Generalized Krawtchouk Matrices

• The generalized Krawtchouk matrices are matrix embodiments of Krawtchouk-type orthogonal polynomial systems, with entries defined by generating functions and basis-change coefficients.
• They exhibit discrete orthogonality, weighted transpose symmetries, and recurrence relations that make them ideal for finite transform and spectral decomposition applications.
• Extensions include multivariate, q-deformed, and super versions, linking these matrices to coding theory, probabilistic models, and representation theory.

Generalized Krawtchouk matrices are matrix-valued realizations of Krawtchouk-type orthogonal polynomial systems in which the entries are organized as coefficients of generating functions, basis-change coefficients, or matrix elements of finite-dimensional representations. In the literature represented here, the term covers several related constructions: the classical univariate Krawtchouk matrix; asymmetric and weighted variants; multivariate matrices attached to multinomial distributions; matrix families induced from Hadamard, orthogonal, or symmetric-power constructions; and $q$-deformed or super extensions. Their common features are discrete orthogonality, weighted transpose symmetries, recurrence and difference structures, and an interpretation as exact finite transforms or spectral decompositions (Feinsilver, 2016).

1. Classical matrix form and the matrix viewpoint

A classical starting point is to regard Krawtchouk polynomials not as isolated scalar polynomials but as the entries of a square matrix indexed by degree and evaluation point. For general parameter $p$, with

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$

the generating function

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$

defines the $(n,j)$-entry of a Krawtchouk matrix of order $N$. In the symmetric case $p=\tfrac12$, so $r=1$, one writes

$$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$

and the resulting matrix $\Phi^N$ is the symmetric Krawtchouk matrix (Feinsilver, 2016).

This matrix viewpoint is closely related to older structural descriptions. In the $p$0 framework, the matrix $p$1 is the transition matrix between two weight bases in the irreducible $p$2-dimensional module $p$3; the identity

$p$4

makes the Krawtchouk entries literal change-of-basis coefficients rather than merely evaluations of special functions (Nomura et al., 2012). A parallel formulation appears in the symmetric-tensor-power picture, where the $p$5-th Krawtchouk matrix is $p$6, the symmetric tensor power of the Sylvester-Hadamard matrix $p$7. In that setting the generating relation

$p$8

is the defining column formula (Kocik, 2016).

The classical matrix already exhibits the duality properties that motivate later generalizations. The univariate Krawtchouk polynomials satisfy the self-duality

$p$9

and the symmetric matrix satisfies weighted transpose relations such as

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$0

with $r=\frac{1-p}{p},\qquad \frac1p=1+r,$1, as well as the involutory identity

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$2

in the symmetric case [(Nomura et al., 2012); (Feinsilver, 2016)]. These formulas make the classical Krawtchouk matrix the prototype for the broader generalized theory.

2. Structural identities, orthogonality, and transform theory

Once Krawtchouk polynomials are written as matrix entries, many of their structural relations become row-wise and column-wise matrix identities. The generating function immediately yields Pascal-type recurrences,

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$3

together with the column-index recurrence

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$4

In the symmetric case this reduces to

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$5

which drives the partial-sum identities developed for columns of $r=\frac{1-p}{p},\qquad \frac1p=1+r,$6 (Feinsilver, 2016).

The same matrix formalism yields explicit orthogonality and inversion formulas for the Krawtchouk transform. With $r=\frac{1-p}{p},\qquad \frac1p=1+r,$7, diagonal weight matrix

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$8

and diagonal norm matrix

$r=\frac{1-p}{p},\qquad \frac1p=1+r,$9

one has

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$0

A refined inversion identity is

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$1

which in the symmetric case $$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$2, $$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$3, collapses to

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$4

This is the finite discrete analogue of a Fourier involution (Feinsilver et al., 2014).

The transform point of view extends beyond orthogonality to convolution. If

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$5

then

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$6

where the convolution coefficients are extracted from the linearization formula for products of Krawtchouk polynomials. In the symmetric case,

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$7

and inversion becomes

$$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$8

This places Krawtchouk matrices within an inherently discrete transform calculus rather than only an orthogonal-polynomial setting (Feinsilver et al., 2014).

3. Multivariate and matrix-defined generalizations

A major generalization replaces the one-dimensional binomial support by multinomial state spaces and replaces scalar degree by multi-indices. One systematic construction begins with a $$(1+z)^{N-j}(1-rz)^j=\sum_{n=0}^N z^n\,\phi_{nj}^{N}$$9 matrix $(n,j)$0, commuting variables $(n,j)$1, and linear forms

$(n,j)$2

At fixed total degree $(n,j)$3, the induced matrix $(n,j)$4 is defined by

$(n,j)$5

This symmetric-power construction is functorial,

$(n,j)$6

and satisfies the weighted transpose identity

$(n,j)$7

where $(n,j)$8 is the diagonal matrix of multinomial coefficients. If $(n,j)$9 with $N$0 orthogonal, then

$N$1

lifts to the multivariate Krawtchouk orthogonality relation

$N$2

and the multivariate Krawtchouk polynomials are the entries $N$3 (Feinsilver et al., 2011). The same framework reappears in Krawtchouk-Griffiths systems, where $N$4 is explicitly identified as the Krawtchouk matrix and its entries as values of multivariate Krawtchouk polynomials orthogonal with respect to a multinomial distribution (Feinsilver, 2016).

A probabilistic generalization appears in the multitype cumulative Bernoulli trial model. There the state is a composition $N$5, the stationary law is multinomial,

$N$6

and the eigenfunctions of the transition kernel are multivariable Krawtchouk polynomials realized as $N$7-type hypergeometric functions. The orthogonality and spectral decomposition depend on parameter constraints

$N$8

and the eigenvalues take the multiplicative form

$N$9

In this setting the generalized Krawtchouk matrix is the spectral matrix of the Markov kernel (Grünbaum et al., 2011).

Another extension replaces the classical binary seed by an arbitrary $p=\tfrac12$0 matrix $p=\tfrac12$1. For weak compositions $p=\tfrac12$2, the $p=\tfrac12$3-polynomials $p=\tfrac12$4 are defined through the generator

$p=\tfrac12$5

When $p=\tfrac12$6 is a generalized Hadamard matrix,

$p=\tfrac12$7

they satisfy the orthogonality relation

$p=\tfrac12$8

For $p=\tfrac12$9 and

$r=1$0

the construction reduces to the classical binary Krawtchouk family. This matrix-defined theory generalizes the generator function, the orthogonality relation, and the finite transform viewpoint simultaneously (Chami et al., 2013).

A different high-rank construction, based on symmetric cones and partition indexing, defines multivariate Krawtchouk polynomials $r=1$1 through generalized binomial coefficients and spherical polynomials. Its determinant formula

$r=1$2

expresses each multivariate polynomial as a determinant of one-variable Krawtchouk polynomials. This provides a precise sense in which a generalized Krawtchouk matrix can be assembled from lower-dimensional Krawtchouk data (Shibukawa, 2014).

4. Group-theoretic, Lie-theoretic, and oscillator realizations

Generalized Krawtchouk matrices often arise as exact matrices of group actions or as transition matrices between distinguished bases. In the $r=1$3 setting, the operators

$r=1$4

form a Leonard pair of Krawtchouk type, and the matrix $r=1$5 simultaneously diagonalizes one operator and tridiagonalizes the other. The matrix equations

$r=1$6

encode the three-term recurrence and the dual difference equation in a purely matrix-theoretic form (Nomura et al., 2012).

For multivariate polynomials, the three-dimensional isotropic oscillator provides a canonical representation-theoretic model. On the fixed-energy subspace of the $r=1$7-dimensional oscillator, a rotation $r=1$8 acts by

$r=1$9

Its matrix elements in the Cartesian basis factor as

$$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$0

where $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$1 are bivariate Krawtchouk polynomials and $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$2 is the trinomial weight factor. In this formulation, the generalized Krawtchouk matrix is the change-of-basis matrix between rotated oscillator bases (Genest et al., 2013).

The same idea extends to the isotropic $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$3D harmonic oscillator in a more explicit interbasis-expansion form. There the matrix

$$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$4

has entries equal to a multinomial normalization factor times general bivariate Krawtchouk polynomials, and its decomposition through spherical bases involves $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$5 Wigner $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$6-matrices and $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$7 Clebsch-Gordan coefficients, hence dual Hahn polynomials. This places the generalized Krawtchouk matrix directly inside a reducible $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$8-representation (Genest et al., 2013).

The oscillator construction generalizes naturally to $$(1+z)^{N-j}(1-z)^j=\sum_{n=0}^N z^n\,\Phi_{nj}^{N},$$9 dimensions. For $\Phi^N$0, one obtains

$\Phi^N$1

with multinomial orthogonality and $\Phi^N$2 effective parameters inherited from the rotation matrix. This shows that multivariate Krawtchouk matrices are not merely coefficient tables; they are matrix elements of unitary reducible representations on oscillator energy shells (Genest et al., 2013).

5. $\Phi^N$3-deformed and super extensions

The matrix-element paradigm persists under deformation. In the one-variable quantum case, a unitary $\Phi^N$4-rotation operator in the Schwinger realization of $\Phi^N$5,

$\Phi^N$6

has matrix elements

$\Phi^N$7

proportional to the quantum $\Phi^N$8-Krawtchouk polynomials. Unitarity gives orthogonality, and the matrix enjoys the self-duality

$\Phi^N$9

The resulting finite unitary matrix is a $p$00-deformed Krawtchouk transform matrix (Genest et al., 2014).

In the multivariate quantum case, successive $p$01-rotations acting on tensor products of $p$02-oscillator states produce matrix elements proportional to multivariate quantum $p$03-Krawtchouk polynomials. For the two-variable case,

$p$04

and the general $p$05-variable case is obtained from a chain of adjacent-plane $p$06-rotations. Orthogonality, duality, recurrence relations, and difference equations all descend from operator identities for these matrices (Genest et al., 2015).

A 2026 extension replaces ordinary polynomial algebras by superalgebras. Super Krawtchouk polynomials are introduced in even, odd, and mixed families using commuting variables $p$07 and Grassmann variables $p$08, with odd relations

$p$09

The central viewpoint is again matrix-theoretic: on the polynomial module $p$10, the standard basis $p$11 and the transformed basis $p$12 are related by a basis-change matrix whose coefficients are the super Krawtchouk polynomials. The odd sector has an explicit determinant formula

$p$13

and the total transition matrix is block diagonal, with classical even, purely odd, and mixed blocks. This gives a Lie-superalgebraic version of the generalized Krawtchouk matrix (Iliev et al., 5 May 2026).

6. Applications, associated algebras, and unifying patterns

Generalized Krawtchouk matrices enter several distinct application domains. In transform theory they provide a discrete alternative to Fourier analysis, with explicit inversion, convolution, shift-operator realizations, and anti-diagonalization in binomial bases (Feinsilver et al., 2014). In coding theory, the same matrix relations underlie MacWilliams-type transforms and Delsarte-style positivity statements; in the symmetric case the matrix acts as a scaled involution (Feinsilver et al., 2011).

In probability, the multivariate Krawtchouk basis diagonalizes the transition kernel of the $p$14-type cumulative Bernoulli trial chain, so the generalized Krawtchouk matrix is the spectral data of a multinomial Markov process (Grünbaum et al., 2011). A different probabilistic reinterpretation comes from discrete path sums: the classical Krawtchouk matrix can be read as a path-integral amplitude on a Galton-board-type lattice, and the weighted extension

$p$15

defines ring-valued or phase-valued generalized Krawtchouk matrices. The same paper identifies the seed Hadamard matrix with a split-quaternion element and derives spectral decompositions from this algebraic model (Kocik, 2016).

A particularly arithmetic line of development appears in the study of sums of squares and Boolean-lattice algebras. For the symmetric Krawtchouk matrix,

$p$16

so complete column sums of squares are governed by binomial expressions that the paper identifies with a Catalan or Super Catalan connection. The same work records special entries equal to Catalan numbers, such as

$p$17

and applies these identities to $p$18-algebras over the Boolean lattice built from zeons. With

$p$19

the eigenvalue of $p$20 on the layer $p$21 is $p$22, exactly the same linear form that appears in Krawtchouk recurrences. For the algebra generated by $p$23 and $p$24, the centralizer dimension is

$p$25

This makes the Krawtchouk matrix a computational device for the representation theory of Boolean-lattice operator algebras rather than only a polynomial table (Feinsilver, 2016).

Across these constructions, several recurring patterns define the subject. First, generalized Krawtchouk matrices are typically coefficient matrices of generating functions or basis-change matrices between two distinguished bases. Second, their orthogonality is expressed through weighted matrix identities such as $p$26 or $p$27. Third, they often arise as exact spectral matrices of finite transforms, Markov kernels, or representation operators. Fourth, deformation is systematic: classical, multivariate, Hadamard, $p$28-deformed, and super versions preserve the same matrix logic while changing the underlying algebraic seed. This suggests a unifying interpretation in which a generalized Krawtchouk matrix is the finite-dimensional matrix avatar of a discrete orthogonal-polynomial system together with its natural symmetries, spectral data, and operator calculus [(Feinsilver et al., 2014); (Iliev et al., 5 May 2026)].