Application of Self-Adjoint Extensions to The
APPLICATION OF SELF-ADJOINT
EXTENSIONS TO THE RELATIVISTIC AND
NON-RELATIVISTIC COULOMB PROBLEM
by
SCOTT BECK
submitted in partial fulfillment for the degree of
Doctor of
Physics
CASE WESTERN RESERVE UNIVERSITY
August, 2016 CASE WESTERN RESERVE UNIVERSITY SCHOOL OF GRADUATE STUDIES
We hereby approve the thesis/dissertation of Scott Beck
candidate for the degree of Physics∗.
Committee Chair Harsh Mathur
Committee Member Walter Lambrecht
Committee Member Mark Meckes
Committee Member Andrew Tolley
Date of Defense 9 June 2016
∗ We also certify that written approval has been obtained
for any proprietary material contained therein.
1 Contents
1 Introduction 1
2 Self-Adjoint Extensions 7 2.1 Particle in a box ...... 9 2.2 Delta Function ...... 10 2.3 Scale Invariance ...... 13
3 Experimental systems 18 3.1 Liquid Helium ...... 18 3.2 Rydberg Atoms ...... 20 3.2.1 First Order Quantum Defect ...... 21 3.2.2 Higher Order Corrections ...... 25 3.3 Graphene ...... 27
4 Non-Relativistic Coulomb Problem 31 4.1 Helium ...... 38 4.2 Alkali atoms ...... 41
5 Graphene 46 5.1 Separation of Variables ...... 47 5.2 Self-Adjoint Extensions ...... 49 5.3 Solutions and Phase Shift ...... 52
2 6 Conclusions 55
A Special Functions 58
B Numerical Fitting 62
C Akaike’s Information Criterion 67
Bibliography 68
3 List of Figures
3.1 Sample spectra of the transition in the electron bound states made by Grimes and Brown in 1976 [15]...... 20 3.2 Observed linearity in the conductivity of the graphene with varying gate
voltages. The point at Vg = 0 corresponds to the fermi energy [20]. . . . 28 3.3 Calculated conductivity in graphene based on Random Phase Approxima- tion screening [19]. The conducivity curve exhibits the same linear pattern observed in the experiments of Geim et. al. in figure 3.2 ...... 29 3.4 Local Density of states (inset) exhibits the same linear behavior as the conductivity curves observed in the experiments, as seen in figure 3.2 [18]. 30
4.1 Graph of λ(κ) described by equation (4.0.14). The x-axis is in units of Hartrees corresponding the κ values. The horizontal line is at λ ≈ .025, the fitted value from the cesium data listed in table (4.2), with the intersections showing the quantized energy levels...... 37
4 List of Tables
4.1 Reproduction of the transitions measured in Lithium 7 [25]. All of the states are in the s orbital, with spin angular momentum 1/2. Measure- ments that included orbitals with ` 6= 0 have been excluded. All units are MHz, converted from the original cm−1...... 42 4.2 Reproduction of the transitions measured in Sodium 23 [25]. All of the states are in the s orbital, with spin angular momentum 1/2. Measure- ments that included orbitals with ` 6= 0 have been excluded. Units are in MHz, converted from the original cm−1 ...... 42 4.3 Reproduction of the transitions measured in Potassium 39 by Lorenzen et. al. [26]. All of the states are in the s orbital, with spin angular momentum 1/2. Measurements have been converted from the original wave numbers reported (cm−1) to frequencies in MHz. The error associated with each measurement is approximately 21 Mhz...... 43 4.4 Reproduction of the transitions measured in Cesium 133 by Goy et. al. [24]. All of the states are in the s orbital, with spin angular momentum 1/2. Measurements that included orbitals with ` 6= 0 have been excluded. 43 4.5 Results of the fitting procedure for the first four Alkali atoms. The residual values reported are the average square residual, determined by dividing the sum of the square residuals by the total number of energy levels fitted. All are in units of MHz2 ...... 44
5 4.6 Comparison of the AIC for the quantum defect model (δ) and the self-
adjoint extension model (λ). The difference ∆ = AICλ − AICδ shows that there is almost no evidence to support the self-adjoint as the preferable model. See Appendix C for further discussion on the AIC...... 45
6 Acknowledgments
I would sincerely like to thank my advisor, Harsh Mathur. His guidance and sugges- tions during the writing and defending of the thesis were invaluable. I would also like to thank David Jacobs, who made significant contributions during the early stages of this project. Finally, I would like to thank my friends and family for their continued support throughout this process.
