1: % ODE.tex
2:
3: \resection{Ordinary differential equations and functional relations}
4: \label{funode}
5:
6: %
7: Surprisingly, the functional equations found in the last section
8: also govern the problems in
9: $\PT$-symmetric quantum mechanics discussed in section~\ref{prelude}.
10: To understand how this comes about, we must first return to the subject
11: of $\PT$-symmetric eigenvalue problems and their generalisations
12: in a little more depth.
13:
14: \subsection{General eigenvalue problems in the complex plane}
15: \label{compeig}
16: We begin with one piece of unfinished business
17: from section~\ref{prelude}:
18: what goes wrong with the Bender-Boettcher problem at
19: $M{=}2$, and what can be done to resolve it?
20: In figures \ref{fig1} and \ref{fig2}, the energy levels
21: continued smoothly past $M{=}2$, but in fact this can only be
22: achieved by implementing a suitable distortion of the problem as
23: originally posed. Consider the situation precisely at $M{=}2$\,:
24: the Hamiltonian is $p^2-x^4$, an `upside-down' quartic oscillator, and a
25: simple WKB analysis (about which more shortly) shows, instead of
26: the exponential growth or decay more generally found, wavefunctions
27: behaving as $x^{-1}\exp(\pm ix^3/3)$ as $x$ tends to plus or minus
28: infinity along the real axis.
29: {\em All}\/ solutions thus decay, albeit algebraically, and this
30: complicates matters significantly. The problem moves from what is called
31: the limit-point to the limit-circle case (see \cite{REL,rich}),
32: and additional boundary conditions should be imposed at infinity if
33: the spectrum is to be discrete.
34:
35: While interesting in its own right, this is clearly
36: not the right eigenproblem
37: if we wish to find a smooth continuation from the region $M<2$.
38: Instead, it is necessary to enlarge the perspective and
39: treat $x$ as a genuinely complex variable. This
40: has been discussed by many authors, and is particularly emphasised in
41: the book by Sibuya \cite{Sha}\,, though
42: the treatment which follows is perhaps closer to that of
43: \cite{BT,BB}.
44:
45: The key is to examine the behaviour of solutions as $|x|\to\infty$
46: along a general ray in the complex plane, in spite of the fact that
47: the only rays involved in the problem as initially posed were
48: the positive and negative real axes.
49: The WKB approximation tells us that
50: \eq
51: \psi(x)\sim P(x)^{-1/4}\,e^{\pm\int^x\!\sqrt{P(t)}dt}
52: \label{wkba}
53: \en
54: as $|x|\to\infty$, with $P(x)=-(ix)^{2M}+l(l{+}1)x^{-2}-E$. (This is
55: easily derived by
56: substituting $\psi(x)=f(x)e^{g(x)}$ into the ODE.)
57: Since the problem was set up with a branch cut running up the
58: positive-imaginary axis, it is natural to define general rays
59: in the complex plane by setting $x=\rho e^{i\theta}\!/i$ with $\rho$
60: real, as illustrated in figure~\ref{ray}.
61: %
62: \smallskip
63:
64:
65: \begin{figure}[ht]
66: \begin{center}
67: \includegraphics[width=0.55\linewidth]{ray.eps}
68: \end{center}
69: \caption{A ray in the complex $x$-plane.\label{ray}}
70: \end{figure}
71: %
72:
73: \noindent
74: For $M>1$, the leading asymptotic predicted by (\ref{wkba})
75: is not changed if $P(x)$ is replaced by $-(ix)^{2M}$, and
76: substituting into the WKB formula we see two possible
77: behaviours, as expected of a second-order ODE:
78: \eq
79: \psi_{\pm}\sim P^{-1/4}\exp\left[\pm\fract{1}{M{+}1}e^{i\theta(1{+}M)}
80: \rho^{1{+}M}\right]\,.
81: \en
82: For most values of $\theta$, one of these solutions
83: will
84: be exponentially growing, the other exponentially decaying. But whenever
85: $\Re e[e^{i\theta(1{+}M)}]=0$, the two solutions swap r\^oles and there
86: is a moment when both oscillate, and neither dominates the other.
87: The relevant values of $\theta$ are
88: \eq
89: \theta=
90: \pm\frac{\pi}{2M{+}2}~,~
91: \pm\frac{3\pi}{2M{+}2}~,~
92: \pm\frac{5\pi}{2M{+}2}~,~\dots~.
93: \en
94: (Confusingly, the rays that these values of $\theta$ define
95: are sometimes called `anti-Stokes lines', and sometimes `Stokes
96: lines'. See, for example, \cite{berry}.)
97:
98:
99: Whenever one of these lines lies along the positive
100: or negative real axis, the eigenvalue
101: problem as originally stated becomes much more delicate.
102: %
103: Increasing $M$ from $1$, the first time that this happens
104: is at $M=2$, the case of the
105: upside-down quartic potential discussed above. But now we see that the
106: problem arose because the line along which the
107: wavefunction was being considered, namely the real axis, happened to
108: coincide with an anti-Stokes line\footnote{as just mentioned, some
109: would have called this a
110: Stokes line.}.
111: We also see how the problem can be averted. Since all functions
112: involved are analytic, there is nothing to stop us from examining
113: the wavefunction along some other contour in the complex plane. In
114: particular, before $M$ reaches $2$, the two ends of the contour can be bent
115: downwards from the real axis without changing the spectrum, so long as
116: their asymptotic directions do not cross any anti-Stokes lines in the
117: process.
118: Having thus distorted the original problem, $M$ can be increased
119: through $2$ without any difficulties.
120: The situation for $M$ just bigger
121: than $2$ is illustrated in figure \ref{sectors},
122: with the anti-Stokes lines shown dashed and the wiggly line a
123: curve along which the wavefunction $\psi(x)$ can be
124: defined.
125: % %
126: \begin{figure}[ht]
127: \begin{center}
128: \includegraphics[width=0.55\linewidth]{sectors.eps}
129: \end{center}
130: \caption{A possible wavefunction contour for $M>2$.\label{sectors}}
131: \end{figure}
132: %
133:
134: The wedges between the dashed lines
135: are called {\em Stokes
136: sectors}, and in directions out to infinity which lie inside these
137: sectors, wavefunctions either grow or decay exponentially, leading to
138: eigenvalue problems with straightforward, and discrete, spectra.
139: Note that once $M$ has passed through $2$, as in figure~\ref{sectors},
140: the real axis is once again a `good' quantisation contour --
141: but for a {\em different}\/
142: eigenvalue problem, which is {\em not}\/ the analytic continuation of the
143: original $M<2$ problem to that value of $M$. (For the analogue of
144: figure~\ref{fig1} for this new problem, see figure~20 of \cite{BBN}.)
145: Going further, we could choose {\em any}\/ pair of Stokes sectors for the
146: start and finish of our contour. A priori, each pair of sectors defines a
147: different problem, though we shall see later that some of these problems
148: are related by simple variable changes.
149:
150: All of the problems do share one feature -- their quantisation contours
151: begin and end in the neighbourhood of the point $x=\infty$. In the
152: terminology of the WKB method,
153: they are related to `lateral' connection problems~\cite{OLV}.
154: There is one other special point for the ordinary differential
155: equation, namely the origin, and this provides another natural place where
156: quantisation contours can end.
157: Contours which join $x=0$ to $x=\infty$ lead to what are called `radial'
158: (or `central') connection problems, and with suitable boundary conditions
159: they can also have interesting, discrete,
160: spectra. However, if
161: both ends of
162: the contour are placed at $x=0$, the resulting eigenvalue problem
163: is always trivial. Some sample quantisation contours are shown in
164: figure~\ref{sectorsPT}.
165:
166:
167: % %
168: \begin{figure}[ht]
169: \begin{center}
170: \includegraphics[width=0.55\linewidth]{sectorsPT.eps}
171: \end{center}
172: \caption{Some further quantisation contours. \label{sectorsPT}}
173: \end{figure}
174: %
175:
176: We should pause for a moment to consider which boundary conditions
177: can be imposed at the origin, in order to understand why $x=0$ and
178: $|x|=\infty$ behave differently as end points of quantisation
179: contours. Even with the angular-momentum-like term $l(l{+}1)x^{-2}$
180: included, the singularity at the origin is much milder than that at
181: $|x|=\infty$, and irrespective of the direction in which it is
182: approached, solutions there behave algebraically, as $x^{l+1}$ or
183: $x^{-l}$. For this reason the complications associated with Stokes
184: sectors do not arise in the neighbourhood of the origin, and there
185: are just two natural boundary conditions to impose there -- we can
186: either demand that the solution behaves as $x^{l+1}$, or as $x^{-l}$
187: (the more singular of these two boundary conditions being defined by
188: analytic continuation). This contrasts with the situation near
189: $|x|=\infty$, where we can ask that a solution be subdominant in any
190: one of the potentially infinitely-many different Stokes sectors.
191: Expressed more technically, the ordinary differential equation has
192: two singular points, one at the origin and one at infinity. The
193: singular point at the origin is {\em regular}, and solutions have
194: straightforward series expansions in its vicinity. These converge in
195: the full neighbourhood of the origin, and can be analytically
196: continued in a simple way\footnote{Cases with $2M$ not an integer
197: fall just outside the treatments in the standard texts, but they
198: behave in essentially the same way -- see, for example,
199: \cite{Cheng}. We have also glossed over some details, such as the
200: logarithms which can arise in certain situations. More background on
201: these issues can be found in \cite{ince}.}. Infinity, on the
202: other hand, is an {\em irregular} singular point, in the
203: neighbourhood of which solutions have asymptotic expansions which
204: only hold in selected Stokes sectors. This makes analytic
205: continuation around the point at infinity much more subtle, and
206: indeed this will be a major theme in the subsequent development.
207:
208:
209: To summarize:
210: associated with an ODE
211: of the type under consideration there
212: are many natural
213: eigenvalue problems, which fall into two classes.
214: Problems in the first, lateral, class are defined
215: by specifying a {\em pair} of Stokes sectors at infinity,
216: and then asking for the values of $E$ at which there exist solutions
217: to the equation which decay exponentially
218: in both sectors simultaneously. Problems in the second, radial, class
219: are defined by demanding decay in
220: a single Stokes sector at infinity, and imposing one of the
221: two simple boundary conditions at the origin.
222: The questions in $\PT$-symmetric quantum mechanics discussed in
223: section~\ref{prelude}
224: are all related to lateral
225: problems, with one particular pair of Stokes sectors selected.
226: Considerations of analytic
227: continuation have led us to put all pairs of sectors
228: on an equal footing, and we completed
229: the story by bringing in the radial problems as well.
230: But at this stage each eigenvalue problem sits on an isolated island,
231: each with its own private spectrum.
232:
233: In the next subsection, we shall start to construct some bridges
234: between the
235: islands, using methods inspired by earlier work of Sibuya and of
236: Voros.
237: Remarkably, these
238: bridges turn out to be precisely the functional equations which had
239: previously arisen in the
240: context of integrable quantum field theory.
241: %
242: %
243:
244:
245: \subsection{A simple example}
246: \label{simp}
247: To illustrate the basic ideas in the simplest possible way,
248: for the time being we set $l(l{+}1)=0$, so we are dealing with the
249: original Bender-Boettcher family of eigenproblems:
250: \eq
251: -\frac{d^2}{dx^2}\psi(x)-(ix)^{2M}\psi(x)=E\psi(x)~~,\qquad\psi\in
252: L^2({\cal C})~.
253: \en
254:
255: Now that the perspective has been widened to encompass
256: eigenvalue problems on general
257: contours, it is convenient to eliminate the factors of $i$ appearing
258: everywhere by making the variable changes
259: \eq
260: x\to x/i~,\quad E\to -E~
261: \label{varchange}
262: \en
263: so that the differential equation becomes
264: \eq
265: \left[
266: -\frac{d^2}{dx^2}+x^{2M}-E\right]\,\psi(x)=0\,.
267: \label{su2}
268: \en
269: This also
270: moves the branch cut onto the negative real axis, and the initial
271: quantisation contour onto the imaginary axis. (This final point
272: explains why the reality of
273: the spectrum remains a non-trivial question, despite the fact that the
274: coefficients in (\ref{su2}) are all real.)
275:
276: Next, we
277: need to develop our treatment of ordinary differential equations in
278: the complex domain, relying largely on the
279: work of Sibuya and co-workers~\cite{HseihSha,Sha}.
280: The key result is the following:\\
281: $\bullet$~The ODE
282: (\ref{su2}) has a `basic' solution $y(x,E)$ such that\\
283: %
284: \noindent
285: {\bf (i)} $y$ is an entire function of $x$ and $E$;
286:
287: \noindent
288: {}~~~~{\small [Though, because of the multivalued potential,
289: $x$ lives on a cover of $\CC \backslash \{0\}$
290: if $2M{\notin} \ZZ$
291: \footnote{The original work of Hseih and
292: Sibuya~\cite{HseihSha} concerned only the case
293: $2M\in\mathbb{N}$, but the result also holds for the more general
294: situation $2M\in\RR^+$ of eq.~(\ref{su2}), so long as the
295: branching at the origin is taken into account. This generalisation
296: was explicitly discussed by Tabara in \cite{Tabara}.}.\,]}
297:
298: \noindent
299: {\bf (ii)} as $|x|\to\infty$ with $|\arg\,x\,|<3\pi/(2M{+}2)$,
300: \bea
301: y\,&\sim&
302: {}~\frac{1}{\sqrt{2i}}\,
303: x^{-M/2}\exp\left[-\fract{1}{M{+}1}\,x^{M{+}1}\right]\,;
304: \label{yas}\\
305: y'&\sim& {-\frac{1}{\sqrt{2i}}}\,
306: x^{M/2}\exp\left[-\fract{1}{M{+}1}\,x^{M{+}1}\right]\,.
307: \eea
308: {}~~~~{\small [Though there are small modifications for $M\leq
309: 1$ -- see, for example, \cite{DMST}\,.\,]}\\
310: Furthermore, properties {\bf (i)} and {\bf (ii)} fix $y$ uniquely.
311:
312: The second property can be understood via
313: the WKB discussion of section~\ref{compeig}.
314: With the shift from $x$ to $x/i$,
315: the anti-Stokes lines for (\ref{su2}) are
316: \eq
317: \arg(x)=
318: \pm\frac{\pi}{2M{+}2}~,~
319: \pm\frac{3\pi}{2M{+}2}~,~
320: \dots
321: \en
322: and in between them lie the Stokes sectors, which we label by defining
323: \eq
324: \CS_k:=\left|\arg(x)-
325: \frac{2\pi k}{2M{+}2}\right|<\frac{\pi}{2M{+}2}\,.
326: \en
327: Three of these sectors
328: are shown in the figure \ref{sibsectors}, a $90^{\rm o}$ rotation of
329: figure~\ref{sectors}.
330:
331: The asymptotic given as property {\bf (ii)} then matches the result of
332: a WKB calculation in $\CS_{-1}\cup\CS_0\cup\CS_1$\,. The determination
333: of the large $|x|$ behaviour of the particular solution $y(x,E)$ beyond
334: these three sectors is a non-trivial matter, since the continuation
335: of a limit is not necessarily the same as the limit of a
336: continuation. This subtlety is related to the so-called
337: Stokes phenomenon, and it can be
338: handled using objects known as {\em Stokes multipliers}, to be introduced
339: shortly.
340:
341: \begin{figure}[ht]
342: \begin{center}
343: \includegraphics[width=0.55\linewidth]{sibsectors.eps}
344: \end{center}
345: \caption{Three Stokes sectors for the ODE (\ref{su2}), with $M=2.1$.
346: \label{sibsectors}}
347: \end{figure}
348: %
349:
350: One more piece of terminology: an
351: exponentially-growing solution in a given sector is called {\em dominant}
352: (in that sector); one which decays there
353: is called {\em subdominant}. It is
354: easy to check that $y$ as
355: defined above is subdominant in $\CS_0$, and dominant in
356: $\CS_{-1}$ and $\CS_1$. A subdominant solution
357: to a second-order ODE in a sector is
358: unique up to a constant multiplier; this is why the quoted
359: asymptotics are enough to pin down $y$ uniquely.
360:
361: Having identified one solution to the ODE, we can now generate
362: a whole family using a trick due to Sibuya. Consider the function
363: $
364: \hat y(x,E):=y(ax,E)
365: $
366: for some (fixed) $a\in\CC$. From (\ref{su2}), $\hat y(x,E)$ satisfies
367: \eq
368: \left[
369: -\frac{d^2}{dx^2}+a^{2M+2}x^{2M}-a^2E\right]\,\hat y(x,E)=0\,.
370: \en
371: (This is sometimes given the rather-grand name of
372: `Symanzik rescaling'.)
373: If $a^{2M+2}{=}1$, shifting $E$ to $a^{-2}E$ shows that
374: $\hat y(x,a^{-2}E)$ again solves (\ref{su2}).
375: Defining
376: \eq
377: \omega:=e^{2\pi i/(2M{+}2)}
378: \en
379: and
380: \eq
381: y_k(x,E):=\omega^{k/2}y(\omega^{-k}x,\omega^{2k}E)
382: \label{ykdef}
383: \en
384: we then have the statements
385:
386: \noindent
387: $\bullet$ $y_k$ solves (\ref{su2}) for all $k\in\ZZ$\,;
388: {}~~{\small [\,This follows since $(\omega^{-k})^{2M{+}2}=1$.\,]}
389:
390: \noindent
391: $\bullet$ up to a constant, $y_k$ is the unique solution to (\ref{su2})
392: subdominant in $\CS_k$.
393: {}~~{\small [\,This follows easily from the asymptotic of $y$.\,]}
394:
395: \noindent
396: $\bullet$ the functions $y_k$, $y_{k+1}$ are linearly independent for
397: all $k$, so
398: each pair $\{y_k,y_{k+1}\}$ forms a {\em basis} of solutions for
399: (\ref{su2}).
400: {}~~{\small [\,This follows on
401: comparing the asymptotics of $y_k$ and $y_{k+1}$
402: in either $\CS_k$ or $\CS_{k+1}$.\,]}
403:
404: We have almost arrived at the TQ relation. Next, the fact that
405: $y_{-1}$ can be expanded in the $\{y_0,y_1\}$ basis shows that a
406: relation of the following form must hold:
407: \eq
408: y_{-1}(x,E)=C(E)y_0(x,E)+\widetilde C(E)y_1(x,E)~.
409: \label{stokesrel}
410: \en
411: This is an example of a {\em Stokes relation}, with
412: the coefficients $C(E)$ and $\widetilde C(E)$ {\em Stokes
413: multipliers}. They can be expressed in terms of Wronskians,
414: where~\cite{CL} the {\em Wronskian} of two functions $f$ and $g$ is
415: \eq
416: W[f,g]:=fg'-f'g\,.
417: \en
418: Given two solutions $f$ and $g$
419: of a second-order ODE with vanishing first-derivative
420: term, their Wronskian $W[f,g]$ is independent of $x$, and vanishes if
421: and only if
422: $f$ and $g$
423: are proportional. As a convenient notation we set
424: \eq
425: W_{k_1,k_2}:=W[y_{k_1},y_{k_2}]=y_{k_1} y_{k_2}' -y_{k_1}' y_{k_2}
426: \label{wprops}
427: \en
428: and record the following two useful
429: properties\footnote{The normalisations in (\ref{yas}) and
430: (\ref{ykdef}), which differ from those adopted by Sibuya,
431: were expressly chosen so as to simplify these two formulae.}:
432: \eq
433: W_{k_1+1,k_2+1}(E)=W_{k_1,k_2}(\omega^2E)~,\quad W_{0,1}(E)=1\,.
434: \en
435: Now `taking Wronskians' of the Stokes relation (\ref{stokesrel})
436: first with $y_1$ and then with $y_0$
437: shows that
438: \eq
439: C=\frac{W_{-1,1}}{W_{0,1}}~~,\quad\tilde C=-\frac{W_{-1,0}}{W_{0,1}}=-1
440: \en
441: and so the relation itself can be rewritten as
442: \eq
443: C(E)y_0(x,E)=y_{-1}(x,E)+y_1(x,E)~,
444: \label{CYa}
445: \en
446: or, in terms of the original function $y$, as
447: \eq
448: C(E)y(x,E)
449: =\omega^{-1/2}y(\omega x,\omega^{-2}E)+
450: \omega^{1/2}y(\omega^{-1} x,\omega^2E)\,.
451: \label{CY}
452: \en
453: This looks very like a TQ relation. The only fly in the ointment is
454: the
455: $x$-dependence of the function $y$. But this is easily fixed: we just
456: set $x$ to zero. We can also take a derivative with respect to
457: $x$ before setting $x$ to zero, which swaps
458: the phase factors $\omega^{\pm1/2}$. So we define
459: \eq
460: D_-(E):=y(0,E)~~,\quad D_+(E):=y'(0,E)\,.
461: \en
462: Then the Stokes relation (\ref{CY}) implies
463: \eq
464: \fl
465: {}~~~~~~C(E)D_{\mp}(E)= \omega^{\mp 1/2}D_{\mp}(\omega^{-2}\!E)+
466: \omega^{\pm 1/2}D_{\mp}(\omega^2\!E)~,
467: \label{CD}
468: \en
469: and precisely matches the forms of the TQ equations (\ref{ctqlat})
470: and (\ref{ctq}) if the twist parameter is set to $\phi=2 \pi
471: p=\pi/(2M{+}2)$. Although the details are at this stage sketchy --
472: more will be provided in later sections -- we can already see
473: how some concepts in the two worlds of integrable models and
474: ordinary differential equations must be related:
475: $$
476: \begin{array}{|lcl|}
477: \hline
478: & & \\[-9pt]
479: {}~\parbox{5.1cm}{Six-vertex model with twist\\[-1pt]
480: $\phi=2 \pi p=\pi/(2M{+}2)$}&
481: &{}\parbox{4.9cm}{Schr\"odinger equation with\\[-1pt] homogeneous
482: potential $x^{2M}$}\\[12pt]
483: \hline
484: & & \\[-10pt]
485: \pppbox{Spectral parameter}&\lra&\mbox{Energy}\\[5pt]
486: \pppbox{Anisotropy}&\lra&\mbox{Degree of potential}\\[5pt]
487: \pppbox{Transfer matrix}&\lra&\mbox{The Stokes multiplier
488: $C$}\\[6pt]
489: \pppbox{Q operator}&\lra&\mbox{$D_-(E)$\,: the value of
490: $y(x,E)$ at $x=0$~~}\\[10pt]
491: \hline
492: \end{array}
493: $$
494: If $y$ on the last line is replaced by $y'$, then the twist changes
495: to $\phi=-\pi/(2M{+}2)$.
496:
497: %%
498: A small puzzle remains at this stage: why should one particular value
499: of the twist in the integrable model be singled out for a link with an
500: ordinary differential equation? This was resolved
501: shortly after the original observation in \cite{DTa}, when
502: Bazhanov, Lukyanov and Zamolodchikov
503: \cite{BLZa} pointed out that including an angular-momentum-like
504: term $l(l{+}1)/x^2$ allowed Q operators at other values of the
505: twist
506: to be matched. The details, and a further small generalisation of
507: the basic ODE (\ref{su2}), will be covered in section~\ref{dict}.
508:
509:
510: \subsection{The spectral interpretation}
511: %
512: How should we think about $C$ and $D$? In fact they are spectral
513: determinants.
514: The {\em spectral determinant} of an eigenvalue problem is
515: a function which vanishes exactly at
516: the eigenvalues of that problem: it generalises
517: to infinite dimensions the characteristic polynomial
518: $\det(M-\lambda {\rm I})$
519: of a finite-dimensional matrix.
520: Recall that $C(E)$ is equal to the Wronskian $W_{-1,1}(E)$.
521: Thus $C(E)$ vanishes if and only if
522: $W[y_{-1},y_1]=0$, in other words if and only if
523: $E$ is such that $y_{-1}$ and $y_1$ are
524: linearly dependent. In turn, this is true
525: if and only if the ODE~(\ref{sh})
526: has a solution decaying in the two sectors $\CS_{-1}$ and $\CS_1$
527: simultaneously, which is exactly the lateral eigenvalue problem
528: discussed in section~\ref{prelude},
529: modulo the trivial redefinitions of $x$ and $E$.
530: This is enough to deduce that, up to a factor of an entire function
531: with
532: no zeros, $C(E)$ is the spectral determinant for the
533: Bender-Boettcher
534: problem\footnote{Since we performed
535: a variable change in this section compared with the
536: discussion in section~\ref{prelude},
537: it is in fact $C(-E)$ which provides the spectral determinant
538: for the Bender-Boettcher problem as originally formulated.}.
539: Even this ambiguity can be eliminated,
540: via Hadamard's factorisation theorem,
541: once the growth properties of the functions involved have been
542: checked;
543: see \cite{DTb} for details.
544: %
545: By its definition,
546: the zeros of $D_{-}(E)$ are the values of $E$ at
547: which the function $y(x)$, vanishing at
548: $x=\infty$, also vanishes at
549: $x=0$. Likewise,
550: the zeros of $D_{+}(E)$ are the values of $E$ at
551: which $y(x)$ has zero first derivative at $x=0$.
552: Thus $D_{\mp}(E)$ are also
553: spectral determinants.
554: Note that the vanishing of $D_{-}$ or $D_{+}$ corresponds
555: to there existing normalisable
556: wave functions for the equation on the full real axis,
557: with potential $|x|^{2M}$, which are
558: odd or even, respectively, as illustrated in figures \ref{even} and
559: \ref{odd}; this explains the labelling convention
560: adopted earlier.
561:
562: \begin{figure}[ht]
563: \begin{center}
564: \includegraphics[width=0.55\linewidth]{even.eps}
565: \end{center}
566: \caption{$|x|$-potential, even sector: the ground state wavefunction.\label{even}}
567: \end{figure}
568: %
569: \begin{figure}[ht]
570: \begin{center}
571: \includegraphics[width=0.55\linewidth]{odd.eps}
572: \end{center}
573: \caption{$|x|$-potential, odd sector: the first excited state wavefunction.\label{odd}}
574: \end{figure}
575: %
576:
577:
578:
579: This insight allows a gap in the correspondence to be filled.
580: We mentioned in section~\ref{aba} that
581: while the TQ relation is very restrictive, it does not have
582: a unique solution. So to claim that $D_-(E)$ is `equal' to
583: $A_+(\lambda,p)$
584: begs the question: which $A_+(\lambda,p)$?
585: To answer, we first note that the radial
586: problem for which $D_-(E)$ is the spectral determinant,
587: in contrast to the lateral
588: Bender-Boettcher problem, {\em is} self-adjoint, and so all of its
589: eigenvalues are guaranteed to be real.
590: Back in the integrable model, we mentioned previously that
591: there is only one solution to the BAE
592: with all roots real, so the question
593: is answered: the relevant $A_+(\lambda)$ is that corresponding to the
594: ground state in the spin-zero sector of the model.
595:
596:
597:
598:
599:
600: %%% Local Variables:
601: %%% mode: latex
602: %%% TeX-master: "review"
603: %%% End:
604: