---
title: Tridiagonal Pair Overview
url: https://www.emergentmind.com/topics/tridiagonal-pair
type: topic
---

# Tridiagonal Pair Overview

A tridiagonal pair is an ordered pair \(A,A^*\) of diagonalizable linear transformations on a finite-dimensional vector space \(V\) such that there exists an ordering \(\{V_i\}_{i=0}^d\) of the eigenspaces of \(A\) with
\[
A^*V_i\subseteq V_{i-1}+V_i+V_{i+1}\qquad (0\le i\le d),
\]
there exists an ordering \(\{V_i^*\}_{i=0}^{\delta}\) of the eigenspaces of \(A^*\) with
\[
AV_i^*\subseteq V_{i-1}^*+V_i^*+V_{i+1}^*\qquad (0\le i\le \delta),
\]
and there is no nonzero proper common invariant subspace for \(A\) and \(A^*\) [1001.1812]. It is known that \(d=\delta\), and this common value is called the diameter. Tridiagonal pairs were motivated by the theory of \(Q\)-polynomial distance-regular graphs, and the subject has developed strong links with representation theory, bispectral problems, integrable systems, and \(q\)-Onsager and related algebras [1001.1812].

## 1. Definition, diameter, and shape

The defining tridiagonal containment conditions are formulated relative to standard orderings of eigenspaces. They do not require the eigenspaces to be one-dimensional, so the most natural matrix interpretation is block tridiagonal rather than scalar tridiagonal. The irreducibility condition excludes a decomposition into smaller common invariant pieces and is part of the definition itself [1001.1812].

A basic structural fact is that if \(\{V_i\}_{i=0}^d\) and \(\{V_i^*\}_{i=0}^d\) are standard orderings, then
\[
\dim V_i=\dim V_i^* \qquad (0\le i\le d).
\]
Writing this common value as \(\rho_i\), the sequence \((\rho_0,\rho_1,\dots,\rho_d)\) is called the shape. The shape is symmetric and unimodal:
\[
\rho_i=\rho_{d-i}\qquad (0\le i\le d),
\]
and
\[
\rho_{i-1}\le \rho_i \qquad (1\le i\le d/2).
\]
Accordingly,
\[
\dim V_i=\dim V_{d-i}=\dim V_i^*=\dim V_{d-i}^*.
\]
These identities are among the fundamental invariants of the theory [1001.1812].

The standard idempotent formalism is packaged in a tridiagonal system
\[
\Phi=\bigl(A;\{E_i\}_{i=0}^d;A^*;\{E_i^*\}_{i=0}^d\bigr),
\]
where \(E_i\) and \(E_i^*\) are the primitive idempotents associated with the chosen standard orderings. For a diagonalizable \(A\) with eigenvalues \(\theta_0,\dots,\theta_d\),
\[
E_i=\prod_{\substack{0\le j\le d\\ j\ne i}}\frac{A-\theta_jI}{\theta_i-\theta_j},
\]
and similarly for \(A^*\). This idempotent language is the natural coordinate-free substitute for choosing explicit eigenbases [1001.1812].

## 2. Sharp systems, split sequences, and classification by parameter arrays

A tridiagonal pair is called sharp when \(\dim V_0=1\). Over an algebraically closed field, every tridiagonal pair is sharp, so the sharp case is the generic classification regime in that setting [1001.1812]. For a sharp tridiagonal system, one introduces the polynomials
\[
\tau_i=(x-\theta_0)(x-\theta_1)\cdots(x-\theta_{i-1}),\qquad
\eta_i=(x-\theta_d)(x-\theta_{d-1})\cdots(x-\theta_{d-i+1}),
\]
and similarly \(\tau_i^*,\eta_i^*\) from the dual eigenvalues. The split sequence \(\{\zeta_i\}_{i=0}^d\) is defined by
\[
E_0^*\tau_i(A)E_0^*=
\frac{\zeta_iE_0^*}{(\theta_0^*-\theta_1^*)(\theta_0^*-\theta_2^*)\cdots(\theta_0^*-\theta_i^*)},
\]
with \(\zeta_0=1\). The parameter array of a sharp tridiagonal system is then
\[
\bigl(\{\theta_i\}_{i=0}^d;\{\theta_i^*\}_{i=0}^d;\{\zeta_i\}_{i=0}^d\bigr),
\]
and two sharp tridiagonal systems are isomorphic if and only if they have the same parameter array [1001.1812].

The classification theorem identifies exactly which parameter arrays occur. A sharp tridiagonal system exists with parameter array
\[
(\{\theta_i\}_{i=0}^d;\{\theta_i^*\}_{i=0}^d;\{\zeta_i\}_{i=0}^d)
\]
if and only if the eigenvalues are pairwise distinct, the two eigenvalue sequences satisfy the common ratio condition
\[
\frac{\theta_{i-2}-\theta_{i+1}}{\theta_{i-1}-\theta_i},\qquad
\frac{\theta_{i-2}^*-\theta_{i+1}^*}{\theta_{i-1}^*-\theta_i^*}
\]
for \(2\le i\le d-1\), and the split sequence satisfies
\[
\zeta_0=1,\qquad \zeta_d\ne 0,\qquad
\sum_{i=0}^d \eta_{d-i}(\theta_0)\eta_{d-i}^*(\theta_0^*)\zeta_i\ne 0.
\]
When these conditions hold, the system is unique up to isomorphism [1001.1812].

The proof runs through the universal algebra \(T(p,F)\) attached to feasible eigenvalue data and the \(\mu\)-conjecture. For feasible
\[
p=(\{\theta_i\}_{i=0}^d;\{\theta_i^*\}_{i=0}^d),
\]
one forms an algebra generated by \(a,\{e_i\},a^*,\{e_i^*\}\) with the tridiagonal vanishing relations
\[
e_i^*a^ke_j^*=0 \quad \text{if } k<|i-j|,\qquad
e_ia^{*k}e_j=0 \quad \text{if } k<|i-j|,
\]
and proves that the corner algebra \(e_0^*Te_0^*\) is a polynomial algebra in \(d\) variables. This yields the existence part of the classification and, over an algebraically closed field, a classification of all tridiagonal pairs up to isomorphism [1001.1812].

## 3. Split decompositions and canonical operators

A central refinement of the eigenspace picture is the first split decomposition
\[
U_i=(E_0^*V+E_1^*V+\cdots+E_i^*V)\cap(E_iV+E_{i+1}V+\cdots+E_dV),
\]
which satisfies
\[
V=\bigoplus_{i=0}^d U_i,
\]
together with
\[
(A-\theta_iI)U_i\subseteq U_{i+1},\qquad
(A^*-\theta_i^*I)U_i\subseteq U_{i-1}.
\]
There is also a second split decomposition
\[
U_i^\Downarrow=(E_0^*V+\cdots+E_i^*V)\cap(E_0V+\cdots+E_{d-i}V),
\]
again giving
\[
V=\bigoplus_{i=0}^d U_i^\Downarrow.
\]
These decompositions share the same lower partial sums with respect to the \(A^*\)-flag but encode different \(A\)-filtrations [1110.3434].

Bockting-Conrad introduced two canonical endomorphisms attached to any tridiagonal pair, \(\Delta\) and \(\Psi\). The operator
\[
\Delta=\sum_{h=0}^d F_h^\Downarrow F_h
\]
is characterized by
\[
(\Delta-I)V_i^*\subseteq V_0^*+\cdots+V_{i-1}^*,\qquad
\Delta(V_i+\cdots+V_d)=V_0+\cdots+V_{d-i},
\]
while \(\Psi\) is the unique operator satisfying
\[
\Psi V_i\subseteq V_{i-1}+V_i+V_{i+1},
\]
and
\[
(\Psi-\Lambda)V_i^*\subseteq V_0^*+\cdots+V_{i-2}^*
\]
with \(\Lambda=(\Delta-I)(\theta_0-\theta_d)^{-1}\). These operators commute:
\[
\Delta\Psi=\Psi\Delta,
\]
and, under the nonvanishing condition on \(\vartheta_1,\dots,\vartheta_d\), both \(\Delta\) and \(\Delta^{-1}\) are degree-\(d\) polynomials in \(\Psi\) [1110.3434].

For tridiagonal pairs of \(q\)-Racah type, the double lowering operator \(\psi\) lowers both split decompositions,
\[
\psi U_i\subseteq U_{i-1},\qquad \psi U_i^\Downarrow\subseteq U_{i-1}^\Downarrow,
\]
and the transition map \(\Delta\) from the first split decomposition to the second admits the factorization
\[
\Delta=\exp_q\!\left(\frac{a}{q-q^{-1}}\psi\right)
\exp_{q^{-1}}\!\left(-\frac{a^{-1}}{q-q^{-1}}\psi\right).
\]
The same work introduces an intermediate diagonalizable operator
\[
\mathcal M=\frac{aK-a^{-1}B}{a-a^{-1}},
\]
whose eigenspace decomposition is a “halfway point” between \(\{U_i\}\) and \(\{U_i^\Downarrow\}\) [1907.01157].

A later refinement distinguishes two raising/lowering formalisms: the decomposition
\[
A=R+F+L
\]
relative to the \(A^*\)-eigenspace decomposition, and the split maps \(\mathcal R,\mathcal L\) defined by
\[
\mathcal R|_{U_i}=A-\theta_iI,\qquad \mathcal L|_{U_i}=A^*-\theta_i^*I.
\]
The comparison is mediated by
\[
\Delta=\sum_{\ell=0}^d F_\ell E_\ell^*,
\]
which identifies \(E_i^*V\) with \(U_i\). Under this identification, \(R\) becomes \(\mathcal R\), \(F\) becomes a quadratic expression in \(\mathcal R,\mathcal L\), and \(L\) becomes a cubic expression in \(\mathcal R,\mathcal L\). The same comparison yields injectivity, surjectivity, and rank formulas for powers of \(R\) and \(L\), paralleling the classical results for \(\mathcal R\) and \(\mathcal L\) [2507.19400].

## 4. Thin cases, Leonard pairs, and concrete matrix models

A tridiagonal pair of shape
\[
(1,1,\dots,1)
\]
is a Leonard pair [1001.1812]. In this case every eigenspace is one-dimensional, the pair is multiplicity-free, and one can use the full parameter-array machinery. Leonard pairs form the sharp finite-dimensional case of the broader tridiagonal-pair notion, and many explicit matrix classifications are first obtained there [1404.6794].

A particularly rigid specialization is the totally bipartite case. A pair \(A,A^*\) is totally bipartite when the tridiagonal containments lose the middle term:
\[
A^*V_i\subseteq V_{i-1}+V_{i+1},\qquad
AV_i^*\subseteq V_{i-1}^*+V_{i+1}^*.
\]
In this regime every eigenspace is one-dimensional, so every totally bipartite tridiagonal pair is a Leonard pair. Its eigenvalue array
\[
\bigl(\{\theta_i\}_{i=0}^d;\{\theta_i^*\}_{i=0}^d\bigr)
\]
is classified by three conditions: distinct eigenvalues, a common recurrence
\[
\theta_{i-1}-\beta\theta_i+\theta_{i+1}=0,\qquad
\theta_{i-1}^*-\beta\theta_i^*+\theta_{i+1}^*=0,
\]
and antisymmetry
\[
\theta_i+\theta_{d-i}=0,\qquad \theta_i^*+\theta_{d-i}^*=0.
\]
The resulting closed forms fall into Krawtchouk, Bannai/Ito, and \(q\)-Racah families [1711.00332].

Another concrete realization is the lower-bidiagonal/tridiagonal, or LB-TD, form for Leonard pairs. Here \(A\) is lower bidiagonal with all subdiagonal entries equal to \(1\), while \(A^*\) is irreducible tridiagonal. Over an algebraically closed field, with \(q\) not a root of unity, the intrinsic criterion is that a Leonard pair has LB-TD form if and only if at least two of \(aa'\), \(bb'\), \(\xi\) are nonzero. In the same non-root-of-unity regime, LB-TD form occurs exactly for Leonard pairs of type \(q\)-Racah, \(q\)-Hahn, or dual \(q\)-Hahn [1404.6794].

Recent work on type II tridiagonal pairs develops a higher-multiplicity analogue of Leonard overlap coefficients. Two bases are constructed: in one, \(A\) is diagonal and \(A^\star\) is block tridiagonal; in the other, \(A^\star\) is diagonal and \(A\) is block tridiagonal. The change-of-basis coefficients are given by nested products of Racah-like special functions involving shift operators. For \(N=1\), the coefficients reduce to ordinary Racah-polynomial form [2503.01231].

## 5. Quantum-algebraic and combinatorial interfaces

A prominent specialization is the \(q\)-Serre type. In this case one may choose standard orderings so that
\[
\theta_i=q^{2i-d},\qquad \theta_i^*=q^{d-2i},
\]
and the pair satisfies the cubic \(q\)-Serre relations
\[
A^3A^*-[3]_qA^2A^*A+[3]_qAA^*A^2-A^*A^3=0,
\]
\[
A^{*3}A-[3]_qA^{*2}AA^*+[3]_qA^*AA^{*2}-AA^{*3}=0.
\]
Then \(V\) becomes an irreducible module for the positive part \(U_q^+\bigl(\widehat{\mathfrak{sl}}_2\bigr)\), with \(W_0=A\) and \(W_1=A^*\). The alternating elements of \(U_q^+\) act on the six canonical decompositions of \(V\) in a tightly controlled way: each acts in either a block diagonal, block upper bidiagonal, block lower bidiagonal, or block tridiagonal fashion [2207.07741].

