On the approximation by regular

arXiv:1908.02936v1 [math-ph] 8 Aug 2019

potentials of Schrödinger operators

with point interactions
Artbazar Galtbayar and Kenji Yajima

Abstract
We prove that the wave operators for Schrödinger operators with
multi-center local point interactions are the scaling limits of the ones
for Schrödinger operators with regular potentials. We simulataneously
present a proof of the corresponding well known result for the resolvent
which substantially simplifies the one by Albeverio et al.
2010 Mathematics Subject Classification 47A10, 81Q10, 81Uxx
Keywords: Point interactions, self-adjoint extensions, scattering theory, wave
operators

1 Introduction
Let Y = {y1 , . . . , yN } be the set of N points in R3 and T0 be the densely
defined non-negative symmetric operator in H = L2 (R3 ) defined by
T0 = −∆|C0∞ (R3 \Y ) .
Any of selfadjoint extensions of T0 is called the Schrödinger operator with
point interactions at Y . Among them, we are concerned with the ones with
local point interactions Hα,Y which are defined by separated boundary con-
ditions at each point yj parameterized by αj ∈ R, j = 1, . . . , N. They can
be defined via the resolvent equation (cf. [2]): With H0 = −∆ being the free
Schrödinger operator and z ∈ C+ = {z ∈ C|ℑz > 0},
N
X
(Hα,Y − z )2 −1 2 −1
= (H0 − z ) + (Γα,Y (z)−1 )jℓ Gzyj ⊗ Gzyℓ , (1)
j,ℓ=1

1
 2

where α = (α1 , . . . , αN ) ∈ RN , Γα,Y (z) is N × N symmetric matrix whose

entries are entire holomorphic functions of z ∈ C given by
 iz  
Γα,Y (z) := αj − δjℓ − Gz (yj − yℓ )δ̂jℓ , (2)
4π j,ℓ=1,...,N

where δjℓ = 1 for j = ℓ and δjℓ = 0 otherwise; δ̂jℓ = 1 − δjℓ ; Gz (x) is the
convolution kernel of (H0 − z 2 )−1 :
eiz|x| eiz|x−y|
Gz (x) = and Gzy (x) = . (3)
4π|x| 4π|x − y|
Since (Hα,Y − z 2 )−1 − (H0 − z 2 )−1 is of rank N by virtue of (1), the wave
±
operators Wα,Y defined by the limits
±
Wα,Y u = lim eitHα,Y e−itH0 u, u∈H (4)
t→±∞
±
exist and are complete in the sense that Image Wα,Y = Hac , the absolutely
continuous (AC for short) subspace of H for Hα,Y . Wave operators are of
fundamental importance in scattering theory.
This paper is concerned with the approximation of the wave operators
±
Wα,Y by the ones for Schrödinger operators with regular potentials and gen-
eralizes a result in [5] for the case N = 1, which immediately implies that
±
Wα,Y are bounded in Lp (R3 ) for 1 < p < 3, see remarks below Theorem 1.1.
We also give a proof of the corresponding well known result for the resolvent
(Hα,Y −z)−1 which substantially simplifies the one in the seminal monograph
[2].
We begin with recalling various properties of Hα,Y (see [2]):
• Equation (1) defines a unique selfadjoint operator Hα,Y in the Hilbert
space H = L2 (R3 ), which is real and local.
• The spectrum of Hα,Y consists of the AC part [0, ∞) and at most N
non-positive eigenvalues. Positive eigenvalues are absent. We define
E = {ik ∈ iR+ : − k 2 ∈ σp (Hα,Y )}. We simply write Hac and Pac
respectively for the AC subspace Hac (Hα,Y ) of H for Hα,Y and for the
projection Pac (Hα,Y ) onto Hac .
• Hα,Y may be approximated by a family of Schrödinger operators with
scaled regular potentials
XN  
λi (ε) x − yi
H Y (ε) = −∆ + 2
Vi , (5)
i=1
ε ε
 3

in the sense that for z ∈ C+

lim(H Y (ε) − z 2 )−1 u = (Hα,Y − z 2 )−1 u, ∀u ∈ H, (6)

ε→0

where Vj , j = 1, . . . , N are such that Hj = −∆ + Vj (x) have threshold

resonances at 0 and λ1 (ε), . . . , λN (ε) are smooth real functions of ε such
that λj (0) = 1 and λ′j (0) 6= 0 (see Theorem 1.1 for more details).

We prove the following theorem (see Section 4 for the definition of the thresh-
old resonance).

Theorem 1.1. Let Y be the set of N points Y = {y1 , . . . , yN }. Suppose that:

(1) V1 , . . . , VN are real-valued functions such that for some p < 3/2 and
q > 3,
hxi2 Vj ∈ (Lp ∩ Lq )(R3 ), j = 1, . . . , N. (7)

(2) λ1 (ε), . . . , λN (ε) are real C 2 functions of ε ≥ 0 such that

λj (0) = 1, λ′j (0) 6= 0, ∀j = 1, . . . , N.

(3) Hj = −∆ + Vj , j = 1, . . . , N admits a threshold resonance at 0.

Then, the following statements are satisfied:

(a) H Y (ε) converges in the strong resolvent sense as in (6) as ε → 0 to

a Schrödinger operator Hα,Y with point interactions at Y with certain
parameters α = (α1 , . . . , αN ) to be specified below .
±
(b) Wave operators WY,ε for the pair (H Y (ε), H0 ) defined by the strong
limits
±
WY,ε u = lim eitH Y (ε) e−itH0 u, u ∈ H (8)
t→±∞
±
exist and are complete. WY,ε satisfy
± ±
lim kWY,ε u − Wα,Y ukH = 0, u ∈ H. (9)
ε→0

Note that Hölder’s inequality implies Vj ∈ Lr (R3 ) for all 1 ≤ r ≤ q under

the condition (7).
 4

±
Remark 1.2. (i) It is known that WY,ε are bounded in Lp (R3 ) for 1 < p < 3
±
([14]) and, if λj (ε) = 1 for all j = 1, . . . , N, kWY,ε kB(Lp ) is independent of
ε > 0 and, the proof of Theorem 1.1 shows that Theorem 1.1 holds with
α = 0. It follows by virtue of (9) that WY,ε converges to Wα=0,Y weakly in Lp
±
and Wα=0,Y are bounded in Lp (R3 ) for 1 < p < 3. Actually, the latter result
is known for general α = (α1 . . . . , αN ) but its proof is long and complicated
([5]). Wave operators satisfy the intertwining property
±∗ ±∗
f (Ha,Y )Hac (Ha,Y ) = Wα,Y f (H0 )Wα,Y .

for Borel functions f on R and, Lp mapping properties of f (Ha,Y )Pac (Ha,Y )

are reduced to those for the Fourier multiplier f (H0 ) for a certain range of
p’s.
(ii) If some of Hj = −∆ + Vj have no threshold resonance then, Theorem
1.1 remains to hold if corresponding points of interactions and parameters
(yj , αj ) are removed from Hα,Y .
(iii) The first statement is long known (see [2]). We shall present here a
simplified proof, providing in particular details of the proof of Lemma 1.2.3
of [2] where [6] is referred to for “a tedious but straightforward calculation”
by using a result from [4] and a simple matrix formula.
±
(iv) The existence and the completeness of wave operators WY,ε are well
known (cf. [12]).
(v) When N = 1 and α = 0, (9) is proved in [5]. The theorem is a general-
ization for general α and N ≥ 2.
(vi) The matrix Γα,Y (k) is non-singular for all k ∈ (0, ∞) by virtue of the
selfadjointness of Hα,Y and H0 . Indeed, if it occurred that det Γα,Y (k0 ) =
0 for some 0 < k0 , then the selfadjointness of Hα,Y and H0 implied that
Γα,Y (k)−1 had a simple pole at k0 and

2k0 Resz=k0 (Γα,Y (z)−1 )jℓ (Gzyj , v)(u, Gzyℓ )

N
X
= lim (z 2 − k02 ) (Γα,Y (z)−1 )jℓ (Gzyj , v)(u, Gzyℓ ) 6= 0 (10)
z=k0 +iε,ε↓0
j,ℓ=1

for some u, v ∈ C0∞ (R3 ). However, the absence of positive eigenvalues of

Hα,Y (see [2], pp. 116-117) and the Lebesgue dominated convergence theorem
 5

imply for all u, v ∈ C0∞ (R3 ) that

lim (z 2 − k02 )((Hα,Y − z 2 )−1 u, v)

z=k0 +iε,ε↓0
Z
2ik0 ε − ε2
= lim (E(dµ)u, v) = (E({k02 })u, v) = 0
z=k0 +iε,ε↓0 R µ − (k0 + iε)2

and the likewise for (z 2 − k02 )((H0 − z 2 )−1 u, v), where E(dµ) is the spectral
projection for Hα,Y , which contradict (10).
For more about point interactions we refer to the monograph [2] or the
introduction of [5] and jump into the proof of Theorem 1.1 immediately.
+
We prove (9) only for WY,ε as H Y (ε) and Hα,Y are real operators and the
−
complex conjugation C changes the direction of the time which implies WY,ε =
+ −1
CWY,ε C .
We write H for L2 (R3 ), (u, v) for the inner product and kuk the norm.
u ⊗ v and |uihv| indiscriminately denote the one dimensional operator
Z
(u ⊗ v)f (x) = |uihv|f i(x) = u(x)v(y)f (y)dy.
R3

Integral operators T and their integral kernels T (x, y) are identified. Thus
we often say that operator T (x, y) satisfies such and such properties and etc.
B2 (H) is the space of Hilbert-Schmidt operators in H and
Z Z 1/2
2
kT kHS = |T (x, y)| dxdy
R3 ×R3

is the norm of B2 (H). hxi = (1 + |x|2 )1/2 and a≤| · | b means |a| ≤ |b|. For
subsets D1 and D2 of the complex plane C, D1 ⋐ D2 means D1 is a compact
subset of the interior of D2 .

2 Scaling
For ε > 0, we let
(Uε f )(x) = ε−3/2 f (x/ε).
This is unitary in H and H0 = ε2 Uε∗ H0 Uε . We define H(ε) by

H(ε) = ε2 Uε∗ H Y (ε)Uε , (H Y (ε) − z 2 )−1 = ε2 Uε (H(ε) − ε2 z 2 )−1 Uε∗ . (11)

 6

Then, H(ε) is written as

N
X  yi 
H(ε) = −∆ + λi (ε)Vi x − ≡ −∆ + V (ε)
i=1
ε

±
and WY,ε are transformed as

± 2 2
WY,ε = lim Uε eitH(ε)/ε e−itH0 /ε Uε∗ = Uε WY± (ε)Uε∗ , (12)
t→±∞
±
WY (ε) = lim Uε eitH(ε) e−itH0 Uε∗ . (13)
t→±∞

We write the translation operator by ε−1 yj by

 yj 
τj,ε f (x) = f x + , j = 1, . . . , N.
ε
When ε = 1, we simply denote τj = τj,1 , j = 1, . . . , N. Then,
 yj  ∗
Vj x − = τj,ε Vj (x)τj,ε .
ε

3 Stationary representation
The following lemma is obvious and well known:

Lemma 3.1. The subspace D∗ = {u ∈ L2 : û ∈ C0∞ (R3 \ {0})} is a dense

linear subspace of L2 (R3 ).
+ +
It is obvious that kWY,ε uk = kWα,Y uk = kuk for every u ∈ H and, for
proving (9) it suffices to show that
+ +
lim(WY,ε u, v) = (Wα,Y u, v), u, v ∈ D∗ . (14)
ε→0

+ +
We express WY,ε and Wα,Y via stationary formulae. We recall from [5] the
+
following representation formula for Wα,Y .

Lemma 3.2. Let u, v ∈ D∗ and let Ωjℓ u be defined for j, ℓ ∈ {1, . . . , N} by

Z Z 
1 ∞ −1


(Γα,Y (−k) )jℓ G−k (x) Gk (y) − G−k (y) u(y)dy kdk. (15)
πi 0 R3
 7

Then,
N
X
+
hWα,Y u, vi = hu, vi + hτj∗ Ωjℓ τℓ u, vi. (16)
j,ℓ=1

Note that for u ∈ D∗ the inner integral in (15) produces a smooth function
of k ∈ R which vanishes outside the compact set {|ξ| : ξ ∈ supp û}.
+
For describing the formula for WY,ε corresponding to (15) and (16), we
(N )
introduce some notation. H = H ⊕ · · · ⊕ H is the N-fold direct sum of H.
(N )
Likewise T = T ⊕ · · · ⊕ T for an operator T on H. For i = 1, . . . , N we
decompose Vi (x) as the product:
Vi (x) = ai (x)bi (x), ai (x) = |Vi (x)|1/2 , bi (x) = |Vi (x)|1/2 sign(Vi (x))
where sign a = ±1 if ±a > 0 and sign a = 0 if a = 0. We use matrix notation
for operators on H(N ) . Thus, we define
     
a1 · · · 0 b1 · · · 0 λ1 (ε) · · · 0
     ..  .
A =  ... . . . ...  , B =  ... . . . ...  , Λ(ε) =  ... ..
. . 
0 · · · aN 0 · · · bN 0 · · · λN (ε)

Since aj , bj and λj (ε), j = 1, . . . , N are real valued, multiplications with A, B

and Λ(ε) are selfadjoint operators on H(N ) . We also define the operator τε
by  
τ1,ε f
 
τε : H ∋ f 7→ τε f =  ...  ∈ H(N )
τN,ε f
so that
N
X  yj 
V (ε) = λj (ε)Vj x− = τε∗ AΛ(ε)Bτε .
j=1
ε
We write for the case ε = 1 simply as τ = τ1 as previously. For z ∈ C, G0 (z)
is the integral operator defined by
Z
1 eiz|x−y|
G0 (z)u(y) = u(y)dy.
4π R3 |x − y|
It is a holomorphic function of z ∈ C+ with values in B(H) and
G0 (z) = (H0 − z 2 )−1 , for z ∈ C+
 8

and, it can be extended to various subsets of C+ when considered as as a

function with values in a space of operators between suitable function spaces.
We also write

Gε (z) = (H(ε) − z 2 )−1 for z ∈ C+ \ {z : z 2 ∈ σ(H(ε))}

+
Lemma 3.3. Let V1 , . . . , VN satisfy the assumption (7) and z ∈ C . Then:

(1) ai , bj ∈ L2 (R3 ), i, j = 1, . . . , N.

(2) ai G0 (z)bj ∈ B2 (H), 1 ≤ i, j ≤ N.

Proof. (1) We have ai , bj ∈ L2 (R3 ) for Vj ∈ L1 (R3 ) as was remarked below

Theorem 1.1.
(2) We also have |aj |2 = |bj |2 = |Vj | ∈ L3/2 (R3 ) and |x|−2 ∈ L3/2,∞ (R3 ).
It follows by the generalized Young inequality that
ZZ
|ai (x)|2 |bj (y)|2
dxdy ≤ CkVi kL3/2 kVj kL3/2 .
R3 ×R3 |x − y|2

Hence, ai G0 (z)bj is of Hilbert-Schmidt type in L2 (R3 ).

Using this notation, we have form (16) that

+



(Wα,Y u, v) = (u, v) + Ωjℓ τ ∗ u, τ ∗ v H(N) . (17)

The resolvent equation for H(ε) may be written as

Gε (z) − G0 (z) = −G0 (z)τε∗ AΛ(ε)Bτε Gε (z)

and the standard argument (see e.g. [13]) yields

Gε (z) = G0 (z) − G0 (z)τε∗ A(1 + Λ(ε)Bτε G0 (z)τε∗ A)−1 Λ(ε)Bτε G0 (z). (18)

Note that τε R0 (z)τε∗ 6= R0 (z) in general unless N = 1.

Under the assumption (7) on V1 , . . . , VN the first two statements of the
following lemma follow from the limiting absorption principle for the free
Schrödinger operator ([1], [11]) and the last from the absence of positive
eigenvalues for H(ε) ([10]). In what follows we often write k for z when we
want emphasize that k can also be real.
 9

Lemma 3.4. Suppose that V1 , . . . VN satisfy the assumption of Theorem 1.1.

Let 0 < ε ≤ 1. Then:
(1) For u ∈ D∗ , limδ↓0 supk∈R kAτε G0 (k + iδ)u − Aτε G0 (k)ukH(N) = 0.
(2) limδ↓0 supk∈R kΛ(ε)Aτε (G0 (k + iδ) − G0 (k))τε∗ AkB(H(N) ) = 0.
+
(3) Define for k ∈ C = {k ∈ ℑk ≥ 0},
Mε (k) = Λ(ε)Bτε G0 (k)τε∗ A. (19)
Then, Mε (k) is a compact operator on H(N ) and 1 + Mε (k) is invertible
for all k 6= 0. (1 + Mε (k))−1 is a locally Hölder continuous function of
+
C \ {0} with values in B(H(N ) ).
Statements (1) and (2) remain to hold when A is replaced by B.
The well known stationary formula for wave operators ([11]) and the
resolvent equation (18) yield
(WY+ (ε)u, v) − (u, v) (20)
Z ∞
1

=− (1 + Mε (−k))−1 Λ(ε)Bτε {G0 (k) − G0 (−k)}u, Aτε G0 (k)v kdk.
πi 0
+
For obtaining the corresponding formula for WY,ε , we scale back (20) by using
the identity (12) and (13). Then
τε Uε∗ = Uε∗ τ,
and change of variable k to εk produce the first statement of the following
lemma. Recall τ = τε=1 . The second formula is proven in parallel with the
first by using (11). The following lemma should need no proof.
Lemma 3.5. (1) For u, v ∈ D ∗ , we have
Z
+ ε2 ∞
(WY,ε u, v) = (u, v) − kdk (21)
πi 0


(1 + Mε (−εk))−1 Λ(ε)B{G0 (kε) − G0 (−kε)}(N ) Uε∗ τ u, AG0 (kε)(N ) Uε∗ τ v .
(2) For k ∈ C+ with sufficiently large ℑk,
(H Y (ε) − k 2 )−1 = G0 (k)
− ε2 τ ∗ Uε G0 (kε)(N ) A(1 + Mε (εk))−1 Λ(ε)BG0(kε)(N ) Uε∗ τ, (22)
where G0 (±kε)(N ) = G0 (±kε) ⊕ · · · ⊕ G0 (±kε) is the N-fold direct sum of
G0 (±kε).
 10

Notice that for u ∈ D∗ , {G0 (kε)−G0 (−kε)}(N ) Uε∗ τ u 6= 0 for R−1 < k < R
for some R > 0 and the integral on the right of (21) is only over [R−1 , R] ⊂
(0, ∞) uniformly for 0 < ε < 1. Indeed, if u ∈ D∗ and û(ξ) = 0 unless
R−1 ≤ |ξ| ≤ R for some R > 1, then, since the translation τ does not change
the support of û(ξ/ε), we have
 
∗ − 23 ξ
F (Uε τ u)(ξ) = ε F (τ u) =0
ε

unless R−1 ε ≤ |ξ| ≤ Rε and

{G0 (kε) − G0 (−kε)}Uε∗ τ u = 2iπδ(ξ 2 − k 2 ε2 )F (Uε∗ τ u)(ξ) = 0.

for k > R or k < R−1 .

4 Limits as ε → 0
We study the small ε > 0 behavior of the right hand sides of (21) and
(22). For (21), the argument above shows that we need only consider the
integral over a compact set K ≡ [R−1 , R] ⊂ R which will be fixed in this
section. Splitting ε2 = ε · ε1/2 · ε1/2 in front of the second term on the right,
we place one ε1/2 each in front of BG0 (±kε)(N ) Uε∗ and AG0 (±kε)(N ) U ∗ or
Uε G0 (kε)(N ) A and the remaining ε in front of (1 + Mε (±εk))−1 . We begin
with the following lemma. Recall the definition (3) of Gk .
Lemma 4.1. Suppose a ∈ L2 (R3 ). Then, following statements are satisfied:
(1) Let u ∈ D∗ . Then, uniformly in k ∈ K, we have
1
lim kε 2 aG0 (±kε)Uε∗ u − |aihG±k , uikL2 = 0. (23)
ε→0

(2) Let u ∈ L2 (R3 ). Then, uniformly on compacts of k ∈ C+ , we have

1
kε 2 aG0 (kε)Uε∗ ukL2 ≤ C(ℑk)−1/2 kakL2 kukL2 (24)

and the convergence (23) with k in place of ±k.

(3) Let u ∈ L2 (R3 ). Then, uniformly on compacts of k ∈ C+ , we have

1
lim kε 2 Uε G0 (kε)au − |Gk iha, uikL2 = 0. (25)
ε→0
 11

Proof. (1) We prove the + case only. The proof for the − case is similar. We
have u ∈ S(R3 ) and
Z Z
1
∗ 1 2 eikε|x−y| 1 eik|y|
ε 2 G0 (kε)Uε u(x) = ε u(εy)dy = u(y + εx)dy.
4π R3 |x − y| 4π R3 |y|
It is then obvious for any R > 0 and a compact K ⊂ R that
1
lim sup |ε 2 G0 (kε)Uε∗ u(x) − hGk , ui| = 0 (26)
ε→0 |x|≤R,k∈K

Moreover, Hölder’s inequality in Lorentz spaces implies that

1
|hGk , ui| + kε 2 G0 (kε)Uε∗ uk∞ ≤ k(4π|x|)−1 k3,∞ kuk 3 ,1 . (27)
2

It follows from (26) that for any R > 0

1
lim sup kε 2 aG0 (kε)Uε∗ u − ahGk , uikL2 (|x|≤R) = 0 (28)
ε→0 k∈K

and, from (27) that

1
kε 2 aG0 (kε)Uε∗ u − ahGk , uikL2 (|x|≥R)
≤ 2kakL2 (|x|≥R) k(4π|x|)−1 k3,∞ kuk 3 ,1 → 0. (29)
2

Combining (26) and (29), we obtain (23) for u ∈ D∗ . (Since D(R3 ) is dense
3
in L3,1 (R3 ), (23) actually holds for u ∈ L 2 ,1 (R3 ).)
(2) We have
Z
|a(x)|2 e−2ℑkε|x−y|
kaG0 (kε)k2HS = dxdy ≤ C(ℑkε)−1 kak2L2 .
R3 ×R3 16|x − y|2
This implies (24) as Uε∗ is unitary in L2 (R3 ) and it suffices to prove the strong
convergence on L2 for u ∈ C0∞ (R3 ). This, however, follows as in the case (1).
(3) We have
Z
1 eik|x−εy|
ε (Uε G0 (kε)au)(x) =
2 a(y)u(y)dy
R3 4π|x − εy|
and Minkowski’s inequality implies
Z
1
kε 2 Uε G0 (kε)au − |Gk iha, uik ≤ kGk (· − εy) − Gk kL2 (R3 ) |a(y)u(y)|dy. (30)
R3
 12

Plancherel’s and Lebesgue’s dominated convergence theorems imply that for

a compact subset K̃ of C+
sup kGk (· + εy) − Gk k = sup k(F −1 Gk )(ξ)(eεyξ − 1)kL2 (R3ξ )
k∈K̃ k∈K̃
Z ! 21 Z  21
= sup |(|ξ|2 − k 2 )−1 (eiεyξ − 1)|2 dξ ≤C hξi−4|(eiεyξ − 1)|2dξ
R3 k∈K̃ R3

is uniformly bounded for y ∈ R3 and converges to 0 as ε → 0. Thus, (25)

follows from (30) by applying Lebesgue’s dominated convergence theorem .

+
We next study ε(1+Mε (εk))−1 for ε → 0 and k ∈ C \{0}. We decompose
Mε (k) = Λ(ε)Bτε G0 (εk)τε∗ A into the diagonal and the off-diagonal parts:
Mε (k) = Dε (εk) + εEε (εk) (31)
where the diagonal part is given by
 
λ1 (ε)b1 G0 (εk)a1 · · · 0
 .. .. .. 
Dε (εk) =  . . .  (32)
0 · · · λN (ε)bN G0 (εk)aN
 
∗
and, the off diagonal part εEε (εk) = λi (ε)bi τi,ε G0 (εk)τj,ε aj δ̂ij by
 
bi (x)eik|ε(x−y)+yi −yj | aj (y)
εEε (εk) = ε λi (ε) δ̂ij . (33)
4π|ε(x − y) + yi − yj | ij

We study Eε (εk) first. Define constant matrix Ĝ(k) by

1 eik|yi−yj |
Ĝij (k) = Gij (k)δ̂ij , Gij (k) = , i 6= j.
4π |yi − yj |
+
Lemma 4.2. Assume (7) and let Ω ⊂ C be compact. We have uniformly
for k ∈ Ω that
lim kEε (±εk) − |BiĜ(±k)hA|kB(H(N) ) = 0. (34)
ε→0

|BiĜ(±k)hA| is an operator of rank at most N on H(N ) :

 
|BiĜ(±k)hA| ≡ bi (x)Gij (±k)aj (y)δ̂ij .
 13

Proof. We prove the + case only. The − case may be proved similarly. Let
k ∈ K. Then,
 
 eik|ε(x−y)+yi −yj | eik|yi −yj | 

 |ε(x − y) + yi − yj | − |yi − yj | 
|k||ε(x − y)| |ε(x − y)|
≤ + (35)
|ε(x − y) + yi − yj | |ε(x − y) + yi − yj ||yi − yj |
C|x − y|
≤ (36)
|(x − y) + (yi − yj )/ε|
for a constant C > 0 and we may estimate as
Z
|b (x)|x − y|a (y)u(y)|
k(Eε,ij (εk) − λi (ε)bi Gij (k)aj )ukL2 ≤ C
i j
3 |(x − y) + (yi − yj )/ε| dy
R
Z
|hxib (x)hyia (y)u(y)|
≤C
i j
3 |(x − y) + (yi − yj )/ε| dy
R
Z
|τi,ε (hxib i )(x)τ j,ε (hyiaj u)(y)|
=C 3 dy .

R |x − y|
6
Since the convolution with the Newton potential |x|−1 maps L 5 (R3 ) to L6 (R3 )
by virtue of Hardy-Littlewood-Sobolev’s inequality, Hölder’s inequality im-
plies that the right hand side is bounded by

Ckhxibi kL3 khyiaj ukL6/5

1 1
≤ Ckhxibi kL3 khxiaj kL3 kukL2 = Ckhxi2 Vi k 2 3 khxi2 Vj k 2 3 kukL2 . (37)
L2 L2

Let BR (0) = {x : |x| ≤ R} for an R > 0. Then, for ε > 0 such that
4Rε < min |yi − yj |, we have
(35) ≤ 4Cε, ∀x, y ∈ BR (0).
Thus, if Vj ∈ C0∞ (R3 ), j = 1, . . . , N are supported by BR (0), then
N
X ε→0
kEε (εk) − Λ(ε)B Ĝ(k)AkB(H(N) ) ≤ 4Cε kVj kL1 −−→ 0.
j=1

Since C0∞ (R3 ) is a dense subspace of the Banach space (hxi−2 L3/2 (R3 )) ∩
L1 (R3 ), (37) implies kEε (εk) − Λ(ε)B Ĝ(k)AkB(H(N) ) → 0 as ε → 0 for gen-
eral Vj ’s which satisfies the assumption (7). The lemma follows because Λ(ε)
 14

converges to the identify matrix.

We have shown in Lemma 3.3 that bi G0 (kε)aj is of Hilbert-Schmidt type

+
for k ∈ C and it is well known that 1 + λj (ε)bj G0 (kε)aj is an isomorphism
of H unless k 2 ε2 is an eigenvalue of Hj (ε) = −∆ + λj (ε)Vj (see [7]). Hence,
the absence of positive eigenvalues for Hj (ε) (see e.g. [10]) implies that
+
1 + λj (ε)bj G0 (kε)aj is an isomorphism in H for all k ∈ C \ (ε−1iEj (ε) ∪ {0})
where Ej (ε) = {k > 0 : − k 2 ∈ σp (Hj (ε))}. Thus, if we fix a compact set
+
Ω ⊂ C \ {0}. 1 + Dε (εk) is invertible in B(H(N ) ) for small ε > 0 and k ∈ Ω
and
1 + Mε (εk) = (1 + Dε (εk))(1 + ε(1 + Dε (εk))−1 Eε (εk)).
It follows that

(1 + Mε (εk))−1 = (1 + ε(1 + Dε (εk))−1 Eε (εk))−1 (1 + Dε (εk))−1 . (38)

and we need study the right hand side of (38) as ε → 0.

We begin by studying ε(1 + Dε (εk))−1 and, since 1 + Dε (εk) is diagonal,
we may do it component-wise. We first study the case N = 1.

4.1 Threshold analysis for the case N = 1

When N = 1, we have Mε (εk) = Dε (εk).
+
Lemma 4.3. Let N = 1, a = a1 and etc and, let Ω be compact in C \ {0}.
Then, for any 0 < ρ < ρ0 , ρ0 = (3 − p)/2p > 1/2, we have following
expansions in Ω in the space of Hilbert-Schmidt operators B2 (H):

bG0 (kε)a = bD0 a + ikεbD1 a + O((kε)1+ρ ), (39)

Mε (εk) = bD0 a + ε λ′ (0)bD0 a + ikbD1 a + O(ε1+ρ) (40)
1 1
D0 = , D1 = . (41)
4π|x − y| 4π

where O((kε)1+ρ) and O(ε1+ρ ) are B2 (H)-valued functions of (k, ε) such that

kO((kε)1+ρ)kHS ≤ C|kε|1+ρ , kO(ε1+ρ)kHS ≤ C|ε|1+ρ, 0 < ε < 1, k ∈ Ω.

 15

Proof. Since ℑk ≥ 0 for k ∈ Ω, Taylor’s formula and the interpolation imply

that for any 0 ≤ ρ ≤ 1 there exists a constant Cρ > 0 such that
|eikε|x−y| − (1 + ikε|x − y|) | ≤ Cρ |εk|1+ρ |x − y|1+ρ.
Hence
 
 
Dε (εk)(x, y) − b(x)a(y) − ikε b(x)a(y)  ≤ Cρ |k|1+ρ ε1+ρ |x − y|ρ|b(x)a(y)|.
 4π|x − y| 4π 
We have shown in Lemma 3.3 that Dε (εk) and bD0 a are Hilbert-Schmidt
operators and bD1 a is evidently so as a, b ∈ L2 (R3 ) (see the remark below
Theorem 1.1). As hxib(x), hyia(y) ∈ L2p (R3 ), we have hxiρ a(x), hxiρ a(y) ∈
L2 (R3 ) for ρ < ρ0 , and
ZZ
|x − y|2ρ|b(x)a(y)|2 dxdy ≤ Ckhxiρ b(x)k2L2 khyiρ a(y)k2L2
R3 ×R3

This prove estimate (39). (40) follows from (39) and Taylor’s expansion of
λ(ε). This completes the proof of the lemma.

We define
Q0 = 1 + bD0 a, Q1 = λ′ (0)bD0 a + ikbD1 a, bD1 a = (4π)−1 |biha|. (42)

The regular case

Definition 4.4. H = −∆ + V (x) is said to be of regular type at 0 if Q0 is
invertible in H. It is of exceptional type if otherwise.
Lemma 4.5. Suppose N = 1 and that H = −∆ + V (x) is of regular type at
+
0. Let Ω be a compact subset of C . Then,
lim sup kε(1 + Mε (εk))−1 kB(H) = 0. (43)
ε→0 k∈Ω

Proof. Since Q0 = 1+bD0 a is invertible, (40) implies the same for 1+Mε (εk)
for k ∈ Ω and small ε > 0 and,
lim sup k(1 + Mε (εk))−1 − Q−1
0 kB(H) = 0.
ε→0 k∈Ω

(43) follows evidently.

 16

An application of Lemma 3.4, Lemma 4.1 and Lemma 4.5 to (21) and
(22) immediately produces following proposition for the case N = 1.

Proposition 4.6. Suppose H = −∆ + V is of regular type at 0. Then:

+
(1) As ε → 0, WY,ε = 1 converges strongly to the identity operator.
+
(2) Let Ω0 ⊂ C be compact. Then, a(H Y (ε) − k 2 )−1 b − aG0 (k)b → 0 in
the norm of B(H) as ε → 0 uniformly with respect to k ∈ Ω0

(3) Let Ω1 ⋐ C+ . Then, lim sup k(H Y (ε) − k 2 )−1 − G0 (k)kB(H) = 0.

ε→0 k∈Ω1

Exceptional case Suppose next that Q0 is not invertible and define

M = : Ker Q0 , N = Ker Q∗0 , Q∗0 = 1 + aD0 b.

By virtue of the Riesz-Schauader theorem dim M = dim N are finite and M

and N are dual spaces of each other with respect to the inner product of H.
Let S be the Riesz projection onto M.

Lemma 4.7. (1) aD0 a is an isomorphism from M onto N and bD0 b from
N onto M. They are inverses of each other.

(2) (aϕ, D0 aϕ) is an inner product on M and (bψ, D0 bψ) on N .

(3) For an orthonormal basis {ϕ1 , . . . , ϕn } of M with respect to the inner

product (aϕ, D0 aϕ), define ψj = aD0 aϕj , j = 1, . . . , n. Then:

(a) {ψ1 , . . . , ψn } is an orthonormal basis of N with respect to (bψ, D0 bψ).

(b) {ϕ1 , . . . , ϕn } and {ψ1 , . . . , ψn } are dual basis of M and N respec-
tively.
(c) Sf = hf, ψ1 iϕ1 + · · · + hf, ψn iϕn , f ∈ H.

Proof. (1) Let ϕ ∈ M. Then, ϕ = −bD0 aϕ and aD0 aϕ = −aD0 b · aD0 aϕ.
Hence aD0 aϕ ∈ N . Likewise bD0 b maps N into M. We have

bD0 b · aD0 aϕ = (bD0 a)2 ϕ = ϕ, ϕ ∈ M,

aD0 a · bD0 bψ = (aD0 b)2 ψ = ψ, ψ ∈ N

and aD0 a and bD0 b are inverses of each other.

 17

(2) Let ϕ ∈ M. Then aϕ ∈ L1 ∩ Lσ for some σ > 3/2 (see the proof of
Lemma 4.9 below) and ac ϕ ∈ L∞ ∩ Lρ for some ρ < 3 by Hausdorff-Young’s
inequality. It follows that
Z
aϕ(ξ)|2
|c
(aϕ, D0 aϕ) = dξ ≥ 0
R3 |ξ|2

and (aϕ, D0 aϕ) = 0 implies aϕ = 0 hence, ϕ = −bD0 aϕ = 0. Thus,

(aϕ, D0 aϕ) is an inner product of M. The proof for (bψ, D0 bψ) is simi-
lar.
(3) We have for any j, k = 1, . . . , n that

(bψj , D0 bψk ) = (baD0 aϕj , D0 baD0 aϕk ) = (−aϕj , −D0 aϕk ) = δjk

and {ψ1 , . . . , ψn } is orthonormal with respect to the inner product (bψ, D0 bψ).
Since n = dim N , it is a basis of N .

(ϕj , ψk ) = (ϕj , aD0 aϕk ) = (aϕj , D0 aϕk ) = δjk , j, k = 1, . . . , n.

Hence {ϕj } and {ψk } are dual basis of each other. Because of this, (c) is a
well known fact for Riesz projections to eigen-spaces of compact operators
([9]). This completes the proof of the lemma.

The following lemma should be known for a long time. We give a proof
for readers’ convenience.
Lemma 4.8. Let 1 < γ ≤ 2 and σ < 3/2 < ρ. Then, the integral operator
Z
hyi−γ u(y)
(Qγ u)(x) = dy (44)
R3 |x − y|

is bounded from (Lσ ∩ Lρ )(R3 ) to the space C∗ (R3 ) of bounded continuous

functions on R3 which converge to 0 as |x| → 0:

kQγ ukL∞ ≤ Ckuk(Lσ ∩Lρ )(R3 ) . (45)

For R ≥ 1, there exists a constant C independent of u such that for |x| ≥ R

  Z
 
(Qγ u)(x) − C(u)  ≤ C kukLσγ∩Lρ , C(u) = hyi−γ u(y)dy. (46)
 |x|  hxi 3
R
 18

Proof. We omit the index γ in the proof. Since |x|−1 ∈ L3,∞ (R3 ), it is obvious
that Qu(x) is a bounded continuous function and that (45) is satisfied. Thus,
it suffices to prove (46) for |x| ≥ 100. Let Kx be the unit cube with center
x. Combining the two integrals on the left hand side of (46), we write it as
Z Z 
C(u) 1 (2yx − y 2)hyi−γ u(y)
(Qγ u)(x)− = + dy ≡ I0 (x)+I1 (x).
|x| |x| Kx R3 \Kx |x − y|(|x − y| + |x|)

When |x − y| ≤ 1 and |x| ≥ 100, |x|, hxi, |y| and |x − y| are comparable in
the sense that 0 < C1 ≤ |x|/hxi ≤ C2 < ∞ and etc. and we may estimate
the integral over Kx as follows:
Z
C |u(y)| C
|I0 (x)| ≤ dy ≤ kukLρ (Kx ) . (47)
|x|hxi γ−1
Kx |x − y| hxiγ

We estimate the integral I1 (x) by splitting it as I1 (x) = I10 (x) + I11 (x):
Z
−1 y 2 hyi−γ u(y)
I10 (x) = dy,
|x| R3 \Kx |x − y|(|x − y| + |x|)
Z
1 2yxhyi−γ u(y)
I11 (x) = dy.
|x| R3 \Kx |x − y|(|x − y| + |x|)

Since |x − y| + |x| ≥ Chxiγ−1 hyi2−γ for |x| ≥ 100, Hölder’s inequality implies
Z
C |u(y)| C
|I10 (x)| ≤ dy ≤ kukLρ (R3 ) . (48)
|x|hxi γ−1
R3 \Kx |x − y| hxiγ

Let σ ′ be the dual exponent of σ. Then, σ ′ > 3 and via Hölder’s inequality
 !σ′ 1/σ′
Z 1−γ
hyi
|I11 (x)| ≤ C  dy  kukLσ (R3 ) . (49)
R3 hx − yi(hxi + hyi)

If |x| < 100|y| then hyiγ−1 (hxi + hyi) ≥ Chxiγ and

 !σ ′ 1/σ′
Z 1−γ
 hyi C −1
dy  ≤ γ khxi kLσ ′ (50)
|x|<100|y| hx − yi(hxi + hyi) hxi
 19

When |x| > 100|y|, we may estimate for 1 < γ ≤ 2 as

hyi1−γ C
≤ .
hx − yi(|x| + |y|) hx − yihxiγ

It follows that
 !σ ′ 1/σ′
Z 1−γ
 hyi C −1
dy  ≤ γ khxi kLσ ′ . (51)
|x|>100|y| hx − yi(hxi + hyi) hxi

Estimates (50) and (51) imply

C
|I11 (x)| ≤ kukLσ (52)
hxiγ

Combining (52) with (48), we obtain (46).

Lemma 4.9. (1) The following is a continuous functional on N :

Z
1 1
N ∋ ϕ 7→ L(ϕ) = a(x)ϕ(x)dx = ha, ϕi ∈ C.
4π R3 4π

(2) For ϕ ∈ N , let u = D0 (aϕ). Then,

(a) u is a sum u = u1 + u2 of u1 ∈ C ∞ (R3 ) ∩ L∞ (R3 ) and u2 ∈

3 3
(W 2 +ε,2 ∩ W 2, 2 +ε )(R3 ) for some ε > 0. It satisfies

(−∆ + V )u(x) = 0. (53)

(b) u is bounded continuous and satisfies

 
L(ϕ) 1
u(x) = +O , |x| → ∞. (54)
|x| |x|2

(c) u is an eigenfunction of H with eigenvalue 0 if and only if L(ϕ) =

0 and it is a threshold resonance of H otherwise.

(3) The space of zero eigenfunctions in N has codimension at most one.

 20

Proof. (1) Since a ∈ L2 , |L(ϕ)| ≤ (4π)−1 kakL2 kϕkL2 .

(2a) Assumption (7) implies a(x) = hxi−1 ã(x) with ã ∈ (L2p ∩ L2q )(R3 ) and
6
1 ≤ 2p < 3 and 2q > 6. It follows by Hölder’s inequality that ãϕ ∈ L 5 −ε ∩
3
L 2 +ε for an ε > 0. Using the the Fourier multiplier χ(D) by χ ∈ C0∞ (R3 )
such that χ(ξ) = 1 for |ξ| ≤ 1,
Z
1
χ(D)u = 3 eixξ χ(ξ)û(ξ)dξ,
(2π) 2 R3

we decompose u:

u = u1 +u2 , u1 = χ(D)D0 (aϕ), u2 = {(1−χ(D))(1−∆)D0}(1−∆)−1 (aϕ).

Since aϕ ∈ L1 (R3 ) it is obvious that

Z
1 u(ξ)
u1 (x) = 3/2
eixξ χ(ξ) 2 dξ ∈ C ∞ (R3 ), lim ∂ α u1 (x) = 0
(2π) R3 |ξ| |x|→∞

for all α. Since (1−χ(ξ))(1+|ξ|2)|ξ|−2 is a symbol of Hörmander class S0 , the

multiplier (1 − χ(D))(1 − ∆)D0 is bounded in any Sobolev space W k,p(R3 )
for 1 < p < ∞ by Mikhlin’s theorem and,
3 3
(1 − ∆)−1 (aϕ) ∈ W 2, 2 +ε (R3 ) ∩ W 2 +ε,2 (R3 )

for an ε > 0 by the Sobolev embedding theorem. It follows that u2 ∈

3 3
W 2, 2 +ε (R3 )∩W 2 +ε,2 (R3 ), in particular, u is bounded and Hölder continuous.
If (1 + bD0 a)ϕ = 0, then

a(1 + bD0 a)ϕ = (1 + V D0 )aϕ = (−∆ + V )D0 aϕ = 0.

and (−∆ + V )u(x) = 0.

(2b) We just proved that u is bounded and Hölder continuous. We use the
notation in the proof of Lemma 4.8. We have aϕ = −V D0 (aϕ) and
Z Z 
1 hyi−1 ã(y)ϕ(y)dy
D0 (aϕ)(x) = + = I1 (x) + I2 (x).
4π Kx R3 \Kx |x − y|

Since hyi is comparable with hxi when |x − y| < 1,

|I1 (x)| ≤ Chxi−1 kãϕkL 23 +ε k|x|−1 kLτ (Kx ) , τ= 3+2ε

1+2ε
<3
 21

6
For estimating the integral over R3 \ Kx , we use that ãϕ ∈ L 5 −ε for some
0 < ε < 1/5. Let δ = (6 − 5ε)/(1 − 5ε). Then, δ > 6 and Hölder’s inequality
implies
Z ! δ1
dy CkãϕkL 65 −ε
|I2 (x)| ≤ CkãϕkL 56 −ε ≤ .
R3 hx − yiδ hyiδ hxi

Hence, aϕ = −V D0 (aϕ) ∈ hxi−3 (Lp ∩ Lq )(R3 ) and Lemma 4.8 with γ = 2

implies statement (2b).
Statements (2a) and (2b) obviously implies (2c). (3) follows from (1) and
(2c).

We distinguish following three cases:

Case (a): N ∩ Ker (L) = {0}. Then, Lemma 4.9 implies dim N = 1, H has
no zero eigenvalue and has only threshold resonances {u = D0 (aϕ) : ϕ ∈ N }.
Case (b): N = Ker (L). Then, {u = D0 (aϕ) : ϕ ∈ N } consists only of
eigenfunctions of H with eigenvalue 0.
Case (c): {0} $ N ∩ Ker (L) $ N . In this case H has both zero eigenvalue
and threshold resonances.
In case (c), we take an orthonormal basis {ϕ1 , ϕ2 , . . . , ϕn } of N such that
ϕ2 , . . . , ϕn ∈ Ker (L) and ϕ1 ∈ Ker (L)⊥ such that L(ϕ1 ) > 0 which uniquely
determines ϕ1 .
We study ε(1+Mε(εk))−1 , Mε (εk) = λ0 (ε)bG0 (εk)a as ε → 0 by applying
the following Lemma 4.10 due to Jensen and Nenciu ([8]). We consider the
case (c) only. The modification for the case (a) and (b) should be obvious.
Lemma 4.10. Let A be a closed operator in a Hilbert space H and S a
projection. Suppose A + S has a bounded inverse. Then, A has a bounded
inverse if and only if
B = S − S(A + S)−1 S
has a bounded inverse in SH and, in this case,

A−1 = (A + S)−1 + (A + S)−1 SB−1 S(A + S)−1 . (55)

We recall (40) and (42). We apply Lemma 4.10 to

A = 1 + Mε (εk) ≡ 1 + λ(ε)bG0 (εk)a. (56)

 22

We take as S the Riesz projection onto the kernel M of Q0 = 1 + bD0 a.

Since bD0 a is compact, Q0 + S is invertible. Hence, by virtue of (40), A + S
is also invertible for small ε > 0 and the Neumann expansion formula yields,
(A + S)−1 = (Q0 + εQ1 + O(ε2) + S)−1
 −1
−1 2
= 1 + ε(Q0 + S) Q1 + O(ε ) (Q0 + S)−1
= (Q0 + S)−1 − ε(Q0 + S)−1 Q1 (Q0 + S)−1 + O(ε2). (57)
Since S(Q0 + S)−1 = (Q0 + S)−1 S = S, the operator B of Lemma 4.10
corresponding to A of (56) becomes
B = εSQ1 S + O(ε2), sup kO(ε2)kB(H) ≤ Cε2 , (58)
k∈Ω
+
where Ω ⋐ C \ {0}. Take the dual basis ({ϕj }, {ψj }) of (M, N ) defined in
Lemma 4.7. Then, bD0 aϕ = −ϕ for ϕ ∈ M, (a, ϕj ) = 0 for 2 ≤ j ≤ n and
(ψj , b) = (aD0 aϕj , b) = −(ϕj , a) imply
ik
SQ1 S = S(λ′ (0)bD0 a + ikbD1 a)S = −λ′ (0)S − |(a, ϕ1 )|2 (ϕ1 ⊗ ψ1 ).
4π
It follows from (58) that uniformly with respect to k ∈ Ω we have
 
k|(a, ϕ )| 2 −1 Xn
−1 1
εB + λ′ (0) + i ϕ1 ⊗ ψ1 + λ′ (0)−1 ϕj ⊗ ψj ≤ Cε
4π
j=2
(59)
Then, since k(A + S)−1 kB(H) is bounded as ε → 0 and k ∈ Ω and
lim sup(kS(A + S)−1 − SkB(H) + k(A + S)−1 S − SkB(H) = 0,
ε→0 k∈Ω

(55), (57) and (59) imply the first statement of the following proposition.
Proposition 4.11. Let N = 1 and the assumption (7) be satisfied. Suppose
that H is of exceptional type at 0 of the case (c). Then, with the notation of
Lemma 4.7, uniformly with respect to k ∈ Ω in the operator norm of H we
have that

lim ε(1 + Dε (εk))−1

ε→0
 −1 Xn
′ k|(a, ϕ1 )|2
= − λ (0) + i ϕ1 ⊗ ψ1 − λ′ (0)−1 ϕj ⊗ ψj ≡ L (60)
4π j=2
 23

and that  −1

  ik λ′ (0)

a| (60) b = − α − , α=− . (61)
4π |(a, ϕ1 )|2
The same result holds for other cases with the following changes: For the
case (a) replace ϕ1 and ψ1 by ϕ and ψ respectively which are normalized as
ϕ1 and ψ1 and, for the case (b) set ϕ1 = ψ1 = 0.

4.2 Proof of Theorem 1.1

Let Lj , j = 1, . . . , N be the L of (60) corresponding to Hj (ε) = −∆+λj (ε)Vj .
Then, applying Proposition 4.11 to Hj (ε), we have

lim ε(1 + Dε (εk))−1 = ⊕N

j=1 Lj ≡ L̃ (62)
ε→0

It follows by combining Lemma 4.2 and (62) that

−1
lim 1 + ε(1 + Dε (εk)) Eε (εk) = 1 + L̃|BiĜ(k)hA| (63)
ε→0

We apply the following lemma due to Deift ([4]) to the right of (63).
Lemma 4.12. Suppose that 1 + hA|L̃|BiĜ(k) is invertible in B(CN ). Then,
1 + L̃|BiĜ(k)hA| is also invertible in B(H(N ) ) and

hA|(1 + L̃|BiĜ(k)hA|)−1 = (1 + hA|L̃|BiĜ(k))−1 hA| (64)

Proof. Since a1 , . . . , aN ∈ L2 (R3 ), |Ai : CN → H(N ) and hA| : H(N ) → CN

are both bounded operators. Then, the lemma is an immediate consequence
of Theorem 2 of [4].

For the next lemma we use the following simple lemma for matrices. Let
   
W X 0 0
A= , B=
Y Z 0 V

be matrices decomposed into blocks.

Lemma 4.13. Suppose V and 1 + V Z are invertible. Then,
   −1
0 0 W X
1+
0 V Y Z
 24

exists and
   −1    
0 0 W X 0 0 0 0
1+ = (65)
0 V Y Z 0 V 0 (V −1 + Z)−1
Proof. It is elementary to see
   −1
0 0 W X
1+
0 V Y Z
 −1  
1 0 1 0
= = (66)
VY 1+VZ −(1 + V Z)−1 V Y (1 + V Z)−1
and the left side of (65) is equal to
   
0 0 0 0
=
0 (1 + V Z)−1 V 0 (V −1 + Z)−1
which proves the lemma.

Lemma 4.14. Let k ∈ Ω. Then, 1 + hA|L̃|BiĜ(k) is invertibe in CN . If

H1 , . . . , HN are arranged in such a way that H1 , . . . , Hn1 have no resonances
and Hn1 +1 , . . . , HN do and, N = n1 + n2 , then
 
−1 On1 n1 On1 n2
(1 + hA|L̃|BiĜ(k)) hA|L̃|Bi = , (67)
On2 n1 −Γ̃(k)−1
where On1 n1 is the zero matrix of size n1 × n1 and etc. and
 ik  
Γ̃(k) = αj − δj,ℓ − Gk (yj − yℓ )δ̂jℓ . (68)
4π j,ℓ=n1 +1,...,N

Proof. We let ϕj1 be the resonance of Hj , j = n1 + 1, . . . , N, corresponding

to ϕ1 of the previous section and define
λ′ (0)
αj = − . (69)
|(aj , ϕj1 )|2
Then, Proposition 4.11 implies that,
 
0
 .. 
 . 
 
 0 
hA|L̃|Bi = 


ik −1
.

 − αn2 +1 − 4π 
 .. 
 . 


ik −1
− αn1 +n2 − 4π
 25

and we obtain (67) by applying Lemma 4.13 to the left of (67) with


ik −1

− αn2 +1 − 4π
 .. 
V = . .


ik −1
− αn1 +n2 − 4π
and with  
W X
= Ĝ(k).
Y Z

Lemma 4.12 and Lemma 4.14 imply that the following limit exists in
B(H) and

−1
−1
lim 1 + ε(1 + Dε (εk))−1 Eε (εk) = 1 + L̃|BiĜ(k)hA|
ε→0

and hence so does

−1
−1
lim ε 1 + Mε (εk) = 1 + L̃|BiĜ(k)hA| L̃ (70)
ε→0

Completion of the proof of Theorem 1.1 By the assumption of the

theorem, we may assume n1 = 0 in Lemma 4.14. Abusing notation, we write
Z
(N ) (N ) 1 eik|x|u(x)
Ĝk u = (Ĝk u) , Ĝk u = dy.
4π R3 |x|
We first prove (9) for the + case. We let u, v ∈ D∗ and R > 0. Then, (23)
and (70) imply that

ε2 ((1 + Mε (−εk))−1 Λ(ε)B(G0 (kε) − G0 (−kε))(N ) Uε u, AG0 (kε)(N ) Uε v) (71)

converges as ε → 0 to
(N ) (N ) (N )
(hA|(1 + L̃|BiĜ(−k)hA|)−1 L̃|Bih(Gk − G−k )u, Gk v) (72)

uniformly with respect to k ∈ [R−1 , R]. Here we have

hA|(1 + L̃|BiĜ(−k)hA|)−1 L̃|Bi

= (1 + hA|L|BiĜ(−k))−1 hA|L|Bi = −Γ̃(−k)−1 (73)
 26

by virtue of (64) and (67). Thus, (71) converges as ε → 0 to

(N ) (N )
−(Γα,Y (−k)−1 Ĝk − Ĝ−k u, Ĝk v)

uniformly on [R−1 , R]. Thus, replacing u and v respectively by τ u and τ v,

+ +
we obtain WY,ε → Wα,Y strongly as ε → 0 in view of (15) and (21).
By virtue of (1) and (22), for proving the convergence (6) of the resolvent,
it suffices to show that as ε → 0 in the strong topology of B(H)

ε2 Uε G0 (kε)(N ) A(1 + Mε (εk))−1 Λ(ε)εBG0(kε)(N ) Uε

(N ) (N )
→ −|Ĝk iΓα,Y (k)−1 hĜk | (74)

for every k ∈ C+ \ E. However, (23), (25) and (70) imply that for k ∈ C+ \ E
the first line of (74) converges strongly in B(H) as ε → 0 to
(N ) (N )
|Gk ihA|(1 + L̃|BiĜ(k)hA|)−1 L̃|BihGk |. (75)

This is equal to the second line by virtue of (73) with k in place of −k. This
completes the proof of the theorem.

References
[1] S. Agmon, Spectral properties of Schrödinger operators and scattering
theory, Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 2 (1975), 151–218.

[2] S. Albeverio, F. Gesztesy, R. Hoegh-Krohn, H. Holden, Solvable Models

in Quantum Mechanics, AMS Chelsea Publishing, (2004),

[3] H. Cornean, A. Michelangeli and K. Yajima, Two dimensional

Schrödinger operators with point interactions: Threshold expansions,
zero modes and Lp -boundedness of wave operators. Reviews in Math-
ematical Physics Vol. 31, No. 04, 1950012 (2019)

[4] P. A. Deift, Applications of a commutation formula, Duke Math. J. bf

45 (1978), 267–309.

[5] G. Dell’Antonio, A. Michelangeli, R. Scandone and K. Yajima, The Lp -

boundedness of wave operators for the three dimensional multi-centre
point interaction, Ann. Inst. H. Poincaré 19 (2018), 283–322.
 27

[6] H. Holden, Konvergens mot punkt-interaksjoner (In Norwegian). Cand.

real.thesis, University of Oslo, Norway, 1981. KK.

[7] A. D. Ionescu and W. Schlag, Agmon-Kato-Kuroda theorems for a large

class of perturbations, Duke Math. J. 131 No. 3 (2006), 397440.

[8] A. Jensen and G. Nenciu, A unified approach to resolvent expansions at

thresholds, Reviews in Mathematical Physics, 13, No. 6 (2001) 717-754.

[9] T.Kato, Perturbation of Linear Operators, Springer Verlag. Heidelberg-

New-York-Tokyo, 1966.

[10] H. Koch and D. Tataru. Carleman estimates and absence of embedded

eigenvalues. Comm. Math. Phys., 267 (2006), 419-449.

[11] S. T. Kuroda, An introduction to scattering theory, vol. 51. Lecture

Notes Series, Aarhus Universitet, Matematisk Institut, Aarhus, 1978.

[12] S. T. Kuroda, On the existence and the unitary property of the scattering
operator, Nuovo Cimento, 12 (1959).

[13] M. Reed and B. Simon, Method of modern mathematical physics IV,

Scattering theory, New York, Academic Press, 1979.

[14] K. Yajima, L1 and L∞ -boundedness of wave operators for three di-

mensional Schrödinger operators with Threshold Singularities, Tokyo J.
Math., 41 (2018), 385-406.

Artbazar Galtbayar
Center of Mathematics for Applications and
Department of Applied Mathematics
National University of Mongolia
University Street 3,
Ulaanbaatar, (Mongolia)

Kenji Yajima
Department of Mathematics
Gakushuin University
1-5-1 Mejiro
Toshima-ku Tokyo 171-8588 (Japan).