Shelah's Conjecture on NIP Fields

Shelah's Conjecture on NIP Fields

Definitions and mathematical statement

A formula $\varphi(x;y)$ has the independence property if, in models of the theory, it can realize every subset of arbitrarily large finite sets: there are $a_1,\ldots,a_n$ and parameters $b_S$ with $\varphi(a_i;b_S)$ exactly when $i\in S$. A theory is NIP if no formula has this property. For an infinite field $K$ in the ring language, the selected conjecture asserts that NIP implies one of: $K$ is separably closed; $K$ is real closed; or $K$ admits a nontrivial henselian valuation. A valuation is henselian when its valuation ring satisfies Hensel's lifting of simple residue-field roots; nontrivial means its valuation ring is not all of $K$. No definability of the valuation is demanded by this formulation. The all-NIP-field quantifier is not replaced by finite-rank or otherwise restricted subclasses.

Formulation and concept references: [S1].

Short English statement

Must every infinite NIP field be separably closed, real closed, or carry a nontrivial henselian valuation?

Research attempt: separating NIP from the missing valuation construction

Status and variants. The statement concerns every infinite NIP field in the pure language of rings. It does not assume finite dp-rank, and it asks for a nontrivial henselian valuation without requiring that valuation to be definable. Johnson's primary paper [E1] proves the dp-finite case, with a definability conclusion in the relevant nonclosed cases. The characterization in [S1] concerns fields already equipped with a henselian valuation; its discussion of all NIP fields is conditional on the conjectural existence statement. These results were checked on 27 September 2026 and are not promoted here to a proof for arbitrary NIP fields.

The attempt develops three elementary parts of a possible strategy: permanence of NIP under field constructions that really are interpretations, the algebraic data needed to obtain a valuation, and the extra argument needed to make that valuation henselian. The gap between these parts is substantial.

Lemma 341.1 (finite extensions preserve NIP). If a field $K$ is NIP and $L/K$ is a finite field extension, then the pure field $L$ is NIP.

Proof. Choose a $K$-basis $e_1,\ldots,e_d$ of $L$. Identify its underlying set with $K^d$. Addition is coordinatewise, while multiplication is given by finitely many structure constants \[ e_i e_j=\sum_{r=1}^d c_{ijr}e_r, \qquad c_{ijr}\in K. \] Thus every ring formula over $L$ translates to a formula over $K$ in tuples of $d$ coordinates, using these finitely many constants as parameters. Any pattern witnessing the independence property in $L$ would translate to the same pattern in $K$. Naming finitely many parameters does not destroy NIP: a formula with those parameters is an instance of a formula with additional parameter variables, and an independence pattern for the instance is one for the original formula. This proves the claim. $\square$

The same translation argument applies to definable subfields, allowing parameters, and to reducts. The direction for reducts is important: forgetting a valuation predicate cannot introduce new formulas, but adding such a predicate can introduce them. The NIP of the pure field cannot simply be transferred to an arbitrarily chosen valued-field expansion.

An algebraic target for the construction. A subring $\mathcal O\subseteq K$ containing $1$ is a valuation ring if \[ \forall x\in K^\times\quad x\in\mathcal O\ \text{or}\ x^{-1}\in\mathcal O. \tag{341.1} \] Here is a direct reconstruction, useful for checking what a proposed ring of "small elements" actually establishes. Let \[ \mathfrak m=\{a\in\mathcal O:a=0\text{ or }a^{-1}\notin\mathcal O\}. \] These are the nonunits of $\mathcal O$. If $a,b\in\mathfrak m$ are nonzero, $341.1$ applied to $a/b$ shows that one of $a,b$ is an $\mathcal O$-multiple of the other. Suppose $b=ac$. Then $a+b=a(1+c)$ cannot be a unit, since a product can be a unit only if both factors are units. Also multiplication of a nonunit by any element of $\mathcal O$ gives a nonunit or zero. Hence $\mathfrak m$ is an ideal. Every element outside it is a unit, making it the unique maximal ideal.

The quotient group $\Gamma_v=K^\times/\mathcal O^\times$ becomes an ordered abelian group by declaring \[ v(x)\geq v(y)\quad\Longleftrightarrow\quad x/y\in\mathcal O. \] This is well defined under multiplication by units; $341.1$ gives totality, and membership in $\mathcal O$ gives transitivity. If $v(x)\geq v(y)$, then $x+y=y(1+x/y)$, so \[ v(x+y)\geq\min\{v(x),v(y)\} \] when $x+y\ne0$. With $v(0)=\infty$, this is a valuation. It is nontrivial exactly when $\mathcal O\ne K$.

Thus a construction must provide much more than an additive subgroup or a proper subring. It must establish $341.1$, properness, and eventually a lifting property for roots. Intersections of candidate rings need not preserve the total comparison in $341.1$, so intersecting all vaguely compatible neighborhoods is not automatically a valuation construction.

Lemma 341.2 (a complete discrete case of henselianity). Let $K$ be complete for a discrete valuation $v:K^\times\to\mathbb Z$, with valuation ring $\mathcal O$ and maximal ideal $\mathfrak m$. If $f\in\mathcal O[X]$ and $a_0\in\mathcal O$ satisfy \[ f(a_0)\in\mathfrak m,\qquad f'(a_0)\in\mathcal O^\times, \] then there is a unique $a\in a_0+\mathfrak m$ with $f(a)=0$.

Proof. Unless a root has already been found, define \[ a_{n+1}=a_n-\frac{f(a_n)}{f'(a_n)}. \] For $a,h\in\mathcal O$, polynomial expansion gives \[ f(a+h)=f(a)+hf'(a)+h^2R(a,h),\qquad f'(a+h)-f'(a)\in h\mathcal O, \] where $R(a,h)\in\mathcal O$. If $r_n=v(f(a_n))\geq1$ and $f'(a_n)$ is a unit, the correction $h_n$ has valuation $r_n$. The two identities show that $f'(a_{n+1})$ remains a unit and $r_{n+1}\geq2r_n$, unless the new value is zero. Hence $r_n\geq2^n$, and the corrections tend to zero rapidly. Completeness gives a limit $a\in a_0+\mathfrak m$, and continuity gives $f(a)=0$.

For uniqueness, if $a,b\in a_0+\mathfrak m$, factor \[ f(a)-f(b)=(a-b)U, \] where the polynomial difference quotient $U$ is congruent to $f'(a_0)$ modulo $\mathfrak m$. It is therefore a unit. Two roots in this residue class must be equal. $\square$

This proves the familiar simple-root form of Hensel's lemma for complete discretely valued fields. It establishes the required kind of root lifting once those stronger topological assumptions are available; it does not construct such a topology on an arbitrary NIP field.

Two stress tests on the proposed route. First, having a nontrivial valuation is not enough. On $\mathbb Q$ with its $3$-adic valuation, the polynomial $f(X)=X^2-7$ has a simple root $1$ modulo $3$, since $f(1)=-6$ and $f'(1)=2$. It has no rational root. Thus this valuation on $\mathbb Q$ is not henselian. Passing to a completion would produce a different field; it would not establish henselianity of the original field in the statement.

Second, henselianity does not by itself imply NIP. The complete discretely valued field $\mathbb F_p((t))$ is henselian by Lemma 341.2, but is not NIP. Here we use the primary theorem of Kaplan--Scanlon--Wagner [E2]: an infinite NIP field of characteristic $p$ has no nontrivial Artin--Schreier extension, equivalently the map $x\mapsto x^p-x$ is surjective.

For this example the failure of surjectivity is elementary. The equation \[ x^p-x=t^{-1} \tag{341.2} \] has no solution in $\mathbb F_p((t))$. If $v(x)\geq0$, its left side has nonnegative valuation. If $v(x)<0$, then $pv(x)<v(x)$, so the two terms have unequal valuations and \[ v(x^p-x)=pv(x), \] which is divisible by $p$ and cannot equal $-1$. The polynomial $X^p-X-t^{-1}$ therefore has no root. Its roots in a splitting field differ by elements of $\mathbb F_p$, and its Galois group is a subgroup of the additive group of order $p$; consequently absence of a root makes it an irreducible Artin--Schreier polynomial of degree $p$. This contradicts the necessary NIP condition from [E2]. The external theorem is used only for that necessary condition, not as a construction of a valuation.

Why finite rank is an additional hypothesis. NIP bounds the shattering complexity of each fixed formula. It does not, from its definition alone, give a single finite numerical rank controlling every family of definable sets. An argument that chooses a maximal finite collection of independent definable directions must justify the finiteness of that collection. In the dp-finite theorem [E1], a finite rank hypothesis supplies precisely additional structure that is absent from the selected statement. Dropping that hypothesis from the beginning of a proof would silently replace a theorem about one subclass by the open general assertion.

Similarly, adjoining parameters does not create a valuation for free. Lemma 341.1 says a specified interpretation preserves NIP; it does not say every abstract subring of an NIP field is definable, or that a definable topology must have a henselian valuation ring. Each of these would require a separate theorem with its own hypotheses.

Checks and outcome. The accompanying exact diagnostic implements Newton lifting of $X^2-7$ from $1$ modulo $3$ to increasing powers of $3$, and checks the valuation obstruction in $341.2$ for several primes and negative valuations. Finite modular lifts illustrate the complete-field argument; they do not turn the resulting $3$-adic root into a rational root. No finite experiment is presented as a test for NIP of an infinite field.

The proved components establish useful permanence, reconstruct a valuation from a total ring comparison, and prove henselianity under completeness. The first unresolved step is constructing a proper ring satisfying $341.1$, with enough control to obtain henselianity, for an arbitrary infinite NIP field outside the separably closed and real closed alternatives. The dp-finite theorem and the characterization of already henselian NIP fields do not remove that step. This attempt leaves the selected conjecture unresolved.

Sources

• S1 Characterizing NIP henselian fields (arXiv abstract). Accessed 2026-09-01. https://arxiv.org/abs/1911.00309. Formulation source.
• E1 Will Johnson, Dp-finite fields VI: the dp-finite Shelah conjecture, arXiv:2005.13989 $2020$. https://arxiv.org/abs/2005.13989. Primary source consulted 27 September 2026; the finite dp-rank hypothesis is essential to the result cited here.
• E2 Itay Kaplan, Thomas Scanlon, and Frank O. Wagner, Artin--Schreier extensions in NIP and simple fields, Israel Journal of Mathematics 185 $2011$, 141--153; arXiv:1009.5421. https://arxiv.org/abs/1009.5421. Primary source consulted 27 September 2026; used for Artin--Schreier closedness of infinite NIP fields in positive characteristic.