1 Introduction

Let E be a field of characteristic different from 2 and let E(X) denote the rational function field in one variable over E.

For \( E = \mathbb C(\!(t)\!)\), the field of Laurent series in one variable over the complex numbers, the quadratic form

$$\begin{aligned} Y_1^2+tY_2^2+tY_3^2+X(Y_1^2+Y_2^2+ t Y_4^2) \end{aligned}$$

in the variables \(Y_1, Y_2,Y_3,Y_4\) over E(X) has no non-trivial zero, but it has a non-trivial zero over the completion of E(X) with respect to any non-trivial valuation on E(X) that is trivial on E. This is in contrast to the situation when E is a finite field, by the Hasse-Minkowski theorem (see [6, Chapter VI, Theorem 66.1]). Note that, in both cases, the field E has a unique extension of each degree in a fixed algebraic closure.

By a \(\mathbb Z\)-valuation we mean a valuation with value group \(\mathbb Z\). A quadratic form is isotropic if it has a non-trivial zero, otherwise it is anisotropic. Without any restrictions on the base field E other than \(\mathsf {char}(E) \ne 2\), any anisotropic quadratic form over E(X) of dimension at most 3 remains anisotropic over the completion of E(X) with respect to some \(\mathbb Z\)-valuation on E(X) that is trivial on E; this follows for example from Milnor’s exact sequence [4, Theorem IX.3.1]. The case of 4-dimensional quadratic forms is the first case over E(X) where the validity of such a local-global principle for isotropy depends on the base field E.

When E is a non-dyadic local field, using a result of Lichtenbaum [5], one obtains that a 4-dimensional anisotropic quadratic form over E(X) remains anisotropic over the completion of E(X) with respect to some \(\mathbb Z\)-valuation on E(X) that is trivial on E (see [1, Remark 3.8]). This resembles the case where E is a finite field.

In contrast to the situations where E is a finite field or a local field, for \(E=\mathbb C(\!(t)\!)\), the example of the quadratic form above shows that the local-global principle for isotropy of 4-dimensional quadratic forms over E(X) fails with respect to \(\mathbb Z\)-valuations that are trivial on E. However, anisotropy of this quadratic form can be detected over the larger field \(\mathbb C(X)(\!(t)\!)\) by using Springer’s theorem (see [4, Proposition VI.1.9]).

Consider the more general situation where the field E is complete with respect to a non-dyadic \(\mathbb Z\)-valuation v. In this case, a local-global principle for isotropy was obtained in [1] using a geometric setup. Let \(\mathcal {O}_v\) denote the valuation ring of v. By a model for E(X) over \(\mathcal {O}_v\) we mean a two-dimensional integral normal projective flat \(\mathcal {O}_v\)-scheme \(\mathscr {X}\) whose function field is isomorphic to E(X). Codimension-one points on a model of E(X) over \(\mathcal {O}_v\) correspond to certain \(\mathbb Z\)-valuations on E(X). For a model \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\), let \(\Omega _\mathscr {X}\) denote the set of \(\mathbb Z\)-valuations given by codimension-one points of \(\mathscr {X}\). Consider the set \(\Omega = \bigcup _{\mathscr {X} } \Omega _\mathscr {X}\) where the union is taken over all models \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\). It follows from [1, Theorem 3.1 and Remark 3.2] that an anisotropic quadratic form over E(X) remains anisotropic over the completion of E(X) with respect to some \(\mathbb Z\)-valuation in \(\Omega \). One may ask whether this remains true if one replaces \(\Omega \) by \(\Omega _{\mathscr {X}}\) for some well-chosen model \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\).

