In the setting of an introduction to functional analysis course, I have read the following statement:

Let $H$ be a Hilbert space and let $A\subseteq H$ be a closed convex set.

Then there exist a projection $$ P_{A}:\, H\to A $$

s.t for every $x\in H$, $P_{A}(x)$ is the only point in $A$ that satisfy $$ d(x,A)=d(P_{A}(x),x) $$

I have three problems with the proof given, I don't understand one technical transition and I don't understand where did the convexity of $A$ came into play, and the uniqueness.

The proof given was:

Take a sequence $\{y_{n}\}\subseteq A$ s.t $d=d(x,A)=\lim_{n\to\infty}d(x,y_{n})$. We will prove that $\{y_{n}\}$is a Cauchy sequence and get both existence and uniqueness. Define $P_{A}(x)=y$, apply the parallelogram equality for $y_{n}-x,y_{m}-x$: $$ 4d^{2}\leftarrow_{\text{as {m,n\to\infty}}}2||y_{n}-x||^{2}+2||y_{m}-x||^{2}=||y_{n}-y_{m}||^{2}+||y_{n}+y_{m}-2x||^{2} $$ $$ =||y_{n}-y_{m}||^{2}+4||\frac{y_{n}+y_{m}}{2}-x||^{2}\geq||y_{n}-y_{m}||^{2}+4d^{2} $$

hence $||y_{n}-y_{m}||$ is as small as we want.

I have $3$ questions that I would appreciate any help with:

1) How was the transition of the first equality made ? I don't see how this transition have anything to to with he parallelogram equality

2) Where did we use that $A$ is convex ? (I can see how we used that $A$ is closed when we took the limit and and it defined an element of $A$, but I don't see where did we need the other condition on $A$ that is the convexity)

3) Why is the point $y$ unique ? at first I thought it was because a normed space is in particular a metric space and thus a Hausdorff space and so the limit is unique, but that just say that the $\{y_{n}\}$ have a unique limit.

But maybe there is a different sequence $\{a_{n}\}$ s.t $$d=d(x,A)=\lim_{n\to\infty}d(x,a_{n})$$ and $a_{n}$ converges to a different element of $A$ rather than $y$.


1 Answer 1


Ad 1), the parallelogram identity is

$$\lVert a+b\rVert^2 + \lVert a-b\rVert^2 = 2\lVert a\rVert^2 + 2\lVert b\rVert^2$$

for all $a,b$. The first line applies that with $a = y_n - x$ and $b = y_m - x$.

2) Where did we use that $A$ is convex?

In the step where it was deduced that

$$\left\lVert \frac{y_n + y_m}{2} - x\right\rVert \geqslant d,$$

because of convexity, the midpoint of $y_m$ and $y_n$ lies in $A$.

3) Why is the point $y$ unique?

Let $p_1, p_2 \in A$ with $\lVert p_i - x\rVert = d$. Then

$$4d^2 = 2\lVert p_1 - x\rVert^2 + 2\lVert p_2 - x\rVert^2 = \lVert p_1 - p_2\rVert^2 + 4 \left\lVert \frac{p_1+p_2}{2} - x\right\rVert^2 \geqslant 4d^2,$$

hence $\lVert p_1 - p_2\rVert^2 = 0$.

  • $\begingroup$ Those were very clear explanations, thank you. $\endgroup$
    – Belgi
    Sep 12, 2013 at 10:39

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.