Reconstructing a function field from small arithmetic invariants

← All digests

Family 009 · Number theory
Importance 4/5 · Provisional editorial rating
Sources consulted 2026-10-10

1. The question

A field comes with addition and multiplication. Suppose much of that information is discarded: keep nonzero elements only modulo powers, together with certain relations connecting a number to one minus that number. Can the original field still be recovered?

For function fields of geometric dimension at least two over algebraically closed constants, family 009 claims a remarkably strong yes. Two manuscripts use only the first two Milnor K-groups modulo one prime and their product. A third uses a small quotient of the absolute Galois group, retaining commutators but discarding further layers of noncommutativity.

These are reconstruction theorems, not just tests distinguishing some examples. They claim that every compatible isomorphism of the reduced data comes from a field isomorphism, with explicitly described scalar and, in some cases, Frobenius ambiguities. The precise answer depends on whether the coefficient prime equals the field’s characteristic.

2. Background and essential definitions

A function field $K/k$ is a finitely generated field extension. The elementary example $k(x,y)$ consists of rational functions in two independent variables. Its transcendence degree over $k$ is two: two algebraically independent parameters are needed before the remaining extension becomes algebraic. Geometrically, this is the dimension of a variety having $K$ as its rational-function field.

The constant field $k$ is assumed algebraically closed, so every nonconstant polynomial over $k$ has a root in $k$. A field has characteristic $p>0$ when adding one to itself $p$ times gives zero, and characteristic zero otherwise.

Fix a prime $\ell$. The first piece of data is

$$V_K=K^\times/(K^\times)^\ell,$$

where $K^\times$ is the multiplicative group of nonzero elements. Two elements have the same class if their ratio is an $\ell$-th power. Written additively, this is a vector space over $\mathbb F_\ell$, with

$$[fg]=[f]+[g],\qquad [f^\ell]=0.$$

Thus its addition records multiplication in the original field, not the original field addition.

The second piece is

$$W_K=(V_K\otimes_{\mathbb F_\ell}V_K)/R_K, \qquad R_K=\operatorname{span}\{[f]\otimes[1-f]:f\ne0,1\}.$$

The tensor product records bilinear combinations; quotienting by $R_K$ imposes the Steinberg relations. These are exactly the degree-one and degree-two Milnor K-groups modulo $\ell$. Their product is the quotient map

$$m_K:V_K\otimes V_K\longrightarrow W_K.$$

Although $V_K$ forgets addition, the relation between $f$ and $1-f$ retains a trace of it. A compatible isomorphism $\Theta:V_K\to V_L$ must carry the entire relation space $R_K$ onto $R_L$, equivalently preserve the full bilinear product after identifying degree two.

In characteristic $p$, Frobenius is the map $z\mapsto z^p$. The perfect closure

$$K^{\mathrm{perf}}=\bigcup_{r\ge0}K^{1/p^r}$$

adjoins every iterated $p$-th root. Frobenius becomes invertible there. In characteristic zero, the perfect closure is just $K$.

A valuation measures orders of zeros or poles, such as the exponent of $x$ in a rational function near $x=0$. Divisorial valuations come from irreducible codimension-one geometric subsets. A central theme of the manuscripts is recovering this geometry from data that do not label any valuations in advance.

3. Development history and previous partial results

Bogomolov proposed reconstructing function fields from a small portion of their Galois groups, retaining the abelian information and its first commutators. This belongs to birational anabelian geometry: recovering rational-function fields from arithmetic symmetry data.

Bogomolov and Tschinkel established reconstruction for surfaces over algebraic closures of finite fields, and subsequently for higher dimensions over those constants. Pop obtained a corresponding reconstruction theorem. These results showed that the small quotient could retain enough geometric information in a substantial setting.

Arbitrary algebraically closed constants introduce another difficulty: locally recovered valuations can act nontrivially on the constants themselves. They need not initially be the geometric divisors needed for reconstruction. Pop’s theorem with divisorial inertia treated broader constants in relative transcendence degree greater than two when additional inertia data were supplied. The challenge here is to recover that information from the invariant itself.

For finite coefficients, Topaz’s mod-$\ell$ reconstruction theorem worked in transcendence degree at least five with extra rational-subfield data. Those data identify images of multiplicative groups of suitable one-variable rational subfields. M1 claims to derive the needed subfields from the product alone and to work already in dimension two.

A related approach reconstructs addition through algebraic dependence and projective geometry. Bogomolov–Tschinkel and Cadoret–Pirutka developed this direction for other multiplicative and Milnor K-theoretic invariants. M2 adapts the projective-line approach to reduction modulo the characteristic, using derivations and degree-$p$ extensions. These historical theorems use different input data; they should not be treated as identical versions of the present claims.

4. What the new manuscripts claim

All three manuscripts assume finitely generated extensions $K/k$ and $L/l$ of algebraically closed fields, with relative transcendence degrees at least two.

M1: one prime different from the characteristic

For any prime $\ell$ different from both characteristics, M1 claims that compatible isomorphisms of $(V,W,m)$, modulo a single global scalar in $\mathbb F_\ell^\times$, correspond bijectively to isomorphisms

$$K^{\mathrm{perf}}\longrightarrow L^{\mathrm{perf}}$$

carrying $k$ onto $l$, modulo integral powers of Frobenius in positive characteristic. Both positive and negative powers are allowed on perfect closures. In characteristic zero there is no Frobenius quotient.

A compatible isomorphism forces equality of the characteristics and of the transcendence degrees. No rational subgroups, valuations, or constant-field identification are supplied as additional input. The scalar is global, rather than chosen separately for each element. At $\ell=2$, the scalar group is trivial.

Scaling the degree-one map by $a$ scales the induced degree-two map by $a^2$. In characteristic $p\ne\ell$, Frobenius acts on degree one by the nonzero scalar $p$ modulo $\ell$, explaining why its powers cannot be distinguished after quotienting by scalars.

M2: reduction modulo the characteristic

M2 assumes both fields have characteristic $p$ and uses $V_K=K^\times/(K^\times)^p$. It claims a bijection

$$\operatorname{Isom}(K,k;L,l) \ \cong\ \operatorname{Isom}_{M}(V_K,V_L)/\mathbb F_p^\times.$$

The left side means field isomorphisms carrying the named constant fields onto one another; the right side means compatible maps modulo scalar multiplication. Here the fields themselves are recovered, and there is no Frobenius ambiguity. This is a different theorem, not M1 with $\ell$ mechanically replaced by $p$. Frobenius sends every class in $V_K$ to zero and so does not supply an invertible compatible map in this setting.

M3: the Galois formulation

Let $G_K^{(\ell)}$ be the maximal pro-$\ell$ quotient of the absolute Galois group. A pro-$\ell$ group is an inverse limit of finite groups of $\ell$-power order. M3 retains

$$\Pi_K^a=G_K^{(\ell)}/[G_K^{(\ell)},G_K^{(\ell)}],$$

and the larger quotient $\Pi_K^c$ in which all commutators are central. The datum includes the resulting commutator pairing on $\Pi_K^a$, not just the abelian group by itself.

For $\ell$ different from the characteristic, M3 claims reconstruction of the perfect closure and constants from this datum. Every continuous bracket-compatible isomorphism is induced contravariantly by a field isomorphism, modulo one $\ell$-adic unit scalar and the same Frobenius ambiguity. “Contravariantly” means the direction of the induced Galois map is reversed. Surfaces and $\ell=2$ are included.

5. Why the advance matters

Provisional importance: 4/5. These claims would complete a substantial reconstruction program while reducing the amount of arithmetic information that needs to be retained. Recovering a field from a large symmetry group is already striking; recovering it from two low-degree finite-coefficient groups and their product is stronger compression of the input.

The significance is also functorial. The statements identify every compatible isomorphism and its ambiguity, not merely assert that some field with the right invariant exists. Such control is important when translating symmetries of invariants back into geometry.

The inclusion of arbitrary algebraically closed constants and dimension two closes gaps between previous settings. M2’s recovery before perfect closure adds a distinct positive-characteristic phenomenon. The rating reflects these structural consequences if the stated theorems hold.

6. The main idea

Recover local geometry from relations

For M1 and M3, the initial step detects pairs of multiplicative characters with a compatibility property. In the finite-coefficient formulation, characters $f,g:K^\times\to\mathbb F_\ell$ are tested by

$$f(x)g(1-x)=f(1-x)g(x) \qquad(x\ne0,1).$$

This condition is visible in the Steinberg relations. Established local theory converts suitable collections of these pairs into valuations and residue-field information. The first recovered valuations may also measure the constants; separating true geometric divisors comes later.

