It is known that some matrices have infinite many square roots, some nilpotent
matrices have no root, and for some matrices we can build certain finite number
of roots according to known algorithms discussed previously. Necessary
and sufficient conditions for exitence of a square root of a matrix can be
found in Higham's monograph.
Theorem: A square n × n matrix A has a square root if and only if ib the
"ascent sequence" of integers \( d_1 , d_2 , d_3 , \ldots , \) defined by
Higham, Nicholas, Functions of Matrices: Theory and Computation
, SIAM, 2008, https://doi.org/10.1137/1.9780898717778
G.W. Cross, P. Lancaster, Square roots of complex matrices, Linear and Multilinear Algebra, 1,
(1974), pp. 289--293.
Åke Björck, Sven Hammarling, A Schur method for the square root of a matrix, Linear Algebra and its Applications,
52-53, July 1983, pp. 127--140.
Roger A. Horn, Charles R. Johnson, Matrix Analysis, Cambridge University Press, 2012.
Although
a square root could be determined for the majority of square matrices, most applications call for matrices with nonnegative
eigenvalues. It is natural to call such matrices positive-definite.
There are known several definitions of positive definite/semi-definite matrices, all of them are equivalent. However,
we start with the definition that is slightly different from universally accepted by mathematicians. Usually positive
definite matrices are symmetric or self-adjoint, but we forfeit this condition. The reason is that for our applications in engineering
it is not necessary. It is usually applied to the product \( {\bf A}^{\mathrm T} {\bf A} \)
or \( {\bf A}^{\ast} {\bf A} , \) which is a symmetric or self-adjoint matrix.
A square matrix A is called positive definite if all its eigenvalues are positive.
Similarly, A is called positive semi-definite if all its eigenvalues are not negative.
Theorem: When a matrix A has independent columns (= full column rank), the
product \( {\bf A}^{\mathrm T} {\bf A} \) is a square, symmetric, positive definite, and
invertible matrix. If A is not a full column rank matrix, then \( {\bf A}^{\mathrm T} {\bf A} \)
is a square, symmetric, and positive semi-definite matrix.
Matrix A is positive definite because its eigenvalues are \( \lambda = 1, 4, 9. \)
On the other hand, matrix B is positive semi-definite because its eigenevalues are
\( \lambda = 6, 1, 0 . \)
Since each of these matrices is diagonalizable (because they have three distinct eigenvalues), we can use the
Sylvester method. First, we calculate Sylvester's auxiliary matrices for A:
It has one defective double eigenvalue λ = 9 and one simple eigenvalue λ = 4. Its minimal polynomial
is the same as the characteristic polynomial \( \chi (\lambda ) = \left( \lambda -9 \right)^2 (\lambda -4) . \)
So we calculate the resolvent