As a part of a larger proof, my text claims that if

$$A\begin{bmatrix}u_1&u_2\\ \end{bmatrix}=\begin{bmatrix}u_1&u_2\\ \end{bmatrix}\begin{bmatrix}\lambda&1\\0&\lambda\\ \end{bmatrix}$$

where $A\in \mathbb R^{n\times n}, u_1, u_2$ are complex vectors, and $\lambda \in \mathbb C \setminus \mathbb R$, then

$$\text{span}\{u_1, u_2 \}\cap \text{span}\{ \overline{u_1}, \overline{u_2} \}=\{ 0 \}$$

I don't quite understand why. So we conjugate both sides:

$$A\begin{bmatrix}\overline{u_1}&\overline{u_2}\\ \end{bmatrix}=\begin{bmatrix}\overline{u_1}&\overline{u_2}\\ \end{bmatrix}\begin{bmatrix}\overline{\lambda}&1\\0&\overline{\lambda}\\ \end{bmatrix}$$

And we know that $\lambda \neq \overline{\lambda}$.

Naively taking an element in the intersection of spans and writing it out as a combination doesn't seem to help.

There must be some simple argument I don't see.

  • $\begingroup$ What are $u_1$ and $u_2$? numbers? vectors? real? complex? $\endgroup$ – Gerry Myerson May 4 '14 at 7:20
  • $\begingroup$ @GerryMyerson: complex vectors. $\endgroup$ – Leo May 4 '14 at 7:24
  • $\begingroup$ What does the notation $A [u_1 u_2]$ mean? I am assuming $u_i \in \mathbb R^{1 \times n}$. $\endgroup$ – user89987 May 4 '14 at 15:43
  • 1
    $\begingroup$ @user89987: $u_i \in \mathbb C^{n \times 1}$ and $A[u_1 \ u_2]=[Au_1 \ Au_2]\in \mathbb C^{n \times 2}$, as far as I understood. $\endgroup$ – Leo May 4 '14 at 16:03

Edit. OK, I have misread the question. Here is a new answer. I've just rushed for a quick fix and haven't spent much time on it. So, I guess the proof below is very clumsy, and more elegant proofs can be found in those textbooks that discuss real Jordan forms.

By assumption, we have \begin{cases} (A-\lambda I)u_1 = 0,\\ (A-\lambda I)u_2 = u_1,\\ (A-\bar{\lambda}I)\bar{u}_1 = 0,\\ (A-\bar{\lambda}I)\bar{u}_2 = \bar{u}_1. \end{cases} Suppose inside $\operatorname{span}\{u_1, u_2\}\cap\operatorname{span}\{\bar{u}_1, \bar{u}_2\}$ there lies a vector $$x = au_1+bu_2=c\bar{u}_1+d\bar{u}_2\tag{$\ast$}$$ where $a,b,c,d$ are some complex numbers. Apply $(A-\lambda I)(A-\bar{\lambda}I)$ on both sides (note that the two multiplicands of this matrix product commute), we get \begin{align} (A-\bar{\lambda}I)bu_1 &= (A-\lambda I)d\bar{u}_1,\\ b(\lambda -\bar{\lambda})u_1 &= d(\bar{\lambda}-\lambda )\bar{u}_1,\\ bu_1 &= -d\bar{u}_1.\tag{$\dagger$} \end{align} Now there are two possibilities:

  1. $u_1=\bar{u}_1=0$. So we have $x=bu_2=d\bar{u}_2$ in $(\ast)$. If $x$ is nonzero, then $u_2=\frac{d}{b}\bar{u}_2$, i.e. $u_2$ is a nonzero multiple of a nonzero real vector, but this is impossible because $A$ is real, $\lambda$ is nonreal and $(A-\lambda I)u_2 = 0$. Hence $x=0$.
  2. $u_1,\bar{u}_1\ne0$. From $(\dagger)$, $b$ and $d$ must be both zero or both nonzero. If they are both nonzero, then $u_1=\frac{-d}{b}\bar{u}_1$. By a similar reasoning to the above, we see that this is impossible. Therefore $b=d=0$ and we get $x = au_1=c\bar{u}_1$ in $(\ast)$. Again, $a$ and $c$ must be zero. Hence $x=0$.

In other words, $x$ must be zero, i.e. the two spans in question have a zero intersection.

  • $\begingroup$ $u_1$ and $u_2$ are in ${\bf C}^n$, not ${\bf C}^2$. $\endgroup$ – Gerry Myerson May 5 '14 at 13:20
  • $\begingroup$ @GerryMyerson Thanks for catching my error. $\endgroup$ – user1551 May 5 '14 at 17:31

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.