Killip-Simon problem on two disjoint intervals
Benjamin Eichinger
Institute of Analysis Johannes Kepler University Linz
July 20, 2015
Killip-Simon theorem
Theorem
Letdσ(x) =σ0(x)dx+dσsing(x) be the spectral measure of a one-sided Jacobi matrixJ+. Then the following are equivalent:
(op) J+−J◦+ ∈HS
(sp) The spectral measure dσ is supported on X∪[−2,2]and Z 2
−2
|logσ0(x)|p
4−x2dx+ X
xk∈X
q xk2−4
3
<∞.
Isospectral torus
LetE = [b0,a0]\Sg
j=1(aj,bj)and for a two-sided Jacobi matrixJ we define
r±(z;J) =D
(J±−z)−1e−1±1
2 ,e−1±1
2
E .
Definition
A Jacobi matrixJ is called reflectionlesson E if
a02r+(x+i0) = 1
r−(x−i0) for almost allx ∈E. The classJ(E)is formed by Jacobi matrices, which are reflectionless on their spectral setE.
Spectral theory of periodic Jacobi matrices
Theorem
LetE = [b0,a0]\Sg
j=1(aj,bj).ThenE is the spectrum of a p periodic two-sided Jacobi matrixJ◦ if and only of there exists a polynomialTp(z)such thatTp−1([−2,2]) =E with some additional properties.
Theorem (Magic formula) LetE = [b0,a0]\Sg
j=1(aj,bj)be the spectrum of a p periodic Jacobi matrix. Then
◦
J ∈J(E)if and only if Tp(
◦
J) =S−p+Sp.
Damanik-Killip-Simon theorem
Theorem (Damanik-Killip-Simon theorem) The following are equivalent:
(op) T(J+)−(S+p+ (S+∗)p)∈HS
(sp) The spectral measure dσ is supported on X∪E and Z
E
|logσ0(x)|p
dist(x,R\E)dx+ X
xk∈X
pdist(xk,E)3<∞.
Why does this Magic Formula only exist in the periodic case?
What can we do in the general case?
The Riemann surface R
EFrom now on, we assume thatE = [b0,a0]\(a1,b1).
We define
s(z) = (z−a0)(z−b0)(z−a1)(z−b1)
and
RE ={(P = (z,w) : w2 =s(z)}∪{∞+,∞−} the compactRiemann surfaceof
s(z).
It is convenient to imagine RE as a two-sheeted Riemann sur- face, where p
s(z) > 0, for z >a0 on R+.
a0
b0 a1 b1
a0
b0 a1 b1
ℛ+
ℛ-
⊕
⊖
⊕
⊕
⊕
⊖
⊖ ⊖
The isospectral torus
By the reflectionless property we extendr+ to a meromorphic function onRE:
a20r+(P) =
(a20r+(z) ifP = (z,w)∈ R+
1
r−(z) ifP = (z,w)∈ R−. Let us define
DE ={(x, ) : x ∈[a1,b1], =±1}.
One can show that for anyJ ∈J(E) we have
a20r+((z,w)) = 1
2(w +w0
z −x0
−(z−q0)),
where(x0, 0)∈ DE,q0=q0((x0, 0))∈Randw0 =w0(x0, 0).
Moreover, each such a function definesJ ∈J(E).
The isospectral torus, Abel map
J(E)aa oo r //
!!
DE
>>
~~ A
T
a0
b0 a1 b1
(x0,1)
(x0,-1)
S+
The Abel map
There exists a conformal map fromRE onto a rectangle with vertices −1−τ2 ,1−τ2 ,1+τ2 ,−1+τ2 inC. It is given by
u(z) =C
z
Z
a0
dz ps(z),
x0
x1 S+
S-
u∞+
u∞-
u0
1/2 uc
τ/2
-1/2
-τ/2
The Abel map
u∞+
u∞-
u∞+
u∞-
We call the map
A:RE →C/L z 7→u(z),
whereL=Z+τZ, the Abel map.
Meromorphic functions on R
Eand elliptic functions
Definition
Meromorphic functions onC with periods 1 andτ bear the name elliptic functions.
Theorem
The properly counted (i.e., counted with multiplicity taking into account) number of poles of a nonconstant elliptic functionf(u) in a period parallelogram is equal to the properly counted number of zeros.
Theorem
Let{αk}Nk=1,{βk}Nk=1 are a system of points in a period parallelogram. Thenαk are the zeros on βk are the poles of an elliptic function, if and only if
N
P
k=1
αk − PN
k=1
βk ∈L.
Meromorphic functions on R
Eand elliptic functions
u∞+
u∞-
u0
uc
u1 τ/2
-1/2 1/2
-τ/2
one-to-one correspondence between meromorphic functions on RE and elliptic functions w.r.t. L.
r+, meromorphic function on RE with poles at {∞−,x0}and zeros at {∞+,x1}.
u∞+ +u1−(u∞−+u0)∈Z
Shift on J(E ) and on T
One can show that
r+(z,w) =− 1
b0−z+r+(1)(z,w). Thus, the zero of r+ is the pole ofr+(1).
Let µ=u∞+ −u∞−. Thenu1 =u0−µ mod Z.
J(E)aa oo r+ //
!!
DE
>>
~~ A
T
μ
u0 u1
Functional models for Jacobi matrices
Theorem (Jacobi theta function)
There exists an analytic functionθ1 onC with the following properties:
θ1(u±1) =−θ1(u)
θ1(u±τ) =−e−iπτe∓2πiuθ1(u) Moreover, the zeros ofθ1 are given by
v =m+nτ, m,n ∈Z.
Recall, thatr+ is a meromorphic functions onRE with poles at {∞−,x0}and zeros at {∞+,x1}. Thus, it is given by
r+(z,w) =Cθ1(u−u∞+)θ1(u−u1) θ1(u−u∞−)θ1(u−u0).
Functional models for Jacobi matrices
Now we set
B(u) = θ1(u−u∞+) θ1(u−u∞−) and
kα(u) =C θ1(u−u0)
θ1(u−uc−), Kα(u) = kα(u) pkα(u∞+), wherec−=x−+iκ2 is a certain normalization point.
This functions are no elliptic functions! We have
B(u+1) =B(u) and Kα(u+1) =Kα(u), and
B(u+τ) = θ1(u−u∞++τ)
θ1(u−u∞−+τ) = θ1(u−u∞+) θ1(u−u∞−)
e−i(u−u∞+) e−i(u−u∞−). Thus, we see that
B(u+τ) =eiµB(u).
Functional models for Jacobi matrices
Similarly, we can show that
Kα(u+τ) =eiαKα(u), whereα=u0−uuc−. In general, we define
Kα−kµ(u) = θ1(u−(u0−kµ)) θ1(u−uc−) . Again, we see that
Kα−kµ(u+τ) =ei(α−kµ)Kα−kµ(u).
The space H
2(α)
LetΩ =C\E. For α∈T, we define H2(α) as the space of multivalued, analyticfunctions fα(z) inΩ, which satisfy:
|fα|2 has a harmonic majorant in Ω,
gα(u) =fα(z), satisfiesgα(u+τ) =eiαgα(u).
Note, that|fα|2 is single-valued. The norm is given by the value at infinity of the least harmonic majorant of|fα|2. The corresponding scalar product is given by
hf1α,f2αi= Z
E
f1α(x)f2α(x)ω(dx,∞),
whereω(dx,∞) denotes the harmonic measure of infinity of the domainΩ.
It is given by
ω(dx,∞) = i 2π
x−c ps(z).
Functional models for Jacobi matrices
Theorem Let
enα(z) =Bn(z)Kα−kµ(z) Then
enn ∈N is a ONB of H2(α) enn ∈Z is a ONB of L2(α)
Theorem
Letenα be defined as above, then the multiplication operatorz in L2(α) with respect to the basis{enα}is a Jacobi matrix J =J(α):
zenα =an(α)en−1α +bn(α)enα+an+1(α)en+1α ,
Sketch of the proof
We only sketch, why the multiplication byz is a Jacobi matrix. We have
r+(z) =− 1
z −b0+a12r(1)(z) =Cθ1(u−u∞+)θ1(u−u1) θ1(u−u∞−)θ1(u−u0). That is,
a0r+ =−BKα−µ
Kα and a1r+(1)=−BKα−2µ Kα−µ. Thus we have
a0Kα
BKα−µ = (z −b0)−a1BKα−2µ Kα−µ ⇔ zBKα−µ=a0Kα+b0BKα−µ+a1B2Kα−2µ⇔ ze1α =a0e0α+b0e1α+a1e2α
Proof of the magic formula
S−2JS =J ⇔2µ∈Z. This is only a property of the spectrum not ofJ! Consider, in this case the basis{enα} has the following form
{. . . ,B−1Kα+µ,Kα,BKα+µ,B2Kα, . . .}.
Recallµ=u∞+ −u∞−. There exists a function ψ(z,w) =
θ1(u−u∞+) θ1(u−u∞−)
2
=B(u)2, on RE. ψ+ψ−1 is a function on Ω =C\E. Thus,
ψ(z,w) + 1
ψ(z,w) =T2(z).
This is the magic formula.
`2 en S2 J(α) L2(α) enα ψ z
Functional models in the general case
Now let us consider the case 2µ /∈Z. In this case we define
ψ= θ1(u−u∞+)θ1(u−uc+)
θ1(u−u∞−)θ1(u−uc−) =B(u)Bc(u), whereµ+µc ∈Z. In this case we have
ψ(z,w) + 1
ψ(z,w) =λ0z +d0+ λ1
d1−z = ∆(z).
Functional models in the general case
Letkα(z,c) =kcα(z) be the reproducing kernels of the space H2(α). We defineKψ(α) =H2(α) ψH2(α). Kψ is spanned by:
f0α(z) =λ(α) kdα
1(z) kkdα
1(z)k and f1α(z) =Bd1(z)kα+µ(z) kkαµk , Thus,H2(α) =Kψ(α)⊕ψKψ(α) +ψ2Kψ(α)⊕. . . .
Theorem
The system of functions
fnα =
(ψmf0α, n=2m ψmf1α, n=2m+1
forms an orthonormal basis in H2(α) forn ∈Nand forms an orthonormal basis in L2(α) forn ∈Z.
Functional models in the general case, SMP-matrices
Definition
We call the operatorA(α)∈L(`2), which corresponds to the multiplication byz in this basis a SMP-matrix.
A(α) is two periodic. σ(A(α)) =E
A=
. ..
. .. ... A A∗ B A
A∗ B A A∗ . .. ...
. ..
,
A=
0 0 p0 p1
, B=
p0q0+d0 p1q0
p1q0 p1q1
.
Byconstruction, we obtain the magic formula for SMP matrices.
`2 en S2 A(α) L2(α) fnα ψ z
Theorem
LetA(E) be the set of all SMP matrices of period two with their spectrum on E. Then we have:
A∈A(E) ⇔ ∆(A) =λ0A+d0+λ1(A−d1)−1 =S2+S−2
The Jacobi flow on SMP-matrices
Definition
We define the mapF : A(E)→J(E), by F(A(α)) =J(α).
Lemma
LetkP+A(α)e−1k2 and{ak,bk}be the Jacobi parameters of J(α).
Then we have
p1q1=b−1, p20+q02=a02 It follows immediately, sincee−1α =f−1α .
The Jacobi flow on SMP-matrices
Definition
The Jacobi flow inA(E)is defined by J(A(α)) =A(α−µ).
A F //
J
J
S
A F //J
Explicit formulas for the Jacobi flow
Theorem
LetU(α)be the periodic 2×2-block diagonal unitary matrix given by
U(α)
e2m e2m+1
=
e2m e2m+1
u(α), (1) where
u(α) = 1
q
p20(α) +p12(α)
p0(α) p1(α) p1(α) −p0(α)
(2)
Then
JA(α) :=A(µ−1α) =S−1U(α)∗A(α)U(α)S. (3)