The aim of this note is to show that this is not the case: if the residue field of v is separably closed, then, for any model \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\), there exists an anisotropic 4-dimensional quadratic form over E(X) which is isotropic over the completion of E(X) with respect to any \(w\in \Omega _{\mathscr {X}}\) (Corollary 2). Let \(\pi \in \mathcal {O}_v\) be a uniformiser of v. For any model \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\), the set \(\{w(\pi ) \mid w \in \Omega _{\mathscr {X}} \}\) is finite and hence it has an upper bound. However, for any positive integer r, the quadratic form

$$\begin{aligned} \varphi _r = (X^r-\pi )Y_1^2+ (X^{r+1}+\pi )Y_2^2 + \pi X Y_3^2+ X(X^r+\pi )Y_4^2 \end{aligned}$$

is anisotropic over E(X), but it is isotropic over the completion of E(X) with respect to any \(\mathbb Z\)-valuation w on E(X) with \(w(\pi ) <r\) (Theorem). The construction of \(\varphi _r\) is inspired by the example in [1, Remark 3.6] of an anisotropic 6-dimensional quadratic form over \(\mathbb Q_p(X)\) where p is an odd prime.

2 Results

We assume some familiarity with basic quadratic form theory over fields, for which we refer to [4]. We first fix some notation and recall some results.

By a quadratic form or simply a form we mean a regular quadratic form. Let E always be a field of characteristic different from 2 and let \({E}^{\times }\) denote its multiplicative group. For \(a_1, \ldots , a_n \in {E}^{\times }\), the diagonal form \(a_1X_1^2+\dots +a_nX_n^2\) is denoted by \(\langle a_1, \ldots ,a_n\rangle \).

Let v be a \(\mathbb Z\)-valuation on E. We denote the corresponding valuation ring, its maximal ideal, and its residue field respectively by \(\mathcal {O}_v, \mathfrak {m}_v\), and \(\kappa _v\). For an element \(a\in \mathcal {O}_v\), let \(\overline{a}\) denote the residue class \(a+\mathfrak {m}_v\) in \(\kappa _v\). The completion of E with respect to v is denoted by \(E_v\). We say that v is henselian if it extends uniquely to every finite field extension of E. Complete discretely valued fields are henselian (see [2, Theorem 1.3.1 and Theorem 4.1.3]). We recall a consequence of Hensel’s lemma:

Lemma

Let v be a henselian \(\mathbb Z\)-valuation on E such that \(v(2) =0\). Then:

  1. (a)

    For \(u_1,u_2\in {\mathcal {O}}^{\times }_v\), the quadratic form \(\langle u_1, u_2 \rangle \) over E is isotropic if and only if \(\overline{u_1u_2} \in -{\kappa }^{\times 2}_v\).

  2. (b)

    If \(\kappa _v\) is separably closed, then every 3-dimensional form over E is isotropic.

Proof

(a) For \(u_1,u_2\in {\mathcal {O}}^{\times }_v\), since v is henselian and \(v(2)=0\), it follows by [2, Theorem 4.1.3(4)] that \(u_1u_2 \in -{E}^{\times 2}\) if and only if \(\overline{u_1u_2} \in -{\kappa }^{\times 2}_v\).

(b) Every 3-dimensional form over E contains a 2-dimensional form isometric to \(\lambda \langle 1 ,u\rangle \) for some \(u \in {\mathcal {O}}^{\times }_v\) and \(\lambda \in {E}^{\times }\). If \(\kappa _v\) is separably closed, then \(\overline{u} \in -{\kappa }^{\times 2}_v\) and hence \(\langle 1, u\rangle \) is isotropic by (a). \(\square \)

The set of all \(\mathbb Z\)-valuations on E(X) is denoted by \(\Omega _{E(X)}\). For \(r\in \mathbb {N}\), we define

$$\begin{aligned} \Omega _r = \{ w\in \Omega _{E(X)} \mid w({E}^{\times }) =i\mathbb Z \text{ for } \text{ some } 0\le i \le r \}. \end{aligned}$$

With this notation, \(\Omega _0\) is the set of all E-trivial \(\mathbb Z\)-valuations on E(X). We recall that any monic irreducible polynomial \(p \in E[X]\) determines a unique \(\mathbb Z\)-valuation \(v_p\) on E(X) which is trivial on E and such that \(v_p(p)=1\). There is further a unique \(\mathbb Z\)-valuation \(v_\infty \) on E(X) such that \(v_\infty (f)=-\deg (f)\) for any \(f\in E[X]\setminus \{0\}\). Moreover, every \(\mathbb Z\)-valuation w on E(X) trivial on E is either equal to \(v_\infty \) or to \(v_p\) for some monic irreducible polynomial \(p \in E[X]\) (see [2, Theorem 2.1.4]), and in either of the two cases the residue field is a finite field extension of E.

Theorem

Let v be a henselian \(\mathbb Z\)-valuation on E such that \(v(2) =0\). Assume that \(\kappa _v\) is separably closed. Let \(\pi \in {E}^{\times }\) be such that \(v(\pi ) =1\) and let \(r \in \mathbb {N}\). Then the quadratic form

$$\begin{aligned} \varphi _r = \langle X^r-\pi , X^{r+1}+\pi , \pi X, X(X^r+\pi ) \rangle \end{aligned}$$

is isotropic over \(E(X)_w\) for every \(\mathbb Z\)-valuation \(w \in \Omega _{r-1}\), but anisotropic over \(E(X)_w\) for some \(w\in \Omega _r\).

Proof

Set \(F=E(X)\). We first show that \(\varphi _r\) is isotropic over \(F_w\) for all \(w\in \Omega _{r-1}\). Consider \(w\in \Omega _{r-1}\).

Case 1: \(w(\pi ) =0 = w(X)\). Then \(\kappa _w\) is a finite extension of E. Since v is henselian, there is a unique extension \(v'\) of v to \(\kappa _w\), and \(v'\) is again henselian. Furthermore, it follows by [2, Theorem 3.3.4] that \(v'({\kappa }^{\times }_w)\) is isomorphic to \(\mathbb Z\) and \(\kappa _{v'}\) is separably closed. It follows by part (b) of the Lemma that every 3-dimensional quadratic form over \(\kappa _w\) is isotropic. We have that \(w =v_p\) for some monic irreducible polynomial \(p \in E[X]\) such that \(p \ne X\). Note that, in this case, at least three diagonal coefficients of \(\varphi _r\) are units in \(\mathcal {O}_w\). It follows by Springer’s theorem [4, Proposition VI.1.9] that \(\varphi _r\) is isotropic over \(F_w\).

Case 2: \(0 \le w (\pi )< r\) and \(1\le w(X)\). Let \( u = (X^r\pi ^{-1}-1)(X^{(r+1)}\pi ^{-1}+1)\). Then \(w(u) =0\) and \(\overline{u} =-1 \in - {\kappa }^{\times 2}_w\). It follows by part (a) of the Lemma that the form \(\pi ^{-1} \langle X^r-\pi , X^{r+1}+\pi \rangle \) is isotropic over \(F_w\). Thus \(\varphi _r\) is isotropic over \({F}_w\).

Case 3: \(w(X)< 0 \le w (\pi )< r\). Note that, if \(w(\pi ) =0\), then \(w =v_\infty \) and \(\kappa _w =E\), and otherwise \(w|_E\) is equivalent to v and \(\kappa _v \subseteq \kappa _w\); since \(-1\in {\kappa }^{\times 2}_v\), we get in either case that \(-1 \in {\kappa }^{\times 2}_w\). Consider \( u = (1+ \pi X^{-{(r+1)}})(1+\pi X^{-r})\). We have that \(w(u) =0\) and \(\overline{ u } = 1 \in {\kappa }^{\times 2}_w = -{\kappa }^{\times 2}_w\). It follows by part (a) of the Lemma that the form \(X^{-(r+1)} \langle X^{r+1}+\pi , X(X^{r}+\pi )\rangle \) is isotropic over \(F_w\). Thus \(\varphi _r\) is isotropic over \(F_w\).

We have thus shown that \(\varphi _r\) is isotropic over \(F_w\) for every \(w \in \Omega _{r-1}\). Now we show that \(\varphi _r\) is anisotropic over \(F_w\) for some \(w\in \Omega _F\).

Let \(E' = E(s)\), where \(s =\root r \of {\pi }\). Then v extends uniquely to a valuation on \(E'\) which we again denote by v. Note that \(s^r = \pi \) in \(E'\) and hence \(v (\pi ) = rv(s)\). Then \(v' = rv\) is a \(\mathbb Z\)-valuation on \(E'\).

Let \(L= E'(X)\) and let \(Y =\frac{X}{s}\). Note that \(L =E'(Y)\). By [2, Corollary 2.2.2], there exists a unique extension of \(v'\) to L such that \(v(Y) =0\) and \(\overline{Y}\) is transcendental of \(\kappa _{v'}\); we further have that \(\kappa _w = \kappa _{v'}(\overline{Y})\) and \(w({L}^{\times }) = v'({E'}^{\times }) = \mathbb Z\). Since \(w(Y) =0\), we have that \(w(X) =w(s) =1\). We get that

$$\begin{aligned} \varphi _r = \langle s^r(Y^r-1), s^r(sY+1), s^{r+1}Y, s^{r+1}Y(Y^r+1) \rangle . \end{aligned}$$

Consider the forms \(\varphi _1 = \langle Y^r-1,~ sY+1\rangle \) and \(\varphi _2 = \langle Y,~ Y(Y^r+1) \rangle \).

Since \( \overline{Y}^r-1, \overline{Y}^{r} +1 \notin - {\kappa }^{\times 2}_w\), it follows by Springer’s theorem [4, Proposition VI.1.9] that the quadratic form \(s^{-r}\varphi _r\) is anisotropic over \(L_w\). Hence \(\varphi _r\) is anisotropic over \(L_w\). We obtain that \(\varphi _r\) is anisotropic over \(F_{w|_F}\). Note that \(w(\pi ) = w(s^r) =rw(s) =r\), thus \(w\in \Omega _r\). \(\square \)

We now provide a different perspective to the above theorem. For a subset \(\Omega \subseteq \Omega _{E(X)}\), we say that \(\Omega \) has the finite support property if for every \(f \in {E(X)}^{\times }\), the set \(\{w\in \Omega \mid w(f) \ne 0\}\) is finite. It is well-known that \(\Omega _{0}\) has the finite support property. When E carries a discrete valuation, the set \(\Omega _{E(X)}\) does not have the finite support property. However, for any model \(\mathscr {X}\) of E(X) over \(\mathcal {O}_v\), the set \(\Omega _\mathscr {X}\) contains \(\Omega _0\) and has the finite support property. We show the following:

Corollary 1

Let v be a henselian \(\mathbb Z\)-valuation on E with \(v(2) =0\). Assume that \(\kappa _v\) is separably closed. Let \(\Omega \subseteq \Omega _{E(X)}\) be a subset with the finite support property. Then there exists an anisotropic 4-dimensional quadratic form over E(X) which is isotropic over \(E(X)_w\) for every \(w\in \Omega \).

Proof

Let \(\pi \in {E}^{\times }\) be such that \(v(\pi ) =1\). Since \(\Omega \) has the finite support property, the set \(\{w\in \Omega \mid w(\pi ) \ne 0 \}\) is finite. Set \(r= 1+\max \{w(\pi ) \mid w\in \Omega \}\). Clearly \(\Omega \subseteq \Omega _{r-1} \). Then the form \(\varphi _r\) in the Theorem is isotropic over \(E(X)_w\) for every \(w\in \Omega \), but anisotropic over E(X). \(\square \)

Corollary 2

Let v be a henselian \(\mathbb Z\)-valuation on E with \(v(2) =0\). Assume that \(\kappa _v\) is separably closed. Let \(\mathscr {X}\) be a regular model of E(X) over \(\mathcal {O}_v\). Then there exists an anisotropic 4-dimensional quadratic form over E(X) which is isotropic over \(E(X)_w\) for every \(w\in \Omega _{\mathscr {X}}\).

Proof

By [3, Chapter  II, Lemma 6.1], for every element \(f\in {E(X)}^{\times }\), the set \(\{w \in \Omega _{\mathscr {X}} \mid w(f) \ne 0 \}\) is finite, hence the statement follows by Corollary 1.

\(\square \)