HAL Id: hal-00431895
https://hal.archives-ouvertes.fr/hal-00431895
Preprint submitted on 13 Nov 2009
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires
Asymptotics near +-m of the spectral shift function for Dirac operators with non-constant magnetic fields
Rafael Tiedra de Aldecoa
To cite this version:
Rafael Tiedra de Aldecoa. Asymptotics near +-m of the spectral shift function for Dirac operators with non-constant magnetic fields. 2009. �hal-00431895�
Asymptotics near±m of the spectral shift function for Dirac operators with non-constant magnetic fields
Rafael Tiedra de Aldecoa
Facultad de Matem´aticas, Pontificia Universidad Cat´olica de Chile, Av. Vicu˜na Mackenna 4860, Santiago, Chile
E-mail: rtiedra@mat.puc.cl
Abstract
We consider a3-dimensional Dirac operatorH0with non-constant magnetic field of constant direction, perturbed by a sign-definite matrix-valued potentialV decaying fast enough at infinity. Then we determine asymptotics, as the energy goes to+mand−m, of the spectral shift function for the pair(H0, H0+V). We obtain, as a by-product, a generalised version of Levinson’s Theorem relating the eigenvalues asymptotics of H0+V near+mand−mto the scattering phase shift for the pair(H0, H0+V).
Contents
1 Introduction 3
2 Unperturbed operator 4
3 Perturbed operator 5
4 Spectral shift function 7
5 Decomposition of the weighted resolvent 8
6 Proof of the main results 12
6.1 The case|λ|< m . . . 13 6.2 The case|λ|> m . . . 18
7 Appendix 22
1 Introduction
It is known [38] that the free Dirac HamiltonianHmacting in the Hilbert spaceH:=L2(R3;C4)is unitarily equivalent to the operatorh(P)⊕ −h(P), whereP :=−i∇andR3 ∋ξ 7→h(ξ) := (ξ2+m2)1/2. For this reason, the set{±m} = h
(∇h)−1({0})
of critical values ofhplays an important role in spectral analysis and scattering theory for Dirac operators. For instance, one cannot prove at±mthe usual limiting absorption principle for operatorsHm+V, even with V a regular perturbation of Hm, by using standard commutator methods. Both the statements and the proofs have to be modified (see e.g. [4, 19]).
In this paper, we provide a new account on the spectral analysis of Dirac operators at the critical values by discussing the behaviour at±mof the spectral shift function associated to sign-definite perturbations of Dirac operators with non-constant magnetic fields. Our work is closely related to [27] where G. D. Raikov treats a similar issue in the case of magnetic Pauli operators. It can also be considered as a complement of [33], where general properties of the spectrum of Dirac operators with variable magnetic fields of constant direction and matrix perturbations are determined. Other related results on the spectrum of3-dimensional magnetic Dirac operators can be found in [2, 3, 5, 7, 11, 13, 15, 16, 17, 20, 23, 34, 36, 37].
Let us describe the content of this paper. We consider a relativistic spin-12particle evolving inR3in pres- ence of a variable magnetic field of constant direction. By virtue of the Maxwell equations, we may assume with no loss of generality that the magnetic field has the form
B(x~ 1, x2, x3) = 0,0, b(x1, x2) . The system is described inHby the Dirac operator
H0:=α1Π1+α2Π2+α3P3+βm,
whereβ ≡ α0, α1, α2, α3 are the usual Dirac-Pauli matrices,m > 0 is the mass of the particle andΠj :=
−i∂j−ajare the generators of the magnetic translations with a vector potential
~a(x1, x2, x3) = a1(x1, x2), a2(x1, x2),0
that satisfiesB=∂1a2−∂2a1. Sincea3= 0, we writeP3=−i∂3instead ofΠ3. We assume that the function b : R2 → Ris continuous (see Section 2 for details), so thatH0, defined onC0∞(R3;C4), can be extended uniquely to a selfadjoint operator inHwith domainD(H0).
Then we consider a bounded positive multiplication operatorV ∈C R3;Bh(C4)
, whereBh(C4)is the set of4×4hermitian matrices, and define the perturbed HamiltonianH± :=H0±V. SinceV is bounded and symmetric, the operatorH±is selfadjoint inHand has domainD(H) =D(H0). We also assume that|V(x)| decays more rapidly than|x|−3as|x| → ∞and that
(H±−z)−3−(H0−z)−3∈S1(H) for each z∈R\ {σ(H0)∪σ(H±)}, (1.1) whereS1(H)denotes the set of trace class operators inH.
Under these assumptions, there exists a unique functionξ(·;H±, H0)∈L1 R; (1 +|λ|)−4dλ
such that the Lifshits-Krein trace formula
Tr
f(H±)−f(H0)
= Z
R
dλ f′(λ)ξ(λ;H±, H0) (1.2) holds for eachf ∈C0∞(R)(see [39, Sec. 8.11]). The functionξ(·;H±, H0)is called the spectral shift function for the pair(H±, H0). It vanishes identically onR\ {σ(H0)∪σ(H±)}, and can be related to the number of eigenvalues ofH± in(−m, m)(see Remark 4.5). Morever, for almost everyλ ∈ σac(H0)the spectral shift function is related to the scattering matrixS(λ;H±, H0)for the pair(H±, H0)by the Birman-Krein formula
detS(λ;H±, H0) = e−2πiξ(λ;H±,H0).
After identification ofξ(·;H±, H0)with some representative of its equivalence class, our results are the following. In Proposition 4.4, we show that there exists a constantζ >0defined in terms ofb(cf. Proposition
2.1) such thatξ(·;H±, H0)is bounded on each compact subset of(−p
m2+ζ,p
m2+ζ)\ {±m} and is continuous on(−p
m2+ζ,p
m2+ζ)\ {±m} ∪σp(H±)
. In Theorem 6.5, we determine the asymptotic behaviour ofξ(λ;H±, H0)asλ→ ±m,|λ|< m, and in Theorem 6.14, we determine the asymptotic behaviour ofξ(λ;H±, H0)as λ → ±m,|λ| > m. In both cases, one hasξ(λ;H±, H0) → ±∞ asλ → ∓m. The divergence of ξ(λ;H±, H0) nearλ = ±mscales as the number of eigenvalues near 0 of certain Berezin- Toeplitz type operators. WhenV admits a power-like or exponential decay at infinity, or when it has a compact support, we give the first term of the asymptotic expansion ofξ(λ,;H±, H0)nearλ =±m(see Proposition 6.10 and Corollary 6.17). In these cases, we show that the limits
εlimց0
ξ m+ε;H−, H0
ξ m−ε;H−, H0 and lim
εց0
ξ −m−ε;H+, H0
ξ −m+ε;H+, H0
exist and are equal to positive constants depending on the decay rate ofV at infinity (see Corollary 6.18 for a precise statement). This can be interpreted as a generalised version of Levinson’s Theorem for the pair(H±, H0) (see [21, 22] for usual versions of Levinson’s Theorem for Dirac operators). The relation between the behaviour of the spectral shift function nearλ= +mand nearλ=−mis explained in Remark 6.15 by using the charge conjugation symmetry.
These results are similar to the results of [27] (where Pauli operators with non-constant magnetic fields are considered) and [12] (where Schr¨odinger operators with constant magnetic field are considered). Part of the interest of this work relies on the fact that we were able to exhibit a non-trivial class of matrix potentialsV satisfying (1.1) even thoughH0 is not a bounded perturbation of the free Dirac operator. We refer to Remark 3.3 and Section 7 for a discussion of this issue.
