betaFeedback

WorksMathematics

Rational lines on pairs of quadrics and cubic hypersurfaces

Abstract

We study rational lines on smooth intersections of two quadrics and on smooth cubic hypersurfaces over number fields. For an intersection of two quadrics in nn variables, we prove the Hasse principle and weak approximation for the variety of lines when n≥11n\geq 11, and more generally for the variety of rr-planes when n≥3r+8n\geq 3r+8. In this range an rr-plane exists over the ground field if and only if one exists over every real completion. For the variety of lines, we also prove the Hasse principle and weak approximation when n≥8n\geq 8, conditional on the finiteness of certain Tate-Shafarevich groups. For a smooth cubic hypersurface in nn variables, we prove that a line exists over the ground field when n≥23n\geq 23.

Introduction

Let kk be a number field and let X⊂Pkn−1X \subset\mathbb{P}^{n-1}_k be a smooth projective variety. For an integer r≥1r \ge1, we write Fr(X)F_r(X) for the Fano variety parametrising rr-dimensional projective linear subspaces Pr⊂X\mathbb{P}^r \subset X. Thus a point of Fr(X)(k)F_r(X)(k) is an (r+1)(r+1)-dimensional vector subspace of knk^n on which the defining equations of XX vanish identically. When r=1r=1, F1(X)F_1(X) is the variety of lines on XX. This paper treats two natural cases, namely smooth complete intersections of two quadrics and smooth cubic hypersurfaces. In each case we use a rational point of XX to reduce the problem to a system of two equations in fewer variables. The following is our first main result.

Theorem 1.1. Let X⊂Pkn−1X \subset\mathbb{P}^{n-1}_k be a smooth complete intersection of two quadrics, with n≥11n \ge11. Then

F1(X)(k)≠∅⟺F1(X)(kv)≠∅for every real place v of k.F_1(X)(k) \ne\varnothing\quad\Longleftrightarrow\quad F_1(X)(k_v) \ne\varnothing\quad\text{for every real place }v\text{ of }k.

If these conditions hold, then F1(X)(k)F_1(X)(k) is dense in F1(X)(Ak)F_1(X)(\mathbb{A}_k). In particular, F1(X)F_1(X) satisfies the Hasse principle and weak approximation.

The real place condition is clearly vacuous when kk is totally imaginary. Hence every smooth intersection of two quadrics in at least 11 variables over such a field contains a line. In general, a concrete way to check the real place condition is provided by Ji and Suzuki [15 Lemma 5.6(1)]. Over Q\mathbb{Q} and for n≥11n \ge11, it follows from Theorem 1.1 that the existence of a rational line is equivalent to the existence of a real line. Consider the smooth intersection of two quadrics defined by

Q0=x12−∑j=2nxj2,Q1=16x22−9x32−∑j=4n(j+10)xj2.Q_0 = x_1^2-\sum_{j=2}^{n}x_j^2,\qquad Q_1 = 16x_2^2-9x_3^2-\sum_{j=4}^{n}(j+10)x_j^2.

This contains the rational point [5:3:4:0:⋯:0][5:3:4:0:\cdots:0], but Q0Q_0 has signature (1,n−1)(1,n-1) and cannot contain a two-dimensional totally isotropic subspace. This therefore gives an example of a smooth intersection over Q\mathbb{Q} with a rational point and no real line.

In fact, we can iterate our fibration argument by induction on rr to handle rr-planes on XX for every r≥1r\ge1.

Corollary 1.2. Let r≥1r\ge1 and let X⊂Pkn−1X\subset\mathbb{P}^{n-1}_k be a smooth complete intersection of two quadrics, with n≥3r+8n\ge3r+8. Then

Fr(X)(k)≠∅⟺Fr(X)(kv)≠∅for every real place v of k.F_r(X)(k)\ne\varnothing\quad\Longleftrightarrow\quad F_r(X)(k_v)\ne\varnothing\quad\text{for every real place }v\text{ of }k.

If these conditions hold, then Fr(X)(k)F_r(X)(k) is dense in Fr(X)(Ak)F_r(X)(\mathbb{A}_k). In particular, Fr(X)F_r(X) satisfies the Hasse principle and weak approximation.

Over a totally imaginary number field, Dietmann and Wooley [9 Equation (2.11)] showed that any pair of quadratic forms in n≥3r+9n\ge3r+9 variables vanishes on an rr-plane defined over the number field. For smooth intersections, Corollary 1.2 saves one variable and also handles arbitrary number fields.

The arithmetic of points on intersections of two quadrics is well understood in sufficiently many variables. The work of Colliot-Thélène, Sansuc and Swinnerton-Dyer [6, 7], followed by Heath-Brown [11], establishes the Hasse principle for smooth intersections in Pn−1\mathbb{P}^{n-1} when n≥8n\ge8. These results form the arithmetic input to Theorem 1.1. In [19 Theorem 3.8], Molyakov has extended this to smooth proper models of geometrically integral, non-conical intersections in P7\mathbb{P}^{7}, but we shall not need this in our work. Our work is inspired by work of Ji and Suzuki [15]. Their Corollaries 3.8 and 3.10 show, for n≥2r+6n\ge2r+6, that Br⁡(k)→Br⁡(Fr(X))\operatorname{Br}(k)\to\operatorname{Br}(F_r(X)) is surjective, and that weak approximation holds as soon as Fr(X)(k)F_r(X)(k) is nonempty. Colliot-Thélène’s conjecture [4] on the Brauer–Manin obstruction therefore predicts the Hasse principle already for n≥2r+6n\ge2r+6. We are able to confirm this in the case r=1r=1 conditionally, assuming the finiteness of certain Tate–Shafarevich groups.

Theorem 1.3. Let n≥8n\ge8 and let X⊂Pkn−1X\subset\mathbb{P}^{n-1}_k be a smooth complete intersection of two quadrics. Assume that

\Sha(\operatorname{Jac}(C)/L)[2^\infty]

is finite for every finite extension L/kL/k and every genus 22 curve C/LC/L. Then F1(X)F_1(X) satisfies the Hasse principle and weak approximation.

The condition in the above theorem is used to control the arithmetic of the intersections of two quadrics arising as fibres in our argument. In particular, work of Morgan–Skorobogatov [20] shows that smooth complete intersections of two quadrics in Pn−1\mathbb{P}^{n-1} for n≥6n\ge6 satisfy the Hasse principle under this assumption.

In smaller dimension the Hasse principle can fail for F1(X)F_1(X). When n=6n=6, the variety of lines is a torsor under the Jacobian of a genus 22 curve [15 Theorem 2.2(2)]. Moreover, Frei and Ji [10] give an explicit smooth intersection in P5\mathbb{P}^{5} over Q\mathbb{Q} which contains a line over every completion of Q\mathbb{Q} but contains no rational line.

Our second main result concerns cubic hypersurfaces.

Theorem 1.4. Let X⊂Pkn−1X \subset\mathbb{P}^{n-1}_k be a smooth cubic hypersurface, with n≥23n \ge23. Then XX contains a kk-line.

Equivalently, if C∈k[x1,…,xn]C \in k[x_1,\ldots,x_n] is a nonsingular cubic form in at least 23 variables, there are linearly independent vectors x,y∈kn\mathbf{x},\mathbf{y} \in k^n such that C(sx+ty)=0C(s\mathbf{x}+t\mathbf{y})=0 for all s,t∈ks,t \in k. Note that there are no local hypotheses in this result.

Conditionally, the Hasse principle holds on F1(X)F_1(X) once n≥7n \ge7. Indeed, F1(X)F_1(X) is a smooth projective Fano variety for n≥7n \ge7 by [13 Chapter 2, Lemma 3.1 and Proposition 3.4]. Thus, Colliot-Thélène’s conjecture [4] predicts that the Brauer–Manin obstruction is the only obstruction to the Hasse principle. Over C\mathbb{C}, Kodaira vanishing gives H2(F1(X)C,OF1(X))=0H^2(F_1(X)_{\mathbb{C}},\mathcal{O}_{F_1(X)})=0, and H3(F1(X)C,Z)H^3(F_1(X)_{\mathbb{C}},\mathbb{Z}) is torsion-free by [13 Chapter 2, Exercise 4.13]. The exponential sequence therefore gives Br⁡(F1(X)C)=0\operatorname{Br}(F_1(X)_{\mathbb{C}})=0, whence Br⁡(F1(X)‾)=0\operatorname{Br}(\overline{F_1(X)})=0 by comparison. Moreover, Pic⁡(F1(X)‾)≅Z\operatorname{Pic}(\overline{F_1(X)}) \cong\mathbb{Z} by [8 Corollary 3.5 and Remark 3.6(5)]. The Galois action is trivial since it fixes the Plücker class. Thus H1(k,Pic⁡(F1(X)‾))=0H^1(k,\operatorname{Pic}(\overline{F_1(X)}))=0 and the Hochschild–Serre sequence implies that Br⁡(k)→Br⁡(F1(X))\operatorname{Br}(k) \to\operatorname{Br}(F_1(X)) is surjective; it follows that F1(X)(Ak)Br⁡=F1(X)(Ak)F_1(X)(\mathbb{A}_k)^{\operatorname{Br}}=F_1(X)(\mathbb{A}_k).

An aspect that is less understood, even conjecturally, is the least number of variables required to guarantee the existence of a line over the non-archimedean completions of kk. Over Qp\mathbb{Q}_p, Brandes and Dietmann [1 Theorems 1.3 and 1.4] show that every cubic form in at least 22 variables vanishes on a line, and that 16 variables suffice for p≥19p \ge19. For each p∈{2,3,5}p \in\{2,3,5\} they also give a smooth cubic in 11 variables with no Qp\mathbb{Q}_p-line.

Wooley [25 Theorem 2(b)] proved that every cubic form in at least 37 variables over a number field vanishes on a rational line. Over Q\mathbb{Q}, Brandes and Dietmann [1 Theorem 1.1] proved that 31 variables suffice for a nonsingular cubic form. Brandes, Dietmann and Leep [2 Theorem 2] subsequently obtained bounds of 35 variables over any number field, 33 over an imaginary quadratic field, and 31 over Q\mathbb{Q}, without assuming nonsingularity. Their argument constructs a line over a suitable quadratic extension and then descends it. As Brandes, Dietmann and Leep observe after [2 Theorem 2], 23 is the limiting number of variables for their method using a general quadratic form. We attain this bound for smooth cubics by choosing the linear space on the quadric so that the cubic restricted to it is nonsingular. We can then apply the ten variable theorem of Browning and Vishe [3 Theorem 1.1].

We proceed by describing the common starting point. Let

I={(a,ℓ)∈X×F1(X):a∈ℓ},π:I⟶X.(1)I=\{(a,\ell)\in X\times F_1(X):a\in\ell\},\qquad\pi:I\longrightarrow X. \tag*{(1)}

The fibre over aa parametrises the lines on XX through aa. If X={Q0=Q1=0}⊂Pn−1X=\{Q_0=Q_1=0\}\subset\mathbb{P}^{n-1} is a smooth intersection of two quadrics over a number field kk and a=[a]a=[\mathbf{a}], write BiB_i for the symmetric bilinear form with Qi(x)=Bi(x,x)Q_i(\mathbf{x})=B_i(\mathbf{x},\mathbf{x}). The conditions on a second vector are

B0(a,y)=B1(a,y)=Q0(y)=Q1(y)=0,y∉ka.B_0(\mathbf{a},\mathbf{y})=B_1(\mathbf{a},\mathbf{y})=Q_0(\mathbf{y})=Q_1(\mathbf{y})=0,\qquad\mathbf{y}\notin k\mathbf{a}.

The linear equations leave n−2n-2 variables. Removing the direction kak\mathbf{a} leaves an intersection of two quadrics Ya⊂Pn−4Y_{\mathbf{a}}\subset\mathbb{P}^{n-4}, which will be smooth for general aa. Thus n=11n=11 is precisely the point at which Heath-Brown’s theorem applies to the fibre. To choose aa with an everywhere locally soluble fibre, we prove that outside a fixed finite set of places, every fibre has a smooth local point. Weak approximation on XX then handles the remaining places. This will be carried out in § 2.

For a cubic hypersurface, choose coordinates at aa so that

a=[1:0:⋯:0],TaX={x2=0}.a=[1:0:\cdots:0],\qquad T_aX=\{x_2=0\}.

Writing z=(x3,…,xn)\mathbf{z}=(x_3,\ldots,x_n), the tangent section has equation

C(x1,0,z)=x1q(z)+c(z),C(x_1,0,\mathbf{z})=x_1q(\mathbf{z})+c(\mathbf{z}),

and the lines through aa are parametrised by q=c=0q=c=0 in Pn−3\mathbb{P}^{n-3}. We choose aa so that qq is nonsingular and this intersection is smooth. Theorem 3 of [2] provides a quadratic extension over which a quadratic form in 2d+12d+1 variables contains a dd-dimensional vector space of zeros. We show that this space can be chosen such that the restriction of the cubic to this space is nonsingular. Taking d=10d=10 gives n−2≥21n-2\geq21, whence n≥23n\geq23. The argument for cubics is explained in § 3.

Acknowledgements. NR would like to thank Professor Colliot-Thélène for pointing him to the reference [8] in regards to the Brauer group of F1(X)F_1(X). During part of this work, NR was funded by the FWF project ESP 441-NBL.

AI statement. The structure of the proofs and most of the ideas in this paper are due to the authors. ChatGPT 6.0 suggested Proposition 2.6 and its proof, as a means of establishing a sufficiently uniform version of local solubility for the intersection of two quadrics Ya⊂Pn−4Y_a\subset\mathbb{P}^{n-4} that emerges in the proof of Theorem 1.1. Moreover, ChatGPT 6.0 suggested the crucial nonsingularity result in Lemma 3.2 that is needed to apply the ten variable theorem for cubics over number fields. Gemini 4 suggested generalising from lines to rr-planes on pairs of quadrics, and helped to provide shorter proofs for Lemmas 2.2, 2.5 and 3.1.

Pairs of quadrics

Lines through a point on two quadrics

In this section let

X={Q0=Q1=0}⊂PKn−1(2)X=\{Q_0=Q_1=0\}\subset\mathbb{P}^{n-1}_K \tag*{(2)}

be a smooth complete intersection over a field KK of characteristic different from 2. Write BiB_i for the associated symmetric bilinear form, with Qi(x)=Bi(x,x)Q_i(\mathbf{x})=B_i(\mathbf{x},\mathbf{x}). All geometric assertions in this section may be checked over an algebraic closure K‾\overline{K}.

Lemma 2.1. If XX in (2) is smooth, then det⁡(sQ0+tQ1)\det(sQ_0+tQ_1) is a nonzero binary form of degree nn with distinct geometric roots. Every member of the geometric pencil has rank at least n−1n-1.

Proof. This is [11 Lemma 3.1]. □

For a=[a]∈X(K)a=[\mathbf{a}]\in X(K), put

Wa={y∈Kn:B0(a,y)=B1(a,y)=0},Ea=Wa/Ka.(3)W_a=\{\mathbf{y}\in K^n:B_0(\mathbf{a},\mathbf{y})=B_1(\mathbf{a},\mathbf{y})=0\},\qquad E_a=W_a/ K\mathbf{a}. \tag*{(3)}

Since XX is a smooth complete intersection, the two gradients ∇Q0(a)\nabla Q_0(\mathbf{a}) and ∇Q1(a)\nabla Q_1(\mathbf{a}) are linearly independent. Thus the equations 2Bi(a,y)=∇Qi(a)⋅y=02B_i(\mathbf{a},\mathbf{y})=\nabla Q_i(\mathbf{a})\cdot\mathbf{y}=0 cut out a space WaW_a of dimension n−2n-2. Moreover, Bi(a,a)=Qi(a)=0B_i(\mathbf{a},\mathbf{a})=Q_i(\mathbf{a})=0 for i=0,1i=0,1, so WaW_a contains the line KaK\mathbf{a}. Taking the quotient by this line gives dim⁡Ea=n−3\dim E_a=n-3. For y∈Wa\mathbf{y}\in W_a and c∈Kc\in K, we clearly have

Qi(y+ca)=Qi(y)+2cBi(a,y)+c2Qi(a)=Qi(y),Q_i(\mathbf{y}+c\mathbf{a})=Q_i(\mathbf{y})+2cB_i(\mathbf{a},\mathbf{y})+c^2Q_i(\mathbf{a})=Q_i(\mathbf{y}),

for i=0,1i = 0, 1. It follows that the restrictions induce quadratic forms R0,a,R1,aR_{0,a}, R_{1,a} on EaE_a.

Define

Ya={R0,a=R1,a=0}⊂P(Ea).(4)Y_a = \{R_{0,a} = R_{1,a} = 0\} \subset\mathbb{P}(E_a). \tag*{(4)}

Choosing a different nonzero representative of aa changes neither WaW_a nor this construction.

Lemma 2.2. For every geometric point a∈Xa \in X and every [s:t]∈P1[s:t] \in\mathbb{P}^1 we have

rank⁡(sR0,a+tR1,a)≥n−5,\operatorname{rank}(sR_{0,a} + tR_{1,a}) \ge n - 5,

and det⁡(sR0,a+tR1,a)\det(sR_{0,a} + tR_{1,a}) is not identically zero. If n≥8n \ge8, then YaY_a is a geometrically integral complete intersection of dimension n−6n - 6. If n=8n = 8 and char⁡K=0\operatorname{char} K = 0, then at the generic point η\eta of XX the polynomial Δη(T)=det⁡(R1,η−TR0,η)\Delta_\eta(T) = \det(R_{1,\eta} - TR_{0,\eta}) is irreducible of degree 55 over K‾(X)\overline{K}(X).

Proof. We work over K‾\overline{K}. By Lemma 2.1, every member of the original pencil has rank at least n−1n - 1. Restricting to WaW_a, which has codimension two, lowers the rank by at most four [6 Lemma 1.8]. Passing to EaE_a leaves the rank unchanged, since aa is already in the kernel. This gives the bound n−5n - 5.

By Lemma 2.1, the pencil has only finitely many singular members over K‾\overline{K}. We may therefore choose two distinct nonsingular members as generators, and denote them again by Q0,Q1Q_0, Q_1. By [12 Proposition 2.1], we may simultaneously diagonalise them and rescale the coordinates. In these new coordinates, we have

Q0(y)=∑i=1nyi2,Q1(y)=∑i=1ndiyi2.Q_0(\mathbf{y}) = \sum_{i=1}^{n} y_i^2,\qquad Q_1(\mathbf{y}) = \sum_{i=1}^{n} d_i y_i^2.

where the did_i are pairwise distinct. To prove that the determinant is not identically zero, it suffices to find one member of the pencil that is nonsingular on EaE_a. Since Ea=Wa/⟨a⟩E_a = W_a/\langle a\rangle, this amounts to finding a member whose kernel on WaW_a is exactly ⟨a⟩\langle a\rangle. Since a≠0a \ne0, we may relabel the coordinates so that an≠0a_n \ne0. We claim that Q=Q1−dnQ0Q = Q_1 - d_nQ_0 has this property. We clearly have

Q(y)=∑i=1n−1(di−dn)yi2.Q(\mathbf{y}) = \sum_{i=1}^{n-1} (d_i - d_n)y_i^2.

Since the did_i are distinct, this is a nonsingular quadratic form in y1,…,yn−1y_1,\ldots,y_{n-1}. Since an≠0a_n \ne0, the equation B0(a,y)=0B_0(a,\mathbf{y}) = 0 determines yny_n. Subtracting dnB0(a,y)d_nB_0(a,\mathbf{y}) from B1(a,y)B_1(a,\mathbf{y}) shows that the first n−1n - 1 coordinates of y∈Wa\mathbf{y} \in W_a satisfy

∑i<n(di−dn)aiyi=0.\sum_{i<n}(d_i-d_n)a_i y_i = 0.

Suppose that y\mathbf{y} lies in the kernel of the restriction of QQ to WaW_a. Then the linear form ∑i<n(di−dn)yizi\sum_{i<n}(d_i-d_n)y_i z_i vanishes whenever ∑i<n(di−dn)aizi=0\sum_{i<n}(d_i-d_n)a_i z_i = 0, so the former is a scalar multiple of the latter. Since di≠dnd_i \ne d_n for i<ni < n, comparison of coefficients gives yi=λaiy_i = \lambda a_i for i<ni < n. Finally,

anyn=−∑i<naiyi=−λ∑i<nai2=λan2,a_n y_n = -\sum_{i<n} a_i y_i = -\lambda\sum_{i<n} a_i^2 = \lambda a_n^2,

since Q0(a)=0Q_0(a) = 0. Thus y=λa\mathbf{y} = \lambda a. As aa already lies in the kernel, the kernel is exactly ⟨a⟩\langle a\rangle, and QQ induces a nonsingular form on EaE_a.

If n≥8n \ge8, every member of the induced pencil has rank at least n−5≥3n - 5 \ge3, and one has rank n−3≥5n - 3 \ge5. The two generators have no common factor. Indeed, a common linear factor would force each to have rank at most two, while a common quadratic factor would make them proportional, contrary to the rank bound for their linear combinations. We may therefore apply [6 Lemma 1.11] to deduce that YaY_a is geometrically integral of codimension two in P(Ea)\mathbb{P}(E_a). It is therefore a complete intersection of dimension n−6n-6.

Finally, suppose that n=8n=8 and char⁡K=0\operatorname{char} K=0. By Jeong [14 Proposition 3.9(i)], the map taking aa to its binary determinant sends [a1:⋯:a8][a_1:\cdots:a_8] to [a12:⋯:a82][a_1^2:\cdots:a_8^2], followed by a linear isomorphism from the image onto the projective space of binary quintics. (Jeong works over C\mathbb{C}, but the same formula holds in characteristic zero by embedding a finitely generated field of definition into C\mathbb{C}.) Since this map is dominant, the generic point η\eta of XX maps to the generic point of the space of binary quintics. Dehomogenising and dividing by the leading coefficient, we therefore obtain

T5+c4T4+c3T3+c2T2+c1T+c0,T^5+c_4T^4+c_3T^3+c_2T^2+c_1T+c_0,

where c0,…,c4c_0,\ldots,c_4 are algebraically independent over K‾\overline{K}. This is the generic monic quintic, which is irreducible over F=K‾(c0,…,c4)F=\overline{K}(c_0,\ldots,c_4). Since the squared coordinates [a12:⋯:a82][a_1^2:\cdots:a_8^2] can be recovered from the coefficients of the quintic by a linear isomorphism, each ratio ai2/a82a_i^2/a_8^2 belongs to FF. To obtain K‾(X)\overline{K}(X) we need only adjoin the ratios ai/a8a_i/a_8 for 1≤i≤71\leq i\leq7. Each is a square root of an element of FF, so [K‾(X):F][\overline{K}(X):F] is a power of two. Put E=K‾(X)E=\overline{K}(X) and let α\alpha be a root of the quintic. Since [F(α):F]=5[F(\alpha):F]=5, the tower law gives

5∣[E(α):F]=[E(α):E][E:F].5\mid[E(\alpha):F]=[E(\alpha):E][E:F].

As [E:F][E:F] is a power of two, 55 divides [E(α):E][E(\alpha):E]. This degree is at most five, since α\alpha satisfies the quintic, so it is exactly five. Thus the quintic is irreducible over EE, and so is its homogenisation, the binary determinant. For the original generators, put

D(U,V)=det⁡(VR1,η−UR0,η),D(U,V)=\det(VR_{1,\eta}-UR_{0,\eta}),

so that D(T,1)=Δη(T)D(T,1)=\Delta_\eta(T). Changing generators amounts to an invertible linear change in U,VU,V, so DD is irreducible over K‾(X)\overline{K}(X). In particular, VV does not divide DD, so its U5U^5 coefficient is nonzero. Thus Δη\Delta_\eta has degree five, and any factorisation of it would homogenise to a factorisation of DD. This proves the lemma. □

Recall the incidence variety II and projection π:I→X\pi:I\to X from (1). The other projection I→F1(X)I\to F_1(X) is a projective line bundle. The fibre π−1(a)\pi^{-1}(a) is naturally isomorphic to YaY_a. Indeed, the point [y]∈Ya[y]\in Y_a corresponds to the line P(Ka+Ky)\mathbb{P}(Ka+Ky). This is an isomorphism of schemes, as is seen by identifying lines through aa with one-dimensional subspaces of Kn/KaK^n/Ka and then imposing the two linear and two quadratic equations for a line to lie on XX. More generally, for any integer r≥1r\geq1, consider the incidence variety

Ir={(a,Λ)∈X×Fr(X):a∈Λ},πr:Ir⟶X(5)I_r=\{(a,\Lambda)\in X\times F_r(X):a\in\Lambda\},\qquad\pi_r:I_r\longrightarrow X \tag*{(5)}

so that I1=II_1=I and π1=π\pi_1=\pi. Sending Λ=P(V)\Lambda=\mathbb{P}(V) to P(V/Ka)⊂P(Ea)\mathbb{P}(V/Ka)\subset\mathbb{P}(E_a) identifies the fibre πr−1(a)\pi_r^{-1}(a) with the variety Fr−1(Ya)F_{r-1}(Y_a) of (r−1)(r-1)-planes on YaY_a, with the convention F0(Ya)=YaF_0(Y_a)=Y_a.

Lemma 2.3. Suppose that char⁡K=0\operatorname{char} K=0 and n≥max⁡{8,2r+5}n\geq\max\{8,2r+5\} for an integer r≥1r\geq1. The varieties Fr(X)F_r(X) and IrI_r are smooth and geometrically integral. There is a nonempty Zariski open subset U⊂XU\subset X, defined over KK, over which πr\pi_r is smooth and YaY_a is a smooth geometrically integral intersection of two quadrics for every a∈Ua\in U.

Proof. For a smooth complete intersection of two quadrics V⊂PNV \subset\mathbb{P}^{N}, the scheme Fs(V)F_s(V) is smooth, projective and nonempty if N≥2s+2N \ge2s+2, and geometrically connected if N≥2s+3N \ge2s+3 by [15 Lemma 2.1]. Thus Fr(X)F_r(X) is smooth and geometrically integral, and the same holds for the Pr\mathbb{P}^{r}-bundle IrI_r.

By [5 Proposition 2.14(f)], there is a nonempty open subset U⊂XU \subset X over which YaY_a is a smooth geometrically integral complete intersection of two quadrics. For each geometric point a∈Ua \in U, the fibre of πr\pi_r is Fr−1(Ya)F_{r-1}(Y_a). This is nonempty by the first assertion above, since n−4≥2(r−1)+2n-4 \ge2(r-1)+2. Hence πr\pi_r is dominant. Shrinking UU once more, generic smoothness in characteristic zero gives the result. □\square

Remark 2.4. The intersection X∩P(Wa)X \cap\mathbb{P}(W_a) is the projective cone over YaY_a with vertex aa. Returning to (3) and (4), we have dim⁡Wa=n−2\dim W_a=n-2 and dim⁡Ea=n−3\dim E_a=n-3, with Ya⊂Pn−4Y_a \subset\mathbb{P}^{n-4}.

Uniform local solubility

The following finite field estimate is the crucial uniform input needed in the fibration argument to come.

Lemma 2.5. Suppose that n≥8n \ge8 and K=FqK=\mathbb{F}_q, where qq is odd and sufficiently large in terms of nn. Then YaY_a has a smooth Fq\mathbb{F}_q-point for every a∈X(Fq)a \in X(\mathbb{F}_q).

Proof. Consider the family over Z\mathbb{Z} of common zero schemes of two quadrics in Pn−4\mathbb{P}^{n-4}, with the coefficients of both quadrics as parameters. Choosing coordinates on EaE_a identifies YaY_a with a fibre of this family. By Lemma 2.2, this fibre is geometrically integral. Thus the uniform Lang–Weil theorem [22 Theorem 7.7.1(iv)] gives a smooth Fq\mathbb{F}_q-point for qq sufficiently large in terms of the family, hence in terms of nn alone. □\square

Proposition 2.6. Let kk be a number field and suppose that n≥8n \ge8. There is a finite set S0S_0 of places of kk, containing every archimedean place, depending only on the equations of XX, such that for every place v∉S0v \notin S_0 and every a∈X(kv)a \in X(k_v), the variety YaY_a has a smooth kvk_v-point.

Proof. Choose S0S_0 to contain the archimedean places, the places above 22, the places where the given equations are not integral or do not define a smooth complete intersection after reduction, and the places whose residue field Fq\mathbb{F}_q fails the bound of Lemma 2.5. For v∉S0v \notin S_0 with valuation ring Ov\mathcal{O}_v, a primitive vector a∈Ovna \in\mathcal{O}_v^n representing a∈X(kv)a \in X(k_v) reduces to a point aˉ\bar{a} on the smooth reduction of XX, so Ea={y∈Ovn:B0(a,y)=B1(a,y)=0}/OvaE_a=\{\mathbf{y}\in\mathcal{O}_v^n:B_0(a,\mathbf{y})=B_1(a,\mathbf{y})=0\}/\mathcal{O}_v a is free of rank n−3n-3 and reduces to E‾a\overline{E}_a. By Lemma 2.5, Y‾a\overline{Y}_a has a smooth Fq\mathbb{F}_q-point, which lifts to a smooth kvk_v-point of YaY_a by Hensel’s lemma [22 Theorem 3.5.63]. □\square

Remark 2.7. Note that the set S0S_0 is chosen before choosing aa; proving that each fixed smooth YaY_a has local points at almost all places would only give a finite exceptional set depending on aa, and would not justify the later use of weak approximation.

Final steps

We now record the results on the arithmetic of intersections of two quadrics that we will need in our work.

Proposition 2.8. Let V⊂PNV \subset\mathbb{P}^{N} be a smooth complete intersection of two quadrics over a number field kk.

  1. If N≥7N \ge7, then VV satisfies the Hasse principle.

  2. If r≥0r \ge0, N≥2r+5N \ge2r+5, and Fr(V)(k)≠∅F_r(V)(k) \ne\varnothing, then Fr(V)F_r(V) satisfies weak approximation.

Proof. The first assertion for N=7N=7 is [11 Theorem 1]; the case N≥8N\ge8 is due to Colliot-Thélène–Sansuc–Swinnerton-Dyer [6 Theorem A(i)(a)]. The second is [15 Corollary 3.10]. For r=0r=0, it follows from [6 Theorem A(ii)]. □

We will use the following standard consequences of the implicit function theorem over local fields; see [22 Proposition 3.5.73(ii) and Remark 3.5.76].

Lemma 2.9. Let KK be a completion of a number field. If VV is a smooth geometrically integral KK-variety and V∘⊂VV^\circ\subset V is a nonempty Zariski open subset, then V∘(K)V^\circ(K) is dense in V(K)V(K). If f:V→Wf:V\to W is a morphism of smooth KK-varieties that is smooth at x∈V(K)x\in V(K), then the image under ff of every sufficiently small neighbourhood of xx contains a neighbourhood of f(x)f(x).

Let us return now to the setting of our smooth intersection of two quadrics X⊂Pkn−1X\subset\mathbb{P}^{n-1}_k over a number field kk. Recall the definition (5) of the incidence variety IrI_r. Armed with the above results, we now describe how to choose our base point aa.

Proposition 2.10. Suppose that r≥1r\ge1, n≥max⁡{8,2r+5}n\ge\max\{8,2r+5\} and Fr(X)(kv)≠∅F_r(X)(k_v)\ne\varnothing for every place vv of kk. Let UU be as in Lemma 2.3. There exists a∈U(k)a\in U(k) such that Fr−1(Ya)(kv)≠∅F_{r-1}(Y_a)(k_v)\ne\varnothing at every archimedean place vv. When r=1r=1, we may take YaY_a to have a point over every completion of kk. If also n=8n=8, we may take Δa(T)\Delta_a(T) to be irreducible over kk.

Proof. Each local rr-plane contains a local point. Thus XX is everywhere locally soluble, and Proposition 2.8 gives a kk-point and weak approximation on XX. When (r,n)=(1,8)(r,n)=(1,8), Lemma 2.2 allows us to shrink UU so that R0,aR_{0,a} is nonsingular throughout UU. We use the monic representative

det⁡(TR0,a−R1,a)det⁡R0,a\frac{\det(TR_{0,a}-R_{1,a})}{\det R_{0,a}}

of Δa(T)\Delta_a(T). Its coefficients are regular on UU, since a change of basis multiplies both determinants by the same factor.

Let S0S_0 be the finite set in Proposition 2.6, and put Ir,U=πr−1(U)I_{r,U}=\pi_r^{-1}(U). For v∈S0v\in S_0, a local rr-plane together with a point on it gives a point of Ir(kv)I_r(k_v). Since IrI_r is smooth and geometrically integral, Lemma 2.9 allows us to move this point into Ir,U(kv)I_{r,U}(k_v). The map πr\pi_r is smooth there, so

Av=πr(Ir,U(kv))A_v=\pi_r\bigl(I_{r,U}(k_v)\bigr)

is a nonempty open subset of X(kv)X(k_v).

When (r,n)=(1,8)(r,n)=(1,8) the generic polynomial is irreducible by Lemma 2.2. Hence the specialisations in U(k)U(k) for which it is reducible lie in a thin subset Z⊂X(k)Z\subset X(k) by [23 Proposition 3.3.5]. In the other cases put Z=∅Z=\varnothing. Since XX is projective, the closure of ZZ in ∏vX(kv)\prod_v X(k_v) has empty interior by [23 Corollary 3.5.4]. By weak approximation we can therefore choose a∈X(k)∖Za\in X(k)\setminus Z lying in AvA_v for every v∈S0v\in S_0. In particular, a∈U(k)a\in U(k) and the fibre Fr−1(Ya)F_{r-1}(Y_a) has a kvk_v-point for every v∈S0v\in S_0. When r=1r=1, Proposition 2.6 supplies a point at every remaining place. □

Proposition 2.11. For n≥11n\ge11, F1(X)F_1(X) satisfies the Hasse principle and weak approximation. Moreover, if n≥8n\ge8 and the hypothesis of Theorem 1.3 holds, then F1(X)F_1(X) satisfies the Hasse principle and weak approximation.

Proof. Suppose that F1(X)F_1(X) has a point over every completion of kk. By Proposition 2.10 with r=1r=1, there is a point a∈X(k)a \in X(k) for which Ya⊂Pkn−4Y_a \subset\mathbb{P}^{n-4}_k is a smooth, everywhere locally soluble intersection of two quadrics. If n≥11n \ge11, then n−4≥7n-4 \ge7, so Ya(k)≠∅Y_a(k) \ne\varnothing by Proposition 2.8(1). For 8≤n≤108 \le n \le10, assume the finiteness hypothesis of Theorem 1.3. If n=9n=9 or 1010, then YaY_a lies in P5\mathbb{P}^5 or P6\mathbb{P}^6 and [20 Theorem B] gives a kk-point. If n=8n=8, then YaY_a is a quartic del Pezzo surface in P4\mathbb{P}^4. In this case Proposition 2.10 allows us to choose aa so that Δa(T)\Delta_a(T) is irreducible, and [20 Theorem A] gives a kk-point. In each case a kk-point of YaY_a gives a kk-line on XX through aa. Weak approximation on F1(X)F_1(X) now follows from Proposition 2.8(2).

Proposition 2.11 proves Theorem 1.3. For n≥11n \ge11, it gives the Hasse principle and weak approximation for F1(X)F_1(X) unconditionally. The real place criterion follows from the next local result, which builds on work of Parimala–Suresh [21 Theorem 4.6] and Leep [18 Theorem 3.4] on the uu-invariant of a pp-adic function field.

Lemma 2.12. Let pp be a prime, let KK be a finite extension of Qp\mathbb{Q}_p, and let r≥1r \ge1. If n≥3r+8n \ge3r+8, then every smooth intersection of two quadrics in PKn−1\mathbb{P}^{n-1}_K contains an rr-dimensional projective linear subspace over KK.

Proof. This follows from [15 Remarks 3.9(3)], taking N=n−1N=n-1; the argument in the proof of [15 Corollary 3.5(2)] applies over every finite extension of Qp\mathbb{Q}_p.

Proof of Theorem 1.1 and Corollary 1.2. Since Theorem 1.1 is the case r=1r=1 of Corollary 1.2 and necessity is clear, we prove sufficiency by induction on r≥1r \ge1. Suppose that n≥3r+8n \ge3r+8 and Fr(X)(kv)≠∅F_r(X)(k_v) \ne\varnothing for every real place vv. Lemma 2.12 supplies rr-planes at every finite place, and rr-planes also exist over C\mathbb{C} by the geometric nonemptiness of Fr(X)F_r(X) in Lemma 2.3. Thus Fr(X)F_r(X) is everywhere locally soluble. When r=1r=1, Proposition 2.11 gives a rational line and weak approximation. When r≥2r \ge2, Proposition 2.10 gives a∈U(k)a \in U(k) such that the smooth complete intersection Ya⊂Pkn−4Y_a \subset\mathbb{P}^{n-4}_k in n−3≥3(r−1)+8n-3 \ge3(r-1)+8 variables satisfies Fr−1(Ya)(kv)≠∅F_{r-1}(Y_a)(k_v) \ne\varnothing for every real place vv. By the induction hypothesis, Fr−1(Ya)(k)≠∅F_{r-1}(Y_a)(k) \ne\varnothing, which gives Fr(X)(k)≠∅F_r(X)(k) \ne\varnothing and weak approximation by Proposition 2.8(2).

Cubic hypersurfaces

Lines through a point on a cubic hypersurface

Let KK be a field of characteristic zero and let X={C=0}⊂PKn−1X=\{C=0\}\subset\mathbb{P}^{n-1}_K be a smooth cubic hypersurface, with n≥5n \ge5. At a point a∈X(K)a \in X(K), choose coordinates in which

a=[1:0:⋯:0],TaX={x2=0}.a=[1:0:\cdots:0],\qquad T_aX=\{x_2=0\}.

Writing z=(x3,…,xn)z=(x_3,\ldots,x_n), we then have

C(x1,0,z)=x1q(z)+c(z),(6)C(x_1,0,z)=x_1q(z)+c(z), \tag*{(6)}

where qq and cc are quadratic and cubic forms, respectively. A line through aa is contained in XX precisely when its direction gives a point on

Ya={q=c=0}⊂Pn−3.Y_a=\{q=c=0\}\subset\mathbb{P}^{n-3}.

Lemma 3.1. There is a nonempty open subset X∘⊂XX^\circ\subset X defined over KK such that, for a∈X∘a \in X^\circ, the quadratic form qq in (6) is nonsingular and YaY_a is a smooth complete intersection.

Proof. Work over K‾\overline{K}. By [13 Chapter 1, Remark 5.20], for general aa the tangent hyperplane section X∩TaXX \cap T_aX has just one singular point, an ordinary double point at aa. Thus the quadratic part qq in (6) is nonsingular. Suppose that [z]∈Ya[z] \in Y_a is a point where the gradients of qq and cc are dependent. Since qq is nonsingular, we have ∇c(z)=λ∇q(z)\nabla c(\mathbf{z})=\lambda\nabla q(\mathbf{z}) for some λ\lambda. Then [−λ:0:z][-\lambda:0:\mathbf{z}] is another singular point of the tangent section. Indeed, the derivatives of its equation x1q(z)+c(z)x_1q(\mathbf{z})+c(\mathbf{z}) are q(z)q(\mathbf{z}) and x1∇q(z)+∇c(z)x_1\nabla q(\mathbf{z})+\nabla c(\mathbf{z}), and both vanish there. This proves that the gradients are independent at every common zero. Such zeros exist in Pn−3\mathbb{P}^{n-3} since n≥5n\geq5. The two conditions in the statement are open and defined over KK, which gives the required X∘X^\circ.

Smooth cubic sections of a quadric

The next lemma supplies the crucial nonsingularity needed to apply the ten variable theorem.

Lemma 3.2. Let FF be a field of characteristic zero, let d≥2d\geq2, and let q,cq,c be forms over FF in m≥2dm\geq2d variables, of degrees two and three, respectively. Suppose that qq is nonsingular and

Y={q=c=0}⊂Pm−1Y=\{q=c=0\}\subset\mathbb{P}^{m-1}

is a smooth complete intersection. If qq vanishes on a dd-dimensional FF-vector space, then such a space VV can be chosen so that c∣Vc|_V is a nonsingular cubic form in dd variables.

Proof. Choose a dd-dimensional FF-space V0V_0 on which qq vanishes. Put Q={q=0}Q=\{q=0\} and G=SO⁡(q)G=\operatorname{SO}(q). Over F‾\overline{F}, GG is connected and acts transitively on QQ. Both P(V0)\mathbb{P}(V_0) and YY are smooth and integral over F‾\overline{F}. By Kleiman’s theorem [16 Corollary 4(ii)], for general g∈Gg\in G the intersection P(gV0)∩Y\mathbb{P}(gV_0)\cap Y is smooth of dimension

(d−1)+(m−3)−(m−2)=d−2.(d-1)+(m-3)-(m-2)=d-2.

It is nonempty since a cubic on Pd−1\mathbb{P}^{d-1} has a geometric zero when d≥2d\geq2. Equivalently, the cubic x↦c(gx)\mathbf{x}\mapsto c(g\mathbf{x}) on V0V_0 has nonzero discriminant for general gg. In an FF-basis of V0V_0 this discriminant is a regular function of gg defined over FF. Its nonvanishing therefore gives a nonempty open subset of GG defined over FF. Since FF has characteristic zero, the set G(F)G(F) is Zariski dense in GG, as explained in [24 Corollary 13.3.10]. Thus we can choose g∈G(F)g\in G(F) in this open subset and take V=gV0V=gV_0.

Conclusion

We record the results that we shall need about the arithmetic of quadratic and cubic forms.

Proposition 3.3. Let kk be a number field.

  1. For every integer d≥1d\geq1, a quadratic form over kk in at least 2d+12d+1 variables vanishes on a dd-dimensional vector space over some quadratic extension L/kL/k, with LL totally imaginary.

  2. A nonsingular cubic form over kk in at least ten variables has a nontrivial zero over kk.

  3. A smooth cubic hypersurface of dimension at least two with a kk-point has Zariski dense kk-points.

Proof. The first assertion is [2 Theorem 3], and the second is [3 Theorem 1.1]. The third follows from unirationality [17 Theorem 1].

Proof of Theorem 1.4. Let X⊂Pkn−1X \subset\mathbb{P}^{n-1}_k be a smooth cubic hypersurface over a number field kk, with n≥23n \ge23. Proposition 3.3(2) gives X(k)≠∅X(k) \ne\varnothing, and part (3) allows us to choose a∈X∘(k)a \in X^\circ(k), with X∘X^\circ as in Lemma 3.1. Choose coordinates as in (6) over kk. We obtain a nonsingular quadratic form qq and a cubic form cc, both in n−2≥21n-2 \ge21 variables, such that

Ya={q=c=0}⊂Pn−3Y_a = \{q=c=0\} \subset\mathbb{P}^{n-3}

is a smooth complete intersection.

By Proposition 3.3(1), with d=10d=10, there is a quadratic extension L/kL/k over which qq vanishes on a 10-dimensional vector space VV. Lemma 3.2, applied over LL, shows that this space can be chosen so that c∣Vc|_V is nonsingular. Applying Proposition 3.3(2) over LL gives 0≠z∈V0 \ne z \in V such that q(z)=c(z)=0q(z)=c(z)=0. The line

ℓ={[s:0:tz]:[s:t]∈PL1}\ell= \{[s:0:tz]:[s:t]\in\mathbb{P}^{1}_{L}\}

is thus contained in XLX_L, by (6), and passes through a=[1:0:⋯:0]∈X(k)a=[1:0:\cdots:0]\in X(k). Since ℓ\ell passes through the kk-point aa, Wooley’s descent lemma [25 Lemma 2.2] supplies a kk-line on XX. □

References

ISTA, Am Campus 1, 3400 Klosterneuburg, Austria Email address: [email protected]

Department of Mathematics, Imperial College London, SW7 2AZ, United Kingdom Email address: [email protected]

Department of Mathematics, University College London, Gower Street, London WC1E 6BT, United Kingdom Email address: [email protected]

Max Planck Institute for Mathematics, Vivatsgasse 7, 53111 Bonn, Germany Email address: [email protected]

References

  1. [1]J. Brandes and R. Dietmann, Rational lines on cubic hypersurfaces. Math. Proc. Cambridge Philos. Soc. 171 (2021), 99–112.DOI
  2. [2]J. Brandes, R. Dietmann and D. B. Leep, Rational lines on cubic hypersurfaces II. Mathematika, to appear.
  3. [3]T. D. Browning and P. Vishe, Cubic hypersurfaces and a version of the circle method for number fields. Duke Math. J. 163 (2014), 1825–1883.
  4. [4]J.-L. Colliot-Thélène, Points rationnels sur les fibrations. Higher dimensional varieties and rational points (Budapest, 2001), 171–221, Springer-Verlag, 2003.
  5. [5]J.-L. Colliot-Thélène, Retour sur l’arithmétique des intersections de deux quadriques. Avec un appendice par A. Kuznetsov. J. reine angew. Math. 806 (2024), 147–185.
  6. [6]J.-L. Colliot-Thélène, J.-J. Sansuc and P. Swinnerton-Dyer, Intersections of two quadrics and Châtelet surfaces. I. J. reine angew. Math. 373 (1987), 37–107.
  7. [7]J.-L. Colliot-Thélène, J.-J. Sansuc and P. Swinnerton-Dyer, Intersections of two quadrics and Châtelet surfaces. II. J. reine angew. Math. 374 (1987), 72–168.
  8. [8]O. Debarre and L. Manivel, Sur la variété des espaces linéaires contenus dans une intersection complète. Math. Ann. 312 (1998), 549–574.DOI
  9. [9]R. Dietmann and T. D. Wooley, Pairs of cubic forms in many variables. Acta Arith. 110 (2003), 125–140.DOI
  10. [10]S. Frei and L. Ji, A threefold violating a local-to-global principle for rationality. Res. Number Theory 10 (2024), Paper No. 39.DOI
  11. [11]D. R. Heath-Brown, Zeros of pairs of quadratic forms. J. reine angew. Math. 739 (2018), 41–80.
  12. [12]D. R. Heath-Brown and L. B. Pierce, Simultaneous integer values of pairs of quadratic forms. J. reine angew. Math. 727 (2017), 85–143.
  13. [13]D. Huybrechts, The geometry of cubic hypersurfaces. Cambridge Studies in Advanced Mathematics 206, Cambridge University Press, 2023.DOI
  14. [14]Y. Jeong, Moduli map of second fundamental forms on a nonsingular intersection of two quadrics. Math. Ann. 372 (2018), 1–54.DOI
  15. [15]L. Ji and F. Suzuki, Arithmetic and birational properties of linear spaces on intersections of two quadrics. J. Éc. polytech. Math. 12 (2025), 1161–1196.arxiv.org/abs/2402.18857
  16. [16]S. L. Kleiman, The transversality of a general translate. Compositio Math. 28 (1974), 287–297.
  17. [17]J. Kollár, Unirationality of cubic hypersurfaces. J. Inst. Math. Jussieu 1 (2002), 467–476.DOI
  18. [18]D. B. Leep, The u-invariant of p-adic function fields. J. reine angew. Math. 679 (2013), 65–73.DOI
  19. [19]A. Molyakov, Le principe de Hasse pour les intersections de deux quadriques dans P⁷. Math. Z. 310 (2025), Paper No. 32, 24 pp.arxiv.org/abs/2305.00313
  20. [20]A. Morgan and A. N. Skorobogatov, Hasse principle for intersections of two quadrics via Kummer surfaces. J. Algebraic Geom., to appear.
  21. [21]R. Parimala and V. Suresh, The u-invariant of the function fields of p-adic curves. Annals of Math. 172 (2010), 1391–1405.arxiv.org/abs/0708.3128
  22. [22]B. Poonen, Rational points on varieties. Graduate Studies in Mathematics 186, American Mathematical Society, 2017.DOI
  23. [23]J.-P. Serre, Topics in Galois theory. Research Notes in Mathematics 1, Jones and Bartlett Publishers, Boston, MA, 1992.
  24. [24]T. A. Springer, Linear Algebraic Groups. Second edition, Progress in Mathematics 9, Birkhäuser, 1998.
  25. [25]T. D. Wooley, Linear spaces on cubic hypersurfaces, and pairs of homogeneous cubic equations. Bull. London Math. Soc. 29 (1997), 556–562.DOI

Paper details

vlnc:W_yg78vj3f
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 5

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT 6.0, Gemini 4

View statement
Looking for related papers…
Contents