1.2
Linear
Operators
and
Matrices
5
every
matrix
A is
unitarily
equivalent
(or
unitarily
similar)
to
an
up-
per
triangular
matrix
T,
i.e., A =
QTQ*,
where Q
is
unitary
and
T is
upper
triangular.
This
triangular
matrix
is called a
Schur
Triangular
Form
for
A.
An
orthonormal basis
with
respect
to
which A is
upper
triangular
is
called a
Schur
basis
for A.
If
A is normal,
then
T
is
diagonal
and
we have
Q*
AQ
=
D,
where D is a diagonal
matrix
whose diagonal entries are
the
eigenvalues of A.
This
is
the
Spectral
Theorem
for normal matrices.
The
Spectral
Theorem
makes
it
easy
to
define functions of normal
matri-
ces.
If
f is any complex function,
and
if D is a diagonal
matrix
with
A!, .
..
,
An
on
its
diagonal,
then
f(D)
is
the
diagonal
matrix
with
f(A1),
...
,
f(An)
on
its
diagonal.
If
A = QDQ*,
then
f(A)
=
Qf(D)Q*.
A special conse-
quence, used very often,
is
the
fact
that
every positive
operator
A
has
a
unique positive
square
root.
This
square
root
will
be
written
as A
1/2.
Exercise
1.2.2
Show that the following statements are equivalent:
(i) A is positive.
(ii) A = B* B for some
B.
(iii) A =
T*T
for some upper triangular
T.
(iv) A =
T*T
for some upper triangular T with nonnegative diagonal
entries.
If
A is positive definite, then the factorisation
in
(iv) is unique. This is
called the
Cholesky
Decomposition
of
A.
Exercise
1.2.3
(i) Let
{AnJ
be
a family
of
mutually commuting operators.
Then, there is a common Schur basis for
{An}.
In
other words, there exists
a unitary
Q such that
Q*
AnQ
is upper triangular for all
Q.
(ii) Let
{An}
be
a family
of
mutually commuting normal operators. Then,
there exists a unitary
Q such that
Q*
AnQ
is diagonal for all
Q.
For any
operator
A
the
operator
A*
A
is
always positive,
and
its
unique
positive square
root
is
denoted by
IAI.
The
eigenvalues of
IAI
counted
with
multiplicities are called
the
singular
values
of A. We will always enu-
merate
these in decreasing order,
and
use for
them
the
notation
s 1 (A)
~
s2(A)
~
...
~
sn(A).
If
rank
A = k,
then
sk(A)
> 0,
but
Sk+1(A) =
...
=
sn(A)
=
O.
Let S
be
the
diagonal
matrix
with
diagonal entries
Sl
(A),
...
,
sn(A)
and
S+
the
k x k
diagonal
matrix
with
diagonal entries
sl(A),
...
,Sk(A).
Let
Q = (Q1,Q2)
be
the
unitary
matrix
in which
Q1
is
the
n x k
matrix
whose columns
are
the
eigenvectors of A* A corresponding
to
the
eigenvalues
si(A),
...
, s%(A)
and
Q2
the
n x
(n
-
k)
matrix
whose columns are
the
eigenvectors of
A*
A
corresponding
to
the
remaining eigenvalues.
Then,
by
the
Spectral
Theorem
Q*(A*
A)Q
=
(SO~
~).