Let us fix the notations that are used in the paper. The norm and scalar product ofH ≡L2(R3;C4)are denoted byk · kandh ·,· i. The symbol⊗stands for the closed tensor product of Hilbert spaces andSp(H), p ∈ [1,∞], denotes the p-th Schatten-von Neumann class of operators inH(S∞(H)is the set of compact operators inH). We denote byk · kpthe corresponding operator norm. The variablex∈R3is often written as x≡(x⊥, x3), withx⊥ ∈R2andx3 ∈R. The symbolQj,j = 1,2,3, denotes the multiplication operator by xjinH,Q:= (Q1, Q2, Q3), andQ⊥:= (Q1, Q2). Sometimes, when the context is unambiguous, we consider the operatorsQjandPjas operators inL2(R)instead ofHwithout changing the notations. Given a selfadjoint operatorAin a Hilbert spaceG, the symbolEA(·)stands for the spectral measure ofA.
2 Unperturbed operator
Throughout this paper we assume that the componentb:R2→Rof the magnetic fieldB~ ≡(0,0, b)belongs to the class of “admissible” magnetic fields defined in [27, Sec. 2.1]. Namely, we assume thatb=b0+eb, where b0>0is a constant while the functioneb:R2→Ris such that the Poisson equation
∆ϕe=eb
admits a solutionϕe : R2 → R, continuous and bounded together with its derivatives of order up to two. We also defineϕ0(x⊥) := 14b0|x⊥|2for eachx⊥ ∈R2 and setϕ:= ϕ0+ϕ. Then we obtain a vector potentiale
~a≡(a1, a2, a3)∈C1(R2;R3)for the magnetic fieldB~ by putting
a1:=∂1ϕ, a2:=∂2ϕ and a3:= 0.
(changing, if necessary, the gauge, we shall always assume that the vector potential~ais of this form). We refer to [27] for further properties and examples of admissible magnetic fields.
Since the vector potential~abelongs toL∞loc(R2;R3), the magnetic Dirac operator H0=α1Π1+α2Π2+α3P3+βm
satisfies all the properties of [33, Sec. 2.1]. The operatorH0 is essentially selfadjoint onC0∞(R3;C4), with domainD(H0)⊂ H1/2loc(R3;C4), the spectrum ofH0satisfies
σ(H0) =σac(H0) = (−∞,−m]∪[m,∞), (2.1)
and we have the identity
H02=
H⊥−⊗1+1⊗(P32+m2) 0 0 0
0 H+⊥⊗1+1⊗(P32+m2) 0 0
0 0 H−⊥⊗1+1⊗(P32+m2) 0
0 0 0 H⊥+⊗1+1⊗(P32+m2)
(2.2)
with respect to the tensorial decompositionL2(R2)⊗L2(R)ofL2(R3). Here the operatorsH⊥±are the compo- nents of the Pauli operatorH⊥:=H⊥−⊕H⊥+inL2(R2;C2)associated with the vector potential(a1, a2).
We recall from [27, Sec. 2.2] thatdim ker(H⊥−) = ∞, that dim ker(H⊥+) = 0 and that we have the following result.
Proposition 2.1. Letb be an admissible magnetic field with b0 > 0. Then0 = infσ(H⊥)is an isolated eigenvalue of infinite multiplicity. More precisely, we have
dim ker(H⊥) =∞ and (0, ζ)⊂R\σ(H⊥), where
ζ:= 2b0e−2osc(ϕ)e and osc(ϕ) := supe
x⊥∈R2ϕ(xe ⊥)− inf
x⊥∈R2ϕ(xe ⊥).
Finally, since(0, ζ)⊂R\σ(H⊥), we know from [33, Thm. 1.2.(d)] that the limits
εlimց0hQ3i−ν3/2(H0−λ∓iε)−1hQ3i−ν3/2, ν3>1, (2.3) exist for eachλ∈(−p
m2+ζ,p
m2+ζ)\ {±m}(note that we use the usual notationh·i:=p
1 +| · |2).
3 Perturbed operator
We consider now the perturbed operatorsH± = H0 ±V, whereV ≡ {Vjk} is the multiplication operator associated to the following matrix-valued functionV.
Assumption 3.1. The functionV ∈ C R3;Bh(C4)
satisfies for eachx ≡ (x⊥, x3) ∈ R3and eachj, k ∈ {1, . . . ,4}
V(x)≥0 and |Vjk(x)| ≤Const.hx⊥i−ν⊥hx3i−ν3 for someν⊥>2andν3>1. (3.1) The potentialV in Assumption 3.1 is short-range alongx3. So we know from [33, Thm. 1.2] that
(i) σess(H±) =σess(H0) = (−∞,−m]∪[m,∞).
(ii) The point spectrum ofH± in −p
m2+ζ,p
m2+ζ
\ {±m}is composed of eigenvalues of finite multiplicity and with no accumulation point.
(iii) H±has no singular continuous spectrum in −p
m2+ζ,p
m2+ζ
. In particular,H0andH±have a common spectral gap in(−m, m).
Using the formula
(A+λ)−γ = Γ(γ)−1 Z ∞
0
dt tγ−1e−t(A+λ), A:D(A)→ H, A≥0, λ, γ >0,
the diamagnetic inequality [1, Thm. 2.3], and the compactness criterion [9, Thm. 5.7.1], we find that
|Vjk|1/2 P
ℓ≤3Π∗ℓΠℓ+m2−1/4
∈S∞[L2(R3)].
Sincebis bounded this implies that
|H0|−1/2V|H0|−1/2≤ |H0|−1/2 P
j,k≤4|Vjk|
|H0|−1/2∈S∞(H).
So|H0|−1/2V|H0|−1/2 also belongs toS∞(H), sinceS∞(H)is an hereditaryC∗-subalgebra ofB(H)[24, Cor. 3.2.3]. One has in particular
V1/2(|H0|+ 1)−1/2∈S∞(H). (3.2)
The standard criterion [31, Thm. XI.20] shows that
|Vjk|1/2 −∆ +m2−γ
∈Sq[L2(R3)] ifq∈[2,∞)andγq >3/2.
This together with arguments as above implies that
V1/2|H0|−γ ∈Sq(H) ifq≥2is even andγq >3. (3.3) So we have in particular that
V1/2EH0(B)∈S2(H) for any bounded borel setB ⊂R. (3.4) In the sequel we shall need a more restrictive assumption onV. For this, we recall that there exists numbers z∈R\ {σ(H0)∪σ(H±)}sinceH0andH±have a common spectral gap in(−m, m). We also setR0(z) :=
(H0−z)−1andR±(z) := (H±−z)−1forz∈C\σ(H0)andz∈C\σ(H±), respectively.
Assumption 3.2. The functionV ∈C R3;Bh(C4)
satisfies for eachx∈R3and eachj, k∈ {1, . . . ,4} V(x)≥0 and |Vjk(x)| ≤Const.hxi−ν for some constantν >3. (3.5) Furthermore,V is chosen such that
R3±(z)−R30(z)∈S1(H) for eachz∈R\ {σ(H0)∪σ(H±)}. (3.6) Note that (3.5) implies (3.1) if one takesν3∈ (1, ν−2)andν⊥ :=ν−ν3. Note also that the choice of functionλ7→(λ−z)−3in the trace class condition (3.6) has been made for convenience. Many other choices would also guarantee the existence of the spectral shift function for the pair(H±, H0)(see e.g. [39, Sec. 8.11]).
Remark 3.3. Since the operatorH0is not a bounded perturbation of the free Dirac operator, we cannot apply the results of [40, Sec. 4] to prove the inclusion (3.6) under the condition (3.5). In general, one has to impose additional assumptions onV to get the result. For instance, ifV verifies (3.5), and
(i) [V, α1] = [V, α2] = 0,
(ii) for eachx∈R3and eachj, k, ℓ∈ {1, . . . ,4}, one has|(∂ℓVjk)(x)| ≤Const.hxi−ςfor someς >3, (iii) for eachj, k, ℓ∈ {1, . . . ,4}, one has(∂ℓ∂3Vjk)∈L∞(R3),
then (3.6) is satisfied. Furthermore, if V is scalar, then the same is true without assuming (iii) (and (i) is trivially satisfied). The proof of these statements can be found in the appendix. Here, we only note that a matrix V∈Bh(C4)satisfying (i) is necessarily of the form
V=
v1 0 v3 0 0 v2 0 v3 v3 0 v2 0
0 v3 0 v1
! , withv1,v2∈Randv3∈C.
4 Spectral shift function
In this section we recall some results due to A. Pushnitski on the representation of the spectral shift function for a pair of not semibounded selfadjoint operators.
Given a a Lebesgue measurable setB⊂R, we setµ(B) := π1R
B dt
1+t2, and note thatµ(R) = 1. Further- more, ifT =T∗is a compact operator in a separable Hilbert spaceG, we set
n±(s;T) := rankE±T (s,∞)
fors >0.
Then we have the following estimates.
Lemma 4.1 (Lemma 2.1 of [26]). LetT1 = T1∗ ∈ S∞(H)andT2 = T2∗ ∈ S1(H). Then one as for each s1, s2>0 Z
R
dµ(t)n±(s1+s2;T1+tT2)≤n±(s1;T1) + 1
πs2kT2k1. Forz∈C\σ(H0), we define the usual weighted resolvent
T(z) :=V1/2(H0−z)−1V1/2 and the corresponding real and imaginary parts
A(z) :=ReT(z) and B(z) :=ImT(z).
Then the next lemma is direct consequence of the inclusions (3.2)-(3.4) and [25, Prop. 4.4.(i)].
Lemma 4.2. LetV satisfy Assumption 3.1. Then, for almost everyλ∈R, the limitsA(λ+i0) := limεց0A(λ+
iε)andB(λ+i0) := limεց0B(λ+iε)≥0exist inS4(H).
Next theorem follows from the inclusions (3.2), (3.4), (3.6), from the equations (1.9), (8.1), (8.2) of [25], and from Theorem 8.1 of [25].
Theorem 4.3. LetV satisfy Assumption 3.2. Then, for almost everyλ∈R,ξ(λ;H±, H0)exists and is given by ξ(λ;H±, H0) =±
Z
R
dµ(t)n∓ 1;A(λ+i0) +tB(λ+i0)
. (4.1)
We know from (2.3) thatA(λ+i0)andB(λ+i0)exist inB(H)for eachλ∈(−p
m2+ζ,p
m2+ζ)\ {±m}. In Propositions 5.2-5.3 and Corollary 5.5 below we show that in factA(λ+i0)∈S4(H)andB(λ+i0)∈ S1(H)for eachλ∈(−p
m2+ζ,p
m2+ζ)\ {±m}. Hence, by Lemma 4.1, the r.h.s. of (4.1) will turn out to be well-defined for everyλ ∈ (−p
m2+ζ,p
m2+ζ)\ {±m}. In the next proposition we state some regularity properties of the function
(−p
m2+ζ,p
m2+ζ)\ {±m} ∋λ7→ξ(λ;e H±, H0) :=± Z
R
dµ(t)n∓ 1;A(λ+i0) +tB(λ+i0) . The proof (which relies on Propositions 5.2-5.3, Lemma 5.4, Corollary 5.5 and the stability result [14, Thm. 3.12]) is similar to the one of [6, Sec. 4.2.1].
Proposition 4.4. LetV satisfy Assumption 3.1. Thenξ(e ·;H±, H0) is bounded on each compact subset of (−p
m2+ζ,p
m2+ζ)\ {±m}and is continuous on(−p
m2+ζ,p
m2+ζ)\ {±m} ∪σp(H±) . In the sequel, we identify the functionsξ(e ·;H±, H0)andξ(·;H±, H0)since they are equal for almost everyλ∈Rdue to Theorem 4.3 (see [35] for a study where the r.h.s. of (4.1) is directly treated as a definition ofξ(λ;H±, H0)).
Remark 4.5. In the interval(−m, m),H0has no spectrum and the spectrum ofH± is purely discrete. Thus the spectral shift functionξ(·;H±, H0)can be related to the number of eigenvalues of H± as follows: for λ1, λ2∈(−m, m)\σ(H±)withλ1< λ2, we have (see [25, Thm. 9.1])
ξ(λ1;H±, H0)−ξ(λ2;H±, H0) = rankEH± [λ1, λ2) .
5 Decomposition of the weighted resolvent
In this section we decompose the weighted resolvent
T(z) =V1/2(H0−z)V1/2, z∈C\σ(H0),
into a sumT(z) = Tdiv(z) +Tbound(z), whereTdiv(z)(respectivelyTbound(z)) corresponds to the diverging (respectively non-diverging) part ofT(z)as z → ±m. Then we estimate the behaviour, in suitable Schatten norms, of each term asz→ ±m. We refer to [12, Sec. 4] and [27, Sec. 4.2] for similar approaches in the case of the Schr¨odinger and Pauli operators.
Letaanda∗be the closures inL2(R2)of the operators given by
aϕ:= (Π1−iΠ2)ϕ and a∗ϕ:= (Π1+iΠ2)ϕ,
forϕ∈C0∞(R2). Then one has (see [38, Sec. 5.5.2] and [28, Sec. 5]) H0=
m 0 1⊗P3 a⊗1 0 m a∗⊗1−1⊗P3
1⊗P3 a⊗1 −m 0 a∗⊗1−1⊗P3 0 −m
!
, (5.1)
with
ker(a∗) = ker(aa∗) = ker(H⊥−)⊂L2(R2). (5.2) Now, let
P:=
P0 0 0
0 0 0 0 0 0P 0 0 0 0 0
be the orthogonal projection onto the union of the eigenspaces ofH0 corresponding to the valuesλ = ±m.
SinceP ≡p⊗1is the orthogonal projection ontoker(H⊥−)⊗L2(R), the equations (5.1) and (5.2) imply that H0andPcommute:
H0−1P=PH0−1. (5.3)
In fact, by using (2.2) and (5.1), one gets for eachz∈C\σ(H0)the equalities (H0−z)−1P
= (H0+z) H02−z2−1
P
=
p⊗R(z2−m2) (z+m) 00 0 00 00
0 0 (z−m) 0
0 0 0 0
! +
p⊗P3R(z2−m2)0 0 1 0
0 0 0 0 1 0 0 0 0 0 0 0
,
whereR(z) := P32−z−1
,z∈C\[0,∞), is the resolvent ofP32inL2(R). This allows us to decomposeT(z) asT(z) =Tdiv(z) +Tbound(z), with
Tdiv(z) :=V1/2
p⊗R(z2−m2) (z+m) 00 0 00 00
0 0 (z−m) 0
0 0 0 0
! V1/2, Tbound(z) :=V1/2
p⊗P3R(z2−m2)0 0 1 0
0 0 0 0 1 0 0 0 0 0 0 0
V1/2+V1/2(H0−z)−1P⊥V1/2 (P⊥ := 1−P).
One may note that this decomposition ofT(z)differs slightly from the simpler decomposition T(z) =V1/2(H0−z)PV1/2+V1/2(H0−z)P⊥V1/2,
since the first term in Tbound(z)is associated to the projectionPand not the projection P⊥. This choice is motivated by the will of distinguishing clearly the contributionTdiv(z), that diverge as z → ±m, from the contributionTbound(z), that stays bounded asz→ ±m.