Mathematical Theory of the Wigner-Weisskopf Atom

Transcription

1 Mathematical Theory of the Wigner-Weisskopf Atom Vojkan Jaksic, Eugene Kritchevski, Claude-Alain Pillet To cite this version: Vojkan Jaksic, Eugene Kritchevski, Claude-Alain Pillet. Mathematical Theory of the Wigner- Weisskopf Atom. J. Derezinski, H. Siedentop. Large Coulomb Systems, Springer/Lecture Notes in Physics 695, pp , 2006, Lecture Notes in Physics 695, < /b >. <hal > HAL Id: hal Submitted on 30 Oct 2007 HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. 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 publics ou privés.

2 Mathematical Theory of the Wigner-Weisskopf Atom V. Jakšić 1, E. Kritchevski 1, and C.-A. Pillet 2 1 Department of Mathematics and Statistics McGill University 805 Sherbrooke Street West Montreal, QC, H3A 2K6, Canada 2 CPT-CNS, UM 6207 Université de Toulon, B.P F La Garde Cedex, France 1 Introduction Non-perturbative theory Basic facts Aronszajn-Donoghue theorem The spectral theorem Scattering theory Spectral averaging Simon-Wolff theorems Fixing ω Examples Digression: the semi-circle law The perturbative theory The adiating Wigner-Weisskopf atom Perturbation theory of embedded eigenvalue Complex deformations Weak coupling limit Examples Fermionic quantization Basic notions of fermionic quantization Fermionic quantization of the WWA Spectral theory Scattering theory Quantum statistical mechanics of the SEBB model Quasi-free states Non-equilibrium stationary states

3 2 V. Jakšić, E. Kritchevski, and C.-A. Pillet 5.3 Non-equilibrium thermodynamics The effect of eigenvalues Thermodynamics in the non-perturbative regime Properties of the fluxes Examples eferences Introduction In these lectures we shall study an atom, S, described by finitely many energy levels, coupled to a radiation field,, described by another set (typically continuum) of energy levels. More precisely, assume that S and are described, respectively, by the Hilbert spaces h S, h and the Hamiltonians h S, h. Let h = h S h and h 0 = h S h. If v is a self-adjoint operator on h describing the coupling between S and, then the Hamiltonian we shall study is h λ h 0 + λv, where λ is a coupling constant. For reasons of space we shall restrict ourselves here to the case where S is a single energy level, i.e., we shall assume that h S C and that h S ω is the operator of multiplication by a real number ω. The multilevel case will be considered in the continuation of these lecture notes [JP3]. We will keep h and h general and we will assume that the interaction has the form v = w + w, where w : C h is a linear map. With a slight abuse of notation, in the sequel we will drop whenever the meaning is clear within the context. Hence, we will write α for α 0, g for 0 g, etc. If w(1) = f, then w = (1 )f and v = (1 )f + (f )1. In physics literature, a Hamiltonian of the form h λ = h 0 + λ((1 )f + (f )1), (1) with λ is sometimes called Wigner-Weisskopf atom (abbreviated WWA) and we will adopt that name. Operators of the type (1) are also often called Friedrichs Hamiltonians [Fr]. The WWA is a toy model invented to illuminate various aspects of quantum physics; see [AJPP1, AM, Ar, B2, CDG, Da1, Da4, DK, Fr, FGP, He, Maa, Mes, PSS]. Our study of the WWA naturally splits into several parts. Non-perturbative and perturbative spectral analysis are discussed respectively in Sections 2 and 3. The fermionic second quantization of WWA is discussed in Sections 4 and 5. In Section 2 we place no restrictions on h and we obtain qualitative information on the spectrum of h λ which is valid either for all or for Lebesgue a.e. λ. Our analysis is based on the spectral theory of rank one perturbations [Ja, Si1]. The theory discussed in this section naturally applies to the cases where describes a quasi-periodic or a random structure, or the coupling constant λ is large. Quantitative information about the WWA can be obtained only in the perturbative regime and under suitable regularity assumptions. In Section 3.2 we assume that the

4 Mathematical Theory of the Wigner-Weisskopf Atom 3 spectrum of h is purely absolutely continuous, and we study spectral properties of h λ for small, non-zero λ. The main subject of Section 3.2 is the perturbation theory of embedded eigenvalues and related topics (complex resonances, radiative life-time, spectral deformations, weak coupling limit). Although the material covered in this section is very well known, our exposition is not traditional and we hope that the reader will learn something new. The reader may benefit by reading this section in parallel with Complement C III in [CDG]. The second quantizations of the WWA lead to the simplest non-trivial examples of open systems in quantum statistical mechanics. We shall call the fermionic second quantization of the WWA the Simple Electronic Black Box (SEBB) model. The SEBB model in the perturbative regime has been studied in the recent lecture notes [AJPP1]. In Sections 4 and 5 we extend the analysis and results of [AJPP1] to the non-perturbative regime. For additional information about the Electronic Black Box models we refer the reader to [AJPP2]. Assume that h is a real Hilbert space and consider the WWA (1) over the real Hilbert space h. The bosonic second quantization of the wave equation 2 t ψ t +h λ ψ t = 0 (see Section 6.3 in [BSZ] and the lectures [DeB, Der1] in this volume) leads to the so called FC (fully coupled) quantum oscillator model. This model has been extensively discussed in the literature. The well-known references in the mathematics literature are [Ar, Da1, FKM]. For references in the physics literature the reader may consult [Br, LW]. One may use the results of these lecture notes to completely describe spectral theory, scattering theory, and statistical mechanics of the FC quantum oscillator model. For reasons of space we shall not discuss this topic here (see [JP3]). These lecture notes are on a somewhat higher technical level than the recent lecture notes of the first and the third author [AJPP1, Ja, Pi]. The first two sections can be read as a continuation (i.e. the final section) of the lecture notes [Ja]. In these two sections we have assumed that the reader is familiar with elementary aspects of spectral theory and harmonic analysis discussed in [Ja]. Alternatively, all the prerequisites can be found in [Ka, Koo, S1, S2, S3, S4, u]. In Section 2 we have assumed that the reader is familiar with basic results of the rank one perturbation theory [Ja, Si1]. In Sections 4 and 5 we have assumed that the reader is familiar with basic notions of quantum statistical mechanics [B1, B2, BSZ, Ha]. The reader with no previous exposure to open quantum systems would benefit by reading the last two sections in parallel with [AJPP1]. The notation used in these notes is standard except that we denote the spectrum of a self-adjoint operator A by sp(a). The set of eigenvalues, the absolutely continuous, the pure point and the singular continuous spectrum of A are denoted respectively by sp p (A), sp ac (A), sp pp (A), and sp sc (A). The singular spectrum of A is sp sing (A) = sp pp (A) sp sc (A). The spectral subspaces associated to the absolutely continuous, the pure point, and the singular continuous spectrum of A are denoted by h ac (A), h pp (A), and h sc (A). The projections on these spectral subspaces are denoted by 1 ac (A), 1 pp (A), and 1 sc (A). Acknowledgment. These notes are based on the lectures the first author gave in the Summer School Large Coulomb Systems QED, held in Nordfjordeid, August

5 4 V. Jakšić, E. Kritchevski, and C.-A. Pillet V.J. is grateful to Jan Dereziński and Heinz Siedentop for the invitation to speak and for their hospitality. The research of V.J. was partly supported by NSEC. The research of E.K. was supported by an FCA scholarship. We are grateful to S. De Bièvre and J. Dereziński for enlightening remarks on an earlier version of these lecture notes. 2 Non-perturbative theory Let ν be a positive Borel measure on. We denote by ν ac, ν pp, and ν sc the absolutely continuous, the pure point and the singular continuous part of ν w.r.t. the Lebesgue measure. The singular part of ν is ν sing = ν pp + ν sc. We adopt the definition of a complex Borel measure given in [Ja, u]. In particular, any complex Borel measure on is finite. Let ν be a complex Borel measure or a positive measure such that dν(t) 1 + t <. The Borel transform of ν is the analytic function dν(t) F ν (z) t z, z C \. Let ν be a complex Borel measure or a positive measure such that dν(t) <. (2) 1 + t2 The Poisson transform of ν is the harmonic function dν(t) P ν (x,y) y (x t) 2 + y 2, x + iy C +, where C ± {z C ± Im z > 0}. The Borel transform of a positive Borel measure is a Herglotz function, i.e., an analytic function on C + with positive imaginary part. In this case P ν (x,y) = Im F ν (x + iy), is a positive harmonic function. The G-function of ν is defined by dν(t) G ν (x) (x t) 2 = lim P ν (x,y), x. y 0 y We remark that G ν is an everywhere defined function on with values in [0, ]. Note also that if G ν (x) <, then lim y 0 ImF ν (x + iy) = 0. If h(z) is analytic in the half-plane C ±, we set

6 Mathematical Theory of the Wigner-Weisskopf Atom 5 h(x ± i0) lim y 0 h(x ± iy), whenever the limit exist. In these lecture notes we will use a number of standard results concerning the boundary values F ν (x ± i0). The proofs of these results can be found in [Ja] or in any book on harmonic analysis. We note in particular that F ν (x ± i0) exist and is finite for Lebesgue a.e. x. If ν is real-valued and nonvanishing, then for any a C the sets {x F ν (x ±i0) = a} have zero Lebesgue measure. Let ν be a positive Borel measure. For later reference, we describe some elementary properties of its Borel transform. First, the Cauchy-Schwartz inequality yields that for y > 0 ν()im F ν (x + iy) y F ν (x + iy) 2. (3) The dominated convergence theorem yields lim y Im F ν(iy) = lim y F ν(iy) = ν(). (4) y y Assume in addition that ν() = 1. The monotone convergence theorem yields y 4 = lim y 2 = lim y lim ( y y2 y Im F ν (iy) y 2 F ν (iy) 2) ( 1 t 2 + y s 2 + y 2 2 (t iy)(s + iy) 1 y 2 y 2 2 t 2 + y 2 s 2 + y 2 (t s)2 dν(t)dν(s) = 1 (t s) 2 dν(t)dν(s). 2 If ν has finite second moment, t2 dν(t) <, then ) dν(t) dν(s) ( 2 1 (t s) 2 dν(t)dν(s) = t 2 dν(t) tdν(t)). (5) 2 If t2 dν(t) =, then it is easy to see that the both sides in (5) are also infinite. Combining this with Equ. (4) we obtain y Im F ν (iy) y 2 F ν (iy) 2 ( 2 lim y F ν (iy) 2 = t 2 dν(t) tdν(t)), (6) where the right hand side is defined to be whenever t2 dν(t) =. In the sequel B denotes the Lebesgue measure of a Borel set B and δ y the delta-measure at y.

7 6 V. Jakšić, E. Kritchevski, and C.-A. Pillet 2.1 Basic facts Let h,f h be the cyclic space generated by h and f. We recall that h,f is the closure of the linear span of the set of vectors {(h z) 1 f z C \ }. Since (C h,f ) is invariant under h λ for all λ and h λ (C h,f ) = h (C h,f ), in this section without loss of generality we may assume that h,f = h, namely that f is a cyclic vector for h. We denote by µ the spectral measure for h and f. By the spectral theorem, w.l.o.g. we may assume that h = L 2 (,dµ ), that h x is the operator of multiplication by the variable x, and that f(x) = 1 for all x. We will write F for F µ, etc. As we shall see, in the non-perturbative theory of the WWA it is very natural to consider the Hamiltonian (1) as an operator-valued function of two real parameters λ and ω. Hence, in this section we will write h λ,ω h 0 + λv = ω x + λ ((f )1 + (1 )f). We start with some basic formulas. The relation yields that A 1 B 1 = A 1 (B A)B 1, (h λ,ω z) 1 1 = (ω z) 1 1 λ(ω z) 1 (h λ,ω z) 1 f, (h λ,ω z) 1 f = (h z) 1 f λ(f (h z) 1 f)(h λ,ω z) 1 1. (7) It follows that the cyclic subspace generated by h λ,ω and the vectors 1,f, is independent of λ and equal to h, and that for λ 0, 1 is a cyclic vector for h λ,ω. We denote by µ λ,ω the spectral measure for h λ,ω and 1. The measure µ λ,ω contains full spectral information about h λ,ω for λ 0. We also denote by F λ,ω and G λ,ω the Borel transform and the G-function of µ λ,ω. The formulas (7) yield F λ,ω (z) = 1 ω z λ 2 F (z). (8) Since F λ,ω = F λ,ω, the operators h λ,ω and h λ,ω are unitarily equivalent. According to the decomposition h = h S h we can write the resolvent r λ,ω (z) (h λ,ω z) 1 in matrix form λ,ω(z) rλ,ω S(z) r λ,ω (z) = rss rλ,ω S (z) r λ,ω (z).

8 Mathematical Theory of the Wigner-Weisskopf Atom 7 A simple calculation leads to the following formulas for its matrix elements r SS λ,ω (z) = F λ,ω(z), r S r S r λ,ω (z) = λf λ,ω(z)1(f (h z) 1 ), λ,ω (z) = λf λ,ω(z)(h z) 1 f(1 ), λ,ω (z) = (h z) 1 + λ 2 F λ,ω (z)(h z) 1 f(f (h z) 1 ). (9) Note that for λ 0, F λ,ω (z) = F λ,0 (z) 1 + ωf λ,0 (z). This formula should not come as a surprise. For fixed λ 0, h λ,ω = h λ,0 + ω(1 )1, and since 1 is a cyclic vector for h λ,ω, we are in the usual framework of the rank one perturbation theory with ω as the perturbation parameter! This observation will allow us to naturally embed the spectral theory of h λ,ω into the spectral theory of rank one perturbations. By taking the imaginary part of elation (8) we can relate the G-functions of µ and µ λ,ω as 1 + λ 2 G (x) G λ,ω (x) = ω x λ 2 F (x + i0) 2, (10) whenever the boundary value F (x + i0) exists and the numerator and denominator of the right hand side are not both infinite. It is important to note that, subject to a natural restriction, every rank one spectral problem can be put into the form h λ,ω for a fixed λ 0. Proposition 1. Let ν be a Borel probability measure on and λ 0. Then the following statements are equivalent: 1. There exists a Borel probability measure µ on such that the corresponding µ λ,0 is equal to ν. 2. tdν(t) = 0 and t2 dν(t) = λ 2. Proof. (1) (2) Assume that µ exists. Then h λ,0 1 = λf and hence tdν(t) = (1,h λ,0 1) = 0, and t 2 dν(t) = h λ,0 1 2 = λ 2. (2) (1) We need to find a probability measure µ such that ( F (z) = λ 2 z 1 ), (11) F ν (z)

9 8 V. Jakšić, E. Kritchevski, and C.-A. Pillet for all z C +. Set H ν (z) z 1 F ν (z). Equ. (3) yields that C + z λ 2 Im H ν (z) is a non-negative harmonic function. Hence, by well-known results in harmonic analysis (see e.g. [Ja, Koo]), there exists a Borel measure µ which satisfies (2) and a constant C 0 such that λ 2 Im H ν (x + iy) = P (x,y) + Cy, (12) for all x + iy C +. The dominated convergence theorem and (2) yield that P (0,y) dµ (t) lim = lim y y y t 2 + y 2 = 0. Note that y Im H ν (iy) = y Im F ν(iy) y 2 F ν (iy) 2 F ν (iy) 2, (13) and so Equ. (6) yields Im H ν (iy) lim = 0. y y Hence, (12) yields that C = 0 and that F (z) = λ 2 H ν (z) + C 1, (14) where C 1 is a real constant. From Equ. (4), (13) and (6) we get µ () = lim y y ImF (iy) = λ 2 lim y y Im H ν(iy) ( = λ 2 t 2 dν(t) and so µ is probability measure. Since ( e H ν (iy) = e F ν(iy) F ν (iy) 2, ) ) 2 tdν(t) Equ. (14), (4) and the dominated convergence theorem yield that λ 2 C 1 = lim y e H ν(iy) Hence, C 1 = 0 and Equ. (11) holds. = lim y y2 e F ν (iy) ty 2 = lim y t 2 + y 2 dν(t) = tdν(t) = 0. = 1,

10 Mathematical Theory of the Wigner-Weisskopf Atom Aronszajn-Donoghue theorem For λ 0 define T λ,ω {x G (x) <, x ω + λ 2 F (x + i0) = 0}, S λ,ω {x G (x) =, x ω + λ 2 F (x + i0) = 0}, (15) L {x Im F (x + i0) > 0}. Since the analytic function C + z z ω + λ 2 F (z) is non-constant and has a positive imaginary part, by a well known result in harmonic analysis T λ,ω = S λ,ω = 0. Equ. (8) implies that, for ω 0, x ω+λ 2 F (x+i0) = 0 is equivalent to F λ,0 (x + i0) = ω 1. Moreover, if one of these conditions is satisfied, then Equ. (10) yields ω 2 G λ,0 (x) = 1 + λ 2 G (x). Therefore, if ω 0, then T λ,ω = {x G λ,0 (x) <, F λ,0 (x + i0) = ω 1 }, S λ,ω = {x G λ,0 (x) =, F λ,0 (x + i0) = ω 1 }. The well-known Aronszajn-Donoghue theorem in spectral theory of rank one perturbations (see [Ja, Si1]) translates to the following result concerning the WWA. Theorem T λ,ω is the set of eigenvalues of h λ,ω. Moreover, µ λ,ω pp = x T λ,ω λ 2 G (x) δ x. (16) If ω 0, then also µ λ,ω pp = x T λ,ω 1 ω 2 G λ,0 (x) δ x. 2. ω is not an eigenvalue of h λ,ω for all λ µ λ,ω sc is concentrated on S λ,ω. 4. For all λ,ω, the set L is an essential support of the absolutely continuous spectrum of h λ,ω. Moreover sp ac (h λ,ω ) = sp ac (h ) and dµ λ,ω ac (x) = 1 π Im F λ,ω(x + i0)dx. 5. For a given ω, {µ λ,ω sing λ > 0} is a family of mutually singular measures. 6. For a given λ 0, {µ λ,ω sing ω 0} is a family of mutually singular measures.

11 10 V. Jakšić, E. Kritchevski, and C.-A. Pillet 2.3 The spectral theorem In this subsection λ 0 and ω are given real numbers. By the spectral theorem, there exists a unique unitary operator U λ,ω : h L 2 (,dµ λ,ω ), (17) such that U λ,ω h λ,ω (U λ,ω ) 1 is the operator of multiplication by x on the Hilbert space L 2 (,dµ λ,ω ) and U λ,ω 1 = 1l, where 1l(x) = 1 for all x. Moreover, where U λ,ω ac U λ,ω pp U λ,ω sc U λ,ω = U λ,ω ac U λ,ω pp U λ,ω sc, : h ac (h λ,ω ) L 2 (,dµ λ,ω ac ), : h pp (h λ,ω ) L 2 (,dµ λ,ω pp ), : h sc (h λ,ω ) L 2 (,dµ λ,ω sc ), are unitary. In this subsection we will describe these unitary operators. We shall make repeated use of the following fact. Let µ be a positive Borel measure on. For any complex Borel measure ν on denote by ν = ν ac + ν sing the Lebesgue decomposition of ν into absolutely continuous and singular parts w.r.t. µ. The adon- Nikodym derivative of ν ac w.r.t. µ is given by P ν (x,y) lim y 0 P µ (x,y) = dν ac dµ (x), for µ-almost every x (see [Ja]). In particular, if µ is Lebesgue measure, then for Lebesgue a.e. x. By Equ. (8), and so (18) yields that lim P ν (x,y) = π dν ac (x), (18) y 0 dx Im F λ,ω (x + i0) = λ 2 F λ,ω (x + i0) 2 Im F (x + i0), (19) dµ λ,ω ac dx = λ2 F λ,ω (x + i0) 2 dµ,ac dx. In particular, since F λ,ω (x + i0) 0 for Lebesgue a.e. x, µ,ac and µ λ,ω ac are equivalent measures. Let φ = α ϕ h and M(z) 1 [ ] (1 (h λ,ω z) 1 φ) (1 (h λ,ω z) 1 φ), z C +. 2i The formulas (7) and (9) yield that

12 and so Mathematical Theory of the Wigner-Weisskopf Atom 11 ( ) (1 (h λ,ω z) 1 φ) = F λ,ω (z) α λ(f (h z) 1 ϕ), (20) ( ) M(z) = Im F λ,ω (z) α λ(f (h z) 1 ϕ) ( ) λf λ,ω (z) y (f ((h x) 2 + y 2 ) 1 ϕ) ( ) = Im F λ,ω (z) α λ(f (h z) 1 ϕ) f(t)ϕ(t) λf λ,ω (z)y (t x) 2 + y 2 dµ (t). This relation and (18) yield that for µ,ac -a.e. x, ( ) M(x + i0) = ImF λ,ω (x + i0) α λ(f (h x i0) 1 ϕ) λf λ,ω (x i0)f(x)ϕ(x)π dµ ac dx (x) ( ) = ImF λ,ω (x + i0) α λ(f (h x i0) 1 ϕ) (21) λf λ,ω (x i0)f(x)ϕ(x)im F (x + i0). On the other hand, computing M(z) in the spectral representation (17) we get (U λ,ω φ)(t) M(z) = y (t x) 2 + y 2 dµλ,ω (t). This relation and (18) yield that for µ λ,ω ac -a.e. x, M(x + i0) = (U λ,ω ac φ)(x)π dµλ,ω ac dx (x) = (Uλ,ω ac φ)(x)im F λ,ω (x + i0). Since µ,ac and µ λ,ω ac are equivalent measures, comparison with the expression (21) and use of Equ. (8) yield Proposition 2. Let φ = α ϕ h. Then (Uac λ,ω φ)(x) = α λ(f (h x i0) 1 ϕ) f(x)ϕ(x) λf λ,ω (x + i0). We now turn to the pure point part U λ,ω pp. ecall that T λ,ω is the set of eigenvalues of h λ,ω. Using the spectral representation (17), it is easy to prove that for x T λ,ω (1 (h λ,ω x iy) 1 φ) lim y 0 (1 (h λ,ω x iy) 1 1) = lim F (U λ,ω φ)µ λ,ω(x + iy) = (U λ,ω φ)(x). (22) y 0 F λ,ω (x + iy)

13 12 V. Jakšić, E. Kritchevski, and C.-A. Pillet The relations (20) and (22) yield that for x T λ,ω the limit H ϕ (x + i0) lim y 0 (f (h x iy) 1 ϕ), (23) exists and that (U λ,ω φ)(x) = α λh ϕ (x + i0). Hence, we have: Proposition 3. Let φ = α ϕ h. Then for x T λ,ω, (U λ,ω pp φ)(x) = α λh ϕ (x + i0). The assumption x T λ,ω makes the proof of (22) easy. However, this formula holds in a much stronger form. It is a deep result of Poltoratskii [Po] (see also [Ja, JL]) that (1 (h λ,ω x iy) 1 φ) lim y 0 (1 (h λ,ω x iy) 1 1) = (Uλ,ω φ)(x) for µ λ,ω sing a.e. x. (24) Hence, the limit (23) exists and is finite for µ λ,ω sing-a.e. x. Thus, we have: Proposition 4. Let φ = α ϕ h. Then, where U λ,ω sing = Uλ,ω pp Usc λ,ω. 2.4 Scattering theory (U λ,ω sing φ)(x) = α λh ϕ(x + i0), ecall that h is the operator of multiplication by the variable x on the space L 2 (,dµ ). U λ,ω h λ,ω (U λ,ω ) 1 is the operator of multiplication by x on the space L 2 (,dµ λ,ω ). Set h,ac h hac(h ), h λ,ω,ac h λ,ω hac(h λ,ω ). Since h ac (h ) = L 2 (,dµ,ac ), h ac (h λ,ω ) = (U λ,ω ac ) 1 L 2 (,dµ λ,ω ac ), and the measures µ,ac and µ λ,ω ac are equivalent, the operators h,ac and h λ,ω,ac are unitarily equivalent. Using (18) and the chain rule one easily checks that the operator (W λ,ω φ)(x) = dµ λ,ω ac dµ,ac (x) (U λ,ω ac φ)(x) = Im F λ,ω (x + i0) ImF (x + i0) (Uλ,ω ac φ)(x), is an explicit unitary which takes h ac (h λ,ω ) onto h ac (h ) and satisfies W λ,ω h λ,ω,ac = h,ac W λ,ω.

14 Mathematical Theory of the Wigner-Weisskopf Atom 13 There are however many unitaries U : h ac (h λ,ω ) h ac (h ) with the intertwining property Uh λ,ω,ac = h,ac U. (25) Indeed, let ϕ h ac (h ) be a normalized cyclic vector for h,ac. Then there is a unique unitary U such that (25) holds and U1 ac (h λ,ω )1 = cϕ, where c = 1 ac (h λ,ω )1 is a normalization constant. On the other hand, any unitary with the property (25) is uniquely specified by its action on the vector 1 ac (h λ,ω )1. Note that (W λ,ω Im F λ,ω (x + i0) 1 ac (h λ )1)(x) = ImF (x + i0). In this subsection we describe a particular pair of unitaries, called wave operators, which satisfy (25). Theorem The strong limits exist and an U ± λ,ω = h ac(h λ,ω ). 2. The strong limits U ± λ,ω s lim t ± eith λ,ω e ith0 1 ac (h 0 ), (26) Ω ± λ,ω s lim t ± eith0 e ith λ,ω 1 ac (h λ,ω ), (27) exist and an Ω ± λ,ω = h ac(h 0 ). 3. The maps U ± λ,ω : h ac(h 0 ) h ac (h λ,ω ) and Ω ± λ,ω : h ac(h λ,ω ) h ac (h 0 ) are unitary. U ± λ,ω Ω± λ,ω = 1 ac(h λ,ω ) and Ω ± λ,ω U ± λ,ω = 1 ac(h 0 ). Moreover, Ω ± λ,ω satisfies the intertwining relation (25). 4. The S-matrix S Ω + λ,ω U λ,ω is unitary on h ac(h 0 ) and commutes with h 0,ac. This theorem is a basic result in scattering theory. The detailed proof can be found in [Ka, S3]. The wave operators and the S-matrix can be described as follows. Proposition 5. Let φ = α ϕ h. Then (Ω ± λ,ω φ)(x) = ϕ(x) λf(x)f λ,ω(x ± i0)(α λ(f (h x i0) 1 ϕ)). (28) Moreover, for any ψ h ac (h 0 ) one has (Sψ)(x) = S(x)ψ(x) with S(x) = 1 + 2πiλ 2 F λ,ω (x + i0) f(x) 2 dµ,ac (x). (29) dx

15 14 V. Jakšić, E. Kritchevski, and C.-A. Pillet emark. The assumption that f is a cyclic vector for h is not needed in Theorem 2 and Proposition 5. Proof. We deal only with Ω + λ,ω. The case of Ω λ,ω is completely similar. The formula for U ± λ,ω follows from the formula for Ω± λ,ω by duality (use Theorem 2). Given these formulas, it is easy to compute the S-matrix. Let ψ h ac (h 0 ) = h ac (h ). We start with the identity (ψ e ith0 e ith λ,ω φ) = (ψ φ) iλ t Note that (ψ φ) = (ψ ϕ), (ψ e ish0 f) = (ψ e ish f), and that Hence, by the Abel theorem, 0 (ψ e ish0 f)(1 e ish λ,ω φ)ds. (30) lim e ith t (ψ eith0 λ,ω φ) = lim (e ith λ,ω e ith0 ψ φ) t = (U + λ,ω ψ φ) = (ψ Ω + λ,ω φ). (ψ Ω + λ,ωφ) = (ψ ϕ) lim iλl(y), (31) y 0 where L(y) = 0 e ys (ψ e ish0 f)(1 e ish λ,ω φ)ds. Now, L(y) = e ys (ψ e ish0 f)(1 e ish λ,ω φ)ds 0 [ ] = ψ(x)f(x) (1 e is(x+iy hλ,ω) φ)ds dµ,ac (x) 0 = i ψ(x)f(x)(1 (h λ,ω x iy) 1 φ)dµ,ac (x) = i ψ(x)f(x)g y (x)dµ,ac (x), where g y (x) (1 (h λ,ω x iy) 1 φ). ecall that for Lebesgue a.e. x, (32) g y (x) g(x) (1 (h λ,ω x i0) 1 φ), (33) as y 0. By the Egoroff theorem (see e.g. Problem 16 in Chapter 3 of [u], or any book on measure theory), for any n > 0 there exists a measurable set n such that \ n < 1/n and g y g uniformly on n. The set {ψ L 2 (,dµ,ac ) supp ψ n }, n>0

16 Mathematical Theory of the Wigner-Weisskopf Atom 15 is clearly dense in h ac (h ). For any ψ in this set the uniform convergence g y g on suppψ implies that there exists a constant C ψ such that ψf(g y g) C ψ ψf L 1 (,dµ,ac ). This estimate and the dominated convergence theorem yield that ψ f(g y g)dµ,ac = 0. lim y 0 On the other hand, Equ. (31) and (32) yield that the limit ψ fg y dµ, exists, and so the relation (ψ Ω + λ,ω φ) = (ψ ϕ) λ lim y 0 holds for a dense set of vectors ψ. Hence, ψ(x)f(x)(1 (h λ,ω x i0) 1 φ)dµ,ac (x), (Ω + λ,ω φ)(x) = ϕ(x) λf(x)(1 (h λ,ω x i0) 1 φ), and the formula (20) completes the proof. 2.5 Spectral averaging We will freely use the standard measurability results concerning the measure-valued function (λ,ω) µ λ,ω. The reader not familiar with these facts may consult [CFKS, CL, Ja]. Let λ 0 and µ λ (B) = µ λ,ω (B)dω, where B is a Borel set. Obviously, µ λ is a Borel measure on. The following (somewhat surprising) result is often called spectral averaging. Proposition 6. The measure µ λ is equal to the Lebesgue measure and for all g L 1 (,dx), [ ] g(x)dx = g(x)dµ λ,ω (x) dω. The proof of this proposition is elementary and can be found in [Ja, Si1]. One can also average with respect to both parameters. It follows from Proposition 6 that the averaged measure µ(b) = 1 π is also equal to the Lebesgue measure. µ λ,ω (B) 1 + λ 2 dλdω, 2

17 16 V. Jakšić, E. Kritchevski, and C.-A. Pillet 2.6 Simon-Wolff theorems ecall that x + λ 2 F (x + i0) and F λ,0 (x + i0) are finite and non-vanishing for Lebesgue a.e. x. For λ 0, Equ. (10) gives that for Lebesgue a.e x, G λ,0 (x) = These observations yield: 1 + λ 2 G (x) x + λ 2 F (x + i0) 2 = F λ,0(x + i0) 2 (1 + λ 2 G (x)). Lemma 1. Let B be a Borel set and λ 0. Then G (x) < for Lebesgue a.e. x B iff G λ,0 (x) < for Lebesgue a.e. x B. This lemma and the Simon-Wolff theorems in rank one perturbation theory (see [Ja, Si1, SW]) yield: Theorem 3. Let B be a Borel set. Then the following statements are equivalent: 1. G (x) < for Lebesgue a.e. x B. 2. For all λ 0, µ λ,ω cont(b) = 0 for Lebesgue a.e. ω. In particular, µ λ,ω 0 for Lebesgue a.e. (λ,ω) 2. cont(b) = Theorem 4. Let B be a Borel set. Then the following statements are equivalent: 1. ImF (x + i0) = 0 and G (x) = for Lebesgue a.e. x B. 2. For all λ 0, µ λ,ω ac (B) + µ λ,ω pp (B) = 0 for Lebesgue a.e. ω. In particular, µ λ,ω ac (B) + µ λ,ω pp (B) = 0 for Lebesgue a.e. (λ,ω) 2. Theorem 5. Let B be a Borel set. Then the following statements are equivalent: 1. ImF (x + i0) > 0 for Lebesgue a.e. x B. 2. For all λ 0, µ λ,ω sing (B) = 0 for Lebesgue a.e. ω. In particular, µλ,ω sing (B) = 0 for Lebesgue a.e. (λ,ω) 2. Note that while the Simon-Wolff theorems hold for a fixed λ and for a.e. ω, we cannot claim that they hold for a fixed ω and for a.e. λ from Fubini s theorem we can deduce only that for a.e. ω the results hold for a.e. λ. This is somewhat annoying since in many applications for physical reasons it is natural to fix ω and vary λ. The next subsection deals with this issue. 2.7 Fixing ω The results discussed in this subsection are not an immediate consequence of the standard results of rank one perturbation theory and for this reason we will provide complete proofs. In this subsection ω is a fixed real number. Let µ ω (B) = µ λ,ω (B)dλ,

18 Mathematical Theory of the Wigner-Weisskopf Atom 17 where B is a Borel set. Obviously, µ ω is a positive Borel measure on and for all Borel measurable g 0, [ ] g(t)dµ ω (t) = g(t)dµ λ,ω (t) dλ, where both sides are allowed to be infinite. We will study the measure µ ω by examining the boundary behavior of its Poisson transform P ω (x,y) as y 0. In this subsection we set Lemma 2. For z C +, h(z) (ω z)f (z). P ω (z) = π h(z) + e h(z). 2 h(z) Proof. We start with [ ] y P ω (x,y) = (t x) 2 + y 2 dµλ,ω (t) dλ = Im F λ,ω (x + iy)dλ. Equ. (8) and a simple residue calculation yield πi F λ,ω (x + iy)dλ =, F (z) ω z F (z) where the branch of the square root is chosen to be in C +. An elementary calculation shows that iπ P ω (x,y) = Im, h(x + iy) where the branch of the square root is chosen to have positive real part, explicitly w 1 2 ( w + e w + isign(imw) w e w ). (34) This yields the statement. Theorem 6. The measure µ ω is absolutely continuous with respect to Lebesgue measure and dµ ω h(x + i0) + e h(x + i0) dx (x) =. (35) 2 h(x + i0) The set E {x Im F (x + i0) > 0} {x (ω x)f (x + i0) > 0}, is an essential support for µ ω and µ λ,ω is concentrated on E for all λ 0.

19 18 V. Jakšić, E. Kritchevski, and C.-A. Pillet Proof. By Theorem 1, ω is not an eigenvalue of h λ,ω for λ 0. This implies that µ ω ({ω}) = 0. By the theorem of de la Vallée Poussin (for detailed proof see e.g. [Ja]), µ ω sing is concentrated on the set By Lemma 2, this set is contained in {x x ω and lim y 0 P ω (x + iy) = }. S {x lim y 0 F (x + iy) = 0}. Since S S λ,ω {ω}, Theorem 1 implies that µ λ,ω sing (S) = 0 for all λ 0. Since S = 0, µ λ,ω ac (S) = 0 for all λ. We conclude that µ λ,ω (S) = 0 for all λ 0, and so µ ω (S) = µ λ,ω (S)dλ = 0. Hence, µ ω sing = 0. From Theorem 1 we now get dµ ω (x) = dµ ω ac(x) = 1 π ImF ω(x + i0)dx, and (35) follows from Lemma 2. The remaining statements are obvious. We are now ready to state and prove the Simon-Wolff theorems for fixed ω. Theorem 7. Let B be a Borel set. Consider the following statements: 1. G (x) < for Lebesgue a.e. x B. 2. µ λ,ω cont(b) = 0 for Lebesgue a.e. λ. Then (1) (2). If B E, then also (2) (1). Proof. Let A {x B G (x) = } E. (1) (2) By assumption, A has zero Lebesgue measure. Theorem 6 yields that µ ω (A) = 0. Since G (x) < for Lebesgue a.e. x B, Im F (x + i0) = 0 for Lebesgue a.e. x B. Hence, for all λ, ImF λ,ω (x + i0) = 0 for Lebesgue a.e. x B. By Theorem 1, µ λ,ω ac (B) = 0 and the measure µ λ,ω sc B is concentrated on the set A for all λ 0. Then, µ λ,ω sc (B)dλ = µ λ,ω sc (A)dλ µ λ,ω (A)dλ = µ ω (A) = 0. Hence, µ λ,ω sc (B) = 0 for Lebesgue a.e λ. (2) (1) Assume that the set A has positive Lebesgue measure. By Theorem 1, µ λ,ω pp (A) = 0 for all λ 0, and µ λ,ω cont(a)dλ = µ λ,ω (A)dλ = µ ω (A) > 0. Hence, for a set of λ s of positive Lebesgue measure, µ λ,ω cont(b) > 0.

20 Mathematical Theory of the Wigner-Weisskopf Atom 19 Theorem 8. Let B be a Borel set. Consider the following statements: 1. ImF (x + i0) = 0 and G (x) = for Lebesgue a.e. x B. 2. µ λ,ω ac (B) + µ λ,ω pp (B) = 0 for Lebesgue a.e. λ. Then (1) (2). If B E, then also (2) (1). Proof. Let A {x B G (x) < } E. (1) (2) Since ImF (x + i0) = 0 for Lebesgue a.e. x B, Theorem 1 implies that µ λ,ω ac (B) = 0 for all λ. By Theorems 1 and 6, for λ 0, µ λ,ω pp B is concentrated on the set A. Since A has Lebesgue measure zero, µ λ,ω pp (A)dλ µ ω (A) = 0, and so µ λ,ω pp (B) = 0 for Lebesgue a.e. λ. (2) (1) If Im F (x+i0) > 0 for a set of x B of positive Lebesgue measure, then, by Theorem 1, µ λ,ω ac (B) > 0 for all λ. Assume that Im F (x+i0) = 0 for Lebesgue a.e. x B and that A has positive Lebesgue measure. By Theorem 1, µ λ,ω cont(a) = 0 for all λ 0 and since A E, Theorem 6 implies µ λ,ω pp (A)dλ = µ λ,ω (A)dλ = µ ω (A) > 0. Thus, we must have that µ λ,ω pp (B) > 0 for a set of λ s of positive Lebesgue measure. Theorem 9. Let B be a Borel set. Consider the following statements: 1. ImF (x + i0) > 0 for Lebesgue a.e. x B. 2. µ λ,ω sing (B) = 0 for Lebesgue a.e. λ. Then (1) (2). If B E, then also (2) (1). Proof. (1) (2) By Theorem 1, for λ 0 the measure µ λ,ω sing B is concentrated on the set A {x B Im F (x + i0) = 0} E. By assumption, A has Lebesgue measure zero and µ λ,ω sing (A)dλ µ λ,ω (A)dλ = µ ω (A) = 0. Hence, for Lebesgue a.e. λ, µ λ sing (B) = 0. (2) (1) Assume that B E and that the set A {x B Im F (x + i0) = 0}, has positive Lebesgue measure. By Theorem 1, µ λ,ω ac (A) = 0 for all λ, and µ λ,ω sing (A)dλ = µ λ,ω (A)dλ = µ ω (A) > 0. Hence, for a set of λ s of positive Lebesgue measure, µ λ,ω sing (B) > 0.

21 20 V. Jakšić, E. Kritchevski, and C.-A. Pillet 2.8 Examples In all examples in this subsection h = L 2 ([a,b],dµ ) and h is the operator of multiplication by x. In Examples 1-9 [a,b] = [0,1]. In Examples 1 and 2 we do not assume that f is a cyclic vector for h. Example 1. In this example we deal with the spectrum outside ]0,1[. Let Λ 0 = 1 0 f(x) 2 dµ (x), Λ 1 = x 1 0 f(x) 2 x 1 dµ (x). Obviously, Λ 0 ]0, ] and Λ 1 [,0[. If λ 2 > ω/λ 0, then h λ,ω has a unique eigenvalue e < 0 which satisfies ω e λ f(x) 2 x e dµ (x) = 0. (36) If λ 2 < ω/λ 0, then h λ,ω has no eigenvalue in ],0[. 0 is an eigenvalue of h λ,ω iff λ 2 = ω/λ 0 and 1 0 f(x) 2 x 2 dµ (x) <. Similarly, if (ω 1)/Λ 1 < λ 2, then h λ,ω has a unique eigenvalue e > 1 which satisfies (36), and if (ω 1)/Λ 1 > λ 2, then h λ,ω has no eigenvalue in ]1, [. 1 is an eigenvalue of h λ,ω iff and 1 0 f(x) 2 (x 1) 2 dµ (x) <. (ω 1)/Λ 1 = λ 2, Example 2. Let dµ (x) dx [0,1], let f be a continuous function on ]0,1[, and let S = {x ]0,1[ f(x) 0}. The set S is open in ]0,1[, and the cyclic space generated by h and f is L 2 (S,dx). The spectrum of h λ,ω (C L2 (S,dx)), is purely absolutely continuous and equal to [0,1]\S. Since for x S, lim y 0 ImF (x+ iy) = π f(x) 2 > 0, the spectrum of h λ,ω in S is purely absolutely continuous for all λ 0. Hence, if S = ]a n,b n [, n is the decomposition of S into connected components, then the singular spectrum of h λ,ω inside [0,1] is concentrated on the set n {a n,b n }. In particular, h λ,ω has no singular continuous spectrum. A point e n {a n,b n } is an eigenvalue of h λ,ω iff

22 1 0 Mathematical Theory of the Wigner-Weisskopf Atom 21 f(x) 2 1 f(x) 2 dx < and ω e λ2 dx = 0. (37) (x e) 2 0 x e Given ω, for each e for which the first condition holds there are precisely two λ s such that e is an eigenvalue of h λ,ω. Hence, given ω, the set of λ s for which h λ,ω has eigenvalues in ]0,1[ is countable. Similarly, given λ, the set of ω s for which h λ,ω has eigenvalues in ]0,1[ is countable. Let Z {x [0,1] f(x) = 0}, and g sup x Z G (x). By (16), the number of eigenvalues of h λ,ω is bounded by 1 + λ 2 g. Hence, if g <, then h λ,ω can have at most finitely many eigenvalues. If, for example, f is δ-hölder continuous, i.e., f(x) f(y) C x y δ, for all x,y [0,1] and some δ > 1/2, then 1 f(t) 2 1 f(t) f(x) 2 g = sup dt = sup x Z (t x) 2 x Z 0 (t x) 2 dt sup x Z C dt <, (t x) 2(1 δ) and h λ,ω has at most finitely many eigenvalues. On the other hand, given λ 0, ω, and a finite sequence E {e 1,...,e n } ]0,1[, one can construct a C function f with bounded derivatives such that E is precisely the set of eigenvalues of h λ,ω in ]0,1[. More generally, let E {e n } ]0,1[ be a discrete set. (By discrete we mean that for all n, inf j n e n e j > 0 the accumulation points of E are not in E). Let λ 0 and ω be given and assume that ω is not an accumulation point of E. Then there is a C function f such that E is precisely the set of eigenvalues of h λ,ω in ]0,1[. Of course, in this case f (x) cannot be bounded. The construction of a such f is somewhat lengthy and can be found in [Kr]. In the remaining examples we take f 1. The next two examples are based on [How]. Example 3. Let µ be a pure point measure with atoms µ (x n ) = a n. Then G (x) = n=1 a n (x x n ) 2. If n an <, then G (x) < for Lebesgue a.e. x [0,1] (see Theorem 3.1 in [How]). Hence, by Simon-Wolff theorems 3 and 7, for a fixed λ 0 and Lebesgue a.e. ω, and for a fixed ω and Lebesgue a.e. λ, h λ,ω has only a pure point spectrum. On the other hand, for a fixed λ 0, there is a dense G δ set of ω such that the spectrum of h λ,ω on ]0,1[ is purely singular continuous [Gor, DMS].

23 22 V. Jakšić, E. Kritchevski, and C.-A. Pillet Example 4 (continuation). Assume that x n = x n (w) are independent random variables uniformly distributed on [0,1]. We keep the a n s deterministic and assume that an =. Then, for a.e. w, G,w (x) = for Lebesgue a.e. x [0,1] (see Theorem 3.2 in [How]). Hence, by Simon-Wolff theorems 4 and 8, for a fixed λ 0 and Lebesgue a.e. ω, and for a fixed ω and Lebesgue a.e. λ, the spectrum of h λ,ω (w) on [0,1] is singular continuous with probability 1. Example 5. Let ν be an arbitrary probability measure on [0,1]. Let dµ (x) = 1 2 ( dx [0,1] + dν(x) ). Since for all x ]0,1[, lim inf y 0 Im F (x + iy) π 2, the operator h λ,ω has purely absolutely continuous spectrum on [0,1] for all λ 0. In particular, the singular spectrum of ν disappears under the influence of the perturbation for all λ 0. Example 6. This example is due to Simon-Wolff [SW]. Let n 2 µ n = 2 n j=1 δ j2 n, and µ = n a nµ n, where a n > 0, n a n = 1 and n 2n a n =. The spectrum of h 0,ω is pure point and equal to [0,1] {ω}. For any x [0,1] there is j x such that j x /2 n x 2 n. Hence, for all n, dµ n (t) (t x) 2 2n, and G (x) = for all x [0,1]. We conclude that for all λ 0 and all ω the spectrum of h λ,ω on [0,1] is purely singular continuous. Example 7. Let µ C be the standard Cantor measure (see [S1]). Set ν j,n (A) µ C (A + j2 n ), and µ cχ [0,1] n=1 n 2 n 2 j=1 ν j,n, where c is the normalization constant. Then G (x) = for all x [0,1] (see Example 5 in Section II.5 of [Si2]), and the spectrum of h λ,ω on [0,1] is purely singular continuous for all λ,ω.

24 Mathematical Theory of the Wigner-Weisskopf Atom 23 Example 8. The following example is due to del io and Simon [DS] (see also Example 7 in Section II.5 of [Si2]). Let {r n } be the set of rationals in [0,1], a n min(3 n 1,r n,1 r n ), and S n (r n a n,r n + a n ). The set S is dense in [0,1] and S 1/3. Let dµ = S 1 χ S dx. The spectrum of h is purely absolutely continuous and equal to [0,1]. The set S is the essential support of this absolutely continuous spectrum. Clearly, for all λ,ω, sp ac (h λ,ω ) = [0,1]. By Theorem 5, for any fixed λ 0, h λ,ω will have some singular spectrum in [0,1] \ S for a set of ω s of positive Lebesgue measure. It is not difficult to show that G (x) < for Lebesgue a.e. x [0,1] \ S. Hence, for a fixed λ, h λ,ω will have no singular continuous spectrum for Lebesgue a.e. ω but it has some point spectrum in [0,1] \ S for a set of ω s of positive Lebesgue measure. For a given ω, h λ,ω has no singular continuous spectrum for Lebesgue a.e. λ. Since the set S is symmetric with respect to the point 1/2, we have that for all z C ±, e F (z) = e F ( z + 1/2). Hence, and if ω 1, then the set e F (x) = e F ( x + 1/2), (38) {x [0,1] \ S (ω x)f (x) > 0}, (39) has positive Lebesgue measure. Theorem 9 yields that for a given ω ]0,1[, h λ,ω will have some point spectrum in [0,1] \ S for a set of λ s of positive Lebesgue measure. If ω ]0,1[, the situation is more complex and depends on the choice of enumeration of the rationals. The enumeration can be always chosen in such a way that for all 0 < ǫ < 1, S [0,ǫ] < ǫ. In this case for any given ω the set (39) has positive Lebesgue measure and h λ,ω will have some singular continuous spectrum in [0,1] \ S for a set of λ s of positive Lebesgue measure. Example 9. This example is also due to del io and Simon [DS] (see also Example 8 in Section II.5 of [Si2]). Let S n = 2 n 1 j=1 ] j 2 n 1 4n 2 2 n, j 2 n + 1 [ 4n 2 2 n, and S = n S n. The set S is dense in [0,1] and S < 1. Let dµ = S 1 χ S dx. Then the absolutely continuous spectrum of h λ,ω is equal to [0,1] for all λ,ω. One easily shows that G (x) = on [0,1]. Hence, for a fixed λ, h λ,ω will have no point spectrum on [0,1] for Lebesgue a.e. ω but it has some singular continuous spectrum in [0,1] \ S for a set of ω s of positive Lebesgue measure. For a given ω, h λ,ω will have no point spectrum inside [0,1] for Lebesgue a.e. λ. The set S is symmetric with respect to 1/2 and (38) holds. Since for any 0 < ǫ < 1, S [0,ǫ] < ǫ, for any given ω the set {x [0,1] \ S (ω x)f (x) > 0},

25 24 V. Jakšić, E. Kritchevski, and C.-A. Pillet has positive Lebesgue measure. Hence, Theorem 9 yields that for a given ω, h λ,ω will have some singular continuous spectrum in [0,1] \ S for a set of λ s of positive Lebesgue measure. Example 10. Proposition 1 and a theorem of del io and Simon [DS] yield that there exist a bounded interval [a,b], a Borel probability measure µ on [a,b] and λ 0 > 0 such that: 1. sp ac (h λ,ω ) = [a,b] for all λ,ω. 2. for a set of ω s of positive Lebesgue measure, h ω,λ0 has embedded point spectrum in [a,b]. 3. for a set of ω s of positive Lebesgue measure, h ω,λ0 has embedded singular continuous spectrum in [a,b]. Example 11. Proposition 1 and a theorem of del io-fuentes-poltoratskii [DFP] yield that there exist a bounded interval [a,b], a Borel probability measure µ on [a,b] and λ 0 > 0 such that: 1. sp ac (h λ,ω ) = [a,b] for all λ,ω. 2. for all ω [0,1], the spectrum of h ω,λ0 is purely absolutely continuous. 3. for all ω [0,1], [a,b] sp sing (h ω,λ0 ). 2.9 Digression: the semi-circle law In the proof of Proposition 1 we have solved the equation (11) for µ. In this subsection we will find the fixed point of the equation (11). More precisely, we will find a finite Borel measure ν whose Borel transform satisfies the functional equation or, equivalently H(z) = 1 z λ 2 H(z), λ 2 H(z) 2 + zh(z) + 1 = 0. (40) The unique analytic solution of this equation is z2 4λ H(z) = 2 z 2λ 2, a two-valued function which can be made single valued by cutting the complex plane along the line segment [ 2 λ, 2 λ ]. Only one branch has the Herglotz property H(C + ) C +. This branch is explicitly given by H(z) = 1 ξ 1 λ ξ + 1, ξ z 2 λ z + 2 λ, where the branch of the square root is determined by e ξ > 0 (the so-called principal branch). In particular, H(x+iy) iy 1 as y +, and by a well known result

26 Mathematical Theory of the Wigner-Weisskopf Atom 25 in harmonic analysis (see e.g. [Ja]) there exists a unique Borel probability measure ν such that F ν (z) = H(z) for z C +. For all x, where lim Im F ν (x + iy) = h λ (x), y 0 4λ2 x 2 h λ (x) = 2λ 2 if x 2 λ, 0 if x > 2 λ. We deduce that the measure ν is absolutely continuous w.r.t. Lebesgue measure and that dν(x) = π 1 h λ (x)dx. Of course, ν is the celebrated Wigner semi-circle law which naturally arises in the study of the eigenvalue distribution of certain random matrices, see e.g. [Meh]. The result of this computation will be used in several places in the remaining part of our lectures. 3 The perturbative theory 3.1 The adiating Wigner-Weisskopf atom In this section we consider a specific class of WWA models which satisfy the following set of assumptions. Assumption (A1) h = L 2 (X,dx;K), where X = (e,e + ) is an open (possibly infinite) interval and K is a separable Hilbert space. The Hamiltonian h x is the operator of multiplication by x. Note that the spectrum of h is purely absolutely continuous and equal to X. For notational simplicity in this section we do not assume that f is a cyclic vector for h. This assumption is irrelevant for our purposes: since the cyclic space h 1 generated by h λ and 1 is independent of λ for λ 0, so is h 1 h and h λ h 1 = h h 1 has purely absolutely continuous spectrum. Assumption (A2) The function g(t) = is in L 1 (,dt). X e itx f(x) 2 K dx, This assumption implies that x f(x) K is a bounded continuous function on X. Note also that for Imz > 0,

27 26 V. Jakšić, E. Kritchevski, and C.-A. Pillet f(x) 2 K F (z) = dx = i e izs g(s)ds. X x z 0 Hence, F (z) is bounded and continuous on the closed half-plane C +. In particular, the function F (x + i0) is bounded and continuous on. If in addition t n g(t) L 1 (,dt) for some positive integer n, then f(x) 2 K and F (x + i0) are n-times continuously differentiable with bounded derivatives. Assumption (A3) ω X and f(ω) K > 0. This assumption implies that the eigenvalue ω of h 0 is embedded in its absolutely continuous spectrum. Until the end of this section we will assume that Assumptions (A1)-(A3) hold. We will call the WWA which satisfies (A1)-(A3) the adiating Wigner-Weisskopf Atom (abbreviated WWA). In contrast to the previous section, until the end of the paper we will keep ω fixed and consider only λ as the perturbation parameter. In the sequel we drop the subscript ω and write F λ for F λ,ω, etc. Since f(x) K is a continuous function of x, the argument of Example 2 in Subsection 2.8 yields that h λ has no singular continuous spectrum for all λ. However, h λ may have eigenvalues (and, if X, it will certainly have them for λ large enough). For λ small, however, the spectrum of h λ is purely absolutely continuous. Proposition 7. There exists Λ > 0 such that, for 0 < λ < Λ, the spectrum of h λ is purely absolutely continuous and equal to X. Proof. By Theorem 1, the singular spectrum of h λ is concentrated on the set S = {x ω x λ 2 F (x + i0) = 0}. Since Im F (ω + i0) = π f(ω) 2 K > 0, there is ǫ > 0 such that ImF (x + i0) > 0, for x ω < ǫ. Let m max x F (x + i0) and Λ (ǫ/m) 1/2. Then, for λ < Λ and x ]ω ǫ,ω+ǫ[, one has ω x > λ 2 F (x+i0). Hence, S is empty for 0 < λ < Λ, and the spectrum of h λ h1 is purely absolutely continuous. We finish this subsection with two examples. Example 1. Assume that h = L 2 ( d,d d x) and let h =, where is the usual Laplacian in d. The Fourier transform 1 ϕ(k) = e ik x ϕ(x)dx, (2π) d/2 d maps unitarily L 2 ( d,d d x) onto L 2 ( d,d d k) and the Hamiltonian h becomes multiplication by k 2. By passing to polar coordinates with r = k we identify

28 Mathematical Theory of the Wigner-Weisskopf Atom 27 L 2 ( d,d d k) with L 2 ( +,r d 1 dr;k), where K = L 2 (S d 1,dσ), S d 1 is the unit sphere in d, and dσ is its surface measure. The operator h becomes multiplication by r 2. Finally, the map ϕ # (x) = 2 1/2 x d 2 4 ϕ( x), maps L 2 ( d,d d x) unitarily onto L 2 (X,dx;K) with X = (0, ), and (h ϕ) # (x) = xϕ # (x). This representation of h and h (sometimes called the spectral or the energy representation) clearly satisfies (A1). The function f # satisfies (A2) iff the function g(t) = (f e ith f) is in L 1 (,dt). If f L 2 ( d,d d x) is compactly supported, then g(t) = O(t d/2 ), and so if d 3, then (A2) holds for all compactly supported f. If d = 1,2, then (A2) holds if f is in the domain of x 2 and its Fourier transform vanishes in a neighborhood of the origin. The proofs of these facts are simple and can be found in [B2], Example Example 2. Let h = l 2 (Z + ), where Z + = {1,2, }, and let h = 1 ) ((δ n )δ n+1 + (δ n+1 )δ n, 2 n Z + where δ n is the Kronecker delta function at n Z +. h is the usual discrete Laplacian on l 2 (Z + ) with Dirichlet boundary condition. The Fourier-sine transform 2 ϕ(k) ϕ(n)sin(kn), π n Z + maps l 2 (Z + ) unitarily onto L 2 ([0,π],dk) and the Hamiltonian h becomes multiplication by cos k. Finally, the map ϕ # (x) = (1 x 2 ) 1/4 ϕ(arccos x), maps l 2 (Z + ) unitarily onto L 2 (X,dx), where X = ( 1,1) and (h ϕ) # (x) = xϕ # (x). If f has bounded support in Z +, then f # (x) 2 = (1 x 2 ) 1/2 P f (x), where P f (x) is a polynomial in x. A simple stationary phase argument yields that g(t) = O( t 3/2 ) and Assumption (A2) holds. 3.2 Perturbation theory of embedded eigenvalue Until the end of this section Λ is the constant in Proposition 7.

29 28 V. Jakšić, E. Kritchevski, and C.-A. Pillet Note that the operator h 0 = ω x has the eigenvalue ω embedded in the absolutely continuous spectrum of x. On the other hand, for 0 < λ < Λ the operator h λ has no eigenvalue the embedded eigenvalue has dissolved in the absolutely continuous spectrum under the influence of the perturbation. In this subsection we will analyze this phenomenon. At its heart are the concepts of resonance and life-time of an embedded eigenvalue which are of profound physical importance. We set D(w,r) {z C z w < r}. In addition to (A1)-(A3) we will need the following assumption. Assumption (A4) There exists ρ > 0 such that the function C + z F (z), has an analytic continuation across the interval ]ω ρ,ω + ρ[ to the region C + D(ω,ρ). We denote the extended function by F + (z). It is important to note that for Imz < 0, F + (z) is different from F (z). This is obvious from the fact that Im F (x + i0) Im F (x i0) = 2π f(x) 2 K > 0, near ω. In particular, if (A4) holds, then ρ must be such that ]ω ρ,ω + ρ[ X. Until the end of this subsection we will assume that Assumptions (A1)-(A4) hold. Theorem The function F λ (z) = (1 (h λ z) 1 1) has a meromorphic continuation from C + to the region C + D(ω,ρ). We denote this continuation by F + λ (z). 2. Let 0 < ρ < ρ be given. Then there is Λ > 0 such that for λ < Λ the only singularity of F + λ (z) in D(ω,ρ ) is a simple pole at ω(λ). The function λ ω(λ) is analytic for λ < Λ and ω(λ) = ω + a 2 λ 2 + O(λ 4 ), where a 2 F (ω + i0). In particular, Im a 2 = π f(ω) 2 K < 0. Proof. Part (1) is simple Assumption A4 and Equ. (8) yield that F + λ (z) = 1 ω z λ 2 F + (z), is the mermorphic continuation of C + z F λ (z) to C + D(ω,ρ). For a given ρ, choose Λ > 0 such that ρ > Λ 2 sup z =ρ F + (z). By ouché s theorem, there is an ǫ > 0 such that for λ < Λ the function ω z λ 2 F + (z) has a unique simple zero ω(λ) inside D(ω,ρ + ǫ) such that ω(λ) ω <

30 Mathematical Theory of the Wigner-Weisskopf Atom 29 ρ ǫ. This yields that F + λ (z) is analytic in C + D(ω,ρ + ǫ) except for a simple pole at ω(λ). The function P(λ) zf + λ (z)dz = ω z =ρ n=0 is analytic for λ < Λ. Similarly, the function Q(λ) F + λ (z)dz = λ 2n ω z =ρ n=0 is analytic and non-zero for λ < Λ. Since ω(λ) = P(λ) Q(λ), ( F λ ω z =ρ 2n + z ω z =ρ ( F + (z) ω z (z) ω z we see that ω(λ) is analytic for λ < Λ with the power series expansion Obviously, a 0 = ω and a 2 = 1 2πi ω z =ρ ω(λ) = λ 2n a 2n. n=0 ) n dz ω z, + F (z) z ω dz = F + (ω) = F (ω + i0). The same formula can be obtained by implicit differentiation of at λ = 0. ω ω(λ) λ 2 F + (ω(λ)) = 0, ) n dz ω z, (41) Theorem 10 explains the mechanism of dissolving of the embedded eigenvalue ω. The embedded eigenvalue ω has moved from the real axis to a point ω(λ) on the second (improperly called unphysical ) iemann sheet of the function F λ (z). There it remains the singularity of the analytically continued resolvent matrix element (1 (h λ z) 1 1), see Figure 1. We now turn to the physically important concept of the life-time of the embedded eigenvalue. Theorem 11. There exists Λ > 0 such that for λ < Λ and all t 0 (1 e ith λ 1) = e itω(λ) + O(λ 2 ). Proof. By Theorem 7 the spectrum of h λ is purely absolutely continuous for 0 < λ < Λ. Hence, by Theorem 1, dµ λ (x) = dµ λ ac(x) = 1 π Im F λ(x + i0)dx = 1 π ImF + λ (x)dx.

31 30 V. Jakšić, E. Kritchevski, and C.-A. Pillet physical iemann sheet ρ ω ω(λ) 2nd iemann sheet Fig. 1. The resonance pole ω(λ). Let Λ and ρ be the constants in Theorem 10, Λ min(λ,λ), and suppose that 0 < λ < Λ. We split the integral representation (1 e ith λ 1) = e itx dµ λ (x), (42) into three parts as Equ. (8) yields X ω ρ ω+ρ e e ω ρ ω+ρ Im F + Im F + λ (x) = (x) λ2 ω x λ 2 F +, (x) 2 and so the first and the third term can be estimated as O(λ 2 ). The second term can be written as I(t) 1 ω+ρ ) e (F itx +λ 2πi (x) F +λ (x) dx. ω ρ The function z F + λ (z) is meromorphic in an open set containing D(ω,ρ) with only singularity at ω(λ). We thus have ( ) I(t) = (λ)e itω(λ) + e itz F + λ (z) F + λ (z) dz, where the half-circle γ = {z z ω = ρ,im z 0} is positively oriented and γ

32 Mathematical Theory of the Wigner-Weisskopf Atom 31 (λ) = es z=ω(λ) F + λ (z). By Equ. (41), (λ) = Q(λ)/2πi is analytic for λ < Λ and Equ. (8) yields that for z γ Since ω is real, this estimate yields (λ) = 1 + O(λ 2 ). F + λ (z) = 1 ω z + O(λ2 ). F + λ (z) F + λ (z) = O(λ2 ). Combining the estimates we derive the statement. If a quantum mechanical system, described by the Hilbert space h and the Hamiltonian h λ, is initially in a pure state described by the vector 1, then P(t) = (1 e ith λ 1) 2, is the probability that the system will be in the same state at time t. Since the spectrum of h λ is purely absolutely continuous, by the iemann-lebesgue lemma lim t P(t) = 0. On physical grounds one often expects more, namely an approximate relation P(t) e tγ(λ), (43) where Γ(λ) is the so-called radiative life-time of the state 1. The strict exponential decay P(t) = O(e at ) is possible only if X =. Since in a typical physical situation X, the relation (43) is expected to hold on an intermediate time scale (for times which are not too long or too short ). Theorem 11 is a mathematically rigorous version of these heuristic claims and Γ(λ) = 2 Im ω(λ). The computation of the radiative life-time is of paramount importance in quantum mechanics and the reader may consult standard references [CDG, He, Mes] for additional information. 3.3 Complex deformations In this subsection we will discuss Assumption (A4) and the perturbation theory of the embedded eigenvalue in some specific situations. Example 1. In this example we consider the case X =]0, [. Let 0 < δ < π/2 and A(δ) = {z C e z > 0, Arg z < δ}. We denote by Hd 2 (δ) the class of all functions f : X K which have an analytic continuation to the sector A(δ) such that f 2 δ = sup f(e iθ x) 2 Kdx <. θ <δ X The class H 2 d (δ) is a Hilbert space. The functions in H2 d (δ) are sometimes called dilation analytic.

33 32 V. Jakšić, E. Kritchevski, and C.-A. Pillet Proposition 8. Assume that f Hd 2 (δ). Then Assumption (A4) holds in the following stronger form: 1. The function F (z) has an analytic continuation to the region C + A(δ). We denote the extended function by F + (z). 2. For 0 < δ < δ and ǫ > 0 one has X sup F + (z) <. z >ǫ,z A(δ ) Proof. The proposition follows from the representation f(x) 2 K F (z) = dx = e iθ (f(e iθ x) f(e iθ x)) K x z e iθ x z X dx, (44) which holds for Imz > 0 and δ < θ 0. This representation can be proven as follows. Let γ(θ) be the half-line e iθ +. We wish to prove that for Imz > 0 X f(x) 2 K x z dx = γ(θ) (f(w) f(w)) K w z dw. To justify the interchange of the line of integration, it suffices to show that lim n r n 0 θ (f(r n e iϕ ) f(r n e iϕ )) K r n e iϕ z dϕ = 0, along some sequence r n. This fact follows from the estimate [ 0 x (f(e iϕ x) f(e iϕ ] x)) K e iϕ dϕ dx C z f 2 x z δ. X θ Until the end of this example we assume that f Hd 2 (δ) and that Assumption (A2) holds (this is the case, for example, if f Hd 2 (δ) and f(0) = 0). Then, Theorems 10 and 11 hold in the following stronger forms. Theorem The function F λ (z) = (1 (h λ z) 1 1), has a meromorphic continuation from C + to the region C + A(δ). We denote this continuation by F + λ (z). 2. Let 0 < δ < δ be given. Then there is Λ > 0 such that for λ < Λ the only singularity of F + λ (z) in A(δ ) is a simple pole at ω(λ). The function λ ω(λ) is analytic for λ < Λ and ω(λ) = ω + λ 2 a 2 + O(λ 4 ), where a 2 = F (ω + i0). In particular, Im a 2 = π f(ω) 2 K < 0.

34 Mathematical Theory of the Wigner-Weisskopf Atom 33 Theorem 13. There exists Λ > 0 such that for λ < Λ and all t 0, (1 e ith λ 1) = e itω(λ) + O(λ 2 t 1 ). The proof of Theorem 13 starts with the identity (1 e ith λ 1) = λ 2 e itx f(x) 2 K F + λ (x) 2 dx. X Given 0 < δ < δ one can find Λ such that for λ < Λ (1 e ith λ 1) = e itω(λ) +λ e 2 e itw (f(w) f(w)) K F + λ (w)f + λ (w)dw, (45) iδ + and the integral on the right is easily estimated by O(t 1 ). We leave the details of the proof as an exercise for the reader. Example 2. We will use the structure of the previous example to illustrate the complex deformation method in study of resonances. In this example we assume that f H 2 d (δ). We define a group {u(θ) θ } of unitaries on h by u(θ) : α f(x) α e θ/2 f(e θ x). Note that h (θ) u( θ)h u(θ) is the operator of multiplication by e θ x. Set h 0 (θ) = ω h (θ), f θ (x) = u( θ)f(x)u(θ) = f(e θ x), and h λ (θ) = h 0 (θ) + λ ((1 )f θ + (f θ )1). Clearly, h λ (θ) = u( θ)h λ u(θ). We set S(δ) {z Im z < δ} and note that the operator h 0 (θ) and the function f θ are defined for all θ S(δ). We define h λ (θ) for λ C and θ S(δ) by h λ (θ) = h 0 (θ) + λ ( (1 )f θ + (f θ )1 ). The operators h λ (θ) are called dilated Hamiltonians. The basic properties of this family of operators are: 1. Dom (h λ (θ)) is independent of λ and θ and equal to Dom (h 0 ). 2. For all φ Dom (h 0 ) the function C S(δ) (λ,θ) h λ (θ)φ is analytic in each variable separately. A family of operators which satisfy (1) and (2) is called an analytic family of type A in each variable separately. 3. If Im θ = Imθ, then the operators h λ (θ) and h λ (θ ) are unitarily equivalent, namely h 0 (θ ) = u( (θ θ))h 0 (θ)u(θ θ). 4. sp ess (h 0 (θ)) = e θ + and sp disc (h 0 (θ)) = {ω}, see Figure 2.

35 34 V. Jakšić, E. Kritchevski, and C.-A. Pillet ¼ µ ÁÑ Ô µµ Fig. 2. The spectrum of the dilated Hamiltonian h λ (θ). The important aspect of (4) is that while ω is an embedded eigenvalue of h 0, it is an isolated eigenvalue of h 0 (θ) as soon as Im θ < 0. Hence, if Imθ < 0, then the regular perturbation theory can be applied to the isolated eigenvalue ω. Clearly, for all λ, sp ess (h λ (θ)) = sp(h 0 (θ)) and one easily shows that for λ small enough sp disc (h λ )(θ) = { ω(λ)} (see Figure 2). Moreover, if 0 < ρ < min{ω,ω tan θ}, then for sufficiently small λ, z(1 (h λ (θ) z) 1 1)dz z ω =ρ ω(λ) =. (1 (h λ (θ) z) 1 1)dz z ω =ρ The reader should not be surprised that the eigenvalue ω(λ) is precisely the pole ω(λ) of F + λ (z) discussed in Theorem 10 (in particular, ω(λ) is independent of θ). To clarify this connection, note that u(θ)1 = 1. Thus, for real θ and Im z > 0, F λ (z) = (1 (h λ z) 1 1) = (1 (h λ (θ) z) 1 1). On the other hand, the function θ (1 h λ (θ) z) 1 1) has an analytic continuation to the strip δ < Imθ < Imz. This analytic continuation is a constant function, and so F + λ (z) = (1 (h λ(θ) z) 1 1), for δ < Im θ < 0 and z C + A( Im θ ). This yields that ω(λ) = ω(λ). The above set of ideas plays a very important role in mathematical physics. For additional information and historical perspective we refer the reader to [AC, BC, CFKS, Der2, Si2, S4]. Example 3. In this example we consider the case X =. Let δ > 0. We denote by Ht 2 (δ) the class of all functions f : X K which have an analytic continuation to the strip S(δ) such that f 2 δ sup f(x + iθ) 2 K dx <. θ <δ X

36 Mathematical Theory of the Wigner-Weisskopf Atom 35 The class H 2 t (δ) is a Hilbert space. The functions in H 2 t (δ) are sometimes called translation analytic. Proposition 9. Assume that f H 2 t (δ). Then the function F (z) has an analytic continuation to the half-plane {z C Im z > δ}. The proposition follows from the relation f(x) 2 K (f(x iθ) f(x + iθ)) K F (z) = dx = x z x + iθ z X X dx, (46) which holds for Imz > 0 and δ < θ 0. The proof of (46) is similar to the proof of (44). Until the end of this example we will assume that f Ht 2 (δ). A change of the line of integration yields that the function g(t) = e itx f(x) 2 K dx, satisfies the estimate g(t) e δ t f 2 δ, and so Assumption (A2) holds. Moreover, Theorems 10 and 11 hold in the following stronger forms. Theorem The function F λ (z) = (1 (h λ z) 1 1), has a meromorphic continuation from C + to the half-plane {z C Im z > δ}. We denote this continuation by F + λ (z). 2. Let 0 < δ < δ be given. Then there is Λ > 0 such that for λ < Λ the only singularity of F + λ (z) in {z C Im z > δ } is a simple pole at ω(λ). ω(λ) is analytic for λ < Λ and ω(λ) = ω + λ 2 a 2 + O(λ 4 ), where a 2 = F (ω + i0). In particular, Im a 2 = π f(ω) 2 K < 0. Theorem 15. Let 0 < δ < δ be given. Then there exists Λ > 0 such that for λ < Λ and all t 0 (1 e ith λ 1) = e itω(λ) + O(λ 2 e δ t ). In this example the survival probability has strict exponential decay. We would like to mention two well-known models in mathematical physics for which analogs of Theorems 14 and 15 holds. The first model is the Stark Hamiltonian which describes charged quantum particle moving under the influence of a

37 36 V. Jakšić, E. Kritchevski, and C.-A. Pillet constant electric field [Her]. The second model is the spin-boson system at positive temperature [JP1, JP2]. In the translation analytic case, one can repeat the discussion of the previous example with the analytic family of operators h λ (θ) = ω (x + θ) + λ ( (1 )f θ + (f θ )1 ), where f θ (x) f(x + θ) (see Figure 3). Note that in this case sp ess (h λ (θ)) = sp ess (h 0 (θ)) = + iim θ. µ ÁÑ Ô µµ Fig. 3. The spectrum of the translated Hamiltonian h λ (θ). Example 4. Let us consider the model described in Example 2 of Subsection 3.1 where f l 2 (Z + ) has bounded support. In this case X =] 1,1[ and 1 1 x 2 F (z) = P f (x)dx, (47) x z 1 where P f (x) is a polynomial in x. Since the integrand is analytic in the cut plane C \ {x x 1}, we can deform the path of integration to any curve γ joining 1 to 1 and lying entirely in the lower half-plane (see Figure 4). This shows that the function F (z) has an analytic continuation from C + to the entire cut plane C \ {x x 1}. Assumption (A4) holds in this case.

38 Mathematical Theory of the Wigner-Weisskopf Atom 37 ½ ½ Fig. 4. Deforming the integration contour in Equ. (47). 3.4 Weak coupling limit The first computation of the radiative life-time in quantum mechanics goes back to the seminal papers of Dirac [Di] and Wigner and Weisskopf [WW]. Consider the survival probability P(t) and assume that P(t) e tγ(λ) where Γ(λ) = λ 2 Γ 2 + O(λ 3 ) for λ small. To compute the first non-trivial coefficient Γ 2, Dirac devised a computational scheme called time-dependent perturbation theory. Dirac s formula for Γ 2 was called Golden ule in Fermi s lectures [Fer], and since then this formula is known as Fermi s Golden ule. One possible mathematically rigorous approach to time-dependent perturbation theory is the so-called weak coupling (or Van Hove) limit. The idea is to study P(t/λ 2 ) as λ 0. Under very general conditions one can prove that lim λ 0 P(t/λ2 ) = e tγ2, and that Γ 2 is given by Dirac s formula (see [Da2, Da3]). In this section we will discuss the weak coupling limit for the WWA. We will prove: Theorem 16. Suppose that Assumptions (A1)-(A3) hold. Then lim (1 e ith λ/λ 2 1) e itω/λ2 e itf(ω+i0) = 0, λ 0 for any t 0. In particular, lim λ 0 (1 e ith λ/λ 2 1) 2 = e 2π f(ω) 2 K t. emark. If in addition Assumption (A4) holds, then Theorem 16 is an immediate consequence of Theorem 11. The point is that the leading contribution to the life-time can be rigorously derived under much weaker regularity assumptions. Lemma 3. Suppose that Assumptions (A1)-(A3) hold. Let u be a bounded continuous function on X. Then

39 38 V. Jakšić, E. Kritchevski, and C.-A. Pillet lim λ 0 λ2 e itx/λ2 u(x) F λ (x + i0) 2 dx u(ω) X f(ω) 2 e it(ω/λ2 F (ω+i0)) = 0, K for any t 0. Proof. We set h ω (x) ω x λ 2 F (ω + i0) 2 and I λ (t) λ 2 e itx/λ2 u(x) F λ (x + i0) 2 dx. We write u(x) F λ (x + i0) 2 as X u(ω)h ω (x) + (u(x) u(ω))h ω (x) + u(x) ( F λ (x + i0) 2 h ω (x) ), and decompose I λ (t) into three corresponding pieces I k,λ (t). The first piece is e+ I 1,λ (t) = λ 2 e itx/λ 2 u(ω) e (ω x λ 2 e F (ω + i0)) 2 + (λ 2 Im F (ω + i0)) 2 dx. The change of variable y = x ω + λ2 e F (ω + i0) λ 2, ImF (ω + i0) and the relation Im F (ω + i0) = π f(ω) 2 K yield that I 1,λ (t) = e it(ω/λ2 e F (ω+i0)) u(ω) 1 e+(λ) itim F(ω+i0)y e f(ω) 2 K π e (λ) y 2 dy, + 1 where as λ 0. From the formula e ± (λ) λ 2 e ± ω π f(ω) 2 + e F (ω + i0) K π f(ω) 2 ±, K we obtain that 1 π itim F(ω+i0)y e y 2 dy = e tim F(ω+i0), + 1 I 1,λ (t) = u(ω) f(ω) 2 e it(ω/λ2 F (ω+i0)) (1 + o(1)), (48) K as λ 0. Using the boundedness and continuity properties of u and h ω, one easily shows that the second and the third piece can be estimated as I 2,λ (t) λ 2 u(x) u(ω) h ω (x)dx, X I 3,λ (t) λ 2 u(x) F λ (x + i0) 2 h ω (x) dx. X

40 Mathematical Theory of the Wigner-Weisskopf Atom 39 Hence, they vanish as λ 0, and the result follows from Equ. (48). Proof of Theorem 16. Let Λ be as in Proposition 7. ecall that for 0 < λ < Λ the spectrum of h λ is purely absolutely continuous. Hence, for λ small, (1 e ith λ/λ 2 1) = 1 e itx/λ2 Im F λ (x + i0)dx π X = 1 e itx/λ2 F λ (x + i0) 2 Im F (x + i0)dx π X = λ 2 e itx/λ2 f(x) 2 K F λ (x + i0) 2 dx, X where we used Equ. (19). This formula and Lemma 3 yield Theorem 16. The next result we wish to discuss concerns the weak coupling limit for the form of the emitted wave. Let p be the orthogonal projection on the subspace h of h. Theorem 17. For any g C 0 (), ( lim(p e ith λ/λ 2 1 g(h )p e ith λ/λ 2 1) = g(ω) 1 e 2π f(ω) 2 t) K. (49) λ 0 Proof. Using the decomposition p g(h )p = (p g(h )p g(h 0 )) + (g(h 0 ) g(h λ )) + g(h λ ) = g(ω)(1 )1 + (g(h 0 ) g(h λ )) + g(h λ ), we can rewrite (p e ith λ/λ 2 1 g(h )p e ith λ/λ 2 1) as a sum of three pieces. The first piece is equal to g(ω) (1 e ith λ/λ 2 1) 2 = g(ω)e 2π f(ω) 2 K t. (50) Since λ h λ is continuous in the norm resolvent sense, we have and the second piece can be estimated as λ 0. The third piece satisfies lim g(h λ) g(h 0 ) = 0, λ 0 (e ith λ/λ 2 1 (g(h 0 ) g(h λ ))e ith λ/λ 2 1) = o(1), (51) (e ith λ/λ 2 1 g(h λ )e ith λ/λ 2 1) = (1 g(h λ )1) = (1 g(h 0 )1) + (1 (g(h λ ) g(h 0 ))1) (52) = g(ω) + o(1), as λ 0. Equ. (50), (51) and (52) yield the statement. Needless to say, Theorems 16 and 17 can be also derived from the general theory of weak coupling limit developed in [Da2, Da3]. For additional information about the weak coupling limit we refer the reader to [Da2, Da3, Der3, FGP, Haa, VH].

41 40 V. Jakšić, E. Kritchevski, and C.-A. Pillet 3.5 Examples In this subsection we describe the meromorphic continuation of F λ (z) = (1 (h λ z) 1 1), across sp ac (h λ ) in some specific examples which allow for explicit computations. Since F λ (z) = F λ (z), we need to consider only λ 0. Example 1. Let X =]0, [ and f(x) π 1/2 (2x) 1/4 (1 + x 2 ) 1/2. Note that f Hd 2 (δ) for 0 < δ < π/2 and so f is dilation analytic. In this specific example, however, one can evaluate F (z) directly and describe the full iemann surface of F λ (z), thus going far beyond the results of Theorem 12. For z C \[0, ) we set w z, where the branch is chosen so that e w > 0. Then iw C + and the integral F (z) = 1 π 0 2t 1 + t 2 dt t z = 2 π t t 4 dt t 2 + w 2, is easily evaluated by closing the integration path in the upper half-plane and using the residue method. We get F (z) = 1 w 2 + 2w + 1. Thus F is a meromorphic function of w with two simple poles at w = e ±3iπ/4. It follows that F (z) is meromorphic on the two-sheeted iemann surface of z. On the first (physical) sheet, where e w > 0, it is of course analytic. On the second sheet, where e w < 0, it has two simple poles at z = ±i. In term of the uniformizing variable w, we have F λ (z) = w 2 + 2w + 1 (w 2 + ω)(w 2 + 2w + 1) λ 2. For λ > 0, this meromorphic function has 4 poles. These poles are analytic functions of λ except at the collision points. For λ small, the poles form two conjugate pairs, one near ±i ω, the other near e ±3iπ/4. Both pairs are on the second sheet. For λ large, a pair of conjugated poles goes to infinity along the asymptote e w = 2/4. A pair of real poles goes to ±. In particular, one of them enters the first sheet at λ = ω and h λ has one negative eigenvalue for λ > ω. Since G (x) = 1 2 dt π 1 + t 2 (t x) 2, 0 is finite for x < 0 and infinite for x 0, 0 is not an eigenvalue of h λ for λ = ω, but a zero energy resonance. Note that the image of the asymptote e w = 2/4

42 Mathematical Theory of the Wigner-Weisskopf Atom 41 on the second sheet is the parabola {z = x + iy x = 2y 2 1/8}. Thus, as λ, the poles of F λ (z) move away from the spectrum. This means that there are no resonances in the large coupling limit. The qualitative trajectories of the poles (as functions of λ for fixed values of ω) are plotted in Figure 5. ¼ ½ ¾ ½ ¾ Ô Ô Ô Ô ½ ¾ ½ Ô Ô ½ Ô Ô Fig. 5. Trajectories of the poles of F λ (z) in w-space for various values of ω in Example 1. Notice the simultaneous collision of the two pairs of conjugate poles when ω = λ = 1/2. The second iemann sheet is shaded. Example 2. Let X = and f(x) π 1/2 (1 + x 2 ) 1/2. Since f H 2 t (δ) for 0 < δ < 1, the function f is translation analytic. Here again we can compute explicitly F (z). For z C +, a simple residue calculation leads to

43 42 V. Jakšić, E. Kritchevski, and C.-A. Pillet F (z) = 1 1 dt π 1 + t 2 t z = 1 z + i. Hence, F λ (z) = z + i λ 2 (z + i)(z ω), has a meromorphic continuation across the real axis to the entire complex plane. It has two poles given by the two branches of ω(λ) = ω i + (ω + i) 2 + 4λ 2, 2 which are analytic except at the collision point ω = 0, λ = 1/2. For small λ, one of these poles is near ω and the other is near i. Since ω(λ) = i 2 + ( ω 2 ± λ ) + O(1/λ), as λ, h λ has no large coupling resonances. The resonance curve ω(λ) is plotted in Figure 6. Clearly, sp(h λ ) = for all ω and λ. Note that for all x, G (x) = and Im F λ (x + i0) = λ 2 (x ω) 2 + (λ 2 x(x ω)) 2. Hence, the operator h λ has purely absolutely continuous spectrum for all ω and all λ 0. Ô µ ¼ ¼ Fig. 6. The poles of F λ (z) for Example 2. Example 3. Let X =] 1,1[ and f(x) 2 π (1 x2 ) 1/4. (ecall Example 2 in Subsection 3.1 and Example 4 in Subsection 3.3 h and h are l 2 (Z + ) and the discrete Laplacian in the energy representation and f = δ # 1.) In Subsection 2.9 we have shown that for z C \ [ 1,1],

44 where Mathematical Theory of the Wigner-Weisskopf Atom 43 F (z) = t 2 π 1 t z dt = 2 ξ 1 ξ + 1, (53) ξ = z 1 z + 1. The principal branch of the square root e ξ > 0 corresponds to the first (physical) Û Þ ¾Ò Ø Fig. 7. Mapping the cut plane C \ [ 1, 1] to the exterior of the unit disk sheet of the iemann surface of F (z). The branch e ξ < 0 corresponds to the second sheet of. In particular, F (x + i0) = 2( x + i 1 x 2 ). For an obvious reason we will call this particular WWA the semi-circle model. To discuss the analytic structure of the Borel transform F λ (z), it is convenient to introduce the uniformizing variable w 2 F (z) = 1 + ξ 1 ξ, which maps the iemann surface to C \ {0}. Note that the first sheet of is mapped on the exterior of the unit disk and that the second sheet is mapped on the punctured disk {z C 0 < z < 1} (see Figure 7). The inverse of this map is z = 1 ( w + 1 ). 2 w

45 44 V. Jakšić, E. Kritchevski, and C.-A. Pillet For z C \ [ 1,1] the function F λ (z) is given by F λ (z) = 2w w 2 2ωw + 1 4λ 2, and thus has a meromorphic continuation to the entire iemann surface. The resonance poles in the w-plane are computed by solving w 2 2ωw + 1 4λ 2 = 0, and are given by the two-valued analytic function w = ω + 4λ 2 + ω 2 1. We will describe the motion of the poles in the case ω 0 (the case ω 0 is completely symmetric). For 0 < λ < 1 ω 2 /2 there are two conjugate poles on the second sheet which, in the w-plane, move towards the point ω on a vertical line. After their collision at λ = 1 ω 2 /2, they turn into a pair of real poles moving towards ± (see Figure 8). The pole moving to the right reaches w = 1 at λ = (1 ω)/2 and enters the first sheet of. We conclude that h λ has a positive eigenvalue ω + (λ) = 1 2 ( ω + 4λ 2 + ω ω + 4λ 2 + ω 2 1 for λ > (1 ω)/2. The pole moving to the left reaches w = 0 at λ = 1/2. This means that this pole reaches z = on the second sheet of. For λ > 1/2, the pole continues its route towards w = 1, i.e., it comes back from z = towards z = 1, still on the second sheet of. At λ = (1 + ω)/2, it reaches w = 1 and enters the first sheet. We conclude that h λ has a negative eigenvalue ω (λ) = 1 2 ( ω 4λ 2 + ω ω 4λ 2 + ω 2 1 for λ > (1 + ω)/2. The trajectory of these poles in the z cut-plane is shown on Figure 9. For clarity, only one pole of the conjugate pair is displayed. Example 4. In all the previous examples, there were no resonances in the large coupling regime, i.e., the second sheet poles of F λ kept away from the continuous spectrum as λ. This fact can be understood as follows. If a resonance ω(λ) approaches the real axis as λ, then it follows from Equ. (8) that Im F (ω(λ)) = o(λ 2 ). Since under Assumptions (A1) and (A2) F is continuous on C +, we conclude that if lim λ ω(λ) = ω, then ImF (ω + i0) = 0. Since f(x) K is also continuous on X, if ω X, then we must have f(ω) = 0. Thus the only possible locations of large coupling resonances are the zeros of f in X. We finish this subsection with an example where such large coupling resonances exist. ), ),

46 Mathematical Theory of the Wigner-Weisskopf Atom 45 Û Ô½ ¾ Ô½ ¾ Fig. 8. The trajectories of the poles of F λ in the w-plane. The second sheet is shaded. ½ Þ ½ ½ µ ¾ Fig. 9. The trajectories of the poles of F λ in the z-plane. Dashed lines are on the second sheet. Let again X =] 1,1[ and set The Borel transform f(x) F (z) = 1 π 1 π x(1 x2 ) 1/ x 2 1 x 2 x z dx, is easily evaluated by a residue calculation and the change of variable x = (u + u 1 )/2. Using the same uniformizing variable w as in Example 3, we get F (z) = 1 ( ) 1 4 w 2 w, (54) and

47 46 V. Jakšić, E. Kritchevski, and C.-A. Pillet F λ (z) = 4w 3 2w 4 4ωw 3 + (2 λ 2 )w 2 λ 2. (55) We shall again restrict ourselves to the case 0 < ω < 1. At λ = 0 the denominator of (55) has a double zero at w = 0 and a pair of conjugated zeros at ω ±i 1 ω 2. As λ increases, the double zero at 0 splits into a pair of real zeros going to ±. The right zero reaches 1 and enters the first sheet at λ = 2(1 ω). At λ = 2(1 + ω), the left zero reaches 1 and also enters the first sheet. The pair of conjugated zeros move from their original positions towards ±i (of course, they remain within the unit disk). For large λ, they behave like w = ±i + 2ω 2ω(2 ± 5iω) λ2 λ 4 + O(λ 6 ). Thus, in the z plane, F λ has two real poles emerging from ± on the second sheet and traveling towards ±1. The right pole reaches 1 at λ = 2(1 ω) and becomes an eigenvalue of h λ which returns to + as λ further increases. The left pole reaches 1 at λ = 2(1 + ω), becomes an eigenvalue of h λ, and further proceeds towards. On the other hand, the eigenvalue ω of h 0 turns into a pair of conjugated poles on the second sheet which, as λ, tend towards 0 as ω(λ) = 2ω 4ω(1 ± 2iω) λ2 λ 4 + O(λ 6 ), see Figure 9. We conclude that h λ has a large coupling resonance approaching 0 as λ. ½ Þ ½ Fig. 10. The trajectories of the poles of F λ in the z-plane. Dashed lines are on the second sheet. 4 Fermionic quantization 4.1 Basic notions of fermionic quantization This subsection is a telegraphic review of the fermionic quantization. For additional information and references the reader may consult Section 5 in [AJPP1]. Let h be a Hilbert space. We denote by Γ(h) the fermionic (antisymmetric) Fock space over h, and by Γ n (h) the n-particle sector in h. Φ h denotes the vacuum in Γ(h) and a(f),a (f) the annihilation and creation operators associated to f h. In the

48 Mathematical Theory of the Wigner-Weisskopf Atom 47 sequel a # (f) represents either a(f) or a (f). ecall that a # (f) = f. The CA algebra over h, CA(h), is the C -algebra of bounded operators on Γ(h) generated by {a # (f) f h}. Let u be a unitary operator on h. Its second quantization Γ(u) Γn(h) u u = u n, defines a unitary operator on Γ(h) which satisfies Γ(u)a # (f) = a # (uf)γ(u). Let h be a self-adjoint operator on h. The second quantization of e ith is a strongly continuous group of unitary operators on Γ(h). The generator of this group is denoted by dγ(h), Γ(e ith ) = e itdγ(h). dγ(h) is essentially self-adjoint on Γ(Dom h), where Dom h is equipped with the graph norm, and one has The maps dγ(h) Γn(Dom h) = n k=1 } I {{ I } h I } {{ I }. k 1 n k τ t (a # (f)) = e itdγ(h) a # (f)e itdγ(h) = a # (e ith f), uniquely extend to a group τ of -automorphisms of CA(h). τ is often called the group of Bogoliubov automorphisms induced by h. The group τ is norm continuous and the pair (CA(h),τ) is a C -dynamical system. We will call it a CA dynamical system. We will also call the pair (CA(h),τ) the fermionic quantization of (h,h). If two pairs (h 1,h 1 ) and (h 2,h 2 ) are unitarily equivalent, that is, if there exists a unitary u : h 1 h 2 such that uh 1 u 1 = h 2, then the fermionic quantizations (CA(h 1 ),τ 1 ) and (CA(h 2 ),τ 2 ) are isomorphic the map σ(a # (f)) = a # (uf) extends uniquely to a -isomorphism such that σ τ t 1 = τ t 2 σ. 4.2 Fermionic quantization of the WWA Let h λ be a WWA on h = C h. Its fermionic quantization is the pair (CA(h),τ λ ), where τ t λ(a # (φ)) = e itdγ(h λ) a # (φ)e itdγ(h λ) = a # (e ith λ φ). We will refer to (CA(h),τ λ ) as the Simple Electronic Black Box (SEBB) model. This model has been discussed in the recent lecture notes [AJPP1]. The SEBB model is the simplest non-trivial example of the Electronic Black Box model introduced and studied in [AJPP2]. The SEBB model is also the simplest non-trivial example of an open quantum system. Set

49 48 V. Jakšić, E. Kritchevski, and C.-A. Pillet τ t S(a # (α)) = a # (e itω α), τ t (a # (g)) = a # (e ith g). The CA dynamical systems (CA(C),τ S ) and (CA(h ),τ ) are naturally identified with sub-systems of the non-interacting SEBB (CA(h),τ 0 ). The system (CA(C),τ S ) is a two-level quantum dot without internal structure. The system (CA(h ),τ ) is a free Fermi gas reservoir. Hence, (CA(h λ ),τ λ ) describes the interaction of a two-level quantum system with a free Fermi gas reservoir. In the sequel we denote H λ dγ(h λ ), H S dγ(ω), H dγ(h ), and V dγ(v) = a (f)a(1) + a (1)a(f). Clearly, H λ = H 0 + λv. 4.3 Spectral theory The vacuum of Γ(h) is always an eigenvector of H λ with eigenvalue zero. The rest of the spectrum of H λ is completely determined by the spectrum of h λ and one may use the results of Sections 2 and 3.2 to characterize the spectrum of H λ. We mention several obvious facts. If the spectrum of h λ is purely absolutely continuous, then the spectrum of H λ is also purely absolutely continuous except for a simple eigenvalue at zero. H λ has no singular continuous spectrum iff h λ has no singular continuous spectrum. Let {e i } i I be the eigenvalues of h λ, repeated according to their multiplicities. The eigenvalues of H λ are given by { sp p (H λ ) = n i e i n i {0,1}, } n i < {0}. i I i I Until the end of this subsection we will discuss the fermionic quantization of the adiating Wigner-Weisskopf Atom introduced in Section 3.2. The point spectrum of H 0 consists of two simple eigenvalues {0,ω}. The corresponding normalized eigenfunctions are Ψ n = a(1) n Φ h, n = 0,1. Apart from these simple eigenvalues, the spectrum of H 0 is purely absolutely continuous and sp ac (H 0 ) is equal to the closure of the set { } n e + x i x i X, e {0,ω}, n 1. i=1 Let Λ be as in Theorem 7. Then for 0 < λ < Λ the spectrum of H λ is purely absolutely continuous except for a simple eigenvalue 0. Note that (Ψ 1 e ith λ Ψ 1 ) = (a(1)φ h e ith λ a(1)φ h ) = (a(1)φ h a(e ith λ )Φ h ) = (1 e ith λ 1).

50 Similarly, Mathematical Theory of the Wigner-Weisskopf Atom 49 (Ψ 1 (H λ z) 1 Ψ 1 ) = (1 (h λ z) 1 1). Hence, one may use directly the results (and examples) of Section 3 to describe the asymptotic of (Ψ 1 e ith λ Ψ 1 ) and the meromorphic continuation of C + z (Ψ 1 (H λ z) 1 Ψ 1 ). (56) 4.4 Scattering theory Let h λ be a WWA on h = C h. The relation yields that for φ h ac (h λ ) the limit τ t 0 τ t λ(a # (φ)) = a # (e ith0 e ith λ φ), lim τ t t 0 τλ(a t # (φ)) = a # (Ω λ φ), exists in the norm topology of CA(h). Denote τ λ,ac τ λ CA(hac(h λ )), τ,ac τ CA(hac(h )). By the intertwining property (25) of the wave operator Ω λ, the map σ + λ (a# (φ)) a # (Ω λ φ), satisfies σ + λ τt λ,ac = τt,ac σ+ λ. Hence, σ+ λ is a -isomorphism between the CA dynamical systems (CA(h ac (h λ )),τ λ,ac ) and (CA(h ac (h )),τ,ac ). This isomorphism is the algebraic analog of the wave operator in Hilbert space scattering theory and is often called the Møller isomorphism. 5 Quantum statistical mechanics of the SEBB model 5.1 Quasi-free states This subsection is a direct continuation of Subsection 4.1. Let h be a given Hilbert space and CA(h) the CA algebra over h. A positive linear functional η : CA(h) C is called a state if η(i) = 1. A physical system P is described by the CA dynamical system (CA(h), τ) if its physical observables can be identified with elements of CA(h) and if their time evolution is specified by the group τ. The physical states of P are specified by states on CA(h). If P is initially in a state described by η and A CA(h) is a given observable, then the expectation value of A at time t is η(τ t (A)). This is the usual framework of the algebraic quantum statistical mechanics in the Heisenberg picture. In the Schrödinger picture one evolves the states and keeps the observables fixed, i.e., if η is an initial state, then the state

51 50 V. Jakšić, E. Kritchevski, and C.-A. Pillet at time t is η τ t. A state η is called τ-invariant (or stationary state, steady state) if η τ t = η for all t. Let T be a self-adjoint operator on h such that 0 T I. The map η T (a (f n ) a (f 1 )a(g 1 ) a(g m )) = δ n,m det{(g i Tf j )} (57) uniquely extends to a state η T on CA(h). This state is usually called the quasi-free gauge-invariant state generated by T. The state η T is completely determined by its two point function η T (a (f)a(g)) = (g Tf). Note that if A j f j(g j ) is a finite rank operator on h, then dγ(a) = j a (f j )a(g j ), and η T (dγ(a)) = Tr (TA) = j (g j Tf j ). (58) Let (CA(h),τ) be the fermionic quantization of (h,h). The quasi-free state η T is τ-invariant iff e ith T = Te ith for all t. In particular, the quasi-free state generated by T = (h), where is a positive bounded function on the spectrum of h, is τ-invariant. The function is the energy density of this quasi-free state. Let β > 0 and µ. Of particular importance in quantum statistical mechanics is the quasi-free state associated with T = βµ (h), where the energy density βµ is given by the Fermi-Dirac distribution βµ (ε) 1 e β(ε µ) + 1. (59) We denote this state by η βµ. The pair (CA(h),τ) and the state η βµ describe free Fermi gas in thermal equilibrium at inverse temperature β and chemical potential µ. 5.2 Non-equilibrium stationary states In this and the next subsection we assume that h λ has a purely absolutely continuous spectrum. We make this assumption in order to ensure that the system will evolve towards a stationary state. This assumption will be partially relaxed in Subsection 5.4, where we discuss the effect of the eigenvalues of h λ. We do not make any assumptions on the spectrum of h. Let η T be a quasi-free state on CA(C h ) generated by T = α T. We denote by η T the quasi-free state on CA(h ) generated by T. We assume that η T is τ -invariant and denote by T,ac the restriction of T to the subspace h ac (h ). Let φ 1,,φ n h and A = a # (φ 1 ) a # (φ n ). (60)

52 Mathematical Theory of the Wigner-Weisskopf Atom 51 Since η T is τ 0 -invariant, Hence η T (τ t λ (A))) = η T(τ t 0 τ t λ (A)) = η T (a # (e ith0 e ith λ φ 1 ) a # (e ith0 e ith λ φ n )). lim η T(τλ(A)) t = η T (a # (Ω t λ φ 1) a # (Ω λ φ n)). Since the set of observables of the form (60) is dense in h, we conclude that for all A CA(h) the limit η + λ (A) = lim η T(τλ(A)), t t exists and defines a state η + λ on CA(h). Note that η+ λ is the quasi-free state generated by T + λ (Ω λ ) TΩ λ. Since an Ω λ = h ac(h 0 ) = h ac (h ), we have and so T + λ = (Ω λ ) T,ac Ω λ, (61) η + λ = η T,ac σ + λ, where σ + λ is the Møller isomorphism introduced in Subsection 4.4. Obviously, η+ λ is independent of the choice of α and of the restriction of T to h sing (h ). Since e ith λ T + λ e ith λ = (Ω λ ) e ith T,ac e ith Ω λ = T + λ, η + λ is τ λ-invariant. The state η + λ is called the non-equilibrium stationary state (NESS) of the CA dynamical system (CA(h),τ λ ) associated to the initial state η T. Note that if A j φ j(ψ j ), then, according to Equ. (58), η + λ (dγ(a)) = Tr (TΩ λ AΩ λ ) = (Ω λ ψ j TΩ λ φ j). (62) By passing to the GNS representation associated to η T one can prove the following more general result. Let N be the set of states on CA(h) which are normal with respect to η T (the set N does not depend on the choice of α). Then for any η N and A CA(h), lim t η(τt λ(a)) = η + λ (A). If T = (h ) for some bounded function on the spectrum of h, then the intertwining property of the wave operator implies that T + λ = (h λ) and hence η + λ = η (h λ ). In particular, if the reservoir is initially in thermal equilibrium at inverse temperature β > 0 and chemical potential µ, then η + λ is the quasifree state associated to (e β(hλ µ) + 1) 1, which is the thermal equilibrium state of (CA(h),τ λ ) at the inverse temperature β and chemical potential µ. This type of phenomenon, often called return to equilibrium, is an important part of the zeroth law of thermodynamics and plays a fundamental role in foundations of statistical mechanics. j

53 52 V. Jakšić, E. Kritchevski, and C.-A. Pillet 5.3 Non-equilibrium thermodynamics In this subsection we again assume that h λ has purely absolutely continuous spectrum. Concerning the reservoir, we assume that h is multiplication by x on h = L 2 (X,dµ;K), where K is a separable Hilbert space. The internal structure of is further specified by an orthogonal decomposition K = M k=1 K k. We set h k = L 2 (X,dµ;K k ) and denote by h k the operator of multiplication by x on h k. Thus, we can write M M h = h k, h = h k. (63) k=1 We interpret (63) as a decomposition of the reservoir into M independent subreservoirs 1,..., M. According to (63), we write f = M k=1 f k and we split the interaction v as v = M k=1 v k, where k=1 v k = (1 )f k + (f k )1. In the sequel we assume that f Dom h. The projection onto the subspace h k is denoted by 1 k. Set f k d dt eith λ h k e ith λ t=0 = i[h λ,h k ] = i[h S + j = λi[h k,v k ] ( hj + λv j ),hk ] (64) = λi((1 )h k f k (h k f k )1), and j k d dt eith λ 1 k e ith λ t=0 = i[h λ,1 k ] = i[h S + j = λi[1 k,v k ] ( hj + λv j ),1k ] (65) = λi((1 )f k (f k )1). The observables describing the heat and particle fluxes out of the k-th subreservoir are F k dγ(f k ) = λi(a (h k f k )a(1) a (1)a(h k f k )), J k dγ(j k ) = λi(a (f k )a(1) a (1)a(f k )). We assume that the initial state of the coupled system S + is the quasi-free state associated to T α T, where

54 Mathematical Theory of the Wigner-Weisskopf Atom 53 M M T = T k = k(h k ), k=1 and k is a bounded positive measurable function on X. Let η + λ be the NESS of (CA(h),τ λ) associated to the initial state η T. According to Equ. (61) and (58), the steady state heat current out of the subreservoir k is k=1 η + λ (F k) = Tr (T + λ f k) = Tr (T Ω λ f k(ω λ ) ) M = Tr ( j(h j )1 j Ω λ f k(ω λ ) 1 j ). j=1 Using Equ. (64) we can rewrite this expression as η + λ (F k) = 2λ Equ. (28) yields the relations M Im (1 j Ω λ h k f k j(h j )1 j Ω λ 1). j=1 ( j(h j )1 j Ω λ 1)(x) = λ j(x)f λ (x i0)f j (x), (1 j Ω λ h k f k )(x) = ( δ kj x + λ 2 F λ (x + i0)h k (x + i0) ) f j (x), where we have set H k (z) X x f k (x) 2 K k x z dµ(x). Since an Ω λ = h ac(h ), it follows that (1 j Ω λ h k f k j(h j )1 j Ω λ 1) is equal to ( λ δkj xf λ (x + i0) λ 2 F λ (x + i0) 2 H k (x + i0) ) f j (x) 2 K j j(x)dµ ac (x). X From Equ. (18) we deduce that for Lebesgue a.e. x X. Equ. (19) yields ImH k (x + i0) = πx f k (x) 2 K k dµ ac dx (x), ImF λ (x + i0) = πλ 2 F λ (x + i0) 2 f(x) 2 dµ ac K dx (x). It follows that Im (1 j Ω λ h k f k j(h j )1 j Ω λ 1) is equal to ) λ 3 ( π δkj f(x) 2 K f k (x) 2 ) 2 K fj k (x) 2 K j F λ (x+i0) x j(x)( 2 dµac X dx (x) dx. Finally, using the fact that f(x) 2 K = j f j(x) 2 K j, we obtain

55 54 V. Jakšić, E. Kritchevski, and C.-A. Pillet where η + λ (F k) = M j=1 X x( k(x) j(x))d kj (x) dx 2π, (66) ( ) 2 D kj (x) 4π 2 λ 4 f k (x) 2 h k f j (x) 2 h j F λ (x + i0) 2 dµac dx (x). (67) Proceeding in a completely similar way we obtain the following formula for the steady particle current η + λ (J k) = M j=1 X ( k(x) j(x))d kj (x) dx 2π. (68) There are several ways to interpret the quantity D kj in Equ. (67). In spectral theoretic terms, we can invoke Equ. (18) and (19) to write D kj = 4π 2 λ 2 dµλ ac dµ k,ac dµ,ac dx dµ j,ac, dx where dµ k denotes the spectral measure of h k for f k. It is also possible to relate D kj to the S-matrix associated to Ω ± λ. According to the decomposition (63), this S-matrix can be written as M M (1 k Sψ)(x) = S kj (x)(1 j ψ)(x) (1 k ψ)(x) + t kj (x)(1 j ψ)(x). j=1 Equ. (29) yields that the t-matrix defined by this relation can be expressed as and we derive that j=1 t kj (x) = 2πiλ 2 dµ,ac dx (x)f λ(x + i0)f k (x)(f j (x) ) Kj, ) D kj (x) = Tr Kj (t kj (x) t kj (x). (69) Equ. (66), (68) together with (69) are the well-known Büttiker-Landauer formulas for the steady currents. It immediately follows from Equ. (66) that M η + λ (F k) = 0, k=1 which is the first law of thermodynamics (conservation of energy). Similarly, particle number conservation M η + λ (J k) = 0, k=1

56 Mathematical Theory of the Wigner-Weisskopf Atom 55 follows from Equ. (68). To describe the entropy production of the system, assume that the k-th subreservoir is initially in thermal equilibrium at inverse temperature β k > 0 and chemical potential µ k. This means that k(x) = F(Z k (x)), where F(t) (e t +1) 1 and Z k (x) β k (x µ k ). The entropy production observable is then given by M σ β k (F k µ k J k ). k=1 The entropy production rate of the NESS η + λ is Ep(η + λ ) = η+ λ (σ) = 1 2 M k,j=1 X (Z j Z k )(F(Z k ) F(Z j ))D kj dx 2π. (70) Since the function F is monotone decreasing, Ep(η + λ ) is clearly non-negative. This is the second law of thermodynamics (increase of entropy). Note that in the case of two subreservoirs with µ 1 = µ 2 the positivity of entropy production implies that the heat flows from the hot to the cold reservoir. For k j let F kj {x X f k (x) hk f j (x) hj > 0}. The subreservoirs k and j are effectively coupled if µ ac (F kj ) > 0. The SEBB model is thermodynamically trivial unless some of the subreservoirs are effectively coupled. If k and j are effectively coupled, then Ep(η + λ ) > 0 unless β k = β j and µ k = µ j, that is, unless the reservoirs k and j are in the same thermodynamical state. 5.4 The effect of eigenvalues In our study of NESS and thermodynamics in Subsections 5.2 and 5.3 we have made the assumption that h λ has purely absolutely continuous spectrum. If X, then this assumption does not hold for λ large. For example, if X =]0, [, ω > 0, and ( 1 λ 2 > ω f(x) 2 h x dµ(x)) 1, 0 then h λ will have an eigenvalue in ],0[. In particular, if 0 f(x) 2 h x 1 dµ(x) =, then h λ will have a negative eigenvalue for all λ 0. Hence, the assumption that h λ has empty point spectrum is very restrictive, and it is important to understand

57 56 V. Jakšić, E. Kritchevski, and C.-A. Pillet the NESS and thermodynamics of the SEBB model in the case where h λ has some eigenvalues. Of course, we are concerned only with point spectrum of h λ restricted to the cyclic subspace generated by the vector 1. Assume that λ is such that sp pp (h λ ) and sp sc (h λ ) =. We make no assumption on the structure of sp pp (h λ ); in particular this point spectrum may be dense in some interval. We also make no assumptions on the spectrum of h. For notational simplicity, in this subsection we write h ac for h ac (h λ ), 1 ac for 1 ac (h λ ), etc. Let T and η T be as in Subsection 5.2 and let φ,ψ h = C h. Then, η T (τ t λ(a (φ)a(ψ))) = (e ith λ ψ Te ith λ φ) = 3 N j (e ith λ ψ,e ith λ φ), j=1 where we have set N 1 (ψ,φ) (1 ac ψ T1 ac φ), Since e ith0 T = Te ith0, we have and so N 2 (ψ,φ) 2e (1 pp ψ T1 ac φ), N 3 (ψ,φ) (1 pp ψ T1 pp φ). N 1 (e ith λ ψ,e ith λ φ) = (e ith0 e ith λ 1 ac ψ Te ith0 e ith λ 1 ac φ), lim t N 1(e ith λ ψ,e ith λ φ) = (Ω λ ψ TΩ λ φ). Since h is separable, there exists a sequence P n of finite rank projections commuting with h λ such that s lim P n = 1 pp. The iemann-lebesgue lemma yields that for all n lim t P nte ith λ 1 ac φ = 0. The relation yields that N 2 (e ith λ ψ e ith λ φ) = (e ith λ 1 pp ψ P n Te ith λ 1 ac φ) + (e ith λ (I P n )1 pp ψ Te ith λ 1 ac φ), lim t N 2(e ith λ ψ,e ith λ φ) = 0. Since N 3 (e ith λ ψ,e ith λ φ) is either a periodic or a quasi-periodic function of t, the limit lim t η T(τ t λ(a (φ)a(ψ))), does not exist in general. The resolution of this difficulty is well known to extract the steady part of a time evolution in the presence of a (quasi-)periodic component one needs to average over time. Indeed, one easily shows that

58 1 lim t t t 0 Mathematical Theory of the Wigner-Weisskopf Atom 57 N 3 (e ish λ ψ,e ish λ φ)ds = e sp p (h λ ) (P e ψ TP e φ), where P e denotes the spectral projection of h λ associated with the eigenvalue e. Hence, 1 lim t t t 0 η T (τ s λ(a (φ)a(ψ)))ds = e sp p (h λ ) (P e ψ TP e φ) + (Ω λ ψ TΩ λ φ). In a similar way one concludes that for any observable of the form the limit 1 lim t t t 0 exists and is equal to the limit 1 lim t t t 0 A = a (φ n ) a (φ 1 )a(ψ 1 ) a(ψ m ), (71) η T (τλ(a))ds s 1 = δ n,m lim t t t 0 det { (e ish λ ψ i Te ish λ φ j )}ds, det { (e ish λ 1 pp ψ i Te ish λ 1 pp φ j ) + (Ω λ 1 acψ i TΩ λ 1 acφ j ) } ds, (72) see [Kat] Section VI.5 for basic results about quasi-periodic function on. Since the linear span of the set of observables of the form (71) is dense in h, we conclude that for all A CA(h) the limit η + 1 λ (A) = lim t t t 0 η T (τ s λ(a))ds, exists and defines a state η + λ on CA(h). By the construction, this state is τ λ- invariant. η + λ is the NESS of (CA(h),τ λ) associated to the initial state η T. Note that this definition reduces to the previous if the point spectrum of h λ is empty. To further elucidate the structure of η + λ we will make use of the decomposition h = h ac h pp. (73) The subspaces h ac and h pp are invariant under h λ and we denote the restrictions of τ λ to CA(h ac ) and CA(h pp ) by τ λ,ac and τ λ,pp. We also denote by η + λ,ac and η + λ,pp the restrictions of η+ λ to CA(h ac) and CA(h pp ). η + λ,ac is the quasi-free state generated by T + λ (Ω λ ) TΩ λ. If A is of the form (71) and φ j,ψ i h pp, then η + λ,pp (A) = δ 1 n,m lim t t t 0 det{(e ish λ ψ i Te ish λ φ j )}ds. Clearly, η + λ,ac is τ λ,ac invariant and η + λ,pp is τ λ,pp invariant. Expanding the determinant in (72) one can easily see that η + λ,ac and η+ λ,pp uniquely determine η+ λ.

59 58 V. Jakšić, E. Kritchevski, and C.-A. Pillet While the state η + λ,pp obviously depends on the choice of α and on T hsing(h ) in T = α T, the state η + λ,ac does not. In fact, if η is any initial state normal w.r.t. η T, then for A CA(h ac ), lim t η(τt λ(a)) = η + λ,ac (A). For a finite rank operator A j φ j(ψ j ) one has η + λ (dγ(a)) = j η + λ (a (φ j )a(ψ j )), and so η + λ (dγ(a)) = j (P e ψ j TP e φ j ) + (Ω λ ψ j TΩ λ φ j). e sp p (h λ ) The conclusion is that in the presence of eigenvalues one needs to add the term (P e ψ j TP e φ j ), j e sp p (h λ ) to Equ. (62), i.e., we obtain the following formula generalizing Equ. (62), η + λ (dγ(a)) = Tr T P e AP e + Ω λ AΩ λ. (74) e sp p (h λ ) Note that if for some operator q, A = i[h λ,q] in the sense of quadratic forms on Dom h λ, then P e AP e = 0 and eigenvalues do not contribute to η + λ (dγ(a)). This is the case of the current observables dγ(f k ) and dγ(j k ) of Subsection 5.3. We conclude that the formulas (66) and (68), which we have previously derived under the assumption sp sing (h λ ) =, remain valid as long as sp sc (h λ ) =, i.e., they are not affected by the presence of eigenvalues. 5.5 Thermodynamics in the non-perturbative regime The results of the previous subsection can be summarized as follows. If sp sc (h λ ) = and sp pp (h λ ) then, on the qualitative level, the thermodynamics of the SEBB model is similar to the case sp sing (h λ ) = 0. To construct NESS one takes the ergodic averages of the states η T τλ t. The NESS is unique. The formulas for steady currents and entropy production are not affected by the point spectra and are given by (66), (68), (70) and (67) or (69) for all λ 0. In particular, the NESS and thermodynamics are well defined for all λ 0 and all ω. One can proceed further along the lines of [AJPP1] and study the linear response theory of the SEBB model (Onsager relations, Kubo formulas, etc) in the non-perturbative

60 Mathematical Theory of the Wigner-Weisskopf Atom 59 regime. Given the results of the previous subsection, the arguments and the formulas are exactly the same as in [AJPP1] and we will not reproduce them here. The study of NESS and thermodynamics is more delicate in the presence of singular continuous spectrum and we will not pursue it here. We wish to point, however, that unlike the point spectrum, the singular continuous spectrum can be excluded in generic physical situations. Assume that X is an open set and that the absolutely continuous spectrum of h is well-behaved in the sense that Im F (x + i0) > 0 for Lebesgue a.e. x X. Then, by the Simon-Wolff theorem 5, h λ has no singular continuous spectrum for Lebesgue a.e. λ. If f is a continuous function and dµ = dx, then h λ has no singular continuous spectrum for all λ. 5.6 Properties of the fluxes In this subsection we consider an SEBB model without singular continuous spectrum, i.e, we assume that sp sc (h λ ) = for all λ and ω. We will study the properties of the steady currents as functions of (λ,ω). For this reason, we will again indicate explicitly the dependence on ω. More precisely, in this subsection we will study the properties of the function (λ,ω) η + λ,ω (F), (75) where F is one of the observables F k or J k for a given k. For simplicity of exposition, besides Assumption (A1) we assume that the functions g j (t) e itx f j (x) 2 K j dx, X are in L 1 (,dt) and that f(x) K is non-vanishing on X. We also assume that the energy densities j(x) of the subreservoirs are bounded continuous functions on X and that the functions (1 + x ) j(x) are integrable on X. According to Equ. (66), (68) and (67), one has η + λ,ω (F) = M 2πλ4 j=1 X f k (x) 2 K k f j (x) 2 K j F λ (x + i0) 2 x n ( k(x) j(x))dx, where n = 0 if F = J k and n = 1 if F = F k. Obviously, the function (75) is real-analytic on \X and for a given ω X, η + λ,ω (F) = O(λ4 ), (76) as λ 0. The function (75) is also real-analytic on \{0}. For ω X, Lemma 3 shows that lim λ 2 η + ω,λ (F) = 2π M f k (ω) 2 K k f j (ω) 2 K j λ 0 f(ω) 2 ω n ( k(ω) j(ω)). (77) j=1 K

61 60 V. Jakšić, E. Kritchevski, and C.-A. Pillet Comparing (76) and (77) we see that in the weak coupling limit we can distinguish two regimes: the conducting regime ω X and the insulating regime ω X. Clearly, the conducting regime coincides with the resonance regime for h λ,ω and, colloquially speaking, the currents are carried by the resonance pole. In the insulating regime there is no resonance for small λ and the corresponding heat flux is infinitesimal compared to the heat flux in the conducting regime. For x X one has Hence, and so λ 4 F λ (x + i0) 2 = (ω x λ 2 e F (x + i0)) 2 + λ 4 π 2 f(x) 4. K sup λ 4 F λ (x + i0) 2 = λ π λ 4 M f j (x) 2 K j j=1 λ 4 f k (x) 2 K k f j (x) 2 K j F λ (x + i0) 2 1 π 2. 2, (78) This estimate and the dominated convergence theorem yield that for all ω, M lim λ η+ λ,ω (F) = 2π f k (x) 2 K k f j (x) 2 K j j=1 X F (x + i0) 2 x n ( k(x) i(x))dx. (79) Thus, the steady currents are independent of ω in the strong coupling limit. In the same way one shows that lim ω η+ λ,ω (F) = 0, (80) for all λ. The cross-over between the weak coupling regime (77) and the large coupling regime (79) is delicate and its study requires detailed information about the model. We will discuss this topic further in the next subsection. We finish this subsection with one simple but physically important remark. Assume that the functions C j (x) 2π f k (x) 2 K k f j (x) 2 K j x n ( k(x) j(x)), are sharply peaked around the points x j. This happens, for example, if all the reservoirs are at thermal equilibrium at low temperatures. Then, the flux (75) is well approximated by the formula M η + λ,ω (F) λ 4 F λ (x j ) 2 j=1 X C j (x)dx, and since the supremum in (78) is achieved at ω = x + λ 2 e F λ (x + i0), the flux (75) will be peaked along the parabolic resonance curves ω = x j + λ 2 e F λ (x j + i0).

62 Mathematical Theory of the Wigner-Weisskopf Atom Examples We finish these lecture notes with several examples of the SEBB model which we will study using numerical calculations. For simplicity, we will only consider the case of two subreservoirs, i.e., in this subsection K = C 2 = C C. We also take f(x) = f 1 (x) f 2 (x) 1 2 ( f0 (x) 0 ) 1 2 ( 0 f 0 (x) ), so that f 1 (x) 2 K 1 = f 2 (x) 2 K 2 = 1 2 f(x) 2 K = 1 2 f 0(x) 2. Example 1. We consider the fermionic quantization of Example 1 in Subsection 3.5, i.e., h = L 2 (]0, [,dx; C 2 ) and f 0 (x) = π 1/2 (2x) 1/4 (1 + x 2 ) 1/2. We put the two subreservoirs at thermal equilibrium j(x) = e βj(x µj), where we set the inverse temperatures to β 1 = β 2 = 50 (low temperature) and the chemical potentials to µ 1 = 0.3, µ 2 = 0.2. We shall only consider the particle flux (n = 0) in this example. The behavior of the heat flux is similar. The function C 2 (x) = 2π f 1 (x) 2 K 1 f 2 (x) 2 K 2 ( 1(x) 2(x)) = x( 1(x) 2(x)) π(1 + x 2 ) 2, plotted in Figure 11, is peaked at x In accordance with our discussion in the previous subsection, the particle current, represented in Figure 12, is sharply peaked around the parabola ω = x + 2λ 2 (1 x)/(1 + x 2 ) (dark line). The convergence to an ω-independent limit as λ and convergence to 0 as ω are also clearly illustrated. Example 2. We consider now the heat flux in the SEBB model corresponding to Example 2 of Subsection 3.5. Here h = L 2 (] 1,1[,dx; C 2 ), 2 f 0 (x) = π (1 x2 ) 1/4, and we choose the high temperature regime by setting β 1 = β 2 = 0.1, µ 1 = 0.3 and µ 2 = 0.2. Convergence of the rescaled heat flux to the weak coupling limit (77) is illustrated in Figure 13. In this case the function C 2 is given by C 2 (x) = 2 π x(1 x2 )( 1(x) 2(x)),

63 62 V. Jakšić, E. Kritchevski, and C.-A. Pillet 0.2 C x Fig. 11. The function C 2(x) in Example 1. Fig. 12. The particle flux in Example 1.

64 Mathematical Theory of the Wigner-Weisskopf Atom 63 3 x Heat flux / λ ω λ 1 Fig. 13. The rescaled heat flux (weak coupling regime) in Example 2. x C x Fig. 14. The function C 2(x) in Example 2.

65 64 V. Jakšić, E. Kritchevski, and C.-A. Pillet Fig. 15. The heat flux in Example 2. and is completely delocalized as shown in Figure 14. Even in this simple example the cross-over between the weak and the strong coupling regime is difficult to analyze. This cross-over is non trivial, as can be seen in Figure 15. Note in particular that the function λ η + λ,ω (F) is not necessarily monotone and may have several local minima/maxima before reaching its limiting value (79) as shown by the section ω = 0.5 in Figure 15. Example 3. In this example we will discuss the large coupling limit. Note that in the case of two subreservoirs Equ. (79) can be written as lim λ η+ λ,ω (F) = 1 sin 2 θ(x)x n ( 1(x) 2(x))dx, (81) 2π X where θ(x) Arg(F (x + i0)). Therefore, large currents can be obtained if one of the reservoir, say 1, has an energy distribution concentrated in a region where Im F (x+i0) e F (x+i0) while the energy distribution of 2 is concentrated in a region where Im F (x + i0) e F (x + i0). As an illustration, we consider the SEBB model corresponding to Example 3 of Subsection 3.5, i.e., h = L 2 (] 1,1[,dx; C 2 ) and 1 f 0 (x) = π x(1 x2 ) 1/4. From Equ. (54) we obtain that