Exact Inversion of Pentadiagonal Matrix for Semi-Analytic Solution of 2D Poisson Equation ()
1. Introduction
Partial differential equations govern most phenomena and systems in physics and engineering. Laplace’s operator intervenes there very often. Once discretized, these differential equations are reduced to a matrix form, where the matrix corresponding to the Laplacian of the quantity to be studied takes the form of a pentadiagonal matrix. This matrix can be considered as a tridiagonal block matrix for the two-dimensional case. The inversion of this matrix makes it possible to obtain the solution of several categories of problems. There are different numerical and other approximation methods for inverting this matrix. The originality of this present work is to propose an exact and direct formula for the inverse of this famous pentadiagonal matrix.
Concretely, it will be a matter of determining the exact formula of the following symmetric block matrix M, which is a Toeplitz matrix:
, where
, and
.
D is a tridiagonal, symmetric, and persymmetric Toeplitz matrix of dimension (
). A is a diagonal Toeplitz matrix. We denote by
the null matrix of the same dimension as D. The parameters d, a, and
are reals.
The matrix M, on the other hand, is pentadiagonal. It can be described as a symmetric Toeplitz block matrix. Its dimension is
.
The inversion method we propose for the block matrix M follows a certain analogy with the approach used in [1] [2] [3] [4] [5] to invert the tridiagonal matrix D. The inverse of M has the following structure:
is a symmetric block matrix and the matrices
which compose it are of the same dimension as D.
(1)
is symmetric and persymmetric. It is composed of symmetric and persymmetric matrices (blocks):
(2)
Inverting M amounts to determining the blocks
. To obtain them, we will first recall the formulas for the inversion of D and those for its eigenvalues and its eigenvectors. We will then exploit these parameters of D using them as a key to determine
. Finally, we will formulate the exact expression for the inverse of M.
2. Inverse of Matrix D
Concerning the matrix D, its inverse is obtained with its cofactors. The inversion of a tridiagonal Toeplitz matrix has been covered in detail in [2] [3] [4] [5].
For the reals d and a, we define the real number
, and the complex number
, as follows:
(3)
Then, the following polynomial
makes it possible to calculate, in an explicit and direct way, the determinants of the sub-matrices
of D:
(4)
or in another form (for
):
(5)
The determinants of these sub-matrices
of D are given, for fixed a and b, by
. We denote by B the inverse of D. This matrix B exists when d is different from one of the following
values:
(6)
Apart from these values
, the matrix D is regular and its inverse B is symmetric, persymmetric, and is given by the following formula [2] [3] [4] [5]:
(7)
It yields
(8)
Which gives, in case
exists and is non-zero:
(9)
This result will be used to determine the inverse of the block matrix M.
3. Block Matrix Inversion
To find the inverse of the block matrix M, one can reason by analogy with the method of inversion of the tridiagonal matrix D. One of the keys to obtain this inverse matrix is to use the polynomial
. This time, we are not going to calculate it for real numbers d and a, but rather for matrices D and A. It is then necessary to determine the matrices
. The matrix D being Hermitian, its spectral decomposition is:
, where
is the diagonal matrix containing on its ith row the eigenvalue
:
.
(10)
X is an unitary matrix containing on its ith line the normalized eigenvector
associated with the eigenvalue
. The jth component of vector
is:
(11)
Applying the rule
, we can calculate
:
(12)
or
(13)
We can notice the spectral decomposition of the matrix
:
(14)
By analogy with the inversion of the tridiagonal matrix D, the inverse of the block matrix M is given by the matrices
, with:
(15)
We can, therefore, notice that the inverse of M exists when the determinant of
is non-zero.
From Equations (14) and (15), the spectral decomposition of the matrix
appears:
(16)
that is:
(17)
is the matrix containing on its ith line the ith eigenvalue of
. These eigenvalues of
are
, with:
(18)
where
:
(19)
Thus, we obtain all the elements of the matrices
composing the block matrix
:
(20)
Equation (20) completely determines the matrix
. Note that these
blocks are symmetric and persymmetric matrices. These different symmetries allow a calculation of the inverse of the pentadiagonal matrix M, in a fast, efficient, precise, and economical way in terms of occupying RAM space. This method allows a very fine implementation of the discretization of the Laplace operator, for 2D problems, even with an inhomogeneous mesh.
4. Conclusion
This work makes it possible to determine the exact formula of the inverse of the symmetric pentadiagonal Toeplitz matrix. This result allows a direct inversion of this matrix of blocks, in an efficient way, precise and economical way, in terms of occupying RAM space. This proposed method will allow simple and accurate resolution of a large category of physics and engineering problems, governed by differential equations involving the two-dimensional Lapalce operator.