The same \(q\)-Serre framework admits a one-parameter perturbation theory. If \(K\) denotes the split-decomposition operator and one defines
\[
B=A,\qquad B^*=tA^*+(1-t)K,
\]
then \((B,B^*)\) is again a tridiagonal pair if and only if
\[
t\neq 0,\qquad P\bigl(t(q-q^{-1})^{-2}\bigr)\neq 0,
\]
where \(P\) is the Drinfel'd polynomial attached to the original pair. In this case the perturbed split sequence is \(\zeta_i'=t^i\zeta_i\) [2107.01430].

For \(q\)-Racah type tridiagonal pairs, the double lowering operator \(\psi\) and the grading operators \(K,B\) yield two \(U_q(\mathfrak{sl}_2)\)-module structures on \(V\). In each, the Chevalley generator \(e\) acts as a scalar multiple of \(\psi\), and comparison of the two Casimir actions implies that \(\psi\) can be written as a rational function of \(K^{\pm1},B^{\pm1}\), while \(K\) and \(B\) satisfy a quadratic relation [1307.7410]. A further development constructs two \(\boxtimes_q\)-module structures, where \(\boxtimes_q\) is the \(q\)-tetrahedron algebra generated by four copies of \(U_q(\mathfrak{sl}_2)\); the two main ingredients are again \(\psi\) and the invertible operator \(W\), and the distinguished central element on the \(\boxtimes_q\)-module coincides with the TD-pair Casimir \(\Lambda\) [2006.02511].

The combinatorial origin remains central. Tridiagonal pairs were motivated by \(Q\)-polynomial distance-regular graphs [1711.00332]. An \(S_3\)-symmetric enlargement of the tridiagonal algebra has six generators arranged on a regular hexagon so that nonadjacent generators commute and adjacent generators satisfy tridiagonal relations. For a \(Q\)-polynomial distance-regular graph \(\Gamma\), the tensor power \(V^{\otimes 3}\) of the standard module becomes a module for this algebra, and the Hamming graph case is analyzed in detail [2407.00551].

## 6. Root-of-unity, higher-order, and tensorial extensions

The ordinary definition uses a linear ordering of eigenspaces. A root-of-unity extension replaces this by cyclic indexing. A cyclic tridiagonal pair of cyclicity \(N\) is an ordered pair \(C,C^*\) with eigenspaces indexed by \(\mathbb Z_N\) such that
\[
C^*V_p\subseteq V_{p-1}+V_p+V_{p+1},\qquad
CV_p^*\subseteq V_{p-1}^*+V_p^*+V_{p+1}^*,
\]
with indices understood modulo \(N\). A broad class arises from \(q\)-Racah tridiagonal pairs after specializing \(q\) to a primitive \(2N\)-th root of unity [1607.00606].

This root-of-unity regime produces divided polynomials
\[
A^{[N]}=\lim_{q\to e^{i\pi/N}}\frac{P_N(A;\{\theta\})}{[N]_q!},\qquad
A^{*[N]}=\lim_{q\to e^{i\pi/N}}\frac{P_N(A^*;\{\theta^*\})}{[N]_q!},
\]
which act as higher-order analogues of the tridiagonal partner. In an eigenbasis of \(A\), the operator \(A^{*[N]}\) becomes block \((2N+1)\)-diagonal, and dually for \(A^{[N]}\). The algebra generated by the divided polynomials is identified as a higher-order generalization of the Onsager algebra and can be viewed as a subalgebra of the \(q\)-Onsager algebra at a primitive \(2N\)-th root of unity [1607.00606].

Tensorial and higher-rank analogues also appear in the \(S_3\)-symmetric tridiagonal algebra, where six generators support multiple overlapping tridiagonal-algebra embeddings, and in type II tridiagonal pairs, where overlap coefficients between canonical eigen-bases take the form of multivariable Racah-like special functions with shift operators [2407.00551; 2503.01231]. This suggests that the standard two-operator framework extends naturally toward cyclic, tensorial, and multivariable settings, while retaining the characteristic tridiagonal constraints in modified form.

Source: https://www.emergentmind.com/topics/tridiagonal-pair