2024 SPRING 27200 HOMEWORK 5, DUE 05/19 23:59

1. Basic Definitions, 8 Pt Total

1. Suppose X is a complex Banach space, A ∈ L (X, X). Let λ1 , · · · , λm be different eigenvalues of A. Prove
that if v1 , · · · , vm are nonzero eigenvectors corresponding to these eigenvalues, then they must be linearly
independent.

2. Let A be an n × n real matrix. Prove that the operator norm of A on Rn (equipped with the usual
Euclidean length) equals to the largest eigenvalue of AAT .

3. Consider the space ℓ2 (Z), namely complex-valued sequences (an )∞

n=−∞ such that
∞
2 X
(an )∞
n=−∞ ℓ2 := |an |2 < ∞.
n=−∞

Let A be the shift operator:

A(an )∞ ∞
n=−∞ = (bn )n=−∞ ,

where bn = an−1 . Prove that the spectral set of A is the unit circle on the complex plane. Does A have
eigenvalue?

4. Let S be the shift operator on complex square summable sequences ℓ2 (N) defined by

S(an )∞


n=1 := a2 , a3 , · · · .

Prove that σ(S) is the closed unit disk on the complex plane, and each z ∈ C with |z| < 1 is an eigenvalue.

5. (1) Let X be a Banach space and J ∈ L (X, X) an invertible bounded operator. We know that if
P∞
A ∈ L (X, X) has sufficiently small norm, then J − A is invertible. Try to find a series expansion k=0 Jk
for (J − A)−1 in the sense that for any n > 0, there holds
n
X
(J − A)−1 − Jk = O ∥A∥n+1 .

k=0

(2) In linear planning, people usually need to solve linear equations with very huge number of unknowns. In
that case, approximate methods are of particular importance. Sometimes this amounts to inverting matrices
with very large size.
Suppose that J is an orthogonal matrix (i.e. J −1 = J T ) of size n × n, where n = 1010 . Suppose the
perturbation A takes the form  
0 a1 0 · · · 0
0 0 a2 · · · 0
 
A =  .. .. . . ,
 
. . . · · · 0
.. .. .. ..
 
..
. . . . .
1
2 2024 SPRING 27200 HOMEWORK 5, DUE 05/19 23:59

namely, only the elements on the first sub-principal diagonal are non-zero. Suppose all the ak ’s are of absolute
value less than 10−2 . Try to estimate the norm of A, and find an approximate solution of the linear system
 
1
0
 
Ax =  .. 

.
0
with error < 10−4 .

6. Let (wn )∞
n=1 be a complex sequence converging to 0. Prove that the operator

T (an )∞ ∞
n=1 := (wn an )n=1

is a compact operator on ℓp for any 1 ≤ p ≤ ∞.

7. Let K ∈ L2 [a, b] × [a, b] be real valued. What is the transpose operator of the integral operator
Z b
f→ K(x, y)f (y)dy
a
2
on L [a, b]?
Rx
8. Consider the integral operator T : f → 0 f (t)dt on Lp [0, 1] with 1 ≤ p < ∞. Prove that T is compact
and determine its spectrum. Find an explicit expression of the inverse of λId − T when λ ∈ ρ(T ).

2. Exercises on Compact Operators, 5 Pt Total

1. Let (aij )∞
i,j=1 be an infinite complex matrix such that
∞
X
|aij |2 < ∞.
i,j=1

Prove that the operator !∞

∞
X
A: (xn )∞
n=1 → ain xi
i=1 n=1
is a compact operator on ℓ2 .

2. Let H be a complex separable Hilbert space, A ∈ L (H, H). Suppose that there is an orthonormal basis
e1 , e2 , · · · of H of eigenvectors of A. Assume further
• Each eigenvalue of A has finite multiplicity.
• The set of eigenvalues of A has zero as its unique accumulation point.
Prove that A is a compact operator.

3. Compute the Green function for the differential operator

d2
L=− −1
dx2
on the interval [0, 1] with Dirichlet boundary condition.

4. Let Ω ⊂ Rn be a bounded open domain. Prove that the first eigenvalue λ1 corresponding to −∆ on Ω is
simple, namely the eigenspace corresponding to λ1 is one-dimensional. Hint. Prove that if φ minimizes the
Rayleigh quotient, then so does |φ|.
 2024 SPRING 27200 HOMEWORK 5, DUE 05/19 23:59 3

5. Compute the eigenvalues of the unit cube (0, 1)n . Hint. Show that

sin(k1 πx1 ) · · · sin(kn πxn ), k1 , · · · , kn ∈ N

exhaust all the eigenvectors of ∆ with Dirichlet boundary condition.

6. Let Ω ⊂ Rn be a bounded open domain. Let N (r) be the number of eigenvalues of Ω less than r, counting
multiplicity. Show that there are constants a, A > 0 such that
N (r) N (r)
a < lim inf n/2
≤ lim sup n/2 < A.
r→∞ r r→∞ r

This is an initial form of the Weyl asymptotic formula, which relates the geometry of the domain to the
distribution of the spectrum of the Laplacian. Hint. Ω contains a cube and is contained in a cube, and the
eigenvalues of a cube are easily counted.