Bondal–Thomsen Generators in Toric Geometry

• Bondal–Thomsen generators are torus‐equivariant line bundles on toric varieties that arise as Frobenius pushforward summands and are indexed by lattice points of half‐open zonotopes.
• They unify combinatorial, birational, and categorical approaches to study toric geometry, offering explicit indexing formulas and stratifications via toric hyperplane arrangements.
• These generators exhibit symmetry under lattice automorphisms and form full strong exceptional collections in Cox categories, with concrete examples in projective and Hirzebruch surfaces.

Bondal–Thomsen generators are torus-equivariant line bundles on a toric variety that arise as direct summands of the pushforward of the structure sheaf under the toric “Frobenius” endomorphism, or, equivalently in recent combinatorial formulations, as line bundles indexed by lattice points of a half-open zonotope or by strata of a periodic toric hyperplane arrangement. In the smooth toric setting they form the stabilized Thomsen collection $T(X)$, and more generally the generalized Thomsen collection $T(X,D)$ attached to a torus-invariant $\mathbb{Q}$-divisor $D$; in class-group language they are indexed by the Bondal–Thomsen set $\Theta_X$; and in the Cox category of a smooth projective toric variety they appear as Cox lifts $\mathcal{O}_{Cox}(-d)$ forming a $G$-invariant full strong exceptional collection (Yi, 30 Jun 2025, Bauermeister et al., 18 Jul 2025, Erman et al., 22 Dec 2025).

1. Frobenius origin and the basic indexing formulas

Let $N$ be a free abelian group of rank $n$, $M=\operatorname{Hom}(N,\mathbb{Z})$, and $T(X,D)$0 the toric variety associated to a fan $T(X,D)$1. For each ray $T(X,D)$2 with primitive generator $T(X,D)$3, there is a torus-invariant prime divisor $T(X,D)$4. For any integer $T(X,D)$5, multiplication by $T(X,D)$6 on the monoid $T(X,D)$7 defines a finite morphism

$T(X,D)$8

which restricts on the open torus $T(X,D)$9 to the character map $\mathbb{Q}$0 (Yi, 30 Jun 2025).

For a smooth toric variety $\mathbb{Q}$1 and any line bundle $\mathbb{Q}$2, Thomsen’s decomposition gives

$\mathbb{Q}$3

where $\mathbb{Q}$4 is the number of integer points $\mathbb{Q}$5 such that the divisor $\mathbb{Q}$6 represents $\mathbb{Q}$7 in $\mathbb{Q}$8 (Yi, 30 Jun 2025). For $\mathbb{Q}$9 a torus-invariant $D$0-divisor and $D$1 sufficiently divisible so that $D$2 is integral, a line bundle $D$3 is a direct summand of $D$4 if and only if there exists $D$5 such that

$D$6

for all $D$7 (Yi, 30 Jun 2025).

This yields the generalized Thomsen collection

$D$8

and for $D$9 the stabilized set

$\Theta_X$0

recovers the usual Thomsen collection (Yi, 30 Jun 2025). In the notation used for the Cox category, the same indexing is often written as

$\Theta_X$1

with corresponding objects $\Theta_X$2 (Erman et al., 22 Dec 2025).

2. Oriented toric hyperplane arrangements and half-open zonotopes

A combinatorial model for Bondal–Thomsen generators begins with a finite ordered set $\Theta_X$3 and the linear map

$\Theta_X$4

Let $\Theta_X$5 be the cokernel projection. Passing to the quotient $\Theta_X$6, one obtains maps

$\Theta_X$7

induced by $\Theta_X$8, and the oriented toric hyperplane arrangement is the collection of codimension-one subtori $\Theta_X$9 (Bauermeister et al., 18 Jul 2025).

The orientation is encoded by the ceiling-based slicing

$\mathcal{O}_{Cox}(-d)$0

For each $\mathcal{O}_{Cox}(-d)$1, the sets $\mathcal{O}_{Cox}(-d)$2 descend to a partition of $\mathcal{O}_{Cox}(-d)$3 into $\mathcal{O}_{Cox}(-d)$4-strata. If $\mathcal{O}_{Cox}(-d)$5 is a stratum, the indices of the hyperplanes containing it are recorded by

$\mathcal{O}_{Cox}(-d)$6

The associated half-open zonotope is

$\mathcal{O}_{Cox}(-d)$7

whose closure $\mathcal{O}_{Cox}(-d)$8 is a lattice polytope with respect to $\mathcal{O}_{Cox}(-d)$9 (Bauermeister et al., 18 Jul 2025).

The central statement is that $G$0 induces a bijection between $G$1-strata and the lattice points of $G$2. If $G$3 is the point corresponding to $G$4, and $G$5 is the minimal face of $G$6 containing $G$7, then $G$8 is determined by

$G$9

The dimension relation is

$N$0

for any lift $N$1 of the stratum and any $N$2; when $N$3 contains a lattice basis of $N$4, so that $N$5, this simplifies to

$N$6

The correspondence $N$7 is inclusion-reversing (Bauermeister et al., 18 Jul 2025).

3. Class groups, effective cones, and the toric meaning of the strata

For a semiprojective toric variety $N$8 with simplicial fan $N$9, character lattice $n$0, and primitive ray generators $n$1, the combinatorial construction is realized by the group homomorphism

$n$2

whose cokernel identifies with $n$3. The half-open zonotope in class-group space is

$n$4

and the Bondal–Thomsen collection is defined as

$n$5

The associated torus-equivariant line bundles are $n$6 with $n$7 (Bauermeister et al., 18 Jul 2025).

In this formulation, the lattice points of $n$8 are the class-group parameters for the generators, and the $n$9-strata give a Bondal stratification. If $M=\operatorname{Hom}(N,\mathbb{Z})$0 is the minimal face of $M=\operatorname{Hom}(N,\mathbb{Z})$1 containing the image of $M=\operatorname{Hom}(N,\mathbb{Z})$2, and $M=\operatorname{Hom}(N,\mathbb{Z})$3 is the corresponding $M=\operatorname{Hom}(N,\mathbb{Z})$4-stratum, then in the semiprojective setting

$M=\operatorname{Hom}(N,\mathbb{Z})$5

Moreover, $M=\operatorname{Hom}(N,\mathbb{Z})$6 is naturally identified with the effective cone of a $M=\operatorname{Hom}(N,\mathbb{Z})$7-dimensional toric variety $M=\operatorname{Hom}(N,\mathbb{Z})$8, and the face of the zonotope supporting $M=\operatorname{Hom}(N,\mathbb{Z})$9 is the zonotope $T(X,D)$00 of the lower-dimensional toric variety obtained by quotienting out the rays indexed by $T(X,D)$01 (Bauermeister et al., 18 Jul 2025).

The effective cone $T(X,D)$02 is the support of the GKZ or secondary fan $T(X,D)$03, so the zonotope stratification sits directly inside toric birational geometry. A further consequence is a vanishing statement for adjacent chambers: if $T(X,D)$04 corresponds to a chamber adjacent to a face $T(X,D)$05 in $T(X,D)$06, then for all $T(X,D)$07 whose image does not lie in $T(X,D)$08,

$T(X,D)$09

where $T(X,D)$10 is induced by viewing the fan of $T(X,D)$11 as a refinement of a generalized fan for $T(X,D)$12 (Bauermeister et al., 18 Jul 2025). The discussion around this result explicitly speculates that, at the level of Cox categories, the dimension of the minimal face containing a generator should control semiorthogonality.

4. Generation results and generalized Thomsen collections

Bondal claimed that for a smooth proper toric variety $T(X,D)$13, the Thomsen collection $T(X,D)$14 generates $T(X,D)$15, and this claim has now been proved in full generality. The generalization developed for a torus-invariant $T(X,D)$16-divisor $T(X,D)$17 defines $T(X,D)$18 as the stabilized set of direct summands of $T(X,D)$19 for $T(X,D)$20 sufficiently divisible, and the main theorem states: for a smooth toric variety $T(X,D)$21 over an algebraically closed field and any torus-invariant $T(X,D)$22-divisor $T(X,D)$23 on $T(X,D)$24, the generalized Thomsen collection $T(X,D)$25 generates $T(X,D)$26; the case $T(X,D)$27 recovers Bondal’s original claim (Yi, 30 Jun 2025).

A key structural point is that the generalized collection admits an $T(X,D)$28-independent description by floor functions. If $T(X,D)$29, then

$T(X,D)$30

for some $T(X,D)$31; if $T(X,D)$32 is integral, then $T(X,D)$33 is $T(X,D)$34 tensored by $T(X,D)$35, so generation by $T(X,D)$36 is equivalent to generation by $T(X,D)$37 (Yi, 30 Jun 2025).

The proof is organized through the notion of a generating system $T(X,D)$38, where $T(X,D)$39 is a real vector space, the $T(X,D)$40 are half-spaces, and the $T(X,D)$41 are disjoint subsets of torus-invariant prime divisors. For each $T(X,D)$42, one defines a torus-invariant effective divisor

$T(X,D)$43

The associated generation theorem shows

$T(X,D)$44

and the proof proceeds by slicing in $T(X,D)$45, reducing to a two-dimensional “peeling” argument using the toric Koszul exact sequence

$T(X,D)$46

for torus-invariant effective divisors $T(X,D)$47 and $T(X,D)$48 with no common components (Yi, 30 Jun 2025).

This framework is combined with Orlov’s blow-up decomposition, the projective bundle theorem, toric weak factorization, and localization. The same work also gives a stacky interpretation: $T(X,D)$49 can be viewed as the pushforward along the coarse moduli map of a Thomsen collection on a toric root stack obtained by taking roots along the torus-invariant divisors with $T(X,D)$50 integral (Yi, 30 Jun 2025). The article’s emphasis, however, is that the generation proof itself does not require stacks.

5. Cox category, symmetry, and strong exceptionality

For a smooth projective toric variety $T(X,D)$51, the Cox ring is

$T(X,D)$52

The secondary fan $T(X,D)$53 has maximal chambers $T(X,D)$54, each corresponding to a projective toric birational model $T(X,D)$55 with simplicial fan $T(X,D)$56. Choosing the associated smooth toric Deligne–Mumford stacks $T(X,D)$57 and a common smooth toric DM stack $T(X,D)$58 with proper birational stacky refinements $T(X,D)$59, the Cox category $T(X,D)$60 is defined as the triangulated hull of the fully faithful subcategories $T(X,D)$61 inside $T(X,D)$62. There is a natural fully faithful functor $T(X,D)$63 (Erman et al., 22 Dec 2025).

If $T(X,D)$64 lies in a chamber $T(X,D)$65, its Cox lift is

$T(X,D)$66

independent of $T(X,D)$67 as long as $T(X,D)$68 lies in the nef cone of $T(X,D)$69. In this language, the Bondal–Thomsen objects are

$T(X,D)$70

and the main theorem of the geometric Merkurjev–Panin study states that if $T(X,D)$71 is the group of lattice automorphisms that permute the rays $T(X,D)$72, then the Bondal–Thomsen collection $T(X,D)$73 forms a $T(X,D)$74-invariant full strong exceptional collection in $T(X,D)$75. More generally, if $T(X,D)$76 is semiprojective and normal, then

$T(X,D)$77

is a $T(X,D)$78-invariant tilting bundle in $T(X,D)$79 (Erman et al., 22 Dec 2025).

The invariance mechanism is explicit. The group $T(X,D)$80 acts on the secondary fan, hence on chambers and the corresponding stacks, and induces a strict action

$T(X,D)$81

Morphisms are controlled by the closed section polytope

$T(X,D)$82

and

$T(X,D)$83

if and only if $T(X,D)$84, with the dimension of the Hom space equal to the number of lattice points in that intersection (Erman et al., 22 Dec 2025). Ordering the Bondal–Thomsen objects by the effective partial order yields strong exceptionality. A useful monotonicity statement is that if such a Hom space is nonzero, then a corresponding “depth set” $T(X,D)$85 is contained in $T(X,D)$86, which supports semi-orthogonal decompositions by “dimension” and “depth” of strata (Erman et al., 22 Dec 2025). This is closely aligned with the face-poset picture of the zonotope construction.

6. Examples, scope, and limitations

Several standard examples make the combinatorics explicit. For projective space,

$T(X,D)$87

and in the Cox-category formulation this is the Beilinson collection $T(X,D)$88, which is $T(X,D)$89-invariant for $T(X,D)$90 permuting the rays (Yi, 30 Jun 2025, Erman et al., 22 Dec 2025). For $T(X,D)$91, the Thomsen collection stabilizes to

$T(X,D)$92

and these two line bundles generate $T(X,D)$93 (Yi, 30 Jun 2025).

For Hirzebruch surfaces, the behavior depends on the perspective. On $T(X,D)$94, the Thomsen bundles obtained from the floor-function formula generate $T(X,D)$95, and in $T(X,D)$96 the resulting finite set is generated by

$T(X,D)$97

where $T(X,D)$98 is the minimal section and $T(X,D)$99 is a fiber class (Yi, 30 Jun 2025). For the specific Hirzebruch surface $\mathbb{Q}$00, the arrangement–zonotope dictionary gives five strata—three of dimension two, one of dimension one, and one of dimension zero—and these correspond bijectively to the five lattice points of $\mathbb{Q}$01; the associated five class-group elements in $\mathbb{Q}$02 determine the five Bondal–Thomsen generators $\mathbb{Q}$03, and

$\mathbb{Q}$04

because $\mathbb{Q}$05 (Bauermeister et al., 18 Jul 2025).

For the weighted blow-up of $\mathbb{Q}$06 at a torus fixed point, the same formalism yields eight strata—six of dimension two, one of dimension one, and one of dimension zero. In the zonotope there are six lattice points on a three-dimensional face, one on a two-dimensional face, and one at a vertex; these eight class-group elements determine the eight Bondal–Thomsen generators, and

$\mathbb{Q}$07

again because $\mathbb{Q}$08 (Bauermeister et al., 18 Jul 2025). For permutohedral varieties, the Cox-category study records that $\mathbb{Q}$09 has $\mathbb{Q}$10 elements for $\mathbb{Q}$11 and $\mathbb{Q}$12 elements for $\mathbb{Q}$13, with the $\mathbb{Q}$14-action respecting decompositions by “dimension” and “depth” of strata (Erman et al., 22 Dec 2025).

The formalism has explicit hypotheses and limits. The clean identification of strata with subsets of the standard torus requires $\mathbb{Q}$15 to contain a lattice basis, and semiprojectivity guarantees $\mathbb{Q}$16; when $\mathbb{Q}$17 has no smooth cone, one works on a finite cover of the usual torus $\mathbb{Q}$18 (Bauermeister et al., 18 Jul 2025). The simplicial assumption is used to streamline the link to the secondary fan and effective cones (Bauermeister et al., 18 Jul 2025). On the homological side, generalized Thomsen collections always generate $\mathbb{Q}$19 for smooth toric $\mathbb{Q}$20, but in general they need not be exceptional or full exceptional collections; the emphasis of the generation theorem is generation rather than exceptionality (Yi, 30 Jun 2025). The combinatorial framework also leaves open further questions: one work explicitly notes that extending the dimension correspondence to resolutions of toric subvarieties requires further control beyond the combinatorics of $\mathbb{Q}$21 (Bauermeister et al., 18 Jul 2025), while another remarks that its proof does not track generation length and poses questions about purely hyperplane-arrangement reformulations and affine-hyperplane generalizations (Yi, 30 Jun 2025).

Taken together, these developments present Bondal–Thomsen generators as a single toric-homological object with three mutually reinforcing descriptions: Frobenius summands of $\mathbb{Q}$22, lattice points of a half-open zonotope paired with strata of an oriented toric hyperplane arrangement, and Cox-category line bundles carrying functorial symmetry under lattice automorphisms. The resulting dictionary is simultaneously combinatorial, birational, and categorical.