Website owner: James Miller
Similarity, Similar matrices, Diagonable matrices, Orthogonal similarity, Real quadratic forms, Hermitian matrices, Normal matrices
Similar matrices. Two n-square matrices A and B over a field F are called similar if there exists a non-singular matrix P over F such that
(1) B = P -1AP
Note. Note that if B = P-1AP then A = PBP-1 . The condition of similarity can be defined by either formula. The formula B = P-1AP becomes B = QAQ -1 if we let Q = P-1.
Similarity is an equivalence relation that separates the set of all n-square matrices into equivalence classes. All matrices similar to a given matrix are similar to each other. What is the significance of this relationship of similarity between matrices? The answer is that any matrix similar to a given matrix represents the same linear transformation as the given matrix, but as referred to a different coordinate system (or basis). Thus any two matrices that are similar to each other represent the same point transformation in n-space i.e. they map points in the same way, they represent the same linear point transformation.
The concept of similarity is thus intricately connected to the concept of a change in basis, a change in coordinate system. Changing the basis for a linear transformation produces similar matrices.
Changes in the expression of a linear transformation due to a change in basis. Let Y = AX be a linear point transformation expressed with respect to the E-basis. What is the expression for this same transformation when expressed with respect to some arbitrary other Z-basis? In the Z-basis the vectors X and Y are
X = ZXZ Y = ZYZ .
Substituting into Y = AX we get
ZYZ = AZXZ
or
YZ = Z-1AZXZ .
Thus we see that the matrix Z-1AZ of the point transformation with respect to the Z-basis is similar to the matrix A of the transformation with respect to the E-basis.
All matrices that are similar to each other represent the same linear point transformation, but as referred to different frames of reference, or basis. Of great importance are those n-square matrices that are similar to diagonal matrices. Any matrix that is similar to a diagonal matrix represents the same linear point transformation as the diagonal matrix. What are the characteristics of the linear point transformation effected by a diagonal matrix? The answer is as follows.
Point transformation effected by a diagonal matrix. The point transformations effected by a diagonal matrix represent a very important class of linear transformations. Given the diagonal matrix
consider the point transformation effected by it in n-space (i.e. the transformations given by Y =
DX ). The transformations of this class have a very simple and intuitive geometric meaning (of
course, only for two and three-dimensional real spaces). If the numbers
are all positive the
effect of this transformation is simply stretching (or compressing) effects directed in the
directions of the different coordinate system axes
with magnitudes given by
. If some of the
are negative, then the deformation of the space is
accompanied by a change of direction. Finally, if for example
, then a projection of the
space parallel to
takes place onto the subspace spanned by
with a subsequent
deformation in these directions. If the coordinate system is an oblique one the transformation
will transform figures in space in this linear way in the directions of the oblique axes.
This class of transformations is important because in spite of its simplicity it is very general. In fact, it can be established that every linear transformation satisfying certain not very severe restrictions belongs to this class i.e. we can find for it a basis in which it is described by a diagonal matrix. [ excerpted from Mathematics, Its Content, Methods and Meaning. Vol. III. p. 80]
Diagonable matrices. A mattrix A which is similar to a diagonal matrix is called diagonable.
Under what conditions is an n-square matrix similar to a diagonal matrix? The following theorems address that question.
Theorems.
1] If all the eigenvalues of a matrix are distinct, then the matrix is similar to a diagonal matrix whose diagonal elements are these eigenvalues.
2] If an n-square matrix has n linearly independent eigenvectors, it is similar to a diagonal matrix.
3] An n-square matrix A is similar to a diagonal matrix if and only if the eigenvectors of A span Vn(F); when this is the case, the eigenvalues of A are the diagonal elements of D.
4] An n-square matrix A is similar to a diagonal matrix if and only if for each eigenvalue
of
A, the multiplicity of
is equal to the dimension of the null space of
I - A.
5] Two diagonal matrices are similar if and only if they differ only in the order of their diagonal elements.
6] Let A be an n-square matrix with n linearly independent eigenvectors p1, p2, ... , pn and let λ1, λ2, ... , λn be the eigenvalues corresponding to p1, p2, ... , pn. Let P be a matrix whose columns 1, 2, ... , n consist of p1, p2, ... , pn i.e.
P = [p1 p2 ... pn]
and let D be the diagonal matrix
whose diagonal elements consist of the eigenvalues λ1, λ2, ... , λn. Then matrix P is non-singular and
2) A = PDP-1
or, equivalently,
3) D = P-1AP .
Significance of Theorem 6. What is the significance of Theorem 6? Consider the linear point transformation
Y = AX
where matrix A has n distinct eigenvalues and thus also n corresponding linearly independent eigenvectors. We would like to know something about the nature of this linear point transformation effected by matrix A, just how it transforms figures in n-space. Could we perhaps get some insight into how it transforms figures by doing a change of coordinate system, transferring over to the coordinate system defined by the eigenvectors? Let us do a change of coordinate system, a change of basis, transferring over to the eigenvector coordinate system. From Theorem 6 we know that if Y = AX then Y = PDP-1X where P is a matrix whose columns consist of the eigenvectors and D is a diagonal matrix whose diagonal elements consist of the corresponding eigenvalues. If X' and Y' are the values of X and Y in the eigenvector coordinate system the equations relating the coordinates in the two systems are
X = PX'
Y = PY' .
Substituting into Y = PDP-1X we get
PY' = PDP-1PX'
or, equivalently,
4) Y' = DX' .
Thus in the eigenvector coordinate system the linear point transformation Y = AX is given by Y' = DX' , a transformation effected by the diagonal matrix D. What does this tell us? It tells us that the transformation of the figure by Y = AX consists of a distortion consisting of stretching (or compressing) in the directions of the various eigenvector axes by amounts given by the associated eigenvalues. It gives us an intuitive insight into the linear point transformation Y = AX.
Similar matrices and eigenvalues.
7] Two similar matrices have the same eigenvalues.
Relationship between the eigenvectors of two similar matrices. Let Xi
be an eigenvector of a matrix A corresponding to an eigenvalue
of A. Then an eigenvector Yi
of a similar matrix B = P -1AP corresponding to the same eigenvalue
of B is given by Yi = P -1X.
Eigenvalues and eigenvectors of diagonal matrices.
8] The eigenvalues of a diagonal matrix D = diag(
) are simply the diagonal
elements.
9] A diagonal matrix D = diag(
) always has n linearly independent eigenvectors.
The elementary unit vectors
..................
constitute a set of n linearly independent eigenvectors for D since DEi = aiEi, ( i =1, 2, ...., n).
General theorems.
10] Any n-square matrix A, similar to a diagonal matrix, has n linearly independent eigenvectors.
11] Every square matrix A is similar to a triangular matrix whose diagonal elements are the eigenvalues of A.
12] If A is any real n-square matrix with real eigenvalues, there exists an orthogonal matrix Q such that Q -1AQ = QTAQ is triangular and has as diagonal elements the eigenvalues of A The matrices A and Q -1AQ are called orthogonally similar.
13] If A is any n-square matrix with complex elements or a real n-square matrix with complex
eigenvalues, there exists a unitary matrix U such that
is triangular and has as
diagonal elements the eigenvalues of A. The matrices A and
are called unitarily
similar.
Synonyms.
Eigenvalue, characteristic root, latent root
Eigenvector, characteristic vector, invariant vector, latent vector
Eigenvector space, characteristic vector space, invariant vector space
Real symmetric matrices.
1] The characteristic roots (i.e. eigenvalues) of a real symmetric matrix are all real.
2] The invariant vectors (i.e. eigenvectors) associated with distinct characteristic roots of a real symmetric matrix are mutually orthogonal.
3] If A is a real n-square symmetric matrix with characteristic roots
, then there
exists a real orthogonal matrix P such that PTAP = P -1AP = diag(
) .
4] If
is a characteristic root of multiplicity ri of a real symmetric matrix, then there is
associated with
an invariant space of dimension ri .
Orthogonal similarity. If P is an orthogonal matrix and B = P -1AP. then B is said to be orthogonally similar to A. Since P -1 = PT, B is also orthogonally congruent and orthogonally equivalent to A.
Every real symmetric matrix A is orthogonally similar to a diagonal matrix whose diagonal elements are the characteristic roots of A.
Canonical set for real symmetric matrices under orthogonal
similarity. Let the characteristic roots of the real symmetric matrix A be arranged so that
. Then diag(
) is a unique diagonal matrix similar to A.
The totality of such diagonal matrices constitutes a canonical set for real symmetric matrices
under orthogonal similarity.
Two real symmetric matrices are orthogonally similar if and only if they have the same characteristic roots, that is if and only if they are similar.
Real quadratic forms.
1] Every real quadratic form q = XTAX can be reduced by an orthogonal transformation X = BY to a canonical form
where r is the rank of A and
are its non-zero characteristic roots.
2] A real symmetric matrix is positive definite if and only if all of its characteristic roots are positive.
3] If XTAX and XTBX are real quadratic forms in (
) and if XTBX is positive
definite, there exists a real non-singular linear transformation X = CY which carries XTAX into
and XTBX into
where
are the roots of
= 0 .
Hermitian matrices.
1] The characteristic roots of an Hermitian matrix are real.
2] The invariant vectors associated with distinct characteristic roots of an Hermitian matrix are mutually orthogonal.
3] If H is an n-square Hermitian matrix with characteristic roots
, there exists a
unitary matrix
such that
= diag(
) . The matrix
is called
unitarily similar to
.
4] If
is a characteristic root of multiplicity ri of the Hermitian matrix H, then there is
associated with
an invariant space of dimension ri .
Canonical set for Hermitian matrices under unitary similarity. Let the
characteristic roots of the Hermitian matrix H be arranged so that
. Then
diag(
) is a unique diagonal matrix similar to H. The totality of such diagonal
matrices constitutes a canonical set for Hermitian matrices under unitary similarity.
Two Hermitian matrices are unitarily similar if and only if they have the same characteristic roots, that is if and only if they are similar.
Normal matrices. An n-square matrix A is called normal if
. Normal
matrices include diagonal, real symmetric, real skew-symmetric, orthogonal, Hermitian, skew-Hermitian and unitary matrices.
1] If
is a normal matrix and
is a unitary matrix, then
is a normal matrix.
2] If
is an invariant vector corresponding to the characteristic root
of a normal matrix
,
then
is also an invariant vector of
corresponding to the characteristic root
.
3] A square matrix A is unitarily similar to a diagonal matrix if and only if A is normal.
4] If A is normal, the invariant vectors corresponding to distinct characteristic roots are orthogonal.
5] If
is a characteristic root of multiplicity ri of a normal matrix A, the associated invariant
space has dimension ri .
6] Two normal matrices are unitarily similar if and only if they have the same characteristic roots, that is if and only if they are similar.
References.
1. Ayres. Matrices (Schaum).
2. Mathematics, Its Content, Methods and Meaning. Vol. III