1: \documentclass{aipproc}

2: 

3: %\usepackage{epsf}

4: \usepackage{graphicx}

5: 

6: \layoutstyle{6x9}

7: 

8: \bibliographystyle{aipproc}

9: 

10: \begin{document}

11: 

12: \title{Simulating the scalar glueball on the lattice}

13: \author{Colin Morningstar}{address={

14: Department of Physics, Carnegie Mellon University,

15: Pittsburgh, PA 15213-3890}}

16: \author{Mike Peardon}{address={School of Mathematics, Trinity College, Dublin

17: 2, Ireland.}}

18: 

19: \begin{abstract}

20: Techniques for efficient computation of the scalar glueball mass on the lattice

21: are described. Directions and physics goals of proposed future calculations

22: will be outlined. 

23: \end{abstract}

24: \maketitle 

25: 

26: \section{Introduction}

27: 

28: Along with the mesons of the quark model, QCD allows for the existence of

29: bound-states of gluons, the glueballs. 

30: The scalar glueball is the lightest state in the spectrum of strongly

31: interacting gluons (described by the $SU(3)$ Yang-Mills theory) and in this

32: consistent, confining quantum field theory, it is stable. The world of gluons

33: alone 

34: thus provides an attractive starting point for investigations of glueballs.

35: Confinement is not described at all within QCD perturbation theory and so

36: computing properties of the glueballs requires a non-perturbative technique for

37: studying strong interactions.  

38: As a result, no

39: unambiguous identification of the scalar glueball has been made, although 

40: there is broad agreement that an extra state in the family of scalar 

41: resonances, unaccounted for by the quark model, exists between 1.5 and 

42: 1.8 GeV. 

43: In QCD, the glueball states will mix with the isoscalar resonances of the 

44: quark model, and will also decay strongly so 

45: to claim a complete understanding of the scalar mesons below 2 GeV 

46: therefore requires reliable knowledge of the masses and other properties of 

47: glueballs. 

48: 

49: In order to disentangle the complex picture presented by experiments much

50: more detailed and precise theoretical prediction from QCD must be made. At 

51: present, the only practical {\it ab initio} method for computing the 

52: non-perturbative properties of strongly interacting field theories such as QCD

53: is through Monte Carlo simulation of the lattice regularisation of the theory.

54: The glueballs of $SU(3)$ Yang-Mills theory have been studied

55: extensively in lattice simulations \cite{Michael:1989jr, Bali:1993fb, 

56: Morningstar:1999rf}, and this spectrum is becoming increasingly

57: accurately determined. This is the first step in understanding the

58: appearance of glueballs among the light scalar resonances of QCD. In spite of 

59: the limitation of these calculations, their output is proving to be very 

60: valuable in the construction and validation of phenomenological models.

61: 

62: In this article, we describe the state-of-the-art calculation of the

63: glueball spectrum of the Yang-Mills theory. Investigations of further 

64: properties of these fascinating states, including their behaviour in QCD (with 

65: the dynamics of quark fields included) are then outlined.

66: 

67: \section{Lattice QCD}

68: 

69: Lattice QCD solves two key problems arising from strongly interacting gauge

70: theories at a stroke; it provides both a non-perturbative regularisation of QCD

71: and a means by which observables in the theory can be predicted by computer.

72: This inherently gauge invariant regularisation is manifest through 

73: a direct cut-off of momenta at the Brillouin zone boundaries of the lattice. 

74: 

75: To begin, the path-integral formulation of the theory in Euclidean space is

76: taken. The four dimensions of space and time are discretised with a regular

77: lattice of points separated by a spacing, $a$. In the gauge-invariant 

78: formulation defined by Wilson nearly 30 years ago \cite{Wilson:1974sk}, the quark fields of QCD are defined

79: at sites of the lattice, while the gluonic degrees of freedom are represented

80: by parallel transporters (usually in the fundamental representation of the

81: gauge group) connecting nearest-neighbouring sites. A definition of the

82: Yang-Mills action is constructed from the trace of path-ordered products of the

83: link variables around small loops on the lattice. The smallest non-trivial loop

84: on the lattice circumnavigates a square of side-length $a$, usually called a

85: plaquette.  Similarly, a number of different definitions of the quark bilinear 

86: action, coupled to the gluons, can be made. 

87: 

88: Path integral expectation values can then be estimated by numerical methods.

89: If the theory is considered for space and time inside a finite-sized box, the

90: path integral at non-zero lattice spacing is reduced to a very-high-dimensional

91: ordinary integral. This type of problem can normally only be addressed by

92: Monte Carlo methods, where a stochastic sampling of points inside the phase

93: space to be integrated over is made.

94: 

95: \subsection{Stochastic estimation methods for quantum field theory}

96: 

97: After Wick rotation, the path integral for field theories such as QCD (at least 

98: at zero chemical potential) can be regarded as a statistical mechanical

99: partition function. The Boltzmann weight for a particular field configuration

100: is real and positive definite, and so can be interpreted as a statistical 

101: probability of the configuration appearing. In these circumstances, the 

102: natural Monte Carlo method to employ is importance sampling. An update 

103: algorithm is defined which generates configurations of the gluon fields, $U$ 

104: with probability density given by the Boltzmann weight, $\exp\{ -S(U)\}$. The 

105: method is employed on the computer to generate an ensemble of gluon field 

106: configurations. Expectation values are then estimated from ensemble averages. 

107: 

108: To compute the physical properties of the theory appropriate observation

109: functions on the underlying fields are defined with the required quantum 

110: numbers, and these observables are measured on all members of the stochastic 

111: ensemble.  Careful analysis allows a reliable determination of statistical 

112: uncertainties.

113: 

114: \subsection{Controlling simulation artefacts}

115: 

116: In order to make contact with the continuum field theory, a number of artefacts 

117: of the method must be controlled. The most obvious of these is the need to use

118: a finite-sized box for numerical work. In practical simulations, boxes with

119: side lengths of the order of 2 fm are feasible. 

120: 

121: In many cases, the qualitative behaviour of states in increasingly 

122: large-but-finite volumes can be predicted. Simulations are run for a range of 

123: box sizes, and the data matched to this predicted form to allow safe 

124: extrapolation to the infinite volume limit. In most studies of glueballs these 

125: effects are now extremely small and do not constitute the dominant systematic 

126: error. 

127: 

128: More difficult is the need to extrapolate data to the limit of zero lattice

129: spacing (usually called the continuum limit). Again, the most natural procedure

130: is to compute the physics of interest on a set of lattices with different grid

131: spacings, and extrapolate. In many cases, the expected leading-order behaviour

132: of these finite lattice spacing effects can be predicted, and this can be used 

133: to control data fitting. The problem with this direct approach is that the cost 

134: of computer simulations rises rapidly as the lattice spacing is diminished, at

135: least in proportion to $a^{-4}$ and generally worse than this. Since the

136: continuum limit exists at a critical point corresponding to a second order

137: phase transition, where the correlation length of the system diverges in units

138: of the lattice spacing, critical slowing down makes this cost higher. It is

139: clear that the cheapest computer simulations are those run on the coarsest

140: grids, which are furthest from the continuum. 

141: 

142: \subsection{Symanzik improvement}

143: 

144: The effects of a finite grid spacing can be understood in the language of

145: quantum field theory as arising from irrelevant operators appearing in the

146: lattice action. These operators are of higher dimension than the relevant 

147: operator of the continuum theory, Tr $F_{\mu\nu} F_{\mu\nu}$ and are 

148: multiplied in the action by dimensionful couplings proportional to (positive)

149: powers of the lattice spacing. For the simplest description of the gauge

150: action, proposed by Wilson and consisting of the trace of the plaquette, the

151: leading irrelevant operator appears at ${\cal O}(a^2)$. Odd powers are

152: prohibited by exact parity symmetry, which is preserved by the discretisation. 

153: 

154: Universality suggests that there is a broad class of lattice

155: representations of the continuum action, whose members all have the identical

156: continuum limit: QCD. This implies that different lattice actions can be 

157: engineered which have smaller contributions from higher-dimensional operators. 

158: For asymptotically free theories, like QCD and the $SU(3)$ Yang-Mills theory of

159: gluons, these discretisations of the action can be designed within the 

160: framework of perturbation theory. This concept is termed Symanzik improvement

161: \cite{Symanzik:1983dc}. 

162: 

163: The idea has been widely exploited in lattice calculations, both in

164: discretisations of gluon fields as well as the quark fields. For the gluon

165: action, taking appropriate linear combinations of traces of larger closed

166: loops, such as the $2\times 1$ rectangle means terms at ${\cal O}(a^2)$ can be

167: eliminated \cite{Luscher:1985zq, Lepage:1996ph}. This allows lattice 

168: investigations to be carried out with grid spacings as coarse as 

169: 0.25 fm, while remaining sufficiently close to the continuum limit to 

170: explore the physics reliably. 

171: 

172: \subsection{The anisotropic lattice}

173: 

174: At the point in the calculation where measurements are made, these coarse 

175: lattices present new disadvantages. The mass of a state is extracted from the 

176: decay of a two-point correlation function in Euclidean space

177: \begin{equation}

178:     C(t) = \langle \Phi(t) \Phi(0) \rangle \propto e^{-M t}

179: \end{equation}

180: where $\Phi(t)$ is a lattice operator that creates or annihiliates the state 

181: of mass $M$ at time $t$. This correlator falls exponentially as the

182: two operators are moved further apart, and for a heavy state (such as the

183: glueballs) the fall-off is rapid. Also, since the correlation function is being

184: estimated in a Monte Carlo simulation, the statistical variance of the

185: operators determines the accuracy to which these correlations can be measured.

186: For the glueballs, these operators are the traces of closed Wilson loops on the

187: underlying gauge fields and have rather large vacuum fluctuations, so a

188: critical signal-to-noise problem arises.  Since the operators can only be

189: measured on points of the grid, little information can be extracted from coarse

190: lattice measurements before the signal is lost in the noisy vacuum.

191: 

192: A solution that combines the economy of the coarse lattice with the good

193: resolution of a fine grid is to use an anisotropic lattice, where the temporal

194: lattice spacing is made small, while the three spatial dimensions are

195: discretised more coarsely. Note that time-like correlation functions are used

196: in extracting the masses of states. This method is particularly efficient for

197: glueball simulations \cite{Morningstar:1997ff}. The two natural length scales 

198: in the problem, the mass and size of the glueball set the optimal temporal and

199: spatial lattice spacings for numerical investigation.

200: 

201: Fig.~\ref{fig:meff} describes the output from just such an investigation. The 

202: figure shows the effective mass,

203: \begin{equation}

204:    a_t M_{\rm eff} = \frac{\ln\; C(t)}{\ln\; C(t+1)}

205: \end{equation}

206: for the scalar glueball measured on a highly anisotropic lattice, where the

207: ratio of scales is $a_s/a_t=6$. The operator, $\Phi$ is constructed from a

208: large basis set of the trace of closed Wilson loops built from smoothed link

209: variables. A single operator for each state is made by taking linear

210: combinations from the basis set of operators, with this combination chosen to

211: optimise the overlap with the state of interest. The graph shows data for the 

212: ground-state scalar and the first excited state with the same quantum numbers.

213: It illustrates the method is capable of precise determinations of the energies 

214: of both these levels. In the simulation presented in Fig.~\ref{fig:meff} these

215: energies are measured to $1\%$ statistical precision.

216: \begin{center}

217: \begin{figure}

218: %\setlength{\epsfxsize}{12cm}\epsfbox{scalar_meff.eps}

219: \includegraphics[width=12cm]{scalar_meff}

220: \caption{The effective mass plot of the scalar glueball channel, including data

221: for the first excited state.\label{fig:meff}}

222: \end{figure}

223: \end{center}

224: 

225: In Fig.~\ref{fig:dip}, the dependence on the lattice cut-off of the scalar

226: glueball mass in units of a physical scale $r_0$, derived from the static

227: potential \cite{Sommer:1994ce}, is presented. The data points

228: represented with crosses are from a range of simulations performed using

229: the Wilson plaquette action \cite{Michael:1989jr, Bali:1993fb}. The circles 

230: are from simulations with an 

231: anisotropic lattice action, improved according to the Symanzik programme with 

232: parameters determined from tree-level perturbation theory after the tadpole 

233: graphs in the lattice weak-coupling expansion have been resummed 

234: \cite{Morningstar:1997ff, Morningstar:1999rf}. The quality 

235: of a lattice action should be judged by how strongly the mass (in physical 

236: units) depends on the cut-off; ``better'' actions should show a weaker 

237: dependence on the lattice spacing. By this assessment, the Symanzik improved 

238: action is superior to the simplest action, however a strong dependence on the 

239: cut-off still remains. The scalar glueball mass (in units of $r_0^{-1}$) falls 

240: as the lattice spacing is increased to reach a minimum from which it begins to 

241: rise again.  We term this effect the ``scalar dip.''

242: 

243: \subsection{Curing the ``scalar dip''}

244: 

245: The lattice theory is not necessarily QCD; the physics of this theory is only 

246: recovered once the discretised version is ``close'' to the critical point on 

247: which QCD exists. If the space of lattice theories contains other critical 

248: points, simulations performed near to these points will be probing the 

249: properties of other continuum quantum field theories. Remarkably, this 

250: apparently abstract problem in non-perturbative renormalisation seems to arise 

251: in the simulation of $SU(3)$ Yang-Mills. 

252: 

253: It has been recognised for some time that there is a line of ``bulk'' 

254: first-order phase transitions in the two-dimensional plane of lattice theories 

255: in which both a coupling to the fundamental and adjoint representations of the 

256: link variables is made \cite{Heller:1995bz}. This line ends in a critical 

257: point at which some

258: unknown continuum theory resides. If the physics of the lattice theory close

259: to this critical point is investigated, it will be predominantly governed by

260: the properties of this other theory. One suggestion is that the continuum

261: theory at this critical point is a free scalar one, and if this was the case,

262: the mass of the scalar particle in the lattice theory would be artificially

263: light in comparison to the higher spin states. Precisely this artefact is

264: observed. 

265: 

266: The data represented by squares in Fig.~\ref{fig:dip} are from a new

267: discretisation of the gluon action on an anisotropic lattice. In this

268: prescription, the lattice action includes terms that trace over the 

269: link variables in the adjoint representation, as well as the fundamental. 

270: The links stored on the computer are in the fundamental representation, but the

271: trace in the adjoint can be computed using the identity

272: \begin{equation}

273:   \mbox{Tr }A(g) = \mbox{Tr }U(g) \;\mbox{Tr }U^\dagger(g) - 1,

274: \end{equation}

275: with $A(g)$ the adjoint representation of an element, $g \in SU(3)$ and $U(g)$

276: its corresponding fundamental representation. 

277: \begin{center}

278: \begin{figure}

279: % \setlength{\epsfxsize}{12cm}\epsfbox{scalar_dip.eps}

280: \includegraphics[width=12cm]{scalar_dip}

281: \caption{Extrapolating simulation data at finite lattice spacing to the

282: continuum limit.\label{fig:dip}}

283: \end{figure}

284: \end{center}

285: 

286: These simulation results are extremely encouraging and clearly show very weak 

287: lattice spacing dependence out to coarse spatial lattices ($a\approx 0.25$ fm).

288: A very reliable extrabolation to the continuum limit can be made.

289: 

290: \subsection{Spin on the lattice}

291: 

292: In a scattering experiment, the spin of resonances is determined from a

293: partial wave analysis. Revealing the spin of states in a lattice

294: calculation requires similar care.

295: Putting quarks and gluons onto a lattice breaks the continuum rotation group 

296: $SO(3)$ to the discrete cubic point group, $O_h$. As a result, states are no

297: longer classified according to a spin quantum number which describes the

298: irreducible representation (irrep) of $SO(3)$ they transform under.

299: They have instead one of five labels corresponding to the five irreps of $O_h$:

300: $A_1,A_2,E,T_1$ and $T_2$. 

301: Table~\ref{tab:subduce}

302: describes how the representation of $O_h$ subduced from spin irreps of 

303: $SO(3)$ subsequently decompose into irreps of $O_h$.

304: \begin{table}[h]

305: \begin{tabular}{c|ccccc}

306:   J & $A_1$ & $A_2$ & $E$ & $T_1$ & $T_2$ \\

307:   \hline

308:   0 &   1   &       &     &       &       \\

309:   1 &       &       &     &   1   &       \\

310:   2 &       &       &  1  &       &  1    \\

311:   3 &       &   1   &     &   1   &  1    \\

312:   4 &   1   &       &  1  &   1   &  1    \\

313:   \hline

314: \end{tabular}

315: \caption{The irreducible content of the representations of $O_h$ subduced from

316: the spin irreps of $SO(3)$}\label{tab:subduce}

317: \end{table}

318: To identify the spin of the state in the continuum,

319: degeneracies across lattice states must be found and compared to the table. For

320: example, tagging a spin 2 state on the lattice requires finding degenerate

321: states in the $E$ and $T_2$ channels (and no others) in the continuum limit. 

322: For the

323: scalar state, this procedure is quite straightforward; a state in the trivial

324: representation, $A_1$, which is not degenerate with any other state in the

325: spectrum can be uniquely identified with a continuum $J=0$ scalar. This spin

326: analysis was carried out in Ref.~\cite{Morningstar:1999rf} for the glueball 

327: spectrum, and the spins of many continuum states were identified. 

328: 

329: \section{The $SU(3)$ Yang-Mills spectrum}

330: 

331: Following the procedures outlined in the previous sections, the simulations of 

332: the Yang-Mills theory of strongly interacting gluons (without quarks) have 

333: reached high levels of precision. Fig.~\ref{fig:spectrum} shows the lightest

334: six states in this spectrum of gluon bound-states. Note that both the 

335: ground-state and first-excited state in the scalar channel has been determined 

336: more reliably after removing the artefacts of the scalar dip and performing a

337: continuum extrapolation. The masses of the scalar glueballs are consistent

338: within errors with earlier determinations. 

339: \begin{center}

340: \begin{figure}[h!]

341: %\setlength{\epsfxsize}{12cm}\epsfbox{spectrum_2.eps}

342: \includegraphics[width=12cm]{spectrum_2}

343: \caption{The lightest six states in the spectrum of the $SU(3)$ Yang-Mills

344: theory.\label{fig:spectrum}}

345: \end{figure}

346: \end{center}

347: 

348: \section{Current and future directions}

349: 

350: While the simulation of the Yang-Mills theory is now well understood, the

351: inclusion of quantum fluctuations of quark fields is significantly more

352: difficult. Since the Grassmann algebra of fermions in a path integral can not 

353: be handled directly by computers, the quark bilinear in the QCD action must be 

354: integrated out analytically, leaving a non-local effective action in the 

355: gluons alone. The non-local nature of these interactions means the simpler 

356: Monte Carlo algorithms used to generate ensembles of gauge field configurations 

357: for the Yang-Mills theories become inefficient. More sophisticated techniques 

358: are used, but they add a large numerical overhead; QCD simulations including 

359: quark dynamics are a factor of 100-1000 times more expensive than those of the 

360: Yang-Mills theory. 

361: 

362: This issue of including the quark field dynamics remains a topic of a good deal

363: of research in the lattice community and optimal simulation strategies are

364: still actively under investigation. It seems likely that within the near

365: future, the physics of glueballs that can decay into and mix with quark mesons

366: will be investigated. Progress in this direction has begun by other lattice

367: groups \cite{Hart:2002sp} although lattice calculations involving unstable 

368: particles are in their infancy and remaind a challenging topic of research. 

369: Also, extending the anisotropic lattice technology, 

370: which proved so useful for scanning the pure gauge theory, to QCD simulations 

371: with dynamical quarks is under investigation.

372: 

373: \section{Conclusions}

374: 

375: Scalar mesons seem to constantly challenge theorists and phenomenologists by

376: turn. At this meeting we heard of the difficulties in finding a convincing

377: picture of the broad range experimental data and in this article, some of the 

378: quite distinct problems presented by the scalar glueball to lattice theorists 

379: have been outlined. Significant progress in this field is still being made and 

380: the Yang-Mills theory has now been mapped out. Full QCD dynamics, with the quark

381: fields playing their role, is a major challenge under investigation by

382: many members of the lattice community \cite{Lat03}. At the same time, more 

383: detailed properties of the gluonic theory are now being examined, including 

384: calculations of the chromoelectric and magnetic matrix elements 

385: \cite{Dong:Lat03} and the structure form factors. 

386: 

387: \bibliography{trinlat_bib/trinlat}

388: 

389: \end{document}

390: