Small proof

1. Jun 17, 2009

evilpostingmong

1. The problem statement, all variables and given/known data

Show that A and AT share the same eigenvalue.

2. Relevant equations

3. The attempt at a solution
let v be the eigenvector
Av=Icv
since ATv=ITcv
and IT=I,
ATv=Icv
so ATv=Icv=Av
so A and AT must have the same eigenvalue.

Last edited: Jun 17, 2009
2. Jun 17, 2009

Cyosis

How did you go from the first to the second step?

3. Jun 17, 2009

evilpostingmong

If Av=Icv, then the transpose of Icv is Icv since the transpose of I is I.
Now (Icv)T=(Acv)T since Icv=Av
since neither c nor v are matrices, they get pushed aside to
get ITcv=ATv

4. Jun 17, 2009

Cyosis

Yes, but a column vector is an mx1-matrix. When you transpose an mx1 matrix you get a 1xm matrix, called a row vector. My point is v and v transposed aren't exactly the same.

5. Jun 17, 2009

evilpostingmong

Oh okay. So (Av)T= (Ivc)T=(Ic)T(v)T=c(I)T(v)T=cI(v)T which has the same eigenvalue
as A, so AT and A both have the same eigenvalue.

6. Jun 17, 2009

Cyosis

I don't quite see how you drew your conclusion. $(Av)^T=v^T A^T$. Just try to compute $I v^T$ with a small vector, you will notice they are incompatible.

7. Jun 17, 2009

evilpostingmong

since Av=cIv, (Av)T=(cIv)T and
factor out c to get c(Iv)T which can be transposed.
wait, never mind....I'm fixing this.

8. Jun 17, 2009

Cyosis

Sure, but how do you conclude from that, that A^T has eigenvalue c?

9. Jun 17, 2009

evilpostingmong

Oh that's right. Let me reprove this. Let c be at position j,j on A. Transposing
the matrix lets c remain on j,j. So multiplying this by
the eigenvector v gives the column matrix Icv. This gives
the same result as not transposing A in the first place, since c will
still remain on j,j no matter what, and as a result, it would wind up at position 1,j on the
column matrix. Now, since Av=ATv,
c must be in both matrices A and AT.
Take note that we know that eigenvalues are found on the diagonals of
invertible matrices, which is why j,j was chosen.

Last edited: Jun 17, 2009
10. Jun 17, 2009

Cyosis

The matrix

$$\left( \begin{matrix} 1&1 \\ 1&1 \end{matrix} \right)$$

Has eigenvalues 2 and 0, neither of them are shown on the diagonal.

I would personally use the inner product to prove this.

Last edited: Jun 17, 2009
11. Jun 17, 2009

Dick

You could also think about using det(A)=det(A^T).

12. Jun 17, 2009

evilpostingmong

Oh, one thing: I am using the Axler book as a main book, and he covers eigenvectors
before inner products, and everything before determinants, so I can't apply those,
but I do appreciate your suggestions. But given cyosis's matrix, it makes sense that
all matrices with entries that equal each other also have eigenvalues besides 0, I messed
up there because I didn't consider that.
So I'll split my proof into 3 cases. Case 1: A is an invertible matrix. Case 2: A is a matrix
where all entries are the same, Case 3: A is noninvertible and not all of its entries are the
same. Is this a good idea? I would use determinants or inner products if I knew what they were. If there is no other way, I won't do this proof until
later on when I know more about them. But thank you for the suggestions.

Take note that when I feel I need to practice a bit more I use other sources, but they vary in topic sequence.

Last edited: Jun 17, 2009
13. Jun 17, 2009

Cyosis

I find it hard to believe you are studying eigenvectors/eigenvalues yet have not been exposed to determinants. This would mean you are not capable of calculating eigenvalues of matrices yet, but your posts suggest otherwise. Could you show me perhaps how to calculate the eigenvalues of the matrix I posted earlier?

14. Jun 17, 2009

Dick

You could also work around not having determinants yet. If c is an eigenvalue, then X=A-c*I is a singular matrix. Can you prove X is singular iff X^(T) is singular? How would that prove what you want to prove?

15. Jun 17, 2009

evilpostingmong

I can see three justifications for X to be singular. First, if A=cI, then X is a zero
matrix, and 0 is an eigenvalue for every vector, and X is singular for being
a zero matrix.

Second, if A-cI is nonzero but singular, then as long as the number of rows
the column matrix (representing a vector) equals the number of columns
in A and cI, Av would equal cIv so Av-cIv=0

Now, if A-cI were to be nonsingular, then (A-cI)v would not be zero,
so we would have Av=/=cIv so in this case v is not an eigenvector of A.

I know this isn't the actual proof, just want to see if I understand the
significance of X being singular.

Last edited: Jun 17, 2009
16. Jun 17, 2009

Dick

Singular doesn't mean A-cI equals 0. It means there is a nonzero vector v such that (A-cI)v=0. Which is what your second and third argument correctly say. You are overcomplicating this already. The question you should be thinking about, instead of trying to find multiple ways to prove the obvious, is why if X=A-cI is singular is X^T also singular?

Last edited: Jun 17, 2009
17. Jun 17, 2009

evilpostingmong

X=A-cI a jxn matrix with j=/=n.

Consider the column with elements aj,1 to aj,n.
If j>n then (at least) this column (lets call it column A) becomes a zero column after row reducing
which results in a matrix with less columns than X so X is singular.
Now transposing causes column A to become a row,
with column A being the one at the bottom. Row reducing X^T
gives a matrix with at least one less row (column A becomes a 0 row) than X^T since t so X^T is singular.
This is a result of A being one of (or the only) the extra columns, so it becomes one of (or the only)
extra rows that get "eliminated" after row reducing.

Last edited: Jun 17, 2009
18. Jun 17, 2009

Dick

You are thinking in the right direction. You proved row rank=column rank, right? So in the nxn case if rank is <n then both matrices are singular. You really don't have to think about any other case. If the matrix isn't square then the concept of an 'eigenvector' doesn't exist. Why not?

19. Jun 17, 2009

evilpostingmong

because I doesn't exist.

20. Jun 17, 2009

evilpostingmong

I be thanks you