1: % Article for J. Phys. A, authors Vincenzo Aquilanti and 

2: % Robert G. Littlejohn

3: %

4: %\documentclass[12pt]{iopart}  % preprint

5: \documentclass{iopart}       % journal version

6: \usepackage{iopams}

7: \usepackage{setstack}

8: 

9: 

10: \begin{document}

11: \title{Semiclassical analysis of Wigner $3j$-symbol}

12: \author{Vincenzo Aquilanti}

13: \address{Dipartimento di Chimica, Universit\`a di Perugia, Perugia, 

14: Italy 06100}

15: \author{Hal M. Haggard, Robert G. Littlejohn and Liang Yu}

16: \address{Department of Physics, University of

17: California, Berkeley, California 94720 USA}

18: 

19: \ead{robert@wigner.berkeley.edu}

20: 

21: \begin{abstract}

22: 

23: We analyze the asymptotics of the Wigner $3j$-symbol as a matrix

24: element connecting eigenfunctions of a pair of integrable systems,

25: obtained by lifting the problem of the addition of angular momenta

26: into the space of Schwinger's oscillators.  A novel element is the

27: appearance of compact Lagrangian manifolds that are not tori, due to

28: the fact that the observables defining the quantum states are

29: noncommuting.  These manifolds can be quantized by generalized

30: Bohr-Sommerfeld rules and yield all the correct quantum numbers.  The

31: geometry of the classical angular momentum vectors emerges in a clear

32: manner.  Efficient methods for computing amplitude determinants in

33: terms of Poisson brackets are developed and illustrated.

34: 

35: \end{abstract}

36: 

37: \pacs{03.65.Sq, 02.20.Qs, 02.30.Ik, 02.40.Yy}

38: 

39: %\maketitle %use this if you want to have title on separate page

40: 

41: \section{Introduction}

42: 

43: This article is a study of the asymptotics of the Wigner $3j$-symbol

44: from the standpoint of semiclassical mechanics, that is, essentially

45: multidimensional WKB theory for integrable systems.  The principal

46: result itself, the leading asymptotic expression for the $3j$-symbol,

47: has been known since Ponzano and Regge (1968).  Nevertheless our

48: analysis presents several novel features.  One is the exploration of

49: Lagrangian manifolds in phase space that are not tori (the usual case

50: for eigenstates of integrable systems).  Instead, one of the states

51: entering into the $3j$-symbol is supported semiclassically on a

52: Lagrangian manifold that is a nontrivial 3-torus bundle over $SO(3)$.

53: This manifold can be quantized by generalized Bohr-Sommerfeld rules,

54: whereupon it yields the exact eigenvalues required by the quantum

55: $3j$-symbol, as well as the correct amplitude and phase of its

56: asymptotic form.  This unusual Lagrangian manifold arises because the

57: quantum state in question is an eigenstate of a set of noncommuting

58: operators.  Other novel features include the expression of the asymptotic

59: phase of the $3j$-symbol in terms of the phases of Schwinger's

60: harmonic oscillators and the determination of stationary phase points

61: by geometrically transparent operations on angular momentum vectors in

62: three-dimensional space.  Yet another is the representation of

63: multidimensional amplitude determinants as matrices of Poisson

64: brackets.  Representations of this type have been known for some time,

65: but they are generalized here to the case of sets of noncommuting

66: operators.  The final result is a one-line derivation of the amplitude

67: of the asymptotic form of the $3j$-symbol.  Similarly brief

68: derivations are possible for the amplitudes of the $6j$- and

69: $9j$-symbols.  

70: 

71: In addition our analysis of the $3j$-symbol may prove to be useful for

72: the asymptotic study of the $3nj$-symbols for higher $n$.  The leading

73: order asymptotics of the $6j$-symbol were derived by Ponzano and Regge

74: (1968), but the understanding of the asymptotics of the $9j$-symbol is

75: still incomplete.  These symbols are important in many applications in

76: atomic, molecular and nuclear physics, for example, the $9j$-symbols

77: are needed in atomic physics to convert from an $LS$-coupled basis to

78: a $jj$-coupled basis.  These symbols are all examples of closed spin

79: networks, of which more elaborate examples occur in applications, each

80: of which presents a challenge to asymptotic analysis.  Moreover in

81: recent years new interest in this subject has arisen from researches

82: into quantum computing (Marzuoli and Rasetti, 2005) and quantum

83: gravity, where new derivations of the asymptotics of the Wigner

84: $6j$-symbol have been produced as well as generalizations to other

85: groups such as the Lorentz group.  The $3nj$-symbols and their

86: asymptotics have also been used recently in algorithms for molecular

87: quantum mechanics (De~Fazio \etal\ 2003 and Anderson and Aquilanti,

88: 2006), which exploit the connections with the theory of discrete

89: orthogonal polynomials (Aquilanti \etal\ 1995, 2001a,b and references

90: therein).

91: 

92: The asymptotic formula for the $3j$-symbol is closely related to that

93: for the $6j$-symbol, being a limiting case of the latter.  These were

94: first derived by Ponzano and Regge (1968), using intuitive methods and

95: building on Wigner's earlier result for the amplitude of the

96: $6j$-symbol (Wigner, 1959).  Later Neville (1971) analyzed the

97: asymptotics of the $3j$- and $6j$- symbols by a discrete version of

98: WKB theory, applied to the recursion relations satisfied by those

99: symbols, without apparently knowing of the work of Ponzano and Regge.

100: His formulas are not presented in a particularly transparent or

101: geometrical manner, but appear to reproduce some of the results of

102: Ponzano and Regge.  The formula for the $3j$-symbol (in the form of

103: Clebsch-Gordan coefficients) was later derived again by Miller (1974),

104: who presented it as an example of his general theory of semiclassical

105: matrix elements of integrable systems.  Miller called on the fact that

106: the phase of the semiclassical matrix element is a generating function

107: of a canonical transformation, and used the classical transformation

108: that most obviously corresponds to the quantum addition of angular

109: momenta to reconstruct the generating function.  The method leads to a

110: difficult integral, which, once done, yields the five terms in the

111: phase of the asymptotic formula for the $3j$-symbol.  Somewhat later

112: Schulten and Gordon (1975a,b) presented a rigorous derivation of the

113: Ponzano and Regge results for the $3j$- and $6j$-symbols, using

114: methods similar to those of Neville but carrying them out in a more

115: thorough and elegant manner.  Schulten and Gordon also provided

116: uniform approximations for the transition from the classical to

117: nonclassical regimes, work that has recently been reanalyzed (Geronimo

118: \etal, 2004) and extended to non-Euclidean and quantum groups (Taylor

119: and Woodward, 2005).  Somewhat later Biedenharn and Louck (1981b)

120: presented a review and commentary of the results of Ponzano and Regge,

121: as well as a proof based on showing that the result satisfies

122: asymptotically a set of defining relations for the $6j$-symbol.  More

123: recently the asymptotics of the $3j$-symbol was derived again by

124: Reinsch and Morehead (1999), working with an integral representation

125: constructed out of Wigner's single-index sum for the Clebsch-Gordan

126: coefficients.  About the same time, Roberts (1999) derived the Ponzano

127: and Regge results for the $6j$-symbol, using methods of geometric

128: quantization.  Finally, Freidel and Louapre (2003) presented a

129: derivation of the asymptotic expression for the square of the

130: $6j$-symbol, based on an analysis of an $SU(2)$ path integral.  This

131: work was part of a larger study of generalizations of the $6j$-symbol

132: to other groups (for example, the $10j$-symbol) that are important in

133: quantum gravity.  See also Barrett and Steele (2003) and Baez,

134: Christensen and Egan (2002).

135: 

136: There are many variations on the calculation of the asymptotic forms

137: of the $3nj$-symbols that have been considered by different authors.

138: There are asymptotic forms inside and outside the classically allowed

139: regions; uniform approximations connecting two or more of these

140: regions; asymptotic forms when only some of the quantum numbers are

141: large and others small; and higher order terms.  Ponzano and Regge

142: (1968) covered many of these issues, while Reinsch and Morehead

143: computed some higher order terms. 

144: 

145: The outline of this paper is as follows.  In Sec.~\ref{scwfis} we

146: review the semiclassical mechanics of integrable systems in the

147: generic case that one has sets of commuting observables, drawing

148: attention to an expression for the amplitude determinant in terms of

149: Poisson brackets.  In Sec.~\ref{Schwingermodel} we review the

150: Schwinger model for representing angular momentum operators in terms

151: of harmonic oscillators.  This model allows us to express angular

152: momentum eigenstates in terms of wave functions on ${\mathbb R}^n$,

153: which we use in Sec.~\ref{3jSchwinger} to express the $3j$-symbols in

154: terms of scalar products of such functions.  In Sec.~\ref{cmSchwinger}

155: we study the Schwinger model from a classical standpoint, in which an

156: important element is the reduction of the Schwinger phase

157: space (the ``large phase space'') by the torus group $T^3$, producing

158: the Poisson manifold ${\mathbb R}^3 \times {\mathbb R}^3 \times

159: {\mathbb R}^3$ (``angular momentum space'') and the reduced phase

160: space $S^2 \times S^2 \times S^2$ (the ``small phase space'').  In

161: Secs.~\ref{jmtori} and \ref{Wignermanifold} we study the two

162: Lagrangian manifolds that support the states whose scalar product is

163: the $3j$-symbol.  One is a conventional invariant torus (the

164: ``$jm$-torus''), but the other, what we call the ``Wigner manifold,''

165: is compact and Lagrangian but not a torus.  This manifold supports

166: Wigner's state of zero total angular momentum that enters into the

167: definition of the $3j$-symbols.  In Secs.~\ref{intersections} we study

168: the intersections of the $jm$-torus and the Wigner manifold, which are

169: the stationary phase points of the $3j$-symbol, and show how these can

170: be found by elementary geometrical considerations in three-dimensional

171: space (that is, by rotating angular momentum vectors).  The

172: intersection of the two manifolds turns out to be a pair of 4-tori.

173: In Sec.~\ref{actionintegrals} we compute the action integrals along

174: the respective Lagrangian manifolds to points on the two 4-tori, whose

175: difference is the Ponzano and Regge phase of the $3j$-symbol.  In

176: Sec.~\ref{BSquant} we apply generalized Bohr-Sommerfeld quantization

177: to the $jm$-torus and the Wigner manifold, a standard procedure for

178: the $jm$-torus, although it leads in an interesting way to the extra

179: $1/2$ in the classical values representing the lengths of the angular

180: momentum vectors.  This extra $1/2$ was guessed by Ponzano and Regge

181: and Miller and derived systematically by Schulten and Gordon, Reinsch

182: and Morehead and by us, although it is missing from the results of

183: Roberts.  In our work it is essentially a Maslov index.  In

184: Sec.~\ref{ampdet} we generalize known expressions for the amplitude

185: determinant of semiclassical matrix elements of integrable systems in

186: terms of Poisson brackets to the case of collections of noncommuting

187: observables (whose level sets nevertheless are Lagrangian).  The

188: result allows us to compute the amplitude of the $3j$-symbol as a

189: $2\times 2$ matrix of Poisson brackets.  We then put all the pieces

190: together to obtain the final asymptotic form.  Finally, in

191: Sec.~\ref{conclusions} we present some comments on the work, prospects

192: for further work, and conclusions.

193: 

194: \section{Semiclassical wave functions for integrable systems}

195: \label{scwfis}

196: 

197: The semiclassical mechanics of integrable systems is well understood

198: (Einstein, 1917; Brillouin, 1926; Keller, 1958; Percival, 1973; Berry

199: and Tabor, 1976; Gutzwiller, 1990; Brack and Bhaduri, 1997; Cargo

200: \etal, 2005a, 2005b).  Here we summarize the basic facts, some of

201: which require modification for our application.

202: 

203: We consider the quantum mechanics of a particle moving in ${\mathbb

204: R}^n$ (with wave function $\psi(x_1,\ldots,x_n)$ and Hilbert space

205: $L^2({\mathbb R}^n)$).  We speak of an integrable system if we have a

206: complete set of commuting observables $\{{\hat A}_1, \ldots, {\hat

207: A_n}\}$ acting on this Hilbert space.  We use hats to distinguish

208: quantum operators from classical quantities with a similar meaning.

209: Sometimes the Hamiltonian is one of these operators or a function of

210: them, but in our application there is no Hamiltonian, or, rather all

211: the ${\hat A}_i$'s are Hamiltonians on an equal footing.  These

212: operators may be converted into their classical counterparts by the

213: Weyl transform (Weyl 1927, Wigner 1932, Groenewold 1946, Moyal 1949,

214: Voros 1977, Berry 1977, Balazs and Jennings 1984, Hillery \etal\ 1984,

215: Littlejohn 1986, McDonald 1988, Estrada \etal\ 1989, Gracia-Bond\'\i a

216: and V\'arilly 1995 and Ozorio de Almeida 1998).  The Weyl transforms

217: (or Weyl ``symbols'') of these operators are functions on the

218: classical phase space ${\mathbb R}^{2n}$, that is, functions of

219: $(x_1,\ldots,x_n; p_1,\ldots,p_n)$.  They are normally even power

220: series in $\hbar$, as we assume, of which the leading term is the

221: ``principal symbol.''  We denote the principal symbols of $\{{\hat

222: A}_1, \ldots, {\hat A_n}\}$ by $\{A_1, \ldots, A_n\}$ (without the

223: hats).  In view of the Moyal star product representation (Moyal 1949)

224: of the vanishing commutators $[{\hat A}_i, {\hat A}_j]=0$, the

225: principal symbols Poisson commute, $\{A_i, A_j\}=0$, thus defining a

226: classically integrable system (Arnold 1989, Cushman and Bates 1997).

227: (We use curly brackets $\{\;\}$ both to denote a set and for Poisson

228: brackets.)  Then according to the Liouville-Arnold theorem (Arnold

229: 1989), the compact level sets of $\{A_1, \ldots, A_n\}$ are

230: generically $n$-tori.  The Hamiltonian vector fields generated by the

231: $A_i$ are commuting and linearly independent on the tori; thus the

232: tori are not only the level sets of the $A_i$, they are also the

233: orbits of the Abelian group generated by the corresponding Hamiltonian

234: flows.  One can define an action function $S$ on a torus as the

235: integral of $\sum_i p_i \, dx_i$ relative to some initial point; it is

236: multivalued because of the topologically distinct paths going from the

237: initial to the final point, but otherwise is independent of the path.

238: 

239: Let $A_i=a_i$ be one of these tori ($A_i$ are the functions, $a_i$ the

240: values).  The torus has a projection onto configuration space defining

241: a classically allowed region in that space; the inverse projection is

242: multivalued.  The function $S$ may be projected onto configuration

243: space, defining a function we shall denote by $S_k(x,a)$ (where for

244: brevity $x$ and $a$ stand for $(x_1, \ldots,x_n)$ and $(a_1, \ldots,

245: a_n)$, etc).  Here $k$ labels the branches of the inverse projection;

246: function $S_k$ has an additional multivaluedness due to the choice of

247: contour connecting initial and final points on the torus.  Then as

248: explained by Arnold (1989), $S(x,a)$ is the generating function of the

249: canonical transformation $(x,p) \to (\alpha,A)$, where

250: $\alpha=(\alpha_1, \ldots, \alpha_n)$ is the set of angle variable

251: conjugate to the conserved quantities $(A_1, \ldots, A_n)$.  Action

252: variables may be defined in the usual way as $(1/2\pi) \oint p\,dx$

253: around the independent basis contours on the torus; these are

254: functions of the $A_i$ and their conjugate variables are the angles

255: that cover the torus once when varying between $0$ and $2\pi$.

256: Sometimes however it is more convenient to work with the $A_i$ instead

257: of the actions (the $A_i$ are not necessarily actions, and their flows

258: are not necessarily periodic on the torus).

259: 

260: Tori are quantized, that is, associated with a consistent solution of

261: the simultaneous Hamiltonian-Jacobi and amplitude transport equations

262: for the operators ${\hat A}_i$, only if they satisfy the

263: Bohr-Sommerfeld or EBK quantization conditions, discussed in

264: Sec.~\ref{BSquant}.  Associated with a quantized torus is a

265: semiclassical wave function in configuration space, which in the

266: classically allowed region is given by

267: \begin{equation}

268: \psi(x) = \langle x \vert a \rangle =

269: \frac{1}{\sqrt V} \sum_k |\Omega_k|^{1/2} 

270: \exp\{i[S_k(x,a) - \mu_k\pi/2]\}.

271: \label{scpsi}

272: \end{equation}

273: The meaning of this formula is the following.  First, here and below

274: we set $\hbar=1$.  Next, given the point $x$ in the classically

275: allowed region, its inverse projection onto the quantized torus is a

276: set of points indexed by $k$.  We assume the projection is nonsingular

277: at these points (we are not at a caustic).  The phase $S_k(x,a)$ is

278: the integral of $p\,dx$ from a given initial point on the torus to the

279: $k$-th point of the inverse projection, and $\mu_k$ is the Maslov

280: index (Maslov 1981, Mishchenko \etal\ 1990, de Gosson 1997) of the

281: same path.  The amplitude determinant $\Omega_k$ is given by

282: \begin{equation}

283: \Omega_k= \det\frac{\partial^2 S_k(x,a)}{\partial x_i \partial a_j}

284: 	=[ \det \{x_i,A_j\} ]^{-1},

285: \label{ampldet}

286: \end{equation}

287: where in the second form the Poisson brackets are evaluated on the

288: $k$-th branch of the inverse mapping from $x$ to the Lagrangian

289: manifold.  The amplitude determinant is a density on configuration

290: space (to within the semiclassical approximation, the probability

291: density corresponding to a single branch), which is the projection

292: onto configuration space of a density on the torus.  The latter

293: density is required to be invariant under the Hamiltonian flows

294: generated by the $A_i$ (this is the meaning of the amplitude transport

295: equations for the $A_i$); in terms of the variables $\alpha_i$

296: conjugate to the $A_i$ this means that the density is constant (it is

297: the $n$-form $d\alpha_1 \wedge \ldots \wedge d\alpha_n$).  Finally,

298: the quantity $V$ in (\ref{scpsi}) is the volume of the torus,

299: measured with respect to this density.  If the $A_i$ are action

300: variables, then $V=(2\pi)^n$.  The overall phase of the wave function

301: (its phase convention) is determined by the choice of the initial

302: point on the torus.

303: 

304: Now let $\{{\hat A}_1, \ldots, {\hat A}_n\}$ and $\{{\hat B}_1,

305: \ldots, {\hat B}_n\}$ be two complete sets of commuting observables,

306: with principal symbols $A_i$ and $B_i$, conjugate angles $\alpha_i$

307: and $\beta_i$ and action functions $S_A(x,a)$ and $S_B(x,b)$, and let

308: $a$ and $b$ refer to two quantized tori (an $A$-torus and a

309: $B$-torus).  We assume initially that the two sets of functions $A_i$

310: and $B_i$ are independent.  We compute $\langle b \vert a \rangle$ 

311: as an integral of the wave functions over $x$, evaluated by the

312: stationary phase approximation.  The stationary phase points are

313: geometrically the intersections of the $A$-torus with the $B$-torus.

314: Generically the two tori intersect in finite set of isolated points

315: that we index by $k$, denoting the corresponding $\alpha$ and $\beta$

316: values by $\alpha_k$ and $\beta_k$.  (For given $k$, $\alpha_k$ and

317: $\beta_k$ refer to the same point in phase space.)  Then the result is

318: \begin{equation}

319: \langle b\vert a\rangle = 

320: \frac{(2\pi i)^{n/2}}{\sqrt{V_A V_B}}

321: \sum_k |\Omega_k|^{1/2}

322: \exp\{ i[S_A(\alpha_k)-S_B(\beta_k)-\mu_k\pi/2]\}.

323: \label{abmatrixelement}

324: \end{equation}

325: Here $V_A$ and $V_B$ are the volumes of the respective tori, as in

326: (\ref{scpsi}), and the actions $S_A$ and $S_B$ are considered

327: functions of the $\alpha$ or $\beta$ coordinates on the respective

328: tori.  

329: 

330: As shown by Littlejohn (1990), the amplitude determinant $\Omega_k$

331: can be written in terms of the Poisson brackets of the observables

332: $A_i$ and $B_i$,

333: \begin{equation}

334: \Omega_k = [\det \{A_i, B_j\}]^{-1}.

335: \label{PBdet}

336: \end{equation}

337: The Maslov index $\mu_k$ in (\ref{abmatrixelement}) is not the

338: same as in (\ref{scpsi}).

339: 

340: Another case considered by Littlejohn (1990) is the one in which some

341: of the $A_i$ are functionally dependent on some of the $B_i$.  For

342: this case it is convenient to assume that the first $r$ of the two

343: sets of variables $A$ and $B$ are functionally independent, while the

344: last $n-r$ are identical, so that $A=\{A_1, \ldots, A_r, A_{r+1},

345: \ldots, A_n\}$ and $B=\{B_1, \ldots, B_r, A_{r+1}, \ldots, A_n\}$.

346: Then the stationary phase points are still the intersections of the

347: two $n$-tori, but now the intersections are generically a finite set

348: of isolated $(n-r)$-tori, upon which linearly independent vector

349: fields are the Hamiltonian vector fields associated with the

350: $(A_{r+1}, \ldots, A_n)$.  Such an $(n-r)$-torus is the orbit of the

351: Abelian group action generated by the corresponding Hamiltonian flows.

352: In this case we find

353: \begin{equation}

354: \langle b \vert a \rangle =

355: \frac{(2\pi i)^{r/2}}{\sqrt{V_A V_B}}

356: \sum_k V_k |\Omega_k|^{1/2} 

357: \exp\{i[S_A(\alpha_k)-S_B(\beta_k)-\mu_k\pi/2]\},

358: \label{abmatrixelement1}

359: \end{equation}

360: where now $V_k$ is the volume of the $k$-th intersection (an

361: $(n-r)$-torus on which the volume measure is $d\alpha_{r+1} \wedge

362: \ldots \wedge d\alpha_n$), and where now the amplitude determinant 

363: $\Omega_k$ is still given by (\ref{PBdet}), except that it is

364: understood that only the first $r$ of the $A$'s and $B$'s enter (thus,

365: it is an $r\times r$ determinant instead of an $n \times n$ one).  The

366: phase difference $S_A - S_B$ for branch $k$ can be evaluated at any

367: point on the $(n-r)$-torus which is the intersection, since the

368: integral of $p\,dx$ back and forth along a path lying in the

369: intersection vanishes.

370: 

371: \section{The Schwinger model}

372: \label{Schwingermodel}

373: 

374: The Schwinger (or $SU(2)$ or boson) model for angular momenta is

375: explained well in Schwinger's original paper (reprinted by Biedenharn

376: and van Dam (1965), the original 1952 paper being unpublished), and

377: reworked in an interesting way by Bargmann (1962).  For further

378: perspective see Biedenharn and Louck (1981a) and Smorodinskii and

379: Shelepin (1972).  Introductions are given by Sakurai (1994) and

380: Schulman (1981).  Here we define the notation for the Schwinger model

381: and emphasize some aspects that will be important for our application.

382: 

383: In the Schwinger model each independent angular momentum vector is

384: associated with two harmonic oscillators.  We shall refer to the

385: $1j$-, $3j$-, etc models, depending on how many independent angular

386: momenta there are.  The number of $j$'s in the model is not

387: necessarily the number of $j$'s in the Wigner symbol; for example,

388: Miller (1974) used a $2j$-model to study the Clebsch-Gordan

389: coefficients, essentially the $3j$-symbols.  

