1: \documentclass[prd,preprint,showpacs,nofootinbib,tightenlines,groupedaddress]{revtex4}
2: \usepackage{epsfig}
3: \usepackage{amssymb}
4: \usepackage{amsfonts}
5: \usepackage{subfigure}
6: \usepackage{float}
7: \usepackage{latexsym,hyperref}
8: \graphicspath{{./Figures/}}
9:
10: \begin{document}
11: \preprint{HU-EP-06/07}
12:
13: \title{Vortex free energy and deconfinement in center-blind discretizations
14: of Yang-Mills theories}
15:
16: \author{G.~Burgio\footnote{Address from September 1$^{\rm st}$ 2006:
17: Universit\"at T\"ubingen, Institut f\"ur Theoretische Physik.}}
18:
19: \author{M.~Fuhrmann}
20:
21: \author{W.~Kerler}
22:
23: \author{M.~M\"uller-Preussker}
24:
25: \affiliation{Humboldt-Universit\"at zu Berlin, Institut f\"ur Physik, Berlin,
26: Germany}
27:
28: \date{\today}%
29:
30: \begin{abstract}
31: Maximal 't~Hooft loops are studied in $SO(3)$ lattice gauge theory
32: at finite temperature $T$. Tunneling barriers among twist sectors
33: causing loss of ergodicity for local update algorithms are overcome
34: through parallel tempering, enabling us to measure the vortex free
35: energy $F$ and to identify a deconfinement transition at some
36: $\beta_A^{crit}$. The behavior of $F$ below $\beta_A^{crit}$ shows
37: however striking differences with what is expected from discretizations
38: in the fundamental representation.
39: \end{abstract}
40:
41: \maketitle
42:
43: Topology plays an important r\^ole in the non-perturbative dynamics of
44: Yang-Mills theories. In particular the vacuum
45: condensation of topological excitations might explain quark confinement and
46: the existence of a mass gap through various scenarios
47: described in the literature
48: \cite{'tHooft:1974qc,Polyakov:1977fu,'tHooft:1979uj}.
49: Some of these models allow to define
50: a topological order parameter for the finite $T$ deconfinement transition;
51: for 't~Hooft magnetic vortices, classified by the
52: first homotopy class $\mathbb{Z}_N$
53: of the continuum Yang-Mills gauge group $SU(N)/\mathbb{Z}_N$
54: \cite{'tHooft:1979uj}, the change in free
55: energy $F = \Delta U -T \Delta S$ for their creation might play such r\^ole
56: and has received broad attention, in particular in lattice discretizations
57: at zero and finite $T$
58: \cite{Kovacs:2000sy,deForcrand:2000fi,deForcrand:2001nd}.
59: The main problem in such non-perturbative regularizations
60: is that creating a vortex is equivalent to the introduction of a non-trivial
61: twist. For discretizations in the fundamental representation,
62: transforming under the {\it enlarged} gauge group $SU(N)$ ($N=2$ in this
63: paper), this cannot be implemented dynamically but only via a modification
64: of their boundary conditions (b.c.), the generalized partition function
65: $\tilde{Z}$ being
66: defined through the weighted sum of partition functions with fixed
67: twisted b.c. Since each of them must be determined by
68: independent simulations their relative weights can only be calculated through
69: indirect means \cite{deForcrand:2000fi,deForcrand:2001nd}.
70:
71: Universality arguments are often cited to claim that results in lattice
72: Yang-Mills theories will not depend on the discretization chosen. A
73: natural alternative in calculating $F$ would therefore be to directly
74: discretize the theory in the $SU(N)/{\mathbb{Z}}_N$ representation ($SO(3)$
75: in our case) with periodic b.c., naturally transforming under
76: the continuum Yang-Mills gauge group \cite{deForcrand:2002vs}. The adjoint
77: partition function $Z({\beta}_A)$ should in fact be equivalent to $\tilde{Z}$
78: provided the $SO(3)$ native constraint
79: $\,\sigma_c=\prod_{\overline{P}\in\partial c}\mathrm{sign}
80: (\mathrm{{Tr}_{F}}\,U_{\overline{P}})=1$ is satisfied for every elementary
81: 3-cube $c$, where $U_{\overline{P}}$ denotes the plaquettes belonging to its
82: surface $\partial c$
83: \cite{Mack:1978rq,deForcrand:2002vs}:
84: ${\mathbb{Z}}_2$ magnetic monopoles are suppressed and only closed
85: ${\mathbb{Z}}_2$ magnetic vortices winding around the boundaries are allowed,
86: i.e. in this limit adjoint actions dynamically allow all topological sectors
87: which in the fundamental case must be fixed through twisted b.c..
88: Moreover, since the standard spontaneous center symmetry breaking argument
89: for the
90: deconfinement transition does not apply to the center-blind adjoint
91: discretization the question whether $F$ behaves as an order parameter
92: is of major interest, also in light of recent studies for alternative
93: descriptions of confinement in centerless theories \cite{Holland:2003jy}.
94:
95: A practical obstacle one needs to overcome in investigating the adjoint
96: theory is the appearance of a bulk transition at some $\bar{\beta}_A$,
97: separating a strong coupling chaotic phase (I) continuously connected
98: with the fundamental action, where $\langle\sigma_c\rangle\simeq0$,
99: from a weak coupling ordered phase (II) where $\langle\sigma_c\rangle\simeq1$
100: \cite{Bhanot:1981eb,Greensite:1981hw,Halliday:1981te}. In phase II, where one
101: wishes to exploit the relation between $Z({\beta}_A)$ and
102: $\tilde{Z}$ mentioned above, high potential barriers separating
103: twist sectors suppress tunneling among them for local update
104: algorithms \cite{deForcrand:2002vs}.
105: On the other hand a well-known result is that a center blind
106: $\mathbb{Z}_2$ monopole suppression term in the action
107: $\lambda \sum_{c}(1-\sigma_{c})$ weakens the order of the bulk transition
108: while moving it down into the strong coupling
109: region (see the curved dashed line in
110: Fig.~\ref{phase}a) \cite{Halliday:1981te}.
111: For asymmetric lattice sizes $~N_{\tau} \times N_s^3, ~N_{\tau} \ll N_s~$
112: indications for a finite temperature critical line
113: $\beta^{crit}_A(\lambda, N_{\tau})$ within phase II
114: (horizontal dashed line in Fig.~\ref{phase}a
115: for $N_{\tau}=4$)
116: have already been found from simulations at {\it fixed} twist
117: \cite{Datta:1999np,Barresi:2003jq,Barresi:2004qa,Barresi:2006gq}.
118: %----------------------------------------------------------
119: \begin{figure}[thb]
120: \begin{center}
121: \subfigure[]{\includegraphics[angle=0,width=0.49\textwidth]{1a.eps}}
122: \subfigure[]{\includegraphics[angle=0,width=0.49\textwidth,%
123: height=0.36\textwidth]{1b.eps}}
124: \end{center}
125: \caption{(a) Phase structure in the $\beta_A$-$\lambda$ plane
126: with paths of couplings indicated as used in our simulations.
127: (b)
128: Adjoint Polyakov loop average
129: $\langle L_A \rangle$ versus $\beta_A$ at fixed $\lambda=0.8$ and
130: $N_{\tau}=4$.}
131: \label{phase}
132: \end{figure}
133: %----------------------------------------------------------
134:
135: In this paper we solve the problem of ergodic updates
136: through parallel tempering (PT) \cite{Hukushima:1995,Marinari:1996dh}
137: for the adjoint Wilson action with $\mathbb{Z}_2$ monopole
138: suppression
139: \begin{eqnarray}
140: S=\beta_{A} \sum_{P}
141: \left(1-\frac{1}{3}\mathrm{Tr}_A U_{P}\right)
142: +\lambda \sum_{c}(1-\sigma_{c})\,,
143: \label{ouraction}
144: \end{eqnarray}
145: where $U_P$ denotes the standard plaquette variable and
146: $\mathrm{Tr}_A O= (\mathrm{Tr}_F O)^2-1 = \mathrm{Tr}_F (O^2)+1$ the adjoint
147: trace. Our PT paths in the $\beta_A-\lambda$ plane
148: extend over the {\it second order}
149: leg of the bulk transition straight into the
150: inner region of phase II
151: as shown in Fig.~\ref{phase}a, with two paths at fixed $\beta_A$, one
152: path with fixed $\lambda$, and one path consisting of a fixed-$\beta_A$
153: and a fixed-$\lambda$ part. In this way we avoid the
154: necessity to cross high potential barriers.