7 Abstract
The Coulomb problem was one of the first successful applications of quantum theory and is a staple topic in textbooks. However there is an ambiguity in the solution to the problem that is seldom discussed in either textbooks or the literature. The ambiguity arises in the boundary conditions that must be applied at the origin where the Coulomb potential is singular. The textbook boundary condition is generally not the only one that is permissible or the one that is most appropriate. Here we revisit the question of boundary conditions using the mathematical method of self-adjoint extensions in context of modern realizations of the Coulomb problem in electrons on helium, Rydberg atoms and graphene. We determine the family of allowed boundary conditions for the non- relativistic Schr¨odingerequation in one and three dimensions and the relativistic Dirac equation in two dimensions. The boundary conditions are found to break the classical SO(4) Runge-Lenz symmetry of the non-relativistic Coulomb problem in three dimensions and to break scale invariance for the two dimensional Dirac problem. The symmetry breaking is analogous to the anomaly phenomenon in quantum field theory. Electrons on helium have been extensively studied for their potential use in quantum computing and as a laboratory for condensed matter physics. The trapped electrons provide a realization of the one dimensional non-relativistic Coulomb problem. Using the method of self-adjoint extensions we are able to reproduce the observed energy levels of electrons on helium which are known to deviate from the textbook Balmer formula. We also study the connection between the method of self-adjoint extensions and an older theoretical model introduced by Cole [1]. Rydberg atoms have potential applications to atomic clocks and precision atomic experiments. They are hydrogen-like in that they have a single highly excited electron that orbits a small positively charged core. We compare the observed spectrum of several species of Rydberg atoms to the predictions of the Coulomb model with self-adjoint extension and to the predictions of the more elaborate quantum defect model which is generally found to be more accurate. The motion of electrons on atomically flat sheets of graphene is governed by the massless Dirac equation. The effect of charged impurities on the electronic states of graphene has been studied using scanning probe microscopy. Here we use the method of self-adjoint extensions to analyze the scattering of electrons from the charged impurities; our results generalize prior theoretical work which considered only one of the family of possible boundary conditions.
2 Chapter 1
Introduction
Historically the Coulomb problem was one of the first significant applications of quan- tum mechanics [2, 3]. Pauli solved the Coulomb problem using algebraic methods before the invention of the wave mechanics. Schr¨odingerthen solved his eponymous equation to obtain the same result. The method followed by Schr¨odingeris now a staple of text- books [4]. However, it is seldom mentioned that there is an ambiguity in the boundary conditions that should be applied to the origin in the ` = 0 channel corresponding to zero angular momentum. The conventional treatment assumes that the wave function should be regular at the origin and this can generally be justified on the grounds that quantum mechanical wave functions should be square integrable. In the ` = 0 channel, the irregular solution has a sufficiently weak divergence that it also remains square in- tegrable and cannot be discarded on that ground alone. In fact it is now understood that there is a one parameter family of boundary conditions which are compatible with the principles of quantum mechanics. Physically these boundary conditions represent a contact interaction experienced by the particle at the origin. The textbook boundary condition that the wave function must be regular at the origin corresponds to the case of no contact interaction. The mathematical methods needed to analyze these subtle boundary problems were
1 developed by von Neumann [5] in the course of his wide ranging study of the mathemati- cal foundations of quantum mechanics. The key mathematical idea is that the boundary conditions must be compatible with the requirement that the Hamiltonian operator for the problem is self-adjoint, instead of merely hermitian. This distinction will be elabo- rated in Chapter 2. The method of self-adjoint extensions allows determination of all the boundary conditions that are compatible with the self-adjointness of the Hamiltonian. However it does not determine which of these boundary conditions should actually be used. That is a question of physics and must be determined by either appeal to experi- ment or by appeal to a more complete theory to which the Hamiltonian being analyzed is an effective approximation. Although the method of self-adjoint extensions was available soon after the discovery of quantum mechanics it has found few applications and is not part of the standard toolbox of most physicists. For example, there is no discussion of the method in any of the undergraduate or graduate textbooks that are in common use. A goal of this thesis is to find new applications of the method to problems that are of current interest. The textbook analysis of the Coulomb problem leads to the celebrated Balmer formula
2 for the energy levels, En = 1/(2n ) [4]. However, even in the 1920’s, it was known that the spectrum of hydrogen had a fine structure that was not captured by Balmer’s formula. This fine structure is due to relativistic effects and was explained by a solution [6] of the Dirac Coulomb problem which does not suffer any ambiguity with regard to boundary conditions, at least for the case of ordinary hydrogen [4] 1. Alternatively, these relativistic effects can be modeled as perturbations to the non-relativistic Coulomb Hamiltonian. Two of these perturbations are fairly straightforward and correspond respectively to the mass increase of moving particles in special relativity and to the spin-orbit coupling of the electron’s magnetic dipole to the Coulomb field when the electron is in motion. The third one is called the Darwin term and corresponds to a contact interaction between the
1Ambiguities do arise if one could create a hydrogen-like ion from an atom with atomic number Z ≥ 118. Needless to day, this remains a distant prospect.
2 electron and the proton that is generally modeled as a delta function potential [7]
Ze2~2 VDarwin(r) = 2 2 δ(r). (1.0.1) 80m c
The standard treatment of fine structure treats the effects of the Darwin term by first order perturbation theory, which gives results consistent with the Dirac equation and the observed fine structure of hydrogen. However, since the Darwin term is a contact interaction, the viewpoint advocated in this thesis is that it would be better to treat it as a modification of the boundary condition that should be applied at the origin. Delta function potentials are inconsistent in more than one dimension: although they produce effects in perturbation theory, if the problem is solved exactly, there are no scattering solutions at all, suggesting that there is no interaction with the delta function potential, a result sometimes called the Beg-Furlong-Huang theorem [8]. The problems associated with delta function potentials can be studied more transparently in the absence of an additional Coulomb potential and it was in this context that these problems were first identified. Pure delta function interactions were studied by Huang in the context of superfluids in the 1960s. Huang wished to consider bosons interaction via a short ranged contact interaction with no additional forces. Later the problem was studied by Beg and Furlong who were motivated by fundamental considerations of quantum field theory. Beg and Furlong showed that two non-relativistic particles that interact via a delta function potential are actually free in two or more dimensions and they argued that this observa- tion supported Wilson’s conjecture regarding the triviality of the φ4 field theory [8]. The consensus that has emerged from these studies is that it is possible to have non-trivial contact interactions in two and higher dimensions and that these interactions should be analyzed by the method of self-adjoint extensions as modified boundary conditions or by renormalization group methods. These studies also established that boundary condi-
3 tions can break symmetries that are present at the classical level in the model and that breaking of these symmetries is analogous to the anomaly phenomenon in quantum field theory. For example, the delta function in two dimensions imposes boundary conditions which anomalously break scale invariance [8]. In Chapter 2, a brief pedagogical review of previous work is provided. This thesis will focus on the study of three new realizations of the Coulomb problem in which potentially contact effects may play a more prominent role than in ordinary hydrogen. The first has to do electrons trapped on the surface of liquid helium by their own image charges [1]. This system is a realization of the one dimensional non- relativistic Coulomb problem. At large distances from the helium one expects the liquid can be treated as a continuous dielectric medium, but at short distances the physics are likely much more complicated. The method of self-adjoint extensions would attempt to condense these complicated short distance interactions into a single boundary contact parameter [9]. The second application covered is to Rydberg atoms [10, 11]. This is a realization of the three dimensional non-relativistic Coulomb problem. Rydberg atoms or molecules are systems where a single electron is highly excited and sees the remaining nuclei and electrons as a small singly charged core. In this case, the goal is to capture the complicated interactions with the core as a single boundary parameter [9]. Finally, the third application considered is the effect of charged impurities in the electronic states of graphene [12]. As graphene is an atomically flat two dimensional crystal, and the electrons are effectively relativistic particles in the system, the motion of the electrons is described by the two dimensional Dirac Coulomb problem. Again, in this case, the charged impurities are clusters of Cobalt or Calcium atom which are far from point- like and hence one might expect significant contact interactions [13]. From a purely theoretical point of view, an interesting find is two examples of anomalous symmetry breaking: the SO(4) Runge-Lenz symmetry for the three dimensional Coulomb problem and scale invariance for the two dimensional Dirac Coulomb problem.
4 In the one dimensional Coulomb problem, application of the self-adjoint extensions led to a modification of the Balmer formula,
1 E ∝ , (1.0.2) n (n − 2λ)2 where λ is the self-adjoint parameter, associated with the modified boundary conditions. Using this formula, the results of the liquid helium experiments performed by Grimes and Brown [14, 15] were successfully recreated. Furthermore, the results using the self-adjoint method are shown to be compatible with the generally accepted Cole [1] model for the system. Application of the method to the Rydberg atoms is only somewhat successful. As the extensions are only present in the s-wave channel, scope is much more confined than in the general Rydberg theory. However, the modified Balmer formula (1.0.2) is also derived in the three dimensional case, and is shown to be equivalent to the first order Rydberg Ritz formula. Unfortunately comparison of the two methods by fitting to known spectra of various alkali metals heavily favors the second order Rydberg-Ritz formula [10]. We conclude that a modified boundary condition does not capture the interaction of the excited electron with the core with accuracy comparable to the quantum defect model used to derive the Rydberg Ritz formula [10, 11]. Prior work on charged impurities in graphene did not take into account the ambiguity in the boundary condition that must be applied at the location of the impurity. By treating boundary conditions using the method of self-adjoint extensions we discover that the contact interaction anomalously breaks scale invariance and leads to a scattering phase shift with a non-trivial energy dependence in contrast to prior work which found a featureless energy independent phase shift that might be expected on the basis of scale invariance. The remainder of the thesis is organized as follows: Chapter 2 covers the mathematics
5 behind the self ajoint extensions and reviews past work on contact potentials through the use of self-adjoint extensions and the renormalization group. Chapter 3 provides a brief review of the experimental systems used in the thesis. In chapter 4, the non- relativistic Coulomb problem in one and three dimensions is discussed and compared to the experimental results of electrons bound on liquid helium and Rydberg atoms. Chapter 5 covers the two dimensional Dirac Coulomb problem and compares it to results in graphene systems. The Conclusions are presented in chapter 6. Technical details are relegated to the appendices. Appendix A covers the finer details of the hypergeometric functions; appendix B provides further details into the fitting algorithms used in the Rydberg section and appendix C describes the statistical method used for comparison of the Rydberg model to the self-adjoint model.
6 Chapter 2
Self-Adjoint Extensions
The underlying mathematical principles which this work was based off are a somewhat subtle generalization of concepts surrounding hermiticity. In the most reductive sense, the self-adjoint extensions that are employed are simply changing the boundary conditions in the system to ensure that the operator is still self-adjoint while simultaneously relaxing the requirements for the accepted wave functions. To fully exploit these extensions, they must have a strong set of definitions as well as a clear distinction from the weaker criterion for hermiticity. As the main distinction between a fully self-adjoint operator and one that is hermitian is fully dependent on the domain of the operators, it is best to first define what constitutes the domain for such an operator. An operator T defined on a Hilbert space H is a linear map from some linear subspace of H into H. The domain of T , denoted D(T ) is simply the subspace of H on which it acts [16]. From this definition of the domain, it is now possible to define the extension of an
operator. Consider two operators T1 and T2 defined on some Hilbert space H, T2 is an
extension of T1 if D(T2) ⊃ D(T1). To write this more specifically, T2 is said to extend T1
if T2φ = T1φ for all φ ∈ D(T1), and is written T2 ⊃ T1 [17]. The final concept to define before differentiating between hermitian and self-adjoint
7 operators is the adjoint of the operator. Let T be an operator on a Hilbert space H with an inner product (·, ·). Let D(T ∗) be the set of φ ∈ H for which there exists a corresponding η ∈ H with
(T ψ, φ) = (ψ, η) for all ψ ∈ D(T ). (2.0.1)
For each such φ ∈ D(T ∗), T ∗ is defined by T ∗φ = η, and T ∗ is called the adjoint of T [17]. For an operator T to be fully self-adjoint it must then meet two criteria. The first is that its adjoint acts in same fashion as the operator itself, namely that T φ = T ∗φ, with φ ∈ D(T ). This can also be written in terms of the inner product as
(T ψ, φ) = (ψ, T φ) for all ψ, φ ∈ D (T ) (2.0.2) which is the colloquial definition of a hermitian operator [17]. The second criterion for the self-adjoint definition is the distinction between simply hermitian and fully self-adjoint. For the operator to be self-adjoint, the domain of the operator and its adjoint must coincide, D(T ) = D(T ∗). If the domains are not equal, D(T ) ⊂ D(T ∗), the operator is then simply hermitian [17]. Since the only difference between hermitian and self-adjoint operators is the require- ment that domain of the adjoint must be equal to the domain of the operator itself, it is sometimes possible to find an operator which is both an extension of a hermitian operator and is itself fully self-adjoint. It is this idea that defines a self-adjoint extension. In most quantum systems, these extensions can arise due to the definition of the inner product
Z (ψ, φ) = dµ ψ∗φ, (2.0.3) where dµ is simply the measure associated with the space. When dealing with any
8 differential operator, showing that the operator is at least hermitian will usually result in a surface term after integration by parts. That is, for some differential operator T , the hermitian definition (2.0.2) has an extra term
(T ψ, φ) = (ψ, T φ) + surface terms.
These are normally forced to be zero through the imposition of boundary conditions on the surface. However these conditions may be over-restrictive on the domain of the operator itself, causing D(T ) ⊂ D(T ∗). The following two systems give concrete examples of the implementation and effects of the self-adjoint extensions to hermitian operators.
2.1 Particle in a box
To better illustrate these concepts, consider the one dimensional free particle re- stricted to the closed interval [0, 1]. The quantization of the momentum for the system is fairly straightforward; to begin, the differential equation given by the momentum op- erator (P = −id/dx),
d −i ψ(x) = pψ(x), (2.1.1) dx has the solution of ψ(x) = C exp(ipx). Imposing periodic boundary conditions of ψ(0) = ψ(1) gives rise to the quantization of the momentum. These force the momentum of the system to be p = 2πn, where n ∈ N. The boundary conditions determine whether the operator is self-adjoint or merely hermitian. As the inner product for the spaces is given by
Z 1 (φ, ψ) = dxφ∗ψ, (2.1.2) 0
9 it is easy to show using integration by parts that the momentum operator P gives rise to
∗ 1 (φ, P ψ) = (P φ, ψ) − i φ ψ|0 . (2.1.3)
If the boundary condition of ψ = 0 at the boundary points x = 0 and x = 1 is adopted, the surface term in (2.1.3) clearly vanishes and the operator P satisfies (φ, P ψ) = (P φ, ψ). The operator P is then clearly hermitian under the standard boundary conditions for a particle in the box. However, the operator is not self-adjoint. When the condition of ψ = 0 at the boundary is imposed, the surface term vanishes regardless of the behavior of φ at the boundary. Thus, D(P ) consists of the functions that vanish at the boundary whereas D(P ∗) consists of functions that do not have to respect any boundary conditions. The domains are not equal and P is not self-adjoint under the standard particle in a box boundary conditions. Indeed one can see that P has no eigenfunctions with these boundary conditions, far from the complete set of eigenfunctions that every self-adjoint operator should possess. Now consider the boundary condition of ψ(0) = λψ(1), where λ is some arbitrary com- plex number. Using this new condition, the requirement for surface term of φ∗(1)ψ(1) − φ∗(0)ψ(0) to vanish is now φ∗(0) = (1/λ)φ∗(1). For the domains of P and P ∗ to coincide, it must be that λ = (1/λ∗), which can be expressed in the simpler form of λ = eiα, where α is some real number. Thus the momentum operator P is self-adjoint on the unit interval if and only if the twisted boundary condition, ψ(0) = eiαψ(1), is imposed.
2.2 Delta Function
As a second example, consider a two dimensional system with a contact interaction at the origin. Here the Hamiltonian would be of the form
H = ∇2 + vδ(x).
10 Due to the rotation invariance the eigenfunctions may expressed as ψ = R(r) exp(i`θ) where R is the radial wavefunction and ` is an integer corresponding to the angular momentum of the system. Away from the origin of the system, the contact interaction can be effectively ignored. Expressing the Laplacian in polar coordinates
∂2 1 ∂ 1 ∂ ∇2 = + + ∂r2 r ∂r r2 ∂θ2
it is apparent that the radial wave function obeys Bessel’s equation and has solutions
which are linear combinations of the Bessel function, J|`|(kr), and Neumann function,
|`| Y|`|(kr). The small r behavior of the Neumann function is Y|`|(kr) ≈ 1/r for ` 6= 0 and
Y|`|(kr) ≈ ln r for ` = 0, while the Bessel function remains regular at the origin. The inner product for two wave functions in the same angular momentum channel is given by
Z ∞ (φ, ψ) = dr rS∗(r)R(r), (2.2.1) 0 where S and R are the radial wave functions corresponding to φ and ψ respectively. It can be quickly noted that that outside the ` = 0 channel the Neumann functions are not square integrable and thus unacceptable as quantum mechanical wavefunctions for the system. The Y0(kr), on the other hand, is allowable; despite diverging at the origin, the divergence is weak enough so that the function remains square integrable. Thus, in the ` = 0 channel, the boundary condition are not determined by square integrability alone, suggesting the existence of self-adjoint extensions. Continuing to employ the polar form of the Laplacian and integrating by parts, it is evident that
11 dS∗ dR∞ (φ, Hψ) = (Hφ, ψ) + r R − rS∗ . (2.2.2) dr dr 0
The Hamiltonian is formally self-adjoint, in the sense that it obeys (φ, Hψ) = (Hφ, ψ) up to surface terms. By locality, it is assumed that separate boundary conditions are applied at r = 0 and r = ∞. As these boundary conditions at infinity are generally determined by square integrability, the focus should remain at the origin. Motivated by the small r behavior of the Bessel and Neumann functions in the ` = 0 channel, the radial components can be expressed as
R = a0 + a1r + ... + b0 ln r + b1r ln r + ...,
S = A0 + A1r + ... + B0 ln r + B1r ln r + ..., (2.2.3)
near the origin. Substitution of these forms into the surface term seen in (2.2.2) yields
∗ dS ∗ dR ∗ ∗ lim r R − rS = a0B0 − b0A0, (2.2.4) r→0 dr dr
where the omitted terms vanish in the r → 0 limit. Imposing the boundary condition of
∗ a0 = λb0 on ψ, then the similar condition of A0 = λ B0 must be imposed on φ for the surface term of (2.2.4) to vanish. By restricting λ to the real numbers, the two boundary conditions will coincide and the domain of H will be the same as for its adjoint. To summarize, in the ` = 0 channel, the radial function R for a non-relativistic particle in two dimensions experiencing a contact or Dirac potential at the origin will have the leading asymptotic behavior of R = a0 + b0 ln r at small r and must obey the boundary condition of a0 = λb0, with λ ∈ R. This condition may be expressed more elegantly as
12 R(r0) R(r) lim R(r) − lim ln r = λ lim . (2.2.5) r→0 r0→0 ln r0 r→0 ln r
In the case of λ = 0, only the non-singular Bessel function will be used for the wavefunc- tion near the origin. This corresponds to the case of a free particle. The case of non-zero λ corresponds to the contact potential. Physically the interaction with the contact po- tential will result in scattering processes and the possibility of a bound state near the origin. These matters are more fully discussed in Jackiw [8].
2.3 Scale Invariance
Aside from rotational invariance the delta function potential also exhibits scale in- variance. Scale invariance powerfully constrains the scattering of the particle. Due to the contact nature of the potential the particle should only exhibit s-wave scattering and scale invariance would imply that the scattering phase shift should be independent of energy. However, the self-adjoint extension analysis shows that the phase shift exhibits a complicated energy dependence. Following the methods of Jackiw, analysis of the contact interaction problem using a renormalization group method shows that the scale invariance is anomalously broken and that a scale enters the problem via “dimensional transmutation” [8]. As will be discussed in Chapter 5, a similar analysis of the physics of a Dirac particle bound to a charged impurity, relevant to graphene, also exhibits an anomalous breaking of scale invariance. The scattering solution for the delta potential is given by