(July 24, 2011)

Compact resolvents
Paul Garrett [email protected] http://www.math.umn.edu/egarrett/

1. Application of perturbation theory

2. Appendix: normal compact operators on Hilbert spaces
Unbounded operators T with compact resolvents (T − λ)−1 are among the most useful among unbounded
operators on Hilbert or Banach spaces. Many important semi-bounded symmetric differential operators are
d2
in this class, the simplest being (regular) Sturm-Liouville operators like T = dx 2 + q(x) on a finite interval

[a, b].
We prove that, for T : X → X a possibly unbounded, but densely-defined, operator on a Banach space, T −1
compact implies that the resolvent (T − λ)−1 is meromorphic, and is compact away from poles. This is an
example of perturbation theory.

The proof uses basic facts about compact operators. The easier case of T a symmetric operator on a Hilbert
space is already useful. In that case, T −1 is a normal compact operator, and the resolvent (T − λ)−1 is
normal, allowing application of simple results about normal compact operators on Hilbert spaces, recalled
in an appendix.

A fuller version of the spectral theory of compact operators on Banach spaces circumvents issues of normality
and of the symmetry of T , and extends the discussion of compact resolvents to Banach spaces. The required
ideas are Fredholm-Riesz theory, from [Fredholm 1900/1903] and [Riesz 1917].
The general Banach space setting is useful, directly addressing intuitive spaces such as C o [a, b] or C k [a, b].

1. Application of perturbation theory

We prove that, if a (not necessarily bounded) densely-defined operator T on a Banach space X has compact
inverse T −1 , then (T − λ)−1 exists and is compact for λ off a discrete set in C, and is meromorphic in λ.
The background on compact operators is more elementary in the interesting sub-case that T is a (not
necessarily bounded) symmetric operator on a Hilbert space X. When T −1 exists and is compact, it is
also normal, and the normal operator (T − λ)−1 exists and is compact for λ off a discrete set in C, and is
meromorphic in λ.
The assertion and argument are standard, especially for Hilbert spaces. E.g., see [Kato 1966], p. 187 and
preceding.

[1.1] Spectrum of possibly-unbounded operators

Recall that specification of a possibly unbounded operator T on a Hilbert or Banach space X includes its
domain DT . We only consider T with DT dense.
The set of eigenvalues or point spectrum of a possibly-unbounded operator T consists of λ ∈ C such that
T − λ fails to be injective.

The continuous spectrum consists of λ with T −λ injective and with dense image, but not surjective. Further,
for possibly unbounded operators, we require a bounded (=continuous) inverse (T − λ)−1 on (T − λ)DT for
λ to be in the continuous spectrum.
The residual spectrum consists of λ with T − λ injective, but (T − λ)DT not dense.

The description of continuous spectrum simplifies for closed T : we claim that for (T − λ)−1 densely defined

1
 Paul Garrett: Compact resolvents (July 24, 2011)

and continuous, (T − λ)DT is the whole space, so (T − λ)−1 is everywhere defined, so λ cannot be in the
residual spectrum. Indeed, the continuity gives a constant C such that |x| ≤ C · |(T − λ)x| for all x ∈ DT .
Then (T − λ)xi Cauchy implies xi Cauchy, and T closed implies T (lim xi ) = lim T xi . Thus, (T − λ)DT is
closed. Then density of (T − λ)DT implies it is the whole space.

[1.2] (T − λ)−1 is compact

Now prove that for T −1 compact on a Banach space the resolvent (T − λ)−1 exists and is compact for λ off
a discrete set, and is meromorphic in λ.

The non-zero spectrum of the compact operator T −1 is point spectrum, from basic Fredholm-Riesz theory
for compact operators. [1]
We claim that the spectrum of T and non-zero spectrum of T −1 are in the obvious bijection λ ↔ λ−1 . From
the algebraic identities

T −1 − λ−1 = T −1 (λ − T )λ−1 T − λ = T (λ−1 − T −1 )λ

failure of either T − λ or T −1 − λ−1 to be injective forces the failure of the other, so the point spectra are
identical.
For (non-zero) λ−1 not an eigenvalue of compact T −1 , T −1 − λ−1 is bijective. by Fredholm-Riesz theory. [2]
Thus, T −1 − λ−1 has a continuous, everywhere-defined inverse. For such λ, inverting T − λ = T (λ−1 − T −1 )λ
gives
(T − λ)−1 = λ−1 (λ−1 − T −1 )−1 T −1

from which (T − λ)−1 is continuous and everywhere-defined. That is, λ is not in the spectrum of T . Finally,
λ = 0 is not in the spectrum of T , because T −1 exists and is continuous. This establishes the bijection.

Thus, for T −1 compact, the spectrum of T is countable, with no accumulation point in C. Letting
Rλ = (T − λ)−1 , the resolvent relation

Rλ = (Rλ − R0 ) + R0 = (λ − 0)Rλ R0 + R0 = (λRλ + 1) ◦ R0

expresses Rλ as the composition of a continuous operator with a compact operator, proving its compactness.
///

2. Appendix: normal compact operators on Hilbert spaces

We prove an easy special case of a more general fact. Our result is that, for a normal, compact operator
T : X → X on a Hilbert space X, for λ 6= 0 not an eigenfunction, (T − λ)X = X.