155: This enables us to account for all twist sectors, restoring
156: ergodicity, and to study $F$ at finite $T$. As a qualitative indicator
157: of deconfinement we have also measured the adjoint
158: Polyakov loop $\langle L_A \rangle = \langle{
159: {\sum_{\vec{x}}\mathrm{Tr}_{A}L(\vec{x})}}\rangle/(3 N_s^3)$,
160: $L(\vec{x})= \prod_t U_4(\vec{x},t)$, which strictly speaking cannot
161: behave as an order parameter for the deconfinement transition,
162: delivering at most information on the screening length for the
163: effective adjoint potential (see Fig.~\ref{phase}b). A preliminary report
164: on the present investigation was presented in \cite{Burgio:2005xe};
165: further results on other observables
166: as well as a detailed description of the implementation of the algorithm
167: will appear in a separate publication \cite{Burgio:2006xj}. To our knowledge this is
168: the first successful attempt to study the deconfinement transition of the
169: $SO(3)$ lattice gauge theory via an ergodic simulation at large volume
170: as well as the first determination of $F$ in the confined
171: phase of a center-blind discretization of Yang-Mills theories.
172:
173: In PT \cite{Hukushima:1995,Marinari:1996dh} we update $K$
174: configurations ${\cal{F}}_i$, $i = 1, \dots, K$, with couplings
175: $(\lambda,\beta_A)_i$ swapping neighboring pairs %$i$, $i+1$
176: from smaller to larger values according to a Metropolis
177: acceptance probability, satisfying detailed balance, $P_{\rm swap}(i,j) =
178: \min[ 1, \mathrm{exp}(-\Delta S_{ij})]$, where
179: \begin{equation}
180: \Delta S_{ij} =
181: S[(\lambda,\beta_A)_i, {\cal{F}}_{i}] + S[(\lambda,\beta_A)_j, {\cal{F}}_{j}]
182: - S[(\lambda,\beta_A)_i, {\cal{F}}_{j}] - S[(\lambda,\beta_A)_j,
183: {\cal{F}}_{i}]\,.
184: \label{eq:Pswap-1}
185: \end{equation}
186: Compared to other methods \cite{Swendsen:1987ce,Berg:1991cf} the appeal of
187: PT is its easy implementation both at criticality and away from it, needing
188: no knowledge of re-weighting factors or other dynamical input. This is a
189: welcome property for us since we wish to go as far as possible
190: from the bulk transition into phase II. In view of
191: various experiences with simulated tempering one also expects PT to be more
192: efficient than multi-canonical simulations. For the success of the method in
193: the case under consideration the softening of the bulk transition to
194: 2$^{\rm nd}$ order is crucial, since we ``transport''
195: tunneling from phase I at lower $\lambda$ into phase II at larger $\lambda$,
196: where twist sectors are well defined but frozen. To work at low
197: $\lambda$, i.e. through a 1$^{\rm st}$ order bulk transition, would make
198: barriers too high and kill any hope of ergodicity at large volume,
199: as experienced in \cite{deForcrand:2002vs} for $N_s > 8$.
200:
201: Some care is of course necessary also with our method.
202: In particular to maintain a sufficient swapping acceptance rate $\omega$ the
203: distance between neighboring couplings must diminish with the volume.
204: On the other hand to keep cross-correlations
205: under control one does not wish the acceptance rate to be too high. We
206: have chosen to tune the parameters for each path and volume at hand
207: so to keep the acceptance rate roughly fixed at around $\omega = 12 \%$,
208: a value for which we empirically find a good balance between auto- (in the
209: sense of freezing of the sectors) and cross-correlations.
210: For the paths in Fig~\ref{phase}a contributing to the left and right branches
211: in Fig~\ref{free1}a details on the statistics, i.e. number of configurations
212: $N$ and number of ensembles $K$, are given in Table~\ref{tab}.
213: The paths at fixed $\beta_A = 0.4$
214: and 0.65 for $N_s = 16$ shown in Fig~\ref{free1}b were calculated with
215: $K=7$ and $N=30000$ configurations.
216: To remain on the safe side, in Fig~\ref{free1}a we have chosen not to quote
217: the ensembles related to the end points of the paths since, having no further
218: configurations to swap with, they might be affected by systematic errors.
219: We wish however to stress that this has never been observed in the PT
220: literature and we also have no indication that this might be the case.
