1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

2: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

3: %%%%%%%%                                                          %%%%%%%%

4: %%%%%%%%     VERSION: March 7, 2007                               %%%%%%%%

5: %%%%%%%%                                                          %%%%%%%%

6: %%%%%%%%     revtex/latex version                                 %%%%%%%%

7: %%%%%%%%                                                          %%%%%%%%

8: %%%%%%%%     Author: C. M. Bender                                 %%%%%%%%

9: %%%%%%%%                                                          %%%%%%%%

10: %%%%%%%%     I am submitting this manuscript for publication      %%%%%%%%

11: %%%%%%%%     in Reports on Progress in Physics:                   %%%%%%%%

12: %%%%%%%%          ROP/189078/REV/7688                             %%%%%%%%

13: %%%%%%%%                                                          %%%%%%%%

14: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

15: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

16: %%%%%%%%                                                          %%%%%%%%

17: %%%%%%%%    Contact Information:                                  %%%%%%%%

18: %%%%%%%%    ADDRESS:   Dr. Carl M. Bender                         %%%%%%%%

19: %%%%%%%%               Dept. of Phys.                             %%%%%%%%

20: %%%%%%%%               Campus Box 1105                            %%%%%%%%

21: %%%%%%%%               Washington University                      %%%%%%%%

22: %%%%%%%%               St. Louis, MO 63130                        %%%%%%%%

23: %%%%%%%%                                                          %%%%%%%%

24: %%%%%%%%    PHONE:     314-935-6216                               %%%%%%%%

25: %%%%%%%%    FAX:       314-935-6219                               %%%%%%%%

26: %%%%%%%%    E-MAIL:    cmb@wuphys.wustl.edu                       %%%%%%%%

27: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

28: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

29: 

30: \documentclass[12pt]{iopart}

31: \usepackage{iopams}

32: \usepackage{setstack}

33: % \usepackage{psfig}

34: 

35: \newcommand{\cC}{\ensuremath{\mathcal{C}}}

36: \newcommand{\cP}{\ensuremath{\mathcal{P}}}

37: \newcommand{\cT}{\ensuremath{\mathcal{T}}}

38: \newcommand{\half}{\mbox{$\textstyle \frac{1}{2}$}}

39: \newcommand{\x}{\ensuremath{\hat x}}

40: \newcommand{\y}{\ensuremath{\hat y}}

41: \newcommand{\z}{\ensuremath{\hat z}}

42: \newcommand{\p}{\ensuremath{\hat p}}

43: \newcommand{\q}{\ensuremath{\hat q}}

44: \newcommand{\rr}{\ensuremath{\hat r}}

45: \newcommand{\pslash}{\partial\raise.3ex\hbox{\kern-.5em /}}

46: \newcommand{\delslash}{\nabla\raise.3ex\hbox{\kern-.7em /}}

47: \newcommand{\ve}{\ensuremath{\varepsilon}}

48: \newcommand{\vf}{\ensuremath{\varphi}}

49: 

50: \begin{document}

51: \rightline{preprint LA-UR-07-1254}

52: 

53: \title[Making Sense of Non-Hermitian Hamiltonians]{Making Sense of Non-Hermitian

54: Hamiltonians}

55: 

56: \author[C M Bender] {Carl~M~Bender\footnote{Permanent address: Department of

57: Physics, Washington University, St. Louis MO 63130, USA}}

58: 

59: \address{Center for Nonlinear Studies, Los Alamos National Laboratory\\

60: Los Alamos, NM 87545, USA\\ \footnotesize\texttt{cmb@wustl.edu}}

61: 

62: \begin{abstract}

63: The Hamiltonian $H$ specifies the energy levels and time evolution of a quantum

64: theory. A standard axiom of quantum mechanics requires that $H$ be Hermitian

65: because Hermiticity guarantees that the energy spectrum is real and that time

66: evolution is unitary (probability-preserving). This paper describes an

67: alternative formulation of quantum mechanics in which the mathematical axiom of

68: Hermiticity (transpose + complex conjugate) is replaced by the physically

69: transparent condition of space-time reflection ($\cP\cT$) symmetry. If $H$ has

70: an unbroken $\cP\cT$ symmetry, then the spectrum is real. Examples of $\cP

71: \cT$-symmetric non-Hermitian quantum-mechanical Hamiltonians are $H=\p^2+i\x^3$

72: and $H=\p^2-\x^4$. Amazingly, the energy levels of these Hamiltonians are all

73: real and positive!

74: 

75: Does a $\cP\cT$-symmetric Hamiltonian $H$ specify a physical quantum theory in

76: which the norms of states are positive and time evolution is unitary? The answer

77: is that if $H$ has an unbroken $\cP\cT$ symmetry, then it has another symmetry

78: represented by a linear operator $\cC$. In terms of $\cC$, one can construct a

79: time-independent inner product with a positive-definite norm. Thus, $\cP

80: \cT$-symmetric Hamiltonians describe a new class of complex quantum theories

81: having positive probabilities and unitary time evolution.

82: 

83: The Lee Model provides an excellent example of a $\cP\cT$-symmetric Hamiltonian.

84: The renormalized Lee-model Hamiltonian has a negative-norm ``ghost'' state

85: because renormalization causes the Hamiltonian to become non-Hermitian. For the

86: past 50 years there have been many attempts to find a physical interpretation

87: for the ghost, but all such attempts failed. The correct interpretation of the

88: ghost is simply that the non-Hermitian Lee Model Hamiltonian is $\cP

89: \cT$-symmetric. The $\cC$ operator for the Lee Model is calculated exactly and

90: in closed form and the ghost is shown to be a physical state having a positive

91: norm. The ideas of $\cP\cT$ symmetry are illustrated by using many

92: quantum-mechanical and quantum-field-theoretic models.

93: 

94: (\today)

95: \end{abstract}

96: 

97: \submitto{\RPP}

98: 

99: \pacs{11.30.Er, 03.65.-w, 03.65.Bz}

100: %\maketitle

101: 

102: \section{Introduction -- New Kinds of Quantum Theories}

103: \label{s1}

104: 

105: The theory of quantum mechanics is nearly one hundred years old and because

106: there have been so many experimental verifications of its theoretical

107: predictions, it has become an accepted component of modern science. In an

108: introductory course on quantum physics, one learns the fundamental axioms that

109: define and characterize the theory. All but one of these axioms are physical

110: requirements. For example, the energy spectrum is required to be real because

111: all measurements of the energy of a system yield real results. Another axiom

112: requires that the energy spectrum be bounded below so that the system has a

113: stable lowest-energy state. Yet another axiom requires that the time evolution

114: of a quantum system be {\em unitary} (probability-conserving) because the

115: expected result of a probability measurement of a state cannot grow or decay in

116: time. A quantum theory of elementary particles must also satisfy the physical

117: axioms of Lorentz covariance and causality. However, there is one axiom that

118: stands out because it is mathematical rather than physical in character, and

119: this is the requirement that the Hamiltonian $H$, which is the operator that

120: expresses the dynamics of the quantum system, be Hermitian. 

121: 

122: The requirement that $H$ be Hermitian dates back to the early days of quantum

123: mechanics. The Hermiticity of $H$ is expressed by the equation

124: \begin{equation}

125: H=H^\dag,

126: \label{e1}

127: \end{equation}

128: where the Dirac Hermitian conjugation symbol $\dag$ represents the combined

129: operations of matrix transposition and complex conjugation. The mathematical

130: symmetry condition (\ref{e1}) is physically obscure but very convenient because

131: it implies that the eigenvalues of $H$ are real and that the time-evolution

132: operator $e^{-iHt}$ is unitary.

133: 

134: Hamiltonians that are non-Hermitian have traditionally been used to describe

135: dissipative processes, such as the phenomenon of radioactive decay. However,

136: these non-Hermitian Hamiltonians are only approximate, phenomenological

137: descriptions of physical processes. They cannot be regarded as fundamental

138: because they violate the requirement of unitarity. A non-Hermitian Hamiltonian

139: whose purpose is to describe a particle that undergoes radioactive decay

140: predicts that the probability of finding the particle gradually decreases in

141: time. Of course, a particle cannot just disappear because this would violate the

142: conservation of probability; rather, the particle transforms into other

143: particles. Thus, a non-Hermitian Hamiltonian that describes radioactive decay

144: can at best be a simplified, phenomenological, and {\em nonfundamental}

145: description of the decay process because it ignores the precise nature of the

146: decay products. In his book on quantum field theory Barton gives the standard

147: reasons for why a non-Hermitian Hamiltonian cannot provide a fundamental

148: description of nature \cite{BARTON}: ``A non-Hermitian Hamiltonian is

149: unacceptable partly because it may lead to complex energy eigenvalues, but

150: chiefly because it implies a non-unitary S matrix, which fails to conserve

151: probability and makes a hash of the physical interpretation.''

152: 

153: The purpose of this paper is to describe at an elementary level the

154: breakthroughs that have been made in the past decade which show that while the

155: symmetry condition (\ref{e1}) is sufficient to guarantee that the energy

156: spectrum is real and that time evolution is unitary, the condition of Dirac

157: Hermiticity is {\em not necessary}. It is possible to describe natural processes

158: by means of non-Hermitian Hamiltonians. We will show that the Hermiticity

159: requirement (\ref{e1}) may be replaced by the analogous but physically

160: transparent condition of space-time reflection symmetry ($\cP\cT$ symmetry)

161: \begin{equation}

162: H=H^{\cP\cT}

163: \label{e2}

164: \end{equation}

165: without violating any of the physical axioms of quantum mechanics. If $H$

166: satisfies (\ref{e2}), it is said to be $\cP\cT$ {\em symmetric}.

167: 

168: The notation used in this paper is as follows: The space-reflection operator, or

169: {\em parity} operator, is represented by the symbol $\cP$. The effect of $\cP$

170: on the quantum-mechanical coordinate operator $\x$ and the momentum operator

171: $\p$ is to change their signs:

172: \begin{equation}

173: \cP\x\cP=-\x\quad{\rm and}\quad\cP\p\cP=-\p.

174: \label{e3}

175: \end{equation}

176: Note that $\cP$ is a linear operator and that it leaves invariant the

177: fundamental commutation relation (the Heisenberg algebra) of quantum mechanics,

178: \begin{equation}

179: \x\p-\p\x=i\hbar\,{\bf 1},

180: \label{e4}

181: \end{equation}

182: where ${\bf 1}$ is the identity matrix. The time-reversal operator is

183: represented by the symbol $\cT$. This operator leaves $\x$ invariant but changes

184: the sign of $\p$:

185: \begin{equation}

186: \cT\x\cT=\x\quad{\rm and}\quad\cT\p\cT=-\p.

187: \label{e5}

188: \end{equation}

189: Like the parity operator $\cP$, the time-reversal operator $\cT$ leaves the

190: commutation relation (\ref{e4}) invariant, but this requires that $\cT$ reverse

191: the sign of the complex number $i$:

192: \begin{equation}

193: \cT i\cT=-i.

194: \label{e6}

195: \end{equation}

196: Equation (\ref{e6}) demonstrates that $\cT$ is not a linear operator; $\cT$ is

197: said to be {\em antilinear}. Also, since $\cP$ and $\cT$ are reflection

198: operators, their squares are the unit operator:

199: \begin{equation}

200: \cP^2=\cT^2={\bf 1}.

201: \label{e7}

202: \end{equation}

203: Finally, the $\cP$ and $\cT$ operators commute:

204: \begin{equation}

205: \cP\cT-\cT\cP=0.

206: \label{e8}

207: \end{equation}

208: In terms of the $\cP$ and $\cT$ operators, we define the $\cP\cT$-reflected

209: Hamiltonian $H^{\cP\cT}$ in (\ref{e2}) as $H^{\cP\cT}\equiv(\cP\cT)H(\cP\cT)$.

210: Thus, if a Hamiltonian is $\cP\cT$ symmetric [that is, if it satisfies

211: (\ref{e2})], then the $\cP\cT$ operator commutes with $H$:

212: \begin{equation}

213: H(\cP\cT)-(\cP\cT)H=0.

214: \label{e9}

215: \end{equation}

216: 

217: A $\cP\cT$-symmetric Hamiltonian need not be Hermitian; that is, it need not

218: satisfy the Hermiticity symmetry condition (\ref{e1}). Thus, it is possible to

219: have a fully consistent quantum theory whose dynamics is described by a

220: non-Hermitian Hamiltonian. Some examples of such non-Hermitian $\cP

221: \cT$-symmetric Hamiltonians are

222: \begin{equation}

223: H=\p^2+i\x^3,

224: \label{e10}

225: \end{equation}

226: and

227: \begin{equation}

228: H=\p^2-\x^4.

229: \label{e11}

230: \end{equation}

231: It is amazing indeed that the eigenvalues of these strange-looking Hamiltonians

232: are all real and positive and that these two Hamiltonians specify a unitary time

233: evolution even though they are non-Hermitian.

234: 

235: The Hamiltonians in (\ref{e10}) and (\ref{e11}) are special cases of the general

236: parametric family of $\cP\cT$-symmetric Hamiltonians 

237: \begin{equation}

238: H=\p^2+\x^2(i\x)^\epsilon,

239: \label{e12}

240: \end{equation}

241: where the parameter $\epsilon$ is real. These Hamiltonians are all $\cP\cT$

242: symmetric because they satisfy the condition in (\ref{e2}). It was shown in 1998

243: that when $\epsilon\geq0$ all of the eigenvalues of these Hamiltonians are

244: entirely real and positive, but when $\epsilon<0$ there are complex eigenvalues

245: \cite{r1}. We say that $\epsilon\geq0$ is the parametric region of {\em

246: unbroken} $\cP\cT$ symmetry and that $\epsilon<0$ is the parametric region of

247: {\em broken} $\cP\cT$ symmetry. (See Fig.~\ref{f1}.)

248: 

249: \begin{figure}[b!]

250: \vspace{3.2in}

251: \special{psfile=Fig1.eps angle=0 hoffset=72 voffset=-37 hscale=38 vscale=38}

252: \caption{Energy levels of the Hamiltonian $H=\p^2+\x^2(i\x)^\epsilon$ as a

253: function of the real parameter $\epsilon$. There are three regions: When

254: $\epsilon\geq0$, the spectrum is real and positive and the energy levels rise

255: with increasing $\epsilon$. The lower bound of this region, $\epsilon=0$,

256: corresponds to the harmonic oscillator, whose energy levels are $E_n=2n+1$. When

257: $-1<\epsilon<0$, there are a finite number of real positive eigenvalues and an

258: infinite number of complex conjugate pairs of eigenvalues. As $\epsilon$

259: decreases from $0$ to $-1$, the number of real eigenvalues decreases; when

260: $\epsilon\leq-0.57793$, the only real eigenvalue is the ground-state energy. As

261: $\epsilon$ approaches $-1^+$, the ground-state energy diverges. For $\epsilon

262: \leq-1$ there are no real eigenvalues. When $\epsilon\geq0$, the $\cP\cT$

263: symmetry is unbroken, but when $\epsilon<0$ the $\cP\cT$ symmetry is broken.}

264: \label{f1}

265: \end{figure}

266: 

267: One can think of the non-Hermitian Hamiltonians in (\ref{e12}) as complex

268: extensions of the harmonic oscillator Hamiltonian $H=\p^2+\x^2$. Indeed, the

269: quantum theories defined by $H$ are complex extensions of the conventional

270: quantum theory of the harmonic oscillator into the complex domain. The general

271: constructive principle that we are using in (\ref{e12}) is to start with a 

272: Hamiltonian that is {\em both} Hermitian and $\cP\cT$ symmetric. One then

273: introduces a real parameter $\epsilon$ in such a way that as $\epsilon$

274: increases from $0$ the Hamiltonian is no longer Hermitian but its $\cP\cT$

275: symmetry is maintained. One need not start with the harmonic oscillator. One

276: can, for example, begin with any of the Hermitian Hamiltonians $H=\p^2+\x^{2N}$,

277: where $N=1,\,2,\,3,\,\ldots$, and introduce the parameter $\epsilon$ as

278: follows: $H=\p^2+\x^{2N}(i\x)^\epsilon$. [The Hamiltonian in (\ref{e12}) is just

279: the special case $N=1$.] The properties of these Hamiltonians are discussed in

280: \cite{BBM}. 

281: 

282: We emphasize that these new kinds of Hamiltonians define valid and consistent

283: quantum theories in which the mathematical condition of Dirac Hermiticity $H=

284: H^\dagger$ has been replaced by the physical condition of $\cP\cT$ symmetry, $H=

285: H^{\cP\cT}$. The condition in (\ref{e2}) that the Hamiltonian is $\cP\cT$

286: symmetric is a physical condition because $\cP$ and $\cT$ are elements of the

287: homogeneous Lorentz group of spatial rotations and Lorentz boosts. The real

288: Lorentz group consists of four disconnected parts \cite{SW}: (i) The first part,

289: called the {\em proper orthochronous Lorentz group}, is a subgroup of the

290: Lorentz group whose elements are continuously connected to the identity. (ii)

291: The second part consists of all of the elements of the proper orthochronous

292: Lorentz group multiplied by the parity operator $\cP$. (iii) The third part

293: consists of all of the elements of the proper orthochronous Lorentz group

294: multiplied by the time-reversal operator $\cT$. (iv) The fourth part consists of

295: all of the elements of the proper orthochronous Lorentz group multiplied by $\cP

296: \cT$. Note that parts (ii) -- (iv) are not subgroups of the Lorentz group

297: because they do not contain the identity element. These four parts of the

298: Lorentz group are disconnected because there is no continuous path in group

299: space from one part to another.

300: 

301: When we say that Lorentz invariance is a physical requirement of a theory, what

302: we really mean is that the theory must be invariant under Lorentz

303: transformations belonging to the proper, orthochronous Lorentz group. We know

304: that the physical world is {\em not} invariant under the full homogeneous

305: Lorentz group because it has been demonstrated experimentally that there exist

306: weak processes that do not respect parity symmetry and other weak processes that

307: do not respect time-reversal symmetry.

308: 

309: One can extend the real Lorentz group to the {\em complex} Lorentz group

310: \cite{SW}. (To perform this extension it is necessary to make the crucial

311: assumption that the eigenvalues of the Hamiltonian are real and bounded below.)

312: The complex Lorentz group consists of {\em two} and not four disconnected parts.

313: In the complex Lorentz group there exists a continuous path in group space from

314: the elements of the real proper, orthochronous Lorentz group to the elements of

315: part (iv) of the real Lorentz group. There also exists a continuous path in

316: group space from the elements of part (ii) to the elements of part (iii) of the

317: real Lorentz group. Thus, while we know that the world is not invariant under

318: $\cP$ reflection or under $\cT$ reflection, we are proposing here to consider

319: the possibility suggested by complex group analysis that a fundamental discrete

320: symmetry of the world is $\cP\cT$ symmetry, or space-time reflection symmetry.

321: 

322: The most important consequence of the discovery that non-Hermitian $\cP

323: \cT$-symmetric Hamiltonians can define acceptable theories of quantum mechanics

324: is that we now can construct many new kinds of Hamiltonians that only a decade 

325: ago would have been rejected as being unphysical because they violate the

326: axiom of Hermiticity. In this paper we will examine some of these new

327: Hamiltonians and discuss the properties and possible physical implications of

328: the theories defined by these Hamiltonians. So far, there have been no

329: experiments that prove clearly and definitively that quantum systems defined by

330: non-Hermitian $\cP\cT$-symmetric Hamiltonians do exist in nature. However, one

331: should keep an open mind regarding the kinds of theories that one is willing to

332: consider. Indeed, Gell-Mann's ``totalitarian principle'' states that among all

333: possible physical theories ``Everything which is not forbidden is compulsory.''

334: 

335: \subsection{Presentation and Scope of this Paper}

336: \label{ss1-1}

337: 

338: Like many research areas in science, the study of non-Hermitian Hamiltonians

339: having real spectra began in a haphazard and diffuse fashion. There are numerous

340: early examples of isolated and disconnected discoveries of such non-Hermitian

341: Hamiltonians. For example, in 1980 Caliceti {\em et al.}, who were studying

342: Borel summation of divergent perturbation series arising from classes of

343: anharmonic oscillators, were astonished to find that the eigenvalues of an

344: oscillator having an imaginary cubic selfinteraction term are real \cite{H1}. In

345: the summer of 1993, when I was visiting CEN Saclay, I learned that Bessis and

346: Zinn-Justin had noticed on the basis of numerical work that some of the

347: eigenvalues of the cubic Hamiltonian in (\ref{e10}) seemed to be real, and they

348: wondered if the spectrum might be entirely real \cite{H2}. [Their interest in

349: the Hamiltonian (\ref{e10}) was piqued by the Lee-Yang edge singularity.] In

350: 1982 Andrianov, who was doing perturbative studies of $-x^4$ potentials, found

351: evidence that such theories might have real eigenvalues \cite{H3}. In 1992

352: Hollowood \cite{H4} and Scholtz {\em et al.} \cite{H5} discovered in their own

353: areas of research surprising examples of non-Hermitian Hamiltonians having real

354: spectra.

355: 

356: The field of $\cP\cT$-symmetric quantum mechanics was established in 1998 with

357: the discovery by Bender and Boettcher that the numerical conjectures of Bessis

358: and Zinn-Justin were not only valid, but were just one instance of a huge class

359: of non-Hermitian Schr\"odinger eigenvalue problems whose spectra are entirely

360: real and positive \cite{r1}. Bender and Boettcher showed that the reality of

361: the spectra was due to a symmetry principle, namely the condition of an unbroken

362: space-time reflection symmetry, and they argued that this symmetry principle

363: could replace the usual requirement of Dirac Hermiticity.

364: 

365: This discovery by Bender and Boettcher relies on two essential mathematical

366: ingredients. First, Bender and Boettcher used the techniques of analytic

367: continuation of eigenvalue problems. These fundamental techniques were developed

368: and used heavily in the early work of Bender and Wu on divergent perturbation

369: series \cite{BW1,BW2} and were later used by Bender and Turbiner \cite{r4}.

370: These techniques are crucial because they show how to analytically continue the

371: boundary conditions of an eigenvalue problem as a function of a parameter in the

372: Hamiltonian. Second, Bender and Boettcher used the delta-expansion techniques

373: that had been discovered and developed by Bender {\em et al}. \cite{delta} as a

374: way to avoid divergent perturbation series. The delta expansion is a powerful

375: perturbation-theory technique in which the small perturbation parameter is a

376: measure of the nonlinearity of the problem. As a result, many crucial properties

377: of the problem are exactly preserved as this parameter varies. In the case of

378: $\cP\cT$-symmetric quantum mechanics, it is the reality of the eigenvalues that

379: is exactly maintained as the perturbation parameter $\epsilon$ in (\ref{e12}) is

380: varied.

381: 

382: Many researchers have contributed immensely to the development of $\cP

383: \cT$-symmetric quantum mechanics by discovering new examples and models, proving

384: theorems, and performing numerical and asymptotic analysis. In the past few

385: years a large and active research community has developed and there have been

386: half a dozen international conferences on the subject of $\cP\cT$ symmetry,

387: pseudo-Hermiticity, and non-Hermitian Hamiltonians. The proceedings of these

388: conferences provide a complete source of references \cite{C1,C2,C3,C4,C5,C6}.

389: By now, there have been so many contributions to the field that it is impossible

390: to describe them all in this one paper.

391: 

392: The purpose of this paper is to give an elementary introduction to this exciting

393: and active field of research. In writing this paper, my hope is that the rate of

394: new discoveries and the development of the field will continue at such a rapid

395: pace that this review will soon become obsolete.

396: 

397: \subsection{Organization of this Paper}

398: \label{ss1-2}

399: 

400: In Sec.~\ref{s2} we show that $\cP\cT$-symmetric Hamiltonians are complex

401: extensions of Hermitian Hamiltonians and we discuss the key property of $\cP

402: \cT$-symmetric Hamiltonians, namely, that their energy eigenvalues are real and

403: bounded below. We discuss techniques for calculating eigenvalues. In

404: Sec.~\ref{s3} we discuss classical $\cP\cT$-symmetric Hamiltonians and show how

405: to continue ordinary classical mechanics into the complex domain. The crucial

406: theoretical questions of conservation of probability and unitary time evolution

407: are addressed in Sec.~\ref{s4}. In Sec.~\ref{s5} we illustrate the theory of

408: $\cP\cT$-symmetric quantum mechanics by using a simple $2\times2$ matrix

409: Hamiltonian. Then in Sec.~\ref{s6} we explain how to calculate the $\cC$

410: operator, which is the central pillar of $\cP\cT$ symmetry and which is needed

411: to construct the Hilbert space for the theory. In the next two sections we

412: discuss the many applications of $\cP\cT$-symmetric quantum theory. We discuss

413: quantum-mechanical applications in Sec.~\ref{s7} and quantum-field-theoretic

414: applications in Sec.~\ref{s8}. Finally, in Sec.~\ref{s9} we make some brief

415: concluding remarks.

416: 

417: \section{Determining the Eigenvalues of a $\cP\cT$-Symmetric Hamiltonian}

418: \label{s2}

419: 

420: In this section we show how to calculate the eigenvalues of a $\cP\cT$-symmetric

421: Hamiltonian. We begin by pointing out that the Hamiltonian operator defines and

422: determines the physical properties of a quantum theory in three important ways:

423: 

424: \begin{itemize}

425: 

426: \item[(i)] {\em The Hamiltonian determines the energy levels of the quantum

427: theory}. To find these energy levels one must solve the time-independent

428: Schr\"odinger eigenvalue problem

429: \begin{equation}

430: H\psi=E\psi.

431: \label{e13}

432: \end{equation}

433: This equation usually takes the form of a differential equation that must be

434: solved subject to boundary conditions on the eigenfunction $\psi$. In the case 

435: of a $\cP\cT$-symmetric Hamiltonian it is crucial that the boundary conditions

436: be imposed properly. We emphasize that for a quantum theory to be physically

437: acceptable the energy eigenvalues of the Hamiltonian must be real and bounded

438: below.

439: 

440: \item[(ii)] {\em The Hamiltonian specifies the time evolution of the states

441: and operators of the quantum theory}. To determine the time evolution of a

442: state $\psi(t)$ in the Schr\"odinger picture we must solve the time-dependent

443: Schr\"odinger equation

444: \begin{equation}

445: i\frac{\partial}{\partial t}\psi(t)=H\psi(t).

446: \label{e14}

447: \end{equation}

448: The solution to this first-order differential equation is straightforward

449: because $H$ is assumed to be independent of time:

450: \begin{equation}

451: \psi(t)=e^{-iHt}\psi(0).

452: \label{e15}

453: \end{equation}

454: We call $e^{-iHt}$ the {\em time-evolution operator}. In conventional quantum

455: mechanics the time-evolution operator is unitary because the Hamiltonian $H$ is

456: Hermitian. As a result, the norm of the state $\psi(t)$ remains constant in

457: time. The constancy of the norm is an essential feature of a quantum system

458: because the norm of a state is a probability, and this probability must remain

459: constant in time. If this probability were to grow or decay in time, we would

460: say that the theory violates unitarity. In $\cP\cT$-symmetric quantum mechanics

461: $H$ is not Dirac Hermitian, but the norms of states are still time independent.

462: 

463: \item[(iii)] {\em The Hamiltonian incorporates the symmetries of the theory}. As

464: an example, suppose that the Hamiltonian $H$ commutes with the parity operator

465: $\cP$. We then say that the Hamiltonian is {\em parity invariant}. Since $\cP$

466: is a linear operator, we know that the eigenstates of the Hamiltonian [the

467: solutions to (\ref{e13})] will also be eigenstates of $\cP$. Thus, the

468: eigenstates of $H$ will have a definite parity; they will be either even or odd

469: under space reflection. (Of course, a general state, which is a linear

470: combination of the eigenstates of $H$, need not have definite parity.)

471: 

472: \end{itemize}

473: 

474: \subsection{Broken and Unbroken $\cP\cT$ Symmetry}

475: \label{ss2-1}

476: 

477: For the case of a $\cP\cT$-symmetric Hamiltonian, the $\cP\cT$ operator commutes

478: with the Hamiltonian $H$ [see (\ref{e9})]. However, $\cP\cT$ symmetry is more

479: subtle than parity symmetry because the $\cP\cT$ operator is not linear. Because

480: $\cP\cT$ is not linear, the eigenstates of $H$ may or may not be eigenstates of

481: $\cP\cT$.

482: 

483: Let us see what may go wrong if we assume that an eigenstate $\psi$ of the

484: Hamiltonian $H$ is also an eigenstate of the $\cP\cT$ operator. Call the

485: eigenvalue $\lambda$ and express the eigenvalue condition as

486: \begin{equation}

487: \cP\cT\psi=\lambda\psi.

488: \label{e16}

489: \end{equation}

490: Multiply (\ref{e16}) by $\cP\cT$ on the left and use the property that $(\cP\cT

491: )^2={\bf 1}$ [see (\ref{e7}) and (\ref{e8})]:

492: \begin{equation}

493: \psi=(\cP\cT)\lambda(\cP\cT)^2\psi.

494: \label{e17}

495: \end{equation}

496: Since $\cT$ is antilinear [see (\ref{e6})], we get

497: \begin{equation}

498: \psi=\lambda^*\lambda\psi=|\lambda|^2\psi.

499: \label{e18}

500: \end{equation}

501: Thus, $|\lambda|^2=1$ and the eigenvalue $\lambda$ of the $\cP\cT$ operator is a

502: pure phase:

503: \begin{equation}

504: \lambda=e^{i\alpha}.

505: \label{e19}

506: \end{equation}

507: 

508: Next, multiply the eigenvalue equation (\ref{e13}) by $\cP\cT$ on the left and

509: again use the property that $(\cP\cT)^2={\bf 1}$:

510: \begin{equation}

511: (\cP\cT)H\psi=(\cP\cT)E(\cP\cT)^2\psi.

512: \label{e20}

513: \end{equation}

514: Using the eigenvalue equation (\ref{e16}) and recalling that $\cP\cT$ commutes

515: with $H$, we get

516: \begin{equation}

517: H\lambda\psi=(\cP\cT)E(\cP\cT)\lambda\psi.

518: \label{e21}

519: \end{equation}

520: Finally, we again use the property that $\cT$ is antilinear to obtain

521: \begin{equation}

522: E\lambda\psi=E^*\lambda\psi.

523: \label{e22}

524: \end{equation}

525: Since $\lambda$ is nonzero [see (\ref{e19})], we conclude that the eigenvalue

526: $E$ is real: $E=E^*$.

527: 

528: In general, this conclusion is false, as Fig.~\ref{f1} clearly demonstrates.

529: When $\epsilon<0$, some of the eigenvalues have disappeared because they are

530: complex. On the other hand, for the restricted region $\epsilon\geq0$ this

531: conclusion is correct; all of the eigenvalues are indeed real. We are then led

532: to make the following definition: If every eigenfunction of a $\cP\cT$-symmetric

533: Hamiltonian is also an eigenfunction of the $\cP\cT$ operator, we say that the

534: $\cP\cT$ symmetry of $H$ is {\em unbroken}. Conversely, if some of the

535: eigenfunctions of a $\cP\cT$-symmetric Hamiltonian are not simultaneously

536: eigenfunctions of the $\cP\cT$ operator, we say that the $\cP\cT$ symmetry of

537: $H$ is {\em broken}.

538: 

539: The correct way to interpret (\ref{e22}) is that if a Hamiltonian has an

540: unbroken $\cP\cT$ symmetry, then all of its eigenvalues are real. Thus, to

541: establish that the eigenvalues of a particular $\cP\cT$-symmetric Hamiltonian

542: are real, it is necessary to prove that the $\cP\cT$ symmetry of $H$ is

543: unbroken. This is difficult to show, and it took several years after the

544: discovery of the family of $\cP\cT$-symmetric Hamiltonians in (\ref{e12}) before

545: a complete and rigorous proof was finally constructed by Dorey {\em et al.} in

546: 2001 \cite{D1,D2}.\footnote{The proof by Dorey {\em et al.} draws from many

547: areas of theoretical and mathematical physics and uses spectral determinants,

548: the Bethe {\em ansatz}, the Baxter $T$-$Q$ relation, the monodromy group, and an

549: array of techniques used in conformal quantum field theory. This proof is too

550: technical to be described in this paper, but it was a significant advance

551: because it establishes a correspondence between ordinary differential equations

552: and integrable models. This correspondence is known as the {\em ODE-IM

553: correspondence} \cite{D3,D4,D5}.} Many others have contributed to the rigorous

554: mathematical development of the theory of $\cP\cT$ symmetry. These include Shin

555: \cite{SHIN}, Pham \cite{PP1}, Delabaere \cite{DD2}, Trinh \cite{TRINH}, Weigert

556: \cite{WWWW,WWWW1}, and Scholtz and Geyer \cite{GEY}. Mostafazadeh generalized

557: $\cP\cT$ symmetry to pseudo-Hermiticity (see Subsec.~\ref{ss4-5}).

558: 

559: \subsection{Boundary Conditions for the Schr\"odinger Eigenvalue Problem}

560: \label{ss2-2}

561: 

562: Our objective in this section is to show how to calculate the eigenvalues of the

563: Schr\"odinger eigenvalue problem (\ref{e13}). The most direct approach is to

564: write (\ref{e13}) as a differential equation in coordinate space. The principal

565: conceptual difficulty that we face in solving this differential equation is in

566: identifying and understanding the boundary conditions on the coordinate-space 

567: eigenfunctions.

568: 

569: To write (\ref{e13}) in coordinate space, we make the standard transcriptions

570: \begin{equation}

571: \x\to x\quad{\rm and}\quad\p\to-i\frac{d}{dx},

572: \label{e23}

573: \end{equation}

574: except that we treat the variable $x$ as complex. The Schr\"odinger eigenvalue

575: problem (\ref{e13}) then takes the form

576: \begin{equation}

577: -\psi''(x)+x^2(ix)^\epsilon\psi(x)=E\psi(x).

578: \label{e24}

579: \end{equation}

580: 

581: Although we cannot solve this equation exactly for arbitrary $\epsilon$, we can

582: easily find the possible asymptotic behaviors of its solutions by using the WKB

583: approximation. In general, for any differential equation of the form $-y''(x)+

584: V(x)y(x)=0$, where $V(x)$ is a function that grows as $|x|\to\infty$, we know 

585: that the exponential component of the asymptotic behavior of $y(x)$ for large

586: $|x|$ has the form

587: \begin{equation}

588: y(x)\sim\exp\left[\pm\int^x ds\sqrt{V(s)}\right].

589: \label{e25}

590: \end{equation}

591: 

592: To identify the appropriate boundary conditions to impose on $\psi(x)$, we

593: consider first the Hermitian case $\epsilon=0$ (the harmonic oscillator). From

594: (\ref{e25}) we can see immediately that the possible asymptotic behaviors of

595: solutions are $\psi(x)\sim\exp\left(\pm\half x^2\right)$. The usual requirement

596: that the eigenfunction be square-integrable implies that we must choose the

597: negative sign in the exponent, and therefore the eigenfunctions are

598: Gaussian-like for large $|x|$. This result extends into the complex-$x$ plane:

599: If the eigenfunctions vanish exponentially on the real-$x$ axis for large $|x|$,

600: they must also vanish in two wedges of opening angle $\half\pi$ in the complex

601: plane centered about the positive-real and negative-real axes. These wedges are

602: called {\em Stokes wedges} \cite{BO}.

603: 

604: What happens as $\epsilon$ increases from $0$? As soon as $\epsilon>0$

605: ($\epsilon$ noninteger), a logarithmic branch point appears at the origin $x=0$.

606: Without loss of generality, we may choose the branch cut to run up the imaginary

607: axis from $x=0$ to $x=i\infty$. In this cut plane the solutions to the

608: differential equation (\ref{e24}) are single-valued. From the asymptotic

609: behavior of $\psi(x)$ in (\ref{e25}), we deduce that the Stokes wedges rotate

610: downward into the complex-$x$ plane and that the opening angles of the wedges

611: decrease as $\epsilon$ increases.

612: 

613: There are many wedges in which $\psi(x)\to0$ as $|x|\to\infty$. Thus, there are

614: many eigenvalue problems associated with the differential equation (\ref{e24}).

615: We choose to continue the eigenvalue differential equation (\ref{e24}) smoothly

616: away from the location of the harmonic oscillator wedges at $\epsilon=0$. (A

617: detailed description of how to extend eigenvalue equations into the complex

618: plane may be found in Ref.~\cite{r4}.) For $\epsilon>0$ the center lines of the

619: left and right wedges lie at the angles

620: \begin{equation}

621: \theta_{\rm left}=-\pi+\frac{\epsilon}{\epsilon+4}~\frac{\pi}{2}\quad{\rm and}

622: \quad \theta_{\rm right}=-\frac{\epsilon}{\epsilon+4}~\frac{\pi}{2}.

623: \label{e26}

624: \end{equation}

625: The opening angle of each of these wedges is $\Delta=2\pi/(\epsilon+4)$. The

626: differential equation (\ref{e24}) may be integrated along any path in the

627: complex-$x$ plane as long as the ends of the path approach complex infinity

628: inside the left wedge and the right wedge. Note that these wedges contain the

629: real-$x$ axis when $-1<\epsilon<2$. However, as soon as $\epsilon$ is larger

630: than 2, the wedges rotate below the real-$x$ axis. These wedges are shown in

631: Fig.~\ref{f2}.

632: 

633: \begin{figure}[t!]

634: \vspace{2.3in}

635: \special{psfile=Fig2.eps angle=0 hoffset=97 voffset=-31 hscale=30 vscale=30}

636: \caption{Stokes wedges in the complex-$x$ plane containing the contour on which

637: the eigenvalue problem for the differential equation (\ref{e24}) for $\epsilon=

638: 2.2$ is posed. In these wedges $\psi(x)$ vanishes exponentially as $|x|\to

639: \infty$. The eigenfunction $\psi(x)$ vanishes most rapidly at the centers of

640: the wedges.}

641: \label{f2}

642: \end{figure}

643: 

644: Notice that the wedges in Fig.~\ref{f2} are mirror images of one another if they

645: are reflected through the imaginary-$x$ axis. This left-right symmetry is the

646: coordinate-space realization of $\cP\cT$ symmetry. If we choose any point $x$ in

647: the complex-$x$ plane and perform a parity reflection, then $x\to-x$. Time

648: reversal replaces $i$ by $-i$ as we saw in (\ref{e6}), and so $\cT$ replace $-x$

649: by its complex conjugate $-x^*$. Thus, in the coordinate representation $\cP\cT$

650: symmetry is left-right symmetry.

651: 

652: \subsection{The Flaw in Dyson's Argument}

653: \label{ss2-3}

654: 

655: The quantum theories that we are considering in this paper are obtained by

656: extending real quantum mechanics into the complex domain, as explained in

657: Subsec.~\ref{ss2-2}. The notion of analytically continuing a Hamiltonian into

658: the complex plane was first discussed in 1952 by Dyson, who argued heuristically

659: that perturbation theory for quantum electrodynamics diverges \cite{DY}. Dyson's

660: argument consists of rotating the electric charge $e$ into the complex-$e$

661: plane: $e\to ie$. Applied to the standard quantum anharmonic-oscillator

662: Hamiltonian $H=\p^2+g\x^4$, Dyson's argument goes as follows: Rotate the

663: parameter $g$ anticlockwise in the complex-$g$ plane from positive $g$ to $-g$.

664: Now, the potential term in the Hamiltonian is {\em no longer bounded below}, so

665: the resulting theory has no ground state. Hence, the ground-state energy $E_0(

666: g)$ has an abrupt transition at $g=0$, which implies that $E_0(g)$ must have a

667: singularity at $g=0$.

668: 

669: Following Dyson's reasoning, one would think that the spectrum of the

670: Hamiltonian (\ref{e11}), which is obtained by setting $\epsilon=2$ in

671: (\ref{e12}), would not be bounded below, and one might conclude that this

672: Hamiltonian is mathematically and physically unacceptable. However, this

673: heuristic argument is flawed. While the ground-state energy of the quantum

674: anharmonic oscillator does indeed have a singularity at $g=0$, the spectrum of

675: the Hamiltonian (\ref{e11}) that is obtained by analytically continuing a

676: parameter in the Hamiltonian remains ambiguous until the boundary conditions 

677: satisfied by the eigenfunctions are specified. The term ``bounded below'' is

678: inappropriate if it relies on an ordering relation applied to a complex

679: potential because ordering relations cannot be used for complex numbers. The

680: concern that the spectrum of $H$ in (\ref{e11}) is not bounded below is

681: unfounded because $\cP\cT$-symmetric boundary conditions on the Schr\"odinger

682: equation (\ref{e24}) prohibit the occurrence of negative eigenvalues.

683: 

684: The eigenvalues of $H$ in (\ref{e11}) depend crucially on the history of how

685: this negative-coupling-constant Hamiltonian constant is obtained. There are two

686: different ways to obtain $H$ in (\ref{e11}): First, one can substitute $g=|g|

687: e^{i\theta}$ into $H=\p^2+g\x^4$ and rotate from $\theta=0$ to $\theta=\pi$.

688: Under this rotation, the ground-state energy $E_0(g)$ becomes complex. Clearly,

689: $E_0(g)$ is real and positive when $g>0$ and complex when $g<

690: 0$.\footnote{Rotating from $\theta=0$ to $\theta =-\pi$, we obtain the same

691: Hamiltonian as in (\ref{e11}), but the spectrum is the complex conjugate of the

692: spectrum obtained when we rotate from $\theta=0$ to $\theta= \pi$.} Second, one

693: can obtain (\ref{e11}) as a limit of the Hamiltonian $H$ in (\ref{e12}) as

694: $\epsilon:\,0\to2$. The spectrum of this complex Hamiltonian is real, positive,

695: and discrete, as is shown in Fig.~\ref{f1}.

696: 

697: How can the Hamiltonian (\ref{e11}) possess two such astonishingly different

698: spectra? The answer lies in the boundary conditions satisfied by the

699: eigenfunctions $\psi(x)$. In the first case, in which $\theta={\rm arg}\,g$ is

700: rotated in the complex-$g$ plane from $0$ to $\pi$, $\psi(x)$ vanishes in the

701: complex-$x$ plane as $|x|\to\infty$ inside the wedges $-\pi/3<{\rm arg}\,x<0$

702: and $-4\pi/3<{\rm arg}\,x<-\pi$. {\em These wedges are not $\cP\cT$-symmetric

703: reflections of one another}. In the second case, in which the exponent

704: $\epsilon$ ranges from $0$ to $2$, $\psi(x)$ vanishes in the complex-$x$ plane

705: as $|x|\to\infty$ inside the $\cP\cT$-symmetric pair of wedges $-\pi/3<{\rm arg}

706: \,x<0$ and $-\pi<{\rm arg}\,x<-2\pi/3$. We emphasize that in this second case

707: the boundary conditions hold in wedges that are symmetric with respect to the

708: imaginary axis; these boundary conditions enforce the $\cP\cT$ symmetry of $H$

709: and are responsible for the reality of the energy spectrum.

710: 

711: \medskip

712: 

713: \begin{itemize}

714: \item[~~]

715: \begin{footnotesize}

716: \noindent{\em Illustrative example:} The harmonic oscillator Hamiltonian

717: \begin{equation}

718: H=\p^2+\omega^2\x^2\quad(\omega>0)

719: \label{e27}

720: \end{equation}

721: is an elementary model that illustrates the dependence of the eigenvalues on the

722: boundary conditions imposed on the eigenfunctions. Let us assume that the

723: eigenfunctions of $H$ vanish on the real-$x$ axis as $x\to\pm\infty$. The

724: eigenfunctions have the form of a Gaussian $\exp\left(-\half\omega x^2\right)$

725: multiplied by a Hermite polynomial. Thus, the Stokes wedges are centered about

726: the positive real-$x$ axis and have angular opening $\half\pi$. The eigenvalues

727: $E_n$ are given exactly by the formula

728: \begin{equation}

729: E_n=\left(n+\half\right)\omega\qquad(n=0,\,1,\,2,\,3,\,\ldots).

730: \label{e28}

731: \end{equation}

732: Now suppose that the parameter $\omega$ is rotated by $180^\circ$ into the

733: complex-$\omega$ plane from the positive axis to the negative axis so that

734: $\omega$ is replaced by $-\omega$. This causes the Stokes wedges in the

735: complex-$x$ plane to rotate by $90^\circ$ so that the eigenfunctions now vanish

736: exponentially on the imaginary-$x$ axis rather than on the real-$x$ axis. Also,

737: as a consequence of this rotation, the {\em eigenvalues change sign}:

738: \begin{equation}

739: E_n=-\left(n+\half\right)\omega\qquad(n=0,\,1,\,2,\,3,\,\ldots).

740: \label{e29}

741: \end{equation}

742: Notice that under the rotation that replaces $\omega$ by $-\omega$ the

743: Hamiltonian remains invariant, and yet the signs of the eigenvalues are

744: reversed! This shows that the eigenspectrum depends crucially on the boundary

745: conditions that are imposed on the eigenfunctions.

746: \end{footnotesize}

747: \end{itemize}

748: 

749: \medskip

750: Apart from the eigenvalues, there is yet another striking difference between the

751: two theories corresponding to $H$ in (\ref{e11}). The expectation value of the

752: operator $\x$ in the ground-state eigenfunction $\psi_0(x)$ is given by

753: \begin{equation}

754: \frac{\langle0|x|0\rangle}{\langle0|0\rangle}\equiv\frac{\int_C dx\,x

755: \psi_0^2(x)}{\int_C dx\,\psi_0^2(x)},

756: \label{e30}

757: \end{equation}

758: where $C$ is a complex contour that lies in the asymptotic wedges described

759: above. The value of $\langle0|x|0\rangle/\langle0|0\rangle$ for $H$ in

760: (\ref{e11}) depends on the limiting process by which we obtain $H$. If we

761: substitute $g=g_0e^{i\theta}$ into the Hamiltonian $H=\p^2+g\x^4$ and rotate

762: $g$ from $\theta=0$ to $\theta=\pi$, we find by an elementary symmetry argument

763: that this expectation value vanishes for all $g$ on the semicircle in the

764: complex-$g$ plane. The expectation value vanishes because this rotation in the

765: complex-$g$ plane preserves parity symmetry ($x\to-x$). However, if we define

766: $H$ in (\ref{e11}) by using the Hamiltonian in (\ref{e12}) and by allowing

767: $\epsilon$ to range from $0$ to $2$, we find that this expectation value is

768: nonzero. In fact, this expectation value is nonvanishing for all $\epsilon>0$.

769: On this alternate path $\cP\cT$ symmetry (reflection about the imaginary axis,

770: $x\to-x^*$) is preserved, but parity symmetry is permanently broken. (We suggest

771: in Subsec.~\ref{ss8-4} that, as a consequence of broken parity symmetry, one

772: might be able to describe the dynamics of the Higgs sector by using a $\cP

773: \cT$-symmetric $-g\vf^4$ quantum field theory.)

774: 

775: \subsection{Using WKB Phase-Integral Techniques to Calculate Eigenvalues}

776: \label{ss2-4}

777: 

778: Now that we have identified the boundary conditions to be imposed on the

779: eigenfunctions of the $\cP\cT$-symmetric Hamiltonian (\ref{e12}), we can use a

780: variety of techniques to calculate the eigenvalues of this Hamiltonian. Not

781: surprisingly, it is impossible to solve the differential-equation eigenvalue

782: problem (\ref{e24}) analytically and in closed form except in two special cases,

783: namely, for $\epsilon=0$ (the harmonic oscillator) and for $\epsilon\to\infty$

784: (the $\cP\cT$-symmetric version of the square-well potential, whose solution is

785: given in Ref.~\cite{SQ}). Thus, it is necessary to use approximate analytic or

786: numerical methods.

787: 

788: The simplest analytic approach uses WKB theory, which gives an excellent

789: approximation to the eigenvalues when $\epsilon>0$. The WKB calculation is

790: interesting because it must be performed in the complex plane rather than on the

791: real-$x$ axis. The turning points $x_{\pm}$ are those roots of $E=x^2(ix

792: )^\epsilon$ that {\em analytically continue} off the real axis as $\epsilon$

793: increases from $0$. These turning points,

794: \begin{equation}

795: x_-=E^\frac{1}{\epsilon+2}e^{i\pi\left(\frac{3}{2}-\frac{1}{\epsilon+2}\right)},

796: \quad x_+=E^\frac{1}{\epsilon+2}e^{-i\pi\left(\frac{1}{2}-\frac{1}{\epsilon+2}

797: \right)},

798: \label{e31}

799: \end{equation}

800: lie in the lower-half (upper-half) $x$ plane in Fig.~\ref{f2} when $\epsilon>0$

801: ($\epsilon<0$).

802: 

803: The leading-order WKB phase-integral quantization condition is given by

804: \begin{equation}

805: (n+1/2)\pi=\int_{x_-}^{x_+}dx\,\sqrt{E-x^2(ix)^\epsilon}.

806: \label{e32}

807: \end{equation}

808: When $\epsilon>0$ this path lies entirely in the lower-half $x$ plane, and when

809: $\epsilon=0$ (the case of the harmonic oscillator) the path lies on the real

810: axis. However, when $\epsilon<0$ the path lies in the upper-half $x$ plane and

811: crosses the cut on the positive imaginary-$x$ axis. In this case there is no

812: {\em continuous path joining the turning points.} Hence, WKB fails when

813: $\epsilon<0$.

814: 

815: When $\epsilon\geq0$, we deform the phase-integral contour so that it follows

816: the rays from $x_-$ to $0$ and from $0$ to $x_+$:

817: \begin{equation}

818: \left(n+\half\right)\pi=2\sin\left(\frac{\pi}{\epsilon+2}\right)

819: E^{\frac{1}{\epsilon+2}+\half}\int_0^1 ds\,\sqrt{1-s^{\epsilon+2}}.

820: \label{e33}

821: \end{equation}

822: We then solve for $E_n$:

823: \begin{equation}

824: E_n\sim\left[\frac{\Gamma\left(\frac{3}{2}+\frac{1}{\epsilon+2}\right)\sqrt{\pi}

825: \left(n+\half\right)}{\sin\left(\frac{\pi}{\epsilon+2}\right)\Gamma\left(1+\frac

826: {1}{\epsilon+2}\right)}\right]^\frac{2\epsilon+4}{\epsilon+4}\quad(n\to\infty).

827: \label{e34}

828: \end{equation}

829: This formula gives a very accurate approximation to the eigenvalues plotted in

830: Fig.~\ref{f1} and it shows, at least in the WKB approximation, that the energy

831: eigenvalues of $H$ in (\ref{e12}) are real and positive (see Table \ref{t1}). We

832: can, in addition, perform a higher-order WKB calculation by replacing the phase

833: integral by a {\em closed contour} that encircles the path joining the turning

834: points (see Refs.~\cite{BBM} and \cite{BO}).

835: 

836: \begin{table}

837: \caption[t1]{Comparison of the exact eigenvalues (obtained with Runge-Kutta)

838: and the WKB result in (\ref{e34}).}

839: \begin{tabular}{llrrllrr}

840: $\epsilon$ & $n$ & $E_{\rm exact}$ & $E_{\rm WKB}$ & $\epsilon$ & $n$ &

841: $E_{\rm exact}$ & $E_{\rm WKB}$ \\ \hline

842: 1 & 0 & $1.156\,267\,072$ & 1.0943& 2 & 0 & $1.477\,149\,753$  &  1.3765 \\

843: &1 & $4.109\,228\,752$ & 4.0895&      & 1 & $6.003\,386\,082$  &  5.9558 \\

844: &2 & $7.562\,273\,854$ & 7.5489&      & 2 & $11.802\,433\,593$ & 11.7690 \\

845: &3 & $11.314\,421\,818$ & 11.3043&    & 3 & $18.458\,818\,694$ & 18.4321 \\

846: &4 & $15.291\,553\,748$ & 15.2832&    & 4 & $25.791\,792\,423$ & 25.7692 \\

847: &5 & $19.451\,529\,125$ & 19.4444&    & 5 & $33.694\,279\,298$ & 33.6746 \\

848: &6 & $23.766\,740\,439$ & 23.7606&    & 6 & $42.093\,814\,569$ & 42.0761 \\

849: &7 & $28.217\,524\,934$ & 28.2120&    & 7 & $50.937\,278\,826$ & 50.9214 \\

850: &8 & $32.789\,082\,922$ & 32.7841&    & 8 & $60.185\,767\,651$ & 60.1696 \\

851: &9 & $37.469\,824\,697$ & 37.4653&    & 9 & $69.795\,703\,031$ & 69.7884 \\

852: \end{tabular}

853: \label{t1}

854: \end{table}

855: 

856: It is interesting that the spectrum of the real $|x|^{\epsilon+2}$ potential

857: strongly resembles that of the $x^2(ix)^\epsilon$ potential. The leading-order

858: WKB quantization condition (accurate for $\epsilon>-2$) is like that in

859: (\ref{e34}) except that the factor of $\sin\left(\frac{\pi}{\epsilon+2}\right)$

860: is absent. However, as $\epsilon\to\infty$, the spectrum of $|x|^{\epsilon+2}$

861: approaches that of the square-well potential [$E_n=(n+1)^2\pi^2/4$], while the

862: energies of the complex $x^2(ix)^\epsilon$ potential diverge, as Fig.~\ref{f1}

863: indicates. The energies of the $|x|^P$ potential are shown in Fig.~\ref{f3}.

864: 

865: \begin{figure}[t!]

866: \vspace{2.6in}

867: \special{psfile=Fig3.ps angle=-90 hoffset=78 voffset=211 hscale=38 vscale=38}

868: \caption{Energy levels of the Hamiltonian $H=p^2+|x|^P$ as a function of the

869: real parameter $P$. This figure is similar to Fig.~\ref{f1}, but the eigenvalues

870: do not pinch off and go into the complex plane because the Hamiltonian is

871: Hermitian. (The spectrum becomes dense at $P=0$.)}

872: \label{f3}

873: \end{figure}

874: 

875: \subsection{Numerical Calculation of Eigenvalues}

876: \label{ss2-5}

877: 

878: There are several highly accurate numerical techniques for computing the

879: energy spectrum that is displayed in Fig.~\ref{f1}. The simplest and most direct

880: method is to integrate the Schr\"odinger differential equation (\ref{e24}) using

881: a Runga-Kutta approach. To do so, we convert this complex differential equation

882: to a system of coupled, real, second-order equations. The convergence is most

883: rapid when we integrate along paths located at the centers of the Stokes wedges

884: and follow these paths out to $\infty$. We then patch the two solutions in each

885: Stokes wedge together at the origin. This procedure, which is described in

886: detail in Ref.~\cite{r1}, gives highly accurate numerical results.

887: 

888: An alternative is to perform a variational calculation in which we determine

889: the energy levels by finding the stationary points of the functional

890: \begin{equation}

891: \langle H\rangle(a,b,c)\equiv\frac{\int_C dx\,\psi(x)H\psi(x)}{\int_C dx\,\psi^2

892: (x)},

893: \label{e35}

894: \end{equation}

895: where

896: \begin{equation}

897: \psi(x)=(ix)^c\exp\left[a(ix)^b\right]

898: \label{e36}

899: \end{equation}

900: is a three-parameter class of $\cP\cT$-invariant trial wave functions

901: \cite{VAN}. The integration contour $C$ used to define $\langle H\rangle(a,b,c)$

902: must lie inside the wedges in the complex-$x$ plane in which the wave function

903: falls off exponentially at infinity (see Fig.~\ref{f2}). Rather than having a

904: local minimum, the functional has a saddle point in $(a,b,c)$-space. At this

905: saddle point the numerical prediction for the ground-state energy is extremely

906: accurate for a wide range of $\epsilon$. This method also determines approximate

907: eigenfunctions and eigenvalues of the excited states of these non-Hermitian

908: Hamiltonians. Handy used the numerical technique of solving the coupled moment

909: problem \cite{HANDY}. This technique, which produces accurate results for the

910: eigenvalues, is the exact quantum-mechanical analog of solving the

911: Schwinger-Dyson equations in quantum field theory.

912: 

913: \subsection{The Remarkable Case of a $\cP\cT$-Symmetric $-x^4$ Potential}

914: \label{ss2-6}

915: 

916: The $\cP\cT$-symmetric $-x^4$ Hamiltonian in (\ref{e11}), which is obtained by

917: setting $\epsilon=2$ in (\ref{e12}), is particularly interesting because it is

918: possible to obtain the energy spectrum by using real analysis alone; that is,

919: one can focus on real $x$ only and avoid having to perform analysis in the

920: complex-$x$ plane. One way to proceed is to deform the integration contour shown

921: in Fig.~\ref{f2} to the upper edges of the wedges so that it lies entirely on

922: the real-$x$ axis. If this is done carefully, the exact eigenvalues for this

923: potential can be obtained by solving the Schr\"odinger equation (\ref{e24})

924: subject to the boundary conditions that the potential be {\em reflectionless}

925: \cite{BERRY}. That is, an incoming incident wave from the left gives rise to an

926: outgoing transmitted wave on the right, but no reflected wave on the left. (This

927: observation may have consequences in cosmological models, as explained in

928: Subsec.~\ref{ss8-7}.)

929: 

930: Another way to proceed is to show that the eigenvalues of the non-Hermitian $-

931: x^4$ Hamiltonian are identical with the eigenvalues of a conventional Hermitian

932: Hamiltonian having a positive $x^4$ potential. A number of authors have observed

933: and discussed this equivalence \cite{H3,BG,JM}. Here, we use elementary

934: differential-equation methods \cite{BBCJM} to prove that the spectrum of the

935: non-Hermitian $\cP\cT$-symmetric Hamiltonian

936: \begin{equation}

937: H=\frac{1}{2m}\p^2-g\x^4\quad(g>0)

938: \label{e37}

939: \end{equation}

940: is identical to the spectrum of the Hermitian Hamiltonian

941: \begin{equation}

942: \tilde H=\frac{1}{2m}\p^2+4g\x^4-\hbar\sqrt{\frac{2g}{m}}\,\x\quad(g>0).

943: \label{e38}

944: \end{equation}

945: (We have included the dimensional constants $m$, $g$, and $\hbar$ because they

946: help to elucidate the physical significance of the spectral equivalence of these

947: two very different Hamiltonians.) To show that $H$ in (\ref{e37}) and $\tilde H$

948: in (\ref{e38}) are equivalent, we examine the corresponding Schr\"odinger

949: eigenvalue equations

950: \begin{equation}

951: -\frac{\hbar^2}{2m}\psi''(x)-gx^4\psi(x)=E\psi(x)

952: \label{e39}

953: \end{equation}

954: and

955: \begin{equation}

956: -\frac{\hbar^2}{2m}\Phi''(x)+\left(-\hbar\sqrt{\frac{2g}{m}}\,x+4gx^4\right)

957: \Phi(x)=E\Phi(x).

958: \label{e40}

959: \end{equation}

960: 

961: We begin by moving the complex integration contour for the Schr\"odinger

962: equation (\ref{e39}) to the real axis. To do so, we parameterize the integration

963: contour using

964: \begin{equation}

965: x=-2iL\sqrt{1+iy/L},

966: \label{e41}

967: \end{equation}

968: where

969: \begin{equation}

970: L=\lambda\left[\hbar^2/(mg)\right]^{1/6}

971: \label{e42}

972: \end{equation}

973: and $y$ is a real parameter that ranges from $-\infty$ to $\infty$. A graph of

974: the contour in (\ref{e41}) is shown in Fig.~\ref{f4}. The transformed

975: differential equation then reads

976: \begin{equation}

977: -\frac{\hbar^2}{2m}\left(1+\frac{iy}{L}\right)\phi''(y)

978: -\frac{i\hbar^2}{4Lm}\phi'(y)-16gL^4\left(1+\frac{iy}{L}

979: \right)^2\phi(y)=E\phi(y).

980: \label{e43}

981: \end{equation}

982: 

983: \begin{figure*}[t!]

984: \vspace{1.9in}

985: \special{psfile=Fig4.eps angle=0 hoffset=55 voffset=-3 hscale=48 vscale=48}

986: \caption{Stokes wedges in the lower-half complex-$x$ plane for the

987: Schr\"odinger equation (\ref{e39}) arising from the Hamiltonian $H$ in

988: (\ref{e37}). The eigenfunctions of $H$ decay exponentially as $|x|\to\infty$

989: inside these wedges. Also shown is the contour in (\ref{e41}).}

990: \label{f4}

991: \end{figure*}

992: 

993: Next, we perform a Fourier transform defined by

994: \begin{equation}

995: \tilde f(p)\equiv\int_{-\infty}^\infty dy\,e^{-iyp/\hbar}f(y).

996: \label{e44}

997: \end{equation}

998: By this definition the Fourier transforms of a derivative and a product are

999: given by

1000: \begin{equation}

1001: f'(y)\to ip\tilde f(p)/\hbar\quad{\rm and}\quad yf(y)\to i\hbar\tilde f'(p).

1002: \label{e45}

1003: \end{equation}

1004: Thus, the transformed version of (\ref{e43}) reads

1005: \begin{eqnarray}

1006: &&\frac{1}{2m}\left(1-\frac{\hbar}{L}\frac{d}{dp}\right)p^2\tilde\phi(p)

1007: +\frac{\hbar}{4Lm}p\tilde\phi(p)\nonumber\\

1008: &&\qquad -16gL^4\left(1-\frac{\hbar}{L}\frac{d}{dp}

1009: \right)^2\tilde\phi(p)=E\tilde\phi(p).

1010: \label{e46}

1011: \end{eqnarray}

1012: We expand and simplify the differential equation in (\ref{e46}) and get

1013: \begin{eqnarray}

1014: &&-16gL^2\hbar^2\tilde\phi''(p)+\left(-\frac{\hbar p^2}{2mL}+32gL^3\hbar\right)

1015: \tilde\phi'(p)\nonumber\\

1016: &&\qquad+\left(\frac{p^2}{2m}-\frac{3p\hbar}{4mL}-16gL^4\right)\tilde\phi(p)

1017: =E\tilde\phi(p).

1018: \label{e47}

1019: \end{eqnarray}

1020: 

1021: Next, we eliminate the one-derivative term in the differential equation

1022: (\ref{e47}) to convert it to the form of a Schr\"odinger equation. To do so,

1023: we substitute

1024: \begin{equation}

1025: \tilde\phi(p)=e^{F(p)}\Phi(p),

1026: \label{e48}

1027: \end{equation}

1028: where

1029: \begin{equation}

1030: F(p)=\frac{L}{\hbar}p-\frac{1}{192gmL^3\hbar}p^3.

1031: \label{e49}

1032: \end{equation}

1033: The resulting equation is

1034: \begin{equation}

1035: -16gL^2\hbar^2\Phi''(p)+\left(\frac{p^4}{256gm^2 L^4}

1036: -\frac{\hbar p}{4mL}\right)\Phi(p)=E\Phi(p).

1037: \label{e50}

1038: \end{equation}

1039: 

1040: Last, we rescale (\ref{e50}) by substituting

1041: \begin{equation}

1042: p=xL\sqrt{32mg}

1043: \label{e51}

1044: \end{equation}

1045: and we obtain the Schr\"odinger differential equation (\ref{e40}). The completes

1046: the proof and verifies that the eigenvalues of the two Hamiltonians (\ref{e37})

1047: and (\ref{e38}) have identical eigenvalues. This demonstration of equivalence is

1048: exact; no approximations were made in this argument.\footnote{A simple and exact

1049: transformation like the one presented in this subsection for mapping a $\cP

1050: \cT$-symmetric non-Hermitian Hamiltonian to a Hermitian Hamiltonian has been

1051: found only for the isolated case $\epsilon=2$ in (\ref{e12}). [A more

1052: complicated spectral equivalence exists for the special case $\epsilon=4$ (see

1053: Ref.~\cite{D1,D2}).]}

1054: 

1055: \subsection{Parity Anomaly}

1056: \label{ss2-7}

1057: 

1058: The proof in Subsec.~\ref{ss2-6} that the Hamiltonians in (\ref{e37}) and

1059: (\ref{e38}) are equivalent helps to clarify some of the physical content of the

1060: non-Hermitian $\cP\cT$-symmetric Hamiltonian (\ref{e37}). The interpretation of

1061: the result in (\ref{e40}) is that the linear term in the potential of the

1062: equivalent quartic Hermitian Hamiltonian in (\ref{e38}) is a parity {\em

1063: anomaly}. In general, an {\em anomaly} is a purely quantum (nonclassical) effect

1064: that vanishes in the classical limit $\hbar\to0$. There is no classical analog

1065: of an anomaly because Planck's constant $\hbar$ does not appear in classical

1066: mechanics.

1067: 

1068: We refer to the linear term in (\ref{e38}) as a {\em parity} anomaly for the

1069: following reason: Even though the $\cP\cT$-symmetric Hamiltonian (\ref{e37})

1070: is symmetric under the parity reflections defined in (\ref{e3}), $H$ does not

1071: respect parity symmetry. The violation of parity symmetry is subtle because it

1072: is contained in the boundary conditions that the eigenfunctions of the

1073: associated Schr\"odinger equation must satisfy. Since these boundary conditions

1074: are given at $|x|=\infty$, the violation of parity symmetry is not detectable

1075: in any finite domain in the complex-$x$ plane. Classical motion is a local

1076: phenomenon. Therefore, a classical particle that is moving under the influence

1077: of this Hamiltonian (see Sec.~\ref{s3}) will act as if it is subject to

1078: parity-symmetric forces; the classical particle cannot feel the influences of

1079: quantum boundary conditions imposed at $|x|=\infty$. In contrast, a quantum

1080: wave function is inherently nonlocal because it must obey boundary conditions

1081: that are imposed at $|x|=\infty$. Thus, only a quantum particle ``knows'' about

1082: the violation of parity symmetry. To establish the equivalence between the

1083: Hamiltonians in (\ref{e37}) and (\ref{e38}) it was necessary to perform a

1084: Fourier transform [see (\ref{e46})]. This transformation maps the point at $x=

1085: \infty$ to the point at $p=0$, and this explains the presence of the linear

1086: parity-violating term in the potential of $\tilde H$ in (\ref{e38}). The

1087: violation of parity is now a visible local effect in the Hamiltonian $\tilde H$.

1088: However, this violation of parity is proportional to $\hbar$ and evaporates in

1089: the classical limit $\hbar\to0$.

1090: 

1091: The Hamiltonian (\ref{e38}) is Hermitian in the usual Dirac sense and its energy

1092: spectrum is bounded below. This Hamiltonian is also $\cP\cT$-symmetric because

1093: at every stage in the sequence of differential-equation transformations in

1094: Subsec.~\ref{ss2-6}, $\cP\cT$ symmetry is preserved. However, the variable

1095: $x$ that gives rise to the parity anomaly in (\ref{e40}) is {\em not} a

1096: coordinate variable. Its behavior is that of a momentum variable because $x$

1097: changes sign under time reversal.

1098: 

1099: The violation of parity symmetry at the quantum level has important physical

1100: implications. It is the lack of parity symmetry that implies that the one-point

1101: Green's function in the corresponding quantum field theory does not vanish. A

1102: possible consequence of this is that the elusive Higgs particle, which is a

1103: fundamental ingredient in the standard model of particle physics, is a quantum

1104: anomaly. In the next subsection, we show that the parity anomaly has a major

1105: impact on the spectrum of bound states in a quantum theory.

1106: 

1107: \subsection{Physical Consequence of the Parity Anomaly: Appearance of Bound

1108: States in a $\cP\cT$-Symmetric Quartic Potential}

1109: \label{ss2-8}

1110: 

1111: A direct physical consequence of the parity anomaly is the appearance of bound

1112: states. To elucidate the connection between the parity anomaly and bound states,

1113: we generalize the Hamiltonian (\ref{e37}) to include a harmonic ($\x^2$) term in

1114: the potential:

1115: \begin{eqnarray}

1116: H=\frac{1}{2m}\p^2+\frac{\mu^2}{2}\x^2-g\x^4.

1117: \label{e52}

1118: \end{eqnarray}

1119: The same differential-equation analysis used in Subsec.~\ref{ss2-6}

1120: straightforwardly yields the equivalent Hermitian Hamiltonian \cite{BG,JM}

1121: \begin{eqnarray}

1122: \tilde H=\frac{\p^2}{2m}-\hbar\sqrt{\frac{2g}{m}}\,\x+4g\left(\x^2-\frac{\mu^2}

1123: {8g}\right)^2.

1124: \label{e53}

1125: \end{eqnarray}

1126: Note that for these more general Hamiltonians the linear anomaly term remains

1127: unchanged from that in (\ref{e38}).

1128: 

1129: It was shown in an earlier paper \cite{BOUND} that the Hamiltonian (\ref{e52})

1130: exhibits bound states. In particle physics a {\em bound state} is a state having

1131: a negative binding energy. Bound states in the context of a quantum mechanics

1132: are defined as follows: Let the energy levels of the Hamiltonian be $E_n$ ($n=0,

1133: 1,2,\ldots$). The {\em renormalized mass} is the mass gap; that is, $M=E_1-E_0$.

1134: Higher excitations must be measured relative to the vacuum energy: $E_n-E_0$

1135: ($n=2,3,4,\ldots$). We say that the $n$th higher excitation is a bound state if

1136: the binding energy

1137: \begin{equation}

1138: B_n\equiv E_n-E_0-nM

1139: \label{e54}

1140: \end{equation}

1141: is negative. If $B_n$ is positive, then we regard the state as {\em unbound}

1142: because this state can decay into $n$ 1-particle states of mass $M$ in the

1143: presence of an external field.

1144: 

1145: In Ref.~\cite{BOUND} it was shown numerically that for small positive values of

1146: $g$ the first few states of $H$ in (\ref{e52}) are bound. As $g$ increases, the

1147: number of bound states decreases until, when $g/\mu^3$ is larger than the

1148: critical value $0.0465$, there are no bound states at all.\footnote{In Ref.~

1149: \cite{BOUND} a heuristic argument was given to explain why there is such a

1150: critical value. This argument is heuristic because the non-Hermitian Hamiltonian

1151: is evaluated for $x$ in the complex plane. As is explained in

1152: Subsec.~\ref{ss2-3}, when $x$ is complex, one cannot use

1153: order relationships such as $>$ or $<$, which only apply to real numbers.}

1154: 

1155: Because $H$ in (\ref{e52}) has the same spectrum as the Hermitian Hamiltonian in

1156: (\ref{e53}), it is easy to explain the appearance of bound states and to show

1157: that the bound states are a direct consequence of the linear anomaly term. To

1158: probe the influence of the anomaly, we generalize (\ref{e53}) by inserting a

1159: dimensionless parameter $\epsilon$ that measures the strength of the anomaly

1160: term:

1161: \begin{eqnarray}

1162: \tilde H=\half\p^2-\epsilon\sqrt{2g}\,\x+4g\left(\x^2-\frac{1}{8g}

1163: \right)^2,

1164: \label{e55}

1165: \end{eqnarray}

1166: where for simplicity we have set $m=\mu=\hbar=1$.

1167: 

1168: If we set $\epsilon=0$, there is no anomaly term and the potential is a {\em

1169: symmetric} double well. The mass gap for a double well is exponentially small

1170: because it is a result of the tunneling between the wells and thus the

1171: renormalized mass $M$ is very small. Therefore, $B_n$ in (\ref{e54}) is positive

1172: and there are no bound states. In Fig.~\ref{f5} we display the double-well

1173: potential and the first several states of the system for the case $g=0.046$ and

1174: $\epsilon=0$. There is a very small splitting between the lowest two states.

1175: 

1176: \begin{figure}[t!]

1177: \vspace{2.05in}

1178: \special{psfile=Fig5.eps angle=0 hoffset=29 voffset=-6 hscale=70 vscale=70}

1179: \caption{Potential of the Hermitian Hamiltonian (\ref{e55}) plotted as a

1180: function of the real variable $x$ for the case $\epsilon=0$ and $g=0.046$. The

1181: energy levels are indicated by horizontal lines. Because $\epsilon=0$, there is

1182: no anomaly and the double-well potential is symmetric. Therefore, the mass gap

1183: is very small and thus there are no bound states.}

1184: \label{f5}

1185: \end{figure}

1186: 

1187: If $\epsilon=1$, the double-well potential is asymmetric and the two lowest

1188: states are not approximately degenerate. Therefore, bound states can occur near

1189: the bottom of the potential well. Higher-energy states eventually become unbound

1190: because, as we know from WKB theory, in a quartic well the $n$th energy level

1191: grows like $n^{4/3}$ for large $n$. As $g$ becomes large, the number of bound

1192: states becomes smaller because the depth of the double well decreases. For large

1193: enough $g$ there are no bound states. In Fig.~\ref{f6} we display the potential

1194: for $\epsilon=1$ and for $g=0.046$; for this $g$ there is one bound state.

1195: 

1196: \begin{figure}[b!]

1197: \vspace{1.95in}

1198: \special{psfile=Fig6.eps angle=0 hoffset=29 voffset=-6 hscale=70 vscale=70}

1199: \caption{Asymmetric potential well plotted as a function of the real variable

1200: $x$ for the Hermitian Hamiltonian (\ref{e55}) with $\epsilon=1$ and $g=0.046$.

1201: The energy levels are indicated by horizontal lines. There is one bound state.

1202: The occurrence of bound states is due to the anomaly.}

1203: \label{f6}

1204: \end{figure}

1205: 

1206: Another way to display the bound states is to plot the value of the binding 

1207: energy $B_n$ as a function of $n$. For example, in Fig.~\ref{f7} we display the

1208: bound states for $\epsilon=1$ and $g=0.008333$. Note that for these values there

1209: are 23 bound states. Observe also that the binding energy $B_n$ is a smooth

1210: function of $n$.

1211: 

1212: \begin{figure}[t!]

1213: \vspace{2.1in}

1214: \special{psfile=Fig7.eps angle=0 hoffset=31 voffset=-6 hscale=68 vscale=68}

1215: \caption{Binding energies $B_n=E_n-E_0-nM$ plotted as a function of $n$ for $g=

1216: 0.008333$ and $\epsilon=1$. A negative value of $B_n$ indicates a bound state.

1217: Observe that there are 23 bound states for these parameter values. Note that

1218: $B_n$ is a smooth function of $n$.}

1219: \label{f7}

1220: \end{figure}

1221: 

1222: It is noteworthy that the bound-state spectrum depends so sensitively on the

1223: strength of the anomaly term in the Hamiltonian (\ref{e55}). If $\epsilon$ is

1224: slightly less than $1$, the first few states become unbound, as shown in

1225: Fig.~\ref{f8}. In this figure $g=0.008333$ and $\epsilon=0.9$. If $\epsilon$ is

1226: slightly greater than $1$, the binding energy $B_n$ is not a smooth function of

1227: $n$ for small $n$. In Fig.~\ref{f9} we plot $B_n$ as a function of $n$ for $g=0.

1228: 008333$ and $\epsilon=1.1$. Note that for these values of the parameters there

1229: are 30 bound states. Figures~\ref{f7}, \ref{f8}, and \ref{f9} are strikingly

1230: different, which demonstrates the extreme sensitivity of the bound-state

1231: spectrum to the anomaly term.

1232: 

1233: \begin{figure}[b!]

1234: \vspace{1.95in}

1235: \special{psfile=Fig8.eps angle=0 hoffset=31 voffset=-6 hscale=68 vscale=68}

1236: \caption{Binding energies $B_n$ plotted as a function of $n$ for $g=0.008333$

1237: and $\epsilon=0.9$. The first five states have now become unbound and $B_n$ is

1238: not a smooth function of $n$ for $n\leq6$. The next 12 states are bound, and in

1239: this region $B_n$ is a smooth function of $n$. Comparison of this figure with

1240: Fig.~\ref{f7} shows that the bound-state spectrum is exquisitely sensitive to

1241: the strength of the linear anomaly term.}

1242: \label{f8}

1243: \end{figure}

1244: 

1245: \begin{figure}[t!]

1246: \vspace{2.15in}

1247: \special{psfile=Fig9.eps angle=0 hoffset=31 voffset=-6 hscale=68 vscale=68}

1248: \caption{Binding energies $B_n$ plotted as a function of $n$ for $g=0.008333$

1249: and $\epsilon=1.1$. Note that there are 30 bound states and that $B_n$ is not a

1250: smooth function of $n$ when $n$ is small.}

1251: \label{f9}

1252: \end{figure}

1253: 

1254: \section{$\cP\cT$-Symmetric Classical Mechanics --- The Strange Dynamics of a

1255: Classical Particle Subject to Complex Forces}

1256: \label{s3}

1257: 

1258: In this section we describe the properties of the $\cP\cT$-symmetric

1259: classical-mechanical theory that underlies the quantum-mechanical theory

1260: described by the Hamiltonian (\ref{e12}). We describe the motion of a particle

1261: that feels complex forces and responds by moving about in the complex plane.

1262: Several papers have been published in this area \cite{BBM,CL1,CL2,CL3,CL4,CL5}

1263: and we summarize here some of the surprising discoveries.

1264: 

1265: One objective here is to explain heuristically how an upside-down potential like

1266: that in (\ref{e11}) can have {\em positive}-energy quantum-mechanical

1267: eigenstates. One might think (incorrectly!) that since a classical particle

1268: would slide down the positive real axis to infinity, the corresponding quantum

1269: states would be unstable and the spectrum of the quantum system would be 

1270: unbounded below. In fact, when $\epsilon\geq0$ (the region of unbroken $\cP\cT$

1271: symmetry), all but a set of measure zero of the possible classical paths are

1272: confined and periodic, and thus the classical particle does not slide off to

1273: infinity. When $\epsilon<0$ (the region of broken $\cP\cT$ symmetry), the

1274: classical trajectories do indeed run off to infinity, and we can begin to

1275: understand why the energy levels of the corresponding quantum system are

1276: complex.

1277: 

1278: The equation of motion of a classical particle described by $H$ in (\ref{e12})

1279: follows from Hamilton's equations:

1280: \begin{equation}

1281: {dx\over dt}={\partial H\over\partial p}=2p,\qquad

1282: {dp\over dt}=-{\partial H\over\partial x}=i(2+\epsilon)(ix)^{1+\epsilon}.

1283: \label{e56}

1284: \end{equation}

1285: Combining these two equations gives

1286: \begin{equation}

1287: {d^2x\over dt^2}=2i(2+\epsilon)(ix)^{1+\epsilon},

1288: \label{e57}

1289: \end{equation}

1290: which is the {\em complex} version of Newton's second law, $F=ma$.

1291: 

1292: We can integrate (\ref{e57}) to give

1293: \begin{equation}

1294: {1\over2}{dx\over dt}=\pm\sqrt{E+(ix)^{2+\epsilon}},

1295: \label{e58}

1296: \end{equation}

1297: where $E$ is the energy of the classical particle (the time-independent value of

1298: $H$). We treat time $t$ as a real variable that parameterizes the complex path

1299: $x(t)$ of this particle. Equation (\ref{e58}) is a complex generalization of

1300: the concept that the velocity is the time derivative of the position ($v=\frac{d

1301: x}{dt}$). Here, $t$ is real, but $v$ and $x$ are complex.

1302: 

1303: We now describe and classify the solutions to Eq.~(\ref{e58}). Because the

1304: corresponding quantum theory possesses ${\cP\cT}$ invariance, we restrict our

1305: attention to real values of $E$. Given this restriction, we can rescale $x$ and

1306: $t$ by real numbers so that without loss of generality Eq.~(\ref{e58}) reduces

1307: to

1308: \begin{equation}

1309: {dx\over dt}=\pm\sqrt{1+(ix)^{2+\epsilon}}.

1310: \label{e59}

1311: \end{equation}

1312: 

1313: \subsection{The Case $\epsilon=0$}

1314: \label{ss3-1}

1315: 

1316: The classical solutions to Eq.~(\ref{e59}) have elaborate topologies, so we

1317: begin by considering some special values of $\epsilon$. For the simplest case,

1318: $\epsilon=0$, there are two turning points and these lie on the real axis at

1319: $\pm1$. To solve Eq.~(\ref{e59}) we must specify the initial condition $x(0)$.

1320: An obvious choice for $x(0)$ is a turning point. If the path begins at $\pm1$,

1321: there is a unique trajectory in the complex-$x$ plane that solves (\ref{e59}).

1322: This trajectory lies on the real axis and oscillates between the turning points.

1323: This is the usual sinusoidal harmonic motion.

1324: 

1325: Choosing the energy determines the locations of the turning points, and choosing

1326: the initial position of the particle determines the initial velocity (up to a

1327: plus or minus sign) as well. So if the path of the particle begins anywhere on

1328: the real axis between the turning points, the initial velocity is fixed up to a

1329: sign and the trajectory of the particle still oscillates between the turning

1330: points.

1331: 

1332: In conventional classical mechanics the only possible initial positions for the

1333: particle are on the real-$x$ axis between the turning points because the

1334: velocity is real; all other points on the real axis belong to the so-called

1335: classically forbidden region. However, because we are analytically continuing

1336: classical mechanics into the complex plane, we can choose any point $x(0)$ in

1337: the complex plane as an initial position. For all complex initial positions

1338: outside of the conventional classically allowed region the classical trajectory

1339: is an ellipse whose foci are the turning points. The ellipses are nested because

1340: no trajectories may cross. These ellipses are shown in Fig.~\ref{f10}.

1341: 

1342: \begin{figure*}[b!]

1343: \vspace{2.2in}

1344: \special{psfile=Fig10.ps angle=-90 hoffset=37 voffset=186 hscale=35 vscale=35}

1345: \caption{Classical trajectories in the complex-$x$ plane for the

1346: harmonic-oscillator Hamiltonian $H=p^2+x^2$. These trajectories are the

1347: complex paths of a particle whose energy is $E=1$. The trajectories are nested

1348: ellipses with foci located at the turning points at $x=\pm1$. The real line

1349: segment (degenerate ellipse) connecting the turning points is the conventional

1350: real periodic classical solution to the harmonic oscillator. All paths are

1351: closed orbits having the same period $2\pi$.}

1352: \label{f10}

1353: \end{figure*}

1354: 

1355: The exact harmonic-oscillator ($\epsilon=0$) solution to (\ref{e59}) is

1356: \begin{equation}

1357: x(t)=\cos[{\rm arccos}\,x(0)\pm t],

1358: \label{e60}

1359: \end{equation}

1360: where the sign of $t$ determines the direction (clockwise or anticlockwise) in

1361: which the particle traces out the ellipse. For {\em any} ellipse the period is

1362: $2\pi$. The period is the same for all trajectories because we can join the

1363: square-root branch points by a single finite branch cut lying along the real

1364: axis from $x=-1$ to $x=1$. The complex path integral that determines the period

1365: can then be shrunk (by Cauchy's theorem) to the usual real integral joining the

1366: turning points.

1367: 

1368: Note that all of the elliptical orbits in Fig.~\ref{f10} are symmetric with

1369: respect to parity $\cP$ (reflections through the origin) and time reversal $\cT$

1370: (reflections about the real axis) as well as to $\cP\cT$ (reflections about the

1371: imaginary axis). Furthermore, $\cP$ and $\cT$ individually preserve the

1372: directions in which the ellipses are traversed.

1373: 

1374: \subsection{The Case $\epsilon=1$}

1375: \label{ss3-2}

1376: 

1377: When $\epsilon=1$, there are three turning points. These turning points solve

1378: the equation $ix^3=1$. Two lie below the real axis and are symmetric with

1379: respect to the imaginary axis:

1380: \begin{equation}

1381: x_-=e^{-5i\pi/6}\quad{\rm and}\quad x_+=e^{-i\pi/6}.

1382: \label{e61}

1383: \end{equation}

1384: Under $\cP\cT$ reflection $x_-$ and $x_+$ are interchanged. The third turning

1385: point lies on the imaginary axis at $x_0=i$.

1386: 

1387: Like the case $\epsilon=0$, the trajectory of a particle that begins at the

1388: turning point $x_-$ follows a path in the complex-$x$ plane to the turning point

1389: at $x_+$. Then, the particle retraces its path back to the turning point at

1390: $x_-$, and it continues to oscillate between these two turning points. This path

1391: is shown on Fig.~\ref{f11}. The period of this motion is $2\sqrt{3\pi}\Gamma

1392: \left(\frac{4}{3}\right)/\Gamma\left(\frac{5}{6}\right)$. A particle beginning

1393: at the third turning point $x_0$ exhibits a completely distinct motion: It

1394: travels up the imaginary axis and reaches $i\infty$ in a finite time $\sqrt{\pi}

1395: \Gamma\left(\frac{4}{3}\right)/\Gamma\left(\frac{5}{6}\right)$. This motion is

1396: not periodic.

1397: 

1398: \begin{figure*}[t!]

1399: \vspace{2.4in}

1400: \special{psfile=Fig11.ps angle=-90 hoffset=37 voffset=186 hscale=35 vscale=35}

1401: \caption{Classical trajectories in the complex-$x$ plane for a particle of

1402: energy $E=1$ described by the Hamiltonian $H=p^2+ix^3$. An oscillatory

1403: trajectory connects the turning points $x_{\pm}$. This trajectory is enclosed by

1404: a set of closed, nested paths that fill the finite complex-$x$ plane except for

1405: points on the imaginary axis at or above the turning point $x_0=i$. Trajectories

1406: that originate on the imaginary axis above $x=i$ either move off to $i\infty$ or

1407: else approach $x_0$, stop, turn around, and then move up the imaginary axis to

1408: $i\infty$.}

1409: \label{f11}

1410: \end{figure*}

1411: 

1412: Paths originating from all other points in the finite complex-$x$ plane follow

1413: closed periodic orbits. No two orbits may intersect; rather they are all nested,

1414: like the ellipses for the case $\epsilon=0$. All of these orbits encircle the

1415: turning points $x_\pm$ and, by virtue of Cauchy's theorem, have the same period

1416: $2\sqrt{3\pi}\Gamma\left(\frac{4}{3}\right)/\Gamma\left(\frac{5}{6}\right)$ as

1417: the oscillatory path connecting $x_\pm$. Because these orbits must avoid

1418: crossing the trajectory that runs up the positive imaginary axis from the

1419: turning point at $x_0=i$, they are pinched in the region just below $x_0$, as 

1420: shown on Fig.~\ref{f11}.

1421: 

1422: \subsection{The Case $\epsilon=2$}

1423: \label{ss3-3}

1424: 

1425: When $\epsilon=2$, there are four turning points, two below the real axis and

1426: symmetric with respect to the imaginary axis, $x_1=e^{-3i\pi/4}$ and $x_2=e^{-i

1427: \pi/4}$, and two more located above the real axis and symmetric with respect to

1428: the imaginary axis, $x_3=e^{i\pi/4}$ and $x_4=e^{3i\pi/4}$. These turning points

1429: are solutions to the equation $-x^4=1$. Classical trajectories that oscillate

1430: between the pair $x_1$ and $x_2$ and the pair $x_3$ and $x_4$ are shown on

1431: Fig.~\ref{f12}. The period of these oscillations is $2\sqrt{2\pi}\Gamma\left(

1432: \frac{5}{4}\right)/\Gamma\left(\frac{3}{4}\right)$. Trajectories that begin

1433: elsewhere in the complex-$x$ plane are also shown on Fig.~\ref{f10}. By virtue

1434: of Cauchy's theorem all these nested nonintersecting trajectories have the same

1435: period. All classical motion is periodic except for the special trajectories

1436: that begin on the real axis. A particle that begins on the real-$x$ axis runs

1437: off to $\pm\infty$; its trajectory is nonperiodic.

1438: 

1439: \begin{figure*}[t!]

1440: \vspace{2.4in}

1441: \special{psfile=Fig12.ps angle=-90 hoffset=37 voffset=186 hscale=35 vscale=35}

1442: \caption{Classical trajectories in the complex-$x$ plane for a particle

1443: described by the Hamiltonian $H=p^2-x^4$ and having energy $E=1$. There are two

1444: oscillatory trajectories connecting the pairs of turning points $x_1$ and $x_2$

1445: in the lower-half $x$-plane and $x_3$ and $x_4$ in the upper-half $x$-plane. [A

1446: trajectory joining any other pair of turning points is forbidden because it

1447: would violate $\cP\cT$ (left-right) symmetry.] The oscillatory trajectories are

1448: surrounded by closed orbits of the same period. In contrast to these periodic

1449: orbits, there is a special class of trajectories having unbounded path length

1450: and running along the real-$x$ axis.}

1451: \label{f12}

1452: \end{figure*}

1453: 

1454: \subsection{Broken and Unbroken Classical $\cP\cT$ Symmetry}

1455: \label{ss3-4}

1456: 

1457: We can now understand heuristically why the energies of the corresponding $\cP

1458: \cT$-symmetric quantum systems are real. In each of Figs.~\ref{f10}, \ref{f11},

1459: and \ref{f12} we can see that all of the orbits are localized and periodic. We

1460: may regard the pictured classical motions as representing particles confined to

1461: and orbiting around {\em complex atoms}! We can then use Bohr-Sommerfeld

1462: quantization to determine the discrete energies of the system:

1463: \begin{equation}

1464: \oint_C dx\,p=\oint_C dx\,\sqrt{E-x^2(ix)^\epsilon}=\left(n+\half\right)\pi,

1465: \label{e62}

1466: \end{equation}

1467: where $C$ represents the orbit of a classical particle in the complex-$x$ plane.

1468: By Cauchy's theorem, any closed orbit leads to the same result for the energy 

1469: $E_n$, and because of $\cP\cT$ symmetry the integral above gives a {\em real}

1470: value for the energy.

1471: 

1472: The key difference between classical paths for $\epsilon>0$ and for $\epsilon<0$

1473: is that in the former case the paths (except for isolated examples) are closed

1474: orbits and in the latter case the paths are open orbits. In Fig.~\ref{f13} we

1475: consider the case $\epsilon=-0.2$ and display two paths that begin on the

1476: negative imaginary axis. Because $\epsilon$ is noninteger, there is a branch cut

1477: and the classical particle travels on a Riemann surface rather than on a single

1478: sheet of the complex plane. In this figure one path evolves forward in time and

1479: the other path evolves backward in time. Each path spirals outward and 

1480: eventually moves off to infinity. Note that the pair of paths forms a $\cP

1481: \cT$-symmetric structure. We remark that the paths do not cross because they are

1482: on different sheets of the Riemann surface. The function $(ix)^{-0.2}$ requires

1483: a branch cut, and we take this branch cut to lie along the positive imaginary

1484: axis. The forward-evolving path leaves the principal sheet (sheet 0) of the

1485: Riemann surface and crosses the branch cut in the positive sense and continues

1486: on sheet 1. The reverse path crosses the branch cut in the negative sense and

1487: continues on sheet $-1$. Figure \ref{f13} shows the projection of the classical

1488: orbit onto the principal sheet.

1489: 

1490: \begin{figure*}[t!]

1491: \vspace{2.6in}

1492: \special{psfile=Fig13.ps angle=0 hoffset=100 voffset=-3 hscale=20 vscale=20}

1493: \caption{Classical trajectories in the complex-$x$ plane for the Hamiltonian in

1494: (\ref{e21}) with $\epsilon=-0.2$. These trajectories begin on the negative

1495: imaginary axis very close to the origin. One trajectory evolves forward in time

1496: and the other goes backward in time. The trajectories are open orbits and show

1497: the particle spiraling off to infinity. The trajectories begin on the principal

1498: sheet of the Riemann surface; as they cross the branch cut on the positive

1499: imaginary axis, they visit the higher and lower sheets of the surface. The

1500: trajectories do not cross because they lie on different Riemann sheets.}

1501: \label{f13}

1502: \end{figure*}

1503: 

1504: Figure \ref{f13} shows why the energies of the quantum system in the broken

1505: $\cP\cT$-symmetric region $\epsilon<0$ are not real. The answer is simply that

1506: the trajectories are not closed orbits; the trajectories are {\em open} orbits,

1507: and all classical particles drift off to $x=\infty$. As explained after

1508: (\ref{e32}), if we attempt to quantize the system for the case $\epsilon<0$

1509: using the Bohr-Sommerfeld integral in (\ref{e62}), the integral does not exist

1510: because the integration contour is not closed.

1511: 

1512: \subsection{Noninteger Values of $\epsilon$}

1513: \label{ss3-5}

1514: 

1515: As $\epsilon$ increases from 0, the turning points at $x=1$ (and at $x=-1$), as

1516: shown in Fig.~\ref{f10} rotate downward and clockwise (anticlockwise) into the

1517: complex-$x$ plane. These turning points are solutions to the equation $1+(ix)^{2

1518: +\epsilon}=0$. When $\epsilon$ is noninteger, this equation has many solutions

1519: that all lie on the unit circle and have the form

1520: \begin{equation}

1521: x=\exp\left(i\pi\frac{4N-\epsilon}{4+2\epsilon}\right)\quad(N~{\rm integer}).

1522: \label{e63}

1523: \end{equation}

1524: These turning points occur in $\cP\cT$-symmetric pairs (pairs that are symmetric

1525: when reflected through the imaginary axis) corresponding to the $N$ values

1526: $(N=-1,~N=0)$, $(N=-2,~N=1)$, $(N=-3,~N=2)$, $(N=-4,~N=3)$, and so on. We label

1527: these pairs by the integer $K$ ($K=0,~1,~2,~3,~\ldots$) so that the $K$th pair

1528: corresponds to $(N=-K-1,~N=K)$. The pair of turning points on the real-$x$ axis

1529: for $\epsilon=0$ deforms continuously into the $K=0$ pair of turning points when

1530: $\epsilon>0$. When $\epsilon$ is rational, there are a finite number of turning

1531: points in the complex-$x$ Riemann surface. For example, when $\epsilon=\frac{12}

1532: {5}$, there are 5 sheets in the Riemann surface and 11 pairs of turning points.

1533: The $K=0$ pair of turning points are labeled $N=-1$ and $N=0$, the $K=1$ pair

1534: are labeled $N=-2$ and $N=1$, and so on. The last ($K=10$) pair of turning

1535: points are labeled $N=-11$ and $N=10$. These turning points are shown in

1536: Fig.~\ref{f14}.

1537: 

1538: \begin{figure}[t!]

1539: \vspace{4.3in}

1540: \special{psfile=Fig14.ps angle=0 hoffset=88 voffset=-12 hscale=49 vscale=49}

1541: \caption{Locations of the turning points for $\epsilon=\frac{12}{5}$. There are

1542: 11 $\cP\cT$-symmetric pairs of turning points, with each pair being mirror

1543: images under reflection through the imaginary-$x$ axis on the principal sheet.

1544: All 22 turning points lie on the unit circle on a five-sheeted Riemann surface,

1545: where the sheets are joined by cuts on the positive-imaginary axis.}

1546: \label{f14}

1547: \end{figure}

1548: 

1549: As $\epsilon$ increases from 0, the elliptical complex trajectories in

1550: Fig.~\ref{f10} for the harmonic oscillator begin to distort but the trajectories

1551: remain closed and periodic except for special singular trajectories that run off

1552: to complex infinity, as we see in Fig.~\ref{f11}. These singular trajectories

1553: only occur when $\epsilon$ is an integer. Nearly all of the orbits that one

1554: finds are $\cP\cT$ symmetric (left-right symmetric), and until very recently it

1555: was thought that all closed periodic orbits are $\cP\cT$ symmetric. This is, in

1556: fact, not so \cite{CL4}. Closed non-$\cP\cT$-symmetric orbits exist, and these

1557: orbits are crucial for understanding the behavior of the periods of the complex

1558: orbits as $\epsilon$ varies.

1559: 

1560: In Fig.~\ref{f15} we display a $\cP\cT$-symmetric orbit of immense topological

1561: intricacy that visits many sheets of the Riemann surface. This figure shows a

1562: classical trajectory corresponding to $\epsilon=\pi-2$. The trajectory starts at

1563: $x(0)=-7.1i$ and visits $11$ sheets of the Riemann surface. Its period is

1564: $T=255.3$. The structure of this orbit near the origin is so complicated that we

1565: provide a magnified version in Fig.~\ref{f16}.

1566: 

1567: \begin{figure*}[t!]

1568: \vspace{2.9in}

1569: \special{psfile=Fig15.ps angle=-90 hoffset=100 voffset=195 hscale=20 vscale=20}

1570: \caption{A classical trajectory in the complex-$x$ plane for the complex

1571: Hamiltonian $H=p^2x^2(ix)^{\pi-2}$. This complicated trajectory begins at $x(0)=

1572: -7.1i$ and visits 11 sheets of the Riemann surface. Its period is approximately

1573: $T=255.3$. This figure displays the projection of the trajectory onto the

1574: principal sheet of the Riemann surface. This trajectory does not cross itself.}

1575: \label{f15}

1576: \end{figure*}

1577: 

1578: \begin{figure*}[t!]

1579: \vspace{2.4in}

1580: \special{psfile=Fig16.ps angle=-90 hoffset=100 voffset=167 hscale=20 vscale=20}

1581: \caption{An enlargement of the classical trajectory $x(t)$ in Fig.~\ref{f15}

1582: showing the detail near the origin in the complex-$x$ plane. We emphasize that

1583: this classical path never crosses itself; the apparent self-intersections are

1584: paths that lie on different sheets of the Riemann surface.}

1585: \label{f16}

1586: \end{figure*}

1587: 

1588: The period of any classical orbit depends on the specific pairs of turning

1589: points that are enclosed by the orbit and on the number of times that the orbit

1590: encircles each pair. As shown in Ref.~\cite{CL2}, any given orbit can be

1591: deformed to a simpler orbit of exactly the same period. This simpler orbit

1592: connects two turning points and oscillates between them rather than encircling

1593: them. For the elementary case of orbits that enclose only the $K=0$ pair of

1594: turning points, the formula for the period of the closed orbit is

1595: \begin{eqnarray}

1596: T_0(\epsilon)=2\sqrt{\pi}\frac{\Gamma\left(\frac{3+\epsilon}{2+\epsilon}\right)}

1597: {\Gamma\left(\frac{4+\epsilon}{r4+2\epsilon}\right)}\cos\left(\frac{\epsilon\pi}

1598: {4+2\epsilon}\right)\qquad(\epsilon\geq0).

1599: \label{e64}

1600: \end{eqnarray}

1601: The derivation of (\ref{e64}) goes as follows: The period $T_0$ is given by a

1602: closed contour integral along the trajectory in the complex-$x$ plane. This

1603: trajectory encloses the square-root branch cut that joins the $K=0$ pair of

1604: turning points. This contour can be deformed into a pair of rays that run from

1605: one turning point to the origin and then from the origin to the other turning

1606: point. The integral along each ray is easily evaluated as a beta function, which

1607: is then written in terms of gamma functions.

1608: 

1609: When the classical orbit encloses more than just the $K=0$ pair of turning

1610: points, the formula for the period of the orbit becomes more complicated

1611: \cite{CL2}. In general, there are contributions to the period integral from many

1612: enclosed pairs of turning points. We label each such pair by the integer $j$.

1613: The formula for the period of the topological class of classical orbits whose

1614: central orbit terminates on the $K$th pair of turning points is

1615: \begin{eqnarray}

1616: T_K(\epsilon)=2\sqrt{\pi}\frac{\Gamma\left(\frac{3+\epsilon}{2+\epsilon}\right)}

1617: {\Gamma\left(\frac{4+\epsilon}{4+2\epsilon}\right)}\sum_{j=0}^{\infty}a_j(K,

1618: \epsilon)\left|\cos\left(\frac{(2j+1)\epsilon\pi}{4+2\epsilon}\right)\right|.

1619: \label{e65}

1620: \end{eqnarray}

1621: In this formula the cosines originate from the angular positions of the turning

1622: points in (\ref{e63}). The coefficients $a_j(K,\epsilon)$ are all nonnegative

1623: integers. The $j$th coefficient is nonzero only if the classical path encloses

1624: the $j$th pair of turning points. Each coefficient is an {\em even} integer

1625: except for the $j=K$ coefficient, which is an odd integer. The coefficients $a_j

1626: (K,\epsilon)$ satisfy

1627: \begin{eqnarray}

1628: \sum_{j=0}^{\infty}a_j(K,\epsilon)=k,

1629: \label{e66}

1630: \end{eqnarray}

1631: where $k$ is the number of times that the central classical path crosses the

1632: imaginary axis. Equation (\ref{e66}) truncates the summation in (\ref{e65}) to

1633: a finite number of terms.

1634: 

1635: The period $T_0$ in (\ref{e64}) of orbits connecting the $K=0$ turning points is

1636: a smoothly decreasing function of $\epsilon$. However, for classical orbits

1637: connecting the $K$th ($K>0$) pair of turning points, the classical orbits

1638: exhibit fine structure that is exquisitely sensitive to the value of $\epsilon$.

1639: Small variations in $\epsilon$ can cause huge changes in the topology and in the

1640: periods of the closed orbits. Depending on $\epsilon$, there are orbits having

1641: short periods as well as orbits having long and possibly arbitrarily long

1642: periods.

1643: 

1644: \subsection{Classical Orbits Having Spontaneously Broken $\cP\cT$ Symmetry}

1645: \label{ss3-6}

1646: 

1647: There is a general pattern that holds for all $K$. For classical orbits that

1648: oscillate between the $K$th pair of turning points, there are three regions of

1649: $\epsilon$. The domain of Region I is $0\leq\epsilon\leq\frac{1}{K}$, the domain

1650: of Region II is $\frac{1}{K}<\epsilon<4K$, and the domain of Region III is $4K<

1651: \epsilon$. In Regions I and III the period is a small and smoothly decreasing

1652: function of $\epsilon$. However, in Region II the period is a rapidly varying

1653: and noisy function of $\epsilon$. We illustrate this behavior for the case $K=1$

1654: and $K=2$ in Figs.~\ref{f17} and \ref{f18}.

1655: 

1656: \begin{figure*}[t!]

1657: \vspace{4.45in}

1658: \special{psfile=Fig17.ps angle=0 hoffset=68 voffset=-12 hscale=45 vscale=45}

1659: \caption{Period of a classical trajectory beginning at the $N=1$ turning point

1660: in the complex-$x$ plane. The period is plotted as a function of $\epsilon$. The

1661: period decreases smoothly for $0\leq\epsilon<1$ (Region I). However, when $1\leq

1662: \epsilon\leq4$ (Region II), the period becomes a rapidly varying and noisy

1663: function of $\epsilon$. For $\epsilon>4$ (Region III) the period is once again a

1664: smoothly decaying function of $\epsilon$. Region II contains short subintervals

1665: where the period is a small and smoothly varying function of $\epsilon$. At the

1666: edges of these subintervals the period suddenly becomes extremely long. Detailed

1667: numerical analysis shows that the edges of the subintervals lie at special

1668: rational values of $\epsilon$. Some of these special rational values of

1669: $\epsilon$ are indicated by vertical line segments that cross the horizontal

1670: axis. At these rational values the orbit does not reach the $N=-2$ turning point

1671: and the $\cP\cT$ symmetry of the classical orbit is spontaneously broken.}

1672: \label{f17}

1673: \end{figure*}

1674: 

1675: \begin{figure*}[t!]

1676: \vspace{4.45in}

1677: \special{psfile=Fig18.ps angle=0 hoffset=67 voffset=-13 hscale=45 vscale=45}

1678: \caption{Period of a classical trajectory joining (except when $\cP\cT$ symmetry

1679: is broken) the $K=2$ pair of turning points. The period is plotted as a function

1680: of $\epsilon$. As in the $K=1$ case shown in Fig.~\ref{f17}, there are three

1681: regions. When $0\leq\epsilon\leq\half$ (Region I), the period is a smooth

1682: decreasing function of $\epsilon$; when $\half<\epsilon\leq8$ (Region II), the

1683: period is a rapidly varying and choppy function of $\epsilon$; when $8<\epsilon$

1684: (Region III), the period is again a smooth and decreasing function of

1685: $\epsilon$.}

1686: \label{f18}

1687: \end{figure*}

1688: 

1689: The abrupt changes in the topology and the periods of the orbits for $\epsilon$

1690: in Region II are associated with the appearance of orbits having {\em

1691: spontaneously broken} $\cP\cT$ symmetry. In Region II there are short patches

1692: where the period is relatively small and is a slowly varying function of

1693: $\epsilon$. These patches are bounded by special values of $\epsilon$ for which

1694: the period of the orbit suddenly becomes extremely long. Numerical studies of

1695: the orbits connecting the $K$th pair of turning points indicate that these

1696: special values of $\epsilon$ are always {\em rational} \cite{CL5}. Furthermore,

1697: at these special rational values of $\epsilon$, the closed orbits are {\em not}

1698: $\cP\cT$-symmetric (left-right symmetric). Such orbits exhibit {\em

1699: spontaneously broken} $\cP\cT$ symmetry. Some special values of $\epsilon$ at

1700: which spontaneously broken $\cP\cT$-symmetric orbits occur are indicated in

1701: Figs.~\ref{f17} and \ref{f18} by short vertical lines below the horizontal axis.

1702: These special values of $\epsilon$ have the form $\frac{p}{q}$, where $p$

1703: is a multiple of 4 and $q$ is odd.

1704: 

1705: A broken-$\cP\cT$-symmetric orbit is a failed $\cP\cT$-symmetric orbit. Figure

1706: \ref{f19} displays a spontaneously-broken-$\cP\cT$-symmetric orbit for $\epsilon

1707: =\frac{4}{5}$. The orbit starts at the $N=2$ turning point, but it never reaches

1708: the $\cP\cT$-symmetric turning point $N=-3$. Rather, the orbit terminates when

1709: it runs into and is reflected back from the complex conjugate $N=4$ turning

1710: point [see (\ref{e63})]. The period of the orbit is short ($T=4.63$). While this

1711: orbit is not $\cP\cT$ (left-right) symmetric, it does possess complex-conjugate

1712: (up-down) symmetry. In general, for a non-$\cP\cT$-symmetric orbit to exist, it

1713: must join or encircle a pair of complex-conjugate turning points.

1714: 

1715: \begin{figure*}[t!]

1716: \vspace{2.60in}

1717: \special{psfile=Fig19.ps angle=0 hoffset=109 voffset=-11 hscale=42 vscale=42}

1718: \caption{A horseshoe-shaped non-$\cP\cT$-symmetric orbit. This orbit is not

1719: symmetric with respect to the imaginary axis, but it is symmetric with respect

1720: to the real axis. The orbit terminates at a complex-conjugate pair of turning

1721: points. For this orbit $\epsilon=\frac{4}{5}$.}

1722: \label{f19}

1723: \end{figure*}

1724: 

1725: If we change $\epsilon$ slightly, $\cP\cT$ symmetry is restored and one can only

1726: find orbits that are $\cP\cT$ symmetric. For example, if we take $\epsilon=

1727: 0.805$, we obtain the complicated orbit in Fig.~\ref{f20}. The period of this

1728: orbit is large ($T=173.36$).

1729: 

1730: \begin{figure*}[ht!]

1731: \vspace{2.95in}

1732: \special{psfile=Fig20.ps angle=0 hoffset=109 voffset=-11 hscale=45 vscale=45}

1733: \caption{$\cP\cT$-symmetric orbit for $\epsilon=0.805$. This orbit connects

1734: the $K=2$ pair of turning points.}

1735: \label{f20}

1736: \end{figure*}

1737: 

1738: Broken-$\cP\cT$-symmetric orbits need not be simple looking, like the orbit

1739: shown in Fig.~\ref{f19}. Indeed, they can have an elaborate topology. As an

1740: example we plot in Fig.~\ref{f21} the complicated orbit that arises when

1741: $\epsilon=\frac{16}{9}$. This orbit is a failed $K=3$ $\cP\cT$-symmetric orbit

1742: that originates at the $N=-4$ turning point, but never reaches the $\cP

1743: \cT$-symmetric $N=3$ turning point. Instead, it is reflected back by the

1744: complex-conjugate $N=-14$ turning point. 

1745: 

1746: \begin{figure*}[t!]

1747: \vspace{3.65in}

1748: \special{psfile=Fig21.eps angle=0 hoffset=95 voffset=-9 hscale=59 vscale=59}

1749: \caption{Non-$\cP\cT$-symmetric orbit for $\epsilon=\frac{16}{9}$. This

1750: topologically complicated orbit originates at the $N=-4$ turning point but

1751: does not reach the $\cP\cT$-symmetric $N=3$ turning point. Instead, it is

1752: reflected back at the complex-conjugate $N=-14$ turning point. The period of

1753: this orbit is $T=186.14$.}

1754: \label{f21}

1755: \end{figure*}

1756: 

1757: This study of classical orbits provides a heuristic explanation of the quantum

1758: transition from a broken to an unbroken $\cP\cT$ symmetry as $\epsilon$

1759: increases past $0$. The quantum transition corresponds to a change from open to

1760: closed classical orbits. Furthermore, we can now see why the quantum theory in

1761: the unbroken region is unitary. At the classical level, particles are bound in a

1762: complex atom and cannot escape to infinity; at the quantum level the probability

1763: is conserved and does not leak away as time evolves. Quantum mechanics is

1764: obtained by summing over all possible classical trajectories, and in the case of

1765: $\cP\cT$-symmetric classical mechanics we have seen some bizarre classical

1766: trajectories. To understand how summing over such trajectories produces $\cP

1767: \cT$-symmetric quantum mechanics will require much more research.

1768: 

1769: \section{$\cP\cT$-Symmetric Quantum Mechanics}

1770: \label{s4}

1771: 

1772: Establishing that the eigenvalues of many $\cP\cT$-symmetric Hamiltonians are

1773: real and positive raises an obvious question: Does a non-Hermitian Hamiltonian

1774: such as $H$ in (\ref{e12}) define a physical theory of quantum mechanics or is

1775: the reality and positivity of the spectrum merely an intriguing mathematical

1776: curiosity exhibited by some special classes of complex eigenvalue problems?

1777: Recall that a {\em physical} quantum theory must (i) have an energy spectrum

1778: that is bounded below; (ii) possess a Hilbert space of state vectors that is

1779: endowed with an inner product having a positive norm; (iii) have unitary time

1780: evolution. The simplest condition on the Hamiltonian $H$ that guarantees that

1781: the quantum theory satisfies these three requirements is that $H$ be real and

1782: symmetric. However, this condition is overly restrictive. One can allow $H$ to

1783: be complex as long as it is Dirac Hermitian: $H^\dagger=H$. In this section we

1784: explain why we can replace the condition of Hermiticity by the condition that

1785: $H$ have an unbroken $\cP\cT$ symmetry and still satisfy the above requirements

1786: for a physical quantum theory.\footnote{All of the $\cP\cT$-symmetric

1787: Hamiltonians considered in this paper are symmetric under matrix transposition.

1788: This matrix symmetry condition is not necessary, but it has the simplifying

1789: advantage that the we do not need to have a biorthogonal set of basis states. We

1790: can consider $\cP\cT$-symmetric Hamiltonians that are not symmetric under matrix

1791: transposition, but only at the cost of introducing a biorthogonal basis

1792: \cite{WWWW1,BIOX}.}

1793: 

1794: \subsection{Recipe for a Quantum-Mechanical Theory Defined by a Hermitian

1795: Hamiltonian}

1796: \label{ss4-1}

1797: 

1798: For purposes of comparison, we summarize in this subsection the standard

1799: textbook procedure that one follows in analyzing a theory defined by a

1800: conventional Hermitian quantum-mechanical Hamiltonian. In the next subsection

1801: we repeat these procedures for a non-Hermitian Hamiltonian.

1802: 

1803: \begin{itemize}

1804: \item[(a)] {\em Eigenfunctions and eigenvalues of $H$}. Given the Hamiltonian

1805: $H$ one can write down the time-independent Schr\"odinger equation associated

1806: with $H$ and calculate the eigenvalues $E_n$ and eigenfunctions $\psi_n(x)$.

1807: Usually, this calculation is difficult to perform analytically, so it must be

1808: done numerically.

1809: 

1810: \item[(b)] {\em Orthogonality of eigenfunctions.} Because $H$ is Hermitian, the

1811: eigenfunctions of $H$ will be orthogonal with respect to the standard Hermitian

1812: inner product:

1813: \begin{equation}

1814: (\psi,\phi)\equiv\int dx\,[\psi(x)]^*\phi(x)

1815: \label{e67}

1816: \end{equation}

1817: {\em Orthogonality} means that the inner product of two eigenfunctions $\psi_m(

1818: x)$ and $\psi_n(x)$ associated with different eigenvalues $E_m\neq E_n$

1819: vanishes:

1820: \begin{equation}

1821: (\psi_m,\phi_n)=0.

1822: \label{e68}

1823: \end{equation}

1824: (We do not discuss here the technical problems associated with degenerate 

1825: spectra.)

1826: 

1827: \item[(c)] {\em Orthonormality of eigenfunctions}. Since the Hamiltonian is

1828: Hermitian, the norm of any vector is guaranteed to be positive. This means that

1829: we can normalize the eigenfunctions of $H$ so that the norm of every

1830: eigenfunction is unity:

1831: \begin{equation}

1832: (\psi_n,\psi_n)=1.

1833: \label{e69}

1834: \end{equation}

1835: 

1836: \item[(d)] {\em Completeness of eigenfunctions}. It is a deep theorem of the

1837: theory of linear operators on Hilbert spaces that the eigenfunctions of a

1838: Hermitian Hamiltonian are {\em complete}. This means that any (finite-norm)

1839: vector $\chi$ in the Hilbert space can be expressed as a linear combination of

1840: the eigenfunctions of $H$:

1841: \begin{equation}

1842: \chi=\sum_{n=0}^\infty a_n\psi_n.

1843: \label{e70}

1844: \end{equation}

1845: The formal statement of completeness in coordinate space is the reconstruction

1846: of the unit operator (the delta function) as a sum over the eigenfunctions:

1847: \begin{equation}

1848: \sum_{n=0}^\infty[\psi_n(x)]^*\psi_n(y)=\delta(x-y).

1849: \label{e71}

1850: \end{equation}

1851: 

1852: \item[(e)] {\em Reconstruction of the Hamiltonian $H$ and the Green's function

1853: $G$, and calculation of the spectral Zeta function}. The Hamiltonian matrix in

1854: coordinate space has the form

1855: \begin{equation}

1856: \sum_{n=0}^\infty [\psi_n(x)]^*\psi_n(y)E_n=H(x,y)

1857: \label{e72}

1858: \end{equation}

1859: and the Green's function is given by

1860: \begin{equation}

1861: \sum_{n=0}^\infty [\psi_n(x)]^*\psi_n(y)\frac{1}{E_n}=G(x,y).

1862: \label{e73}

1863: \end{equation}

1864: The Green's function is the matrix inverse of the Hamiltonian in the sense that

1865: \begin{equation}

1866: \int dy\,H(x,y)G(y,z)=\delta(x-z).

1867: \label{e74}

1868: \end{equation}

1869: The formula for the Green's function in (\ref{e73}) allows us to calculate the

1870: sum of the reciprocals of the energy eigenvalues. We simply set $x=y$ in

1871: (\ref{e73}), integrate with respect to $x$, and use the normalization condition

1872: in (\ref{e69}) to obtain the result that

1873: \begin{equation}

1874: \int dx\,G(x,x)=\sum_{n=0}^\infty\frac{1}{E_n}.

1875: \label{e75}

1876: \end{equation}

1877: The summation on the right side of (\ref{e75}) is called the {\em spectral zeta

1878: function}. This sum is convergent if the energy levels $E_n$ rise faster than

1879: linearly with $n$. Thus, the spectral zeta function for the harmonic oscillator

1880: is divergent, but it exists for the $|x|^{\epsilon+2}$ and $x^2(ix)^\epsilon$

1881: potentials if $\epsilon>0$.

1882: 

1883: \item[(f)] {\em Time evolution and unitarity}. For a Hermitian Hamiltonian the

1884: time-evolution operator $e^{-iHt}$ [see (\ref{e15})] is unitary, and it

1885: automatically preserves the inner product:

1886: \begin{equation}

1887: \Big(\chi(t),\chi(t)\Big)=\Big(\chi(0)e^{iHt},e^{-iHt}\chi(t)\Big)=

1888: \Big(\chi(0),\chi(0)\Big).

1889: \label{e76}

1890: \end{equation}

1891: 

1892: \item[(g)] {\em Observables}. An observable is represented by a linear Hermitian

1893: operator. The outcome of a measurement is one of the {\em real} eigenvalues of

1894: this operator.

1895: 

1896: \item[(h)] {\em Miscellany}. One can study a number of additional topics, such

1897: as the classical and semiclassical limits of the quantum theory, probability

1898: density and currents, perturbative and nonperturbative calculations, and so on.

1899: We do not address these issues in depth in this paper.

1900: 

1901: \end{itemize}

1902: 

1903: \subsection{Recipe for $\cP\cT$-Symmetric Quantum Mechanics}

1904: \label{ss4-2}

1905: 

1906: Let us follow the recipe outlined in Subsec.~\ref{ss4-1} for the case of a

1907: non-Hermitian $\cP\cT$-symmetric Hamiltonian having an unbroken $\cP\cT$

1908: symmetry. For definiteness, we will imagine that the non-Hermitian Hamiltonian

1909: has the form in (\ref{e12}). The novelty here is that we do not know {\em a

1910: priori} the definition of the inner product, as we do in the case of ordinary

1911: Hermitian quantum mechanics. We will have to discover the correct inner product

1912: in the course of our analysis. The inner product is determined by the

1913: Hamiltonian itself, so $\cP\cT$-symmetric quantum mechanics is a kind of

1914: ``bootstrap'' theory. The Hamiltonian operator chooses its own Hilbert space

1915: (and associated inner product) in which it prefers to live!

1916: 

1917: \begin{itemize}

1918: \item[(a)] {\em Eigenfunctions and eigenvalues of $H$}. In Sec.~\ref{s2} we

1919: discussed various techniques for determining the coordinate-space eigenfunctions

1920: and eigenvalues of a non-Hermitian Hamiltonian. We assume here that we have

1921: found the eigenvalues $E_n$ by using either analytical or numerical methods and

1922: that these eigenvalues are all real. [This is equivalent to assuming that the

1923: $\cP\cT$ symmetry of $H$ is unbroken; that is, all eigenfunctions $\psi_n(x)$ of

1924: $H$ are also eigenfunctions of $\cP\cT$.].

1925: 

1926: \item[(b)] {\em Orthogonality of eigenfunctions}. To test the orthogonality of

1927: the eigenfunctions, we must specify an inner product. (A pair of vectors can be

1928: orthogonal with respect to one inner product and not orthogonal with respect to

1929: another inner product.) Since we do not yet know what inner product to use, one

1930: might try to guess an inner product. Arguing by analogy, one might think that

1931: since the inner product in (\ref{e67}) is appropriate for Hermitian Hamiltonians($H=H^\dag$), a good choice for an inner product associated with a $\cP

1932: \cT$-symmetric Hamiltonian ($H=H^{\cP\cT}$) might be

1933: \begin{equation}

1934: (\psi,\phi)\equiv\int_C dx\,[\psi(x)]^{\cP\cT}\phi(x)=\int_C dx\,[\psi(-x)]^*

1935: \phi(x),

1936: \label{e77}

1937: \end{equation}

1938: where $C$ is a contour in the Stokes wedges shown in Fig.~\ref{f2}. With this

1939: inner-product definition one can show by a trivial integration-by-parts argument

1940: using the time-independent Schr\"odinger equation (\ref{e24}) that pairs of

1941: eigenfunctions of $H$ associated with different eigenvalues are orthogonal.

1942: However, this guess for an inner product is not acceptable for formulating a

1943: valid quantum theory because the norm of a state is not necessarily positive.

1944: 

1945: \item[(c)] {\em The $\cC\cP\cT$ inner product}. To construct an inner product

1946: with a positive norm for a complex non-Hermitian Hamiltonian having an {\em

1947: unbroken} $\cP\cT$ symmetry, we will construct a new linear operator $\cC$ that

1948: commutes with both $H$ and $\cP\cT$. Because $\cC$ commutes with the

1949: Hamiltonian, it represents a {\em symmetry} of $H$. We use the symbol $\cC$ to

1950: represent this symmetry because, as we will see, the properties of $\cC$ are

1951: similar to those of the charge conjugation operator in particle physics. The

1952: inner product with respect to $\cC\cP\cT$ conjugation is defined as

1953: \begin{equation}

1954: \langle\psi|\chi\rangle^{\cC\cP\cT}=\int dx\,\psi^{\cC\cP\cT}(x)\chi(x),

1955: \label{e78}

1956: \end{equation}

1957: where $\psi^{\cC\cP\cT}(x)=\int dy\,{\cC}(x,y)\psi^*(-y)$. We will show that

1958: this inner product satisfies the requirements for the quantum theory defined by

1959: $H$ to have a Hilbert space with a positive norm and to be a unitary theory of

1960: quantum mechanics. We will represent the $\cC$ operator as a sum over the

1961: eigenfunctions of $H$, but before doing so we must first show how to normalize

1962: these eigenfunctions.

1963: 

1964: \item[(d)] {\em $\cP\cT$-symmetric normalization of the eigenfunctions and the

1965: strange statement of completeness}. We showed in (\ref{e19}) that the

1966: eigenfunctions $\psi_n(x)$ of $H$ are also eigenfunctions of the $\cP\cT$

1967: operator with eigenvalue $\lambda=e^{i\alpha}$, where $\lambda$ and $\alpha$

1968: depend on $n$. Thus, we can construct $\cP\cT$-normalized eigenfunctions

1969: $\phi_n(x)$ defined by

1970: \begin{equation}

1971: \phi_n(x)\equiv e^{-i\alpha/2}\psi_n(x).

1972: \label{e79}

1973: \end{equation}

1974: By this construction, $\phi_n(x)$ is still an eigenfunction of $H$ and it is

1975: also an eigenfunction of $\cP\cT$ with eigenvalue $1$. One can also

1976: show both numerically and analytically that the algebraic sign of the $\cP\cT$

1977: norm in (\ref{e77}) of $\phi_n(x)$ is $(-1)^n$ for all $n$ and for all values

1978: of $\epsilon>0$ \cite{BBJ1}. Thus, we {\em define} the eigenfunctions so that

1979: their $\cP\cT$ norms are exactly $(-1)^n$:

1980: \begin{equation}

1981: \int_C dx\,[\phi_n(x)]^{\cP\cT}\phi_n(x)=\int_C dx\,[\phi_n(-x)]^*\phi_n(x)=

1982: (-1)^n,

1983: \label{e80}

1984: \end{equation}

1985: where the contour $C$ lies in the Stokes wedges shown in Fig.~\ref{f2}. In terms

1986: of these $\cP\cT$-normalized eigenfunctions there is a simple but unusual

1987: statement of completeness:

1988: \begin{equation}

1989: \sum_{n=0}^\infty(-1)^n\phi_n(x)\phi_n(y)=\delta(x-y).

1990: \label{e81}

1991: \end{equation}

1992: This unusual statement of completeness has been verified both numerically and

1993: analytically to great precision for all $\epsilon>0$ \cite{rr2,rr3} and a

1994: mathematical proof has been given \cite{WWWW1}. It is easy to verify using

1995: (\ref{e80}) that the left side of (\ref{e81}) satisfies the integration rule for

1996: delta functions: $\int dy\,\delta(x-y)\delta(y-z)=\delta(x-z)$.

1997: \end{itemize}

1998: 

1999: \begin{itemize}

2000: \begin{footnotesize}

2001: \item[~~] {\em Example: $\cP\cT$-symmetric normalization of harmonic-oscillator

2002: eigenfunctions.} For the harmonic-oscillator Hamiltonian $H=\p^2+\x^2$, the

2003: eigenfunctions are Gaussians multiplied by Hermite polynomials: $\psi_0(x)=\exp

2004: \left(-\half x^2\right)$, $\psi_1(x)=x\exp\left(-\half x^2\right)$, $\psi_2(x)=

2005: (2x^2-1)\exp\left(-\half x^2\right)$, $\psi_3(x)=(2x^3-3x)\exp\left(-\half x^2

2006: \right)$, and so on. To normalize these eigenfunctions so that they are

2007: also eigenfunctions of the $\cP\cT$ operator with eigenvalue 1, we choose

2008: \begin{eqnarray}

2009: \phi_0(x)&=&a_0\exp\left(-\half x^2\right),\nonumber\\

2010: \phi_1(x)&=&a_1ix\exp\left(-\half x^2\right),\nonumber\\

2011: \psi_2(x)&=&a_2(2x^2-1)\exp\left(-\half x^2\right),\nonumber\\

2012: \psi_3(x)&=&a_3i(2x^3-3x)\exp\left(-\half x^2\right),

2013: \label{e82}

2014: \end{eqnarray}

2015: and so on, where the real numbers $a_n$ are chosen so that the integral in

2016: (\ref{e80}) evaluates to $(-1)^n$ for all $n$. It is easy to verify that if the

2017: eigenfunctions $\phi_n(x)$ are substituted into (\ref{e81}) and the summation

2018: is performed, then the result is the Dirac delta function on the right side of

2019: (\ref{e81}).

2020: \end{footnotesize}

2021: \end{itemize}

2022: 

2023: \begin{itemize}

2024: \item[(e)] {\em Coordinate-space representation of $H$ and $G$ and the spectral

2025: Zeta function}. From the statement of completeness in (\ref{e81}) we can

2026: construct coordinate-space representations of the linear operators. For example,

2027: since the coordinate-space representation of the parity operator is $\cP(x,y)=

2028: \delta(x+y)$, we have

2029: \begin{equation}

2030: \cP(x,y)=\sum_{n=0}^\infty(-1)^n\phi_n(x)\phi_n(-y).

2031: \label{e83}

2032: \end{equation}

2033: We can also construct the coordinate-space representations of the Hamiltonian

2034: and the Green's function,

2035: \begin{eqnarray}

2036: H(x,y)&=&\sum_{n=0}^\infty(-1)^nE_n\phi_n(x)\phi_n(y),\nonumber\\

2037: G(x,y)&=&\sum_{n=0}^\infty(-1)^n\frac{1}{E_n}\phi_n(x)\phi_n(y),

2038: \label{e84}

2039: \end{eqnarray}

2040: and using (\ref{e80}) it is straightforward to show that $G$ is the functional

2041: inverse of $H$: $\int dy\,H(x,y)G(y,z)=\delta(x-z)$. For the class of $\cP

2042: \cT$-symmetric Hamiltonians in (\ref{e12}) this equation takes the form of a

2043: differential equation satisfied by $G(x,y)$:

2044: \begin{equation}

2045: \left(-\frac{d^2}{dx^2}+x^2(ix)^\epsilon\right)G(x,y)=\delta(x-y).

2046: \label{e85}

2047: \end{equation}

2048: Equation (\ref{e85}) can be solved in terms of associated Bessel functions in

2049: each of two regions, $x>y$ and $x<y$. The solutions can then be patched together

2050: at $x=y$ to obtain a closed-form expression for $G(x,y)$ \cite{rr2,rr3}. One

2051: then uses (\ref{e75}) to find an exact formula for the spectral zeta function

2052: for all values of $\epsilon$:

2053: \begin{equation}

2054: \sum_n\frac{1}{E_n}=\left[1+\frac{\cos\left(\frac{3\epsilon\pi}{2\epsilon+8}

2055: \right)\sin\left(\frac{\pi}{4+\epsilon}\right)}{\cos\left(\frac{\epsilon\pi}

2056: {4+2\epsilon}\right)\sin\left(\frac{3\pi}{4+\epsilon}\right)}\right]

2057: \frac{\Gamma\left(\frac{1}{4+\epsilon}\right)\Gamma\left(\frac{2}{4+\epsilon}

2058: \right)\Gamma\left(\frac{\epsilon}{4+\epsilon}\right)}{(4+\epsilon)^\frac{4+

2059: 2\epsilon}{4+\epsilon}\Gamma\left(\frac{1+\epsilon}{4+\epsilon}\right)

2060: \Gamma\left(\frac{2+\epsilon}{4+\epsilon}\right)}.

2061: \label{e86}

2062: \end{equation}

2063: 

2064: \item[(f)] {\em Construction of the $\cC$ operator}. The situation here in which

2065: half of the energy eigenstates have positive norm and half have negative norm

2066: [see (\ref{e80})] is analogous to the problem that Dirac encountered in

2067: formulating the spinor wave equation in relativistic quantum theory \cite{rf17}.

2068: Following Dirac, we attack the problem of an indefinite norm by finding an

2069: interpretation of the negative-norm states. For any Hamiltonian $H$ having an

2070: unbroken $\cP\cT$ symmetry there exists an additional symmetry of $H$ connected

2071: with the fact that there are equal numbers of positive- and negative-norm

2072: states. The linear operator $\cC$ that embodies this symmetry can be represented

2073: in coordinate space as a sum over the $\cP\cT$-normalized eigenfunctions of the

2074: $\cP\cT$-symmetric Hamiltonian in (\ref{e12}):

2075: \begin{equation}

2076: \cC(x,y)=\sum_{n=0}^\infty\phi_n(x)\phi_n(y).

2077: \label{e87}

2078: \end{equation}

2079: Notice that this equation is identical to the statement of completeness in 

2080: (\ref{e81}) except that the factor of $(-1)^n$ is absent. We can use (\ref{e80})

2081: and (\ref{e81}) to verify that the square of $\cC$ is unity ($\cC^2={\bf 1}$):

2082: \begin{equation}

2083: \int dy\,\cC(x,y)\cC(y,z)=\delta(x-z).

2084: \label{e88}

2085: \end{equation}

2086: Thus, the eigenvalues of $\cC$ are $\pm1$. Also, $\cC$ commutes with $H$.

2087: Therefore, since $\cC$ is linear, the eigenstates of $H$ have definite values of

2088: $\cC$. Specifically,

2089: \begin{eqnarray}

2090: \cC\phi_n(x)&=&\int dy\,\cC(x,y)\phi_n(y)\nonumber\\

2091: &=&\sum_{m=0}^\infty\phi_m(x)\int dy\,\phi_m(y)\phi_n(y)=(-1)^n\phi_n(x).

2092: \label{e89}

2093: \end{eqnarray}

2094: This new operator $\cC$ resembles the charge-conjugation operator in quantum

2095: field theory. However, the precise meaning of $\cC$ is that it represents the

2096: measurement of the sign of the $\cP\cT$ norm in (\ref{e80}) of an eigenstate.

2097: 

2098: The operators $\cP$ and $\cC$ are distinct square roots of the unity operator

2099: $\delta(x-y)$. That is, $\cP^2=\cC^2={\bf 1}$, but $\cP\neq\cC$ because $\cP$ is

2100: real, while $\cC$ is complex. The parity operator in coordinate space is

2101: explicitly real ($\cP(x,y)=\delta(x+y)$), while the operator $\cC(x,y)$ is

2102: complex because it is a sum of products of complex functions. The two operators

2103: $\cP$ and $\cC$ do not commute. However, $\cC$ {\em does} commute with $\cP\cT$.

2104: 

2105: \item[(g)] {\em Positive norm and unitarity in $\cP\cT$-symmetric quantum

2106: mechanics}. Having constructed the operator $\cC$, we can now use the new $\cC

2107: \cP\cT$ inner-product defined in (\ref{e78}). Like the $\cP\cT$ inner product,

2108: this new inner product is phase independent. Also, because the time-evolution

2109: operator (as in ordinary quantum mechanics) is $e^{-iHt}$ and because $H$

2110: commutes with $\cP\cT$ and with $\cC\cP\cT$, both the $\cP\cT$ inner product and

2111: the $\cC\cP\cT$ inner product remain time independent as the states evolve.

2112: However, unlike the $\cP\cT$ inner product, the $\cC\cP\cT$ inner product is

2113: {\em positive definite} because $\cC$ contributes a factor of $-1$ when it acts

2114: on states with negative $\cP\cT$ norm. In terms of the $\cC\cP\cT$ conjugate,

2115: the completeness condition reads

2116: \begin{equation}

2117: \sum_{n=0}^\infty\phi_n(x)[\cC\cP\cT\phi_n(y)]=\delta(x-y).

2118: \label{e90}

2119: \end{equation}

2120: \end{itemize}

2121: 

2122: \subsection{Comparison of Hermitian and $\cP\cT$-Symmetric Quantum Theories}

2123: \label{ss4-3}

2124: 

2125: We have shown in Subsecs.~\ref{ss4-1} and \ref{ss4-2} how to construct quantum

2126: theories based on Hermitian and on non-Hermitian Hamiltonians. In the

2127: formulation of a conventional quantum theory defined by a Hermitian Hamiltonian,

2128: the Hilbert space of physical states is specified even before the Hamiltonian is

2129: known. The inner product (\ref{e67}) in this vector space is defined with

2130: respect to Dirac Hermitian conjugation (complex conjugate and transpose). The

2131: Hamiltonian is then chosen and the eigenvectors and eigenvalues of the

2132: Hamiltonian are determined. In contrast, the inner product for a quantum theory

2133: defined by a non-Hermitian $\cP\cT$-symmetric Hamiltonian depends on the

2134: Hamiltonian itself and thus is determined {\em dynamically}. One must solve for

2135: the eigenstates of $H$ before knowing the Hilbert space and the associated inner

2136: product of the theory because the $\cC$ operator is defined and constructed in

2137: terms of the eigenstates of the Hamiltonian. The Hilbert space, which consists

2138: of all complex linear combinations of the eigenstates of $H$, and the $\cC\cP

2139: \cT$ inner product are determined by these eigenstates.

2140: 

2141: The operator $\cC$ does not exist as a distinct entity in ordinary Hermitian

2142: quantum mechanics. Indeed, if we allow the parameter $\epsilon$ in (\ref{e12})

2143: to tend to 0, the operator $\cC$ in this limit becomes identical to $\cP$ and

2144: the $\cC\cP\cT$ operator becomes $\cT$, which performs complex conjugation.

2145: Hence, the inner product defined with respect to $\cC\cP\cT$ conjugation reduces

2146: to the inner product of conventional quantum mechanics and (\ref{e81}) reduces

2147: to the usual statement of completeness $\sum_n\phi_n(x)\phi_n^*(y)=\delta(x-y)$.

2148: 

2149: The $\cC\cP\cT$ inner product is independent of the choice of integration

2150: contour $C$ as long as $C$ lies inside the asymptotic wedges associated with the

2151: boundary conditions for the eigenvalue problem. In ordinary quantum mechanics,

2152: where the positive-definite inner product has the form $\int dx\,f^*(x)g(x)$,

2153: the integral must be taken along the real axis and the path of integration

2154: cannot be deformed into the complex plane because the integrand is not analytic.

2155: The $\cP\cT$ inner product shares with the $\cC\cP\cT$ inner-product the

2156: advantage of analyticity and path independence, but it suffers from

2157: nonpositivity. It is surprising that we can construct a positive-definite metric

2158: by using $\cC\cP\cT$ conjugation without disturbing the path independence of the

2159: inner-product integral.

2160: 

2161: Time evolution is expressed by the operator $e^{-iHt}$ whether the theory is

2162: determined by a $\cP\cT$-symmetric Hamiltonian or just an ordinary Hermitian

2163: Hamiltonian. To establish unitarity we must show that as a state vector evolves,

2164: its norm does not change in time. If $\psi_0(x)$ is any given initial vector

2165: belonging to the Hilbert space spanned by the energy eigenstates, then it

2166: evolves into the state $\psi_t(x)$ at time $t$ according to $\psi_t(x)=e^{-iHt}

2167: \psi_0(x)$. With respect to the $\cC\cP\cT$ inner product the norm of $\psi_t

2168: (x)$ does not change in time because $H$ commutes with $\cC\cP\cT$.

2169: 

2170: \subsection{Observables}

2171: \label{ss4-4}

2172: 

2173: How do we represent an observable in $\cP\cT$-symmetric quantum mechanics?

2174: Recall that in ordinary quantum mechanics the condition for a linear operator

2175: $A$ to be an observable is that $A=A^\dagger$. This condition guarantees that

2176: the expectation value of $A$ in a state is real. Operators in the Heisenberg

2177: picture evolve in time according to $A(t)=e^{iHt}A(0)e^{-iHt}$, so this

2178: Hermiticity condition is maintained in time. In $\cP\cT$-symmetric quantum

2179: mechanics the equivalent condition is that at time $t=0$ the operator $A$ must

2180: obey the condition $A^{\rm T}=\cC\cP\cT A\,\cC\cP\cT$, where $A^{\rm T}$ is the

2181: {\em transpose} of $A$ \cite{BBJ1}. If this condition holds at $t=0$, then it

2182: will continue to hold for all time because we have assumed that $H$ is symmetric

2183: ($H=H^{\rm T}$). This condition also guarantees that the expectation value of

2184: $A$ in any state is real.\footnote{The requirement given here for $A$ to be an

2185: observable involves matrix transposition. This condition is more restrictive

2186: than is necessary and it has been generalized by Mostafazadeh. See

2187: Refs.~\cite{Z4,P7x,Jobs}. Note that if the matrix transpose symmetry condition

2188: on the Hamiltonian is removed we must introduce a biorthogonal basis. See

2189: Refs.~\cite{WWWW1} and \cite{BIOX}.}

2190: 

2191: The operator $\cC$ itself satisfies this requirement, so it is an observable.

2192: The Hamiltonian is also an observable. However, the $\x$ and $\p$ operators are

2193: not observables. Indeed, the expectation value of $\x$ in the ground state is a

2194: negative imaginary number. Thus, there is no position operator in $\cP

2195: \cT$-symmetric quantum mechanics. In this sense $\cP\cT$-symmetric quantum

2196: mechanics is similar to fermionic quantum field theories. In such theories the

2197: fermion field corresponds to the $\x$ operator. The fermion field is complex and

2198: does not have a classical limit. One cannot measure the position of an electron;

2199: one can only measure the position of the {\em charge} or the {\em energy} of the

2200: electron!

2201: 

2202: One can see why the expectation of the $x$ operator in $\cP\cT$-symmetric

2203: quantum mechanics is a negative imaginary number by examining a classical

2204: trajectory like that shown in Fig.~\ref{f15}. Note that this classical

2205: trajectory has left-right ($\cP\cT$) symmetry, but not up-down symmetry. Also,

2206: the classical paths favor (spend more time in) the lower-half complex-$x$ plane.

2207: Thus, the average classical position is a negative imaginary number. Just as the

2208: classical particle moves about in the complex plane, the quantum probability

2209: current flows about in the complex plane. It may be that the correct

2210: interpretation is to view $\cP\cT$-symmetric quantum mechanics as describing the

2211: interaction of extended, rather than pointlike objects.

2212: 

2213: \subsection{Pseudo-Hermiticity and $\cP\cT$ Symmetry}

2214: \label{ss4-5}

2215: 

2216: The thesis of this paper -- replacing the mathematical condition of Hermiticity

2217: by the more physical condition of $\cP\cT$ symmetry -- can be placed in a more

2218: general mathematical context known as pseudo-Hermiticity. A linear operator $A$

2219: is {\em pseudo-Hermitian} if there is a Hermitian operator $\eta$ such that

2220: \begin{equation}

2221: A^\dag=\eta A\eta.

2222: \label{e91}

2223: \end{equation}

2224: The operator $\eta$ is often called an {\em intertwining} operator. The

2225: condition in (\ref{e91}) obviously reduces to ordinary Hermiticity when the

2226: intertwining operator $\eta$ is the identity ${\bf 1}$. The concept of

2227: pseudo-Hermiticity was introduced in the 1940s by Dirac and Pauli, and later

2228: discussed by Lee, Wick, and Sudarshan, who were trying to resolve the problems

2229: that arise in quantizing electrodynamics and other quantum field theories in

2230: which negative-norm states states appear as a consequence of renormalization

2231: \cite{P1,P2,P3,P4,P5}. These problems are illustrated very clearly by the Lee

2232: model, which is discussed in Subsec.~\ref{ss8-3}.

2233: 

2234: Mostafazadeh observed that because the parity operator $\cP$ is Hermitian, it

2235: may be used as an intertwining operator. The class of Hamiltonians $H$ in

2236: (\ref{e12}) is pseudo-Hermitian because the parity operator $\cP$ changes the

2237: sign of $\x$ while Dirac Hermitian conjugation changes the sign of $i$

2238: \cite{M1,P6,P7,P7x}:

2239: \begin{equation}

2240: H^\dag=\cP H\cP.

2241: \label{e92}

2242: \end{equation}

2243: For some other references on the generalization of $\cP\cT$ symmetry to

2244: pseudo-Hermiticity see \cite{P8}.

2245: 

2246: \section{Illustrative $2\times2$ Matrix Example of a $\cP\cT$-Symmetric

2247: Hamiltonian}

2248: \label{s5}

2249: 

2250: It is always useful to study exactly solvable models when one is trying to

2251: understand a formal procedure like that discussed in Sec.~\ref{s4}. One

2252: exactly solvable model, which has been used in many papers on $\cP\cT$ symmetry,

2253: is due to Swanson \cite{Swan}. This model is exactly solvable because it is

2254: quadratic in $\x$ and $\p$. In this section we use an even simpler model to

2255: illustrate the construction of a quantum theory described by a $\cP

2256: \cT$-symmetric Hamiltonian. We consider the elementary $2\times2$ Hamiltonian

2257: matrix

2258: \begin{equation}

2259: H=\left(\begin{array}{cc} re^{i\theta} & s \cr s & re^{-i\theta}

2260: \end{array}\right),

2261: \label{e93}

2262: \end{equation}

2263: where the three parameters $r$, $s$, and $\theta$ are real \cite{AJP}. The

2264: Hamiltonian in (\ref{e93}) is not Hermitian, but it is easy to see that it is

2265: $\cP\cT$ symmetric, where we define the parity operator as

2266: \begin{equation}

2267: \cP=\left(\begin{array}{cc} 0 & 1 \cr 1 & 0\end{array}\right)

2268: \label{e94}

2269: \end{equation}

2270: and we define the operator $\cT$ to perform complex conjugation.

2271: 

2272: As a first step in analyzing the Hamiltonian (\ref{e93}), we calculate its two

2273: eigenvalues:

2274: \begin{equation}

2275: E_\pm=r\cos\theta\pm(s^2-r^2\sin^2\theta)^{1/2}.

2276: \label{e95}

2277: \end{equation}

2278: There are clearly two parametric regions to consider, one for which the square

2279: root in (\ref{e95}) is real and the other for which it is imaginary. When $s^2<

2280: r^2\sin^2\theta$, the energy eigenvalues form a complex conjugate pair. This is

2281: the region of broken $\cP\cT$ symmetry. On the other hand, when $s^2\geq r^2

2282: \sin^2\theta$, the eigenvalues $\varepsilon_\pm=r\cos\theta\pm(s^2-r^2\sin^2

2283: \theta)^{1/2}$ are real. This is the region of unbroken $\cP\cT$ symmetry. In

2284: the unbroken region the simultaneous eigenstates of the operators $H$ and

2285: $\cP\cT$ are

2286: \begin{equation}

2287: |E_+\rangle=\frac{1}{\sqrt{2\cos\alpha}}

2288: \left(\begin{array}{c} e^{i\alpha/2}\cr e^{-i\alpha/2}\end{array}\right)\quad

2289: {\rm and}\quad|E_-\rangle=\frac{i}{\sqrt{2\cos\alpha}}\left(\begin{array}{c}

2290: e^{-i\alpha/2}\cr-e^{i\alpha/2}\end{array}\right),

2291: \label{e96}

2292: \end{equation}

2293: where

2294: \begin{equation}

2295: \sin\alpha =\frac{r}{s}\,\sin\theta.

2296: \label{e97}

2297: \end{equation}

2298: The $\cP\cT$ inner product gives

2299: \begin{equation}

2300: (E_{\pm},E_{\pm})=\pm1\quad{\rm and}\quad (E_{\pm},E_{\mp})=0,

2301: \label{e98}

2302: \end{equation}

2303: where $(u,v)=(\cP\cT u)\cdot v$. With respect to the $\cP\cT$ inner product, the

2304: vector space spanned by the energy eigenstates has a metric of signature $(+,-

2305: )$. If the condition $s^2>r^2\sin^2\theta$ for an unbroken $\cP\cT$ symmetry is

2306: violated, the states (\ref{e96}) are no longer eigenstates of $\cP\cT$ because

2307: $\alpha$ becomes imaginary. When $\cP\cT$ symmetry is broken, the $\cP\cT$ norm

2308: of the energy eigenstate vanishes.

2309: 

2310: Next, we construct the operator $\cC$ using (\ref{e87}):

2311: \begin{equation}

2312: \cC=\frac{1}{\cos\alpha}\left(\begin{array}{cc} i\sin\alpha & 1 \cr 1 & -i\sin

2313: \alpha\end{array}\right).

2314: \label{e99}

2315: \end{equation}

2316: Note that $\cC$ is distinct from $H$ and $\cP$ and it has the key property that

2317: \begin{equation}

2318: \cC|E_{\pm}\rangle=\pm|E_{\pm}\rangle.

2319: \label{e100}

2320: \end{equation}

2321: The operator $\cC$ commutes with $H$ and satisfies $\cC^2=1$. The eigenvalues of

2322: $\cC$ are precisely the signs of the $\cP\cT$ norms of the corresponding

2323: eigenstates. Using the operator $\cC$ we construct the new inner product

2324: structure $\langle u|v\rangle= (\cC\cP\cT u)\cdot v$. This inner product is

2325: positive definite because $\langle E_{\pm}|E_{\pm}\rangle=1$. Thus, the

2326: two-dimensional Hilbert space spanned by $|E_\pm\rangle$, with inner product

2327: $\langle\cdot|\cdot\rangle$, has signature $(+,+)$.

2328: 

2329: Finally, we prove that the $\cC\cP\cT$ norm of any vector is positive. For the

2330: % arbitrary vector $\psi=\left(\begin{array}{c}a\cr b\end{array}\right)$, where 

2331: arbitrary vector $\psi=\left({a\atop b}\right)$, where 

2332: $a$ and $b$ are any complex numbers, we see that

2333: \begin{eqnarray}

2334: &&\cT\psi=\left(\begin{array}{c}a^*\cr b^*\end{array}\right),\qquad

2335: \cP\cT\psi=\left(\begin{array}{c} b^*\cr a^*\end{array}\right),\nonumber\\

2336: &&\cC\cP\cT\psi=\frac{1}{\cos\alpha}\,\left(\begin{array}{c}a^*+ib^*\sin\alpha

2337: \cr b^*-ia^*\sin\alpha\end{array}\right).

2338: \label{e101}

2339: \end{eqnarray}

2340: Thus, $\langle\psi|\psi\rangle=(\cC\cP\cT\psi)\cdot\psi=\frac{1}{\cos\alpha}

2341: [a^*a+b^*b+i(b^*b-a^*a)\sin\alpha]$. Now let $a=x+iy$ and $b=u+iv$, where $x$,

2342: $y$, $u$, and $v$ are real. Then

2343: \begin{equation}

2344: \langle\psi|\psi\rangle=\frac{1}{\cos\alpha}\left(x^2+v^2+2xv\sin\alpha

2345: +y^2+u^2-2yu\sin\alpha\right),

2346: \label{e102}

2347: \end{equation}

2348: which is explicitly positive and vanishes only if $x=y=u=v=0$.

2349: 

2350: Since $\langle u|$ denotes the $\cC\cP\cT$-conjugate of $|u\rangle$, the

2351: completeness condition reads

2352: \begin{equation}

2353: |E_+\rangle\langle E_+|+|E_-\rangle\langle E_-|=\left(\begin{array}{cc} 1 & 0

2354: \cr 0 & 1\end{array}\right).

2355: \label{e103}

2356: \end{equation}

2357: Furthermore, using the $\cC\cP\cT$ conjugate $\langle E_\pm|$, we can represent

2358: $\cC$ as

2359: \begin{equation}

2360: \cC=|E_+\rangle\langle E_+|-|E_-\rangle\langle E_-|.

2361: \label{e104}

2362: \end{equation}

2363: 

2364: In the limit $\theta\to0$, the Hamiltonian (\ref{e93}) for this two-state system

2365: becomes Hermitian and $\cC$ reduces to the parity operator $\cP$. Thus, $\cC\cP

2366: \cT$ invariance reduces to the standard condition of Hermiticity for a symmetric

2367: matrix; namely, $H=H^*$.

2368: 

2369: \section{Calculation of the $\cC$ Operator in Quantum Mechanics}

2370: \label{s6}

2371: 

2372: The distinguishing feature of $\cP\cT$-symmetric quantum mechanics is the $\cC$

2373: operator. In ordinary Hermitian quantum mechanics there is no such operator.

2374: Only a non-Hermitian $\cP\cT$-symmetric Hamiltonian possesses a $\cC$ operator

2375: distinct from the parity operator $\cP$. Indeed, if we were to sum the series in

2376: (\ref{e87}) for a $\cP\cT$-symmetric Hermitian Hamiltonian, the result would be

2377: $\cP$, which in coordinate space is $\delta(x+y)$. [See (\ref{e83}).]

2378: 

2379: While the $\cC$ operator is represented formally in (\ref{e87}) as an infinite

2380: series, it is not easy to evaluate the sum of this series. Calculating $\cC$ by

2381: direct brute-force evaluation of the sum in (\ref{e87}) is not easy in quantum

2382: mechanics because it is necessary to find all the eigenfunctions $\phi_n(x)$ of

2383: $H$. Furthermore, such a procedure cannot be used in quantum field theory

2384: because in field theory there is no simple analog of the Schr\"odinger

2385: eigenvalue differential equation and its associated coordinate-space

2386: eigenfunctions.

2387: 

2388: The first attempt to calculate $\cC$ relied on a perturbative approach

2389: \cite{BMW}. In this paper the $\cP\cT$-symmetric Hamiltonian

2390: \begin{equation}

2391: H=\half\p^2+\half\x^2+i\epsilon\x^3

2392: \label{e105}

2393: \end{equation}

2394: is considered, where $\epsilon$ is treated as a small real parameter. When

2395: $\epsilon=0$, the Hamiltonian reduces to the Hamiltonian for the quantum

2396: harmonic oscillator, all of whose eigenfunctions can be calculated exactly.

2397: Thus, it is possible to express the eigenfunctions of $H$ in (\ref{e105}) in the

2398: form of perturbation series in powers of $\epsilon$. For each of the

2399: eigenfunctions the first few terms in the perturbation series were calculated.

2400: These perturbation series were then substituted into (\ref{e87}) and the

2401: summation over the $n$th eigenfunction was performed. The result is a

2402: perturbation-series expansion of the $\cC$ operator. This calculation is long

2403: and tedious and the final result is quite complicated. However, the calculation

2404: turned out to be of great value because while the final answer is complicated,

2405: it was discovered that the answer in coordinate space simplified dramatically if

2406: the $\cC$ operator is written as the exponential of a derivative operator $Q$

2407: multiplying the parity operator $\cP$:

2408: \begin{equation}

2409: \cC(x,y)=\exp\left[Q\left(x,-i\frac{d}{dx}\right)\right]\delta(x+y).

2410: \label{e106}

2411: \end{equation}

2412: The simplification that occurs when $\cC$ is written in the form $\cC=e^Q\cP$ is

2413: that while the expression for $\cC$ is a series in all positive integer powers

2414: of $\epsilon$, $Q$ is a series in {\em odd} powers of $\epsilon$ only. Since $Q$

2415: is a series in odd powers of $\epsilon$, in the limit $\epsilon\to0$ the

2416: function $Q$ vanishes. Thus, in this limit the $\cC$ operator tends to the

2417: parity operator $\cP$.

2418: 

2419: The expression in (\ref{e106}) need not be limited to coordinate space. A more

2420: general way to represent the $\cC$ operator is to express it generically in

2421: terms of the fundamental dynamical operators $\x$ and $\p$:

2422: \begin{equation}

2423: \cC=e^{Q(\x,\p)}\cP.

2424: \label{e107}

2425: \end{equation}

2426: Written in this form, $Q$ is a {\em real} function of its two variables. By

2427: seeking the $\cC$ operator in the form (\ref{e107}) we will be able to devise

2428: powerful analytic tools for calculating it.

2429: 

2430: We illustrate the representation $\cC=e^Q\cP$ by using two elementary

2431: Hamiltonians. First, consider the shifted harmonic oscillator $H=\half\p^2+\half

2432: \x^2+i\epsilon\x$. This Hamiltonian has an unbroken $\cP\cT$ symmetry for all

2433: real $\epsilon$. Its eigenvalues $E_n=n+\half+\half\epsilon^2$ are all real. The

2434: exact formula for $\cC$ for this theory is given exactly by $\cC=e^Q\cP$, where

2435: \begin{equation}

2436: Q=-\epsilon\p.

2437: \label{e108}

2438: \end{equation}

2439: Note that in the limit $\epsilon\to0$, where the Hamiltonian becomes Hermitian,

2440: $\cC$ becomes identical with $\cP$.

2441: 

2442: As a second example, consider the non-Hermitian $2\times2$ matrix Hamiltonian

2443: (\ref{e93}). The associated $\cC$ operator in (\ref{e99}) can be rewritten in

2444: the form $\cC=e^Q\cP$, where

2445: \begin{equation}

2446: Q=\half\sigma_2\ln\left(\frac{1-\sin\alpha}{1+\sin\alpha}\right)

2447: \label{e109}

2448: \end{equation}

2449: and

2450: \begin{equation}

2451: \sigma_2=\left(\begin{array}{cc}0 & -i\cr i & 0\end{array}\right)

2452: \label{e110}

2453: \end{equation}

2454: is the Pauli sigma matrix. Again, observe that in the limit $\theta\to0$, where

2455: the Hamiltonian becomes Hermitian, the $\cC$ operator becomes identical with

2456: $\cP$.

2457: 

2458: \subsection{Algebraic Equations Satisfied by the $\cC$ Operator}

2459: \label{ss6-1}

2460: 

2461: Fortunately, there is a relatively easy algebraic way to calculate the $\cC$

2462: operator, and the procedure circumvents the difficult problem of evaluating the

2463: sum in (\ref{e87}). As a result, the technique readily generalizes from quantum

2464: mechanics to quantum field theory. In this subsection we show how to use this

2465: technique to calculate $\cC$ for the $\cP\cT$-symmetric Hamiltonian in

2466: (\ref{e105}). We explain how to calculate $\cC$ perturbatively to high order in

2467: powers of $\epsilon$ for this cubic Hamiltonian. Calculating $\cC$ for other

2468: kinds of interactions is more difficult and may require the use of semiclassical

2469: approximations \cite{rf20.5}.

2470: 

2471: To calculate $\cC$ we make use of its three crucial algebraic properties. First,

2472: $\cC$ commutes with the space-time reflection operator $\cP\cT$,

2473: \begin{eqnarray}

2474: [\cC,\cP\cT]=0,

2475: \label{e111}

2476: \end{eqnarray}

2477: although $\cC$ does not, in general, commute with $\cP$ or $\cT$ separately.

2478: Second, the square of $\cC$ is the identity,

2479: \begin{eqnarray}

2480: \cC^2={\bf 1},

2481: \label{e112}

2482: \end{eqnarray}

2483: which allows us to interpret $\cC$ as a reflection operator. Third, $\cC$

2484: commutes with $H$,

2485: \begin{eqnarray}

2486: [\cC,H]=0,

2487: \label{e113}

2488: \end{eqnarray}

2489: and thus is time independent. To summarize, $\cC$ is a time-independent $\cP

2490: \cT$-symmetric reflection operator.

2491: 

2492: The procedure for calculating $\cC$ is simply to substitute the generic operator

2493: representation in (\ref{e107}) into the three algebraic equations (\ref{e111})

2494: -- (\ref{e113}) in turn and to solve the resulting equations for the function

2495: $Q$. First, we substitute (\ref{e107}) into the condition (\ref{e111}) to obtain

2496: \begin{equation}

2497: e^{Q(\x,\p)}=\cP\cT e^{Q(\x,\p)}\cP\cT=e^{Q(-\x,\p)},

2498: \label{e114}

2499: \end{equation}

2500: from which we conclude that $Q(\x,\p)$ is an {\em even} function of $\x$.

2501: 

2502: Second, we substitute (\ref{e107}) into the condition (\ref{e112}) and find that

2503: \begin{equation}

2504: e^{Q(\x,\p)}\cP e^{Q(\x,\p)}\cP=e^{Q(\x,\p)}e^{Q(-\x,-\p)}=1,

2505: \label{e115}

2506: \end{equation}

2507: which implies that $Q(\x,\p)=-Q(-\x,-\p)$. Since we already know that $Q(\x,\p)$

2508: is an even function of $\x$, we conclude that it is also an {\em odd} function

2509: of $\p$.

2510: 

2511: The remaining condition (\ref{e113}) to be imposed is that the operator $\cC$

2512: commutes with $H$. While the first two conditions are, in effect, {\em

2513: kinematic} conditions on $Q$ that are generally true for any Hamiltonian,

2514: condition (\ref{e113}) is equivalent to imposing the specific dynamics of the

2515: particular Hamiltonian that defines the quantum theory. Substituting $\cC=e^{Q(

2516: \x,\p)}\cP$ into (\ref{e113}), we get

2517: \begin{equation}

2518: e^{Q(\x,\p)}[\cP,H]+[e^{Q(\x,\p)},H]\cP=0.

2519: \label{e116}

2520: \end{equation}

2521: This equation is difficult to solve in general, and to do so we must use

2522: perturbative methods, as we explain in the next subsection.

2523: 

2524: \subsection{Perturbative Calculation of $\cC$.}

2525: \label{ss6-2}

2526: 

2527: To solve (\ref{e116}) for the Hamiltonian in (\ref{e105}), we express this

2528: Hamiltonian in the form $H=H_0+\epsilon H_1$. Here, $H_0$ is the

2529: harmonic-oscillator Hamiltonian $H_0=\half\p^2+\half\x^2$, which commutes with

2530: the parity operator $\cP$, and $H_1=i\x^3$, which {\em anticommutes} with $\cP$.

2531: Thus, the condition (\ref{e116}) becomes

2532: \begin{eqnarray}

2533: 2\epsilon e^{Q(\x,\p)}H_1=[e^{Q(\x,\p)},H].

2534: \label{e117}

2535: \end{eqnarray}

2536: 

2537: We expand the operator $Q(\x,\p)$ as a perturbation series in odd powers of

2538: $\epsilon$:

2539: \begin{eqnarray}

2540: Q(\x,\p)=\epsilon Q_1(\x,\p)+\epsilon^3Q_3(\x,\p)+\epsilon^5Q_5(\x,\p)+\cdots\,.

2541: \label{e118}

2542: \end{eqnarray}

2543: Substituting the expansion in (\ref{e118}) into the exponential $e^{Q(\x,\p)}$,

2544: we get after some algebra a sequence of equations that can be solved 

2545: systematically for the operator-valued functions $Q_n(\x,\p)$ $(n=1,3,5,\ldots)$

2546: subject to the symmetry constraints that ensure the conditions (\ref{e111}) and 

2547: (\ref{e112}). The first three of these equations are

2548: \begin{eqnarray}

2549: \left[H_0,Q_1\right] &=& -2H_1,\nonumber\\

2550: \left[H_0,Q_3\right] &=& -{\textstyle\frac{1}{6}}[Q_1,[Q_1,H_1]],\nonumber\\

2551: \left[H_0,Q_5\right] &=& {\textstyle\frac{1}{360}}[Q_1,[Q_1,[Q_1,[Q_1,H_1]]]]

2552: -{\textstyle\frac{1}{6}}[Q_1,[Q_3,H_1]]\nonumber\\

2553: &&\qquad+{\textstyle\frac{1}{6}}[Q_3,[Q_1,H_1]].

2554: \label{e119}

2555: \end{eqnarray}

2556: 

2557: Let us solve these equations for the Hamiltonian in (\ref{e105}), for which

2558: $H_0=\half\p^2+\half\x^2$ and $H_1=i\x^3$. The procedure is to substitute the

2559: most general polynomial form for $Q_n$ using arbitrary coefficients and then to

2560: solve for these coefficients. For example, to solve $\left[H_0,Q_1\right]=-2i

2561: \x^3$, the first of the equations in (\ref{e119}), we take as an {\em ansatz}

2562: for $Q_1$ the most general Hermitian cubic polynomial that is even in $\x$ and

2563: odd in $p$:

2564: \begin{eqnarray}

2565: Q_1(\x,\p)=M\p^3+N\x\p\x,

2566: \label{e120}

2567: \end{eqnarray}

2568: where $M$ and $N$ are numerical coefficients to be determined. The operator

2569: equation for $Q_1$ is satisfied if $M=-\frac{4}{3}$ and $N=-2$.

2570: 

2571: It is algebraically tedious but completely straightforward to continue this

2572: process. In order to present the solutions for $Q_n(\x,\p)$ ($n>1$), it is

2573: convenient to introduce the following notation: Let $S_{m,n}$ represent the {\em

2574: totally symmetrized} sum over all terms containing $m$ factors of $\p$ and $n$

2575: factors of $\x$. For example,

2576: \begin{equation}

2577: S_{0,0}=1,~S_{0,3}=\x^3,~S_{1,1}=\half\left(\x\p+\p\x\right),~

2578: S_{1,2}={\textstyle\frac{1}{3}}\left(\x^2\p+\x\p\x+\p\x^2\right),

2579: \label{e121}

2580: \end{equation}

2581: and so on. (The properties of the operators $S_{m,n}$ are summarized in

2582: Ref.~\cite{rf23}.)

2583: 

2584: In terms of the symmetrized operators $S_{m,n}$ the first three functions

2585: $Q_{2n+1}$ are

2586: \begin{eqnarray}

2587: Q_1 &=& -{\textstyle\frac{4}{3}}\p^3-2S_{1,2},\nonumber\\

2588: Q_3 &=& {\textstyle\frac{128}{15}}\p^5+{\textstyle\frac{40}{3}}S_{3,2}+8S_{1,4}

2589: -12\p,\nonumber\\

2590: Q_5 &=& -{\textstyle\frac{320}{3}}\p^7-{\textstyle\frac{544}{3}}S_{5,2}-

2591: {\textstyle\frac{512}{3}}S_{3,4}-64S_{1,6}+{\textstyle\frac{24\,736}{45}}\p^3

2592: +{\textstyle\frac{6\,368}{15}}S_{1,2}.

2593: \label{e122}

2594: \end{eqnarray}

2595: Together, (\ref{e107}), (\ref{e118}), and (\ref{e122}) represent an explicit

2596: perturbative expansion of $\cC$ in terms of the operators $\x$ and $\p$, correct

2597: to order $\epsilon^6$.

2598: 

2599: To summarize, using the {\em ansatz} (\ref{e107}) we can calculate $\cC$ to high

2600: order in perturbation theory. We are able to perform this calculation because

2601: this {\em ansatz} obviates the necessity of calculating the $\cP\cT$-normalized

2602: wave functions $\phi_n(x)$. We show how use these same techniques for quantum

2603: field theory in Sec.~\ref{s8}.

2604: 

2605: \subsection{Perturbative Calculation of $\cC$ for Other Quantum-Mechanical

2606: Hamiltonians}

2607: \label{ss6-3}

2608: 

2609: The $\cC$ operator has been calculated perturbatively for a variety of

2610: quantum-mechanical models. For example, let us consider first the case of the

2611: Hamiltonian

2612: \begin{equation}

2613: H=\half\left(\p^2+\q^2\right)+\half\left(\x^2+\y^2\right)+i\epsilon \x^2\y,

2614: \label{e123}

2615: \end{equation}

2616: which has two degrees of freedom. The energy levels of this complex

2617: H\'enon-Heiles theory were studied in Ref.~\cite{BDMS}. The $\cC$ operator for 

2618: this Hamiltonian was calculated in \cite{BBRR,BBJ2,BBJ3}. The perturbative

2619: result for the $Q=Q_1\epsilon+Q_3\epsilon^3+\ldots$ operator is

2620: \begin{eqnarray}

2621: Q_1(\x,\y,\p,\q) &=& -{\textstyle\frac{4}{3}}\p^2\q-{\textstyle\frac{1}{3}}

2622: S_{1,1}y-{\textstyle\frac{2}{3}}\x^2\q,\nonumber\\

2623: Q_3(\x,\y,\p,\q) &=& {\textstyle\frac{512}{405}}\p^2\q^3+{\textstyle\frac{512}

2624: {405}}\p^4\q+{\textstyle\frac{1088}{405}}S_{1,1}T_{2,1}\nonumber\\

2625: && -{\textstyle\frac{256}{405}}\p^2T_{1,2}+{\textstyle\frac{512}{405}}S_{3,1}\y

2626: +{\textstyle\frac{288}{405}}S_{2,2}\q -{\textstyle\frac{32}{405}}\x^2\q^3

2627: \nonumber\\

2628: &&+{\textstyle\frac{736}{405}}\x^2T_{1,2}-{\textstyle\frac{256}{405}}S_{1,1}\y^3

2629: +{\textstyle\frac{608}{405}}S_{1,3}\y-{\textstyle\frac{128}{405}}\x^4\q

2630: -{\textstyle\frac{8}{9}}q\,,

2631: \label{e124}

2632: \end{eqnarray}

2633: where $T_{m,n}$ represents a totally symmetric product of $m$ factors of $\q$

2634: and $n$ factors of $\y$.

2635: 

2636: For the Hamiltonian

2637: \begin{eqnarray}

2638: H=\half\left(\p^2+\q^2+\rr^2\right)+\half\left(\x^2+\y^2+\z^2\right)+i

2639: \epsilon\x\y\z,

2640: \label{e125}

2641: \end{eqnarray}

2642: which has three degrees of freedom, we have \cite{BBRR,BBJ2,BBJ3}

2643: \begin{eqnarray}

2644: Q_1(\x,\y,\z,\p,\q,\rr) &=& -{\textstyle\frac{2}{3}}(\y\z\p+\x\z\q+\x\y\rr)

2645: -{\textstyle\frac{4}{3}}\p\q\rr,\nonumber\\

2646: Q_3(\x,\y,\z,\p,\q,\rr) &=& {\textstyle\frac{128}{405}}\left(\p^3\q\rr+\q^3\p\rr

2647: +\rr^3\q\p\right)\nonumber\\

2648: &&+{\textstyle\frac{136}{405}}[\p\x\p(\y\rr+\z\q)+\q\y\q(\x\rr+\z\p)+\rr\z\rr

2649: (\x\q+\y\p)]\nonumber\\

2650: &&-{\textstyle\frac{64}{405}}(\x\p\x\q\rr+\y\q\y\p\rr+\z\rr\z\p\q)+

2651: {\textstyle\frac{184}{405}}(\x\p\x\y\z+\y\q\y\x\z\nonumber\\

2652: &&+\z\rr\z\x\y)-{\textstyle\frac{32}{405}}\big[\x^3(\y\rr+\z\q)+\y^3(\x\rr+\z\p)

2653: \nonumber\\

2654: &&+\z^3(\x\q+\y\p)\big]-{\textstyle\frac{8}{405}}\left(\p^3\y\z+\q^3\x\z+\rr^3

2655: \x\y\right).

2656: \label{e126}

2657: \end{eqnarray}

2658: 

2659: An extremely interesting and deceptively simple quantum-mechanical model is the 

2660: $\cP\cT$-symmetric square well, whose Hamiltonian on the domain $0<x<\pi$ is

2661: given by

2662: \begin{equation}

2663: H=\p^2+V(\x),

2664: \label{e127}

2665: \end{equation}

2666: where in the coordinate representation $V(x)=\infty$ for $x<0$ and $x>\pi$ and

2667: \begin{equation} V(x) = \left\{ \begin{array}{cl}

2668: i\epsilon \quad & \mbox{for $\frac{\pi}{2}<x<\pi,$} \\

2669: -i\epsilon \quad & \mbox{for $0<x<\frac{\pi}{2}.$} \\

2670: \end{array}\right.

2671: \label{e128}

2672: \end{equation}

2673: The $\cP\cT$-symmetric square-well Hamiltonian was invented by Znojil \cite{Z1}

2674: and it has been heavily studied by many researchers \cite{Z2,Z3,Z4,Z5}. This

2675: Hamiltonian reduces to the conventional Hermitian square well in the limit as

2676: $\epsilon\to0$. For $H$ in (\ref{e127}) the parity operator $\cP$ performs a

2677: reflection about $x=\frac{\pi}{2}$: $\cP:x\to\pi-x$.

2678: 

2679: In all the examples discussed so far the coordinate-space representation of the

2680: $\cC$ operator is a combination of integer powers of $x$ and integer numbers of

2681: derivatives multiplying the parity operator $\cP$. Hence, the $Q$ operator is a

2682: polynomial in the operators $\x$ and $\p=-i\frac{d}{dx}$. The novelty of the

2683: $\cP\cT$-symmetric square-well Hamiltonian (\ref{e127}) and (\ref{e128}) is that

2684: $\cC$ contains {\em integrals} of $\cP$. Thus, the $Q$ operator, while it is a

2685: simple function, is {\em not} a polynomial in $\x$ and $\p$ and therefore it

2686: cannot be found easily by the algebraic perturbative methods introduced in

2687: Subsec.~\ref{ss6-1}.

2688: 

2689: Instead, in Ref.~\cite{BT} the $\cC$ operator for the square-well Hamiltonian

2690: was calculated by using the brute-force approach of calculating the $\cP

2691: \cT$-normalized eigenfunctions $\phi_n(x)$ of $H$ and summing over these

2692: eigenfunctions. The eigenfunctions $\phi_n(x)$ were obtained as perturbation

2693: series to second order in powers of $\epsilon$. The eigenfunctions were then

2694: normalized according to (\ref{e80}), substituted into the formula (\ref{e87}),

2695: and the sum was evaluated directly to obtain the $\cC$ operator accurate to

2696: order $\epsilon^2$. The advantage of the domain $0<x<\pi$ is that this sum

2697: reduces to a set of Fourier sine and cosine series that can be evaluated in

2698: closed form. After evaluating the sum, the result was translated to the

2699: symmetric region $-\frac{\pi}{2}<x<\frac{\pi}{2}$. On this domain the parity

2700: operator in coordinate space is $\cP(x,y)=\delta(x+y)$.

2701: 

2702: The last step in the calculation is to show that the $\cC$ operator to order

2703: $\epsilon^2$ has the form $e^Q\cP$ and then to evaluate the function $Q$ to

2704: order $\epsilon^2$. The final result for $Q(x,y)$ on the domain $-\frac{\pi}

2705: {2}<x<\frac{\pi}{2}$ has the relatively simple structure

2706: \begin{equation}

2707: Q(x,y)=\textstyle{\frac{1}{4}}i\epsilon[x-y+\varepsilon(x-y)\,(|\,x+y\,|-\pi)]+

2708: \mathcal{O}(\epsilon^3),

2709: \label{e129}

2710: \end{equation}

2711: where $\varepsilon(x)$ is the standard step function

2712: \begin{equation}

2713: \varepsilon(x)=\left\{\begin{array}{cl}1\quad&\mbox{($x>0$),}\\

2714: 0\quad&\mbox{($x=0$),}\\ -1\quad&\mbox{($x<0$).}\\ \end{array}\right.

2715: \label{e130}

2716: \end{equation}

2717: 

2718: When we say that the formula for $Q(x,y)$ has a relatively simple structure, we

2719: mean that this structure is simple in comparison with the formula for the

2720: $C$ operator expressed as an expansion $C(x,y)=C^{(0)}+\epsilon C^{(1)}+

2721: \epsilon^2C^{(2)}+\mathcal{O}(\epsilon^3)$. The formulas for the first three

2722: coefficients in this series for $\cC$ are

2723: \begin{eqnarray}

2724: \cC^{(0)}(x,y)&=&\delta(x+y),\nonumber\\

2725: \cC^{(1)}(x,y)&=&{\textstyle\frac{1}{4}}i[x+y+\varepsilon(x+y)\,(|x-y|-\pi)],

2726: \nonumber\\

2727: \cC^{(2)}(x,y)&=&{\textstyle\frac{1}{96}}\pi^3-{\textstyle\frac{1}{24}}(x^3+y^3)

2728: \,\varepsilon(x+y)-{\textstyle\frac{1}{24}}(y^3-x^3)\,\varepsilon(y-x)

2729: \nonumber\\

2730: &&+{\textstyle\frac{1}{8}}xy\pi-{\textstyle\frac{1}{16}}\pi^2(x+y)\,\varepsilon(

2731: x+y)+{\textstyle\frac{1}{8}}\pi(x|x|+y|y|)\,\varepsilon(x+y)\nonumber\\

2732: &&-{\textstyle\frac{1}{4}}xy\{|x|[\theta(x-y)\,\theta(-x-y)+\theta(y-x)\,\theta

2733: (x+y)]\nonumber\\

2734: &&+|y|[\theta(y-x)\,\theta(-x-y)+\theta(x-y)\,\theta(x+y)]\}.

2735: \label{e131}

2736: \end{eqnarray}

2737: We plot the imaginary part of $\cC^{(1)}(x,y)$ in Fig.~\ref{f22}, and we plot

2738: $\cC^{(2)}(x,y)$ in Fig.~\ref{f23}. These three-dimensional plots show

2739: $\cC^{(1)}(x,y)$ and $\cC^{(2)}(x,y)$ on the symmetric domain $-\frac{\pi}{2}<

2740: (x,y)<\frac{\pi}{2}$.

2741: 

2742: The most noteworthy property of the $\cC$ operator for the square-well model is

2743: that the associated operator $Q$ is a nonpolynomial function, and this kind of

2744: structure had not been seen in previous studies of $\cC$. It was originally

2745: believed that for such a simple $\cP\cT$-symmetric Hamiltonian it would be

2746: possible to calculate the $\cC$ operator in closed form. It is a surprise that

2747: for this elementary model the $\cC$ operator is so nontrivial.

2748: 

2749: \begin{figure}[th]\vspace{2.65in}

2750: \special{psfile=Fig22.ps angle=0 hoffset=80 voffset=-26 hscale=57 vscale=57}

2751: \caption{Three-dimensional plot of the imaginary part of $\cC^{(1)}(x,y)$, the

2752: first-order perturbative contribution in (\ref{e131}) to the $\cC$ operator in

2753: coordinate space. The plot is on the symmetric square domain $-\frac{\pi}{2}<(x,

2754: y)<\frac{\pi}{2}$. Note that $\cC^{(1)}(x,y)$ vanishes on the boundary of this

2755: square domain because the eigenfunctions $\phi_n(x)$ are required to vanish at

2756: $x=0$ and $x=\pi$.}

2757: \label{f22}

2758: \end{figure}

2759: 

2760: \begin{figure}[th]\vspace{2.65in}

2761: \special{psfile=Fig23.ps angle=0 hoffset=80 voffset=-26 hscale=57 vscale=57}

2762: \caption{Three-dimensional plot of $\cC^{(2)}(x,y)$ in (\ref{e131}) on the

2763: symmetric square domain $-\frac{\pi}{2}<(x,y)<\frac{\pi}{2}$. The function

2764: $\cC^{(2)}(x,y)$ vanishes on the boundary of this square domain because the

2765: eigenfunctions $\phi_n(x)$ from which it was constructed vanish at the

2766: boundaries of the square well.}

2767: \label{f23}

2768: \end{figure}

2769: 

2770: \subsection{Mapping from a $\cP\cT$-Symmetric Hamiltonian to a Hermitian

2771: Hamiltonian}

2772: \label{ss6-4}

2773: 

2774: Mostafazadeh observed that the square root of the positive operator $e^Q$ can be

2775: used to construct a similarity transformation that maps a non-Hermitian $\cP

2776: \cT$-symmetric Hamiltonian $H$ to an equivalent Hermitian Hamiltonian $h$

2777: \cite{M1,M2}:

2778: \begin{equation}

2779: h=e^{-Q/2}He^{Q/2}.

2780: \label{e132}

2781: \end{equation}

2782: He noted that $h$ is equivalent to $H$ because it has the same eigenvalues as

2783: $H$.

2784: 

2785: To understand why (\ref{e132}) is valid, recall from (\ref{e107}) that the $\cC$

2786: operator has the general form $\cC=e^Q\cP$, where $Q=Q(\x,\p)$ is a Hermitian

2787: function of the fundamental dynamical operator variables of the quantum theory.

2788: Multiplying $\cC$ on the right by $\cP$ gives an expression for $e^Q$:

2789: \begin{equation}

2790: e^Q=\cC\cP,

2791: \label{e133}

2792: \end{equation}

2793: which indicates that $\cC\cP$ is a positive and invertible operator.

2794: 

2795: To verify that $h$ in (\ref{e132}) is Hermitian, take the Hermitian conjugate of

2796: (\ref{e132})

2797: \begin{equation}

2798: h^\dag=e^{Q/2}H^\dag e^{-Q/2},

2799: \label{e134}

2800: \end{equation}

2801: which can be rewritten as

2802: \begin{equation}

2803: h^\dag=e^{-Q/2}e^QH^\dag e^{-Q}e^{Q/2}.

2804: \label{e135}

2805: \end{equation}

2806: Use (\ref{e133}) to replace $e^Q$ by $\cC\cP$ and $e^{-Q}$ by $\cP\cC$

2807: \begin{equation}

2808: h^\dag=e^{-Q/2}\cC\cP H^\dag\cP\cC e^{Q/2}

2809: \label{e136}

2810: \end{equation}

2811: and recall from (\ref{e92}) that $H^\dag$ can be replaced by $\cP H\cP$. This

2812: gives

2813: \begin{equation}

2814: h^\dag=e^{-Q/2}\cC\cP\cP H\cP\cP\cC e^{Q/2}=e^{-Q/2}\cC H\cC e^{Q/2}.

2815: \label{e137}

2816: \end{equation}

2817: Finally, one use the fact that $\cC$ commutes with $H$ [see (\ref{e113})] and

2818: that the square of $\cC$ is unity [see (\ref{e112})] to reduce the right side 

2819: of (\ref{e136}) to the right side of (\ref{e132}). This verifies that $h$ is

2820: Hermitian in the Dirac sense.

2821: 

2822: We conclude from this calculation that for every non-Hermitian $\cP

2823: \cT$-symmetric Hamiltonian $H$ whose $\cP\cT$ symmetry is unbroken, it is,

2824: in principle, possible to construct via (\ref{e132}) a Hermitian Hamiltonian $h$

2825: that has exactly the same eigenvalues as $H$. Note that it is crucial that $H$

2826: have an unbroken $\cP\cT$ symmetry because this allows us to construct $\cC$,

2827: which in turn allows us to construct the similarity operator $\e^{Q/2}$.

2828: 

2829: This construction poses the following question: Are $\cP\cT$-symmetric

2830: Hamiltonians physically new and distinct from ordinary Hermitian Hamiltonians,

2831: or do they describe exactly the same physical processes that ordinary

2832: Hermitian Hamiltonians describe?

2833: 

2834: There are two answers to this question, the first being technical and practical

2835: and the second being an answer in principle. First, while it is theoretically

2836: possible to construct the Hermitian Hamiltonian $h$ whose spectrum is identical

2837: to that of $H$, it cannot in general be done except at the perturbative level

2838: because the transformation is so horribly complicated. (See, for example, the

2839: discussion of the square well in Subsec.~\ref{ss6-3}.) Furthermore, while the

2840: $\cP\cT$-symmetric Hamiltonian $H$ is simple in structure and easy to use in

2841: calculations because the interaction term is local, it is shown in

2842: Ref.~\cite{k1} that $h$ is {\em nonlocal} (its interaction term has arbitrarily

2843: high powers of the variables $\x$ and $\p$.) Thus, it is not just difficult to

2844: calculate using $h$; it is almost impossible because the usual regulation

2845: schemes are hopelessly inadequate.

2846: 

2847: There is only one known nontrivial example for which there is actually a

2848: closed-form expression for both $H$ and $h$, and this is the case of the quartic

2849: Hamiltonian discussed in Subecs.~\ref{ss2-6} -- \ref{ss2-8}. Even this case, it

2850: is not possible to construct the $\cC$ operator in closed form because this

2851: operator is nonlocal (it contains a Fourier transform) and it performs a

2852: transformation in the complex plane. This is the only known example for which it

2853: is practical to calculate with both $H$ and $h$. Hence, while the mapping from

2854: $H$ to $h$ is of great theoretical interest, it does not have much practical

2855: value.

2856: 

2857: The second answer is of greater importance because it leads immediately to

2858: physical considerations. The transformation from $H$ to $h$ in (\ref{e132}) is

2859: a {\em similarity} and not a unitary transformation. Thus, while the eigenvalues

2860: of $H$ and $h$ are the same, relationships between vectors are changed; pairs of

2861: vectors that are orthogonal are mapped into pairs of vectors that are not

2862: orthogonal. Thus, experiments that measure vectorial relationships can

2863: distinguish between $H$ and $h$. One plausible experiment, which involves the

2864: speed of unitary time evolution, is described in detail in Subsec.~\ref{ss7-3}.

2865: 

2866: \section{Applications of $\cP\cT$-Symmetric Hamiltonians in Quantum Mechanics}

2867: \label{s7}

2868: 

2869: It is not yet known whether non-Hermitian $\mathcal{PT}$-symmetric Hamiltonians

2870: describe phenomena that can be observed experimentally. However, non-Hermitian

2871: $\cP\cT$-symmetric Hamiltonians have {\em already} appeared in the literature

2872: very often and their remarkable properties have been noticed and used by many 

2873: authors. For example, in 1959 Wu showed that the ground state of a Bose system

2874: of hard spheres is described by a non-Hermitian Hamiltonian \cite{A1}. Wu found

2875: that the ground-state energy of this system is real and he conjectured that all

2876: the energy levels were real. Hollowood showed that the non-Hermitian Hamiltonian

2877: of a complex Toda lattice has real energy levels \cite{H4}. Cubic non-Hermitian

2878: Hamiltonians of the form $H=\p^2+i\x^3$ (and also cubic quantum field theories)

2879: arise in studies of the Lee-Yang edge singularity \cite{A3,A4,A5,A6} and in

2880: various Reggeon field theory models \cite{A7,A8}. In all of these cases a

2881: non-Hermitian Hamiltonian having a real spectrum appeared mysterious at the

2882: time, but now the explanation is simple: In every case the non-Hermitian

2883: Hamiltonian is $\cP\cT$ symmetric. In each case the Hamiltonian is constructed

2884: so that the position operator $\x$ or the field operator $\phi$ is always

2885: multiplied by $i$. Hamiltonians having $\cP\cT$ symmetry have also been used to

2886: describe magnetohydrodynamic systems \cite{G1,G2} and to study nondissipative

2887: time-dependent systems interacting with electromagnetic fields \cite{Frig}.

2888: 

2889: In this section we describe briefly four areas of quantum mechanics in which

2890: non-Hermitian $\cP\cT$-Hamiltonians play a crucial and significant role. 

2891: 

2892: \subsection{New $\cP\cT$-Symmetric Quasi-Exactly Solvable Hamiltonians}

2893: \label{ss7-1}

2894: 

2895: A quantum-mechanical Hamiltonian is said to be {\em quasi-exactly solvable}

2896: (QES) if a finite portion of its energy spectrum and associated eigenfunctions

2897: can be found exactly and in closed form \cite{A9}. QES potentials depend on an

2898: integer parameter $J$; for positive values of $J$ one can find exactly the first

2899: $J$ eigenvalues and eigenfunctions, typically of a given parity. QES systems are

2900: classified using an algebraic approach in which the Hamiltonian is expressed in

2901: terms of the generators of a Lie algebra \cite{A10}. This approach generalizes

2902: the dynamical-symmetry analysis of {\em exactly solvable} quantum-mechanical

2903: systems whose {\em entire} spectrum may be found in closed form by algebraic

2904: means.

2905: 

2906: Prior to the discovery of non-Hermitian $\cP\cT$-symmetric Hamiltonians the

2907: lowest-degree one-dimensional QES polynomial potential that was known was a

2908: sextic potential having one continuous parameter as well as a discrete parameter

2909: $J$. A simple case of such a potential is \cite{BD}

2910: \begin{equation}

2911: V(x)=x^6-(4J-1)x^2.

2912: \label{e138}

2913: \end{equation}

2914: For this potential, the Schr\"odinger equation $-\psi''(x)+[V(x)-E]\psi(x)=0$

2915: has

2916: $J$ even-parity solutions of the form

2917: \begin{equation}

2918: \psi(x)=e^{-x^4/4}\sum_{k=0}^{J-1}c_k x^{2k}.

2919: \label{e139}

2920: \end{equation}

2921: The coefficients $c_k$ for $0\leq k\leq J-1$ satisfy the recursion relation

2922: \begin{equation}

2923: 4(J-k)c_{k-1}+Ec_k+2(k+1)(2k+1)c_{k+1}=0,

2924: \label{e140}

2925: \end{equation}

2926: where $c_{-1}=c_{J}=0$. The linear equations (\ref{e140}) have a nontrivial

2927: solution for $c_0,\,c_1,\,...,\,c_{J-1}$ if the determinant of the coefficients

2928: vanishes. For each integer $J$ this determinant is a polynomial of degree $J$ in

2929: the variable $E$. The roots of this polynomial are all real and they are the $J$

2930: quasi-exact energy eigenvalues of the potential (\ref{e138}).

2931: 

2932: The discovery of $\cP\cT$ symmetry allows us to introduce an entirely new class

2933: of QES {\em quartic} polynomial potentials having {\em two} continuous

2934: parameters in addition to the discrete parameter $J$. The Hamiltonian has the

2935: form \cite{A11}

2936: \begin{equation}

2937: H=\p^2-\x^4+2ia\x^3+(a^2-2b)\x^2+2i(ab-J)\x,

2938: \label{e141}

2939: \end{equation}

2940: where $a$ and $b$ are real and $J$ is a positive integer. The spectra of this

2941: family of Hamiltonians are real, discrete, and bounded below. Like the

2942: eigenvalues of the potential (\ref{e138}), the lowest $J$ eigenvalues of $H$ are

2943: the roots of a polynomial of degree $J$.

2944: 

2945: The eigenfunction $\psi(x)$ satisfies $\cP\cT$-symmetric boundary conditions;

2946: it vanishes in the Stokes wedges shown in Fig.~\ref{f4}. The eigenfunction

2947: satisfies

2948: \begin{equation}

2949: -\psi''(x)+\left[-x^4+2iax^3+(a^2-2b)x^2+2i(ab-J)x\right]\psi(x)=E\psi(x).

2950: \label{e142}

2951: \end{equation}

2952: 

2953: We obtain the QES portion of the spectrum of $H$ in (\ref{e141}) by making

2954: the {\em ansatz} $\psi(x)=\exp\left(-\frac{1}{3}ix^3-\half ax^2-ibx\right)

2955: P_{J-1}(x)$, where 

2956: \begin{equation}

2957: P_{J-1}(x)=x^{J-1}+\sum_{k=0}^{J-2}c_k x^k

2958: \label{e143}

2959: \end{equation}

2960: is a polynomial in $x$ of degree $J-1$. Substituting $\psi(x)$ into the

2961: differential equation (\ref{e142}), dividing off the exponential, and collecting

2962: powers of $x$, we obtain a polynomial in $x$ of degree $J-1$. Setting the

2963: coefficients of $x^k$ ($1\leq k\leq J-1$) to $0$ gives a system of $J-1$

2964: simultaneous linear equations for the coefficients $c_k$ ($0\leq k \leq J-2$).

2965: We solve these equations and substitute the values of $c_k$ into the coefficient

2966: of $x^0$. This gives a polynomial $Q_J(E)$ of degree $J$ in the eigenvalue $E$.

2967: The coefficients of this polynomial are functions of the parameters $a$ and $b$.

2968: The first two of these polynomials are

2969: \begin{eqnarray}

2970: Q_1 &=& E -b^2 -a,\nonumber\\

2971: Q_2 &=& E^2 -(2b^2+4a)E+b^4+4ab^2-4b+3a^2.

2972: \label{e144}

2973: \end{eqnarray}

2974: The roots of $Q_J(E)$ are the QES portion of the spectrum of $H$.

2975: 

2976: The polynomials $Q_J(E)$ simplify dramatically when we substitute

2977: \begin{equation}

2978: E=F+b^2+Ja\qquad{\rm and}\qquad K=4b+a^2.

2979: \label{e145}

2980: \end{equation}

2981: The new polynomials then have the form

2982: \begin{eqnarray}

2983: Q_1 &=& F,\nonumber\\

2984: Q_2 &=& F^2-K,\nonumber\\

2985: Q_3 &=& F^3-4KF-16,\nonumber\\

2986: Q_4 &=& F^4-10KF^2-96F+9K^2,\nonumber\\

2987: Q_5 &=& F^5-20KF^3-336F^2+64K^2F+768K.

2988: \label{e146}

2989: \end{eqnarray}

2990: The roots of these polynomials are all real as long as $K\geq K_{\rm critical}$,

2991: where $K_{\rm critical}$ is a function of $J$. 

2992: 

2993: Had $\cP\cT$-symmetric quantum mechanics not been discovered, this beautiful

2994: family of quartic QES Hamiltonians would never have been considered because the

2995: quartic term has a negative sign. Lacking the discovery that $\cP\cT$-symmetric

2996: Hamiltonians have positive spectra, the Hamiltonians in (\ref{e141}) would have

2997: been rejected because it would have been assumed that the spectra of such

2998: Hamiltonians would be unbounded below. Quasi-exactly solvable sextic $\cP

2999: \cT$-symmetric Hamiltonians are studied in \cite{MONOU}.

3000: 

3001: \subsection{Complex Crystals}

3002: \label{ss7-2}

3003: 

3004: An experimental signal of a complex Hamiltonian might be found in the context of

3005: condensed matter physics. Consider the complex crystal lattice whose potential

3006: is $V(x)=i\sin\,x$. The optical properties of complex crystal lattices were

3007: first studied by Berry and O'Dell, who referred to them as complex diffraction

3008: gratings \cite{MVB}.

3009: 

3010: While the Hamiltonian $H=\p^2+i\sin\x$ is not Hermitian, it is $\cP\cT$

3011: symmetric and all of its energy bands are {\em real}. At the edge of the bands

3012: the wave function of a particle in such a lattice is bosonic ($2\pi$-periodic),

3013: and unlike the case of ordinary crystal lattices, the wave function is never

3014: fermionic ($4\pi$-periodic). The discriminant for a Hermitian $\sin(x)$

3015: potential is plotted in Fig.~\ref{f24} and the discriminant for a non-Hermitian

3016: $i\sin(x)$ potential is plotted in Fig.~\ref{f25}. The difference between these

3017: two figures is subtle. In Fig.~\ref{f25} the discriminant does not go below $-

3018: 2$, and thus there are half as many gaps \cite{A12}. Direct observation of such

3019: a band structure would give unambiguous evidence of a $\cP\cT$-symmetric

3020: Hamiltonian \cite{ZZ}. Complex periodic potentials having more elaborate band

3021: structures have also been found \cite{KS1,KS2,KS3,KS4}.

3022: 

3023: \begin{figure*}[t!]

3024: \vspace{2.5in}

3025: \special{psfile=Fig24.ps angle=0 hoffset=-15 voffset=-547 hscale=100 vscale=100}

3026: \caption{Discriminant $\Delta(E)$ plotted as a function of $E$ for the real

3027: periodic potential $V(x)=\sin(x)$. Although it cannot be seen in the figure, all

3028: local maxima lie {\em above} the line $\Delta=2$ and all local minima lie {\em

3029: below} the line $\Delta=-2$. The regions of energy $E$ for which $|\Delta|\leq2$

3030: correspond to allowed energy bands and the regions where $|\Delta|>2$ correspond

3031: to gaps. There are infinitely many gaps and these gaps become exponentially

3032: narrow as $E$ increases. It is clear in this figure that the first maximum lies

3033: above $2$. The first minimum occurs at $E=2.313\,8$, where the discriminant has

3034: the value $-2.003\,878\,7$. The second maximum occurs at $E=4.033\,6$, where the

3035: discriminant is $2.000\,007$. Similar behavior is found for other potentials in

3036: the class $V(x)=\sin^{2N+1}(x)$.}

3037: \label{f24}

3038: \end{figure*}

3039: 

3040: \begin{figure*}[t!]

3041: \vspace{2.65in}

3042: \special{psfile=Fig25.ps angle=0 hoffset=-15 voffset=-537 hscale=100 vscale=100}

3043: \caption{Discriminant $\Delta(E)$ plotted as a function of $E$ for the complex

3044: periodic potential $V(x)=i\sin(x)$. For $E>2$ it is not possible to see any

3045: difference between this figure and Fig.~\ref{f24}. However, although it cannot

3046: be seen, all local maxima in this figure lie {\em above} the line $\Delta=2$ and

3047: all local minima lie {\em above} the line $\Delta =-2$. This behavior is

3048: distinctly different from the behavior in Fig.~\ref{f24} exhibited by the real

3049: periodic potential $V(x)=\sin x$. In stark contrast with real periodic

3050: potentials, for the potentials of the form $V(x)=i\sin^{2N+1}(x)$, while the

3051: maxima of the discriminant lie above $+2$, the minima of the discriminant lie

3052: above $-2$. Thus, for these potentials there are no antiperiodic wave functions.

3053: Lengthy and delicate numerical analysis verifies that for the potential $V(x)=i

3054: \sin(x)$ the first three maxima of the discriminant $\Delta(E)$ are located at

3055: $E=3.966\,428\,4$, $E=8.985\,732\,0$, and $E=15.992\,066\,213\,46$. The value of

3056: the discriminant $\Delta(E)$ at these energies is $2.000\,007$, $2.000\,000\,000

3057: \,000\,69$, and $2.000\,000\,000\,000\,000\,000\,04$. The first two minima of

3058: the discriminant are located at $E=2.191\,6$ and $E=6.229\,223$ and at these

3059: energies $\Delta(E)$ has the values $-1.995\,338\,6$ and $-1.999\,999\,995\,27$.

3060: Similar behavior is found for the other complex periodic potentials in the

3061: class.}

3062: \label{f25}

3063: \end{figure*}

3064: 

3065: \subsection{Quantum Brachistochrone}

3066: \label{ss7-3}

3067: 

3068: We pointed out in Subsec.~\ref{ss6-4} that there is a similarity transformation

3069: that maps a non-Hermitian $\cP\cT$-symmetric Hamiltonian $H$ to a Hermitian

3070: Hamiltonian $h$ [see (\ref{e132})]. The two Hamiltonians, $H$ and $h$, have the 

3071: same eigenvalues, but this does not mean they describe the same physics. To

3072: illustrate the differences between $H$ and $h$, we show how to solve the quantum

3073: brachistochrone problem for $\cP\cT$-symmetric and for Hermitian quantum

3074: mechanics, and we show that the solution to this problem in these two

3075: formulations of quantum mechanics is not the same.

3076: 

3077: The fancy word {\em brachistochrone} means ``shortest time.'' Thus, the {\em

3078: quantum brachistochrone} problem is defined as follows: Suppose we are given

3079: initial and final quantum states $|\psi_I\rangle$ and $|\psi_F\rangle$ in a

3080: Hilbert space. There exist infinitely many Hamiltonians $H$ under which $|\psi_I

3081: \rangle$ evolves into $|\psi_F\rangle$ in some time $t$:

3082: \begin{equation}

3083: |\psi_F\rangle=e^{-iHt/\hbar}|\psi_I\rangle.

3084: \label{e147}

3085: \end{equation}

3086: The problem is to find the Hamiltonian $H$ that minimizes the evolution time $t$

3087: subject to the constraint that $\omega$, the difference between the largest and

3088: smallest eigenvalues of $H$, is held fixed. The shortest evolution time is

3089: denoted by $\tau$.

3090: 

3091: In Hermitian quantum mechanics there is an unavoidable lower bound $\tau$ on the

3092: time required to transform one state into another. Thus, the minimum time

3093: required to flip unitarily a spin-up state into a spin-down state of an electron

3094: is an important physical quantity because it gives an upper bound on the

3095: speed of a quantum computer.

3096: 

3097: In this paper we have shown that Hermitian quantum mechanics can be extended

3098: into the complex domain while retaining the reality of the energy eigenvalues,

3099: the unitarity of time evolution, and the probabilistic interpretation. It

3100: has recently been discovered that within this complex framework a spin-up state

3101: can be transformed {\em arbitrarily quickly} to a spin-down state by a simple

3102: non-Hermitian Hamiltonian \cite{BR1}.

3103: 

3104: Let us first show how to find the value of $\tau$ for the case of a Hermitian

3105: Hamiltonian: This problem is easy because finding the optimal evolution time

3106: $\tau$ requires only the solution to the simpler problem of finding the optimal

3107: evolution time for the $2\times2$ matrix Hamiltonians acting on the

3108: two-dimensional subspace spanned by $|\psi_I\rangle$ and $|\psi_F\rangle$

3109: \cite{rrr2}. To solve the Hermitian version of the two-dimensional quantum

3110: brachistochrone problem, we choose the basis so that the initial and final

3111: states are given by

3112: \begin{equation}

3113: |\psi_I\rangle=\left( \begin{array}{c} 1\\0\end{array} \right)\quad{\rm and}

3114: \qquad |\psi_F\rangle=\left(\begin{array}{c} a\\b\end{array}\right),

3115: \label{e148}

3116: \end{equation}

3117: where $|a|^2+|b|^2=1$. The most general $2\times2$ Hermitian Hamiltonian is

3118: \begin{equation}

3119: H=\left(\begin{array}{cc} s & r{\rm e}^{-i\theta} \cr r{\rm e}^{i\theta} & u

3120: \end{array}\right),

3121: \label{e149}

3122: \end{equation}

3123: where the parameters $r$, $s$, $u$, and $\theta$ are real. The eigenvalue

3124: constraint is

3125: \begin{equation}

3126: \omega^2=(s-u)^2+4r^2.

3127: \label{e150}

3128: \end{equation}

3129: The Hamiltonian $H$ in (\ref{e149}) can be expressed in terms of the Pauli

3130: matrices

3131: \begin{equation}

3132: \sigma_1=\left( \begin{array}{cc} 0&1\\ 1&0\end{array}\right),\quad

3133: \sigma_2=\left( \begin{array}{cc} 0&-i\\ i&0\end{array}\right),\quad

3134: \sigma_3=\left( \begin{array}{cc} 1&0\\ 0&-1\end{array}\right),

3135: \label{e151}

3136: \end{equation}

3137: as $H=\half(s+u){\bf 1}+\half\omega{\boldsymbol{\sigma}}\!\cdot\!{\bf n}$, where

3138: ${\bf n}=\frac{2}{\omega}\left(r\cos\theta,r\sin\theta,\frac{s-u}{2}\right)$ is

3139: a unit vector. The matrix identity

3140: \begin{equation}

3141: \exp(i\phi\,{\boldsymbol{\sigma}}\!\cdot\!{\bf n})=\cos\phi\,{\bf

3142: 1} + i \sin\phi\,{\boldsymbol{\sigma}}\!\cdot\!{\bf n}

3143: \label{e152}

3144: \end{equation}

3145: then simplifies the relation $|\psi_F\rangle=e^{-iH\tau/\hbar}|\psi_I\rangle$ to

3146: \begin{equation}

3147: \left(\begin{array}{c} a\\ b\end{array}\right)=e^{-i(s+u)t/(2\hbar)}\left(

3148: \begin{array}{c}\cos\left(\frac{\omega t}{2\hbar}\right)-i\frac{s-u}{\omega}

3149: \sin\left(\frac{\omega t}{2\hbar}\right) \\ {}\\ -i\frac{2r}{\omega}e^{i\theta}

3150: \sin\left(\frac{\omega t}{2\hbar}\right)\end{array}\right).

