Fujita Freeness for Toric Projective Bundles: Sharp Bounds

Fujita Freeness for Toric Projective Bundles: Sharp Bounds

Introduction: Fujita freeness meets toric geometry, with a clean new proof

If you’ve ever tried to answer “when do global sections actually show up everywhere?”, you already know the frustration: positivity statements sound great, but turning them into global generation often needs careful numerics.

This new work—based on Antonio Laface’s paper on arXiv:2608.28438—pushes Fujita’s freeness conjecture into a broad toric setting. The main characters are: a smooth projective toric variety (X) of dimension (n), a toric vector bundle (\mathcal E) of rank (r\ge 2), and its projectivization
[
\pi: Y=\mathbb{P}_X(\mathcal{E})\to X,
]
where the fibers parametrize one-dimensional quotients (Grothendieck’s convention). The payoff: explicit, uniform bounds ensuring that certain adjoint-type line bundles become globally generated.

Concretely, the paper studies ample line bundles on (Y) of the form
[
A=\mathcal{O}Y(a)\otimes \pi^*L,
\quad a\ge 1,
]
and proves global generation of
[
KY+mA + mA \text{ in the precise form } KY+mA KY + mA
]
(meaning the bundle (KY + mKY + mA) is globally generated under the given inequalities; the statement is spelled out precisely in the paper). The conditions involve two competing numerical constraints: one coming from the fiber direction (through (a) and (r)), and another coming from how positivity behaves on torus-invariant curves (through a positive integer (\delta(A))). Even better: the uniform bounds are shown sharp.

Why This Matters: global generation bounds you can actually trust

This result is significant right now because “global generation from positivity” keeps showing up across algebraic geometry’s practical workflows—especially when someone needs to produce morphisms, control base loci, or reduce questions to manageable fibers.

A concrete scenario where this can be applied today: imagine you’re building explicit toric models of moduli problems (or running a pipeline that replaces geometric data by section-generated line bundles). Global generation is the technical ingredient that lets you turn numerical positivity into a map to projective space without surprises on some lurking subvariety. Laface’s theorem gives an actionable threshold: for every projectivized toric vector bundle over a smooth projective toric base, Fujita’s freeness holds with a uniform bound depending only on dimensions and a simple inequality involving the bundle rank and the parameter (a).

And there’s a broader methodological value. Prior work often handled special polarizations or used external jet-separation / Seshadri-constant machinery in a more indirect way. Here, the proof is built around a blow-up argument at torus-fixed points, designed to work for arbitrary (a) and arbitrary rank. In other words: instead of translating everything into a different formalism and hoping the numerics line up, the proof stays on the projective bundle and makes the toric structure do the heavy lifting—through torus invariance of the base locus and explicit curve checks. That’s a very “engineer-friendly” style of mathematics.

Finally, this builds on and connects to AI-era mathematical thinking in a surprising way: recent research about Seshadri constants and jet separation has a flavor of “local-to-global rules.” Laface’s blow-up realizes an analogous mechanism integrally in the toric setting, so you get an implementation-style proof rather than a purely conceptual one. If you’ve seen how positivity criteria can be algorithmically checked on invariant curves, this paper is essentially showing: yes, and we can even make the full Fujita bound work uniformly.

Theorem 1: how Laface turns Fujita freeness into explicit inequalities

Let’s unpack the key numerical setup. We start with:

• (X): smooth projective toric variety, (\dim X=n\ge 1)
• (\mathcal E): toric vector bundle, rank (r\ge 2)
• (Y=\mathbb{P}_X(\mathcal E)) (Grothendieck quotient convention)
• (A=\mathcal{O}_Y(a)\otimes\pi^*L) with (a\ge 1) and (A) ample

Then the theorem introduces a positive integer (\delta(A)). Roughly: (\delta(A)) measures how positivity behaves when you look at invariant quotient sections over torus-invariant curves in (X). More precisely, (\delta(A)) comes from degrees along sections determined by quotients (\mathcal E|C\twoheadrightarrow \mathcal O{\mathbb P^1}(b_{C,j})) on each torus-invariant curve (C\cong \mathbb P^1).

The key global generation statement (numerical conditions)

A line bundle of the form
[
KY + mA + mKY\quad\text{(as written precisely in the paper)}
]
turns out to be globally generated if (m) satisfies two inequalities:

1. Fiber-direction constraint:
   [
   ma \ge r \quad\text{(and equivalently the paper assumes } ma \ge r \ge 2\text{)}
   ]
2. Toric-curve / Seshadri-type constraint:
   [
   m\delta(A) > n\,m\delta(A) > n
   ]
   In the cleanest corollary form, this is guaranteed once (m\ge n+1) because (\delta(A)) is a positive integer.

The paper states it in a streamlined form: if an integer (m) satisfies
[
ma\ge r
\quad\text{and}\quad
m\delta(A)>n,
]
in a way made precise with (n\,m\delta(A)>n) in the theorem, then the adjoint-type bundle is globally generated.

The “uniform bound” version becomes wonderfully simple

Because (\delta(A)) is a positive integer, you immediately get:
- If (m\ge n+1), then (m\delta(A)>n).
- If also (ma\ge r), then both constraints are met.

So Laface obtains global generation whenever
[
m\ge n+1
\quad\text{and}\quad
ma\ge r.
]

Equivalently, the theorem yields a bound of the form:
[
m\ge \max{n+1,\lceil r/a\rceil}.
]

This is the headline: every projectivized toric vector bundle satisfies Fujita’s freeness conjecture, with a uniform bound, and the paper proves that the bound is sharp.

How the blow-up proof actually works (and why torus symmetry is the magic trick)

The proof is built from three moves that fit together extremely cleanly:

1. Blow up a torus-fixed point in the base (X)
2. Use a toric positivity criterion to keep control after subtracting an exceptional divisor
3. Use Kodaira vanishing to make restriction maps surjective, then use torus invariance to kill base loci globally

Step 1: blow up the fiber over a fixed point

Take (x\in X) a torus-fixed point. Blow it up:
[
p:\widetilde X=\operatorname{Bl}_x X\to X
]
with exceptional divisor (F).

Because the projection (\pi:Y\to X) is flat, the geometry commutes with this base change: you can form the induced blow-up (\widetilde Y) sitting over (\widetilde X). The exceptional divisor for (\rho:\widetilde Y\to Y) is denoted (G), and the paper shows (G=\pĩ^*F).

Step 2: a positivity inequality checks ampleness on invariant curves

Here’s where the toric structure becomes computationally powerful. The strategy is to prove that for suitable rational (\lambda),
[
\rho^*A - \lambda G
]
is ample on (\widetilde Y). Rather than proving this by some abstract global criterion, the paper uses an invariant-curve check: in toric geometry, ampleness of toric bundles (even with (\mathbb Q)-twists) can be tested on torus-invariant curves.

The paper splits torus-invariant curves in the blow-up into three types:

• strict transforms of invariant curves not passing through (x)
• strict transforms of invariant curves passing through (x)
• curves contained in the exceptional divisor (F)

For each type, the degrees of the relevant summands can be computed, and the definition of (\delta_x(A)) is used exactly to guarantee positivity in the middle case. For curves inside the exceptional divisor, the twisting gives degrees (\lambda/a>0), so those are safe too.

That’s the role of (\delta(A)): it packages the “worst-case positivity along curves through (x)” into a number you can compare to (n) and (m).

Step 3: Kodaira vanishing makes restriction maps surjective

Once ampleness is established, the paper applies Kodaira vanishing in a standard way: it wants surjectivity of
[
H^0(Y,Dm)\to H^0(Yx, Dm|{Yx}),
]
where (Yx=\pi^{-1}(x)) is the fiber over the fixed point and
[
Dm := KY + mA.
]

The restriction to the fiber turns out to be explicitly computable:
[
Dm|{Yx}\cong \mathcal O{\mathbb P^{r-1}}(ma-r).
]
So the inequality (ma\ge r) directly implies that (\mathcal O_{\mathbb P^{r-1}}(ma-r)) is globally generated on the fiber.

Step 4: torus invariance kills the global base locus

Now comes the neat final step. Since the line bundle (D_m) is torus-linearized, its base locus is a torus-invariant closed subset. If there were a base point somewhere, you could follow a torus orbit closure and land at a torus-fixed point in the base locus.

But the paper already showed that over every torus-fixed point (x\in X^T), the restriction is globally generated—so there can’t be fixed points in the base locus. Hence the torus-invariant base locus must be empty globally.

This is the “local-to-global” mechanism: positivity and vanishing ensure no trouble on fixed fibers; torus symmetry prevents trouble anywhere else.

(And yes—this is exactly where the toric version of Fujita freeness becomes unusually checkable. The same paper even explicitly relates this to the Seshadri-constant approach of Hering–Mustaţă–Payne and the adjoint jet-separation theorem of Fulger–Murayama; Laface’s blow-up is presented as an integral toric realization of that mechanism.)

A related payoff: adjoint symmetric powers also become globally generated

The paper doesn’t stop at base-point-freeness of (K_Y+mA)-type bundles. It also upgrades to a statement about adjoint symmetric powers on the base variety (X).

Theorem 2: evaluation surjectivity at fixed points

Under the same hypotheses (with a specific choice connecting (m) and the symmetric power exponent), the paper proves that the bundle
[
\omegaX \otimes \det\mathcal E \otimes \operatorname{Sym}^{q}\mathcal E
]
twisted appropriately by powers of (L) and by (\omegaX) is globally generated.

A practical way to think about it:
- “global generation” means: at each point, enough global sections exist to generate the fiber.
- in the toric setting, it’s enough to prove this at torus-fixed points, because torus invariance forces uniform behavior.
- the restriction map surjectivity becomes an evaluation map surjectivity at (x).

The proof again uses the same pipeline: positivity (\Rightarrow) Kodaira-type vanishing (\Rightarrow) surjective restriction/evaluation at (x) (\Rightarrow) torus invariance kills the non-generation locus.

Special clean case: polarization (\mathcal O_Y(1))

When you choose the “tautological” polarization (A=\mathcal O_Y(1)\otimes \pi^*L) (so (a=1)), you get a particularly readable bound. If (\mathcal E) is ample of rank (r) on (X) of dimension (n), then global generation holds for
[
q+r\ge n+1
]
(up to the paper’s (\delta(\mathcal E))-weighted inequality, which implies this for free because (\delta(\mathcal E)\ge 1)).

So for many readers, the takeaway is: you really do recover Fujita’s “(d+1)” style bound in this toric projectivized setting, and you can even see exactly where the symmetric-power exponent fits.

Sharpness: why the bounds can’t be improved uniformly

A key quality check in math results like this is sharpness: could a smaller bound still work always?

Here, Laface proves the answer is no: the uniform conditions are sharp.

Sharpness for Fujita’s freeness bound

The paper constructs explicit examples where global generation holds exactly when both numerical conditions hold.

Take:
- (X=\mathbb P^n)
- (\mathcal E=\mathcal O_{\mathbb P^n}^{\oplus r}) (trivial bundle)
Then
[
Y=\mathbb P^n\times \mathbb P^{r-1}.
]

In this product case, line bundles look like (\mathcal O(u,v)), and global generation is easy to characterize: (\mathcal O(u,v)) is globally generated iff (u\ge 0) and (v\ge 0). When translated into the parameters from the theorem, the conditions become precisely:
- (m\ge n+1)
- (ma\ge r)

So neither part can be dropped uniformly.

Sharpness for symmetric powers (determinant matters)

There’s also a sharpness story for the symmetric power variant: the determinant factor (\det\mathcal E) is essential. The paper discusses explicit constructions by Di Rocco–Jabbusch–Smith and Nødland showing that even for ample toric vector bundles, the symmetric powers (without the determinant twist) may fail to be globally generated.

This isn’t just a minor caveat—it’s a warning sign for anyone hoping to simplify the statement further: positivity alone isn’t enough; the correct adjoint structure matters.

What’s new compared to earlier approaches? (a quick comparison table)

A lot of previous work proves Fujita-type results for special toric constructions or for special polarizations. Laface’s contribution is that the argument works directly on (\mathbb P_X(\mathcal E)) and handles arbitrary (a) and arbitrary rank.

Here’s a high-level comparison based on what the paper emphasizes:

This is also why the paper is framed as an “integral realization” of the Seshadri-constant mechanism: it’s not just re-proving something known in a disguised form; it’s giving a proof strategy that works broadly without translating to formal (\mathbb Q)-twists.

Key Takeaways

• Fujita’s freeness holds for all projectivized toric vector bundles over smooth projective toric varieties, not just special cases.
• For (Y=\mathbb P_X(\mathcal E)) with (\dim X=n) and (\operatorname{rank}\mathcal E=r\ge 2), global generation of the relevant adjoint-type line bundles holds once
  [
  m\ge n+1
  \quad\text{and}\quad
  ma\ge r
  ]
  (with the paper’s precise (\delta(A))-refined inequalities giving the same conclusion because (\delta(A)\ge 1)).
• The proof is constructive in spirit: blow up a torus-fixed point, check positivity on invariant curves, use Kodaira vanishing for surjectivity on fibers, then use torus invariance to eliminate global base loci.
• There’s a parallel result for adjoint symmetric powers on (X): global generation is obtained under related numeric constraints (in the clean case (a=1), it boils down to conditions like (q+r\ge n+1)).
• The bounds are sharp: the paper gives explicit examples (notably with (X=\mathbb P^n)) where global generation holds only when the stated numeric thresholds are met.
• The determinant twist is not cosmetic: without it, even ample toric bundles can fail to have globally generated symmetric powers (supported by prior explicit constructions referenced in the paper).

If you want, tell me what level of background you prefer (beginner-friendly intuition vs. more geometric details), and I can rewrite the blow-up/positivity part with an even more concrete “picture” of what (\delta(A)) is measuring.