🔢 mathematics

Failure of the invariant cycle theorem over $\mathbb Z$

This paper establishes that while the local invariant cycle theorem with integral coefficients holds for $H^1$ and for $H^2$ when the general fiber has a trivial Albanese variety, it can fail for $H^2$ in semistable families of algebraic surfaces with non-trivial Albanese varieties, as demonstrated by a newly constructed counterexample generalizing the Shioda–Inose construction.

Original authors: Donu Arapura, François Greer, Yilong Zhang

Published 2026-05-18

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Donu Arapura, François Greer, Yilong Zhang

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

The Big Picture: A Broken Rule of Thumb

Imagine you are a detective trying to solve a mystery about shapes and their hidden patterns. In the world of mathematics, specifically geometry, there is a famous rule called the Invariant Cycle Theorem.

Think of this rule like a promise: "If you have a family of shapes that change smoothly over time (like a clay pot being slowly reshaped), any pattern that stays the same (is 'invariant') during the changes must have come from the original, unbroken object."

For a long time, mathematicians knew this rule worked perfectly if you were counting with fractions (rational numbers). It was like saying, "If you can split your clues into halves or thirds, the rule holds."

However, this paper asks a harder question: Does this rule still hold if we only use whole numbers (integers)? In math, this is called working "over $\mathbb{Z}$ ."

The authors, Donu Arapura, François Greer, and Yilong Zhang, discovered that the rule breaks when you use whole numbers. They found a specific family of shapes where a pattern stays the same throughout the changes, but it cannot be traced back to the original object using whole numbers. It's like finding a fingerprint at a crime scene that matches the suspect, but the suspect claims, "I was never there," and the math proves him right because the fingerprint doesn't fit the whole-number mold of his alibi.

The Cast of Characters

To understand how they broke the rule, we need to meet the "actors" in their story:

The Family of Surfaces: Imagine a movie where a 3D object (a surface) slowly morphs. Most of the time, it looks like a smooth, complex shape. But at the very end (the "central fiber"), it collapses into a messy pile of pieces.
The "Elliptic-Elliptic" Surfaces: These are the main characters. They are special 2D shapes (surfaces) that look like a stack of donuts (elliptic curves) arranged over a base that is also a donut. They are complex, but they have a very specific, rigid structure.
The K3 Surfaces: Think of these as the "ancestors" or the "blueprints." They are a famous type of shape in geometry, known for being incredibly symmetric and "algebraic" (built from equations). The authors use a special type of K3 surface discovered by a mathematician named Vinberg, which is like the "most algebraic" K3 surface possible.

The Plot: How They Built the Counterexample

The authors didn't just find a mistake; they built a machine to create one. Here is the step-by-step process they used, explained with analogies:

1. The Shapeshifting Machine (Quadratic Base Change)
Imagine you have a K3 surface (the blueprint). The authors take this blueprint and run it through a "shapeshifting machine" called a quadratic base change.

Analogy: Imagine taking a map of a city and doubling the scale, but only along specific roads. This creates a new, slightly different map (a new surface).
The Trick: They chose to do this "doubling" in a way that creates a defect. Usually, when you double a map, the complexity doubles perfectly. But here, they branched the doubling at specific "star-shaped" intersections (singular fibers). This caused the complexity to drop unexpectedly.

2. The Constant Period Map (The Frozen Heart)
As they changed the parameters of their machine (moving a variable $t$ ), the resulting surfaces changed shape. However, the authors noticed something magical: the "heart" of these surfaces (their transcendental Hodge structure) remained frozen.

Analogy: Imagine a kaleidoscope. As you twist the tube, the colored glass pieces move around, but the central pattern in the middle stays exactly the same.
Why this matters: Because the "heart" is frozen, any pattern that stays the same (invariant) in the changing surfaces should be able to be found in the original family.

3. The Collapse (The Degeneration)
The authors then pushed the machine to a breaking point. They let the variable $t$ hit a specific value where the surface collapses into a singular mess (a "degeneration").

The Twist: When the surface collapsed, the "frozen heart" didn't just shrink; it changed its internal structure in a way that whole numbers couldn't handle.
The Math: They calculated the "discriminant" (a number that measures the complexity of the pattern).
- For the smooth surfaces, the discriminant was 3.
- For the collapsed surface, the discriminant was 48.
- Because 48 is not a simple multiple of 3 in the world of whole numbers (it's a factor of 16 difference), the pattern from the smooth surface cannot be lifted back to the collapsed surface using whole numbers.

The Verdict: The Rule is Broken

The paper proves two main things:

Local Failure: If you look at a small neighborhood around the collapse, the rule fails. You have a pattern that is invariant (stays the same), but it cannot be explained by the total space of the family using whole numbers.
Global Failure: Even if you look at the entire family over a whole curve (not just a small spot), the rule still fails.

The "Why" in Simple Terms:
The authors show that the "invariant" pattern exists in the smooth parts, but when you try to pull it back to the singular part using whole numbers, it gets "stuck" or "lost" because the geometry of the collapse is too coarse for whole numbers to bridge the gap. It's like trying to fit a square peg (the invariant pattern) into a round hole (the singular fiber) when you are forced to use only integer-sized tools.

What About the Good News?

The paper isn't just about breaking things; it also confirms where the rule does work.

For $H^1$ (First level of complexity): The rule always holds, even with whole numbers.
For $H^2$ (Second level): The rule holds IF the shapes have no "holes" that look like a donut (specifically, if the Albanese variety is trivial). This covers many famous shapes like K3 surfaces and Calabi-Yau varieties.
The Exception: The rule only breaks for surfaces that have a "donut-like" structure ( $q=1$ ) and a specific type of complexity (transcendental classes).

Summary

This paper is a mathematical detective story. The authors constructed a specific family of geometric shapes (elliptic-elliptic surfaces) that morphs smoothly but collapses into a singular form. They proved that while a "frozen" pattern exists in the smooth parts, it cannot be traced back to the whole family using whole numbers.

They used a clever construction involving Vinberg's "most algebraic" K3 surface and a defective double cover to create a scenario where the "discriminant" (the complexity score) jumps from 3 to 48. This jump proves that the Invariant Cycle Theorem, which was thought to be a universal law for these families, actually fails when you restrict yourself to whole numbers.

It's a reminder that in the world of geometry, what works for fractions doesn't always work for integers, and sometimes, the most beautiful patterns are the ones that break the rules.

Technical Summary: Failure of the Invariant Cycle Theorem over $\mathbb{Z}$

Problem Statement
The paper investigates the validity of the Local Invariant Cycle Theorem (LICT) and the Global Invariant Cycle Theorem (GICT) when coefficients are taken in the integers ( $\mathbb{Z}$ ) rather than the rationals ( $\mathbb{Q}$ ). Classically, for a flat family of complex projective varieties $f: X \to \Delta$ over a disk with a smooth total space and smooth fibers over the punctured disk $\Delta^*$ , the restriction map $H^n(X, \mathbb{Q}) \to H^n(X_t, \mathbb{Q})^{\pi_1(\Delta^*)}$ is surjective onto the monodromy-invariant part. This result, established by Clemens, Schmid, and Deligne, relies on the exactness of the Clemens–Schmid sequence or the Decomposition Theorem.

The central question addressed (Question 1.2) is whether this surjectivity holds integrally for semistable families (where the central fiber is a reduced normal crossing divisor). While the rational case is well-understood, the integral case had remained open, with partial positive results known for specific classes of surfaces (K3 surfaces and I-surfaces) by Friedman, Griffiths, and Scattone.

Methodology
The authors employ a combination of Hodge theory, algebraic geometry, and lattice theory to address the problem in two phases: establishing positive results in low degrees and constructing counterexamples in degree 2.

Positive Results (Low Degrees):
- $H^1$ : The authors prove that the LICT holds for $H^1$ with integral coefficients for any semistable family. The proof strategy involves reducing the general case to curves via a relative Lefschetz hyperplane theorem. For curves, they utilize the relative compactified Jacobian and the Mumford construction of semistable models for abelian varieties. By analyzing the mixed Hodge structure of the central fiber (a semiabelian variety) and the associated tropicalization (moment maps), they demonstrate the exactness of the relevant sequences over $\mathbb{Z}$ .
- $H^2$ (Vanishing Albanese): They prove that if the general fiber $X_t$ has a trivial Albanese variety (equivalently $q(X_t) = 0$ ), the LICT holds for $H^2$ . This generalizes previous results for K3 surfaces and I-surfaces. The proof relies on slicing the Clemens–Schmid sequence and utilizing the exactness of the Wang sequence and the pair sequence $(X, X \setminus X_0)$ over $\mathbb{Z}$ .
Negative Results (Counterexamples):
- Construction: To disprove the theorem for $H^2$ when $q(X_t) = 1$ , the authors construct a specific semistable family of "elliptic-elliptic" surfaces (surfaces with $p_g = q = 1$ ). These surfaces are constructed as rational double covers of elliptic K3 surfaces via a "defective" quadratic base change.
- Shioda–Inose Analogy: The construction is modeled on the Shioda–Inose structure for K3 surfaces. The authors start with a specific K3 surface $X$ (Vinberg's "most algebraic" K3 surface with discriminant 3) admitting an elliptic fibration with star fibers. They perform a quadratic base change branched at three star fibers and one smooth fiber.
- Hodge Theory Analysis: The resulting family of surfaces $S_t$ has a constant Variation of Hodge Structure (VHS) on the transcendental part of $H^2$ . The authors analyze the degenerations of this family at specific points where the branch point collides with the star fibers.
- Lattice Discriminants: The core of the counterexample lies in comparing the discriminants of the transcendental lattices. The smooth fibers $S_t$ have a transcendental lattice with discriminant 3. However, at a specific singular fiber (associated with a different limiting K3 surface), the transcendental lattice of the central fiber has discriminant 48.
- Specialization Map: The authors show that the specialization map induces an injection on the transcendental parts that preserves the cup product pairing. Because the discriminant jumps from 3 to 48, the image of the restriction map $H^2(X, \mathbb{Z}) \to H^2(X_t, \mathbb{Z})^{\pi_1}$ is a proper sublattice of index 4. Thus, the map is not surjective.

Key Contributions and Results

Theorem 1.3 & 1.4: The LICT holds over $\mathbb{Z}$ for $H^1$ in any semistable family, and for $H^2$ if the general fiber has trivial Albanese variety ( $q=0$ ). This covers hypersurfaces, Calabi-Yau varieties, and hyperkähler varieties.
Theorem 1.5 (Local Failure): The LICT fails over $\mathbb{Z}$ for semistable families of surfaces with $p_g = q = 1$ . The authors provide the first explicit example of a semistable degeneration where the restriction map on $H^2$ is not surjective. The failure is driven by the behavior of transcendental classes, not torsion classes (the cohomology is torsion-free).
Theorem 1.7 (Global Failure): The Global Invariant Cycle Theorem (GICT) also fails over $\mathbb{Z}$ for $H^2$ in general. The authors construct a smooth projective threefold fibered over a curve where the global invariant cycles do not lift to the total space cohomology, despite the Leray spectral sequence degenerating over $\mathbb{Z}$ for the smooth part.
Corollary 1.6: As a consequence, the Clemens–Schmid sequence is not exact over $\mathbb{Z}$ for these families, and the Decomposition Theorem fails integrally.
Counterexample 2 (Section 9): A second counterexample is constructed using Vinberg's second "most algebraic" K3 surface (discriminant 4), demonstrating the robustness of the failure mechanism.

Significance
The paper establishes that the classical invariant cycle theorems, which are foundational in the study of degenerations of algebraic varieties, do not hold integrally in cohomological degree 2 for families with non-trivial Albanese varieties. This result sharpens the understanding of the limitations of the Clemens–Schmid sequence and the Decomposition Theorem when torsion-free integral coefficients are required.

The construction of these counterexamples relies on a novel application of quadratic base changes to elliptic surfaces, creating a family with constant period maps but varying lattice discriminants upon degeneration. This highlights a subtle obstruction in integral Hodge theory: even when the Hodge structure varies trivially, the integral lattice structure can degenerate in a way that prevents the lifting of invariant cycles. The work generalizes the Shioda–Inose construction and provides a new class of "elliptic-elliptic" surfaces with maximal Picard rank and minimal discriminant, closely related to Vinberg's most algebraic K3 surfaces.