Mathematics.

functional analysis

Compact Operators

Real Analysis110 minDifficulty9 out of 10

You should know: banach spaces, spectral theory

Overview

A compact operator is a bounded linear operator between Banach spaces that maps bounded sets to precompact sets (sets whose closure is compact). Compact operators are in many respects the closest infinite-dimensional analogue of finite-rank (matrix) operators: their spectrum behaves like that of a finite matrix (at most countably many nonzero eigenvalues accumulating only at 0), and the Fredholm alternative holds—either \(T - \lambda I\) is invertible or \(\lambda\) is an eigenvalue with finite-dimensional eigenspace. Compact operators arise naturally as integral operators, as solution operators for elliptic PDEs via Sobolev embeddings, and in the spectral theory of differential operators.

Intuition

A compact operator 'almost has finite rank': it compresses any bounded input set into a set that looks essentially finite-dimensional (precompact). Concretely, any bounded sequence \((x_n)\) yields a subsequence \((Tx_{n_k})\) that converges. This is why the spectral theory of compact operators so closely mirrors finite-dimensional linear algebra—the Fredholm alternative and the countable-spectrum theorem both follow from this near-finite-dimensionality.

Formal Definition

Definition

A linear operator \(T: X \to Y\) between Banach spaces is compact if the image \(T(B)\) of the closed unit ball \(B = \{x \in X : \|x\| \le 1\}\) is precompact in \(Y\) (i.e., its closure is compact). Equivalently, every bounded sequence \((x_n) \subset X\) has a subsequence \((x_{n_k})\) such that \((Tx_{n_k})\) converges in \(Y\). The space of compact operators is denoted \(\mathcal{K}(X,Y)\).

TK(X,Y)    T(BX) is compact in YT \in \mathcal{K}(X,Y) \iff \overline{T(B_X)} \text{ is compact in } Y
Compact operator definition
K(X,Y)B(X,Y)\mathcal{K}(X,Y) \subset \mathcal{B}(X,Y)
Compact operators are bounded

Properties

Compact operators form a closed two-sided ideal

K(X)isaclosedsubspaceofB(X)andsatisfies:ifTK(X)andSB(X),thenST,TSK(X).\mathcal{K}(X) is a closed subspace of \mathcal{B}(X) and satisfies: if T \in \mathcal{K}(X) and S \in \mathcal{B}(X), then ST, TS \in \mathcal{K}(X).

Finite-rank operators are compact

Everyboundedoperatorwithfinitedimensionalrangeiscompact.Thecompactoperatorsarethenormclosureofthefiniterankoperators(inHilbertspacesandmanyBanachspaces).Every bounded operator with finite-dimensional range is compact. The compact operators are the norm-closure of the finite-rank operators (in Hilbert spaces and many Banach spaces).

Spectrum of a compact operator

IfTK(X)andXisinfinitedimensional,then0σ(T).Thenonzerospectrumconsistsentirelyofeigenvalues,eachwithfinitedimensionaleigenspace,and0istheonlypossibleaccumulationpoint.If T \in \mathcal{K}(X) and X is infinite-dimensional, then 0 \in \sigma(T). The nonzero spectrum consists entirely of eigenvalues, each with finite-dimensional eigenspace, and 0 is the only possible accumulation point.

Theorems

Theorem 8.1: Spectral Theorem for Compact Self-Adjoint Operators
LetTbeacompactselfadjointoperatoronaseparableHilbertspaceH.ThenHhasanorthonormalbasis{en}consistingofeigenvectorsofT,withcorrespondingrealeigenvaluesλnsatisfyingλn0asn.Let T be a compact self-adjoint operator on a separable Hilbert space H. Then H has an orthonormal basis \{e_n\} consisting of eigenvectors of T, with corresponding real eigenvalues \lambda_n satisfying \lambda_n \to 0 as n \to \infty.
Theorem 8.2: Fredholm Alternative
LetTK(X)andλ0.Either(i)TλIisbijective(withboundedinverse),or(ii)λisaneigenvalueofTwithfinitedimensionaleigenspace,anddimker(TλI)=dimker(TλI)<(indexzero).Let T \in \mathcal{K}(X) and \lambda \neq 0. Either (i) T - \lambda I is bijective (with bounded inverse), or (ii) \lambda is an eigenvalue of T with finite-dimensional eigenspace, and \dim \ker(T-\lambda I) = \dim \ker(T^*-\lambda I^*) < \infty (index zero).
Theorem 8.3: Schauder's Theorem
TB(X,Y)iscompactifandonlyifitsadjointTB(Y,X)iscompact.T \in \mathcal{B}(X,Y) is compact if and only if its adjoint T^* \in \mathcal{B}(Y^*,X^*) is compact.
Theorem 8.4: Hilbert–Schmidt operators are compact
AnintegraloperatorT:L2(Ω)L2(Ω)withkernelKL2(Ω×Ω)iscompact(infactHilbertSchmidt),withTKKL2(Ω×Ω).An integral operator T: L^2(\Omega) \to L^2(\Omega) with kernel K \in L^2(\Omega \times \Omega) is compact (in fact Hilbert–Schmidt), with \|T\|_{\mathcal{K}} \le \|K\|_{L^2(\Omega\times\Omega)}.

Worked Examples

  1. Write \(Tf(x) = \int_0^1 K(x,t)f(t)\,dt\) with kernel \(K(x,t) = \mathbf{1}_{\{t \le x\}}\).

  2. Check \(K \in L^2([0,1]^2)\): \(\int_0^1\int_0^1 |K(x,t)|^2\,dt\,dx = \int_0^1 x\,dx = 1/2 < \infty\).

    KL2([0,1]2)2=12<\|K\|_{L^2([0,1]^2)}^2 = \frac{1}{2} < \infty
  3. Since the kernel is in \(L^2\), \(T\) is a Hilbert–Schmidt operator. By Theorem 8.4, \(T\) is compact.

  4. Estimate \(\|T\|_{\mathcal{B}} \le \|K\|_{L^2} = 1/\sqrt{2}\).

    T12\|T\| \le \frac{1}{\sqrt{2}}

Answer: The Volterra operator is Hilbert–Schmidt (kernel in \(L^2\)), hence compact, with \(\|T\| \le 1/\sqrt{2}\).

Practice Problems

Difficulty 8/10

Prove that the composition \(ST\) of a compact operator \(T \in \mathcal{K}(X,Y)\) with a bounded operator \(S \in \mathcal{B}(Y,Z)\) is compact.

Difficulty 9/10

Prove the Fredholm alternative for compact operators: if \(T \in \mathcal{K}(X)\) and \(\lambda \neq 0\), then \(\ker(T - \lambda I)\) is finite-dimensional.

Difficulty 8/10

Give an example of a bounded operator that is NOT compact, and verify it fails the compactness criterion.

Common Mistakes

Common Mistake

Compact operators on infinite-dimensional spaces can be surjective

A compact operator on an infinite-dimensional Banach space is never surjective (its range cannot be all of the infinite-dimensional space, since the image of the unit ball is precompact hence 'small'). Equivalently, \(0\) is always in the spectrum.

Common Mistake

Every bounded operator is compact

The identity operator on any infinite-dimensional space is bounded but not compact. Compactness is a strict strengthening: the image of the unit ball must be precompact, which fails for the identity in infinite dimensions.

Quiz

Which of the following is always true for a compact operator \(T: X \to X\) on an infinite-dimensional Banach space?
The Fredholm alternative states that for \(T\) compact and \(\lambda \neq 0\):
A Hilbert–Schmidt operator (square-integrable kernel) is always:

Summary

  • A compact operator maps bounded sets to precompact sets; equivalently, every bounded sequence has a subsequence whose image converges.
  • Finite-rank operators are compact; compact operators are the norm-closure of finite-rank operators in Hilbert spaces.
  • The spectrum of a compact operator on an infinite-dimensional space has 0 as the only possible accumulation point; every nonzero spectral value is a finite-multiplicity eigenvalue.
  • The Fredholm alternative: for compact \(T\) and \(\lambda \neq 0\), either \((T-\lambda I)^{-1}\) is bounded or \(\lambda\) is a finite-multiplicity eigenvalue (and the index is zero).
  • Compact self-adjoint operators on Hilbert spaces admit a complete orthonormal eigenbasis with eigenvalues tending to 0—the exact infinite-dimensional analogue of diagonalisation.

References

  1. BookRudin, W. Functional Analysis. 2nd ed., McGraw-Hill, 1991. Chapter 4.
  2. BookConway, J. B. A Course in Functional Analysis. 2nd ed., Springer, 1990. Chapter 6.