Compact Operators — Mathematics Knowledge System

Overview

A compact operator is a bounded linear operator that maps bounded sets to precompact (relatively compact) sets. Compact operators are the closest infinite-dimensional analogues of matrices: they can be approximated by finite-rank operators, they behave well under spectral analysis, and they appear naturally as integral operators. The theory of compact operators underpins Fredholm theory and the spectral theorem for compact self-adjoint operators.

Intuition

In finite dimensions, every bounded operator is 'compact' in the sense that bounded sets map to bounded (hence precompact) sets. In infinite dimensions, the unit ball is not compact, so this fails for most operators. A compact operator 'collapses' the infinite-dimensional structure enough that the image of the unit ball has compact closure. Think of Fourier projections: projecting onto the first N modes gives a finite-rank (hence compact) operator, and every compact operator is a limit of such finite-rank approximations.

Formal Definition

Definition

A linear operator T: X -> Y between Banach spaces is compact if for every bounded sequence (x_n) in X, the sequence (Tx_n) has a convergent subsequence in Y. Equivalently, T maps the closed unit ball of X to a precompact set in Y. Every compact operator is bounded. A finite-rank operator (range is finite-dimensional) is compact. The set K(X,Y) of compact operators is a closed subspace (ideal) of B(X,Y).

T \text{ compact} \iff T(B_X) \text{ is precompact in } Y

Compact operator definition

\mathcal{K}(X,Y) \subseteq \mathcal{B}(X,Y) \text{ closed ideal}

Compact operators form a closed ideal

\|T - T_n\| \to 0,\; T_n \text{ finite rank} \implies T \text{ compact}

Approximation by finite rank

\|T\|_{\mathrm{HS}}^2 = \sum_{m,n} |\langle Te_m, f_n\rangle|^2 = \sum_n \|Te_n\|^2

Hilbert-Schmidt norm

Notation

Notation	Meaning
$\mathcal{K}(X,Y)$	Space of compact linear operators from X to Y
$\\|T\\|_{\mathrm{HS}}$	Hilbert-Schmidt norm of T
$\mathrm{rank}(T)$	Dimension of the range of T

Theorems

Theorem 10.1: Compact Operators are Limits of Finite-Rank Operators (on Hilbert Spaces)

On a Hilbert space H, every compact operator T: H -> H is the norm limit of finite-rank operators. More precisely, if (e_n) is an orthonormal basis of H and P_n is the projection onto span{ e_1,...,e_n }, then ||TP_n - T|| -> 0.

Theorem 10.2: Composition with Bounded Operators

K(X,Y) is a closed two-sided ideal in B(X,X): if T is compact and S is bounded, then ST and TS are compact. Also, norm-limits of compact operators are compact.

Theorem 10.3: Riesz's Theorem on Compact Operators

Let T: X -> X be compact on a Banach space X and lambda != 0. Then ker(T - lambda I) is finite-dimensional, range(T - lambda I) is closed, and either (T - lambda I) is bijective or lambda is an eigenvalue.

Theorem 10.4: Hilbert-Schmidt Operators are Compact

An operator T in B(H) is Hilbert-Schmidt if ||T||_HS^2 = sum_n ||Te_n||^2 < infinity for an orthonormal basis (e_n). Every Hilbert-Schmidt operator is compact, and ||T|| <= ||T||_HS.

Worked Examples

1
Compute the Hilbert-Schmidt norm:
$\|T\|_{\mathrm{HS}}^2 = \int_0^1\int_0^1 |K(s,t)|^2\,ds\,dt = \|K\|_{L^2}^2 < \infty$
2
Since T is Hilbert-Schmidt, it is compact by Theorem 10.4.
3
Alternatively, approximate K in L^2 by finite linear combinations of rank-1 kernels K_n(s,t) = sum_{j=1}^N phi_j(s) psi_j(t), which give finite-rank (hence compact) operators T_n, and ||T_n - T|| <= ||T_n - T||_HS -> 0.

✓ Answer

Any integral operator with an L^2 kernel is Hilbert-Schmidt, hence compact.

Practice Problems

Mediumproof writing

Prove that K(X,Y) is a closed subspace of B(X,Y): if T_n are compact and T_n -> T in operator norm, then T is compact.

Mediumfree response

Explain why compact operators are natural generalisations of matrices (finite-rank operators).

MediumMultiple choice

Which of the following operators on l^2 is compact?

The identity operator IThe right-shift operator R(x_1, x_2, ...) = (0, x_1, x_2, ...)The operator T(x_1, x_2, ...) = (x_1/1, x_2/2, x_3/3, ...)The multiplication operator M_b(x_n) = (b_n x_n) with b_n = 1 for all n

Common Mistakes

Common Mistake

Thinking that bounded operators are always compact.

Compact is a strictly stronger condition than bounded. The identity on any infinite-dimensional space is bounded but not compact.

Common Mistake

Assuming compact operators have discrete spectrum in all Banach spaces.

The spectral theorem for compact operators (nice spectral decomposition) requires additional structure — typically self-adjointness on a Hilbert space. On a general Banach space, compact operators still have discrete spectrum away from 0, but eigenvectors don't form a basis.

Quiz

Historical Background

Compact operators (originally called 'completely continuous' operators) were studied by David Hilbert and Erhard Schmidt in the context of integral equations around 1906-1907. Frigyes Riesz formalised the theory in 1918, introducing the term 'completely continuous' and proving the Fredholm alternative. The modern term 'compact operator' was popularised by Grothendieck in the 1950s. Hilbert-Schmidt operators (a subclass of compact operators) were central to Hilbert's spectral theory.

1906
Hilbert and Schmidt study integral operators, precursors to compact operators
David Hilbert, Erhard Schmidt
1918
Riesz formalises the theory of completely continuous operators
Frédéric Riesz
1930s
Schauder develops the theory in abstract Banach spaces
Juliusz Schauder
1955
Grothendieck's thesis revolutionises the study of tensor products and compact operators
Alexander Grothendieck

Summary

Compact operators map bounded sets to precompact sets; equivalently, they send bounded sequences to sequences with convergent subsequences.
Every finite-rank operator is compact. On Hilbert spaces, every compact operator is a limit of finite-rank operators.
K(X,Y) is a closed two-sided ideal in B(X,X): compositions with bounded operators remain compact.
Hilbert-Schmidt operators (finite Hilbert-Schmidt norm) are a natural class of compact operators.
Compact operators have discrete spectrum: nonzero eigenvalues are isolated with finite-dimensional eigenspaces.

References

BookRudin, W. — Functional Analysis (2nd ed.), Chapter 4
BookConway, J. — A Course in Functional Analysis, Chapter 2
WebsiteWikipedia: Compact operator