1: \documentclass[usenatbib]{mn2e}

2: 

3: \usepackage{epsf}

4: 

5: \newcommand\be{{\bmath e}}

6: \newcommand\bu{{\bmath u}}

7: \newcommand\bcdot{{\bmath\cdot}}

8: \newcommand\bnabla{{\bmath\nabla}}

9: \newcommand\bfT{\mathbf{T}}

10: \newcommand\bfone{\mathbf{1}}

11: \newcommand\real{\mathrm{Re}}

12: \newcommand\imag{\mathrm{Im}}

13: \newcommand\half{{\textstyle\frac{1}{2}}}

14: \newcommand\rmD{\mathrm{D}}

15: \newcommand\rmb{\mathrm{b}}

16: \newcommand\rmd{\mathrm{d}}

17: \newcommand\rme{\mathrm{e}}

18: \newcommand\rmi{\mathrm{i}}

19: \newcommand\rmK{\mathrm{K}}

20: \newcommand\rmm{\mathrm{m}}

21: \newcommand\rmp{\mathrm{p}}

22: \newcommand\rmr{\mathrm{r}}

23: \newcommand\rms{\mathrm{s}}

24: \newcommand\rmT{\mathrm{T}}

25: \newcommand\f{\frac}

26: \newcommand\p{\partial}

27: \newcommand\cst{\mathrm{constant}}

28: \newcommand\he{\mathrm{He}}

29: 

30: \title[Three-dimensional eccentric discs around Be stars]

31: {Three-dimensional eccentric discs around Be stars}

32: 

33: \author[Gordon I. Ogilvie]{Gordon I. Ogilvie\\

34: Department of Applied Mathematics and Theoretical Physics,

35: University of Cambridge, Centre for Mathematical Sciences,\\

36: Wilberforce Road, Cambridge CB3 0WA

37: }

38: 

39: \begin{document}

40: 

41: \maketitle

42: 

43: \label{firstpage}

44:  

45: \begin{abstract}

46:   One-armed oscillation modes in the circumstellar discs of Be~stars

47:   may explain the cyclical variations in their emission lines.  We

48:   show that a three-dimensional effect, involving vertical

49:   motion and neglected in previous treatments, profoundly influences

50:   the dynamics.  Using a secular theory of eccentric discs that

51:   reduces the problem to a second-order differential equation, we show

52:   that confined prograde modes are obtained for all reasonable disc

53:   temperatures and stellar rotation rates.  We confirm these results

54:   using a numerical analysis of the full set of linearized equations

55:   for three-dimensional isothermal discs including viscous

56:   terms that couple the horizontal motions at different altitudes.  In

57:   order to make these modes grow, viscous damping must be overcome by

58:   an excitation mechanism such as viscous overstability.

59: \end{abstract}

60: 

61: \begin{keywords}

62:   accretion, accretion discs --- circumstellar matter --- hydrodynamics ---

63:   stars: emission-line, Be

64: \end{keywords}

65: 

66: \section{Introduction}

67: 

68: Classical Be~stars \citep{2003PASP..115.1153P} are rapidly rotating

69: early-type stars that exhibit Balmer emission lines.  It is widely

70: agreed that these lines, which are generally double-peaked, originate

71: in a relatively thin circumstellar disc that is in approximately

72: Keplerian rotation \citep[and references

73: therein]{2007ASPC..361..230O}.  While the precise mechanism by which

74: the disc is formed remains controversial, it is likely to resemble a

75: viscous decretion disc that is expelled by the action of a torque at

76: its inner boundary \citep{1991MNRAS.250..432L,2003PASP..115.1153P}.

77: 

78: Many Be~stars show cyclical variations in their double-peaked

79: emission lines over years or decades, with the red and blue

80: peaks alternately becoming more prominent \citep[and references

81: therein]{1997A&A...318..548O}.  An explanation of this phenomenon was

82: given by \citet{1991PASJ...43...75O}, who proposed that a

83: low-frequency, one-armed oscillation mode \citep{1983PASJ...35..249K}

84: occurs in the disc.  This is equivalently to saying that the disc

85: becomes eccentric and the slow precession of its elliptical shape

86: gives rise to the cyclical changes in the observed emission lines.

87: 

88: Eccentric discs, in which fluid elements, solid particles or stars

89: follow elliptical orbits of variable eccentricity around a central

90: mass, have further applications in systems as diverse as planetary

91: rings, protoplanetary systems, close binary stars and galactic nuclei.

92: A detailed understanding of the origin of eccentricity and the rates

93: of precession in the circumstellar discs of Be~stars would therefore

94: be of general interest.

95: 

96: \citet{1991PASJ...43...75O} originally considered a disc that orbits

97: in a point-mass potential and obtained a sequence of retrograde modes

98: in which the precession of the disc is in a direction opposite to its

99: rotation.  Retrograde precession is a natural consequence of the

100: pressure forces in the disc, which cause a small departure from

101: Keplerian rotation and allow the eccentricity to propagate in a

102: wavelike manner.  The global modes found by

103: \citet{1991PASJ...43...75O} are weighted towards the outer part of the

104: disc and their periods become extremely long as the outer radius of

105: the disc is increased to realistic values.

106: 

107: \citet{1992A&A...265L..45P} and \citet{1993A&A...276..409S} considered

108: the effect of the quadrupole gravitational potential associated with

109: the rotational deformation of the star, which tends to cause a

110: prograde precession of elliptical orbits.  They showed that, when the

111: quadrupole effect is taken into account, prograde modes can be

112: obtained that are naturally confined in the inner part of the disc and

113: are insensitive to the outer boundary condition.  Subsequent

114: observations confirmed that the precession is indeed prograde

115: \citep{1994A&A...288..558T}.

116: 

117: However, \citet{1997A&A...318..548O} found that confined prograde

118: modes can be obtained only when the disc is sufficiently cool, so that

119: the quadrupole effect dominates over the tendency of pressure to

120: produce retrograde precession and extended modes.  He concluded that a

121: different mechanism is required in the hotter discs of early-type

122: Be~stars such as the prototype $\gamma$~Cas, and proposed

123: that radiative line forces could explain the required prograde

124: precession.  Unfortunately the modelling of radiative forces is

125: subject to considerable uncertainty.  \citet{2006A&A...447..277F}

126: found that the resulting model has little predictive power owing to

127: the sensitivity of the results to the parameters.

128: 

129: More recently, \citet{2006A&A...456.1097P} investigated an alternative

130: way to obtain prograde modes in hotter discs.  This involves replacing

131: the rigid inner boundary condition at the stellar surface with a free

132: boundary condition, on the basis that a gap is formed between the star

133: and the disc.  Such a gap is not generally expected in the scenario of

134: the viscous decretion disc but might be possible in alternative

135: models.

136: 

137: All of the treatments described so far are based on two-dimensional

138: models of the disc that neglect aspects of its vertical structure and

139: motion.  At first sight, such an approach seems reasonable for

140: studying eccentric modes in thin discs, where the motion might be

141: assumed to be purely horizontal and independent of height.  However,

142: in presenting a three-dimensional, non-linear theory of eccentric

143: discs, we have previously argued that three-dimensional effects are of

144: considerable importance \citep{2001MNRAS.325..231O}.  In particular,

145: the variation of the vertical gravitational acceleration around an

146: elliptical orbit excites an oscillatory vertical motion in an

147: eccentric disc that should not be neglected.

148: 

149: In this paper we show that, in fact, three-dimensional effects are

150: essential to understanding the precession of eccentric discs around

151: Be~stars.  They allow confined prograde modes to be obtained even when

152: the stellar quadrupole moment is negligible and when the inner

153: boundary is rigid.  This property allows us to give a unified

154: description of eccentric discs of Be~stars of all stellar types

155: without introducing uncertain radiative forces or modifying the inner

156: boundary condition.  While we do not deny that in some cases radiative

157: forces might be important, or that the inner boundary condition might

158: differ from a rigid one, we show that these innovations are

159: unnecessary.

160: 

161: Most previous analyses have not discussed the processes that

162: could cause eccentric modes to grow or decay, and which are therefore

163: relevant to explaining the occurrence of eccentric discs around

164: Be~stars.  Viscous overstability

165: \citep{1978MNRAS.185..629K,2006MNRAS.372.1829L} provides a possible

166: explanation; \citet{2001A&A...369..117N} have applied this idea to

167: Be~stars and estimated the associated growth rate.  We defer to a

168: future investigation a detailed analysis of the effects of viscous or

169: turbulent stresses.  A non-linear treatment, which could address the

170: saturation of the growth mechanism and attempt to predict the observed

171: amplitudes and precession rates of the eccentric modes, also remains

172: to be undertaken.

173: 

174: The structure of this paper is as follows.  In Section~\ref{s:basic}

175: we describe the basic state of the disc.  We then discuss an

176: approximate, secular theory of three-dimensional eccentric discs in

177: Section~\ref{s:secular}, comparing it with the two-dimensional

178: theories used by other authors.  In Section~\ref{s:full} we solve the

179: full linearized equations accurately using a spectral method,

180: including some effects of viscosity, and compare the results with

181: those of the secular theory.  Conclusions are given in

182: Section~\ref{s:conc}.

183: 

184: \section{Basic state of the disc}

185: \label{s:basic}

186: 

187: We consider a basic state consisting of a steady, axisymmetric disc

188: around a rotating star.  Adopting cylindrical polar coordinates

189: $(r,\phi,z)$, we write the gravitational potential of the star as

190: \begin{equation}

191:   \Phi=-GM(r^2+z^2)^{-1/2}\left[1+

192:   \f{Q}{3}\f{(r^2-2z^2)R^2}{(r^2+z^2)^2}\right],

193: \end{equation}

194: where $M$ and $R$ are its mass and (equatorial) radius.  This

195: potential consists of monopole and quadrupole components, the latter

196: being of dimensionless strength $Q$ and arising from the rotational

197: deformation of the star.  (The parameter $Q$ is related to the

198: gravitational moment $J_2$ used in planetary science by $Q=3J_2/2$.)

199: For a star with uniform angular velocity $\Omega_*$, we have

200: $Q=k_2\Omega_*^2R^3/GM$, where $k_2$ is the apsidal motion constant.

201: We neglect higher-order multipole components as well as the

202: self-gravitation of the disc.

203: 

204: In common with most other treatments, we assume that the stellar

205: radiation maintains the disc at a constant temperature $T_\rmd$,

206: slightly less than the effective temperature $T_\rme$ of the star.

207: The isothermal sound speed $c_\rms=(\mathcal{R}T_\rmd/\mu)^{1/2}$ is

208: therefore constant, both in the equilibrium state and for the

209: perturbations.  We neglect any viscous or turbulent stresses and any

210: resulting meridional motions in the disc.

211: 

212: The basic state then has velocity $\bu=r\Omega(r)\,\be_\phi$.  The

213: angular velocity $\Omega$ is independent of $z$ because the basic

214: state is isothermal and therefore barotropic.  The balance of forces

215: requires $r\Omega^2\,\be_r=\bnabla(\Phi+h)$, where $h=c_\rms^2\ln\rho$

216: is the (pseudo-)enthalpy of an isothermal gas.

217: 

218: The potential and enthalpy in the midplane are

219: $\Phi_\rmm(r)=\Phi(r,0)=-(GM/r)[1+(Q/3)(R/r)^2]$ and

220: $h_\rmm(r)=h(r,0)$, and the radial force balance implies

221: \begin{equation}

222:   \Omega^2=\f{GM}{r^3}\left[1+Q\left(\f{R}{r}\right)^2\right]+\f{1}{r}\p_rh_\rmm.

223: \label{omsq}

224: \end{equation}

225: The associated epicyclic frequency $\kappa(r)$ is given by

226: \begin{eqnarray}

227:   \kappa^2&=&\f{1}{r^3}\p_r(r^4\Omega^2)\nonumber\\

228:   &=&\f{GM}{r^3}\left[1-Q\left(\f{R}{r}\right)^2\right]+\f{1}{r^3}\p_r(r^3\p_rh_\rmm).

229: \label{kasq}

230: \end{eqnarray}

231: We also refer to the Keplerian angular velocity $\Omega_\rmK(r)$

232: defined by

233: \begin{equation}

234:   \Omega_\rmK^2=\f{GM}{r^3}.

235: \end{equation}

236: 

237: For a thin disc we may expand the potential about the midplane to

238: obtain $\Phi\approx\Phi_\rmm+\half\Omega_z^2z^2$, where the vertical

239: frequency $\Omega_z(r)$ is given by

240: \begin{equation}

241:   \Omega_z^2=\f{GM}{r^3}\left[1+3Q\left(\f{R}{r}\right)^2\right].

242: \end{equation}

243: The vertical force balance then implies $h=h_\rmm-\half\Omega_z^2z^2$

244: and so $\rho=\rho_\rmm\exp[-(z^2/2H^2)]$, where $\rho_\rmm(r)$ is the

245: midplane density and $H(r)=c_\rms/\Omega_z$ is the scaleheight.  The

246: surface density is $\Sigma(r)=(2\pi)^{1/2}\rho_\rmm H$.

247: 

248: Since the basic state is steady and axisymmetric, wave modes may be

249: considered in which the dependence on azimuth and time is of the form

250: $\exp(\rmi m\phi-\rmi\omega t)$, where $m$ is the azimuthal wavenumber

251: and $\omega$ is the wave frequency.  In the case $m=1$ that is of

252: interest here, $\omega$ is also the angular pattern speed of the mode

253: in an inertial frame of reference.  This can be identified with the

254: precession rate of the eccentric disc, which is positive if the

255: precession is prograde (i.e.\ in the same direction as the rotation of

256: the disc).

257: 

258: \section{Secular treatment}

259: \label{s:secular}

260: 

261: \subsection{Comparison of two- and three-dimensional theories}

262: 

263: An approximate theory can be derived in which the disc is assumed to

264: be nearly Keplerian and the frequency $\omega$ of the wave is much

265: less than the orbital frequency $\Omega$ of the disc.  This type of

266: approximation, which is related to the secular theory of celestial

267: mechanics, has been discussed in the two-dimensional linear case by

268: \citet{2001AJ....121.1776T}, \citet{2002A&A...388..615P} and

269: \citet{2006MNRAS.368.1123G}, and in the three-dimensional non-linear

270: case by \citet{2001MNRAS.325..231O}.  The derivation of this theory in

271: the case of a three-dimensional isothermal disc around a Be~star is

272: presented in Appendix~\ref{s:appendix}, which draws on the analysis of

273: Section~\ref{s:full} below.  The two-dimensional approximation is also

274: discussed there.

275: 

276: In the secular theory the radial velocity $u(r)$ satisfies the

277: equation

278: \begin{equation}

279:   (\omega-f)u=\f{1}{2r^2}\p_r\left[\f{c_\rms^2r^3}{\Sigma}\p_r\left(\f{\Sigma u}{r\Omega}\right)\right],

280: \label{secular}

281: \end{equation}

282: where $f(r)$ is given by

283: \begin{equation}

284:   f=\f{\Omega^2-\kappa^2}{2\Omega}

285: \label{f2d}

286: \end{equation}

287: in a two-dimensional disc, but

288: \begin{equation}

289:   f=\f{\Omega^2-\kappa^2}{2\Omega}+\f{9c_\rms^2}{4r^2\Omega}

290: \label{f3d}

291: \end{equation}

292: in a three-dimensional disc.  With the dependence

293: $\rme^{\rmi\phi-\rmi\omega t}$ assumed above, the radial velocity is

294: related to the complex eccentricity $E(r)=e\,\rme^{\rmi\varpi}$, where

295: $e(r)$ is the eccentricity and $\varpi(r)$ is the longitude of

296: periastron measured in a frame of reference that rotates with the

297: angular pattern speed $\omega$ of the mode, by $u^*=\rmi r\Omega E$

298: \citep{2001MNRAS.325..231O}.  It can be seen from

299: equation~(\ref{secular}) that $f$ is a local contribution to the

300: global precession rate $\omega$ of the eccentric mode.  Indeed, an

301: integral expression for the mode frequency in the case of rigid

302: boundary conditions\footnote{Although there is no physical

303: justification for a rigid outer boundary condition, a confined mode

304: decays sufficiently fast with increasing $r$ that the same integral

305: relation is recovered in the limit of large $r_\mathrm{out}$.  Such

306: modes are in fact insensitive to the outer boundary condition.} ($u=0$

307: at $r=r_\mathrm{in}$ and $r=r_\mathrm{out}$) is

308: \begin{eqnarray}

309:   \lefteqn{\omega\int_{r_\mathrm{in}}^{r_\mathrm{out}}\f{\Sigma r|u|^2}{\Omega}\,\rmd r=\int_{r_\mathrm{in}}^{r_\mathrm{out}}f\,\f{\Sigma r|u|^2}{\Omega}\,\rmd r}&\nonumber\\

310:   &&-\int_{r_\mathrm{in}}^{r_\mathrm{out}}\f{c_\rms^2r^3}{2\Sigma}\left|\p_r\left(\f{\Sigma u}{r\Omega}\right)\right|^2\,\rmd r,

311: \label{vp}

312: \end{eqnarray}

313: which follows from equation~(\ref{secular}) after multiplication by

314: $\Sigma ru^*/\Omega$ and an integration by parts.  The first term on

315: the right-hand side shows the contribution to the precession rate from

316: $f$ in the form of an integral weighted by the structure of the mode,

317: while the second term shows a retrograde contribution associated with

318: pressure.  (Note that the pressure also contributes indirectly through

319: its effect on $f$.)

320: 

321: In a two-dimensional disc $f$ is given by equation~(\ref{f2d}) and

322: corresponds to the expected expression for the local apsidal

323: precession rate.  Note that, in this case, $f\approx\Omega-\kappa$ if,

324: as assumed, the disc is nearly Keplerian

325: ($|\Omega^2-\kappa^2|\ll\Omega^2$).  There is, in fact, more than one

326: version of the two-dimensional theory.  \citet{1993A&A...276..409S}

327: work throughout with vertically averaged equations and find

328: \begin{equation}

329:   f=Q\f{GMR^2}{r^5\Omega}-\f{c_\rms^2}{2r^2\Omega}\p_r(r^2\p_r\ln\Sigma).

330: \label{2ds}

331: \end{equation}

332: \citet{1997A&A...318..548O} calculates $\Omega$ and $\kappa$ using a

333: three-dimensional equilibrium disc but then applies vertically

334: averaged equations for the perturbations.  He therefore uses

335: \begin{equation}

336:   f=Q\f{GMR^2}{r^5\Omega}-\f{c_\rms^2}{2r^2\Omega}\p_r(r^2\p_r\ln\rho_\rmm),

337: \label{2do}

338: \end{equation}

339: which agrees with equations~(\ref{omsq}) and~(\ref{kasq}) above.

340: 

341: In a three-dimensional disc, however, equation~(\ref{f3d}) gives the

342: relevant expression for $f$ as

343: \begin{equation}

344:   f=Q\f{GMR^2}{r^5\Omega}-\f{c_\rms^2}{2r^2\Omega}\p_r(r^2\p_r\ln\rho_\rmm)+

345:   \f{9c_\rms^2}{4r^2\Omega}.

346: \label{3d}

347: \end{equation}

348: The last term represents an additional local contribution to the

349: prograde precession of the disc that arises from the three-dimensional

350: dynamics including the vertical motion; it is discussed further in

351: Section~\ref{s:interp} below.  For the parameters relevant to

352: Be~stars, this term is never negligible (it corresponds to a

353: precession period of order $1\,\mathrm{yr}$ at the stellar surface)

354: and declines much more slowly with radius than the quadrupole term.

355: 

356: We refer to the three theories described above as `2DS'

357: \citep[equation~\ref{2ds}

358: above;][]{1993A&A...276..409S,2006A&A...456.1097P}, `2DO'

359: \citep[equation~\ref{2do}

360: above;][]{1991PASJ...43...75O,1997A&A...318..548O} and `3D'

361: (equation~\ref{3d}).  Note, however, that the secular approximations

362: were usually not employed in the cited papers.  The 3D~secular theory

363: is based on similar assumptions and approximations to the non-linear

364: analysis of \citet{2001MNRAS.325..231O}.  In particular, some viscous

365: or other stress is required to couple different layers in the disc so

366: that they tend to adopt the same eccentricity.  In the absence of such

367: stresses the eccentricity may vary significantly with $z$ and the

368: results can differ.  This complication is discussed in

369: Section~\ref{s:full}, where the full linearized equations are solved.

370: 

371: \subsection{Application to power-law discs}

372: \label{s:powerlaw}

373: 

374: It is consistent with the spirit of the secular approximation to

375: neglect the differences between $\Omega$, $\kappa$, $\Omega_z$ and

376: $\Omega_\rmK$ except where essential (i.e.\ in the quantity

377: $\Omega^2-\kappa^2$).  Accordingly, we may take $\Omega\propto

378: r^{-3/2}$ and $H\propto r^{3/2}$.  For a surface density profile

379: $\Sigma\propto r^{-\sigma}$ the density in the midplane varies as

380: $\rho_\rmm\propto r^{-\sigma-3/2}$.  We introduce the dimensionless

381: radial coordinate

382: \begin{equation}

383:   x=\f{r}{R}

384: \end{equation} 

385: and the small parameter

386: \begin{equation}

387:   \epsilon=c_\rms\left(\f{R}{GM}\right)^{1/2},

388: \end{equation}

389: which is a measure of the angular semithickness $H/r$ of the disc at the

390: stellar surface.  The three theories then give

391: \begin{equation}

392:   \mathrm{2DS:\phantom{O3D}}\f{f}{\Omega}=Qx^{-2}+\half\sigma\epsilon^2x,

393: \end{equation}

394: \begin{equation}

395:   \mathrm{2DO:\phantom{S3D}}\f{f}{\Omega}=Qx^{-2}+\half(\sigma+{\textstyle\f{3}{2}})\epsilon^2x,

396: \end{equation}

397: \begin{equation}

398:   \mathrm{3D:\phantom{S2DO}}\f{f}{\Omega}=Qx^{-2}+\half(\sigma+6)\epsilon^2x.

399: \end{equation}

400: We therefore consider the generic form

401: \begin{equation}

402:   \f{f}{\Omega}=Qx^{-2}+s\epsilon^2x,

403: \end{equation}

404: where $s$ is a constant.  It is convenient to work with dimensionless,

405: rescaled values of the mode frequency and quadrupole strength,

406: $\tilde\omega$ and $\tilde Q$, defined by

407: \begin{equation}

408:   \omega=\epsilon^2\left(\f{GM}{R^3}\right)^{1/2}\tilde\omega,

409: \end{equation}

410: \begin{equation}

411:   Q=\epsilon^2\tilde Q.

412: \end{equation}

413: Equation~(\ref{secular}) then becomes

414: \begin{eqnarray}

415:   \lefteqn{\left[\tilde\omega-(\tilde Qx^{-7/2}+sx^{-1/2})\right]u}&\nonumber\\

416:   &&=\f{1}{2x^2}\p_x\left[x^{\sigma+3}\p_x(x^{-\sigma+1/2}u)\right],

417: \label{secular2}

418: \end{eqnarray}

419: which is an eigenvalue problem of Sturm--Liouville form, when

420: considered together with appropriate boundary conditions.  Note that

421: the parameter $\epsilon$ drops out of the equation under the above

422: rescalings.  For prograde modes ($\tilde\omega>0$), one solution of

423: this equation decays exponentially as $x\to\infty$, while the other

424: solution grows exponentially.  We therefore consider the equation in

425: the domain $1<x<\infty$, requiring solutions to satisfy a rigid

426: boundary condition $u=0$ at the stellar surface $x=1$ and to decay as

427: $x\to\infty$.

428: 

429: Reasonable values of $\tilde Q$ for Be~stars can be estimated

430: following \citet{2006A&A...447..277F}.  Their Table~4 is based on

431: stellar models of intermediate main-sequence age.  There is

432: considerable uncertainty in the applicability of these models because

433: of the rapid rotation of Be~stars.  Nevertheless, using these data, we

434: estimate that $\tilde Q$ ranges from approximately~$8$ at the lower

435: end ($M=2.51\,M_\odot$) to approximately~12 at the upper end

436: ($M=15.85\,M_\odot$).  These values assume, somewhat arbitrarily, a

437: disc temperature of $T_\rmd=(2/3)T_\rme$ and a stellar rotation rate

438: that is $95\%$ of the critical value.  Similarly, for the stellar

439: models of $\gamma$~Cas and $59$~Cyg described in Tables~2 and~3 of

440: \citet{2006A&A...447..277F}, we estimate $\tilde Q\approx9$ and~$13$,

441: respectively.  It is clear that significantly larger values of $\tilde

442: Q$ cannot be obtained by further increasing the stellar rotation rate,

443: and in fact smaller values may be more appropriate.  The values of

444: $\epsilon$ for all these models are in the range $0.021$--$0.026$, again

445: assuming that $T_\rmd=(2/3)T_\rme$.

446: 

447: \subsection{Solutions in the absence of a quadrupole}

448: 

449: Consider first the case without a quadrupole term, $\tilde Q=0$.  The

450: solution that decays as $x\to\infty$ is then

451: \begin{equation}

452:   u=x^{(\sigma-3)/2}K_\nu(z),

453: \label{knu}

454: \end{equation}

455: with

456: \begin{equation}

457:   \nu=2[(\sigma+2)^2-8s]^{1/2},\qquad z=4(2\tilde\omega)^{1/2}x^{1/4},

458: \end{equation}

459: where $K_\nu$ is the modified Bessel function of the second kind of

460: order $\nu$ \citep{1965hmfw.book.....A}.  Note that $z$ is real and

461: positive for the prograde modes of interest.  This solution has no

462: zeros in $z>0$ if $\nu$ is real, but does so if $\nu$ is imaginary.

463: To match a rigid boundary condition at $x=1$ we therefore require

464: $\nu^2<0$.  The 2DS~theory gives $8s=4\sigma$, so

465: $\nu^2=4(\sigma^2+4)$ and modes are never confined.  The 2DO~theory

466: gives $8s=4\sigma+6$, so $\nu^2=4(\sigma^2-2)$ and modes may be

467: confined for $\sigma^2<2$.  The 3D~theory gives $8s=4\sigma+24$, so

468: $\nu^2=4(\sigma^2-20)$ and modes may be confined for $\sigma^2<20$.

469: 

470: This analysis is somewhat misleading because the modes may not be

471: adequately confined to be applicable to the discs of Be~stars.  In

472: Fig.~\ref{f:knu} we plot $K_\nu(z)$ versus $x$ for $\nu=8\rmi$ and

473: $\nu=\sqrt{8}\rmi$, choosing $\tilde\omega$ such that $K_\nu(z)=0$ at

474: $x=1$.  Only the former case provides an adequately confined mode.

475: This shows that the 2DO~theory cannot in fact produce confined

476: prograde modes in the absence of a quadrupole, even for the most

477: optimistic choice of surface density profile, because $|\nu|$ is never

478: large enough.  However, sufficiently large imaginary values of $\nu$

479: are obtained in the 3D~theory.

480: 

481: \begin{figure}

482: \centerline{\epsfysize8cm\epsfbox{fig1.eps}}

483: \caption{Confinement of prograde modes in the absence of a quadrupole.

484: The modified Bessel function $K_\nu(z)$ (multiplied by a constant for

485: convenient normalization) is plotted against $x=r/R$ for the cases

486: $\nu=8\rmi$ (solid line) and $\nu=\sqrt{8}\rmi$ (dashed line).

487: According to equation~(\ref{knu}), the modified Bessel function, which

488: decays exponentially as $x\to\infty$ if $\nu$ is imaginary, describes

489: the confinement of the eigenfunction.  The solid line is relevant to

490: the 3D~theory for a surface density index $\sigma=2$, and shows a good

491: confinement of the mode.  The dashed line is relevant to the

492: 2DO~theory in the (optimal) case $\sigma=0$, but here the confinement

493: is very poor.}

494: \label{f:knu}

495: \end{figure}

496: 

497: When $\nu$ is imaginary, confined solutions are obtained, in

498: principle, when $\tilde\omega=z_n^2/32$ where $z_{n,\nu}$ is the $n$th

499: zero of $K_\nu(z)$ in $z>0$.  For example, consider the 3D~theory with

500: $\sigma=2$, which is the value expected for a steady decretion disc

501: with constant alpha viscosity parameter, far inside its outer radius.

502: Then $s=4$ and $\nu=8\rmi$, and zeros occur at $z=4.802$, $3.067$,

503: $2.029$, $\dots$\ Therefore, in principle, $\tilde\omega=0.7207$,

504: $0.2940$, $0.1286$, $\dots$ are the scaled frequencies of confined

505: modes.  However, only the first of these, corresponding to an

506: eigenfunction with no nodes (Fig.~\ref{f:knu}, solid line), gives rise

507: to a mode that is adequately confined in a disc of modest radial

508: extent.

509: 

510: \subsection{Critical quadrupole strength}

511: \label{s:qcrit}

512: 

513: The integral expression for the dimensionless frequency eigenvalue of

514: a mode that satisfies a rigid boundary condition at the stellar

515: surface and decays as $r\to\infty$ is (cf.\ equation~\ref{vp})

516: \begin{eqnarray}

517:   \lefteqn{\tilde\omega\int_1^\infty x^{-\sigma+5/2}u^2\,\rmd x=

518:   \int_1^\infty(\tilde Qx^{-\sigma-1}+sx^{-\sigma+2})u^2\,\rmd x}&\nonumber\\

519:   &&-\f{1}{2}\int_1^\infty x^{\sigma+3}\left[\p_x(x^{-\sigma+1/2}u)\right]^2\,\rmd x,

520: \end{eqnarray}

521: where we take $u$ to be real.  Following \citet{2006A&A...456.1097P},

522: we note that this expression has the usual variational property

523: associated with self-adjoint eigenvalue problems.  In this case, a

524: confined prograde mode exists if and only the right-hand side of this

525: equation can be made positive by a trial function $u(x)$ satisfying

526: the appropriate boundary conditions.  This is clearly possible if

527: either $\tilde Q$ or $s$ is large enough.  What is the minimum value

528: of $\tilde Q$ (for a given value of $s$) such that the right-hand side

529: can just be made to vanish for a non-trivial $u$?  This happens when

530: \begin{eqnarray}

531:   \lefteqn{\tilde Q\int_1^\infty x^{-\sigma-1}u^2\,\rmd x=

532:   \f{1}{2}\int_1^\infty

533:   x^{\sigma+3}\left[\p_x(x^{-\sigma+1/2}u)\right]^2\,\rmd

534:   x}&\nonumber\\ &&-\int_1^\infty sx^{-\sigma+2}u^2\,\rmd x.

535: \end{eqnarray}

536: The minimum value of $\tilde Q$ for which this equation can be

537: satisfied is given by the corresponding Euler--Lagrange equation,

538: which is just equation~(\ref{secular2}) with $\tilde\omega$ set to

539: zero.  The solutions are

540: \begin{equation}

541:   u=x^{(\sigma-3)/2}J_{\pm\mu}(w),

542: \end{equation}

543: with

544: \begin{equation}

545:   \mu=\f{\nu}{6}=\f{1}{3}[(\sigma+2)^2-8s]^{1/2},\qquad w=\f{\tilde Q^{1/2}}{3}\left(\f{x}{2}\right)^{-3/2}.

546: \end{equation}

547: If $\mu^2<0$ we have seen that confined modes can be found, in

548: principle, even if $\tilde Q=0$.  Consider then the case $\mu^2>0$.

549: The solution that decays\footnote{Since $J_{\mu}(w)\propto w^\mu$ for

550: small $w$, $u\propto x^{[\sigma-3(1+\mu)]/2}$ for large $x$.  This

551: solution satisfies the condition that

552: $x^{7/2}u\p_x(x^{-\sigma+1/2}u)\to0$ as $x\to\infty$, which is

553: required to carry out the integration by parts and derive the

554: variational principle and Euler--Lagrange equation, while $J_{-\mu}$

555: does not.} as $x\to\infty$ is $J_{\mu}(w)$.  The critical condition

556: for a rigid inner boundary condition is therefore $\tilde

557: Q=9w_\mu^2/8$, where $w_\mu$ is the first zero of $J_\mu(w)$ in $w>0$.

558: (This is an increasing function of $\mu$ and therefore a decreasing

559: function of $s$.)

560: 

561: For example, if $\sigma=2$, we require $\tilde Q>10.8$ for confinement

562: in the 2DO~theory, or $\tilde Q>15.9$ in the 2DS~theory.  These values

563: are not very sensitive to $\sigma$; \citet{2006A&A...456.1097P} quote

564: $\tilde Q>17.3$ for $\sigma=5/2$ and a rigid inner boundary condition,

565: which agrees with this analysis.

566: 

567: \subsection{Schr\"odinger analogy}

568: 

569: A useful description of confined modes uses an analogy with bound

570: states in quantum mechanics.  If $y=x^{1/4}$ and

571: $u=x^{\sigma/2-13/8}\psi(y)$, then $\psi(y)$ satisfies the

572: Schr\"odinger equation

573: \begin{equation}

574:   -\f{\rmd^2\psi}{\rmd y^2}+[V(y)-E]\psi=0

575: \end{equation}

576: with an effective potential

577: \begin{equation}

578:   V=\f{1}{4}(16\sigma^2+64\sigma+63-128s)y^{-2}-32\tilde Qy^{-14}

579: \label{v}

580: \end{equation}

581: and an effective energy eigenvalue $E=-32\tilde\omega$.  The

582: coefficient of $1/4y^2$ in $V$ is $(16\sigma^2+63)$ for the

583: 2DS~theory, $(16\sigma^2-33)$ for 2DO and $(16\sigma^2-321)$ for 3D.

584: It is therefore likely to be negative only in the 3D~theory.  Note

585: that the disc occupies the region $y>1$.  A bound state of negative

586: energy, equivalent to a confined prograde mode, can be obtained if

587: there is a sufficiently deep and wide potential well.  The $\tilde Q$

588: term tends to create a deep well close to the stellar surface.  In the

589: 2D~theories it competes with the $y^{-2}$ term, which contributes a

590: repulsive potential.  In the 3D~theory, however, the $y^{-2}$ term

591: creates a much wider well, so allowing broader modes with slower

592: precession rates, even if $\tilde Q=0$.  These potentials are plotted

593: in Fig.~\ref{f:v} for the case $\sigma=2$ and $\tilde Q=10$.  For

594: these parameters the 3D~potential is deep and wide enough to support a

595: bound state in the case of a rigid inner boundary condition,

596: while the others are not.

597: 

598: \begin{figure}

599: \centerline{\epsfysize8cm\epsfbox{fig2.eps}}

600: \caption{Effective potentials $V(y)$ (equation~\ref{v}) describing the

601: confinement of prograde modes by analogy with bound states in quantum

602: mechanics.  Potentials are shown for the three theories for the case

603: $\sigma=2$ and $\tilde Q=10$.  The bottom of the potential well at

604: $y=1$ (i.e.\ at the stellar surface) in each case is considerably

605: deeper than can be shown on the graph.  However, only in the 3D~theory

606: is the well deep and wide enough in this case to support a bound state

607: in the case of a rigid inner boundary condition.}

608: \label{f:v}

609: \end{figure}

610: 

611: \subsection{Precession rates and mode shapes}

612: 

613: \begin{figure*}

614: \centerline{\epsfysize8.5cm\epsfbox{fig3a.eps}\qquad\epsfysize8.5cm\epsfbox{fig3b.eps}}

615: \caption{Dimensionless precession rates of confined prograde modes in

616: the three theories, for the case $\sigma=2$.  The left and

617: right panels are for rigid and free inner boundary conditions,

618: respectively.  These results are insensitive to the outer boundary

619: condition provided that the disc is reasonably large.  In the

620: 2D~theories, confined prograde modes are found only for sufficiently

621: large values of $\tilde Q$.  The dotted line indicates a

622: second confined mode, with one radial node in its eigenfunction, found

623: in the 3D~theory; this and higher-order modes are less well confined

624: and therefore more sensitive to the outer boundary condition.}

625: \label{f:ot}

626: \end{figure*}

627: 

628: We now compute the eigenvalues $\tilde\omega$ of

629: equation~(\ref{secular2}) by a shooting method, adopting a rigid inner

630: boundary condition at the stellar surface and seeking prograde

631: confined modes that decay as $x\to\infty$.  (In practice this is done

632: by imposing a rigid outer boundary condition and verifying that the

633: eigenvalue is completely insensitive to the value of the outer radius

634: provided it is sufficiently large.)  The results are shown in the left

635: panel of Fig.~\ref{f:ot}.  Here we confirm that confined prograde

636: modes exist in the 2D~theories only for sufficiently large values of

637: $\tilde Q$, as described in Section~\ref{s:qcrit}.  In contrast, the

638: 3D~theory allows such modes to be obtained for any value of $\tilde

639: Q$.  Furthermore, the precession rates are much larger in the

640: 3D~theory.  Since the realistic values of $\tilde Q$ are probably in

641: the vicinity of $10$ or smaller, it is clear that the

642: three-dimensional effects are of essential importance in the

643: case of a rigid inner boundary condition.

644: 

645: In the right panel of Fig.~\ref{f:ot} we show comparable

646: results for the free inner boundary condition considered by

647: \citet{2006A&A...456.1097P}, meaning that the Lagrangian pressure

648: perturbation is zero at $r=R$.  In the secular theory this is

649: equivalent to $\p_rE=0$, or $u+2r\p_r u=0$.  As described by

650: \citet{2006A&A...456.1097P}, a free inner boundary condition allows

651: prograde modes to be found for smaller values of $\tilde Q$ in the

652: 2DS~theory.  It is still true, however, that the three-dimensional

653: effects have an important effect on the results.  Since the

654: eigenfunctions obtained with a free inner boundary condition are

655: generally peaked at the inner radius, there may also be a conflict

656: between the nonlinear development of such an eccentric mode and the

657: existence of a stellar surface, unless there is a wide gap between the

658: star and the disc.

659: 

660: To convert these eigenvalues into physical units, we again make use of

661: the stellar models in \citet{2006A&A...447..277F}.  We then find that

662: the precession period is

663: \begin{equation}

664:   P=\f{C_P}{\tilde\omega}\,\mathrm{yr},

665: \end{equation}

666: where $C_P$ ranges from approximately $1.6$ at the lower end to

667: approximately $2.3$ at the upper end (or $2.4$ for $\gamma$~Cas and

668: $1.8$ for $59$~Cyg).  This conversion factor is independent of

669: assumptions regarding the stellar rotation rate, except inasmuch as

670: the rotation affects the equatorial radius, but $C_P$ is inversely

671: proportional to the assumed disc temperature.

672: 

673: The shapes of the confined modes in the 3D~theory are illustrated in

674: Fig.~\ref{f:ef}.  For larger values of $\tilde Q$, the mode is

675: increasingly confined in the inner part of the disc.  Although

676: higher-order modes may exist, one of which is referred to in

677: Fig.~\ref{f:ot}, we focus here on the fundamental confined mode with

678: the simplest radial structure.

679: 

680: \begin{figure}

681: \centerline{\epsfysize8cm\epsfbox{fig4.eps}}

682: \caption{Eigenfunctions of confined prograde modes in the 3D~secular

683: theory with a rigid inner boundary condition.  The eccentricity

684: (proportional to $u/r\Omega$, and with arbitrary normalization) is

685: plotted versus radius for the case $\sigma=2$, and for relative

686: quadrupole strengths $\tilde Q=0$ (solid line), $5$ (dotted line),

687: $10$ (dashed line) and $20$ (dot-dashed line).  The first case

688: corresponds to the solid line in Fig.~\ref{f:knu}.}

689: \label{f:ef}

690: \end{figure}

691: 

692: Since $\tilde\omega$ is close to~$1$ in the 3D~theory for reasonable

693: values of~$\tilde Q$ and with a rigid inner boundary

694: condition, precession rates in the vicinity of $1$--$3\,\mathrm{yr}$

695: are obtained.  Observed cycle times, which can vary significantly even

696: for the same star, are more usually in the range

697: $5$--$10\,\mathrm{yr}$ \citep{1997A&A...318..548O}.  This discrepancy

698: might occur because the discs have different temperature or density

699: profiles from those assumed here, either of which would affect the

700: precession period.  Non-linearity of the eccentric mode may also

701: increase the precession period by altering the distribution of

702: eccentricity so that it is less peaked in the inner part of the disc.

703: Furthermore, non-isothermal effects or radiative forces might be

704: important.  In some cases, including $\gamma$~Cas and $59$~Cyg, the

705: presence of a relative close binary companion may also affect the

706: dynamics of the eccentric mode, although it would seem most

707: likely to contribute to the prograde precession and therefore to

708: decrease the precession period.

709: 

710: \subsection{Physical interpretation of the three-dimensional dynamics}

711: \label{s:interp}

712: 

713: What is the origin of the additional prograde precession that occurs

714: in a three-dimensional disc?  As described in

715: Appendix~\ref{s:appendix} and Section~\ref{s:full} below, two

716: different types of motion are involved in an eccentric disc.  One

717: (corresponding to the $n=0$ mode in later sections) consists of

718: horizontal velocities and enthalpy perturbations that are independent

719: of $z$; this describes the eccentric orbital motion of the gas.  The

720: other (corresponding to $n=2$) involves a vertical velocity

721: proportional to $z$ and an enthalpy perturbation proportional to

722: $z^2-H^2$; this is a vertical `breathing' mode of the disc.  Coupling

723: of these motions occurs because of the variation of the vertical

724: gravitational acceleration (or, equivalently, the vertical frequency

725: $\Omega_z$, or the scaleheight $H$) with radius.  Vertical hydrostatic

726: equilibrium cannot be maintained in an eccentric disc because a fluid

727: element in an elliptical orbit experiences a vertical gravitational

728: acceleration that oscillates with the orbital frequency.  The

729: breathing mode is excited and the associated enthalpy perturbation

730: affects the horizontal dynamics, contributing to the precession of the

731: eccentric mode.  This contribution is found to be always

732: positive.  In the more general situation of a non-isothermal disc

733: undergoing adiabatic perturbations, the three-dimensional contribution

734: to $f$ (in which we compare a 3D-type theory with a 2DS-type theory)

735: is

736: \begin{equation}

737:   \f{3(\gamma+1)}{2\gamma}\f{P}{\Sigma r^2\Omega},

738: \end{equation}

739: where $\gamma$ is the adiabatic index and $P$ is the

740: vertically integrated pressure (Ogilvie \& Goodchild, in preparation),

741: but the derivation of this expression is beyond the scope of the

742: present paper.

743: 

744: \section{Full treatment}

745: \label{s:full}

746: 

747: \subsection{Inviscid dynamics}

748: 

749: The dynamical equations governing an isothermal, inviscid disc are

750: \begin{equation}

751:   (\p_t+\bu\bcdot\bnabla)\bu=-\bnabla(\Phi+h),

752: \label{motion}

753: \end{equation}

754: \begin{equation}

755:   (\p_t+\bu\bcdot\bnabla)h=-c_\rms^2\bnabla\bcdot\bu.

756: \end{equation}

757: When these are linearized about the basic state described in

758: Section~\ref{s:basic}, we obtain

759: \begin{equation}

760:   -\rmi\hat\omega u_r'-2\Omega u_\phi'=-\p_r h',

761: \end{equation}

762: \begin{equation}

763:   -\rmi\hat\omega u_\phi'+\f{\kappa^2}{2\Omega}u_r'=-\f{\rmi mh'}{r},

764: \end{equation}

765: \begin{equation}

766:   -\rmi\hat\omega u_z'=-\p_z h',

767: \end{equation}

768: \begin{eqnarray}

769:   \lefteqn{-\rmi\hat\omega h'+u_r'\p_r h+u_z'\p_z h}&\nonumber\\

770:   &&=-c_\rms^2\left[\f{1}{r}\p_r(ru_r')+\f{\rmi mu_\phi'}{r}+\p_zu_z'\right],

771: \end{eqnarray}

772: where $\hat\omega=\omega-m\Omega$ is the Doppler-shifted wave

773: frequency and the perturbations, denoted by primed quantities, have

774: the dependence $\exp(-\rmi\omega t+\rmi m\phi)$.

775: 

776: Following \citet{1987PASJ...39..457O}, we decompose the vertical

777: structure of the mode into the basis of Hermite polynomials defined by

778: \begin{equation}

779:   \he_n(\zeta)=\rme^{\zeta^2/2}\left(-\f{\rmd}{\rmd \zeta}\right)^n\rme^{-\zeta^2/2},

780: \end{equation}

781: where $\zeta=z/H$ is a dimensionless vertical coordinate and

782: $n=0,1,2,\dots$\  These polynomials satisfy the differential equation

783: \begin{equation}

784:   \he_n''(\zeta)-\zeta\,\he_n'(\zeta)+n\,\he_n(\zeta)=0,

785: \end{equation}

786: the recurrence relations

787: \begin{equation}

788:   \he_n'(\zeta)=n\,\he_{n-1}(\zeta),

789: \end{equation}

790: \begin{equation}

791:   \zeta\,\he_n(\zeta)=\he_{n+1}(\zeta)+n\,\he_{n-1}(\zeta)

792: \end{equation}

793: and the orthogonality relation

794: \begin{equation}

795:   \int_{-\infty}^\infty\rme^{-\zeta^2/2}\,\he_m(\zeta)\,\he_n(\zeta)\,\rmd \zeta=(2\pi)^{1/2}n!\,\delta_{mn}.

796: \end{equation}

797: The first three are $\he_0(\zeta)=1$, $\he_1(\zeta)=\zeta$ and

798: $\he_2(\zeta)=\zeta^2-1$.

799: 

800: We therefore expand

801: \begin{eqnarray}

802:   u_r'(r,z)&=&\sum_nu_n(r)\,\he_n(\zeta),\\

803:   u_\phi'(r,z)&=&\sum_nv_n(r)\,\he_n(\zeta),\\

804:   u_z'(r,z)&=&\sum_nw_n(r)\,\he_{n-1}(\zeta),\\

805:   h'(r,z)&=&\sum_nh_n(r)\,\he_n(\zeta).

806: \end{eqnarray}

807: with $u_n=v_n=h_n=0$ for $n<0$ and $w_n=0$ for $n<1$.  Bearing in mind

808: that $H$ depends on $r$, we have

809: \begin{equation}

810:   \p_r\,\he_n(\zeta)=-(\p_r\ln H)[n\,\he_n(\zeta)+n(n-1)\,\he_{n-2}(\zeta)].

811: \end{equation}

812: The projected equations are then

813: \begin{eqnarray}

814:   \lefteqn{-\rmi\hat\omega u_n-2\Omega v_n=-\p_r h_n}&\nonumber\\

815:   &&+(\p_r\ln H)[nh_n+(n+1)(n+2)h_{n+2}],

816: \label{un}

817: \end{eqnarray}

818: \begin{equation}

819:   -\rmi\hat\omega v_n+\f{\kappa^2}{2\Omega}u_n=-\f{\rmi mh_n}{r},

820: \label{vn}

821: \end{equation}

822: \begin{equation}

823:   -\rmi\hat\omega w_n=-\f{nh_n}{H},

824: \label{wn}

825: \end{equation}

826: \begin{eqnarray}

827:   \lefteqn{-\rmi\hat\omega\f{h_n}{c_\rms^2}+\f{1}{r\Sigma}\p_r(r\Sigma u_n)+\f{\rmi mv_n}{r}-\f{w_n}{H}}&\nonumber\\

828:   &&+(\p_r\ln H)(nu_n+u_{n-2})=0,

829: \label{hn}

830: \end{eqnarray}

831: \citep[cf.][]{2002ApJ...565.1257T,2006MNRAS.368..917Z}.

832: 

833: This approach corresponds to a (Galerkin) spectral treatment of the

834: partial differential equations governing the linearized dynamics,

835: which is much preferable to a finite-difference treatment.  In

836: practice this system of equations must be truncated by setting $u_n$,

837: etc., to zero for $n>N$ for some integer $N$.  The 2DO~theory is

838: obtained, in fact, by considering a radical truncation, $N=0$, of the

839: equations.

840: 

841: \subsection{Selected viscous effects}

842: \label{s:viscous}

843: 

844: To include viscosity, a term

845: \begin{equation}

846:   \f{1}{\rho}\bnabla\bcdot\bfT

847: \end{equation}

848: should be added to the right-hand side of the equation of

849: motion~(\ref{motion}), where

850: \begin{equation}

851:   \bfT=\rho\nu[\bnabla\bu+(\bnabla\bu)^\rmT]+\rho(\nu_\rmb-{\textstyle\f{2}{3}}\nu)(\bnabla\bcdot\bu)\bfone

852: \end{equation}

853: is the viscous stress tensor.  In the context of an isothermal disc it

854: is reasonable to assume that the kinematic shear and bulk viscosities

855: $\nu$ and $\nu_\rmb$ depend only on $r$.  We parametrize them as

856: \begin{equation}

857:   \nu=\alpha c_\rms H,\qquad\nu_\rmb=\alpha_\rmb c_\rms H.

858: \end{equation}

859: 

860: A full treatment of the effects of viscosity is complicated, not only

861: because the above expression for the viscous force must be evaluated

862: in cylindrical polar coordinates and then projected on to the basis of

863: Hermite polynomials, but also because the basic state is modified to

864: include a meridional flow driven by viscous stresses, which should be

865: considered in the linearized equations.  This problem is therefore

866: deferred to a future investigation.

867: 

868: In the present paper we adopt a simpler approach in which only

869: selected viscous effects are included.  We consider what might

870: be assumed to be the dominant viscous terms, i.e.\ those involving two

871: derivatives with respect to~$z$.  Since

872: \begin{equation}

873:   \f{1}{\rho}\p_z[\rho\p_z\,\he_n(\zeta)]=-\f{n}{H^2}\,\he_n(\zeta),

874: \end{equation}

875: the inviscid perturbation equations (\ref{un})--(\ref{wn}) are

876: modified by the addition of the viscous terms

877: \begin{equation}

878:   -\rmi\hat\omega u_n=\cdots-\f{\nu}{H^2}nu_n,

879: \label{unviscous}

880: \end{equation}

881: \begin{equation}

882:   -\rmi\hat\omega v_n=\cdots-\f{\nu}{H^2}nv_n,

883: \end{equation}

884: \begin{equation}

885:   -\rmi\hat\omega w_n=\cdots-\f{(\nu_\rmb+{\textstyle\f{4}{3}}\nu)}{H^2}(n-1)w_n,

886: \label{wnviscous}

887: \end{equation}

888: while equation~(\ref{hn}) is unchanged.  These terms act to damp the

889: mode, but have most effect on components of large $n$.  They have no

890: effect on $u_0$ and $v_0$, which represent horizontal motions

891: independent of $z$.  These viscous terms can also be thought of as

892: providing a coupling between different layers of the disc and thereby

893: encouraging it to adopt a horizontal motion independent of $z$.  We

894: show below that this effect is of considerable importance.

895: 

896: \subsection{Numerical solutions}

897: 

898: We solve the system of ordinary differential equations in $r$ for

899: modes with $m=1$ using a Chebyshev collocation (i.e.\ pseudospectral)

900: method.  This approach converts the differential equations and

901: boundary conditions into an algebraic generalized eigenvalue problem

902: for the frequency $\omega$, which we solve using a standard direct

903: method.  Specifically, equations~(\ref{un})--(\ref{hn}),

904: supplemented by the viscous terms

905: (\ref{unviscous})--(\ref{wnviscous}), are solved for $n=0,2,4,\dots,N$

906: with $u_n$, etc., set to zero for $n>N$.  Rigid boundary conditions

907: $u_n=0$ are adopted at both inner and outer boundaries, but it is

908: ensured that the modes obtained are completely insensitive to the

909: value of the outer radius and therefore to the choice of outer

910: boundary condition.

911: 

912: For comparison with the results in Section~\ref{s:powerlaw}, we

913: consider a disc with a midplane density profile $\rho_\rmm\propto

914: r^{-\sigma-3/2}$.  We also include a shear viscosity corresponding to

915: a constant $\alpha$ parameter, but no bulk viscosity.

916: 

917: Sample results are shown in Table~\ref{t:ot}.  The convergence of the

918: eigenfrequency with increasing truncation order $N$ of the Hermite

919: polynomial basis is remarkable.  The case $N=0$ corresponds exactly to

920: the two-dimensional theory considered by \citet{1991PASJ...43...75O},

921: and therefore agrees well with the 2DO~secular approximation.  Here

922: $\tilde Q=20$ is large enough to support a confined prograde mode.

923: The precession rate is much larger in the case $N=2$ and hardly varies

924: as further Hermite polynomials are included.  It agrees reasonably

925: well with the 3D~secular theory for an inviscid disc.  The slight

926: offset of the precession frequency is attributable partly to errors in

927: the secular approximation, which is valid only to leading order in

928: $\epsilon$, and partly to the effects of viscosity.  As described in

929: Appendix~\ref{s:appendix}, the viscous damping of vertical motions

930: considered in the full model can be represented within the 3D~secular

931: theory by multiplying the coefficient $9/4$ in the three-dimensional

932: expression~(\ref{f3d}) for $f$ by $(1-\rmi\beta)/(1+\rmi\beta)$, where

933: $\beta=\alpha_\rmb+{\textstyle\f{4}{3}}\alpha$.  Table~\ref{t:ot}

934: shows that this viscous secular theory gives good agreement with the

935: full model for $\alpha=0.1$.

936: 

937: \begin{table}

938: \caption{Scaled frequency eigenvalues obtained from the full

939: linearized equations for a disc with $\epsilon=0.02$, $\alpha=0.1$,

940: $\tilde Q=20$ and $\sigma=2$.  Rapid convergence is seen with

941: increasing values of the vertical truncation number $N$.  Comparable

942: results from the secular theories are shown below.  A negative value

943: of $\mathrm{Im}(\tilde\omega)$ represents the (scaled) exponential

944: damping rate of the mode.}

945: \begin{tabular}{lll}

946: Version&$\mathrm{Re}(\tilde\omega)$&$\mathrm{Im}(\tilde\omega)$\\

947: \hline

948: Full, $N=0$&1.189356&\\

949: Full, $N=2$&2.782638&-0.449908\\

950: Full, $N=4$&2.782585&-0.449510\\

951: Full, $N=6$&2.782585&-0.449510\\

952: \\

953: 2DS secular, inviscid&0.636435&\\

954: 2DO secular, inviscid&1.193322&\\

955: 3D secular, inviscid&2.890299&\\

956: 3D secular, viscous&2.829523&-0.448547\\

957: \hline

958: \end{tabular}

959: \label{t:ot}

960: \end{table}

961: 

962: The viscous damping rate of the modes is considerable.  Although the

963: dominant motion is horizontal and independent of $z$, and therefore

964: does not incur any viscous forces in our approximation, the

965: accompanying vertical motion is damped.  To excite eccentric modes in

966: a three-dimensional disc, this damping must be overcome.  Viscous

967: overstability may be able to do this, but detailed calculations are

968: required and there is uncertainty in the applicability of a

969: Navier--Stokes viscosity to turbulent stresses in the disc.  A

970: simple estimate can be made as follows.  In a two-dimensional,

971: isothermal, Keplerian shearing sheet with constant kinematic shear

972: viscosity $\nu$ and no bulk viscosity, the maximum local growth rate

973: of the overstability is $\approx0.034\,\alpha\Omega$ and occurs for a

974: radial wavelength $\approx13\,H$ \citep{2006MNRAS.372.1829L}.

975: Although our eigenfunctions do not have an obviously wavelike form,

976: this suggests that overstability may be able to compensate for the

977: damping rate found in Table~\ref{t:ot}, which corresponds to only

978: $\approx0.0018\,\alpha\Omega$ for the parameters adopted there.

979: 

980: A sufficiently large viscosity is required to couple different layers

981: in the disc effectively.  If the viscosity is reduced to

982: $\alpha=0.01$, with other parameters as in Table~\ref{t:ot}, a

983: frequency eigenvalue of $\tilde\omega=2.7831-0.1793\,\rmi$ is

984: obtained, and a slightly larger value of $N$ is required to obtain the

985: same convergence.  While the precession rate agrees well with the case

986: of $\alpha=0.1$, the damping rate is now larger than predicted by the

987: viscous secular 3D~theory.  This happens because of the increasing

988: $z$-dependence of the horizontal motion in the absence of a strong

989: coupling between layers; although $u_2$ is still much smaller than

990: $u_0$ for $\alpha=0.01$, the ratio $u_2/u_0$ is several times larger

991: than in the case $\alpha=0.1$.  The viscous damping of this shearing

992: motion is more important, for $\alpha=0.01$, than that of the vertical

993: motion considered in the viscous secular theory.

994: 

995: If $\alpha$ is reduced further, the precession rate starts to

996: deviate from the 3D~secular theory and more vertical structure

997: develops in the eigenfunction, requiring a larger value of $N$ for

998: convergence.  Similar behaviour was found by

999: \citet{2006MNRAS.372.1829L}.  The behaviour of a three-dimensional

1000: inviscid eccentric disc could be very difficult to describe.

1001: 

1002: We also note that the stresses associated with tangled

1003: magnetic fields in a disc in which the magnetorotational instability

1004: occurs may provide an elastic, or viscoelastic, coupling between

1005: different layers \citep{2001MNRAS.325..231O}.  The associated damping

1006: rate may be smaller than that estimated on the basis of a

1007: Navier--Stokes viscosity.

1008: 

1009: \section{Conclusions}

1010: \label{s:conc}

1011: 

1012: In this paper we have examined the linear dynamics of one-armed

1013: oscillation modes in the circumstellar discs of Be~stars.  A

1014: three-dimensional effect, first identified by

1015: \citet{2001MNRAS.325..231O} but neglected in previous treatments of

1016: Be~stars, makes a crucial positive contribution to the precession

1017: rates of such modes.  It allows confined prograde modes to be obtained

1018: for all reasonable disc temperatures and stellar rotation rates.  This

1019: property allows us to give a unified description of eccentric discs of

1020: Be~stars of all stellar types without introducing uncertain radiative

1021: forces or modifying the inner boundary condition.  While we do not

1022: deny that in some cases radiative forces might be important, or that

1023: the inner boundary condition might differ from a rigid one, we have

1024: shown that these innovations are unnecessary.  We obtained these

1025: results using a secular theory of eccentric discs and confirmed them

1026: using a spectral treatment of the full linearized equations for

1027: three-dimensional isothermal discs including viscous terms that couple

1028: the horizontal motions at different altitudes.  In order to

1029: make these modes grow, viscous damping must be overcome by an

1030: excitation mechanism such as viscous overstability, which will be

1031: investigated in a subsequent paper.

1032: 

1033: The three-dimensional dynamics that we have described may also

1034: have important consequences for the behaviour of eccentric discs in

1035: other circumstances.  For example, in cataclysmic variable stars

1036: exhibiting superhumps, the relation between the precession rate of the

1037: disc and the binary mass ratio is an important observational property

1038: that is not fully explained by current theoretical models

1039: \citep{2006MNRAS.368.1123G,2007MNRAS.378..785S}.

1040: 

1041: The physical model adopted in this paper is idealized.  To explain in

1042: detail the observed cyclical behaviour of the emission lines of

1043: Be~stars within this theoretical framework is likely to require a

1044: treatment of non-isothermal and non-linear effects as well as a better

1045: understanding of the time-dependent behaviour of the density

1046: distribution of the circumstellar disc.  In addition, future work

1047: should attempt to identify and assess mechanisms, such as viscous

1048: overstability, by which one-armed oscillation modes may be excited in

1049: circumstellar discs.  Nevertheless, the effect investigated in this

1050: paper enhances the credibility of the one-armed oscillation model by

1051: providing a natural explanation of confined prograde modes.

1052: 

1053: \section*{acknowledgments}

1054: 

1055: I thank Atsuo Okazaki, John Papaloizou and an anonymous

1056: referee for useful comments.

1057: 

1058: \appendix

1059: 

1060: \section{Derivation of the secular theory}

1061: \label{s:appendix}

1062: 

1063: We consider equations~(\ref{un})--(\ref{hn}) together with the viscous

1064: effects described in Section~\ref{s:viscous}.  When $v_n$ and $w_n$

1065: are eliminated, we have

1066: \begin{eqnarray}

1067:   \lefteqn{\f{(\hat\omega_{\mathrm{h}n}^2-\kappa^2)}{\rmi\hat\omega_{\mathrm{h}n}}u_n=-\p_rh_n+\f{2m\Omega}{r\hat\omega_{\mathrm{h}n}}h_n}&\nonumber\\

1068:   &&+(\p_r\ln H)[nh_n+(n+1)(n+2)h_{n+2}],

1069: \label{unelim}

1070: \end{eqnarray}

1071: \begin{eqnarray}

1072:   \lefteqn{\left(-\rmi\hat\omega-\f{n\Omega_z^2}{\rmi\hat\omega_{\mathrm{v}n}}-\f{m^2c_\rms^2}{r^2}\f{1}{\rmi\hat\omega_{\mathrm{h}n}}\right)\f{h_n}{c_\rms^2}+\f{m\kappa^2}{2r\Omega\hat\omega_{\mathrm{h}n}}u_n}&\nonumber\\

1073:   &&+\f{1}{r\Sigma}\p_r(r\Sigma u_n)+(\p_r\ln H)(nu_n+u_{n-2})=0,

1074: \label{hnelim}

1075: \end{eqnarray}

1076: where $\hat\omega_{\mathrm{h}n}=\hat\omega+\rmi\alpha n\Omega_z$ and

1077: $\hat\omega_{\mathrm{v}n}=\hat\omega+\rmi(\alpha_\rmb+{\textstyle\f{4}{3}}\alpha)(n-1)\Omega_z$

1078: are the Doppler-shifted wave frequency modified to allow for the

1079: viscous damping of horizontal and vertical motions, respectively.  We

1080: are interested in the case $m=1$.

1081: 

1082: This system of equations is not closed because the $u_0$ equation

1083: refers to $h_2$, which depends on $u_2$.  In turn, the $u_2$ equation

1084: refers to $h_4$, which depends on $u_4$, and so on.  However, suppose

1085: for the time being that $u_2$ can be neglected in the $h_2$ equation

1086: (i.e.\ equation~\ref{hnelim} with $n=2$).  Then we have a

1087: closed system

1088: \begin{equation}

1089:   \f{(\hat\omega^2-\kappa^2)}{\rmi\hat\omega}u_0=-\p_rh_0+\f{2\Omega}{r\hat\omega}h_0+2(\p_r\ln H)h_2,

1090: \label{closed1}

1091: \end{equation}

1092: \begin{equation}

1093:   \left(-\rmi\hat\omega-\f{c_\rms^2}{r^2}\f{1}{\rmi\hat\omega}\right)\f{h_0}{c_\rms^2}+\f{\kappa^2}{2r\Omega\hat\omega}u_0+\f{1}{r\Sigma}\p_r(r\Sigma u_0)=0,

1094: \label{closed2}

1095: \end{equation}

1096: \begin{equation}

1097:   \left(-\rmi\hat\omega-\f{2\Omega_z^2}{\rmi\hat\omega_{\mathrm{v}2}}-\f{c_\rms^2}{r^2}\f{1}{\rmi\hat\omega_{\mathrm{h}2}}\right)\f{h_2}{c_\rms^2}+(\p_r\ln H)u_0\approx0.

1098: \label{closed3}

1099: \end{equation}

1100: These equations agree with a two-dimensional theory except for the

1101: additional $h_2$ term in equation~(\ref{closed1}), which can be

1102: related to $u_0$ through the approximate equation~(\ref{closed3}).

1103: 

1104: In the secular approximation we neglect the differences between

1105: $\Omega$, $\kappa$, $\Omega_z$ and $\Omega_\rmK$ except where

1106: essential, so that $\Omega\propto r^{-3/2}$ and $H\propto r^{3/2}$,

1107: and we assume a low frequency $|\omega|\ll\Omega$.  The leading

1108: approximations to the above equations in the inviscid case are then

1109: \begin{equation}

1110:   2\rmi\left(\f{\Omega^2-\kappa^2}{2\Omega}-\omega\right)u_0=-\p_rh_0-\f{2h_0}{r}+\f{3h_2}{r},

1111: \end{equation}

1112: \begin{equation}

1113:   \rmi\Omega\f{h_0}{c_\rms^2}-\f{u_0}{2r}+\f{1}{r\Sigma}\p_r(r\Sigma u_0)=0,

1114: \end{equation}

1115: \begin{equation}

1116:   -\rmi\Omega\f{h_2}{c_\rms^2}+\f{3u_0}{2r}=0.

1117: \label{h2}

1118: \end{equation}

1119: These combine to give equation~(\ref{secular}) for $u=u_0$, with the

1120: three-dimensional expression~(\ref{f3d}) for $f$.  If $h_2$ is

1121: neglected altogether, equation~(\ref{secular}) is obtained, but with

1122: the two-dimensional expression~(\ref{f2d}) for $f$.  It can be

1123: seen from the above equations that $h_2$ always makes a positive

1124: contribution to the precession rate $\omega$.

1125: 

1126: Neglecting $u_2$ compared to $u_0$ is equivalent to assuming that the

1127: eccentricity is independent of $z$, since $\he_0(\zeta)=1$ and

1128: $\he_2(\zeta)=\zeta^2-1$.  Under what conditions is this assumption

1129: reasonable?  Rough estimates based on equations~(\ref{unelim})

1130: and~(\ref{hnelim}) show that $u_2$ can be neglected in the $h_2$

1131: equation if $\alpha\Omega$ is much larger than the precession

1132: frequencies $\omega$ or $f$, meaning that the shear viscosity prevents

1133: the significant development of a $z$-dependent horizontal motion.

1134: This condition is readily satisfied in the circumstellar discs of

1135: Be~stars for reasonable values of $\alpha$.  Ultimately, however, the

1136: validation of this approximation comes from the numerical solution of

1137: the full system of linearized equations.

1138: 

1139: If viscosity is retained in this analysis, within the secular

1140: approximation, then the coefficient of $h_2$ in equation~(\ref{h2}) is

1141: multiplied by $(1+\rmi\beta)/(1-\rmi\beta)$, where

1142: $\beta=\alpha_\rmb+{\textstyle\f{4}{3}}\alpha$.  In this case,

1143: equation~(\ref{secular}) is again obtained, but in

1144: expression~(\ref{f3d}) for $f$ the coefficient $9/4$ of the

1145: three-dimensional term is multiplied by $(1-\rmi\beta)/(1+\rmi\beta)$.

1146: In this case the mode decays because of the viscous damping of the

1147: vertical motion.  For sufficiently small $\alpha$, however,

1148: the decay rate is enhanced by the viscous damping of the $z$-dependent

1149: horizontal motion.

1150: 

1151: \begin{thebibliography}{}

1152: \bibitem[Abramowitz \& Stegun(1965)]{1965hmfw.book.....A}

1153: Abramowitz, M.\ \& Stegun, I.~A., 1965, Handbook of Mathematical Functions, Dover, New York

1154: \bibitem[Fi{\v r}t \& Harmanec(2006)]{2006A&A...447..277F}

1155: Fi{\v r}t, R.\ \& Harmanec, P., 2006, A\&A, 447, 277 

1156: \bibitem[Goodchild \& Ogilvie(2006)]{2006MNRAS.368.1123G}

1157: Goodchild, S.\ \& Ogilvie, G., 2006, MNRAS, 368, 1123 

1158: \bibitem[Kato(1978)]{1978MNRAS.185..629K}

1159: Kato, S., 1978, MNRAS, 185, 629

1160: \bibitem[Kato(1983)]{1983PASJ...35..249K}

1161: Kato, S., 1983, PASJ, 35, 249

1162: \bibitem[Latter \& Ogilvie(2006)]{2006MNRAS.372.1829L}

1163: Latter, H.~N.\ \& Ogilvie, G.~I., 2006, MNRAS, 372, 1829

1164: \bibitem[Lee et al.(1991)]{1991MNRAS.250..432L}

1165: Lee, U., Osaki, Y.\ \& Saio, H., 1991, MNRAS, 250, 432 

1166: \bibitem[Negueruela et al.(2001)]{2001A&A...369..117N}

1167: Negueruela, I., Okazaki, A.~T., Fabregat, J., Coe, M.~J., Munari, U.\ \& Tomov, T., 2001, A\&A, 369, 117

1168: \bibitem[Ogilvie(2001)]{2001MNRAS.325..231O}

1169: Ogilvie, G.~I., 2001, MNRAS, 325, 231 

1170: \bibitem[Okazaki(1991)]{1991PASJ...43...75O}

1171: Okazaki, A.~T., 1991, PASJ, 43, 75

1172: \bibitem[Okazaki(1997)]{1997A&A...318..548O}

1173: Okazaki, A.~T., 1997, A\&A, 318, 548 

1174: \bibitem[Okazaki(2007)]{2007ASPC..361..230O}

1175: Okazaki, A.~T., 2007, in Active OB-Stars: Laboratories for Stellar and Circumstellar Physics, eds S.~Stefl, S.~P.~Owocki \& A.~T.~Okazaki, ASP Conf.\ Ser., 361, 230

1176: \bibitem[Okazaki et al.(1987)]{1987PASJ...39..457O}

1177: Okazaki, A.~T., Kato, S.\ \& Fukue, J., 1987, PASJ, 39, 457 

1178: \bibitem[Papaloizou(2002)]{2002A&A...388..615P}

1179: Papaloizou, J.~C.~B., 2002, A\&A, 388, 615 

1180: \bibitem[Papaloizou \& Savonije(2006)]{2006A&A...456.1097P}

1181: Papaloizou, J.~C.~B.\ \& Savonije, G.~J., 2006, A\&A, 456, 1097

1182: \bibitem[Papaloizou et al.(1992)]{1992A&A...265L..45P}

1183: Papaloizou, J.~C., Savonije, G.~J.\ \& Henrichs, H.~F., 1992, A\&A, 265, L45 

1184: \bibitem[Porter \& Rivinius(2003)]{2003PASP..115.1153P}

1185: Porter, J.~M.\ \& Rivinius, T., 2003, PASP, 115, 1153 

1186: \bibitem[Savonije \& Heemskerk(1993)]{1993A&A...276..409S}

1187: Savonije, G.~J.\ \& Heemskerk, M.~H.~M., 1993, A\&A, 276, 409 

1188: \bibitem[Smith et al.(2007)]{2007MNRAS.378..785S}

1189: Smith, A.~J., Haswell, C.~A., Murray, J.~R., Truss, M.~R.\ \& Foulkes, S.~B.\ 2007, MNRAS, 378, 785

1190: \bibitem[Tanaka et al.(2002)]{2002ApJ...565.1257T}

1191: Tanaka, H., Takeuchi, T.\ \& Ward, W.~R., 2002, ApJ, 565, 1257 

1192: \bibitem[Telting et al.(1994)]{1994A&A...288..558T}

1193: Telting, J.~H., Heemskerk, M.~H.~M., Henrichs, H.~F.\ \& Savonije, G.~J., 1994, A\&A, 288, 558 

1194: \bibitem[Tremaine(2001)]{2001AJ....121.1776T}

1195: Tremaine, S., 2001, AJ, 121, 1776 

1196: \bibitem[Zhang \& Lai(2006)]{2006MNRAS.368..917Z}

1197: Zhang, H.\ \& Lai, D., 2006, MNRAS, 368, 917

1198: \end{thebibliography}

1199: 

1200: \label{lastpage}

1201: 

1202: \end{document}

1203: