Semiclassical Propagators of Wigner Function: A comparative study.
Jose Mauricio Sevilla Moreno
Universidad Nacional de Colombia Facultad, Departamento de F´ısica Ciudad, Colombia 2020
Semiclassical Propagators of Wigner Function: A comparative study.
Jose Mauricio Sevilla Moreno
Tesis presentada como requisito parcial para optar al t´ıtulode: Magister en Ciencias: F´ısica
Director: Carlos Leonardo Viviescas Ram´ırez
L´ıneade Investigaci´on: Caos Cu´anticoy M´etodos Semicl´asicos
Grupo de Investigaci´on: Caos y complejidad
Universidad Nacional de Colombia Facultad de Ciencias, Departamento de F´ısica Bogot´a,Colombia 2020
To my parents, Jos´eand Esperanza.
Acknowledgments
I would like to thank my thesis advisor, Dr. C. Viviescas, for welcoming me into the Caos y Complejidad group, working with me closely as a collaborator more than a superior, listening to all the comments and ideas I had al along the time I spent in the group. His support , patience and encouragement were essential on the development of this project as well as myself as a physicist.
I also would like to thank all the Caos and Complejidad group members whom were always open to listen and discuss ideas on the office.
But most of all, I need to thank my friends that were always there for me, listening, guiding and advising specially when I was feeling lost and without force to go on. Thanks to Alejandro Monta˜na, C´esar C´elis,Daniela Garz´on,Daniel Padilla, Frank Bautista, Juli´anTriana and Nicol´as Medina for all the things that have done for me that I can’t even try to describe in just a few words.
Last but not least, I want to thank my family, my parents, my brother and my sister, who always have been an inspiration source for my life.
ix Abstract
Semiclassical approximations for the dynamics have been widely used in different represen- tations of quantum mechanics. In particular, phase space representations exhibit a clear way to implement those approximations by direct comparison with classical mechanics theory. During the last years, such approximations have been recovering interest, due to the fact that numerical applications suit the modern computational architectures. In this work different semiclassical approximations are built and compared on performance and complexity, showing that they are a suitable way to calculate the quantum dynamics. A study over the caustics is presented on a the Morse potential showing the similarities of the final Wigner function considering and not considering them. Following this, we introduced an initial and final value representations (IVR and FVR) to compare the dynamical properties with the center-center representation proposed by Dittrich et al. [Dittrich et al., 2006] showing that the later can be used as a initial value representation in numerical applications, getting better performance with less complexity than the IVR and FVR using different criteria in such comparison. Contents
Acknowledgments vii
Abstract ix
0 Introduction 1
1 Theoretical Background 6
1.1 Wigner Function ...... 8
1.1.1 Dynamics ...... 9
1.2 Semiclassical Propagation ...... 11
1.2.1 Wigner function propagator ...... 12
1.2.2 Initial and Final Value Representations of the Propagator ...... 16
2 Numerical Implementation 23
2.1 Center-Center Representation ...... 24
2.1.1 Algorithm ...... 27
2.2 *Caustics Counting ...... 28
2.3 Initial Value Representation ...... 30 CONTENTS xi
2.3.1 Algorithm ...... 31
2.4 Final Value Representation ...... 32
2.4.1 Algorithm ...... 33
2.5 Comparison Criteria ...... 34
2.5.1 Observable Calculation ...... 34
2.5.2 Autocorrelation Functions ...... 34
2.5.3 Normalized Inner Product ...... 35
2.5.4 Marginals ...... 35
3 Results 37
3.1 Model: Morse Oscillator ...... 38
3.2 Caustics ...... 39
3.2.1 Initial State ...... 40
3.2.2 Determinant ...... 41
3.2.3 Reconstruction of the Final State ...... 43
3.2.4 Final Wigner and Wave Functions ...... 43
3.3 Center-Center Representation ...... 47
3.3.1 Marginals ...... 49
3.4 Initial Value Representation ...... 51
3.4.1 Wigner Function ...... 51
3.5 Final Value Representation ...... 53
3.5.1 Wigner Function ...... 53 xii CONTENTS
3.6 Comparison ...... 55
3.6.1 Autocorrelation ...... 56
3.6.2 Observables ...... 57
3.6.3 Normalized Inner Product ...... 61
3.6.4 Complexity ...... 61
3.7 Parallelization ...... 63
3.7.1 IVR-FVR ...... 63
3.7.2 Center-Center ...... 63
4 Conclusions and Perspectives 64
A Coherent States 66
A.1 Wigner Function for Coherent States ...... 66
A.2 Relationship with the Husimi Function ...... 69
B Wigner Function Calculations 71
B.1 Marginal Probabilities and Area ...... 71
B.2 Weyl Correspondence ...... 73
C Caustics 76
D Classical Evolution 79
D.1 Dynamics ...... 80
D.2 Distributions ...... 80 CONTENTS xiii
D.3 Numerical Implementation ...... 81
E Quantum Propagation 84
E.1 State Dynamics ...... 84
E.1.1 Algorithm ...... 87
E.2 Winger Function Dynamics ...... 88
E.2.1 Algorithm ...... 90
Bibliography 92 Chapter 0
Introduction
Form the very beginning of the development of quantum mechanics an interpretation of the theory was sought. The straightforward path was comparing this new, at that time, theory with the well known results developed so far, i.e., the classical mechanics. This comparison was not clear as the quantum theory has a mathematical structure which is clearly way different than the classical theory and therefore this had to be done using particular results such as expectation values. The quantum theory resulted to have different results than the classical mechanics. This was one of the biggest issues of quantum mechanics, as the community was not accepting the theory because of the lack of interpretation or the strangeness of some already existing. As the continuous comparison of classical and quantum mechanics showed that they are indeed no compatible, the question of the relationship between them remained. it was until 1928 when after studying an statistical interpretation of works done by Dirac [Dirac and Fowler, 1927], that van Vleck [Van Vleck, 1928] built up an expression of the quantum dynamics in terms of classical quantities which is used as a basis of the so called semiclassical methods. It uses the idea of classical propagation of a set of classical particles in order to get the time evolution of wave functions.
This idea of semiclassics, relies on the dynamics1, even when the dynamical rules of classical and quantum mechanics are not the same and even worse, they are not even comparable by some important reasons. First, the Schr¨odingerequation which is the quantum mechanical equation of motion is complex as well as a wave function ψ[Landau and E.M., 1977], and the Hamilton’s equations are not[Arnold et al., 2013]. Second, the representation of single particles is completely different, as in quantum mechanics a state is represented by a non localized function2, while in
1This means that these methods are not constructed to find other quantum characteristics, such as ground states. 2In fact, it is localized but not on a single point. 2 0 Introduction
classical mechanics, a state is defined to be a single point on phase space3, and so on.
The idea behind the van Vleck formula can be understood using several ways to proceed, for instance, from the Feynman’s path integral point of view [Feynman, 1948, Feynman, 1966], which consists on the usage to trajectories to construct the dynamics of the quantum mechanical state. This trajectories do not necessary represent classically possible trajectories, but the classical trajectories are also included. We may do some comments about this way to proceed, first is that for the evolution of a single quantum mechanical particle represented via a wave function cannot be evolved using a single classical particle but a distribution of particles instead, which leads to the question Shall we use a set of classical particles to compare a single quantum mechanical particle? and the answer is yes, but this brings up more questions about what happen with the amount of quantum particles on a limit where quantum mechanics goes to classical mechanics, but we will not discuss about this concern4, a second comment is that, the evolution is described by two terms, an amplitude and the exponential of a phase. The two of them depend on the particular trajectory taken but these two quantities are calculated from an action corresponding to the chosen path. The semiclassics here enters on reducing the complete set of trajectories to just the classical ones.
This result has huge implications. In particular that, the van Vleck propagator itself is not enough to describe certain quantum mechanical features such as tunnelling, due to the fact that as we are only considering the classical trajectories, there are forbidden areas so cannot be transport of probability or density classically speaking, but it occurs quantum mechanically.
The van Vleck formula has another issue which was pointed out by Gutzwiller [Gutzwiller, 1967], as the amplitude term has some insights that play a very important role on the dynamics an ad- ditional treatment shall to be added to the van Vleck result. This is basically what are called caustics (For more on caustics, see appendix C), the caustics are points (lines, surfaces and so on depending the order and the number of degrees of freedom of the system) where many trajectories concentrate, this generates a divergence on the amplitude that is not real. As semiclassics is not valid on that exact point5, but it is just before and after in such a way that the only difference is that the trajectory gains a phase of π/2 on the caustic[Maslov and Fedoriuk, 1981], this lead to a big problem and it is that the caustics must be counted, so that at the end of the propagation the complete phase is added to the contribution of each particular trajectory.
This, now complete way of propagate is known as the van Vleck-Gutzwiller propagator, and it has been widely spread used after its development, for example Gutzwiller got an expression to calculate using this semiclassics scheme, an approximation for the spectrum of a quantum
3There are some definitions of phase space on the classical theory, but we refer to the Hamilton’s phase-space. 4We will see what kind of approach we may take in this context to get to a Classical limit. 5Which does not mean that one cannot get the information of the dynamics on that specific point, but means that the trajectory which is on a caustic, shall not count be counted 3 mechanical system which is known as the Gutzwiller trace formula [Gutzwiller, 1969]. This formula is very important as this is the path to go into quantum chaos.
The van Vleck-Gutzwiller formula requires a certain level of computational resources, and thereby initially it was not possible to use it as a tool to calculate the dynamics, as it demands a lot of calculations that is only possible to be done using computers. Nowadays, this kind of semiclassical schemes have arisen to be a suitable way to add quantum effects to simulation of systems where quantum mechanics is present but very expensive to be calculated, such as it occur in molecular dynamics simulations.
All of these semiclassical structures use as a basis the van Vleck propagator, even the propa- gations based on the Herman Kluk [Herman and Kluk, 1984] and Frozen Gaussians [Heller, 1981]6 which commonly are thought to be independent of it, have a strong and deep relationship on the sense that they both are based on the idea on expanding on coherent states (Gaussians) which are not a complete basis of functions but an overcomplete one, it means that the expansion of a given function is not unique and it can depend on several features of the specific set of Gaussians chosen.
The relationship relies on the fact that the Herman Kluk propagator smooths the functions with a Gaussian kernel7 in such a way that if we take the limit of the width of the Gaussians γ → 0 we recover the van Vleck propagator.
This means that, fundamentally, the Herman Kluk propagator lacks the contributions made by Gutzwiller to the van Vleck propagator, which initially was considered by Heller [Heller, 1981] but missed out by Herman and Kluk [Herman and Kluk, 1984] on their justification of Heller’s method. This way to propagate is reduced to calculate a Gaussian smoothed version of the final wave function, that may be troublesome as the width of the Gaussians is fixed beforehand the evolution.
Adding quantum corrections via the van Vleck-Gutzwiller propagation is still not good enough for the limitations that we have already mentioned among others, so to solve those problems, there is an interest on using a different representation which lacks these issues. The quan- tum phase space representations have been proved to work successfully[Dittrich et al., 2006, Ozorio de Almeida et al., 2013, Ozorio de Almeida et al., 2019]. The main idea behind phase space representation of quantum mechanics is using functions instead of operators, and that can be done on many different ways, but the important feature of these representations is that there exists a clearer way to make direct comparison with the classical mechanics, as they can be
6Even this work, which is constructed from a conceptual point of view without any proof or derivation, just intuition. 7This is exactly the same idea of the Husimi function, which is not the best option when looking for the dynamics as it will be explained on the following chapter on in appendix A 4 0 Introduction
written on some similar spaces[Zachos et al., 2005, P. Schleich, 2001].
As phase space of quantum mechanics and classical mechanics can be built up on similar contexts, it is worth to think about the links between them, and in particular the role that semiclassical approximations take on the phase space representations of quantum mechanics. This comparison have led to a preferred phase space which is the Wigner phase space where the Wigner function is the representation of the density operator and therefore of the state of the system. This because of its dynamics has very interesting implications and it is the most suitable compared with other where the dynamics cannot even be defined completely such as Husimi function.
The Wigner function has a wide range of applications, for instance, in semiclassical methods [Ozorio de Almeida et al., 2019, Ozorio de Almeida et al., 2013, Dittrich et al., 2006, Dittrich et al., 2010, Koda, 2015], as in this the case, but also in signal process and analysis [Daniela, 2005], classical and quantum optics [P. Schleich, 2001], molecular, atomic and nuclear physics [Wigner, 1932], quantum statistical mechanics [Wigner, 1932], decoherence or quantum chaos[P. Schleich, 2001].
The semiclassical approximations on the context of Wigner function has been growing on the recent years [Dittrich et al., 2006, Dittrich et al., 2010, Ozorio de Almeida et al., 2013, Ozorio de Almeida et al., 2019, Koda, 2015], using a different kind of systems to prove their flexibility and robustness. The implementation of these propagators have led to comparisons at the numerical level that have shown to lack justification in the sense that one stands that the van Vleck approach proposed in [Ozorio de Almeida et al., 2013] is not valid for long times while the Herman Kluk approach does[Koda, 2015]. While in the other hand a more recent work showed that in fact the van Vleck approach [Ozorio de Almeida et al., 2013] can be used for very long times8 [Ozorio de Almeida et al., 2019]. This basically shows that the comparison previously done is unfair and poorly implemented.
In this work we implemented a series of semiclassical propagators based on the van Vleck propagator [Dittrich et al., 2006, Dittrich et al., 2010, Ozorio de Almeida et al., 2013, Ozorio de Almeida et al., 2019] as lately they have been strongly criticized [Koda, 2015] but with not very strong arguments, which seems a misinterpretation of the way the methods must be implemented. This docu- ment is organized as follows: A brief introduction to quantum mechanics on phase space and its dynamics, after we construct and discuss a semiclassical propagator [Dittrich et al., 2006, Dittrich et al., 2010], then we follow the way to proceed of [Ozorio de Almeida et al., 2013] to show that the previously presented propagator can be written on some center-center represen- tation so that we eliminate some divergences, just as it happens for the mixed propagators shown in [Ozorio de Almeida et al., 2013]. After that, we explain the numerical implementa-
8We reefer as long times as the Ehrenfest time, which is used to characterize quantum chaotic systems and its correlations. 5 tion and how the propagator [Dittrich et al., 2006, Dittrich et al., 2010] can be understood as an initial value representation, even when it is said not to be one on the literature[Koda, 2015, Ozorio de Almeida et al., 2013]. Finally we present the results of the dynamics for different times using a typical system and lastly some conclusions and perspectives. Chapter 1
Theoretical Background
Phase space representation of quantum mechanics started from the development of the theory itself in the sense that it was under constant comparison with the classical theory, in the attempt to have a better interpretation of the theoretical and experimental results. It was then, during 1932, that Eugine Wigner [Wigner, 1932] built up the first so called phase space representation, when trying to calculate quantum observables just as it is done in classical statistical mechanics. This comparison led to a function which behaves just as a density probability distribution to the calculation of observables, but the function resulted to be negative in some parts of the domain, and as a consequence it could not be called a probability distribution. Nevertheless, as it satisfied all the other properties a distribution does, it is called a pseudo probability distribution. This issue notwithstanding, the function is smooth and well behaved in all the domain, so it has been widely studied as a suitable representation of quantum mechanics.
This is not the only possible phase space representation, and actually there are an infinite number of them, because the key point is to construct functions from operators andthis can be done in several ways. For instance, one can consider the following way to construct a phase space representation P (p, q) [Hillery et al., 1997, Ozorio de Almeida, 1998] of the quantum state: re- placing operators qˆ by the variable q and doing exactly the same for the momenta, then tracing in order to get a function instead of an operator Pˆ(pˆ, qˆ), h i P (p, q) = Tr Pˆ(pˆ, qˆ) δ(qˆ − q, pˆ − p) . (1-1)
It is possible to choose the representation of the operator or the delta function before the tracing on many different ways, each one will induce a basis in which the trace is calculated, and each one of them generates a particular phase space with plenty different properties.
This can be translated to the second quantization formalism and, therefore, uses a ordering 7 rule indicating a clear way to proceed. Usually three ordering recipes are followed, normal ordering, antinormal ordering and mixed ordering, each one of them leading to one of the most commonly used representations [P. Schleich, 2001]:
• Normal Ordering: Husimi Phase Space.
• Anti-Normal Ordering: Glauber Phase Space.
• Mixed Ordering: Wigner Phase Space.
In principle, all of these three options can be used in order to represent quantum mechanical systems, but each one of them has its unique properties and hence advantages and disadvantages regarding an specific problem. For instance, the Q or Husimi function [Husimi, 1940], which results of constructing the Husimi phase space representation of the density operator is well known and used because it is defined to be always positive; it can be interpreted as a probability distribution. It also presents some deep problems as it is defined from the density operator and a particular basis of coherent states (see appendix A) which form an overcomplete basis, and leads to losing some features of the state after using a Gaussian kernel to smooth the function. Regarding the dynamics, the Husimi function has plenty of troubles, mainly because as it is a smooth version of the function that represents the state, or in other words, it is constructed from a reduced version of the density operator where the coherences get lost, it is not possible find the Husimi function on a time t > t0 from the Husimi function of a time 1 t0 [O’Connell and Wigner, 1981b] . So the Husimi function offers an easy interpretation of the phase space representations of quantum mechanics, but is very restricted when applied to specific problems such as the dynamics of the quantum mechanical system. It is also worth mention that it performs well when looking for probability concentrations or a straightforward way to interpret and compare with the corresponding classical phase space. Something similar occurs when describing quantum mechanics in the Glauber phase space, where the density operator is know as the P or Glauber-Sudarshan function [Glauber, 1963]. This function is also built from a coherent states basis but in contrast to the Husimi function, the P function can take negative values. It has some interesting properties such as, in the field of quantum optics, the P function represents a true probability density for classical states of light2. However, this implies that the P function has problems regarding non classicality. This idea of a probability distribution breaks when calculating probabilities of mutually exclusive states, which cannot be described by the P function, due to the coherent states not being orthogonal among themselves [Leonhardt et al., 1997].
1It is possible to find a equation of motion for the Husimi function from the equation of motion of the Wigner function, but it is not possible to guarantee that the evolved Husimi function from a time t = t0 is the same that the one transformed from the Wigner function at a time t0. 2The definition is done usually the other way around, a classical state of light is one such that the P function can be considered a probability density. 8 1 Theoretical Background
1.1 Wigner Function
The Wigner function was the first of the phase space representations ever proposed, and can be defined from the Weyl transform. For any operator Aˆ it returns its Weyl symbol, Z i D y yE A(p, q, t) = dy exp − p · y q − Aˆ(t) q + , (1-2) h¯ 2 2 where (p, q) are the coordinates of phase space, and h¯ is Planck constant divided by 2π. This relation allows us to construct the Wigner phase space representation of any quantum mechanical operator, a necessary step if we want to calculate quantities such as expectation values or the time evolution in the phase space representation.
If we consider the specific case of transforming the density operator, so that we obtain the function that represents the state of the quantum mechanical system, we get the Wigner function,
1 Z i D y yE W (p, q, t) = dy exp − p · y q − ρˆ(t) q + . (1-3) (πh¯)d h¯ 2 2 One of the most important features of this phase space representation is that it is equivalent to any other representation of quantum mechanics, for instance, the position or momentum representations. This relation can be inverted and the density matrix can be reconstructed from the Wigner function (See appendix B).
The Weyl representation has many different characteristics and properties that make it a very useful representation of quantum mechanics, specially when a comparison with classical mechanics is wanted, as it is the case in this work. We now list some of the principal characteristics of the Wigner function[Ozorio de Almeida, 1989, Hillery et al., 1997, P. Schleich, 2001]:
• As a consequence of the hermiticity of the density operator, the Wigner function is real.
• It is possible to calculate the position probability density just by applying one integral, Z |ψ(q)|2 = dpW (q, p). (1-4)
• The same can be done for the momentum representation, Z |φ(p)|2 = dqW (q, p). (1-5)
• it is normalized, Z Z dq dpW (p, q) = Tr(ˆρ) = 1. (1-6) 1.1 Wigner Function 9
• For systems up to quadratic potentials, the evolution corresponds to the classical Liouville equation.
• The previous feature can be extended when considering the classical limit of the dynamics. The evolution of the Wigner function in the limit h¯ → 0 is given by the Liouville equation. This entails huge consequences. For example, if a comparison to classical mechanics is wanted, it is not possible to compare a single quantum mechanical particle with a single classical particle, but a comparison of a single quantum mechanical particle with a complete classical distribution is necessary 3.
• The only possible state in which the Wigner function is always positive is a Gaussian sate (e.g., coherent or thermal states), otherwise it will take negative values [Hudson, 1974].
1.1.1 Dynamics
We have previously mentioned properties of the dynamics in phase space yet, so far, we have not describe it properly. In this section we will mention some features and properties of the time evolution.
The dynamics of the Wigner function can be found by evaluating the time derivative of the Weyl transform of the density operator (1-3) and then using the von-Neumann equation for the dynamics of the density operator. Hence, it becomes necessary to Weyl transform the commutator of the density operator and the Hamiltonian function. This procedure leads to a Bracket-like equation that defines a lie algebra called the Moyal bracket,
∂W (p, q) = {H,W } . (1-7) ∂t Moyal The Moyal bracket is defined as,
{A, B}Moyal = A?B − B ? A, (1-8) in terms of the ? product, also called the Moyal product, ←− −→ ←− −→ !! ih¯ ∂ ∂ ∂ ∂ ? = exp − , (1-9) 2 ∂p ∂q ∂q ∂p where the direction of the arrows over the operators indicate the direction in which they operate.
3The idea of taking ¯h → 0 seems nonsensical, but what is really happening is that the typical actions on the system are large enough, so that the ratio between the actions and ¯h increases rapidly, which is equivalent to take this limit. 10 1 Theoretical Background
This differential equation is, in general, an infinite order partial differential equation. It corresponds exactly to the Poisson bracket if up to quadratic Hamiltonians are considered, so the quantum information shows up in cases of Non-linear Hamiltonians.
Propagator
Even though this equation contains high order derivatives, it is linear, and therefore it can be solved by using a Kernel, Z 0 0 0 0 0 0 W (p, q, t) = dq dp G(p , q , t0; p, q, t)W (p , q , t0). (1-10)
0 0 The propagator G(p , q , t0; p, q, t) is called the Wigner function propagator [Dittrich et al., 2006] or Wigner propagator, for short, and shares some properties with the quantum mechanical prop- agator on position representation, e.g. it also form a group.
Contrary to what one would expect, the Wigner propagator is not found by doing the Weyl transform of the position representation propagator. As the Wigner function is built from the density operator (i.e., two waves functions), it is necessary to consider the convolution of the Weyl symbols of two time evolution operators instead. Adopting the notation for a point on phase-space r = (p, q), we obtain for the Wigner propagator
1 Z i G (r00, r0, t) = dr0 exp r ∧ (r00 − r0) W hd h¯ r00 + r0 r r00 + r0 r U − U + , (1-11) W 2 2 W 2 2
where the UW are the Weyl transform of the time evolution operator in position representation or Feynman propagator K,
Z i q0 q0 U (r, t00, t0) = dq0 exp − p · q K q + , t00, q − , t0 . (1-12) W h¯ 2 2
Hence, the propagator of the Wigner function in terms of the Feynman propagator is,
2 d Z i G (r00, r0, t) = dQ0dQ00 exp (p0 · Q0 − p00 · Q00) W h h¯ Q00 Q0 Q00 Q0 K∗ q00 − , t00, q0 − , t0 K q00 + , t00, q0 + , t0 . (1-13) 2 2 2 2
This last expression will be used in the following section in order to derive the semiclassical Wigner function propagator, the main object of this work. 1.2 Semiclassical Propagation 11
1.2 Semiclassical Propagation
The principal ingredient to the semiclassical approximations presented here is the van Vleck propagator [Van Vleck, 1928]. It was developed with a very similar goal than the Wigner function, both were done while looking for an interpretation of quantum mechanics calculating averages. The derivation done by van Vleck is not complete, because it does not consider the caustics (Apendix C) that come out of a folding on the Lagrangian manifold where the trajectories exist. These were added later by Gutzwiller [Gutzwiller, 1967, Gutzwiller, 1969]. Gutzwiller was also one of the first to adopt this strategy of semiclassical propagation for doing dynamics, and use it to calculate the energy spectrum of a quantum mechanical systems [Gutzwiller, 1970], proving the validity of the van Vleck propagation, which until that point in time was consider a purely theoretical artefact due to the computational power required to be able to use it.
The derivation of the van-Vleck-Gutzwiller propagator can be done following different paths, for instance, Berry and Mount in [Berry and Mount, 1972] did a very complete revision on this by using the Feynman path integral strategy, Littlejohn [Littlejohn, 1992] and Keller [Keller, 1985], used the WKB approximation which allowed them to make a clearer geometrical interpretation of it including the caustics.
This problem of the caustics is a matter of representation, in the sense that the points that are caustics on a particular representation are not necessary caustics on a different representation. The work done by Littlejohn [Littlejohn, 1992] clearly illustrates that the caustics on position representation are not in momentum, and therefore a phase is gain after crossing a single caustic depending the order, this issue will be discussed later on the appendix C.
The semiclassical van Vleck-Gutzwiller propagator is [Van Vleck, 1928, Gutzwiller, 1967],
X m d/2 i π K (q0, q, t) = exp R (q0, q, t) − i µ , (1-14) vV 2πiht¯ h¯ j 2 j j where R(q0, q, t) is Hamilton’s principal function of the classical trajectory connecting q and q0 in time t, µ is the Maslov index [Ozorio de Almeida, 2009, Maslov and Fedoriuk, 1981] and the summation over j’s takes into account all the possible trajectories connecting the two points; there may be more than one because in this case the momentum is a free parameter that can lead to different trajectories and therefore different values of R and µ.
This propagator can be rewritten in terms of Hamilton’s principal function only, s 2 0 0 X 1 ∂ Rj(q , q, t) i 0 π KvV (q , q, t) = exp Rj(q , q, t) − i µj , (1-15) hd ∂q0∂q00 h¯ 2 j 12 1 Theoretical Background
This expression is richer on interpretation, as it makes clear how µj must be calculated in terms of the change of sign of the Hessian matrix of Rj. This procedure is, indeed, very heavy compu- tationally as there is no a clear way to count the number of caustics a particular trajectory has crossed.
This propagator is exact for up to quadratic potentials, e.g., for free particles, harmonic wells or quadratic barriers. This is one of the most challenging issues this scheme of propagation has: classically there is no such concept as tunnelling but on a quadratic barriers in quantum mechanics there is. Works on this direction have been done, successfully explaining why this propagator should reproduce tunnelling on quadratic barriers [Jaubert and de Aguiar, 2007, Kay, 2013].
In addition, the evaluation of this propagator is numerically demanding. As the propagation from one specific position to other position is required, it is necessary to find all trajectories that connect these two points on a given time t. This problem is known as the root-search problem.
1.2.1 Wigner function propagator
The van Vleck-Gutzwiller version of the Wigner function propagator is done by replacing (1-15) into (1-13), but this is not as simple as it seems, as some considerations have to be taken in order to get a good approximation. The propagator of the Wigner function has to be built from two Weyl symbols of the time evolution operator[Dittrich et al., 2006, Dittrich et al., 2010], so it is given by
2d Z 0 −1/2 0 −1/2 00 0 2 X r r GW (r , r , t) = dr det Mj r + + I det Mj0 r − + I h 2 2 j,j0 i 00 0 i r r × exp r ∧ (r − r ) exp A ¯r + − A 0 ¯r − h¯ h¯ j 2 j 2 i r r 00 0 iπ × exp − H ¯r + − H 0 ¯r − (t − t ) + (µ − µ 0 ) , (1-16) h¯ j 2 j 2 2 j j
where the Mj’s are the stability matrices of the corresponding j trajectory, H the system Hamil- tonian, and Aj represents the symplectic area enclosed by the j classical trajectory as we explain now.
The symplectic area is constructed as follows: Consider a given trajectory on phase space
from point rj(ti) to point rj(ti) and the straight line joining them, the area enclosed is called the symplectic area, as shown in figure 1-1. 1.2 Semiclassical Propagation 13
p
rj(tf )
rj(ti)
q
Figure 1-1: Representation of the symplectic area of a single trajectory. It is constructed with the area enclosed between the trajectory and a straight line from the final to the initial point (On phase space).
In fact, it is possible to use these definitions to write the propagator in terms of the actions S of the trajectories,
2d Z 0 −1/2 0 −1/2 00 0 2 X r r GW (r , r , t) = dr det Mj r + + I det Mj0 r − + I h 2 2 j,j0 i 00 0 i r r iπ × exp r ∧ (r − r ) exp S ¯r + − S 0 ¯r − + (µ − µ 0 ) . (1-17) h¯ h¯ j 2 j 2 2 j j
This way of writing the propagator highlights some of its most relevant features. Let us start by pointing out the double sum over trajectories. Each set of trajectories provides the support for each one of the time evolution operators. In particular, if the same set of trajectories (j = j0) is chosen for both propagators, a selection that goes by the name of the diagonal approximation, and an integration is done using the WKB approximation up to first order on the phase, the Wigner propagator reduces to the Liouville propagator, corresponding to the classical distribution propagator,
00 00 00 0 0 ∂rcl 0 0 00 0 00 00 00 0 00 0 GW (r , t , r , t ) ≈ det 0 δ (r − rcl(r , t , t )) = δ (r − rcl(r , t , t )) . (1-18) ∂rcl
This result recalls the van Vleck propagation on position representation, where the evolution is done following classical trajectories, but without changing the amplitudes. Though, in this approximation the Wigner propagator corresponds to the Liouville propagator, the comparison of the results with the classical mechanics case must be done using a single quantum mechanical particle and a complete distribution of classical particles, i.e., Wigner functions with density of particles. This is particularly strange, because the Wigner function is not a probability density 14 1 Theoretical Background
distribution as it takes negative values in some parts, yet it is precisely this feature what will help us identify the regions on phase space where the quantum effects take place more prominently.
We can, however, go beyond the diagonal approximation in (1-17), improving on the van Vleck propagation. The way to proceed is by considering the j 6= j0 contributions, and solving the integral in stationary phase approximation considering terms up to second order4. This imposes a strong rule over the trajectories that are going to be used to do the propagation. This chord rule demands that trajectories must be now consider in pairs [Dittrich et al., 2006, Dittrich et al., 2010], and these pairs evolve following the classical evolution given by Hamilton’s equations with the corresponding classical Hamiltonian. So, let us consider the case of just two of these trajectories j 6= j0 and, for simplicity sake, label them as j = 1 and j0 = 2, hence the chord rule states that 1 J(r00 − r0 − (r00 − r0 + r00 + r0 )) = 0, (1-19) 2 2 2 1 1 where J is the symplectic matrix. This is not a condition on the trajectories themselves, but over the midpoint between both trajectories, which can be make evident by writting r r = ¯r − (1-20a) 1 2 and r r = ¯r + , (1-20b) 2 2 where ¯r represents the midpoint at each time5. To conclude, we notice that the amplitude term given by the second order approximation yields
∂ Mj − Mj0 0 2 (Sj (r1, t) − Sj (r2)) = J . (1-21) ∂r (Mj + I)(Mj0 + I)
The van Vleck-Gutzwiller version of the propagator of the Wigner function is then given by