ITALIAN JOURNAL OF PURE AND APPLIED MATHEMATICS – N.

48–2022 (1235–1241) 1235

Matrices of infinite dimensions and their applications

Lukasz Matysiak∗
Kazimierz Wielki University
Kazimierz Wielki University
ul. Powstańców Wielkopolskich 2
85-090 Bydgoszcz
Poland
lukmat@ukw.edu.pl
Weronika Przewoźniak
Kazimierz Wielki University
ul. Powstańców Wielkopolskich 2
85-090 Bydgoszcz
Poland
weronika.przewozniak@student.ukw.edu.pl
Natalia Rulińska
ul. Powstańców Wielkopolskich 2
85-090 Bydgoszcz
Poland
natalia.rulinska@student.ukw.edu.pl

Abstract. Matrices are very popular and widely used in mathematics and other fields
of science. Every mathematician has known the properties of finite-sized matrices since
the time of study. In this paper, we consider the basic theory of infinite matrices. So
far, there have been references and few results in certain scientific fields, but they have
not been thoroughly researched.
Keywords: matrix, determinant, inverse matrix, rank.

1. Introduction

Infinite matrices, the forerunner and a main constituent of many branches of

classical mathematics (infinite quadratic forms, integral equations, differential
equations, etc.) and of the modern operator theory, is revisited to demonstrate
its deep influence on thedevelopment of many branches of mathematics, classical
and modern, replete with applications.
It is known that we can add matrices, multiply by scalar, multiply matrices,
calculate determinant, calculate inverse matrix, determine rank matrix. We can
find all these properties of matrices in many basic academic books, for example
in [1], [2], [3].

*. Corresponding author
1236 LUKASZ MATYSIAK, WERONIKA PRZEWOŹNIAK and NATALIA RULIŃSKA

Matrices with an infinite number of rows and/or columns are also considered
- formally, it is sufficient that for any elements indexing rows and columns there
is a well-defined matrix element (index sets do not even have to be subsets of
natural numbers). Similarly to the finite case, we can define addition, subtrac-
tion, multiplication by scalar or matrix shifting, although matrix multiplication
requires some assumptions.
Applications of matrices are found in most scientific fields ([5]). In every
branch of physics, including classical mechanics, optics, electromagnetism, quan-
tum mechanics, and quantum electrodynamics, they are used to study physical
phenomena, such as the motion of rigid bodies. In computer graphics, they are
used to manipulate 3D models and project them onto a 2-dimensional screen. In
probability theory and statistics, stochastic matrices are used to describe sets of
probabilities. For example, they are used within the PageRank algorithm that
ranks the pages in a Google search. ([4]) Matrix calculus generalizes classical
analytical notions such as derivatives and exponentials to higher dimensions.
Matrices are used in economics to describe systems of economic relationships.
A major branch of numerical analysis is devoted to the development of effi-
cient algorithms for matrix computations, a subject that is centuries old and is
today an expanding area of research. Matrix decomposition methods simplify
computations, both theoretically and practically. Algorithms that are tailored
to particular matrix structures, such as sparse matrices and near-diagonal ma-
trices, expedite computations in finite element method and other computations.
Infinite matrices occur in planetary theory and in atomic theory. A simple ex-
ample of an infinite matrix is the matrix representing the derivative operator,
which acts on the Taylor series of a function.
In this paper, we formalize and develop the basic theory of infinite matrices.
So far, they have not been thoroughly researched, despite their significant use
in some fields of science.

2. Results
By an infinite dimension matrix we call a matrix for which the number of rows
is infinite or the number of columns is infinite.
We define zero, triangular, diagonal, unitary and transposed matrices of an
infinite dimension very analogously.
A square matrix of an infinite dimension is a matrix in which the number of
rows is equinumerous to the number of columns.
Matrix sum and by scalar multiplication are also analogous.
Corollary 2.1. If we try to multiply matrix Am×n with matrix Bn×k , we get
the following conclusions:
(a) If m = ∞, k = ∞, then AB = C∞×∞ .
P∞
(b) If n = ∞, then AB = Cm×k = [cij ], where cij = l=1 ail blj (1 ⩽ i ⩽ m,
1 ⩽ j ⩽ k) be a convergent series.
MATRICES OF INFINITE DIMENSIONS AND THEIR APPLICATIONS 1237

(c) If A, B be square matrices of an infinite dimension, then AB = C holds.

(d) If A, B be matrices of an infinite dimension and the number of rows in

A is equinumerous to the number of columns in B, then we can multiply
matrices A and B only if rows of A and columns of B be a convergent
series.

Let M1 (∞, R) = M1 (R) be denote the set of all square matrices of an in-
finite dimension with coefficients from any integral domain R, where all rows
and columns are convergent series. Then M1 (∞, R) be a ring. Easy to check
that {A ∈ M1 (∞, Z) : det A ∈ {−1, 1}} and {A ∈ M1 (∞, Z) : det A = 1} are
multiplicative groups.
The determinant of a square matrix A of finite dimension can be easily
determined by the formula:

det A = det(exp(log A)) = exp(tr(log A)),

P∞ Ak k+1
where log A = k=1 (−1). For an infinite dimension we must add the
k
assumption that tr(log A) be a convergent series.

Proposition 2.1. Let A be an m × n matrix, and let B be an matrix n × m,

where m, n ∈ N ∪ {∞}. Let 1 ⩽ j1 , j2 , . . . , jm ⩽ n. Let Aj1 j2 ...jm denote the
m × m matrix consisting of columns j1 , j2 , . . . , jm of A. Let Bj1 j2 ...jm denote the
m × m matrix consisting of rows j1 , j2 , . . . , jm of B. Then

X
det(AB) = det(Aj1 j2 ...jm ) det(Bj1 j2 ...jm ).
1⩽j1 <j2 <···<jm ⩽n

Proof. First we will show the proof in the finite version.

Let (k1 , k2 , . . . , km ) be an ordered m-tuple of integers. Let η(k1 , k2 , . . . , km ) de-
note the sign of (k1 , k2 , . . . , km ). Let (l1 , l2 , . . . , lm ) be the same as (k1 , k2 , . . . , km )
except for ki and kj having been transposed. Then, from Transposistion is of
Odd Parity:
η(l1 , l2 , . . . , lm ) = −η(k1 , k2 , . . . , km ).

Let (j1 , j2 , . . . , jm ) be the same as (k1 , k2 , . . . , km ) by arranged into non-decreasing

order. That is j1 ⩽ j2 ⩽ · · · ⩽ jm . Then, it follows that:

det(Bk1 ...km ) = η(k1 , k2 , . . . , km ) det(Bj1 ...jm ).

1238 LUKASZ MATYSIAK, WERONIKA PRZEWOŹNIAK and NATALIA RULIŃSKA

Hence:
X Xn Xn
det(AB) = η(l1 , . . . , lm )( a1k bkl1 ) . . . ( amk bklm ) =
1⩽l1 ,...,lm ⩽m k=1 k=1
X X
= a1k1 . . . amkm η(l1 , . . . , lm )bk1 l1 . . . bkm lm =
1⩽k1 ,...,km ⩽n 1⩽l1 ,...,lm ⩽m
X
= a1k1 . . . amkm det(Bk1 ...km ) =
1⩽k1 ,...,km ⩽n
X
= a1k1 η(k1 , . . . , km ) . . . amkm det(Bj1 ...jm ) =
1⩽k1 ,...,km ⩽n
X
= det(Aj1 ...jm ) det(Bj1 ...jm ).
1⩽j1 ⩽j2 ⩽···⩽jm ⩽n

If two js are equal:

det(Aj1 ...jm ) = 0.
For an infinite matrices we put ∞-tuple in proof in the form (k1 , k2 , k3 , . . . ).
And put 1 ⩽ j1 , j2 , j3 , · · · < n = ∞.

Corollary 2.2. If m = n (m, n ∈ N ∪ {∞}), then

det(AB) = det(A) det(B).

The following two Propositions give us a way to compute the inverse matrix.

Proposition 2.2. Let A be a matrix in which every rows and colums form
convergent series such that ||I − A|| < 1, where || · || is a submultiplicative norm.
Then
A−1 = I + (I − B) + (I − B)2 + . . .

Proof. A matrix A ∈ M1 (K) (where n ∈ N ∪ {∞}, K be a field) is invertible

if and only if the map f : K n → K n defined by f (x) = Ax is invertible, where
elements of K n are considered as column vectors.

3. Applications
Let A be an m × n matrix over an arbitrary field F (m, n ∈ N ∪ {∞}). There
is an associated linear mapping f : F n → F m defined by f (x) = Ax. The rank
of A is the dimension of the image f . This definition has the advantage that it
can be applied to any linear map without need for a specific matrix.
Let 


 a11 x1 + a12 x2 + a13 x3 + . . . = b1

a x + a x + a x + . . . = b
21 1 22 2 23 3 2


 a31 x1 + a32 x2 + a33 x3 + . . . = b3

. . .
MATRICES OF INFINITE DIMENSIONS AND THEIR APPLICATIONS 1239

(AX = B) be a system of equations. By Cramer’s system we mean a system

in which the number of equations is equinumerous to the number of unknowns.
Then, Cramer’s theorem states that in finite case the system has a unique so-
lution provided we have n equations. Hence individual values for the unknowns
are given by:
det Ai
xi = ,
det A
for i = 1, 2, . . . , where Ai is the matrix formed by replacing the i-th column of
A by the column vector B. If tr Ai , tr A are convergent series, then Cramer’s
formula holds for infinity case.
In other hand, system of equations AX = B (A, X, B can have an infinite
dimension) implies X = A−1 B.

Theorem 3.1 (Rouché-Capelli Theorem (Kronecker-Capelli Theorem)). Let

m, n ∈ N ∪ {∞}. A system of m linear equations in n variables Ax = b is
compatible if and only if both the incomplete and complete matrices (A and
[A|b] respectively) are characterised by the same rank A = rank [A|b].

Proof. Let m, n ∈ N ∪ {∞}. The system of linear equations Ax = b can

be interpreted as a linear mapping f : F n → F m , by f (x) = Ax, such that
A ∈ M (n, R), R be an integral domain.
This system is determined if one solution exists, t.e. if there exists x0 such
that f (x0 ) = b. This means that the system is determined if b ∈ Im(f ).
The basis spanning the image vector space (Im(f ), +, ·) is composed of the
column vectors of the matrix A:

BIm(f ) = {I1 , I2 , . . . , In }, A = (I1 I2 . . . In ).

Thus, the fact that b ∈ Im(f ) is equivalent to the fact that b belongs to the
span of the column vectors of the matrix A:

b = (I1 , I2 , I3 , . . . ).

This is equivalent to say that the rank of

A = (I1 I2 I3 . . . )

and
[A|b] = (I1 I2 I3 . . . b)
have the same rank. Thus, the system is compatible if rank A = rank [A|B].

Let B = {v1 , v2 , v3 , . . . }, B ′ = {u1 , u2 , u3 , . . . }. Then, for i = 1, 2, 3, . . . we

(i) (i) (i)
compute coordinates α1 , α2 , α3 , . . . of the basis vector B ′ in basis B:
∞
(i)
X
ui = αj vj .
j=1
1240 LUKASZ MATYSIAK, WERONIKA PRZEWOŹNIAK and NATALIA RULIŃSKA

Hence a transition matrix is of the form:

 (1) (2) (3)

α1 α1 α1 ...
 (1) (2) (3) 
 α2 α2 α2 ... 
 (1) (2) (3)
.
 α α α . . . 
 3 3 3 
.. .. .. ..
. . . .

Let L : U → V be a linear transformation, where U , V be a linear spaces

such that dim U = m, dim V = n (m, n can be ∞) and a basis of U be
{u1 , u2 , . . . , um }, a basis of V be {v1 , v2 , . . . , vn }. For i = 1, 2, . . . , m and
j = 1, 2, . . . , n compute
n
(i)
X
L(ui ) = αj vj .
j=1

Then a transformation matrix of transformation L is of the form:

 (1) (2) (n)

α1 α1 . . . α1
 (1) (2) (n) 
 α2 α2 . . . α2 
.
 . .. .. 
 . ..
 . . . . 
(1) (2) (n)
αm αm . . . αm

We will try to find the eigenvalues and eigenvectors of the infinity matrix.
Solve a characteristic equation:

det(A − λI) = 0.

So, we have to calculate

det(A − λI) = exp(tr(log(A − λI))),

where tr log(A − λI) be a convergent series.

For an appropriate eigenvalue λ, we find the corresponding eigenvector v =
(x1 , x2 , x3 , . . . ) from the system of equations:
   
x1 0
 x2   0 
(A − λI)  = .
   
 x3   0 
.. ..
. .

Let A be a matrix of any dimension, whose rows are given linearly indepen-
dent vectors. We are building a block matrix [AAT | A]. Applying elementary
row operations we bring it to the block matrix of the form [G | A′ ], where G be
the upper triangular matrix. The rows of A′ form orthogonal vectors.
MATRICES OF INFINITE DIMENSIONS AND THEIR APPLICATIONS 1241

References
[1] A. Cayley, A memoir on the theory of matrices, 1855.

[2] Z. Opial, Algebra wyższa, PWN, 1975.

[3] A. Bialynicki-Birula, Algebra liniowa z geometria, PWN, BM, 48 (1979).

[4] K. Bryan, T. Leise, The 25,000,000,000 eigenvector: the linear algebra

behind Google, SIAM Review, 48 (2006), 569-581.

[5] Matrix | mathematics, Encyclopedia Britannica, Retrieved 2020-08-19.

Accepted: August 18, 2021