CUBO A Mathematical Journal Vol.14, No¯ 03, (01–39). October 2012 Fundamentals of scattering theory and resonances in quantum mechanics Peter D. Hislop Department of Mathematics, University of Kentucky, Lexington, Kentucky 40506-0027, USA email: hislop@ms.uky.edu ABSTRACT We present the basics of two-body quantum-mechanical scattering theory and the the- ory of quantum resonances. The wave operators and S-matrix are constructed for smooth, compactly-supported potential perturbations of the Laplacian. The meromor- phic continuation of the cut-off resolvent is proved for the same family of Schrödinger operators. Quantum resonances are defined as the poles of the meromorphic con- tinuation of the cut-off resolvent. These are shown to be the same as the poles of the meromorphically continued S-matrix. The basic problems of the existence of resonances and estimates on the resonance counting function are described and recent results are presented. RESUMEN Presentamos los conceptos básicos de la teoŕıa de dispersión cuanto-mecánica de dos cuerpos y la teoŕıa de resonancias cuánticas. El operador de ondas y la matriz S se construyen para perturbaciones del potencial suaves y de soporte compacto del Lapla- ciano. La continuación meromórfica de la resolvente truncada se prueba para la misma familia de operadores de Schrdinger. Las resonancias cuánticas se definen como los po- los de la continuación meromórifca de la resolvente truncada. Se muestra que ellas son las mismas que los polos de la matriz S continuada meromórficamente. Los problemas básicos de la existencia de resonancias y las estimaciones de la función de conteo de la resonancia se describen y resultados recientes se presentan. Keywords and Phrases: Scattering theory, resonances, Schrödinger equation, wave operators, quantum mechanics 2010 AMS Mathematics Subject Classification: 35J10, 35P25, 35Q40,47A40, 47A55, 81U05, 81U20 2 Peter D. Hislop CUBO 14, 3 (2012) 1 Introduction: Schrödinger operators The purpose of these notes is to present the necessary background and the current state-of-the- art concerning quantum resonances for Schrödinger operators in a simple, but nontrivial, setting. The unperturbed Hamiltonian H0 = −∆ is the Laplacian on L 2(Rd). In quantum mechanics, the Schrödinger operator or Hamiltonian H0 represents the kinetic energy operator of a free quantum particle. Many interactions are represented by a potential V that is a real-valued function with V ∈ L∞0 (R d), the essentially bounded functions of compact support. Occasionally, we need the potential to have some derivatives and this will be indicated. If, for example, the potential V ∈ C∞0 (Rd), then all the results mentioned here hold true. The perturbed Hamiltonian is HV = −∆ + V. A fundamental property shared by both Hamiltonians is self-adjointness. The unperturbed Hamiltonian H0 is self-adjoint on its natural domain H 2(Rd), the Sobolev space of order two, which is dense in L2(Rd). The self-adjoint operator H0 is the generator of a one-parameter strongly- continuous unitary group t ∈ R → U0(t) = e−iH0t. The potential V is relatively H0-bounded with relative bound zero. By the Kato-Rellich Theorem [14, Theorem 13.5], the perturbed operator HV is self-adjoint on the same domain H 2(Rd). This self-adjoint operator generates a one-parameter strongly-continuous unitary group t ∈ R → UV(t) = e −iHV t. The unitary groups U0(t) and UV(t) provide solutions to the initial value problem for the Schrödinger operator in L2(Rd). For example, the solution to i ∂ψ(t) ∂t = HVψ(t), ψ(0) = ψ0 ∈ H2(Rd), (1.1) is formally given by ψ(t) = UV(t)ψ0. In this way, the unitary group UV(t) provides the time- evolution of the initial state ψ0. Scattering theory seeks to provide a description of the perturbed time-evolution UV(t) in terms of the simpler (as we will show below) time-evolution U0(t). Although we will work on the Hilbert space L2(Rd), much of scattering theory can be formulated in a more abstract setting. Consequently, we will often write H for a general Hilbert space. Suppose we take a state f ∈ H and consider the interacting time-evolution UV(t)f. What is the behavior of UV(t)f as t → ±∞? There is one exactly solvable case, although, as we will see, it is not too interesting. Suppose that f is an eigenfunction of HV with eigenvalue E so that f satisfies the eigenvalue equation HVf = Ef. Then, the time evolution is rather simple since UV(t)f = e −itEf, as is easily verified by differentiation. We do not expect this simple oscillating state to be approximated by the free dynamics so we should eliminate these states from our consideration. Let Hcont(HV) be the closed subspace of H orthogonal to the span of all the eigenfunctions of HV. We will call these states the scattering states of HV. Given f ∈ Hcont(HV), can we find a state f+ ∈ H so that as time runs to plus infinity, the state UV(t)f looks approximately like the free time-evolved state U0(t)f+? In particular, we ask if given f ∈ Hcont(HV), does there CUBO 14, 3 (2012) Scattering theory and resonances ... 3 exist a state f+ ∈ H so that UV(t)f − U0(t)f+ → 0, as t → +∞. (1.2) When it is possible to find such a vector f+, we have a simpler description of the dynamics UV(t) generated by HV in terms of the free dynamics U0(t) generated by H0. We can also pose the question concerning the existence of a state f− so that (1.2) holds for t → −∞ with f− replacing f+. We understand (1.2) to mean convergence as a vector in H, that is lim t→+∞ ‖UV(t)f − U0(t)f+‖H = 0. (1.3) Note that if f+ is an eigenfunction of H0 with eigenvalue E, that is H0f+ = Ef+, then U0(t)f+ = e−itEf+, we would not expect the limit (1.3) to exist. Hence, we want f+ to be a state with nontrivial free time evolution. This means that we want f+ to be a scattering state for H0, that is, f+ ∈ Hcont(H0). For our specific example, H0 = −∆, there are no eigenfunctions so Hcont(H0) = H. Because the operators U0(t) and UV(t) are unitary, the limit in (1.3) is equivalent to lim t→+∞ ‖f − UV(t)∗U0(t)f+‖H = 0. (1.4) Since H0 = −∆ has no eigenvalues and only continuous spectrum, we expect that the limit lim t→+∞ UV(t) ∗U0(t)f+ = f, (1.5) if it exists, should exist for all states f+ ∈ H. Similarly, we might expect that the limit lim t→−∞ UV(t) ∗U0(t)f− = f, (1.6) exists for all f ∈ H. We will prove in section 2 that these limits do exist and define bounded operators Ω±(HV,H0) on H called the wave operators for the pair (H0,HV). If we consider the original problem: Given f ∈ Hcont(HV), find f± so that the limit in (1.2), and the similar limit for t → −∞, it might seem strange that we consider Ω±(HV,H0) rather than the limit of the operators in the other order, namely, U0(t) ∗UV(t) on the scattering states of HV. As we will see, it is much more difficult to prove the existence of the latter limit. Let us consider, however, the inner product (g,Ω±(HV,H0)f) for g in the range of the wave operator Ω±(HV,H0). Using the definition and unitarity of the time evolution groups, we have (g,Ω±(HV,H0)f) = lim t→±∞ (g,UV(t) ∗U0(t)f) = lim t→±∞ (U0(t) ∗UV(t)g,f) = (Ω±(HV,H0) ∗g,f). (1.7) Since this holds for all f ∈ H, it follows that for g ∈ Ran Ω±(HV,H0), lim t→±∞ U0(t) ∗UV(t)g = Ω±(HV,H0) ∗g. (1.8) 4 Peter D. Hislop CUBO 14, 3 (2012) Comparing this to (1.3), it is clear that we obtain the desired states by f± = Ω±(HV,H0) ∗f. As we will see in Proposition 4, the existence of the strong limits of U0(t) ∗UV(t) on the scattering states of HV as t → ±∞ is related to asymptotic completeness. The existence of the wave operators Ω±(HV,H0) allow us to define states f± for any scattering state f ∈ Hcont(HV). The map S : f− → f+ plays an important role in scattering theory. This map is called the S-operator for the pair (H0,HV). Two technical remarks. 1) The subspace of scattering states Hcont(HV) is technically the absolutely continuous spectral subspace of HV (see section 8.1). The unperturbed operator H0 = −∆ has spectrum equal to the half-line [0,∞) and is purely absolutely continuous. In our setting, the perturbed operator HV has only absolutely continuous spectrum and possibly eigenvalues. In general, it is a difficult task to prove the absence of singular continuous spectrum. There is an orthogonal spectral projector Econt(HV) so that Hcont(HV) = Econt(HV)H. We will use either notation interchangeably. 2) The type of convergence described in (1.5) and (1.6) is called strong convergence of operators. We say that a sequence of bounded operators An on H converges strongly to A ∈ B(H) if for all f ∈ H, we have limn→∞ Anf = Af. 2 Fundamentals of two-body scattering theory The basic objects of scattering theory are the wave operators and the scattering operator. The crucial property of the wave operators Ω±(HV,H0) is called asymptotic completeness. This condi- tion guarantees the unitarity of the scattering operator. On the level of spectral theory, asymptotic completeness means that the restrictions of the operators H0 and HV to their absolutely continuous subspaces are unitarily equivalent. From this viewpoint, scattering theory is a tool for studying the absolutely continuous spectral components of the pair (H0,HV) of self-adjoint operators. The theory has been developed to a very abstract level and the reader is referred to the references for further details (for example, [32, 45]). 2.1 Wave operators Another way to write (1.4) is lim t→∞ UV(t) ∗U0(t)f+ = f, (2.1) so one of our first tasks is to ask whether the limit on the left side of (2.1) exists. Proposition 1. Suppose that the real-valued potential V ∈ L∞0 (Rd) and that d ≥ 3. For any f ∈ H, the limit lim t→∞ UV(t) ∗U0(t)f (2.2) exists. This limit defines a bounded linear transformation Ω+(HV,H0) with ‖Ω+(HV,H0)‖ = 1. CUBO 14, 3 (2012) Scattering theory and resonances ... 5 The linear operator Ω+(HV,H0) is called a wave operator. We can also consider the limit in (2.1) as time runs to minus infinity. We introduce another wave operator Ω−(HV,H0) defined by s − lim t→−∞ UV(t) ∗U0(t) ≡ Ω−(HV,H0), (2.3) when the strong limit exists. Of course, we can introduce another pair of wave operators by interchanging the order of HV and H0. We will consider these wave operators Ω±(H0,HV) in section 2.3 when we discuss asymptotic completeness. We will see that it is much more difficult to prove the existence of these wave operators. We prove Proposition 1 using the classic Cook-Hack method (see, for example, [31, section XI.4]). In the following proof, we drop the Hamiltonians from the notation for the wave operators and simply write Ω± for the wave operators Ω±(HV,H0). Proof. 1. The proof of Proposition 1 relies on an explicit estimate for the free propagation given by U0(t). For any f ∈ L1(Rd) ∩ L2(Rd), and for t 6= 0, we have ‖U0(t)f‖∞ ≤ Cd td/2 ‖f‖1. (2.4) This estimate is proved (see [1, Lemma 3.12]) using an explicit formula for U0(t)f, t 6= 0. For any f ∈ L1(Rd) ∩ L2(Rd), we have (U0(t)f)(x) = ( 1 4πit )d/2 ∫ Rd ei|x−y| 2/(4t) f(y) ddy. (2.5) This representation is based on the fact that the Fourier transform (see (3.4) and (3.4)) of the action of the free propagation group is (F(U0(t)f))(k) = e −i|k|2t(Ff)(k). (2.6) Formally, formula (2.5) is obtained by computing the inverse Fourier transform. This involves a singular integral: ∫ Rd eik·(x−y)e−i|k| 2t ddk. (2.7) This integral can be done by first regularizing the integrand by replacing t by t − iǫ, for ǫ > 0. This results in a Gaussian function of k, and the Fourier transform is explicitly computable. It is also a Gaussian function. One can then take ǫ → 0 and recover the formula (2.5) since f ∈ L2(Rd) ∩ L1(Rd) guarantees convergence of the integral. 2. Given this result (2.4), we proceed as follows. Let us define Ω(t) by Ω(t) ≡ UV(t)∗U0(t). (2.8) From this definition, we compute for any f ∈ L1(Rd) ∩ L2(Rd) (Ω(t) − 1)f = ∫t 0 d ds UV(s) ∗U0(s)f ds = i ∫t 0 UV(s) ∗VU0(s)f ds. (2.9) 6 Peter D. Hislop CUBO 14, 3 (2012) Since U0(t) maps L 2(Rd) to itself and V ∈ L∞0 (Rd), the integral on the right is well-defined. To prove the existence of the limit, consider 0 << t1 < t2 and note that from (2.9) and the estimate (2.4), we have ∥ ∥ ∥ ∥ ∫t2 t1 UV(s) ∗VU0(s)f ds ∥ ∥ ∥ ∥ ≤ ‖V‖L2(Rd) ∫t2 t1 ‖U0(s)f‖L∞(Rd) ds ≤ Cd‖V‖L2(Rd) ‖f‖1 ∫t2 t1 s−d/2 ds ≤ C̃d‖V‖L2(Rd)‖f‖1(t 1−d/2 1 − t 1−d/2 2 ). (2.10) It follows that for d ≥ 3, we have the bound ‖(Ω(t2) − Ω(t1)f‖ ≤ C̃d‖V‖‖f‖1(t1−d/21 − t 1−d/2 2 ). (2.11) Consequently, for any sequence tn → ∞, the sequence of vectors Ω(tn)f is a norm-convergent Cauchy sequence so limt→∞ Ω(t)f ≡ f̃+ exists. We must show that the map f ∈ L1(Rd)∩L2(Rd) → f̃+ defines a linear bounded operator. Since ‖Ω(tn)f‖ ≤ ‖f‖L2(Rd), for any tn, it follows that ‖f̃+‖ ≤ ‖f‖. This defines Ω+ : f → f̃+ on a dense domain L1(Rd) ∩ L2(Rd). A densely-defined bounded linear operator can be extended to H without increasing the norm. Finally, one verifies that s − limt→∞ Ω(t) = Ω+ by approximating any g ∈ H by a sequence in L1(Rd) ∩ L2(Rd) and using a triangle inequality argument. The simplicity of this proof relies on the estimate (2.4) for the group U0(t). It is more difficult to consider the strong limit of U0(t) ∗UV(t) since no general formula is available for UV(t)f. 2.2 Properties of wave operators The wave operators Ω± are bounded operators on H with ‖Ω±‖ = 1. They satisfy a number of important properties. First, they are partial isometries in the sense that E± ≡ Ω∗±Ω± are orthogonal projections. In our case, E± = I, the identity operator on H. In the general case, the operator E± is the projection onto the continuous subspace of H0. For any f,g ∈ H, we have (Ω±f,Ω±g) = (f,E±g) = (E±f,E±g), (2.12) so that ‖Ω±f‖ = ‖E±f‖. (2.13) It follows that Ω± are isometries on E±H and that the kernel of Ω± is (1 − E±)H. We have that Ω±E± = Ω±. The subspaces of H given by E±H are called the initial spaces of the partial isometries Ω±. Second, the adjoints Ω∗± are partial isometries. Since (Ω ∗ ±) ∗Ω∗± = Ω±Ω ∗ ±, the operator F± ≡ Ω±Ω∗± satisfies F2± = Ω±(Ω∗±Ω±)Ω∗± = Ω±E±Ω∗± = F±, and in a similar manner F∗± = F±, CUBO 14, 3 (2012) Scattering theory and resonances ... 7 so F± are orthogonal projections. It follows that F±Ω ∗ ± = Ω ∗ ± and that ‖Ω∗±f‖ = ‖F±f‖. One can show that F± are the orthogonal projections onto the closed ranges of the wave operators Ran Ω± = F±H. The subspaces F±H are called the final subspaces of the partial isometries Ω±. Proposition 2. The wave operators satisfy the following intertwining relations: Ω±U0(t) = UV(t)Ω± U0(t)Ω ∗ ± = Ω ∗ ±UV(t). (2.14) Proof. These relations follow from the existence of the wave operators and the simple properties of the unitary evolution groups. For any f ∈ H, we have UV(t)Ω+f = lim s→∞ UV(t)UV (s) ∗U0(s)f = lim s→∞ [UV(s − t) ∗U0(s − t)]U0(t)f = lim u→∞ [UV(u) ∗U0(u)]U0(t)f = Ω+U0(t)f, (2.15) proving the first intertwining relation. The second is proven in the same manner. 2.3 Asymptotic completeness The existence of the wave operators Ω±(HV,H0) means the existence of a orthogonal projectors onto the initial space E± ≡ Ω±(HV,H0)∗Ω±(HV,H0) = I and final subspaces F± ≡ Ω±(HV,H0)Ω±(HV,H0) ∗ that are the ranges of the wave operators Ω±(HV,H0). The range of the wave operators must be contained in the continuous spectral subspace of HV. Definition 3. The pair of self-adjoint operators (H0,HV) is said to be asymptotically complete if F−H = F+H = Econt(HV)H, that is, if Ran Ω− = Ran Ω+ = Econt(HV)H. In our situation, with H0 = −∆, the spectrum of H0 is purely absolutely continuous and Econt(H0)H = H. In particular, E± = 1H. Also, neither operator H0 nor HV has singular continuous spectrum. In more general situations, one needs to prove that the perturbed operator HV has no singular continuous spectrum. In these more general cases, the subspace Hcont(HV) must be taken as the absolutely continuous spectral subspace. One can also consider wave operators Ω±(H0,HV) defined by switching the order of the unitary operators in (2.2): Ω±(H0,HV) ≡ s − lim t→±∞ U0(−t)UV(t)Econt(HV). (2.16) At first sight, it would seem that the existence of these wave operators would be equivalent to the existence of Ω±(HV,H0). However, we have no explicit control over the dynamics generated by HV such as formula (2.5). Consequently, it is difficult to use the Cook-Hack method to prove the existence of the wave operators Ω±(H0,HV). In fact, the existence of the wave operators Ω±(H0,HV) is equivalent to asymptotic completeness. 8 Peter D. Hislop CUBO 14, 3 (2012) Proposition 4. Suppose that the wave operators Ω±(HV,H0) exist. Then the pair of operators (H0,HV) are asymptotically complete if and only if the wave operators Ω±(H0,HV) exist. Proof. 1. Suppose that both sets of wave operators exist. Then, we know that the projection Econt(HV) = Ω±(HV,HV). But, we have UV(−t)UV(t) = UV(−t)U0(t) · U0(−t)UV(t), (2.17) from which it follows that Ω±(HV,HV) = Ω±(HV,H0)Ω∓(H0,HV). (2.18) This implies that Hcont(HV) ⊂ Ran Ω±(HV,H0). Since the existence of Ω±(HV,H0) means that Ran Ω±(HV,H0) ⊂ Hcont(HV), these two inclusions mean that Ran Ω+(HV,H0) = Ω−(HV,H0) = Hcont(HV). 2. To prove the other implication, we assume that the wave operators Ω±(HV,H0) exist and are asymptotically complete. Then, for any φ ∈ Hcont(HV), there exists a ψ ∈ H so that φ = Ω+(HV,H0)ψ. This means that U0(t)ψ − UV(t)φ converges to zero as t → +∞. By unitarity of the operator U0(t), this means that limt→+∞ U0(−t)UV(t)φ = ψ for all φ ∈ Hcont(HV). This implies the existence of Ω+(H0,HV). The proof of the existence of the other wave operator is analogous. We now turn to proving the existence of the wave operators Ω±(H0,HV). Many methods have been developed over the years in order to do this. The classic result of Birman [31, Theorem XI.10] is perhaps the simplest to apply to our simple two-body situation. There are more elegant and far-reaching methods. The Enss method, in particular, is based on a beautiful phase-space analysis of the scattering process. A thorough account of the Enss method may be found in Perry’s book [27]. Perry combined the Enss method with the Melin transform in [26] to present a new, clear, and short proof of asymptotic completeness for two-body systems more general than those considered here. Finally, the problem of asymptotic completeness for N-body Schrödinger operators with short-range, two-body potentials, was solved by Sigal and Soffer [38]. They developed a very useful technique of local decay estimates. In preparation, we recall that a bounded operator K is in the trace class if the following condition is satisfied. The singular values of a compact operator A are given by µj(A) = √ λj(A ∗A), where {λj(B)} are the eigenvalues of B. We say that K is in the trace class if ∑ j µj(K) < ∞. We say that K is in the Hilbert-Schmidt class if ∑ j µj(K) 2 < ∞. We refer to [29] or [39] for details concerning the von Neumann-Schatten trace ideals of bounded of operators. Theorem 5. Let V ∈ L∞0 (Rd) be a real-valued potential and d ≥ 3. Then the pair (H0,HV)is asymptotically complete. Proof. 1. By Proposition 4, it suffices to prove that Ω±(H0,HV) exist since we know from Proposi- tion 1 that the wave operators Ω±(HV,H0) exist. For any interval I ⊂ R and self-adjoint operator CUBO 14, 3 (2012) Scattering theory and resonances ... 9 A, let EI(A) denote the spectral projection for A and the interval I. In the first step, we note that EI(H0)VEI(HV),EI(HV)VEI(H0) ∈ I1. (2.19) The trace class property of these operators is easily demonstrated by proving that |V|1/2R0(i) k is a Hilbert-Schmidt operator for k > d/2 and noting that EI(H0)R0(i) −k is a bounded operator. 2. Next, we need the following result called Pearson’s Theorem in [31, Theorem XI.7]. Let a > 0 and define the bounded operator Ja ≡ E(−a,a)(H0)E(−a,a)(HV). The trace class property (2.19) means that H0Ja − JaHV ∈ I1. The main result of [31, Theorem XI.7] is that s − lim t→±∞ U0(t) ∗JaUV(t)Econt(HV) (2.20) exists. Let 0 < a0 < a and choose φ ∈ E(−a0,a0)(HV)Econt(HV)H. We then have U0(t) ∗E(−a,a)(H0)UV(t)φ = U0(t) ∗JaUV(t)φ, (2.21) so by (2.20), the strong limit of the term on the left in (2.21) exists. 3. We can now write the expression that gives the wave operator acting on any φ ∈ E(−a0,a0)(HV)Econt(HV)H: U0(t) ∗UV(t)φ = U0(t) ∗[E(−a,a)(H0) + ER\(−a,a)(H0)]UV(t)φ. (2.22) Since the strong limit of the first term on the right in (2.22) exists by (2.21), it suffices to prove that lim a→∞ { sup t∈R ‖U0(t)∗ER\(−a,a)(H0)UV(t)φ‖ } = 0. (2.23) Once this is proven, we can first take a → ∞ and then a0 → ∞ so that the limit in (2.22) holds for any φ ∈ Econt(HV)H. 4. To prove (2.23), we need some estimates. Let f(s) = s2 + 1 ≥ 1. The fact that V is relatively H0-bounded means that ‖f(HV)f(H0)−1‖ < C1 < ∞. (2.24) Next, recall that φ ∈ E(−a0,a0)(HV)H, for 0 < a0 < a, so that ‖f(HV)UV(t)φ‖ ≤ sup |s|≤a0 f(s) = a20 + 1 < ∞. (2.25) Finally, since f is invertible, we have ‖f(H0)−1ER\(−a,a)(H0) ≤ [ inf |s|≥a0 f(s) ]−1 = (a2 + 1)−1. (2.26) Note that this vanishes as a → ∞. 10 Peter D. Hislop CUBO 14, 3 (2012) 5. Returning to (2.23), we write the norm as ‖U0(t)∗ER\(−a,a)(H0)UV(t)φ‖ ≤ ‖U0(t)∗ · f(H0)−1ER\(−a,a)(H0) · f(H0)f(HV)−1 · f(HV)UV(t)φ‖ ≤ ‖f(H0)−1ER\(−a,a)(H0)‖ ‖f(H0)f(HV)−1‖ ‖f(HV)UV(t)φ‖ ≤ C1(a20 + 1)(a2 + 1)−1, (2.27) independently of t. Taking a → ∞ proves (2.23). The asymptotic completeness of (H0,HV) means that the absolutely continuous parts of each operator are unitarily equivalent. Recall that our condition on the real-valued potential V ∈ L∞0 (Rd) means that V(H0 + i)−1 is compact. By Weyl’s Theorem (see, for example, [14, Theorem 14.6]), the essential spectrum of HV is the same as the essential spectrum of H0 that is [0,∞). Hence, the perturbation can add at most a discrete set of isolated eigenvalues with finite multiplicities. The property of asymptotic completeness goes beyond this and establishes the unitary equivalence of the absolutely continuous components. 3 The scattering operator The existence of the wave operators Ω±(HV,H0) guarantees the existence of the asymptotic states f±. For any f ∈ Ran Ω±(HV,H0) ⊂ Econt(HV)H, we have f± = Ω±(HV,H0)∗f. The S-operator maps f− to f+. It is a bounded operator on L 2(Rd). Furthermore, the S-operator commutes with the free time evolution U0(t). This allows for a reduction of the S-operator to a family of operators S(λ) defined on L2(Sd−1) called the S-matrix. 3.1 Basic properties of the S-operator An important use of the wave operators is the construction of the S-operator on H. For any f ∈ Ran Ω±(HV,H0), we have from section 2.1 that f± = Ω±(HV,H0)∗f, or, for example, f = Ω−f−. As a result, we can compute a formula for the map f− → f+ in terms of the wave operators: Sf− = f+ = Ω ∗ +f = Ω ∗ +Ω−f−. Consequently, the S-operator is defined as the bounded operator S ≡ Ω∗+Ω− : H → H. (3.1) Proposition 6. Suppose Ran Ω− ⊂ Ran Ω+. Then, the scattering operator is a partial isometry on L2(Rd). Proof. To prove this, we need to show that S∗S is an orthogonal projection. This follows from the properties of the wave operators: S∗S = (Ω∗+Ω−) ∗ (Ω∗+Ω−) = Ω ∗ −[Ω+Ω ∗ +]Ω− = Ω ∗ −F+Ω−. (3.2) CUBO 14, 3 (2012) Scattering theory and resonances ... 11 Since we assume that Ran Ω− ⊂ F+H, we have F+Ω− = Ω−, so from (3.2), S∗S = F−, an orthogonal projection. If H0 = −∆, this operator F− is the identity operator on H. Since Ran S ⊂ Ran Ω∗+ ⊂ Econt(H0)H, we have that S : Econt(H0)H → Econt(H0)H. An essential property of the S-operator is that it commutes with the free time evolution, as stated in the following proposition. Proposition 7. The S-operator commutes with the free time evolution: [S,U0(t)] ≡ SU0(t) − U0(t)S = 0. Consequently, the S-operator satisfies E0(I)S = SE0(I), where E0(I) is the spectral projector for H0 and any Lebesgue measurable I ⊂ R. Proof. This follows from the definition S = Ω∗+Ω− and the intertwining properties (2.14) of the wave operators. We compute: SU0(t) = Ω ∗ +UV(t)Ω− = (UV(−t)Ω+) ∗Ω− = (Ω+U0(−t)) ∗Ω− = U0(t)S. (3.3) It follows from Proposition 7 that for a wide class of reasonable functions φ, we have the general result Sφ(H0) = φ(H0)S. The key property of the equality of the ranges of the wave operators (part of asymptotic completeness) has important consequences for the S-operator. Theorem 8. Suppose that for a pair of self-adjoint operators (H0,HV), we have Ran Ω−(HV,H0) = Ran Ω+(HV,H0). Then, the S-operator is a unitary operator on L 2(Rd). To prove the unitarity of the S-operator, we recall from (3.2) that, in general, S∗S = Ω∗−F+Ω−. If Ran Ω− = Ran Ω+, we have F+Ω− = Ω−. Furthermore, under our hypotheses, we have Ω∗−Ω− = 1L2(Rd), so that S ∗S = 1. As for SS∗, a similar calculation gives SS∗ = Ω∗+F−Ω+. It could happen that Ran Ω+ is strictly larger that Ran Ω−. In this case, the kernel of SS ∗ is nontrivial and consists of any element of Ran Ω+ orthogonal to Ran Ω−. In this case, SS ∗ is not invertible. Our condition that Ran Ω− = Ran Ω+ eliminates this possibility and we find SS∗ = Ω∗+F−Ω+ = Ω ∗ +Ω+ = 1. Hence, the S-operator S is invertible and S −1 = S∗. 3.2 The S-matrix Because the S-operator commutes with spectral family for H0, both operators admit a simultane- ous spectral decomposition. This is achieved with the Fourier transform. We define the Fourier transform of f ∈ S(Rd) by (Ff)(k) ≡ (2π)−d/2 ∫ Rd e−ik·xf(x) ddx. (3.4) 12 Peter D. Hislop CUBO 14, 3 (2012) The inverse Fourier transform is defined, for any g ∈ S(Rd), by (F−1g)(x) ≡ (2π)−d/2 ∫ Rd eik·xg(k) ddk. (3.5) The Fourier transform extends to a unitary map on L2(Rd). Note that for H0 = −∆, and f ∈ S(Rd), we have (F(H0f))(k) = |k| 2(Ff)(k). (3.6) It is convenient to write k = λω ∈ Rd, where λ ∈ [0,∞) and ω ∈ Sd−1. With this decompo- sition a function f(k) may be viewed as a function on Sd−1 parameterized by λ ∈ [0,∞). We need a family of maps from L2(Rd) → L2(Sd−1) parameterized by the energy λ. These maps E±(λ) can be defined via the Fourier transform (3.4). For λ ∈ R, and any f ∈ S(Rd), we define (E±(λ)f)(ω) ≡ (2π)−d/2 ∫ Rd e±iλx·ωf(x) ddx, ω ∈ Sd−1. (3.7) The transpose of these maps, tE±(λ) : L 2(Sd−1) → L2(Rd). The formula for the S-matrix involves the resolvent RV(λ) ≡ (HV − λ2) of HV. We will study the resolvent in detail in section 4. Provided ℑλ2 6= 0 and −λ2 is not an eigenvalue of HV, the resolvent RV(λ) is a bounded operator. We need to understand the behavior of VRV(λ + iǫ)V, for λ ∈ R, in the limit as ǫ → 0. That this limit exists as a compact operator is part of the limiting absorption principle that is discussed in section 4.1. We will write VRV(λ + i0)V for this limit. Recall from section 3.2 that the singular values of a compact operator A are given by µj(A) = √ λj(A ∗A), where {λj(B)} are the eigenvalues of B, and that K is in the trace class if ∑ j µj(K) < ∞. Theorem 9. Assume that the pair (H0,HV) is asymptotically complete with H0 = −∆. Then the S-matrix is the unitary family of operators S(λ), for λ ∈ R, on L2(Sd−1) given by S(λ) = 1L2(Sd−1) − πiλ d−2E−(λ)(V − VRV(λ + i0)V) tE+(λ) = 1L2(§d−1) − A(λ). (3.8) The operator A(λ) is the scattering amplitude. It is given by A(λ) ≡ −πiλd−2E−(λ)(V − VRV(λ + i0)V)tE+(λ), (3.9) and is in the trace class. We can also express the S-matrix in terms of localization operators in the case the support of V is compact. We assume that suppV ⊂ B(0,R1). We choose two other length scales so that 0 < R1 < R2 < R3 < ∞. Let 0 ≤ χj ∈ C20(Rd) have the property that χjV = V and suppχ2 ⊂ B(0,R2) and suppχ3 ⊂ B(0,R3). Finally, let W(φ) denote the commutator W(φ) ≡ [−∆,φ], for any φ ∈ C2(Rd). The following representation is due to Petkov and Zworski [28]. Theorem 10. Let V ∈ C20(Rd) and consider the S-matrix S(λ), λ ∈ R, as a unitary operator on L2(Sd−1). Then, the S-matrix has the form S(λ) = 1L2(Sd−1) + A(λ), λ ∈ R, (3.10) CUBO 14, 3 (2012) Scattering theory and resonances ... 13 where A(λ) is in the trace class. Explicitly, the scattering amplitude A(λ) has the form A(λ) = cdλ d−2E−(λ)W(χ2)RV(λ)W(χ1) tE+(λ), (3.11) where the constant cd = −i(2π) −d2(1−d)/2. 4 The resolvent and resonances We now switch our perspective and return to the study of the resolvent of the Schrödinger operator HV. We will connect these results with the S-matrix in section 4.5. We recall from section 2 that the spectrum of a self-adjoint operator A, denoted by σ(A), is a closed subset of the real line. The discrete spectrum of A, denoted σdisc(A), is the subset of the spectrum consisting of all isolated eigenvalues with finite multiplicity. The complement of the spectrum is called the resolvent set of A, denoted by ρ(A) ≡ C\σ(A). The resolvent of a self-adjoint operator A is defined, for any z ∈ ρ(A), as the bounded operator RA(z) = (A − z)−1. It is a bounded operator-valued analytic function on ρ(A). This means that about any point z0 ∈ ρ(A), the resolvent RA(z) has a norm convergent power series of the form RA(z) = ∞∑ j=0 Aj(z − z0) j, (4.1) for bounded operators Aj depending on z0. We note that for a self-adjoint operator A, the set C\R is always in the resolvent set. For a Schrödinger operator HV = −∆+V, we reparameterized the spectrum by setting z = λ 2 and write RHV (z) = RV(λ). Under this change of energy parameter, the spectrum in the complex λ-plane is the union of the line ℑλ = 0 and at most finitely-many points of the form iλj on the positive imaginary axis λj > 0. These points correspond to the negative eigenvalues of HV so that z = −λ2j ∈ σdisc(A). Let χV ∈ C∞0 (R) be a compactly-supported function so that χVV = V. We are most concerned with the properties of the localized resolvent RV(λ) ≡ χVRV(λ)χV. The operator-valued function RV(λ) is defined for ℑλ > 0 and λ 6= iλj, with λj > 0 and −λ2j an eigenvalue of HV. We would like to find the largest region in the complex λ-plane on which RV(λ) can be defined. 4.1 Limiting absorption principle One might first ask if the bounded operator RV(λ) has a limit as ℑλ → 0, from ℑλ > 0. That is, does the boundary-value of this operator-valued meromorphic function exist as a bounded operator for λ ∈ R? Because of the weight functions χV the answer to this question is yes. In more general settings, this result is part of what is referred to as the limiting absorption principle (LAP). The LAP plays an important role in scattering theory. 14 Peter D. Hislop CUBO 14, 3 (2012) Theorem 11. The meromorphic bounded operator-valued function RV(λ) on the open set ρ+(HV) ≡ {λ ∈ C | ℑλ > 0,−λ2 6∈ σdisc(HV)} admits continuous boundary values RV(λ) for λ ∈ R, except pos- sibly at λ = 0. That is, limǫ→0+ RV(λ + iǫ) exists for all λ ∈ R\{0}, and is a bounded, continuous operator-valued function on that set. The proof of this is given for more general potentials and N-body Schrödinger operators in, for example, [9, chapter 4]. The key ingredient is a local commutator estimate called the Mourre estimate, due to E. Mourre [22]. Let A = (1/2)(x · ∇ + ∇ ·x) be the generator of the unitary group implementing the dilations x → eθx, for θ ∈ R, on L2(Rd). One formally computes the following commutator, assuming ∇V exists: [HV,A] = 2H0 − x · ∇V = 2HV − (2V + x · ∇V). (4.2) Let I ⊂ R be a closed interval. Let EV(I) be the projector for HV and the interval I. We conjugate the commutator in (4.2) by this spectral projector: EV(I)[HV,A]EV(I) = 2EV(I)HVEV(I) − K(V,I), (4.3) where K(V,I) ≡ EV(I)(2V + x · ∇V)EV(I) is a compact, self-adjoint operator due to the properties of V. We now assume that there are no eigenvalues of HV in the interval I. For I ⊂ R+, this means that there are no positive eigenvalues of HV. In our situation, this is true (see [9, chapter 4]). Then, the spectral theorem implies that s−lim|I|→0 EV(I) = 0. Since K(V,I) is a compact operator and K(V,I) = K(V,I)EV(I), it follows that lim|I|→0 ‖K(V,I)‖ = 0. Furthermore, if I = [E1,E2], then 2E(I)HVE(I) ≥ 2E1. Given any ǫ > 0, we choose I so that |I| is so small that ‖K(V,I)‖ ≤ ǫ. Consequently, the commutator on the left in (2.9) is strictly nonnegative and bounded below: EV(I)[HV,A]EV(I) ≥ (2E1 − ǫ)EV(I) ≥ 0, |I| = E2 − E1 sufficiently small. (4.4) One of the main results of Mourre theory is that for any interval I for which a positive commutator estimate of the form (4.4) holds, the boundary value of the weighted resolvent exists. More precisely, for any α > 1, one has lim ǫ→0+ { sup E∈I ‖(A2 + 1)−α/2(HV − E − iǫ)−1(A2 + 1)−α/2‖ } < ∞. (4.5) This technical estimate is the heart of the LAP. Estimate (4.5) is proved using a differential inequality-type argument. In our case, the function χV serves as the weight for the resolvent. One also proves that the limit in (4.5) is continuous in E ∈ I. If there are no embedded eigenvalues, as in our case, this holds for all E > 0. Let us summarize what we have proved so far. The cut-off resolvent RV(λ) is meromorphic on C + with poles having finite-rank residues at at most finitely-many values iλj, with λj > 0 such that −λ2j is an eigenvalue of HV. Using the LAP, we can extend the cut-off resolvent RV(λ) onto the real axis as a bounded operator RV(λ), for λ ∈ R\{0}. This extension is continuous in λ. Hence, the cut-off resolvent is meromorphic on C+ and continuous on C+\{0}. CUBO 14, 3 (2012) Scattering theory and resonances ... 15 4.2 Analytic continuation of the cut-off resolvent of H0 Our cut-off resolvent RV(λ) is meromorphic on C+ and continuous on the real axis, except possibly at zero. It is now natural to ask if the operator has a meromorphic extension to the entire complex λ-plane as a bounded operator. We first consider the simpler case when V = 0. In this case, let χ ∈ C∞0 (Rd) be any compactly-supported cut-off function and consider the compact operator R0(λ) ≡ χR0(λ)χ. We mention that the kernel of this operator is known explicitly: R0(λ)(x,y) = i 4 χ(x) ( λ 2π|x − y| )(d−2)/2 H (1) (d−2)/2 (λ|x − y|)χ(y), (4.6) where H (1) j (s) is the Hankel function of the first kind with index j. We remark that the LAP is not necessary in order to construct an analytic continuation of the free cut-off resolvent R0(λ). An alternate and very nice method, based on the explicit formula (4.6), is presented in Vodev’s review article [44]. We are tempted to define the continuation R̃0(λ) of R0(λ) for ℑλ < 0 as the operator χR0(−λ)χ since, if λ ∈ C−, then −λ ∈ C+ and χR0(−λ)χ is well defined away from σdisc(HV). Clearly, R̃0(λ) ≡ χR0(−λ)χ for ℑλ < 0 is a meromorphic function in C−. The problem with this extension is that the two functions R0(λ) and R̃0(λ) do not match up on the real axis. In order to understand this, recall that in the z-plane, the resolvent (H0 − z) −1 is analytic on C\[0,∞). For λ0 > 0 and ǫ > 0, we are interested in the discontinuity of the resolvent across the positive z-axis at the point λ20 > 0. We can measure this by computing the following limit of the difference of the resolvents from above and below the point λ20 > 0: (H0 − (λ 2 0 + iǫ)) −1 − (H0 − (λ 2 0 − iǫ)) −1, (4.7) as ǫ → 0. The point z+ = λ 2 0 + iǫ has two square roots in the λ-plane. Let λ̃0 ≡ √ λ40 + ǫ 2. For z+, let θ be the angle in the first quadrant so that 0 ≤ θ < π/2. Then, the the two square roots are ±λ̃0[cos(θ/2) + isin(θ/2)]. The positive square root lies in C+ in the λ-plane so we work with this root λ+(ǫ) ≡ λ̃0[cos(θ/2) + isin(θ/2)]. Similarly, the point z− = λ20 − iǫ has two square roots ±λ̃0[cos(θ/2)−isin(θ/2)]. Note that because z− lies in the fourth quadrant, the imaginary part is negative. We choose the negative square root of z− because it lies in C + and call it λ−(ǫ). Finally, for ǫ small, we may write λ±(ǫ) = ±λ0 + iǫ ∈ C+. Consequently, the jump discontinuity in (4.7) across R+ in the z-plane corresponds to studying lim ǫ→0 [R0(λ0 + iǫ) − R0(−λ0 + iǫ)], (4.8) in the λ-plane. Both terms in (4.8) are well-defined since the points ±λ+iǫ have positive imaginary parts ǫ > 0. We will compute the limit as ǫ → 0 in (4.8) and show that it is nonzero. Furthermore, we will see that the limit extends to an analytic function on C. This is the term that must be added to 16 Peter D. Hislop CUBO 14, 3 (2012) R0(−λ), for ℑλ < 0, in order to obtain a function that is continuous (and actually analytic) across ℑλ = 0. We follow a calculation in [19, sections 1.5-1.6]. For f ∈ C∞0 (Rd) and ℑλ > 0, we have (R0(λ)f)(x) = (2π) −d/2 ∫ Rd eiξ·x (Ff)(ξ) (ξ2 − λ2) ddξ. (4.9) The Fourier transform Ff is a Schwartz function so it decays rapidly in |ξ| (see, for example, [30, section IX.1, Theorem IX.1]). Since ℑλ > 0, this guarantees that the integral in (4.9) is absolutely convergent. Switching to polar coordinates ξ = ρω, with ρ ≥ 0 and ω ∈ Sd−1, we obtain for the integral ∫ Rd eiξ·x (Ff)(ξ) (ξ2 − λ2) ddξ = ∫ Sd−1 dω ∫ ∞ 0 dρ eiρω·x ρn−1(Ff)(ρω) (ρ2 − λ2) . (4.10) In order to compute R0(λ0 + iǫ), we deform the ρ-contour into the lower-half complex ρ- plane in a small, counter-clockwise oriented semicircle centered at λ0. The Fourier transform Ff extends to an analytic function (see, for example [30, section IX.3]) so there is no difficulty with this. Similarly, in order to compute R0(−λ0 + iǫ), we note that this is the same as computing the integral in (4.10) with λ = λ0 − iǫ. This allows us to deform the ρ-integral into the upper-half complex ρ-plane and integrate around a small, clockwise semicircle centered at λ0. Subtracting the two terms as in (4.8), we obtain R0(λ0 + iǫ) − R0(−λ0 + iǫ) = ∫ Sd−1 dω ∫ Γ(λ0) dρ eiρω·x ρd−1(Ff)(ρω) (ρ2 − (λ0 + iǫ) 2) (4.11) where Γ(λ0) is a counter-clockwise oriented circle about λ0 > 0. Evaluating the integral by the residue theorem, we obtain for λ0 > 0, lim ǫ→0 [(R0(λ0 + iǫ) − R0(−λ0 + iǫ))f)(x)] = i 2 λd−20 (2π)(d−1)/2 ∫ Sd−1 dω (Ff)(λ0ω) e iλ0ω·x. (4.12) We define the kernel M(λ;x,y) by M(λ;x,y) ≡ i 2 1 (2π)d−1/2 ∫ Sd−1 dω eiλω·(x−y). (4.13) Undoing the Fourier transform in (4.12), we can write the limit in (4.12) as lim ǫ→0 [(R0(λ0 + iǫ) − R0(−λ0 + iǫ))f)(x)] = λ d−2 0 ∫ Rd M(λ0;x,y)f(y) d dy. (4.14) Because the integration is over a compact set, the sphere, the kernel M(λ;x,y) extends to an analytic function on C. Furthermore, recalling that we have compactly supported cut-off functions, the localized kernel M(λ;x,y) ≡ χ(x)M(λ;x,y)χ(y), (4.15) is square integrable for any λ ∈ C. Hence, the operator M(λ) is an analytic, operator-valued function on C with values in the Hilbert-Schmidt class of operators (see [29, section VI.6]). CUBO 14, 3 (2012) Scattering theory and resonances ... 17 We can now define an extension R̃0(λ) of the cut-off resolvent R0(λ) from ℑλ > 0 to C−\(−∞,0] by R̃0(λ) ≡ χR̃0(λ)χ = χR0(−λ)χ + λd−2χM(λ)χ, ℑλ < 0. (4.16) We then have for λ > 0, lim ǫ→0 χR̃0(λ − iǫ)χ = lim ǫ→0 [χR0(−λ + iǫ)χ + λ d−2χM(λ − iǫ)χ] = χR0(λ)χ, (4.17) and thus we have continuity across the positive λ half-axis. It can be checked that this actually gives analyticity in a neighborhood of R\(−∞,0]. As for the open negative real axis (−∞,0), we note that M(−λ) = M(λ) since the sphere is invariant under the antipodal map ω → −ω. A similar analysis can be performed for d ≥ 2 even. We summarize the main results on the analytic continuation for the free cut-off resolvent. Proposition 12. Suppose that the dimension d ≥ 3 is odd. The cut-off resolvent χR0(λ)χ of the Lapalcian admits an analytic continuation as a compact operator-valued function to the entire complex plane. In the case d = 1, there is an isolated pole of order one at λ = 0. When the dimension d ≥ 4 is even, the cut-off resolvent admits an analytic continuation as a compact operator-valued function to the infinite-sheeted Riemann surface of the logarithm Λ. In the case d = 2, there is a logarithmic singularity at λ = 0. 4.3 Meromorphic continuation of the cut-off resolvent of HV We can use Proposition 12 and the second resolvent formula to obtain a meromorphic continuation of the resolvent RV(λ). First, we write the second resolvent equation for λ ∈ C+, RV(λ) = R0(λ) − RV(λ)VR0(λ). (4.18) Conjugating this equation by the cut-off function χV and using the fact that χVV = V, we obtain RV(λ) = χR0(λ)χ − RV(λ)VχR0(λ)χ. (4.19) Solving this for RV(λ), we obtain RV(λ)(1 + VχVR0(λ)χV) = χVR0(λ)χV. (4.20) We use this equality in order to construct the meromorphic continuation of RV(λ). The right side of (4.20) has an analytic continuation as does the second factor on the left. We need to prove that this factor (1 + VχVR0(λ)χV) has a continuation that is boundedly invertible, at least away from a discrete set of λ. Recall that an operator of the form 1 + K, for a bounded operator K, is boundedly invertible if, for example, ‖K‖ < 1. The inverse is constructed as a norm convergent geometric series. 18 Peter D. Hislop CUBO 14, 3 (2012) There is another sufficient condition for invertibility. If the operator K is compact, then the Fredholm Alternative Theorem [14, Theorem 9.12] states that either K has an eigenvalue −1, and consequently, the operator 1 + K is not injective, or 1 + K is boundedly invertible. It follows from section 4.2 that the operator K(λ) = VχVR0(λ)χV in our expression (4.20) extends to a compact operator-valued analytic function. In this setting, the Analytic Fredholm Theorem [29, Theorem VI.14] is most useful. Theorem 13. Suppose that K(λ) is a compact operator-valued analytic function on a open connected set Ω ⊂ C. Then, either the operator 1+K(λ) is not invertible for any λ ∈ Ω, or else it is boundedly invertible on Ω except possibly on a discrete set D of points having no accumulation point in Ω. The operator is meromorphic on Ω\D At those points, the inverse has a residue that is a finite-rank operator. This theorem tells us that 1+K(λ), the first factor on the right of (4.20), is boundedly invertible for λ ∈ C except at a discrete set of points. Since we know that Rχ(λ) is invertible for ℑλ > 0, except for a finite number of points on the positive imaginary axis corresponding to eigenvalues, it also follows from (4.20) that the discrete set of points at which 1 + K(λ) fails to be invertible lies in C− if d is odd, or on Λ\C+ if d is even. Consequently, the Analytic Fredholm Theorem allows us to establish the existence of a meromorphic extension of RV(λ). Proposition 14. Let V ∈ C20(Rd) be a real-valued potential and let χV ∈ C∞0 (Rd) be any function such that χVV = V. Then the cut-off resolvent RV(λ) admits a meromorphic extension to C if d is odd and to Λ if d is even. The poles have finite-rank residues. 4.4 Resonances of HV Having constructed the meromorphic continuation of the cut-off resolvent RV(λ), we can now define the resonances of HV. Definition 15. Let V ∈ C20(Rd) be a real-valued potential. The resonances of HV are the poles of the meromorphic continuation of the compact operator RV(λ) occurring in C− for d odd, or on Λ\C+ for d even. This definition can also be extended to complex-valued potentials. The residues of the exten- sion of RV(λ) at the poles are finite rank operators. If λ0 ∈ C− is a resonance, then a resonance state ψλ0 ∈ H corresponding to λ0 is a solution to (1 + VχVR0(λ0)χV)ψλ0 = 0. (4.21) The poles are independent of the cut-off function used provided it has compact support and satisfies χV = V. CUBO 14, 3 (2012) Scattering theory and resonances ... 19 4.5 Meromorphic continuation of the S-matrix The meromorphic continuation of the cut-off resolvent RV(λ) permits us to mermorphically con- tinue the S-matrix S(λ) as a bounded operator on L2(Sd) from λ ∈ C+ to all of C or Λ depending on the parity of d. This follows from formula (3.8) of Theorem 9. Because of the compactness of the support of V, the operators E±(λ), and their transposes, admit analytic continuations. This property, together with the continuation properties of RV(λ) and formula (3.8), establish the meromorphic continuation of S(λ). For complex λ, the S-matrix is no longer unitary. The relation S(λ)S(λ)∗ = 1, however, does continue to hold for λ ∈ C (or Λ). Theorem 16. The S-matrix S(λ) admits a mermorphic continuation to C if d is odd, or to the Riemann surface Λ, if d is even, with poles precisely at the resonances of HV. The order of the poles are the same as the order of the poles for HV and the residues at these poles have the same finite rank. For the Schrödinger operator HV, the resonances may be defined as the poles of the meromor- phic continuation of the cut-off resolvent RV(λ), or in terms of the meromorphic continuation of the S-matrix S(λ). From formula (3.8), it follows that the poles of the meromorphic continuation of the S-matrix are included in the poles of the continuation of the resolvent. It is not always true that the scattering poles, defined via the S-matrix, are the same as the resolvent poles. A striking example where the scattering poles differ from the resolvent poles occurs for hyperbolic spaces. However, in the Schrödinger operator case considered here, these are the same. A proof is given by Shenk and Thoe [37]. 5 Resonances: Existence and the counting function The resonance set RV for a Schrödinger operator was defined in Definition 15 as the poles of the meromorphic continuation of the cut-off resolvent RV(λ) to C for d ≥ 3 odd or to the Riemann surface Λ for d ≥ 4 even, together with their multiplicities. There are two basic questions that arise: (1) Existence: Do resonances exist for Schrödinger operators HV with our class of potentials? (2) Counting: How many resonances exist? 5.1 Existence of resonances There are many different proofs of the existence of resonances for various quantum mechanical systems. Resonances are considered as almost bound states or long-lived states that eventually decay to spatial infinity. To understand this physical description, let us consider the time evolution of a resonance state ψ0 corresponding to a resonance energy z0 = E0−iΓ (in the z-parametrization), with Γ > 0. A resonance state ψ0 solves the partial differential equation HVψ0 = z0ψ0 and is 20 Peter D. Hislop CUBO 14, 3 (2012) purely outgoing. Since V has compact support, the function ψ0 satisfies −∆ψ0 = z0ψ0 for |x| large enough. The outgoing condition means that the radial behavior of a component of ψ0 with angular momentum ℓ ≥ 0 is a Hankel function of the first kind H(1) (d−2)/2+ℓ ( √ z0|x|). Such a function ψ0 grows exponentially as |x| → ∞ and is not in H. We can formally compute UV(t)ψ0 by expressing the time evolution group as an integral of the resolvent over the energy UV(t)ψ0 = −1 2πi ∫ R e−itERV(E) dE. (5.1) Performing a deformation of the contour to capture the pole of the resolvent at z0 and applying the residue theorem, one finds that the time evolution behaves like e−iz0tψ0 = e −Γte−iE0tψ0. The factor e−itE0ψ0 has an oscillatory time evolution similar to that of a bound state with energy E0, whereas the factor e −tΓ is an exponentially decaying amplitude. The lifetime of the state is τ = Γ−1. This is, roughly, the time it takes the amplitude to decay to e−1 times its original size. As noted above, there is no such state ψ0 ∈ H corresponding to a resonance z0 in the sense that HVψ0 = z0ψ0. Since HV is self-adjoint and z0 has a nonzero imaginary part, the solutions of this eigenvalue equation are not in H. There are, however, approximate resonance states obtained by truncating such ψ0 to bounded regions, say K ⊂ Rd. The truncated state χKψ0 ∈ H has an approximate time evolution like e−Γte−iE0tχKψ0 showing that the amplitude of the resonance state in the bounded region K decays exponentially to zero. A typical situation for which resonances are expected to exist is the hydrogen atom Hamil- tonian HV = −∆ − |x| −1 acting on L2(R3), perturbed by an external constant electric field Vpert(x) = Ex1 in the x1-direction. The total Stark hydrogen Schrödinger operator is HV(E) = −∆ − |x|−1 + Ex1. When E = 0, the spectrum of HV consists of an infinite sequence of eigenvalues En = −1/4n 2 plus the half line [0,∞). When E is turned on, the spectrum of HV(E) is purely absolutely continuous and equal to exactly the entire real line. There are no eigenvalues! It is expected that the bound states En of the hydrogen atom have become resonances for E 6= 0. These finite-lifetime states are observed in the laboratory. These resonances, in the z- parametrization, have their real parts close to the eigenvalues En. Their imaginary parts are exponentially small behaving like e−α/E. This means their lifetime is very long. The proof of the existence of these resonances for the the Stark hydrogen Hamiltonian was given by Herbst [13] in 1979. The method of proof is perturbative in that the electric field strength is assumed to be very small. More generally, there are various models for which one can prove the existence of resonances using the smallness of some parameter. The semiclassical approximation is the most common regime. The quantum Hamiltonian is written as HV(h) = −h 2∆ + V0 + V1 and h is considered as a small parameter. For a discussion of resonances in the semiclassical regime, see, for example, [14, Chapters 20–23]. For more information on the semiclassical approximation for eigenvalues, eigenfunctions and resonances, see, for example, the monographs [10, 18, 33]. CUBO 14, 3 (2012) Scattering theory and resonances ... 21 If we inquire about the existence of resonances for the models studied here, HV = −∆ + V, with V ∈ C∞0 (Rd), with no parameters, the proof is much harder and requires different techniques. Melrose [19] gave perhaps the first proof of the existence of infinity many resonances for such HV. The proof holds for smooth, real-valued, compactly-supported potentials V ∈ C∞0 (Rd), for d ≥ 3 odd. The proof requires two ingredients that will be presented here without proof. 5.1.1 Small time expansion of the wave trace The wave group WV(t) associated with the Schrödinger operator HV is defined as follows. Let ∂t denote the partial derivative ∂/∂t. Consider the wave equation associated with HV: (∂2t − HV)u = 0, u(t = 0) = u0, ∂tu(t = 0) = u1. (5.2) The solution can be expressed in terms of the initial conditions (u0,u1). The time evolution occurs on a direct sum of two Hilbert spaces HFE = {(u0,u1) | ∫ [|∇u0|2 + |u1|2] < ∞}. This is the space of finite energy solutions. In two-by-two matrix notation, the time evolution acts as WV(t) ( u0 u1 ) = ( u ∂tu ) (5.3) The infinitesimal generator of the wave group WV(t) is the two-by-two matrix-valued operator AV ≡ ( 0 1 HV 0 ) . (5.4) The evolution group WV(t) is unitary on HFE. Similarly, the free wave group W0(t) is generated by A0 that is expressed as in (5.4) with V = 0. If HV ≡ −∆ + V ≥ 0, then this operator can be diagonalized. The diagonal form is ( √ HV 0 0 − √ HV ) . (5.5) In this case, the wave group WV(t) can be considered as two separate unitary groups e ±i √ HV t each acting on a single component Hilbert space. The basic fact that we need is that the map t ∈ R → Tr[WV(t) − W0(t)] is a distribution. This means that for any ρ(t), a smooth, compactly-supported function, the integral ∫ R dt ρ(t)Tr[WV(t) − W0(t)] (5.6) is finite and bounded above by an appropriate sum of semi-norms of ρ. The distribution has a singularity at t = 0 and the behavior of the distribution at t = 0 has been well-studied. For d ≥ 3 22 Peter D. Hislop CUBO 14, 3 (2012) odd, the wave trace has the following expansion as t → 0: Tr[WV(t) − W0(t)] = (d−1)/2∑ j=1 wj(V)(−i) d−1−2jδ(d−1−2j)(t) + N∑ j=(d+1)/2 wj(V)|t| 2j−d + rN(t), (5.7) where the remainder rN(t) ∈ C2N−d(R). The first sum consists of derivatives of the delta function δ(t) at zero. We recall that for any smooth function f, these distributions are defined as 〈δj,f〉 = (−1)jf(j)(0). The second part of the sum consists of distributions that are polynomial in t. The coefficients wj(V) are integrals of the potential V and its derivatives. These are often called the ‘heat invariants’. The first three are: w1(V) = c1,d ∫ Rd V(x) ddx w2(V) = c2,d ∫ Rd V2(x) ddx w3(V) = c3,d ∫ Rd (V3(x) + |∇V(x)|2) ddx, (5.8) where the constants cj,d are nonzero and depend only on the dimension d. For some insight as to why the trace in (5.7) exists, note that for ρ ∈ C∞0 (R), a formal calculation gives ∫ R ρ(t)Tr[WV(t) − W0(t)] dt = Tr( (Fρ)(AV ) − (Fρ)(A0)). (5.9) The Fourier transform Fρ is a smooth, rapidly decreasing function. Because V has compact support, the difference (Fρ)(AV )−(Fρ)(A0) is in the trace class. This follows from the fact that the difference of the resolvents RV(z) k − R0(z) k is in the trace class for ℑz 6= 0 and k > d/2. 5.1.2 Poisson formula The key formula that links the resonances with the trace of the difference of the wave groups is the Poisson formula. In our context it was proved by Melrose [20]. It is named this because of the analogy with the classical Poisson summation formula. Let f ∈ C∞(Rd) be a Schwarz function meaning that the function and all of its derivatives decay faster that 〈‖x‖〉−N, for any N ∈ N. The classical Poisson summation formula states that ∑ k∈Zd f(x + k) = ∑ k∈Zd (Ff)(k)e2πix·k. (5.10) The Poisson formula for the wave group states that Tr[WV(t) − W0(t)] = ∑ ξ∈RV m(ξ)ei|t|ξ, t 6= 0, (5.11) CUBO 14, 3 (2012) Scattering theory and resonances ... 23 where m(ξ) is the algebraic multiplicity of the resonance ξ. This multiplicity is defined as the rank of the residue of the resolvent at the pole ξ or, equivalently, by the rank of the contour integral: m(ξ) = Rank (∫ γξ R(s) ds ) , (5.12) where γξ is a small contour enclosing only the pole ξ of the resolvent. It is important to note that the Poisson formula (5.11) is not valid at t = 0. 5.1.3 Melrose’s proof of the existence of resonances Melrose [19, section 4.3] observed that the Poisson formula (5.11) and the trace formula (5.7) can be used together to prove the existence of infinitely many resonances for Schrödinger operators. Theorem 17. Let us suppose that d ≥ 3 is odd and that V ∈ C∞0 (Rd; R). Suppose also that wj(V) 6= 0 for some j ≥ (d + 1)/2. Then HV has infinitely many resonances. In particular, for d = 3, since w2(V) = c2 ∫ V2(x) dx, for a positive constant c2 > 0, if V ∈ C∞0 (R3; R) is nonzero, then HV has an infinite number of resonances. Proof. 1. Suppose that HV has no resonances. Then the right side of the Poisson formula (5.11) is zero. On the other hand, it follows from the small time expansion (5.7) and the assumption that wj(V) 6= 0 for some j ≥ (d + 1)/2 that for t > 0 the right side of the expansion (5.7) is nonzero. Note that for t > 0 all the contributions from the delta functions vanish. Hence we obtain a contradiction. Consequently, there must be at least one resonance. 2. If there are only finitely-many resonances, then the sum on the right in (5.11) is finite and the formula can be extended to t = 0. In particular, at t = 0, it is a finite positive number greater than or equal to the number of resonances. On the other hand, looking at the trace formula (5.7), if only one or more of the coefficients wj(V) 6= 0 for j > (d+1)/2, then the trace is zero at t = 0 (the coefficients of the derivatives of the delta functions being zero), so we get a contradiction. Hence, at least one of the coefficients of a delta function term is nonzero. Then the trace formula indicates that the distribution Tr[WV(t) − W0(t)] is not continuous at t = 0 whereas the Poisson formula indicates that it is continuous through t = 0, and we again obtain a contradiction. Consequently, there must be an infinite number of resonances. We remark that in the even dimensional case for d ≥ 4, Sá Barreto and Tang [36] proved the existence of at least one resonance for a real-valued, compactly-supported, smooth nontriv- ial potential. Having settled the question of existence, we now turn to counting the number of resonances. 24 Peter D. Hislop CUBO 14, 3 (2012) 5.2 The one-dimensional case: Zworski’s asymptotics As with many problems, the one-dimensional case is special since many techniques of ordinary differential equations can be used. The most complete result on resonances for HV = −d 2/dx2 +V on L2(R) with a compactly-supported potential was proven by Zworski [46]. Theorem 18. Let V ∈ L∞0 (R). Then the number of resonances NV(r) with modulus less that r > 0 satisfies: NV(r) = 2 π ( sup x,y∈supp V |x − y| ) r + o(r). (5.13) There are extensions of this result to a class of super-exponentially decaying potentials due to R. Froese [11]. We will not comment further on the one-dimensional case. 5.3 Estimates on the number of resonances: Upper bounds The resonance counting function counts the number of poles, including multiplicities, of the mero- morphic continuation of the cut-off resolvent in C− for d odd, and on Λ for d even. We will concentrate on the odd d-dimensional case, although we will give comments on the even dimen- sional case too. For any r > 0, we define NV(r) as NV(r) = #{j | λj(V) satisfies |λj(V)| ≤ r and ℑλj(V) < 0}. (5.14) This function is monotone increasing in r. It is the analogue of the eigenvalue counting function NM(r) studied by Weyl to count the number of eigenvalues of the Laplace-Beltrami operator on a compact Riemannian manifold M with size less than r > 0. The Weyl upper bound on the eigenvalue counting function is NM(r) ≤ cdVol(M)〈r〉d, (5.15) where 〈r〉 = √ 1 + r2. It is natural to ask if the resonance counting function NV(r) satisfies a similar upper bound. Since Melrose’s early work [21], many people have established upper bounds on NV(r) with in- creasing optimality. Zworski [49] presents a good survey of the state-of-the-art up to 1994. The optimal upper bound, having the same polynomial behavior as Weyl’s eigenfunction counting func- tion (5.15), was achieved by Zworski [47]. A significant simplification of the proof was given by Vodev [41]. Theorem 19. For d ≥ 3 odd, the resonance counting function NV(r) satisfies NV(r) ≤ C(d,V)〈r〉d, (5.16) for a constant 0 < C(d,V) < ∞ depending on d and V. A sketch of the proof of this theorem will be given following the beautiful exposition of Zworski [49, section 5], using Vodev’s simplification [41]. One basic idea of the proof is to find a suitable CUBO 14, 3 (2012) Scattering theory and resonances ... 25 analytic or meromorphic function that has zeros exactly at the resonances. Suppose h(λ) is one such function analytic on C. Then one can count the number of zeros using Jensen’s formula. This formula relates the number of zeros of h to growth properties of h. If a circle of radius r > 0 crosses no zero of h, if h(0) 6= 0, and if ak are the zeros of h inside the circle, then Jensen’s formula states that Nh(r)∑ k=1 log ( r |ak| ) = 1 2π ∫2π 0 log |h(reiθ)| dθ − log |h(0)|. (5.17) If we only sum over those zeros inside the circle of radius r/2, we have that log(r/|ak|) ≥ log2, so that Nh(r/2)[log2] ≤ 1 2π ∫2π 0 | log |h(reiθ)|| dθ + | log |h(0)||. (5.18) This inequality shows that it suffices to bound h on circles of radius 2r in order to count the number of zeros inside the circle of radius r > 0. We will use some inequalities for singular values, the proofs of which can be found in [39]. Proof. 1. The first observation is that the operator (VR0(λ)χV) d+1 is in the trace class for ℑλ ≥ 0. Consequently, the following determinant is well-defined: h(λ) ≡ det(1 − (VR0(λ)χV)d+1). (5.19) This function is analytic on C+ with at most a finite number of zeros corresponding to the eigen- values of HV. It follows from section 4.3 that this function has an analytic continuation to all of C. Furthermore, the zeros of this function for ℑλ < 0 include with the resonances of HV that are given as the zeros of the analytic continuation of 1 + VR0(λ)χV according to (4.20). The problem, then, is to count the number of zeros of the analytic function h(λ) inside a ball of radius r > 0 in C. By Jensen’s inequality (5.18), it suffices to obtain a growth estimate on h of the form |h(λ)| ≤ C1ec2|λ| d . (5.20) 2. We first estimate h in the half space ℑλ ≥ 0 using the fact that V has compact support contained inside of a bounded region Ω. Let −∆Ω ≥ 0 denote the Dirichlet Laplacian on Ω. By Weyl’s bound (5.15), the jth eigenvalue λj(Ω) of −∆Ω grows like λj(Ω) ∼ j 2/d. Furthermore, we have ∆ΩV = ∆V. Using these ideas and the simple inequality for the singular values µj(AB) ≤ ‖A‖µj(B), we have µj(χVR0(λ)χV) = µj((−∆Ω + 1) −1/2(−∆Ω + 1) 1/2χVR0(λ)χV) ≤ ‖(−∆Ω + 1)1/2χVR0(λ)χV‖ µj((−∆Ω + 1)−1/2) ≤ Cj−1/d. (5.21) It is important to note that the operator χVR0(λ)χV : L 2(Rd) → H1(Rd) is bounded uniformly in λ, for ℑλ ≥ 0. Consequently, the norm ‖(−∆Ω + 1)1/2χVR0(λ)χV‖ is bounded uniformly in λ in the upper half space. Upon squaring this norm, the operator −∆Ω can be replaced by −∆ 26 Peter D. Hislop CUBO 14, 3 (2012) because of the support of χV. Since µm+k−1(AB) ≤ µk(A)µm(B), we have µ2j−1(A2) ≤ µj(A)2, and consequently, for all large j µj((χVR0(λ)χV) d+1) ≤ Cj−(d+1)/d. (5.22) It follows that |h(λ)| ≤ C for ℑλ ≥ 0. 3. For ℑλ < 0, we make use of the following formula from scattering theory used already in section 4.2. For λ ∈ R, we have χV(R0(λ) − R0(−λ))χV = cd(λ d−2) tEχ(λ)Eχ(λ), (5.23) where Eχ(λ) : L 2(Rd) → L2(Sd−1) is given by (Eχ(λ)g)(ω) ≡ ∫ Rd eiλω·xχV(x)g(x) d dx, (5.24) and tEχ(λ) denotes the transpose of this operator. This formula can be extended to all of C. We compute the singular values of the operator on the left in (5.23): µj(χV(R0(λ) − R0(−λ))χV) ≤ C|λ|d−2ec2|λ|µj(Eχ(λ)). (5.25) Since Eχ(λ) ∗ : L2(Sd−1) → L2(Rd), the operator Eχ(λ) ∗Eχ(λ) : L 2(Sd−1) → L2(Sd−1). This is a crucial observation since the operator acts on a d − 1 dimensional space. Without this reduction, one obtains an upper bound but with exponent d + 1 rather than the optimal exponent d. In a manner similar to (5.21), we compute for any m > 0, µj(Eχ(λ)) ≤ µj((−∆Sd−1 + 1)−m(−∆Sd−1 + 1)mEχ(λ)) ≤ ‖(−∆Sd−1 + 1)mEχ(λ))‖L2(Sd−1) µj((−∆Sd−1 + 1)−m) ≤ Cm(2m)!j−2m/(d−1)ec|λ|. (5.26) This follows from the explicit formula for the kernel of Eχ(λ)), Eχ(λ)(ω,x) = e −iλω·xχV(x). (5.27) In particular, the factor (2m)! comes from differentiating the exponential factor. Using Stirling’s formula for the factorial, we obtain from (5.25)–(5.26) µj(χV(R0(λ) − R0(−λ))χV) ≤ |λ|d−2ec2|λ|Cm(2m + 1)2m+(1/2)(j−1/(d−1))2m. (5.28) We now optimize over the free parameter m by choosing m ∼ j−1/(d−1). As a result, we obtain µj(χV(R0(λ) − R0(−λ))χV) ≤ ec|λ|e−cj 1/(d−1) . (5.29) 4. We now combine (5.21) with (5.29). For this, we need Fan’s inequality for singular values [39, Theorem 1.7] that states that µn+m+1(A + B) ≤ µm+1(A) + µn+1(B). (5.30) CUBO 14, 3 (2012) Scattering theory and resonances ... 27 For ℑλ < 0, we write µj(χVR0(λ)χV) = µj([χV(R0(λ) − R0(−λ))χV] + χVR0(−λ)χV). (5.31) Applying Fan’s inequality (5.30) to the right side of (5.31), we find that for ℑλ ≤ 0, the singular values satisfy µj(χVR0(λ)χV) ≤ ec|λ|e−cj 1/(d−1) + cj−1/d. (5.32) Taking the (d + 1)st power of the operators, as in (5.22), we find µj((χVR0(λ)χV) d+1 ) ≤ ec|λ|e−cj 1/(d−1) + cj−(d+1)/d, (5.33) for a constant c > 0. As j → ∞, the first term dominates until j ∼ [|λ|d−1], where [·] denotes the integer part. We then use the Weyl estimate for the determinant (see [39]), factorize the product using the first estimate in (5.33) for j ≤ [d|λ|d−1], to obtain |h(λ)| ≤ | det(1 + (VR0(λ)χV)d+1)| ≤ Π∞j=1(1 + µj((VR0(λ)χV)d+1)) ≤ ( Π [d|λ|d−1] j=1 e C|λ| ) ( Πj≥[d|λ|d−1](1 + c2j −(d+1)/d) ) ≤ cec|λ| d . (5.34) This establishes (5.20) so by Jensen’s inequality (5.18) we obtain the optimal upper bound on the resonance counting function. Upper bounds for super-exponentially decaying potentials in d ≥ 3 odd dimensions were proved by R. Froese [12]. There are fewer results in even dimensions. We refer to [7] for a discussion and the papers [15, 42, 43]. 5.4 Estimates on the number of resonances: Lower bounds One might hope to have a lower bound on the number of resonances of the form NV(r) ≥ Cdrd. (5.35) This is known to hold in two cases. The first case is Zworski’s result for d = 1. The second is for a class of spherically symmetric potentials in dimension d ≥ 3 odd. Zworski proved that if V(r) has the property that V ′(a) 6= 0, where a > 0 is the radius of the support of V, then an asymptotic expansion holds for the number of resonances: NV(r) = cda drd + o(rd), d ≥ 3 and odd. (5.36) In general, for V ∈ L∞0 (Rd) (or, even V ∈ C∞0 (Rd)), there is presently no known proof of the optimal lower bound (5.35). There are some partial results for d ≥ 3 odd. These include nonoptimal lower bounds, estimates on the number of purely imaginary poles for potentials with fixed sign, and counterexamples made from certain complex potentials. 28 Peter D. Hislop CUBO 14, 3 (2012) 5.4.1 Nonoptimal lower bounds For the case of d ≥ 3 odd, the first quantitative lower bounds for the resonance counting function for nontrivial, smooth, real-valued V ∈ C∞0 (Rd), not of fixed sign, were proved in [2]. In particular, it was proved there that lim sup r→∞ nV(r) r(logr)−p = ∞, (5.37) for all p > 1. For the same family of potentials, Sá Barreto [34] improved this to lim sup r→∞ nV(r) r > 0. (5.38) We mention that, in particular, all these lower bounds require the potential to be smooth. Concerning lower bounds in the even dimensional case for d ≥ 4, Sá Barreto [35] studied the resonance counting function NSaB(r) defined to be the number of resonances λj with 1/r < |λj| < r and | argλj| < logr. As r → ∞, this region in the Riemann surface Λ opens like logr. Sá Barreto proved that for even d ≥ 4, lim sup r→∞ NSaB(r) (logr)(log logr)−p = ∞, (5.39) for all p > 1. 5.4.2 Purely imaginary poles Lax and Phillips [17] noticed that for odd dimensions d ≥ 3, the wave operator associated with exterior obstacle scattering has an infinite number of purely imaginary resonances. They remarked that their proof held for Schrödinger operators with nonnegative, compactly-supported, nontrivial potentials. Vasy [40] used their method to prove that a Schrödinger operator HV with a compactly- supported, bounded, real-valued potential with fixed sign (either positive or negative) has an infinite number of purely imaginary resonances. These resonances are poles of the meromorphic continuation of the resolvent of the form −iµj(V), with µj(V) > 0. In the z = λ 2 plane, these are located on the second Riemann sheet of the square root function. Furthermore, Vasy is able to count these poles and prove the following lower bound NV(r) ≥ Cdrd−1. (5.40) This is not an optimal lower bound on the total number of resonances. Recently, the author and T. J. Christiansen [7] proved that in even dimension there are no purely imaginary resonances on any sheet for HV with bounded, positive, real-valued potentials with compact support. 5.4.3 Complex potentials Most surprisingly, Christiansen [3] gave examples of compactly supported, bounded complex-valued potentials having no resonances in any dimension d ≥ 2! This result, while interesting in its own CUBO 14, 3 (2012) Scattering theory and resonances ... 29 right, means that any technique that provides a result of the type (5.35) must be sensitive to whether the potential is real- or complex-valued. 6 Maximal order of growth is generic for the resonance counting function There is one general result that is a weak form of (5.35) due to the author and T. J. Christiansen [5]. This result states that for almost all potentials V ∈ L∞0 (K), for a compact subset K ⊂ Rd, real- or complex-valued, the lower bound holds in the following sense as determined by the order of growth of the resonance counting function NV(r). Definition 20. The order of growth of the monotone increasing function NV(r) is defined by ρV ≡ lim r→∞ logNV(r) logr , (6.1) if the limit exists and is finite. Because of the upper bound (5.16), the order of growth of the resonance counting function is bounded from above as ρV ≤ d. We say that the order of growth is maximal for a potential V if ρV = d. By “almost all potentials” referred to above, we mean that the set of potentials in L∞0 (K), for a fixed compact subset K ⊂ Rd with nonempty interior, for which the resonance counting function has maximal order of growth, is a dense Gδ-set. Recall that a Gδ-set is a countable intersection of open sets. One sometimes says that a property that holds for all elements in a dense Gδ-set is generic. (Added in proof: For some recent developments, see Dinh and Vu arXiv:1207.4273v1.) 6.1 Generic behavior: odd dimensions The basic theorem on generic behavior is the following. Theorem 21. [5] Let d ≥ 3 be odd and K ⊂ Rd be a fixed, compact set with nonempty interior. Let MF(K) ⊂ L∞0 (K), for F = R or F = C, be the set of all real-valued, respectively, complex-valued potentials in L∞0 (K) such that the resonance counting function NV(r) has maximal order of growth. Then, the set MF(K) is a dense Gδ set for F = R or F = C. This holds for both real-valued and complex-valued potentials. By [3], we know there are complex potentials with zero order of growth. An interesting open question is whether there exist real-valued potentials in L∞0 (R d) for which the resonance counting function has less than maximal order of growth. The proof of this theorem uses the S-matrix and its continuation to the entire complex plane. In section 3, we defined the scattering matrix for the pair H0 = −∆ and HV = H0 + V. The S-matrix S(λ), acting on L2(Sd−1), is the bounded operator defined in (3.8). In the case that V is 30 Peter D. Hislop CUBO 14, 3 (2012) real-valued, this is a unitary operator for λ ∈ R. Under the assumption that supp V is compact, the scattering amplitude A(λ) : L2(Sd−1) → L2(Sd−1), defined in (3.9), is a trace class operator. Hence, the function fV(λ) ≡ detS(λ), (6.2) is well-defined, at least for ℑλ > 0 sufficiently large. What are the meromorphic properties of fV(λ)? As proved in Theorem 16, the S-matrix has a meromorphic continuation to the entire complex plane with finitely many poles for ℑλ > 0 corresponding to eigenvalues of HV, and poles in ℑλ < 0 corresponding to resonances. We recall that if ℑλ0 ≥ c0〈‖V‖L∞〉, the multiplicity of λ0, as a zero of detSV(λ), and of −λ0, as a pole of the cut-off resolvent RV(λ), coincide. Consequently, the function fV(λ) is holomorphic for ℑλ > c0〈‖V‖L∞〉, and well-defined for ℑλ ≥ 0 with finitely many poles corresponding to the eigenvalues of HV. Hence, the problem of estimating the number of zeros of fV(λ) in the upper half plane is the same as estimating the number of resonances in the lower half plane. The estimates on fV(λ) are facilitated in the odd dimensional case by the well-known rep- resentation of fV(λ) in terms of canonical products. Let G(λ;p) be defined for integer p ≥ 1, by G(λ;p) = (1 − λ)eλ+λ 2/2+···+λp/p, (6.3) and define P(λ) = Πλj∈RV ,λj 6=0 G(λ/λj;d − 1). (6.4) Then the function fV(λ) may be written as fV(λ) = αe ig(λ)P(−λ) P(λ) , (6.5) for some constant α > 0 and where g(λ) is a polynomial of order at most d. Careful study of the scattering matrix and the upper bound of Theorem 19 may be used to show that fV(λ) is of order at most d in the half-plane ℑλ > c0〈‖V‖∞〉, see [48]. We consider a fixed compact set K ⊂ Rd with nonempty interior. Let M(K) be the subset of potentials in L∞0 (K) having a resonance counting function with maximal order of growth. We can separately consider real- or complex-valued potentials. The proof of Theorem 21 requires that we prove 1) that M(K) is a Gδ-set, and 2) that M(K) is dense in L∞0 (K). The proof that M(K) is a Gδ-set uses standard estimates on the S-matrix as in [5]. For N,M,j ∈ N with j > 2N+1, and for q > 0, we define sets of potentials A(N,M,q,j) ⊂ L∞0 (K) by A(N,M,q,j) ≡ {V ∈ L∞0 (K) | ‖V‖L∞ ≤ N, log | det(SV(λ))| ≤ M|λ|q, for ℑλ ≥ 2N + 1 and |λ| ≤ j} (6.6) One proves that these sets are closed. More importantly, we can use these sets to characterize the set of potentials having a resonance counting function with an order of growth strictly less that d. CUBO 14, 3 (2012) Scattering theory and resonances ... 31 For this, we define sets B(N,M,q) by B(N,M,q) ≡ ⋂ j≥2N+1 A(N,M,q,j). (6.7) One proves that if NV(r) has order of growth strictly less than d, then one can find (N,M,ℓ) ∈ N3 so that V ∈ B(N,M,d − 1/ℓ). Since the sets A(N,M,j,q) are closed, so are the sets B(N,M,j). One notes that ∪(N,M,j)∈N3B(N,M,j) is an Fσ set. The final step of the proof is to show that M(K) is the complement of this set. It follows that M(K) is a Gδ-set. The proof of the density of M(K) is more involved and relies on machinery from complex analysis as developed in [4]. The basic idea is to consider a wider family of potentials V(x;z) holomorphic in the complex variable z ∈ Ω ⊂ C, for some open subset Ω. The construction of the S-matrix goes through for these complex-valued potentials. The key result is that if for some z0 ∈ Ω the order of growth ρV(z0) for NV(z0) is equal to d, then there is a pluripolar subset E ⊂ Ω so that the order of growth for all potentials V(z), with z ∈ Ω\E, is equal to d. A pluripolar set is very small, in particular, the Lebesgue measure of E ∩ R is zero. How do we know there is a potential for which NV(r) has maximal order of growth? For d ≥ 3 odd, we can use the result of Zworski [47]. As mentioned in section 5.4, Zworski proved the an asymptotic expansion for NV(r) for a class of radially symmetric potentials with compact support. Let V0 be one of these potentials so that V0 ∈ M(K). To prove the density of M(K) in L∞0 (Rd), we take any V1 ∈ L∞0 (K) and form V(z) = zV0 + (1 − z)V1. This is a holomorphic function of z for z ∈ Ω = C. We apply the result described above to this family of holomorphic potentials. In particular, for z0 = 1, we have V(z0) = V0 and ρV(z0) = d by Zworski’s result. Let E ⊂ C be the pluripolar set so that for z ∈ C\E, the order of growth ρV(z) = d. Since the Lebesgue measure of E ∩ R is zero, we can find z ∈ R\(E ∩ R), with |z| as small as desired, for which ρV(z) = d. So, given ǫ > 0, we take z̃ ∈ R\(E ∩ R) so that |z̃| ≤ ǫ(1 + ‖V1‖L∞ + ‖V0‖L∞)−1. With this choice, we have ‖V1 − V(z̃)‖L∞ ≤ |z̃| (‖V1‖L∞‖V0‖L∞) ≤ ǫ. (6.8) This proves the density of M(K) in L∞0 (Rd). Note that we can take V0 real-valued and so V(z̃) is real-valued. We remark that the representation (6.5) is not available in the even dimensional case. 6.2 Generic behavior: even dimensions We now summarize the corresponding results in the even dimensional case. Let χV ∈ C∞0 (Rd) be a smooth, compactly-supported function satisfying χVV = V, and denote the resolvent of HV by RV(λ) = (HV −λ 2)−1. In the even-dimensional case, the operator-valued function χVRV(λ)χV has a meromorphic continuation to the infinitely-sheeted Riemann surface of the logarithm Λ. We denote by Λm the m th open sheet consisting of z ∈ Λ with mπ < argz < (m + 1)π. The physical sheet corresponds to Λ0 and it is identified with the upper half complex plane. We denote the number 32 Peter D. Hislop CUBO 14, 3 (2012) of the poles NV,m(r) of the meromorphic continuation of the truncated resolvent χVRV(λ)χV on each sheet Λm, counted with multiplicity, and with modulus at most r > 0. The order of growth of the resonance counting function NV,m(r) for HV on the m th-sheet is defined by ρV,m ≡ lim sup r→∞ logNV,m(r) logr . (6.9) It is known that ρV,m ≤ d for d ≥ 2 even [41, 42]. As in the odd dimensional case, it is proved that generically (in the sense of Baire typical) the resonance counting function has the maximal order of growth d on each non-physical sheet. Theorem 22. Let d ≥ 2 be even, and let K ⊂ Rd be a fixed, compact set with nonempty interior. Let MF(K) ⊂ L∞0 (K), for F = R or F = C, be the set of all real-valued, respectively, complex- valued potentials in L∞0 (K) such that the resonance counting functions NV,m(r), for m ∈ Z\{0}, have maximal order of growth. Then, the set MF(K) is a dense Gδ set for F = R or F = C. This theorem states that for a generic family of real or complex-valued potentials in L∞0 (K), the order of growth of the resonance counting function is maximal on each sheet, ρV,m = d, for m ∈ Z\{0}. This implies that there are generically infinitely many resonances on each nonphysical sheet. There are two challenges in proving Theorem 22. The first is to construct a function whose analytic extension to the mth-sheet Λm has zeros at the resonances of HV. This function will substitute for (6.2). The second problem is prove a lower bound (5.35) for some potential in L∞0 (K) in even dimensions. To resolve the first problem, we use the following key identity, that follows from (4.16) and the formulas for the meromorphic continuation of Hankel functions (see [6, section 6] or [23, chapter 7]), relating the free resolvent on Λm to that on Λ0, for any m ∈ Z, R0(e imπλ) = R0(λ) − m(d)T(λ), where m(d) = { m mod 2 d odd m d even. (6.10) The operator T(λ) on L2(Rd) has a Schwartz kernel T(λ;x,y) = iπ(2π)−dλd−2 ∫ Sd−1 eiλ(x−y)·ωdω, (6.11) see [19, Section 1.6]. This operator is related to M(λ) introduced in section 4.2 in (4.13) (see also (5.24)). We note that for any χ ∈ C∞0 (Rd), χT(λ)χ is a holomorphic trace-class operator for λ ∈ C. The operator T(λ) has a kernel proportional to |x−y|(−d+2)/2J(d−2)/2(λ|x−y|) when d is odd, and to |x − y|(−d+2)/2N(d−2)/2(λ|x − y|) when d is even. The different behavior of the free resolvent for d odd or even is encoded in (6.10). By the second resolvent formula (4.20), the poles of RV(λ) with multiplicity, correspond to the zeros of I+VR0(λ)χV. We can reduce the analysis of the zeros of the continuation of I+VR0(λ)χV CUBO 14, 3 (2012) Scattering theory and resonances ... 33 to Λm to the analysis of zeros of a related operator on Λ0 using (6.10). If 0 < argλ < π and m ∈ Z, then eimπλ ∈ Λm, and I + VR0(e imπλ)χ = I + V(R0(λ) − mT(λ))χV = (I + VR0(λ)χV)(I − m(I + VR0(λ)χV) −1VT(λ)χV). For any fixed V ∈ L∞0 (Rd), there are only finitely many poles of (I+VR0(λ)χV)−1 with 0 < argλ < π. Thus fV,m(λ) = det(I − m(I + VR0(λ)χV) −1VT(λ)χV) (6.12) is a holomorphic function of λ when 0 < argλ < π and |λ| > c0〈‖V‖L∞〉. Moreover, with at most a finite number of exceptions, the zeros of fV,m(λ), with 0 < argλ < π correspond, with multiplicity, to the poles of RV(λ) with mπ < argλ < (m + 1)π. Henceforth, we will consider the function fV,m(λ), for m ∈ Z∗ ≡ Z\{0}, on Λ0. For d odd, we are only interested in m = −1. In this case, the zeros of fV,−1(λ), for λ ∈ Λ0, correspond to the resonances. This provides an alternative to the S-matrix formalism, as presented in section 6.1, for estimating the resonance counting function in the odd dimensional case. The second problem in even dimensions is to prove that there are some potentials in L∞0 (K) for which the resonance counting function has the correct lower bound on each sheet. This is done by an explicit calculation. We prove (5.35) in even dimensions d ≥ 2 for Schrödinger operators HV with radial potentials V(x) = V0χBR(0)(x), with V0 > 0, using separation of variables and uniform asymptotics of Bessel and Hankel functions due to Olver [23, 24, 25]. This method can also be used in odd dimensions as an alternative to [47] thus providing examples as required in section 6.1. 7 Topics not covered and some literature This notes focussed on perturbations of the Laplacian on Rd by real-valued, smooth, compactly supported potentials. This is just one family of examples where resonances arise. Other topics concerning resonances include: (1) Complex-spectral deformation method and resonances (2) Obstacle scattering (3) Resonance free regions (4) Resonances for the wave equation (5) Resonances for elastic media (6) Resonances for manifolds hyperbolic at infinity (7) Semiclassical theory of resonances 34 Peter D. Hislop CUBO 14, 3 (2012) (8) Description of resonance wave functions (9) Approximate exponential decay of resonance states (10) Local energy decay estimates There are some reviews on resonances that cover many aspects of the theory in this list. These reviews include: (1) The long discussion by M. Zworski [49] that covers the complex scaling method developed by Sjöstrand and Zworski (inspired by the Baslev-Combes method) and its applications. (2) G. Vodev has written an expository article in Cubo [44]. Many aspects of resonances for elastic bodies and obstacle scattering are described there. (3) The proof of the generic properties of the resonance counting function for even and odd dimensions is described in Christiansen and Hislop [8], an expository summary written for les Journées EDP 2008 Evian available on the arXiv and from Cedram. (4) Text book versions of spectral deformation and quantum resonances, with an emphasis on the semiclassical regime, can be found in [9] and [14]. Finally, for a lighter and intuitive discussion of resonances, the reader is referred to Zworski’s article Resonances in physics and geometry that appeared in the Notices of the American Mathe- matical Society [50]. 7.1 Acknowledgements These notes are an extended version of lectures on scattering theory and resonances given as a mini-course during the Penn State-Göttingen International Summer School in Mathematics at the Pennsylvania State University in August 2010. I would like to thank the organizers Juan Gil, Thomas Krainer, Gerardo Mendoza, and Ingo Witt for the invitation to present this mini-course. I would like to thank Gerardo Mendoza and Peter A. Perry for some useful discussions. I also thank Tanya Christiansen for our enjoyable collaboration. I was partially supported by NSF grant DMS 0803379 during the time this work was done. 8 Appendix: Assorted results Two groups of results that are related to material in the text are summarized here. The first is a synopsis of the spectral theory of linear self-adjoint operators. The second is an analysis of the time evolution of states lying in various spectral subspaces of a self-adjoint operator. Detailed discussions of these topics may be found in the Reed-Simon series [29]-[32], for example, and many other texts. CUBO 14, 3 (2012) Scattering theory and resonances ... 35 8.1 Spectral theory Let A be a self-adjoint operator on a separable Hilbert space H. Then, there is a direct sum decomposition H = Hac(A)⊕Hsc(A)⊕Hpp(A) into three orthogonal subspaces that are A-invariant in that A : D(A) ∩ HX(A) → HX(A) for X = ac,sc,pp. The pure point subspace Hpp(A) is the closure of the span of all the eigenfunctions of A. The continuous subspace Hcon(A) ≡ Hac(A) ⊕ Hsc(A) is the orthogonal complement of Hpp(A). For most Schrödinger operators HV = −∆ + V, one has Hsc(HV) = ∅. The proof of the absence of singular continuous spectrum is one of the main applications of the Mourre estimate, see the discussion in section 4.1, [9, chapter 4], and the original paper [22]. As the names suggest, there is a measure associated with a self-adjoint operator and this measure has a Lebesgue decomposition into pure point and continuous measures. The continuous measure admits a decomposition relative to Lebesgue measure into a singular continuous and absolutely continuous parts. 8.2 The RAGE Theorem The RAGE Theorem (Ruelle, Amrein, Georgescu, Enss) (see, for example, [9, section 5.4]) is a general result about the averaged time evolution of states in the continuous subspace Hcont(A) of a self-adjoint operator A. Theorem 23. Let A be a self-adjoint operator and φ ∈ Hcont(A), where Hcont(A) is the continuous spectral subspace of A. Suppose that C is a bounded operator and that C(A + i)−1 is compact. Then, we have lim T→∞ 1 2T ∫T −T ‖Ce−itAφ‖ dt = 0. (8.1) Furthermore, if φ ∈ H satisfies (8.1), then φ ∈ Hcont(A). Let A = HV be a Schrödinger operator of the type considered here, and C = χBR(0), the characteristic function on a ball of radius R > 0 centered at the origin. The RAGE Theorem (8.1) states that a state, initially localized near the origin, and in the continuous spectral subspace of HV, will eventually leave this neighborhood of the origin in this time-averaged sense. The con- tinuous spectral subspace Hcont(HV) has a further decomposition into the singular and absolutely continuous subspaces. It is the possible recurrent behavior of states in the singular continuous subspace that requires the time averaging in (8.1). Corollary 24. Let HV be a self-adjoint operator on L 2(Rd). Let φ ∈ Hac(HV), where Hac(HV) is the absolutely continuous spectral subspace of HV. Let χK be the characteristic function for a compact subset K ⊂ Rd. Then, we have lim t→∞ ‖χKUV(t)φ‖ = 0. (8.2) As one might expect, if φ is a finite linear combination of eigenfunctions, then the state χKUV(t)φ remains localized for all time. Indeed, for any ǫ > 0, there is a compact subset Kǫ ⊂ Rd so that ‖χRd\KǫUV(t)φ‖ < ǫ, for all t. 36 Peter D. Hislop CUBO 14, 3 (2012) Received: November 2010. Revised: Enero 2012. References [1] W. O. Amrein, J. M. Jauch, K. B. Sinha, Scattering theory in quantum mechanics. Reading, MA: W. A. Benjamin, Inc., 1977. [2] T. Christiansen, Some lower bounds on the number of resonances in Euclidean scattering, Math. Res. Lett. 6 (1999), no. 2, 203–211. [3] T. Christiansen, Schrödinger operators with complex-valued potentials and no resonances, Duke Math. Journal 133, no. 2 (2006), 313-323. [4] T. Christiansen, Several complex variables and the distribution of resonances for potential scattering, Commun. Math. Phys 259 (2005), 711-728. [5] T. Christiansen and P. D. Hislop, The resonance counting function for Schrödinger operators with generic potentials, Math. Research Letters 12 (6) (2005), 821-826. [6] T. Christiansen and P. D. Hislop, Maximal order of growth for the resonance counting func- tion for generic potentials in even dimensions, Indiana University Mathematics Journal 59, 621–660 (2010). [7] T. J. Christiansen, P. D. Hislop, Some remarks on scattering resonances in even dimensions, in preparation. [8] T. J. Christiansen, P. D. Hislop, Resonances for Schrödinger operators with compactly sup- ported potentials, Journées Équations aux dérivées partielles, Évian, 2-6 juin 2008, GDR 2434 CNRS, Exposé no. III, 18 pages (available on Cedram). [9] H. L. Cycon, R. G. Forese, W. Kirsch, B. Simon, Schrödinger operators with applications to quantum mechanics and global geometry, Berlin: Springer-Verlag, 1987. [10] M. Dimassi, J. Sjöstrand, Spectral Asymptotics in the Semi-Classical Limit, London Math. Soc. Lecture Note Ser. 268, Cambridge, England: Cambridge Univ. Press, 1999. [11] R. Froese, Asymptotic distribution of resonances in one dimension, J. Differential Equations 137 (1997), no. 2, 251272. [12] R. Froese, Upper bounds for the resonance counting function of Schrdinger operators in odd dimensions, Canad. J. Math. 50 (1998), no. 3, 538546. [13] I. Herbst, Dilation analyticity in a constant electric field I. The two body problem, Commun. Math. Phys. 67, 279–298 (1979). CUBO 14, 3 (2012) Scattering theory and resonances ... 37 [14] P. D. Hislop, I. M. Sigal, Introduction to spectral theory with applications to Schrödinger operators, Applied Mathematics Series volume 113, New York: Springer, 1996. [15] A. Intissar, A polynomial bound on the number of the scattering poles for a potential in even dimensional spaces Rn, Comm. Partial Diff. Eqns. 11, No. 4 (1986), 367–396. [16] P. Lelong and L. Gruman, Entire functions of several complex variables, Berlin: Springer Verlag, 1986. [17] P. D. Lax and R. S. Phillips, Decaying modes for the wave equation in the exterior of an obstacle, Comm. Pure Appl. Math. 22 (1969), 737–787. [18] A. Martinez, An introduction to semiclassical and microlocal analysis, Universitext, New York: Springer-Verlag, 2002. [19] R. Melrose, Geometric scattering theory, Cambridge: Cambridge University Press, 1995. [20] R. Melrose, Scattering theory and the trace of the wave group, J. Funct. Anal. 45 (1982), 25–40. [21] R. Melrose, Polynomial bounds on the number of scattering poles, J. Funct. Anal. 53 (1983), 287–303. [22] E. Mourre, Absence of singular continuous spectra for certain selfadjoint operators, Commun. Math. Phys. 78, 391–408 (1981). [23] F. W. J. Olver, Asymptotics and Special Functions, Academic Press, San Deigo, 1974. [24] F. W. J. Olver, The asymptotic solution of linear differential equations of the second order for large values of a parameter, Phil. Trans. Royal Soc. London Ser. A 247, 307–327 (1954). [25] F. W. J. Olver, The asymptotic expansion of Bessel functions of large order, Phil. Trans. Royal Soc. London ser. A 247, 328–368 (1954). [26] P. A. Perry, Mellin transforms and scattering theory. I. Short range potentials, Duke Math. J. 47 (1980), no. 1, 187–193. [27] P. A. Perry, Scattering theory by the Enss method. Edited by B. Simon. Mathematical Re- ports, 1, Part 1. Harwood Academic Publishers, Chur, 1983. [28] V. Petkov, M. Zworski, Semi-classical estimates on the scattering determinant, Ann. H. Poincaré 2 (2001), 675–711. [29] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Volume I: Functional Analysis (revised and enlarged edition). New York: Academic Press, 1980. [30] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Volume II: Fourier Analysis, Self-Adjointness. New York: Academic Press, 1979. 38 Peter D. Hislop CUBO 14, 3 (2012) [31] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Volume III: Scattering The- ory. New York: Academic Press, 1979. [32] M. Reed, B. Simon, Methods of Modern Mathematical Physics, Volume IV: Analysis of Operators. New York: Academic Press, 1984. [33] D. Robert, Autour de l’approximation semi-classique, Progr. Math. 68, Boston: Birkhauser, 1987. [34] A. Sá Barreto, Remarks on the distribution of resonances in odd dimensional Euclidean scattering, Asymptot. Anal. 27 (2001), no. 2, 161–170. [35] A. Sá Barreto, Lower bounds for the number of resonances in even dimensional potential scattering, J. Funct. Anal. 169 (1999), 314–323. [36] A. Sá Barreto, S.-H. Tang, Existence of resonances in even dimensional potential scattering, Commun. Part. Diff. Eqns. 25 (2000), no. 5-6, 1143–1151. [37] N. Shenk, D. Thoe, Resonant states and poles of the scattering matrix for perturbations of −∆, J. Math. Anal. Appl. 37 (1972), 467–491. [38] I. M. Sigal, A. Soffer, The N-particle scattering problem: asymptotic completeness for short- range systems, Ann. of Math. (2) 126 (1987), no. 1, 35–108. [39] B. Simon, Trace ideals and their applications, second edition, Providence, RI: American Math. Soc., 2005. [40] A. Vasy, Scattering poles for negative potentials, Comm. Partial Differential Equations 22 (1997), no. 1-2, 185–194 [41] G. Vodev, Sharp polynomial bounds on the number of scattering poles for perturbations of the Laplacian, Commun. Math. Phys. 146 (1992), 39–49. [42] G. Vodev, Sharp bounds on the number of scattering poles in even-dimensional spaces, Duke Math. J. 74 (1) (1994), 1–17. [43] G. Vodev, Sharp bounds on the number of scattering poles in the two-dimensional case, Math. Nachr. 170 (1994), 287–297. [44] G. Vodev, Resonances in Euclidean scattering, Cubo Matemática Educacional 3 No. 1, Enero 2001, 319–360. [45] D. Yafaev, Mathematical scattering theory, Providence, RI: AMS, 2000. [46] M. Zworski, Distribution of poles for scattering on the real line, J. Funct. Anal. 73 (1987), 277–296. CUBO 14, 3 (2012) Scattering theory and resonances ... 39 [47] M. Zworski, Sharp polynomial bounds on the number of scattering poles of radial potentials, J. Funct. Anal. 82 (1989), 370–403. [48] M. Zworski, Poisson formulae for resonances, Séminaire sur les Equations aux Dérivées Partielles, 1996–1997, Exp. No. XIII, 14 pp., Ecole Polytech., Palaiseau, 1997. [49] M. Zworski, Counting scattering poles, In: Spectral and scattering theory (Sanda, 1992), 301–331, Lectures in Pure and Appl. Math. 161, New York: Dekker, 1994. [50] M. Zworski, Resonances in physics and geometry, Notices Amer. Math. Soc. 46, no. 3 (1999), 319–328.