5
$\begingroup$

Suppose $A$ and $B$ are two commuting square matrices over an algebraically closed field. It is true that the spectrum of $A+B$ is contained in the set $\{\lambda_1+\lambda_2:\lambda_1 \in \sigma(A), \lambda_2 \in \sigma(B)\}$ where $\sigma(A)$ denotes the spectrum of $A$?

If so, can you provide a reference for this? can more be said?

Thanks

$\endgroup$
0
9
$\begingroup$

Diagonalizable matrices

Let's say that $A$ and $B$ are diagonalizable. Then, since they commute, there exists an invertible matrix $T$ such that both $T^{-1}AT = D_A$ and $T^{-1}BT = D_B$ are diagonal matrices (Horn & Johnson, Matrix Analysis, page 52). The elements of the diagonal matrices are the eigenvalues. Thus we get:

$$A + B = TD_AT^{-1} + TD_BT^{-1} = T(D_A + D_B)T^{-1}$$

and so the eigenvalues of $A+B$ have the desired form.

General case

In the general case you can use simultaneous triangularization. The idea is the same, since $A$ and $B$ commute, there exists a unitary $U$ such that $U^*AU = T_A$ and $U^*BU = T_B$ where $T_A$ and $T_B$ are upper triangular matrices with the eigenvalues on the diagonal (Horn & Johnson, Matrix Analysis, page 81). Now we get:

$$A+B = U(T_A+T_B)U^*$$

and thus the eigenvalues have the desired form.

$\endgroup$
1
  • 1
    $\begingroup$ The OP is about matrices "over an algebraically closed field", not over $\mathbb C$. So replace "unitary" and "$^*$" by "invertible" and "$^{-1}$" (and probably use another reference that actually proves this). $\endgroup$ – darij grinberg Mar 24 '19 at 22:56
5
$\begingroup$

Frobenius has proved that commuting matrices over an algebraically closed field can be simultaneously brought into upper triangular form (this holds more generally for a solvable Lie algebras of matrices by Lie's theorem). To prove this for commuting matrices, show that they have a simultaneous eigenvector and proceed by induction on the dimension.

One way to prove this last fact (again for commuting matrices) is to use the Hilbert Nullstellensatz, see here, a nice proof of which you can find online e.g. on p.116 of Pete L. Clark's notes.

$\endgroup$
3
  • 1
    $\begingroup$ @anonymous: I'm slightly confused: Why do you ask your question in terms of algebraically closed fields if you're then accepting an answer that presupposes that you work over the complex numbers? $\endgroup$ – t.b. Jan 29 '11 at 17:33
  • $\begingroup$ i tried looking at page number 116 of your reference but it doesn't look like its on that page ? can u check and let me know . Thanks . $\endgroup$ – Theorem Jan 21 '13 at 20:25
  • $\begingroup$ @Theorem: Section 11 in the current version is called Nullstellensätze and in any case the pdf is fully searchable... $\endgroup$ – t.b. Feb 3 '13 at 9:41
1
$\begingroup$

Yes.

Matrix Analysis by Horn and Johnson has a proof.

$\endgroup$
1
  • $\begingroup$ The proof uses the fact that commuting matrices can be simultaneously upper-triangularized. They follow with an example where the other direction doesn't hold. $\endgroup$ – Brian Jan 29 '11 at 17:03
0
$\begingroup$

Yes see this(article about commutating matrices). In particular, if $\alpha_i$ and $\beta_j$ are the eigenvalues of $A$ and $B$, then one can choose an eigenbasis such that the eigenvalues of $A+B$ are $\alpha_i+ \beta_j$.

$\endgroup$
3
  • $\begingroup$ thanks, I saw this. The problem is that the reference for the fact ultimitely leads to an article of Frobenius, which is 1. In german. 2. I could not locate an online version of it. I wonder if there is a more accessible reference for this fact. $\endgroup$ – the L Jan 29 '11 at 16:54
  • $\begingroup$ Can we say if, Spectrum of $A+ B$ is same as $\{ λ_1+ λ_2 : λ_1 ∈ σ(A) , λ_2 ∈ σ(B) \}$ ? In which case this is true? $\endgroup$ – Akash Patalwanshi Sep 24 '17 at 14:17
  • $\begingroup$ Eigenbasis? There isn't always an eigenbasis. $\endgroup$ – darij grinberg Mar 24 '19 at 22:55

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.