This is also the equivalence in [6 Proposition 2.4 (i) and (iii)].

For rational direction spaces, the uniquely balanced condition is also equivalent to strong extremality of balanced weighted complexes in the sense of [7 Definition 2.11].

Filling cycles on deleted facets

Our goal is to bound the number of connected components remaining after deleting facets by the dimension of the top-cycle space. The idea is to construct an independent cycle from each surviving component. Restricting a positive cycle to a surviving component may destroy the balancing condition, but the resulting boundary is supported on the union of the deleted facets. We first show that such boundaries can be “filled”. The bound of d−1d-1 deleted facets arises from two complementary constraints: one topological and the other geometric.

Lemma 5. Let P1,…,PmP_1,\ldots,P_m be pointed closed convex polyhedra, m≥1m \ge1. For any finite-dimensional real vector space AA,

HqBM(⋃i=1mPi;A)=0H^{\mathrm{BM}}_q\left(\bigcup_{i=1}^{m} P_i; A\right)=0

for q≥mq \ge m.

Proof. We first consider a single convex polyhedron. A bounded nonempty convex polyhedron is contractible, so its Borel-Moore homology vanishes in positive degrees. We show that an unbounded pointed polyhedron PP has zero Borel-Moore homology in every degree. To show this, choose hh strictly positive on rec⁡(P)∖{0}\operatorname{rec}(P)\setminus\{0\}, and RR larger than its values at all vertices. The hyperplane {h=R}\{h=R\} cuts off the unbounded parts without removing any vertices. Then, P∩{h≤R}P\cap\{h\le R\} and P∩{h=R}P\cap\{h=R\} are nonempty compact convex polyhedra. The cellular Borel-Moore chain complex of PP is isomorphic to the relative cellular chain complex of (P∩{h≤R},P∩{h=R})(P\cap\{h\le R\},P\cap\{h=R\}) since the faces of PP correspond to the truncated faces not contained in P∩{h=R}P\cap\{h=R\}. Because both P∩{h≤R}P\cap\{h\le R\} and P∩{h=R}P\cap\{h=R\} are contractible, their relative homology vanishes. The same argument applies to any finite subdivision by choosing RR above all its vertices.

We now induct on mm. Let D=P1∪⋯∪Pm−1D=P_1\cup\cdots\cup P_{m-1}. The cellular Mayer-Vietoris sequence for D∪PmD\cup P_m contains

HqBM(D;A)⊕HqBM(Pm;A)⟶HqBM(D∪Pm;A)⟶Hq−1BM(D∩Pm;A).H^{\mathrm{BM}}_q(D;A)\oplus H^{\mathrm{BM}}_q(P_m;A)\longrightarrow H^{\mathrm{BM}}_q(D\cup P_m;A)\longrightarrow H^{\mathrm{BM}}_{q-1}(D\cap P_m;A).

For q≥mq\ge m, both HqBM(D;A)H^{\mathrm{BM}}_q(D;A) and HqBM(Pm;A)H^{\mathrm{BM}}_q(P_m;A) vanish by the induction hypothesis and the base case, respectively. Moreover, D∩Pm=⋃i=1m−1(Pi∩Pm)D\cap P_m=\bigcup_{i=1}^{m-1}(P_i\cap P_m) is a union of at most m−1m-1 pointed convex polyhedra, after discarding empty intersections. Thus, its (q−1)(q-1)-th Borel-Moore homology vanishes by induction. The result follows. □

In our situation, if DD is the union of at most d−1d-1 deleted dd-dimensional facets, then Hd−1BM(D;A)=0H^{\mathrm{BM}}_{d-1}(D;A)=0 by Lemma 5. Hence, for any (d−1)(d-1)-cycle bb on DD with coefficients in AA, there exists a dd-chain yy supported on DD such that ∂y=b\partial y=b. However, when A=⋀dVA=\bigwedge^d V, such a filling need not have coefficients in the lines ⋀dTσ\bigwedge^d T_\sigma required by (1). The following lemma rules this out.

Lemma 6. Let Δ\Delta be a pure pointed dd-dimensional polyhedral complex with d≥2d\ge2, and let G\mathcal{G} consist of at most d−1d-1 facets. Each σ∈G\sigma\in\mathcal{G} has two ridges with distinct direction hyperplanes, neither contained in another member of G\mathcal{G}.

Proof. Fix σ∈G\sigma\in\mathcal{G}. At a vertex of σ\sigma, which exists as σ\sigma is pointed, choose dd incident ridges whose normal covectors in Tσ∗T^*_\sigma are linearly independent. Another deleted facet can contain at most one of these ridges since its intersection with σ\sigma is a proper face of σ\sigma.

There are at most d−2d-2 other deleted facets. Hence, at least two of the selected ridges are not contained in any other member of G\mathcal{G}. Their direction hyperplanes are distinct by construction. □

These two lemmas yield the following bound.

Proposition 7. Let Δ\Delta be a finite pure pointed dd-dimensional real polyhedral complex admitting a positive top cycle. After strong deletion of at most d−1d-1 facets, the number of nonempty components of its surviving facet-ridge hypergraph is at most dim⁡ker⁡∂d\dim\ker\partial_d.

Proof. Fix a positive top cycle z=∑σ∈Δdwσξσ[σ]z = \sum_{\sigma\in\Delta^d} w_\sigma\xi_\sigma[\sigma] with wσ>0w_\sigma> 0. If no facets are deleted, the restriction of zz to each facet-ridge component is already a cycle. These cycles are linearly independent, proving the claim. This also covers d=1d = 1.

Let G\mathcal{G} be a nonempty collection of at most d−1d - 1 deleted facets, and let DD be their union, including all faces. For each surviving component CC, define the chain zC=∑σ∈Cwσξσ[σ]z_C = \sum_{\sigma\in C} w_\sigma\xi_\sigma[\sigma]. This may not be a cycle because restricting zz to CC may destroy the balancing equations.

However, at every ridge outside DD, all incident facets survive and belong to the same component. Thus, the original balancing equation still holds there. Hence, bC:=∂zCb_C := \partial z_C is supported on DD. Since ∂bC=0\partial b_C = 0, it is a (d−1)(d-1)-cycle on DD with constant coefficients A=⋀dVA = \bigwedge^d V.

By Lemma 5, Hd−1BM(D;A)=0H^{\mathrm{BM}}_{d-1}(D; A) = 0. Therefore, there exists a filling yC=∑σ∈Gaσ[σ]y_C = \sum_{\sigma\in G} a_\sigma[\sigma] with aσ∈⋀dVa_\sigma\in\bigwedge^d V such that ∂yC=bC\partial y_C = b_C.

We show that the filling coefficients satisfy aσ∈⋀dTσa_\sigma\in\bigwedge^d T_\sigma. Fix σ∈G\sigma\in\mathcal{G} and choose the two ridges τ1,τ2\tau_1,\tau_2 as in Lemma 6. At either ridge, σ\sigma is the only deleted facet contributing to ∂yC\partial y_C. On the other hand, every contribution to bCb_C at that ridge belongs to (⋀d−1Tτi)∧V(\bigwedge^{d-1} T_{\tau_i}) \wedge V. Since ∂yC=bC\partial y_C = b_C, we obtain

aσ∈((⋀d−1Tτ1)∧V)∩((⋀d−1Tτ2)∧V).a_\sigma\in\left((\bigwedge^{d-1} T_{\tau_1}) \wedge V\right) \cap\left((\bigwedge^{d-1} T_{\tau_2}) \wedge V\right).

The two direction hyperplanes are distinct in TσT_\sigma. Choosing a basis adapted to both hyperplanes shows that

((⋀d−1Tτ1)∧V)∩((⋀d−1Tτ2)∧V)=⋀dTσ.\left((\bigwedge^{d-1} T_{\tau_1}) \wedge V\right) \cap\left((\bigwedge^{d-1} T_{\tau_2}) \wedge V\right) = \bigwedge^d T_\sigma.

Thus, every coefficient of yCy_C belongs to the required top exterior line.

Hence, z~C=zC−yC\widetilde{z}_C = z_C - y_C is a top cycle since ∂z~C=bC−bC=0\partial\widetilde{z}_C = b_C - b_C = 0. Since yCy_C is supported only on deleted facets, z~C\widetilde{z}_C agrees with the original positive cycle on CC and vanishes on every other surviving component. If ∑CλCz~C=0\sum_C \lambda_C\widetilde{z}_C = 0, then comparison of coefficients on any facet σ∈C\sigma\in C gives λCwσ=0\lambda_C w_\sigma= 0. Since wσ>0w_\sigma> 0, we have λC=0\lambda_C = 0 for every CC. Hence, these cycles are linearly independent. We conclude the proof. □

The following example shows that the bound in Proposition 7 is sharp and illustrates how the boundary created by restricting a cycle to a surviving component can be filled using the deleted facet.

Example 8. For linearly independent u,v,wu,v,w, let P(u,v,w)P(u,v,w) be the fan whose rays are generated by u,v,w,−u−v−wu,v,w,-u-v-w and whose facets are the cones on every pair of these rays. Set X=P(e1,e2,e3)∪P(e1,2e2,e4)X = P(e_1,e_2,e_3) \cup P(e_1,2e_2,e_4), σ=pos⁡(e1,e2)\sigma= \operatorname{pos}(e_1,e_2) and ξ=e1∧e2\xi= e_1 \wedge e_2 (see Figure 1). The two fans meet exactly in σ\sigma and its faces. Each fan is a linear image of the standard tropical plane. Normalize its positive cycle ziz_i to have exterior coefficient ξ\xi on σ\sigma, and take z=z1+z2z = z_1 + z_2. Thus, the coefficient of zz on σ\sigma is 2ξ2\xi. Strong deletion of σ\sigma leaves precisely two components C1C_1 and C2C_2. With orientations chosen so that ∂[σ]=[a]−[b]\partial[\sigma] = [a] - [b], we have ∂zCi=−ξ∂[σ]\partial z_{C_i} = -\xi\partial[\sigma] and z~Ci=zCi+ξ[σ]=zi\widetilde{z}_{C_i} = z_{C_i} + \xi[\sigma] = z_i. The top-cycle space has dimension exactly two: the balancing equations at the two nonshared rays of each plane force its five surviving facet coefficients to be proportional to those of ziz_i. After subtracting these two multiples, any remaining cycle is supported on σ\sigma alone and hence is zero. Thus, the bound in Proposition 7 is attained.

Lemma 5 also gives a lower bound on the number of facets supporting a nonzero top cycle.

Constructing a top cycle from a surviving component

Figure 1. Constructing a top cycle from a surviving component in Proposition 7. Vertices represent rays and edges represent two-dimensional facets. The dashed edge represents the deleted facet σ\sigma. (a) Deleting σ\sigma separates the two components. (b) Restricting the original cycle to one component creates boundary coefficients −ξ-\xi and +ξ+\xi. (c) Adding ξ[σ]\xi[\sigma] cancels this boundary and completes the cycle.

Lemma 9. A pointed pure dd-dimensional complex admitting a nonzero top cycle has at least d+1d+1 facets.

Proof. If there were m≤dm \le d facets, Lemma 5 would give HdBM(∣Δ∣;⋀dV)=0H_d^{\mathrm{BM}}(|\Delta|;\bigwedge^d V)=0. But a nonzero top cycle represents a nonzero homology class since there are no (d+1)(d+1)-cells. □

To the skeleta

We show how connectivity through codimension one descends to lower-dimensional skeleta. Related higher-connectivity results for skeleta of arbitrary convex polytopes were studied in [4, 12, 16]. The main difficulty is that deleting a lower-dimensional cell may remove either a vertex or a hyperedge from the facet-ridge hypergraph of a higher-dimensional cell. The following lemma shows that either deletion drops strong vertex connectivity by at most one.

Lemma 10. Let HH be a strongly pp-vertex-connected hypergraph with at least p+1p+1 vertices for p≥2p \ge2. Then, H−hH-h is strongly (p−1)(p-1)-vertex-connected for every hyperedge hh. Consequently, deleting aa vertices strongly and then bb hyperedges leaves HH connected whenever a+b≤p−1a+b \le p-1.

Proof. Fix a hyperedge hh of HH. First, we delete at most p−2p-2 vertices and call the resulting hypergraph KK. KK has at least three vertices and remains connected even after one additional strong vertex deletion. If hh was removed by the vertex deletions, then K−h=KK-h=K which is connected. Otherwise, if K−hK-h were disconnected, hh would meet every component because KK is connected. Choose v∈hv \in h so that at least two components of K−hK-h remain nonempty after removing vv: when there are two components, choose vv in one with at least two vertices, and when there are at least three components, any choice works. Strong deletion of vv also deletes hh, contradicting connectivity of K−v=(K−h)−vK-v=(K-h)-v. Iterating this observation after the first aa vertex deletions proves the claim. □

:::

We next apply this observation to the facet-ridge graphs of pointed polyhedra.

Lemma 11. Let QQ be a pointed polyhedron of dimension q≥2q \ge2. Its facet-ridge hypergraph is (q−1)(q-1)-connected and has at least qq vertices.

Proof. Homogenize QQ to a pointed cone in one higher dimension and cut by a strictly positive affine section. The resulting compact qq-polytope Q‾\overline{Q} has a distinguished face F∞F_{\infty} at height zero, possibly empty. Faces of Q‾\overline{Q} not contained in F∞F_{\infty} correspond to the faces of QQ by projective transformation.

The facet-ridge graph of Q‾\overline{Q} is the vertex-edge graph of its polar, hence is qq-connected by Balinski’s theorem [5]. The ridges of Q‾\overline{Q} contained in F∞F_{\infty} are not faces of QQ, and every other ridge lies in exactly two facets of QQ, so the facet-ridge hypergraph of QQ is a graph. Hence, if dim⁡F∞=q−1\dim F_{\infty}=q-1, passing to QQ deletes one vertex and its incident edges from the facet-ridge graph of Q‾\overline{Q}. If dim⁡F∞=q−2\dim F_{\infty}=q-2, passing to QQ deletes one edge. If its dimension is smaller, or QQ is bounded, it deletes neither vertices nor edges. Lemma 10 proves (q−1)(q-1)-connectivity. A qq-polytope has at least q+1q+1 facets, and at most one is lost, proving the second claim.

We can descend from the connectivity of a polyhedral complex to that of its lower-dimensional skeleta.

Proposition 12. Let Δ\Delta be a finite pure pointed dd-dimensional polyhedral complex with at least d+1d+1 facets. If Δ\Delta is dd-connected through codimension one, then its kk-skeleton is kk-connected through codimension one for every 1≤k≤d1 \le k \le d.

Proof. Use descending induction, assuming the (k+1)(k+1)-skeleton is (k+1)(k+1)-connected. Delete at most k−1k-1 closed kk-cells, denoted G\mathcal{G}. In the facet-ridge hypergraph of the (k+1)(k+1)-skeleton, these kk-cells are hyperedges. Deleting them preserves connectivity by Lemma 10. This hypergraph has at least k+2k+2 vertices: when k=d−1k=d-1, this is assumed; for k≤d−2k \le d-2, we apply Lemma 11 to any (k+2)(k+2)-cell.

To turn this connectivity of (k+1)(k+1)-cells into connectivity of their kk-dimensional faces, we need to connect surviving kk-faces within each (k+1)(k+1)-cell. Fix such a (k+1)(k+1)-cell QQ. By Lemma 11, its facet-ridge graph is kk-connected and has at least k+1k+1 vertices. For each γ∈G\gamma\in\mathcal{G}, the intersection γ∩Q\gamma\cap Q is a single proper face or is empty. If it has dimension kk, it deletes one graph vertex and its incident edges in the facet-ridge graph. If it has dimension k−1k-1, it deletes one edge. A smaller intersection deletes neither. Thus, as we deleted at most k−1k-1 kk-cells, at most k−1k-1 vertex or edge deletions occur. By Lemma 10, the surviving kk-faces of QQ remain connected through allowed (k−1)(k-1)-faces.

Given two surviving kk-cells α\alpha and β\beta, choose (k+1)(k+1)-cells QαQ_{\alpha} and QβQ_{\beta} containing α\alpha and β\beta, respectively. Because Δ\Delta is pure, such cells exist. By connectivity of the (k+1)(k+1)-skeleton after the hyperedge deletions, QαQ_{\alpha} and QβQ_{\beta} can be joined by a sequence of (k+1)(k+1)-cells, with consecutive cells sharing a surviving kk-cell (see Figure 2). Within each successive (k+1)(k+1)-cell along this sequence, connect the incoming and outgoing kk-faces using the local connectivity established above. Concatenating these paths gives a path between the original two kk-cells. This also works for k=1k=1, when there are no deletions, and proves the induction.

Schematic illustration of the proof of Proposition 12

Figure 2. Schematic illustration of the proof of Proposition 12. Connectivity of the (k+1)(k+1)-skeleton gives a path Q0,Q1,Q2Q_0,Q_1,Q_2 where consecutive cells share surviving kk-faces ρ1,ρ2\rho_1,\rho_2. Within each QiQ_i, local facet-ridge connectivity provides a path between the relevant kk-faces. Concatenating these paths connects α\alpha to β\beta in the kk-skeleton.

We now combine the connectivity bound of Proposition 7 with the descent result of Proposition 12. When the top-cycle space is one-dimensional, the resulting connectivity extends to every skeleton after quotienting by the common linearity space.

Proof of Theorem 1. Restricting a positive top cycle to each facet-ridge component gives a nonzero top cycle since all facets incident to a ridge belong to the same component. These cycles have disjoint supports, and so they are linearly independent. Since dim⁡ker⁡∂dΔ=1\dim\ker\partial_d^\Delta= 1, there can be only one component. A nonempty face of a polyhedron has the same lineality as that polyhedron. Hence, facets sharing a ridge have the same lineality space. Then, connectivity through codimension one and purity imply that all cells of Δ\Delta share a common lineality space UU. By Lemma 3, the pointed quotient has a positive top cycle and a one-dimensional top-cycle space. Apply Propositions 7 and 12 and lemma 9 in dimension d−dim⁡Ud-\dim U. When d−dim⁡U=0d-\dim U=0, the connectivity assertion is vacuous. □

For complexes with rational direction spaces, unique balancing is equivalent to tropical irreducibility by Lemma 4. Thus, we obtain the following consequence.

Corollary 13. A finite pure tropically irreducible balanced complex of dimension dd, with common lineality of dimension ℓ\ell, is (d−ℓ)(d-\ell)-connected through codimension one. Its (ℓ+k)(\ell+k)-skeleton is kk-connected through codimension one for 1≤k≤d−ℓ1 \leq k \leq d-\ell.

Proof. Apply Lemma 4 and theorem 1. □

The general theorem also applies to arbitrary pointed polyhedral subdivisions of Rn\mathbb{R}^n.

Corollary 14. Let Δ\Delta be a finite pointed polyhedral complex with support Rn\mathbb{R}^n. For every 1≤k≤n1 \leq k \leq n, its kk-skeleton is kk-connected through codimension one.

Proof. The complex Δ\Delta is pure of dimension nn, and every facet has direction space Rn\mathbb{R}^n. Hence, ker⁡∂nΔ=HnBM(Δ;⋀nRn)≅R\ker\partial_n^\Delta= H_n^{\mathrm{BM}}(\Delta; \bigwedge^n \mathbb{R}^n) \cong\mathbb{R}. The fundamental cycle induced by an orientation of Rn\mathbb{R}^n is positive. Since Δ\Delta is pointed, Theorem 1 applies with U=0.□U=0. \square

Applied to the normal fan of a dd-polytope PP, whose (d−k)(d-k)-skeleton has the kk-faces of PP as facets and the (k+1)(k+1)-faces as ridges, Corollary 14 recovers [12 Theorem 1.1], and hence [15 Corollary 2]. It also covers non-polytopal complete fans and pointed subdivisions of Rn\mathbb{R}^n that are not fans.

Tropical linear spaces and Bergman fans

Theorem 1 requires the global condition that dim⁡ker⁡∂dΔ=1\dim\ker\partial_d^\Delta=1, involving balancing equations across all facets and ridges. However, it can be obtained locally when the balancing relations at individual ridges are unique and the complex is connected through codimension one. The following proposition makes this observation precise.

Proposition 15. Let Δ\Delta be a finite pure dd-dimensional complex with a positive top cycle

z=∑σwσξσ[σ].z=\sum_\sigma w_\sigma\xi_\sigma[\sigma].

For each ridge τ\tau, define a linear map

Rτ:R{σ∈Δd∣τ⊂σ}⟶⋀dVR_\tau:\mathbb{R}^{\{\sigma\in\Delta_d\mid\tau\subset\sigma\}}\longrightarrow\bigwedge^d V

with (aσ)σ↦∑τ⊂σϵστaσξσ(a_\sigma)_\sigma\mapsto\sum_{\tau\subset\sigma}\epsilon_{\sigma\tau}a_\sigma\xi_\sigma. Thus, ker⁡Rτ\ker R_\tau consists of the weightings balanced at τ\tau. Suppose dim⁡ker⁡Rτ=1\dim\ker R_\tau=1 for every ridge. Then, dim⁡ker⁡∂dΔ\dim\ker\partial_d^\Delta equals the number of facet-ridge components of Δ\Delta. In particular, if Δ\Delta is connected through codimension one, then ker⁡∂dΔ=Rz\ker\partial_d^\Delta=\mathbb{R}z, and the conclusions of Theorem 1 hold.

Proof. Let a=∑σaσξσ[σ]a=\sum_\sigma a_\sigma\xi_\sigma[\sigma] be a top cycle. At a ridge τ\tau, both (aσ)τ⊂σ(a_\sigma)_{\tau\subset\sigma} and the strictly positive vector (wσ)τ⊂σ(w_\sigma)_{\tau\subset\sigma} belong to ker⁡Rτ\ker R_\tau. Hence, aσwσ\frac{a_\sigma}{w_\sigma} has the same value for all facets σ\sigma containing τ\tau. These ratios are constant on each facet-ridge component.

Conversely, restricting zz to any such component gives a top cycle because all facets incident to any ridge belong to the same component. These restrictions have disjoint facet supports and form a basis of ker⁡∂dΔ\ker\partial_d^\Delta. □\square

Remark 16. Under the assumptions of Proposition 15, the existence of a positive top cycle and the condition dim⁡ker⁡Rτ=1\dim\ker R_\tau=1 imply that the transverse star

Star⁡Δ(τ)/Tτ=⋃τ⊂σR≥0uσ/τ\operatorname{Star}_\Delta(\tau)/T_\tau=\bigcup_{\tau\subset\sigma}\mathbb{R}_{\geq0}u_{\sigma/\tau}

is linearly isomorphic to the image of the standard tropical line in Rm/R1\mathbb{R}^m/\mathbb{R}\mathbf{1} where mm denotes the number of facets containing τ\tau. More precisely, after rescaling by the positive balancing coefficients, the vectors uσ/τu_{\sigma/\tau} are the images of the standard basis vectors of Rm\mathbb{R}^m under an injective linear map from Rm/R1\mathbb{R}^m/\mathbb{R}\mathbf{1} to V/TτV/T_\tau. Note that this linear isomorphism may not preserve the integral lattice, so the ridge condition does not imply tropical smoothness.

We apply this criterion to tropical linear spaces of valuated matroids (see [14 Section 4.4] or [17], for example). Their local geometry is governed by initial matroids, which allows us to verify the ridge condition explicitly. Let p=(pB)p=(p_B) be a valuated matroid of rank rr on [n][n] with loopless underlying matroid MM. For w∈Rnw\in\mathbb{R}^n, write w(B)=∑i∈Bwiw(B)=\sum_{i\in B}w_i, and let MwM_w be the matroid whose bases maximize w(B)−pBw(B)-p_B. The associated tropical linear space is

T(p)={w∈Rn∣Mw is loopless}\mathcal{T}(p)=\{w\in\mathbb{R}^n\mid M_w\text{ is loopless}\}

which is pure of dimension rr. The Bergman fan B(M)\mathcal{B}(M) of a matroid MM is the tropical linear space associated with the trivial valuation. For its combinatorial structure, see [1, 10].

For F⊆[n]F\subseteq[n], write eF=∑i∈Feie_F=\sum_{i\in F}e_i. Let c(M)c(M) denote the number of connected components of MM, and set

L=span⁡{eC∣C is a connected component of M}.L=\operatorname{span}\{e_C\mid C\text{ is a connected component of }M\}.

Thus, dim⁡L=c(M)\dim L=c(M).

For a polyhedral support X⊂VX \subset V and x∈Xx \in X, write

Tan⁡xX={v∈V∣x+εv∈X for all sufficiently small ε>0}.\operatorname{Tan}_{x}X=\{v\in V\mid x+\varepsilon v\in X\text{ for all sufficiently small }\varepsilon>0\}.

The coarse subdivision of T(p)\mathcal{T}(p) consists of the closures of the loci on which MwM_w is constant. Its cells are defined by comparisons of the basis scores w(B)−pBw(B)-p_B.

Lemma 17. Let pp be a valuated matroid of rank rr with a loopless underlying matroid. Then, every finite pure rr-dimensional polyhedral complex Δ\Delta with support T(p)\mathcal{T}(p) admits a positive top cycle, and dim⁡ker⁡∂rΔ=1\dim\ker\partial_r^\Delta=1.

Proof. We first claim that we have Tan⁡wT(p)=B(Mw)\operatorname{Tan}_w\mathcal{T}(p)=\mathcal{B}(M_w) for every w∈T(p)w\in\mathcal{T}(p). To see this, for any fixed v∈Rnv\in\mathbb{R}^n and sufficiently small ε>0\varepsilon>0, the bases maximizing

(w+εv)(B)−pB=w(B)−pB+εv(B)(w+\varepsilon v)(B)-p_B=w(B)-p_B+\varepsilon v(B)

are precisely those that maximize v(B)v(B) among the bases of MwM_w. Thus, Mw+εv=(Mw)vM_{w+\varepsilon v}=(M_w)_v, which gives the claim.

Let σ\sigma be a cell of the coarse subdivision and w∈relint⁡(σ)w\in\operatorname{relint}(\sigma). A vector vv belongs to Tσ=Span⁡(σ−σ)T_\sigma=\operatorname{Span}(\sigma-\sigma) precisely when v(B)=v(C)v(B)=v(C) for all bases B,CB,C of MwM_w. These vectors are exactly those constant on each connected component of MwM_w. Hence, TσT_\sigma is the lineality space of B(Mw)\mathcal{B}(M_w), and dim⁡σ=c(Mw)\dim\sigma=c(M_w).

Let τ\tau be a ridge of the coarse subdivision and w∈relint⁡(τ)w\in\operatorname{relint}(\tau). Then, c(Mw)=r−1c(M_w)=r-1, so MwM_w has exactly one component SS of rank two, and all other components have rank one. Let P1,…,PmP_1,\ldots,P_m be the parallel classes of Mw∣SM_w|_S. Since SS is connected, m≥3m\geq3.

Near ww, the facets containing τ\tau correspond to the rays generated by eP1,…,ePme_{P_1},\ldots,e_{P_m} modulo TτT_\tau. These satisfy ∑iePi=eS∈Tτ\sum_i e_{P_i}=e_S\in T_\tau, and any m−1m-1 of them are linearly independent modulo TτT_\tau. Hence, this is their unique linear relation up to scaling. Moreover, the vectors ePie_{P_i} give the primitive normal vectors in the quotient lattice. Consequently, choosing the primitive lattice generators ξσ\xi_\sigma, the cycle ∑σξσ[σ]\sum_\sigma\xi_\sigma[\sigma] with all coefficients one is balanced at every ridge, and dim⁡ker⁡Rτ=1\dim\ker R_\tau=1 at every ridge.

Tropical linear spaces are connected through codimension one [3 Corollary 1.6]. By Proposition 15, the top-cycle space is spanned by the weight-one cycle. Finally, Lemma 3 transfers this to any finite polyhedral complex with support T(p)\mathcal{T}(p) through a common refinement. □\square

We verify that LL lies in the lineality space of every cell of the coarse subdivision. Indeed, any two bases of MM have the same cardinality in each connected component of MM. Thus, translation by a vector in LL adds the same constant to every basis score w(B)−pBw(B)-p_B, leaving the initial matroid unchanged. Consequently, every cell is invariant under translation by LL.

Theorem 18. Let WW be a linear subspace of LL and let Σ\Sigma be any pointed polyhedral complex with support T(p)/W\mathcal{T}(p)/W. For every 1≤k≤r−dim⁡W1\leq k\leq r-\dim W, the kk-skeleton of Σ\Sigma is kk-connected through codimension one.

Proof. By Lemma 17, the coarse subdivision of T(p)\mathcal{T}(p) admits a positive top cycle and has a one-dimensional top cycle space. Since W⊂LW\subset L, quotienting by WW preserves these properties by Lemma 3. Then, a finite common refinement of the resulting quotient complex and Σ\Sigma transfers the same properties to Σ\Sigma. Since Σ\Sigma is pointed and has dimension d=r−dim⁡Wd=r-\dim W, the conclusion follows from Theorem 1. □\square

Corollary 19. Let MM be a loopless matroid of rank rr on [n][n], and let LL be the lineality space of B(M)\mathcal{B}(M). Let W⊂LW \subset L and let Σ\Sigma be a pointed fan with support B(M)/W\mathcal{B}(M)/W. Put d=r−dim⁡W≥1d=r-\dim W\geq1. Then, its kk-skeleton is kk-connected through codimension one for every 1≤k≤d1\leq k\leq d. In particular, the coarse subdivision in Rn/L\mathbb{R}^{n}/L and every finer fan structure have (r−c(M))(r-c(M))-connected facet-ridge hypergraphs whenever r>c(M)r>c(M). For connected MM of rank at least two, the coarse fan in Rn/R1\mathbb{R}^{n}/\mathbb{R}1 is (r−1)(r-1)-connected.

Proof. Apply Theorem 18 to the trivial valuation. □

Connectivity and coordinate-projection matroids

For a balanced complex with support X⊂RnX\subset\mathbb{R}^{n}, let I(X)={S⊂[n]∣dim⁡πS(X)=∣S∣}\mathcal{I}(X)=\{S\subset[n]\mid\dim\pi_S(X)=|S|\} where πS:Rn→RS\pi_S:\mathbb{R}^{n}\to\mathbb{R}^{S} is the coordinate projection. If XX is the tropicalization of an irreducible variety, then I(X)\mathcal{I}(X) is the collection of independent sets of its algebraic matroid [Yu17]. The following examples show that higher connectivity does not force I(X)\mathcal{I}(X) to be a matroid, and that tropical irreducibility, even together with higher connectivity, does not force it to be algebraic.

Example 20. Let Lij=span⁡(ei,ej)⊂R4L_{ij}=\operatorname{span}(e_i,e_j)\subset\mathbb{R}^{4} and X=L12∪L23∪L34X=L_{12}\cup L_{23}\cup L_{34} with each plane subdivided into its four coordinate quadrants. Since each plane is balanced, this pointed fan is balanced with all weights one. It is 2-connected through codimension one. In fact, after one quadrant is deleted, the other three quadrants of its plane remain connected, every other plane loses at most one boundary ray and remains connected, and consecutive planes still share a ray of their common axis. However, the two-element sets in I(X)\mathcal{I}(X) are exactly {1,2}\{1,2\}, {2,3}\{2,3\} and {3,4}\{3,4\}. Thus, {1}\{1\} cannot be augmented from {3,4}\{3,4\}, and I(X)\mathcal{I}(X) is not a matroid. This fan is not tropically irreducible because each plane carries its own balanced weighting.

Example 21. Let V8V_8 be the Vámos matroid, a connected rank-four matroid on eight elements that is not algebraic over any field [13], and let X=B(V8)⊂R8X=\mathcal{B}(V_8)\subset\mathbb{R}^{8}. By Lemmas 4 and 17, XX is tropically irreducible, and by Corollary 19, every pointed fan structure on X/R1X/\mathbb{R}1 is 3-connected through codimension one. However, I(X)\mathcal{I}(X) is the collection of independent sets of V8V_8 [Yu17, Lemma 3], which is not algebraic. In particular, XX is not the tropicalization of any irreducible variety.

We do not know whether I(X)\mathcal{I}(X) always defines a matroid when XX is tropically irreducible.

Acknowledgments

We thank Josephine Yu for suggesting this problem when [15] first appeared, and Cvetelina Hill for discussions at an early stage of this project. Department of Mathematical Sciences, Korea Advanced Institute of Science and Technology

Email address: [email protected]

URL: https://klee669.github.io

References

  1. [1]F. Ardila and C. J. Klivans. The Bergman complex of a matroid and phylogenetic trees. Journal of Combinatorial Theory, Series B, 96(1):38–49, 2006.DOI
  2. [2]E. Aksnes. Tropical Poincaré duality spaces. Advances in Geometry, 23(3):345–370, 2023.
  3. [3]N. Anderson. Matroid products in tropical geometry. Research in the Mathematical Sciences, 11(2):38, 2024.arxiv.org/abs/2306.14771
  4. [4]C. A. Athanasiadis. On the graph connectivity of skeleta of convex polytopes. Discrete & Computational Geometry, 42(2):155–165, 2009.arxiv.org/abs/0801.0939
  5. [5]M. L. Balinski. On the graph structure of convex polyhedra in n-space. Pacific Journal of Mathematics, 11(2):431–434, 1961.
  6. [6]F. Babaee, S. Dewar, and J. Maxwell. Extremal decompositions of tropical varieties and relations with rigidity theory. Journal of Symbolic Computation, 132:102461, 2026.
  7. [7]F. Babaee and J. Huh. A tropical approach to a generalized Hodge conjecture for positive currents. Duke Mathematical Journal, 166(14):2749 – 2813, 2017.DOI
  8. [9]D. Cartwright and S. Payne. Connectivity of tropicalizations. Mathematical Research Letters, 19(5):1089–1095, 2013.DOI
  9. [10]E. M. Feichtner and B. Sturmfels. Matroid polytopes, nested sets and Bergman fans. Portugaliae Mathematica, 62(4):437–468, 2005.
  10. [12]D. Hathcock and J. Yu. On the hypergraph connectivity of skeleta of polytopes. Discrete & Computational Geometry, 69(2):593–596, 2023.arxiv.org/abs/2010.05053
  11. [13]A. W. Ingleton and R. A. Main. Non-algebraic matroids exist. Bulletin of the London Mathematical Society, 7(2):144–146, 1975.DOI
  12. [14]D. Maclagan and B. Sturmfels. Introduction to tropical geometry, volume 161. American Mathematical Society, Providence, RI, 2015.
  13. [15]D. Maclagan and J. Yu. Higher connectivity of tropicalizations. Mathematische Annalen, 384:775–788, 2022.arxiv.org/abs/1908.05988
  14. [16]G. T. Sallee. Incidence graphs of convex polytopes. Journal of Combinatorial Theory, 2(4):466–506, 1967.DOI
  15. [17]D. E. Speyer. Tropical linear spaces. SIAM Journal on Discrete Mathematics, 22(4):1527–1558, 2008.DOI

Paper details

vlnc:W_ntmj6mqn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents