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

Would you help me to solve this problem. Show that an idempotent operator on hilbert space is compact if and only if it has finite rank.

share|cite|improve this question
up vote 7 down vote accepted

Okay. If you require an answer that uses the sequential definition of a compact operator, here it goes. :)

A finite-rank operator is clearly compact, so one direction is established.

Now, an operator $ T: \mathcal{H} \rightarrow \mathcal{H} $ is compact if and only if it maps every bounded sequence in $ \mathcal{H} $ to a sequence in $ \mathcal{H} $ that contains a convergent subsequence. This is the basic sequential characterization of compact operators.

Let $ T: \mathcal{H} \rightarrow \mathcal{H} $ be an idempotent and compact operator. We want to show that it is a finite-rank operator. By way of contradiction, assume that it is not finite-rank. Let $ R(T) $ denote the infinite-dimensional range space of $ T $. We claim that $ R(T) $ is a closed subspace of $ \mathcal{H} $. To prove this claim, let $ (T(\mathbf{x}_{n}))_{n \in \mathbb{N}} $ be a sequence in $ R(T) $ that converges to some $ \mathbf{y} \in \mathcal{H} $. Then $$ T(\mathbf{y}) = T \left( \lim_{n \rightarrow \infty} T(\mathbf{x}_{n}) \right) = \lim_{n \rightarrow \infty} T(T(\mathbf{x}_{n})) = \lim_{n \rightarrow \infty} T(\mathbf{x}_{n}) = \mathbf{y}. $$ Hence, $ \mathbf{y} \in R(T) $, and as $ (\mathbf{x}_{n})_{n \in \mathbb{N}} $ is arbitrary, we are done.

As $ R(T) $ is now seen to be a closed (hence complete) subspace of $ \mathcal{H} $, it follows that $ R(T) $ is a Hilbert subspace of $ \mathcal{H} $. By the infinite-dimensionality of $ R(T) $, pick an orthonormal sequence of vectors $ (\mathbf{e}_{n})_{n \in \mathbb{N}} $ in $ R(T) $ (we can do so by extracting an orthonormal sequence from some infinite orthonormal basis for $ R(T) $). By construction, $ \mathbf{e}_{n} \in R(T) $ for each $ n \in \mathbb{N} $, so we can associate a $ \mathbf{v}_{n} \in \mathcal{H} $ such that $ T(\mathbf{v}_{n}) = \mathbf{e}_{n} $. Then $$ \forall n \in \mathbb{N}: \quad T(\mathbf{e}_{n}) = T(T(\mathbf{v}_{n})) = T(\mathbf{v}_{n}) = \mathbf{e}_{n}. $$ Hence, $ T $ maps the sequence $ (\mathbf{e}_{n})_{n \in \mathbb{N}} $ identically to itself. This sequence is clearly bounded. However, $ (T(\mathbf{e}_{n}))_{n \in \mathbb{N}} = (\mathbf{e}_{n})_{n \in \mathbb{N}} $ does not contain a convergent subsequence! (Simply observe that $ \| \mathbf{e}_{m} - \mathbf{e}_{n} \|_{\mathcal{H}} = \sqrt{2} $ for distinct $ m,n \in \mathbb{N} $.) It follows from the sequential definition of compactness that $ T $ is not compact, which contradicts the fact that we started our argument with $ T $ being compact.

Conclusion $ T $ must be a finite-rank operator.

I noticed that people have been asking questions about the exercises from the chapter on compact operators in Conway's A Course in Functional Analysis. For the benefit of the OP and also of other people who are interested in those exercises, the following link may be useful: Compact operator. It contains the full solution to another exercise from the same chapter.

share|cite|improve this answer

Let $ T \in B(\mathcal{H}) $ be idempotent.

If $ T $ is of finite rank, then it is clearly compact, so one direction is proven.

Conversely, suppose that $ T $ is compact. By way of contradiction, assume that $ T $ has infinite rank. Then $ T $ acts as the identity operator on its infinite-dimensional range space $ R(T) $. Letting $ \mathbb{B}_{R(T)} $ denote the open unit ball of $ R(T) $, we see that $ T $ maps $ \mathbb{B}_{R(T)} $ identically to itself. However, as $ \mathbb{B}_{R(T)} $ is not relatively compact in $ \mathcal{H} $ (i.e., $ {\text{cl}_{\mathcal{H}}}(\mathbb{B}_{R(T)}) $ is not compact; see below), this contradicts the compactness of $ T $. Therefore, $ T $ must have finite rank.


As mentioned, $ {\text{cl}_{\mathcal{H}}}(\mathbb{B}_{R(T)}) $ is not compact. If it were compact, then as $ R(T) $ is a closed subspace of $ \mathcal{H} $ (this follows from the idempotence of $ T $) and $ {\text{cl}_{R(T)}}(\mathbb{B}_{R(T)}) = {\text{cl}_{\mathcal{H}}}(\mathbb{B}_{R(T)}) \cap R(T) $, it would follow that $ {\text{cl}_{R(T)}}(\mathbb{B}_{R(T)}) $ is compact. However, this contradicts Riesz's Lemma, which implies that the closed unit ball in any infinite-dimensional normed space is never compact.

share|cite|improve this answer
How about compact definition by sequence? – beginner Oct 23 '12 at 6:11

Let $P$ be idempotent on a Hilbert Space $H$. Put $R(x) = x - P(x)$; then $$R(R(x)) = R(x - P(x)) = R(x) - R(P(x)) = x - P(x) - (P(x) - P(P(x)) = x - P(x) - P(x) + P(P(x)) = x - P(x).$$

From this we see that $H = \ker(P) + \ker(I - P)).$ We have represented $H$ as the direct sum of two closed subspaces. If $P$ is compact, its range is a locally-compact Banach Space, so the range is finite dimensional.

share|cite|improve this answer

If $P$ is finite-rank, then it is compact.

If $P$ is compact, then its spectrum consists only of eigenvalues, with $0$ as the only possible accumulation point. The equality $P^2=P$ implies that the only eigenvalues are $0$ and $1$. Now take any $x$ in the range of $P$; then $$ Px=x, $$ so $x$ is an eigenvector with eigenvalue $1$. The multiplicity of $1$ has to be finite, as otherwise $P$ would not be compact. Thus the range of $P$ is finite-dimensional.

share|cite|improve this answer
It is a very elegant proof. Could you please expand on how the infinite multiplicity of $1$ would imply that $1$ is an accumulation point? By definition of accumulation point I believe the set $\{0,1\}$ cannot have any accumulation points at all.(?) – user167889 Sep 1 '14 at 4:26
It's a matter of language. The point is that the nonzero eigenvalues of a compact operator can only have finite multiplicity. – Martin Argerami Sep 1 '14 at 4:29
I'm sorry I still don't understand. For example assume that $1$ had infinite multiplicity. Then it should be possible to construct a sequence of $\lambda_n \neq 1$ (and therefore $\lambda_n = 0$) with $\lambda_n \to 1$. But how? – user167889 Sep 1 '14 at 4:33
No it shouldn't. Why would that be possible? As you already said, the spectrum of a projection is contained in $\{0,1\}$. – Martin Argerami Sep 1 '14 at 4:38
Because you wrote ''The multiplicity of $1$ has to be finite, as otherwise it would be an accumulation point.'' Am I misunderstanding the sentence? (I'm sorry if my question is stupid, I've been thinking about it for days and just can't work it out on my own) – user167889 Sep 1 '14 at 4:39

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.