SCHRÖDINGER OPERATORS

WITH SINGULAR BOUNDARY CONDITIONS

JAN DEREZIŃSKI
Department of Mathematical Methods in Physics
 Plan of the talk:
1. One-dimensional Schrödinger operators −∂x2 + V (x).
J.D., V.Georgescu, also many classic authors.
1 1
2 2


2. Bessel operators −∂x + m − 4 x2 .
L.Bruneau, J.D., V.Georgescu, S.Richard
3. Whittaker operators −∂x + m − 4 x2 − βx . J.D., J.Faupin, S.Richard,
1 1
2 2

Q.N.Nguyen
1 1


4. Perturbed Bessel operators −∂x2 2
+ m − 4 x2 + Q(x). J.D., J.Faupin
 ONE-DIMENSIONAL
SCHRÖDINGER OPERATORS
Suppose that R+ 3 x 7→ V (x) is a function in L1loc(R+) bounded near
infinity, possibly complex valued. Consider the one-dimensional Schrödinger
operator
L := −∂x2 + V (x).
We would like to describe closed (if possible self-adjoint) realizations of L
on L2(R+). There are two obvious closed realizations: the minimal Lmin
and the maximal Lmax with domains given by

D(L ) := f ∈ L2(R+) | Lf ∈ L2(R+) ,

max


D(Lmin) := the closure of {f ∈ D(Lmax) | f = 0 near 0},

the closure taken with respect to the graph norm of Lmax.

 One can show that dim D(Lmax)/D(Lmin) is either 0 or 2.
In the latter case there exists a one-parameter family of operators L• that
satisfy Lmin ⊂ L• ⊂ Lmax defined by boundary conditions (b.c.) near zero.
If V ∈ L1 near zero, these b.c are easy to describe:

D(Lκ) := f ∈ D(Lmax) | f (0) = κf 0(0) .



κ = 0 is called the Dirichlet b.c., κ = ∞ the Neumann b.c., the remain-

ing are mixed or Robin b.c.. If V 6∈ L1 near zero, the situation is more
complicated.
 In most of the literature it is assumed that V is real. Then the operator L
with the domain Cc∞(R+) is Hermitian (symmetric), but not necessarily self-
adjoint. The closure of L with this domain coincides with Lmin, its adjoint
L∗ coincides with Lmax. If Lmin 6= Lmax, then self-adjoint extensions of L
are in between Lmin and Lmax.
One can apply von Neumann’s method of defining self-adjoint extentions.
One looks for eigenvectors in L2(R+) of

L∗f± = ±if±.

If dim D(Lmax)/D(Lmin) = 2, Then

D(Lα) = D(Lmin) + C(eiαf+ + e−iαf−)

gives all self-adjoint extensions of L. Not an optimal approach—it requires

solving an unnecessarily difficult eigenvalue problem.
 Let us go back to a possibly complex V .
Let N (L − λ) denote the space of eigenfunctions of L with eigenvalue
λ ∈ C (not necessarily square integrable). One can show the alternative

dimD(Lmax)/D(Lmin) = 0
⇔ dim f ∈ N (L − λ) | f ∈ L2 near 0 ≤ 1,

∀λ ∈ C;
dimD(Lmax)/D(Lmin) = 2
⇔ dim f ∈ N (L − λ) | f ∈ L2 near 0 = 2,

∀λ ∈ C.

Usually, the following approach to define closed extensions of Lmin is much

more convenient than von Neumann’s:
Fix λ (e.g. λ = 0). Choose a cutoff function ξ ∈ C ∞(R+) equal 1 near 0
and 0 near ∞. For h ∈ N (L − λ) set

D(L•) := D(Lmin) + Cξh.

Note that L• does not depend on the cutoff ξ.

 Introduce the Wronskian of two functions h1 and h2:

W(h1, h2, x) = h1(x)h02(x) − h01(x)h2(x)

Note that for h1, h2 ∈ N (L − λ), the Wronskian W(h1, h2, x) does not
depend on x. Thus it defines a symplectic form on N (L − λ).
Moreover, for f, g ∈ D(Lmax) the Wronskian at 0 is well-defined:

lim W(f, g; x) =: W(f, g; 0)

x&0

exists and defines a continuous bilinear form on D(Lmax).

Here is an even better approach to defining close realizations of L:

D(L•) = {f ∈ D(Lmax) | W(h0, f ; 0) = 0},

where h0 is an approximate eigenfunction of L.

 Later we will need a few integral kernels naturally associated with the
operator L. Let h1, h2 ∈ N (L − λ) be linearly independent. The canonical
bisolution of L − λ defined by the integral kernel
1

G↔(λ; x, y) = h1(x)h2(y) − h2(x)h1(y) ,
W(h1, h2)
does not depend on the choice of h1, h2. G↔(λ) is usually unbounded on
L2(R+). It satisfies

∂x2


− + V (x) − λ G↔(λ; x, y) = 0.

We will use the term Green’s operator as a synonym for a right inverse
of L − λ (not necessarily bounded). In other words, the integral kernel
G•(λ; x, y) of Green’s operator G•(λ) satisfies
2


− ∂x + V (x) − λ G•(λ; x, y) = δ(x − y).
 We have various types of Green’s operators:
1. the forward Green’s operator

G→(λ; x, y) := θ(x − y)G↔(λ; x, y),

2. the backward Green’s operator

G←(λ; x, y) := −θ(y − x)G↔(λ; x, y).

3. Choose h1, h2 ∈ N (L−λ). The two-sided Green’s operator correspond-

ing to b.c. near 0 given by h1, resp. near ∞ given by h2:
h1(x)h2(y)θ(x − y) + h2(x)h1(y)θ(y − x)
G•(λ; x, y) :=
W(h1, h2)
 Suppose we have two Schrödinger operators

L0 = −∂x2 + V 0(x),
L = −∂x2 + V (x) = L0 + Q(x).

Suppose we know an eigenfunction of the unperturbed operator u0 ∈

N (L0 −λ). Then one can try to construct an eigenfunction of the perturbed
operator u ∈ N (L − λ) by applying
0

−1 0
u = 1l + G•(λ)Q u ,

where G0•(λ) is one of the Green’s operators of L0 − λ. This is a gener-

alization of the Lippmann-Schwinger equation well-known from Quantum
Mechanics.
 BESSEL OPERATORS
One of the most important families of exactly solvable 1-dimensional
Schrödinger operators is the family of Bessel operators
2 c
−∂x + 2 .
x
We will allow c to be complex.
As is well-known, it is convenient to set c = m2 − 41 , so that the Bessel
operator is rewritten as
 1 1
L0m2
:= −∂x2
+ m − 2
2
.
4 x
We will often assume that Re(m) ≥ 0, because L0m2 depends only on m2.
 Many operators in mathematics and physics can be reduced to Bessel
operators:
1. the usual Laplacian on the halfline, Dirichlet m = 21 , Neumann m = − 12 ;
2. the usual Laplacian in dimension d ≥ 3, m = d2 − 1 + `, ` ∈ N0;
θ
3. the 2d Aharonov-Bohm Hamiltonian with flux θ, m = 2π + n, n ∈ Z,
2πn
4. the Laplacian on a conical surface of angle α, m = α , n ∈ Z;
5. the Laplacian on a wedge of angle α with Dirichlet or Neumann b.c.,
πn
m= α, n ∈ Z;
6. perturbed Bessel operators with m complex define Regge poles,
7. three-body systems with contact interactions.
8. generators of sl(2, R).
 The zero energy eigenvalue problem L0m2 f = 0 is easy:
1 +m 1 1 1
x 2 , x 2 −m , m 6= 0; x ,
2 x ln(x),
2 m = 0.

One can distinguish 3 regimes with a different behavior of eigensolutions:

1. Re(m) > 0. Eigensolutions of L0m2 can be divided into principal, and
1 +m
non-principal, all the others. Principal solutions behave as x 2 and are
1
more regular than non-principal ones, which behave as x 2 −m.
2. Re(m) = 0, m 6= 0. Eigensolutions of L0m2 are spanned by eigensolu-
1 1
tions with a comparable oscillating behavior x 2 +m and x 2 −m near zero.
3. m = 0. Eigensolutions of L0m2 are spanned by a principal solution
behaving like x0. All others are non-principal, behave like x0 ln(x), and
are less regular.
 Let us now sketch the theory of closed realizations of L0m2 on L2(R+).
First of all, the minimal and maximal realization of L0m2 denoted by L0,min
m2
resp. L0,max
m2
, satisfy

|Re(m)| ≥ 1 implies L0,min

m 2 = L0,max
m2
,
|Re(m)| < 1 implies dim D(L0,max
m2
)/D(L 0,min
m2
) = 2.

Thus for |Re(m)| < 1 there exists a 1-parameter family of closed realisations
of L0m2 between L0,min
m 2 and L 0,max
m2
defined by b.c. at zero. To describe all
closed realizations of Bessel operators one can introduce the following three
holomorphic families of operators:
0
{−1 < Re(m)} 3 m 7→ Hm ,
0


{−1 < Re(m) < 1} × C ∪ {∞} 3 (m, κ) 7→ Hm,κ ,
C ∪ {∞} 3 ν 7→ H00,ν .


 0
The family Hm is the most basic one. It is called homogeneous. For
1 ≤ Re(m) it is the unique closed realization of L0m2 . Then it is extended
to the strip −1 < Re(m) < 1 by analytic continuation. Its domain is
1
defined by the boundary condition ∼ x 2 +m at zero.
0
The operator Hm,κ is defined by
0 1 1 −m
D(Hm,κ ) = D(Lmin
m2 ) + C(x 2 +m + κx 2 )ξ
1 1 −m
= {f ∈ D(Lmax
m2 ) | W(f, x 2 +m + κx 2 ; 0) = 0}.

It is holomorphic except for a singularity at (m, κ) = (0, −1).

Finally, for the special case m = 0, H00,ν is defined by
0,ν 1 1
D(Hm ) = D(Lmin
m2 ) + C(x ln(x) + νx )ξ
2 2

1 1
= {f ∈ D(Lmax
m2 ) | W(f, x ln(x) + νx ; 0) = 0}.
2 2
 Let Uτ be the group of dilations:

(Uτ f )(x) = eτ /2f (eτ x).

For any m with |Re(m)| < 1 and any κ, ν ∈ C ∪ {∞}, we have

Uτ Hm,κU−τ = e−2τ Hm,e−2τ mκ,

Uτ H0ν U−τ = e−2τ H0ν+τ .

In particular, only

Hm,0 = Hm, Hm,∞ = H−m, H0∞ = H0 are homogeneous (of degree 2).

If A is an operator, then the transformation Rτ (A) := e2τ Uτ AU−τ . will

be called the renormalization group action. Operator is homogeneous iff it
is its fixed point.
 Self-adjoint extensions of the Hermitian operator
2 1
1
Lα = −∂x + − + α 2 .
4 x
K—Krein, F—Friedrichs, dashed line—single bound state, dotted line—
infinite sequence of bound states.
 For k 6= 0 the eigenvalue problem L0m2 f = k 2f reduces to the Bessel
equation. N (L0m2 + k 2) is spanned by
1
 2 m √ x 2 +m
u0m(x, k) := xIm(kx), 0
um(x) ' near 0;
k Γ(m + 1)
 k m √
0
vm(x, k) := xKm(kx), decaying exponentially
2
where Im is the modified Bessel function and Km the Macdonald function.
u0m is principal for Re(m) > 0 and for m = 0.
0
vm is always non-principal.
u0m for Re(m) < 0 and m 6= −1, −2, . . . is a distinguished non-principal
eigenfunction. (For m = −1, −2, . . . it is proportional to u0−m).
 We have
1
2 +m
x
u0m(x, 0) = ,
Γ(m + 1)
1
Γ(m)x 2 −m
0
vm (x, 0) = , Re(m) ≥ 0, m 6= 0.
2
It is convenient to introduce another eigensolution
0
wm 0
(x, k) = w−m (x, k) ∼ e−kx x → ∞,
0
which differs from vm (x, k) only by a different normalization:
r   r
0 2k 2 m
0 2xk
wm(x, k) = vm(x, k) = Km(kx).
π k π
 The Bessel operator for m = 0 often needs a separate treatment. This
case is actually very important – it corresponds to the 2-dimensional Lapla-
cian in the s-wave sector.
For instance, v00(·, k) does not have a limit at k = 0. To treat the case
m = 0 in a satisfactory way it is useful to introduce a family of non-principal
eigenfunctions of L00:
 k  
p00(x, k) := −v00(x, k) − ln + γ u00(x, k),
2
where γ denotes Euler’s constant. At k = 0 it coincides with the logarithmic
solution:
1
p00(x, 0) = x 2 ln(x).
 As explained in the previous subsection, with L0m2 + k 2 one can associate
various Green’s operators: The most important are
1. the forward Green’s operator G0m2,→(−k 2);
2. the backward Green’s operator G0m2,←(−k 2);
3. the two-sided Green’s operator with homogeneous boundary conditions,
G0m2,./(−k 2), using u0m and vm
0
, for brevity often called two-sided;
4. for m = 0, additionally, the two-sided Green’s operator logarithmic near
zero G00,(−k 2), using u00, p00, for brevity called logarithmic.
For Re(m) > −1, Re(k) > 0 the two-sided Green’s operator with homo-
geneous b.c. is bounded on L2(R+) and is with the resolvent of Hm
0
:
0 2 0 2 −1


Gm2,./(−k ) = Hm + k .

However, the integral kernel G0m2,./(−k 2; x, y) is well defined and useful also
for other values of k and m, when it does not define a bounded operator.
 THE WHITTAKER OPERATOR
The radial Schrödinger operator with the Coulomb potential
2

2 1 1 β
Lβ,m2 := −∂x + m − −
4 x2 x
will be called the Whittaker operator. It is an example of an perturbed
Bessel operator.
It has a distinguished family of closed realizations

C × {m ∈ C | Re(m) > −1} 3 (β, m) 7→ Hβ,m ,

holomorphic except for a singularity at (0, − 12 ). We will call them pure.

They are generalizations of the homogeneous family Hm to the Whittaker
case. In particular, for β = 0 they coincide with Bessel operators:

H0,m = Hm.
 If Re(m) ≥ 1 the boundary condition is not needed. For Re(m) > −1
Hβ,m can be defined by analytic continuation.
Alternatively, if Re(m) ≥ − 21 , we can use a simplified boundary condition:
1
D(Hβ,m) = {f ∈ D(Lmax
β,m2 ) | W(f, x 2 +m ; 0) = 0}.

For all Re(m) > −1 pure boundary conditions are defined by

n  1 +m
 o
max β
D(Hβ,m) = f ∈ D(Lβ,m2 ) | W f, x 2 1 − 1+2m x ; 0 = 0 .
 The singularity at (β, m) = (0, − 12 ) is quite curious: it is invisible when
we consider just the variable m. In fact, the Bessel operator

{Re(m) > −1} 3 m 7→ Hm = H0,m

is holomorphic.
H− 1 is the Laplacian with the Neumann boundary condition;
2
H 1 is the Laplacian with the Dirichlet boundary condition.
2
Thus one has
H0,− 1 6= H0, 1 .
2 2
 If we introduce the Coulomb potential, then

whenever β 6= 0, Hβ,− 1 = Hβ, 1 .

2 2

The function

(β, m) 7→ Hβ,m is holomorphic around (0, 12 ).

In particular, in the sense of the strong resolvent limit

lim Hβ, 1 = H0, 1 .

β→0 2 2

But
lim Hβ,− 1 = H0, 1 6= H0,− 1
β→0 2 2 2

Thus (β, m) 7→ Hβ,m is not even continuous near (0, − 12 ).

 PERTURBED BESSEL OPERATORS
Consider now m ∈ C and complex Q ∈ L1loc(R+)
 1 1
Lm2 := −∂x2 + m2 − + Q(x).
4 x2
Proposition 1. Let Re(m) ≥ 0, k ∈ C and suppose that
Z 1
x|Q(x)|dx < ∞, if m 6= 0;
0
Z 1
x(1 + |ln(x)|)|Q(x)|dx < ∞, if m = 0.
0
1
Suppose that g 0 ∈ N (L0m2 + k 2) such that g 0(x) = O(x 2 +Re(m)) near 0.
Then, there exists a unique g ∈ N (Lm2 + k 2) such that,
0 1
g(x) − g (x) = o(x 2 +Re(m) ),
1
∂xg(x) − ∂xg 0(x) = o(x− 2 +Re(m)), x → 0.
 If we want to well approximate all unperturbed solutions, including the
more singular ones, we need to strengthen the assumption on the perturba-
tion.
Proposition 2. Let Re(m) ≥ 0, k ∈ C and suppose that
Z 1
x1−2Re(m)|Q(x)|dx < ∞, if m 6= 0;
0
Z 1
2


x 1 + (ln(x)) |Q(x)|dx < ∞, if m = 0.
0

Suppose that g 0 ∈ N (L0m2 + k 2). Then, there exists a unique

g ∈ N (Lm2 + k 2) such that
0 1
g(x) − g (x) = o(x 2 +Re(m) ),
1
∂xg(x) − ∂xg 0(x) = o(x− 2 +Re(m)), x → 0.
 Here are consequences of Propositions 1 and 2:
Corollary 3. Let m ∈ C, k ∈ C and suppose that
Z 1
ε
x1−ε|Q(x)|dx < ∞, ε ≥ 0, Re(m) ≥ − , m 6= 0;
0 2
Z 1
x(1 + |ln(x)|)|Q(x)|dx < ∞, m = 0.
0

Then there exists a unique um(·, k) ∈ N (Lm2 + k 2) that satisfies

1 +|Re(m)|
um(x, k) − u0m(x, k) = o(x 2 ),
− 12 +|Re(m)|
∂xum(x, k) − ∂xu0m(x, k) = o(x ), x → 0.

Note that if |Q(x)| . |x|α near 0, then the above condition is satisfied for
α > −2 + ε.
Corollary 4. Let k ∈ C and suppose that
Z 1
x(1 + (ln(x))2)|Q(x)|dx < ∞, m = 0.
0

Then there exists a unique p0(·, k) ∈ N (Lm2 + k 2) such that

1
p0(x, k) − p00(x, k) = o(x 2 ),
1
∂xp0(x, k) − ∂xp00(x, k) = o(x− 2 ), x → 0.
 Conditions of Proposition 1 are the minimal assumptions near zero for our
purposes. They guarantee the existence of distinguished eigenfunctions um
with Re(m) ≥ 0, principal for Re(m) > 0 and m = 0. They also imply
that the behavior near zero of non-principal eigenfunctions is roughly as in
the unperturbed case:
Proposition 5. Let Re(m) ≥ 0, Re(k) ≥ 0. Under the assumptions of
Proposition 1 for all g ∈ N (Lm2 + k 2), we have
1 − 12 −Re(m)
g(x) = O(x 2 −Re(m) ), ∂xg(x) = O(x ),
1 − 12
g(x) = O(x ln(x)),
2 ∂xg(x) = O(x ln(x)), x → 0.
 It seems that to have distinguished non-principal eigenfunctions one needs
to impose stronger conditions on Q. In particular, under the conditions of
Corollary 3 um is constructed only in the region Re(m) ≥ − 2ε . This suggests
the following question:
Open Problem 6. Let Q satisfy the condition of Corollary 3. Does it
imply that the function m 7→ um(·, k) extends holomorphically (or at
least meromorphically) to the whole C? (This is true for the Coulomb
potential, see S.Richard, J.D.).
 Let us now consider the behavior near infinity. To prove the existence of
solutions well approximating exponentially decaying solutions, called Jost
solutions, we need the so-called short-range condition on the potential.
Proposition 7. Let m ∈ C. Suppose that
Z ∞
|Q(x)|dx < ∞.
1

Let k 6= 0 be such that Re(k) ≥ 0. Then there exists a unique wm(·, k) =

w−m(·, k) ∈ N (Lm2 + k 2) such that
0
wm(x, k) − wm (x, k) = o(e−xRe(k)),
0
∂xwm(x, k) − ∂xwm (x, k) = o(e−xRe(k)), x → ∞.

Proposition 7 does not cover the zero energy, that is, k = 0.

Proposition 8. Let m ∈ C. Suppose that
Z ∞
xδ |Q(x)|dx < ∞, if m 6= 0,


with δ = 1 + 2 max Re(m), 0 ;
Z ∞ 1

x(1 + ln(x))|Q(x)|dx < ∞, if m = 0.

1

Then there exists a unique qm ∈ N (Lm2 ) such that

1 1
qm(x) − x 2 +m = o(x 2 −Re(m)),
1 +m − 21 −Re(m)
∂xqm(x) − ∂xx 2 = o(x ), x → ∞.
Proposition 9. Let m = 0. Suppose that
Z ∞
x(1 + (ln(x))2)|Q(x)|dx < ∞.
1

Then there exists a unique q0,ln ∈ N (L0) such that

1 1
q0,ln(x) − x 2 ln(x) = o(x 2 ),
1 1
∂xq0,ln(x) − ∂xx 2 ln(x) = o(x− 2 ), x → ∞.
 The zero energy eigenequation near infinity is equivalent to the zero
energy eigenequation near zero:
 1 1   1 1 
−∂x2 + m2 − 2
+ Q(x) = y 3
− ∂y
2
+ m2
− 2
+ Q̃(y) y,
4 x 4 y
1
y= , Q̃(y) := y −4Q(y −1).
x
Note also a simple relationship between the integral conditions near zero on
Q and near infinity on Q̃:
Z 1 Z ∞
x1−ε|Q(x)|dx = y 1+ε|Q̃(y)|dy,
0
Z 1 Z1 ∞
x(1 + | ln(x)|α)|Q(x)|dx = y(1 + | ln(y)|α)|Q̃(y)|dy.
0 1
Thus one can derive Propositions 8 and 9 from the k = 0 case of Corollaries
3 and 4.
 The main tools used in the construction of eigenfunctions are various
Green’s operators for the unperturbed Bessel operator. The forward Green’s
operator is used in Propositions 1, 2 and their corollaries. For instance,
0 2

−1 0
um(·, k) = 1l + Gm2,→(−k )Q um(·, k),
0 2

−1 0
p0(·, k) = 1l + G0,→(−k )Q p0(·, k).

The backward Green’s operator is used in Propositions 7, 8 and 9:

0 2

−1 0
wm(·, k) = 1l + Gm2,←(−k )Q wm(·, k),
0

−1 0
qm = 1l + Gm2,←(0)Q um(·, 0),
0

−1 0
q0,ln = 1l + G0,←(0)Q p0(·, 0).

In quantum physics the equation for the Jost solution w(·, k) is called the
Lippmann–Schwinger Equation.
 ε
If assumptions of Proposition 1 holds and 2 < Re(m), then Corollary 3
guarantees the existence only of um(·, k), but not of u−m(·, k). Therefore,
in this case it is more complicated to describe non-principal solutions. One
way to do this is to use compressed two-sided Green’s operators:
Proposition 10. Suppose the assumptions of Proposition 1 hold. If a
is small enough, the following functions are well defined and belong to
N (Lm2 + k 2) on ]0, a[:
./(a) 0(a) 2

−1 0
u−m (·, k) := 1l + Gm,./(−k )Q u−m(·, k),
(a) 0(a) 2

−1 0
p0 (·, k) := 1l + G0, (−k )Q p0(·, k).

Here,
G(a)
• (x, y) = 1l[0,a] (x)G• (x, y)1l[0,a] (y).
 Unfortunately, the construction of Proposition 10 involves inverting a com-
plicated integral operator. Alternatively, choose a non-negative integer n,
expand the denominator into a power series retaining n first terms, fix a = 1
(quite arbitrarily) and set
n
0[n]
X
u−m(x, k) = (−G0(1) j 0
./ (0)Q) u−m (x, k).
j=0

Proposition 11. Let Re(k) ≥ 0. Let n ∈ N such that

Z 1
x1−ε|Q(x)|dx < ∞, ε ≥ 0,
0

is satisfied for − 2ε (n + 1) ≤ Re(−m) ≤ 0. Then there exists a unique

[n]
u−m(·, k) ∈ N (Lm2 + k 2) such that
[n] 0[n] 1
u−m(x, k) − u−m(x, k) = o(x 2 +Re(m)),
[n] 0[n] 1
∂xu−m(x, k) − ∂xu−m(x, k) = o(x− 2 +Re(m)), x → 0.
 0[n]
Boundary conditions determined by u−m(·, k) still have an unpleasant
feature – they depend on k. If we want to have boundary conditions inde-
pendent of k we need to assume that |Re(m)| < 1. Then it is reasonable
to choose k = 0, which we do setting
0[n] 0[n]
u−m(x) := u−m(x, 0). (1)

In particular, under the condition |Re(m)| < 1 in Proposition 11 we can

0[n] 0[n]
replace u−m(·, k) with u−m(·).
We have seen the condition |Re(m)| < 1 already in the L2 theory of
Bessel operators.
 An important object of our analysis is the Jost function Wm(k), that is
the Wronskian of the two main solutions um(·, k) and vm(·, k).
Proposition 12. Assume Re(m) > −1, as well as assumptions of
Corollary 3. Then

lim Wm(k) = 1, Re(k) ≥ 0. (2)

|k|→∞

Note the assumption Re(m) > −1 that appears in the above proposition—
again anticipating the basic condition needed in the L2 analysis.
 Let us now dscribe close realizations of perturbed Bessel operators. As
usual, we can introduce the minimal and maximal Bessel perturbed Bessel
operator Lmin
m2 and Lmax
m2
. Under the assumptions of Propositions 1 the basic
picture is the same as in the unperturbed case:

|Re(m)| ≥ 1 implies Lmin

m2 = Lmax
m2 ,
|Re(m)| < 1 implies dim D(Lmax
m2 )/D(L min
m2 ) = 2.

In particular, for |Re(m)| < 1, beside the minimal and maximal realiza-
tions, there exists a 1-parameter family of closed realizations of Lm2 defined
by b.c. at zero. They can be fixed by specifying continuous linear function-
als on D(Lmax
m2 ) vanishing on D(L min
m2
), called boundary functionals, forming
the boundary space
max min 0


Bm2 := D(Lm2 )/D(Lm2 ) , where the prime denotes the dual.
 As we discussed in the general theory, boundary functionals can be effi-
ciently described by the Wronskian at zero with an eigenfunctions of Lm2 .
Zero-energy eigenfunctions are the simplest. In practice we can use approx-
imate ones.
One can ask about distinguished bases of the boundary space. Under the
assumptions of Proposition 1 we have the principal boundary functional,
1
which for 0 ≤ Re(m) < 1 can be defined as W(x 2 +m, ·; 0). There are
also non-principal boundary functionals, which lead to boundary conditions
1 1
roughly of the type x 2 −m for m 6= 0, or x 2 ln(x) for m = 0.
 Let us now impose the assumption
Z 1
x1−ε|Q(x)|dx < ∞.
0

If 2 > ε > 0, then for 0 ≤ Re(m) ≤ ε/2 we have a distinguished

1
non-principal boundary functional given by W(x 2 −m, ·; 0) if m 6= 0 and
1
W(x 2 ln(x), ·; 0) if m = 0. We obtain three families of perturbed Bessel
operators
n ε o
− < Re(m) 3 m 7→ Hm,
2
n ε o

|Re(m)| < × C ∪ {∞} 3 (m, κ) 7→ Hm,κ,
2
C ∪ {∞} 3 ν 7→ H0ν ,

analogous to the families of the unperturbed case. All three families are
holomorphic except for a singularity at (m, κ) = (0, −1).
 They are defined as the restrictions of Lm2 to the domains:
n 1
o
D(Hm) := f ∈ D(Lmax m 2 ) | W(x 2 +m , f ; 0) = 0 ,
n 1 +m 1 −m
o
max


D(Hm,κ) := f ∈ D(Lm2 ) | W x 2 + κx 2 , f ; 0 = 0 , κ ∈ C,
n 1
o
D(Hm,∞) := f ∈ D(Lmax 2 −m , f ; 0 = 0 ,


m2 ) | W x
n 1 1
o
D(H0ν ) := f ∈ D(Lmax


0 ) | W νx + x ln(x), f ; 0 = 0 , ν ∈ C,
2 2

D(H0∞) := D(H0).

Open Problem 13. Under the conditions of Proposition 1, does the

family m 7→ Hm extend meromorphically from {Re(m) > 1} to {Re(m) > −1}?
 Let us now consider a nonnegative integer n. Under the assumptions
0[n]
of Proposition 11 we can use the function u−m. Then every non-principal
boundary functional can be written as
0[n] 1
W(Γ(1 − m)u−m + κx 2 +m , ·; 0)
1 −m 1
for some κ ∈ C. Clearly, it is proportional to W(x 2 + κx 2 +m , ·; 0) for
n = 0. For n ≥ 1 it is less canonical. The set of non-principal boundary
conditions can be viewed as a 1-dimensional affine space, where we can use
0[n]
W(Γ(1 − m)u−m, ·; 0) as a possible “reference point”.
 Physicists often prefer to fix the operator not by a b.c. bear zero, but by
the behavior of the zero-energy eigenfunction near infinity. The behavior of
zero energy eigenfunctions at large distances is responsible for large scale
properties of quantum systems. It is described by a parameter called the
scattering length, which at least in dimension 2 and 3 is popular in physics.
THANK YOU FOR YOUR ATTENTION