Try   HackMD

Rayleigh quotient of a real symmetric matrix

Problem

Let

A be an
n×n
real symmetric matrix with its eigenvalues
λ1λn
. Show that for any
xRn
with
x=1
, we have

λ1xAxλn.

Moreover, when

xAx=λ1,
x
is an eigenvector of
A
with respect to
λ1
; when
xAx=λn
,
x
is an eigenvector of
A
with respect to
λn
.

Thought

According to the spectral theorem, a real symmetric matrix is similar to a diagonal matrix by an orthonormal basis. That is, there is an orthonormal basis

β such that the matrix
A
, when observed by
β
, has its matrix representation diagonal. Intuitively, the proof should be similar to the case when
A
is a diagonal matrix.

Sample answer

By the spectral theorem, there is an orthonormal basis

β={u1,,un} composed of eigenvectors of
A
. We may assume that
ui
is the eigenvector corresponding to
λi
for
i=1,,n
.

Since

β is an orthonormal basis of
Rn
, we may write

x=c1u1++cnun

for some scalars

c1,,cn. As
ui
's are eigenvectors, we also have

Ax=c1λ1u1++cnλnun.

The fact that

β is orthonormal means that
ui,uj=0
for
ij
and
ui,ui=1
for any
i
. By the distributive law of the inner product, we may compute

xAx=x,Ax=c1u1++cnun,c1λ1u1++cnλnun=i=1nci2λiui,ui+ijcicjλjui,uj=c12λ1++cn2λn

and

x2=x,x=c1u1++cnun,c1u1++cnun=i=1nci2ui,ui+ijcicjui,uj=c12++cn2.

By replacing

ci2=pi for
i=1,,n
, we have

xAx=p1λ1++pnλn,x2=p1++pn=1,

and

pi0 for
i=1,,n
. Thus,
xAx
is the weighted average of
λ1,,λn
with respect to the weights
p1,,pn
. Consequently, the value of
xAx
is between the minimum and the maximum of
{λ1,,λn}
, which are
λ1
and
λn
, respectively. When
xAx=λ1
, necessarily all the weight is attributed to those
λi
with
λi=1
, and
x
is a linear combination of eigenvectors with respect to
λ1
, which is again an eigenvector with respect to
λ1
. Similarly, when
xAx=λn
,
x
is an eigenvector with respect to
λn
.

This note can be found at Course website > Learning resources.