221: Further details on the algorithm and a detailed analysis of correlations
222: will be reported in a forthcoming paper \cite{Burgio:2006xj}.
223: \begin{table}[thb]
224: \begin{center}
225: \subfigure[]{
226: \begin{tabular}{|c|c|c|}
227: \hline
228: $N_s$ & $K$& $N$\\
229: \hline
230: 12 & 10& 30000\\
231: 16 & 10& 30000\\
232: 20 & 10& 30000\\
233: 24 & 10& 30000\\
234: \hline
235: \end{tabular}}
236: \hspace*{1cm}
237: \subfigure[]{
238: \begin{tabular}{|c|c|c|}
239: \hline
240: $N_s$ & $K$& $N$\\
241: \hline
242: 12 & 14 & 100000\\
243: 16 & 14 & 100000\\
244: 20 & 14 & 100000\\
245: 24 & 10 & 100000\\
246: \hline
247: \end{tabular}}
248: \end{center}
249: \caption{Statistics achieved for the data given in Fig~\ref{free1}a.
250: The left branches correspond to (a) while the right branches together
251: with pieces at fixed $\beta_A=0.95$ and varying $\lambda$ correspond to
252: (b).}
253: \label{tab}
254: \end{table}
255: For the fixed $\lambda$ paths of Fig.~\ref{phase}a, along which main
256: simulations have been performed, we can fit very well
257: the step $\delta \beta_A$ needed to keep $\omega$ fixed with a
258: law of the form
259: \begin{equation}
260: \delta \beta_{A}(\omega,N_s) \simeq \frac{\alpha(\omega)}{N_s^2}\,,
261: \end{equation}
262: where $\alpha(12\%)=2.15(3)$ in the $\beta_A = 0.95-1.09$ range considered,
263: although we
264: expect it to change with the $(\lambda,\beta_A)$ window. Such scaling
265: implies that in order to explore a fixed region $\Delta \beta_A$ of
266: parameter space the number of ensembles will scale like
267: \begin{equation}
268: K \simeq \frac{\Delta \beta_A}{\alpha} N_s^2\,.
269: \label{scale}
270: \end{equation}
271: Therefore we cannot go too deep into phase II since the number of PT
272: simulations will eventually become too large for the computational means
273: at our disposal.
274:
275: The native $SO(3)$ temporal twists are given by $z_{i} \equiv N_s^{-2}
276: \sum_{x_{j},\,x_{k}}\,\prod_{P\;\in\; \mathrm{plane}\; i,4}
277: \mathrm{sign Tr}_F U_P$, ($\epsilon_{ijk4}=1$) for $i=x,y,z$
278: \cite{deForcrand:2002vs,Barresi:2003jq}. Within phase II they are
279: well defined having values close to either 1 or -1 for each configuration.
280: The partition functions restricted to fixed twist
281: are given as expectation values
282: $\langle \nu_k\rangle \,=\,{Z|_{z=k}}/Z$ of the projectors
283: \cite{'tHooft:1979uj}
284: \begin{eqnarray}
285: \nu_0 &=& {\frac{1}{8}}
286: \prod_{i=x,y,z}\lbrack1+\mathrm{sign}(z_{i})\rbrack\,,
287: \qquad
288: \nu_1 ={\frac{1}{8}} \sum_{j={x,y,z}}\;
289: \prod_{i=x,y,z} \lbrack1+(1-2\delta_{i,j})\,\mathrm{sign}(z_{i})\rbrack\,,
290: \nonumber\\
291: \nonumber\\
292: \nu_2 &=&{\frac{1}{8}}\sum_{j={x,y,z}}\;
293: \prod_{i=x,y,z} \lbrack1-(1-2\delta_{i,j})\,\mathrm{sign}(z_{i})\,,
294: \qquad
295: \nu_3 ={\frac{1}{8}}
296: \prod_{i=x,y,z}\lbrack1-\mathrm{sign}(z_{i})\rbrack\,.
297: \label{fractions}
298: \end{eqnarray}
299: The 2- and 3-twist sectors can of course
300: only exist on $\mathbb{T}^3$ \cite{'tHooft:1979uj,deForcrand:2001nd}.
301: From Eq.~(\ref{fractions}), since for an adjoint theory a change of twist
302: sector leaves the action unchanged, $\Delta U=0$ and:
303: \begin{equation}
304: F = -T \log{ {{Z_1}\over{3 Z_0}}}=
305: -{{1}\over{a N_\tau}} \log{ {{\langle\nu_1\rangle}
306: \over{3 \langle\nu_0\rangle}}}\,.
307: \label{eq6}
308: \end{equation}
309: The factor three in the denominator is again due to the
310: three equivalent 1-twist sectors on $\mathbb{T}^3$ rather
311: than one as on $\mathbb{R}^3$ \cite{deForcrand:2001nd,deForcrand:2002vs}.
312: With such a choice $F$ will be
313: zero if all twists are equally probable, i.e. on top of the bulk transition
314: and everywhere in phase I, where twist sectors are however
315: ill-defined due to the presence of open vortices. Eq.~(\ref{eq6})
316: obviously implies $F=0$ in the $T\to 0$ limit
317: \cite{'tHooft:1979uj,Kovacs:2000sy,deForcrand:2000fi} as long as the
318: $\Delta S$ contribution remains bounded.
319: Fig.~\ref{free1}a shows numerical results for $F/T$ in lattice
320: units at fixed $\lambda=0.8$ obtained
321: along the two separated paths indicated in Fig.~\ref{phase}a.
322: Errors are given combining statistical errors with
323: auto and cross correlations.
324: %-----------------------------------------------------------------
325: \begin{figure}[thb]
326: \begin{center}
327: \subfigure[]{\includegraphics[angle=0,width=0.49\textwidth,height=0.40\textwidth]%
328: {2a.eps}}
329: \subfigure[]{\includegraphics[angle=0,width=0.49\textwidth,height=0.40\textwidth]%
330: {2b.eps}}
331: \end{center}
332: \caption{(a) $a N_\tau F$ versus $\beta_A$ at
333: $\lambda=0.8$ and various $N_s$. (b) $a N_\tau F$ versus $\lambda-\lambda_c$
334: at fixed $\beta_A$ for $N_s=16$. $\lambda_c(\beta_A)$ denotes the position
335: of the bulk transition.}
336: \label{free1}
337: \end{figure}
338: %-----------------------------------------------------------------
339: The data start on top of the 2$^{\rm nd}$ order bulk
340: transition and go upward to what we interpret as the finite $T$
341: deconfinement transition -- given the rapid growth of $\langle L_A \rangle$
342: (cf. Fig.~\ref{phase}b) as well as of $F$ and taking fixed-twist
343: results for varying $N_\tau$ into account \cite{Barresi:2003jq,Barresi:2004qa,Barresi:2006gq}.
344:
345: The behavior in phase I and just at the bulk transition as well as in the
346: deconfined phase at large $\beta_A$ (upper phase II) is in agreement with the
347: standard vortex arguments for confinement \cite{'tHooft:1979uj}:
348: if vortices behave "chaotically" then $F$ should be zero and the theory
349: confines, while as deconfinement occurs $F$ should rise as
350: $F \sim \tilde{\sigma} N_s^2$, $\tilde{\sigma}(T)$ being the dual string
351: tension. This is qualitatively in agreement with our data. While $F \approx 0$
352: close to the bulk transition, we find a strong rise of $F$ for
353: $\beta_A \ge \beta_A^{crit} = 1.01(1)$ allowing to locate the
354: finite-temperature transition in phase II. As already explained above,
355: we are not able to go too deep into the deconfined phase with the computing
356: facilities at our disposal. Thus, we cannot really check whether the data for
357: $\beta_A \gg \beta_A^{crit}$ are consistent with the expected
358: $O(N_s^2)$ plateaus or whether they get saturated with respect to the
359: thermodynamic limit, i.e. whether $\tilde{\sigma}$ can indeed be calculated.
360: The effort necessary to this purpose, assuming that the estimate in
361: Eq.~(\ref{scale}) still works at higher $\beta_A$ and even taking into
362: account that for higher volumes the asymptotic behavior should kick in
363: earlier, we would need to simulate around 50 parallel ensembles for each
364: volume, again for a statistics of at least $O(10^5)$ per configuration in
365: each ensemble. For volumes with $N_s \geq 20$, for which finite size effects
366: start to be reasonably small, this goes beyond the computational power at
367: our disposal, although it should be manageable with a medium sized PC
368: cluster.
369:
370: For $\beta_A \le \beta_A^{crit}$, i.e. throughout the confining region of
371: phase II, we find negative values for $F$ which stabilize at large
372: 3-volumes ($N_s \geq 20$). $F < 0$ comes as a surprise, meaning that vortex
373: production is enhanced as compared to phase I. This is in contrast
374: to what expected from arguments valid within the fundamental representation,
375: i.e. $F(T)=0$ throughout the confined phase.
376: For an independent check we have carried out simulations at a few (fixed)
377: lower
378: $\beta_A$-values, varying $\lambda$ i.e. at somewhat lower temperatures
379: (see the horizontal paths drawn in Fig.~\ref{phase}a).
380: The results are plotted in Fig.~\ref{free1}b. Again we
381: find $F<0$, but the plateau values do not behave monotonously
382: as a function of $\beta_A$. Passing some minimum they increase again for
383: decreasing $\beta_A$. This is compatible with the expectation that the free
384: energy should go to zero in the zero-temperature limit
385: \cite{'tHooft:1979uj,Kovacs:2000sy,deForcrand:2000fi}. A systematic
386: extrapolation for different volumes and $N_\tau$ would be required to confirm
387: this behavior and to decide whether $F/T$ itself vanishes or goes to a
388: constant value.
389:
390: Let us draw the conclusions.
391: The main result of the present paper is the success in sampling the
392: {\it full} partition function via ergodic PT Monte Carlo simulations
393: and determining the free energy
394: $F$ for the creation of a $\mathbb{Z}_2$ vortex in pure $SO(3)$
395: Yang-Mills theory at finite $T$. We have seen a clear indication
396: for a deconfinement transition consistent with earlier findings of a
397: second order transition at fixed twist. Furthermore we
398: find that $F$ does not vanish in the confined phase
399: at $T \neq 0$, vortex creation being enhanced throughout it;
400: $F$ cannot therefore serve as an order parameter in a strict sense.
401: This implies
402: that the adjoint theory is unable to exhibit an order parameter
403: for center symmetry breaking in any form,
404: much like in the case of strictly centerless groups \cite{Holland:2003jy}.
405: This is not in contradiction
406: with the confining properties of the model, $F=0$ being a
407: sufficient but not necessary condition for confinement away from the $T= 0$
408: limit \cite{'tHooft:1979uj,Tomboulis:1985}. Moreover while vortex
409: suppression for $T<T_c$ would have been difficult to justify in light of
410: the literature \cite{Greensite:2003bk} the
411: vortex enhancement we observe does not contradict
412: that they can play a r\^ole in describing confinement,
413: although one cannot speak of vortex condensation in the usual understanding.
414:
415: Let us finish with a short remark on the universality problem we seem
416: to face. When identifying the partition functions
417: $\tilde{Z}\simeq Z({\beta}_A)$, invoking universality for
418: observables, one should be cautious. First of all, although
419: ${\mathbb{Z}}_2$ monopoles become suppressed in the continuum limit
420: of $\tilde{Z}$, $\langle\sigma_c\rangle$ is still far from unity for
421: the range of parameters commonly used in the simulations
422: \cite{Kovacs:2000sy,deForcrand:2000fi,deForcrand:2001nd}
423: and open vortices might still dominate the partition function.
424: Moreover $Z({\beta}_A)$ simply does not allow to define physical
425: observables in the fundamental representation. Expectation
426: values of fundamental Wilson and Polyakov loops and all their correlators
427: vanish identically for all $\beta_A$, i.e.
428: a fundamental string tension cannot be defined in a straightforward way.
429: To our knowledge bounds for $F$ from reflection positivity
430: or the connection with the electric flux have only been derived within the
431: fundamental representation \cite{'tHooft:1979uj,Tomboulis:1985,
432: deForcrand:2001nd}.
433: An interpretation of universality implying that any observable will
434: assume the same value in any discretization is therefore trivially
435: contradicted by the above considerations. However, a slightly more
436: conservative reading can agree with our findings: the truly physical
437: properties measurable in ``experiments'' like glueball masses and the
438: critical exponents at the transition should be reflected by physical
439: observables which can be defined {\it irrespective} of the discretization
440: chosen.
441:
442: We thank Ph. de Forcrand, J. Greensite, E.-M.
443: Ilgenfritz, T. Kovacs, M. Pepe and U. J. Wiese for comments and discussions.
444: G. B. acknowledges support from INFN.
445:
446: \bibliography{bib.bib}
447: \end{document}
448: