🔢 mathematics

On the canonical bundle formula in positive characteristic

Assuming the Minimal Model Program and the existence of log resolutions in dimension $n$ , this paper establishes the nefness of the moduli part in the canonical bundle formula for fibrations over curves in positive characteristic $p>2$ , thereby proving the formula unconditionally for 3-dimensional dlt pairs when $p>5$ .

Original authors: Marta Benozzo

Published 2026-05-25

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marta Benozzo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand a complex, multi-layered cake (a high-dimensional mathematical shape called a variety). You want to know how the ingredients of the whole cake relate to the ingredients of just one slice (a fibration or a map from the cake to a simpler base).

In the world of algebraic geometry, there is a famous recipe called the Canonical Bundle Formula. It tells us how to reconstruct the "flavor" (the canonical divisor) of the whole cake based on the flavor of the slice and the base it sits on.

For a long time, mathematicians knew this recipe worked perfectly in a world called Characteristic 0 (think of this as a world where numbers behave like standard real numbers). However, in Positive Characteristic (a world where numbers wrap around like a clock, common in computer science and cryptography), the recipe was broken. The ingredients didn't mix right, and the "flavor" of the slice didn't seem to predict the whole cake.

Marta Benozzo's paper is like a master chef who has figured out how to fix this broken recipe for a specific type of cake (where the base is a simple curve) in this tricky "clock-number" world.

Here is a breakdown of her work using simple analogies:

1. The Problem: The "Wild" Ingredients

In the standard world (Characteristic 0), if you slice a cake, the slice looks smooth and predictable. But in the "clock-number" world (Positive Characteristic), slicing the cake can sometimes produce wild, jagged edges or strange, singular points that don't behave like normal slices.

Mathematicians call these "wild fibres." Because of these wild edges, the old recipe failed. You couldn't just look at the slice and say, "Ah, this is what the whole cake tastes like."

2. The Solution: The "Frobenius" Magic Trick

Benozzo's main trick is a technique called Frobenius Base Change.

The Analogy: Imagine you have a blurry photo of a cake slice. You can't see the details. Instead of trying to sharpen the photo directly, you take a magical "x-ray" of the cake (a specific mathematical operation involving powers of the prime number $p$ ).
The Result: This x-ray reveals that the "wild" jagged edges on the slice actually correspond to smooth, normal features on the whole cake. It's like realizing that the jagged edge was just a shadow cast by a smooth object. By looking at the cake through this "x-ray" lens, the wild problems disappear, and the slice looks normal again.

3. The "Moduli Part": The Secret Sauce

The formula has two main parts:

The Discriminant Part: This accounts for the obvious scars and cuts on the cake (the singularities).
The Moduli Part: This is the "secret sauce." It measures how much the cake changes as you move along the base. If every slice is identical, the secret sauce is zero. If the slices change flavor, the sauce is strong.

The big question was: Is this secret sauce "positive"?
In math, "positive" (or nef) means the sauce is stable and well-behaved. If it's not positive, the whole theory falls apart. In the "clock-number" world, people thought this sauce might be spoiled.

4. The Breakthrough

Benozzo proves that yes, the secret sauce is positive, provided you follow her new steps:

Clean up the cake: First, she uses a process called a $(*)$ -modification. Think of this as smoothing out the frosting and rearranging the layers of the cake so that the "wild" parts are hidden or fixed.
Use the X-ray: She applies the Frobenius trick to show that even if the cake looks weird, the underlying structure is actually very orderly.
The Bend and Break: She uses a clever geometric argument (like bending a stick until it breaks) to prove that if the sauce weren't positive, you would find a contradiction—a "broken stick" that shouldn't exist. Since the stick doesn't break, the sauce must be positive.

5. The Result: A Working Recipe for 3D Cakes

The paper concludes with a powerful result for 3-dimensional cakes (threefolds) in a world where the clock numbers are greater than 5 ( $p > 5$ ).

The Claim: If you have a 3D cake that is "log canonical" (a specific type of well-behaved, though slightly imperfect, cake), and you slice it over a curve, the Canonical Bundle Formula works unconditionally.
What this means: You can now reliably predict the flavor of the whole 3D cake just by looking at the slice and the base, even in this tricky "clock-number" world.

Summary

Think of this paper as a repair manual for a broken recipe book.

The Problem: The recipe for mixing cake layers failed in "clock-number" math because of wild, jagged slices.
The Fix: Benozzo showed that by using a special "x-ray" view (Frobenius) and smoothing the cake layers (modifications), the wild slices turn out to be normal.
The Payoff: She proved that the "secret sauce" (the moduli part) is always stable and positive. This allows mathematicians to finally use this powerful formula to study complex 3D shapes in positive characteristic, a feat that was previously thought impossible or unreliable.

In short, she took a broken tool, figured out why it was broken, and showed us exactly how to fix it so we can build better mathematical structures in the future.

Technical Summary: On the Canonical Bundle Formula in Positive Characteristic

Problem Statement
The classification of algebraic varieties relies heavily on the Minimal Model Program (MMP) and the study of the positivity properties of the canonical divisor. A central tool in this context is the canonical bundle formula, which relates the canonical divisor of a fibration $f: X \to Z$ to the canonical divisors of the base and the fibers, decomposing the relation into a discriminant part (measuring singularities) and a moduli part (measuring variation in moduli).

While this formula is well-established in characteristic 0 (e.g., via variations of Hodge structures or the Cone theorem for foliations), its validity in positive characteristic $p > 0$ has been obstructed by several factors:

Failure of Generic Smoothness: In positive characteristic, the geometric generic fiber of a fibration may be singular even if the total space is smooth, preventing the direct definition of a discriminant part.
Wild Ramification: The presence of "wild" fibers (where multiplicities are divisible by $p$ ) complicates the relationship between the canonical divisor of the total space and the foliation induced by the fibration.
Failure of the Cone Theorem for Foliations: The standard proof of the positivity of the moduli part in characteristic 0 relies on the Cone theorem for algebraically integrable foliations, a result known to fail in positive characteristic.
Lack of Semistable Reduction: The construction of "good" models (satisfying Property $(*)$ ) used in characteristic 0 often relies on semistable reduction theorems that do not hold in the same generality in positive characteristic.

This paper addresses these issues by establishing a canonical bundle formula for fibrations over a curve in positive characteristic, specifically proving the nefness of the moduli part under specific assumptions.

Methodology
The author adapts the strategy introduced by Acosta, Cascini, Svaldi, and Spicer (ACSS21) for characteristic 0, modifying it to overcome the specific obstacles of positive characteristic. The methodology proceeds through the following key steps:

Reduction to Property $(*)$ : The proof strategy involves first establishing the result for a specific class of fibrations satisfying "Property $(*)$ ," where the moduli part coincides with the canonical divisor of the induced foliation (up to correction terms). The general case is then reduced to this class via birational modifications.
Handling Non-Normal Generic Fibers: In positive characteristic, the geometric generic fiber may not be normal. To resolve this, the author employs a Frobenius base change. By considering the diagram involving the $e$ -th power of the Frobenius morphism ( $F^e$ ), the singularities of the geometric generic fiber are "lifted" to the total space $X^{(e)}$ . This allows the application of adjunction and MMP techniques on a model where the fibers are normal.
Foliations and Purely Inseparable Maps: The paper exploits the correspondence between foliations and purely inseparable morphisms in positive characteristic. Since the differential of the Frobenius morphism vanishes, purely inseparable maps define foliations. The author derives a formula relating the canonical divisor of the Frobenius base change $X^{(e)}$ to the moduli part of the original fibration, accounting for wild ramification terms ( $W(f)$ ).
Bend and Break for the Moduli Divisor: Since the Cone theorem for foliations fails in positive characteristic, the author cannot directly construct rational curves tangent to the foliation. Instead, the proof utilizes the Bend and Break theorem applied to the variety $X^{(e)}$ obtained after Frobenius base change. By relating the canonical divisor of $X^{(e)}$ to the moduli part, the author constructs vertical rational curves that demonstrate the negativity of the moduli part if it were not nef, leading to a contradiction.
Induction on Dimension: The proof of the main theorem proceeds by induction on the dimension of $X$ . This requires performing adjunction on log canonical centers of the geometric generic fiber. To ensure these centers exist and behave well, the author constructs "geometric $(*)$ -modifications" that extract divisors with discrepancy $-1$ over these centers, utilizing log resolutions and the dlt MMP.

Key Contributions and Results

The Canonical Bundle Formula (Theorem 7.9): Assuming the log MMP, inversion of adjunction, and the existence of log resolutions in dimension up to $n$ , the paper proves that for a fibration $f: X \to Z$ from a normal projective variety of dimension $n$ to a normal curve $Z$ over a perfect field of characteristic $p > 2$ , if $(X, B)$ is a dlt pair with log canonical general fibers and $K_X + B$ is $f$ -nef, then there exists a birational model $(Y, C)$ satisfying Property $(*)$ such that the moduli part $M_Y$ is nef.
Positivity in the K-Trivial Case (Theorem 7.10): As a corollary, the paper proves that if $K_X + B \sim_{\mathbb{Q}} f^*L$ (the K-trivial case), the moduli part $M_Z = L - (K_Z + B_Z)$ is nef.
Unconditional Result in Dimension 3 (Corollary 7.11): For threefolds over an algebraically closed field of characteristic $p > 5$ , the assumptions of the log MMP, inversion of adjunction, and log resolutions are known to hold. Consequently, the canonical bundle formula and the nefness of the moduli part hold unconditionally for dlt pairs with log canonical general fibers.
Formulas for Wild Ramification: The paper provides explicit formulas (Theorem 3.6, Theorem 3.25) for the canonical divisor of the foliation induced by a fibration in positive characteristic, introducing a correction term $W(f)$ supported on wild components. It further relates the canonical divisor of the Frobenius base change $X^{(e)}$ to the moduli part (Corollary 5.4).

Significance and Claims
The paper claims to extend the canonical bundle formula to positive characteristic, a setting where previous techniques (Hodge theory, standard Cone theorems) were inapplicable. The significance lies in:

Overcoming Generic Smoothness Failure: By using Frobenius base changes, the author bypasses the issue of non-normal generic fibers, a major hurdle in positive characteristic birational geometry.
Bypassing the Cone Theorem: The use of the Bend and Break theorem on the Frobenius base change provides a viable alternative to the Cone theorem for foliations, which is known to fail in positive characteristic.
Completing the Picture in Dimension 3: The results provide a complete, unconditional canonical bundle formula for threefolds in characteristic $p > 5$ , filling a gap in the classification theory of varieties in positive characteristic.

The author explicitly notes that the results rely on the existence of log resolutions and the validity of the MMP in the relevant dimensions, which are currently known only for dimensions $\le 3$ (and partially for dimension 4 with standard coefficients). The paper does not claim to solve the problem for all dimensions or all characteristics without these assumptions, nor does it propose new experimental applications, but rather establishes the theoretical foundation necessary for inductive arguments in the classification of varieties in positive characteristic.