betaFeedback

WorksMathematics

A counterexample to a stability question for unramified multiplicity one symmetric pairs

Abstract

We construct a split reductive symmetric pair over Q17\mathbb{Q}_{17} with multiplicity at most one for every irreducible unramified representation and which is not a stable symmetric pair.

Introduction

Let kk be a pp-adic local field of characteristic zero with ring of integers o\mathfrak{o}. For a symmetric pair (G,H,θ)(G,H,\theta), fix a hyperspecial compact subgroup K⊂G(k)K \subset G(k). An irreducible smooth complex representation π\pi is unramified with respect to KK if πK≠0\pi^K \ne0. We say that the pair has unramified multiplicity one if

dim⁡Hom⁡G(k)(π,C∞(H(k)\G(k)))≤1\dim\operatorname{Hom}_{G(k)}\left(\pi,C^\infty(H(k)\backslash G(k))\right) \le1

for every such π\pi. Here G(k)G(k) acts by right translation.

Aizenbud and Gourevitch call a symmetric pair satisfying Definition 3.2 a good pair [1 Definition 7.1.6]. We use Carmeli’s term stable for the same property [2]. We consider the following question.

Does unramified multiplicity one imply stability of a symmetric pair? $$ {#eq:1.1}

We take

k=Q17,(n1,…,n6)=(4,4,8,8,4,8),hj=nj/2.k = \mathbb{Q}_{17}, \qquad(n_1,\ldots,n_6) = (4,4,8,8,4,8), \qquad h_j = n_j/2.

Let T=Gm6T = \mathbb{G}_m^6, with coordinates t=(s1,t1,s2,t2,u,v)t=(s_1,t_1,s_2,t_2,u,v), and put

τ(t)=(t1,s1,t2,s2,u,v).\tau(t)=(t_1,s_1,t_2,s_2,u,v).

Write pi:T→Gmp_i:T\to\mathbb{G}_m for the ii-th coordinate projection. Define χj=cjhj\chi_j=c_j^{h_j} using the characters

c1=p1p2p3p6,c2=p3,(1)c_1=p_1p_2p_3p_6,\qquad c_2=p_3, \tag*{(1)}
c3=p1p6,c4=p1,(2)c_3=p_1p_6,\qquad c_4=p_1, \tag*{(2)}
c5=p1p3p5p6,c6=p1p3p5.(3)c_5=p_1p_3p_5p_6,\qquad c_6=p_1p_3p_5. \tag*{(3)}

We introduce the group GG as

G={((Aj,Bj)j,t)∈∏j=16GLnj×T | det⁡Aj=χj(t), det⁡Bj=χj(τt) (1≤j≤6)}.(4)G=\left\{((A_j,B_j)_j,t)\in\prod_{j=1}^{6}\mathrm{GL}_{n_j} \times T \,\middle|\, \det A_j=\chi_j(t),\ \det B_j=\chi_j(\tau t)\ (1\le j\le6)\right\}. \tag*{(4)}

The involution

θ((Aj,Bj)j,t)=((Bj,Aj)j,τt)(5)\theta((A_j,B_j)_j,t)=((B_j,A_j)_j,\tau t) \tag*{(5)}

has fixed subgroup H=GθH=G^\theta.

Theorem 1.1. The symmetric pair defined by (1.5)–(1.6) has unramified multiplicity one (Proposition 2.3) and fails stability (Proposition 3.6).

Unramified multiplicity one

Write S=Tτ={(a,a,b,b,u,v)}S=T^\tau=\{(a,a,b,b,u,v)\} and let p:G→Tp:G\to T be the torus projection. For a unit xx in a coefficient ring set δn(x)=diag⁡(x,1,…,1)\delta_n(x)=\operatorname{diag}(x,1,\ldots,1). The homomorphism

ι(t)=((δnj(χj(t)),δnj(χj(τt)))j,t)\iota(t)=\left((\delta_{n_j}(\chi_j(t)),\delta_{n_j}(\chi_j(\tau t)))_j,t\right)

splits pp over Z17\mathbb{Z}_{17} and restricts to a section S→HS\to H. The kernels of G→TG\to T and H→SH\to S are respectively ∏jSL⁡nj\prod_j \operatorname{SL}_{n_j} and its blockwise diagonal copy ∏jSL⁡nj\prod_j \operatorname{SL}_{n_j}. These descriptions give smooth group schemes with connected reductive geometric fibers. Indeed, the semidirect product descriptions give smoothness and connectedness, and in each fiber a connected normal unipotent subgroup maps trivially to the torus and hence lies in the semisimple kernel. The diagonal matrices give a split maximal torus in each group, since in each matrix factor the diagonal torus of SL⁡nj\operatorname{SL}_{n_j} already has diagonal centralizer. In particular GG is split reductive and K=G(Z17)K=G(\mathbb{Z}_{17}) is hyperspecial.

For context we recall the general criterion of Sakellaridis. If XX is a homogeneous affine spherical variety for a connected split reductive group LL, write A∗A^* for the unramified character torus of a maximal split torus of LL, and use AX∗A_X^*, δ(X)\delta(X), WW, and WXW_X for the corresponding objects of [4 §8]. The restriction map in that criterion is

C[A∗]W⟶C[δ(X)1/2AX∗]WX.\mathbb{C}[A^*]^W\longrightarrow\mathbb{C}[\delta(X)^{1/2}A_X^*]^{W_X}.

Theorem 2.1 ([4 Corollary 8.0.4]). Assume the standing hypotheses of [4 §§1.1, 1.7, 8], including its integral orbit axioms and the single rational open Borel orbit hypothesis, and assume Statements 6.3.1 and 7.1.5 of loc. cit. for the spherical roots and virtual weighted colors. If the restriction map is surjective, then every irreducible unramified representation π\pi of L(k)L(k) satisfies

dim⁡Hom⁡L(k)(π,C∞(X(k)))≤1.\dim\operatorname{Hom}_{L(k)}\left(\pi,C^\infty(X(k))\right)\le1.

Proof. Let KLK_L be the hyperspecial subgroup and HH its spherical Hecke algebra. By [4 Corollary 8.0.4(1)], C∞(X(k))KLC^\infty(X(k))^{K_L} is cyclic over HH. Pair it with C∞(X(k))KLC^\infty(X(k))^{K_L} by integration against an invariant measure. This pairing separates functions, since the compact open KLK_L-orbits have positive volume. If ff is a simultaneous eigenfunction with character ω\omega, its pairing with hh applied to a cyclic generator is its pairing with the generator multiplied by ω(h∨)\omega(h^\vee), where h∨(g)=h(g−1)h^\vee(g)=h(g^{-1}). Thus each eigenspace has dimension at most one. An L(k)L(k)-equivariant map out of π\pi is determined by the image of a nonzero spherical vector, which spans πKL\pi^{K_L} and must map into the eigenspace for its Hecke character. □

For the pair (1.5)–(1.6) we give a direct proof. It rests on a choice of double-coset representatives for which averaging a spherical vector determines every H(k)H(k)-invariant functional from its value on that vector.

Lemma 2.2. Every double coset in H(k)\G(k)/KH(k)\backslash G(k)/K has a representative g0g_0 such that Kg0K⊂H(k)g0KKg_0K\subset H(k)g_0K.

Proof. Put ϖ=17\varpi=17 and let ν\nu be the valuation with ν(ϖ)=1\nu(\varpi)=1. For t=(s1,t1,s2,t2,u,v)t=(s_1,t_1,s_2,t_2,u,v) the class of tt modulo S(k)T(o)S(k)T(\mathfrak{o}) is determined by the two integers x=ν(t1/s1)x=\nu(t_1/s_1) and y=ν(t2/s2)y=\nu(t_2/s_2). Indeed, division by an element of S(k)S(k) allows us to set s1,s2,u,vs_1,s_2,u,v equal to 11, and T(o)T(\mathfrak{o}) removes the remaining unit factors. Thus, using ι(S(k))\iota(S(k)) on the left and ι(T(o))\iota(T(\mathfrak{o})) on the right, we may take the torus component to be

t=(1,ϖx,1,ϖy,1,ϖe),e={1x odd and y even,0otherwise.t=(1,\varpi^x,1,\varpi^y,1,\varpi^e),\qquad e=\begin{cases}1 & x\text{ odd and }y\text{ even},\\0 & \text{otherwise}.\end{cases}

The last coordinate may be chosen freely because SS contains the last coordinate subtorus. The valuations of the six characters in (1.2)–(1.4) are

(νcj(t))j=(x+e,0,e,0,e,0),(\nu_{c_j}(t))_j=(x+e,0,e,0,e,0),
(νcj(τt))j=(x+y+e,y,x+e,x,x+y+e,x+y).(\nu_{c_j}(\tau t))_j=(x+y+e,y,x+e,x,x+y+e,x+y).

In each column at least one entry is even. For xx even this holds throughout the first row. For x,yx,y both odd only its first entry is odd, and x+yx+y is even. In the remaining case e=1e=1, the only odd entries of the first row are in columns 3 and 5, where the second row has even entries.

Write the matrix components as (Aj,Bj)(A_j,B_j). Since tt has pure uniformizer powers as coordinates and χj=cjnj/2\chi_j=c_j^{n_j/2}, at least one matrix in each pair has determinant λjnj\lambda_j^{n_j} for some λj∈k×\lambda_j\in k^\times. If that matrix is MjM_j, multiplication on the left by λjMj−1\lambda_jM_j^{-1} on both blocks is allowed in the diagonal SLnj(k)\mathrm{SL}_{n_j}(k) and makes MjM_j scalar. We can also make the other block diagonal. Indeed, elementary divisors over o\mathfrak{o} give P,Q∈GLnj(o)P,Q\in\mathrm{GL}_{n_j}(\mathfrak{o}) with PMQPMQ diagonal for the other block MM. Multiplying PP on the left and QQ on the right by diagonal unit matrices makes both determinants one and preserves diagonality. Apply PP on the left of both blocks, then P−1P^{-1} on the right of the scalar block and QQ on the right of the other. These operations belong to H(k)H(k) and KK, respectively. Denote the resulting representative by g0g_0.

The subgroup ι(T(o))\iota(T(\mathfrak{o})) commutes with g0g_0, since all the matrices are diagonal. As K=(∏jSLnj(o)2)ι(T(o))K=(\prod_j\mathrm{SL}_{n_j}(\mathfrak{o})^2)\iota(T(\mathfrak{o})), it remains to move a pair (U,V)∈SLnj(o)2(U,V)\in\mathrm{SL}_{n_j}(\mathfrak{o})^2 past the jjth block (A,B)(A,B). If AA is scalar, then

(UA,VB)=(V,V)(A,B)(V−1U,I).(UA,VB)=(V,V)(A,B)(V^{-1}U,I).

If BB is scalar, then

(UA,VB)=(U,U)(A,B)(I,U−1V).(UA,VB)=(U,U)(A,B)(I,U^{-1}V).

In each identity the left factor belongs to the diagonal subgroup and the right factor is integral of determinant one. Taking the product over jj proves the inclusion. □

Proposition 2.3. The symmetric pair (G,H)(G,H) defined by (1.5)–(1.6) has unramified multiplicity one with respect to K=G(Z17)K=G(\mathbb{Z}_{17}).

Proof. Let π\pi be irreducible and KK-spherical. By the spherical representation theorem, πK=Cw\pi^K=\mathbb{C}w for any nonzero spherical vector ww [3 §6]. Fix ℓ∈Hom⁡H(k)(π,C)\ell\in\operatorname{Hom}_{H(k)}(\pi,\mathbb{C}). For g0g_0 as in Lemma 2.2, the function g↦ℓ(π(g)w)g\mapsto\ell(\pi(g)w) is constant on Kg0KKg_0K, since it is left H(k)H(k)-invariant and right KK-invariant. Give KK volume one. Then

ℓ(π(g0)w)=ℓ(∫Kπ(k′g0)w dk′)=ag0ℓ(w),\ell(\pi(g_0)w)=\ell\left(\int_K\pi(k'g_0)w\,\mathrm{d}k'\right)=a_{g_0}\ell(w),

where the averaged vector lies in πK\pi^K and equals ag0wa_{g_0}w. The integral is a finite sum after passage to a sufficiently small open subgroup of KK. If ℓ(w)=0\ell(w)=0, the matrix coefficient therefore vanishes on every H(k)g0KH(k)g_0K. The G(k)G(k)-translates of ww span π\pi, so ℓ=0\ell=0. Evaluation at ww thus embeds Hom⁡H(k)(π,C)\operatorname{Hom}_{H(k)}(\pi,\mathbb{C}) in C\mathbb{C}. Finally, evaluation of a G(k)G(k)-map at the identity coset identifies the Hom space in the introduction with Hom⁡H(k)(π,C)\operatorname{Hom}_{H(k)}(\pi,\mathbb{C}). The inverse sends ℓ\ell to v↦[H(k)g↦ℓ(π(g)v)]v\mapsto[H(k)g\mapsto\ell(\pi(g)v)]. □

Verification of Sakellaridis’s criterion

Proposition 2.4. For k=Q17k=\mathbb{Q}_{17} and the integral model of X=H\GX=H\backslash G over o=Z17\mathfrak{o}=\mathbb{Z}_{17}, all the hypotheses of Theorem 2.1 hold. In particular δ(X)=1\delta_{(X)}=1, and the restriction

C[A∗]W⟶C[AX∗]WX\mathbb{C}[A^*]^W\longrightarrow\mathbb{C}[A_X^*]^{W_X}

is surjective. Consequently Theorem 2.1 also gives Proposition 2.3.

Proof. The open orbit and its integral model. Take the Borel BB lower triangular in the first matrix of each pair and upper triangular in the second. Write AA for its diagonal torus. Identify Q=T/SQ=T/S with Gm2G_m^2 by z(t)=(t1/s1,t2/s2)z(t)=(t_1/s_1,t_2/s_2). Define six characters of QQ by

(ηj(z))j=(z2,z2,z1,z1,z1z2,z1z2).(\eta_j(z))_j=(z_2,z_2,z_1,z_1,z_1z_2,z_1z_2).

Indeed, cj(τt)/cj(t)=ηj(z(t))c_j(\tau t)/c_j(t)=\eta_j(z(t)). Over o\mathfrak{o} the map g↦((Aj−1Bj)j,z(t))g\mapsto((A_j^{-1}B_j)_j,z(t)) identifies XX with

{((Yj)j,z)∈∏jGLnj×Q | det⁡Yj=ηj(z)hj}.(6)\left\{((Y_j)_j,z)\in\prod_j\mathrm{GL}_{n_j}\times Q\ \middle|\ \det Y_j=\eta_j(z)^{h_j}\right\}. \tag*{(6)}

To see this over any o\mathfrak{o}-algebra, lift zz to t=(1,z1,1,z2,1,1)t=(1,z_1,1,z_2,1,1), set Aj=δnj(χj(t))A_j=\delta_{n_j}(\chi_j(t)), and take Bj=AjYjB_j=A_jY_j. This gives a section, and two lifts differ by a unique left multiplier in HH. In particular X(k)=H(k)\G(k)X(k)=H(k)\backslash G(k). The model is smooth and affine, since multiplying YjY_j on the left by δnj(ηj(z)−hj)\delta_{n_j}(\eta_j(z)^{-h_j}) identifies it as a scheme with Q×∏jSLnjQ\times\prod_j\mathrm{SL}_{n_j}. The derived group of GG is ∏jSLnj2\prod_j\mathrm{SL}_{n_j}^2 and is simply connected.

The open orbit through x0=Hx_0=H is given by nonvanishing of all proper leading principal minors of the YjY_j. Gaussian elimination factors each matrix there as lower unitriangular, diagonal, and upper unitriangular. The diagonal locus is AX=A/(A∩H)A_X=A/(A\cap H): the section just constructed takes diagonal tuples to AA. Thus the open orbit is a product of AXA_X with the two families of unitriangular groups over o\mathfrak{o}. It is smooth, has homogeneous special fiber, and its kk-points are one B(k)B(k)-orbit. The group GG and the stabilizer HH are unimodular, so X(k)X(k) has an invariant measure. The open stabilizer B∩H=A∩HB\cap H=A\cap H is a split torus, and δB\delta_B is trivial on it because the two triangular factors have opposite roots. This gives a B(k)B(k)-invariant measure on the open orbit. Every simple minimal parabolic enlarges that orbit, so P(X)=BP(X)=B, its Levi is AA, and δ(X)=1\delta_{(X)}=1.

Bruhat strata and Axiom 2.2.1. In (2.1) the BB-strata are indexed by tuples w=(wj)∈∏jSnjw=(w_j)\in\prod_j S_{n_j}: the jjth matrix lies in Bnj−w˙jBnj+B^-_{n_j}\dot{w}_jB^+_{n_j}. Here Bn±B^\pm_n are the triangular Borels of GLn\mathrm{GL}_n, and we take signed permutation representatives of determinant one. The torus QQ is free to vary. Indeed, the diagonal factor used to trivialize (2.1) preserves these cells, the section ι\iota moves between fibers, and the determinant-one triangular groups act transitively on each cell with fixed determinant. The integral cells are products of split tori and affine spaces by Bruhat elimination. Their frontiers are the usual unions of Bruhat cells, defined by the same northwest submatrix rank conditions over o\mathfrak{o} and in both fibers. In particular the boundaries, also for a union of two cells related by a minimal parabolic, acquire no extra special-fiber stratum: every boundary cell is flat over o\mathfrak{o} and its generic fiber is dense in it.

To identify the rank-one quotient, take the representative ((w˙j)j,1)((\dot{w}_j)_j,1) of a stratum YY and enlarge one triangular factor to PαP_\alpha. The stabilizer projected to Pα/rad⁡(Pα)≃PGL2P_\alpha/\operatorname{rad}(P_\alpha)\simeq\mathrm{PGL}_2 preserves one coordinate line: the other flag in the same block is a full coordinate flag in permutation order, and it singles out one of the two lines in the affected two-dimensional flag quotient. The image is its entire Borel. It contains the corresponding elementary unipotent matrices, acting on both flags with the permutation conjugacy, and the diagonal SL2\mathrm{SL}_2-torus maps onto the adjoint torus as a group scheme for the faithfully flat topology. These matrix descriptions work over o\mathfrak{o}. Thus YPα/rad⁡(Pα)YP_\alpha/\operatorname{rad}(P_\alpha) is a split Borel quotient of PGL2\mathrm{PGL}_2, of type U in [4 §2.2]. Together with the boundary description this gives Axiom 2.2.1 and excludes all type-N edges.

The Cartan and Iwahori axioms. Let Λ=X∗(A)\Lambda=X_*(A) and ΛX=X∗(AX)\Lambda_X=X_*(A_X). Write ℓj(ξ)=⟨cj,ξ⟩\ell_j(\xi)=\langle c_j,\xi\rangle for ξ∈X∗(T)\xi\in X_*(T). An element of Λ\Lambda consists of an exponent vector ξ\xi and two integer vectors aj,bja_j,b_j of length njn_j in each block with

∑iaji=hjℓj(ξ),∑ibji=hjℓj(τξ).\sum_i a_{ji}=h_j\ell_j(\xi),\qquad\sum_i b_{ji}=h_j\ell_j(\tau\xi).

An element of ΛX\Lambda_X consists of q∈X∗(Q)q \in X_*(\mathbb{Q}) and integer vectors mjm_j with ∑imji=hj⟨ηj,q⟩\sum_i m_{ji}=h_j\langle\eta_j,q\rangle. The projection of lattices is

pr⁡X(a,b,ξ)=((bj−aj)j,z∗(ξ)),(7)\operatorname{pr}_X(a,b,\xi)=((b_j-a_j)_j,z_*(\xi)), \tag*{(7)}

where z∗(ξ)z_*(\xi) denotes the image of ξ\xi in X∗(Q)X_*(\mathbb{Q}). The antidominant cone consists of the mj1≤⋯≤mjnjm_{j1}\leq\cdots\leq m_{jn_j} in each block. Indeed for a symmetric variety it is the image of the ambient antidominant cone over Q\mathbb{Q} [4 §2.1]. Here aja_j is decreasing and bjb_j increasing in that cone. Conversely any increasing mjm_j lifts rationally by taking aja_j constant with its prescribed sum and bj=aj+mjb_j=a_j+m_j.

The invariants of a KK-orbit are the two valuations of zz and the ordered elementary-divisor exponents of every YjY_j. They determine the orbit. First normalize zz to pure powers of 1717 using ι(T(o))\iota(T(\mathfrak{o})). Smith normal form in block jj gives integral invertible P,RP,R with PYjR=Dj=diag⁡(17mji)iPY_jR=D_j=\operatorname{diag}(17^{m_{ji}})_i. At the normalized zz the determinants of YjY_j and DjD_j agree. Thus det⁡Pdet⁡R=1\det P\det R=1, and replacing (P,R)(P,R) by (EP,RE−1)(EP,RE^{-1}) for a diagonal unit matrix EE makes both determinants one while fixing DjD_j. These left and right operations on YjY_j are induced by the two SL⁡nj(o)\operatorname{SL}_{n_j}(\mathfrak{o}) factors of KK. Conversely the invariants are unchanged under KK. Hence X(k)/KX(k)/K is precisely the set of antidominant integral parameters just described (Axiom 2.4.1).

For Axiom 2.4.2 take any diagonal point in AX+A_X^+, with block D=diag⁡(d1,…,dn)D=\operatorname{diag}(d_1,\ldots,d_n), so that ν(di)≤ν(di+1)\nu(d_i)\leq\nu(d_{i+1}). For an element of the Iwahori JJ write (C,E)(C,E) for its blocks. They are respectively lower and upper triangular modulo 1717. Gaussian elimination over o\mathfrak{o} gives C−1=LUC^{-1}=LU with LL integral lower triangular invertible and UU upper unitriangular congruent to II modulo 1717. Then (D−1UD)E(D^{-1}UD)E is integral and upper triangular modulo 1717, so it has a factorization L′RL'R with L′L' lower unitriangular and RR upper triangular integral invertible. Now

C−1DE=L(DL′D−1)DR.C^{-1}DE=L(DL'D^{-1})DR.

The left factor is integral lower triangular invertible by the order of the valuations. Its determinant is (det⁡C)−1(\det C)^{-1}, and det⁡R=det⁡E\det R=\det E. Taking its inverse as the first block, RR as the second block, and the original torus component in T(o)T(\mathfrak{o}) produces an element of B(o)B(\mathfrak{o}) with exactly the same action on the given point. Thus xJ⊂xB(o)xJ\subset xB(\mathfrak{o}) for every x∈AX+x\in A_X^+.

Spherical roots and virtual weighted colors. On the Bruhat indices the ambient Weyl group acts separately by left and right permutations, since each simple step exchanges adjacent positions or adjacent values. The open index is the identity tuple. Its stabilizer is the blockwise diagonal WX=∏jSnjW_X=\prod_j S_{n_j}, using P(X)=BP(X)=B and [4 §2.2]. The simple reflection for positions i,i+1i,i+1 in block jj is the product of the corresponding left and right simple reflections. Their length-two path passes through the same codimension-one cell reached from the open orbit on either side. Thus the normalized spherical root is the sum of the two corresponding simple roots, of type G in the notation of [4 §6.2]. Its supporting Levi has derived group SL⁡2×SL⁡2\operatorname{SL}_2\times\operatorname{SL}_2. Thus [4 Theorem 6.3.2] applies. In this case the rank-one computation is exactly the two-sided group computation in [4 Example 5.4.1]: the adjacent root subgroups act by left and right multiplication on the same two-dimensional matrix block. This verifies Statement 6.3.1.

For clarity, the color in this path is the divisor of the leading ii-rowed principal minor Δji\Delta_{ji} in block jj. Its valuation on AXA_X is the cocharacter βji=eji−ej,i+1\beta_{ji}=e_{ji}-e_{j,i+1}, with zero torus component, where ejie_{ji} denotes a standard coordinate vector in the jjth block. Indeed the other leading minors and the QQ-coordinates are units at its generic point, and Δji\Delta_{ji} vanishes there to order one. For the order of vanishing, take z=(1,1)z=(1,1) and the identity matrices with one two-by-two block replaced by (s−1110)\begin{pmatrix}s^{-1}&1\\1&0\end{pmatrix} at positions i,i+1i,i+1, where Δji=s\Delta_{ji}=s and all the other leading minors equal one. On the diagonal locus successive quotients of the leading minors are the diagonal coordinates. These evaluations give βji\beta_{ji}, and the Bruhat cells of codimension one exhaust the colors. Each color is defined over kk, belongs to its single type-G root and has weight ⟨βji,ρP(X)⟩=1\langle\beta_{ji},\rho_{P(X)}\rangle=1, as seen by lifting βji\beta_{ji} to the right simple coroot. The definition in [4 §7.1] therefore gives, with multiplicity one,

Θ={(eji−ejl+1,1):1≤j≤6, 1≤i,l≤nj, i≠l}.\Theta=\{(e_{ji}-e_{jl+1},1):1\leq j\leq6,\ 1\leq i,l\leq n_j,\ i\ne l\}.

The cone of color valuations is generated by the adjacent differences βji\beta_{ji}, so the positive members have i<li<l. Each simple transposition sends exactly its own positive simple coroot to a negative one. This proves both assertions of Statement 7.1.5.

Surjectivity on invariants. The coordinate rings of A∗A^* and AX∗A_X^* are the group algebras C[Λ]\mathbb{C}[\Lambda] and C[ΛX]\mathbb{C}[\Lambda_X]. Restriction sends eλe^\lambda to epr⁡X(λ)e^{\operatorname{pr}_X(\lambda)} by (7). Fix μ=((mj)j,(x,y))∈ΛX\mu=((m_j)_j,(x,y))\in\Lambda_X. Lift (x,y)(x,y) to ξ=(0,x,0,y,0,e)\xi=(0,x,0,y,0,e) with ee chosen as in Lemma 2.2. The two valuation rows in its proof say, for each jj, that either ℓj(ξ)\ell_j(\xi) or ℓj(τξ)\ell_j(\tau\xi) is even. In the first case choose the constant vector aj=(ℓj(ξ)/2)(1,…,1)a_j=(\ell_j(\xi)/2)(1,\ldots,1) and put bj=aj+mjb_j=a_j+m_j. In the other case choose bj=(ℓj(τξ)/2)(1,…,1)b_j=(\ell_j(\tau\xi)/2)(1,\ldots,1) and put aj=bj−mja_j=b_j-m_j. In each case the constant vector has length njn_j. Since nj=2hjn_j=2h_j and the sum of mjm_j is prescribed, both determinant equations hold. These choices give a lift λ∈Λ\lambda\in\Lambda of μ\mu with a constant vector on at least one side of every block.

The two symmetric groups of each ambient block permute aj,bja_j,b_j independently, whereas WXW_X permutes the entries of mjm_j. With one vector constant, subtraction gives a bijection of the distinct ambient and relative orbit points in that block, including when entries repeat. The Weyl groups fix the torus coordinates. Taking products yields

∑λ′∈Wλeλ′∣AX∗=∑μ′∈WXμeμ′.\sum_{\lambda'\in W\lambda}e^{\lambda'}\bigg|_{A_X^*}=\sum_{\mu'\in W_X\mu}e^{\mu'}.

The sums on the right form a basis of C[ΛX]WX\mathbb{C}[\Lambda_X]^{W_X}. This proves surjectivity and completes the verification of the hypotheses, independently of the multiplicity conclusion. □\square

Stable symmetric pairs

Symmetric pairs

Definition 3.1. A symmetric pair is a triple (G,H,θ)(G,H,\theta), where GG and H⊂GH\subset G are reductive groups, θ\theta is an involution of GG, and H=GθH=G^\theta. We call the pair connected if H\GH\backslash G is connected as an algebraic variety.

For a symmetric pair (G,H,θ)(G,H,\theta), define the anti-involution σ(g)=θ(g−1)\sigma(g)=\theta(g^{-1}) and the symmetrization map

s:G⟶Gσ,s(g)=gσ(g).s:G\longrightarrow G^\sigma,\qquad s(g)=g\sigma(g).

For g=Lie⁡(G)\mathfrak{g}=\operatorname{Lie}(G), we write gσ=ker⁡(dθ+1)\mathfrak{g}^\sigma=\ker(d\theta+1), since dσ=−dθd\sigma=-d\theta. The adjoint action of HH preserves this space.

Definition 3.2. The pair (G,H,θ)(G,H,\theta) is stable if σ(H(k)gH(k))=H(k)gH(k)\sigma(H(k)gH(k))=H(k)gH(k) for every g∈G(k)g\in G(k) whose double coset is closed in the kk-adic topology.

An unstable symmetric pair

In this section, we let

k=Q17,(n1,…,n6)=(4,4,8,8,4,8),(d1,…,d6)=(3,3,17,17,51,51).k=\mathbb{Q}_{17},\qquad(n_1,\ldots,n_6)=(4,4,8,8,4,8),\qquad(d_1,\ldots,d_6)=(3,3,17,17,51,51).

Choose roots of unity i,ζ∈ki,\zeta\in k characterized by

i2=−1,i≡4(mod17),i^2=-1,\qquad i\equiv4\pmod{17},

and

ζ4=−1,ζ≡2(mod17).\zeta^4=-1,\qquad\zeta\equiv2\pmod{17}.

These roots exist uniquely by Hensel’s lemma, and ζ2=i\zeta^2=i. Write

ζ4=i,ζ8=ζ.\zeta_4=i,\qquad\zeta_8=\zeta.

For N∈{4,8}N \in\{4,8\} and d∈{3,17,51}d \in\{3,17,51\}, let

R(N,d)=(00⋯0d10⋯0001⋯00⋮⋱⋱⋮⋮00⋯10).R(N,d)=\begin{pmatrix} 0 & 0 & \cdots& 0 & d\\ 1 & 0 & \cdots& 0 & 0\\ 0 & 1 & \cdots& 0 & 0\\ \vdots& \ddots& \ddots& \vdots& \vdots\\ 0 & 0 & \cdots& 1 & 0 \end{pmatrix}.

In the basis 1,ρ,…,ρN−11,\rho,\ldots,\rho^{N-1} of k[X]/(XN−d)k[X]/(X^N-d), with ρ\rho the class of XX, this is multiplication by ρ\rho. Its characteristic polynomial is XN−dX^N-d, and R(N,d)N=dINR(N,d)^N=dI_N.

Define

Z(N,d)=R(N,d)N/2(IN+R(N,d))(IN+ζNR(N,d))−1.Z(N,d)=R(N,d)^{N/2}(I_N+R(N,d))(I_N+\zeta_N R(N,d))^{-1}.

Writing R=R(N,d)R=R(N,d), its characteristic polynomial gives det⁡R=−d\det R=-d and det⁡(IN+aR)=1−aNd\det(I_N+aR)=1-a^N d. In particular both det⁡(IN+R)\det(I_N+R) and det⁡(IN+ζNR)\det(I_N+\zeta_N R) equal 1−d≠01-d\ne0. Thus Z(N,d)Z(N,d) is defined over kk and

det⁡Z(N,d)=(−d)N/2=dN/2.(8)\det Z(N,d)=(-d)^{N/2}=d^{N/2}. \tag*{(8)}

The six matrices used in the construction are

Z1=Z2=Z(4,3),Z3=Z4=Z(8,17),Z_1=Z_2=Z(4,3),\qquad Z_3=Z_4=Z(8,17),
Z5=Z(4,51),Z6=Z(8,51).Z_5=Z(4,51),\qquad Z_6=Z(8,51).

We set

A1=diag⁡(172,1,1,1),Aj=Inj (j=2,…,6),Bj=AjZj,A_1=\operatorname{diag}(17^2,1,1,1),\qquad A_j=I_{n_j}\ (j=2,\ldots,6),\qquad B_j=A_jZ_j,

and put

t∗=(1,17,1,3,1,1),g∗=((Aj,Bj)j=16,t∗).t_*=(1,17,1,3,1,1),\qquad g_*=\left((A_j,B_j)_{j=1}^6,t_*\right).

Writing χ(t)=(χj(t))j=16\chi(t)=(\chi_j(t))_{j=1}^6 as defined in the introduction, we obtain

(det⁡Aj)j=(172,1,1,1,1,1)=χ(t∗),(\det A_j)_j=(17^2,1,1,1,1,1)=\chi(t_*),
(det⁡Bj)j=(512,32,174,174,512,514)=χ(τt∗),(\det B_j)_j=(51^2,3^2,17^4,17^4,51^2,51^4)=\chi(\tau t_*),

by (8) and the characters (1.2)–(1.4). Thus g∗∈G(Q17)g_*\in G(\mathbb{Q}_{17}).

For the remainder of this section Ej=k(ρj)E_j=k(\rho_j), where ρjnj=dj\rho_j^{n_j}=d_j. We first determine the centralizers that enter the double-coset calculation.

Lemma 3.3. Each Ej/kE_j/k is cyclic of degree njn_j. For dj=3d_j=3 it is unramified, and for dj=17,51d_j=17,51 it is totally ramified. The matrix ZjZ_j has irreducible characteristic polynomial of degree njn_j. Its centralizer in GLnj(k)\mathrm{GL}_{n_j}(k) is Ej×E_j^\times acting by multiplication, and the determinants of its elements form NEj/k(Ej×)N_{E_j/k}(E_j^\times).

Proof. Write (N,d)=(nj,dj)(N,d)=(n_j,d_j), ρ=ρj\rho=\rho_j, and E=EjE=E_j. For (N,d)=(4,3)(N,d)=(4,3), a root α\alpha of X4−3X^4-3 over F17\mathbb{F}_{17} satisfies α17=34α=13α\alpha^{17}=3^4\alpha=13\alpha. The element 1313 has order four in F17\mathbb{F}_{17}, so the four roots form one Frobenius orbit. The polynomial has irreducible separable reduction, giving an unramified extension of degree four. For d=17d=17 or 5151 the polynomial XN−dX^N-d is Eisenstein at 1717, so E/kE/k is totally ramified of degree NN and ρ\rho is a uniformizer. Since ζN∈k\zeta_N\in k has order NN, in all cases the extension is cyclic with generator ρ↦ζNρ\rho\mapsto\zeta_N\rho.

In the power basis of EE, the matrix Z(N,d)Z(N,d) represents multiplication by

z=ρN/21+ρ1+ζNρ.z=\rho^{N/2}\frac{1+\rho}{1+\zeta_N\rho}.

The unique involution of Gal⁡(E/k)\operatorname{Gal}(E/k) sends ρ\rho to −ρ-\rho. It fixes ρN/2\rho^{N/2} because N/2N/2 is even. If it fixed zz, cross-multiplication would give 2(1−ζN)ρ=02(1-\zeta_N)\rho=0, which is impossible. Every nontrivial subgroup of this cyclic 22-group contains the involution, so the stabilizer of zz is trivial and k[z]=Ek[z]=E.

Multiplication by zz therefore has irreducible minimal and characteristic polynomials of degree NN. A kk-linear map commuting with it is EE-linear on the one-dimensional EE-space EE, hence multiplication by an element of EE. The determinant of multiplication by any such element is its field norm. □\square

Lemma 3.4. The rational double coset H(k)g∗H(k)H(k)g_*H(k) is closed in G(k)G(k).

Proof. For g=((Aj,Bj)j,t)\mathbf{g}=((A_j,B_j)_j,t) put Yj(g)=Aj−1BjY_j(\mathbf{g})=A_j^{-1}B_j. Under left multiplication by H(k)H(k) the matrices YjY_j are unchanged. A right multiplier with blocks (Dj,Dj)(D_j,D_j) sends YjY_j to Dj−1YjDjD_j^{-1}Y_jD_j. Let FF consist of the g\mathbf{g} with t∈t∗S(k)t\in t_*S(k) such that each Yj(g)Y_j(\mathbf{g}) has the characteristic polynomial of ZjZ_j. These conditions are closed in the kk-adic topology and invariant under both copies of H(k)H(k). Hence FF is a closed union of double cosets containing g∗g_*.

We show that each double coset in FF is relatively open. Fix g0=((Aj0,Bj0)j,t0)∈F\mathbf{g}^0=((A_j^0,B_j^0)_j,t^0)\in F and put Yj0=Yj(g0)Y_j^0=Y_j(\mathbf{g}^0). By Lemma 3.3 its characteristic polynomial is irreducible, so any nonzero vector wjw_j is cyclic. For YY near Yj0Y_j^0 the matrix

Pj(Y)=[wj,Ywj,…,Ynj−1wj]P_j(Y)=[w_j,Yw_j,\ldots,Y^{n_j-1}w_j]

is invertible. In these cyclic bases Y=Yj(g)Y=Y_j(\mathbf{g}) and Yj0Y_j^0 have the same companion matrix whenever g∈F\mathbf{g}\in F. Consequently Dj=Pj(Yj0)Pj(Y)−1D_j=P_j(Y_j^0)P_j(Y)^{-1} satisfies Dj−1Yj0Dj=YD_j^{-1}Y_j^0D_j=Y and tends to the identity as g→g0\mathbf{g}\to\mathbf{g}^0. For g\mathbf{g} sufficiently close to g0\mathbf{g}^0, Hensel’s lemma gives an njn_jth root of (det⁡Dj)−1(\det D_j)^{-1} in kk, since 17∤nj17\nmid n_j. Multiplying DjD_j by this scalar preserves the conjugacy and gives det⁡Dj=1\det D_j=1 in every block.

Right multiplication of g0\mathbf{g}^0 by ((Dj,Dj)j,1)∈H(k)((D_j,D_j)_j,1)\in H(k) now gives the required relative matrices Yj(g)Y_j(\mathbf{g}). For the left multiplier take torus component t/t0∈S(k)t/t^0\in S(k) and common blocks Cj=Aj(Aj0Dj)−1C_j=A_j(A_j^0D_j)^{-1}. Their determinants are χj(t/t0)\chi_j(t/t^0), so the multiplier belongs to H(k)H(k). This gives g\mathbf{g} itself, because Bj=AjYj(g)B_j=A_jY_j(\mathbf{g}). All double cosets in FF are thus relatively open. Their complements in FF are unions of relatively open double cosets, so each is also relatively closed. Since FF is closed in G(k)G(k), the assertion follows. □\square

It remains to decide whether the anti-involution preserves this closed orbit. To record the necessary norm conditions, let m(x)=ν(x) mod 2m(x)=\nu(x)\bmod2 and let ϵ(x)∈F2\epsilon(x)\in\mathbb{F}_2 record the square class of the residue of the unit 17−ν(x)x17^{-\nu(x)}x, with 00 for a square and 11 for a nonsquare. Both are homomorphisms from k×k^\times to the additive group F2\mathbb{F}_2.

Lemma 3.5. Suppose xN/2x^{N/2} is a norm from E=k(dN)E=k(\sqrt[N]{d}) for one of the pairs (N,d)(N,d) occurring in the construction. Then

ddnecessary condition on xx
3m(x)=0m(x)=0
17ϵ(x)=0\epsilon(x)=0
51ϵ(x)+m(x)=0\epsilon(x)+m(x)=0

Table 1.

Proof. If d=3d=3 then N=4N=4 and the extension is unramified. The valuation of a norm is divisible by four, since all four conjugates have the same integral valuation. Thus 4∣2ν(x)4\mid2\nu(x).

In the ramified cases put h=N/2h=N/2, d=17Dd=17D, r=ν(x)r=\nu(x), and x=17rwx=17^r w with w∈O×w\in\mathcal{O}^\times. Every element of E×E^\times is ρlb\rho^l b for a unit bb of EE, where ρN=d\rho^N=d and NE/k(ρ)=−17DN_{E/k}(\rho)=-17D. If NE/k(ρlb)=xhN_{E/k}(\rho^l b)=x^h, valuation gives l=hrl=hr. Total ramification means all conjugates of bb have the same residue in F17×\mathbb{F}_{17}^\times, whence

(w‾(−D‾)−r)h=NE/k(b)=w‾ N.\left(\overline{w}(-\overline{D})^{-r}\right)^h=N_{E/k}(b)=\overline{w}^{\,N}.

In the cyclic group F17×\mathbb{F}_{17}^\times of order 1616, with N=2hN=2h dividing 1616, an hhth power aha^h is an NNth power only if aa is a square. Indeed, writing aa as the qqth power of a generator gives hq≡2hs(mod16)hq\equiv2hs\pmod{16} for some ss, so qq is even. Finally −1-1 is a square modulo 1717 and 33 is a nonsquare (38≡−13^8\equiv-1). For D=1D=1 this gives ϵ(x)=0\epsilon(x)=0, and for D=3D=3 it gives ϵ(x)+m(x)=0\epsilon(x)+m(x)=0. □\square

Proposition 3.6. The symmetric pair defined by (1.5)–(1.6) is not stable.

Proof using Carmeli’s second obstruction. Put r=s(g∗)r=s(g_*) and Pj=BjAj−1=AjZjAj−1P_j=B_jA_j^{-1}=A_jZ_jA_j^{-1}. In our coordinates,

r=((Pj−1,Pj)j,t∗/τ(t∗)).r=((P_j^{-1},P_j)_j,t_*/\tau(t_*)).

By Lemma 3.3, ZjZ_j has separable characteristic polynomial, so the matrix components of rr are semisimple. Its torus component is semisimple as well. Hence rr is a semisimple element of s(G(k))s(G(k)), and the double coset of g∗g_* is closed by Lemma 3.4.

Let C=ZG(r)(k)C=Z_G(r)(k). Since θ(r)=r−1\theta(r)=r^{-1}, the involution θ\theta preserves CC and rr represents a cocycle for its action on CC. Carmeli identifies the second obstruction with the class represented by rr and proves that rr is stable exactly when this class is neutral in H1(θ,C)H^1(\theta,C) [2 Propositions 4.18–4.19]. Here neutrality means that

r=cθ(c)−1for some c∈C.(9)r=c\theta(c)^{-1}\qquad\text{for some }c\in C. \tag*{(9)}

Indeed, in the convention y−1θ(y)y^{-1}\theta(y) one takes y=c−1y=c^{-1}. In G(k)G(k) the equation already has the solution g∗g_*.

We determine when it has a solution in CC. Set Fj=Ajk[Zj]Aj−1≃EjF_j=A_jk[Z_j]A_j^{-1}\simeq E_j and Nj=NEj/k(Ej×)N_j=N_{E_j/k}(E_j^\times). The two blocks Pj−1P_j^{-1} and PjP_j have the same centralizer Fj×F_j^\times in GLnj(k)\mathrm{GL}_{n_j}(k), and the determinant on Fj×F_j^\times is the field norm (Lemma 3.3). Thus an element c=((Uj,Vj)j,w)c=((U_j,V_j)_j,w) of CC has Uj,Vj∈Fj×U_j,V_j\in F_j^\times, with det⁡Uj=χj(w)\det U_j=\chi_j(w) and det⁡Vj=χj(τw)\det V_j=\chi_j(\tau w).

The torus component of (3.2) requires w/τ(w)=t∗/τ(t∗)w/\tau(w)=t_*/\tau(t_*), or equivalently w=t∗sw=t_*s with s∈S(k)s\in S(k). The first matrix components require UjVj−1=Pj−1U_jV_j^{-1}=P_j^{-1}, hence Vj=PjUjV_j=P_jU_j. It follows that (3.2) is solvable exactly when there exists a single s∈S(k)s\in S(k) such that

χj(t∗s)∈Nj(1≤j≤6).(10)\chi_j(t_*s)\in N_j\qquad(1\leq j\leq6). \tag*{(10)}

For sufficiency, choose Uj∈Fj×U_j\in F_j^\times with that determinant and set Vj=PjUjV_j=P_jU_j. Since det⁡Pj=χj(τt∗)/χj(t∗)\det P_j=\chi_j(\tau t_*)/\chi_j(t_*) and τ(s)=s\tau(s)=s, we get det⁡Vj=χj(τw)\det V_j=\chi_j(\tau w). Also VjUj−1=PjV_jU_j^{-1}=P_j, so these choices solve the full equation in CC.

Write s=(a,a,b,b,u,v)s=(a,a,b,b,u,v). The six bases of the powers χj(t∗s)=cj(t∗s)nj/2\chi_j(t_*s)=c_j(t_*s)^{n_j/2} in (3.3) are

(17a2bv, b, av, a, abuv, abu).(17a^2bv,\ b,\ av,\ a,\ abuv,\ abu).

Apply Lemma 3.5 in each pair of blocks. For j=1,2j=1,2 the quotient of the bases gives m(17a2v)=0m(17a^2v)=0, so m(v)=1m(v)=1. For j=3,4j=3,4 the quotient gives ϵ(v)=0\epsilon(v)=0. For j=5,6j=5,6 it gives ϵ(v)+m(v)=0\epsilon(v)+m(v)=0. These three equalities contradict one another in F2\mathbb{F}_2. Hence (3.3) has no solution and [r]≠1[r]\ne1 in H1(θ,C)H^1(\theta,C). By Carmeli’s Proposition 4.19 the semisimple symmetrization rr is unstable. Thus the closed double coset of g∗g_* is moved by σ\sigma, proving the proposition. □

Proof of Theorem 1.1. Proposition 2.3 gives the multiplicity bound for every irreducible KK-unramified representation of G(k)G(k). Proposition 3.6 exhibits a closed rational double coset for this same pair that is moved by σ\sigma. Together they give a negative answer to (1.1). □

References

  1. [1]A. Aizenbud and D. Gourevitch, Generalized Harish-Chandra descent, Gelfand pairs, and an Archimedean analog of Jacquet–Rallis’s theorem, Duke Math. J. 149 (2009), 509–567.
  2. [2]S. Carmeli, On the stability and Gelfand property of symmetric pairs, arXiv:1511.01381.
  3. [3]B. H. Gross, On the Satake isomorphism, in Galois Representations in Arithmetic Algebraic Geometry, Cambridge University Press, 1998, 223–238.DOI
  4. [4]Y. Sakellaridis, Spherical functions on spherical varieties, Amer. J. Math. 135 (2013), 1291–1381.

Paper details

vlnc:W_s3rxpgwn
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 9

downloads · 0

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

  • Mathematics

OpenAI ChatGPT

View statement
Looking for related papers…
Contents