Most proofs that a compact Hilbert-space operator is the limit (in the operator topology) of finite-rank operators seem to restrict to the case that is separable. But this is unnecessary hypothesis.
Indeed, let . Since the image has compact closure–some authors use the term relatively compact— is necessarily totally bounded: for any , there exist finitely many elements , where , such that
We define to be the orthogonal projection onto the finite-dimensional subspace spanned by , and we can define a finite-rank operator by .
I claim that as . Recall that for all , is the unique element in such that , so that
Taking the supremum over all , we obtain the estimate , which tends to as .
I wonder whether a similar characterization was true for compact operators between possibly nonseparable Banach spaces, but there answer is no. Apparently, this was an open question for many years until the Swedish mathematician Per Enflo showed that there are compact operators on separable, reflexive Banach spaces which are not the limit of finite-rank operators. The interested reader can consult Enflo, Per. “A counterexample to the approximation problem in Banach spaces.” Acta Mathematica 130.1 (1973): 309-317.