[2.1] Im(T − λ) is closed for λ 6= 0

We claim that, for a compact operator T : X → X on a Hilbert space X, for λ 6= 0, the image (T − λ)X of
T − λ is closed.

[1] This is an easy part of Fredholm-Riesz theory. There is a simpler proof that non-zero spectrum is point spectrum
for T a symmetric operator on a Hilbert space, since then the assumed-compact operator T −1 is normal. This easier
discussion is recalled in an appendix.
[2] Again, for T −1 a normal operator on a Hilbert space, there is an easier argument for this bijection, as in the
appendix.

2
 Paul Garrett: Compact resolvents (July 24, 2011)

To see this, let (T − λ)xn → y. First consider the situation that {xn } is bounded. Compactness of T yields
a convergent subsequence of T xn , and we replace xn by this subsequence. Then −λxn = y − T xn converges
to y − lim T xn , so xn is convergent to xo ∈ X, since λ 6= 0, and T xo = y.
Next, when the distance of xn from ker(T − λ) is bounded by b, write xn = x0n + x00n with x00n ∈ ker(T − λ)
and x0n ∈ ker(T − λ)⊥ . Then

|(T − λ)xn | = |(T − λ)x0n | ≤ |T − λ| · b < ∞

That is, (T − λ)xn is bounded.

In general, let X 0 = X/ ker(T − λ) and q : X → X 0 the quotient map. Then T − λ factors through q, by
some continuous S : X 0 → X. There is also the canonical map j : X 0 → ker(T − λ)⊥ so that q ◦ j is the
identity on X 0 .
We claim that there is δ > 0 such that |Sξ| ≥ δ for |ξ| = 1 in X 0 . To see this, suppose Sξn → 0. Then
(T − λ)jξn → 0. Since jξn is bounded, we can replace it by a subsequence so that T jξn is convergent. Then
−λjξn = T jξn is convergent, so jξn is convergent. Thus, ξn is convergent to some ξo , with |ξo | = lim |ξn | = 1.
Apparently, Sξo = lim Sξn = 0, contradiction, proving that |Sξ| ≥ δ > 0 for |ξ| = 1. Returning to the main
argument, suppose that (T − λ)xn → yo . With ξn = qxn , Sξn → yo , and S(ξm − ξn ) → 0. By the claim,
ξm − ξn → 0, so ξn is bounded. That is, the distance from xn to ker(T − λ) is bounded, reducing to the
previous case. ///

[2.2] Normal operators

For normal operators T : X → X, compact or not, for λ not an eigenvalue, T − λ has dense image. To see
this, let y be in the orthogonal complement to the image. Then

0 = h(T − λ)x, yi = hx, (T ∗ − λ̄)yi (for all x ∈ X)

Thus, (T ∗ − λ̄)y = 0. Then

|(T − λ)y|2 = h(T − λ)y, (T − λ)yi = h(T ∗ − λ̄)(T − λ)y, yi = h(T − λ)(T ∗ − λ̄)y, yi = 0

Since λ was not an eigenvalue, y = 0. ///

[2.2.1] Remark: Recall that the residual spectrum of a bounded operator T : X → X is the collection of
λ such that T − λ is injective but (T − λ)X is not dense. Thus, the previous result asserts that (bounded)
normal operators have empty residual spectrum.

[2.2.2] Corollary: For 0 6= λ not an eigenvector of compact, normal T , T − λ is surjective.

Proof: We saw that (T − λ)X is dense for compact T and λ 6= 0 not an eigenvalue. For normal T that
image is also closed, so must be the whole space. ///

[2.2.3] Remark: The continuous spectrum of a bounded operator T is λ with T − λ injective and dense
image, but not closed image. Thus, the corollary asserts that normal compact operators have empty non-zero
continuous spectrum (and empty residual spectrum, as for any bounded, normal operator).

3
 Paul Garrett: Compact resolvents (July 24, 2011)

Bibliography
[Fredholm 1900] I. Fredholm, Sur une nouvelle méthode pour la résolution du problème de Dirichlet, Kong.
Betenskaps-Akad. Förh. Stockholm (1900), 39-46.
[Fredholm 1903] I. Fredholm, Sur une classe d’équations fonctionelles, Acta Math. 27 (1903), 365-390.

[Kato 1966] T. Kato, Perturbation theory for linear operators, Springer, 1966, second edition, 1976, reprinted
1995.
[Riesz 1917] F. Riesz, Über lineare Funktionalgleichungen, Acta Math. 41 (1917), 71-98.
[Riesz-Nagy 1952] F. Riesz, B. Szökefalvi.-Nagy, Functional Analysis, English translation, 1955, L. Boron,
from Lecons d’analyse fonctionelle 1952, F. Ungar, New York, 1955