Discrete spectrum (mathematics)

In mathematics, specifically in spectral theory, a discrete spectrum of a closed linear operator is defined as the set of isolated points of its spectrum such that the rank of the corresponding Riesz projector is finite.

The discrete spectrum can also be defined as the set of normal eigenvalues.

Definition

A point ฮปโˆˆโ„‚ in the spectrum ฯƒ(A) of a closed linear operator A:๐”…โ†’๐”… in the Banach space ๐”… with domain ๐”‡(A)โŠ‚๐”… is said to belong to the discrete spectrum ฯƒd(A) of A if the following two conditions are satisfied:

  1. ฮป is an isolated point in ฯƒ(A);
  2. The rank of the corresponding Riesz projector Pฮป=12ฯ€iโˆฎฮ“(zI๐”…โˆ’A)โˆ’1dz is finite.

Here, I๐”… is the identity operator in the Banach space ๐”…, and ฮ“โŠ‚โ„‚ is a simple closed counterclockwise-oriented curve bounding an open region ฮฉโŠ‚โ„‚ such that ฮป is the only point of the spectrum of A in the closure of ฮฉ; that is, ฯƒ(A)โˆฉฮฉโ€พ={ฮป}.

Normal eigenvalues

The set of points in the discrete spectrum is equal to the set of normal eigenvalues.

ฯƒd(A)={normal eigenvalues ofย A}.

Root lineal

Let ๐”… be a Banach space. Consider a partially defined linear operator A:๐”…โ†’๐”… with domain ๐”‡(A). The root lineal ๐”ฮป(A) corresponding to an eigenvalue ฮปโˆˆฯƒp(A) is defined as the set of elements x such that x,(Aโˆ’ฮปI๐”…)x,(Aโˆ’ฮปI๐”…)2x,โ€ฆ all belong to ๐”‡(A), and that after finitely many steps, we end up with zero: (Aโˆ’ฮปI๐”…)kx=0.

This set is a linear manifold but is not necessarily closed. If it is closed (for example, when it is finite-dimensional), it is called the generalized eigenspace of A corresponding to the eigenvalue ฮป.

Normal eigenvalue

An eigenvalue ฮปโˆˆฯƒp(A) of a closed linear operator A:๐”…โ†’๐”… in the Banach space ๐”… with domain ๐”‡(A)โŠ‚๐”… is called normal (in the original terminology, ฮป corresponds to a normally splitting finite-dimensional root subspace) if the following two conditions are satisfied:

  1. The algebraic multiplicity of ฮป is finite: ฮฝ=dimโก๐”ฮป(A)<โˆž, where ๐”ฮป(A) is the root lineal of A corresponding to the eigenvalue ฮป;
  2. The space ๐”… can be decomposed into a direct sum ๐”…=๐”ฮป(A)โŠ•๐”‘ฮป, where ๐”‘ฮป is an invariant subspace of A in which Aโˆ’ฮปI๐”… has a bounded inverse.

Equivalent characterizations

Equivalent characterizationsย (:โ€ŠTheorem III.88โ€Š)โ€”Let A:๐”…โ†’๐”… be a closed linear densely defined operator in the Banach space ๐”…, then the following are equivalent:

  1. ฮปโˆˆฯƒ(A) is a normal eigenvalue;
  2. ฮปโˆˆฯƒ(A) is an isolated point in ฯƒ(A) and Aโˆ’ฮปI๐”… is semi-Fredholm;
  3. ฮปโˆˆฯƒ(A) is an isolated point in ฯƒ(A) and Aโˆ’ฮปI๐”… is Fredholm;
  4. ฮปโˆˆฯƒ(A) is an isolated point in ฯƒ(A) and Aโˆ’ฮปI๐”… is Fredholm of index zero;
  5. ฮปโˆˆฯƒ(A) is an isolated point in ฯƒ(A) and the rank of the corresponding Riesz projector Pฮป is finite;
  6. ฮปโˆˆฯƒ(A) is an isolated point in ฯƒ(A), its algebraic multiplicity ฮฝ=dimโก๐”ฮป(A) is finite, and the range of Aโˆ’ฮปI๐”… is closed.

In all such cases, the root lineal ๐”ฮป(A) is closed, and equals the range of the Riesz projector.

Relation to other spectra

Isolated eigenvalues of finite algebraic multiplicity

In general, the rank of the Riesz projector can be larger than the dimension of the root lineal ๐”ฮป of the corresponding eigenvalue, and in particular it is possible to have dim๐”ฮป<โˆž, rankPฮป=โˆž. So, there is the following inclusion:

ฯƒd(A)โŠ‚{isolated points of the spectrum ofย Aย with finite algebraic multiplicity}.

In particular, for a quasinilpotent operator

Q:l2(โ„•)โ†’l2(โ„•),Q:(a1,a2,a3,โ€ฆ)โ†ฆ(0,a1/2,a2/22,a3/23,โ€ฆ),

one has ๐”ฮป(Q)={0}, rankPฮป=โˆž. Therefore, ฮป=0 is an isolated eigenvalue of finite algebraic multiplicity, but it is not in the discrete spectrum: ฯƒ(Q)={0}, ฯƒd(Q)=โˆ…๏ธ€.

Point spectrum

The discrete spectrum ฯƒd(A) of an operator A is not to be confused with the point spectrum ฯƒp(A), which is defined as the set of eigenvalues of A. Each point of the discrete spectrum is an eigenvalue, so

ฯƒd(A)โŠ‚ฯƒp(A).

However, they may be unequal. An eigenvalue may not be an isolated point of the spectrum, or it may be isolated, but with an infinite-rank Riesz projector. For example, for the left shift operator, L:l2(โ„•)โ†’l2(โ„•),L:(a1,a2,a3,โ€ฆ)โ†ฆ(a2,a3,a4,โ€ฆ), the point spectrum is the open unit disc ๐”ป1 in the complex plane, the full spectrum is the closed unit disc ๐”ป1โ€พ, and the discrete spectrum is empty:

ฯƒp(L)=๐”ป1,ฯƒ(L)=๐”ป1โ€พ,ฯƒd(L)=โˆ…๏ธ€.

This is because ฯƒp(L) has no isolated points.

Spectral decomposition

The spectrum of a closed operator A:๐”…โ†’๐”… in a Banach space ๐”… can be decomposed into the union of two disjoint sets: the discrete spectrum and the fifth type of the essential spectrum (see page for the definition of each type):

ฯƒ(A)=ฯƒd(A)โˆชฯƒess,5(A).

See also