1

If $A$ is idempotent (defined by $AA=A$) then the eigenvalues of $A$ are $0$ or $1$.

Proof: $$Ax = \lambda x$$ $$\Rightarrow Ax = AAx = \lambda Ax = \lambda^{2}x.$$

Then in the lecture notes, it says that the above implies that $\lambda^{2} = \lambda$ which implies $\lambda = 0$ or $\lambda = 1$.

I don't quite follow how it is deduced that $\lambda^{2} = \lambda$ from the above calculations.

Veliko
  • 1,104
OGC
  • 2,363
  • 1
    They should've started by letting $x$ be non-zero. ​ ​ –  May 27 '17 at 14:51
  • 6
    The proof should, perhaps, have been this: Let $x$ be a nonzero eigenvector for $\lambda$. Then $Ax = \lambda x$. Multiplying both sides by $A$, we get $A^2 x = \lambda Ax = \lambda (Ax) = \lambda (\lambda x) = \lambda^2 x$. On the other hand, $A^2 = A$, so $A^2 x = Ax = \lambda x$. Hence $\lambda x = \lambda^2 x$. Because $x$ is nonzero, we get $\lambda = \lambda^2$. – John Hughes May 27 '17 at 14:55
  • @JohnHughes Got it! You can post your comment as an answer and I'll accept it. – OGC May 27 '17 at 15:02

3 Answers3

10

In addition to my comment, let me suggest another proof that leads to a slightly larger insight.

For any matrix $M$, and $k \ge 1$, if $x$ is an eigenvector for $\lambda$ then $$ M^k x = \lambda^k x. $$

The proof is just induction, so I won't write it out.

Now start from $A^2 = A$. Rewrite this as $$ A^2 - A = 0 $$ and then let $x$ be an eigenvector for $\lambda$, an eigenvalue of $A$. We then get \begin{align} (A^2 - A)x &= 0 \\ A^2x - Ax &= 0 \\ \lambda^2x - \lambda x &= 0 & \text{by applying the lemma}\\ (\lambda^2 - \lambda) x &= 0 \\ \lambda^2 - \lambda &= 0 & \text{because $x$ is nonzero}\\ \end{align}

Now the polynomial that $A$ satisfied --- $A^2 - A = 0$ --- was nothing special. Suppose instead that we knew that $A^3 - 3A^2 - A + I = 0$. We can write this as $p(A)$, where $p(x) = x^3 - 3x^2 - x + 1$. By exactly the same kind of argument, we'd find that if $\lambda$ was an eigenvalue for $A$, then $p(\lambda) = 0$.

Summary: if a matrix $A$ satisfies $p(A) = 0$ for some polynomial $p$, then for any eignevalue $\lambda$ of $A$, we also have $p(\lambda) = 0$.

Arnaldo
  • 21,758
John Hughes
  • 100,827
  • 4
  • 86
  • 159
2

By definition eigenvector is vector, which $\neq 0$: $Av = \lambda v $

So, we have: $\lambda^2v=\lambda v = > (\lambda^2 - \lambda)v = 0$.

For non zero $v_i$ you have (over field of C you don't have divisors of zero.)

$\lambda^2-\lambda = \lambda(\lambda - 1) = 0 $. And that's why only $\lambda = 0$ or $\lambda = 1$. And eigenvector $\neq 0$

duncan
  • 415
1

Because at the beginning you have $Ax$ which is equal to $\lambda x $ because x is an eigenvector.

Bargabbiati
  • 2,311