On the defect in the generalized Grunwald--Wang problem

This paper demonstrates that, contrary to the classical Grunwald–Wang theorem's special case where the obstruction is a group of order 2, the obstruction in the generalized setting for valued fields is not always finite nor bounded independently of the number of places, even for rational function fields.

The Gist

Imagine you are a detective trying to solve a mystery about numbers and symmetry. This paper is about a famous rule in mathematics called the Grunwald-Wang Theorem, which acts like a "local-to-global" law.

Here is the simple breakdown of what the authors, David Harari and Tamás Szamely, discovered.

1. The Big Idea: The "Local" vs. "Global" Puzzle

Think of a country (let's call it Field K).

• Global View: You want to know if a specific pattern exists across the entire country.
• Local View: You check the pattern in specific towns (called valuations or points).

The Grunwald-Wang Theorem is a promise: "If you can find a pattern in every single town you check, then that pattern must exist in the whole country."

For a long time, mathematicians thought this promise was almost always true, except for a few weird, famous exceptions (like when dealing with the number 8 and the number 2).

2. The New Question: How Big is the "Gap"?

The authors asked a new question. They knew the promise usually holds, but they wondered: What happens when it doesn't?

Imagine you have a list of towns where a pattern exists.

• The Ideal World: The pattern exists everywhere in the country. (The "Gap" is zero).
• The Real World: Sometimes, the pattern exists in all the towns you checked, but not in the whole country.

The authors wanted to measure the size of this Gap (the "Defect").

• Question 1: Is this Gap always a small, manageable size (finite)?
• Question 2: If you check more and more towns, does the Gap stay small, or does it grow out of control?

3. The Discovery: The Gap Can Explode

The authors found that in many cases, the answer is NO. The Gap can be huge, and it can grow infinitely large.

They used a clever construction to prove this:

• The Setup: They imagined a world where the "country" is a line of numbers (like a rational function field), and the "towns" are specific points on that line.
• The Trick: They found a specific type of symmetry (related to the number 8) that behaves strangely.
• The Result:
  • If you pick a finite number of towns, the Gap is finite, but it can be arbitrarily large. If you pick the right towns, the Gap can be as big as you want.
  • If you pick an infinite number of towns, the Gap can become infinite. The pattern exists in every single town you check, but it never exists in the whole country.

4. The Analogy: The "Missing Key" Mystery

Let's use a metaphor to make this concrete.

Imagine you are trying to open a giant vault (the Global Field). You have a master key (the Global Solution).

• You go to many small safes (the Local Towns) scattered around the country.
• In every single small safe you visit, you find a key that fits perfectly.
• The Theorem says: "If you have keys for all the small safes, you must have the master key for the big vault."

The Authors' Discovery:
They found a scenario where you have keys for every small safe you check, but you still don't have the master key.

• The "Defect": This is the collection of all the "fake" keys that work locally but don't unlock the global vault.
• The Shock: They showed that you can have an infinite number of these fake keys. You can keep finding new towns with working keys, and the pile of "fake keys" (the defect) keeps growing forever.

5. Why Does This Matter?

In the world of math, this is a big deal because:

1. It breaks a rule of thumb: It shows that "checking the parts" doesn't always guarantee "understanding the whole," even in very well-behaved mathematical worlds.
2. It reveals hidden complexity: It proves that the "gap" between local and global isn't just a tiny, fixed error. It can be a massive, unbounded chasm.
3. It connects to other problems: This "gap" is related to how well we can approximate numbers (a concept called "weak approximation"). If the gap is huge, it means our ability to approximate numbers globally is much weaker than we thought.

Summary in One Sentence

The authors proved that in certain mathematical worlds, you can have a pattern that works perfectly in every single local neighborhood you check, yet fails completely in the big picture, and the "failure" can be infinitely large.

Technical Summary

Here is a detailed technical summary of the paper "On the Defect in the Generalized Grunwald–Wang Problem" by David Harari and Tamás Szamuely.

1. Problem Statement and Context

The paper investigates the Grunwald–Wang problem in the context of general valued fields, specifically focusing on the "defect" or the failure of the local-global principle for cohomology groups.

The Setup:

• Let $K$ be a field and $T$ a set of valuations of $K$.
• Let $M$ be a finite étale commutative group scheme over $K$.
• Consider the restriction map from the global cohomology to the product of local cohomologies:
  $$(r_v)_{v \in T}: H^1(K, M) \to \prod_{v \in T} H^1(K_v, M)$$
  where $K_v$ is the completion of $K$ at $v$.
• The authors define $\overline{H^1(K, M)}$ as the closure of the image of the global group in the product (equipped with the product topology).
• The Grunwald–Wang theorem holds if the map is surjective, i.e., $\overline{H^1(K, M)} = \prod_{v \in T} H^1(K_v, M)$.

The Classical Case:
For number fields ($K$) and $M = \mathbb{Z}/n\mathbb{Z}$ with finite $T$, the theorem holds except for the "special case" (related to the non-cyclicity of the Galois group of the extension generated by $2^r$-th roots of unity, where$2^r$is the highest power of 2 dividing$n$). In this classical setting, the cokernel of the map is known to be finite (order at most 2).

The Central Questions:
The authors ask whether these finiteness properties hold in the generalized setting (arbitrary fields $K$, infinite sets $T$, or general $M$):

1. Is the quotient $\left(\prod_{v \in T} H^1(K_v, M)\right) / \overline{H^1(K, M)}$ always finite?
2. If finite, is its order bounded independently of the set $T$?

2. Methodology

The authors employ Saltman's method via generic Galois extensions, as adapted and reinterpreted by Colliot-Thélène and Sansuc. The core technical tool is the flasque resolution of group schemes.

• Flasque Resolution: For a group $M$ of multiplicative type, there exists an exact sequence:
  $$1 \to M \to F \to P \to 1$$
  where $F$ is a flasque torus and $P$ is a quasi-trivial torus.
• Key Lemma (Lemma 2.1): The authors establish a canonical isomorphism between the cokernels of the restriction maps for $M$ and $F$:
  $$\text{Coker}\left(H^1(K, M) \to \prod_{v \in T} H^1(K_v, M)\right) \cong \text{Coker}\left(H^1(K, F) \to \prod_{v \in T} H^1(K_v, F)\right)$$
  This reduction is crucial because the cohomology of quasi-trivial tori vanishes (Hilbert's Theorem 90), simplifying the analysis to the behavior of the flasque torus $F$.
• Geometric Construction: The authors construct counterexamples using rational function fields $K = k(t)$, where valuations in $T$ correspond to closed points (or rational points) of the projective line $\mathbb{P}^1_k$. They utilize the property that for a flasque torus $F$, $H^1(K, F) \cong H^1(k, F)$ under certain conditions, allowing them to transfer cohomological properties from the base field $k$ to the function field $K$.

3. Key Contributions and Results

The paper provides negative answers to both central questions, demonstrating that the defect can be infinite and unbounded even in relatively simple geometric settings.

A. Infinite Cokernels (Theorem 1.2)

• Result: There exists a field $k$ of characteristic 0 such that for $K = k(t)$ and any finite set $T$ of discrete valuations coming from rational points of $\mathbb{P}^1_k$ (with $|T| \ge 2$), the cokernel of the map $H^1(K, \mathbb{Z}/8\mathbb{Z}) \to \prod H^1(K_v, \mathbb{Z}/8\mathbb{Z})$ is infinite.
• Condition: This requires $k$ to have infinite transcendence degree over $\mathbb{Q}$. The construction relies on finding a base field $k_0$ (e.g., $\mathbb{Q}_2$) where $H^1(k_0, F) \neq 0$ for the flasque torus associated with $\mathbb{Z}/8\mathbb{Z}$, and then extending $k_0$ to an infinitely generated field $k$ where $H^1(k, F)$ becomes infinite.

B. Unbounded Cokernels (Theorem 1.3)

• Result: Even for "small" base fields like $k = \mathbb{Q}$ or $k = \mathbb{Q}_2$, the size of the cokernel is unbounded as the set $T$ varies.
• Mechanism:
  • For $K = \mathbb{Q}_2(t)$, if $T$ is an infinite set of rational points, the quotient is infinite.
  • For $K = \mathbb{Q}(t)$, the authors use a lemma stating that for any finite separable extension $L/k$, there are infinitely many closed points in $\mathbb{P}^1_k$ with residue field $L$. By choosing $T$ to consist of points with a specific residue field $L$ where the local cohomology is non-trivial, the dimension of the cokernel over $\mathbb{F}_2$ grows arbitrarily large as $|T|$ increases.

C. Implications for Weak Approximation (Corollary 1.5)

• The results have direct consequences for the group $X^2_\omega(\mu_8^{\otimes 2})$, which consists of elements in $H^2(K, \mu_8^{\otimes 2})$ that vanish locally almost everywhere.
• Significance: The authors prove that for $K = \mathbb{Q}_2(t)$, this group is infinite. This answers a question posed by M. L. Nguyen regarding whether such groups can be non-trivial. These groups control weak approximation on quotients of semisimple simply connected $K$-groups by finite abelian constant subgroups.

4. Significance

1. Refutation of General Finiteness: The paper demonstrates that the finiteness of the Grunwald–Wang defect, a hallmark of number fields, does not extend to function fields over local or global fields, nor to fields of infinite transcendence degree.
2. Unboundedness: It shows that even when the defect is finite for a fixed $T$, its size is not uniformly bounded. This contrasts sharply with the classical number field case where the defect is bounded by 2.
3. Geometric Insight: The work highlights the deep connection between the arithmetic of the base field $k$ (specifically the cohomology of flasque tori) and the arithmetic of the function field $K=k(t)$.
4. Application to Algebraic Groups: By linking the defect to the group $X^2_\omega$, the paper provides new examples of failures of weak approximation in the context of algebraic groups, expanding the known landscape of obstructions in arithmetic geometry.

In summary, Harari and Szamuely utilize the machinery of flasque resolutions to construct explicit counterexamples showing that the generalized Grunwald–Wang problem can fail dramatically in terms of both the finiteness and the boundedness of the local-global defect.