3151: \label{e153}

3152: \end{equation}

3153: 

3154: From the second component of (\ref{e153}) we obtain $|b|=\frac{2r}{\omega}\sin

3155: \left(\frac{\omega t}{2\hbar}\right)$, which gives the time required to

3156: transform the initial state: $t=\frac{2\hbar}{\omega}\arcsin\big(\frac{\omega|b|

3157: }{2r}\big)$. To optimize this relation over all $r>0$, we note that (\ref{e150})

3158: implies that the maximum value of $r$ is $\half\omega$ and that at this maximum

3159: $s=u$. The optimal time is thus

3160: \begin{equation}

3161: \tau=\frac{2\hbar}{\omega}\arcsin|b|.

3162: \label{e154}

3163: \end{equation}

3164: Note that if $a=0$ and $b=1$ we have $\tau=2\pi\hbar/\omega$ for the smallest

3165: time required to transform $\left({1\atop0}\right)$ to the orthogonal state

3166: $\left({0\atop1}\right)$.

3167: 

3168: For general $a$ and $b$, at the optimal time $\tau$ we have $a=e^{i\tau s/\hbar}

3169: \sqrt{1-|b|^2}$ and $b=ie^{i\tau s/\hbar}|b|e^{i\theta}$, which satisfies the

3170: condition $|a|^2+|b|^2=1$ that the norm of the state does not change under

3171: unitary time evolution. The parameters $s$ and $\theta$ are determined by the

3172: phases of $a$ and $b$. We set $a=|a|e^{i{\rm arg}(a)}$ and $b=|b|e^{i{\rm arg}

3173: (b)}$ and find that the optimal Hamiltonian is

3174: \begin{equation}

3175: H=\left(\begin{array}{cc}\frac{\omega{\rm arg}(a)}{2\arcsin|b|} & \frac{\omega}

3176: {4}e^{-i[{\rm arg}(b)-{\rm arg}(a)-\frac{\pi}{2}]}\cr\frac{\omega}{4}e^{i[{\rm

3177: arg}(b)-{\rm arg}(a)-\frac{\pi}{2}]}&\frac{\omega{\rm arg}(a)}{2\arcsin|b|}

3178: \end{array}\right).

3179: \label{e155}

3180: \end{equation}

3181: The overall phase of $|\psi_{\rm F}\rangle$ is not physically relevant, so the

3182: quantity ${\rm arg}(a)$ may be chosen arbitrarily; we may thus assume that it is

3183: $0$. We are free to choose ${\rm arg}(a)$ because there is no absolute energy in

3184: quantum mechanics; one can add a constant to the eigenvalues of the Hamiltonian

3185: without altering the physics. Equivalently, this means that the value of $\tau$

3186: cannot depend on the trace $s+u$ of $H$.

3187: 

3188: How do we interpret the result for $\tau$ in (\ref{e154})? While this equation

3189: resembles the time-energy uncertainty principle, it is really the much simpler

3190: statement that {\em rate$\times$time$=$distance}. The constraint (\ref{e150}) on

3191: $H$ is a bound on the standard deviation $\Delta H$ of the Hamiltonian, where

3192: $\Delta H$ in a normalized state $|\psi\rangle$ is given by $(\Delta H)^2=

3193: \langle\psi|H^2|\psi\rangle-\langle\psi|H|\psi\rangle^2$. The maximum value of

3194: $\Delta H$ is $\omega/2$. From the Anandan-Aharonov relation~\cite{rrr4}, the

3195: speed of evolution of a quantum state is given by $\Delta H$. The distance

3196: between the initial state $|\psi_I\rangle$ and the final state $|\psi_F\rangle$

3197: is $\delta=2\arccos(|\langle\psi_F|\psi_I\rangle|)$. Thus, the shortest time

3198: $\tau$ for $|\psi_I\rangle$ to evolve to $|\psi_F\rangle=e^{-iH\tau/\hbar}|

3199: \psi_I\rangle$ is bounded below because the speed is bounded above while the

3200: distance is held fixed. The Hamiltonian $H$ that realizes the shortest time

3201: evolution can be understood as follows: The standard deviation $\Delta H$ of

3202: the Hamiltonian in (\ref{e149}) is $r$. Since $\Delta H$ is bounded by $\omega/

3203: 2$, to maximize the speed of evolution (and minimize the time of evolution) we

3204: choose $r=\omega /2$.

3205: 

3206: We now perform the same optimization for the complex $2\times2$ non-Hermitian

3207: $\cP\cT$ Hamiltonian in (\ref{e93}). Following the same procedure used for

3208: Hermitian Hamiltonians, we rewrite $H$ in (\ref{e93}) in the form $H=(r\cos

3209: \theta){\bf 1}+\half\omega{\boldsymbol{\sigma}}\!\cdot\!{\bf n}$, where ${\bf n}

3210: =\frac{2}{\omega}(s,0,ir\sin\theta)$ is a unit vector and the squared difference

3211: between the energy eigenvalues [see (\ref{e95}) is

3212: \begin{equation}

3213: \omega^2=4s^2-4r^2\sin^2\theta.

3214: \label{e156}

3215: \end{equation}

3216: The condition of unbroken $\cP\cT$ symmetry ensures the positivity of

3217: $\omega^2$. This equation emphasizes the key difference between Hermitian and

3218: $\cP\cT$-symmetric Hamiltonians: The corresponding equation (\ref{e150}) for the

3219: Hermitian Hamiltonian has a {\em sum} of squares, while this equation for

3220: $\omega^2$ has a {\em difference} of squares. Thus, Hermitian Hamiltonians

3221: exhibit elliptic behavior, which leads to a nonzero lower bound for the optimal

3222: time $\tau$. The hyperbolic nature of (\ref{e156}) allows $\tau$ to approach

3223: zero because, as we will see, the matrix elements of the $\cP\cT$-symmetric

3224: Hamiltonian can be made large without violating the energy constraint

3225: $E_+-E_-=\omega$.

3226: 

3227: The $\cP\cT$-symmetric analog of the evolution equation (\ref{e153}) is

3228: \begin{equation}

3229: e^{-iHt/\hbar}\left(\begin{array}{c}1\\ 0\end{array}\right)=\frac{e^{-itr\cos

3230: \theta/\hbar}}{\cos\alpha}\left(\begin{array}{c}\cos\left(\frac{\omega t}{2\hbar

3231: }-\alpha\right) \\ {}\\ -i\sin\left(\frac{\omega t}{2\hbar}\right)\end{array}

3232: \right).

3233: \label{e157}

3234: \end{equation}

3235: We apply this result to the same pair of vectors examined in the Hermitian case:

3236: $|\psi_I\rangle=\left(1\atop0\right)$ and $|\psi_F\rangle=\left(0\atop1\right)$.

3237: (Note that these vectors are not orthogonal with respect to the $\cC\cP\cT$

3238: inner product.) Equation (\ref{e157}) shows that the evolution time to reach $|

3239: \psi_F\rangle$ from $|\psi_I\rangle$ is $t=(2\alpha-\pi)\hbar/\omega$.

3240: Optimizing this result over allowable values for $\alpha$, we find that as

3241: $\alpha$ approaches $\half\pi$ the optimal time $\tau$ tends to zero.

3242: 

3243: Note that in the limit $\alpha\to\half\pi$ we get $\cos\alpha\to0$. However, in

3244: terms of $\alpha$, the energy constraint (\ref{e156}) becomes $\omega^2=4s^2

3245: \cos^2\alpha$. Since $\omega$ is fixed, in order to have $\alpha$ approach

3246: $\half\pi$ we must require that $s\gg1$. It then follows from the relation $\sin

3247: \alpha=(r/s)\sin\theta$ that $|r|\sim|s|$, so we must also require that $r\gg1$.

3248: Evidently, in order to make $\tau\ll1$, the matrix elements of the $\cP

3249: \cT$-symmetric Hamiltonian (\ref{e93}) must be large.

3250: 

3251: The result demonstrated here does not violate the uncertainty principle. Indeed,

3252: Hermitian and non-Hermitian $\cP\cT$-symmetric Hamiltonians share the properties

3253: that (i) the evolution time is given by $2\pi\hbar/\omega$, and (ii) $\Delta H

3254: \leq\omega/2$. The key difference is that a pair of states such as $\left({1

3255: \atop0}\right)$ and $\left({0\atop1}\right)$ are orthogonal in a Hermitian

3256: theory, but have separation $\delta=\pi-2|\alpha|$ in the $\cP\cT$-symmetric

3257: theory. This is because the Hilbert space metric of the $\cP\cT$-symmetric

3258: quantum theory {\em depends on the Hamiltonian}. Hence, it is possible to

3259: choose the parameter $\alpha$ to create a wormhole-like effect in the Hilbert

3260: space.

3261: 

3262: A {\em gedanken} experiment to realize this effect in a laboratory might work as

3263: follows: A Stern-Gerlach filter creates a beam of spin-up electrons. The beam

3264: then passes through a `black box' containing a device governed by a $\cP

3265: \cT$-symmetric Hamiltonian that flips the spins unitarily in a very short time.

3266: The outgoing beam then enters a second Stern-Gerlach device which verifies that

3267: the electrons are now in spin-down states. In effect, the black-box device is

3268: applying a magnetic field in the complex direction $(s,0,ir\sin\theta)$. If the

3269: field strength is sufficiently strong, then spins can be flipped unitarily in

3270: virtually no time because the complex path joining these two states is arbitrary

3271: short without violating the energy constraint. The arbitrarily short alternative

3272: complex pathway from a spin-up state to a spin-down state, as illustrated by

3273: this thought experiment, is reminiscent of the short alternative distance

3274: between two widely separated space-time points as measured through a wormhole in

3275: general relativity.

3276: 

3277: The results established here provide the possibility of performing experiments

3278: that distinguish between Hermitian and $\cP\cT$-symmetric Hamiltonians. If

3279: practical implementation of complex $\cP\cT$-symmetric Hamiltonians were

3280: feasible, then identifying the optimal unitary transformation would be

3281: particularly important in the design and implementation of fast quantum

3282: communication and computation algorithms. Of course, the wormhole-like effect we

3283: have discussed here can only be realized if it is possible to switch rapidly

3284: between Hermitian and $\cP\cT$-symmetric Hamiltonians by means of similarity

3285: transformations. It is conceivable that so much quantum noise would be generated

3286: that there is a sort of quantum protection mechanism that places a lower bound

3287: on the time required to switch Hilbert spaces. If so, this would limit the

3288: applicability of a Hilbert-space wormhole to improve quantum algorithms.

3289: 

3290: \subsection{Supersymmetric $\cP\cT$-Symmetric Hamiltonians}

3291: \label{ss7-4}

3292: 

3293: After the discovery of $\cP\cT$-symmetric Hamiltonians in quantum mechanics, it

3294: was proposed that $\cP\cT$ symmetry might be combined with supersymmetry

3295: \cite{BM7} in the context of quantum field theory. In Ref.~\cite{BM7} it was

3296: shown that one can easily construct two-dimensional supersymmetric quantum field

3297: theories by introducing a $\cP\cT$-symmetric superpotential of the form ${\cal

3298: S}(\phi)=-ig(i\phi)^{1+\epsilon}$. The resulting quantum field theories exhibit

3299: a broken parity symmetry for all $\delta>0$. However, supersymmetry remains

3300: unbroken, which is verified by showing that the ground-state energy density

3301: vanishes and that the fermion-boson mass ratio is unity. Many papers have

3302: subsequently worked on the subject of $\cP\cT$-symmetric supersymmetric quantum

3303: mechanics. (See Ref.~\cite{ZX}.) A particularly interesting paper by Dorey {\em

3304: et al.} examines the connection between supersymmetry and broken $\cP\cT$

3305: symmetry in quantum mechanics \cite{D6}.

3306: 

3307: \subsection{Other Quantum-Mechanical Applications}

3308: \label{ss7-5}

3309: 

3310: There are many additional quantum-mechanical applications of non-Hermitian ${\cP

3311: \cT}$-invariant Hamiltonians. In condensed matter physics Hamiltonians rendered

3312: non-Hermitian by an imaginary external field have been introduced to study

3313: delocalization transitions in condensed matter systems such as vortex flux-line

3314: depinning in type-II superconductors \cite{HN}. In this Hatano-Nelson model

3315: there is a critical value of the anisotropy (non-Hermiticity) parameter below

3316: which all eigenvalues are real \cite{RS}. In the theory of Reaction-Diffusion

3317: systems, many models have been constructed for systems described by matrices

3318: that can be non-Hermitian \cite{MH1,MH2} and with the appropriate definition of

3319: the $\cP$ and $\cT$ operators, these systems can be shown to be $\cP\cT$

3320: symmetric. Finally, we mention a recent paper by Hibberd {\em et al.} who found

3321: a transformation that maps a Hamiltonian describing coherent superpositions of

3322: Cooper pairs and condensed molecular bosons to one that is $\cP\cT$-symmetric

3323: \cite{HDL}.

3324: 

3325: \section{$\cP\cT$-Symmetric Quantum Field Theory}

3326: \label{s8}

3327: 

3328: Quantum-mechanical theories have only a finite number of degrees of freedom.

3329: Most of the $\cP\cT$-symmetric quantum-mechanical models discussed so far in

3330: this paper have just one degree of freedom; that is, the Hamiltonians for these

3331: theories are constructed from just one pair of dynamical variables, $\x$ and

3332: $\p$. In a quantum field theory the operators $\x(t)$ and $\p(t)$ are replaced

3333: by the quantum fields $\vf({\bf x},t)$ and $\pi({\bf x},t)$, which represent

3334: a continuously infinite number of degrees of freedom, one for each value of the

3335: spatial variable ${\bf x}$. Such theories are vastly more complicated than

3336: quantum-mechanical theories, but constructing quantum field theories that are

3337: non-Hermitian and $\cP\cT$ symmetric is straightforward. For example, the

3338: quantum-field-theoretic Hamiltonians that are analogous to the

3339: quantum-mechanical Hamiltonians in (\ref{e10}) and (\ref{e11}) are

3340: \begin{equation}

3341: \mathcal{H}=\half\pi^2({\bf x},t)+\half[\nabla_{\bf x}\vf({\bf x},t)]^2+

3342: \half\mu^2\vf^2({\bf x},t)+ig\vf^3({\bf x},t)

3343: \label{e158}

3344: \end{equation}

3345: and

3346: \begin{equation}

3347: \mathcal{H}=\half\pi^2({\bf x},t)+\half[\nabla_{\bf x}\vf({\bf x},t)]^2+

3348: \half\mu^2\vf^2({\bf x},t)-{\textstyle\frac{1}{4}}g\vf^4({\bf x},t).

3349: \label{e159}

3350: \end{equation}

3351: As in quantum mechanics, where the operators $\x$ and $\p$ change sign under

3352: parity reflection $\cP$, we assume that the fields in these Hamiltonians are

3353: {\em pseudoscalars} so that they also change sign:

3354: \begin{equation}

3355: \cP\vf({\bf x},t)\cP=-\vf(-{\bf x},t),\qquad\cP\pi({\bf x},t)\cP=-\pi(-{\bf

3356: x},t).

3357: \label{e160}

3358: \end{equation}

3359: 

3360: Quantum field theories like these that possess $\cP\cT$ symmetry exhibit a rich

3361: variety of behaviors. Cubic field-theory models like that in (\ref{e158}) are of

3362: interest because they arise in the study of the Lee-Yang edge singularity

3363: \cite{A3,A4,A5} and in Reggeon field theory \cite{A7,A8}. In these papers it was

3364: asserted that an $i\vf^3$ field theory is nonunitary. However, by constructing

3365: the $\cC$ operator, we argue in Subsec.~\ref{ss8-1} that this quantum field

3366: theory is, in fact, unitary. We show in Subsec.~\ref{ss8-3} how $\cP\cT$

3367: symmetry eliminates the ghosts in the Lee Model, another cubic quantum field

3368: theory. The field theory described by (\ref{e159}) is striking because it is

3369: asymptotically free, as explained in Subsec.~\ref{ss8-4}. We also examine $\cP

3370: \cT$-symmetric quantum electrodynamics in Subsec.~\ref{ss8-5}, the $\cP

3371: \cT$-symmetric Thirring and Sine-Gordon models in Subsec.~\ref{ss8-6}, and

3372: gravitational and cosmological theories in Subsec.~\ref{ss8-7}. Last, we look

3373: briefly at $\cP\cT$-symmetric classical field theories in Subsec.~\ref{ss8-8}.

3374: 

3375: \subsection{$i\vf^3$ Quantum Field Theory}

3376: \label{ss8-1}

3377: 

3378: In courses on quantum field theory, a scalar $g\vf^3$ theory is used as a

3379: pedagogical example of perturbative renormalization even though this model is

3380: not physically realistic because the energy is not bounded below. However, by

3381: calculating $\cC$ perturbatively for the case when the coupling constant $g=i

3382: \epsilon$ is pure imaginary, one obtains a fully acceptable Lorentz invariant

3383: quantum field theory. This calculation shows that it is possible to construct

3384: perturbatively the Hilbert space in which the Hamiltonian for this cubic scalar

3385: field theory in $(D+1)$-dimensional Minkowski space-time is {\em self-adjoint}.

3386: Consequently, such theories have positive spectra and exhibit unitary time

3387: evolution. 

3388: 

3389: In this subsection we explain how to calculate perturbatively the $\cC$ operator

3390: for the quantum-field-theoretic Hamiltonian in (\ref{e158}) \cite{BBJ2}. We

3391: apply the powerful algebraic techniques explained in Subsecs.~\ref{ss6-1} and

3392: \ref{ss6-2} for the calculation of the $\cC$ operator in quantum mechanics. As

3393: in quantum mechanics, we express $\cC$ in the form $\cC=\exp\left(\epsilon Q_1+

3394: \epsilon^3Q_3+\ldots\right)\cP$, where now $Q_{2n+1}$ ($n=0,\,1,\,2,\,\ldots$)

3395: are real functionals of the fields $\vf_{\bf x}$ and $\pi_{\bf x}$. To find

3396: $Q_n$ for $\mathcal{H}$ in (\ref{e158}) we must solve a system of operator

3397: equations.

3398: 

3399: We begin by making an {\em ansatz} for $Q_1$ analogous to the {\em ansatz} used

3400: in (\ref{e120}):

3401: \begin{equation}

3402: Q_1=\int\!\!\int\!\!\int d{\bf x}\,d{\bf y}\,d{\bf z}\left(M_{({\bf xyz})}\pi_{

3403: \bf x}\pi_{\bf y}\pi_{\bf z}+N_{{\bf x}({\bf yz})}\vf_{\bf y}\pi_{\bf x}

3404: \vf_{\bf z}\right).

3405: \label{e161}

3406: \end{equation}

3407: In quantum mechanics $M$ and $N$ are constants, but in field theory they are

3408: functions. The notation $M_{({\bf x}{\bf y}{\bf z})}$ indicates that this

3409: function is totally symmetric in its three arguments, and the notation $N_{{\bf

3410: x}({\bf y}{\bf z})}$ indicates that this function is symmetric under the

3411: interchange of the second and third arguments. The unknown functions $M$ and $N$

3412: are form factors; they describe the spatial distribution of three-point

3413: interactions of the field variables in $Q_1$. The nonlocal spatial interaction

3414: of the fields is an intrinsic property of $\cC$. (Note that we have suppressed

3415: the time variable $t$ in the fields and that we use subscripts to indicate the

3416: spatial dependence.)

3417: 

3418: To determine $M$ and $N$ we substitute $Q_1$ into the first equation in

3419: (\ref{e119}), namely $\left[H_0,Q_1\right]=-2H_1$, which now takes the form

3420: \begin{equation}

3421: \left[\int d{\bf x}\,\pi^2_{\bf x}+\int\!\!\int d{\bf x}\,d{\bf y}\,\vf_{\bf

3422: x}G_{\bf xy}^{-1}\vf_{\bf y},Q_1\right]=-4i\int d{\bf x}\,\vf^3_{\bf x},

3423: \label{e162}

3424: \end{equation}

3425: where the inverse Green's function is given by $G_{{\bf x}{\bf y}}^{-1}\equiv(

3426: \mu^2-\nabla_{\bf x}^2)\delta({\bf x}-{\bf y})$. We obtain the following

3427: coupled system of partial differential equations:

3428: \begin{eqnarray}

3429: &&(\mu^2-\nabla_{\bf x}^2)N_{{\bf x}({\bf y}{\bf z})}

3430: +(\mu^2-\nabla_{\bf y}^2)N_{{\bf y}({\bf x}{\bf z})}

3431: +(\mu^2-\nabla_{\bf z}^2)N_{{\bf z}({\bf x}{\bf y})}\nonumber\\

3432: &&\qquad\qquad\qquad\,=-6\delta({\bf x}-{\bf y})\delta({\bf x}-{\bf z}),

3433: \nonumber\\

3434: &&N_{{\bf x}({\bf y}{\bf z})}+N_{{\bf y}({\bf x}{\bf z})}=

3435: 3(\mu^2-\nabla_{\bf z}^2) M_{({\bf w}{\bf x}{\bf y})}.

3436: \label{e163}

3437: \end{eqnarray}

3438: 

3439: We solve these equations by Fourier transforming to momentum space and get

3440: \begin{equation}

3441: M_{({\bf xyz})}=-\frac{4}{(2\pi)^{2D}}\int\!\!\int d{\bf p}\,

3442: d{\bf q}\frac{e^{i({\bf x}-{\bf y})\cdot{\bf p}+i({\bf x}-{\bf z})\cdot{\bf q}}}

3443: {D({\bf p},{\bf q})},

3444: \label{e164}

3445: \end{equation}

3446: where $D({\bf p},{\bf q})=4[{\bf p}^2{\bf q}^2-({\bf p}\cdot{\bf q})^2]+4\mu^2(

3447: {\bf p}^2+{\bf p}\cdot{\bf q}+{\bf q}^2)+3\mu^4$ is positive, and

3448: \begin{eqnarray}

3449: N_{{\bf x}({\bf yz})}&=&-3\left(\nabla_{\bf y}\cdot\nabla_{\bf

3450: z}+\half\mu^2\right)M_{({\bf xyz})}.

3451: \label{e165}

3452: \end{eqnarray}

3453: For the special case of a $(1+1)$-dimensional quantum field theory the integral

3454: in (\ref{e165}) evaluates to $M_{({\bf x}{\bf y}{\bf z})}=-{\rm K}_0(\mu R)/

3455: \left(\sqrt{3}\pi\mu^2\right)$, where ${\rm K}_0$ is the associated Bessel

3456: function and $R^2=\half[({\bf x}-{\bf y})^2+({\bf y}-{\bf z})^2+({\bf z}-{\bf x}

3457: )^2]$. This completes the calculation of $\cC$ to first order in perturbation

3458: theory.

3459: 

3460: We mention finally that the $\cC$ operator for this cubic quantum field theory

3461: transforms as a scalar under the action of the homogeneous Lorentz group

3462: \cite{BBCW}. In Ref.~\cite{BBCW} it was argued that because the Hamiltonian

3463: $H_0$ for the unperturbed theory ($g=0$) commutes with the parity operator

3464: $\cP$, the intrinsic parity operator $\cP_{\rm I}$ in the noninteracting theory

3465: transforms as a Lorentz scalar. (The {\em intrinsic} parity operator $\cP_{\rm

3466: I}$ and the parity operator $\cP$ have the same effect on the fields, except

3467: that $\cP_{\rm I}$ does not reverse the sign of the spatial argument of the

3468: field. In quantum mechanics $\cP$ and $\cP_{\rm I}$ are indistinguishable.) When

3469: the coupling constant $g$ is nonzero, the parity symmetry of $H$ is broken and

3470: $\cP_{\rm I}$ is no longer a scalar. However, $\cC$ {\em is} a scalar. Since

3471: $\lim_{g\to0}\cC=\cP_{\rm I}$, one can interpret the $\cC$ operator in quantum

3472: field theory as the complex extension of the intrinsic parity operator when the

3473: imaginary coupling constant is turned on. This means that $\cC$ is

3474: frame-invariant and it shows that the $\cC$ operator plays a truly fundamental

3475: role in non-Hermitian quantum field theory.

3476: 

3477: \subsection{Other Quantum Field Theories Having Cubic Interactions}

3478: \label{ss8-2}

3479: 

3480: We can repeat the calculations done in Subsec.~\ref{ss8-1} for cubic quantum

3481: field theories having several interacting scalar fields \cite{BBJ2,BBJ3}. For

3482: example, consider the case of {\em two} scalar fields $\vf_{\bf x}^{(1)}$

3483: and $\vf_{\bf x}^{(2)}$ whose interaction is governed by

3484: \begin{equation}

3485: H=H_0^{(1)}+H_0^{(2)}+i\epsilon\int d{\bf x}\,\big[\vf_{\bf x}^{(1)}\big]^2

3486: \vf_{\bf x}^{(2)},

3487: \label{e166}

3488: \end{equation}

3489: which is the analog of the quantum-mechanical theory described by $H$ in

3490: (\ref{e123}). Here,

3491: \begin{equation}

3492: H_0^{(j)}=\half\int d{\bf x}\,\big[\pi_{\bf x}^{(j)}\big]^2+\half\int\!\!\int

3493: d{\bf x}\,d{\bf y}\,\big[G_{\bf xy}^{(j)}\big]^{-1}\vf_{\bf x}^{(j)}

3494: \vf_{\bf y}^{(j)}.

3495: \label{e167}

3496: \end{equation}

3497: To determine $\cC$ to order $\epsilon$ we introduce the {\em ansatz}

3498: \begin{eqnarray}

3499: Q_1&=&\int\!\!\int\!\!\int d{\bf x}\,d{\bf y}\,d{\bf z}\Big[N_{{\bf x}({\bf yz})

3500: }^{(1)}\left(\pi_{\bf z}^{(1)}\vf_{\bf y}^{(1)}+\vf_{\bf y}^{(1)}\pi_{

3501: \bf z}^{(1)}\right)\vf_{\bf x}^{(2)}\nonumber\\

3502: &&+N_{{\bf x}({\bf yz})}^{(2)}\pi_{\bf x}^{(2)}\vf_{\bf y}^{(1)}\vf_{\bf

3503: z}^{(1)}+M_{{\bf x}({\bf yz})}\pi_{\bf x}^{(2)}\pi_{\bf y}^{(1)}\pi_{\bf z}^{(1)

3504: }\Big],

3505: \label{e168}

3506: \end{eqnarray}

3507: where $M_{{\bf x}({\bf y}{\bf z})}$, $N_{{\bf x}({\bf y}{\bf z})}^{(1)}$, and

3508: $N_{{\bf x}({\bf y}{\bf z})}^{(2)}$ are unknown functions and the parentheses

3509: indicate symmetrization. We get

3510: \begin{eqnarray}

3511: M_{{\bf x}({\bf yz})}&=&- G_m(R_1)\,G_{\mu_2}(R_2),\nonumber\\

3512: N_{{\bf x}({\bf yz})}^{(1)}&=&-\delta(2{\bf x}-{\bf y}-{\bf z})G_m(R_1),

3513: \nonumber\\

3514: N_{{\bf x}({\bf yz})}^{(2)}&=& \half\delta(2{\bf x}-{\bf y}-{\bf z})G_m(R_1)

3515: -\delta({\bf y}-{\bf z})G_{\mu_2}(R_2),

3516: \label{e169}

3517: \end{eqnarray}

3518: where $G_\mu(r)=\frac{r}{\mu}\big(\frac{\mu}{2\pi r}\big)^{D/2}{\rm K}_{-1+D/2}

3519: (\mu r)$ with $r=|{\bf r}|$ is the Green's function in $D$-dimensional space,

3520: $m^2=\mu_1^2-\frac{1}{4}\mu_2^2$, $R_1^2=({\bf y}-{\bf z})^2$, and $R_2^2=

3521: \frac{1}{4}(2{\bf x}-{\bf y}-{\bf z})^2$.

3522: 

3523: For {\em three} interacting scalar fields whose dynamics is described by

3524: \begin{equation}

3525: H=H_0^{(1)}+H_0^{(2)}+H_0^{(3)}+i\epsilon\int d{\bf x}\,\vf_{\bf x}^{(1)}

3526: \vf_{\bf x}^{(2)}\vf_{\bf x}^{(3)},

3527: \label{e170}

3528: \end{equation}

3529: which is the analog of $H$ in (\ref{e125}), we make the {\em ansatz}

3530: \begin{eqnarray}

3531: Q_1&=&\int\!\!\int\!\!\int d{\bf x}\,d{\bf y}\,d{\bf z}\Big[N_{{\bf x}{\bf y}

3532: {\bf z}}^{(1)}\pi_{\bf x}^{(1)}\vf_{\bf y}^{(2)}\vf_{\bf z}^{(3)}+N_{

3533: {\bf x}{\bf y}{\bf z}}^{(2)}\pi_{\bf x}^{(2)}\vf_{\bf y}^{(3)}\vf_{\bf

3534: z}^{(1)}\nonumber\\

3535: &&\quad +N_{{\bf x}{\bf y}{\bf z}}^{(3)}\pi_{\bf x}^{(3)}\vf_{\bf y}^{(1)}

3536: \vf_{\bf z}^{(2)}+M_{{\bf x}{\bf y}{\bf z}}\pi_{\bf x}^{(1)}\pi_{\bf y}^{(2)

3537: }\pi_{\bf z}^{(3)}\Big].

3538: \label{e171}

3539: \end{eqnarray}

3540: 

3541: The solutions for the unknown functions are as follows: $M_{\bf xyz}$ is given

3542: by the integral (\ref{e164}) with the more general formula $D({\bf p},{\bf q})=

3543: 4[{\bf p}^2{\bf q}^2-({\bf p}\cdot{\bf q})^2]+4[\mu_1^2({\bf q}^2+{\bf p}\cdot{

3544: \bf q})+\mu_2^2({\bf p}^2+{\bf p}\cdot{\bf q})-\mu_3^2{\bf p}\cdot{\bf q}]+\mu^4

3545: $ with $\mu^4=2\mu_1^2\mu_2^2+2\mu_1^2\mu_3^2+2\mu_2^2\mu_3^2-\mu_1^4-\mu_2^4-

3546: \mu_3^4$. The $N$ coefficients are expressed as derivatives acting on $M$:

3547: \begin{eqnarray}

3548: N_{\bf xyz}^{(1)}&=&\left[4\nabla_{\bf y}\cdot\nabla_{\bf z}+2(\mu_2^2

3549: +\mu_3^2-\mu_1^2)\right] M_{\bf xyz},\nonumber\\

3550: N_{\bf xyz}^{(2)}&=&\left[-4\nabla_{\bf y}\cdot\nabla_{\bf z}

3551: -4\nabla_{\bf z}^2+2(\mu_1^2+\mu_3^2-\mu_2^2)\right] M_{\bf xyz},\nonumber\\

3552: N_{\bf xyz}^{(3)}&=&\left[-4\nabla_{\bf y}\cdot\nabla_{\bf z}

3553: -4\nabla_{\bf y}^2+2(\mu_1^2+\mu_2^2-\mu_3^2)\right] M_{\bf xyz}.

3554: \label{e172}

3555: \end{eqnarray}

3556: Once again, the calculation of $\cC$ shows that these cubic field theories are

3557: fully consistent quantum theories.\footnote{In Refs.~\cite{BBCW,LEE} it is shown

3558: that the $\cC$ operator in quantum field theory has the form $\cC=e^Q\cP_{\rm

3559: I}$, where $\cP_{\rm I}$ is the {\em intrinsic} parity reflection operator. The

3560: difference between $\cP$ and $\cP_{\rm I}$ is that $\cP_{\rm I}$ does not

3561: reflect the spatial arguments of the fields. For a cubic interaction Hamiltonian

3562: this distinction is technical. It does not affect the final result for the $Q$

3563: operator in (\ref{e161}), (\ref{e168}), and (\ref{e171}).}

3564: 

3565: \subsection{The Lee model}

3566: \label{ss8-3}

3567: 

3568: In this subsection we will show how to use the tools that we have developed to

3569: study non-Hermitian $\cP\cT$-symmetric quantum theories to examine a famous

3570: model quantum field theory known as the {\em Lee model}. In 1954 the Lee model

3571: was proposed as a quantum field theory in which mass, wave function, and charge

3572: renormalization could be performed exactly and in closed form

3573: \cite{BARTON,L1,L2,L3}. We discuss the Lee model here because when the

3574: renormalized coupling constant is taken to be larger than a critical value, the

3575: Hamiltonian becomes non-Hermitian and a (negative-norm) ghost state appears.

3576: The appearance of the ghost state was assumed to be a fundamental defect of the

3577: Lee model. However, we show that the non-Hermitian Lee-model Hamiltonian is

3578: actually $\cP\cT$ symmetric. When the states of this model are examined using

3579: the $\cC$ operator, the ghost state is found to be an ordinary physical state

3580: having positive norm \cite{LEE}.

3581: 

3582: The idea for studying the Lee model as a non-Hermitian Hamiltonian is due to

3583: Kleefeld, who was the first to point out this transition to $\cP\cT$ symmetry

3584: \cite{KLE}. His work gives a comprehensive history of non-Hermitian

3585: Hamiltonians.

3586: 

3587: The Lee model has a cubic interaction term that describes the interaction of

3588: three spinless particles called $V$, $N$, and $\theta$. The $V$ and $N$

3589: particles are fermions and behave roughly like nucleons, and the $\theta$

3590: particle is a boson and behaves roughly like a pion. In the model a $V$ may emit

3591: a $\theta$, but when it does so it becomes an $N$: $V\rightarrow N\,+\,\theta$.

3592: Also, an $N$ may absorb a $\theta$, but when it does so it becomes a $V$: $N\,+

3593: \,\theta\rightarrow V$.

3594: 

3595: The solvability of the Lee model is based on the fact that there is no crossing

3596: symmetry. That is, the $N$ is forbidden to emit an anti-$\theta$ and become a

3597: $V$. Eliminating crossing symmetry makes the Lee model solvable because it

3598: introduces two conservation laws. First, {\em the number of $N$ quanta plus the

3599: number of $V$ quanta is fixed}. Second, {\em the number of $N$ quanta minus the

3600: number of $\theta$ quanta is fixed.} These two highly constraining conservation

3601: laws decompose the Hilbert space of states into an infinite number of

3602: noninteracting sectors. The simplest sector is the vacuum sector. Because of the

3603: conservation laws, there are no vacuum graphs and the bare vacuum is the

3604: physical vacuum. The next two sectors are the one-$\theta$-particle and the

3605: one-$N$-particle sector. These two sectors are also trivial because the two

3606: conservation laws prevent any dynamical processes from occurring there. As a

3607: result, the masses of the $N$ particle and of the $\theta$ particle are not

3608: renormalized; that is, the physical masses of these particles are the same as

3609: their bare masses.

3610: 

3611: The lowest nontrivial sector is the $V/N\theta$ sector. The physical states in

3612: this sector of the Hilbert space are linear combinations of the bare $V$ and the

3613: bare $N\theta$ states, and these states consist of the one-physical-$V$-particle

3614: state and the physical $N$-$\theta$-scattering states. To find these states one

3615: can look for the poles and cuts of the Green's functions. The renormalization in

3616: this sector is easy to perform. Following the conventional renormalization

3617: procedure, one finds that the mass of the $V$ particle is renormalized; that is,

3618: the mass of the physical $V$ particle is different from its bare mass. In the

3619: Lee model one calculates the unrenormalized coupling constant as a function of

3620: the renormalized coupling constant in closed form. There are many ways to define

3621: the renormalized coupling constant. For example, in an actual scattering

3622: experiment one could define the square of the renormalized coupling constant

3623: $g^2$ as the value of the $N\theta$ scattering amplitude at threshold.

3624: 

3625: The intriguing aspect of the Lee model is the appearance of a ghost state in the

3626: $V/N\theta$ sector. This state appears when one performs coupling-constant

3627: renormalization. Expressing $g_0^2$, the square of the unrenormalized coupling

3628: constant, in terms of $g^2$, the square of the renormalized coupling constant,

3629: one obtains the graph in Fig.~\ref{f1}. In principle, the $g$ is a physical

3630: parameter whose numerical value is determined by a laboratory experiment. If

3631: $g^2$ is measured to be near 0, then from Fig.~\ref{f26} we see that $g_0^2$ is

3632: also small. However, if the experimental value of $g^2$ is larger than this

3633: critical value, then the square of the unrenormalized coupling constant is

3634: negative. In this regime $g_0$ is imaginary and the Hamiltonian is

3635: non-Hermitian. Moreover, a new state appears in the $V/N\theta$ sector, and

3636: because its norm is negative, the state is called a {\em ghost}. Ghost states

3637: are unacceptable in quantum theory because their presence signals a violation of

3638: unitarity and makes a probabilistic interpretation impossible.

3639: 

3640: \begin{figure}[b!]\vspace{2.6in}

3641: \special{psfile=Fig26.eps angle=0 hoffset=-10 voffset=-440 hscale=85 vscale=85}

3642: \caption{Square of the unrenormalized coupling constant, $g_0^2$, plotted as a

3643: function of the square of the renormalized coupling constant $g^2$. Note that

3644: $g^2=0$ when $g_0^2=0$, and as $g^2$ increases from $0$ so does $g_0^2$.

3645: However, as $g^2$ increases past a critical value, $g_0^2$ abruptly becomes

3646: negative. In this regime $g_0$ is imaginary and the Hamiltonian is

3647: non-Hermitian.}

3648: \label{f26}

3649: \end{figure}

3650: 

3651: There have been many unsuccessful attempts to make sense of the Lee model as a

3652: physical quantum theory in the ghost regime \cite{BARTON,L2,L3}. However, the

3653: methods of $\cP\cT$-symmetric quantum theory enable us to give a physical

3654: interpretation for the $V/N\theta$ sector of the Lee model when $g_0$ becomes

3655: imaginary and $H$ becomes non-Hermitian. The Lee model is a cubic interaction

3656: and we have already shown in Subsecs.~\ref{ss8-1} and \ref{ss8-2} how to make

3657: sense of a Hamiltonian in which there is a cubic interaction multiplied by an

3658: imaginary coupling constant. The procedure is to calculate the $\cC$ operator

3659: and to use it to define a new inner product when the Hamiltonian is

3660: non-Hermitian.

3661: 

3662: For simplicity, we focus here on the {\em quantum-mechanical} Lee model; the

3663: results for the field-theoretic Lee model in Ref.~\cite{LEE} are qualitatively

3664: identical. The Hamiltonian for the quantum-mechanical Lee model is

3665: \begin{equation}

3666: H=H_0+g_0 H_1=m_{V_0}V^\dag V+m_N N^\dag N+m_\theta a^\dag a

3667: +\left(V^\dag Na+a^\dag N^\dag V\right).

3668: \label{e173}

3669: \end{equation}

3670: The bare states are the eigenstates of $H_0$ and the physical states are the

3671: eigenstates of the full Hamiltonian $H$. The mass parameters $m_N$ and $m_

3672: \theta$ represent the {\em physical} masses of the one-$N$-particle and

3673: one-$\theta$-particle states because these states do not undergo mass

3674: renormalization. However, $m_{V_0}$ is the {\em bare} mass of the $V$ particle.

3675: 

3676: We treat the $V$, $N$, and $\theta$ particles as pseudoscalars. To understand

3677: why, recall that in quantum mechanics the position operator $x=(a+a^\dag)/\sqrt{

3678: 2}$ and the momentum operator $p=i(a^\dag -a)/\sqrt{2}$ both change sign under

3679: parity reflection $\cP$:

3680: \begin{equation}

3681: \cP x\cP=-x,\quad\cP p\cP=-p.

3682: \label{e174}

3683: \end{equation}

3684: Thus, $\cP V\cP=-V$, $\cP N\cP=-N$, $\cP a\cP=-a$, $\cP V^\dag\cP=-V^\dag$,

3685: $\cP N^\dag\cP=-N^\dag$, $\cP a^\dag\cP=-a^\dag$. Under time reversal $\cT$, $p$

3686: and $i$ change sign but $x$ does not:

3687: \begin{equation}

3688: \cT p\cT=-p,\quad\cT i\cT=-i,\quad \cT x\cT=x.

3689: \label{e175}

3690: \end{equation}

3691: Thus, $\cT V\cT=V$, $\cT N\cT=N$, $\cT a\cT=a$, $\cT V^\dag\cT=V^\dag$, $\cT N^

3692: \dag\cT=N^\dag $, $\cT a^\dag\cT=a^\dag$.

3693: 

3694: When the bare coupling constant $g_0$ is real, $H$ in (\ref{e173}) is Hermitian:

3695: $H^\dag=H$. When $g_0$ is imaginary, $g_0=i\lambda_0\quad(\lambda_0~{\rm real}

3696: )$, $H$ is not Hermitian, but by virtue of the above transformation properties,

3697: $H$ is $\cP\cT$-symmetric: $H^{\cP\cT}=H$.

3698: 

3699: We assume first that $g_0$ is real so that $H$ is Hermitian and we examine the

3700: simplest nontrivial sector of the quantum-mechanical Lee model; namely, the

3701: $V/N\theta$ sector. We look for the eigenstates of the Hamiltonian $H$ in the

3702: form of linear combinations of the bare one-$V$-particle and the bare

3703: one-$N$-one-$\theta$-particle states. There are two eigenfunctions. We interpret

3704: the eigenfunction corresponding to the lower-energy eigenvalue as the physical

3705: one-$V$-particle state, and we interpret the eigenfunction corresponding with

3706: the higher-energy eigenvalue as the physical one-$N$-one-$\theta$-particle

3707: state. (In the field-theoretic Lee model this higher-energy state corresponds to

3708: the continuum of physical $N$-$\theta$ scattering states.) Thus, we make the

3709: {\em ansatz}

3710: \begin{equation}

3711: |V\rangle=c_{11}|1,0,0\rangle+c_{12}|0,1,1\rangle,\qquad

3712: |N\theta\rangle=c_{21}|1,0,0\rangle+c_{22}|0,1,1\rangle,

3713: \label{e176}

3714: \end{equation}

3715: and demand that these states be eigenstates of $H$ with eigenvalues $m_V$ (the

3716: renormalized $V$-particle mass) and $E_{N\theta}$. The eigenvalue problem

3717: reduces to a pair of elementary algebraic equations:

3718: \begin{equation}

3719: c_{11}m_{V_0}+c_{12}g_0=c_{11}m_V,\qquad

3720: c_{21}g_0+c_{22}\left(m_N+m_\theta\right)=c_{22}E_{N\theta}.

3721: \label{e177}

3722: \end{equation}

3723: The solutions to (\ref{e177}) are

3724: \begin{eqnarray}

3725: m_V&=&\frac{1}{2}\left(m_N+m_\theta+m_{V_0}-\sqrt{\mu_0^2+4g_0^2}\right),

3726: \nonumber\\

3727: E_{N\theta}&=&\frac{1}{2}\left(m_N+m_\theta+m_{V_0}+\sqrt{\mu_0^2+4g_0^2}

3728: \right),

3729: \label{e178}

3730: \end{eqnarray}

3731: where $\mu_0\equiv m_N+m_\theta-m_{V_0}$. Notice that $m_V$, the mass of the

3732: physical $V$ particle, is different from $m_{V_0}$, the mass of the bare $V$

3733: particle, because the $V$ particle undergoes mass renormalization.

3734: 

3735: Next, we perform wave-function renormalization. Following Barton \cite{BARTON}

3736: we define the wave-function renormalization constant $Z_V$ by $\sqrt{Z_V}=

3737: \langle 0|V|V\rangle$. This gives

3738: \begin{equation}

3739: Z_V^{-1}=\frac{1}{2}g_0^{-2}\sqrt{\mu_0^2+4g_0^2}\left(\sqrt{\mu_0^2+4g_0^2}

3740: -\mu_0\right).

3741: \label{e179}

3742: \end{equation}

3743: 

3744: Finally, we perform coupling-constant renormalization. Again, following Barton

3745: we note that $\sqrt{Z_V}$ is the ratio between the renormalized coupling

3746: constant $g$ and the bare coupling constant $g_0$ \cite{BARTON}. Thus, $g^2/

3747: g_0^2=Z_V$. Elementary algebra gives the bare coupling constant in terms of the

3748: renormalized mass and coupling constant:

3749: \begin{equation}

3750: g_0^2=g^2/\left(1-g^2/\mu^2\right),

3751: \label{e180}

3752: \end{equation}

3753: where $\mu$ is defined as $\mu\equiv m_N+m_\theta-m_V$. We cannot freely choose

3754: $g$ because the value of $g$ is, in principle, taken from experimental data.

3755: Once $g$ has been determined experimentally, we can use (\ref{e21}) to determine

3756: $g_0$. The relation in (\ref{e21}) is plotted in Fig.~\ref{f26}. This figure

3757: reveals a surprising property of the Lee model: If $g$ is larger than the

3758: critical value $\mu$, then the square of $g_0$ is negative and $g_0$ is

3759: imaginary.

3760: 

3761: As $g$ approaches its critical value from below, the two energy eigenvalues in

3762: (\ref{e178}) vary accordingly. The energy eigenvalues are the two zeros of the

3763: secular determinant $f(E)$ obtained from applying Cramer's rule to (\ref{e177}).

3764: As $g$ (and $g_0$) increase, the energy of the physical $N\theta$ state

3765: increases. The energy of the $N\theta$ state becomes infinite as $g$ reaches its

3766: critical value. As $g$ increases past its critical value, the upper energy

3767: eigenvalue goes around the bend; it abruptly jumps from being large and positive

3768: to being large and negative. Then, as $g$ continues to increase, this energy

3769: eigenvalue approaches the energy of the physical $V$ particle from below.

3770: 

3771: When $g$ increases past its critical value, the Hamiltonian $H$ in (\ref{e173})

3772: becomes non-Hermitian, but its eigenvalues in the $V/N\theta$ sector remain

3773: real. (The eigenvalues remain real because $H$ becomes $\cP\cT$ symmetric. All

3774: cubic $\cP\cT$-symmetric Hamiltonians that we have studied have been shown to

3775: have real spectra.) However, in the $\cP\cT$-symmetric regime it is no longer

3776: appropriate to interpret the lower eigenvalue as the energy of the physical $N

3777: \theta$ state. Rather, it is the energy of a new kind of state $|G\rangle$

3778: called a {\em ghost}. As is shown in Refs.~\cite{L2,L3,BARTON}, the Hermitian

3779: norm of this state is {\em negative}.

3780: 

3781: A physical interpretation of the ghost state emerges easily when we use the

3782: procedure developed in Ref.~\cite{LEE}. We begin by verifying that in the $\cP

3783: \cT$-symmetric regime, where $g_0$ is imaginary, the states of the Hamiltonian

3784: are eigenstates of the $\cP\cT$ operator, and we then choose the multiplicative

3785: phases of these states so that their $\cP\cT$ eigenvalues are unity:

3786: \begin{equation}

3787: \cP\cT|G\rangle=|G\rangle,\qquad\cP\cT|V\rangle=|V\rangle.

3788: \label{e181}

3789: \end{equation}

3790: It is then straightforward to verify that the $\cP\cT$ norm of the $V$ state is

3791: positive, while the $\cP\cT$ norm of the ghost state is negative.

3792: 

3793: We now follow the procedures described in Sec.~\ref{s6} to calculate $\cC$. We

3794: express the $\cC$ operator as an exponential of a function $Q$ multiplying the

3795: parity operator: $\cC=\exp\left[Q(V^\dag,V;N^\dag,N;a^\dag,a)\right]\cP$. We

3796: then impose the operator equations $\cC^2={\bf 1}$, $[\cC,\cP\cT]=0$, and $[\cC,

3797: H]=0$. The condition $\cC^2={\bf 1}$ gives

3798: \begin{equation}

3799: Q(V^\dag,V;N^\dag,N;a^\dag,a)=-Q(-V^\dag,-V;-N^\dag,-N;-a^\dag,-a).

3800: \label{e182}

3801: \end{equation}

3802: Thus, $Q(V^\dag,V;N^\dag,N;a^\dag,a)$ is an odd function in total powers of

3803: $V^\dag$, $V$, $N^\dag$, $N$, $a^\dag$, and $a$. Next, we impose the condition

3804: $[\cC,\cP\cT]=0$ and obtain

3805: \begin{equation}

3806: Q(V^\dag,V;N^\dag,N;a^\dag,a)=Q^*(-V^\dag,-V;-N^\dag,-N;-a^\dag,-a),

3807: \label{e183}

3808: \end{equation}

3809: where $*$ denotes complex conjugation.

3810: 

3811: Last, we impose the condition that $\cC$ commutes with $H$, which requires that

3812: \begin{equation}

3813: \left[e^Q, H_0\right]= g_0\left[e^Q,H_1\right]_+.

3814: \label{e184}

3815: \end{equation}

3816: Although in Subsecs.~\ref{ss8-1} and \ref{ss8-2} we were only able to find the

3817: $\cC$ operator to leading order in perturbation theory, for the Lee model one

3818: can calculate $\cC$ exactly and in closed form. To do so, we seek a solution for

3819: $Q$ as a formal Taylor series in powers of $g_0$:

3820: \begin{equation}

3821: Q=\sum_{n=0}^{\infty}g_0^{2 n+1}Q_{2n+1}.

3822: \label{e185}

3823: \end{equation}

3824: Only odd powers of $g_0$ appear in this series, and $Q_{2n+1}$ are all

3825: anti-Hermitian: $Q_{2n+1}^\dag=-Q_{2n+1}$. From (\ref{e185}) we get

3826: \begin{eqnarray}

3827: Q_{2n+1}=(-1)^n\frac{2^{2n+1}}{(2n+1)\mu_0^{2n+1}}\left(V^\dag Nan_\theta^n

3828: -n_\theta^na^\dag N^\dag V\right),

3829: \label{e186}

3830: \end{eqnarray}

3831: where $n_\theta=a^\dag a$ is the number operator for $\theta$-particle quanta.

3832: 

3833: We then sum over all $Q_{2n+1}$ and obtain the {\em exact} result that

3834: \begin{equation}

3835: Q=V^\dag Na\frac{1}{\sqrt{n_\theta}}{\rm arctan}\left(\frac{2g_0\sqrt{n_\theta}}

3836: {\mu_0}\right)-\frac{1}{\sqrt{n_\theta}}{\rm arctan}\left(\frac{2g_0

3837: \sqrt{n_\theta}}{\mu_0}\right) a^\dag N^\dag V.

3838: \label{e187}

3839: \end{equation}

3840: We exponentiate this result to obtain the $\cC$ operator. The exponential of $Q$

3841: simplifies considerably because we are treating the $V$ and $N$ particles as

3842: fermions and therefore we can use the identity $n_{V,N}^2=n_{V,N}$. The {\em

3843: exact} result for $e^Q$ is

3844: \begin{eqnarray}

3845: e^Q&=&\left[1-\frac{2g_0\sqrt{n_\theta}}{\sqrt{\mu_0^2+4g_0^2n_\theta}}a^\dag

3846: N^\dag V+\frac{\mu_0n_N\left(1-n_V\right)}{\sqrt{\mu_0^2+4g_0^2n_\theta}}+\frac{

3847: \mu_0n_V\left(1-n_N\right)}{\sqrt{\mu_0^2+4g_0^2\left(n_\theta+1\right)}}\right.

3848: \nonumber\\

3849: &&\left.+V^\dag Na\frac{2g_0\sqrt{n_\theta}}{\sqrt{\mu_0^2+4g_0^

3850: 2n_\theta}}- n_V-n_N+n_Vn_N \right].

3851: \label{e188}

3852: \end{eqnarray}

3853: 

3854: We can also express the parity operator $\cP$ in terms of number operators:

3855: \begin{equation}

3856: \cP=e^{i\pi\left(n_V+n_N+n_\theta\right)}=\left(1-2n_V\right)\left(1-2n_N

3857: \right)e^{i\pi n_\theta}.

3858: \label{e189}

3859: \end{equation}

3860: Combining $e^Q$ and $\cP$, we obtain the exact expression for $\cC$:

3861: \begin{eqnarray}

3862: \cC&=&\left[1-\frac{2g_0\sqrt{n_\theta}}{\sqrt{\mu_0^2+4g_0^2n_\theta}}a^\dag

3863: N^\dag V +\frac{\mu_0n_N\left(1-n_V\right)}{\sqrt{\mu_0^

3864: 2+4g_0^2n_\theta}}+\frac{\mu_0n_V\left(1-n_N\right)}{\sqrt{\mu_0^2+4g_0^2\left(

3865: n_\theta+1\right)}}\right.\nonumber\\

3866: &&\!\!\!\!\!\!\!\!\!\!\!\!\!

3867: \left.+V^\dag Na\frac{2g_0\sqrt{n_\theta}}{\sqrt{\mu_0^2+4g_0^2n_\theta}

3868: }- n_V-n_N+n_Vn_N\right]\!\!\left(1-2n_V\right)\!\left(1-2n_N\right)

3869: e^{i\pi n_\theta}.

3870: \label{e190}

3871: \end{eqnarray}

3872: 

3873: Using this $\cC$ operator to calculate the $\cC\cP\cT$ norm of the $V$ state and

3874: of the ghost state, we find that these norms are both positive. Furthermore, the

3875: the time evolution is unitary. This establishes that with the proper definition

3876: of the inner product the quantum-mechanical Lee model is a physically acceptable

3877: and consistent quantum theory, even in the ghost regime where the unrenormalized

3878: coupling constant is imaginary. The procedure of redefining the inner product to

3879: show that the ghost state is a physical state is a powerful technique that has

3880: been used by Curtright {\em et al.} and by Ivanov and Smilga for more advanced

3881: problems \cite{CM,SMIL}.

3882: 

3883: \subsection{The Higgs Sector of the Standard Model of Particle Physics}

3884: \label{ss8-4}

3885: 

3886: The distinguishing features of the $-g\vf^4$ quantum field theory in 

3887: (\ref{e159}) are that its spectrum is real and bounded below, it is

3888: perturbatively renormalizable, it has a dimensionless coupling constant in

3889: four-dimensional space-time, and it is {\em asymptotically free} \cite{AF}.

3890: The property of asymptotic freedom was established many years ago by Symanzik

3891: \cite{Sy}, as has been emphasized in a recent paper by Kleefeld \cite{KKK}. As

3892: Kleefeld explains, a $+g\vf^4$ theory in four-dimensional space-time is trivial

3893: because it is not asymptotically free. However, Symanzik proposed a

3894: ``precarious'' theory with a negative quartic coupling constant as a candidate

3895: for an asymptotically free theory of strong interactions. Symanzik used the term

3896: ``precarious'' because the negative sign of the coupling constant suggests that

3897: this theory is energetically unstable. However, as we have argued in this paper,

3898: imposing $\cP\cT$-symmetric boundary conditions (in this case on the

3899: functional-integral representation of the quantum field theory) gives a spectrum

3900: that is bounded below. Thus, Symansik's proposal of a nontrivial theory is

3901: resurrected.

3902: 

3903: The $-g\vf^4$ quantum field theory has another remarkable property. Although the

3904: theory seems to be parity invariant, the $\cP\cT$-symmetric boundary conditions

3905: violate parity invariance, as explained in Subsec.~\ref{ss2-7}. Hence, the

3906: one-point Green's function (the expectation value of the field $\vf$) does not

3907: vanish. (Techniques for calculating this expectation value are explained in

3908: \cite{BMY}.) Thus, a nonzero vacuum expectation value can be achieved without

3909: having to have spontaneous symmetry breaking because parity symmetry is {\em

3910: permanently} broken. These properties suggest that a $-g\vf^4$ quantum field

3911: theory might be useful in describing the Higgs sector of the standard model.

3912: 

3913: Perhaps the Higgs particle state is a consequence of the field-theoretic parity

3914: anomaly in the same way that the quantum-mechanical parity anomaly described in

3915: Subsecs.~\ref{ss2-7} and \ref{ss2-8} gives rise to bound states. Recent research

3916: in this area has been done by Jones {\em et al.}, who studied transformations of

3917: functional integrals \cite{JMR}, and by Meisinger and Ogilvie, who worked on

3918: large-$N$ approximations and matrix models \cite{MO}.

3919: 

3920: \subsection{$\cP\cT$-Symmetric Quantum Electrodynamics}

3921: \label{ss8-5}

3922: 

3923: If the unrenormalized electric charge $e$ in the Hamiltonian for quantum

3924: electrodynamics were imaginary, then the Hamiltonian would be non-Hermitian.

3925: However, if one specifies that the potential $A^\mu$ in this theory transforms

3926: as an {\em axial} vector instead of a vector, then the Hamiltonian becomes

3927: $\cP\cT$ symmetric \cite{QED2}. Specifically, we assume that the four-vector

3928: potential and the electromagnetic fields transform under $\cP$ like

3929: \begin{equation}

3930: \cP:\quad {\bf E\to E},\quad {\bf B\to-B}, \quad {\bf A\to A}, \quad A^0\to-A^0.

3931: \label{e191}

3932: \end{equation}

3933: Under time reversal, the transformations are assumed to be conventional:

3934: \begin{equation}

3935: \cT:\quad {\bf E\to E},\quad{\bf B\to-B},\quad{\bf A\to-A},\quad A^0\to A^0.

3936: \label{e192}

3937: \end{equation}

3938: The Lagrangian of the theory possesses an imaginary coupling constant in order

3939: that it be invariant under the product of these two symmetries:

3940: \begin{equation}

3941: \mathcal{L}=-\textstyle{\frac{1}{4}}F^{\mu\nu}F_{\mu\nu}+\half\psi^\dagger\gamma

3942: ^0\gamma^\mu\frac1i\partial_\mu\psi+\half m\psi^\dagger\gamma^0\psi+ie\psi^\dag

3943: \gamma^0\gamma^\mu\psi A_\mu.

3944: \label{e193}

3945: \end{equation}

3946: The corresponding Hamiltonian density is then

3947: \begin{equation}

3948: \mathcal{H}=\half(E^2+B^2)+\psi^\dagger\left[\gamma^0\gamma^k\left(-i\nabla_k+ie

3949: A_k\right)+m\gamma^0\right]\psi.

3950: \label{e194}

3951: \end{equation}

3952: The Lorentz transformation properties of the fermions are unchanged from the

3953: usual ones. Thus, the electric current appearing in the Lagrangian and

3954: Hamiltonian densities, $j^\mu=\psi^\dagger\gamma^0\gamma^\mu\psi$, transforms

3955: conventionally under both $\cP$ and $\cT$:

3956: \begin{equation}

3957: \cP j^\mu({\bf x},t)\cP=\left(\begin{array}{c}j^0\\-{\bf j}\end{array}\right)

3958: (-{\bf x},t),\qquad \cT j^\mu({\bf x},t)\cT=\left(\begin{array}{c} j^0\\-{\bf j}

3959: \end{array}\right)({\bf x},-t).

3960: \label{e195}

3961: \end{equation}

3962: 

3963: Because its interaction is cubic, this non-Hermitian theory of

3964: ``electrodynamics'' is the analog of the spinless $i\vf^3$ quantum field theory

3965: discussed in Subsec.~\ref{ss8-1}. ${\cP\cT}$-symmetric electrodynamics is

3966: especially interesting because it is an asymptotically free theory (unlike

3967: ordinary electrodynamics) and because the sign of the Casimir force is the

3968: opposite of that in ordinary electrodynamics \cite{QED2,QED1}. This theory is

3969: remarkable because finiteness conditions enable it to determine its own coupling

3970: constant \cite{QED1}. 

3971: 

3972: The $\cC$ operator for $\cP\cT$-symmetric quantum electrodynamics has been

3973: constructed perturbatively to first order in $e$ \cite{QED3}. This construction

3974: is too technical to describe here, but it demonstrates that non-Hermitian

3975: quantum electrodynamics is a viable and consistent unitary quantum field theory.

3976: $\cP\cT$-symmetric quantum electrodynamics is more interesting than an $i\phi^3$

3977: quantum field theory because it possesses many of the features of conventional

3978: quantum electrodynamics, including Abelian gauge invariance. The only

3979: asymptotically free quantum field theories described by Hermitian Hamiltonians

3980: are those that possess a {\em non-Abelian} gauge invariance; $\cP\cT$ symmetry

3981: allows for new kinds of asymptotically free theories, such as the $-\vf^4$

3982: theory discussed in Subsec.~\ref{ss8-4}, that do not possess a non-Abelian gauge

3983: invariance.

3984: 

3985: \subsection{Dual $\cP\cT$-Symmetric Quantum Field Theories}

3986: \label{ss8-6}

3987: 

3988: Until now we have focused on bosonic $\cP\cT$-symmetric Hamiltonians, but it is

3989: just as easy to construct fermionic $\cP\cT$-symmetric Hamiltonians. We look

3990: first at free theories. The Lagrangian density for a conventional Hermitian free

3991: fermion field theory is

3992: \begin{equation}

3993: {\cal L}({\bf x},t)=\bar\psi({\bf x},t)(i\pslash-m)\psi({\bf x},t)

3994: \label{e196}

3995: \end{equation}

3996: and the corresponding Hamiltonian density is

3997: \begin{equation}

3998: {\cal H}({\bf x},t)=\bar\psi({\bf x},t)(-i\delslash+m)\psi({\bf x},t),

3999: \label{e197}

4000: \end{equation}

4001: where $\bar\psi({\bf x},t)=\psi^\dagger({\bf x},t)\gamma_0$.

4002: 

4003: In $(1+1)$-dimensional space-time we adopt the following conventions:

4004: $\gamma_0=\sigma_1$ and $\gamma_1=i\sigma_2$, where the Pauli $\sigma$ matrices

4005: are given in (\ref{e151}). With these definitions, we have $\gamma_0^2=1$ and

4006: $\gamma_1^2=-1$. We also define $\gamma_5=\gamma_0\gamma_1=\sigma_3$, so that

4007: $\gamma_5^2=1$. The parity-reflection operator $\cP$ has the effect

4008: \begin{equation}

4009: \cP\psi(x,t)\cP=\gamma_0\psi(-x,t),\qquad\cP\bar\psi(x,t)\cP=\bar\psi(-x,t)

4010: \gamma_0.

4011: \label{e198}

4012: \end{equation}

4013: The effect of the time-reversal operator $\cT$,

4014: \begin{equation}

4015: \cT\psi(x,t)\cT=\gamma_0\psi(x,-t),\qquad\cT\bar\psi(x,t)\cT=\bar\psi(x,-t)

4016: \gamma_0,

4017: \label{e199}

4018: \end{equation}

4019: is similar to that of $\cP$, except that $\cT$ is antilinear and takes the

4020: complex conjugate of complex numbers. From these definitions the Hamiltonian $H=

4021: \int dx\,{\cal H}(x,t)$, where $\cal H$ is given in (\ref{e197}), is Hermitian:

4022: $H=H^\dag$. Also, $H$ is separately invariant under parity reflection and under

4023: time reversal: $\cP H\cP=H$ and $\cT H\cT=H$.

4024: 

4025: We can construct a non-Hermitian fermionic Hamiltonian by adding a

4026: $\gamma_5$-dependent mass term to the Hamiltonian density in (\ref{e197}):

4027: \begin{equation}

4028: {\cal H}(x,t)=\bar\psi(x,t)(-i\delslash+m_1+m_2\gamma_5)\psi(x,t)\quad(m_2~{\rm

4029: real}).

4030: \label{e200}

4031: \end{equation}

4032: The Hamiltonian $H=\int dx\,{\cal H}(x,t)$ associated with this Hamiltonian

4033: density is not Hermitian because the $m_2$ term changes sign under Hermitian

4034: conjugation. This sign change occurs because $\gamma_0$ and $\gamma_5$

4035: anticommute. Also, $H$ is not invariant under $\cP$ or under $\cT$ separately

4036: because the $m_2$ term changes sign under each of these reflections. However,

4037: $H$ {\em is} invariant under combined $\cal P$ and $\cal T$ reflection. Thus,

4038: $H$ is $\cP\cT$-symmetric.

4039: 

4040: To see whether the $\cP\cT$ symmetry of $H$ is broken or unbroken, we must check

4041: to see whether the spectrum of $H$ is real. We do so by solving the field

4042: equations. The field equation associated with $\cal H$ in (\ref{e200}) is

4043: \begin{equation}

4044: \left(i\pslash-m_1-m_2\gamma_5\right)\psi(x,t)=0.

4045: \label{e201}

4046: \end{equation}

4047: If we iterate this equation and use $\pslash^2=\partial^2$, we obtain

4048: \begin{equation}

4049: \left(\partial^2+\mu^2\right)\psi(x,t)=0,

4050: \label{e202}

4051: \end{equation}

4052: which is the two-dimensional Klein-Gordon equation with $\mu^2=m_1^2-m_2^2$. The

4053: physical mass that propagates under this equation is real when the inequality

4054: \begin{equation}

4055: m_1^2\geq m_2^2

4056: \label{e203}

4057: \end{equation}

4058: is satisfied. This condition defines the two-dimensional parametric region of

4059: {\em unbroken} $\cP\cT$ symmetry. When (\ref{e203}) is not satisfied, $\cP\cT$

4060: symmetry is {\em broken}. At the boundary between the regions of broken and

4061: unbroken $\cP\cT$ symmetry (the line $m_2=0$), the Hamiltonian is Hermitian.

4062: [The same is true in quantum mechanics. Recall that for the Hamiltonian in

4063: (\ref{e12}) the region of broken (unbroken) $\cP\cT$ symmetry is $\epsilon<0$

4064: ($\epsilon>0$). At the boundary $\epsilon=0$ of these two regions the

4065: Hamiltonian is Hermitian.]

4066: 

4067: The $\cC$ operator associated with the $\cP\cT$-symmetric Hamiltonian density

4068: $\cal H$ in (\ref{e200}) is given by $\cC=e^Q\cP$, where \cite{BJR}

4069: \begin{eqnarray}

4070: Q&=&-\tanh^{-1}\!\ve\!\int dx\,\bar\psi(x,t)\gamma_1\psi(x,t)\nonumber\\

4071: &=&-\tanh^{-1}\!\ve\!\int dx\,\psi^\dag(x,t)\gamma_5\psi(x,t).

4072: \label{e204}

4073: \end{eqnarray}

4074: The inverse hyperbolic tangent in this equation requires that $|\ve|\leq1$, or

4075: equivalently that $m_1^2\geq m_2^2$, which corresponds to the unbroken region of

4076: $\cP\cT$ symmetry. We use (\ref{e204}) to construct the equivalent Hermitian

4077: Hamiltonian $h$ as in (\ref{e132}):

4078: \begin{eqnarray}

4079: h&=&\exp\left[\half\tanh^{-1}\!\ve\int\!dx\,\psi^\dag(x,t)\gamma_5\psi(x,t)

4080: \right]H\nonumber\\

4081: &&\qquad\times\exp\left[-\half\tanh^{-1}\!\ve\int\!dx\,\psi^\dag(x,t)\gamma_5

4082: \psi(x,t)\right].

4083: \label{e205}

4084: \end{eqnarray}

4085: 

4086: The commutation relations $[\gamma_5,\gamma_0]=-2\gamma_1$ and $[\gamma_5,

4087: \gamma_1]=-2\gamma_0$ simplify $h$ in (\ref{e205}):

4088: \begin{equation}

4089: h=\int dx\,\bar\psi(x,t)(-i\delslash+\mu)\psi(x,t),

4090: \label{e206}

4091: \end{equation}

4092: where $\mu^2=m^2(1-\ve^2)=m_1^2-m_2^2$, in agreement with (\ref{e202}).

4093: Replacing $H$ by $h$ changes the $\gamma_5$-dependent mass term $m\bar\psi(1+\ve

4094: \gamma_5)\psi$ to a conventional fermion mass term $\mu\bar\psi\psi$. Thus, the

4095: non-Hermitian $\cP\cT$-symmetric Hamiltonian density in (\ref{e200}) is

4096: equivalent to the Hermitian Hamiltonian density in (\ref{e197}) with $m$

4097: replaced by $\mu$.

4098: 

4099: If we introduce a four-point fermion interaction term in (\ref{e196}), we obtain

4100: the Lagrangian density for the massive Thirring model in $(1+1)$ dimensions:

4101: \begin{equation}

4102: {\cal L}=\bar\psi(i\pslash-m)\psi+\half g(\bar\psi\gamma^\mu\psi)(\bar\psi

4103: \gamma_\mu\psi),

4104: \label{e207}

4105: \end{equation}

4106: whose corresponding Hamiltonian density is

4107: \begin{equation}

4108: {\cal H}=\bar\psi(-i\delslash+m)\psi-\half g(\bar\psi\gamma^\mu\psi)(\bar\psi

4109: \gamma_\mu\psi),

4110: \label{e208}

4111: \end{equation}

4112: We can then construct the $\cP\cT$-symmetric Thirring model

4113: \begin{equation}

4114: {\cal H}=\bar\psi(-i\delslash+m+\ve m\gamma_5)\psi-\half g(\bar\psi\gamma^\mu

4115: \psi)(\bar\psi\gamma_\mu\psi)

4116: \label{e209}

4117: \end{equation}

4118: by introducing a $\gamma_5$-dependent mass term. The additional term is

4119: non-Hermitian but $\cP\cT$-symmetric because it is odd under both parity

4120: reflection and time reversal. Remarkably, the $Q$ operator for the interacting

4121: case $g\neq0$ is {\em identical} to the $Q$ operator for the case $g=0$ because

4122: in $(1+1)$-dimensional space the interaction term $(\bar\psi\gamma^\mu\psi)(\bar

4123: \psi\gamma_\mu\psi)$ commutes with the $Q$ in (\ref{e204}) \cite{BJR}. Thus, the

4124: non-Hermitian $\cal PT$-symmetric Hamiltonian density in (\ref{e209}) is

4125: equivalent to the Hermitian Hamiltonian density in (\ref{e208}) with the mass

4126: $m$ replaced by $\mu$, where $\mu^2=m^2(1-\ve^2)=m_1^2-m_2^2$.

4127: 

4128: The same holds true for the $(3+1)$-dimensional interacting Thirring model by

4129: virtue of the commutation relation $[\gamma_5,\gamma_0\gamma_\mu]=0$, but

4130: because this higher-dimensional field theory is nonrenormalizable, the $Q$

4131: operator may only have a formal significance.

4132: 

4133: In $(1+1)$ dimensions the massive Thirring Model (\ref{e207}) is {\em dual} to

4134: the $(1+1)$-dimensional Sine-Gordon model \cite{DUAL}, whose Lagrangian density

4135: is

4136: \begin{equation}

4137: {\cal L}=\half(\p_\mu\vf)^2+m^2\lambda^{-2}(\cos\lambda\vf-1),

4138: \label{e210}

4139: \end{equation}

4140: whose corresponding Hamiltonian density is

4141: \begin{equation}

4142: {\cal H}=\half\pi^2+\half(\nabla\vf)^2+m^2\lambda^{-2}(1-\cos\lambda\vf),

4143: \label{e211}

4144: \end{equation}

4145: where $\pi(x,t)=\partial_0\vf(x,t)$ and where in $(1+1)$-dimensional space

4146: $\nabla\vf(x,t)$ is just $\p_1\vf(x,t)$. The duality between the Thirring model

4147: and the Sine-Gordon model is expressed as an algebraic relationship between the

4148: coupling constants $g$ and $\lambda$:

4149: \begin{equation}

4150: \frac{\lambda^2}{4\pi}=\frac{1}{1-g/\pi}.

4151: \label{e212}

4152: \end{equation}

4153: Note that the free fermion theory ($g=0$) is equivalent to the Sine-Gordon model

4154: with the special value for the coupling constant $\lambda^2=4\pi$.

4155: 

4156: The $\cP\cT$-symmetric extension (\ref{e209}) of the modified Thirring model is,

4157: by the same analysis, dual to a modified Sine-Gordon model with the Hamiltonian

4158: density

4159: \begin{equation}

4160: {\cal H}=\half\pi^2+\half(\nabla\vf)^2+m^2\lambda^{-2}(1-\cos\lambda\vf-i

4161: \ve\sin\lambda\vf),

4162: \label{e213}

4163: \end{equation}

4164: which is $\cP\cT$-symmetric and not Hermitian. The $Q$ operator for this

4165: Hamiltonian is

4166: \begin{equation}

4167: Q=\frac{2}{\lambda}\tanh^{-1}\!\ve\!\int\!dx\,\pi(x,t).

4168: \label{e214}

4169: \end{equation}

4170: Thus, the equivalent Hermitian Hamiltonian $h$ is given by

4171: \begin{equation}

4172: h=\exp\left[-\frac{1}{\lambda}\tanh^{-1}\!\ve\!\int\!dx\,\pi(x,t)\right]H

4173: \exp\left[\frac{1}{\lambda}\tanh^{-1}\!\ve\!\int\!dx\,\pi(x,t)\right].

4174: \label{e215}

4175: \end{equation}

4176: Note that the operation that transforms $H$ to $h$ has the same effect as

4177: shifting the boson field $\vf$ by an imaginary constant:

4178: \begin{equation}

4179: \vf\to\vf+\frac{i}{\lambda}\tanh^{-1}\!\ve.

4180: \label{e216}

4181: \end{equation}

4182: Under this transformation the interaction term $m^2\lambda^{-2}(1-\cos\lambda\vf

4183: -i\ve\sin\lambda\vf)$ in (\ref{e213}) becomes $-m^2\lambda^{-2}(1-\ve^2)\cos

4184: \lambda\vf$, apart from an additive constant. Hence, $h$ is the Hamiltonian for

4185: the Hermitian Sine-Gordon model, but with mass $\mu$ given by $\mu^2=m^2(1-

4186: \ve^2)=m_1^2-m_2^2$. This change in the mass is the same as for the Thirring

4187: model. Being Hermitian, $h$ is even in the parameter $\ve$ that breaks the

4188: Hermiticity of $H$.

4189: 

4190: The idea of generating a non-Hermitian but $\cP\cT$-symmetric Hamiltonian from a

4191: Hermitian Hamiltonian by shifting the field operator as in (\ref{e216}), first

4192: introduced in the context of quantum mechanics in Ref.~\cite{ZNOJ}, suggests a

4193: way to construct solvable fermionic $\cP\cT$-invariant models whenever there is

4194: a boson-fermion duality.

4195: 

4196: \subsection{$\cP\cT$-Symmetric Gravitational and Cosmological Theories}

4197: \label{ss8-7}

4198: 

4199: In Subsec.~\ref{ss8-5} we showed that to construct a $\cP\cT$-symmetric model of

4200: quantum electrodynamics, one need only replace the electric charge $e$ by $ie$

4201: and then replace the vector potential $A^\mu$ by an axial-vector potential. The 

4202: result is a non-Hermitian but $\cP\cT$-symmetric Hamiltonian. An interesting 

4203: classical aspect of this model is that the sign of the Coluomb force is

4204: reversed, so that like charges feel an attractive force and unlike charges

4205: feel a repulsive force.

4206: 

4207: One can use the same idea to construct a $\cP\cT$-symmetric model of massless

4208: spin-2 particles (gravitons). One simply replaces the gravitational coupling

4209: constant $G$ by $iG$ and then requires the two-component tensor field to behave

4210: like an axial tensor under parity reflection. The result is a non-Hermitian $\cP

4211: \cT$-symmetric Hamiltonian, which at the classical level describes a {\em

4212: repulsive} gravitational force. It would be interesting to investigate the

4213: possible connections between such a model and the notion of dark energy and the

4214: recent observations that the expansion of the universe is accelerating.

4215: 

4216: The connection discussed in Subsec.~\ref{ss2-6} between reflectionless

4217: potentials and $\cal PT$ symmetry may find application in quantum cosmology.

4218: There has been much attention given to anti-de Sitter cosmologies \cite{ADS} and

4219: de Sitter cosmologies \cite{DS1,DS2}. In the AdS description the universe

4220: propagates reflectionlessly in the presence of a wrong-sign potential ($-x^6$,

4221: for example). In the dS case the usual Hermitian quantum mechanics must be

4222: abandoned and be replaced by a non-Hermitian one in which there are

4223: `meta-observables'. The non-Hermitian inner product that is used in the dS case

4224: is based on the $\cC\cP\cT$ theorem in the same way that the $\cC\cP\cT$ inner

4225: product is used in $\cal PT$-symmetric quantum theory \cite{BBJ1}. The

4226: condition of reflectionless, which is equivalent to the requirement of $\cP\cT$

4227: symmetry, is what allows the wrong-sign potential to have a positive spectrum.

4228: Calculating the lowest energy level in this potential would be equivalent to

4229: determining the cosmological constant \cite{Moffat}.

4230: 

4231: \subsection{Classical Field Theory}

4232: \label{ss8-8}

4233: 

4234: The procedure for constructing $\cP\cT$-symmetric Hamiltonians is to begin with

4235: a Hamiltonian that is both Hermitian and $\cP\cT$ symmetric, and then to

4236: introduce a parameter $\epsilon$ that extends the Hamiltonian into the complex

4237: domain while maintaining its $\cP\cT$ symmetry. This is the procedure that was

4238: used to construct the new kinds Hamiltonians in (\ref{e12}). We can follow the

4239: same procedure for classical nonlinear wave equations because many wave

4240: equations are $\cP\cT$ symmetric.

4241: 

4242: As an example, consider the Korteweg-de Vries (KdV) equation

4243: \begin{equation}

4244: u_t+uu_x+u_{xxx}=0.

4245: \label{e217}

4246: \end{equation}

4247: To demonstrate that this equation is $\cP\cT$ symmetric, we define a classical

4248: parity reflection $\cP$ to be the replacement $x\to-x$, and since $u=u(x,t)$ is

4249: a velocity, the sign of $u$ also changes under $\cP$: $u\to-u$. We define a

4250: classical time reversal $\cT$ to be the replacement $t\to-t$, and again, since

4251: $u$ is a velocity, the sign of $u$ also changes under $\cT$: $u\to-u$. Following

4252: the quantum-mechanical formalism, we also require that $i\to-i$ under time

4253: reversal. Note that the KdV equation is not symmetric under $\cP$ or $\cT$

4254: separately, but that it {\it is} symmetric under combined $\cP\cT$ reflection.

4255: The KdV equation is a special case of the Camassa-Holm equation \cite{r8}, which

4256: is also $\cP\cT$ symmetric. Other nonlinear wave equations such as the

4257: generalized KdV equation $u_t+u^ku_x+u_{xxx}=0$, the Sine-Gordon equation $u_{tt

4258: }-u_{xx}+g\sin u=0$, and the Boussinesq equation are $\cP\cT$ symmetric as well.

4259: 

4260: The observation that there are many nonlinear wave equations possessing $\cP\cT$

4261: symmetry suggests that one can generate rich and interesting families of new

4262: complex nonlinear $\cP\cT$-symmetric wave equations by following the same

4263: procedure that was used in quantum mechanics and one can try to discover which

4264: properties (conservation laws, solitons, integrability, stochastic behavior) of

4265: the original wave equations are preserved and which are lost. One possible

4266: procedure for generating new nonlinear equations from the KdV equation is to

4267: introduce the real parameter $\epsilon$ as follows:

4268: \begin{equation}

4269: u_t-iu(iu_x)^\epsilon+u_{xxx}=0.

4270: \label{e218}

4271: \end{equation}

4272: Various members of this family of equations have been studied in

4273: Ref.~\cite{KDV}. Of course, there are other ways to extend the KdV equation into

4274: the complex domain while preserving $\cP\cT$ symmetry; see, for example 

4275: Ref.~\cite{Fring}.

4276: 

4277: \section{Final Remarks}

4278: \label{s9}

4279: 

4280: In this paper we have shown how to extend physical theories into the complex

4281: domain. The complex domain is huge compared with the real domain, and therefore

4282: there are many exciting new theories to explore. The obvious potential problem

4283: with extending a real theory into complex space is that one may lose some of the

4284: characteristics that a valid physical theory must possess. Thus, it is necessary

4285: that this complex extension be tightly constrained. We have shown that the

4286: essential physical axioms of a quantum theory are maintained if the complex

4287: extension is done in such a way as to preserve $\cP\cT$ symmetry.

4288: 

4289: The complex theories that we have constructed are often far more elaborate and

4290: diverse than theories that are restricted to the real domain. Upon entering the

4291: complex world we have found a gold mine of new physical theories that have

4292: strange and fascinating properties. We have just begun to study the vast new

4293: panorama that has opened up and we can hardly begin to guess what new kinds of

4294: phenomena have yet to be discovered.

4295: 

4296: \vspace{0.5cm}

4297: 

4298: \begin{footnotesize}

4299: \noindent As an Ulam Scholar, CMB receives financial support from the Center for

4300: Nonlinear Studies at the Los Alamos National Laboratory and he is supported by a

4301: grant from the U.S. Department of Energy.

4302: \end{footnotesize}

4303: \vspace{0.5cm}

4304: 

4305: \begin{thebibliography}{999}

4306: 

4307: \bibitem{BARTON} G.~Barton, {\em Introduction to Advanced Field Theory} (John

4308: Wiley \& Sons, New York, 1963), Chap.~12.

4309: 

4310: \bibitem{r1} C.~M.~Bender and S.~Boettcher, Phys.~Rev.~Lett.~{\bf 80}, 5243

4311: (1998).

4312: % ``Real Spectra in Non-Hermitian Hamiltonians Having {\cal PT} Symmetry''

4313: 

4314: \bibitem{BBM} C.~M.~Bender, S.~Boettcher, and P.~N. Meisinger,

4315: J.~Math.~Phys.~{\bf 40}, 2201 (1999).

4316: % ``${\cal PT}$-Symmetric Quantum Mechanics''

4317: 

4318: \bibitem{SW} R.~F.~Streater and A.~S.~Wightman, {\em PCT, Spin and Statistics,

4319: and All That} (Benjamin, New York, 1964).

4320: 

4321: \bibitem{H1} E.~Caliceti, S.~Graffi, and M.~Maioli, Comm.~Math.~Phys.~{\bf 75},

4322: 51 (1980).

4323: 

4324: \bibitem{H2} D.~Bessis, private communication, 1993.

4325: 

4326: \bibitem{H3} A.~A.~Andrianov, Ann.~Phys.~{\bf 140}, 82 (1982).

4327: 

4328: \bibitem{H4} T.~Hollowood, Nucl.~Phys.~B {\bf 384}, 523 (1992).

4329: % "Solitons in affine Toda field theories" (523-540)

4330: 

4331: \bibitem{H5} F.~G.~Scholtz, H.~B.~Geyer, and F.~J.~H.~Hahne, Ann.~Phys.~{\bf

4332: 213}, 74 (1992).

4333: % "Quasi-Hermitian Operators in Quantum Mechanics and the Variational Principle"

4334: 

4335: \bibitem{BW1} C.~M.~Bender and T.~T.~Wu, Phys.~Rev.~Lett.~{\bf 21}, 406 (1968).

4336: % ``Analytic Structure of Energy Levels in a Field--Theory Model''

4337: 

4338: \bibitem{BW2} C.~M.~Bender and T.~T.~Wu, Phys.~Rev.~{\bf 184}, 1231 (1969).

4339: 

4340: \bibitem{r4} C.~M.~Bender and A.~Turbiner, Phys.~Lett.~A{\bf 173}, 442 (1993).

4341: % Analytic Continuation of Eigenvalue Problems

4342: 

4343: \bibitem{delta} C.~M.~Bender, K.~A.~Milton, S.~S.~Pinsky, and L.~M.~Simmons,

4344: Jr., J.~Math.~Phys.~{\bf 30}, 1447 (1989).

4345: 

4346: \bibitem{C1} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4347: Quantum Physics,'' Prague, June 2003, Ed.~by M.~Znojil, Czech.~J.~Phys.~{\bf 54}

4348: (2004).

4349: 

4350: \bibitem{C2} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4351: Quantum Physics II,'' Prague, June 2004, Ed.~by M.~Znojil, Czech.~J.~Phys.~{\bf

4352: 54} (2004).

4353: 

4354: \bibitem{C3} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4355: Quantum Physics III,'' Istanbul, June 2005, Ed.~by M.~Znojil,

4356: Czech.~J.~Phys.~{\bf 55} (2005).

4357: 

4358: \bibitem{C4} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4359: Quantum Physics IV,'' Stellenbosch, November 2005, Ed.~by H.~Geyer, D.~Heiss,

4360: and M.~Znojil, J.~Phys.~A: Math.~Gen.~{\bf 39} (2006).

4361: 

4362: \bibitem{C5} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4363: Quantum Physics V,'' Bologna, July 2006, Ed. by M.~Znojil, Czech.~J.~Phys.~{\bf 

4364: 56} (2006).

4365: 

4366: \bibitem{C6} Proceedings of the Workshop, ``Pseudo-Hermitian Hamiltonians in

4367: Quantum Physics VI,'' London, July 2007, Ed.~by A.~Fring, H.~Jones, and

4368: M.~Znojil, J.~Phys.~A (to be published).

4369: 

4370: \bibitem{D1} P.~Dorey, C.~Dunning and R.~Tateo, J.~Phys.~A: Math. Gen.~{\bf 34},

4371: 5679 (2001).

4372: % "SPECTRAL EQUIVALENCES, BETHE ANSATZ EQUATIONS, AND REALITY PROPERTIES IN

4373: % PT-SYMMETRIC QUANTUM MECHANICS"

4374: 

4375: \bibitem{D2} P.~Dorey, C.~Dunning and R.~Tateo, Czech.~J.~Phys.~{\bf 54}, 35

4376: (2004).

4377: % "A REALITY PROOF IN PT-SYMMETRIC QUANTUM MECHANICS"

4378: 

4379: \bibitem{D3} P.~Dorey, C.~Dunning and R.~Tateo, arXiv: hep-th/0201108.

4380: % "THE ODE/IM CORRESPONDENCE AND PT-SYMMETRIC QUANTUM MECHANICS"

4381: 

4382: \bibitem{D4} P.~Dorey, C.~Dunning, A.~Millican-Slater, and R.~Tateo,

4383: arXiv: hep-th/0309054.

4384: % "DIFFERENTIAL EQUATIONS AND THE BETHE ANSATZ"

4385: 

4386: \bibitem{D5} P.~Dorey, A.~Millican-Slater, and R.~Tateo,

4387: J.~Phys.A: Math.~Gen.~{\bf 38}, 1305 (2005).

4388: % "BEYOND THE WKB APPROXIMATION IN PT-SYMMETRIC QUANTUM MECHANICS"

4389: 

4390: \bibitem{SHIN} K.~C.~Shin, J.~Math.~Phys.~{\bf 42}, 2513 (2001);

4391: % "On the eigenproblems of PT-symmetric oscillators"

4392: Commun.~Math.~Phys.~{\bf 229}, 543 (2002);

4393: % "On the reality of the eigenvalues for a class of PT-symmetric oscillators"

4394: J.~Phys.~A: Math.~Gen~{\bf 37}, 8287 (2004);

4395: % "On the shape of spectra for non-self-adjoint periodic Schroedinger operators"

4396: J.~Math.~Phys.~{\bf 46}, 082110 (2005);

4397: % "The potential (iz)^m generates real eigenvalues only, under symmetric

4398: % rapid decay conditions"

4399: J.~Phys.~A: Math.~Gen~{\bf 38}, 6147 (2005).

4400: % "Eigenvalues of PT-symmetric oscillators with polynomial potentials"

4401: 

4402: \bibitem{PP1} F.~Pham and E.~Delabaere, Phys.~Lett.~A{\bf 250}, 25-28 (1998).

4403: % ``Eigenvalues of Complex Hamiltonians with ${\cal PT}$ Symmetry. I,''

4404: 

4405: \bibitem{DD2} E.~Delabaere and F.~Pham, Phys.~Lett.~A{\bf 250}, 29-32 (1998);

4406: % ``Eigenvalues of Complex Hamiltonians with ${\cal PT}$ Symmetry. II,''

4407: E.~Delabaere and D.~T.~Trinh, J.~Phys.~A: Math.~Gen.~{\bf 33}, 8771-8796 (2000).

4408: % ``Spectral analysis of the cubic oscillator,''

4409: 

4410: \bibitem{TRINH} D.~T.~Trinh, PhD Thesis, University of Nice-Sophia Antipolis

4411: (2002).

4412: 

4413: \bibitem{WWWW} S.~Weigert, J.~Opt.~B {\bf 5}, S416 (2003);

4414: % "PT-symmetry and its spontaneous breakdown explained by anti-linearity"

4415: J.~Phys.~A: Math.~Gen.~{\bf 39}, 235 (2006);

4416: % "An Algorithmic Test for Diagonalizability of Finite-Dimensional PT-Invariant

4417: % Systems"

4418: J.~Phys.~A: Math.~Gen.~{\bf 39}, 10239 (2006).

4419: % "Detecting Broken PT-Symmetry"

4420: \bibitem{BO} Bender~C~M and Orszag~S~A 1978 {\em Advanced Mathematical

4421: Methods for Scientists and Engineers} (McGraw Hill, New York, 1978).

4422: 

4423: \bibitem{WWWW1} S.~Weigert, Phys.~Rev.~A {\bf 68}, 062111 (2003).

4424: % "Completeness and Orthonormality in PT-symmetric Quantum Systems"

4425: 

4426: \bibitem{GEY} F.~G.~Scholtz and H.~B.~Geyer, Phys.~Lett~B{\bf 634}, 84 (2006);

4427: % "Operator equations and Moyal products metrics in quasi-hermitian quantum

4428: % mechanics"

4429: F.~G.~Scholtz and H.~B.~Geyer, arXiv: quant-ph/0602187.

4430: % Moyal products -- a new perspective on quasi-hermitian quantum mechanics"

4431: 

4432: \bibitem{DY} F.~J.~Dyson, Phys.~Rev.~{\bf 85}, 631 (1952).

4433: 

4434: \bibitem{SQ} C.~M.~Bender, S.~Boettcher, H.~F.~Jones, and V.~M.~Savage,

4435: J.~Phys.~A: Math.~Gen.~{\bf 32}, 6771 (1999).

4436: 

4437: \bibitem{VAN} C.~M.~Bender, F.~Cooper, P.~N.~Meisinger, and V.~M.~Savage,

4438: Phys.~Lett.~A{\bf 259}, 224 (1999).

4439: 

4440: \bibitem{HANDY} C.~R.~Handy, J.~Phys.~A: Math.~Gen.~{\bf 34}, 5065 (2001);

4441: % "Generating converging bounds to the (complex) discrete states of the

4442: % $P^2+iX^3+iX$ Hamiltonian"

4443: C.~R.~Handy, D.~Khan, S.~Okbagabir, and T.~Yarahmad, J.~Phys.~A: Math.~Gen.~{\bf

4444: 36}, 1623 (2003);

4445: % "Moment problem quantization within a generalized scalet Wigner (autoscaling)

4446: % transform representation"

4447: Z.~Yan and C.~R.~Handy, J.~Phys.~A: Math.~Gen.~{\bf 34}, 9907 (2003).

4448: % "Extension of a spectral bounding method to the PT invariant states of

4449: % the -(iX)**N non-Hermitian potential"

4450: 

4451: \bibitem{BERRY} Z.~Ahmed, C.~M.~Bender, and M.~V.~Berry, J.~Phys.~A:

4452: Math.~Gen.~{\bf 38}, L627 (2005).

4453: % "Reflectionless Potentials and $\cal PT$ Symmetry"

4454: 

4455: \bibitem{BG} V.~Buslaev and V.~Grecchi, J.~Phys.~A: Math.~Gen.~{\bf 26}, 5541

4456: (1993).

4457: 

4458: \bibitem{JM} H.~F.~Jones and J.~Mateo, Phys.~Rev.~D {\bf 73}, 085002 (2006).

4459: 

4460: \bibitem{BBCJM} C.~M.~Bender, D.~C.~Brody, J.-H.~Chen, H.~F. Jones,

4461: K.~A.~Milton, and M.~C.~Ogilvie, Phys. Rev. D {\bf 74}, 025016 (2006).

4462: 

4463: \bibitem{BOUND} C.~M.~Bender, S.~Boettcher, H.~F.~Jones, P.~N.~Meisinger,

4464: and M.~\d{S}im\d{s}ek, Phys.~Lett.~A{\bf 291}, 197 (2001).

4465: 

4466: \bibitem{CL1} A.~Nanayakkara, Czech.~J.~Phys.~{\bf 54}, 101 (2004) and

4467: J.~Phys.~A: Math.~Gen.~{\bf 37}, 4321 (2004).

4468: 

4469: \bibitem{CL2} C.~M.~Bender, J.-H.~Chen, D.~W.~Darg, and K.~A.~Milton,

4470: J.~Phys.~A: Math.~Gen.~{\bf 39}, 4219 (2006).

4471: % "Classical Trajectories for Complex Hamiltonians"

4472: 

4473: \bibitem{CL3} C.~M.~Bender, D.~D.~Holm, and D.~W.~Hook,

4474: J.~Phys.~A: Math.~Theor.~{\bf 40}, F81 (2007).

4475: % Pendulum

4476: 

4477: \bibitem{CL4} C.~M.~Bender and D.~W.~Darg, submitted.

4478: % Spontaneous Breaking of Classical PT Symmetry

4479: 

4480: \bibitem{CL5} C.~M.~Bender, D.~D.~Holm, and D.~W.~Hook, in preparation.

4481: % Euler rotation

4482: 

4483: \bibitem{BIOX} T.~Curtright and L.~Mezincescu, arXiv: quant-ph/0507015.

4484: % "BIORTHOGONAL QUANTUM SYSTEMS"

4485: 

4486: \bibitem{BBJ1} C. M. Bender, D. C. Brody, and H. F. Jones, Phys.~Rev.~Lett.~{\bf

4487: 89}, 270401 (2002) [Erratum: {\em Ibid.}~{\bf 92}, 119902 (2004)].

4488: % ``Complex Extension of Quantum Mechanics''

4489: 

4490: \bibitem{rr2} G.~A.~Mezincescu, J.~Phys.~A: Math.~Gen.~{\bf 33}, 4911 (2000).

4491: 

4492: \bibitem{rr3} C.~M.~Bender and Q.~Wang, J.~Phys.~A: Math.~Gen.~{\bf 34},

4493: 3325 (2001).

4494: 

4495: \bibitem{rf17} P.~A.~M.~Dirac, Proc.~R.~Soc.~London A {\bf 180}, 1 (1942).

4496: 

4497: \bibitem{Z4} A.~Mostafazadeh and A.~Batal, J.~Phys.~A: Math.~Gen.~{\bf 37},

4498: 11645 (2004).

4499: 

4500: \bibitem{P7x} A.~Mostafazadeh, J.~Phys.~A: Math.~Gen.~{\bf 38}, 3213 (2005).

4501: 

4502: \bibitem{Jobs} H.~F.~Jones, J.~Phys.~A: Math.~Gen.~{\bf 38}, 1741 (2005).

4503: % "On pseudo-Hermitian Hamiltonians and their Hermitian counterparts"

4504: 

4505: \bibitem{P1} W.~Pauli, Rev.~Mod.~Phys.~{\bf 15}, 175 (1943).

4506: 

4507: \bibitem{P2} S.~N.~Gupta, Phys.~Rev.~{\bf 77}, 294 (1950) and

4508: Proc.~Phys.~Soc.~London {\bf 63}, 681 (1950).

4509: 

4510: \bibitem{P3} K.~Bleuler, Helv.~Phys.~Act.~{\bf 23}, 567 (1950).

4511: 

4512: \bibitem{P4} E.~C.~G.~Sudarshan, Phys.~Rev.~{\bf 123}, 2183 (1961).

4513: 

4514: \bibitem{P5} T.~D.~Lee and G.~C.~Wick, Nucl.~Phys.~B {\bf 9}, 209 (1969).

4515: 

4516: \bibitem{M1} A.~Mostafazadeh, J.~Math.~Phys.~{\bf 43}, 205 (2002).

4517: 

4518: \bibitem{P6} A.~Mostafazadeh, J.~Math.~Phys.~{\bf 43}, 2814, 3944, and 6343

4519: (2002) [Erratum: {\em Ibid.}~{\bf 44}, 943 (2003)]; {\bf 44}, 974 (2003).

4520: 

4521: \bibitem{P7} A.~Mostafazadeh, Nucl.~Phys.~B {\bf 640}, 419 (2002).

4522: 

4523: \bibitem{P8} B.~Bagchi and C.~Quesne, Phys.~Lett.~A{\bf 301}, 173 (2002);

4524: % "Pseudo-hermiticity, weak pseudo-Hermiticity and \eta-orthogonality condition"

4525: Z.~Ahmed, Phys.~Lett. A{\bf 294}, 287 (2002);

4526: % "Pseudo-Hermiticity of Hamiltonians under gauge-like transformation:

4527: % real spectrum of non-Hermitian Hamiltonians"

4528: G.~S.~Japaridze, J.~Phys.~A: Math.~Gen.~{\bf 35}, 1709 (2002);

4529: % "Space of state vectors in PT-symmetric quantum mechanics"

4530: Z.~Ahmed, Phys.~Lett.~A{\bf 308}, 140 (2003) and {\bf 310}, 139 (2003);

4531: Z.~Ahmed and S.~R.~Jain, Phys.~Rev.~E {\bf 67}, 045106 (2003) and

4532: J.~Phys.~A: Math.~Gen.~{\bf 36}, 3349 (2003);

4533: Z.~Ahmed, J.~Phys.~A: Math.~Gen.~{\bf 36}, 9711 and 10325 (2003);

4534: % "C-, PT- and CPT-invariance of pseudo-Hermitian Hamiltonians"

4535: % "Pseudo-reality and pseudo-adjointness of Hamiltonians"

4536: A.~Blasi, G.~Scolarici and L.~Solombrino, J.~Phys.~A: Math.~Gen.~{\bf 37}, 4335

4537: (2004);

4538: % "PSEUDO-HERMITIAN HAMILTONIANS, INDEFINITE INNER PRODUCT SPACES & THEIR

4539: % SYMMETRIES"

4540: B.~Bagchi, C.~Quesne, and R.~Roychoudhury, J.~Phys.~A: Math.~Gen.~{\bf 38}, L647

4541: (2005).

4542: % "Pseudo-Hermiticity and some consequences of a generalized quantum condition"

4543: 

4544: \bibitem{Swan} M.~S.~Swanson, J.~Math.~Phys.~{\bf 45}, 585 (2004).

4545: % "Transition elements for a non-Hermitian quadratic Hamiltonian"

4546: 

4547: \bibitem{AJP} C.~M.~Bender, D.~C.~Brody, and H.~F.~Jones, Am.~J.~Phys.~{\bf 71},

4548: 1095 (2003).

4549: % ``Must a Hamiltonian be Hermitian?''

4550: 

4551: \bibitem{BMW} C.~M.~Bender, P.~N.~Meisinger, and Q.~Wang, J.~Phys.~A: Math.

4552: Gen. {\bf 36}, 1973 (2003).

4553: % ``Perturbative Calculation of the Hidden Symmetry Operator in

4554: % PT-Symmetric Quantum Mechanics''

4555: 

4556: \bibitem{rf20.5} C.~M.~Bender and H.~F.~Jones, Phys.~Lett.~A{\bf 328}, 102

4557: (2004).

4558: 

4559: \bibitem{rf23} C.~M.~Bender and G.~V.~Dunne Phys. Rev. D {\bf 40}, 2739 and

4560: 3504 (1989).

4561: 

4562: \bibitem{BDMS} C.~M.~Bender, G.~V.~Dunne, P.~N.~Meisinger, and

4563: M.~\d{S}im\d{s}ek, Phys.~Lett.~A{\bf 281}, 311-316 (2001).

4564: % ``Quantum Complex H\'enon-Heiles Potentials''

4565: 

4566: \bibitem{BBRR} C.~M.~Bender, J.~Brod, A.~Refig, and M.~E.~Reuter, J.~Phys.~A:

4567: Math.~Gen. {\bf 37}, 10139-10165 (2004).

4568: % ``The C Operator in PT-Symmetric Quantum Theories''

4569: 

4570: \bibitem{BBJ2} C.~M.~Bender, D.~C.~Brody, and H.~F.~Jones, Phys.~Rev.~Lett.~{\bf

4571: 93}, 251601 (2004).

4572: % ``Scalar Quantum Field Theory with Cubic Interaction''

4573: 

4574: \bibitem{BBJ3} C.~M.~Bender, D.~C.~Brody, and H.~F.~Jones, Phys.~Rev.~D {\bf

4575: 70}, 025001 (2004).

4576: % ``Extension of PT-Symmetric Quantum Mechanics to Quantum Field Theory with

4577: % Cubic Interaction''

4578: 

4579: \bibitem{Z1} M.~Znojil, Phys.~Lett.~A{\bf 285}, 7 (2001).

4580: 

4581: \bibitem{Z2} M.~Znojil and G.~L\'evai, Mod.~Phys.~Lett.~A {\bf 16}, 2273 (2001).

4582: 

4583: \bibitem{Z3} B.~Bagchi, S.~Mallik, and C.~Quesne, Mod.~Phys.~Lett.~A{\bf17},

4584: 1651 (2002).

4585: 

4586: \bibitem{Z5} M.~Znojil, J.~Math.~Phys.~{\bf 46}, 062109 (2005).

4587: 

4588: \bibitem{BT} C.~M.~Bender and B.~Tan, J.~Phys.~A: Math.~Gen.~{\bf 39}, 1945

4589: (2006).

4590: % ``Calculation of the Hidden Symmetry Operator for a PT-Symmetric Square Well''

4591: 

4592: \bibitem{M2} A.~Mostafazadeh, J.~Phys.~A: Math.~Gen.~{\bf 36}, 7081 (2003).

4593: 

4594: \bibitem{k1} C.~M.~Bender, J.~Chen, K.~A.~Milton, J.~Phys.~A: Math.~Gen.~{\bf

4595: 39}, 1657 (2006).

4596: % ``$PT-Symmetric Versus Hermitian Formulation of Quantum Mechanics''

4597: 

4598: \bibitem{A1} T.~T.~Wu, Phys.~Rev.~{\bf 115}, 1390 (1959).

4599: 

4600: \bibitem{A3} M.~E.~Fisher, Phys.~Rev.~Lett.~{\bf 40}, 1610 {1978}.

4601: 

4602: \bibitem{A4} J.~L.~Cardy, Phys.~Rev.~Lett.~{\bf 54}, 1345 {1985}.

4603: 

4604: \bibitem{A5} J.~L.~Cardy and G.~Mussardo, Phys.~Lett.~B{\bf 225}, 275 {1989}

4605: 

4606: \bibitem{A6} A.~B.~Zamolodchikov, Nucl.~Phys.~B {\bf 348}, 619 (1991).

4607: 

4608: \bibitem{A7} R.~Brower, M.~Furman, and M.~Moshe, Phys.~Lett.~B{\bf 76}, 213

4609: (1978).

4610: 

4611: \bibitem{A8} B.~Harms, S.~Jones, and C.-I Tan, Nucl.~Phys.~{\bf 171}, 392 (1980)

4612: and Phys.~Lett.~B{\bf 91}, 291 (1980).

4613: 

4614: \bibitem{G1} U.~Guenther, F.~Stefani, and M.~Znojil, J.~Math.~Phys.~{\bf 46},

4615: 063504 (2005).

4616: 

4617: \bibitem{G2} U.~Guenther, B.~F.~Samsonov, and F.~Stefani, J.~Phys.~A:

4618: Math.~Theor.~{\bf 40}, F169 (2007).

4619: 

4620: \bibitem{Frig} C.~F.~de M.~Faria and A.~Fring, to appear in Laser Phys.,

4621: arXiv: quant-ph/0609096.

4622: % "NON-HERMITIAN HAMILTONIANS WITH REAL EIGENVALUES COUPLED TO ELECTRIC FIELDS:

4623: % FROM THE TIME-INDEPENDENT TO THE TIME DEPENDENT QUANTUM MECHANICAL FORMULATION

4624: 

4625: \bibitem{A9} A.~G.~Ushveridze, {\sl Quasi-Exactly Solvable Models in Quantum

4626: Mechanics} (Institute of Physics, Bristol, 1993) and references therein.

4627: 

4628: \bibitem{BD} C.~M.~Bender and G.~V.~Dunne, J.~Math.~Phys.~{\bf 37}, 6 (1996) and

4629: C.~M.~Bender, G.~V.~Dunne, and M.~Moshe, Phys.~Rev.~A {\bf 55}, 2625 (1997).

4630: 

4631: \bibitem{A10} A.~V.~Turbiner, Sov.~Phys., JETP {\bf 67}, 230 (1988),

4632: Contemp.~Math.~{\bf 160}, 263 (1994), and M.~A.~Shifman, Contemp.~Math.~{\bf

4633: 160}, 237 (1994).

4634: 

4635: \bibitem{A11} C.~M.~Bender and S.~Boettcher, J.~Phys.~A: Math.~Gen.~{\bf 31},

4636: L273 (1998).

4637: % "Quasi-Exactly Solvable Quartic Potential"

4638: 

4639: \bibitem{MONOU} C.~M.~Bender and M.~Monou, J.~Phys.~A: Math.~Gen.~{\bf 38}, 2179

4640: (2005).

4641: % ``New Quasi-Exactly Solvable Sextic Polynomial Potentials''

4642: 

4643: \bibitem{MVB} M.~V.~Berry and D.~H.~J.~O'Dell, J.~Phys.~A: Math.~Gen.~{\bf 31},

4644: 2093 (1998).

4645: % "Diffraction by volume gratings with imaginary potentials"

4646: 

4647: \bibitem{A12} C.~M.~Bender, G.~V.~Dunne, and P.~N.~Meisinger, Phys.~Lett.~A{\bf

4648: 252}, 272 (1999).

4649: % ``Complex Periodic Potentials with Real Band Structure''

4650: 

4651: \bibitem{ZZ} We hope that some of the delicate experiments performed by

4652: A.~Zeilinger on models of this type (private communication, M.~V.~Berry) will

4653: eventually be able to verify by direct observation the theoretical band-edge

4654: predictions that are illustrated in Figs.~\ref{f24} and \ref{f25}.

4655: 

4656: \bibitem{KS1} A.~Khare and U.~P.~Sukhatme, Phys.~Lett.~A{\bf 324}, 406 (2004).

4657: % "Analytically Solvable PT-invariant Periodic Potentials"

4658: 

4659: \bibitem{KS2} A.~Khare and U.~P.~Sukhatme, J.~Math.~Phys.~{\bf 46}, 082106

4660: (2005).

4661: % "PT-Invariant Periodic Potentials With a Finite Number of Band Gaps"

4662: 

4663: \bibitem{KS3} A.~Khare and U.~P.~Sukhatme, J.~Phys.~A: Math.~Gen~{\bf 39}, 10133

4664: (2006).

4665: % "Periodic Potentials and PT Symmetry"

4666: 

4667: \bibitem{KS4} A.~Khare and U.~P.~Sukhatme, J.~Math.~Phys.~{\bf 47}, 062103

4668: (2006).

4669: % "Complex Periodic Potentials With a Finite Number of Band Gaps"

4670: 

4671: \bibitem{BR1} C.~M.~Bender, D.~C.~Brody, H.~F. Jones, and B.~Meister,

4672: Phys.~Rev.~Lett.~{\bf 98}, 040403 (2007).

4673: 

4674: \bibitem{rrr2} D.~C.~Brody and D.~W.~Hook, J.~Phys.~A: Math.~Gen.~{\bf 39} L167

4675: (2006).

4676: 

4677: \bibitem{rrr4} J.~Anandan and Y.~Aharonov, Phys.~Rev.~Lett. {\bf 65}, 1697

4678: (1990).

4679: 

4680: \bibitem{BM7} C.~M.~Bender and K.~A.~Milton, Phys.~Rev.~D {\bf 57}, 3595 (1998).

4681: % ``Model of Supersymmetric Quantum Field Theory with Broken Parity Symmetry''

4682: 

4683: \bibitem{ZX} M.~Znojil, J.~Phys.~A: Math.~Gen.~{\bf 33}, L61 (2000);

4684: % "Shape invariant potentials with PT symmetry"

4685: B.~Bagchi, F.~Cannata, C.~Quesne, Phys.~Lett.~A{\bf 69}, 79 (2000);

4686: % "PT-symmetric sextic potentials"

4687: B.~Bagchi, S.~Mallik, C.~Quesne, Int.~J.~Mod.~Phys.~A {\bf 16}, 2859 (2001);

4688: % "Generating Complex Potentials with Real Eigenvalues in Supersymmetric Quantum

4689: % Mechanics"

4690: G.~L\'evai and M.~Znojil, Int.~J.~Mod.~Phys.~A {\bf 17} 51 (2002);

4691: % "The interplay of supersymmetry and PT symmetry in quantum

4692: % mechanics: a case study for the Scarf II potential"

4693: B.~Bagchi and C.~Quesne, Mod.~Phys.~Lett.~A {\bf 17}, 463 (2002);

4694: % "PT-symmetric non-polynomial oscillators and hyperbolic potential

4695: % with two known real eigenvalues in a SUSY framework"

4696: A.~Sinha, G.~L\'evai, P.~Roy, Phys.~Lett.~A{\bf 322}, 78 (2004); 

4697: % "Image symmetry of a conditionally exactly solvable potential"

4698: A.~Sinha and P.~Roy, J.~Phys.~A: Math.~Gen.~{\bf 37}, 2509 (2004);

4699: % "New Exactly Solvable Isospectral Partners for PT Symmetric Potentials"

4700: E.~Caliceti, F.~Cannata, M.~Znojil, A.~Ventura, Phys.~Lett.~A{\bf 335}, 26

4701: (2005);

4702: % "Construction of PT-asymmetric non-Hermitian Hamiltonians with CPT-symmetry"

4703: B.~Bagchi, A.~Banerjee, E.~Caliceti, F.~Cannata, H.~B.~Geyer, C.~Quesne, and

4704: M.~Znojil, Int.~J.~Mod.~Phys.~A {\bf 20} 7107 (2005);

4705: % "CPT-conserving Hamiltonians and their nonlinear

4706: % supersymmetrization using differential charge-operators C"

4707: B.~F.~Samsonov, J.~Phys.~A: Math.~Gen.~{\bf 38}, L397 (2005);

4708: % "SUSY transformations between diagonalizable and non-diagonalizable

4709: % Hamiltonians"

4710: A.~Gonzalez-Lopez and T.~Tanaka, J.~Phys.~A: Math.~Gen.~{\bf 39}, 3715 (2006);

4711: T.~Curtright, L.~Mezincescu, and D.~Schuster, J.~Math.~Phys.~(to appear),

4712: arXiv: quant-ph/0603170.

4713: % "Supersymmetric Biorthogonal Quantum Systems"

4714: 

4715: \bibitem{D6} P.~Dorey, C.~Dunning and R.~Tateo, J.~Phys.~A: Math.~Gen.~{\bf 34},

4716: L391 (2001).

4717: % "SUPERSYMMETRY AND THE SPONTANEOUS BREAKDOWN OF PT SYMMETRY"

4718: 

4719: \bibitem{HN} N.~Hatano and D.~R.~Nelson, Phys.~Rev.~Lett.~{\bf 77}, 570 (1996),

4720: and Phys.~Rev.~B {\bf 56}, 8651 (1997).

4721: 

4722: \bibitem{RS} R.~Scalettar (private communication).

4723: 

4724: \bibitem{MH1} M.~Henkel, in {\em Classical and Quantum Nonlinear Integrable

4725: Systems: Theory and Applications} (Institute of Physics Publishing, Bristol,

4726: 2003), ed. by A. Kundu.

4727: % ``Reaction-diffusion processes and their connection with integrable

4728: % quantum spin chains''

4729: 

4730: \bibitem{MH2} F.~C.~Alcaraz, M.~Droz, M.~Henkel and V.~Rittenberg,

4731: Ann.~Phys.~{\bf 230}, 250 (1994).

4732: 

4733: \bibitem{HDL} K.~E.~Hibberd, C.~Dunning, and J.~Links, Nucl.~Phys.~B {\bf 748},

4734: 458 (2006).

4735: % "A BETHE ANSATZ SOLVABLE MODEL FOR SUPERPOSITIONS OF COOPER PAIRS AND

4736: % CONDENSED MOLECULAR BOSONS"

4737: 

4738: \bibitem{BBCW} C.~M.~Bender, S.~F.~Brandt, J.-H.~Chen, and Q.~Wang, Phys.~Rev.~D

4739: {\bf 71}, 065010 (2005).

4740: % ``The C Operator in PT-Symmetric Quantum Field Theory Transforms as a Lorentz

4741: % Scalar''

4742: 

4743: \bibitem{L1} T.~D.~Lee, Phys.~Rev.~{\bf 95}, 1329 (1954).

4744: 

4745: \bibitem{L2} G.~K\"all\'en and W.~Pauli, Mat.-Fys.~Medd.~{\bf 30}, No.~7 (1955).

4746: 

4747: \bibitem{L3} S.~S.~Schweber, {\em An Introduction to Relativistic Quantum Field

4748: Theory} (Row, Peterson and Co., Evanston, 1961), Chap.~12.

4749: 

4750: \bibitem{LEE} C.~M.~Bender, S.~F.~Brandt, J.-H.~Chen, and Q.~Wang, Phys.~Rev.~D

4751: {\bf 71}, 025014 (2005).

4752: % ``Ghost Busting: $\cal{PT}$-Symmetric Interpretation of the Lee Model''

4753: 

4754: \bibitem{KLE} F.~Kleefeld: hep-th/0408028 and hep-th/0408097.

4755: 

4756: \bibitem{CM} T.~Curtright, E.~Ivanov, L.~Mezincescu, and P.~K.~Townsend, arXiv:

4757: hep-th/0612300;

4758: % "Planar Super-Landau Models Revisited"

4759: T.~Curtright and A.~Veitia, arXiv: quant-ph/0701006.

4760: % "QUASI-HERMITIAN QUANTUM MECHANICS IN PHASE SPACE"

4761: 

4762: \bibitem{SMIL} E.~A.~Ivanov and A.~V.~Smilga, manuscript in preparation

4763: (private communication).

4764: 

4765: \bibitem{AF} C.~M.~Bender, K.~A.~Milton, and V.~M.~Savage, Phys.~Rev.~D~{\bf

4766: 62}, 85001 (2000).

4767: % ``Solution of Schwinger-Dyson Equations for ${\cal PT}$-Symmetric Quantum

4768: % Field Theory''

4769: 

4770: \bibitem{Sy} K.~Symanzik, Springer Tracts Mod.~Phys.~{\bf 57}, 222 (1971);

4771: Commun.~Math.~Phys.~{\bf 23}, 49 (1971); Nuovo Cim.~{\bf 6}, 77 (1973).

4772: 

4773: \bibitem{KKK} F.~Kleefeld, J.~Phys.~A: Math.~Gen.~{\bf 39}, L9 (2006).

4774: 

4775: \bibitem{BMY} C.~M.~Bender, P.~Meisinger, and H.~Yang, Phys.~Rev.~D {\bf 63},

4776: 45001 (2001).

4777: % ``Calculation of the One-Point Green's Function for a $-g\phi^4$ Quantum Field

4778: % Theory''

4779: 

4780: \bibitem{JMR} H.~F.~Jones, J.~Mateo, and R.~J.~Rivers, Phys.~Rev.~D {\bf 74},

4781: 125022 (2006).

4782: % "ON THE PATH-INTEGRAL DERIVATION OF THE ANOMALY FOR THE HERMITIAN EQUIVALENT

4783: % OF THE COMPLEX $PT$-SYMMETRIC QUARTIC HAMILTONIAN"

4784: 

4785: \bibitem{MO} P.~N.~Meisinger and M.~C.~Ogilvie, arXiv: hep-th/0701207.

4786: % "PT-SYMMETRIC MATRIX QUANTUM MECHANICS"

4787: 

4788: \bibitem{QED2} K.~A.~Milton, Czech.~J.~Phys.~{\bf 54}, 85 (2004).

4789: 

4790: \bibitem{QED1} C.~M.~Bender and K.~A.~Milton, J.~Phys.~A: Math.~Gen.~{\bf 32},

4791: L87 (1999).

4792: % ``A Nonunitary Version of Massless Quantum Electrodynamics Possessing a 

4793: % Critical Point''

4794: 

4795: \bibitem{QED3} C.~M.~Bender, I.~Cavero-Pelaez, K.~A.~Milton, and K.~V.~Shajesh,

4796: Phys.~Lett.~B{\bf 613}, 97 (2005).

4797: % ``$\cal{PT}$-Symmetric Quantum Electrodynamics''

4798: 

4799: \bibitem{BJR} C.~M.~Bender, H.~F.~Jones, and R.~J.~Rivers, Phys.~Lett.~B{\bf

4800: 625}, 333-340 (2005).

4801: % ``Dual $\cal PT$-Symmetric Quantum Field Theories''\br

4802: 

4803: \bibitem{DUAL} E.~Abdalla, M.~G.~B.~Abdalla and K.~D.~Rothe, {\em

4804: Non-perturbative Methods in 2 Dimensional Quantum Field Theory}, (World

4805: Scientific, New York, 1991).

4806: 

4807: \bibitem{ZNOJ} M.~F.~Fern\'andez, R.~Guardiola, J.~Ros and M.~Znojil,

4808: J.~Phys.~A: Math.~Gen.~{\bf 32}, 3105 (1999).

4809: % ``A family of complex potentials with real spectrum''

4810: 

4811: \bibitem{ADS} T.~Hertog and G.~T.~Horowitz, JHEP {\bf 04}, 005 (2005).

4812: % "Holographic description of AdS cosmologies"

4813: 

4814: \bibitem{DS1} E.~Witten, arXiv: hep-th/016109.

4815: % "Quantum gravity in de Sitter space"

4816: 

4817: \bibitem{DS2} R.~Bousso, A.~Maloney, and A.~Strominger, Phys.~Rev.~D {\bf 65},

4818: 104039 (2000).

4819: % "Conformal vacua and entropy in de Sitter space"

4820: 

4821: \bibitem{Moffat} J.~W.~Moffat, Phys.~Lett.~B{\bf 627}, 9 (2005) and

4822: % "Charge conjugation invariance of the vacuum and the cosmological constant

4823: % problem"

4824: arXiv: hep-th/0610162.

4825: % "POSITIVE AND NEGATIVE ENERGY SYMMETRY AND THE COSMOLOGICAL CONSTANT PROBLEM"

4826: 

4827: \bibitem{r8} R.~Camassa and D.~D.~Holm, Phys.~Rev.~Lett.~{\bf 71}, 1661 (1993).

4828: 

4829: \bibitem{KDV} C.~M.~Bender, D.~C.~Brody, J.-H.~Chen, and E.~Furlan,

4830: J.~Phys.~A.: Math.~Theor.~{\bf 40}, F153 (2007).

4831: % ``$\cal{PT}$-Symmetric Extension of the Korteweg-de Vries Equation''

4832: 

4833: \bibitem{Fring} A.~Fring, arXiv: math-ph/0701036

4834: % "PT-SYMMETRIC DEFORMATIONS OF THE KORTEWEG-DE VRIES EQUATION"

4835: 

4836: \end{thebibliography}

4837: \end{document}

4838: 

4839: