The universal DAHA of type (C∨,C ) and 1 1 Leonard pairs of q-Racah type 7 1 0 Kazumasa Nomura and Paul Terwilliger 2 n a Abstract J 1 ALeonardpairisapairofdiagonalizablelineartransformationsofafinite-dimensional 2 vector space, each of which acts in an irreducible tridiagonal fashion on an eigenbasis for the other one. Let F denote an algebraically closed field, and fix a nonzero q ∈ F ] that is not a root of unity. The universal double affine Hecke algebra (DAHA) Hˆ of A q Q (tyi)petit(−iC11∨,=C1t)−i 1istith=e a1s;so(icii)attiiv+eFt-−ia1lgiesbrcaendterfialn;ed(iibi)ytg0etn1te2rta3to=rsq{−t±i1.1}3iW=0eacnodnsriedleartiothnes h. elements X = t3t0 and Y = t0t1 of Hˆq. Let V denote a finite-dimensional irreducible at Hˆq-moduleonwhicheachofX,Y isdiagonalizableandt0 hastwodistincteigenvalues. m Then V is a direct sum of the two eigenspaces of t0. We show that the pair X +X−1, Y +Y−1 acts on each eigenspace as a Leonard pair, and each of these Leonard pairs [ falls into a class said to have q-Racah type. Thus from V we obtain a pair of Leonard 1 pairs of q-Racah type. It is known that a Leonard pair of q-Racah type is determined v up to isomorphism by a parameter sequence (a,b,c,d) called its Huang data. Given a 9 8 pair of Leonard pairs of q-Racah type, we find necessary and sufficient conditions on 0 their Huang data for that pair to come from the above construction. 6 0 . 1 Introduction 1 0 7 Throughout the paper F denotes an algebraically closed field. Fix a nonzero q ∈ F that is 1 not a root of unity. An F-algebra is meant to be associative and have a 1. : v WebeginbyrecallingthenotionofaLeonardpair. Weusethefollowingterms. Asquare i X matrix is said to betridiagonal whenever each nonzeroentry lies on either thediagonal, the r subdiagonal, or the superdiagonal. A tridiagonal matrix is said to be irreducible whenever a each entry on the subdiagonal is nonzero and each entry on the superdiagonal is nonzero. Definition 1.1 (See [12, Definition 1.1].) Let V denote a vector space over F with fi- nite positive dimension. By a Leonard pair on V we mean an ordered pair of F-linear ∗ transformations A:V → V and A : V → V that satisfy (i) and (ii) below: (i) thereexistsabasisforV withrespecttowhichthematrixrepresentingAisirreducible ∗ tridiagonal and the matrix representing A is diagonal; ∗ (ii) there exists a basis for V with respect to which the matrix representing A is irre- ducible tridiagonal and the matrix representing A is diagonal. We say that A,A∗ is over F. By the diameter of A,A∗ we mean dimV −1. 1 Note 1.2 According to a common notational convention, A∗ denotes the conjugate-trans- pose of A. We are not using this convention. In a Leonard pair A,A∗ the F-linear trans- ∗ formations A and A are arbitrary subject to (i) and (ii) above. We refer the reader to [2,3,8,12–14] for background on Leonard pairs. ′ We recall the notion of an isomorphism of Leonard pairs. Let V (resp. V ) denote a vector space over F with finite positive dimension, and let A,A∗ (resp. A′,A∗′) denote a ′ ∗ ′ ∗′ Leonard pair on V (resp. V ). By an isomorphism of Leonard pairs from A,A to A,A we mean an F-linear bijection f :V → V′ such that both fA= A′f and fA∗ = A∗′f. We recall some facts about the eigenvalues of a Leonard pair. We use the following terms. LetV denoteavector spaceover Fwithfinitepositive dimensionandlet A:V → V denote an F-linear transformation. Then A is said to be diagonalizable whenever V is spannedby theeigenspaces of A. Wesay Ais multiplicity-free whenever Ais diagonalizable ∗ and each eigenspace of A has dimension one. Let A,A denote a Leonard pair on V. Then each of A,A∗ is multiplicity-free (see [12, Lemma 1.3]). Let {θ }d denote an ordering r r=0 of the eigenvalues of A. For 0 ≤ r ≤ d let 0 6= v ∈ V denote an eigenvector of A r associated with θ . Observe that {v }d is a basis for V. The ordering {θ }d is said to r r r=0 r r=0 be standard whenever the basis {v }d satisfies Definition 1.1(ii). A standard ordering of r r=0 the eigenvalues of A∗ is similarly defined. Let {θ }d denote a standard ordering of the r r=0 eigenvalues of A. Then the ordering {θd−r}dr=0 is also standard and no further ordering is ∗ standard. A similar result applies to A . Definition 1.3 Let d ≥ 0 denote an integer and let {θ }d denote a sequence of scalars r r=0 in F. The sequence {θ }d is said to be q-Racah whenever there exists a nonzero α ∈ F r r=0 such that θ = αq2r−d +α−1qd−2r for 0 ≤ r ≤ d. The scalar α is uniquely determined by r {θ }d if d ≥ 1, and determined up to inverse if d = 0. We call α the parameter of the r r=0 q-Racah sequence. For a q-Racah sequence {θr}dr=0 with parameter α, the inverted sequence {θd−r}dr=0 is q-Racah with parameter α−1. Definition 1.4 Let A,A∗ denote a Leonard pair over F with diameter d. Then A,A∗ is ∗ said to have q-Racah type whenever for each of A,A a standard orderingof theeigenvalues forms a q-Racah sequence. Referring to Definition 1.4, assume that A,A∗ has q-Racah type. Let {θ }d (resp. r r=0 {θ∗}d ) denote a standard ordering of the eigenvalues of A (resp. A∗). Let a (resp. b) r r=0 denote the parameter of the q-Racah sequence {θ }d (resp. {θ∗}d ). It is known that r r=0 r r=0 A,A∗ is determined up to isomorphism by a, b, d and one more nonzero scalar c ∈ F. The ∗ sequence (a,b,c,d) is called a Huang data of A,A . The scalar c is defined up to inverse if ∗ d ≥ 1, and arbitrary if d = 0 (see Section 2). For a Huang data (a,b,c,d) of A,A , each of (a±1,b±1,c±1,d) is a Huang data of A,A∗. Moreover A,A∗ has no further Huang data, provided that d ≥1. Next we recall the universal double affine Hecke algebra Hˆ . The double affine Hecke q ∨ algebra(DAHA)wasintroducedbyCherednik[1]. TheDAHAoftype(C ,C )wasstudied 1 1 2 by Macdonald [7, Ch. 6], Noumi-Stokman [9], Sahi [10,11], Koornwinder [5,6], and Ito- Terwilliger [4]. The algebra Hˆ was introduced by the second author as a generalization of q theDAHAoftype(C∨,C ). WenowrecallthedefinitionofHˆ . Fornotationalconvenience 1 1 q define I= {0,1,2,3}. Definition 1.5 (See [16, Definition 3.1].) Let Hˆ denote the F-algebra defined by genera- q tors {t±i 1}i∈I and relations t t−1 = t−1t = 1 i ∈ I, (1) i i i i t +t−1 is central i ∈ I, (2) i i t t t t = q−1. (3) 0 1 2 3 The algebra Hˆ is called the universal DAHA of type (C∨,C ). q 1 1 Referring to Definition 1.5, for i ∈ I define T = t +t−1. i i i Note that T is central in Hˆ . i q Definition 1.6 Let V denote a finite-dimensional irreducible Hˆ -module. By Schur’s q lemma each T acts on V as a scalar. Write this scalar as k + k−1 with 0 6= k ∈ F. i i i i Thus T = k +k−1 on V. (4) i i i We refer to {ki}i∈I as a parameter sequence of V. Referring to Definition 1.6, note that each k is defined up to inverse. So each of the 16 i sequences {ki±1}i∈I is a parameter sequence of V, and V has nofurther parameter sequence. Observe by (4) that (t −k )(t −k−1)V = 0. Thus the eigenvalues of t are among k , k−1. i i i i i i i We consider the following elements of Hˆ : q X = t t , Y = t t , A = Y +Y−1, B = X +X−1. (5) 3 0 0 1 It is known that each of A, B commutes with t (see Lemma 3.6). By an XD (resp. YD) 0 Hˆ -module we mean a finite-dimensional irreducible Hˆ -module on which X (resp. Y) is q q diagonalizable. An Hˆ -module is said to be feasible whenever (i) it is both XD and YD; q (ii) t has two distinct eigenvalues. Let V denote a feasible Hˆ -module with parameter 0 q sequence {ki}i∈I. Then k02 6= 1. Moreover t0 is diagonalizable on V with eigenvalues k0 and k−1. Observe that V is a direct sum of the two eigenspaces of t , and each eigenspace 0 0 is invariant under A, B. We remark that t does not commute with X, Y, and so the 0 eigenspaces of t may not be invariant under X, Y. 0 We now state our first main result. 3 Theorem 1.7 Let V denote a feasible Hˆ -module. Then A,B act on each eigenspace of t q 0 as a Leonard pair of q-Racah type. Let V denote a feasible Hˆ -module. By Theorem 1.7 we obtain a pair of Leonard pairs q of q-Racah type. In order to describe how these Leonard pairs are related, we use the ∗ ′ ∗′ following notion. Let A,A denote a Leonard pair on V and let A,A denote a Leonard ′ ′ pair on V . We say that these Leonard pairs are linked whenever the direct sum V ⊕V supports a feasible Hˆ -module structure such that V, V′ are the eigenspaces of t and q 0 A|V = A, B|V = A∗, A|V′ = A′, B|V′ = A∗′. (6) We now state our second main result. Theorem 1.8 Suppose we are given two Leonard pairs A,A∗ and A′,A∗′ over F that have q-Racah type. Then these Leonard pairs are linked if and only if there exist a Huang data ∗ ′ ′ ′ ′ ′ ∗′ (a,b,c,d) of A,A and a Huang data (a,b,c,d) of A,A that satisfy one of (i)–(vii) below: ′ ′ ′ ′ Case d −d a/a b/b c/c Inequalities (i) −2 1 1 1 (ii) −1 q q q a2 6= q−2d b2 6= q−2d (iii) 0 q2 1 1 b2 6= q±2d a2 6= q−2 (iv) 0 1 q2 1 a2 6= q±2d b2 6= q−2 (v) 0 1 1 q2 a2 6= q±2d b2 6= q±2d c2 6= q−2 (vi) 1 q−1 q−1 q−1 a2 6= q−2d b2 6= q−2d (vii) 2 1 1 1 Remark 1.9 Referring to Theorem 1.8, in each of (ii)–(vi) there appear some inequalities. ForeachoftheseinequalitiesweexplaintheroleinourconstructionofafeasibleHˆ -module. q The inequalities in the first column are needed to make Y diagonalizable. The inequalities in the second column are needed to make X diagonalizable. The inequalities in the third column are needed to make t have two distinct eigenvalues. 0 Remark 1.10 Referring to Theorem 1.8, for d ≥ 2 we have a2 6= q−2 and b2 6= q−2 by Lemma 2.8(i), so these inequalities can be deleted from the table. Remark 1.11 Suppose we exchange our two Leonard pairs in Theorem 1.8. In the con- ′ ′ ′ ′ ditions (i)–(vii), the Huang data (a,b,c,d) and (a,b,c,d) are exchanged. After this exchange, the conditions (i), (ii), (vi), (vii) become the original conditions (vii), (vi), (ii), ′ ′ ′ (i) respectively. Concerning the conditions (iii)–(v), we replace each of a, b, c, a, b, c with its inverse after the above exchange. Thenthe conditions (iii)–(v) become the original conditions (iii)–(v) respectively. In a moment we will summarize the proof of Theorems 1.7 and 1.8. Prior to that we explain the significance of the cases that show up in Theorem 1.8. Let V denote an XD 4 Hˆ -module. For µ ∈ F, let the subspace V (µ) consist of the vectors v ∈ V such that q X Xv = µv. Thus V (µ) is nonzero if and only if µ is an eigenvalue of X, and in this case X V (µ) is the corresponding eigenspace. As we will see in Lemmas 6.1 and 6.2, for µ, ν ∈ F X with µν = 1 (resp. µν = q−2) the subspace V (µ)+V (ν) is invariant under t , t (resp. X X 0 3 t , t ). Motivated by this, we consider the following diagram. 1 2 Let µ, ν ∈ F. We say µ, ν are 1-adjacent (resp. q-adjacent) whenever their product is 1 (resp. q−2). For a subset M of F we define the diagram of M, that has vertex set M, and µ, ν ∈ M are connected by a single bond (resp. double bond) whenever µ, ν are 1-adjacent (resp. q-adjacent). If µ = ν then the single bond(resp.double bond)becomes a single loop (resp. double loop). The reduced diagram of M is obtained by deleting all the loops in the diagram of M. Let V denote an XD Hˆ -module. By the X-diagram of V we mean the diagram of M, q where M is the set of eigenvalues of X on V. As we will see in Lemma 6.7, the reduced X-diagram is a path. So the reduced X-diagram has one of the following types: DS: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) DD: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (7) SS: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) If the above diagram has only one vertex, we interpret it to be DS. For each of (7) we choose an ordering of the eigenvalues {µ }n of X as follows: r r=0 µ0 µ1 µ2 µ3 µn−1 µn DS: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) µ0 µ1 µ2 µ3 µn−1 µn DD: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (8) µ0 µ1 µ2 µ3 µn−1 µn SS: (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:10)(cid:13)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:11)(cid:12) ······ (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) (cid:8)(cid:15)(cid:9)(cid:14)(cid:13)(cid:10)(cid:12)(cid:11) It turns out that each eigenspace of X has dimension one (see Proposition 7.8). For each end-vertex µ of the diagram, we consider the action of {ti}i∈I on VX(µ). As we will see in Lemma 8.1, Case V (µ ) is invariant under V (µ ) is invariant under X 0 X n DS t , t t , t 0 3 1 2 (9) DD t , t t , t 0 3 0 3 SS t , t t , t 1 2 1 2 For each t , consider the action of t on V. For this action any two distinct eigenvalues are i i reciprocals. For the reduced X-diagram DD one of the following cases occurs (see Lemma 8.2): Eigenvalues of t on Eigenvalues of t on Case 0 3 V (µ ) and V (µ ) V (µ ) and V (µ ) X 0 X n X 0 X n (10) DDa same reciprocals DDb reciprocals same 5 Similarly, for the reduced X-diagram SS one of the following cases occurs (see Lemma 8.3): Eigenvalues of t on Eigenvalues of t on Case 1 2 V (µ ) and V (µ ) V (µ ) and V (µ ) X 0 X n X 0 X n (11) SSa same reciprocals SSb reciprocals same We refer to each case DS, DDa, DDb, SSa, SSb as the X-type of V. An XD Hˆ -module is q determined up to isomorphism by its dimension, its parameter sequence, and its X-type (see Note 9.3). The cases (i)–(v) in Theorem 1.8 correspond to the X-types as follows: (i) (ii) (iii) (iv) (v) (12) DDa DS SSa DDb SSb The cases (vi) and (vii) are reduced to (ii) and (i) by exchanging our two Leonard pairs (see Remark 1.11). Recall that V has 16 parameter sequences {ki±1}i∈I. In view of (9)–(11) we adopt the following convention for most of the paper: Case Rule k (resp. k ) is the eigenvalue of t (resp. t ) on V (µ ) DS 0 3 0 3 X 0 k1 (resp. k2) is the eigenvalue of t1 (resp. t2) on VX(µn) (13) DD k (resp. k ) is the eigenvalue of t (resp. t ) on V (µ ) 0 3 0 3 X 0 SS k (resp. k ) is the eigenvalue of t (resp. t ) on V (µ ) 1 2 1 2 X 0 Under this convention, the following equation holds (see Lemma 12.1): X-type of V Equation DS k k k k = q−n−1 0 1 2 3 DDa k2 = q−n−1 0 (14) DDb k2 = q−n−1 3 SSa k2 = q−n−1 1 SSb k2 = q−n−1 2 Below wesummarize ourproof of themain theorems. Until starting theproofsummary of Theorem 1.8, the following notation is in effect. Let V denote an XD Hˆ -module with q dimension n+1, n ≥ 1. We consider the reduced X-diagram of V. Let {µ }n denote r r=0 the eigenvalues of X on V, as shown in (8). Choose a parameter sequence {ki}i∈I of V that satisfies (13). Assume that k 6= k−1. Let V(k ) denote the subspace of V consisting of 0 0 0 v ∈V such that t v = k v. The subspace V(k−1) is similarly defined. 0 0 0 Towards Theorem 1.7, we make the following observations (A)–(C). (A) The eigenvalues {µr}nr=0 are determined by {ki}i∈I, using (13) and the shape of the diagram. (B) We indicated earlier that for each single bond µ, ν the subspace V (µ)+V (ν) X X is invariant under t . It turns out that the intersections of this subspace with V(k ) and 0 0 6 V(k−1) each have dimension one. Call these intersections bond subspaces. For an endvertex 0 µ that is incident to a double bond, by (9) V (µ) is invariant under t , so it is contained X 0 in one of V(k ), V(k−1). Call V (µ) a bond subspace. Note that each of V(k ), V(k−1) is 0 0 X 0 0 a direct sum of its bond subspaces. Also note that t has only one eigenvalue on V if and 0 only if n = 1 and V has X-type DDa. (C)ForeachvertexµthesubspaceV (µ)isaneigenspaceofX witheigenvalueµ. Since X B = X +X−1, V (µ) is invariant under B with eigenvalue µ+µ−1. For each single bond X µ, ν we have µν = 1, so on V (µ) and V (ν) the eigenvalue of B is the same. Therefore B X X acts on V (µ)+V (ν) as µ+ν times the identity. X X We summarize our proof of Theorem 1.7. Assume that V is feasible. First consider the action of A,B on V(k ). We pick a nonzero vector in each bond subspace of V(k ). By 0 0 (B) these vectors form a basis for V(k ). We order these vectors along with the ordering 0 {µ }n , and denote them by {w }d . By (C) the vectors {w }d are eigenvectors of B. r r=0 r r=0 r r=0 By (A) the corresponding eigenvalues are represented in terms of {ki}i∈I, and we find that these eigenvalues form a q-Racah sequence. We indicated earlier that for each single (resp. double) bond µ, ν the subspace V (µ)+V (ν) is invariant under t (resp. t ). By this X X 0 1 we see that the matrix representing A with respect to {w }d is tridiagonal, and it turns r r=0 out that this tridiagonal matrix is irreducible. We have described the action of A,B on V(k ), and the action of A,B on V(k−1) is similar. So far for each of V(k±1) there exists 0 0 0 a basis with respect to which the matrix representing A is irreducible tridiagonal and the matrix representing B is diagonal whose diagonal entries form a q-Racah sequence. There is an automorphism σ of Hˆ that fixes t and swaps A, B. Applying the above fact to the q 0 twisted Hˆ -module Vσ, we find that for each of V(k±1) there exists a basis with respect to q 0 which the matrix representing B is irreducible tridiagonal and the matrix representing A is diagonal whose diagonal entries form a q-Racah sequence. Therefore A,B act on each V(k±1) as a Leonard pair of q-Racah type. 0 Towards Theorem 1.8, we do the following (D)–(G). (D) We construct a certain basis {u }n for V such that for 1 ≤ r ≤ n we have r r=0 Yeur−1 ∈ F(ur−1−ur), where e= 1 if µr−1, µr are 1-adjacent, and e = −1 if µr−1, µr are q-adjacent. (E) We obtain the action of X±1 and Y±1 on the basis {u }n . The results show that r r=0 with respect to the basis {u }n the matrices representing X±1, B are lower tridiagonal, r r=0 and the matrices representing Y±1, A are upper tridiagonal. (F) We obtain some inequalities for {ki}i∈I from the facts that X is diagonalizable on V, {µ }n are mutually distinct, and each V (µ ) has dimension one (see Lemma 12.4). r r=0 X r Also, using the basis {ur}nr=0 we obtain necessary and sufficient conditions on {ki}i∈I for Y to be diagonalizable on V (see Corollary 14.8). (G) Assume that V is feasible, and consider the two Leonard pairs of q-Racah type from Theorem 1.7. We represent a Huang data of each Leonard pair in terms of {ki}i∈I as follows. First consider the Leonard pair A,B on V(k ). Using the basis {u }n from 0 r r=0 (D) we construct a basis for V(k ) with respect to which the matrix representing A (resp. 0 B) is lower bidiagonal (resp. upper bidiagonal). By construction the diagonal entries of this matrix give an ordering of the eigenvalues of A (resp. B) on V(k ), and it turns out 0 (see Lemma 2.4) that this ordering is standard. This standard ordering forms a q-Racah 7 sequence. Let a (resp. b) denote the parameter of this q-Racah sequence. We represent a (resp.b)intermsof{ki}i∈I. Weremarkthattheabovestandardorderingoftheeigenvalues of B coincides with the one from the proof summary of Theorem 1.7 (see Proposition 11.12 and Corollary 19.6). We define a certain nonzero c ∈ F in terms of {ki}i∈I. Using the equitable Askey-Wilson relations (see Lemma 20.1), we show that (a,b,c,d) is a Huang data of the Leonard pair A,B on V(k ). So far, we have represented a Huang data of the 0 Leonard pair A,B on V(k0) in terms of {ki}i∈I. We similarly represent a Huang data of the Leonard pair A,B on V(k0−1) in terms of {ki}i∈I. We now summarize our proof of Theorem 1.8. First consider the “only if” direction. ∗ ′ ∗′ ′ Suppose we are given two Leonard pairs A,A on V and A,A on V that have q-Racah type. Assume that these Leonard pairs are linked. So we have a feasible Hˆ -module q structure on V := V ⊕ V′ such that V, V′ are the eigenspaces of t and (6) holds. Let 0 {ki}i∈I denote a parameter sequence of V that satisfies (13). First assume that V = V(k0) and V′ = V(k−1). In (G) we described a Huang data (a,b,c,d) of A,B on V(k ) and a 0 0 Huang data (a′,b′,c′,d′) of A,B on V(k−1). These Huang data are displayed in Proposition 0 ′ ′ ′ ′ 21.1. For these Huang data the value of d −d and the ratios a/a, b/b, c/c are as follows. X-type of V d′−d a′/a b′/b c′/c DS −1 q q q DDa −2 1 1 1 DDb 0 1 q2 1 SSa 0 q2 1 1 SSb 0 1 1 q2 Using (F), (14) and k2 6= 1 we get the following inequalities: 0 X-type of V inequalities DS a2 6= q−2d b2 6= q−2d DDa DDb a2 6= q±2d b2 6= q−2 SSa b2 6= q±2d a2 6= q−2 SSb a2 6= q±2d b2 6= q±2d c2 6= q−2 Now we find that the following case in Theorem 1.8 occurs: X-type of V DS DDa DDb SSa SSb Case (ii) (i) (iv) (iii) (v) For the case V = V(k−1) and V′ = V(k ) the argument is similar. 0 0 ∗ Next consider the “if” direction. Suppose we are given two Leonard pairs A,A on V ′ ∗′ ′ and A,A on V that have q-Racah type. Assume that there exist a Huang data (a,b,c,d) ∗ ′ ′ ′ ′ ′ ∗′ of A,A and aHuangdata (a,b,c,d) of A,A that satisfy one of theconditions (i)–(vii). We may assume that the Huang data satisfy one of (i)–(v) by exchanging our two Leonard 8 pairs if necessary. For each case (i)–(v), we define scalars {ki}i∈I as follows: Case k k k k 0 1 2 3 (i) q−d a c b (ii) (abcq1−d)1/2 aq−dk−1 cq−dk−1 bq−dk−1 0 0 0 (15) (iii) aq q−d−1 b c (iv) bq c a q−d−1 (v) cq b q−d−1 a In case (ii), take any one of the squareroots of abcq1−d for the valueof k . We constructan 0 XD Hˆq-moduleVwith dimensiond+d′+1 and parameter sequence {ki}i∈I in the following way. Set n = d+d′ +1 and let V denote a vector space over F with basis {v }n . We r r=0 define scalars {µ }n as follows: r r=0 Case Definition of µ r (i), (ii), (iv) µ = k k qr if r is even, µ = k k qr+1 −1 if r is odd r 0 3 r 0 3 (iii), (v) µ = k k qr+1 −1 if r is even, µ = k k qr if r is odd r 1 2 (cid:0) r 1 (cid:1)2 We find that the reduced d(cid:0)iagram of(cid:1){µ }n is as in (8) with r r=0 (i) (ii) (iii) (iv) (v) DD DS SS DD SS Using this diagram we define the action of {ti}i∈I on {vr}nr=0 as follows. For 0 ≤ r ≤ n−1 such that µ , µ are 1-adjacent (resp. q-adjacent), we define the action of t , t (resp. r r+1 0 3 t , t ) on Fv + Fv by (46)–(49) (resp.(50)–(53)). The remaining actions are defined 1 2 r r+1 by (43). It turns out that these actions give an Hˆ -module structure of V. Moreover, this q Hˆ -module V is XD and has X-type q (i) (ii) (iii) (iv) (v) (16) DDa DS SSa DDb SSb The inequalities in Theorem 1.8 yield some inequalities for {ki}i∈I (see Lemma 24.3). By these inequalities we find that {ki}i∈I satisfy the conditions in (F) that make Y diagonaliz- ableonV,soVisYD.Wealsofindthatt hastwodistincteigenvalues onV,soVisfeasible. 0 By Theorem 1.7 A,B act on each eigenspace of t as a Leonard pair of q-Racah type. These 0 Leonard pairs have Huang data that are displayed in Proposition 21.1. Using (15) we find ′ ′ ′ ′ that these Huang data coincide with (a,b,c,d) and (a,b,c,d). Thus the Leonard pair A,B on V(k ) (resp. V(k−1)) and the Leonard pair A,A∗ (resp. A′,A∗′) have a common 0 0 Huang data. So the Leonard pair A,B on V(k ) (resp. V(k−1)) and the Leonard pair A,A∗ 0 0 on V (resp. A′,A∗′ on V′) are isomorphic. Let f : V(k ) → V (resp. f′ : V(k−1) → V′) 0 0 denote the corresponding isomorphism of Leonard pairs. Consider the F-linear bijection f ⊕f′ : V = V(k )+V(k−1) → V ⊕V′. We define an Hˆ -module structure on V ⊕V′ so 0 0 q that f ⊕f′ is an isomorphism of Hˆ -modules. By construction the Hˆ -module V ⊕V′ is q q ′ feasible, the subspaces V and V arethe eigenspaces of t , and (6)holds. Thusthe Leonard 0 ∗ ′ ∗′ pairs A,A and A,A are linked. 9 The paper is organized as follows. In Section 2 we recall some materials concerning Leonard pairs. In Section 3 we recall some basic facts about Hˆ . In Sections 4 and 5 q we consider certain elements of Hˆ and study their properties. In Section 6 we study the q X-diagram. In Section 7 we show that X is multiplicity-free on an XD Hˆ -module. In q Section 8 we study the cases DS, DDa, DDb, SSa, SSb, and we do (A). In Section 9 we investigate the structureof an XD Hˆ -module. InSection 10 we investigate the eigenspaces q of t , and we do (B). In Section 11 we prove Theorem 1.7. In Section 12 we obtain some 0 (in)equalities for {ki}i∈I. In Sections 13 we do (D). In Section 14 we obtain the action of Y±1 on the basis {ur}nr=0, and we do (F). In Section 15 we obtain the action of {ti}i∈I on the basis {u }n . In Section 16 we obtain the action of X±1 on the basis {u }n . In r r=0 r r=0 Sections 17–21 we do (G). In Section 22 we prove the “only if” direction of Theorem 1.8. In Sections 23 and 24 we prove the “if” direction of Theorem 1.8. 2 Preliminaries for Leonard pairs Inthissection werecall somematerials concerningLeonardpairs. We firstrecall thenotion of a parameter array, and next recall the notion of a Huang data. Fix an integer d≥ 0. Let M denote a (d+1)×(d+1) matrix with all entries in F. We index the rows and columns by 0,1,...,d. Let V denote a vector space over F with basis {v }d , and consider a linear transformation A : V → V. We say M represents A with r r=0 respect to {v }d whenever Av = d M v for 0 ≤ s ≤ d. r r=0 s r=0 r,s r Lemma 2.1 (See [12, Theorem 3.2P].) Let V denote a vector space over F with dimension d + 1, and let A,A∗ denote a Leonard pair on V. Let {θ }d (resp. {θ∗}d ) denote a r r=0 r r=0 ∗ standard ordering of the eigenvalues of A (resp. A ). Then there exists a basis for V with ∗ respect to which the matrices representing A,A are θ 0 θ∗ ϕ 0 0 0 1 ∗ 1 θ θ ϕ 1 1 2    ∗  1 θ θ · A : 2 , A∗ : 2  · ·   · ·       · ·   · ϕ     d 0 1 θ  0 θ∗  d  d     forsome scalars {ϕ }d inF. The sequence{ϕ }d isuniquelydetermined bythe ordering r r=1 r r=1 ({θ }d ,{θ∗}d ). Moreover ϕ 6= 0 for 1≤ r ≤ d. r r=0 r r=0 r With reference to Lemma 2.1 we refer to {ϕ }d as the first split sequence of A,A∗ r r=1 associated with the ordering ({θ }d ,{θ∗}d ). By the second split sequence of A,A∗ r r=0 r r=0 associated with the ordering ({θ }d ,{θ∗}d ) we mean the first split sequence of A,A∗ r r=0 r r=0 associate with the ordering ({θd−r}dr=0,{θr∗}dr=0). 10