The next task is global. Fix a nonconstant rational function $t$. Its expressions $t-a$, as $a$ runs through the constants, describe the fibers of a map to a projective line. The manuscripts study the transported classes of this entire family, called a pencil.

They claim a uniform bound on the degrees of the divisors supporting those classes. The argument tests finite collections on residual curves of bounded genus and uses the Riemann–Hurwitz formula, which relates the genus and ramification of a map between curves. Five fixed members of the pencil control every further member. Uniformity over the whole pencil is stronger than a separate finite bound for each class.

Recover curves and then addition

Infinitely many divisors of bounded degree lie in an algebraic parameter family. Studying which points belong to which divisors produces an incidence construction and eventually a one-variable subfield. The argument must use one parameter curve for all relevant classes and descend it from a finite extension without losing the information being reconstructed.

M1 and M3 handle that descent differently. M1 works modulo one prime, where taking a norm can annihilate information, and retains an exact compositum relation. M3 uses completed multiplicative groups and norm arguments appropriate to that setting. Both aim to recover the curve subfields that distinguish true divisors and rational quotients.

M3 then invokes Pop’s global reconstruction theorem with the newly recovered geometric data. M1 instead reconstructs field operations from simultaneous values of rational functions at points. Compatible value maps align to one map of constants and preserve algebraic relations among generators, producing the perfect-field isomorphism.

Why characteristic $p$ permits a different route

For M2, put $C=K^p$. Then

$$K^\times/(K^\times)^p=K^\times/C^\times$$

is the set of projective points of the vector space $K$ over $C$: a point is a one-dimensional subspace $Cf$. Its projective lines come from two-dimensional $C$-subspaces. This projective geometry is over $K^p$, not over the original constants $k$.

The degree is $[K:K^p]=p^d$, where $d=\operatorname{trdeg}(K/k)$. Derivations, additive maps satisfying the product rule, detect independence over $K^p$. M2 uses determinants of logarithmic derivatives to show that a vanishing symbol $\{s,t\}$, with $s\notin K^p$, forces $t\in K^p(s)$.

Those degree-$p$ subfields are still larger than projective lines in general. An intersection calculation recovers lines after a power correction. Uniqueness propagates the correction to one global scalar. A projective lifting theorem then provides a semilinear map, meaning an additive map linear up to a field automorphism of the scalars. The multiplicative structure turns its normalized lift into a field isomorphism.

Finally,

$$\bigcap_{m\ge1}K^{p^m}=k$$

recovers the original algebraically closed constants. This explains why their identification need not be supplied separately.

7. Scope, limitations, and open questions

The dimension-at-least-two and algebraically closed constant-field hypotheses are part of all three statements. The manuscripts do not claim these results unchanged for arbitrary fields, curve function fields, or finite constant fields.

The full bilinear Milnor product or the full specified commutator datum is required. Two abstract vector spaces without that relationship would not be the input described here. The pro-$\ell$ theorem also includes continuity and its $\ell$-adic scalar ambiguity.

Perfect-closure recovery in M1 and M3 is deliberately weaker than recovering a particular inseparable presentation of a field in positive characteristic. M2 avoids that ambiguity because its coefficient prime equals the characteristic, using different information and a different method.

These are structural reconstruction theorems, not practical complexity bounds for an algorithm operating on a finite encoding of the invariants. Extensions to other base fields, lower dimension, or still smaller input data remain separate questions.

8. Sources and further reading

Each manuscript’s Theorem 1.1 states its own isomorphism theorem and ambiguity. M1’s introduction explains the finite-coefficient pencil and incidence arguments; M2’s Sections 2–5 develop derivations, projective-line recovery, and field lifting; M3 identifies the geometric data it reconstructs before applying Pop’s theorem.

The primary works linked in Section 3 trace the Galois and Milnor K-theoretic approaches. Optional background entries on function fields, valuations, and perfect fields may help with terminology; the differing reconstruction conventions are stated above.

Source manuscripts

• M1. OpenAI, Reconstruction of Function Fields from Mod-ℓ Milnor K-Theory (October 5, 2026).
• M2. OpenAI, Reconstruction from Milnor K-theory modulo the characteristic (October 5, 2026).
• M3. OpenAI, The Bogomolov-Pop reconstruction theorem (September 23, 2026).

Sources consulted on 10 October 2026 at repository revision fd4aeeb. Download the manuscript citations.