Eigenvalues of Gramian Matrix

  • #1
677
16

Main Question or Discussion Point

if x is a column vector, then a matrix G = x*xT is a Gramian Matrix.
When I tried calculating the matrix G and its eigenvalues for cases when x = [x1 x2]' and [x1 x2 x3]'
by actually working out the algebra, it turned out (if I didn't do any mistakes) that the eigen values are all zeros except one which is equal to (x12+x22 OR x12 + x22 + x32) depending upon the case.
Is this a standard result for a Gramian Matrix to have a single non-zero eigenvalue? If, yes, is there a simpler proof?

Thank you.
 

Answers and Replies

  • #2
177
61
Yes that is a standard and pretty simple fact. How did you compute the eigenvalues? Did you compute characteristic polynomials and then find their roots? If so, then yes, there is a much simpler proof.
 
  • #3
677
16
Yes that is a standard and pretty simple fact. How did you compute the eigenvalues? Did you compute characteristic polynomials and then find their roots? If so, then yes, there is a much simpler proof.
Yes, I solved the roots of characteristic equation, and it was a nasty business for even a 3x3 matrix. :D Would love to know the simpler method.
 
  • #4
177
61
Let ##\lambda \ne 0## be an eigenvalue, and ##\mathbf v\ne\mathbf 0## be he corresponding eigenvector. That means ##G\mathbf v =\lambda\mathbf v##. But $$G \mathbf v = \mathbf x (\mathbf x^T \mathbf v)$$ and ##(\mathbf x^T \mathbf v)## is a scalar. Therefore $$(\mathbf x^T \mathbf v) \mathbf x = \lambda \mathbf v,$$ so the eigenvector ##\mathbf v## must be a non-zero multiple of ##\mathbf x##. Substituting ##a\mathbf x## (where ##a\ne0## is a scalar) in the above equation, we get that indeed ## a\mathbf x ## is an eigenvector corresponding to ##\lambda= \mathbf x^T\mathbf x##.
 
  • #5
677
16
Let ##\lambda \ne 0## be an eigenvalue, and ##\mathbf v\ne\mathbf 0## be he corresponding eigenvector. That means ##G\mathbf v =\lambda\mathbf v##. But $$G \mathbf v = \mathbf x (\mathbf x^T \mathbf v)$$ and ##(\mathbf x^T \mathbf v)## is a scalar. Therefore $$(\mathbf x^T \mathbf v) \mathbf x = \lambda \mathbf v,$$ so the eigenvector ##\mathbf v## must be a non-zero multiple of ##\mathbf x##. Substituting ##a\mathbf x## (where ##a\ne0## is a scalar) in the above equation, we get that indeed ## a\mathbf x ## is an eigenvector corresponding to ##\lambda= \mathbf x^T\mathbf x##.
I cannot see why the bold part should be true?
But, doing the underlined part, i.e. substitution v=ax, I can see that it gives a solution, is that from this you infered that the bold part should hold?

Thank you for your help.
 
  • #6
177
61
I cannot see why the bold part should be true?
But, doing the underlined part, i.e. substitution v=ax, I can see that it gives a solution, is that from this you infered that the bold part should hold?

Thank you for your help.
No, the "bold" part is true independently of the "underlined" part, they both prove different parts of the statement.

For the "bold" part: we know that ##(\mathbf x^T\mathbf v)## is a number, let call it ##\beta##. Then the equation is rewritten as ##\beta\mathbf x = \lambda\mathbf v##, and solving it for ##\mathbf v ## gives us ##\mathbf v = (\beta/\lambda) \mathbf x##.

Now, the constant ##\beta## depends on the unknown ##\mathbf v##, and we do not know what ##\lambda## is, so we cannot say from here that ##\mathbf v = (\beta/\lambda) \mathbf x= a\mathbf x## is an eigenvector. But what we can say is that if ##\mathbf v## is an eigenvector corresponding to a non-zero eigenvalue ##\lambda##, then it must be a non-zero multiple of ##\mathbf x##.

Substituting then ##\mathbf v=a\mathbf x## we get that it is indeed an eigenvector and find ##\lambda##. So the "underlined" part give you that ##\mathbf v=a\mathbf x## is an eigenvector, and gives the corresponding eigenvalue. The "bold"" part shown that there are no other eigenvectors corresponding to a non-zero eigenvalue.
 

Related Threads on Eigenvalues of Gramian Matrix

Replies
2
Views
927
Replies
4
Views
794
  • Last Post
Replies
8
Views
5K
  • Last Post
Replies
2
Views
3K
  • Last Post
Replies
5
Views
4K
  • Last Post
Replies
8
Views
1K
  • Last Post
Replies
7
Views
10K
Replies
8
Views
3K
Replies
8
Views
15K
  • Last Post
Replies
3
Views
1K
Top