390: 

391: We start with the $1j$-model, for which there are two harmonic

392: oscillators indexed by Greek indices $\mu, \nu, \ldots = 1,2$.  (These

393: are just labels of the two oscillators; sometimes other labels such as

394: $1/2$, $-1/2$ are more suitable.)  The wave functions are $\psi(x_1,x_2)$

395: and the Hilbert space is ${\cal H} = L^2({\mathbb R}^2)$.  We write

396: ${\hat H}_\mu=(1/2)({\hat x}_\mu^2 + {\hat p}_\mu^2)$ for the two

397: oscillator Hamiltonians, and we define ${\hat H}=\sum_\mu {\hat

398: H}_\mu$.  The eigenvalues of ${\hat H}$ are $n+1$, with

399: $n=0,1,\ldots$, and energy level $E_n$ is $(n+1)$-fold degenerate.  We

400: introduce usual annihilation and creation operators $a_\mu= ({\hat

401: x}_\mu + i{\hat p}_\mu)/\sqrt{2}$, $a^\dagger_\mu = ({\hat x}_\mu -i

402: {\hat p}_\mu)/\sqrt{2}$, omitting the hats on the $a$'s and

403: $a^\dagger$'s since these will always be understood to be operators.

404: We define operators

405: \begin{equation}

406:   {\hat I} = \frac{1}{2} \sum_\mu a^\dagger_\mu a_\mu 

407:   =\frac{1}{2} ({\hat H} - 1)

408: \label{Imudef}

409: \end{equation}

410: and

411: \begin{equation}

412:   {\hat J}_i = \frac{1}{2} 

413:   \sum_{\mu\nu} a^\dagger_\mu \, \sigma^i_{\mu\nu} \, a_\nu,

414: \label{Jidef}

415: \end{equation}

416: where $\sigma^i$ is the $i$-th Pauli matrix.  Here and below we use

417: indices $i,j,\ldots = 1,2,3$ (or $x,y,z$ if that is more clear) to

418: denote the Cartesian components of a 3-vector.  Notice that ${\hat I}$

419: and ${\hat J}_i$ are quadratic functions of the $x$'s and $p$'s of the

420: system.  The eigenvalues of ${\hat I}$ are $n/2$ for $n=0,1,\ldots$.

421: These operators satisfy the commutation relations $[{\hat I},{\hat

422: J}_i] =0$ and $[{\hat J}_i, {\hat J}_j]= i\sum_k \epsilon_{ijk} {\hat

423: J}_k$.  We also define ${\hat{\bf J}}^2 = \sum_i {\hat J}_i^2$, so

424: that $[{\hat I},{\hat{\bf J}}^2]=0$ and $[{\hat J_i}, {\hat{\bf

425: J}}^2]=0$.  It avoids some confusion with indices to always denote the

426: square of a vector by a bold face symbol, as we have done here.  We

427: note the important operator identity ${\hat{\bf J}}^2 = {\hat I}({\hat

428: I} +1)$, expressing the quartic operator ${\hat{\bf J}}^2$ as a

429: function of the quadratic operator ${\hat I}$.  

430: 

431: From this identity and the known eigenvalues of ${\hat I}$ it follows

432: that the eigenvalues of ${\hat{\bf J}}^2$ are $(n/2)[(n/2)+1]$, for

433: $n=0,1,\ldots$, which leads us to identify $n/2$ with

434: $j=0,1/2,1,\ldots$, the usual angular momentum quantum number.  The

435: $n$-th (or $j$-th) eigenspace of ${\hat H}$ or ${\hat I}$ is

436: $(2j+1)$-dimensional, and so must contain a single copy of the $j$-th

437: irrep of $SU(2)$.  Each irrep (both integer and half-integer values of

438: $j$) occurs precisely once in the Hilbert space ${\cal H}$.  We denote

439: these subspaces by ${\cal H}_j$, and write ${\cal H} = \sum_j \oplus

440: {\cal H}_j$.  The standard basis in ${\cal H}_j$ is the eigenbasis of

441: ${\hat J}_z =\frac{1}{2}({\hat H}_1 - {\hat H}_2)$, with the usual

442: quantum number $m$, so that if $n_\mu$ are the usual quantum numbers

443: of the oscillators ${\hat H}_\mu$, then $n_1=j+m$, $n_2=j-m$.  The

444: simultaneous eigenstates of ${\hat{\bf J}}^2$ and ${\hat J}_z$ are

445: $\vert jm\rangle$ or $\vert n_1 n_2

446: \rangle$.

447: 

448: In the $Nj$-model we index the angular momenta with indices $r,s,\dots

449: = 1, \ldots, N$.  The oscillators are now labelled ${\hat H}_{r\mu}$

450: with coordinates and momenta ${\hat x}_{r\mu}$ and ${\hat p}_{r\mu}$

451: and annihilation and creation operators $a_{r\mu}$ and

452: $a^\dagger_{r\mu}$.  The wave functions are now $\psi(x_{11}, x_{12},

453: x_{21}, \ldots, x_{N2})$ and the Hilbert space is $L^2({\mathbb

454: R}^{2N})$.  We define operators

455: \begin{eqnarray}

456:   {\hat I}_r &= \frac{1}{2} \sum_\mu a^\dagger_{r\mu} a_{r\mu},

457:   \label{Irdef} \\

458:   {\hat J}_{ri} &= \frac{1}{2} \sum_{\mu\nu}

459:    a^\dagger_{r\mu} \, \sigma^i_{\mu\nu} \, a_{r\nu},

460:    \label{Jridef} \\

461:    {\hat{\bf J}}^2_r &= \sum_i {\hat J}_{ri}^2,

462:    \label{Jrsquareddef} \\

463:    {\hat J}_i &= \sum_r {\hat J}_{ri},

464:    \qquad \hbox{\rm or} \qquad

465:    {\hat{\bf J}} = \sum_r {\hat{\bf J}}_r, 

466:    \label{Jtotalidef} \\

467:    {\hat{\bf J}}^2 & = \sum_i {\hat J}_i^2,

468:    \label{Jtotalsquareddef}

469: \end{eqnarray}

470: most of which are obvious generalizations from the case $N=1$.  These

471: satisfy the identity

472: \begin{equation}

473:   {\hat{\bf J}}_r^2 = {\hat I}_r({\hat I}_r+1),

474:   \label{JIidentity}

475: \end{equation}

476: and the commutation relations $[{\hat I}_r, {\hat J}_{si}] = [{\hat

477: I}_r, {\hat{\bf J}}^2_s]=0$.  Each angular momentum vector ${\hat{\bf

478: J}}_r$ also obeys the standard commutation relations among its

479: components and square, which we omit, as does any sum of these angular

480: momenta (partial or total).

481: 

482: The angular momenta generate an action of $[SU(2)]^N$ on the Hilbert

483: space (one copy for each pair of oscillators).  Here we discuss only

484: the simultaneous rotation of all oscillator degrees of freedom by the

485: same element of $SU(2)$, which is generated by the total angular

486: momentum, but  partial rotation operators can also be defined and are

487: useful.  We begin with the commutation relations,

488: \begin{eqnarray}

489: [{\hat J}_i, a_{r\mu}] &= -\frac{1}{2} \sum_\nu \sigma^i_{\mu\nu} \,

490: a_{r\nu},

491: \label{Jacomrel} \\ \relax

492: [{\hat J}_i, a^\dagger_{r\mu}] &=

493: +\frac{1}{2} \sum_\nu a^\dagger_{r\nu} \,\sigma^i_{\nu\mu},

494: \label{Jadaggercomrel}

495: \end{eqnarray}

496: which define the transformation properties of the operators

497: $a_{r\mu}$, $a^\dagger_{r\mu}$ under infinitesimal rotations.  We

498: define a finite rotation operator in axis-angle or Euler angle form by

499: \begin{eqnarray}

500: U({\bf n},\theta) &= \exp(-i\theta {\bf n} \cdot {\bf J}),

501: \label{Uaxisangledef} \\

502: U(\alpha,\beta,\gamma) &= U({\bf z},\alpha) U({\bf y},\beta)

503: U({\bf z},\gamma),

504: \label{UEulerangledef}

505: \end{eqnarray}

506: where ${\bf n}$ is a unit vector defining an axis and $\theta$ an

507: angle of rotation about that axis, and where ${\bf x}$, ${\bf y}$ and

508: ${\bf z}$ are respectively the unit vectors along the three coordinate

509: axes.  The $U$ operators form a faithful representation of $SU(2)$. 

510: 

511: We use the symbol $u({\bf n},\theta)$ or $u(\alpha,\beta,\gamma)$

512: for the $2 \times 2$ matrices belonging to $SU(2)$, in axis-angle or

513: Euler angle parameterization (not to be confused with the $U$

514: operators that act on the Hilbert space of the $2N$ oscillators).  Thus

515: \begin{equation}

516: u({\bf n},\theta) = 

517: \exp(-i\theta{\bf n}\cdot \bsigma/2)=

518: \cos\theta/2 - i{\bf n}\cdot {\bsigma}

519: \sin\theta/2.

520: \label{udef}

521: \end{equation}

522: The exponentiated versions of equations (\ref{Jacomrel}) and

523: (\ref{Jadaggercomrel}) are

524: \begin{equation}

525:   \eqalign{

526:     U^\dagger a_{r\mu} U &= \sum_\nu u_{\mu\nu} \, a_{r\nu}, \\

527:     U^\dagger a^\dagger_{r\mu} U &= \sum_\nu

528:     a^\dagger_{r\nu}\, (u^{-1})_{\nu\mu},}

529:   \label{rotatea}

530: \end{equation}

531: where both $U$ and $u$ have the same parameterization. In the language

532: of irreducible tensor operators the pair of operators

533: $(a_{r1},a_{r2})$ transforms as a spin-$1/2$ operator.

534: 

535: Similarly, vector operators are the angular momenta themselves, which

536: satisfy the conjugation relations,

537: \begin{equation}

538:   U^\dagger {\hat J}_{ri} U = \sum_j R_{ij} {\hat J}_{rj},

539:   \label{rotateJr}

540: \end{equation}

541: where $R$ is the $3\times 3$ orthogonal rotation matrix with the same

542: axis and angle as $U$.  The relation between $R$ and $u$ (with the

543: same axis and angle) is

544: \begin{equation}

545:   R_{ij} = \frac{1}{2} \mathop{\rm tr}(U^\dagger \sigma_i U \sigma_j).

546:   \label{Ruproj}

547: \end{equation}

548: This is the usual projection from $SU(2)$ to $SO(3)$, in which the

549: inverse image of a given $R \in SO(3)$ is a pair $(u,-u)$ in $SU(2)$.

550: 

551: \section{The Wigner $3j$-symbols in the Schwinger Model}

552: \label{3jSchwinger}

553: 

554: We now define the $3j$-symbols in the context of the Schwinger

555: model.  We take the $3j$-model, $N=3$.  One complete set of commuting

556: observables on the Hilbert space ${\cal H} \otimes {\cal H}

557: \otimes {\cal H}$ is $({\hat I}_1, {\hat I}_2, {\hat I}_3, {\hat

558: J}_{1z}, {\hat J}_{2z}, {\hat J}_{3z})$, with  corresponding eigenstates

559: $\vert j_1j_2j_3m_1m_2m_3 \rangle = \vert j_1 m_1 \rangle \vert j_2

560: m_2 \rangle \vert j_3 m_3\rangle$.  Another complete set arises in the

561: usual problem of addition of three angular momenta, in which we

562: consider the values of $j$ and $m$ (the quantum numbers of ${\hat{\bf

563: J}}^2$ and ${\hat J}_z$) that occur in the product space ${\cal

564: H}_{j_1} \otimes {\cal H}_{j_2} \otimes {\cal H}_{j_3}$ for fixed

565: values of $(j_1,j_2,j_3)$, a subspace of ${\cal H} \otimes {\cal H}

566: \otimes {\cal H}$.  The set of the five commuting operators $({\hat

567: I}_1, {\hat I}_2, {\hat I}_3, {\hat J}_3, {\hat{\bf J}}^2)$ that

568: arises in this way is however not complete (the simultaneous

569: eigenstates in general possess degeneracies), so to resolve these we

570: introduce a sixth commuting operator, conventionally taken to be

571: ${\hat{\bf J}}^2_{12} = ({\hat{\bf J}}_1 + {\hat{\bf J}}_2)^2$ with

572: quantum number $j_{12}$ (${\hat{\bf J}}_{23}^2$ or ${\hat{\bf

573: J}}_{13}^2$ will also work).

574: 

575: The Wigner $3j$-symbols only involve the case $j=0$, but we mention

576: the others anyway because the foliation of the classical phase space

577: into Lagrangian manifolds involves the other values.  The usual rules

578: for the addition of angular momenta show that if $(j_1,j_2,j_3)$

579: satisfy the triangle inequality, then there exists precisely a

580: one-dimensional subspace of ${\cal H}_{j_1} \otimes {\cal H}_{j_2}

581: \otimes {\cal H}_{j_3}$ with $j=0$; if they do not, then no such

582: subspace exists.  If we enlarge our point of view to the full Hilbert

583: space ${\cal H} \otimes {\cal H} \otimes {\cal H}$, then there is an

584: infinite dimensional subspace with $j=0$, a basis in which is

585: specified by all triplets $(j_1,j_2,j_3)$ that satisfy the triangle

586: inequalities.  If $j=0$, then the quantum number $m$ is superfluous,

587: since $m=0$; the quantum number $j_{12}$ is superfluous as well, since

588: $j_{12}=j_3$.  

589: 

590: We note that if $\langle \psi \vert {\hat{\bf J}}^2

591: \vert \psi \rangle=0$ for any state $\vert \psi \rangle$, then ${\hat

592: J}_i \vert \psi \rangle=0$ for $i=1,2,3$.  Although the components of

593: ${\hat{\bf J}}$ do not commute and so do not possess simultaneous

594: eigenstates in general, the case of a state with $j=0$ is an

595: exception, since it is a simultaneous eigenstate of all three

596: components of ${\hat{\bf J}}$ with eigenvalues 0.  With this in mind

597: we denote the basis of states in the subspace of the full Hilbert

598: space with $j=0$ by $\vert j_1j_2j_3 {\bf 0} \rangle$, where the zero

599: vector ${\bf 0}$ indicates the vanishing eigenvalues of ${\hat{\bf

600: J}}$.  These basis states are also eigenstates of the operators

601: ${\hat{\bf J}}_{ij}^2$, for example, ${\hat{\bf J}}_{12}^2 \vert

602: j_1j_2j_3 {\bf 0}\rangle = j_3(j_3+1) \vert j_1j_2j_3 {\bf

603: 0}\rangle$.  

604: 

605: When the phase of the state $\vert j_1j_2j_3 {\bf 0} \rangle$ is

606: chosen to agree with Wigner's convention for the phases of the

607: $3j$-symbols, we have

608: \begin{equation}

609:   \left(

610:   \begin{array}{ccc}

611:     j_1 & j_2 & j_3 \\

612:     m_1 & m_2 & m_3 

613:   \end{array}

614:   \right) = \langle j_1j_2j_3 m_1m_2m_3 \vert j_1 j_2 j_3 {\bf 0}

615:   \rangle.

616: \label{3jdef}

617: \end{equation}

618: In this manner we have expressed the $3j$-symbol as a matrix element

619: connecting the eigenstates of two sets of observables, $({\hat I}_1,

620: {\hat I}_2, {\hat I}_3, {\hat J}_{1z}, {\hat J}_{2z}, {\hat J}_{3z})$

621: on the left and $({\hat I}_1, {\hat I}_2, {\hat I}_3, {\hat J}_x,

622: {\hat J}_y, {\hat J}_z)$ on the right.  Since the second set is

623: noncommuting, we will require a generalization of

624: (\ref{abmatrixelement}) to compute the semiclassical approximation to

625: the $3j$-symbols.

626: 

627: \section{Classical mechanics of the Schwinger model}

628: \label{cmSchwinger}

629: 

630: The classical mechanics of the Schwinger model must be well understood

631: in order to carry out a semiclassical analysis.  A general reference

632: on the classical mechanics of integrable systems from the modern point

633: of view is Cushman and Bates (1997), where harmonic oscillators in

634: particular are treated.  

635: 

636: \subsection{The $1j$-model}

637: 

638: We start with the $1j$-model, defining two classical oscillators

639: $H_\mu = (1/2)(x_\mu^2 + p_\mu^2)$, and $H=\sum_\mu H_\mu$, as in the

640: quantum case.  The classical configuration space is ${\mathbb R}^2$

641: and the phase space is ${\mathbb R}^4$.  We introduce complex

642: coordinates on phase space $z_\mu = (x_\mu +i p_\mu)/\sqrt{2}$ and

643: ${\bar z}_\mu = (x_\mu - i p_\mu)/\sqrt{2}$, where we use an overbar

644: for complex conjugation.  These are the Weyl symbols of the operators

645: $a_\mu$, $a^\dagger_\mu$.  The complex coordinates $z_\mu$, ${\bar

646: z}_\mu$ allow us to identify the phase space ${\mathbb R}^4$ with

647: ${\mathbb C}^2$, that is, knowledge of $z_1$ and $z_2$ allows us to

648: find all four real coordinates $(x_1,x_2,p_1,p_2)$, since the ${\bar

649: z}$'s are complex conjugates of the $z$'s.  As we shall see,

650: coordinates $(z_1,z_2)$, arranged as a 2-component column vector,

651: transform as a spinor under certain $SU(2)$ transformations.

652: Variables $z_\mu$ and $i{\bar z}_\mu$ are canonically conjugate ($q$'s

653: and $p$'s respectively), so that the Poisson bracket of two functions

654: $f$ and $g$ on phase space can be written,

655: \begin{eqnarray}

656:   \{f,g\} &= \sum_\mu \left(

657:   \frac{\partial f}{\partial x_\mu}

658:   \frac{\partial g}{\partial p_\mu}-

659:   \frac{\partial f}{\partial p_\mu}

660:   \frac{\partial g}{\partial x_\mu}\right) \nonumber \\

661:   &=

662:   \sum_\mu \left(

663:   \frac{\partial f}{\partial z_\mu}

664:   \frac{\partial g}{\partial (i{\bar z}_\mu)}-

665:   \frac{\partial f}{\partial (i{\bar z}_\mu)}

666:   \frac{\partial g}{\partial z_\mu}\right).

667:   \label{PB}

668: \end{eqnarray}

669:    

670: The basic building blocks of the classical Schwinger model are the function  

671: \begin{equation}

672:   I=\frac{1}{2} \sum_\mu {\bar z}_\mu z_\mu = 

673:   \frac{1}{2}\sum_\mu |z_\mu|^2,

674:   \label{1jIdef}

675: \end{equation}

676: and the three functions

677: \begin{equation}

678: J_i = \frac{1}{2} \sum_{\mu\nu} {\bar z}_\mu \, \sigma^i_{\mu\nu} \,

679: z_\nu,

680: \label{classJidef}

681: \end{equation}

682: for $i=1,2,3$, which define a classical angular momentum vector.  We

683: also define ${\bf J}^2 = \sum_i J_i^2$.  These functions satisfy the

684: identity ${\bf J}^2 = I^2$ and the Poisson bracket relations

685: $\{I,J_i\}=0$, $\{J_i, J_j\} = \sum_k \epsilon_{ijk} J_k$, $\{J_i,

686: {\bf J}^2\}=0$.

687: 

688: There are two groups of interest that act on the phase space ${\mathbb

689: R}^4$ or ${\mathbb C}^2$. The first is $U(1)$, generated by $I$.

690: Hamilton's equations for $I$ are

691: \begin{equation}

692:   \frac{d z_\mu}{d \psi} = \frac{\partial I}{\partial (i{\bar z}_\mu)} =

693:   -\frac{i}{2} z_\mu,

694:   \qquad

695:   \frac{d (i{\bar z}_\mu)}{d\psi} = 

696:   -\frac{\partial I}{\partial z_\mu} = -\frac{1}{2} {\bar z}_\mu,

697:   \label{Iflowodes}

698: \end{equation}

699: where $\psi$ is the parameter of the orbits.  These have the solution

700: \begin{equation}

701:   z_\mu(\psi) = \exp(-i\psi/2) \,z_\mu(0), \qquad

702:   {\bar z}_\mu(\psi) = \exp(i\psi/2) \,{\bar z}_\mu(0).

703:   \label{Iflow}

704: \end{equation}

705: Under the $I$-flow, the two-component spinor $(z_1,z_2)$ just gets

706: multiplied by an overall phase $\exp(-i\psi/2)$.  Except for

707: the special initial condition $(z_1,z_2)=(0,0)$ (the origin of phase

708: space ${\mathbb R}^4$ or ${\mathbb C}^2$), the orbits are circles with

709: period $4\pi$ with respect to the variable $\psi$.  Henceforth when

710: citing equations such as (\ref{Iflowodes}) or (\ref{Iflow}) we shall

711: omit the second half, when it is simply the complex conjugate of the

712: first half.

713: 

714: We denote a value of $I$ by $j\ge0$.  This is convenient notation, but

715: in this classical context $j$ is a continuous variable not to be

716: identified with the quantum number of any operator (see

717: Sec.~\ref{BSquant}).  Except for the origin $j=0$, the level set $I=j$

718: (or equivalently, ${\bf J}^2 = j^2$) is the sphere $S^3$, which is

719: foliated into circles by the action (\ref{Iflow}). This foliation is

720: precisely the Hopf fibration (Frankel 1997, Nakahara 2003), yielding

721: the quotient space $S^2 = S^3/S^1$.

722: 

723: The second group acting on phase space is $SU(2)$, whose action is

724: generated by the $J_i$.  Explicitly, if ${\bf n}$ is a unit vector and

725: $\theta$ an angle, then the solutions of Hamilton's equations

726: \begin{equation}

727:   \frac{d z_\mu}{d\theta} = 

728:   \frac{\partial ({\bf n}\cdot {\bf J})}{\partial (i{\bar z}_\mu)}=

729: 	-\frac{i}{2} \sum_\nu ({\bf n}\cdot\bsigma)_{\mu\nu} \,

730: 	z_\nu

731:   \label{ndotJeqns}

732: \end{equation}

733: and its complex conjugate are

734: \begin{equation}

735:     z_\mu(\theta) = \sum_\nu u({\bf n},\theta)_{\mu\nu}\, z_\nu(0)

736:   \label{ndotJflow}

737: \end{equation}

738: and its complex conjugate.  These are the obvious classical analogs of

739: equations (\ref{rotatea}); notice that the period in $\theta$ is

740: $4\pi$.  It is because of this $SU(2)$ action that we say that

741: coordinates $(z_1,z_2)$ form a spinor.  This classical action of

742: $SU(2)$ can be understood as a subgroup of the classical group of

743: linear canonical transformations, $Sp(4)$ (Littlejohn 1986); in

744: general, $Sp(2N)$ possesses a subgroup $Sp(2N) \cap O(2N)$ that is

745: isomorphic to $U(N)$, which contains the subgroup $SU(N)$ (in this

746: case, $N=2$).  When the symplectic matrices lying in the $SU(2)$

747: subgroup are expressed in the complex basis $(z_\mu, i{\bar z}_\mu)$,

748: they block diagonalize with $u$ multiplying the $z$'s and ${\bar u}$

749: multiplying the ${\bar z}$'s.

750: 

751: Equation~(\ref{classJidef}) defines a map (a projection) $\pi:{\mathbb

752: R}^4 \;(\hbox{\rm or}\; {\mathbb C}^2) \to {\mathbb R}^3$, where

753: ${\mathbb R}^3$ is ``angular momentum space,'' the space with

754: coordinates $(J_1,J_2,J_3)$.  Here and below we use $\pi$ to denote

755: this map or its generalization to the $Nj$-model.  The map $\pi$ maps

756: a larger space onto a smaller one, and so is not one-to-one.  The

757: inverse image of a point ${\bf J}$ of angular momentum space is a set

758: of spinors that differ by an overall phase.  It is easy to see that

759: the definition (\ref{classJidef}) does not depend on the overall phase

760: of the spinor.  Thus, the inverse image is a circle, except in the

761: case ${\bf J}=0$ when it is a single point (the origin of

762: phase space ${\mathbb C}^2$ or ${\mathbb R}^4$).

763: 

764: These circles are precisely the orbits of the $I$-flow (\ref{Iflow}).

765: Any function $f$ that is constant on these circles projects onto a

766: well defined function on angular momentum space.  But such functions

767: are those that Poisson commute with $I$, $\{f,I\}$=0.  This includes

768: $I$ itself as well as the three $J_i$.  We can write such a function

769: as $f(z_1,z_2,{\bar z}_1,{\bar z}_2)$ or $f({\bf J})$.  Now if $f$ and

770: $g$ are any two such functions, then so is their Poisson bracket

771: $\{f,g\}$, as follows from the Jacobi identity, $\{\{f,g\},I\} =

772: \{f,\{g,I\}\} + \{g,\{I,f\}\}=0$.  Thus, this Poisson bracket can be

773: computed directly in angular momentum space without going back to the

774: bracket (\ref{PB}); the result is the Lie-Poisson bracket,

775: \begin{equation}

776:   \{f,g\} = {\bf J} \cdot \left(

777:   \frac{\partial f}{\partial {\bf J}} \times

778:   \frac{\partial g}{\partial {\bf J}} \right).

779:   \label{LiePoisson}

780: \end{equation}

781: 

782: Interpretations of these spaces may be given in terms of the theory of

783: ``reduction'' (Marsden and Ratiu, 1999).  Angular momentum space is

784: the Poisson manifold that results from Poisson reduction of the phase

785: space ${\mathbb R}^4$ under the $U(1)$ action (\ref{Iflow}) generated

786: by $I$.  It is not by itself an ordinary phase space (symplectic

787: manifold), which would have an even dimensionality, but it is foliated

788: into symplectic submanifolds (the symplectic leaves).  In this case

789: the symplectic leaves are the 2-spheres in angular momentum space,

790: that is, the level sets ${\bf J}^2 = j^2$, the images under $\pi$ of

791: the 3-spheres $I=j$ in ${\mathbb R}^4$ or ${\mathbb C}^2$.  Canonical

792: coordinates on a given 2-sphere ${\bf J}^2 = j^2$ are $(\phi,J_z)$, a

793: $(q,p)$ pair, where $J_z=j\cos\theta$ and where $(\theta,\phi)$ are

794: the usual spherical angles in angular momentum space.  Thus we have

795: \begin{equation}

796:   dq\wedge dp = d\phi \wedge d(j\cos\theta) = j\,\sin\theta\,d\theta

797:   \wedge d\phi = j\,d\Omega,

798:   \label{S2symplform}

799: \end{equation}

800: and the symplectic form on a given sphere is $j\,d\Omega$, where

801: $d\Omega$ is the element of solid angle. This is not the geometrical

802: solid angle in a Euclidean geometry on angular momentum space, which

803: would be $j^2 \, d\Omega$.  Another interpretation of angular momentum

804: space is that it is the dual of the Lie algebra of $SU(2)$, while

805: $\pi$, given by (\ref{classJidef}), is the momentum map of the

806: $SU(2)$ action (\ref{ndotJflow}).

807: 

808: We now have three spaces, the ``large phase space'' ${\mathbb R}^4$ or

809: ${\mathbb C}^2$, its image under $\pi$, ``angular momentum space''

810: ${\mathbb R}^3$, and its symplectic leaves, the ``small phase

811: spaces,'' the 2-spheres ${\bf J}^2 = j^2$. Angular momentum space is

812: useful for visualizing classical angular momentum vectors, but by

813: considering inverse projections under $\pi$ the corresponding

814: geometrical objects in the large phase space can be constructed.

815: Angular momentum space has been used since the time of the old quantum

816: theory for visualizing the classical limit of quantum angular momentum

817: operators; for example, one spoke of an angular momentum vector

818: ``precessing'' around the $z$-direction.  In reality, the

819: ``precession'' defines a manifold of classical states in the small

820: phase space that is a level set of a complete set of commuting

821: observables, that is, it is an invariant torus of an integrable system

822: (just a circle in the $1j$-model, where the commuting observables are

823: $I$ and $J_z$).

824: 

825: \subsection{The $Nj$-model}

826: 

827: We now consider the classical mechanics of the $Nj$-model, which is

828: mostly a simple generalization of the $1j$-model. We have $2N$

829: classical oscillators $H_{r\mu} = (1/2)(x_{r\mu}^2 + p_{r\mu}^2)$; the

830: configuration space is $({\mathbb R}^2)^N ={\mathbb R}^{2N}$ and the

831: ``large'' phase space is $({\mathbb R}^4)^N={\mathbb R}^{4N}$ or

832: $({\mathbb C}^2)^N={\mathbb C}^{2N}$.  We define $z_{r\mu}=(x_{r\mu} +

833: i p_{r\mu})/\sqrt{2}$, ${\bar z}_{r\mu} = (x_{r\mu} -i

834: p_{r\mu})/\sqrt{2}$, so a point in phase space can be thought of as a

835: collection of $N$ 2-spinors, $(z_{r1},z_{r2})$, $r=1,\ldots,N$.  We

836: make the obvious definitions (classical versions of equations

837: (\ref{Irdef})--(\ref{Jtotalsquareddef})),

838: \begin{eqnarray}

839:  I_r &=\frac{1}{2}\sum_\mu |z_{r\mu}|^2,

840:  \label{classIrdef} \\

841:  J_{ri} &= \frac{1}{2}\sum_{\mu\nu} {\bar z}_{r\mu}

842:  \, \sigma^i_{\mu\nu} \, z_{r\nu},

843:  \label{classJridef}

844: \end{eqnarray}

845: as well as ${\bf J}_r^2 = \sum_i J_{ri}^2$, $J_i = \sum_r J_{ri}$ or

846: ${\bf J} = \sum_r {\bf J}_r$, and ${\bf J}^2 = \sum_i J_i^2$.

847: 

848: We denote a value of the functions $I_r$ by $j_r \ge 0$; for positive

849: values $j_r>0$, $r=1,\ldots,N$, the level set $I_r = j_r$ (or ${\bf

850: J}_r^2 = j_r^2$) in the phase space $({\mathbb R}^4)^N$ is $S^3 \times

851: \ldots \times S^3= (S^3)^N$.  The flow generated by $I_r$ for a

852: specific value of $r$ is just multiplication of the $r$-th spinor

853: $(z_{r1},z_{r2})$ by a phase factor $\exp(-i\psi_r/2)$, as in

854: (\ref{Iflow}); the other spinors are not affected.  Thus the $N$

855: commuting flows generated by all the $I_r$'s constitute a $U(1)^N =

856: T^N$ action on the large phase space ($T^N$ is the $N$-torus).

857: 

858: Equation~(\ref{classJridef}) defines the projection map $\pi:({\mathbb

859: C}^2)^N \to ({\mathbb R}^3)^N$, the latter space being ``angular

860: momentum space'' for the $Nj$-model, with one copy of ${\mathbb R}^3$

861: for each classical angular momentum vector ${\bf J}_r$.  In view of

862: its importance, we write out the components of this map explicitly: 

863: \begin{eqnarray}

864:   J_{rx} &= \frac{1}{2}({\bar z}_{r1} z_{r2} + {\bar z}_{r2} z_{r1})

865:   = {\mathop {\rm Re}}({\bar z}_{r1} z_{r2}), 

866:   \label{Jxeqn} \\

867:   J_{ry} &= -\frac{i}{2}({\bar z}_{r1} z_{r2} - {\bar z}_{r2} z_{r1})

868:   ={\mathop {\rm Im}}({\bar z}_{r1} z_{r2}),

869:   \label{Jyeqn} \\

870:   J_{rz} &= \frac{1}{2}(|z_{r1}|^2 - |z_{r2}|^2).

871:   \label{Jzeqn}

872: \end{eqnarray}

873: Points of angular momentum space can be visualized as $N$ classical angular

874: momentum vectors, each living in its own angular momentum space, or $N$

875: such vectors all in the same 3-dimensional angular momentum space. The

876: inverse image under $\pi$ of a set of $N$ nonvanishing classical

877: angular momentum vectors is an $N$-torus in the large phase space,

878: generated by taking any point in the inverse image (a collection of

879: $N$ 2-spinors), and multiplying them by $N$ independent, overall phase

880: factors.  We denote the angles on this torus by $\psi_r$,

881: $r=1,\ldots,N$, which are the evolution parameters corresponding to

882: the $I_r$, as in (\ref{Iflow}); thus their periods are $4\pi$.  As

883: in the $1j$-model, angular momentum space $({\mathbb

884: R}^3)^N$ is a Poisson manifold, now with Poisson bracket

885: \begin{equation}

886:   \{f,g\} = \sum_r {\bf J}_r \cdot \left(

887:   \frac{\partial f}{\partial {\bf J}_r} \times

888:   \frac{\partial g}{\partial {\bf J}_r} \right).

889:   \label{NjLiePoisson}

890: \end{equation}

891: The symplectic leaves (the ``small phase spaces'') are the spaces $S^2

892: \times \ldots \times S^2 = (S^2)^N$ obtained by fixing the values of

893: $j_1, \ldots, j_N$, with canonical coordinates $(\phi_r,J_{rz})$ on

894: each sphere.

895: 

896: In the classical $Nj$-model any partial or total sum of the angular

897: momenta ${\bf J}_r$ generates an $SU(2)$ action on the large phase

898: space, generalizing equations~(\ref{ndotJeqns}) and (\ref{ndotJflow}) in

899: the $1j$-model, in that the $SU(2)$ matrix $u$ is applied to all

900: spinors $(z_{r1},z_{r2})$ whose $r$ values lie in the sum.  For

901: example, the total ${\bf J}$ rotates all spinors.

902: 

903: These $SU(2)$ actions on the large phase space project to $SO(3)$ actions

904: on angular momentum space.  Consider, for example, the $SU(2)$ action

905: generated by the total ${\bf J}$.  Along an orbit in the large phase

906: space generated by ${\bf n}\cdot {\bf J}$, parameterized by $\theta$,

907: we can follow the value of ${\bf J}_r$, giving us ${\bf J}_r(\theta)$,

908: an orbit in the small phase space (the projection under $\pi$ of the

909: first orbit).  The latter orbit is

910: \begin{equation}

911:   J_{ri}(\theta) = \sum_j R({\bf n},\theta)_{ij} \, J_{rj}(0),

912:   \label{Jrrotation}

913: \end{equation}

914: where $R({\bf n},\theta)$ is the $3 \times 3$ rotation associated with

915: $u({\bf n},\theta)$ according to (\ref{Ruproj}).  This is the

916: classical analog of (\ref{rotateJr}).  It follows

917: from (\ref{classJridef}) and the spinor adjoint equation,

918: $u^\dagger \sigma^i u = \sum_j R_{ij} \, \sigma^j$,

919: itself equivalent to (\ref{Ruproj}).  Thus, under the $SU(2)$

920: action on the large phase space generated by ${\bf J}$, the individual

921: vectors ${\bf J}_r$ rotate in their individual angular momentum spaces

922: by the corresponding $3 \times 3$ rotation.  For example, $J_z$

923: rotates all vectors ${\bf J}_r$ about the $z$-axis.  Because of the

924: two-to-one relation between $SU(2)$ and $SO(3)$, when the orbit in the

925: large phase space goes around once ($\theta$ goes from 0 to $4\pi$),

926: the angular momentum vectors go around twice in their individual

927: angular momentum spaces.  

928: 

929: We may visualize this action as in Fig.~\ref{SU2action}, where $A$

930: represents a point of angular momentum space (a set of $N$ classical

931: angular momentum vectors ${\bf J}_r$ in the $Nj$-model).  To obtain

932: the generic case we assume these vectors are linearly independent (in

933: particular, none of them vanishes).  In the figure, $T$ is the inverse

934: image of $A$ under $\pi$, an $N$-torus.  Point $a$ is any specific

935: point in the large phase space on this $N$-torus, to which the $SU(2)$

936: rotation $u({\bf n},\theta)$ is applied for $0\le\theta < 4\pi$.  That

937: is, we treat $a$ as initial conditions for the Hamiltonian flow

938: generated by ${\bf n}\cdot {\bf J}$, with $\theta$ as the parameter.

939: This generates the circle $C$ in the large phase space, which amounts

940: to rotating all $N$ spinors by the same $u({\bf n},\theta)$.  The

941: projection of the circle $C$ is a circle $D$ in angular momentum space

942: $({\mathbb R}^3)^N$, that is, all classical vectors ${\bf J}_r$ rotate

943: about ${\bf n}$ by angle $\theta$.  However, when the circle $C$ is

944: covered once, circle $D$ is covered twice.  This is because when

945: $\theta=2\pi$, the spinor rotation $u({\bf n},\theta)=-1$, so all

946: spinors in the large phase space are just multiplied by $-1$.  This is

947: illustrated as point $a'$ in the figure, where all spinors are $-1$

948: times their values at $a$.  Since $-1$ is just a phase factor, both

949: $a$ and $a'$ project onto the same point $A$ in angular momentum

950: space.  These are the only two points on $C$ that project onto $A$;

951: for $\theta$ not a multiple of $2\pi$, the spinor rotation $u({\bf

952: n},\theta)$ is not a multiplication by a phase factor.

953: 

954: Alternatively, we may apply the entire group $SU(2)$ to the original

955: point $a$ (not just rotations along a fixed axis).  Then the manifold

956: $C$ is the orbit of the $SU(2)$ action which is diffeomorphic to

957: $SU(2)$. The point $a'$ is the image of $a$ under $u=-1$, a specific

958: element in $SU(2)$, and once again it projects onto the original point

959: $A$ in the small phase space.  The manifold $D$ is the orbit of point

960: $A$ under the group $SO(3)$.  In the $1j$-model, it is just a sphere

961: in angular momentum space (all vectors that can be reached from the

962: original one by applying all rotations), while in the $Nj$-model for

963: $N>1$ $D$ is generically diffeomorphic to $SO(3)$ (it is the set of

964: all classical configurations of $N$ angular momentum vectors that can

965: be obtained from the original one by applying rigid rotations).

966: 

967: \section{The invariant $jm$-tori}

968: \label{jmtori}

969: 

970: In this section we continue with the classical point of view,

971: examining the classical manifolds corresponding to the left side of

972: the matrix element (\ref{3jdef}).  For this exercise and the rest of

973: the paper we adopt a $3j$-model ($N=3$).  The manifolds in question

974: are the level sets of the commuting functions $I_r$, $J_{rz}$,

975: $r=1,2,3$, or, equally well, of the functions $I_{r\mu} = (1/2)

976: |z_{r\mu}|^2$ for $r=1,2,3$, $\mu=1,2$, since $I_r = I_{r1}+I_{r2}$

977: and $J_{rz} = I_{r1} - I_{r2}$.  We denote the level sets by $I_r =

978: j_r$, $J_{rz} = m_r$ for contour values $j_r$, $m_r$, $r=1,2,3$, or,

979: equivalently, by

980: \begin{equation}

981: 	I_{r1} = \frac{1}{2}(j_r+m_r), \qquad

982: 	I_{r2} = \frac{1}{2}(j_r-m_r).

983: 	\label{ia1ia2}

984: \end{equation}

985: In spite of the notation, $j_r$ and $m_r$ take on continuous values

986: and are not necessarily the eigenvalues of any quantum operators.

987: Since the $I_{r\mu}$ are all nonnegative, we have

988: \begin{equation}

989:   j_r \ge 0, \qquad -j_r \le m_r \le j_r,

990:   \label{jmranges}

991: \end{equation}

992: the classical analogs of the usual inequalities in quantum mechanics.

993: 

994: Since each of the six $I_{r\mu}$ is a harmonic oscillator (times

995: $1/2$), the level set of the $I_{r\mu}$'s is an invariant torus of a

996: collection of harmonic oscillators.  Generically (for nonzero

997: amplitude in each oscillator, that is, when none of the quantities

998: $j_r \pm m_r$ vanishes) this is a 6-torus, upon which the coordinates

999: may be taken to be the six angles $\theta_{r\mu}$, the variables of

1000: evolution of the $I_{r\mu}$.  The Hamiltonian flow generated by

1001: $I_{r\mu}$ for a specific value of $r$ and $\mu$ just multiplies

1002: $z_{r\mu}$ for the same values of $r$ and $\mu$ by

1003: $\exp(-i\theta_{r\mu}/2)$, while leaving all other $z$'s unaffected.

1004: This is not an overall spinor rotation since the other half of the

1005: spinor containing the given $z_{r\mu}$ is not affected.  If viewed in

1006: the Cartesian $x_{r\mu}$-$p_{r\mu}$ phase plane, this flow is a

1007: clockwise rotation by angle $\theta_{r\mu}/2$, as illustrated in

1008: Fig.~\ref{xpplane}.  The period of the angles $\theta_{r\mu}$ is

1009: $4\pi$.  We agree to measure the angles $\theta_{r\mu}$ from the

1010: positive $x_{r\mu}$ axis, as in the figure, where $z_{r\mu}$ is real

1011: and positive (or zero); this is a specific convention for a set of

1012: canonical coordinates $(\theta_{r\mu}, I_{r\mu})$, $r=1,2,3$,

1013: $\mu=1,2$ on the large phase space.  The volume of the 6-torus with

1014: respect to the measure $d\theta_{11} \wedge \ldots

1015: \wedge d\theta_{32}$ is $(4\pi)^6$.

1016: 

1017: These tori are also the orbits of the flows generated by the

1018: observables $I_r$, $J_{rz}$.  We denote the evolution variables of the

1019: $I_r$ and the $J_{rz}$ by $\psi_r$ (as above) and $\phi_r$,

1020: respectively.  Each $J_{rz}$ generates an $SU(2)$ rotation about the

1021: $z$-axis on the spinor with the given value of $r$; thus each of the

1022: six angles $(\psi_r,\phi_r)$ has period $4\pi$.  However, when we

1023: allow all six angles $(\psi_r,\phi_r)$ to range from 0 to $4\pi$, the

1024: torus is actually covered eight times.  This can be seen from

1025: Fig.~\ref{SU2action}: a rotation by $2\pi$ in one of the $\phi$'s and

1026: one of the $\psi$'s returns us to the initial point (the path is $a$

1027: to $a'$ along $C$ in Fig.~\ref{SU2action}, then $a'$ to $a$ along

1028: $T$.)  Alternatively, we may consider the canonical transformation

1029: $(\theta_{r\mu}; I_{r\mu}) \to (\psi_r, \phi_r; I_r, J_{rz})$,

1030: generated by

1031: \begin{equation}

1032:   F_2(\theta_{r1}, \theta_{r2}, I_r, J_{rz}) =

1033:   \frac{1}{2}[\theta_{r1}(I_r + J_{rz}) + 

1034:     \theta_{r2}(I_r - J_{rz})],

1035:   \label{F2gf}

1036: \end{equation}

1037: which generates $I_r=I_{r1}+I_{r2}$, $J_{rz}=I_{r1}-I_{r2}$ and 

1038: \begin{equation}

1039:   \psi_r = \frac{1}{2}(\theta_{r1} + \theta_{r2}), \qquad

1040:   \phi_r = \frac{1}{2}(\theta_{r1} - \theta_{r2}),

1041:   \label{psiphiCT}

1042: \end{equation}

1043: so that the Jacobian in the angles is $(1/2)^3=1/8$.  To cover the

1044: torus precisely once we may let the $\psi_r$'s range from $0$ to

1045: $4\pi$ and the $\phi_r$'s from $0$ to $2\pi$, or vice versa; thus the

1046: volume of the torus with respect to $d\psi_1 \wedge d\psi_2 \wedge

1047: d\psi_3 \wedge d\phi_1 \wedge d\phi_2 \wedge d\phi_3$ is

1048: \begin{equation}

1049: 	V_{jm} = (2\pi)^3 (4\pi)^3.

1050: 	\label{Vjm}

1051: \end{equation} 

1052: 

1053: The angles $\phi_r$ defined in this way on the large phase space can

1054: be projected onto the small phase space, whereupon they coincide with

1055: the usual azimuthal spherical angles in the individual angular

1056: momentum spaces.  It is clear this must be so to within an additive

1057: constant, since the $J_{rz}$-flow is just an $SO(3)$ rotation about the

1058: $z$-axis in the $r$-th angular momentum space, but by our conventions

1059: even the additive constant comes out right.  To see this we note first

1060: of all that $\phi_r$ for a given $r$ is constant along the $I_s$-flows

1061: for all $s$, since the variables in question are members of a

1062: canonical coordinate system on the large phase space and satisfy

1063: $\{\phi_r, I_s\}=0$.  Thus, $\phi_r$, defined in the large phase

1064: space, projects onto a meaningful function in angular momentum space.

1065: Next, to compute the value of $\phi_r$ for a specific angular momentum

1066: vector ${\bf J}_r$, it suffices to take any point in the 3-torus that

1067: is the inverse image, that is, any value of the angles $\psi_s$ may be

1068: chosen.  For simplicity we take $\psi_r=0$, which implies

1069: $\theta_{r1}=\phi_r$ and $\theta_{r2}=-\phi_r$.  This in turn implies

1070: $z_{r1} = |z_{r1}| \exp(-i\phi_r/2)$, $z_{r2} = |z_{r2}|

1071: \exp(i\phi_r/2)$.  But by equations~(\ref{Jxeqn}) and (\ref{Jyeqn}), these

1072: imply $J_{rx} = J_{r\perp} \cos\phi_r$, $J_{ry} = J_{r\perp}

1073: \sin\phi_r$, where $J_{r\perp} = |z_{r1}||z_{r2}|$.

1074: 

1075: We shall henceforth call the level set $I_r=j_r$, $J_{rz}=m_r$ the

1076: ``$jm$-torus.''  This torus can be projected onto angular momentum

1077: space; we consider the generic case when $j_r \pm m_r \ne 0$ for all

1078: $r$, in which case the $jm$-torus is a 6-torus.  In this case, its

1079: image in angular momentum space is a 3-torus, which, since it is a

1080: surface on which $I_r=j_r$, is also a submanifold of the small phase

1081: space.  This is because the three $\psi_r$ angles just change the

1082: overall phases of the three spinors, without changing their image

1083: under $\pi$, so the three coordinates  on the projected 3-torus are the

1084: angles $\phi_r$.  The 3-torus in angular momentum space can be

1085: visualized as three classical vectors ${\bf J}_r$ in a single angular

1086: momentum space, with specified values of $m_r = J_{rz}$,

1087: ``precessing'' about the $z$-axis.  See Fig.~\ref{3Jcones}.  This is

1088: an example of how we shall visualize manifolds in the large phase

1089: space: The $jm$-torus, a six-dimensional manifold in the large phase

1090: space (itself with twelve dimensions), is visualized as three angular

1091: momentum vectors in three dimensional space, as in Fig.~\ref{3Jcones},

1092: defining a 3-torus by varying their azimuthal angles independently,

1093: and each point of this 3-torus is associated with another 3-torus, the

1094: inverse projection under $\pi$ of the given point, which consists of

1095: independently changing the overall phases of the three spinors.  The

1096: 6-dimensional $jm$-torus is thus conceived of as the Cartesian product

1097: $T^3 \times T^3$ (it is actually a trivial bundle).

1098: 

1099: \section{The Wigner manifold}

1100: \label{Wignermanifold}

1101: 

1102: Now we turn to the right hand side of the matrix element

1103: (\ref{3jdef}), containing the state $\vert j_1j_2j_3 {\bf 0}\rangle$.

1104: This state suggests that we examine the classical manifold upon which

1105: the $I_r$ have definite values, say, $I_r=j_r$, and upon which ${\bf

1106: J}^2=0$.  Again, we do not necessarily identify the $j_r$ with any

1107: quantum numbers, but it is convenient in the following to assume that

1108: none of the $j_r$'s vanishes.  

1109: 

1110: \subsection{Properties of the Wigner manifold}

1111: 

1112: Usually the dimensionality of a manifold can be guessed by counting

1113: the constraints that define it, for example, we expect the manifold in

1114: the large (twelve-dimensional) phase space upon which $I_r=j_r$, ${\bf

1115: J}_{12}^2 = j_{12}^2$, $J_z = m$ and ${\bf J}^2 = j^2$, for given

1116: contour values, to be six-dimensional (six constraints on twelve

1117: variables).  Indeed, for most values of $j$ this is correct, and the

1118: manifold in question is a 6-torus (by the Liouville-Arnold theorem,

1119: for certain ranges of the contour values).  These are the invariant

1120: tori that would be involved in the semiclassical treatment of the

1121: addition of three angular momenta, producing a nonzero result (the

1122: case $j\ne0$).  But this naive dimension count only works when the

1123: differentials of the functions in question are linearly independent

1124: (in particular, nonvanishing) on the manifold.  This condition breaks

1125: down when ${\bf J}^2 = j^2 = 0$, since ${\bf J}^2=0$ and ${\bf J}=0$

1126: imply one another, and $d({\bf J}^2)= 2{\bf J}\cdot d{\bf J}=0$.  In

1127: fact, just the four conditions $I_r=j_r>0$, ${\bf J}^2=j^2=0$ define a

1128: six-dimensional manifold in the twelve-dimensional large phase space

1129: (for certain ranges of the $j_r$).  To see this we notice first that

1130: since ${\bf J}^2=0$ implies $J_i=0$, $i=1,2,3$, and $J_z=m=0$ in

1131: particular, the $J_z$ constraint is not independent; neither is the

1132: ${\bf J}_{12}^2=j_{12}^2$ constraint, since when ${\bf J}=0$, $j_{12}

1133: = j_3$.  This is just as in the quantum case.  In fact, the manifold

1134: $I_r=j_r$, ${\bf J}^2=0$ is characterized equivalently but better by

1135: $I_r=j_r$, ${\bf J}=0$, since the six differentials $dI_r$ and $dJ_i$

1136: are linearly independent on it.  (Although the $J_i$ vanish on the

1137: manifold in question, their differentials do not.)  Thus, the naive

1138: count of dimensions works with the set $I_r$, $J_i$.

1139: 

1140: We shall call the manifold $I_r=j_r$, ${\bf J}=0$ in the

1141: large phase space the ``Wigner manifold,'' because

1142: it corresponds to the rotationally invariant state $\vert

1143: j_1j_2j_3 {\bf 0} \rangle$ introduced by Wigner in his definition of

1144: the $3j$-symbols.  The dimensionality of this manifold (six, in the

1145: appropriate ranges of the $j_r$'s) is the same as that of the

1146: invariant tori of any integrable system of six degrees of freedom, and

1147: indeed the same as that of the nearby invariant tori in phase space

1148: corresponding to the level sets of the functions $(I_1,I_2,I_3,{\bf

1149: J}_{12}^2,J_z,{\bf J}^2)$ when $j>0$.  The Wigner manifold, however,

1150: is not a torus.  This is not a contradiction of the Liouville-Arnold

1151: theorem, which requires that the classical observables making up the

1152: level set should Poisson commute.  In the case of the Wigner manifold,

1153: we do have $\{I_r,I_s\}=0$ and $\{I_r,J_i\}=0$, but $\{J_i, J_j\} =

1154: \sum_k \epsilon_{ijk}\, J_k$.  The Hamiltonian vector fields

1155: corresponding to the functions $I_r$, $J_i$ are linearly independent

1156: on the Wigner manifold, but the three $J_i$-flows do not commute (two

1157: Hamiltonian flows commute if and only if their Poisson bracket is a

1158: constant).  The Wigner manifold is, however, an orbit of the

1159: collective action of these Hamiltonian flows (any point can be reached

1160: from any other point by following the flows in some order).  These

1161: facts, information about the topology of the Wigner manifold, and the

1162: required ranges on the contour values $j_r$ will be clarified

1163: momentarily.

1164: 

1165: The Wigner manifold is also a Lagrangian manifold, like the invariant

1166: tori of an integrable system.  This means that the integral of $p\,dx$

1167: along the manifold is locally independent of path, so an action

1168: function $S(x,A)$ can be defined.  This function in turn is the

1169: solution of the simultaneous Hamilton-Jacobi equations for the

1170: observables $(I_1,I_2,I_3,J_x,J_y,J_z)$, call them $A_i$,

1171: $i=1,\ldots,6$ for short, of which the Wigner manifold is the level

1172: set.  

1173: 

1174: To prove that the Wigner manifold is Lagrangian, we note that the 

1175: differentials $dA_i$ are linearly independent, so the vector fields

1176: $X_i$ are, too, and span the tangent space to the Wigner manifold at

1177: each point.  Evaluating the symplectic form on these vector fields, we

1178: have $\omega(X_i,X_j) = -\{A_i,A_j\}$.  These Poisson brackets all

1179: vanish except for the $\{J_i,J_j\}$; the latter are nonzero at most

1180: points in phase space, but on the Wigner manifold where ${\bf J}=0$,

1181: these also vanish.  Thus the symplectic form restricted to the Wigner

1182: manifold vanishes, the condition that the Wigner manifold be

1183: Lagrangian.

1184:   

1185: To visualize the Wigner manifold we work our way up from angular

1186: momentum space to the large phase space.  First we attempt to

1187: construct three angular momentum vectors of given positive lengths

1188: $j_1,j_2,j_3$ that add up to the zero vector.  This can be done if and

1189: only if the $j_r$ satisfy the triangle inequalities, whereupon the

1190: values of the $j_r$'s (the lengths of the sides) specify a triangle

1191: that is unique to within orientation.  If we choose a standard or

1192: reference orientation for the triangle, then the three desired vectors

1193: are the vectors running along its sides.  Let us therefore assume the

1194: triangle inequalities are satisfied, and let us choose a standard

1195: orientation for the triangle by placing the ${\bf J}_3$ along the

1196: $z$-axis, ${\bf J}_1$ in the $x$-$z$ plane with $J_{1x}>0$, and ${\bf

1197: J}_2$ in the $x$-$z$ plane with $J_{2x}<0$, as illustrated in

1198: Fig.~\ref{Jtriangle}.  Given any two triangles with the same

1199: (positive) sides, there exists a unique rotation that maps one into

1200: the other; this fact and others regarding triangles are discussed in

1201: the context of the 3-body problem by Littlejohn and Reinsch (1995).

1202: In the present context this means that all classical configurations of

1203: three classical angular momentum vectors of fixed lengths that add up

1204: to the zero vector are related to any one such configuration, such as

1205: the one shown in Fig.~\ref{Jtriangle}, by a unique rotation.  Thus the

1206: manifold of such classical configurations in angular momentum space

1207: ${\mathbb R}^3 \times {\mathbb R}^3 \times {\mathbb R}^3$ or in the

1208: small phase space $S^2 \times S^2 \times S^2$ is diffeomorphic to

1209: $SO(3)$.

1210: 

1211: The Wigner manifold in the large phase space is now the inverse

1212: projection under $\pi$ of this $SO(3)$ manifold in angular momentum

1213: space.  Since the inverse image of any given point of angular momentum

1214: space is a 3-torus in the large phase space (obtained by varying the

1215: overall phases of the three spinors), the Wigner manifold is a 3-torus

1216: bundle over $SO(3)$, and is six-dimensional.  The bundle is

1217: nontrivial.

1218: 

1219: The Wigner manifold may also be visualized with the help of

1220: Fig.~\ref{Wman}, an elaboration of Fig.~\ref{SU2action}.  It is

1221: assumed that the three $j$'s are positive and satisfy the triangle

1222: inequality.  The lower part of this figure refers to angular momentum

1223: space, while the upper part refers to the large phase space.

1224: Projection $\pi$ maps between the two spaces.  Point $A$ in angular

1225: momentum space is a state of three classical angular momenta of the

1226: given lengths $j_r$ whose vector sum is zero (that is, the angular

1227: momenta define a triangle), in a definite orientation.  To be

1228: specific, let us say that $A$ is the configuration shown in

1229: Fig.~\ref{Jtriangle}.  By applying all $SO(3)$ rotations to $A$ we

1230: generate all orientations of the triangle, of which $B$ in the figure

1231: is one.  The lower circle in the figure represents the manifold of

1232: such configurations, diffeomorphic to $SO(3)$.

1233: 

1234: The inverse image of any point on this manifold under $\pi$ is a

1235: 3-torus in the large phase space.  The 3-tori above points $A$ and $B$

1236: are indicated schematically as lines $T_A$ and $T_B$ in the figure.

1237: Let $a$ be some point on $T_A$.  To be specific, if $A$ is the

1238: configuration of angular momentum vectors shown in

1239: Fig.~\ref{Jtriangle} and the stated conditions on the $j_r$ hold, then

1240: none of the three vectors lies on the negative $z$-axis.  This means

1241: that for any point on $T_A$, $|z_{r1}|^2$ is never zero, since by

1242: equations~(\ref{classIrdef}) and (\ref{Jzeqn}), we have 

1243: \begin{equation}

1244: 	|z_{r1}|^2 = j_r + J_{rz}, \qquad

1245:   	|z_{r2}|^2 = j_r - J_{rz},

1246:  	 \label{zrsquare}

1247: \end{equation}

1248: for $r=1,2,3$.  Thus by adjusting the overall phases of the three

1249: spinors $(z_{r1},z_{r2})$, we can make $z_{r1}$ real and positive for

1250: all $r=1,2,3$.  Let this be the point $a$ in Fig.~\ref{Wman}. 

1251: 

1252: It is notationally tempting to write $m_r$ for the value of $J_{rz}$,

1253: but we shall not do this in the context of the Wigner manifold,

1254: instead reserving the symbol $m_r$ for the contour value of $J_{rz}$

1255: on the $jm$-torus.

1256: 

1257: Now we apply spinor rotations to point $a$, that is, simultaneous

1258: multiplication of all three spinors $(z_{r1}, z_{r2})$ by the same

1259: element of $SU(2)$.  The orbit thereby generated is a manifold

1260: diffeomorphic to $SU(2)$, as indicated in Fig.~\ref{Wman}.  The

1261: projection of this manifold onto angular momentum space is the surface

1262: $SO(3)$ shown in the figure, that is, all orientations of the triangle

1263: are generated.  For example, in the 3-torus $T_B$ over the angular

1264: momentum triangle with orientation $B$, there is a point $b$ that can

1265: be reached from the given point $a$ by some spinor rotation.  The

1266: spinor rotation in question is one of the two that projects onto the

1267: $SO(3)$ rotation that maps $A$ into $B$, according to (\ref{Ruproj}).

1268: The orbit of reference point $a$ under the $SU(2)$ action therefore

1269: passes through the 3-tori over every possible orientation of the

1270: triangle.  In fact, it passes through each 3-torus twice, since the

1271: $SU(2)$ rotation $u=-1$ is just a phase factor.  This is the meaning

1272: of points $a'$ and $b'$ in the figure, which are related to points $a$

1273: and $b$ by multiplying all three spinors by $-1$.

1274: 

1275: Thus any point on the Wigner manifold can be reached from the

1276: reference point $a$ by applying some $SU(2)$ rotation, and then

1277: adjusting the overall phases of the three spinors $(z_{r1},z_{r2})$.

1278: The first step is equivalent to following along the Hamiltonian flows

1279: in the large phase space generated by the three $J_i$ (this creates the

1280: $SU(2)$ rotation), while the second is equivalent to following the

1281: Hamiltonian flows generated by the three $I_r$ (this changes the

1282: overall phases of the three spinors).  By letting the rotations range

1283: over all of $SU(2)$ and the three angles $\psi_r$ range from $0$ to

1284: $4\pi$, the Wigner manifold is covered twice.  Thus we obtain

1285: coordinates on the Wigner manifold $(\alpha,\beta,\gamma,

1286: \psi_1,\psi_2,\psi_3)$ (the first three of which are Euler angles on

1287: $SU(2)$).

1288: 

1289: Solving the simulataneous amplitude transport equations for the six

1290: observables $I_r$, $J_i$, $r,i=1,2,3$ requires us to find an invariant

1291: measure on the Wigner manifold, that is, one invariant under all of

1292: the corresponding Hamiltonian flows.  Details are presented in

1293: Sec.~\ref{ampdet}; for now we just guess that this measure is the

1294: Haar measure on the group $SU(2)$ times the obvious measure on the

1295: 3-tori generated by the $I_r$, namely,    

1296: \begin{equation}

1297:   \sin\beta \, d\alpha \wedge d\beta \wedge d\gamma \wedge d\psi_1

1298:   \wedge d\psi_2 \wedge d\psi_3,

1299:   \label{Wignermeasure}

1300: \end{equation}

1301: where $(\alpha,\beta,\gamma)$ are Euler angles on $SU(2)$.  The

1302: integral of this measure over the Wigner manifold is

1303: \begin{equation}

1304:   V_W = \frac{1}{2} (16\pi^2) (4\pi)^3 = 2^9 \pi^5,

1305:   \label{Wignervolume}

1306: \end{equation}

1307: where the $1/2$ compensates for the fact that the Wigner manifold is

1308: covered twice when the Euler angles run over $SU(2)$ and each $\psi_r$

1309: runs from 0 to $4\pi$.

1310: 

1311: \subsection{Angles related to the shape of the triangle}

1312: 

1313: Figure~\ref{Jtriangle} defines the angles $\eta_r$ as the angles

1314: opposite vectors ${\bf J}_r$.  Under our assumptions, these angles lie

1315: in the range $0 \le \eta_r \le \pi$.  By projecting all three vectors

1316: onto the directions parallel and orthogonal to each of the vectors in

1317: turn, we obtain a series of identities,

1318: \begin{eqnarray}

1319:   j_1 \cos\eta_2 + j_2 \cos \eta_1 + j_3 &=0, 

1320:   \label{jcosident} \\

1321:   j_1 \sin\eta_2 -j_2 \sin\eta_1 &=0,

1322:   \label{jsinident}

1323: \end{eqnarray}

1324: and four more obtained by cycling indices $1,2,3$.  These allow us to

1325: solve for the cosines of the angles $\eta_r$,

1326: \begin{equation}

1327:   \cos\eta_1 = \frac{j_1^2 -j_2^2 -j_3^2}{2j_2j_3},

1328:   \label{cosetadef}

1329: \end{equation}

1330: and cyclic permutations, which in view of the stated ranges on the

1331: angles allows all three angles $\eta_r$ to be uniquely determined as

1332: functions of $(j_1,j_2,j_3)$.  We shall regard angles $\eta_r$ as

1333: convenient substitutions for these definite functions of the lengths

1334: of the angular momentum vectors.

1335: 

1336: For the stated ranges on the $\eta_r$, the sines of the angles are

1337: nonnegative and are related to the area $\Delta$ of the triangle, as

1338: follows:

1339: \begin{eqnarray}

1340:   \fl\Delta&= \frac{1}{2} |{\bf J}_1 \times {\bf J}_2| =

1341:   \frac{1}{2} j_1j_2 \sin\eta_3 \nonumber \\

1342:   \fl &= \frac{1}{4} \sqrt{(j_1+j_2+j_3) (-j_1+j_2+j_3) (j_1-j_2+j_3)

1343:     (j_1+j_2-j_3)},

1344:   \label{Deltadef}

1345: \end{eqnarray}

1346: and cyclic permutations.  Some authors define $\Delta$ as the final

1347: square root (without the $1/4$).

1348: 

1349: \section{Intersections of manifolds}

1350: \label{intersections}

1351: 

1352: The stationary phase points of the matrix element (\ref{3jdef}) are

1353: the intersections of the $jm$-manifold and the Wigner manifold in the

1354: large phase space.  Thus we must use a version of

1355: (\ref{abmatrixelement1}) for the matrix element, rather than

1356: (\ref{abmatrixelement}).  In this section we study the intersections

1357: of the manifolds, continuing with a classical picture.

1358: 

1359: If the $jm$-torus and the Wigner manifold in the large phase space

1360: have a common point of intersection, then the projections of these two

1361: manifolds onto angular momentum space must have a common point of

1362: intersection.  The converse is also true:  if the projections have a

1363: point in common, then the inverse image of this point under $\pi$, a

1364: 3-torus which is the orbit of the three $I_r$-flows, must contain two

1365: points, one of which belongs to the $jm$-torus, and the other to the

1366: Wigner manifold.  But the 3-torus is the orbit of the $I_r$-flows, and

1367: these flows confine one to both the $jm$-torus and the Wigner

1368: manifold.  Therefore the entire 3-torus is common to both the

1369: $jm$-torus and the Wigner manifold.  Therefore to find intersections

1370: of the $jm$-torus and the Wigner manifold, we may first find the

1371: intersections of their projections under $\pi$.

1372: 

1373: The $jm$-torus projects onto a set of configurations of

1374: three angular momenta with given lengths and fixed values of $m_r =

1375: J_{rz}$, with arbitrary azimuthal angles, while the Wigner manifold

1376: projects onto configurations with the same lengths in which the vector

1377: sum of the angular momenta vanishes, forming a triangle, with

1378: arbitrary orientation.  Therefore to find a common point between these

1379: two sets of classical angular momenta configurations, we can either

1380: adjust the azimuthal angles of the angular momenta with given $m$

1381: values until a triangle is formed (total ${\bf J}=0$), or we can

1382: rotate a triangle from a given, reference orientation until the $m$

1383: values are the desired ones.  We choose the latter procedure.

1384: 

1385: Our reference orientation of the triangle is shown in

1386: Fig.~\ref{Jtriangle}, which is indicated schematically as the point

1387: $A$ in Fig.~\ref{Wman}.  We must rotate this reference orientation

1388: to obtain some prescribed values of $m_r$.  These values satisfy the

1389: relations (\ref{jmranges}), so in particular $|m_3|\le j_3$.  Thus by

1390: rotating the triangle in the reference orientation about the $y$-axis

1391: by a unique angle $\beta$, $0\le \beta \le \pi$, defined by

1392: \begin{equation}

1393:   m_3=j_3\cos\beta,

1394:   \label{betadef}

1395: \end{equation}

1396: we guarantee that ${\bf J}_3$ has the right projection.  The result of

1397: this rotation is shown in Fig.~\ref{beta}, for a certain negative

1398: value of $m_3$.

1399: 

1400: Next we rotate the triangle about the axis ${\bf J}_3$ by an angle

1401: $\gamma$, which does not change ${\bf J}_3$ or its projection, but

1402: which rotates ${\bf J_1}$ and ${\bf J}_2$ in a cone, as illustrated in

1403: Fig.~\ref{gamma}.  We wish to choose the angle $\gamma$ so that ${\bf

1404: J}_2$ has the desired projection $m_2$ onto the $z$-axis.  In

1405: Fig.~\ref{gamma}, ${\bf J}_1$ is not shown, but ${\bf J}_2$ rotates

1406: about the ${\bf J}_3$ direction, its tip sweeping out circle $C_3$.

1407: The circle $C_z$ in the figure is swept out by a vector of length

1408: $j_2$ and projection $m_2$ onto the $z$-axis (this vector is not

1409: shown).  Circles $C_3$ and $C_z$ intersect in two points $Q$ and $Q'$

1410: in the figure, which represent two orientations of the triangle that

1411: have the correct values of both $m_3$ and $m_2$.  Now the orientation

1412: of the triangle is fixed, so there is no more freedom to rotate ${\bf

1413: J}_1$.  In this final orientation the value of $J_{1z}$ is

1414: $-J_{2z}-J_{3z} = -m_2-m_3$, since ${\bf J}=0$ for the triangle.

1415: Either this value of $J_{1z}$ equals the value of $m_1$ associated

1416: with the $jm$-torus, or it does not.  If it does not, then there are

1417: no intersections between the $jm$-torus and the Wigner manifold.

1418: This is just the classical expression of the condition that the matrix

1419: element (\ref{3jdef}) vanishes unless $\sum_r m_r=0$.  Henceforth we

1420: assume that the $m_r$ values for the $jm$-torus do satisfy this

1421: condition.

1422: 

1423: In this case we may solve for the values of $\gamma$ associated with

1424: points $Q$ and $Q'$ in Fig.~\ref{gamma}.  Writing $R({\bf n},\theta)$

1425: for a 3-dimensional rotation by angle $\theta$ about axis ${\bf n}$,

1426: we have applied the rotation

1427: \begin{equation}

1428:   R({\bf j}_3, \gamma) R({\bf y},\beta) = R({\bf y},\beta)

1429:   R({\bf z},\gamma)

1430:   \label{Rprod}

1431: \end{equation}

1432: to the reference orientation in Fig.~\ref{Jtriangle}, where ${\bf

1433: j}_3$ is the unit vector in the direction ${\bf J}_3$ shown in

1434: Fig.~\ref{beta} (after the first rotation).  In the reference

1435: orientation the vectors are

1436: \begin{equation}

1437:   \fl{\bf J}_1 = j_1 \left(

1438:   \begin{array}{c}

1439:     \sin\eta_2 \\

1440:     0 \\

1441:     \cos\eta_2

1442:   \end{array}\right), \qquad

1443:   {\bf J}_2 = j_2 \left(

1444:   \begin{array}{c}

1445:     -\sin\eta_1 \\

1446:     0 \\

1447:     \cos\eta_1

1448:   \end{array} \right), \qquad

1449:   {\bf J_3}=j_3 \left(

1450:   \begin{array}{c}

1451:     0 \\ 0 \\ 1

1452:   \end{array}\right).

1453:   \label{Jreference}

1454: \end{equation}

1455: After applying rotation (\ref{Rprod}) these become

1456: \begin{eqnarray}

1457:   {\bf J}_1 &= j_1 \left(

1458:   \begin{array}{c}

1459:     \cos\beta\cos\gamma\sin\eta_2 + \sin\beta\cos\eta_2 \\

1460:     \sin\gamma \sin\eta_2 \\

1461:     -\sin\beta\cos\gamma\sin\eta_2 + \cos\beta\cos\eta_2

1462:   \end{array}\right),

1463: 	\nonumber \\

1464:   {\bf J}_2 &= j_2 \left(

1465:   \begin{array}{c}

1466:     -\cos\beta\cos\gamma\sin\eta_1 + \sin\beta\cos\eta_1 \\

1467:     -\sin\gamma\sin\eta_1 \\

1468:     \sin\beta\cos\gamma\sin\eta_1 + \cos\beta\cos\eta_1

1469:   \end{array}\right),

1470:   \nonumber \\

1471:   {\bf J}_3 &= j_3 \left(

1472:   \begin{array}{c}

1473:     \sin\beta \\

1474:     0 \\

1475:     \cos\beta

1476:   \end{array}\right).

1477:   \label{Jrotated}

1478: \end{eqnarray}

1479: We have already solved for $\beta$ in (\ref{betadef}); we may now

1480: solve for $\gamma$ by demanding either $J_{1z}=m_1$ or $J_{2z}=m_2$.

1481: These lead to

1482: \begin{equation}

1483:   \cos\gamma = 

1484:   \frac{j_1 \cos\beta\cos\eta_2 -m_1}{j_1\sin\beta\sin\eta_2} =

1485:   \frac{m_2-j_2 \cos\beta\cos\eta_1}{j_2\sin\beta\sin\eta_1}.

1486:   \label{gammadef}

1487: \end{equation}

1488: These two conditions are equivalent (under the assumption $\sum_r

1489: m_r=0$), as follows from the identities

1490: (\ref{jcosident})--(\ref{cosetadef}).  If the common value of the two

1491: expressions on the right hand side of (\ref{gammadef}) lies in the

1492: range $(-1,+1)$, then there are two real angles $\gamma$ satisfying

1493: (\ref{gammadef}), corresponding to the two points $Q$ and $Q'$ in

1494: Fig.~\ref{gamma}.  In this case the two manifolds have real

1495: intersections, and we are in the classically allowed region for the

1496: $3j$-symbol.  We let $\gamma$ represent the root (the ``principal

1497: branch'') in the range $[0,\pi]$, and $-\gamma$ the root (the

1498: ``secondary branch'') in the range $[-\pi,0]$.  Note that

1499: $\sin\gamma\ge0$ ($\le0$) on the principal (secondary) branch.  If the

1500: right hand side of (\ref{gammadef}) lies outside the range

1501: $[-1,1]$, then there are two complex roots for $\gamma$.  In this case

1502: the two manifolds have no real intersections, but they do have complex

1503: ones.  Only one of the two complex roots is picked up by the contour

1504: of integration used in obtaining the matrix element

1505: (\ref{abmatrixelement}) or (\ref{abmatrixelement1}), resulting in an

1506: exponentially decaying expression for the matrix element.  In this

1507: case we are in the classically forbidden region of the $3j$-symbol.

1508: In the following for simplicity we assume we are in the classically

1509: allowed region.

1510: 

1511: The points $Q$ and $Q'$ in Fig.~\ref{gamma} represent values of ${\bf

1512: J}_2$ in a single angular momentum space ${\mathbb R}^3$.  Taken with

1513: the values of ${\bf J}_1$ and ${\bf J}_3$, they specify points, call

1514: them $P$ and $P'$, in the combined angular momentum space ${\mathbb

1515: R}^3 \times {\mathbb R}^3 \times {\mathbb R}^3$ that lie on the

1516: intersection of the projections of the $jm$-torus and the Wigner

1517: manifold onto that space.  Then by applying rotations about the

1518: $z$-axis to $P$ and $P'$, we generate a pair of circles in angular

1519: momentum space.  Such rotations change neither the $z$-projections

1520: of the three vectors nor their vector sum (zero).  This is obviously a

1521: reflection of the fact that the operator $J_z$ defining the state on

1522: the right side of (\ref{3jdef}) is a function of the operators

1523: $(J_{1z}, J_{2z}, J_{3z})$ defining the state on the left.  Thus the

1524: projections of the $jm$-manifold and Wigner manifold under $\pi$

1525: intersect generically in a pair of circles. 

1526: 

1527: Thus the intersection of the $jm$-torus and the Wigner manifold in

1528: the large phase space is the inverse image of this pair of circles

1529: under $\pi$, generically a pair of 3-torus bundles over a circle.

1530: Since the $I_r$-flows and the $J_z$-flow commute, these bundles are

1531: trivial, in fact each is a 4-torus, on which coordinates are

1532: $(\psi_1,\psi_2,\psi_3,\phi)$, where $\phi$ is the angle of evolution

1533: along the $J_z$-flow.  The volume of either one of the 4-tori with

1534: respect to the measure $d\psi_1 \wedge d\psi_2 \wedge d\psi_3 \wedge

1535: d\phi$ is

1536: \begin{equation}\

1537: 	V_I = \frac{1}{2} (4\pi)^4,

1538: 	\label{VIdef}

1539: \end{equation}

1540: the factor of $1/2$ being explained by Fig.~\ref{SU2action}.  

1541: 

1542: The $jm$-torus and the Wigner manifold, both six-dimensional,

1543: intersect in a 4-torus because the the lists of functions defining the

1544: two manifolds, $(I_1,I_2,I_3,J_{1z},J_{2z},J_{3z})$ and

1545: $(I_1,I_2,I_3,J_x,J_y,J_z)$, have three functions in common while

1546: $J_z$ in the second list is a function of $(J_{1z},J_{2z},J_{3z})$ in

1547: the first list.  Below we will transform the functions to make both

1548: lists have explicitly four variables in common (see Eq.~(\ref{JzCT})).

1549:   

1550: \section{Action integrals}

1551: \label{actionintegrals}

1552: 

1553: Action integrals on the $jm$-torus and the Wigner manifold are needed

1554: for the phases in expresssions such as (\ref{abmatrixelement1}).  We

1555: only need the action function at some point on the intersection

1556: between the two manifolds, which gives us a lot of choice since that

1557: intersection is a 4-torus.  We continue with a classical picture in

1558: this section.

1559: 

1560: \subsection{Choosing reference points} 

1561: 

1562: Action integrals are defined relative to some initial or reference

1563: point on each manifold.  For the $jm$-torus, a convenient point is the

1564: one where $\theta_{r\mu}=0$, $r=1,2,3$, $\mu=1,2$, that is, the point

1565: where each $z_{r\mu}$ is real and nonnegative, as explained in

1566: Sec.~\ref{jmtori}.  According to (\ref{ia1ia2}), the spinors at

1567: this reference point are given explicitly by

1568: \begin{equation}

1569: 	\left(\begin{array}{c}

1570: 	z_{r1} \\ z_{r2}

1571: 	\end{array}\right) =

1572: 	\left(\begin{array}{c}

1573: 	\sqrt{j_r + m_r} \\

1574: 	\sqrt{j_r - m_r}

1575: 	\end{array}\right).

1576: 	\label{jmref}

1577: \end{equation}

1578: The projection of this point onto angular momentum space is a set of

1579: vectors ${\bf J}_r$, $r=1,2,3$ of given lengths $j_r$ that lie in the

1580: $x$-$z$ plane, with $J_{rz}=m_r$ and $J_x \ge 0$, as shown by

1581: (\ref{Jxeqn})--(\ref{Jzeqn}).  Such vectors are illustrated in

1582: Fig.~\ref{3Jcones}.

1583: 

1584: As for the Wigner manifold, it is convenient to take the reference point

1585: to be point $a$ in Fig.~\ref{Wman}, which is discussed in

1586: Sec.~\ref{Wignermanifold}.   This point projects onto the standard

1587: orientation of the triangle, point $A$ in Fig.~\ref{Wman}, where the

1588: angular momentum vectors have the values shown in

1589: (\ref{Jreference}).  At the point $a$, $z_{r1}$ is real and

1590: positive for all $r$, as explained in Sec.~\ref{Wignermanifold}.  For

1591: example, for $r=1$ this assumption combined with (\ref{zrsquare})

1592: implies $z_{11}=\sqrt{j_1+J_{1z}}$, which by (\ref{Jreference})

1593: becomes $z_{11}=\sqrt{j_1(1+\cos\eta_2)}=\sqrt{2j_1}\cos\eta_2/2$.

1594: Then (\ref{Jyeqn}) and $J_{1y}=0$ imply that $z_{12}$ is purely

1595: real, and (\ref{Jxeqn}) allows us to solve for $z_{12}$ in terms

1596: of $J_{1x}$, given by (\ref{Jreference}), producing finally

1597: $z_{12}=\sqrt{2j_1}\sin\eta_2$.  Proceeding similarly with the other

1598: two spinors $r=2,3$, we obtain the three spinors at the reference

1599: point $a$ on the Wigner manifold,

1600: \begin{eqnarray}

1601: 	\fl\left(\begin{array}{c}

1602: 	z_{11} \\ z_{12}

1603: 	\end{array}\right) &=

1604: 	\sqrt{2j_1} \left(\begin{array}{c}

1605: 	\cos\eta_2/2 \\

1606: 	\sin\eta_2/2

1607: 	\end{array}\right), \qquad

1608: 	\left(\begin{array}{c}

1609: 	z_{21} \\ z_{22}

1610: 	\end{array}\right) =

1611: 	\sqrt{2j_2} \left(\begin{array}{c}

1612: 	\cos\eta_1/2 \\

1613: 	-\sin\eta_1/2

1614: 	\end{array}\right), 

1615: 	\nonumber \\

1616: 	[3pt]

1617: 	\fl\left(\begin{array}{c}

1618: 	z_{31} \\ z_{32}

1619: 	\end{array}\right) &=

1620: 	\sqrt{2j_3} \left(\begin{array}{c}

1621: 	1 \\ 0

1622: 	\end{array}\right).

1623: 	\label{Wref}

1624: \end{eqnarray}

1625: 

1626: Now to obtain a point common to both the $jm$-torus and the Wigner

1627: manifold, we apply the spinor rotation 

1628: \begin{equation}

1629:   u({\bf y},\beta) u({\bf z},\gamma)= \left(

1630:   \begin{array}{cc}

1631:     e^{-i\gamma/2}\cos\beta/2 & -e^{i\gamma/2}\sin\beta/2 \\

1632:     e^{-i\gamma/2}\sin\beta/2 & e^{i\gamma/2}\cos\beta/2

1633:   \end{array} \right)

1634:   \label{uprod}

1635: \end{equation}

1636: to the reference spinors $(z_{r1},z_{r2})$, $r=1,2,3$,

1637: in (\ref{Wref}), where Euler angles $\beta$

1638: and $\gamma$ are defined by (\ref{betadef}) and

1639: (\ref{gammadef}).  We obtain either the principal branch or the

1640: secondary one by taking $\gamma\ge0$ or $\gamma\le0$, respectively.  

1641: The spinor rotation (\ref{uprod}) induces the $3\times 3$

1642: rotation on the angular momentum vectors shown in (\ref{Rprod}).

1643: Thus we obtain the spinors at the common point of intersection between

1644: the $jm$-torus and the Wigner manifold,

1645: \begin{eqnarray}

1646:   	\fl\left(\begin{array}{c} 

1647: 	z_{11} \\ z_{12} \end{array} \right) &=

1648:   	\sqrt{2j_1}  \left( \begin{array}{c}

1649: 	e^{-i\gamma/2} \cos\beta/2 \cos\eta_2/2 - 

1650:     	e^{i\gamma/2} \sin\beta/2 \sin\eta_2/2 

1651: 	\label{ip1} \\

1652:     	[3pt]

1653:     	e^{-i\gamma/2} \sin\beta/2 \cos\eta_2/2 + 

1654:     	e^{i\gamma/2 } \cos\beta/2 \sin\eta_2/2

1655:   	\end{array}\right), \\

1656:   	\fl\left(\begin{array}{c} 

1657: 	z_{21} \\ z_{22} \end{array}\right) &=

1658:   	\sqrt{2j_2}  \left( \begin{array}{c}

1659:     	e^{-i\gamma/2} \cos\beta/2 \cos\eta_1/2 + 

1660: 	e^{i\gamma/2} \sin\beta/2 \sin\eta_1/2 

1661: 	\label{ip2} \\

1662: 	[3pt]

1663:     	e^{-i\gamma/2} \sin\beta/2 \cos\eta_1/2 - 

1664: 	e^{i\gamma/2} \cos\beta/2 \sin\eta_1/2

1665:   	\end{array}\right), \\

1666:   	\fl\left(\begin{array}{c} 

1667: 	z_{31} \\ z_{32} \end{array}\right) &=

1668:   	e^{-i\gamma/2} \sqrt{2j_3}  \left(

1669:   	\begin{array}{c}

1670:     	\cos\beta/2 \\ \sin\beta/2

1671:   	\end{array}\right).

1672: 	\label{ip3}

1673: \end{eqnarray}

1674: One can easily check using (\ref{Jxeqn})--(\ref{Jzeqn}) that

1675: these spinors project onto the angular momentum vectors in

1676: (\ref{Jrotated}). 

1677: 

1678: \subsection{Computing the actions} 

1679: 

1680: In computing action integrals we use the identity,

1681: \begin{equation}

1682: 	\sum_{r\mu} p_{r\mu} \, dx_{r\mu} =

1683: 	\frac{i}{2} \sum_{r\mu} 

1684: 	({\bar z}_{r\mu} \, dz_{r\mu} -

1685: 	z_{r\mu} \, d{\bar z}_{r\mu}) +

1686: 	\frac{1}{2} d \sum_{r\mu} 

1687: 	x_{r\mu} p_{r\mu}.

1688: 	\label{diffformident}

1689: \end{equation}

1690: The integral of the left hand side is the usual action one would need

1691: for wave functions $\psi(x_{11}, \ldots, x_{32})$, but it can be

1692: replaced by the integral of the first differential form on the right,

1693: for the following reason.  First, the integral of the exact

1694: differential on the right contributes the difference in the function

1695: $(1/2)\sum_{r\mu} x_{r\mu} p_{r\mu}$ between the initial and final

1696: points.  But the final point is the common point of intersection

1697: between the $jm$-torus and the Wigner manifold, so this contribution

1698: cancels when we subtract actions as in (\ref{abmatrixelement1}).  As

1699: for the initial points on the two manifolds, these have been chosen

1700: (see Eqs.~(\ref{jmref}) and (\ref{Wref})) so that all $z_{r\mu}$ are

1701: purely real, or $p_{r\mu}=0$.  Thus the function in question vanishes

1702: at the initial points.  As for the integral of the first term on the

1703: right of (\ref{diffformident}), it can be written

1704: \begin{equation}

1705: 	S=

1706: 	\mathop{\rm Im} \int \sum_{r\mu}

1707: 	z_{r\mu} \, d{\bar z}_{r\mu}.

1708: 	\label{zaction}

1709: \end{equation}

1710: 

1711: For the action on the $jm$-torus between initial point (\ref{jmref})

1712: and final point (\ref{ip1})--(\ref{ip3}), we follow a path consisting

1713: of flows of the functions $I_{r\mu} = (1/2)|z_{r\mu}|^2$ taken one at

1714: a time by angles $\theta_{r\mu}$.  Along the $I_{r\mu}$-flow we have

1715: $d{\bar z}_{r\mu}/ d\theta_{r\mu} = (i/2) {\bar z}_{r\mu}$, so the

1716: contribution to $S$ is

1717: \begin{equation}

1718: 	\mathop{\rm Im} \int_0^{\theta_{r\mu}}

1719: 	\frac{i}{2} |z_{r\mu}|^2 \, d\theta_{r\mu} =

1720: 	I_{r\mu} \theta_{r\mu},

1721: 	\label{Irmucontrib}

1722: \end{equation}

1723: since $I_{r\mu}$ is constant along its own flow and since

1724: $\theta_{r\mu}=0$ at the reference point.  Thus the total action

1725: between initial and final points on the $jm$-torus is

1726: \begin{equation}

1727: 	S_{jm}=\sum_{r\mu} I_{r\mu} \theta_{r\mu}.

1728: 	\label{jmaction}

1729: \end{equation}

1730: Under the canonical transformation (\ref{F2gf}) this becomes

1731: \begin{equation}

1732: 	S_{jm} = \sum_r (I_r \psi_r + J_{rz} \phi_r) =

1733: 	\sum_r (j_r \psi_r + m_r \phi_r),

1734: 	\label{jmaction1}

1735: \end{equation}

1736: where in the final form we replace $I_r$ and $J_{rz}$ by their values

1737: on a given $jm$-torus.  

1738: 

1739: The angles $\theta_{r\mu}$ or $(\psi_r,\phi_r)$ are the coordinates

1740: of the final point specified by Eqs.~(\ref{ip1})--(\ref{ip3}).  The

1741: solutions of Hamilton's equations for the $I_r$-flow can be written

1742: $z_{r\mu}(\theta_{r\mu}) = z_{r\mu}(0) \exp(-i\theta_{r\mu}/2)$ (see

1743: Fig.~\ref{xpplane}) where the initial conditions are real and

1744: nonnegative, so we have $\theta_{r\mu} = 2 \mathop{\rm arg} {\bar

1745: z}_{r\mu}$.  Combining this and (\ref{psiphiCT}), we can write the

1746: action on the $jm$-torus as 

1747: \begin{equation}

1748: 	\fl S_{jm} = 2 \sum_{r\mu} I_{r\mu} 

1749: 	\mathop{\rm arg} {\bar z}_{r\mu} =

1750: 	\sum_r j_r \mathop{\rm arg}

1751: 	({\bar z}_{r1} {\bar z}_{r2}) + 

1752: 	\sum_r m_r \mathop{\rm arg}

1753: 	({\bar z}_{r1} z_{r2}).

1754: 	\label{jmaction2}

1755: \end{equation}

1756: Using Eqs.~(\ref{ip1})--(\ref{ip3}), this can be written,

1757: \begin{eqnarray}

1758: 	S_{jm} &= j_3\gamma

1759: 	+j_1 \mathop{\rm arg}(

1760: 	\cos\beta \sin\eta_2 + 

1761: 	\sin\beta \cos\gamma \cos\eta_2 +

1762: 	i\sin\beta \sin\gamma) \nonumber \\

1763: 	&+j_2 \mathop{\rm arg}(

1764: 	-\cos\beta \sin\eta_1 +

1765: 	\sin\beta \cos\gamma \cos\eta_1 +

1766: 	i\sin\beta \sin\gamma) \nonumber \\

1767: 	&+m_1 \mathop{\rm arg}(

1768: 	\sin\beta \cos\eta_2 +

1769: 	\cos\beta \cos\gamma \sin\eta_2 +

1770: 	i\sin\gamma \sin\eta_2) \nonumber \\

1771: 	&+m_2 \mathop{\rm arg}(

1772: 	\sin\beta \cos\eta_1 -

1773: 	\cos\beta \cos\gamma \sin\eta_1 -

1774: 	i\sin\gamma\sin\eta_1).

1775: 	\label{jmaction3}

1776: \end{eqnarray}

1777: 

1778: Here we have used the rule $\mathop{\rm arg}(ab) = \mathop{\rm arg}a +

1779: \mathop{\rm arg} b$, which is only valid for certain choices of branch

1780: of the arg function.  A more careful analysis shows that

1781: (\ref{jmaction3}) is the correct action along a certain path from the

1782: initial to final point (the principal branch) on the $jm$-torus if the

1783: range of the arg function is taken to be $[-\pi,\pi)$.  (The path is

1784: defined by $\gamma \le \theta_{11}, \theta_{22} \le 2\pi$, $-\gamma

1785: \le \theta_{12}, \theta_{21} \le \gamma$, that is, one integrates from

1786: 0 to these final $\theta$ values.)  In particular, this means that

1787: $\psi_1, \psi_2,\psi_3,\phi_1$ all lie in $[0,\pi]$, while $\phi_2$

1788: lies in $[-\pi,0]$.  These ranges on angles $\phi_1$, $\phi_2$ are

1789: also evident from Fig.~\ref{gamma}.  Similarly, for the secondary

1790: branch ($-\pi \le \gamma \le 0$, $\sin\gamma \le 0$) there exists a

1791: path such that with the same range on the arg function

1792: (\ref{jmaction3}) is still correct.  With these understandings, the

1793: values of $S_{jm}$ on the two branches differ by a sign.  We shall

1794: henceforth write $S_{jm}$ ($-S_{jm}$) for the principal (secondary)

1795: branch.

1796: 

1797: Equation~(\ref{jmaction}) can also be written in terms of $\cos^{-1}$

1798: functions.  We note that $\mathop{\rm arg}({\bar z}_{11} {\bar

1799: z}_{12}) = \cos^{-1}[ \mathop{\rm Re} ({\bar z}_{11} {\bar z}_{12})

1800: /|z_{11} z_{12}|]$ and that $|z_{11} z_{12}| = (j_1^2 -m_1^2)^{1/2}$,

1801: etc.  We can also use (\ref{gammadef}) to eliminate $\cos\gamma$.  For

1802: the principal branch ($\gamma\ge0$) this gives

1803: \begin{eqnarray}

1804: 	\fl S_{jm} &= j_1 \cos^{-1} \left(

1805: 	\frac{j_1 \cos\beta - m_1 \cos\eta_2}

1806: 	{\sin\eta_2 J_{1\perp}}\right) +

1807: 	j_2 \cos^{-1} \left(

1808: 	\frac{m_2\cos\eta_1 - j_2 \cos\beta}

1809: 	{\sin\eta_1 J_{2\perp}}\right) \nonumber \\

1810: 	[3pt]

1811: 	\fl &+j_3 \cos^{-1} \left(

1812: 	\frac{j_1\cos\beta\cos\eta_2 -m_1}

1813: 	{j_1\sin\beta\sin\eta_2}\right) +

1814: 	m_1\cos^{-1}\left(

1815: 	\frac{j_1\cos\eta_2 -m_1\cos\beta}

1816: 	{\sin\beta J_{1\perp}}\right) \nonumber \\

1817: 	[3pt]

1818: 	\fl &-m_2 \cos^{-1} \left(

1819: 	\frac{j_2\cos\eta_1 -m_2 \cos\beta}

1820: 	{\sin\beta J_{2\perp}}\right),

1821: 	\label{jmaction4}

1822: \end{eqnarray}

1823: where

1824: \begin{equation}

1825: 	J_{r\perp} = \sqrt{j_r^2 - m_r^2},

1826: 	\label{Jrperpdef}

1827: \end{equation}

1828: and where the range of the $\cos^{-1}$ function is $[0,\pi]$.

1829: Finally, by using Eqs.~(\ref{cosetadef}) and (\ref{Deltadef}) these can

1830: be written explicitly in terms of the parameters $j_r$, $m_r$.  The

1831: result has the form of (\ref{jmaction1}), where

1832: \begin{equation}

1833: 	\psi_1 = \cos^{-1} \left(

1834: 	\frac{j_1^2(m_3 - m_2) + m_1(j_3^2 - j_2^2)}

1835: 	{4\Delta J_{1\perp}}\right),

1836: 	\label{psi1def}

1837: \end{equation}

1838: and cyclic permutations of indices, and where

1839: \begin{equation}

1840: 	\fl \phi_1 = \cos^{-1}\left(

1841: 	\frac{j_2^2 - j_3^2 -j_1^2 - 2m_1m_3}

1842: 	{2J_{1\perp} J_{3\perp}}\right), 

1843: 	\; \phi_2 = -\cos^{-1}\left(

1844: 	\frac{j_1^2 -j_3^2 -j_2^2 -2m_2m_3}

1845: 	{2 J_{2\perp} J_{3\perp}}\right),

1846: 	\label{phi12def}

1847: \end{equation}

1848: and where $\phi_3=0$.  

1849: 

1850: Now we consider the action on the Wigner manifold between the initial

1851: point (\ref{Wref}) and the final point (\ref{ip1})--(\ref{ip3}).  The

1852: path between these points is made up of the product of rotations

1853: (\ref{uprod}), so we consider the action integral (\ref{zaction})

1854: along a rotation by angle $\theta$ generated by ${\bf n}\cdot {\bf

1855: J}$.  Hamilton's equations (see Eq.~(\ref{ndotJeqns})) are $d{\bar

1856: z}_{r\mu}/d\theta = (i/2) \sum_\nu {\bar z}_{r\nu} ({\bf n} \cdot

1857: \bsigma)_{\nu\mu}$, so by (\ref{zaction}) we have

1858: \begin{equation}

1859: 	\fl S=\mathop{\rm Im} \int_0^\theta 

1860: 	\frac{i}{2} \sum_{r\mu\nu} {\bar z}_{r\nu}

1861: 	({\bf n} \cdot \bsigma)_{\nu\mu} z_{r\mu} \,

1862: 	d\theta =

1863: 	\int_0^\theta ({\bf n} \cdot {\bf J})\, 

1864: 	d\theta =

1865: 	({\bf n} \cdot {\bf J})\theta=0,

1866: 	\label{rotaction}

1867: \end{equation}

1868: where we use (\ref{classJridef}), the fact that ${\bf n} \cdot

1869: {\bf J}$ is constant along its own flow, and the fact that ${\bf J}=0$

1870: on the Wigner manifold.  The rotational action vanishes.

1871: 

1872: Thus the phase of the matrix element (\ref{3jdef}) is determined

1873: entirely by the action integral along the $jm$-torus, that is, to

1874: within a sign it is given by Eqs.~(\ref{jmaction})--(\ref{phi12def}).

1875: This is the phase function determined previously by Ponzano and Regge,

1876: Miller, and others, and we see that it is essentially a simple

1877: combination of the phases of the Schwinger oscillators.  We have, however,

1878: determined this phase function entirely within a classical model, that

1879: is, without imposing any quantization conditions on the manifolds.  

1880: 

1881: \section{Bohr-Sommerfeld quantization}

1882: \label{BSquant}

1883: 

1884: We do not need Bohr-Sommerfeld approximations to the eigenvalues of

1885: the operators involved in the $3j$-symbols because those eigenvalues

1886: are known exactly.  We must, however, quantize the $jm$-torus and the

1887: Wigner manifold, to obtain the wave functions whose scalar product is

1888: the $3j$-symbol.  We also need the Bohr-Sommerfeld rules to make the

1889: connection between the contour values for various classical functions

1890: and the standard quantum numbers of the associated operators.

1891: 

1892: To quantize a Lagrangian manifold we must first find the generators of

1893: the fundamental or first homotopy group of the manifold, that is, a

1894: set of closed contours in terms of which all closed contours can be

1895: generated by concatening curves.  In the following we shall call these

1896: generators ``basis contours,'' although technically the fundamental

1897: group, even when Abelian, is a group and not a vector space.  For

1898: example, in the familiar case of the invariant $n$-tori of integrable

1899: systems of $n$ degrees of freedom, the fundamental group is ${\mathbb

1900: Z}^n$, that is, an arbitrary closed contour is expressed as a ``linear

1901: combination'' of the $n$ basis contours with integer coefficients.

1902: The $n$ basis contours themselves go around the torus once in the $n$

1903: different directions.

1904: 

1905: After finding the basis contours, we compute the total phase

1906: associated with each of them, the sum of an action, the integral of

1907: $p\,dq$ around the contour, and a Maslov phase, which is $-\pi/2$

1908: times the Maslov index of the loop.  Both these phases are topological

1909: invariants and are additive when loops are concatenated.  Then we

1910: demand that the total phase be a multiple of $2\pi$; this is the

1911: consistency condition on the semiclassical wave function that selects

1912: out certain manifolds as being ``quantized.''

1913: 

1914: The Lagrangian manifolds we are interested in are level sets of a set

1915: of classical functions that are the principal symbols of a set of

1916: operators, which in our application need not commute.  Quantized

1917: Lagrangian manifolds support wave functions that are approximate

1918: eigenfunctions of the set of operators.  The corresponding eigenvalues

1919: are the contour values of the principal symbols, to within errors of

1920: order $\hbar^2$.

1921: 

1922: \subsection{Quantizing the $jm$-tori}

1923: 

1924: In the case of the $jm$-tori, whose fundamental group is ${\mathbb

1925: Z}^6$, we are dealing with the eigenfunctions of a set of independent

1926: harmonic oscillators, so the problem could not be more elementary from

1927: a semiclassical standpoint.  There are, however, interesting issues

1928: that arise.  Let the complete set of commuting quantum observables be

1929: $({\hat I}_r, {\hat J}_{rz})$, $r=1,2,3$, defined in (\ref{Irdef})

1930: and (\ref{Jridef}).  Let us denote the Weyl symbol of an operator

1931: ${\hat A}$ by $\mathop{\rm sym}({\hat A})$.  Then we have

1932: \begin{equation}

1933: 	\mathop{\rm sym}({\hat I}_r) =

1934: 	    I_r -\frac{1}{2}, \qquad

1935: 	\mathop{\rm sym}({\hat J}_{rz}) =

1936: 	J_{rz},

1937: 	\label{IJsymbols}

1938: \end{equation}

1939: where $I_r$ and $J_{rz}$ (the classical functions) are defined by

1940: Eqs.~(\ref{classIrdef}) and (\ref{classJridef}).  The operator ${\hat

1941: I}_r$ violates our assumption in Sec.~\ref{scwfis} that the commuting

1942: operators defining our integrable system should have Weyl symbols that

1943: are even power series in $\hbar$ (since the $-1/2$ is of order

1944: $\hbar$).  Our assumption is valid for the harmonic oscillators ${\hat

1945: H}_r = (1/2)\sum_\mu ({\hat x}_{r\mu}^2 + {\hat p}_{r\mu}^2)$, but in

1946: defining ${\hat I}_r$ in (\ref{Irdef}) we have subtracted the zero

1947: point energy, a constant of order $\hbar$, ${\hat I}_r = (1/2)({\hat

1948: H}_r - 1)$, so that the eigenvalues of ${\hat I}_r$ would be the

1949: conventional quantum numbers $j_r$ for an angular momentum, and so

1950: that the identity (\ref{JIidentity}) would have a familiar form.  In

1951: the following we shall take the principal symbol of ${\hat I}_r$ to be

1952: the whole symbol, including the $-1/2$.  This achieves the same

1953: results we would have had if we had worked with ${\hat H}_r$ instead

1954: of ${\hat I}_r$ and defined the principal symbol as the leading term

1955: in $\hbar$, as in Sec.~\ref{scwfis}, since $\mathop{\rm sym}({\hat

1956: H}_r) = 2I_r$.  We must be careful, however, since the principal

1957: symbol of ${\hat I}_r$ is not $I_r$.  For most of the other operators

1958: we shall use, the principal symbol is obtained simply by removing the

1959: hat (for example, $J_{rz}$ above).

1960: 

1961: The basis contours on the $jm$-torus are most easily expressed as the

1962: contours on which each of the angles $\theta_{r\mu}$ is allowed to go

1963: from 0 to $4\pi$ while all other $\theta_{r\mu}$'s are held fixed.  We

1964: may also use any linear combination of these contours with integer

1965: coefficients and unit determinant.  In terms of the angles $\psi_r$,

1966: $\phi_r$, given by (\ref{psiphiCT}), a convenient choice is to

1967: take one basis contour as the path on which one $\psi_r$ goes from $0$

1968: to $4\pi$ while all others $\psi_r$'s and all $\phi_r$'s are held

1969: fixed; this is following the $I_r$-flow for elapsed angle $\psi_r=4\pi$.

1970: A second basis contour may be taken to be the path on which $\psi_r$

1971: goes from 0 to $2\pi$, and then $\phi_r$ goes from 0 to $2\pi$; the

1972: two legs involve following the $I_r$-flow and then the $J_{rz}$-flow,

1973: each for elapsed angle $2\pi$.  Doing this for $r=1,2,3$ gives us 

1974: six basis contours on the $jm$-torus.

1975: 

1976: The action along the first basis contour is computed as in

1977: (\ref{Irmucontrib}).  Hamilton's equations for $I_r$ are $d{\bar

1978: z}_{r\mu}/d\psi_r = (i/2){\bar z}_{r\mu}$, so we obtain $S_{r1}=4\pi

1979: I_r$, where $S_{r1}$ refers to the action along the first basis

1980: contour.  For the second basis contour, the first leg contributes an

1981: action $2\pi I_r$, while the second leg, which follows the flow

1982: generated by $J_{rz}$, is a rotation whose action may be computed as

1983: in (\ref{rotaction}), but with ${\bf J}$ replaced by ${\bf J}_r$ since

1984: we do not sum over $r$.  Thus the final answer does not vanish (${\bf

1985: J}_r$ is nonzero on the $jm$-torus), and the contribution from the

1986: second leg is $2\pi J_{rz}$, or $S_{r2} = 2\pi(I_r + J_{rz})$.

1987: Altogether, we have

1988: \begin{equation}

1989: 	S_{r1} = 4\pi j_r, \qquad

1990: 	S_{r2} = 2\pi(j_r + m_r),

1991: 	\label{jmactions}

1992: \end{equation}

1993: where the 1 and 2 refer to the first and second basis contours

1994: associated with a particular value of $r$, and where we have replaced

1995: $I_r$ and $J_{rz}$ by their contour values $j_r$ and $m_r$ on the

1996: $jm$-torus.

1997: 

1998: Next we need the Maslov indices along the two basis contours.  Here we

1999: follow the computational method described in Littlejohn and Robbins

2000: (1987), which uses the determinant of complex matrices and which is

2001: based ultimately on Arnold (1967).  Similar techniques are discussed

2002: by Mishchenko \etal\ (1990). The method works for finding Maslov

2003: indices along closed curves on orientable Lagrangian manifolds in

2004: ${\mathbb R}^{2n}$.  To describe the method we adopt a general

2005: notation, in which global coordinates on phase space are $(q_1,

2006: \ldots, q_n,p_1, \ldots, p_n)$.  We suppose that there exists a set of

2007: $n$ vector fields on the Lagrangian manifold, linearly independent at

2008: each point, so that they span the Lagrangian tangent

2009: plane at each point.  In our applications, these are the Hamiltonian

2010: vector fields associated with a set of functions $(A_1, \ldots, A_n)$.

2011: We consider the rate of change of the quantities $q_i -i p_i$ along

2012: the $j$-th vector field, which is the Poisson bracket

2013: $\{q_i-ip_i,A_j\}$.  The set of these Poisson brackets forms an $n

2014: \times n$ complex matrix $M_{ij}$ that is never singular, so $\det M$

2015: traces out a closed loop in the complex plane without passing through

2016: the origin when we go around a closed loop on the Lagrangian manifold.

2017: Then the Maslov index $\mu$ associated with this loop is given by

2018: \begin{equation}

2019: 	\mu = 2 \mathop{\rm wn} \det M,

2020: 	\label{mudef}

2021: \end{equation}

2022: where wn refers to the winding number of the loop in the complex

2023: plane, reckoned as positive in the counterclockwise direction.  The

2024: winding number is invariant when $M_{ij}$ is multiplied by any nonzero

2025: complex constant (or constant matrix), so such constants can be

2026: dropped in the calculation.

2027: 

2028: For the $jm$-torus, we identify the $q$'s and $p$'s with the

2029: coordinates $x_{r\mu}$ and $p_{r\mu}$, and the $A$'s with the

2030: functions $(I_1,I_2,I_3, J_{1z},J_{2z},J_{3z})$.  Then we can replace

2031: $q_i-ip_i$ by ${\bar z}_{r\mu}$, dropping the $1/\sqrt{2}$.  The

2032: needed matrix elements are

2033: \begin{equation}

2034: 	\eqalign{

2035: 	\{{\bar z}_{r\mu}, I_s\} &= 

2036: 	i\frac{\partial I_s}{\partial z_{r\mu}} =

2037: 	\frac{i}{2} \delta_{rs} \, {\bar z}_{r\mu}, \\

2038: 	\{{\bar z}_{r1}, J_{sz}\} &=

2039: 	i\frac{\partial J_{sz}}{\partial z_{r1}} =

2040: 	\frac{i}{2} \delta_{rs} \, {\bar z}_{r1}, \\

2041: 	\{{\bar z}_{r2}, J_{sz}\} &=

2042: 	i\frac{\partial J_{sz}}{\partial z_{r2}} =

2043: 	-\frac{i}{2} \delta_{rs} \, {\bar z}_{r2}.}

2044: 	\label{Mijjm}

2045: \end{equation}

2046: We drop the constant $i/2$ on the right hand side, and choose the

2047: ordering $(I_1,J_{1z}, I_2,J_{2z}, I_3,J_{3z})$ for the functions.

2048: Then the $6\times 6$ matrix block diagonalizes into three $2 \times 2$

2049: blocks, and we find

2050: \begin{equation}

2051: 	\det M = {\bar z}_{11} {\bar z}_{12} {\bar z}_{21}

2052: 	{\bar z}_{22} {\bar z}_{31} {\bar z}_{32},

2053: 	\label{detMjm}

2054: \end{equation}

2055: to within a constant.  

2056: 

2057: Along the flow of one of the $I_r$'s we have ${\bar z}_{r\mu}(\psi_r)

2058: = \exp(i\psi_r/2) {\bar z}_{r\mu}(0)$, or $\det M(\psi_r) =

2059: \exp(i\psi_r) \det M(0)$.  Therefore when the given $\psi_r$ goes

2060: from 0 to $4\pi$, the other $\psi_r$'s being held fixed (this is the

2061: first basis contour), $\det M$ circles the origin twice and we have

2062: $\mu=4$.  Along the flow of one of the $J_{rz}$'s, however, we have

2063: ${\bar z}_{r1}(\phi_r) = \exp(i\phi_r/2) {\bar z}_{r1}(0)$ and ${\bar

2064: z}_{r2}(\phi_r) = \exp(-i\phi_r/2) {\bar z}_{r2}(0)$, or $\det

2065: M(\phi_r) = \det M(0)$.  Therefore along the second basis contour the

2066: $I_r$-flow takes $\det M$ once around the origin (elapsed parameter

2067: $\psi_r=2\pi$), while the $J_{rz}$-flow does nothing.  Therefore the

2068: Maslov index of the second basis contour is $\mu=2$.  There are easier

2069: ways to find the Maslov indices of harmonic oscillators, but this

2070: calculation is useful practice for the case of the Wigner manifold

2071: that we take up momentarily.

2072: 

2073: Now we apply the quantization conditions.  For the first basis

2074: contour the total phase is $4\pi j_r -4(\pi/2)$, which we set to

2075: $2n_r\pi$ where $n_r$ is an integer.  Thus the quantized tori must

2076: satisfy $j_r = (n_r+1)/2$.  The allowed values of $n_r$ are determined

2077: by the fact $n_r < -1$ is impossible in view of the fact that $I_r$ is

2078: nonnegative definite, and $n_r=-1$ corresponds to a torus of less than

2079: full dimensionality (six), so the wave function (\ref{scpsi}) is not

2080: meaningful.  Thus we must have $n_r=0,1,\ldots$.  The $j_r$ in these

2081: formulas, and throughout all of the classical analysis from

2082: Sec.~\ref{cmSchwinger} up to this point, has referred to a contour

2083: value for the function $I_r$; the only difference now is that we are

2084: restricting the value of $j_r$ in order that the torus be quantized.

2085: This $j_r$, however, is not value of the principal symbol of the

2086: operator ${\hat I}_r$ (see Eq.~(\ref{IJsymbols})), so the

2087: Bohr-Sommerfeld or EBK quantization rule gives the semiclassical

2088: eigenvalue of ${\hat I}_r$, call it $j_r^{\rm qu}$, as

2089: \begin{equation}

2090: 	j_r^{\rm qu} = j_r-\frac{1}{2} = \frac{n_r}{2}.

2091: 	\label{jrquant}

2092: \end{equation}

2093: The semiclassical eigenvalues of ${\hat I}_r$ are nonnegative integers

2094: or half-integers, the exact answer (not surprising in view of the fact

2095: that semiclassical quantization of quadratic Hamiltonians is exact).

2096: If we use the operator identity (\ref{JIidentity}) to find the

2097: eigenvalues of operators ${\bf J}_r^2$, these are also exact.

2098: 

2099: Equation~(\ref{jrquant}) shows that the classical level set

2100: corresponding to quantum number $j_r^{\rm qu}$ is $j_r = j_r^{\rm qu}

2101: + 1/2$.  The extra $1/2$ in this formula has caused some discussion in

2102: the past and merits a little more now.  Ponzano and Regge (1968) used

2103: intuition and numerical evidence to argue for the presence of the

2104: $1/2$.  Miller, without knowing about Ponzano and Regge, also included

2105: the $1/2$, referring to the ``usual'' semiclassical replacement for

2106: angular momenta.  Presumably he was referring to the similar

2107: replacement that occurs in the treatment of radial wave equations (the

2108: Langer modification, see Berry and Mount (1972), Morehead (1995)).  It

2109: is not obvious to us what the Langer modification has to do with the

2110: $1/2$ that occurs in the present context, nor are we aware of any

2111: general rules about when in the asymptotics of angular momentum theory

2112: it is correct to replace a classical $j$ by $j+1/2$ (instead of

2113: $[j(j+1)]^{1/2}$ or something else).  Schulten and Gordon (1975b) and

2114: Reinsch and Morehead (1999) obtain the $1/2$ as a part of their proper

2115: semiclassical analyses.  Biedenharn and Louck (1981b) also speculate

2116: on the significance of the $1/2$.  Roberts' (1999) derivation of the

2117: asymptotics of the $6j$-symbols does not produce the $1/2$.  He argues

2118: that in the asymptotic limit there is no confusion about whether a

2119: given point lies in the classically allowed or forbidden region,

2120: whether the $1/2$ is included or not.  The omission of the $1/2$ does,

2121: however, cause an error in the phase function that is of order unity,

2122: so the oscillations are not even approximately represented (this point

2123: was also made by Biedenharn and Louck, 1981b).  We suspect that a

2124: modification of Roberts' method would produce the $1/2$'s.  According

2125: to Girelli and Livine (2005), different choices for the semiclassical

2126: replacement for the quantum number $j$ have been made by various

2127: researchers in the field of quantum gravity.  Here we have shown that

2128: the extra $1/2$ is a necessary consequence of standard semiclassical

2129: theory.  We remark in addition that with the inclusion of the $1/2$,

2130: the quantized spheres in angular momentum space are those with an area

2131: of $(2j^{\rm qu}+1)2\pi$, that is, they contain a number of Planck

2132: cells exactly equal to the dimension of the irrep, obviously a form of

2133: geometric quantization.  In particular, the $s$-wave $j^{\rm qu}=0$ is

2134: represented by a sphere of nonzero radius, a case for which the

2135: replacement $j_r = j_r^{\rm qu} + 1/2$ is declared by Biedenharn and

2136: Louck (1981b) to be ``clearly invalid.''

2137: 

2138: For the second basis contour on the $jm$-torus, the quantization

2139: condition is $2\pi(j_r + m_r) - 2(\pi/2) = 2\pi n'_r$, where $n'_r$ is

2140: an integer.  With (\ref{jrquant}), this implies $m_r = -j_r^{\rm

2141: qu} + n'_r$.  Combined with the classical restriction

2142: (\ref{jmranges}), this gives the usual range on magnetic quantum

2143: numbers.  Again the semiclassical quantization is exact.  In the case

2144: of $J_{rz}$, the eigenvalue of the operator is equal to the classical

2145: contour value $m_r$ on the quantized torus (without any correction such as

2146: we see in (\ref{jrquant})).

2147: 

2148: \subsection{Quantizing the Wigner manifold}

2149: 

2150: We begin the quantization of the Wigner manifold by guessing the basis

2151: contours of the fundamental group by inspection of Fig.\ref{Wman}.

2152: Taking the base (initial) point of the loops to be point $a$ in the

2153: figure, we get three independent basis contours (call them $C_1$,

2154: $C_2$, $C_3$) by going around the 3-torus $T_A$ in the three different

2155: directions.  A fourth contour (call it $C_4$) is created by following

2156: an $SU(2)$ rotation about some axis by angle $2\pi$, taking us along

2157: the path $aba'$, which puts us half way around the torus $T_A$ from

2158: the starting point, and then by applying half rotations along each of

2159: the three directions on the torus, taking us down along $T_A$ in the

2160: diagram back to the starting point $a$.  These four contours are not

2161: independent, since

2162: \begin{equation}

2163: 	2C_4 = C_1 + C_2 + C_3, 

2164: 	\label{homotopyrelation}

2165: \end{equation}

2166: where addition of contours means concatenation, but they are

2167: convenient for studying the quanitzation conditions since a minimal

2168: set of three contours (not $(C_1,C_2,C_3)$, but for example

2169: $(C_1,C_2,C_4)$) is less symmetrical.  The fundamental group is

2170: ${\mathbb Z}^3$.

2171: 

2172: We may show the correctness of this guess by a topological argument.

2173: First, the Wigner manifold is the orbit of a $SU(2) \times T^3$ group

2174: action on the large phase space.  Let $(\psi_1,\psi_2,\psi_3)$ be

2175: coordinates on $T^3$, where $0\le \psi_r \le 4\pi$.  The action of

2176: $(u,(\psi_1, \psi_2, \psi_3)) \in SU(2) \times T^3$ on the large phase

2177: space is to multiply each spinor by $u$ and then by

2178: $\exp(-i\psi_r/2)$.  The isotropy subgroup of this action consists of the

2179: identity $(1,(0,0,0))$ and the element $(-1,(2\pi,2\pi,2\pi))$, a

2180: normal subgroup isomorphic to ${\mathbb Z}_2$.  Thus the Wigner

2181: manifold is diffeomorphic to $(SU(2) \times T^3)/{\mathbb Z}_2$,

2182: itself a group manifold, of which the original group $SU(2) \times

2183: T^3$ is a double cover.  This cover is topologically simple since

2184: $SU(2)$ is simply connected (we can go to the universal cover if we

2185: wish by replacing $T^3$ by ${\mathbb R}^3$).  Therefore the homotopy

2186: classes on the Wigner manifold are in one-to-one correspondence with

2187: classes of topologically inequivalent curves that go from the identity

2188: in $SU(2) \times T^3$ to one of the elements of the isotropy

2189: subgroup.  Since $SU(2)$ is simply connected, such paths are

2190: characterized by choice of the end point (the element of the isotropy

2191: subgroup), and the winding numbers around the torus $T^3$.  They are

2192: thus all ``linear combinations'' of the four contours $C_k, k=1,

2193: \ldots, 4$ defined above.  Because of the relation

2194: (\ref{homotopyrelation}), however, the fundamental group is not ${\mathbb

2195: Z}^4$, but only ${\mathbb Z}^3$.

2196: 

2197: It is easy to compute the action along these contours.  Contours

2198: $C_r$, $r=1,2,3$ follow the flows of the $I_r$'s, and the action is

2199: the same as on the $jm$-torus, namely, $S_r = 4\pi j_r$.

2200: Along contour $C_4$ the spinor rotation makes no contribution to the

2201: action while the half rotation around the torus in all three angles

2202: gives the action

2203: \begin{equation}

2204: 	S_4 = 2\pi \sum_r j_r.

2205: 	\label{2Zaction}

2206: \end{equation}

2207: 

2208: As for the Maslov indices, we compute the complex matrix of Poisson

2209: brackets whose rows are indexed by the functions

2210: $(I_1,I_2,I_3,J_x,J_y,J_z)$, and whose columns are indexed by $({\bar

2211: z}_{11}, {\bar z}_{12}, {\bar z}_{21}, {\bar z}_{22}, {\bar z}_{31},

2212: {\bar z}_{32})$.  To within a multiplicative constant that we drop, the

2213: determinant is

2214: \begin{eqnarray}

2215: 	&\left| \begin{array}{cccccc}

2216: 	{\bar z}_{11} & {\bar z}_{12} & 0 & 0 & 0 & 0 \\

2217: 	0 & 0 & {\bar z}_{21} & {\bar z}_{22} & 0 & 0 \\

2218: 	0 & 0 & 0 & 0 & {\bar z}_{31} & {\bar z}_{32} \\

2219: 	{\bar z}_{12} & {\bar z}_{11} & {\bar z}_{22}  &

2220: 	{\bar z}_{21} & {\bar z}_{32} & {\bar z}_{31} \\

2221: 	-{\bar z}_{12} & {\bar z}_{11} & -{\bar z}_{22} &

2222: 	{\bar z}_{21} & -{\bar z}_{32} & {\bar z}_{31} \\

2223: 	{\bar z}_{11} & -{\bar z}_{12} & {\bar z}_{21}  &

2224: 	-{\bar z}_{22} & {\bar z}_{31} & -{\bar z}_{32}

2225: 	\end{array}\right| \nonumber \\

2226: 	&={\rm const.} \times

2227: 	\left| \begin{array}{cc}

2228: 	{\bar z}_{11} & {\bar z}_{12} \\

2229: 	{\bar z}_{21} & {\bar z}_{22}

2230: 	\end{array}\right|

2231: 	\left| \begin{array}{cc}

2232: 	{\bar z}_{21} & {\bar z}_{22} \\

2233: 	{\bar z}_{31} & {\bar z}_{32}

2234: 	\end{array}\right|

2235: 	\left| \begin{array}{cc}

2236: 	{\bar z}_{31} & {\bar z}_{32} \\

2237: 	{\bar z}_{11} & {\bar z}_{12}

2238: 	\end{array}\right|.

2239: 	\label{2ZMi}

2240: \end{eqnarray}

2241: The final product of determinants is interesting, since these are the

2242: $SU(2)$ invariants that Schwinger (Biedenharn and van Dam, 1965) used

2243: to construct the rotationally invariant state $\vert j_1 j_2 j_3 {\bf

2244: 0} \rangle$ (with ${\bar z}_{r\mu}$ replaced by $a^\dagger_{r\mu}$).

2245: Bargmann (1962) and Roberts (1999) make use of the same invariants.

2246: For our purposes we need the winding number of the loop traced out in

2247: the complex plane by the product of the three determinants as we

2248: follow the four basis contours on the Wigner manifold.  

2249: 

2250: Proceeding as we did on the $jm$-torus, we find

2251: that the Maslov indices along the contours $C_r, r=1,2,3$ are 4.

2252: For example, along the $I_1$-flow ${\bar z}_{11}$ and ${\bar z}_{12}$

2253: get multiplied by $\exp(i\psi_1/2)$, which causes two of the three

2254: determinants to be multiplied by the same factor, so the product gets

2255: multiplied by $\exp(i\psi_1)$, which has winding number 2 and hence

2256: Maslov index 4 when $\psi_1$ goes from 0 to $4\pi$.  This is the same

2257: answer we found along the $I_r$-flows on the $jm$-torus; this was not

2258: exactly a foregone conclusion, even though the contours are the same,

2259: because the tangent planes are different.  On the other hand, in both

2260: cases we find the result (\ref{jrquant}) for the eigenvalues $j_r^{\rm

2261: qu}$ of the operators ${\hat I}_r$, which of course must not depend on

2262: how we compute them.  As for contour $C_4$ the first leg, a rotation

2263: by angle $2\pi$ about some axis, leaves all three $2 \times 2$

2264: determinants in (\ref{2ZMi}) invariant, so the big determinant in

2265: the complex plane does not move.  As for the second leg, since each

2266: $\psi_r$ only goes from 0 to $2\pi$ we get a winding number of 1 along

2267: each $I_r$-flow, but since there are three of them the total winding

2268: number is 3 and Maslov index is 6.

2269: 

2270: Combining this result with (\ref{2Zaction}), we obtain the

2271: Bohr-Sommerfeld quantization condition for contour $C_4$

2272: in the form

2273: \begin{equation}

2274: 	2\pi \sum_r j_r - 6\frac{\pi}{2} = 2\pi \times {\rm integer},

2275: 	\label{2Zquant}

2276: \end{equation}

2277: or, with (\ref{jrquant}), 

2278: \begin{equation}

2279: 	\sum_r j_r^{\rm qu} = {\rm integer}.

2280: 	\label{2zquant1}

2281: \end{equation}

2282: This is precisely the condition that the three quantum angular momenta

2283: must satisfy, in addition to the triangle inequalities, that they may

2284: add up to zero.  It emerges in a semiclassical analysis because the

2285: Wigner manifold is not quantized otherwise.

2286: 

2287: In conclusion, the Bohr-Sommerfeld quantization conditions applied to

2288: the $jm$-torus and the Wigner manifold give us a complete (and exact)

2289: accounting of all the quantum numbers and the restrictions on them that

2290: appear in the coupling of three angular momenta with a resultant of

2291: zero.  It also allows us to identify the classical manifold (that is,

2292: its contour values) with a given set of quantum numbers.  

2293: 

2294: \section{The amplitude determinant}

2295: \label{ampdet}

2296: 

2297: The generic semiclassical eigenfunction of a complete set of commuting

2298: observables is given by (\ref{scpsi}), with the amplitude

2299: determinant expressed in terms of Poisson brackets by

2300: (\ref{ampldet}).  These formulas apply in particular to the state

2301: $\vert j_1 j_2 j_3 m_1 m_2 m_3\rangle$ on the left of the matrix

2302: element (\ref{3jdef}), which is supported by the $jm$-torus in the

2303: large phase space.  The state on the right, $\vert j_1 j_2 j_3 {\bf

2304: 0}\rangle$, however, which is supported by the Wigner manifold, is an

2305: eigenfunction of observables that do not commute.  Therefore we must

2306: rethink the derivation of Eqs.~(\ref{scpsi}) and (\ref{ampldet}) to

2307: see what changes in this case.  In particular, we must see what

2308: happens to the Poisson bracket expression for the amplitude

2309: determinant, which is the solution of the simultaneous amplitude

2310: transport equations for the collection of observables.  As it turns

2311: out, nothing changes, the wave function is still given by

2312: Eqs.~(\ref{scpsi}) and (\ref{ampldet}), with the (now noncommuting)

2313: observables used in the amplitude determinant.  In addition, there is

2314: a certain understanding about how the volume $V$ in (\ref{scpsi})

2315: is computed, since the angles $\alpha$ conjugate to the $A$'s are no

2316: longer meaningful.

2317: 

2318: Once this is done, we must evaluate the scalar product of the two wave

2319: functions by stationary phase.  If both states were eigenstates of

2320: complete sets of commuting observables, then the answer would be

2321: (\ref{abmatrixelement1}) with amplitude determinant (\ref{PBdet}), but

2322: again we must rethink the derivation of this result since the

2323: observables for one of the wave functions do not commute.  Again, the

2324: answer turns out to be given by formulas (\ref{abmatrixelement1}) and

2325: (\ref{PBdet}) of Sec.~\ref{scwfis}, with a proper understanding of the

2326: meanings of the volume factors.

2327: 

2328: Having established these facts, we can then proceed to the (easy)

2329: calculation of the amplitude determinant for the $3j$-symbol in terms

2330: of Poisson brackets, and finally put the remaining pieces together to

2331: get the leading asymptotic form of the $3j$-symbols. 

2332: 

2333: \subsection{Amplitude determinant for noncommuting observables}

2334: 

2335: We begin showing that Eqs.~(\ref{scpsi}) and (\ref{ampldet}) are valid

2336: for the state $\vert j_1 j_2 j_3 {\bf 0} \rangle$, with a proper

2337: definition of the volume factors.  The classical functions defining

2338: the Wigner manifold are $(I_1, I_2, I_3, J_x, J_y, J_z)$.  Let us

2339: refer to these collectively as $A_k$, $k=1, \ldots, 6$, let us write

2340: $x^i$, $i=1, \ldots, 6$ for the configuration space coordinates

2341: instead of the notation $(x_{11}, \ldots, x_{32})$ used above, and let

2342: us adopt the summation convention.  The functions $A_k$ form a Lie

2343: algebra, that is, $\{A_k, A_l \} = c^m_{kl} \, A_m$, where $c^m_{kl}$

2344: are the structure constants.  The Wigner manifold is a compact group

2345: manifold with this Lie algebra, on which the Haar measure is both

2346: left- and right-invariant.  This density is also invariant under the

2347: flows generated by the right-invariant vector fields, which in our

2348: case are the Hamiltonian flows of the functions $A_i$.  The projection

2349: of this density onto configuration space is the density that provides

2350: the solution of the simultaneous amplitude transport equations for the

2351: functions $A_i$.  These are the basic geometrical facts, which we now

2352: present more explicitly in coordinate language.

2353: 

2354: The amplitude transport equations for the functions $A_k$, $k=1,

2355: \ldots, 6$ are

2356: \begin{equation}

2357: 	\frac{\partial}{\partial x^i}

2358: 	\left[ \Omega(x) 

2359: 	\frac{\partial A_k}{\partial p_i}\right]=0,

2360: 	\label{ATeqn}

2361: \end{equation}

2362: where $p_i$ are the momenta conjugate to $x^i$.  These are six

2363: simultaneous equations that must be solved for the density $\Omega(x)$

2364: on configuration space.  Notice that $\partial A_k/\partial p_i =

2365: \{x^i,A_k\} = {\dot x}^i_{(k)}$, the latter being notation we shall

2366: use for the velocity in configuration space along the Hamiltonian flow

2367: generated by $A_k$.  The amplitude transport equation is a continuity

2368: equation, which is form-invariant under general coordinate

2369: transformations.

2370: 

2371: Let us pick one of the branches of the inverse projection from

2372: configuration space onto the Lagrangian (Wigner) manifold.  We shall

2373: suppress the branch index in the following.  Let $u^i$, $i=1,\ldots,6$

2374: be an arbitrary set of local coordinates on the Wigner manifold, which

2375: we extend in a smooth but arbitrary manner into some small

2376: neighborhood of the Wigner manifold, so that partial derivatives of

2377: the $u^i$ with respect to all phase space coordinates are defined.

2378: Assuming we are not at a caustic, the transformation from $x^i$ to

2379: $u^i$ is locally one-to-one, and the Jacobian $\partial u^i/\partial

2380: x^j$ is nonsingular.  Under the inverse projection or coordinate

2381: transformation $x\to u$, the flow velocity transforms according to

2382: \begin{equation}

2383: 	{\dot u}^i_{(k)} = 

2384: 	\frac{\partial u^i}{\partial x^j} \,

2385: 	{\dot x}^j_{(k)} = \{ u^i, A_k\} = X^i_{(k)},

2386: 	\label{velxfm}

2387: \end{equation}

2388: which defines the quantities $X^i_{(k)}$.  As a matrix, $X^i_{(k)}$ is

2389: nonsingular because the flow vectors are linearly independent on the

2390: Wigner manifold.  As for the density, it transforms according to

2391: \begin{equation}

2392: 	\sigma(u) = \Omega(x) \left| \det

2393: 	\frac{\partial x^l}{\partial u^m}\right|,

2394: 	\label{densityxfm}

2395: \end{equation}

2396: so that the amplitude transport equations, lifted to the the Wigner

2397: manifold, become

2398: \begin{equation}

2399: 	\frac{\partial}{\partial u^i}

2400: 	[ \sigma(x) X^i_{(k)}]=0.

2401: 	\label{ATWm}

2402: \end{equation}

2403: Now define $\Lambda^{(k)}_j$ as the matrix inverse to $X^i_{(k)}$, 

2404: \begin{equation}

2405: 	\Lambda^{(k)}_i X^i_{(l)} = \delta^k_l.

2406: 	\label{Lambdadef}

2407: \end{equation}

2408: As we will prove momentarily, the solution of Eqs.~(\ref{ATWm}) is 

2409: \begin{equation}

2410: 	\sigma(u)= |\det \Lambda^{(k)}_j|,

2411: 	\label{sigmasoln}

2412: \end{equation}

2413: which, by (\ref{densityxfm}), gives us the solution of

2414: (\ref{ATeqn}),

2415: \begin{eqnarray}

2416: 	\Omega(x) &= \left| {\mathop{\rm det}\nolimits}_{kl} 

2417: 	\left(

2418: 	\Lambda^{(k)}_i 

2419: 	\frac{\partial u^i}{\partial x^l}\right)\right| =

2420: 	\left| {\mathop{\rm det}\nolimits}_{kl} 

2421: 	\left(

2422: 	X^i_{(k)} 

2423: 	\frac{\partial x^l}{\partial u^i}\right)\right|^{-1}

2424: 	\nonumber \\

2425: 	&= \left| {\mathop{\rm det}\nolimits}_{kl}\,

2426: 	\left({\dot u}^i_{(k)} 

2427: 	\frac{\partial x^l}{\partial u^i}\right)\right|^{-1}

2428: 	=\left| {\mathop{\rm det}\nolimits}_{kl}\,

2429: 	{\dot x}^l_{(k)} \right|=

2430: 	\left| {\mathop{\rm det}\nolimits}_{kl}\, 

2431: 	\{x^l,A_k \} \right|^{-1}.

2432: 	\label{Omegasoln}

2433: \end{eqnarray}

2434: In carrying out these manipulations it is important to note that

2435: $\partial u^i/\partial x^j$ is taken at constant $A_k$, not $p_k$.

2436: Thus the amplitude determinant for the wave function $\psi(x)$

2437: associated with the Wigner manifold has the same Poisson bracket form

2438: shown in (\ref{ampldet}), that is, in spite of the fact that the

2439: $A_i$ do not commute.  

2440: 

2441: The essential differential geometry of these manipulations is that

2442: $X_{(k)} = X^i_{(k)} \partial/\partial u^i$ are the Hamiltonian vector

2443: fields on the Wigner manifold associated with functions $A_k$,

2444: $\lambda^{(k)} = \Lambda^{(k)}_i \, du^i$ are the dual forms,

2445: $\lambda^{(k)} X_{(l)} = \delta^k_l$, and $\sigma=\lambda^1 \wedge

2446: \ldots \wedge \lambda^6 = \sigma(u) \,du^1 \wedge \ldots \wedge du^6$ 

2447: is the Haar measure.  The condition (\ref{ATWm}) is equivalent to

2448: $L_{X_{(k)}} \sigma = 0$.

2449: 

2450: To prove (\ref{sigmasoln}) in coordinates we substitute it into

2451: (\ref{ATWm}) and expand out the derivative, obtaining an

2452: expression proportional to

2453: \begin{equation}

2454: 	X^i_{(k),i} - X^i_{(k)} X^l_{(m),i} \Lambda^{(m)}_l

2455: 	\label{intqty}

2456: \end{equation}

2457: using commas for derivatives.  Then we use the Lie bracket of the

2458: vector fields $X_{(k)}$,

2459: \begin{equation}

2460: 	[X_{(k)}, X_{(m)}]^l = X^i_{(k)} X^l_{(m),i} -

2461: 	X^i_{(m)} X^l_{(k),i} = -c^n_{km} X^l_{(n)},

2462: 	\label{Liebracket}

2463: \end{equation}

2464: where $c^n_{km}$ are the structure constants.  Here we use the

2465: identity expressing the Lie bracket of Hamiltonian vector fields for

2466: two functions in terms of the Hamiltonian vector field of their

2467: Poisson bracket, $[X_H, X_K] = -X_{\{H,K\}}$ (Arnold, 1989).  Thus

2468: (\ref{intqty}) becomes simply $c^m_{km}$, which vanishes since for the

2469: group in question the structure constants are completely

2470: antisymmetric.

2471: 

2472: Finally to normalize the semiclassical eigenfunction supported by the

2473: Wigner manifold we use the stationary phase approximation to compute

2474: the integral

2475: \begin{equation}

2476: 	\int dx \, \left| \sum_{\rm br} \Omega(x)^{1/2}

2477: 	\exp[iS(x)-i\mu\pi/2]\right|^2,

2478: 	\label{Wignorm}

2479: \end{equation}

2480: where the sum is over branches and the branch index is suppressed.

2481: Cross terms do not contribute, and when the integral is lifted to the

2482: Wigner manifold it just gives the volume of that manifold with respect

2483: to the Haar measure,

2484: \begin{equation}	

2485: 	\sum_{\rm br} \int \frac{dx}

2486: 	{|{\mathop{\rm det}\nolimits} \{x^i, A_j \}|}

2487: 	= \int du\,\sigma(u) = V_W,

2488: 	\label{Wignorm1}

2489: \end{equation}

2490: where $V_W$ is given by (\ref{Wignervolume}).

2491: 

2492: \subsection{Matrix elements for noncommuting observables}

2493: 

2494: Now we write the $3j$ matrix element (\ref{3jdef}) as $\langle b \vert

2495: a \rangle$, where $A=(I_1, I_2, I_3, J_x, J_y, J_z)$ and $B=(I_1, I_2, I_3,

2496: J_{1z}, J_{2z}, J_{3z})$.  Actually this is not the most convenient

2497: form, since $J_z$ in the $A$-list is a function of the $J_{rz}$ in the

2498: $B$-list.  We fix this by performing a canonical transformation

2499: $(\phi_1, \phi_2, \phi_3, J_{1z}, J_{2z}, J_{3z}) \to ({\tilde

2500: \phi}_1, {\tilde \phi}_2, {\tilde \phi}_3, {\tilde J}_{1z}, {\tilde

2501: J}_{2z}, J_z)$ on the functions in the $B$-list, generated by

2502: \begin{equation}

2503: 	\fl F_2(\phi_1,\phi_2,\phi_3, {\tilde J}_{1z}, {\tilde J}_{2z},

2504: 	J_z) = \phi_1 {\tilde J}_{1z} + \phi_2 {\tilde J}_{2z}

2505: 	+ \phi_3(J_z - {\tilde J}_{1z} - {\tilde J}_{2z}).

2506: 	\label{JzCT} 

2507: \end{equation}

2508: This gives ${\tilde J}_{1z} = J_{1z}$, ${\tilde J}_{2z} = J_{2z}$,

2509: $J_z = J_{1z} + J_{2z} + J_{3z}$, and ${\tilde \phi}_1 =

2510: \phi_1 - \phi_3$, ${\tilde \phi}_2 = \phi_2-\phi_3$, ${\tilde \phi}_3

2511: = \phi_3$.  The linear transformation in the angles has unit

2512: determinant, so the volume of the $jm$-torus is still given by

2513: (\ref{Vjm}).  Dropping the tildes, the $B$-list is now

2514: $(I_1,I_2,I_3, J_{1z}, J_{2z}, J_z)$, which has four functions in

2515: common with the $A$-list.

2516: 

2517: Now the integral we must evaluate is

2518: \begin{eqnarray}

2519: 	\langle b \vert a \rangle &=

2520: 	\frac{1}{\sqrt{V_A V_B}} \int dx

2521: 	\frac{1}{|{\mathop {\rm det}} \{ x^i, A_j \}|^{1/2}}

2522: 	\frac{1}{|{\mathop {\rm det}} \{ x^i, B_j \}|^{1/2}}

2523: 	\nonumber \\

2524: 	&\qquad \times

2525: 	\sum_{\rm br}

2526: 	\exp\{i[S_A(x)-S_B(x) - \mu\pi/2]\},

2527: 	\label{abintegral}

2528: \end{eqnarray}

2529: where the sum is over all branches of the projections of the two

2530: manifolds, and where $\mu$ just stands for whatever Maslov index

2531: appears in a given term (different $\mu$'s are not necessarily

2532: equal).  An integral like this was evaluated by Littlejohn (1990),

2533: using the angles conjugate to the $A$'s and $B$'s, but those do not

2534: all exist in the present circumstances and we must evaluate the

2535: integral in a different way.  

2536: 

2537: Let us write $A=(C,D)$ and $B=(C,E)$, where $C=(I_1, I_2, I_3, J_z)$

2538: are the four observables in common in the $A$- and $B$-lists, and

2539: where $D=(J_x, J_y)$ and $E=(J_{1z}, J_{2z})$ are the two pairs of

2540: observables that are distinct.  The stationary phase set of the

2541: integral (\ref{abintegral}) consists of points $x$ where

2542: $\partial(S_A-S_B)/\partial x^i=0$, that is, it is the projection onto

2543: configuration space of the intersection of the $A$-manifold and the

2544: $B$-manifold.  That intersection, which we denote by $I$, was

2545: studied in Sec.~\ref{intersections} (it is a 4-torus).  It is the

2546: simultaneous level set of all of the $A$'s and $B$'s, and at the same

2547: time the orbit of the commuting Hamiltonian flows generated by the

2548: $C$'s.  Its projection onto configuration space is a 4-dimensional

2549: region.

2550: 

2551: We introduce a local coordinate transformation in configuration space

2552: $x \to (y,z)$ where the four $y$'s are coordinates along the stationary

2553: phase set and the two $z$'s are transverse to it.  We let the stationary

2554: phase set itself be specified by $z=0$.  We let $(u,v)$ be the momenta

2555: conjugate to $(y,z)$. Then the two amplitude determinants in

2556: (\ref{abintegral}) may be combined with the Jacobian of

2557: the coordinate transformation to result in the square root of the

2558: product of two determinants, one of which is

2559: \begin{equation}

2560: 	\det \frac{\partial(y,z)}{\partial x}

2561: 	\det \{ x, A \} =

2562: 	\det \left(\begin{array}{cc}

2563: 	\{ y, C \} & \{ y, D \} \\

2564: 	\{ z, C \} & \{ z, D \}

2565: 	\end{array}

2566: 	\right),

2567: 	\label{yzCDPB}

2568: \end{equation}

2569: and the other of which is the same but with the substitutions $A\to

2570: B$, $D \to E$.  But since the $C$'s generate flows along $I$, we have

2571: $\{z^i,C_j\}=0$, and the lower left block of the two matrices

2572: vanishes.  Thus, the product of the two determinants becomes

2573: \begin{equation}

2574: 	[\det \{ y, C \}]^2 \det \{ z, D\} \det \{ z, E\},

2575: 	\label{detprod}

2576: \end{equation}

2577: the square root of which appears in the denominator of the integrand.

2578: Evaluating the final two Poisson brackets in the $(y,z;u,v)$ canonical

2579: coordinates, we have

2580: \begin{equation}

2581: 	\{z^i, D_j \} = \frac{\partial D_j}{\partial v_i},

2582: 	\qquad

2583: 	\{z^i, E_j \} = \frac{\partial E_j}{\partial v_i}.

2584: 	\label{zDzEPBs}

2585: \end{equation}

2586: 

2587: We perform the $z$-integration by stationary phase, expanding $S_A$

2588: and $S_B$, regarded as functions of $(y,z)$, to second order in $z$

2589: for a fixed value of $y$, and simply evaluating the amplitude at $z=0$

2590: (that is, on $I$).  To within a phase, the $z$-integration gives

2591: \begin{equation}

2592: 	2\pi

2593: 	\left| \det \left(

2594: 	\frac{\partial^2 S_A}{\partial z\partial z}

2595: 	-\frac{\partial^2 S_B}{\partial z\partial z}\right) 

2596: 	\right|^{-1/2}.

2597: 	\label{zintegral}

2598: \end{equation}

2599: The determinant in this result must be multiplied by the determinants

2600: of the matrices (\ref{zDzEPBs}) to get the overall determinant in the

2601: denominator after the $z$-integration.  The product of these three

2602: determinants is the determinant of the matrix

2603: \begin{equation}

2604: 	\frac{\partial D}{\partial v}\left[

2605: 	\left(\frac{\partial v}{\partial z}\right)_{yCD} -

2606: 	\left(\frac{\partial v}{\partial z}\right)_{yCE}\right]

2607: 	\left(\frac{\partial E}{\partial v}\right)^T,

2608: 	\label{3matrixprod}

2609: \end{equation}

2610: where it is understood that a partial derivative stands for a matrix

2611: whose row index is given by the numerator and column index by the

2612: denominator, unless the matrix transpose or inverse is indicated, in

2613: which case the rule is reversed.  Also, if a partial derivative is

2614: shown without subscripts, then it is assumed that it is computed in

2615: the canonical coordinates $(y,z;u,v)$, and otherwise the variables to

2616: be held fixed are explicitly indicated.  In the two middle matrices in

2617: (\ref{3matrixprod}), the variables held fixed amount to

2618: differentiating $v$ with respect to $z$ along the $A$- and

2619: $B$-manifolds, respectively, since $\partial S_A/\partial z = v(x,A)$

2620: and $\partial S_B/\partial z=v(x,B)$.   Notice that these two matrices

2621: are symmetric.

2622: 	

2623: Now we express the two matrices in the middle of

2624: (\ref{3matrixprod}) purely in terms of partial derivatives

2625: computed in the canonical $(y,z;u,v)$ coordinates.  We do this by

2626: writing out the Jacobian matrix $\partial(y,z;C,D)/ \partial(y,z;u,v)$

2627: and the inverse Jacobian $\partial(y,z;u,v)/\partial(y,z;C,D)$,

2628: multiplying the two together to obtain a series of identities

2629: connecting the forward and inverse Jacobian blocks, and then solving

2630: for the inverse Jacobian blocks in terms of the forward ones.  We note

2631: that the block $\partial C/\partial v$ of the forward Jacobian

2632: vanishes, since it is $\{z,C\}$.  Thus we find

2633: \begin{equation}

2634: 	\fl\eqalign{

2635: 	\left(\frac{\partial v}{\partial z}\right)_{yCD} &=

2636: 	\left(\frac{\partial D}{\partial v}\right)^{-1} \left[

2637: 	\frac{\partial D}{\partial u}

2638: 	\left(\frac{\partial C}{\partial u}\right)^{-1}

2639: 	\frac{\partial C}{\partial z}-

2640: 	\frac{\partial D}{\partial z}\right], \cr

2641: 	\left(\frac{\partial v}{\partial z}\right)_{yCE} &=

2642: 	\left[\left(\frac{\partial C}{\partial z}\right)^T

2643: 	\left(\frac{\partial C}{\partial u}\right)^{-1T}

2644: 	\left(\frac{\partial E}{\partial u}\right)^T -

2645: 	\left(\frac{\partial E}{\partial z}\right)^T\right]

2646: 	\left(\frac{\partial E}{\partial v}\right)^{-1T}. \cr}

2647: 	\label{vzidents}

2648: \end{equation}

2649: Upon substituting these into (\ref{3matrixprod}), that matrix

2650: becomes

2651: \begin{eqnarray}

2652: 	&\frac{\partial D}{\partial v}

2653: 	\left(\frac{\partial E}{\partial z}\right)^T -

2654: 	\frac{\partial D}{\partial z}

2655: 	\left(\frac{\partial E}{\partial v}\right)^T 

2656: 	+\frac{\partial D}{\partial u}

2657: 	\left(\frac{\partial C}{\partial u}\right)^{-1}

2658: 	\frac{\partial C}{\partial z}

2659: 	\left(\frac{\partial E}{\partial v}\right)^{T}

2660: 	\nonumber \\

2661: 	&-\frac{\partial D}{\partial v}

2662: 	\left(\frac{\partial C}{\partial z}\right)^T

2663: 	\left(\frac{\partial C}{\partial u}\right)^{-1T}

2664: 	\left(\frac{\partial E}{\partial u}\right)^T,

2665: 	\label{firsthalfDEPB}

2666: \end{eqnarray}

2667: where the first two terms are the beginning of the Poisson bracket

2668: $\{E,D\}$.  As for the last two terms, we write out the vanishing

2669: Poisson brackets $\{C,E\}$ and $\{C,D\}$ in the $(y,z;u,v)$

2670: coordinates, making use of $\partial C/\partial v=0$, to obtain

2671: \begin{equation}

2672: 	\eqalign{

2673: 	\frac{\partial C}{\partial z}

2674: 	\left(\frac{\partial E}{\partial v}\right)^T

2675: 	&= \frac{\partial C}{\partial u}

2676: 	\left(\frac{\partial E}{\partial y}\right)^T -

2677: 	\frac{\partial C}{\partial y}

2678: 	\left(\frac{\partial E}{\partial u}\right)^T, \cr

2679: 	\frac{\partial D}{\partial v}

2680: 	\left(\frac{\partial C}{\partial z}\right)^T

2681: 	&= \frac{\partial D}{\partial y}

2682: 	\left(\frac{\partial C}{\partial u}\right)^T -

2683: 	\frac{\partial D}{\partial u}

2684: 	\left(\frac{\partial C}{\partial y}\right)^T. \cr}

2685: 	\label{CEPBident}

2686: \end{equation}

2687: Actually the matrix of Poisson brackets $\{C,D\}$ does not vanish

2688: everywhere in phase space, just on the $A$- (or Wigner) manifold, and

2689: in particular on the intersection $I$ which is where we are evaluating

2690: them.  Now substituting Eqs.~(\ref{CEPBident}) into the last two terms

2691: of (\ref{firsthalfDEPB}), those terms become

2692: \begin{eqnarray}

2693: 	&\frac{\partial D}{\partial u}

2694: 	\left(\frac{\partial E}{\partial y}\right)^T -

2695: 	\frac{\partial D}{\partial y}

2696: 	\left(\frac{\partial E}{\partial u}\right)^T \\

2697: 	&+\frac{\partial D}{\partial u} \left[

2698: 	\left(\frac{\partial C}{\partial y}\right)^T

2699: 	\left(\frac{\partial C}{\partial u}\right)^{-1T} -

2700: 	\left(\frac{\partial C}{\partial u}\right)^{-1}

2701: 	\frac{\partial C}{\partial y}\right]

2702: 	\left(\frac{\partial E}{\partial u}\right)^T,

2703: 	\label{secondhalfDEPB}

2704: \end{eqnarray}

2705: in which the first two terms give us the remainder of the Poisson

2706: bracket $\{E,D\}$.  As for the last major term, the factor in the

2707: square brackets vanishes, as we see by writing out the vanishing

2708: Poisson bracket $\{C,C\}$ in coordinates $(y,z;u,v)$ and using

2709: $\partial C/\partial v=0$.  

2710: 

2711: As a result the integral (\ref{abintegral}) becomes

2712: \begin{equation}

2713: 	\langle b \vert a \rangle =

2714: 	\frac{2\pi}{\sqrt{V_A V_B}}

2715: 	\sum_{\rm br}

2716: 	\int \frac{dy}{|\det\{y,C\}|}

2717: 	\frac{e^{i(S_I - \mu\pi/2)}}

2718: 	{|\det\{E,D\}|^{1/2}},

2719: 	\label{abyintegral}

2720: \end{equation}

2721: where the branch sum runs over all branches of the projection of $I$

2722: onto configuration space as well as the two disconnected components of $I$

2723: (the two 4-tori discussed in Sec.~\ref{intersections}), and where

2724: $S_I$ is the phase on a given connected component of $I$ (this is the

2725: phase $\pm S_{jm}$ computed in Sec.~\ref{actionintegrals}).  We have also

2726: dropped an overall phase, and we are not attempting to compute the

2727: Maslov indices in detail.  The amplitude determinant has been reduced

2728: to a $2\times 2$ matrix of Poisson brackets of the observables in the

2729: $A$- and $B$-lists that differ, exactly as in (\ref{PBdet}).

2730: Calculating this matrix explicitly, we find 

2731: \begin{eqnarray}

2732: 	|\det\{E,D\}| &= \left|\begin{array}{cc}

2733: 	\{ J_{1z}, J_x \} & \{ J_{1z}, J_y \} \\

2734: 	\{ J_{2z}, J_x \} & \{ J_{2z}, J_y \}

2735: 	\end{array}\right| =

2736: 	\left|\begin{array}{cc}

2737: 	J_{y1} & -J_{x1} \\

2738: 	J_{y2} & -J_{x2} 

2739: 	\end{array}\right|

2740: 	\nonumber \\

2741: 	[3pt]

2742: 	&= |J_{x1} J_{y2} - J_{x2} J_{y1}|

2743: 	= | {\bf z} \cdot ({\bf J}_1 \times {\bf J}_2)|

2744: 	= 2 \Delta_z,

2745: 	\label{3jampldet}

2746: \end{eqnarray}

2747: where $\Delta_z$ is the projection of the area of the triangle

2748: $\Delta$ onto the $x$-$y$ plane (see Eq.~(\ref{Deltadef}) and Fig.~2

2749: of Ponzano and Regge (1968)).  This quantity is invariant under

2750: rotations about the $z$-axis, that is, it Poisson commutes with $J_z$.

2751: It also Poisson commutes with the other three variables in the

2752: $C$-list, $(I_1,I_2,I_3)$, and so is constant on the intersection

2753: $I$ and can be taken out of the $y$-integral.  The same applies to the

2754: phase factor, since $S_I$ is also constant on the $I$-manifold.  Then

2755: the $y$-integral can be done, since $|\det\{y,C\}|$ is just the

2756: Jacobian connecting $y$ with the angle variables conjugate to

2757: $C=(I_1,I_2,I_3,J_z)$, denoted above by $(\psi_1,\psi_2,\psi_3,\phi)$.

2758: Thus the $y$-integral just gives the volume $V_I$ of the intersection $I$

2759: with respect to these angles, see (\ref{VIdef}).  In fact, had the

2760: variables $C$ not been commuting, but if they had formed a Lie

2761: algebra, then $V_I$ would be the volume of $I$ with respect to the

2762: Haar measure of the corresponding group.  This circumstance arises,

2763: for example, in a similar treatment of the $6j$-symbol.

2764: 

2765: As a result of these rather lengthy manipulations of amplitude

2766: determinants, we obtain the final, simple result,

2767: \begin{equation}

2768: 	\langle b \vert a \rangle =

2769: 	\frac{2\pi}{\sqrt{V_A V_B}} \sum_{\rm br} V_I

2770: 	\frac{e^{i(S_I-\mu\pi/2)}}{|\det\{E,D\}|^{1/2}},

2771: 	\label{integraldone}

2772: \end{equation} 

2773: where the branches now run over just the two disconnected pieces of

2774: the intersection $I$.  This is a version of

2775: (\ref{abmatrixelement1}), with the right understanding of the

2776: volume measures, generalized to the case at hand in which the

2777: observables do not commute.  The actual calculation of the final amplitude

2778: determinant takes just one line, Eq.~(\ref{3jampldet}).  

2779: 

2780: In fact, for our application the volume $V_I$ and the remaining

2781: amplitude determinant are the same for both branches and can be taken

2782: out of the sum.  The relative Maslov index between the two branches is

2783: 1; we will not belabor this point since the answer is already known.

2784: We simply note that by splitting the Maslov phase $i\pi/2$ between

2785: the two branches and subsitituting $V_A=V_W$, $V_B=V_{jm}$, we obtain

2786: to within an overall phase the result of Ponzano and Regge,

2787: \begin{equation}

2788: 	\left(

2789:   \begin{array}{ccc}

2790:     j_1 & j_2 & j_3 \\

2791:     m_1 & m_2 & m_3 

2792:   \end{array}

2793:   \right) = ({\rm phase}) \times 

2794: 	\frac{\cos(S_{jm}+\pi/4)}{\sqrt{2\pi\Delta_z}}.

2795: 	\label{PRresult}

2796: \end{equation}

2797: 

2798: \section{Conclusions}

2799: \label{conclusions}

2800: 

2801: In many ways the $3j$-symbol is not as interesting as the $6j$-symbol,

2802: of which it is a limiting case.  We intended our work on the

2803: $3j$-symbol as a warm-up exercise, expecting a routine application of

2804: semiclassical methods for integrable systems.  The nongeneric

2805: Lagrangian (Wigner) manifold was a surprise.  Similar nongeneric

2806: Lagrangian manifolds occur also in the semiclassical analysis of the

2807: $6j$- and $9j$-symbols.

2808: 

2809: If all one wants is a derivation of an asymptotic formula, then there

2810: are many ways to proceed.  For example, one can simply take the

2811: expression for the symbol due to Wigner ($3j$) or Racah ($6j$) as a

2812: sum over a single index, and apply standard asymptotic methods

2813: (Stirling's approximation, Poisson sum rule, etc).  But if one wants a

2814: derivation that reveals the geometrical meaning of the classical

2815: objects that emerge (the triangle, the tetrahedron, etc), then an

2816: approach such as ours may be preferable.

2817: 

2818: Our approach is more geometrical than earlier ones, and in that

2819: respect is closer in spirit to the work of Roberts (1999), Freidel and

2820: Louapre (2003) and later authors.  It is likely that at some deeper

2821: level all these methods are the same, although superficially we see

2822: only a little similarity between our work and these others.  

2823: 

2824: One may also desire a method that makes the symmetries of the symbol

2825: manifest.  Our analysis does not do this for the $3j$-symbol, but

2826: those symmetries are not manifest in Wigner's definition of the

2827: $3j$-symbol that we employ as our starting point, either.  To bring

2828: the symmetries out it seems necessary to employ some construction

2829: related to Schwinger's generating functions, which involve lifting the

2830: definitions into higher dimensional spaces.

2831: 

2832: Our method of calculating amplitude determinants in terms of Poisson

2833: brackets may have computational advantages in other applications, as

2834: well.  The method can be remarkably easy to use.  For example, the

2835: $6j$-symbol can be defined as a matrix element,

2836: \begin{equation}

2837: 	\left\{ \begin{array}{ccc}

2838: 	j_1 & j_2 & j_{12} \cr

2839: 	j_4 & j_3 & j_{23} 

2840: 	\end{array}\right\}

2841: 	= {\rm const.} \times \langle j_1j_2j_3j_4 j_{23} {\bf 0} \vert

2842:           j_1 j_2 j_3 j_4 j_{12} {\bf 0} \rangle,

2843: 	\label{6jdef}

2844: \end{equation}

2845: which is the unitary matrix in $(j_{12},j_{23})$ defining a change of

2846: basis in the subspace in which four angular momenta of given lengths

2847: add up to zero (${\bf 0}$ means ${\bf J}={\bf 0}$).  In this case

2848: there are eight observables on each side of the matrix element, of

2849: which seven are common and one is different.  Thus the amplitude of

2850: the $6j$-symbol is the inverse square root of the single Poisson

2851: bracket,

2852: \begin{equation}

2853: 	\{{\bf J}_{23}^2, {\bf J}_{12}^2\} = 4

2854: 	{\bf J}_1 \cdot ({\bf J}_2 \times {\bf J}_3),

2855: 	\label{6jPB}

2856: \end{equation}

2857: as follows immediately from (\ref{NjLiePoisson}).  One sees

2858: immediately that it is proportional to the volume of the tetrahedron.

2859: A similarly easy calculation is possible for the $9j$-symbol.  It is

2860: harder, however, to express these amplitudes in terms of the quantum

2861: numbers (the magnitudes $j_r$), that is, to translate these magnitudes

2862: into vectors ${\bf J}_r$ that lie on the stationary phase set.  We

2863: shall report on these and other extensions of our work in future

2864: publications.  

2865: 

2866: 

2867: 

2868: \section*{References}

2869: \begin{harvard}

2870: 

2871: \item[] Anderson R W and Aquilanti V 2006 {\it J. Chem. Phys.} {\bf

2872: 124} 214104

2873: 

2874: \item[] Aquilanti V, Cavalli S and De~Fazio D 1995 {\it J. Phys. Chem.}

2875: {\bf 99} 15694

2876: 

2877: \item[] Aquilanti V, Cavalli S and Coletti C 2001 {\it Chem. Phys. Lett.}

2878: {\bf 344} 587

2879: 

2880: \item[] Aquilanti V and Coletti C 2001 {\it Chem. Phys. Lett.} {\bf

2881: 344} 601 

2882: 

2883: \item[] Arnold V I 1967 {\it Functional Anal.\ Appl.} {\bf 1} 1

2884: 

2885: \item[] \dash 1989 {\it Mathematical Methods of Classical

2886: Mechanics} (New York: Springer-Verlag)

2887: 

2888: \item[] Barrett J W and Steele C M 2002 {\it Class. Quant. Grav.} {\bf

2889: 20} 1341

2890: 

2891: \item[] Baez J C, Christensen J D and Egan G 2002 {\it

2892: Class. Quant. Grav.} {\bf 19} 6489

2893: 

2894: \item[] Balazs N L and Jennings B K 1984 {\it Phys. Reports} {\bf 104}

2895: 347

2896: 

2897: \item[] Bargmann V 1962 {\it Rev. Mod. Phys.} {\bf 34} 829

2898: 

2899: \item[] Berry M V 1977 {\it Phil. Trans. Roy. Soc.} {\bf 287} 237

2900: 

2901: \item[] Berry M V and Mount K E 1972 {\it Rep.\ Prog.\ Phys.} {\bf 35}

2902: 315

2903: 

2904: \item[] Berry M V and Tabor M 1976 {\it Proc.\ Roy.\ Soc.\ Lond.\ A}

2905: {\bf 349} 101

2906: 

2907: \item[] Biedenharn L C and Louck J D 1981a {\it Angular Momentum in 

2908: Quantum Physics} (Reading, Massachusetts: Addison-Wesley)

2909: 

2910: \item[] \dash 1981b {\it The Racah-Wigner Algebra in Quantum Theory}

2911: (Reading, Massachusetts: Addison-Wesley)

2912: 

2913: \item[] Biedenharn L C and van Dam H 1965 {\it Quantum Theory of

2914:   Angular Momentum} (New York: Academic Press)

2915: 

2916: \item[] Brack Matthias and Bhaduri Rajat K 1997 {\it Semiclassical

2917: Physics} (Reading, Massachusetts: Addison-Wesley)

2918: 

2919: \item[] Brillouin M L 1926 {\it J.\ Phys.} {\bf 7} 353

2920: 

2921: \item[] Cargo Matthew, Gracia-Saz Alfonso, Littlejohn R  G , Reinsch

2922:   M W and de M. Rios P 2005,  {\it J. Phys.\ A} {\bf 38} 1977

2923: 

2924: \item[] Cargo Matthew, Gracia-Saz Alfonso and Littlejohn R G 2005,

2925:   preprint math-ph/0507032.

2926: 

2927: \item[] Cushman R H and Bates L 1997 {\it Global Aspects of Classical 

2928: Integrable Systems} (Basel: Birkh\"auser Verlag)

2929: 

2930: \item[] De~Fazio D, Cavalli S and Aquilanti V 2003 {\it

2931: Int. J. Quant. Chem.} {\bf 93} 91

2932: 

2933: \item[] de Gosson, M 1997 {\it Maslov Classes, Metaplectic

2934: Representation and Lagrangian Quantization (Mathematical Research

2935: Vol. 5)} (Berlin: Akademischer Verlag)

2936: 

2937: \item[] Einstein A 1917 {\it Verh.\ dt.\ Phys.\ Ges.} {\bf 19} 82

2938: 

2939: \item[] Estrada Ricardo, Gracia-Bond\'\i a J M and V\'arilly J C 1989

2940: {\it J. Math. Phys.} {\bf 30} 2789 

2941: 

2942: \item[] Frankel Theodore 1997 {\it The Geometry of Physics} (Cambridge)

2943: 

2944: \item[] Freidel Laurent and Louapre David 2003 {\it Class. Quantum Grav.}

2945: {\bf 20} 1267

2946: 

2947: \item[] Geronimo J S, Bruno O and Van Assche W 2004 {\it Oper. Theory

2948: Adv. Appl.} {\bf 154} 101

2949: 

2950: \item[] Girelli F and Livine E R 2005 {\it Class. Quantum Grav.} 

2951: {\bf 22} 3295

2952: 

2953: \item[] Gracia-Bond\'\i a J M and V\'arilly J C 1995 {\it

2954: J. Math. Phys.} {\bf 36} 2691

2955: 

2956: \item[] Groenewold H J 1946 {\it Physica} {\bf 12} 405

2957: 

2958: \item[] Gutzwiller Martin C 1990 {\it Chaos in Classical and Quantum

2959: Mechanics} (New York: Springer-Verlag)

2960: 

2961: \item[] Hillery M, O'Connell R F, Scully M O and Wigner E P 1984 {\it

2962: Phys. Reports} {\bf 106} 123

2963: 

2964: \item[] Keller J B 1958 {\it Ann.\ Phys.} {\bf 4} 180

2965: 

2966: \item[] Littlejohn R G 1986 {\it Phys. Reports} {\bf 138} 193

2967: 

2968: \item[] \dash 1990 {\it  J. Math.\ Phys.} {\bf 31} 2952

2969: 

2970: \item[] Littlejohn R G and Reinsch M 1995 {\it Phys.\ Rev.\ A} {\bf52}

2971: 2035

2972: 

2973: \item[] Littlejohn R G and Robbins J M 1987 {\it Phys.\ Rev.\ A}

2974: {\bf 36} 2953

2975:  

2976: \item[] Marsden J E and Ratiu T 1999 {\it Introduction to Mechanics and

2977:   Symmetry} (New York: Springer-Verlag)

2978: 

2979: \item[] Marzuoli A and Rasetti M 2005 {\it Ann. Phys.} {\bf 318} 345

2980: 

2981: \item[] Maslov V P and Fedoriuk M V 1981 {\it Semi-Classical

2982: Approximations in Quantum Mechanics} (Dordrecht: D. Reidel)

2983: 

2984: \item[] McDonald S 1988 {\it Phys. Reports} {\bf 158} 377

2985: 

2986: \item[] Miller W H 1974 {\it Adv. Chem. Phys.} {\bf 25} 69

2987: 

2988: \item[] Mishchenko A S, Shatalov V E and Sternin B Yu 1990 {\it Lagrangian

2989: manifolds and the Maslov operator} (Berlin: Springer Verlag)

2990: 

2991: \item[] Morehead J J 1995 {\it J.\ Math.\ Phys.} {\bf 36} 5431

2992: 

2993: \item[] Moyal J E 1949 {\it Proc. Camb. Phil. Soc.} {\bf 45} 99

2994: 

2995: \item[] Nakahara M 2003 {\it Geometry, Topology and Physics} (Bristol:

2996:   IOP Publishing)

2997: 

2998: \item[] Neville Donald 1971 {\it J. Math. Phys.} {\bf 12} 2438

2999: 

3000: \item[] Ozorio de Almeida Alfredo M 1998 {\it Phys.\ Reports} {\bf

3001: 295} 265

3002: 

3003: \item[] Percival I C 1973 {\it J.\ Phys.\ B} {\bf 6} L229

3004: 

3005: \item[] Ponzano G and Regge T 1968 in {\it Spectroscopy and Group

3006: Theoretical Methods in Physics} ed F Bloch \etal\ (Amsterdam:

3007: North-Holland) p~1

3008: 

3009: \item[] Reinsch M W and Morehead J J 1999 {\it J.\ Math.\ Phys.} {\bf

3010: 40} 4782

3011: 

3012: \item[] Roberts J 1999 {\it Geometry and Topology} {\bf 3} 21

3013: 

3014: \item[] Sakurai J J 1994 {\it Modern Quantum Mechanics} (New York:

3015:   Addison-Wesley)

3016: 

3017: \item[] Schulman L S 1981 {\it Techniques and Applications of Path

3018:   Integration} (New York:  John Wiley \& Sons)

3019: 

3020: \item[] Schulten K and Gordon R G 1975a {\it J. Math. Phys.} {\bf 16} 1961

3021: 

3022: \item[] \dash 1975b {\it J. Math. Phys.} {\bf 16} 1971

3023: 

3024: \item[] Smorodinskii Ya A and Shelepin L A 1972 {\it Sov. Phys. Usp.}

3025:   {\bf 15} 1

3026: 

3027: \item[] Taylor Y U and Woodward C T 2005 {\it Selecta Math. (N.S.)}

3028: {\bf 11} 539

3029: 

3030: \item[] Voros A 1977 {\it Ann. Inst. Henri Poincar\'e} {\bf 4} 343

3031: 

3032: \item[] Weyl H 1927 {\it Z. Phys.} {\bf 46} 1

3033: 

3034: \item[] Wigner E P 1932 {\it Phys. Rev.} {\bf 40} 749

3035: 

3036: \item[] \dash 1959 {\it Group Theory} (Academic Press, New York)

3037: 

3038: \end{harvard}

3039: 

3040: \Figures

3041: 

3042: \begin{figure}

3043: \caption{\label{SU2action} The action of an $SU(2)$ rotation about a

3044:   fixed axis on a point of the large phase space, and its projection

3045:   onto angular momentum space.}

3046: \end{figure} 

3047: 

3048: \begin{figure}

3049: \caption{\label{xpplane} Harmonic oscillator motion in the

3050:   $x_{r\mu}$-$p_{r\mu}$ plane, generated by $I_{r\mu}$.  Definition of

3051:   angle $\theta_{r\mu}$ is shown.}

3052: \end{figure}

3053: 

3054: \begin{figure}

3055: \caption{\label{3Jcones} The quantum state $\vert j_1j_2j_3 m_1m_2m_3

3056:   \rangle$ corresponds classically to a 3-torus in the small phase

3057:   space $S^2 \times S^2 \times S^2$, which can be visualized as three

3058:   ${\bf J}$ vectors lying on three cones with fixed values of $J_z$.  The

3059:   three azimuthal angles are independent.}

3060: \end{figure}

3061: 

3062: \begin{figure}

3063: \caption{\label{Jtriangle} If $j_1$, $j_2$ and $j_3$ satisfy the

3064:   triangle inequalities, then they define a triangle that is unique

3065:   apart from its orientation.  A standard orientation places the

3066:   triangle in the $x$-$z$ plane with sides ${\bf J}_1$, ${\bf J}_2$,

3067:   ${\bf J}_3$ oriented as shown.  The angle opposite ${\bf J}_r$ is

3068:   $\eta_r$.}

3069: \end{figure}

3070: 

3071: \begin{figure}

3072: \caption{\label{Wman} An schematic illustration showing how the Wigner

3073:   manifold in the large phase space is the inverse image under $\pi$

3074:   of a set of triangles formed from three angular momentum vectors

3075:   with vanishing sum, all related by rigid rotations.}

3076: \end{figure}

3077: 

3078: \begin{figure}

3079: \caption{\label{beta} By rotating the reference orientation of the

3080:   triangle about the $y$-axis, we can give ${\bf J}_3$ the desired

3081:   projection $m_3$ onto the $z$-axis.}

3082: \end{figure}

3083: 

3084: \begin{figure}

3085: \caption{\label{gamma} Once vector ${\bf J_3}$ has the desired

3086:   projection $m_3$, we rotate the triangle by angle $\gamma$ about the

3087:   axis ${\bf J}_3$ to make ${\bf J}_2$ have its desired projection

3088:   $m_2$.  This cannot always be done for real angles $\gamma$, but

3089:   when it can be done there are generically two angles that work,

3090:   illustated by points $Q$ and $Q'$ in the figure.}

3091: \end{figure}

3092: 

3093: 

3094: \end{document}

3095: