This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Arithmetic and birational properties of linear spaces on intersections of two quadrics

Lena Ji Department of Mathematics, University of Michigan, 530 Church Street, Ann Arbor, MI 48109-1043 [email protected] https://www-personal.umich.edu/~lenaji  and  Fumiaki Suzuki Institute of Algebraic Geometry, Leibniz University Hannover, Welfengarten 1, 30167, Hannover, Germany [email protected] https://sites.google.com/view/fumiaki-suzuki/home
Abstract.

We study rationality questions for Fano schemes of linear spaces on smooth complete intersections of two quadrics, especially over non-closed fields. Our approach is to study hyperbolic reductions of the pencil of quadrics associated to XX. We prove that the Fano schemes Fr(X)F_{r}(X) of rr-planes are birational to symmetric powers of hyperbolic reductions, generalizing results of Reid and Colliot-Thélène–Sansuc–Swinnerton-Dyer, and we give several applications to rationality properties of Fr(X)F_{r}(X).

For instance, we show that if XX contains an (r+1)(r+1)-plane over a field kk, then Fr(X)F_{r}(X) is rational over kk. When XX has odd dimension, we show a partial converse for rationality of the Fano schemes of second maximal linear spaces, generalizing results of Hassett–Tschinkel and Benoist–Wittenberg. When XX has even dimension, the analogous result does not hold, and we further investigate this situation over the real numbers. In particular, we prove a rationality criterion for the Fano schemes of second maximal linear spaces on these even-dimensional complete intersections over \mathbb{R}; this may be viewed as extending work of Hassett–Kollár–Tschinkel.

2020 Mathematics Subject Classification:
Primary: 14E08, Secondary: 14G20, 14C25, 14D10
L.J. is partially supported by NSF MSPRF grant DMS-2202444. F.S. is supported by the ERC grant “RationAlgic” (grant no. 948066) and partially by the DFG project EXC-2047/1 (project no. 390685813).

1. Introduction

Over an arbitrary field kk of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. There is an extensive literature on the Fano schemes Fr(X)F_{r}(X) of rr-dimensional linear spaces on such complete intersections [Gau55, Rei72, Tyu75, Don80, Wan18], and there have been many applications to arithmetic problems [CTSSD87, BGW17, IP22, CT24], moduli theory [DR77, Ram81], and rationality questions [ABB14, HT21b, BW23]. In this paper, we study birational properties of the Fano schemes of linear spaces on XX via hyperbolic reductions of the pencil of quadrics 𝒬1\mathcal{Q}\rightarrow\mathbb{P}^{1} associated to XX, with an eye toward applications to rationality questions over non-closed fields.

A variety over a field kk is kk-rational if it is birationally equivalent to projective space over kk. It is classically known that if a smooth complete intersection of two quadrics XNX\subset\mathbb{P}^{N} contains a line defined over kk, then XX is kk-rational by projection from this line (see e.g. [CTSSD87, Proposition 2.2]). Our first result is a generalization of this to Fano schemes of higher-dimensional linear spaces on XX:

Theorem 1.1.

Over a field kk of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. Let 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2. If Fr+1(X)(k)F_{r+1}(X)(k)\neq\emptyset, then Fr(X)F_{r}(X) is kk-rational.

An immediate corollary of Theorem 1.1 over algebraically closed fields is that the Fano schemes of non-maximal linear spaces on XX are all rational (Corollary 3.5(1)); that is, Fr(X)F_{r}(X) is rational for all 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2. To the authors’ knowledge, this result was previously only known for r=0r=0 and r=N22r=\lfloor\frac{N}{2}\rfloor-2 (the latter case by combining works of [DR77, New75, New80, Bau91, Cas15]). For 0<r<N220<r<\lfloor\frac{N}{2}\rfloor-2, only unirationality of Fr(X)F_{r}(X) for general XX was previously known by Debarre–Manivel [DM98].

One may wonder whether the converse of Theorem 1.1 holds. In general, the answer to this question is no: counterexamples for (r,N)=(0,4),(0,6)(r,N)=(0,4),(0,6) are known over \mathbb{R} by [HT21b, Remark 37] and [HKT22, Propositions 6.1 and 6.2]. We additionally show that there are counterexamples for (r,N)=(g2,2g)(r,N)=(g-2,2g) for any g2g\geq 2 (see Corollary 5.7(2)).

However, for (r,N)=(0,5)(r,N)=(0,5), the above question has a positive answer over any field. In this case, Hassett–Tschinkel (over \mathbb{R}) [HT21b] and Benoist–Wittenberg (over arbitrary kk) [BW23] show that F0(X)=X5F_{0}(X)=X\subset\mathbb{P}^{5} is kk-rational if and only if F1(X)(k)F_{1}(X)(k)\neq\emptyset. In this case, lines are the maximal linear subspaces on the threefold XX. More generally, when N=2g+1N=2g+1 is odd, the Fano schemes of linear subspaces on XX encode a lot of interesting arithmetic and geometric data. Over k¯{\overline{k}}, Weil first observed that the Fano scheme Fg1(X)F_{g-1}(X) of maximal linear subspaces is isomorphic to the Jacobian of the genus gg hyperelliptic curve obtained as the Stein factorizaton of Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1} [Gau55] (see also [Rei72, DR77, Don80]). Over kk, Wang studied the torsor structure of Fg1(X)F_{g-1}(X) [Wan18]. For the second maximal linear subspaces, Fg2(X)F_{g-2}(X) is isomorphic over k¯{\overline{k}} to the moduli space of rank 2 vector bundles on the hyperelliptic curve [DR77].

The main result of this paper proves a partial converse to Theorem 1.1 in the case when N=2g+1N=2g+1 and r=g2r=g-2, generalizing Hassett–Tschinkel and Benoist–Wittenberg’s results to arbitrary g2g\geq 2. To state this result, we first need to introduce a definition. If Fr(X)(k)F_{r}(X)(k)\neq\emptyset, choose Fr(X)(k)\ell\in F_{r}(X)(k) and define 𝒬(r)1\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} to be the hyperbolic reduction of 𝒬1\mathcal{Q}\rightarrow\mathbb{P}^{1} with respect to \ell (see Section 2.2). The hyperbolic reduction 𝒬(r)1\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} is itself a quadric fibration and may be regarded as the relative Fano scheme of isotropic (r+1)(r+1)-planes of 𝒬1\mathcal{Q}\rightarrow\mathbb{P}^{1} containing \ell; moreover, the kk-birational equivalence class of 𝒬(r)\mathcal{Q}^{(r)} does not depend on \ell (see Section 3.1 for these and additional properties). We prove:

Theorem 1.2.

Over a field kk of characteristic 2\neq 2, fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g+1\mathbb{P}^{2g+1}. Then Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset and 𝒬(g2)\mathcal{Q}^{(g-2)} is kk-rational if and only if Fg1(X)(k)F_{g-1}(X)(k)\neq\emptyset.

In every even dimension, the analogous statement fails (see Section 1.1). The reason Theorem 1.2 gives a partial converse of Theorem 1.1 is the following birational description of the Fano schemes, which relates Fr(X)F_{r}(X) to the hyperbolic reduction 𝒬(𝓇)\cal Q^{(r)}:

Theorem 1.3.

Over a field kk of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. Let 0rN210\leq r\leq\lfloor\frac{N}{2}\rfloor-1. If Fr(X)(k)F_{r}(X)(k)\neq\emptyset, then one of the following conditions holds.

  1. (1)

    Fr(X)F_{r}(X) is kk-birational to Symr+1𝒬(r)\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)}.

  2. (2)

    N=2gN=2g and r=g1r=g-1 for some g1g\geq 1. In this case, the subscheme of Fg1(X)F_{g-1}(X) parametrizing (g1)(g-1)-planes on XX disjoint from \ell is kk-isomorphic to the subscheme of Symg𝒬(g1)\operatorname{Sym}^{g}\mathcal{Q}^{(g-1)} parametrizing gg-tuples of distinct points of 𝒬(g1)\mathcal{Q}^{(g-1)}, and they are 0-dimensional schemes of length (2g+1g)\binom{2g+1}{g}.

Two special cases of Theorem 1.3 were previously known. When N=2g+15N=2g+1\geq 5 is odd, r=g1r=g-1, and the field is algebraically closed, Reid proved the birational equivalence of Fg1(X)F_{g-1}(X) and SymgC\operatorname{Sym}^{g}C, where CC is the genus gg hyperelliptic curve obtained as the Stein factorization of Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1} [Rei72, Section 4]. The other previously known case is when N>2N>2 and r=0r=0: Colliot-Thélène–Sansuc–Swinnerton-Dyer proved that XX is kk-birational to 𝒬(0)\mathcal{Q}^{(0)} [CTSSD87, Theorem 3.2].

Theorem 1.3 shows that Theorem 1.2 is a partial converse to Theorem 1.1 because symmetric powers of kk-rational varieties are also kk-rational [Mat69]. However, it does not give the full converse because, in general, it is possible for a symmetric power of an irrational variety to be rational (see Remark 4.14). For Theorem 1.1, we show that the statement follows from Theorem 1.3 by proving that a kk-point on Fr+1(X)F_{r+1}(X) gives a section of the quadric fibration 𝒬(𝓇)1\cal Q^{(r)}\to\mathbb{P}^{1} and hence a kk-rationality construction.

As another application of Theorem 1.3, we prove the following (separable) kk-unirationality criterion:

Theorem 1.4.

Over a field kk of characteristic 2\neq 2, fix N6N\geq 6, and let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. The following are equivalent:

  1. (1)

    F1(X)F_{1}(X) is separably kk-unirational;

  2. (2)

    F1(X)F_{1}(X) is kk-unirational;

  3. (3)

    F1(X)(k)F_{1}(X)(k)\neq\emptyset.

In addition, if k=k=\mathbb{R} is the real numbers, then the above result holds for all Fano schemes of non-maximal linear subspaces. That is, for every 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2, Fr(X)F_{r}(X) is \mathbb{R}-unirational if and only if it has an \mathbb{R}-point.

The bound on NN in Theorem 1.4 is crucial because F1(X)F_{1}(X) is never k¯{\overline{k}}-unirational for N5N\leq 5. Our result extends previous results for F0(X)=XF_{0}(X)=X due to Manin [Man86, Theorems 29.4 and 30.1], Knecht [Kne15, Theorem 2.1], Colliot-Thélène–Sansuc–Swinnerton-Dyer [CTSSD87, Remark 3.28.3], and Benoist–Wittenberg [BW20, Theorem 4.8]. Using Theorem 1.3, the proof of Theorem 1.4 is reduced to showing separable kk-unirationality of the hyperbolic reduction 𝒬(1)\mathcal{Q}^{(1)} (and, if k=k=\mathbb{R}, for 𝒬(𝓇)\cal Q^{(r)}; here we use a result of Kollár over local fields [Kol99]). Part of the difficulty in generalizing the result for higher rr from \mathbb{R} to other fields lies in the discrepancy between kk-points and 0-cycles of degree 11.

In Section 3.3, we also apply Theorems 1.1 and 1.3 to establish kk-rationality results for Fr(X)F_{r}(X) for certain fields kk, extending earlier results that were previously known only for XX. More precisely, we prove results for CiC_{i}-fields, pp-adic fields, totally imaginary number fields, and finite fields, generalizing results of Colliot-Thélène–Sansuc–Swinnerton-Dyer [CTSSD87, Theorem 3.4]. Over algebraically closed fields, using work of Ramanan [Ram81] relating Fano schemes of odd-dimensional XX and moduli spaces of certain vector bundles on hyperelliptic curves, we also prove rationality results for these moduli spaces (Corollary 3.7), partially extending the work of Newstead [New75, New80] and King–Schofield [KS99].

1.1. Second maximal linear spaces on even-dimensional complete intersections over \mathbb{R}

In the latter part of the paper, we focus on rationality over the field \mathbb{R} of real numbers. For a smooth complete intersection of two quadrics defined over \mathbb{R}, Theorem 1.1 implies that its Fano schemes of non-maximal linear spaces are \mathbb{C}-rational. One may further ask when these Fano schemes are rational over \mathbb{R}.

The locus of real points encodes additional obstructions to rationality over \mathbb{R}: if YY is an \mathbb{R}-rational smooth projective variety, then Y()Y(\mathbb{R}) is necessarily connected and nonempty. In dimensions 11 and 22, this topological obstruction characterizes rationality for \mathbb{C}-rational varieties [Com13]. In higher dimensions, however, this fails in general: in dimension 3\geq 3, there are \mathbb{C}-rational varieties, with nonempty connected real loci, that are irrational over \mathbb{R} [BW20, Theorem 5.7]. Among complete intersections of quadrics X5X\subset\mathbb{P}^{5}, Hassett–Tschinkel showed that there exist examples that are irrational over \mathbb{R} despite X()X(\mathbb{R}) being non-empty and connected [HT21b]. (See also [FJS+22, JJ24] for other examples in dimension 33.)

Hassett–Kollár–Tschinkel studied \mathbb{R}-rationality for even-dimensional complete intersections of quadrics. In particular, they showed that a 44-fold X6X\subset\mathbb{P}^{6} is \mathbb{R}-rational if and only if its real locus is non-empty and connected. We prove an analogous result for the second maximal linear spaces on X2gX\subset\mathbb{P}^{2g}:

Theorem 1.5 (Theorem 5.1).

Over the real numbers, fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g\mathbb{P}^{2g}. Then Fg2(X)F_{g-2}(X) is \mathbb{R}-rational if and only if Fg2(X)()F_{g-2}(X)(\mathbb{R}) is non-empty and connected. Furthermore, this is equivalent to \mathbb{R}-rationality of the hyperbolic reduction 𝒬(g2)\mathcal{Q}^{(g-2)}.

Thus, in this case, the aforementioned necessary condition for \mathbb{R}-rationality is in fact sufficient. Moreover, we show that this property is determined by the real isotopy class of XX (Section 5.3). One can apply Theorem 1.5 to construct examples in any even dimension where 𝒬(g2)\mathcal{Q}^{(g-2)} is \mathbb{R}-rational but Fg1(X)()=F_{g-1}(X)(\mathbb{R})=\emptyset (see Corollary 5.7(2)), contrasting the odd-dimensional case as shown by Theorem 1.2. Furthermore, the analogue of Theorem 1.5 fails in every odd dimension (see Example 5.10).

As an application, in the case of 44-dimensional X6X\subset\mathbb{P}^{6}, combining Theorems 1.4 and 1.5 with earlier results in [CTSSD87, HKT22], we completely determine \mathbb{R}-rationality and \mathbb{R}-unirationality of the Fano schemes of non-maximal linear spaces on a 44-fold complete intersection XX of two real quadrics using an isotopy invariant of Krasnov [Kra18] (see Section 5.2). This invariant, which was used in the rationality classifications in [HT21b, HKT22], is defined using the signatures of the quadrics in the associated pencil 𝒬1\cal Q\to\mathbb{P}^{1}. In particular, we show that these \mathbb{R}-(uni)rationality properties of Fr(X)F_{r}(X) for r=0,1r=0,1 are controlled solely by the real isotopy class of XX, extending the results of [HKT22] on \mathbb{R}-rationality of XX:

Corollary 1.6.

Over the real numbers, let XX be a smooth complete intersection of two quadrics in 6\mathbb{P}^{6}.

  1. (1)

    F1(X)F_{1}(X) is \mathbb{R}-rational if and only if the Krasnov invariant is one of (1)(1), (3)(3), (1,1,1)(1,1,1), (2,2,1)(2,2,1), (1,1,1,1,1)(1,1,1,1,1), (2,1,2,1,1)(2,1,2,1,1), or (1,1,1,1,1,1,1)(1,1,1,1,1,1,1).

  2. (2)

    F1(X)F_{1}(X) is \mathbb{R}-unirational if and only if the Krasnov invariant is one of those listed in 1, or is (3,1,1)(3,1,1), (3,2,2)(3,2,2), (3,1,1,1,1)(3,1,1,1,1), or (2,2,1,1,1)(2,2,1,1,1).

  3. (3)

    [HKT22, Theorem 1.1] XX is \mathbb{R}-rational if and only if the Krasnov invariant is one of those listed in 2, or is (5)(5), (4,2,1)(4,2,1), or (3,3,1)(3,3,1).

  4. (4)

    XX is \mathbb{R}-unirational if and only if the Krasnov invariant is one of those listed in 3 or is (5,1,1)(5,1,1).

1.2. Outline

In Section 2, we recall preliminary results on Fano schemes of linear subspaces on complete intersections of quadrics, the definition and properties of hyperbolic reductions, lemmas on pencils of quadrics from Reid’s thesis [Rei72], and Benoist–Wittenberg’s codimension 2 Chow scheme for threefolds. In Section 3, we describe the hyperbolic reduction of a pencil of quadrics. We prove Theorems 1.1, 1.3, and 1.4 in this section, and as consequences, we derive rationality results over certain fields in Section 3.3. Section 4, which is the most technical part of the paper, is devoted to the proof of Theorem 1.2 on odd-dimensional complete intersections. Finally, in Section 5 we turn to the case of even-dimensional complete intersections over \mathbb{R}. Here we prove Theorem 1.5 and Corollary 1.6, and we give examples contrasting the behavior in the even- and odd-dimensional cases.

Notation

Throughout kk is a field of characteristic 2\neq 2. For a variety XX, we use xXx\in X to denote a scheme-theoretic point. For i0i\geq 0, we let CHi(X)\operatorname{CH}^{i}(X) denote the Chow group of codimension ii cycles on XX. We denote the subgroup of algebraically trivial cycles by CHi(X)algCHi(X)\operatorname{CH}^{i}(X)_{\operatorname{alg}}\subset\operatorname{CH}^{i}(X), and we denote the quotient by NSi(X)=CHi(X)/CHi(X)alg\operatorname{NS}^{i}(X)=\operatorname{CH}^{i}(X)/\operatorname{CH}^{i}(X)_{\operatorname{alg}}. For non-negative integers rr, SymrX\operatorname{Sym}^{r}X denotes the rrth symmetric power. We say a variety XX over kk is kk-rational to emphasize that the rationality construction is defined over kk; when we say rational without specifying the ground field, we usually mean k¯{\overline{k}}-rational over an algebraically closed field. For a curve CC over kk, we use Pic(C)\operatorname{Pic}(C) to denote the Picard group over kk, and 𝐏𝐢𝐜C/k\operatorname{\mathbf{Pic}}_{C/k} to denote the relative Picard scheme over kk. If 1,,rn\ell_{1},\ldots,\ell_{r}\subset\mathbb{P}^{n} are linear subspaces, we denote their span by 1,,r\langle\ell_{1},\ldots,\ell_{r}\rangle.

Acknowledgements

We thank Brendan Hassett and Jerry Wang for interesting discussions. We thank Olivier Debarre for feedback on an earlier version of the paper, and Jean-Louis Colliot-Thélène for comments, in particular for pointing out Remark 3.8 and for noting that Theorem 4.13 does not require the assumption that Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset. This work started during a visit of the first author to the University of California, Los Angeles, and she thanks Joaquín Moraga and Burt Totaro for their hospitality and for providing a welcoming environment.

2. Preliminaries

2.1. Fano schemes of rr-planes on a smooth complete intersection of two quadrics

Let XNX\subset\mathbb{P}^{N} be a smooth complete intersection of two quadrics. For non-negative integers r0r\geq 0, Fr(X)F_{r}(X) denotes the Fano scheme of rr-planes on XX. In this section, we recall some preliminary results about these Fano schemes Fr(X)F_{r}(X).

Lemma 2.1 ([CT24, Lemmas A.3 and A.4]).

Over a field kk of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. The following hold.

  1. (1)

    If r>N21r>\lfloor\frac{N}{2}\rfloor-1, Fr(X)F_{r}(X) is empty.

  2. (2)

    If 0rN210\leq r\leq\lfloor\frac{N}{2}\rfloor-1, Fr(X)F_{r}(X) is non-empty, smooth, projective, and of dimension (r+1)(N2r2)(r+1)(N-2r-2).

  3. (3)

    If 0r<N210\leq r<\frac{N}{2}-1 (or equivalently if dimFr(X)>0\dim F_{r}(X)>0), then Fr(X)F_{r}(X) is geometrically connected.

For the sake of completeness, we add the following results, which will not be used in the rest of the paper.

Theorem 2.2.

In the setting of Lemma 2.1, the following hold.

  1. (1)

    If N=2gN=2g and r=g1r=g-1 for some g1g\geq 1, Fg1(X)F_{g-1}(X) is a torsor under the finite group scheme 𝐏𝐢𝐜C/k0[2]\operatorname{\mathbf{Pic}}^{0}_{C/k}[2], where CC is a certain curve of genus gg associated to XX. The curve CC depends on the choice of two quadrics defining XX, but 𝐏𝐢𝐜C/k0[2]\operatorname{\mathbf{Pic}}^{0}_{C/k}[2] does not depend on this choice.

  2. (2)

    If N=2g+1N=2g+1 and r=g1r=g-1 for some g1g\geq 1, Fg1(X)F_{g-1}(X) is a torsor under 𝐏𝐢𝐜C/k0\operatorname{\mathbf{Pic}}_{C/k}^{0}, where CC is the curve of genus gg obtained as the Stein factorization of Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1}.

  3. (3)

    If 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2, Fr(X)F_{r}(X) is a Fano variety, i.e., the anti-canonical divisor KFr(X)-K_{F_{r}(X)} is ample.

Proof.

(1), (2), and (3) respectively follow from [BG14, Section 6], [Wan18, Theorem 1.1], and [DM98, Remarque 3.2]. ∎

2.2. Quadric bundles and hyperbolic reductions

In this section, we recall the definition and basic properties of quadric fibrations and of the hyperbolic reductions of a quadric fibrations; see, e.g., [Kuz21, Section 2], [ABB14, Section 1.3] for more details.

Let SS be an integral separated Noetherian scheme over a field kk of characteristic 2\neq 2. A quadric fibration over SS is a morphism 𝒬𝒮\cal Q\to S that can be written as a composition 𝒬𝒮()Proj¯𝒮Sym()𝒮\cal Q\hookrightarrow\mathbb{P}_{S}(\cal E)\coloneqq\underline{\operatorname{Proj}}_{S}\operatorname{Sym}^{\bullet}(\cal E^{\vee})\to S where \cal E is a vector bundle and 𝒬𝒮()\cal Q\hookrightarrow\mathbb{P}_{S}(\cal E) is a divisor of relative degree 2 over SS. A quadric fibration is determined by a quadratic form q:Sym2q\colon\operatorname{Sym}^{2}\cal E\to\cal L^{\vee} with values in a line bundle \cal L.

Lemma 2.3 ([ABB14, Proposition 1.2.5]).

Let SS be a smooth scheme over a field of characteristic 2\neq 2, and let π:𝒬𝒮\pi\colon\cal Q\to S be a flat quadric fibration with smooth generic fiber. Then the degeneration divisor of π\pi is smooth over kk if and only if 𝒬\cal Q is smooth over kk and π\pi has simple degeneration (i.e. the fibers of π\pi have corank at most 1).

A subbundle \cal F\subset\cal E is isotropic if q|=0q|_{\cal F}=0 (or, equivalently, if S()𝒬\mathbb{P}_{S}(\cal F)\subset\cal Q). An isotropic subbundle \cal F\subset\cal E is regular if, for every (closed) point sSs\in S, the fiber S()𝓈\mathbb{P}_{S}(\cal F)_{s} is contained in the smooth locus of the fiber 𝒬𝓈\cal Q_{s}. If \cal F is regular isotropic, then \cal F is contained in the subbundle

Ker()\cal F^{\perp}\coloneqq\text{Ker}(\cal E\twoheadrightarrow\cal F^{\vee}\otimes\cal L^{\vee})

of \cal E. Moreover, \cal F is in the kernel of the restriction q|q|_{\cal F^{\perp}}, so we have an induced quadratic form on /\cal F^{\perp}/\cal F.

Definition 2.4.

The induced quadratic form q¯:Sym2(/)\overline{q}\colon\operatorname{Sym}^{2}(\cal F^{\perp}/\cal F)\to\cal L^{\vee} is the hyperbolic reduction of q:Sym2q\colon\operatorname{Sym}^{2}\cal E\to\cal L^{\vee} with respect to the regular isotropic subbundle \cal F. We also say that 𝒬¯(q¯=0)\overline{\cal Q}\coloneqq(\overline{q}=0) is the hyperbolic reduction of 𝒬=(𝓆=0)\cal Q=(q=0) with respect to S()\mathbb{P}_{S}(\cal F).

The process of hyperbolic reduction along a regular isotropic subbundle preserves the degeneration divisor of a quadric fibration [ABB14, Corollary 1.3.9]. (More generally, it preserves the locus of corank i\geq i fibers for each ii [KS18, Lemma 2.4].)

Hyperbolic reduction can be described geometrically in terms of the linear projection of 𝒬𝒮()\cal Q\subset\mathbb{P}_{S}(\cal E) from the linear subbundle S()𝒬𝒮()\mathbb{P}_{S}(\cal F)\subset\cal Q\subset\mathbb{P}_{S}(\cal E) [KS18, Proposition 2.5].

2.3. Lemmas of Reid on pencils of quadrics

Next, over algebraically closed fields, we recall several results proven by Reid [Rei72] that we will use in the proof of Theorem 1.3.

Lemma 2.5 ([Rei72, Lemma 2.2]).

Over an algebraically closed field of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}, and let \ell be an rr-plane on XX. Then there exist coordinate points p0,,pNNp_{0},\cdots,p_{N}\in\mathbb{P}^{N} such that p0,,pr=\langle p_{0},\dots,p_{r}\rangle=\ell and XX is defined by two quadrics which correspond in these coordinates to symmetric matrices of the form

(0I0I0000I),(0MM0),\begin{pmatrix}0&I&0\\ I&0&0\\ 0&0&I\end{pmatrix},\quad\begin{pmatrix}0&M&*\\ M&0&*\\ *&*&*\end{pmatrix},

where MM is a diagonal (r+1)×(r+1)(r+1)\times(r+1) matrix with distinct diagonal entries.

Lemma 2.6 ([Rei72, Lemma 3.4]).

Over an algebraically closed field of characteristic 2\neq 2, let ,m\ell,m be disjoint rr-planes in 2r+1\mathbb{P}^{2r+1} and choose coordinate points p0,,pr,q0,,qr2r+1p_{0},\dots,p_{r},q_{0},\dots,q_{r}\in\mathbb{P}^{2r+1} such that p0,,pr=\langle p_{0},\dots,p_{r}\rangle=\ell and q0,,qr=m\langle q_{0},\dots,q_{r}\rangle=m. Furthermore, let Q1,Q2Q_{1},Q_{2} be quadrics in 2r+1\mathbb{P}^{2r+1} which correspond in these coordinates to symmetric matrices of the form

(0II0),(0MM0),\begin{pmatrix}0&I\\ I&0\end{pmatrix},\quad\begin{pmatrix}0&M\\ M&0\end{pmatrix},

where MM is a diagonal (r+1)×(r+1)(r+1)\times(r+1) matrix with distinct diagonal entries. Then the set of rr-planes on the singular complete intersection of quadrics YQ1Q2Y\coloneqq Q_{1}\cap Q_{2} coincides with

{pi,qjiI,jII{0,,r}}.\left\{\langle p_{i},q_{j}\rangle_{i\in I,j\not\in I}\mid I\subset\left\{0,\dots,r\right\}\right\}.
Lemma 2.7.

Over an algebraically closed field of characteristic 2\neq 2, fix g1g\geq 1, and let XX be a smooth complete intersection of two quadrics in 2g\mathbb{P}^{2g}. Let ,m\ell,m be disjoint (g1)(g-1)-planes on XX. Then there exist coordinate points p0,,p2g2gp_{0},\cdots,p_{2g}\in\mathbb{P}^{2g} such that p0,,pg1=\langle p_{0},\dots,p_{g-1}\rangle=\ell, pg,,p2g1=m\langle p_{g},\dots,p_{2g-1}\rangle=m, and XX is defined by two quadrics which correspond in these coordinates to symmetric matrices of the form

(0II0),(0MM0),\begin{pmatrix}0&I&*\\ I&0&*\\ *&*&*\end{pmatrix},\quad\begin{pmatrix}0&M&*\\ M&0&*\\ *&*&*\end{pmatrix},

where MM is a diagonal g×gg\times g matrix with distinct diagonal entries.

For the proof, we need the following result stated in [Rei72, page 44]:

Lemma 2.8.

Over an algebraically closed field of characteristic 2\neq 2, let A,BA,B be n×nn\times n matrices. Then there exist an invertible 2n×2n2n\times 2n matrix LL and a diagonal n×nn\times n matrix MM with distinct diagonal entries such that

LT(0AAT0)L=(0II0),LT(0BBT0)L=(0MM0)L^{T}\begin{pmatrix}0&A\\ A^{T}&0\\ \end{pmatrix}L=\begin{pmatrix}0&I\\ I&0\\ \end{pmatrix},\quad L^{T}\begin{pmatrix}0&B\\ B^{T}&0\\ \end{pmatrix}L=\begin{pmatrix}0&M\\ M&0\\ \end{pmatrix}

if and only if the polynomial det(λA+B)\det(\lambda A+B) has nn distinct roots.

Proof.

The forward direction is immediate. As for the backward direction, the assumption implies that AA is invertible and BA1BA^{-1} has nn distinct eigenvalues. Let CC be an invertible n×nn\times n matrix such that C1BA1CC^{-1}BA^{-1}C equals some diagonal n×nn\times n matrix MM with distinct diagonal entries. Then we may take

L:=((CT)100A1C).L:=\begin{pmatrix}(C^{T})^{-1}&0\\ 0&A^{-1}C\end{pmatrix}.

Proof of Lemma 2.7.

The proof is outlined in [Rei72, page 44]. Here is a detailed argument. Take coordinate points p0,,p2g2gp_{0},\cdots,p_{2g}\in\mathbb{P}^{2g} such that p0,,pg1=\langle p_{0},\dots,p_{g-1}\rangle=\ell and pg,,p2g1=m\langle p_{g},\dots,p_{2g-1}\rangle=m. In these coordinates, any two quadrics defining XX correspond to symmetric matrices of the form

S=(0AAT0),T=(0BBT0),S=\begin{pmatrix}0&A&*\\ A^{T}&0&*\\ *&*&*\end{pmatrix},\quad T=\begin{pmatrix}0&B&*\\ B^{T}&0&*\\ *&*&*\end{pmatrix},

where A,BA,B are g×gg\times g matrices, and, by the smoothness of XX, we may choose those quadrics so that SS is invertible. By [Rei72, Corollary 3.7], the polynomial det(λA+B)\det(\lambda A+B) divides the polynomial det(λS+T)\det(\lambda S+T), where the latter has distinct 2g+12g+1 roots by the smoothness of XX [Rei72, Proposition 2.1]. Hence det(λA+B)\det(\lambda A+B) has distinct gg roots, which implies that after a suitable coordinate change we may take A=IA=I and BB to be diagonal with distinct diagonal entries by Lemma 2.8. Accordingly, we obtain coordinate points with the desired properties. ∎

2.4. 𝐂𝐇2\operatorname{\mathbf{CH}}^{2}-scheme of Benoist–Wittenberg

Throughout this section, let XX be a smooth, proper, geometrically connected, geometrically rational threefold over kk. We recall several key properties of Benoist–Wittenberg’s codimension 2 Chow scheme of XX [BW23], which we will use in the proof of Theorem 1.2.

For such a threefold XX, Benoist and Wittenberg use K-theory to define a functor CHX/k,fppf2\operatorname{CH}^{2}_{X/k,\operatorname{fppf}} for codimension 2 cycles that is analogous to the Picard functor PicX/k,fppf\operatorname{Pic}_{X/k,\operatorname{fppf}}. They show that this functor is represented by a smooth group scheme 𝐂𝐇X/k2\operatorname{\mathbf{CH}}^{2}_{X/k} over kk with the following properties [BW23, Theorem 3.1]:

  1. (1)

    The identity component (𝐂𝐇X/k2)0(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0} is an abelian variety, which we refer to as the intermediate Jacobian of XX. Its base change to the perfect closure of kk agrees with the intermediate Jacobian as defined by Murre [Mur85, ACMV17].

  2. (2)

    There is a GkG_{k}-equivariant isomorphism CH2(Xk¯)𝐂𝐇X/k2(k¯)\operatorname{CH}^{2}(X_{{\overline{k}}})\cong\operatorname{\mathbf{CH}}^{2}_{X/k}({\overline{k}}).

  3. (3)

    The component group 𝐂𝐇X/k2/(𝐂𝐇X/k2)0\operatorname{\mathbf{CH}}^{2}_{X/k}/(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0} is identified with the GkG_{k}-module NS2(Xk¯)\operatorname{NS}^{2}(X_{\overline{k}}).

  4. (4)

    If XX is kk-rational, then there is a smooth projective (not necessarily connected) curve BB over kk such that 𝐂𝐇X/k2\operatorname{\mathbf{CH}}^{2}_{X/k} is a principally polarized direct factor of 𝐏𝐢𝐜B/k\operatorname{\mathbf{Pic}}_{B/k}.

For each class γNS2(Xk¯)Gk=(𝐂𝐇X/k2/(𝐂𝐇X/k2)0)(k)\gamma\in\operatorname{NS}^{2}(X_{{\overline{k}}})^{G_{k}}=(\operatorname{\mathbf{CH}}^{2}_{X/k}/(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0})(k), its inverse image (𝐂𝐇X/k2)γ(\operatorname{\mathbf{CH}}^{2}_{X/k})^{\gamma} in 𝐂𝐇X/k2\operatorname{\mathbf{CH}}^{2}_{X/k} is a(n étale) (𝐂𝐇X/k2)0(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0}-torsor. The quotient map is a group homomorphism, so we have an equality of (𝐂𝐇X/k2)0(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0}-torsors

[(𝐂𝐇X/k2)γ]+[(𝐂𝐇X/k2)γ]=[(𝐂𝐇X/k2)γ+γ][(\operatorname{\mathbf{CH}}^{2}_{X/k})^{\gamma}]+[(\operatorname{\mathbf{CH}}^{2}_{X/k})^{\gamma^{\prime}}]=[(\operatorname{\mathbf{CH}}^{2}_{X/k})^{\gamma+\gamma^{\prime}}]

for any γ,γNS2(Xk¯)Gk\gamma,\gamma^{\prime}\in\operatorname{NS}^{2}(X_{{\overline{k}}})^{G_{k}}.

Hassett–Tschinkel, over k=k=\mathbb{R} [HT21b], and Benoist–Wittenberg, over arbitary fields [BW23], (see also [HT21a] for kk\subset\mathbb{C}) observed that these intermediate Jacobian torsors can be used to refine the rationality obstruction (4). We call their refined obstruction the intermediate Jacobian torsor obstruction to rationality. For simplicity, we state a special case of this obstruction, which is enough for our application:

Theorem 2.9 (Special case of [BW23, Theorem 3.11]).

For XX as above, assume that there exists an isomorphism (𝐂𝐇X/k2)0𝐏𝐢𝐜C/k0(\operatorname{\mathbf{CH}}^{2}_{X/k})^{0}\cong\operatorname{\mathbf{Pic}}^{0}_{C/k} of principally polarized abelian varieties for some smooth, projective, geometrically connected curve CC of genus 2\geq 2. If XX is kk-rational, then for every γ(NS2Xk¯)Gk\gamma\in(\operatorname{NS}^{2}X_{{\overline{k}}})^{G_{k}} there exists an integer dd such that (𝐂𝐇X/k2)γ(\operatorname{\mathbf{CH}}^{2}_{X/k})^{\gamma} and 𝐏𝐢𝐜C/kd\operatorname{\mathbf{Pic}}^{d}_{C/k} are isomorphic as 𝐏𝐢𝐜C/k0\operatorname{\mathbf{Pic}}^{0}_{C/k}-torsors.

3. Fano schemes of linear spaces and hyperbolic reductions of pencils of quadrics

In this section, we construct the hyperbolic reduction of a pencil of quadrics 𝒬1\cal Q\to\mathbb{P}^{1} with respect to a linear subspace in the base locus XX. We show several properties about the hyperbolic reductions, relating them to linear spaces on XX, and we prove Theorems 1.1, 1.3, and 1.4.

3.1. Construction of φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} and its properties

Throughout Section 3.1, we will work in the following setting. Fix an arbitrary field kk of characteristic 2\neq 2 and an integer N2N\geq 2. Let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}, and let φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} be the associated pencil of quadrics. Let 0rN210\leq r\leq\lfloor\frac{N}{2}\rfloor-1. By Lemma 2.1, the Fano scheme Fr(X)F_{r}(X) of rr-planes on XX is non-empty. We will assume Fr(X)(k)F_{r}(X)(k)\neq\emptyset in what follows.

Choose Fr(X)(k)\ell\in F_{r}(X)(k). The projection π:NNr1\pi_{\ell}\colon\mathbb{P}^{N}\dashrightarrow\mathbb{P}^{N-r-1} away from \ell induces diagrams

Nr1(𝒪r+1𝒪(1)){\mathbb{P}_{\mathbb{P}^{N-r-1}}(\mathcal{O}^{\oplus r+1}\oplus\mathcal{O}(1))}N{\mathbb{P}^{N}}Nr1,{\mathbb{P}^{N-r-1},}bl\scriptstyle{\operatorname{bl}_{\ell}}π\scriptstyle{\pi_{\ell}}X~{\widetilde{X}}X{X}Nr1,{\mathbb{P}^{N-r-1},}bl\scriptstyle{\operatorname{bl}_{\ell}}f\scriptstyle{f}π\scriptstyle{\pi_{\ell}}𝒬~{\widetilde{\mathcal{Q}}}𝒬{\mathcal{Q}}1×Nr1.{\mathbb{P}^{1}\times\mathbb{P}^{N-r-1}.}bl1×\scriptstyle{\operatorname{bl}_{\mathbb{P}^{1}\times\ell}}h\scriptstyle{h}id×π\scriptstyle{\operatorname{id}\times\pi_{\ell}}

To be more precise, choose homogeneous coordinates x0,,xNx_{0},\dots,x_{N} for N\mathbb{P}^{N} so that

={xr+1==xN=0}.\ell=\left\{x_{r+1}=\dots=x_{N}=0\right\}.

Then there exist forms lij,qik[xr+1,,xN]l_{ij},q_{i}\in k[x_{r+1},\dots,x_{N}] with deglij=1,degqi=2\deg l_{ij}=1,\deg q_{i}=2 such that

X={(l00l0rq0l10l1rq1)(x0xr1)=0}N.\displaystyle X=\left\{\begin{pmatrix}l_{00}&\dots&l_{0r}&q_{0}\\ l_{10}&\dots&l_{1r}&q_{1}\end{pmatrix}\begin{pmatrix}x_{0}\\ \vdots\\ x_{r}\\ 1\end{pmatrix}=0\right\}\subset\mathbb{P}^{N}.

Choosing a section zz whose zero set equals the divisor Nr1(𝒪r+1)Nr1(𝒪r+1𝒪(1))\mathbb{P}_{\mathbb{P}^{N-r-1}}(\mathcal{O}^{\oplus r+1})\subset\mathbb{P}_{\mathbb{P}^{N-r-1}}(\mathcal{O}^{\oplus r+1}\oplus\mathcal{O}(1)), there exist homogeneous coordinates yr+1,,yNy_{r+1},\dots,y_{N} for Nr1\mathbb{P}^{N-r-1} such that xl=ylzx_{l}=y_{l}z, and

X~={(l00l0rq0l10l1rq1)(x0xrz)=0}Nr1(𝒪r+1𝒪(1)),\widetilde{X}=\left\{\begin{pmatrix}l_{00}&\dots&l_{0r}&q_{0}\\ l_{10}&\dots&l_{1r}&q_{1}\end{pmatrix}\begin{pmatrix}x_{0}\\ \vdots\\ x_{r}\\ z\end{pmatrix}=0\right\}\subset\mathbb{P}_{\mathbb{P}^{N-r-1}}(\mathcal{O}^{r+1}\oplus\mathcal{O}(1)),

where lij,qil_{ij},q_{i} are in yr+1,,yNy_{r+1},\dots,y_{N}. We also have

𝒬={(st)(l00l0rq0l10l1rq1)(x0xr1)=0}1×N,\displaystyle\mathcal{Q}=\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0r}&q_{0}\\ l_{10}&\dots&l_{1r}&q_{1}\end{pmatrix}\begin{pmatrix}x_{0}\\ \vdots\\ x_{r}\\ 1\end{pmatrix}=0\right\}\subset\mathbb{P}^{1}\times\mathbb{P}^{N},
𝒬~={(st)(l00l0rq0l10l1rq1)(x0xrz)=0}1×Nr1(𝒪r+1𝒪(1)).\displaystyle\widetilde{\mathcal{Q}}=\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0r}&q_{0}\\ l_{10}&\dots&l_{1r}&q_{1}\end{pmatrix}\begin{pmatrix}x_{0}\\ \vdots\\ x_{r}\\ z\end{pmatrix}=0\right\}\subset\mathbb{P}^{1}\times\mathbb{P}_{\mathbb{P}^{N-r-1}}(\mathcal{O}^{\oplus r+1}\oplus\mathcal{O}(1)).

We now define

𝒫(r){(st)(l00l0rl10l1r)=0}1×Nr1,\displaystyle\mathcal{P}^{(r)}\coloneqq\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0r}\\ l_{10}&\dots&l_{1r}\end{pmatrix}=0\right\}\subset\mathbb{P}^{1}\times\mathbb{P}^{N-r-1},
𝒬(r){(st)(l00l0rq0l10l1rq1)=0}1×Nr1.\displaystyle\mathcal{Q}^{(r)}\coloneqq\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0r}&q_{0}\\ l_{10}&\dots&l_{1r}&q_{1}\end{pmatrix}=0\right\}\subset\mathbb{P}^{1}\times\mathbb{P}^{N-r-1}.

The first projection defines a N2r2\mathbb{P}^{N-2r-2}-bundle 𝒫(r)1\mathcal{P}^{(r)}\rightarrow\mathbb{P}^{1}, which restricts to a quadric hypersurface fibration φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1}. (By convention, 𝒬(1)=𝒬\mathcal{Q}^{(-1)}=\mathcal{Q} and φ(1)=φ\varphi^{(-1)}=\varphi.) Furthermore, define

E(r)h1(𝒬(r))=𝒬(r)(𝒪r+1𝒪(0,1)),E^{(r)}\coloneqq h^{-1}(\mathcal{Q}^{(r)})=\mathbb{P}_{\mathcal{Q}^{(r)}}(\mathcal{O}^{\oplus r+1}\oplus\mathcal{O}(0,1)),

and let π(r):E(r)𝒬(r)\pi^{(r)}\colon E^{(r)}\rightarrow\mathcal{Q}^{(r)} be the projection.

Lemma 3.1.

𝒬(r)\mathcal{Q}^{(r)} satisfies the following properties.

  1. (1)

    φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} is the hyperbolic reduction of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} with respect to \ell.

  2. (2)

    𝒬(r)\mathcal{Q}^{(r)} is smooth of dimension N2r2N-2r-2. If N2r2>0N-2r-2>0, 𝒬(r)\mathcal{Q}^{(r)} is geometrically connected.

  3. (3)

    π(r):E(r)𝒬(r)\pi^{(r)}\colon E^{(r)}\rightarrow\mathcal{Q}^{(r)} induces an embedding 𝒬(r)Fr+1(𝒬/1)\mathcal{Q}^{(r)}\hookrightarrow F_{r+1}(\mathcal{Q}/\mathbb{P}^{1}) over 1\mathbb{P}^{1}, whose image is the relative Fano scheme of isotropic (r+1)(r+1)-planes of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} containing \ell.

  4. (4)

    The restriction π(r)|E(r)(1×X~):E(r)(1×X~)𝒬(r)\pi^{(r)}|_{E^{(r)}\cap(\mathbb{P}^{1}\times\widetilde{X})}\colon E^{(r)}\cap(\mathbb{P}^{1}\times\widetilde{X})\rightarrow\mathcal{Q}^{(r)} induces an isomorphism

    𝒬(r){lij=0(i=0,1,j=1,,r)}{mFr(X)dimm=r1 and ,mX}.\mathcal{Q}^{(r)}\setminus\left\{l_{ij}=0\,(i=0,1,j=1,\dots,r)\right\}\xrightarrow{\sim}\left\{m\in F_{r}(X)\mid\dim\ell\cap m=r-1\text{ and }\langle\ell,m\rangle\not\subset X\right\}.
  5. (5)

    If N2r2>0N-2r-2>0, the kk-birational equivalence class of 𝒬(r)\mathcal{Q}^{(r)} does not depend on \ell.

  6. (6)

    If N>2N>2, 𝒬(0)\mathcal{Q}^{(0)} is kk-birational to XX. If N=2gN=2g for some g1g\geq 1, 𝒬(g1)\mathcal{Q}^{(g-1)} is of dimension zero and length 2g+12g+1. If N=2g+1N=2g+1 for some g1g\geq 1, 𝒬(g1)\mathcal{Q}^{(g-1)} is kk-isomorphic to the curve CC of genus gg obtained as the Stein factorization of Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1}.

  7. (7)

    If N2r2>0N-2r-2>0, 𝒬(r)\mathcal{Q}^{(r)} has a 0-cycle of degree 11, in other words, the index of 𝒬(r)\mathcal{Q}^{(r)} is 11.

  8. (8)

    If N3r30N-3r-3\geq 0, 𝒬(r)\mathcal{Q}^{(r)} has a kk-point.

Proof.

(1): For x1x\in\mathbb{P}^{1}, 𝒬κ(x)\mathcal{Q}_{\kappa(x)} corresponds to the symmetric matrix

(0AATB),\begin{pmatrix}0&A\\ A^{T}&B\end{pmatrix},

where AA is the Jacobian matrix for {sl00+tl10==sl0r+tl1r=0}\{sl_{00}+tl_{10}=\dots=sl_{0r}+tl_{1r}=0\} and BB is the Hessian matrix for sq0+tq1sq_{0}+tq_{1}. This in turn shows that

={(st)(l00l0rl10l1r)=0}κ(x)N\ell^{\perp}=\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0r}\\ l_{10}&\dots&l_{1r}\end{pmatrix}=0\right\}\subset\mathbb{P}^{N}_{\kappa(x)}

and the hyperbolic reduction of 𝒬κ(x)\mathcal{Q}_{\kappa(x)} with respect to \ell equals 𝒬κ(x)(r)𝒫κ(x)(r)=/κ(x)Nr1=κ(x)N/\mathcal{Q}^{(r)}_{\kappa(x)}\subset\mathcal{P}^{(r)}_{\kappa(x)}=\ell^{\perp}/\ell\subset\mathbb{P}^{N-r-1}_{\kappa(x)}=\mathbb{P}^{N}_{\kappa(x)}/\ell.

(2): The degeneracy locus of φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} is the same as that of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1}, defined by a separable polynomial of degree N+1N+1 (Section 2.2). Hence 𝒬(r)\mathcal{Q}^{(r)} is smooth by Lemma 2.3. Moreover, we deduce from (1) that we have dim𝒬(r)=N2r2\dim\mathcal{Q}^{(r)}=N-2r-2. This implies that 𝒬(r)\mathcal{Q}^{(r)} is a complete intersection of r+2r+2 ample divisors in 1×Nr1\mathbb{P}^{1}\times\mathbb{P}^{N-r-1}. So if N2r2>0N-2r-2>0, then 𝒬(r)\mathcal{Q}^{(r)} is geometrically connected.

(3), (4): Using the equations for 𝒬\mathcal{Q} and 𝒬(r)\mathcal{Q}^{(r)}, we may observe that π(r):E(r)𝒬(r)\pi^{(r)}\colon E^{(r)}\rightarrow\mathcal{Q}^{(r)} is a r+1\mathbb{P}^{r+1}-bundle whose fibers are the (r+1)(r+1)-planes of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} containing \ell. The induced morphism 𝒬(r)Fr+1(𝒬/1)\mathcal{Q}^{(r)}\rightarrow F_{r+1}(\mathcal{Q}/\mathbb{P}^{1}) over 1\mathbb{P}^{1} is an isomorphism onto its image because the blow-up of 𝒬\mathcal{Q} along 1×\mathbb{P}^{1}\times\ell yields the inverse map.

Similarly, using the equations for XX and 𝒬(r)\mathcal{Q}^{(r)}, it is direct to see that

X{lij=0(i=0,1,j=0,,r)}={lij=qi=0(i=0,1,j=0,,r)}NX\cap\left\{l_{ij}=0\,(i=0,1,j=0,\cdots,r)\right\}=\left\{l_{ij}=q_{i}=0\,(i=0,1,j=0,\cdots,r)\right\}\subset\mathbb{P}^{N}

is the union of all linear spaces on XX containing \ell, and that over the complement of the proper closed subscheme 𝒬(r){lij=0(i=0,1,j=1,,r)}𝒬(r)\mathcal{Q}^{(r)}\cap\left\{l_{ij}=0\,(i=0,1,j=1,\dots,r)\right\}\subset\mathcal{Q}^{(r)}, the morphism π(r)|E(r)(1×X~):E(r)(1×X~)𝒬(r)\pi^{(r)}|_{E^{(r)}\cap(\mathbb{P}^{1}\times\widetilde{X})}\colon E^{(r)}\cap(\mathbb{P}^{1}\times\widetilde{X})\rightarrow\mathcal{Q}^{(r)} is a r\mathbb{P}^{r}-bundle whose fibers are the rr-planes mm on XX such that dimm=r1\dim\ell\cap m=r-1 and ,mX\langle\ell,m\rangle\not\subset X. The induced morphism

𝒬(r){lij=0(i=0,1,j=1,,r)}{mFr(X)dimm=r1 and ,mX}\mathcal{Q}^{(r)}\setminus\left\{l_{ij}=0\,(i=0,1,j=1,\dots,r)\right\}\rightarrow\left\{m\in F_{r}(X)\mid\dim\ell\cap m=r-1\text{ and }\langle\ell,m\rangle\not\subset X\right\}

is an isomorphism because the blow-up of XX along \ell yields the inverse map.

(5): It is enough for us to show that the k(1)k(\mathbb{P}^{1})-isomorphism class of the generic fiber of 𝒬(r)1\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} does not depend on \ell. For ,Fr(X)(k)\ell,\ell^{\prime}\in F_{r}(X)(k), denote by 𝒬(r)1,𝒬(r)1\mathcal{Q}^{(r)}_{\ell}\rightarrow\mathbb{P}^{1},\mathcal{Q}^{(r)}_{\ell^{\prime}}\rightarrow\mathbb{P}^{1} the corresponding hyperbolic reductions. Letting HH denote a split quadric, we have

Qk(1)(𝒬(r))k(1)H(𝒬(r))k(1)H.Q_{k(\mathbb{P}^{1})}\simeq(\mathcal{Q}^{(r)}_{\ell})_{k(\mathbb{P}^{1})}\perp H\simeq(\mathcal{Q}^{(r)}_{\ell^{\prime}})_{k(\mathbb{P}^{1})}\perp H.

The Witt cancellation theorem [EKM08, Theorem 8.4] then shows (𝒬(r))k(1)(𝒬(r))k(1)(\mathcal{Q}^{(r)}_{\ell})_{k(\mathbb{P}^{1})}\simeq(\mathcal{Q}^{(r)}_{\ell^{\prime}})_{k(\mathbb{P}^{1})}.

(6): If N>2N>2, then 𝒬(0)\mathcal{Q}^{(0)} is kk-birational to the image of XX under the projection away from a kk-point on XX, hence kk-birational to XX. If N=2gN=2g for some g1g\geq 1, 𝒬(g1)\mathcal{Q}^{(g-1)} is a complete intersection of gg divisors of type (1,1)(1,1) and a divisor of type (1,2)(1,2) in 1×g\mathbb{P}^{1}\times\mathbb{P}^{g}, hence 𝒬(g1)\mathcal{Q}^{(g-1)} is of dimension zero and length 2g+12g+1. If N=2g+1N=2g+1 for some g1g\geq 1, the composition 𝒬(g1)Fg(𝒬/1)C\mathcal{Q}^{(g-1)}\hookrightarrow F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow C is an isomorphism over 1\mathbb{P}^{1} since the double covers φ(g1):𝒬(g1)1\varphi^{(g-1)}\colon\mathcal{Q}^{(g-1)}\rightarrow\mathbb{P}^{1} and C1C\rightarrow\mathbb{P}^{1} are both branched over the degeneracy locus of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1}.

(7): Since 𝒬(r)\mathcal{Q}^{(r)} is a complete intersection of r+1r+1 divisors of type (1,1)(1,1) and a divisor of type (1,2)(1,2) in 1×Nr1\mathbb{P}^{1}\times\mathbb{P}^{N-r-1}, the zero-cycle 𝒬(r)(1×r+1)(r+1)𝒬(r)(×r+2)\mathcal{Q}^{(r)}\cdot(\mathbb{P}^{1}\times\mathbb{P}^{r+1})-(r+1)\mathcal{Q}^{(r)}\cdot(*\times\mathbb{P}^{r+2}) is of degree 11.

(8): If N3r30N-3r-3\geq 0, then {lij=0(i=0,1,j=1,,r)}Nr1\left\{l_{ij}=0\,(i=0,1,j=1,\cdots,r)\right\}\subset\mathbb{P}^{N-r-1} is non-empty by a dimension count. Using the equations for 𝒬(r)\mathcal{Q}^{(r)} and the fact that lijl_{ij} are linear forms, 𝒬(r){lij=0(i=0,1,j=0,,r)}\mathcal{Q}^{(r)}\cap\left\{l_{ij}=0\,(i=0,1,j=0,\cdots,r)\right\} has a kk-point. This concludes the proof. ∎

The following result is well-known to the experts; see [CTSSD87, Remark 3.4.1(a)] and [HT22, Section 6.1], where a special case is stated.

Proposition 3.2.

The following conditions are equivalent:

  1. (1)

    Fr+1(X)(k)F_{r+1}(X)(k)\neq\emptyset;

  2. (2)

    φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} has a relative (r+1)(r+1)-plane containing 1×\mathbb{P}^{1}\times\ell;

  3. (3)

    φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} has a section.

If any of these equivalent conditions holds, then 𝒬(r)\mathcal{Q}^{(r)} is kk-rational.

Proof.

For (1)\Leftrightarrow(2), the Amer–Brumer theorem [Lee07, Theorem 2.2] shows that Fr+1(X)(k)F_{r+1}(X)(k)\neq\emptyset if and only if Fr+1(𝒬k(1))(k(1))F_{r+1}(\mathcal{Q}_{k(\mathbb{P}^{1})})(k(\mathbb{P}^{1}))\neq\emptyset; by the Witt extension theorem [EKM08, Theorem 8.3], the latter is equivalent to the existence of an (r+1)(r+1)-plane on 𝒬k(1)\mathcal{Q}_{k(\mathbb{P}^{1})}, containing \ell, and defined over k(1)k(\mathbb{P}^{1}). Next, the equivalence (2)\Leftrightarrow(3) follows from Lemma 3.1(3). Finally, if (3) holds, then the generic fiber of φ(r)\varphi^{(r)} is a smooth quadric with a k(1)k(\mathbb{P}^{1})-point, so 𝒬(𝓇)\cal Q^{(r)} is kk-rational. ∎

Remark 3.3.

If we allow XX to be singular and assume that Fr(X)(k)\ell\in F_{r}(X)(k) is entirely contained in the smooth locus of XX, most of the results in this section, after some suitable fixes, still hold. An analogue of Proposition 3.2 is also true. The point is that if 𝒬k(1)\mathcal{Q}_{k(\mathbb{P}^{1})} is singular, then 𝒬k(1)(r)\mathcal{Q}^{(r)}_{k(\mathbb{P}^{1})} is singular, but any singular quadric contains a rational point. We do not state and prove results in this generality in this paper because it is not necessary for our purposes.

3.2. Proofs of Theorems 1.1, 1.3, and 1.4

Proof of Theorem 1.3.

Under the assumptions of Theorem 1.3, choose Fr(X)(k)\ell\in F_{r}(X)(k) so that 𝒬(r)\mathcal{Q}^{(r)} is defined. For mFr(X)m\in F_{r}(X), not necessarily a kk-point, we say that mm is general with respect to \ell if: (1) m=\ell\cap m=\emptyset; and (2) X,m,m=2r+1X\cap\langle\ell,m\rangle\subset\langle\ell,m\rangle=\mathbb{P}^{2r+1} is defined by two quadrics which, over an algebraic closure of the residue field of mm and for some choice of coordinates, correspond to symmetric matrices of the form

(0II0),(0MM0),\begin{pmatrix}0&I\\ I&0\end{pmatrix},\quad\begin{pmatrix}0&M\\ M&0\end{pmatrix},

where MM is diagonal with distinct diagonal entries. By Lemma 2.8, property (2) is equivalent to separability of a certain polynomial associated to \ell and mm. Define

U{mFr(X)m is general with respect to }.U\coloneqq\left\{m\in F_{r}(X)\mid m\text{ is general with respect to }\ell\right\}.

UU is a priori only defined over k¯{\overline{k}}, but it actually descends to kk as \ell is defined over kk. Clearly, UU is open in Fr(X)F_{r}(X), and moreover, it is non-empty by Lemma 2.5. For every mUm\in U, Lemma 2.6 implies that the (singular) intersection X,m,m=2r+1X\cap\langle\ell,m\rangle\subset\langle\ell,m\rangle=\mathbb{P}^{2r+1} contains exactly r+1r+1 distinct rr-planes m1,,mr+1m_{1},\dots,m_{r+1} that intersect \ell along an (r1)(r-1)-plane. Using the isomorphism in Lemma 3.1(4), we now define a morphism

ψ:USymr+1𝒬(r),m(m1,,mr+1).\psi\colon U\rightarrow\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)},\quad m\mapsto(m_{1},\dots,m_{r+1}).

The morphism ψ\psi is defined over kk by the same reasoning as above. Moreover, ψ\psi is one-to-one onto its image on the level of k¯{\overline{k}}-points because we have m1,,mr+1=,m\langle m_{1},\cdots,m_{r+1}\rangle=\langle\ell,m\rangle and mm is the unique rr-plane on X,mX\cap\langle\ell,m\rangle that is disjoint from \ell. Since dimFr(X)=dimSymr+1𝒬(r)=(r+1)(N2r2)\dim F_{r}(X)=\dim\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)}=(r+1)(N-2r-2) by Lemma 3.1(2) and ψ\psi factors through the smooth locus of Symr+1𝒬(r)\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)}, it then follows that ψ\psi is an open immersion. If N2r2>0N-2r-2>0, then Fr(X)F_{r}(X) and Symr+1𝒬(r)\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)} are both geometrically integral by Lemmas 2.1 and 3.1, so Fr(X)F_{r}(X) is kk-birational to Symr+1𝒬(r)\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)}.

It remains to consider the case N2r2=0N-2r-2=0. Set g=r+1g=r+1. We have dimFg1(X)=dimSymg𝒬(g1)=0\dim F_{g-1}(X)=\dim\operatorname{Sym}^{g}\mathcal{Q}^{(g-1)}=0. In this case, Lemma 2.7 shows that if mFg1(X)m\in F_{g-1}(X) satisfies property (1), then property (2) is automatic. Hence UU coincides with the subscheme of Fg1(X)F_{g-1}(X) parametrizing (g1)(g-1)-planes on XX disjoint from \ell. It remains to show that ψ\psi gives an isomorphism between UU and the subscheme of Symg𝒬(g1)\operatorname{Sym}^{g}\mathcal{Q}^{(g-1)} parametrizing gg-tuples of distinct points of 𝒬(g1)\mathcal{Q}^{(g-1)}. The former is of length (2g+1g)\binom{2g+1}{g} by the analysis on the configuration of the (g1)(g-1)-planes on XX due to Reid [Rei72, Theorem 3.8], and the latter is also of length (2g+1g)\binom{2g+1}{g} by Lemma 3.1(6). The claim thus follows, and this concludes the proof of the theorem. ∎

Proof of Theorem 1.1.

If Fr+1(X)(k)F_{r+1}(X)(k)\neq\emptyset, then Fr(X)(k)F_{r}(X)(k)\neq\emptyset, and hence 𝒬(r)\mathcal{Q}^{(r)} is defined. Proposition 3.2 shows that 𝒬(r)\mathcal{Q}^{(r)} is kk-rational, and Theorem 1.3 shows that Fr(X)F_{r}(X) is kk-birational to Symr+1𝒬(r)\operatorname{Sym}^{r+1}\mathcal{Q}^{(r)}, hence to Symr+1N2r2\operatorname{Sym}^{r+1}\mathbb{P}^{N-2r-2}. Symmetric powers of a projective space over kk are kk-rational by a result of Mattuck [Mat69], completing the proof. ∎

Proof of Theorem 1.4.

The implications (1)\Rightarrow(2)\Rightarrow(3) are standard, so it remains to show (3)\Rightarrow(1).

First we address the case of F1(X)F_{1}(X) and arbitrary kk. For this, by Theorem 1.3 and [Mat69], it is enough to show that if F1(X)(k)F_{1}(X)(k)\neq\emptyset then 𝒬(1)\mathcal{Q}^{(1)} is separably kk-unirational. We show this statement by induction on NN. By [BW23, Lemma 4.9], we may and will assume that kk is infinite.

As for the base case N=6N=6, by Lemma 3.1(8), 𝒬(1)\mathcal{Q}^{(1)} has a kk-point. Since φ(1):𝒬1\varphi^{(1)}\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} is a conic bundle with 77 singular fibers, a result of Kollár–Mella [KM17, Theorem 7] shows that 𝒬(1)\mathcal{Q}^{(1)} is kk-unirational with a degree 88 dominant rational map 2𝒬(1)\mathbb{P}^{2}\dashrightarrow\mathcal{Q}^{(1)}. Since kk is of characteristic 2\neq 2, 𝒬(1)\mathcal{Q}^{(1)} is separably kk-unirational.

For N>6N>6, choose F1(X)(k)\ell\in F_{1}(X)(k) and consider a general pencil X1X\dashrightarrow\mathbb{P}^{1} of hyperplane sections containing \ell. By the proof of [CT24, Lemma A.4], a general member of the pencil is smooth, hence the resolution of the indeterminacy X1X^{\prime}\rightarrow\mathbb{P}^{1} has a smooth generic fiber. The induced pencil 𝒬(1)1\mathcal{Q}^{(1)}\dashrightarrow\mathbb{P}^{1} yields the resolution of the indeterminacy (𝒬(1))1(\mathcal{Q}^{(1)})^{\prime}\rightarrow\mathbb{P}^{1}, whose generic fiber is the hyperbolic reduction of the generic fiber of X1X^{\prime}\rightarrow\mathbb{P}^{1} with respect to the line \ell and is separably k(1)k(\mathbb{P}^{1})-unirational by induction. This implies that (𝒬(1))(\mathcal{Q}^{(1)})^{\prime} is separably kk-unirational, so is 𝒬(1)\mathcal{Q}^{(1)}. This completes the proof for F1(X)F_{1}(X).

It remains to show for 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2 that if Fr(X)()F_{r}(X)(\mathbb{R})\neq\emptyset, then Fr(X)F_{r}(X) is \mathbb{R}-unirational. Choose Fr(X)()\ell\in F_{r}(X)(\mathbb{R}) so that 𝒬(r)\mathcal{Q}^{(r)} is defined. Then 𝒬(r)\mathcal{Q}^{(r)} has an \mathbb{R}-point by Lemma 3.1(7), so it is \mathbb{R}-unirational by [Kol99, Collorary 1.8] applied to the quadric bundle φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1}. Therefore Fr(X)F_{r}(X) is \mathbb{R}-unirational by Theorem 1.3. ∎

Remark 3.4.

A potential strategy for further extending Theorem 1.4 to Fr(X)F_{r}(X) for arbitrary r>1r>1 over arbitrary fields would be to reduce to showing separable kk-unirationality of the surface 𝒬(r)\mathcal{Q}^{(r)} for X2r+4X\subset\mathbb{P}^{2r+4}. Nevertheless, for r>1r>1, it is not even clear whether 𝒬(r)\mathcal{Q}^{(r)} has a kk-point. The difficulty lies in the fact that a conic bundle over 1\mathbb{P}^{1} with a 0-cycle of degree 11 does not necessarily have a kk-point, as first observed by Colliot-Thélène–Coray [CTC79].

3.3. Rationality results for Fr(X)F_{r}(X) over certain fields

In this section, we give some consequences of Theorems 1.1 and 1.3 over some specific fields kk. First, we prove the following result, which generalizes the r=0r=0 case due to Colliot-Thélène–Sansuc–Swinnerton-Dyer [CTSSD87, Theorem 3.4]. Note that over pp-adic fields, for r=0r=0, our bound is N10N\geq 10, while their bound is N>10N>10.

Corollary 3.5.

Over a field kk of characteristic 2\neq 2, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}.

  1. (1)

    If kk is a CiC_{i}-field for some i0i\geq 0, then, for every 0rN22i10\leq r\leq\lfloor\frac{N}{2}\rfloor-2^{i}-1, Fr(X)F_{r}(X) is kk-rational. In particular, if kk is algebraically closed, then Fr(X)F_{r}(X) is rational for all 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2.

  2. (2)

    If kk is a pp-adic field, then, for every 0rN250\leq r\leq\lfloor\frac{N}{2}\rfloor-5, Fr(X)F_{r}(X) is kk-rational.

  3. (3)

    If kk is a totally imaginary number field, then, for every 0rN350\leq r\leq\lfloor\frac{N}{3}\rfloor-5, Fr(X)F_{r}(X) is kk-rational.

Proof.

(1): The function field k(1)k(\mathbb{P}^{1}) is a Ci+1C_{i+1}-field by the Lang–Nagata theorem. Since the morphism φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} is a fibration into quadrics in N2r1>2i+1N-2r-1>2^{i+1} variables, the assumption implies it has a section. Hence, 𝒬(r)\mathcal{Q}^{(r)} is kk-rational, so Fr(X)F_{r}(X) is kk-rational by Theorem 1.3.

(2): Let kk be a pp-adic field. By theorems of Parimala–Suresh [PS10, Theorem 4.6] and Leep [Lee13, Theorem 3.4] on the uu-invariant of quadratic forms over pp-adic function fields, any quadric bundle of relative dimension 7\geq 7 over a curve defined over kk has a section. Applying this to φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1}, together with Theorem 1.3, yields the result.

(3): Let kk be a totally imaginary number field. If 1rN35-1\leq r\leq\lfloor\frac{N}{3}\rfloor-5, the result of Leep [Lee84, Theorem 2.7] implies that {lij=qi=0(i=0,1,j=0,,r)}Nr1\left\{l_{ij}=q_{i}=0\,(i=0,1,j=0,\cdots,r)\right\}\subset\mathbb{P}^{N-r-1} has a kk-point, which yields a section for φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1}, showing that 𝒬(r)\mathcal{Q}^{(r)} is kk-rational. Theorem 1.3 then implies Fr(X)F_{r}(X) is kk-rational. ∎

If k=𝔽qk=\mathbb{F}_{q} is a finite field, the i=1i=1 case of Corollary 3.5(1) applies, so the only case when 𝔽q\mathbb{F}_{q}-rationality of Fr(X)F_{r}(X) is undetermined is when r=N22r=\lfloor\frac{N}{2}\rfloor-2. When NN is odd, we show 𝔽q\mathbb{F}_{q}-rationality for the remaining case.

Corollary 3.6.

Let 𝔽q\mathbb{F}_{q} be a finite field of characteristic 2\neq 2. Fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g+1\mathbb{P}^{2g+1}. Then, for every 0rg20\leq r\leq g-2, Fr(X)F_{r}(X) is 𝔽q\mathbb{F}_{q}-rational.

Proof.

By Corollary 3.5(1), it remains to consider the case r=g2r=g-2. It follows from [Rei72, Theorem 4.8] that Fg1(X)F_{g-1}(X) is a torsor under an abelian variety over 𝔽q\mathbb{F}_{q}, so it has a 𝔽q\mathbb{F}_{q}-point by Lang’s theorem. Now the result is immediate from Theorem 1.1. ∎

When NN is even, the analogue of Corollary 3.6 does not hold: the second maximal Fano scheme FN/22(X)F_{N/2-2}(X) is not necessarily rational over finite fields. Indeed, for N=4N=4, over any finite field 𝔽q\mathbb{F}_{q} there exist degree 4 del Pezzo surfaces that are not 𝔽q\mathbb{F}_{q}-rational [Ryb05, Theorem 3.2].

Finally, over an algebraically closed field of characteristic 2\neq 2, let CC be a hyperelliptic curve of genus gg. One can then associate to CC a smooth complete intersection XX of two quadrics in 2g+1\mathbb{P}^{2g+1}, uniquely determined up to isomorphism, such that the Stein factorization of Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1} yields CC. In [Ram81], Ramanan established an isomorphism between various moduli spaces of orthogonal and spin bundles on CC and the Fano schemes of linear spaces on XX. More precisely, let 1-1 be the hyperelliptic involution on CC and fix an (1)(-1)-invariant line bundle \mathcal{L} of degree 2g12g-1 on CC. For 1ng1\leq n\leq g, let Un,ξU_{n,\xi} be the moduli space of (1)(-1)-invariant orthogonal bundles \mathcal{E} of rank 2n2n with Γ+(2n)\Gamma^{+}(2n)-structure such that, for every Weierstrass point xCx\in C, the eigenspace associated to the eigenvalue 1-1 on (x)\mathcal{E}\otimes\mathcal{L}(x) has dimension 11. Here Γ+(2n)\Gamma^{+}(2n) is a certain subgroup of the group of units of the Clifford algebra introduced in [Ram81, Section 2]. The moduli space Un,ξU_{n,\xi} is isomorphic to Fgn(X)F_{g-n}(X) by [Ram81, Theorem 3], so using Corollary 3.5(1), we obtain:

Corollary 3.7.

Over an algebraically closed field of characteristic 2\neq 2, let CC be a hyperelliptic curve of genus gg. Then, for every 2ng2\leq n\leq g, the moduli space Un,ξU_{n,\xi} is rational.

Remark 3.8.

Corollary 3.5(1) has the following further consequence:

Corollary 3.9.

Over a field kk of characteristic 0, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}. Let 0r<N220\leq r<\frac{N}{2}-2. Then Br(k)Br(Fr(X))\operatorname{Br}(k)\rightarrow\operatorname{Br}(F_{r}(X)) is surjective.

Proof.

The Leray spectral sequence yields an exact sequence:

Br(k)Ker(Br(Fr(X))Br(Fr(Xk¯)))H1(k,Pic(Fr(Xk¯))).\operatorname{Br}(k)\rightarrow\text{Ker}(\operatorname{Br}(F_{r}(X))\rightarrow\operatorname{Br}(F_{r}(X_{{\overline{k}}})))\rightarrow H^{1}(k,\operatorname{Pic}(F_{r}(X_{\overline{k}}))).

Corollary 3.5(1) shows that Fr(Xk¯)F_{r}(X_{{\overline{k}}}) is rational, hence Br(Fr(Xk¯))=0\operatorname{Br}(F_{r}(X_{{\overline{k}}}))=0. Moreover, we have Pic(Fr(Xk¯))=\operatorname{Pic}(F_{r}(X_{{\overline{k}}}))=\mathbb{Z} by [DM98, Corollaire 3.5], where the Galois group acts trivially, thus H1(k,Pic(Fr(Xk¯)))=0H^{1}(k,\operatorname{Pic}(F_{r}(X_{\overline{k}})))=0. The result is now immediate from the above exact sequence. ∎

A conjecture of Colliot-Thélène states that the Brauer–Manin obstruction is the only obstruction to the Hasse principle for smooth projective rationally connected varieties over a number field. By Corollary 3.9, the conjecture predicts the Hasse principle for rational points on Fr(X)F_{r}(X) for smooth complete intersections of two quadrics XNX\subset\mathbb{P}^{N} and 0r<N220\leq r<\frac{N}{2}-2.

4. The odd-dimensional case

We now focus on the odd-dimensional case and prove Theorem 1.2. Throughout this section, fix g2g\geq 2 and let XX be a smooth complete intersection of two quadrics in 2g+1\mathbb{P}^{2g+1}. By Lemma 2.1, the maximal linear subspaces on XX are (g1)(g-1)-planes. The Stein factorization of the relative Fano scheme Fg(𝒬/1)1F_{g}(\mathcal{Q}/\mathbb{P}^{1})\rightarrow\mathbb{P}^{1} yields a hyperelliptic curve CC of genus gg. Before starting the proof of Theorem 1.2, we outline the main ideas.

For arbitrary g2g\geq 2, the hyperbolic reduction 𝒬(2)\cal Q^{(g-2)} is a threefold. We show its intermediate Jacobian (Section 2.4) is the Jacobian of CC. A key result we need to prove is to identify Fg1(X)F_{g-1}(X) with an intermediate Jacobian torsor given by a certain algebraic curve class on 𝒬(2)\cal Q^{(g-2)} (Proposition 4.10), which we do by studying actions of correspondences coming from hyperbolic reductions. Using the explicit description of the curve classes on 𝒬(2)\cal Q^{(g-2)}, we prove directly that 2[Fg1(X)]=[𝐏𝐢𝐜C/k1]2[F_{g-1}(X)]=[\operatorname{\mathbf{Pic}}^{1}_{C/k}] (Proposition 4.7 is an important step toward this). To prove Theorem 1.2, if the threefold 𝒬(2)\cal Q^{(g-2)} is kk-rational, then the vanishing of the intermediate Jacobian torsor obstruction over kk (Theorem 2.9) implies that the torsor Fg1(X)F_{g-1}(X) is 𝐏𝐢𝐜C/kd\operatorname{\mathbf{Pic}}^{d}_{C/k} for some dd. Combining this with the property that 2[Fg1(X)]=[𝐏𝐢𝐜C/k1]2[F_{g-1}(X)]=[\operatorname{\mathbf{Pic}}^{1}_{C/k}] shows Fg1(X)(k)F_{g-1}(X)(k)\neq\emptyset.

When g=2g=2, 𝒬(0)\cal Q^{(0)} is kk-birational to the threefold X5X\subset\mathbb{P}^{5}. In this case, Hassett–Tschinkel and Benoist–Wittenberg showed that kk-rationality of XX implies F1(X)(k)F_{1}(X)(k)\neq\emptyset [HT21b, BW23]. In their case, the universal line on XX can be used to show that F1(X)F_{1}(X) is an intermediate Jacobian torsor, and a result of Wang shows that 2[F1(X)]=[𝐏𝐢𝐜C/k1]2[F_{1}(X)]=[\operatorname{\mathbf{Pic}}^{1}_{C/k}] as torsors [Wan18]. In the present paper, we do not rely on Wang’s result to prove Theorem 1.2, as we give a direct argument that 2[Fg1(X)]=[𝐏𝐢𝐜C/k1]2[F_{g-1}(X)]=[\operatorname{\mathbf{Pic}}^{1}_{C/k}]. In fact, using Wang’s result does not significantly simplify the proof of Theorem 1.2, because we still need to prove that Fg1(X)F_{g-1}(X) is an intermediate Jacobian torsor for arbitrary gg (Proposition 4.10).

4.1. Preparation for the proof of Theorem 1.2

First, we consider the Fano scheme Fg1(X)F_{g-1}(X) of maximal linear spaces. By [Wan18, Theorem 1.1], Fg1(X)F_{g-1}(X) is a torsor under 𝐏𝐢𝐜C/k0\operatorname{\mathbf{Pic}}^{0}_{C/k}. The following weaker statement, which follows from a theorem of Reid [Rei72, Theorem 4.8], is enough for our application.

Lemma 4.1 (Corollary of [Rei72, Theorem 4.8]).

Fg1(X)F_{g-1}(X) is a torsor under 𝐀𝐥𝐛Fg1(X)/k\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}. In particular, dim𝐀𝐥𝐛Fg1(X)/k=g\dim\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}=g.

Proof.

We will show that the natural map Fg1(X)𝐀𝐥𝐛Fg1(X)/k1F_{g-1}(X)\rightarrow\operatorname{\mathbf{Alb}}^{1}_{F_{g-1}(X)/k} is an isomorphism. (See [Wit08, Section 2] for the definition and basic properties of the Albanese torsor 𝐀𝐥𝐛Fg1(X)/k1\operatorname{\mathbf{Alb}}^{1}_{F_{g-1}(X)/k}.) By the universality of the map Fg1(X)𝐀𝐥𝐛Fg1(X)/k1F_{g-1}(X)\rightarrow\operatorname{\mathbf{Alb}}^{1}_{F_{g-1}(X)/k}, it is enough for us to show that Fg1(Xk¯)F_{g-1}(X_{{\overline{k}}}) is isomorphic to an abelian variety over k¯{\overline{k}}. This follows from the fact that Fg1(Xk¯)F_{g-1}(X_{{\overline{k}}}) is isomorphic to 𝐏𝐢𝐜Ck¯/k¯0\operatorname{\mathbf{Pic}}_{C_{{\overline{k}}}/{\overline{k}}}^{0} by [Rei72, Theorem 4.8]. ∎

For the remainder of this section, fix 0rg10\leq r\leq g-1 and assume Fr(X)(k)F_{r}(X)(k)\neq\emptyset. Choose Fr(X)(k)\ell\in F_{r}(X)(k), and let 𝒬(r)\mathcal{Q}^{(r)} and E(r)E^{(r)} be as defined in Section 3.1. We study the action (E(r))(E^{(r)})_{*} on algebraic cycles. For this purpose, it is useful to regard the hyperbolic reduction with respect to \ell as iterations of the hyperbolic reduction with respect to a point, as follows.

Choose kk-points p0,,prp_{0},\cdots,p_{r}\in\ell with p0,,pr=\langle p_{0},\cdots,p_{r}\rangle=\ell and coordinate points for 2g+1\mathbb{P}^{2g+1} that extend p0,,prp_{0},\cdots,p_{r}, so that we get coordinate expressions of the key varieties as in Section 3. For each 0ir0\leq i\leq r, let

𝒫(i)¯{(st)(l00l0,i1l10l1,i1)=0}1×2gi\overline{\mathcal{P}^{(i)}}\coloneqq\left\{\begin{pmatrix}s&t\end{pmatrix}\begin{pmatrix}l_{00}&\dots&l_{0,i-1}\\ l_{10}&\dots&l_{1,i-1}\end{pmatrix}=0\right\}\subset\mathbb{P}^{1}\times\mathbb{P}^{2g-i}

where the linear forms are in k[xi+1,,x2g+1]k[x_{i+1},\ldots,x_{2g+1}], and let 𝒬(i)\mathcal{Q}^{(i)} be the hyperbolic reduction of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} with respect to ip0,,pi\ell_{i}\coloneqq\langle p_{0},\cdots,p_{i}\rangle. We have a blow-up diagram (see [KS18, Remark 2.6])

𝒬(i1)~{\widetilde{\mathcal{Q}^{(i-1)}}}Ei{E_{i}}𝒬(i1){\mathcal{Q}^{(i-1)}}𝒫(i)¯{\overline{\mathcal{P}^{(i)}}}𝒬(i),{\mathcal{Q}^{(i)},}bl1×pi\scriptstyle{\operatorname{bl}_{\mathbb{P}^{1}\times p_{i}}}bl𝒬(i)\scriptstyle{\operatorname{bl}_{\mathcal{Q}^{(i)}}}πi\scriptstyle{\pi_{i}}id×πpi\scriptstyle{\operatorname{id}\times\pi_{p_{i}}}

where EiE_{i} is the exceptional divisor of bl𝒬(i)\operatorname{bl}_{\mathcal{Q}^{(i)}} and πi:Ei𝒬(i)\pi_{i}\colon E_{i}\rightarrow\mathcal{Q}^{(i)} is the projection.

Lemma 4.2.

The following statements hold.

  1. (1)

    E(r)=ErE0E^{(r)}=E_{r}\circ\dots\circ E_{0} as correpondences.

  2. (2)

    For every 0ir0\leq i\leq r, there is an isomorphism

    (Ei):CHgi+1(𝒬(i1))algCHgi(𝒬(i))alg,α(πi)((bl1×piα)|Ei).(E_{i})_{*}\colon\operatorname{CH}^{g-i+1}(\mathcal{Q}^{(i-1)})_{\operatorname{alg}}\xrightarrow{\sim}\operatorname{CH}^{g-i}(\mathcal{Q}^{(i)})_{\operatorname{alg}},\quad\alpha\mapsto(\pi_{i})_{*}((\operatorname{bl}_{\mathbb{P}^{1}\times p_{i}}^{*}\alpha)|_{E_{i}}).

    The inverse, which we denote by (EiT)-(E_{i}^{T})_{*}, is given by β(bl1×pi)(πi)β\beta\mapsto-(\operatorname{bl}_{\mathbb{P}^{1}\times p_{i}})_{*}(\pi_{i})^{*}\beta.

Proof.

(1): Using Lemma 3.1(3) to identify 𝒬(i)\mathcal{Q}^{(i)} with the relative Fano scheme of isotropic (i+1)(i+1)-planes of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} containing i=p0,,pi\ell_{i}=\langle p_{0},\dots,p_{i}\rangle, we may regard E(i)E^{(i)} as a subscheme of 𝒬×𝒬(i)\mathcal{Q}\times\mathcal{Q}^{(i)} as follows:

E(i)={(x,L)xL}𝒬×𝒬(i).E^{(i)}=\left\{(x,L)\mid x\in L\right\}\subset\mathcal{Q}\times\mathcal{Q}^{(i)}.

Similarly, identifying 𝒬(i)\mathcal{Q}^{(i)} with the relative Fano scheme of isotropic lines of φ(i1):𝒬(i1)1\varphi^{(i-1)}\colon\mathcal{Q}^{(i-1)}\rightarrow\mathbb{P}^{1} containing pip_{i}, we may regard EiE_{i} as a subscheme of 𝒬(i1)×𝒬(i)\mathcal{Q}^{(i-1)}\times\mathcal{Q}^{(i)} as follows:

Ei={(y,m)ym}𝒬(i1)×𝒬(i).E_{i}=\left\{(y,m)\mid y\in m\right\}\subset\mathcal{Q}^{(i-1)}\times\mathcal{Q}^{(i)}.

A line on a fiber of φ(i1):𝒬(i1)1\varphi^{(i-1)}\colon\mathcal{Q}^{(i-1)}\rightarrow\mathbb{P}^{1} and containing pip_{i} corresponds to an (i+1)(i+1)-plane on a fiber of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} and containing i\ell_{i}, so E(i)=EiE(i1)E^{(i)}=E_{i}\circ E^{(i-1)}. Since E(0)=E0E^{(0)}=E_{0}, we have E(r)=ErE(r1)==ErE0E^{(r)}=E_{r}\circ E^{(r-1)}=\dots=E_{r}\circ\dots\circ E_{0}.

(2): For 0ig20\leq i\leq g-2, the blow-up formula for Chow groups implies

CHgi+1(𝒬(i1)~)\displaystyle\operatorname{CH}^{g-i+1}(\widetilde{\mathcal{Q}^{(i-1)}}) =CHgi+1(𝒬(i1))CH0(1)CH1(1)\displaystyle=\operatorname{CH}^{g-i+1}(\mathcal{Q}^{(i-1)})\oplus\operatorname{CH}^{0}(\mathbb{P}^{1})\oplus\operatorname{CH}^{1}(\mathbb{P}^{1})
=CHgi+1(𝒫(i)¯)CHgi(𝒬(i)),\displaystyle=\operatorname{CH}^{g-i+1}(\overline{\mathcal{P}^{(i)}})\oplus\operatorname{CH}^{g-i}(\mathcal{Q}^{(i)}),

where (bl1×P)?(\operatorname{bl}_{\mathbb{P}^{1}\times P})_{*}? and π(?|Ei)-\pi_{*}(?|_{E_{i}}) respectively define the projectors onto CHgi+1(𝒬(i1))\operatorname{CH}^{g-i+1}(\mathcal{Q}^{(i-1)}) and CHgi(𝒬(i))\operatorname{CH}^{g-i}(\mathcal{Q}^{(i)}). For i=g1i=g-1, we have the same decomposition except that CH0(1)\operatorname{CH}^{0}(\mathbb{P}^{1}) does not appear due to dimension reasons. Finally, note that CH0(1)alg=CH1(1)alg=CHgi+1(𝒫(i)¯)alg=0\operatorname{CH}^{0}(\mathbb{P}^{1})_{\operatorname{alg}}=\operatorname{CH}^{1}(\mathbb{P}^{1})_{\operatorname{alg}}=\operatorname{CH}^{g-i+1}(\overline{\mathcal{P}^{(i)}})_{\operatorname{alg}}=0 since 𝒫(i)¯\overline{\mathcal{P}^{(i)}} is a projective bundle over 1\mathbb{P}^{1}. ∎

In particular, Lemma 4.2 implies:

Corollary 4.3.

There is an isomorphism

(E(r)):CHg+1(𝒬)algCHgr(𝒬(r))alg(E^{(r)})_{*}\colon\operatorname{CH}^{g+1}(\mathcal{Q})_{\operatorname{alg}}\xrightarrow{\sim}\operatorname{CH}^{g-r}(\mathcal{Q}^{(r)})_{\operatorname{alg}}

with inverse given by (1)r+1(E(r)T)(-1)^{r+1}(E^{(r)T})_{*}.

4.2. Proof of Theorem 1.2

In the following, we assume Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset. For any Fg2(X)(k)\ell\in F_{g-2}(X)(k), Lemma 3.1(2) and (5) show that 𝒬(g2)\mathcal{Q}^{(g-2)} is a smooth projective geometrically connected threefold, whose kk-birational equivalence class does not depend on \ell. We start by showing that the kk-isomorphism class of the intermediate Jacobian (𝐂𝐇𝒬(g2)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0} (see Section 2.4) does not depend on \ell.

Lemma 4.4.

For ,Fg2(X)(k)\ell,\ell^{\prime}\in F_{g-2}(X)(k), let 𝒬(g2),𝒬(g2)\mathcal{Q}^{(g-2)}_{\ell},\mathcal{Q}^{(g-2)}_{\ell^{\prime}} denote the corresponding hyperbolic reductions of 𝒬\mathcal{Q}, and let E(g2),E(g2)E^{(g-2)}_{\ell},E^{(g-2)}_{\ell^{\prime}} denote the exceptional subschemes. Then the composition (1)g1E(g2)E(g2)T(-1)^{g-1}E^{(g-2)}_{\ell^{\prime}}\circ E^{(g-2)T}_{\ell} induces an isomorphism of abelian varieties

(𝐂𝐇𝒬(g2)/k2)0(𝐂𝐇𝒬(g2)/k2)0.(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell}/k})^{0}\xrightarrow{\sim}(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell^{\prime}}/k})^{0}.
Proof.

Setting Γ(1)g1E(g2)E(g2)T\Gamma\coloneqq(-1)^{g-1}E^{(g-2)}_{\ell^{\prime}}\circ E^{(g-2)T}_{\ell}, we will show that the morphisms

Γ:(𝐂𝐇𝒬(g2)/k2)0(𝐂𝐇𝒬(g2)/k2)0,(ΓT):(𝐂𝐇𝒬(g2)/k2)0(𝐂𝐇𝒬(g2)/k2)0\displaystyle\Gamma_{*}\colon(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell}/k})^{0}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell^{\prime}}/k})^{0},\quad(\Gamma^{T})_{*}\colon(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell^{\prime}}/k})^{0}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell}/k})^{0}

are inverse to each other. By Corollary 4.3, the compositions (ΓT)Γ(\Gamma^{T})_{*}\Gamma_{*} and Γ(ΓT)\Gamma_{*}(\Gamma^{T})_{*} are the identities on the groups CH2(𝒬(g1))alg\operatorname{CH}^{2}(\mathcal{Q}^{(g-1)}_{\ell})_{\operatorname{alg}} and CH2(𝒬(g1))alg\operatorname{CH}^{2}(\mathcal{Q}^{(g-1)}_{\ell^{\prime}})_{\operatorname{alg}}, respectively. Since the intermediate Jacobian of a geometrically rational threefold geometrically agrees with Murre’s intermediate Jacobian [BW23, Theorem 3.1(vi)], the universal property [Mur85, Section 1.8] implies that (ΓT)Γ(\Gamma^{T})_{*}\Gamma_{*} and Γ(ΓT)\Gamma_{*}(\Gamma^{T})_{*} are automorphisms of the abelian varieties (𝐂𝐇𝒬(g1)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-1)}_{\ell}/k})^{0} and (𝐂𝐇𝒬(g2)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}_{\ell^{\prime}}/k})^{0}. This concludes the proof. ∎

Lemma 4.5.

dim(𝐂𝐇𝒬(g2)/k2)0=g\dim(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}=g.

Proof.

By passing to a finite extension of kk, we may assume Fg1(X)(k)F_{g-1}(X)(k)\neq\emptyset. Using Lemma 4.4, we may further assume that \ell is contained in some LFg1(X)(k)L\in F_{g-1}(X)(k). Choosing a kk-point in LL\setminus\ell, by Lemma 3.1(6) and the blow-up formula of Benoist–Wittenberg [BW23, Proposition 3.10], we have an isomorphism of (principally polarized) abelian varieties

(Eg1):(𝐂𝐇𝒬(g2)/k2)0𝐏𝐢𝐜C/k0.(E_{g-1})_{*}\colon(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}\xrightarrow{\sim}\operatorname{\mathbf{Pic}}_{C/k}^{0}.

We conclude dim(𝐂𝐇𝒬(g2)/k2)0=dim𝐏𝐢𝐜C/k0=g\dim(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}=\dim\operatorname{\mathbf{Pic}}_{C/k}^{0}=g. ∎

Lemma 4.6.

The inclusion ι:𝒬(g2)1×g+2\iota\colon\mathcal{Q}^{(g-2)}\rightarrow\mathbb{P}^{1}\times\mathbb{P}^{g+2} induces an isomorphism

ι:NS2(𝒬k¯(g2))NSg+2(k¯1×k¯k¯g+2).\iota_{*}\colon\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}})\xrightarrow{\sim}\operatorname{NS}^{g+2}(\mathbb{P}^{1}_{{\overline{k}}}\times_{{\overline{k}}}\mathbb{P}^{g+2}_{{\overline{k}}}).

In particular, if s,fNS2(𝒬k¯(g2))s,f\in\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}) denote the classes defined by s1×s\mapsto\mathbb{P}^{1}\times* and f×1f\mapsto*\times\mathbb{P}^{1}, we have NS2(𝒬k¯(g2))=sf\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}})=\mathbb{Z}s\oplus\mathbb{Z}f.

Proof.

Over k¯{\overline{k}}, let ss^{\prime} denote the class of a section of the quadric surface fibration φk¯(g2):𝒬k¯(g2)k¯1\varphi^{(g-2)}_{\overline{k}}\colon\mathcal{Q}^{(g-2)}_{{\overline{k}}}\rightarrow\mathbb{P}^{1}_{{\overline{k}}} (which exists by Tsen’s theorem), and let ff denote the class of a k¯{\overline{k}}-line ff on a fiber of φk¯(g2)\varphi^{(g-2)}_{\overline{k}}. We show that

(1) 2NS2(𝒬k¯(g2)),(a,b)as+bf\displaystyle\mathbb{Z}^{2}\rightarrow\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}),\quad(a,b)\mapsto as^{\prime}+bf

is an isomorphism. Since 𝒬k¯(1)(g2)\mathcal{Q}^{(g-2)}_{{\overline{k}}(\mathbb{P}^{1})} is an isotropic quadric, CH2(𝒬k¯(1)(g2))=\operatorname{CH}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}(\mathbb{P}^{1})})=\mathbb{Z} is generated by the image of ss^{\prime}. Moreover, the Chow group of 11-cycles on any fiber of φk¯(g2):𝒬k¯(g2)k¯1\varphi^{(g-2)}_{\overline{k}}\colon\mathcal{Q}^{(g-2)}_{{\overline{k}}}\rightarrow\mathbb{P}^{1}_{{\overline{k}}} is generated by the classes of k¯{\overline{k}}-lines on it. The localization exact sequence then yields that every class αCH2(𝒬k¯(g2))\alpha\in\operatorname{CH}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}) may be written as

α=deg(α/1)s+b1f1++bnfn,\alpha=\deg(\alpha/\mathbb{P}^{1})s^{\prime}+b_{1}f_{1}+\dots+b_{n}f_{n},

where b1,,bnb_{1},\dots,b_{n}\in\mathbb{Z} and f1,,fnf_{1},\dots,f_{n} are k¯{\overline{k}}-lines on fibers of φk¯(g2)\varphi^{(g-2)}_{\overline{k}}. Since F1(𝒬k¯(g2)/k¯1)F_{1}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}/\mathbb{P}^{1}_{{\overline{k}}}) is a 1\mathbb{P}^{1}-bundle over CC, we have f1==fn=ff_{1}=\dots=f_{n}=f in NS2(𝒬k¯(g2))\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}) and the map (1) is surjective. The injectivity of (1) follows from the fact that the images of s,fs^{\prime},f in NSg+2(k¯1×k¯k¯g+2)\operatorname{NS}^{g+2}(\mathbb{P}^{1}_{{\overline{k}}}\times_{{\overline{k}}}\mathbb{P}^{g+2}_{{\overline{k}}}) are linearly independent. Finally, this implies ι\iota_{*} is an isomorphism because ιs\iota_{*}s^{\prime} and ιf\iota_{*}f freely generate NSg+2(k¯1×k¯k¯g+2)\operatorname{NS}^{g+2}(\mathbb{P}^{1}_{{\overline{k}}}\times_{{\overline{k}}}\mathbb{P}^{g+2}_{{\overline{k}}}). ∎

Now we can describe the intermediate Jacobian (𝐂𝐇𝒬(g2)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0} and the torsor associated to the class ff.

Proposition 4.7.

There exists an isomorphism of principally polarized abelian varieties

𝐏𝐢𝐜C/k0(𝐂𝐇𝒬(g2)/k2)0,\operatorname{\mathbf{Pic}}_{C/k}^{0}\xrightarrow{\sim}(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0},

which underlies an isomorphism of torsors

𝐏𝐢𝐜C/k1(𝐂𝐇𝒬(g2)/k2)f.\operatorname{\mathbf{Pic}}_{C/k}^{1}\xrightarrow{\sim}(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{f}.
Proof.

Denote FF1(𝒬(g2)/1)F\coloneqq F_{1}(\mathcal{Q}^{(g-2)}/\mathbb{P}^{1}), and let UU1(𝒬(g2)/1)U\coloneqq U_{1}(\mathcal{Q}^{(g-2)}/\mathbb{P}^{1}) be the universal family. Let p:FCp\colon F\rightarrow C be the morphism induced by the Stein factorization of the natural morphism F1F\rightarrow\mathbb{P}^{1}, and let 1:CC-1\colon C\rightarrow C be the hyperelliptic involution. We have morphisms of torsors

p:𝐀𝐥𝐛F/k1𝐏𝐢𝐜C/k1,U:𝐀𝐥𝐛F/k1(𝐂𝐇𝒬(g2)/k2)fp_{*}\colon\operatorname{\mathbf{Alb}}_{F/k}^{1}\rightarrow\operatorname{\mathbf{Pic}}_{C/k}^{1},\quad U_{*}\colon\operatorname{\mathbf{Alb}}_{F/k}^{1}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{f}

which induce morphisms of abelian varieties

p:𝐀𝐥𝐛F/k𝐏𝐢𝐜C/k0,U:𝐀𝐥𝐛F/k(𝐂𝐇𝒬(g2)/k2)0.p_{*}\colon\operatorname{\mathbf{Alb}}_{F/k}\rightarrow\operatorname{\mathbf{Pic}}_{C/k}^{0},\quad U_{*}\colon\operatorname{\mathbf{Alb}}_{F/k}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}.

Since p:FCp\colon F\rightarrow C is geometrically a 1\mathbb{P}^{1}-bundle, p:𝐀𝐥𝐛F/k𝐏𝐢𝐜C/k0p_{*}\colon\operatorname{\mathbf{Alb}}_{F/k}\rightarrow\operatorname{\mathbf{Pic}}_{C/k}^{0} is an isomorphism of abelian varieties. It now remains for us to show that the composition U(p)1:𝐏𝐢𝐜C/k0(𝐂𝐇𝒬(g2)/k2)0U_{*}\circ(p_{*})^{-1}\colon\operatorname{\mathbf{Pic}}_{C/k}^{0}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0} is an isomorphism of principally polarized abelian varieties. For this, we may assume that kk is an algebraically closed. We have the following diagram.

𝐏𝐢𝐜C/k0{\operatorname{\mathbf{Pic}}^{0}_{C/k}}𝐀𝐥𝐛F/k{\operatorname{\mathbf{Alb}}_{F/k}}(𝐂𝐇𝒬(g2)/k2)0{(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}}𝐏𝐢𝐜C/k0{\operatorname{\mathbf{Pic}}^{0}_{C/k}}𝐏𝐢𝐜F/k0{\operatorname{\mathbf{Pic}}_{F/k}^{0}}p\scriptstyle{p_{*}}U\scriptstyle{U_{*}}UU\scriptstyle{U^{*}\circ U_{*}}U\scriptstyle{U^{*}}p\scriptstyle{p^{*}}

The composition (p)1UU(p)1(p^{*})^{-1}\circ U^{*}\circ U_{*}\circ(p_{*})^{-1} is equal to (1)(-1)^{*}, and thus is an isomorphism. Since Lemma 4.5 shows dim𝐏𝐢𝐜C/k0=dim(𝐂𝐇𝒬(g2)/k2)0=g\dim\operatorname{\mathbf{Pic}}_{C/k}^{0}=\dim(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}=g, all the arrows in the above diagram are isomorphisms of abelian varieties. To show that U(p)1U_{*}\circ(p_{*})^{-1} respects the principal polarizations on 𝐏𝐢𝐜C/k0\operatorname{\mathbf{Pic}}_{C/k}^{0} and (𝐂𝐇𝒬(g2)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}, consider the following diagram of \ell-adic cohomology where all arrows are isomorphisms.

H1(C){H^{1}(C)}H3(F){H^{3}(F)}H3(𝒬(g2)){H^{3}(\mathcal{Q}^{(g-2)})}H1(C){H^{1}(C)}H1(F){H^{1}(F)}(1)\scriptstyle{(-1)^{*}}p\scriptstyle{p_{*}}U\scriptstyle{U_{*}}UU\scriptstyle{U^{*}\circ U_{*}}U\scriptstyle{U^{*}}p\scriptstyle{p^{*}}

For α,βH3(F)\alpha,\beta\in H^{3}(F),

(Uα)(Uβ)=α(UUβ)=(pα)((1)pβ)=(pα)(pβ),(U_{*}\alpha)\cup(U_{*}\beta)=\alpha\cup(U^{*}U_{*}\beta)=(p_{*}\alpha)\cup((-1)^{*}p_{*}\beta)=-(p_{*}\alpha)\cup(p_{*}\beta),

where we have used that the pullback (1)(-1)^{*} on H1(C)H^{1}(C) is equal to multiplication by 1-1. The characterization of the principal polarization on the intermediate Jacobian (𝐂𝐇2)0(\operatorname{\mathbf{CH}}^{2})^{0} in [BW20, Property 2.4, the following comments, and Identity (2.9)] then concludes the proof. ∎

Next, we will use the following lemmas to relate Fg1(X)F_{g-1}(X) to the intermediate Jacobian (𝐂𝐇𝒬(g2)/k2)0(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}.

Lemma 4.8.

Let LNSg(Xk¯)L\in\operatorname{NS}^{g}(X_{{\overline{k}}}) be the class of a (g1)(g-1)-plane on XX. Then there exists aa\in\mathbb{Z} such that

(E(g2))(1×L)=s+af(E^{(g-2)})_{*}(\mathbb{P}^{1}\times L)=s+af

in NS2(𝒬k¯(g2))\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}), where ss and ff are the classes defined in Lemma 4.6.

Proof.

On the generic fiber level, each (Ei)(E_{i})_{*} corresponds to taking the intersection with the tangent hyperplane HiH_{i} to 𝒬k(1)(i1)\mathcal{Q}^{(i-1)}_{k(\mathbb{P}^{1})} at pip_{i} and then projecting to the base of the cone 𝒬k(1)(i1)Hi\mathcal{Q}^{(i-1)}_{k(\mathbb{P}^{1})}\cap H_{i}. Hence a linear space on 𝒬k(1)(i1)\mathcal{Q}^{(i-1)}_{k(\mathbb{P}^{1})} maps to a linear space of one dimension lower on 𝒬k(1)(i)\mathcal{Q}^{(i)}_{k(\mathbb{P}^{1})}. In particular, under (E(g2))(E^{(g-2)})_{*}, a (g1)(g-1)-plane on 𝒬k(1)\mathcal{Q}_{k(\mathbb{P}^{1})} maps to a k(1)k(\mathbb{P}^{1})-point on 𝒬k(1)\mathcal{Q}_{k(\mathbb{P}^{1})}. (Note: since we are interested in algebraic equivalence classes over k¯{\overline{k}}, we may assume that a linear space on a quadric in question does not contain a blown-up point.) ∎

We will also need the following variant of Lemma 4.8.

Lemma 4.9.

Assume Fg1(X)(k)F_{g-1}(X)(k)\neq\emptyset. Choose LFg1(X)(k)L\in F_{g-1}(X)(k) so that 𝒬(g1)\mathcal{Q}^{(g-1)} is defined, and, using Lemma 3.1(3), identify 𝒬(g1)\mathcal{Q}^{(g-1)} with the relative Fano scheme of isotropic gg-planes of 𝒬1\mathcal{Q}\rightarrow\mathbb{P}^{1} containing LL. There exists a divisor class DD on 1×g+1\mathbb{P}^{1}\times\mathbb{P}^{g+1} such that for any (g1)(g-1)-plane MM on XX with dimLM=g2\dim L\cap M=g-2,

(E(g1))(1×M)=(1)g1L,M+𝒬(g1)D(E^{(g-1)})_{*}(\mathbb{P}^{1}\times M)=(-1)^{g-1}\langle L,M\rangle+\mathcal{Q}^{(g-1)}\cdot D

in CH1(𝒬(g1))\operatorname{CH}^{1}(\mathcal{Q}^{(g-1)}).

Proof.

Fix a (g1)(g-1)-plane MM on XX such that dimLM=g2\dim L\cap M=g-2. Choose kk-points p0,,pg1Lp_{0},\dots,p_{g-1}\in L such that LM=p0,,pg2L\cap M=\langle p_{0},\dots,p_{g-2}\rangle and L=p0,,pg1L=\langle p_{0},\dots,p_{g-1}\rangle. We inductively show that for every i{0,,g2}i\in\left\{0,\dots,g-2\right\}

(E(i))(1×M)=(1)i+1VM(i)+𝒬(i)(j=0i(1)j+1((j+1)H1H2gi1H2gi))(E^{(i)})_{*}(\mathbb{P}^{1}\times M)=(-1)^{i+1}V^{(i)}_{M}+\mathcal{Q}^{(i)}\cdot(\sum_{j=0}^{i}(-1)^{j+1}((j+1)H_{1}\cdot H_{2}^{g-i-1}-H_{2}^{g-i}))

in CHgi(𝒬(i))\operatorname{CH}^{g-i}(\mathcal{Q}^{(i)}), where VM(i)V^{(i)}_{M} is the subvariety of 𝒬(i)\mathcal{Q}^{(i)} corresponding to 1×M\mathbb{P}^{1}\times M, and H1,H2H_{1},H_{2} are the pull-backs of 𝒪(1)\mathcal{O}(1) on 1,2gr\mathbb{P}^{1},\mathbb{P}^{2g-r} respectively. This is a consequence of Lemma 4.2 and the equalities

(Ei)VM(i1)=VM(i)+𝒬(i)((i+1)H1gi1H2+H2gi),\displaystyle(E_{i})_{*}V_{M}^{(i-1)}=-V_{M}^{(i)}+\mathcal{Q}^{(i)}\cdot(-(i+1)H_{1}^{g-i-1}\cdot H_{2}+H_{2}^{g-i}),
(Ei)(𝒬(i1)(H1H2gi))=𝒬(i)(H1H2gi1),\displaystyle(E_{i})_{*}(\mathcal{Q}^{(i-1)}\cdot(H_{1}\cdot H_{2}^{g-i}))=\mathcal{Q}^{(i)}\cdot(H_{1}\cdot H_{2}^{g-i-1}),
(Ei)(𝒬(i1)H2gi+1)=𝒬(i)H2gi,\displaystyle(E_{i})_{*}(\mathcal{Q}^{(i-1)}\cdot H_{2}^{g-i+1})=\mathcal{Q}^{(i)}\cdot H_{2}^{g-i},

which may be directly verified. Note that the last two formulas also hold for i=g1i=g-1. Finally, using that there is a unique point p1p\in\mathbb{P}^{1} such that L,M𝒬p\langle L,M\rangle\subset\mathcal{Q}_{p},

L,M=(Eg1)VM(g2),\langle L,M\rangle=(E_{g-1})_{*}V_{M}^{(g-2)},

and we get the desired formula with

D=j=0g2(1)j+1((j+1)H1H2).D=\sum_{j=0}^{g-2}(-1)^{j+1}((j+1)H_{1}-H_{2}).

We are now ready to identify Fg1(X)F_{g-1}(X) with an intermediate Jacobian torsor of 𝒬(2)\cal Q^{(g-2)}.

Proposition 4.10.

For the integer aa in Lemma 4.8, there exists an isomorphism of kk-schemes

Fg1(X)(𝐂𝐇𝒬(g2)/k2)s+af.F_{g-1}(X)\xrightarrow{\sim}(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{s+af}.
Proof.

Let Ug1(X)U_{g-1}(X) denote the universal family associated to Fg1(X)F_{g-1}(X). By Lemma 4.8, E(g2)(1×X)Ug1(X)E^{(g-2)}\circ(\mathbb{P}^{1}\times X)\circ U_{g-1}(X) induces a morphism of kk-schemes

E(g2)(1×X)Ug1(X):Fg1(X)(𝐂𝐇𝒬(g2)/k2)s+af.E^{(g-2)}\circ(\mathbb{P}^{1}\times X)\circ U_{g-1}(X)\colon F_{g-1}(X)\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{s+af}.

We aim to show that this is an isomorphism. By Lemma 4.1, it is enough for us to show that

(E(g2)(1×X)Ug1(X)):𝐀𝐥𝐛Fg1(X)/k(𝐂𝐇𝒬(g2)/k2)0(E^{(g-2)}\circ(\mathbb{P}^{1}\times X)\circ U_{g-1}(X))_{*}\colon\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}\rightarrow(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}

is an isomorphism of abelian varieties. To verify this, we may assume kk is algebraically closed. As in the proof of Lemma 4.5, we may assume that \ell is contained in some (g1)(g-1)-plane LL on XX, and choosing a point in LL\setminus\ell yields an isomorphism of principally polarized abelian varieties

(Eg1):(𝐂𝐇𝒬(g2)/k2)0𝐏𝐢𝐜C/k0.(E_{g-1})_{*}\colon(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}\xrightarrow{\sim}\operatorname{\mathbf{Pic}}_{C/k}^{0}.

Moreover, Lemma 3.1(4) yields a morphism CFg1(X)C\rightarrow F_{g-1}(X), which induces

𝐏𝐢𝐜C/k0𝐀𝐥𝐛Fg1(X)/k.\operatorname{\mathbf{Pic}}_{C/k}^{0}\rightarrow\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}.

We claim that the composition

𝐏𝐢𝐜C/k0𝐀𝐥𝐛Fg1(X)/k(E(g2)(1×X)Ug1(X))(𝐂𝐇𝒬(g2)/k2)0(Eg1)𝐏𝐢𝐜C/k0,\operatorname{\mathbf{Pic}}^{0}_{C/k}\rightarrow\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}\xrightarrow{(E^{(g-2)}\circ(\mathbb{P}^{1}\times X)\circ U_{g-1}(X))_{*}}(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{0}\xrightarrow{(E_{g-1})_{*}}\operatorname{\mathbf{Pic}}^{0}_{C/k},

which by Lemma 4.2 equals the composition

(2) 𝐏𝐢𝐜C/k0𝐀𝐥𝐛Fg1(X)/k(E(g1)(1×X)Ug1(X))𝐏𝐢𝐜C/k0,\displaystyle\operatorname{\mathbf{Pic}}^{0}_{C/k}\rightarrow\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}\xrightarrow{(E^{(g-1)}\circ(\mathbb{P}^{1}\times X)\circ U_{g-1}(X))_{*}}\operatorname{\mathbf{Pic}}^{0}_{C/k},

is (1)g1(-1)^{g-1} times the identity, hence an isomorphism. Since dim𝐏𝐢𝐜C/k0=dim𝐀𝐥𝐛Fg1(X)/k=g\dim\operatorname{\mathbf{Pic}}_{C/k}^{0}=\dim\operatorname{\mathbf{Alb}}_{F_{g-1}(X)/k}=g by Lemma 4.1, this will conclude the proof.

We may identify a given point of CC with a (g1)(g-1)-plane MM on XX such that dimLM=g2\dim L\cap M=g-2, and also with a gg-plane on a fiber of 𝒬1\mathcal{Q}\rightarrow\mathbb{P}^{1} and containing LL, where they correspond by ML,MM\mapsto\langle L,M\rangle. The map in (2) is well-defined on the Picard group of CC, and by Lemma 4.9, the map may be described as

M(1)g1L,M+CD,M\mapsto(-1)^{g-1}\langle L,M\rangle+C\cdot D,

where DD is a constant divisor class on 1×g+1\mathbb{P}^{1}\times\mathbb{P}^{g+1}. The see-saw theorem then shows that the map in (2) is induced by a correspondence of the form

(1)g1ΔC+β×C+C×γ,(-1)^{g-1}\Delta_{C}+\beta\times C+C\times\gamma,

where β,γPic(C)\beta,\gamma\in\operatorname{Pic}(C). (See also [Rei72, Corollary 4.12].) Since β×C\beta\times C and C×γC\times\gamma act trivially on 𝐏𝐢𝐜C/k0\operatorname{\mathbf{Pic}}^{0}_{C/k}, this finishes the proof. ∎

The following shows that the obstruction to the existence of a (g1)(g-1)-plane over kk is of order 44.

Lemma 4.11.

The following statements hold.

  1. (1)

    The cokernel of ι:NS2(k¯1×k¯k¯g+2)NS2(𝒬k¯(g2))\iota^{*}\colon\operatorname{NS}^{2}(\mathbb{P}^{1}_{\overline{k}}\times_{\overline{k}}\mathbb{P}^{g+2}_{\overline{k}})\rightarrow\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}) is generated by ss and is isomorphic to /4\mathbb{Z}/4. The equivalence f2sf\equiv 2s holds in this cokernel.

  2. (2)

    As torsors, (𝐂𝐇𝒬(g2)2)f(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}})^{f} and (𝐂𝐇𝒬(g2)2)2s(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}})^{2s} are kk-isomorphic, and (𝐂𝐇𝒬(g2)2)2f(𝐂𝐇𝒬(g2)2)4s(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}})^{2f}\cong(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}})^{4s} is split over kk.

Proof.

Part (1) indeed follows by observing

1×g2s+(2g1)f,×g+12f\mathbb{P}^{1}\times\mathbb{P}^{g}\mapsto 2s+(2g-1)f,\quad*\times\mathbb{P}^{g+1}\mapsto 2f

under NS2(k¯1×k¯k¯g+2)NS2(𝒬k¯(g2))\operatorname{NS}^{2}(\mathbb{P}^{1}_{\overline{k}}\times_{\overline{k}}\mathbb{P}^{g+2}_{\overline{k}})\rightarrow\operatorname{NS}^{2}(\mathcal{Q}^{(g-2)}_{{\overline{k}}}). Part (2) is immediate from part (1) and properties of the 𝐂𝐇2\operatorname{\mathbf{CH}}^{2}-scheme (see Section 2.4). ∎

Finally, we are ready to prove Theorem 1.2.

Proof of Theorem 1.2.

The backward direction follows from Proposition 3.2, applied for r=g2r=g-2. We show the forward direction. If 𝒬(g2)\mathcal{Q}^{(g-2)} is kk-rational, then Theorem 2.9 and Propositions 4.7 and 4.10 imply that there exists dd\in\mathbb{Z} such that

(3) [Fg1(X)]=[(𝐂𝐇𝒬(g2)/k2)s+af]=[𝐏𝐢𝐜C/kd].\displaystyle[F_{g-1}(X)]=[(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{s+af}]=[\operatorname{\mathbf{Pic}}_{C/k}^{d}].

On the other hand, by Propositions 4.7 and 4.10, Lemma 4.11(2), and additivity of (𝐂𝐇2)0(\operatorname{\mathbf{CH}}^{2})^{0}-torsors (see Section 2.4), we have

2[Fg1(X)]=[(𝐂𝐇𝒬(g2)/k2)2s+2af]=[(𝐂𝐇𝒬(g2)/k2)f]=[𝐏𝐢𝐜C/k1],2[F_{g-1}(X)]=[(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{2s+2af}]=[(\operatorname{\mathbf{CH}}^{2}_{\mathcal{Q}^{(g-2)}/k})^{f}]=[\operatorname{\mathbf{Pic}}_{C/k}^{1}],

which in turn implies

(4) [Fg1(X)]=[𝐏𝐢𝐜C/k1d].\displaystyle[F_{g-1}(X)]=[\operatorname{\mathbf{Pic}}_{C/k}^{1-d}].

. Since the parities of dd and 1d1-d are distinct, the equalities (3) and (4) imply [Fg1(X)]=0[F_{g-1}(X)]=0, i.e. Fg1(X)F_{g-1}(X) has a kk-point, completing the proof. ∎

Remark 4.12.

Let αXBr(C)\alpha_{X}\in\operatorname{Br}(C) be the Brauer class associated to the even Clifford algebra of φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1}. Such Brauer classes of even Clifford algebras arise in the context of derived categories and rationality problems, see e.g. [ABB14]. We now state and prove another version of Theorem 1.2 involving this class αX\alpha_{X}:

Theorem 4.13.

Over a field kk of characteristic 2\neq 2, fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g+1\mathbb{P}^{2g+1}. The following conditions are equivalent:

  1. (1)

    Fg1(X)(k)F_{g-1}(X)(k)\neq\emptyset;

  2. (2)

    αX=0\alpha_{X}=0 in Br(C)\operatorname{Br}(C);

  3. (3)

    Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset and 𝒬(g2)\mathcal{Q}^{(g-2)} is kk-rational;

  4. (4)

    Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset and φ(g2):𝒬(g2)1\varphi^{(g-2)}\colon\mathcal{Q}^{(g-2)}\rightarrow\mathbb{P}^{1} has a section;

  5. (5)

    Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset and p:F1(𝒬(g2)/1)Cp\colon F_{1}(\mathcal{Q}^{(g-2)}/\mathbb{P}^{1})\rightarrow C has a section.

Proof.

Theorem 1.2 shows that (1)\Leftrightarrow(3), and (1)\Leftrightarrow(4) holds by Proposition 3.2. Next, (4)\Leftrightarrow(5) follows from the following geometric argument. If S𝒬(g2)S\subset\mathcal{Q}^{(g-2)} is a section for φ(g2)\varphi^{(g-2)}, then the variety of isotropic lines of φ(g2)\varphi^{(g-2)} intersecting SS gives a section for pp. Conversely, if TF1(𝒬(g2)/1)T\subset F_{1}(\mathcal{Q}^{(g-2)}/\mathbb{P}^{1}) is a section for pp, define a rational map T𝒬(g2)T\dashrightarrow\mathcal{Q}^{(g-2)} as follows. Recall that 1-1 denotes the hyperelliptic involution on CC. For each (k¯{\overline{k}}-)line λ\lambda in a smooth fiber of φ(g2)\varphi^{(g-2)}, we send λ\lambda to the intersection point λ(1)λ\lambda\cap(-1)^{*}\lambda. This map T𝒬(g2)T\dashrightarrow\mathcal{Q}^{(g-2)} is defined over kk because the involution 1-1 is, and the image of TT gives a section for φ(g2)\varphi^{(g-2)}.

We show (5)\Rightarrow(2). By [ABB14, Proposition B.6], αX\alpha_{X} equals the Brauer class corresponding to the smooth conic fibration p:F1(𝒬(g2)/1)Cp\colon F_{1}(\mathcal{Q}^{(g-2)}/\mathbb{P}^{1})\rightarrow C. Since the vanishing of the latter class is equivalent to the existence of the section, (5) is equivalent to:

  1. (6)

    Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset and αX=0\alpha_{X}=0 in Br(C)\operatorname{Br}(C).

Clearly, we have (6)\Rightarrow(2), hence the implication that we want.

It remains for us to show (2)\Rightarrow(1). To achieve this, we use the “generic point” trick (see [CT24, Théorème 5.10]). First, if we assume Fg2(X)(k)F_{g-2}(X)(k)\neq\emptyset, then by (1)\Leftrightarrow(5)\Leftrightarrow(6), we get Fg1(X)F_{g-1}(X)\neq\emptyset. For the general case, we consider the base change X×kk(Fg2(X))X\times_{k}k(F_{g-2}(X)) to the function field of k(Fg2(X))k(F_{g-2}(X)). Then the previous case implies that X×kk(Fg2(X))X\times_{k}k(F_{g-2}(X)) contains a (g1)(g-1)-plane over k(Fg2(X))k(F_{g-2}(X)); equivalently, the torsor Fg1(X)F_{g-1}(X) splits over k(Fg2(X))k(F_{g-2}(X)). Since Fg2(X)F_{g-2}(X) is k¯{\overline{k}}-rational by Corollary 3.5(1), the map H1(k,𝐏𝐢𝐜C0)H1(k(Fg2(X)),𝐏𝐢𝐜C0)H^{1}(k,\operatorname{\mathbf{Pic}}^{0}_{C})\to H^{1}(k(F_{g-2}(X)),\operatorname{\mathbf{Pic}}^{0}_{C}) is injective, and hence Fg1(X)F_{g-1}(X) is split over kk. This completes the proof. ∎

Remark 4.14.

By Theorem 1.2 and Theorem 1.3, the converse to Theorem 1.1 for (r,N)=(g2,2g+1)(r,N)=(g-2,2g+1) would imply that kk-rationality of 𝒬(g2)\mathcal{Q}^{(g-2)} is equivalent to kk-rationality of its (g1)(g-1)th symmetric power. Constructing counterexamples to this latter statement seems to be a subtle problem. For instance, for any Severi–Brauer variety VV of dimension nn over kk, Symn+1V\operatorname{Sym}^{n+1}V is kk-rational [KS04, Theorem 1.4]. However, we can show that 𝒬(g2)\mathcal{Q}^{(g-2)} is not birational to any non-trivial Severi–Brauer variety. Indeed, by Lemma 3.1(7), there exists a zero-cycle of degree 11 on 𝒬(g2)\mathcal{Q}^{(g-2)}. Since non-trivial Severi–Brauer varieties never admit a zero-cycle of degree 11, it is now enough to note that the existence of a zero-cycle of degree 11 is a birational invariant of smooth projective varieties over a field.

5. The even-dimensional case over \mathbb{R}

In this section, we focus on the case of even-dimensional X2gX\subset\mathbb{P}^{2g} defined over the real numbers \mathbb{R}. First, we prove Theorem 1.5. Then, in Section 5.2, we recall an isotopy invariant of Krasnov [Kra18] that was previously used in [HT21b, HKT22] to study \mathbb{R}-rationality of XX. In Section 5.3, we use this invariant to observe several consequences that the isotopy class has for linear subspaces on XX. Finally, we use this to prove Corollary 1.6 and give examples.

5.1. Rationality criterion for Fg2(X)F_{g-2}(X)

The maximal linear subspaces that X2gX\subset\mathbb{P}^{2g} contains over \mathbb{C} are (g1)(g-1)-planes. We first study the \mathbb{R}-rationality of the Fano scheme Fg2(X)F_{g-2}(X) of second maximal linear subspaces, by proving the following more precise version of Theorem 1.5.

Theorem 5.1.

Over the real numbers, fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g\mathbb{P}^{2g}. The following are equivalent:

  1. (1)

    Fg2(X)F_{g-2}(X) is \mathbb{R}-rational;

  2. (2)

    Fg2(X)()F_{g-2}(X)(\mathbb{R}) is non-empty and connected;

  3. (3)

    Fg2(X)()F_{g-2}(X)(\mathbb{R}) is non-empty and 𝒬(g2)\mathcal{Q}^{(g-2)} is \mathbb{R}-rational;

  4. (4)

    Fg2(X)()F_{g-2}(X)(\mathbb{R}) is non-empty and 𝒬(g2)()\mathcal{Q}^{(g-2)}(\mathbb{R}) is non-empty and connected.

One key difference between the even- and odd-dimensional cases is that here 𝒬(2)\cal Q^{(g-2)} is a surface, whereas it is a threefold when dimX\dim X is odd (Section 3.1). As one might expect, the conclusion of Theorem 5.1 fails when XX has odd dimension (see Example 5.10 and the preceding discussion). To prove Theorem 5.1, we will use properties of \mathbb{R}-rationality for surfaces. We will first need several lemmas about the images of real points under quadric fibrations.

Lemma 5.2.

Let φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} be a quadric fibration of positive relative dimension over \mathbb{R}. Then 𝒬()\mathcal{Q}(\mathbb{R}) is connected if and only if the image of the induced map φ():𝒬()1()\varphi(\mathbb{R})\colon\mathcal{Q}(\mathbb{R})\rightarrow\mathbb{P}^{1}(\mathbb{R}) is connected.

Proof.

The forward direction is immediate. As for the backward direction, assume that 𝒬()\mathcal{Q}(\mathbb{R}) is disconnected and let U,V𝒬()U,V\subset\mathcal{Q}(\mathbb{R}) be two disjoint non-empty open subsets that cover and disconnect 𝒬()\mathcal{Q}(\mathbb{R}). For every pp in the image of φ()\varphi(\mathbb{R}), the fiber φ()1(p)\varphi(\mathbb{R})^{-1}(p) is connected, hence is either fully contained in UU or in VV. This shows φ()(U)\varphi(\mathbb{R})(U) and φ()(V)\varphi(\mathbb{R})(V) are disjoint. Moreover, φ()\varphi(\mathbb{R}) is a closed map since 𝒬\cal Q is proper, so φ()(U)\varphi(\mathbb{R})(U) and φ()(V)\varphi(\mathbb{R})(V) are open in the image of φ()\varphi(\mathbb{R}). We conclude that the image of φ()\varphi(\mathbb{R}) is disconnected by φ()(U)\varphi(\mathbb{R})(U) and φ()(V)\varphi(\mathbb{R})(V). ∎

Lemma 5.3.

Let φ:Q1\varphi\colon Q\to\mathbb{P}^{1} be a quadric fibration over \mathbb{R}. If the image of the induced map φ():Q()1()\varphi(\mathbb{R})\colon Q(\mathbb{R})\to\mathbb{P}^{1}(\mathbb{R}) on \mathbb{R}-points is disconnected, then so is the image of the induced map (SymrQ)()(Symr1)()r()(\operatorname{Sym}^{r}Q)(\mathbb{R})\to(\operatorname{Sym}^{r}\mathbb{P}^{1})(\mathbb{R})\cong\mathbb{P}^{r}(\mathbb{R}) on real points of the symmetric powers for any r1r\geq 1. In particular, (SymrQ)()(\operatorname{Sym}^{r}Q)(\mathbb{R}) is disconnected.

Proof.

Let φr:Symr𝒬Sym𝓇1𝓇\varphi^{r}\colon\operatorname{Sym}^{r}\cal Q\to\operatorname{Sym}^{r}\mathbb{P}^{1}\cong\mathbb{P}^{r} be the induced morphism on the symmetric powers, and let ϖ:1××1Symr1r\varpi\colon\mathbb{P}^{1}\times\dots\times\mathbb{P}^{1}\to\operatorname{Sym}^{r}\mathbb{P}^{1}\cong\mathbb{P}^{r} be the quotient by the symmetric action. Let I1,,InI_{1},\dots,I_{n} be the connected components of the complement of the image of φ()\varphi(\mathbb{R}) in 1()\mathbb{P}^{1}(\mathbb{R}), and let Vir()V_{i}\subset\mathbb{P}^{r}(\mathbb{R}) denote the image of Ii×1()r1I_{i}\times\mathbb{P}^{1}(\mathbb{R})^{r-1} under ϖ():1()××1()(Symr1)()r()\varpi(\mathbb{R})\colon\mathbb{P}^{1}(\mathbb{R})\times\dots\times\mathbb{P}^{1}(\mathbb{R})\rightarrow(\operatorname{Sym}^{r}\mathbb{P}^{1})(\mathbb{R})\xrightarrow{\sim}\mathbb{P}^{r}(\mathbb{R}).

First, we claim that the image of φr()\varphi^{r}(\mathbb{R}) is equal to the complement r()(i=1nVi)\mathbb{P}^{r}(\mathbb{R})\setminus(\bigcup_{i=1}^{n}V_{i}). For this, let pr()p\in\mathbb{P}^{r}(\mathbb{R}). If pi=1nVip\notin\bigcup_{i=1}^{n}V_{i}, then the fiber of φr\varphi^{r} over pp is a finite product of \mathbb{R}-varieties, where each component either is the fiber φ1(q)\varphi^{-1}(q) over a real point q1()(i=1nIi)q\in\mathbb{P}^{1}(\mathbb{R})\setminus(\bigcup_{i=1}^{n}I_{i}) or is the Weil restriction R/(φ1(q))R_{\mathbb{C}/\mathbb{R}}(\varphi^{-1}(q)) for some q1()1()q\in\mathbb{P}^{1}(\mathbb{C})\setminus\mathbb{P}^{1}(\mathbb{R}). In both cases, these varieties have \mathbb{R}-points, so the fiber (φr)1(p)(\varphi^{r})^{-1}(p) has an \mathbb{R}-point. On the other hand, if pVip\in V_{i}, then the fiber (φr)1(p)(\varphi^{r})^{-1}(p) over pp is a finite product of \mathbb{R}-varieties, where at least one component is of the form φ1(q)\varphi^{-1}(q) for some qIiq\in I_{i}, and hence the fiber has no \mathbb{R}-points.

It remains to show that r()(i=1nVi)\mathbb{P}^{r}(\mathbb{R})\setminus(\bigcup_{i=1}^{n}V_{i}) is disconnected. For this, choose points s1I1,s2I2s_{1}\in I_{1},s_{2}\in I_{2}. For i=1,2i=1,2, since Ii1()I_{i}\subset\mathbb{P}^{1}(\mathbb{R}) is an open interval, it deformation retracts to sis_{i}. Let Hir()H_{i}\subset\mathbb{P}^{r}(\mathbb{R}) be the image of si×1()r1s_{i}\times\mathbb{P}^{1}(\mathbb{R})^{r-1} under ϖ()\varpi(\mathbb{R}). Each HiH_{i} is the real locus of a hyperplane, and ViV_{i} deformation retracts to HiH_{i}. Since the complement r()(H1H2)\mathbb{P}^{r}(\mathbb{R})\setminus(H_{1}\cup H_{2}) is disconnected, we see that r()(V1V2)\mathbb{P}^{r}(\mathbb{R})\setminus(V_{1}\cup V_{2}) is as well. This implies that the image of φr()\varphi^{r}(\mathbb{R}) is disconnected, since the image of φr()\varphi^{r}(\mathbb{R}) is contained in this latter set and intersects each of the two connected components. (For this latter claim, we may reduce to the r=2r=2 case. Then for any p1()p\in\mathbb{P}^{1}(\mathbb{R}), the image of p×1()p\times\mathbb{P}^{1}(\mathbb{R}) under the quotient ϖ()\varpi(\mathbb{R}) is a line tangent to the image of the diagonal, so this line properly intersects both V1V_{1} and V2V_{2} if pI1I2p\notin I_{1}\cup I_{2}.) ∎

Proof of Theorem 5.1.

We may assume throughout the proof that Fg2(X)()F_{g-2}(X)(\mathbb{R})\neq\emptyset. First, since 𝒬(2)\cal Q^{(g-2)} is a surface, a result of Comessatti [Com13, pages 54–55] shows that (3)\Leftrightarrow(4).

We next show the implications (1)\Rightarrow(2)\Rightarrow(3)\Rightarrow(1). First, (1)\Rightarrow(2) is [DK81, Theorem 13.3]. For (2)\Rightarrow(3), assume 𝒬(2)\cal Q^{(g-2)} is irrational over \mathbb{R}. Then [Com13] implies 𝒬(2)()\cal Q^{(g-2)}(\mathbb{R}) is disconnected, so Lemmas 5.2 and 5.3 applied to φ(g2):𝒬(2)1\varphi^{(g-2)}\colon\cal Q^{(g-2)}\rightarrow\mathbb{P}^{1} imply that (Symg1𝒬(2))()(\operatorname{Sym}^{g-1}\cal Q^{(g-2)})(\mathbb{R}) is disconnected. The Hilbert–Chow morphism ρ:Hilbg1𝒬(2)Sym1𝒬(2)\rho\colon\operatorname{Hilb}^{g-1}\cal Q^{(g-2)}\to\operatorname{Sym}^{g-1}\cal Q^{(g-2)} is a resolution of singularities [Fog68], and the image of ρ()\rho(\mathbb{R}) intersects every connected component of (Symg1𝒬(2))()(\operatorname{Sym}^{g-1}\cal Q^{(g-2)})(\mathbb{R}), so the real locus (Hilbg1𝒬(2))()(\operatorname{Hilb}^{g-1}\cal Q^{(g-2)})(\mathbb{R}) is disconnected. The number of real connected components is a birational invariant of smooth projective varieties [DK81, Theorem 13.3], so by Theorem 1.3 this implies Fg2(X)()F_{g-2}(X)(\mathbb{R}) is disconnected. Finally, (3)\Rightarrow(1) by Theorem 1.3 and [Mat69]. ∎

5.2. Krasnov’s isotopy classification for real complete intersections of quadrics

In this section, we recall an invariant studied by Krasnov to give an isotopy classification of real complete intersections of two quadrics in N\mathbb{P}^{N} for any N3N\geq 3 [Kra18] (see also [HT21b, Section 11.2] and [HKT22, Section 4.1]).

Over \mathbb{R}, let X={Q0=Q1=0}NX=\{Q_{0}=Q_{1}=0\}\subset\mathbb{P}^{N} be a smooth complete intersection of two quadrics. The degeneracy locus Δ\Delta of the pencil 𝒬={𝓈𝒬0+𝓉𝒬1=0}1\cal Q=\{sQ_{0}+tQ_{1}=0\}\to\mathbb{P}^{1} contains rr real points for some integer 0rN+10\leq r\leq N+1. Consider the 02\mathbb{Z}_{\geq 0}^{2}-valued function defined by the signatures of the real quadratic forms

{sQ0+tQ1(s,t)2 such that s2+t2=1}\{sQ_{0}+tQ_{1}\mid(s,t)\in\mathbb{R}^{2}\text{ such that }s^{2}+t^{2}=1\}

as (s,t)(s,t) varies counterclockwise over the unit circle 𝕊12\mathbb{S}^{1}\subset\mathbb{R}^{2}. This function has 2r2r points of discontinuity, given by the preimage of Δ()\Delta(\mathbb{R}) under the quotient 𝕊11()\mathbb{S}^{1}\to\mathbb{P}^{1}(\mathbb{R}). At each of these points, the number of positive eigenvalues either increases (denoted by ++) or decreases (denoted by -) by exactly 1. This gives an odd partition

r=r1++r2u+1r=r_{1}+\dots+r_{2u+1}

where each rir_{i} is the length of a maximal sequence of consecutive ++’s. Following [HKT22], we call the sequence (r1,,r2u+1)(r_{1},\ldots,r_{2u+1}) the Krasnov invariant of XX. It is well defined up to cyclic permutations and reversal of the order. This invariant determines the rigid isotopy class of XX:

Theorem 5.4 ([Kra11, Theorem 1.1], see also [DIK00, Appendix A.4.2]).

For N3N\geq 3, isotopy classes of smooth complete intersections of two real quadrics in N\mathbb{P}^{N} correspond to equivalence classes of odd decompositions r1++r2u+1=rr_{1}+\cdots+r_{2u+1}=r where 0rN+10\leq r\leq N+1 is an integer with parity equal to N+1N+1.

In particular, for each (r1,,r2u+1)(r_{1},\ldots,r_{2u+1}) as above, there exist smooth complete intersections of quadrics XNX\subset\mathbb{P}^{N} with this given Krasnov invariant.

Definition 5.5.

In the above setting, for the Krasnov invariant (r1,,r2u+1)(r_{1},\cdots,r_{2u+1}), [Kra18, Section 2] defines IminI_{\min} to be the minimum number of negative eigenvalues that occurs for sQ0+tQ1sQ_{0}+tQ_{1} on the unit circle 𝕊1\mathbb{S}^{1}. We define the height hN+12Iminh\coloneqq N+1-2I_{\min}, and we define the frequency ff to be the number of distinct intervals, i.e. components of 𝕊1\mathbb{S}^{1} after removing the preimage of Δ()\Delta(\mathbb{R}), where IminI_{\min} is achieved.

One can check that the Krasnov invariant uniquely (up to cyclic permutations and order reversal) determines the following sign sequence

++ri,ri+u+1,++ri+1,ri+u+2,++ri+2,,ri+2u,++ri+u,\displaystyle\overbrace{+\dots+}^{r_{i}},\overbrace{-\dots-}^{r_{i+u+1}},\overbrace{+\dots+}^{r_{i+1}},\overbrace{-\dots-}^{r_{i+u+2}},\overbrace{+\dots+}^{r_{i+2}},\dots,\overbrace{-\dots-}^{r_{i+2u}},\overbrace{+\dots+}^{r_{i+u}},
ri,++ri+u+1,ri+1,++ri+u+2,ri+2,,++ri+2u,ri+u,\displaystyle\underbrace{-\dots-}_{r_{i}},\underbrace{+\dots+}_{r_{i+u+1}},\underbrace{-\dots-}_{r_{i+1}},\underbrace{+\dots+}_{r_{i+u+2}},\underbrace{-\dots-}_{r_{i+2}},\dots,\underbrace{+\dots+}_{r_{i+2u}},\underbrace{-\dots-}_{r_{i+u}},

where the subscripts are modulo 2u+12u+1, so hh and ff in Definition 5.5 are well defined.

Notice that h>N1h>N-1 if and only if the Krasnov invariant is (N+1)(N+1).

5.3. Consequences of the isotopy class for linear subspaces

Hassett–Kollár–Tschinkel and Krasnov observed that Krasnov invariant determines when XX, F1(X)F_{1}(X), and Fg1(X)F_{g-1}(X) have \mathbb{R}-points [HKT22, Proposition 5.1], [Kra18, Theorems 3.1 and 3.6]. We extend their analysis to all linear subspaces on XX:

Lemma 5.6.

Over \mathbb{R}, let XX be a smooth complete intersection of two quadrics in N\mathbb{P}^{N}.

  1. (1)

    Fr(X)()F_{r}(X)(\mathbb{R})\neq\emptyset if and only if hN2r1h\leq N-2r-1.

  2. (2)

    Assume hN2r1h\leq N-2r-1. Then 𝒬(r)()\mathcal{Q}^{(r)}(\mathbb{R}) is non-empty and connected if and only if either hN2r3h\leq N-2r-3 or f=1f=1.

Proof.

(1): For r=0r=0, the Amer–Brumer theorem [Lee07, Theorem 2.2] shows that X()X(\mathbb{R})\neq\emptyset if and only if φ:𝒬1\varphi\colon\mathcal{Q}\rightarrow\mathbb{P}^{1} has a section; this is further equivalent to surjectivity of the induced map φ()\varphi(\mathbb{R}) on real points by a result of Witt [Wit37, Satz 22]. The latter happens exactly when neither (N+1,0)(N+1,0) nor (0,N+1)(0,N+1) appears as the signature of a real fiber of φ\varphi, which is equivalent to hN1h\leq N-1. For r>0r>0, Proposition 3.2 and [Wit37, Satz 22] show that Fr(X)()F_{r}(X)(\mathbb{R})\neq\emptyset if and only if Fr1(X)()F_{r-1}(X)(\mathbb{R})\neq\emptyset and φ(r1)():𝒬(r1)()1()\varphi^{(r-1)}(\mathbb{R})\colon\mathcal{Q}^{(r-1)}(\mathbb{R})\rightarrow\mathbb{P}^{1}(\mathbb{R}) is surjective. Surjectivity of φ(r1)()\varphi^{(r-1)}(\mathbb{R}) happens exactly when (N2r+1,0),(0,N2r+1)(N-2r+1,0),(0,N-2r+1) never appear as the signatures of real fibers of φ(r1)\varphi^{(r-1)}, which is equivalent to hN2r1h\leq N-2r-1.

(2): If hN2r1h\leq N-2r-1, then (1) shows Fr(X)()F_{r}(X)(\mathbb{R})\neq\emptyset. Hence φ(r):𝒬(r)1\varphi^{(r)}\colon\mathcal{Q}^{(r)}\rightarrow\mathbb{P}^{1} is defined, and 𝒬(r)()\mathcal{Q}^{(r)}(\mathbb{R})\neq\emptyset by Lemma 3.1(7). If hN2r3h\leq N-2r-3, then Fr+1(X)()F_{r+1}(X)(\mathbb{R})\neq\emptyset by (1), so by Proposition 3.2, 𝒬(r)()\mathcal{Q}^{(r)}(\mathbb{R}) is \mathbb{R}-rational and in particular has nonempty and connected real locus by [DK81, Theorem 13.3]. Finally, if h=N2r1h=N-2r-1, then (N2r1,0)(N-2r-1,0) appears as the signature of a real fiber of φ(r)\varphi^{(r)}, and the induced map φ(r)()\varphi^{(r)}(\mathbb{R}) on real points is not surjective. By Lemma 5.2, 𝒬(r)()\mathcal{Q}^{(r)}(\mathbb{R}) is connected if and only if the image of φ(r)()\varphi^{(r)}(\mathbb{R}) is connected, and the latter is equivalent to f=1f=1. ∎

Lemma 5.6 and (the proof of) Theorem 1.4 show that the Krasnov invariant determines \mathbb{R}-unirationality for Fano schemes of non-maximal linear subspaces and the corresponding hyperbolic reductions. That is, for XNX\subset\mathbb{P}^{N} and 0rN220\leq r\leq\lfloor\frac{N}{2}\rfloor-2, the \mathbb{R}-unirationality of (resp. 𝒬(𝓇)\cal Q^{(r)}) is determined by the isotopy class of XX.

Furthermore, in the case when N=2gN=2g is even and r=g2r=g-2, Theorem 5.1 and Lemma 5.6 imply that the isotopy class further determines the following \mathbb{R}-(uni)rationality properties. In particular, we can find the isotopy classes of even-dimensional XX violating the conclusion of Theorem 1.2.

Corollary 5.7.

Over \mathbb{R}, fix g2g\geq 2, and let XX be a smooth complete intersection of two quadrics in 2g\mathbb{P}^{2g}.

  1. (1)

    Then \mathbb{R}-rationality of Fg2(X)F_{g-2}(X) (resp. 𝒬(g2)\mathcal{Q}^{(g-2)}) is determined by the isotopy class of XX.

  2. (2)

    Fg1(X)()=F_{g-1}(X)(\mathbb{R})=\emptyset, Fg2(X)F_{g-2}(X)\neq\emptyset, and 𝒬(2)\cal Q^{(g-2)} is \mathbb{R}-rational if and only if h=3h=3 and f=1f=1.

  3. (3)

    Fg2(X)F_{g-2}(X) is \mathbb{R}-unirational but not \mathbb{R}-rational if and only if h=3h=3 and f>1f>1.

In particular, for any g2g\geq 2, there exist smooth complete intersections of quadrics X2gX\subset\mathbb{P}^{2g} with Krasnov invariant (3)(3) by Theorem 5.4, so examples satisfying Corollary 5.7(2) exist in any even dimension.

Remark 5.8.

In general, for N7N\geq 7, it is not known whether the isotopy class of XX determines \mathbb{R}-rationality of Fr(X)F_{r}(X). See [HKT22, Section 6.2] for an isotopy class in the case N=8N=8 where the \mathbb{R}-rationality of its members is unknown.

As a sample application of Lemma 5.6, for the case N=6N=6 we show the following properties for the Fano schemes Fr(X)F_{r}(X), extending the analysis in [HKT22]. One could similarly carry out an analysis for any NN.

Corollary 5.9.

Over \mathbb{R}, let XX be a smooth complete intersection of two quadrics in 6\mathbb{P}^{6}.

  1. (1)

    F2(X)()F_{2}(X)(\mathbb{R}) is non-empty if and only if the Krasnov invariant is one of (1)(1), (1,1,1)(1,1,1), (1,1,1,1,1)(1,1,1,1,1), or (1,1,1,1,1,1,1)(1,1,1,1,1,1,1).

  2. (2)

    F1(X)()F_{1}(X)(\mathbb{R}) is non-empty and 𝒬(1)()\mathcal{Q}^{(1)}(\mathbb{R}) is non-empty and connected if and only if the Krasnov invariant is one of those listed in item 1, or is (3)(3), (2,2,1)(2,2,1), or (2,1,2,1,1)(2,1,2,1,1).

  3. (3)

    F1(X)()F_{1}(X)(\mathbb{R}) is non-empty if and only if Krasnov invariant is one of those listed in item 2, or is (3,1,1)(3,1,1), (3,2,2)(3,2,2), (3,1,1,1,1)(3,1,1,1,1), or (2,2,1,1,1)(2,2,1,1,1).

  4. (4)

    X()X(\mathbb{R}) is non-empty and 𝒬(0)()\mathcal{Q}^{(0)}(\mathbb{R}) is non-empty and connected if and only if the Krasnov invariant is one of those listed in item 3, or is (5)(5), (4,2,1)(4,2,1), or (3,3,1)(3,3,1).

  5. (5)

    X()X(\mathbb{R}) is non-empty if and only if the Krasnov invariant is one of those listed in item 4 or is (5,1,1)(5,1,1).

Proof.

There are only finitely many possible Krasnov invariants for a given smooth complete intersection of two real quadrics. For the list when N8N\leq 8 is even, see the r7r\leq 7 cases in [HKT22, Section 4.1, Figure 1]. Lemma 5.6 and a direct computation then conclude the proof. ∎

Proof of Corollary 1.6.

The result follows by combining Corollary 5.9 with [CTSSD87, Remark 3.28.3], [HKT22, Theorem 1.1], and Theorems 1.4 and 5.1. Since 𝒬(0)\mathcal{Q}^{(0)} is \mathbb{R}-birational to XX (Lemma 3.1(6)), 𝒬(0)()\mathcal{Q}^{(0)}(\mathbb{R}) is non-empty and connected if and only if X()X(\mathbb{R}) is [DK81, Theorem 13.3]. ∎

Finally, we end the paper by returning to odd-dimensional X2g+1X\subset\mathbb{P}^{2g+1}. For any g2g\geq 2, Theorem 1.2 and Lemma 5.6(2) imply that Krasnov invariants with h=4h=4 and f=1f=1 correspond to X2g+1X\subset\mathbb{P}^{2g+1} where 𝒬(2)\cal Q^{(g-2)} has nonempty and connected real locus but is irrational over \mathbb{R}. In particular, applying Theorem 5.4 to the isotopy class (4)(4), we see that Theorem 1.5 fails in every odd dimension. For concreteness, we list the possible Krasnov invariants here for g=2,3g=2,3:

Example 5.10.

For g=2g=2 and N=5N=5, the Krasnov invariants with h=4h=4 and f=1f=1 are (4)(4), (4,1,1)(4,1,1), and (3,2,1)(3,2,1). For g=3g=3 and N=7N=7, the Krasnov invariants with h=4h=4 and f=1f=1 are (4)(4), (4,1,1)(4,1,1), (3,2,1)(3,2,1), (3,1,2,1,1)(3,1,2,1,1), (2,2,2,1,1)(2,2,2,1,1), and (3,3,2)(3,3,2). For each of these, 𝒬(2)()\cal Q^{(g-2)}(\mathbb{R})\neq\emptyset is connected but Fg1(X)()=F_{g-1}(X)(\mathbb{R})=\emptyset, so 𝒬(2)\cal Q^{(g-2)} is irrational over \mathbb{R}.

References

  • [ABB14] Asher Auel, Marcello Bernardara, and Michele Bolognesi. Fibrations in complete intersections of quadrics, Clifford algebras, derived categories, and rationality problems. J. Math. Pures Appl. (9), 102(1):249–291, 2014.
  • [ACMV17] Jeffrey D. Achter, Sebastian Casalaina-Martin, and Charles Vial. On descending cohomology geometrically. Compos. Math., 153(7):1446–1478, 2017.
  • [Bau91] Stefan Bauer. Parabolic bundles, elliptic surfaces and SU(2){\rm SU}(2)-representation spaces of genus zero Fuchsian groups. Math. Ann., 290(3):509–526, 1991.
  • [BG14] Manjul Bhargava and Benedict H. Gross. Arithmetic invariant theory. In Symmetry: representation theory and its applications, volume 257 of Progr. Math., pages 33–54. Birkhäuser/Springer, New York, 2014.
  • [BGW17] Manjul Bhargava, Benedict H. Gross, and Xiaoheng Wang. A positive proportion of locally soluble hyperelliptic curves over \mathbb{Q} have no point over any odd degree extension. J. Amer. Math. Soc., 30(2):451–493, 2017. With an appendix by Tim Dokchitser and Vladimir Dokchitser.
  • [BW20] Olivier Benoist and Olivier Wittenberg. The Clemens-Griffiths method over non-closed fields. Algebr. Geom., 7(6):696–721, 2020.
  • [BW23] Olivier Benoist and Olivier Wittenberg. Intermediate Jacobians and rationality over arbitrary fields. Ann. Sci. Éc. Norm. Supér. (4), 56(4):1029–1086, 2023.
  • [Cas15] Cinzia Casagrande. Rank 22 quasiparabolic vector bundles on 1\mathbb{P}^{1} and the variety of linear subspaces contained in two odd-dimensional quadrics. Math. Z., 280(3-4):981–988, 2015.
  • [Com13] Annibale Comessatti. Fondamenti per la geometria sopra le superficie razionali dal punto di vista reale. Mathematische Annalen, 73:1–72, 1913.
  • [CT24] Jean-Louis Colliot-Thélène. Retour sur l’arithmétique des intersections de deux quadriques, avec un appendice par A. Kuznetsov. J. Reine Angew. Math., 806:147–185, 2024.
  • [CTC79] Jean-Louis Colliot-Thélène and Daniel Coray. L’équivalence rationnelle sur les points fermés des surfaces rationnelles fibrées en coniques. Compositio Math., 39(3):301–332, 1979.
  • [CTSSD87] Jean-Louis Colliot-Thélène, Jean-Jacques Sansuc, and Peter Swinnerton-Dyer. Intersections of two quadrics and Châtelet surfaces. I. J. Reine Angew. Math., 373:37–107, 1987.
  • [DIK00] Alexander Degtyarev, Ilia Itenberg, and Viatcheslav Kharlamov. Real Enriques surfaces, volume 1746 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 2000.
  • [DK81] Hans Delfs and Manfred Knebusch. Semialgebraic topology over a real closed field. II. Basic theory of semialgebraic spaces. Math. Z., 178(2):175–213, 1981.
  • [DM98] Olivier Debarre and Laurent Manivel. Sur la variété des espaces linéaires contenus dans une intersection complète. Math. Ann., 312(3):549–574, 1998.
  • [Don80] Ron Donagi. Group law on the intersection of two quadrics. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 7(2):217–239, 1980.
  • [DR77] Usha V. Desale and Sundararaman Ramanan. Classification of vector bundles of rank 22 on hyperelliptic curves. Invent. Math., 38(2):161–185, 1976/77.
  • [EKM08] Richard Elman, Nikita Karpenko, and Alexander Merkurjev. The algebraic and geometric theory of quadratic forms, volume 56 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 2008.
  • [FJS+22] Sarah Frei, Lena Ji, Soumya Sankar, Bianca Viray, and Isabel Vogt. Curve classes on conic bundle threefolds and applications to rationality. arXiv e-prints, page arXiv:2207.07093, July 2022.
  • [Fog68] John Fogarty. Algebraic families on an algebraic surface. Amer. J. Math., 90:511–521, 1968.
  • [Gau55] Luc Gauthier. Footnote to a footnote of André Weil. Univ. e Politec. Torino Rend. Sem. Mat., 14:325–328, 1954/55.
  • [HKT22] Brendan Hassett, János Kollár, and Yuri Tschinkel. Rationality of even-dimensional intersections of two real quadrics. Comment. Math. Helv., 97(1):183–207, 2022.
  • [HT21a] Brendan Hassett and Yuri Tschinkel. Cycle class maps and birational invariants. Comm. Pure Appl. Math., 74(12):2675–2698, 2021.
  • [HT21b] Brendan Hassett and Yuri Tschinkel. Rationality of complete intersections of two quadrics over nonclosed fields. Enseign. Math., 67(1-2):1–44, 2021. With an appendix by Jean-Louis Colliot-Thélène.
  • [HT22] Brendan Hassett and Yuri Tschinkel. Equivariant geometry of odd-dimensional complete intersections of two quadrics. Pure Appl. Math. Q., 18(4):1555–1597, 2022.
  • [IP22] Jaya N. Iyer and Raman Parimala. Period -Index problem for hyperelliptic curves. arXiv e-prints, page arXiv:2201.12780, January 2022.
  • [JJ24] Lena Ji and Mattie Ji. Rationality of Real Conic Bundles With Quartic Discriminant Curve. Int. Math. Res. Not. IMRN, (1):115–151, 2024.
  • [KM17] János Kollár and Massimiliano Mella. Quadratic families of elliptic curves and unirationality of degree 1 conic bundles. Amer. J. Math., 139(4):915–936, 2017.
  • [Kne15] Amanda Knecht. Degree of unirationality for del Pezzo surfaces over finite fields. J. Théor. Nombres Bordeaux, 27(1):171–182, 2015.
  • [Kol99] János Kollár. Rationally connected varieties over local fields. Ann. of Math. (2), 150(1):357–367, 1999.
  • [Kra11] Vyacheslav A. Krasnov. Real four-dimensional biquadrics. Izv. Ross. Akad. Nauk Ser. Mat., 75(2):151–176, 2011.
  • [Kra18] Vyacheslav A. Krasnov. On the intersection of two real quadrics. Izv. Ross. Akad. Nauk Ser. Mat., 82(1):97–150, 2018.
  • [KS99] Alastair King and Aidan Schofield. Rationality of moduli of vector bundles on curves. Indag. Math. (N.S.), 10(4):519–535, 1999.
  • [KS04] Daniel Krashen and David J. Saltman. Severi-Brauer varieties and symmetric powers. In Algebraic transformation groups and algebraic varieties, volume 132 of Encyclopaedia Math. Sci., pages 59–70. Springer, Berlin, 2004.
  • [KS18] Alexander Kuznetsov and Evgeny Shinder. Grothendieck ring of varieties, D- and L-equivalence, and families of quadrics. Selecta Math. (N.S.), 24(4):3475–3500, 2018.
  • [Kuz21] Alexander Kuznetsov. Quadric bundles and hyperbolic equivalence. arXiv e-prints, page arXiv:2108.01546, August 2021.
  • [Lee84] David B. Leep. Systems of quadratic forms. J. Reine Angew. Math., 350:109–116, 1984.
  • [Lee07] David B. Leep. The Amer–Brumer theorem over arbitrary fields. https://www.ms.uky.edu/~leep/Amer-Brumer_theorem.pdf, 2007.
  • [Lee13] David B. Leep. The uu-invariant of pp-adic function fields. J. Reine Angew. Math., 679:65–73, 2013.
  • [Man86] Yuri I. Manin. Cubic forms, volume 4 of North-Holland Mathematical Library. North-Holland Publishing Co., Amsterdam, second edition, 1986. Algebra, geometry, arithmetic, Translated from the Russian by M. Hazewinkel.
  • [Mat69] Arthur Mattuck. On the symmetric product of a rational surface. Proc. Amer. Math. Soc., 21:683–688, 1969.
  • [Mur85] Jacob P. Murre. Applications of algebraic KK-theory to the theory of algebraic cycles. In Algebraic geometry, Sitges (Barcelona), 1983, volume 1124 of Lecture Notes in Math., pages 216–261. Springer, Berlin, 1985.
  • [New75] Peter E. Newstead. Rationality of moduli spaces of stable bundles. Math. Ann., 215:251–268, 1975.
  • [New80] Peter E. Newstead. Correction to: “Rationality of moduli spaces of stable bundles” [Math. Ann. 215 (1975), 251–268; MR 54 #7468]. Math. Ann., 249(3):281–282, 1980.
  • [PS10] Raman Parimala and Venapally Suresh. The uu-invariant of the function fields of pp-adic curves. Ann. of Math. (2), 172(2):1391–1405, 2010.
  • [Ram81] Sundararaman Ramanan. Orthogonal and spin bundles over hyperelliptic curves. Proc. Indian Acad. Sci. Math. Sci., 90(2):151–166, 1981.
  • [Rei72] Miles Reid. The complete intersection of two or more quadrics. http://homepages.warwick.ac.uk/~masda/3folds/qu.pdf, 1972. Ph.D thesis (Trinity College, Cambridge).
  • [Ryb05] Sergey Rybakov. Zeta functions of conic bundles and Del Pezzo surfaces of degree 4 over finite fields. Mosc. Math. J., 5(4):919–926, 974, 2005.
  • [Tyu75] Andrei N. Tyurin. On intersections of quadrics. Russian Mathematical Surveys, 30(6):51, December 1975.
  • [Wan18] Xiaoheng Wang. Maximal linear spaces contained in the base loci of pencils of quadrics. Algebr. Geom., 5(3):359–397, 2018.
  • [Wit37] Ernst Witt. Theorie der quadratischen Formen in beliebigen Körpern. J. Reine Angew. Math., 176:31–44, 1937.
  • [Wit08] Olivier Wittenberg. On Albanese torsors and the elementary obstruction. Math. Ann., 340(4):805–838, 2008.