Spectral flow and Levinson’s theorem for Schrödinger operators

Angus Alexander, Adam Rennie∗

School of Mathematics and Applied Statistics, University of Wollongong,
arXiv:2405.19571v1 [math-ph] 29 May 2024

Wollongong, Australia

May 31, 2024

Abstract
We use spectral flow to present a new proof of Levinson’s theorem for Schrödinger
operators on Rn with smooth compactly supported potential. Our proof is valid in all
dimensions and in the presence of resonances. The statement is expressed in terms of the
spectral shift function and the “high energy corrected time delay” following Guillopé and
others.

1 Introduction

Much work has been done in recent years investigating the topological nature of Levinson’s
theorem from quantum scattering theory, both as an index theorem [3, 5, 24, 25, 33, 34, 35]
and as an index pairing [1, 2, 4]. In this paper we prove Levinson’s theorem for Hamiltonians
H0 , H on Rn by using spectral flow from H0 to H. By applying the operator pseudodifferential
calculus to the spectral flow formula of [15], we obtain a proof of the integral form of Levin-
son’s theorem in all dimensions and in the presence of zero energy resonances. The dominant
contribution is from the eta invariants of the endpoints H0 , H, and can be computed using
the Birman-Kreı̆n formula. In particular, we give a new approach to the relationship between
the spectral shift function and spectral flow, extending work of Azamov, Carey, Dodds and
Sukochev [8, 9]. Our main result (see Theorem 4.2) is
Theorem (Levinson’s theorem). Suppose that V ∈ Cc∞ (Rn ). Then the number N of eigenval-
ues (counted with multiplicity) of H = H0 + V is given by
Z ∞
1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ − βn (V ) + Nres


−N =
2πi 0
where Nres is the contribution from resonances as defined in Theorem 2.15 and the polynomial
pn and constant βn are defined in Lemma 2.10 and Definition 2.11.

The layout of the paper is as follows. In Section 2.1 we recall the definition of spectral flow due
to Phillips [30, 31] and the general formula for the spectral flow along a path of unbounded
operators from [15]. In Section 2.2 we summarise the stationary scattering theory for the
Hamiltonians H0 , H and in Section 2.3 we recall the spectral shift functon and its defining
properties, including the Birman-Kreı̆n trace formula. In Section 2.4 we describe the high-
energy behaviour of the spectral shift function from [1] and the pseudodifferential expansion
of the resolvent from [14].
∗
email: [email protected], [email protected]

1
In Section 3 we use the scattering techniques of Section 2 to analyse the spectral flow formula in
two components. The first is an ‘integral of one-form’ type term in Section 3.1 and the second
is a Birman-Kreı̆n contribution in Section 3.2. Finally, in Section 4 we obtain a formula for
the spectral flow in terms of scattering data and as a consequence prove Levinson’s theorem.
Acknowledgements This work is supported by the ARC Discovery grant DP220101196 and
AA was supported by an Australian Government RTP scholarship. We would like to thank
Alan Carey and Galina Levitina for useful input.

2 Background and notations

2.1 Spectral flow

The concept of spectral flow was used by Atiyah, Patodi and Singer in [6, 7] as a tool to
develop APS index theory. Spectral flow is intuitively defined as the net number of eigenvalues
which change sign along a path of self-adjoint operators, with the convention that an eigenvalue
changing from negative to positive will provide a contribution of 1 to the spectral flow. We use
the definition due to Phillips [30, 31]. Phillips’ definition of spectral flow is valid in the much
broader setting of semifinite von Neumann algebras with faithful normal semifinite traces, and
while we do not need the full power of such a definition, we do need the ability to handle
operators with continuous spectrum.
Consider the compact operators K(H) ⊂ B(H) with trace Tr and let π : B(H) → B(H)/K(H)
denote the projection onto the Calkin algebra. Let χ = χ[0,∞) be the characteristic function
of the interval [0, ∞). Let (Tt )t∈[0,1] be any norm-continuous path of bounded self-adjoint
Fredholm operators in B(H), so that π(Tt ) is a norm continuous path of invertibles. Then
π (χ (Tt )) = χ (π (Tt )). Since the spectrum of the π (Tt ) are bounded away from zero, the path
χ (π (Tt )) is continuous. By compactness (and [11, Lemma 4.1]) we can choose a partition
0 = t0 < t1 < · · · < tk = 1 such that
1
||π (χ (Tt )) − π (χ (Ts ))|| <
2
for all t, s ∈ [ti−1 , ti ] and 1 ≤ i ≤ k. Defining the projection Pi = χ (Tti ) we find that
Pi−1 Pi : Pi H → Pi−1 H is Fredholm. We recall the following definition, due to Phillips [30, 31].
Definition 2.1. Let H be a Hilbert space. For t ∈ [0, 1] let (Tt ) be any norm-continuous path
of bounded self-adjoint Fredholm operators in B(H). For a partition 0 = t0 < t1 < · · · < tk = 1
of the interval [0, 1] define the operators Pi = χ (Tti ). Then we define the spectral flow of the
path (Tt ) by
k
X
sf(Tt ) := Index (Pi−1 Pi ) .
i=1

We note that the above definition of spectral flow is independent of the choice of partition
[27, 30, 31] and agrees with the topological definition used in [6, 7] when both make sense. For
unbounded operators, we make the following definition of spectral flow [13].
Definition 2.2. Let H be a Hilbert space with trace Tr. Let (Dt ) be a graph norm continuous
path of unbounded self-adjoint Fredholm operators on H. Define the function F : R → [−1, 1]

2
 1
by F (x) = x(1 + x2 )− 2 . The spectral flow along the path (Dt ) is defined by

sf(Dt ) := sf(F (Dt )).

Throughout the rest of this section [0, 1] ∋ t 7→ Dt stands for a path of unbounded self-
adjoint linear Fredholm operators acting on some dense domain in H = L2 (Rn ). We denote by
(Ft ) = (F (Dt )) the bounded transform of the path (Dt ). We must also impose a smoothness
assumption on Dt to use analytic formulae for the spectral flow.
Definition 2.3. 1. A path [0, 1] ∋ t 7→ Dt is called Γ-differentiable at the point t = t0 if
and only if there exists a bounded linear operator T such that
1
lim t−1 (Dt − Dt0 )(Id + Dt20 )− 2 − T = 0.
t→t0

1
In this case we set Ḋt0 = T (Id + Dt20 ) 2 . The operator Ḋt is a symmetric linear operator
with domain Dom(Dt ) [15, Lemma 25].
1
2. If the mapping t 7→ Ḋt (Id+ Dt2 )− 2 is defined and continuous with respect to the operator
norm, then we call the path t 7→ Dt a continuously Γ-differentiable or a CΓ1 path.

The most general analytic spectral flow formula for the case of unbounded operators on a
Hilbert space is given by the following theorem [15, Theorem 9]. The sign of the second term
in (2.1) below appears incorrectly in [15, Theorem 9].
Theorem 2.4. Let [0, 1] ∋ t 7→ Dt be a piecewise CΓ1 path of linear operators and Ft ∈ B(H)
be Fredholm with ||Ft || ≤ 1. Let g : R → R be a positive C 2 function such that
R
1. R g(x) dx = 1;
R1
2. 0 Ḋt g(Dt ) dt < ∞; and
1

3. G(D1 ) − 12 B1 − G(D0 ) + 21 B0 ∈ L1 (H), where Bj = 2χ[0,∞) (Dj ) − 1, and G is the

antiderivative of g such that G(±∞) = ± 21 .

Then
1  
1 1
Z  
sf(Dt ) = Tr Ḋt g(Dt ) dt − Tr G(D1 ) − B1 − G(D0 ) + B0 . (2.1)
0 2 2

In our applications of this formula we will take Dt = H0 + αId + tV where H0 is the free
Hamiltonian, α a carefully chosen constant and V a suitable potential. We now describe these
ingredients.

2.2 Stationary scattering theory

We consider the scattering theory on Rn associated to the operators

n
X ∂2
H0 = − = −∆ and H = H0 + V,
j=1
∂x2j

3
where the (multiplication operator by the) potential V is a smooth compactly supported and
real-valued function. With h·, ·i the Euclidean inner product on Rn , we denote the Fourier
transform by
Z
n
F : L2 (Rn ) → L2 (Rn ), [Ff ](ξ) = (2π)− 2 e−ihx,ξi f (x) dx.
Rn

Note that the Fourier transform F is an isomorphism from H s,t to H t,s for any s, t ∈ R.
We denote by B(H1 , H2 ), K(H1 , H2 ) and L1 (H1 , H2 ) the bounded, compact and trace class
operators from H1 to H2 . For z ∈ C \ R, we let

R0 (z) = (H0 − z)−1 , R(z) = (H − z)−1 .

The operator H0 has purely absolutely continuous spectrum, and in particular no kernel. The
operator H can have finitely many eigenvalues which are negative, or zero [36, Theorem 6.1.1].
Several Hilbert spaces recur, and we adopt the notation (following [21, Section 2] which contains
an excellent discussion on the relations between the spaces and operators we introduce here)

H = L2 (Rn ), P = L2 (Sn−1 ), Hspec = L2 (R+ , P) ∼

= L2 (R+ ) ⊗ P.

Here Hspec provides the Hilbert space on which we can diagonalise the free Hamiltonian H0 .
Since V is bounded, H = H0 +V is self-adjoint with Dom(H) = Dom(H0 ). Since V ∈ Cc∞ (Rn ),
the wave operators
W± = s-lim eitH e−itH0
t→±∞

exist and are asymptotically complete [36, Theorem 1.6.2]. The wave operators are partial
isometries satisfying W±∗ W± = Id and W± W±∗ = Pac , the projection onto the absolutely
continuous subspace for H. The scattering operator is the unitary operator

S = W+∗ W− , (2.2)

which commutes strongly with the free Hamiltonian H0 . For our analysis of the scattering
operator, we describe the explicit unitary which diagonalises the free Hamiltonian.

Definition 2.5. Define the operator which diagonalises the free Hamiltonian H0 as
1 n−2 1
F0 : H → Hspec by [F0 f ](λ, ω) = 2− 2 λ 4 [Ff ](λ 2 ω).

By [21, p. 439] the operator F0 is unitary and for λ ∈ [0, ∞), ω ∈ Sn−1 and f ∈ Hspec we have

[F0 H0 F0∗ f ](λ, ω) = λf (λ, ω) =: λf (λ, ω).

As a consequence of the relation SH0 = H0 S, there exists a family {S(λ)}λ∈R+ of unitary

operators on P = L2 (Sn−1 ) such that for all λ ∈ R+ , ω ∈ Sn−1 and f ∈ H we have

[F0 Sf ](λ, ω) = S(λ)[F0 f ](λ, ω).

For historical reasons, we refer to S(λ) as the scattering matrix at energy λ ∈ R+ since in
dimension n = 1 the operator S(λ) is an M2 (C)-valued function.

4
Note that the operators H0 and H are not Fredholm, since 0 is in the essential spectrum of
both. To use the spectral flow formula of Theorem 2.4 we make the following adjustment for
the rest of this article. Let ν ≤ 0 be the furthest eigenvalue of H from zero. We fix α > −2ν +1,
so that the operators H0 (α) = H0 + α and H(α) = H + α define Fredholm operators. As a
consequence, the path

[0, 1] ∋ t 7→ H0 + tV + α =: Ht (α)

defines a CΓ1 path of Fredholm operators with Ḣt (α) = V . The operator H0 (α) has purely
absolutely continuous spectrum σ(H0 (α)) = [α, ∞) and the operator H(α) has absolutely
continuous spectrum σac (H(α)) = σ(H0 (α)). In addition, the operator H(α) has a finite
number of distinct eigenvalues 0 < λ1 (α) < λ2 (α) < · · · < λK (α) ≤ α of finite multiplicity.
The eigenvalues satisfy λj (α) = λj +α, with λ1 < λ2 < · · · < λK ≤ 0 the distinct eigenvalues of
H. We write M (λj ) = M (λj (α)) for the multiplicity of the eigenvalue λj and use the notation
N0 for the multiplicity of the zero eigenvalue for H. We also write
K
X
N= M (λj )
j=1

for the total number of eigenvalues of H (counted with multiplicity). Let Pac (H0 (α)) denote
the projection onto the absolutely continuous spectrum for H0 (α). The wave operators

W± (α) = s-lim eitH(α) e−itH0 (α) Pac (H0 (α)) = W±

t→±∞

exist and are asymptotically complete by the invariance principle [32, Theorem XI.11]. Direct
calculation gives the following diagonalisation for H0 (α).

Lemma 2.6. The operator Fα : H → L2 ([α, ∞)) ⊗ L2 (Sn−1 ) given by

[Fα f ](λ, ω) = [F0 f ](λ − α, ω)

satisfies

[Fα H0 (α)f ](λ, ω) = λ[Fα f ](λ, ω).

The scattering operator S = W+∗ W− is unitary and commutes with H0 (α) and so there exists
a family {Sα (λ)}λ∈[α,∞) of unitary operators on L2 (Sn−1 ) such that

[Fα Sf ](λ, ω) = Sα (λ)[Fα f ](λ, ω).

In fact, we have Sα (λ) = S(λ − α) for all λ ∈ [α, ∞). Pointwise we have Sα (λ) − Id ∈
L1 (L2 (Sn−1 )), [36, Proposition 8.1.5].

2.3 The spectral shift function and the Birman-Kreı̆n trace formula

We now recall the spectral shift function [12, 26] for the pair (H(α), H0 (α)) and some of its
defining properties (see [36, Theorems 0.9.2 and 0.9.7]). The proofs in [36] only consider α = 0,
however extend directly to α > 0 by translation.

5
Theorem 2.7. Suppose that V ∈ Cc∞ (Rn ), α ≥ 0 and let S be the corresponding scattering
operator. Then there exists a unique (up to an additive constant) real-valued piecewise-C 1
function ξα (·, H, H0 ) : R → R such that
Z
Tr(f (H(α)) − f (H0 (α))) = ξα (λ, H, H0 )f ′ (λ) dλ, (2.3)
R

at least for all f ∈ C 2 (R) with two locally bounded derivatives and satisfying

d
λm+1 f ′ (λ) = O(λ−1−ε )


(2.4)
dλ
as λ → ∞, for some ε > 0 and m > n2 . We specify ξα (·, H, H0 ) uniquely by the convention
ξα (λ, H, H0 ) = 0 for λ sufficiently negative. Thus for λ < α, ξα (·, H, H0 ) satisfies the relation
K
X
ξα (λ, H, H0 ) = − M (λk (α)) χ[λk (α),∞) (λ),
k=1

where we we have indexed the distinct eigenvalues of H(α) as λ1 (α) < · · · < λK (α) and each
λj (α) has multiplicity M (λj (α)). Furthermore, we have ξα (·, H, H0 )|(α,∞) ∈ C 1 (α, ∞) and for
λ > α the relations

Det(Sα (λ)) = e−2πiξα (λ) and Tr Sα (λ)∗ Sα′ (λ) = −2πiξα′ (λ)

hold. Furthermore, we have ξα (λ) = ξ0 (λ − α) for almost all λ ∈ R.

We call ξα (·, H, H0 ) the spectral shift function for the pair (H(α), H0 (α)) and will often just
write ξα = ξα (·, H, H0 ). We also write ξ = ξ0 . Using integration by parts we can rewrite the
definining property (2.3) in a sometimes more convenient fashion, known as the Birman-Kreı̆n
trace formula [18, Theorem III.4].

Lemma 2.8. Suppose that V ∈ Cc∞ (Rn ), α ≥ 0 and let Sα , ξα be the corresponding scattering
operator and spectral shift function. Then for all f ∈ Cc∞ (R) we have

∞ K
1
Z X
∗
Sα′ (λ)


Tr(f (H(α)) − f (H0 (α))) = f (λ)Tr Sα (λ) dλ + f (λk (α))M (λk (α))
2πi α k=1
+ f (α) (ξα (α−) − ξα (α+) − M (α)) ,

where we have defined ξα (α±) = lim ξα (α ± ε).

ε→0+

In fact by Theorem 2.7 we have, with N the total number of eigenvalues of H counted with
multiplicity and N0 = M (α) the number of zero eigenvalues for H, the relation ξα (α−) =
−N + N0 . We can then rewrite the Birman-Kreı̆n trace formula as

∞ K
1
Z X
∗
Sα′ (λ)


Tr(f (H(α)) − f (H0 (α))) = f (λ)Tr Sα (λ) dλ + f (λk (α))M (λk (α))
2πi α k=1
+ f (α) (−N − ξα (α+)) .

6
2.4 Resolvent expansions and limiting behaviour of the spectral shift function

For k ∈ N ∪ {0} and f ∈ Cc∞ (Rn ) we introduce the notation f (k) = [H0 , [H0 , [· · · , [H0 , f ] · · · ]]],
where the expression has k commutators. We recall the following pseudodifferential expansion
of the resolvent [1, Lemma 4.8] (see also [14, Lemma 6.11]).

Lemma 2.9. Suppose that V ∈ Cc∞ (Rn ). For all M, K ≥ 0 and z ∈ / σ(H) we have the
expansion
 
M
X K
X
R(z) = (H − z)−1 =  Cm (k)(−1)m+|k| V (k1 ) · · · V (km ) R0 (z)m+|k|+1 + Pm,K (z)
m=0 |k|=0

+ (−1)M +1 (R0 (z)V )M +1 R(z),

where the remainder Pm,K (z) is a pseudodifferential operator of order at most −2m − K − 3.
The combinatorial coefficients Cm (k) are given by

(m + |k|)!
Cm (k) = .
k1 ! · · · km !(k1 + 1)(k1 + k2 + 2) · · · (|k| + m)

Note that the operator V (k1 ) · · · V (km ) is a differential operator of order at most |k| with smooth
compactly supported coefficients and thus we may write
|k|
X
(k1 ) (km )
V ···V = gk,β ∂ β , (2.5)
|β|=0

where the multi-indices β are of length n and gk,β ∈ Cc∞ (Rn ).

We now recall the high-energy behaviour of the spectral shift function and its derivative [1,
Lemma 2.15, Theorem 4.15 and Remark 4.16].

Lemma 2.10. Suppose that V ∈ Cc∞ (Rn ). Then for 1 ≤ ℓ ≤ ⌊ n2 ⌋ there exist coefficients
Cℓ (n, V ), cℓ (n, V ), βn (V ) such that

⌊ n−1
 
2
⌋
X n
−ℓ
0 = lim −2πiξ(λ) − 2πiβn (V ) − Cℓ (n, V )λ 2 
λ→∞
ℓ=1
⌊ n−1
 
2
⌋
X n
= lim −2πiξ ′ (λ) − cℓ (n, V )λ 2 −ℓ−1  .
λ→∞
ℓ=1

The coefficients are related by cℓ (n, V ) = n2 − ℓ Cℓ (n, V ). For 1 ≤ ℓ ≤ ⌊ n−1

2 ⌋ and M, K ∈ N
with M + K ≥ n we define the set

QM,K (ℓ) = (m, k, β) ∈ {0, 1, . . . , M } × {0, 1, . . . , K}m × {0, 1, . . . , K}n : |β| ≤ |k|,


|β|
and m + |k| + 1 − =ℓ .
2

7
The coefficients Cℓ (n, V ) are given by
   
X (−1)m+|k|+1 (2πi)Cm (k)(−i)|β| Γ β12+1 · · · Γ βn2+1 Z
Cℓ (n, V ) =   V (x)gk,β (x) dx,
n |β|
(m,k,β)∈QM,K (j) (m + 1)(m + |k|)!Γ 2 − m − |k| + 2 (2π)n Rn

(2.6)

and
(
0, if n is odd,
βn (V ) = 1
(2.7)
2πi C 2 (n, V ), if n is even.
n

Definition 2.11. Define the functions Pn , pn : (0, ∞) → C by

⌊ n−1
2
⌋
X n
Pn (λ) = 2πiβn (V ) + Cℓ (n, V )λ 2 −ℓ ,
ℓ=1
⌊ n−1
2
⌋
X n
pn (λ) = cℓ (n, V )λ 2 −ℓ−1 = Pn′ (λ).
ℓ=1

We call Pn the high-energy polynomial for ξ and pn the high-energy polynomial for ξ ′ .
Remark 2.12. Recall the spectral shift functions ξ, ξα for the pairs (H, H0 ) and (H(α), H0 (α)).
Since ξα (λ) = ξ(λ − α) for all almost all λ ∈ R we have that the high-energy polynomial for
ξα is Pn (· − α) and likewise for ξ ′ and pn (· − α).

We can explicitly compute the lowest order polynomials (see [10, 16]), finding P1 = 0, and

(2πi)Vol(S1 ) 2πi
Z Z
P2 (λ) = − V (x) dx = − V (x) dx,
2(2π)2 R2 4π R2
1 1 Z
(2πi)λ 2 Vol(S2 ) (2πi)λ 2
Z
P3 (λ) = − V (x) dx = − V (x) dx,
2(2π)3 R3 4π 2 R3
(2πi)λVol(S3 ) (2πi)Vol(S3 )
Z Z
P4 (λ) = − V (x) dx + V (x)2 dx.
2(2π)4 R 4 4(2π) 4
R 4

The integrability properties of the derivative of the spectral shift function on R+ are well-
known, see [22, Theorem 5.2] and [1, Lemma 4.12].
Lemma 2.13. Suppose that V ∈ Cc∞ (Rn ). Then the function Tr (S(·)∗ S ′ (·)) − pn is integrable
on R+ . In particular, if n = 1, 2 we have Tr (S(·)∗ S ′ (·)) ∈ L1 (R+ ).

We now define zero-energy resonances, a low-energy phenomena known to provide obstructions

to generic behaviour in scattering theory in low dimensions.
Definition 2.14. Suppose that V ∈ Cc∞ (Rn ). If n 6= 2 we say there is a resonance if there exists
a non-zero bounded distributional solution to Hψ = 0. If n = 2 we say there is a p-resonance if
there exists a non-zero distributional solution ψ to Hψ = 0 with ψ ∈ Lq (R2 )∩L∞ (R2 ) for some
q > 2. We say that there is an s-resonance if there exists a non-zero bounded distributional
solution ψ to Hψ = 0 with ψ ∈ / Lq (R2 ) for all q < ∞.

8
General bounds on the resolvent of H [20] show that there can be no resonances for dimension
n ≥ 5.
We now recall the value of the spectral shift function at zero in all dimensions from [1, Corollary
5.11].
Theorem 2.15. Suppose V ∈ Cc∞ (Rn ). Then the value of the spectral shift function at zero
is given by ξ(0+) = −N − Nres , where Nres = 0 unless

1
2, if n = 1 and there are no resonances,




N , if n = 2 and there are Np = 0, 1, 2 p-resonances,
p
Nres = 1
2,
 if n = 3 and there are resonances,


1, if n = 4 and there are resonances.

We note that the proof of Theorem 2.15 in [1] is as a corollary of Levinson’s theorem, however
the result can be obtained directly using perturbation determinant methods in odd dimensions
(see [28], [29] and [18, Theorem 3.3]).

3 Spectral flow for Schrödinger operators

In this section we analyse the spectral flow formula of Theorem 2.4 applied to the path Ht (α)
by making a particular choice of the function g and then taking residues.
Define for Re(s) > 21 the constants Cs = R (1 + u2 )−s du. The functions s 7→ Cs have a pole
R

at s = 12 with residue equal to one. For Re(s) > 12 we define the eta function ηs : R → C by
Z ∞ Z ∞
1 2 −s − 12 2
ηs (x) = x(1 + wx ) w dw = (1 + v 2 )−s ds,
Cs 1 Cs x
where the second expression is valid only for x > 0. We can now use the function ηs to obtain
a useful form of Theorem 2.4.
Lemma 3.1. Let [0, 1] ∋ t 7→ Dt be a piecewise CΓ1 path of linear operators with Ḋt (1 + Dt2 )−s
trace-class for all s > n4 . Then
sf(Dt ) =
Z 1  
 Cs
Tr Ḋt (Id + Dt2 )−s dt +


Res Tr ηs (D1 ) − ηs (D0 ) + PKer(D1 ) − PKer(D0 ) . (3.1)
s= 12 0 2

Proof. For s > n4 , let gs : R → R be given by gs (x) = Cs−1 (1 + x2 )−s . Note that the
antiderivative Gs of gs with G(±∞) = ± 12 is given by
Z x
1 1
Gs (x) = − + (1 + u2 )−s du.
2 Cs −∞
The function gs is even and so Gs is odd. For x > 0 we have Gs (x) = 12 − 12 ηs (x), while for
x < 0 we have Gs (x) = − 12 − 21 ηs (x). Thus applying Theorem 2.4 to gs , Gs and multiplying
both sides by Cs yields
Z 1   Cs
Tr Ḋt (Id + Dt2 )−s dt +


Cs sf(Dt ) = Tr ηs (D1 ) − ηs (D0 ) + PKer(D1 ) − PKer(D0 ) .
0 2
(3.2)

9
The left-hand side of Equation (3.2) is a meromorphic function of s with a simple pole at s = 21
and thus defines a meromorphic continuation of the right-hand side of Equation (3.2) with a
simple pole at s = 12 . As a result, taking the residue at s = 21 gives Equation (3.1).

Equation (3.1) is the starting point for our analysis of the spectral flow along the path Ht (α).
There are two separate types of terms to be considered. The first is the “integral of one-form”
term which is evaluated in Section 3.1 using the pseudodifferential expansion of Lemma 2.9
and the second is the η contribution which is evaluated in Section 3.2 using the Birman-Kreı̆n
trace formula.

3.1 The “integral of one form” term

We use the pseudodifferential expansion of Lemma 2.9 to compute an expansion for the integral
of one form term in Theorem 2.4. After a fixed number of terms (depending on the dimension
n) the remainder term will be holomorphic at s = 21 and can be discarded. We begin with a
residue computation.
Lemma 3.2. For ℓ ∈ N, α > 0 we have
 
ℓ−1 ℓ− j2 −1 ℓ−j−1 j+1
Z ∞  
X ℓ−1  (−1) α Γ 2
Res uℓ−1 (1 + (u + α)2 )−s du =   .
s= 21 0 j 4Γ j + 1 Γ 1


j=0 2 2
j even

Proof. Fix s ∈ C with Re(s) > ℓ+1 2 . We make the substitution v = u + α and apply the
binomial expansion to obtain
Z ∞ Z ∞
ℓ−1 2 −s
u (1 + (u + α) ) du = (v − α)ℓ−1 (1 + v 2 )−s dv
0 α
ℓ−1
X  Z ∞
ℓ−1
= (−α)ℓ−j−1 v j (1 + v 2 )−s dv
j α
j=0
ℓ−1   ∞
ℓ−1
X Z
= (−α)ℓ−j−1 v j (1 + v 2 )−s dv
j 0
j=0
ℓ−1   α
ℓ−1
X Z
− (−α)ℓ−j−1 v j (1 + v 2 )−s dv.
j 0
j=0

Since the integrals from 0 to α are over a finite region, they are holomorphic at s = 12 and thus
have vanishing residue. So we compute for 0 ≤ j ≤ ℓ − 1 that
   
j+1 j+1
Z ∞
1
Z ∞
j+1
Γ 2 Γ s − 2
v j (1 + v 2 )−s dv = w 2 −1 (1 + w)−s dw = .
0 2 0 2Γ(s)
1
Taking the residue at s = 2 we find
 j
Z ∞   (−1) 2 j+1
2
Γ( j+1
2 )
, if j is even,
Res v j (1 + v 2 )−s dv = j
2Γ( 2 +1)Γ( 12 )
s= 21 0 0, otherwise,


from which the result follows.

10
To evaluate some further traces, we need to be able to integrate polynomials over Sn−1 . We
use the following result [17].

Lemma 3.3. Let β be a multi-index of length n and let Pβ : Rn → C be given by Pβ (x) =

xβ = xβ1 1 · · · xβnn . Then

Z 0, 


if some βj is odd,
Pβ (ω) dω = 2Γ β12+1 ···Γ( βn2+1 )
Sn−1    , if all βj are even.
Γ n+|β|

2

We now contribute the residue of the contribution from the “integral of one form” term to the
spectral flow.

Proposition 3.4. Suppose that V ∈ Cc∞ (Rn ) and α > −2ν. Then for n odd we have
Z 1  


2 −s
Res Tr V Id + (H0 + tV + α) dt = 0.
s= 21 0

If n is even we have
Z 1 


2 −s
Res Tr V Id + (H0 + tV + α) dt
s= 21 0
 
n
2
n
−ℓ 
2 n  (−1) n2 −ℓ− 2j −1 α n2 −ℓ−j Γ j+1
2 −ℓ 2
X X
=   Cℓ (n, V ),
j 4(2πi)Γ j
+1 Γ 1


ℓ=1 j=0 2 2
j even

with the Cℓ (n, V ) the high-energy coefficients defined in Equation (2.6).

n
Proof. For α > 0 and Re(s) > 4 we define the function ϕα,s : R → C by

−s
ϕα,s (x) = 1 + (x + α)2 ,

using the principal branch of the

logarithm. The function ϕα,s is holomorphic in the half-plane
α
Re(z) > −α. Let a ∈ − 2 , 0 so that a < λ for all t ∈ [0, 1] and λ ∈ σ(H0 + tV ) and define
the vertical line γ = {a + iv : v ∈ R}. For t ∈ [0, 1] we use Cauchy’s integral formula to write

1
Z
ϕα,s (H0 + tV ) = − ϕα,s (z)(H0 + tV − z)−1 dz. (3.3)
2πi γ

Denoting Rt (z) = (H0 + tV − z)−1 we have by Lemma 2.9 that for all M, K ≥ 0 that
 
M
X XK
Rt (z) = tm (−1)m+|k| Cm (k)V (k1 ) · · · V (km ) R0 (z)m+|k|+1 + Pm,K,t (z)
m=0 |k|=0

+ (−1)M +1 tM +1 (R0 (z)V )M +1 Rt (z),

11
where Pm,K,t (z) has order (at most) −2m − K − 3. We can now write Equation (3.3) as
M K
1 X X Z
m m+|k| (k1 ) (km )
ϕα,s (H0 + tV ) = − t (−1) Cm (k)V ···V ϕα,s (z)R0 (z)m+|k|+1 dz
2πi γ
m=0 |k|=0
!
(−1)M tM +1
Z Z
+ ϕα,s (z)Pm,K,t (z) dz + ϕα,s (z)(R0 (z)V )M +1 Rt (z) dz
γ 2πi γ
M K
1 X X Z
m m+|k| (k1 ) (km )
:= − t (−1) V ···V ϕα,s (z)R0 (z)m+|k|+1 dz
2πi γ
m=0 |k|=0

+ E(M, K, t, α, s).

Using again Cauchy’s integral formula we can compute that

1 1 dm+|k| ϕα,s
Z
ϕα,s (z)R0 (z)m+|k|+1 dz = − |z=H0 .
2πi γ (m + |k|)! dz m+|k|

Thus we have the expression

M X
K
X Cm (k)(−1)m+|k| tm (k1 ) dm+|k| ϕα,s
ϕα,s (H0 + tV ) = V · · · V (km ) |z=H0 + E(M, K, t, α, s).
m=0 |k|=0
(m + |k|)! dz m+|k|

Choose M = ⌊n⌋ and for 0 ≤ m ≤ M let K = M − m. Since V (k1 ) · · · V (km ) is a differential

operator of order |k|, we can write
|k|
X
(k1 ) (km )
V ···V = gk,β ∂ β ,
|β|=0

where β is a multi-index of length n and gk,β ∈ Cc∞ (Rn ). Then we can use Lemma 3.3 to
compute
|k|
! !
m+|k| ϕ m+|k| ϕ
(km ) d α,s βd α,s
X
(k1 )
Tr V V ···V |z=H0 = Tr V gk,β ∂ |z=H0
dz m+|k| dz m+|k|
|β|=0
|k|
!
X Z dm+|k|ϕα,s
 Z
= (2π)−n V (x)gk,β (x) dx (−i)|β| ξ β (|ξ|2 ) dξ
Rn Rn dz m+|k|
|β|=0
   
β1 +1
|k| 2(−i)|β| Γ · · · Γ βn2+1 Z
!
∞
dm+|k| ϕα,s 2
 Z
X 2
=   V (x)gk,β (x)dx r n+|β|−1 (r ) dr
(2π)n Γ n+|β| Rn 0 dz m+|k|
|β|=0 2
β even
   
β1 +1
|k| (−i)|β| Γ · · · Γ βn2+1 Z
!
∞
dm+|k| ϕα,s
 Z
X 2 n+|β|
−1
=   V (x)gk,β (x) dx u 2 (u) du ,
(2π)n Γ n+|β| Rn 0 dum+|k|
|β|=0 2
β even

12
 n+|β|
where the sum is over multi-indices β with all βj even. First, suppose that m + |k| ≤ 2 − 1.
Integrating by parts in the u integral (m + |k|) times we find
!
m+|k| ϕ
(k1 ) (km ) d α,s
Tr V V ···V |z=H0
dz m+|k|
   
|k| (−i)|β| (−1)m+|k| Γ β1 +1 · · · Γ βn +1 Z 
X 2 2
=   V (x)gk,β (x) dx
|β|=0 (2π)n Γ n+|β| 2
Rn
β even
!
∞
dm+|k|  n+|β| −1 
Z
× u 2 ϕα,s (u) du
0 dum+|k|
   
|k| (−i)|β| (−1)m+|k| Γ β1 +1 · · · Γ βn +1 Z 
X 2 2
=   V (x)gk,β (x) dx
|β|=0 (2π)n Γ n+|β|
2 − m − |k| Rn
β even
Z ∞ n+|β|

−m−|k|−1
× u 2 ϕα,s (u) du .
0

Note that the boundary terms from the integration by parts vanish since n+|β|2 −m−|k|−1 > 0.
We first consider n odd and make the estimate
Z ∞ Z ∞
n+|β| n+|β|
−m−|k|−1
u 2 ϕα,s (u) du ≤ u 2 −m−|k|−1 (1 + u2 )−Re(s) du
0 0
1 ∞ n+|β| − m+|k| −1
Z
= v 4 2 (1 + v)−Re(s) dv
2 0
   
Γ n+|β|
4 − m+|k|
2 Γ Re(s) + m+|k|
2 − n+|β|
4
= .
2Γ (Re(s))
Since n is odd, we have
1 m + |k| n + |β|
+ − ∈
/Z
2 2 4
for all possible m, k, β and thus we find
Z ∞ 
n+|β|
−m−|k|−1
Res u 2 ϕα,s (u) du = 0,
s= 21 0

from which we deduce that

Z 1   Z 1 


2 −s
Res Tr V Id + (H0 + tV + α) dt = Res Tr (V E(M, K, t, α, s) dt ,
s= 21 0 s= 21 0

which we show is zero below. We now consider n even. An application of Lemma 3.2 gives
Z ∞ 
n+|β|
Res (u 2 −m−|k|−1 ϕα,s (u) du
s= 21 0
n+|β| n+|β| n+|β|
 
−m−|k|−1 −m−|k|−1 j+1
2 X  n+|β|
− m − |k| − 1 (−1) 2
 α 2 −m−|k|−j−1 Γ 2
= 2   .
j 4Γ 2j + 1 Γ 21


j=0

13
 n+|β| n+|β|
Next, we consider the case m + |k| > 2 − 1. If n is even, we integrate by parts 2 −1
times in the u-integral to obtain
n+|β|
m+|k|+1− n+|β|
! !
∞ Z ∞ −1
dm+|k| ϕα,s d d ϕ
Z
n n+|β| 2 n+|β| 2 α,s
−1
u 2 (u) = (−1) 2 −1 u 2 −1 (u) du
0 dum+|k| 0 du 2
n+|β|
−1
du m+|k|+1−
n+|β|
2

  Z ∞ m+|k|+1− n+|β|
n+|β| n + |β| d 2 ϕα,s
= (−1) 2 Γ n+|β|
(u) du
2 0 dum+|k|+1− 2
n+|β|
n + |β| dm+|k|− 2 ϕα,s
 
n+|β|
= (−1) 2 +1 Γ n+|β|
,
2 dum+|k|+1− 2 u=0
n+|β|
which is holomorphic at s = 21 . If n is odd, then a similar estimate to the case m+|k| ≤ 2 −1
shows that the contribution is holomorphic at s = 21 . So we find
Z 1 ! !
d m+|k| ϕ
α,s
Res tm Tr V V (k1 ) · · · V (km ) |z=H0 dt
s= 21 0 dz m+|k|
n+|β|
|k| 2
−m−|k|−1 Z   n+|β| 
X X
2 − m − |k| − 1
= V (x)gk,β (x) dx
Rn j
|β|=0 j=0
β even j even
n+|β| n+|β|
     
−m−|k|− j+1
(−i)|β| (−1) 2 α 2 −m−|k|−j−1 Γ β12+1 · · · Γ βn2+1 Γ j+1
2
2
×     .
(2π)n (m + 1)Γ 2j + 1 Γ 12 Γ n+|β|


2 − m − |k|

For ℓ ∈ N define the set

QM,K (ℓ) = {(m, k, β) ∈ {0, 1, . . . , M } × {0, 1, . . . , K}m × {0, 1, . . . , K}n : |β| ≤ |k|,
|β|
m + |k| + 1 − = ℓ}.
2
Recalling the coefficients Cℓ (n, V ) of the high-energy polynomial for ξ of Equation (2.6) we
have that
 ! 
M Z 1 K m+|k| m+|k|
X X Cm (k)(−1) d ϕα,s
Res  tm Tr V V (k1 ) · · · V (km ) |
m+|k| z=H0
dt
1
s= 2 0 (m + |k|)! dz
m=0 |k|=0
n+|β|
M X
K |k| 2
−m−|k|−1 Z   n+|β| 
X X X
2 − m − |k| − 1
= V (x)gk,β (x) dx
m=0 |k|=0 |β|=0 Rn j
j=0
β even j even
n+|β| n+|β|
     
−m−|k|− j+1
(−i)|β| (−1) 2 α 2 −m−|k|−j−1 Γ β12+1 · · · Γ βn2+1 Γ j+1
2
2
×    
j n+|β|
(2π)n (m + 1)Γ 2 + 1 Γ 12 Γ


2 − m − |k|
 
n n n
−ℓ− 2j −1 n
2
X 2
X
−ℓ 
n
− ℓ (−1) 2
 α 2 −ℓ−j−1 Γ j+1
2
= 2   Cℓ (n, V ).
j 4(2πi)Γ j + 1 Γ 1


ℓ=1 j=0 2 2
j even

14
We now consider the contribution from the remainder term E(M, K, t, α, s). There are two
types of terms to consider. The first are those involving Pm,K,t (z). Since V ∈ Cc∞ (Rn ) we can
factorise V = q1 q2 with q1 , q2 ∈ Cc∞ (Rn ). Since Pm,K,t (z) has order at most −2m − K − 3,
there exists C > 0 such that
K 3
R0 (z)−m− 2 − 2 q2 Pm,K,t (z) ≤ C.

Note also that

K 3 m K 3 n+1
q1 R0 (z)m+ 2 + 2 ≤ (a2 + v 2 )− 2 − 4 − 4 + 4 ,
1

which follows from a careful application of the Rellich lemma. Combining these we make the
estimate
Z Z
Re(s) m K 3 n+1
ϕα,s (z)Pm,K,t (z) dz ≤ C (1 + ((a + α)2 + v 2 ))− 2 (a2 + v 2 )− 2 − 4 − 4 + 4 dv,
γ 1 R

which is finite for Re(s) + m + K 3 n+1 1

2 + 2 > 2 + 1. Recalling that Re(s) > 2 and we have chosen
M = n and K = M − m guarantees convergence. A similar argument to [14, Lemma 7.4]
shows that this contribution is holomorphic at s = 21 , as is the contribution from the terms
containing Rt (z). Thus for both n even and odd, we find
Z 1 
Res Tr (V E(M, K, t, α, s) dt = 0,
s= 21 0

which completes the proof.

3.2 The Birman-Kreı̆n term

In this subsection we use the Birman-Kreı̆n trace formula to determine the kernel and η con-
tributions to the spectral flow.

Lemma 3.5. By construction, the projections PKer(H(α)) and PKer(H0 (α)) are both zero.

Since the kernel terms both vanish we are now able to evaluate the η contributions. We note
that by Proposition 3.4 the residue of the integral of one form contribution to Equation (3.1)
at s = 21 exists, and thus so does the residue of the Birman-Kreı̆n contribution at s = 21 .

Lemma 3.6. Suppose that V ∈ Cc∞ (Rn ). Then the η contribution to the Hamiltonian spectral
flow is given by


Res Cs Tr(ηs (H(α)) − ηs (H0 (α))
s= 21
 ∞ 
1
Z
∗
= N + Nres + Res Cs ηs (λ)Tr(Sα (λ) Sα′ (λ)) dλ ,
s= 12 2πi α

where N is the number of eigenvalues of H = H0 + V , counted with multiplicity, and Nres is

the contribution from resonances as defined in Theorem 2.15.

15
Proof. Choose s > n2 + 1, so that Equation (2.4) is satisfied for ηs and thus the Birman-
Kreı̆n trace formula can be applied to ηs . Enumerate the distinct eigenvalues of H(α) as
0 < λ1 (α) < · · · < λK (α) ≤ α. We prove the result in the case λK (α) = α, that is in the case
that zero is an eigenvalue for H = H0 + V . Apply the Birman-Kreĭn trace formula to obtain

∞ K−1
1
Z X
Tr (ηs (H(α)) − ηs (H0 (α))) = ηs (λ)Tr(Sα (λ)∗ Sα′ (λ)) dλ + M (λk (α))ηs (λk (α))
2πi α k=1
+ M (α)ηs (α) + ηs (α) (ξα (α−) − ξα (α+) − M (α)) ,

where ξα is the spectral shift function for the pair (H(α), H0 (α)) and M (λj (α)) denotes the
multiplicity of the eigenvalue λj (α) for the operator H(α). Recall that by construction, we
have ξα (λ) = ξ(λ − α) so that ξα (α±) = ξ(0±). Thus after multiplying by Cs we have

Cs ∞
Z
Cs Tr (ηs (H(α)) − ηs (H0 (α))) = ηs (λ)Tr(Sα (λ)∗ Sα′ (λ)) dλ (3.4)
2πi α
K
X
+ Cs M (λk (α))ηs (λk (α)) + Cs ηs (α) (ξ(0−) − ξ(0+) − N0 ) .
k=1

Observe that for x 6= 0 we have Res (Cs ηs (x)) = sign(x).

s= 21

The left hand side of Equation (3.4) has a residue at s = 12 if and only if the first term on
the right hand side does. Note that by Theorem 2.15 we have ξ(0−) − ξ(0+) − N0 = Nres . It
remains to take the residue at s = 21 .
By construction we have λj (α) > 0 for all j and thus

Res (Cs Tr (ηs (H(α)) − ηs (H0 (α))))

s= 21
K
!
∞
Cs
Z X
= Res ηs (λ)Tr(Sα (λ)∗ Sα′ (λ)) dλ + M (λk )Cs ηs (λk (α)) + Cs ηs (α)Nres
s= 21 2πi α k=1
 ∞ 
1
Z
∗
= N + Nres + Res Cs ηs (λ)Tr(Sα (λ) Sα′ (λ)) dλ ,
s= 21 2πi α

as claimed.

We can now compute the residue of the Birman-Kreı̆n integral contribution to the spectral
flow with the aid of a technical result.

Lemma 3.7. Fix β ≥ 0 and f, g : R+ → C with f − g ∈ L1 (R+ ). Suppose in addition that

 Z ∞ 
Res Cs ηs (λ)f (λ − β) dλ
s= 21 β

exists. Then
 Z ∞  Z ∞  Z ∞ 
Res Cs ηs (λ)f (λ − β) dλ = (f (λ) − g(λ)) dλ + Res Cs ηs (λ)g(λ − β) dλ .
s= 21 β 0 s= 12 β

16
Proof. Adding zero gives
 Z ∞ 
Res Cs ηs (λ)f (λ − β) dλ (3.5)
s= 21 β
 Z ∞ Z ∞ 
= Res Cs ηs (λ)(f (λ − β) − g(λ − β)) dλ + Cs ηs (λ)g(λ − β) dλ .
s= 21 β β

One straightforwardly checks that Ress=1/2 ηs (x) = 1 for all x > 0. Then an application of the
dominated convergence theorem allows us to compute that
Z ∞  Z ∞
Res ηs (λ)(f (λ − β) − g(λ − β)) dλ = Res ηs (λ)(f (λ − β) − g(λ − β)) dλ
s= 21 β β s= 12
Z ∞
= (f (λ − β) − g(λ − β)) dλ.
β

Since the residue of the first term on the right-hand side of Equation (3.5) exists, so does the
residue of the second term on the right-hand side. Making the substitution u = λ−β completes
the proof.

Proposition 3.8. Suppose that V ∈ Cc∞ (Rn ) and α > −2ν. Then for n odd we have
Cs ∞
  Z ∞
1
Z
∗ ′
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ.


Res Tr(Sα (λ) Sα (λ))ηs (λ) dλ =
s= 12 2πi α 2πi 0
If n is even we have
C Z ∞  1
Z ∞
s ∗ ′
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ


Res Tr(Sα (λ) Sα (λ))ηs (λ) dλ =
s= 2 2πi α
1 2πi 0
 
n−1 n
⌊ 2 ⌋ 2 −ℓ  n
−ℓ− 2j n −ℓ−j j+1
n  (−1) 2 α 2 Γ
2 −ℓ 2
X X
+  
Cℓ (n, V ),
j 2(2πi)Γ j + 1 Γ 1
ℓ=1 j=0 2 2
j even

where the Cℓ (n, V ) are the high-energy coefficients for Pn defined in Equation (2.6)

Proof. Note that by Lemma 2.13 the map [0, ∞) ∋ λ 7→ Tr(S(λ)∗ S ′ (λ)) − pn (λ) is integrable
on [0, ∞) and thus we can apply Lemma 3.7 to obtain
Cs ∞
 Z 
∗ ′


Res Tr(Sα (λ) Sα (λ)) ηs (λ) dλ
s= 21 2πi α
Z ∞  Z ∞ 
1 1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ +


= Res Cs pn (λ − α)ηs (λ) dλ .
2πi 0 2πi s= 21 α

Thus it remains to compute

 n−1 
⌊ 2 ⌋
 ∞  X cℓ (n, V ) Z ∞
Cs
Z
n
Res pn (λ − α)ηs (λ) dλ = Res  Cs (λ − α) 2 −ℓ−1 ηs (λ) dλ
s= 21 2πi α s= 12 2πi α
ℓ=1
 n−1 
⌊ 2 ⌋
X cℓ (n, V ) Z ∞ n
= Res  Cs u 2 −ℓ−1 ηs (u + α) du .
1
s= 2 2πi 0ℓ=1

17
We show that the residue of each of the terms in the sum exist individually, so that the
summation can be passed through the residue. First we consider n odd. We integrate by parts
to find
Z ∞ Z ∞Z ∞
n n
Cs u 2 −ℓ−1 ηs (u + α) du = u 2 −ℓ−1 (1 + v 2 )−s dv du
0 0 u+α
" n Z #∞
−ℓ ∞ Z ∞
u2 2 −s 1 n
= n (1 + v ) dv + n u 2 −ℓ (1 + (u + α)2 )−s du
2 − ℓ u+α 0 2 −ℓ 0
Z ∞ Z ∞
1 n
−ℓ 2 −s 1 n
= n u 2 (1 + u ) du + n u 2 −ℓ (1 + (u + α)2 )−s − (1 + u2 )−s du
2 −ℓ 0 −ℓ 0
n ℓ 1

n ℓ 1

2
Γ 4 − 2 + 2 Γ s− 4 + 2 − 2
= + holo(s)
2Γ(s)
where holo is a function holomorphic at s = 12 . Since n is odd, we have 1 − n4 + 2ℓ ∈
/ Z and thus
n−1
 ∞  ⌊X2
⌋  Z ∞ 
Cs cℓ (n, V )
Z
n
Res pn (λ − α)ηs (λ) dλ = Res Cs u 2 −ℓ−1 ηs (u + α) du
s= 21 2πi α 2πi s= 12 0
ℓ=1
= 0.
Now we consider n even. In this case we integrate by parts to write
Z ∞ Z ∞
n
−ℓ−1 1 n d
(λ − α) 2 ηs (λ) dλ = − n (λ − α) 2 −ℓ ηs (λ) dλ
α 2 −ℓ α dλ
Z ∞
1 n
= n (λ − α) 2 −ℓ (1 + λ2 )−s dλ.
2 − ℓ α
We now use the binomial expansion to obtain
n
Z ∞ −ℓ 
2 n  Z ∞
n
−ℓ−1 1 X 2 − ℓ (−α) 2 −ℓ−j
n
(λ − α) 2 ηs (λ) dλ = n λj (1 + λ2 )−s dλ.
α 2 − ℓ j α
j=0

Returning to the residue calculation we find

 Z ∞ 
n
−ℓ−1
Res Cs (λ − α) 2 ηs (λ) dλ
s= 21 α
n
−ℓ
1 X n2 − ℓ
2  Z ∞ 
n
−ℓ−j j 2 −s
= n (−α) 2 Res λ (1 + λ ) dλ
2 −ℓ j s= 21 α
j=0
n
−ℓ
1 X n2 − ℓ
2   Z ∞ 
n
= n (−α) 2 −ℓ−j Res λj (1 + λ2 )−s dλ
2 −ℓ j s= 21 0
j=0
n     
−ℓ  j+1 j+1
2
1 X 2 −ℓ n 
n
Γ 2 Γ s − 2
= n (−α) 2 −ℓ−j Res  
2 −ℓ j s= 21 2Γ(s)
j=0
 
n
2
−ℓ 
n  (−1) n2 −ℓ− 2j α n2 −ℓ−j Γ j+1
1 X
2 −ℓ 2
= n   ,
2 −ℓ j 2Γ j
+1 Γ 1


j=0 2 2
j even
n


from which the statement follows by observing the relation cℓ (n, V ) = 2 − ℓ Cℓ (n, V ).

18
4 The spectral flow formula and Levinson’s theorem

In this section we return to the spectral flow formula of Equation (3.1) applied to the path
Ht (α) and, using the results of Section 3 we can prove Levinson’s theorem in all dimensions.

Theorem 4.1. Suppose that V ∈ Cc∞ (Rn ) and α > −2ν. Then the spectral flow along the
path Ht (α) is given by
Z ∞
1 1 1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ − βn (V ).


sf(Ht (α)) = (N + Nres ) +
2 4πi 0 2

Proof. Lemma 3.1 and Lemma 3.6 give that

 Z 1 
2 −s

Cs

sf(Ht (α)) = Res Cs Tr V (Id + Ht (α) ) dt + Tr ηs (H(α)) − ηs (H0 (α)) .
s= 21 0 2

Suppose first that n is odd. Applying Proposition 3.4 to the first term on the right-hand side
and Proposition 3.8 to the second term gives
Z ∞
1 1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ.


sf(Ht (α)) = (N + Nres ) +
2 4πi 0

Now we consider n even. Applying again Propositions 3.4 and 3.8 gives
Z ∞
1 1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ


sf(Ht (α)) = (N + Nres ) +
2 4πi 0
 
n−1 n
⌊ 2 ⌋ 2 −ℓ 
n  (−1) n2 −ℓ− 2j α n2 −ℓ−j Γ j+1
1 X X
2 −ℓ 2
+  
Cℓ (n, V )
2 j 2(2πi)Γ 2j + 1 Γ 12
ℓ=1 j=0
j even
 
n n
−ℓ n
−ℓ− 2j −1 n −ℓ−j j+1
2
 n 2 (−1) 2 α 2 Γ
2 −ℓ 2
X X
+   Cℓ (n, V ).
j 4(2πi)Γ j + 1 Γ 1


ℓ=1 j=0 2 2
j even

It remains to observe that for n even we have ⌊ n−1 n

2 ⌋ = 2 − 1 and thus
 
n
2
n
2
−ℓ 
n  (−1) n2 −ℓ− 2j −1 α n2 −ℓ−j Γ j+1
2 −ℓ 2
X X
  Cℓ (n, V )
j 4(2πi)Γ j
+1 Γ 1


ℓ=1 j=0 2 2
j even
 
⌊ n−1 ⌋ n
−ℓ n
−ℓ− 2j n −ℓ−j j+1
1 X 2 2  n  (−1) 2 α 2 Γ
2 −ℓ 2 1
X
=−   Cℓ (n, V ) − βn (V ),
2 j 2(2πi)Γ j + 1 Γ 1

2
ℓ=1 j=0 2 2
j even

where we have used the definition of βn (V ) in Equation (2.7).

We are now able to prove Levinson’s theorem as a consequence of spectral flow along the path
Ht (α).

19
Theorem 4.2 (Levinson’s theorem). Suppose that V ∈ Cc∞ (Rn ). Then the number N of
eigenvalues (counted with multiplicity) of H = H0 + V is given by
Z ∞
1
Tr(S(λ)∗ S ′ (λ)) − pn (λ) dλ − βn (V ) + Nres ,


−N =
2πi 0
where Nres is as defined in Theorem 2.15.

Proof. By construction we know that for α > −2ν we have

sf(Ht (α)) = 0, (4.1)

since there is no spectrum which moves through zero from right to left as the path is traversed
from H0 (α) to H(α). Substituting Equation (4.1) into the result of Theorem 4.1 and solving
for N completes the proof.

References

[1] A. Alexander. Trace formula and Levinson’s theorem as an index pairing in the presence
of resonances, arXiv preprint: https://arxiv.org/abs/2402.15979, 2024.

[2] A. Alexander. Topological Levinson’s theorem via index pairings and spectral flow, PhD
thesis, University of Wollongong, 2024.

[3] A. Alexander, D. T. Nguyen, A. Rennie, S. Richard. Levinson’s theorem for two-

dimensional scattering systems: it was a surprise, it is now topological!, to appear in
Journal of Spectral Theory, arXiv preprint: https://arxiv.org/abs/2311.09650, 2023.

[4] A. Alexander, A. Rennie. Levinson’s theorem as an index pairing, J. Funct. Anal., 286
(5), 2024.

[5] A. Alexander, A. Rennie. The structure of the wave operator in four dimensions in the
presence of resonances, arXiv preprint: https://arxiv.org/abs/2311.16438, 2023.

[6] M. F. Atiyah, V. K. Patodi, I. M. Singer. Spectral asymmetry and Riemannian geometry.

I, Math. Proc. Cambridge Philos. Soc., 77, 1975, 45–69.

[7] M. F. Atiyah, V. K. Patodi, I. M. Singer. Spectral asymmetry and Riemannian geometry.

III, Math. Proc. Cambridge Philos. Soc., 79, 1976, 71–99.

[8] N. Azamov, A. Carey, P. Dodds, F. Sukochev. Operator integrals, spectral shift, and spec-
tral flow, Canad. J. Math., 61 (2), 2009, 241–263.

[9] N. Azamov, A. Carey, F. Sukochev. The spectral shift function and spectral flow, Comm.
Math. Phys., 276 (1), 2007, 51–91.

[10] R. Bañuelos, A. Sá Barreto. On the heat trace of Schrödinger operators, Comm. Partial
Differential Equations, 20 (11-12), 1995, 2153–2164.

[11] M. T. Benameur, A. Carey, J. Phillips, A. Rennie, F. Sukochev, K. Wojciechowski. An

analytic approach to spectral flow in von Neumann algebras, in ‘Analysis, geometry and
topology of elliptic operators’, World Sci. Publ., 2006, 297–352.

20
[12] M. Š. Birman, M. G. Kreı̆n. On the theory of wave operators and scattering operators,
Dokl. Akad. Nauk SSSR, 144, 1962, 475–478.

[13] A. Carey, J. Phillips. Unbounded Fredholm modules and spectral flow, Canad. J. Math. 50
(4), 1998, 673–718.

[14] A. Carey, J. Phillips, A. Rennie, F. Sukochev. The local index formula in semifinite von
Neumann algebras I: Spectral flow, Adv. Math., 202 (2), 2006, 451–516.

[15] A. Carey, D. Potatov, F. Sukochev. Spectral flow is the integral of a one form on the
Banach manifold of self-adjoint Fredholm operators, Adv. Math., 222 (5), 2009, 1809–
1849.

[16] Y. Colin de Verdière. Une formule de traces pour l’opérateur de Schrödinger dans R3 ,
Ann. Sci. École Norm. Sup., 14 (1), 1981, 27–30.

[17] G. B. Folland. How to integrate a polynomial over a sphere, Amer. Math. Monthly, 108
(5), 2001.

[18] L. Guillopé. Une formule de trace pour l’opérateur de Schrödinger,

PhD Thesis, Université Joseph Fourier Grenoble, 1981. Available at
https://www.math.sciences.univ-nantes.fr/∼guillope/LG/these 1981.pdf

[19] A. Jensen, T. Kato. Spectral properties of Schrödinger operators and time-decay of the
wave functions, Duke Math. J., 46 (3), 1979, 583–611.

[20] A. Jensen. Spectral properties of Schrödinger operators and time-decay of the wave func-
tions results in L2 (Rm ), m ≥ 5, Duke Math. J., 47 (1), 1980, 57–80.

[21] A. Jensen. Time-delay in potential scattering theory. Some “geometric” results, Comm.
Math. Phys., 82 (3), 1981, 435–456.

[22] X. Jia, F. Nicoleau, X.P. Wang. A new Levinson’s theorem for potentials with critical
decay, Ann. Henri Poincaré, 13 (1), 2012, 41-84.

[23] J. Kellendonk, S. Richard. Levinson’s theorem for Schrödinger operators with point inter-
action: a topological approach, J. Phys. A, 39 (46), 2006, 14397–14403.

[24] J. Kellendonk, S. Richard. On the structure of the wave operators in one dimensional
potential scattering, Math. Phys. Electron. J., 14, 1-21, 2008.

[25] J. Kellendonk, S. Richard. On the wave operators and Levinson’s theorem for potential
scattering in R3 , Asian-Eur. J. Math., 5 (1), 2012.

[26] M. G. Kreı̆n. On the trace formula in perturbation theory, Mat. Sbornik N.S., 33 /75,
1953, 597–626.

[27] M. Lesch. The uniqueness of the spectral flow on spaces of unbounded self-adjoint Fred-
holm operators, in ‘Spectral geometry of manifolds with boundary and decomposition of
manifolds’, ser. Contemp. Math. 366, Amer. Math. Soc., 2005, 193–224.

[28] N. Levinson. On the uniqueness of the potential in a Schrödinger equation for a given
asymptotic phase, Danske Vid. Selsk. Mat.-Fys. Medd., 25 (9), 1949.

21
[29] R. G. Newton. Noncentral potentials: the generalized Levinson theorem and the structure
of the spectrum, J. Mathematical Phys., 18 (7), 1977, 1348-1357.

[30] J. Phillips. Self-adjoint Fredholm operators and spectral flow, Canad. Math. Bull. 39 (4),
1996, 460–467.

[31] J. Phillips. Spectral flow in type I and II factors — a new approach, in ‘Cyclic cohomology
and noncommutative geometry’, Fields Inst. Commun., 17, 1997, 137–153.

[32] M. Reed, B. Simon. Methods of modern mathematical physics III: Scattering theory, Aca-
demic Press, New York-London, 1979.

[33] S. Richard, R. Tiedra de Aldecoa. New expressions for the wave operators of Schrödinger
operators in R3 , Lett. Math. Phys., 103 (11), 2013, 1207–1221.

[34] S. Richard, R. Tiedra de Aldecoa. Explicit formulas for the Schrödinger wave operators
in R2 , C. R. Math. Acad. Sci. Paris, 351 (5-6), 2013, 209–214.

[35] S. Richard, R. Tiedra de Aldecoa, L. Zhang. Scattering operator and wave operators for
2D Schrödinger operators with threshold obstructions, Complex Anal. Oper. Theory, 15
(6) 2021.

[36] D. R. Yafaev. Mathematical scattering theory: Analytic theory, 158, Mathematical surveys
and monographs, American Mathematical Society, 2010.

22