Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $H$ be a Hilbert space with inner product $\langle\cdot,\cdot\rangle$ and let $V,W$ be two closed subspaces. For $x_0\in H$ we may define the sequence of projections $$x_{2n+1}=P_W(x_{2n}), \qquad x_{2n+2}=P_V(x_{2n+1}).$$ I intend to prove that if $V\cap W=\{0\}$, then $x_n\rightarrow 0$.

In the first place, one can easily show that $||x_{n+1}||^2=\langle x_{n+1},x_n\rangle$, from which one easily deduces that the sequence of norms is decreasing, so that $\lVert x_n\rVert\to\ell\geq 0$.

Then one can also show that if $V\cap W=\{0\}$ and a subsequemce $\{x_{2n_k}\}_k$ converges weakly to some $x$, then the sequence $\{x_{2n_k+1}\}_k$ also converges weakly to $x$, whence $x=0$.

So now it suffices to prove that a subsequence $\{x_{2n_k+1}\}$ converges weakly to something. For this, I first note that $$\lVert x_n\rVert^2=\langle x_{n+1},x_{n-2}\rangle=\cdots=\langle x_{2n-1},x_0\rangle$$ And now I would like to use that the sequence of norms converges and Riesz's representation theorem to conclude weak convergence but I am unsure as to how to proceed, since while it is true that any functional in $H$ is of the for $\langle \cdot,x_0\rangle$, varying $x_0$ also varies the sequence $\{x_{2n-1}\}$.

Thanks in advance for any insight.

share|cite|improve this question
Alaoglu's theorem will guarantee a weakly convergent subsequence. – Nate Eldredge May 1 '12 at 23:24
@NateEldredge: why a subsequence as opposed to a subnet? – user16299 May 2 '12 at 1:03
@Yemon: because we're only working in the span of the $x_{n+1}$, so we have metrizability of the unit ball in the weak$^\ast$ topology by separability. – t.b. May 2 '12 at 1:05
@t.b. of course - thanks! – user16299 May 2 '12 at 3:58

(disclaimer: my professional (de)-formation makes me dislike "vector computations" and makes me prefer "operator computations". I feel that in many cases algebraic arguments over the operators are neater than vector manipulations)

Your result is a particular case of something more general: if $P,Q$ are projections onto closed subspaces $M$ and $N$ of $H$ respectively, then $$ \lim_{n\to\infty}(QPQ)^n=P\wedge Q, $$ where $P\wedge Q$ is the projection onto $M\cap N$ and the limit is taken in the strong operator topology (i.e. with the pointwise convergence).

To see this, fix any $z\in H$; then $$ \langle (QPQ)^{2n+1}z,z\rangle=\langle (QPQ)^n(QPQ)(QPQ)^{n}z,z\rangle=\langle (QPQ)(QPQ)^{n}z,(QPQ)^nz\rangle=\langle PQ(QPQ)^{n}z,PQ(QPQ)^nz\rangle= \|PQ(QPQ)^nz\|^2\leq\|Q(QPQ)^nz\|^2=\|(QPQ)^nz\|^2 =\langle(QPQ)^nz,(QPQ)^nz\rangle=\langle(QPQ)^{2n}z,z\rangle. $$ This shows that $(QPQ)^{2n+1}\leq(QPQ)^{2n}$ as operators. A very similar argument shows that $(QPQ)^{2n+2}\leq(QPQ)^{2n+1}$. So the sequence of selfadjoint operators $\{(QPQ)^n\}$ is monotone, and so it converges in the strong operator topology to a selfadjoint operator $R$.

Edit: here's a short justification of the last sentence. That $\{(QPQ)^n\}_n$ is monotone decreasing (and positive) implies that for any $z\in H$ the sequence of numbers $\{\langle(QPQ)^nz,z\rangle\}_n$ is positive and decreasing, so convergent to a number $R_{x,x}$; using polarization, one gets that $\langle(QPQ)^nz,w\rangle\to R_{z,w}$ in such a way that the map $(z,w)\mapsto R_{z,w}$ is a sesquilinear form; it is then standard that any sesquilinear form defines an operator, which we call $R$ in this case.

We also have, since the sequence is bounded, that the product of the limit is the limit of the product, so $$ R^2=\lim_{n\to\infty}(QPQ)^n\lim_{n\to\infty}(QPQ)^n =\lim_{n\to\infty}(QPQ)^n(QPQ)^n=\lim_{n\to\infty}(QPQ)^{2n}=R. $$ This shows that $R$ is a projection. Also, since it is the infimum of the sequence, $$ R\leq QPQ\leq Q^2=Q, $$ and similarly $R\leq P$. So $R\leq P\wedge Q$. On the other hand, $$ P\wedge Q=(P\wedge Q)P=(P\wedge Q)Q, $$ and from this $P\wedge Q\leq R$. Thus $R=P\wedge Q$.

share|cite|improve this answer
Thanks for the edit. – t.b. May 2 '12 at 2:25
@t.b.: sorry I didn't answer you directly, I was about to do it after I edited but then I had to leave for a while. – Martin Argerami May 2 '12 at 2:39

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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