Page 1

arXiv:hep-lat/9205027v1 27 May 1992

High-Temperature series for the RPn−1lattice spin model

(generalized Maier-Saupe model of nematic liquid crystals)

in two space dimensions and with general spin dimensionality n

P. Butera and M. Comi

Istituto Nazionale di Fisica Nucleare

Dipartimento di Fisica, Universit` a di Milano

Via Celoria 16, 20133 Milano, Italy

High temperature series expansions of the spin-spin correlation functions of the RPn−1spin model

on the square lattice are computed through order β8for general spin dimensionality n.

Tables are reported for the expansion coefficients of the energy per site, the susceptibility and the

second correlation moment.

PACS numbers: 05.50.+q,64.60.Cn, 64.70.Md, 75.10.Hk

I. INTRODUCTION

Interest in the classical RPn−1spin systems [ [1]] on a two-dimensional lattice has been revived recently by the

results of a MonteCarlo simulation[ [2]] interpreted as evidence of a second order ”topological” phase transition, taking

place for values of the spin dimensionality n ≥ 3. This is unexpected according to renormalization group ideas. Indeed

the RPn−1models have the same formal continuum limit as the conventional O(n) symmetric n−vector spin models,

therefore they should belong to the same universality class and should not behave differently for n ≥ 3 ( when n = 2

the RPn−1model trivially reduces to the n−vector model). However the global topologies of the spin manifolds: the

hypersphere Sn−1with antipodal points identified in the case of the RPn−1model and simply Sn−1in the case of

the n-vector model, are different and it has long been known that this might be a reason for different phase diagrams[

[3]].

MonteCarlo studies of these systems, mainly in the n = 3 case, sometimes with conflicting or not completely

convincing results are by now numerous [ [4,5,6,7,8,9,10]], and they have been augmented by recent more extensive

simulations on large lattices [ [2,11,12]] using cluster algorithms[ [13]] in order to reduce the critical slowing-down.

On the other hand, high temperature expansion (HTE) studies are still practically absent, the only exceptions being,

to the best of our knowledge, a series through order β9for the internal energy and a series for the mass gap through

order β5in the n = 3 case [ [7]]. These expansions have been helpful for a first check of MonteCarlo simulation codes,

and series for other quantities and for other values of n would be equally welcome.

We have extended to every value of the spin dimensionality n through order β8the computation of the internal

energy , and for the first time we have computed series for the susceptibility and the second correlation moment.

These series are probably not long enough to provide, by their own, convincing evidence about the existence, the

location and the nature of a possible critical point, but we believe it is useful to make them promptly available so that

they can serve not only to check MonteCarlo data, but also for future more extensive high temperature calculations.

We shall explain later why our computational method, based on the Schwinger-Dyson recursion equations[ [14]],

although very transparent, becomes rapidly cumbersome and therefore is unable, in its present form, to produce

substantially longer series.

II. THE HIGH TEMPERATURE SERIES

Let us briefly describe the model and fix our conventions.

The partition function of the model is

Z =

??

x

ds(x)δ(s(x)2− 1)exp[β

2

?

x

?

µ=1,2

(s(x) · s(x + eµ))2] (1)

1

Page 2

The variables of the model are n-component classical spins s(x) of unit length associated to each site x = x1e1+x2e2

of a 2-dimensional square lattice, e1and e2are the two elementary lattice vectors.

The Hamiltonian and the integration measure have a global O(n)/Z2 and a local Z2 invariance. Since in two

dimensions continuous symmetries are unbroken[ [15]], the most general correlation function < φ(C) > can be written

as

< φ(C) >=< φ(x1,x2,...,xn;{bi,j}) >=<

?

1≤i<j≤n

(s(xi) · s(xj))bij> (2)

with integer bi,j ≥ 0. The local invariance under Z2, which also cannot break[ [16]], implies the further restriction

that each s(xi) has to appear in φ(C) an even number of times.

The correlation function (2) may be represented graphically as follows: the lattice points x1,x2,...,xn are taken

as vertices and a line connecting the vertices xi and xj is associated to each factor s(xi) · s(xj) in φ(C). In terms

of graphs the local Z2invariance requires that the degree of each vertex be even. Thus, for instance, the correlation

< s(x1) · s(x2) > vanishes trivially.

The fundamental two-spin correlation is then G(x2− x1;β,n) =< (s(x1) · s(x2))2>.

In particular we have −G(e1;β,n) = E, the energy per site.

We also have computed the moments m(l)(β,n) of the connected correlations

C(x2− x1;β,n) =

?

a,b

<

?

sa(x1)sb(x1)− < sa(x1)sb(x1) >

??

sa(x2)sb(x2)− < sa(x2)sb(x2) >

?

>= G(x2− x1;β,n) − 1/n

(3)

which are defined as follows

m(l)(β,n) =

?

x

| x |lC(x;β,n) =

?

r

a(l)

rβr

(4)

The HTE coefficients for G(e1;β,n) =?

rgr(n)βrare:

g0(n) =1

n

g1(n) =

n − 1

n2(n + 2)

g2(n) =

(n − 1)(n − 2)

n3(n + 2)(n + 4)

g3(n) =(n − 1)(72 + 18n − 11n2− n3+ n4)

n4(n + 2)3(n + 4)(n + 6)

g4(n) =(n − 1)(n − 2)(528 + 130n − 17n2− 3n3+ n4)

n5(n + 2)3(n + 4)(n + 6)(n + 8)

g5(n) = ((n − 1)(284160+ 130496n− 104032n2− 53344n3+ 6888n4+ 5496n5+ 474n6− 56n7− 2n8+ n9)

/(n6(n + 2)5(n + 4)2(n + 6)(n + 8)(n + 10))

g6(n) = (n − 1)(n − 2)(11704320+ 8093952n− 1233088n2− 1863104n3− 200776n4+ 103840n5

+26210n6+ 1386n7− 100n8− 2n9+ n10)/(n7(n + 2)5(n + 4)3(n + 6)(n + 8)(n + 10)(n + 12))

g7(n) = (n − 1)(341118812160+ 428301582336n+ 17644511232n2− 191549657088n3− 76694446080n4

+17276826240n5+ 16424658272n6+ 1926697808n7− 951227456n8− 295105184n9

−5505626n10+ 10001781n11+ 1876337n12+ 133277n13+ 1527n14− 171n15+ 9n16+ n17)

/(n8(n + 2)7(n + 4)4(n + 6)3(n + 8)(n + 10)(n + 12)(n + 14))

2

Page 3

g8(n) = (n − 1)(n − 2)(1271577968640+ 1237547925504n− 87404783616n2− 441393059328n3

−107082739328n4+ 37546256480n5+ 17834481104n6+ 440575008n7− 777645296n8

−105547274n9+ 10134853n10+ 3591697n11+ 316891n12+ 7749n13− 289n14− n15+ n16)

/(n9(n + 2)7(n + 4)3(n + 6)3(n + 8)(n + 10)(n + 12)(n + 14)(n + 16))

For n = 3 we have (compare with Ref.[ [7]]):

G(e1;β,3) =1

3+

2

45β +

2

945β2+

2

7875β3+

34

467775β4+

12179386

142468185234375β7+

13402

2280403125β5+

10702

47888465625β6

33996598

4872411935015625β8+ ...+

The HTE coefficients for m(0)(β,n), also called the susceptibility, are:

a(0)

0(n) =n − 1

n

a(0)

1(n) =

4(n − 1)

n2(2 + n)

a(0)

2(n) =4(n − 1)(8 + 3n + n2)

n3(2 + n)2(4 + n)

a(0)

3(n) =4(n − 1)(96 + 64n + 32n2+ 5n3+ n4)

n4(2 + n)3(4 + n)(6 + n)

a(0)

4(n) =4(n − 1)(1 + n)(3456+ 1968n + 570n2+ 89n3+ 9n4+ n5)

n4(2 + n)4(4 + n)2(6 + n)(8 + n)

a(0)

5(n) = (4(n − 1)(122880+ 101888n+ 40640n2+ 42528n3+ 35696n4+ 11094n5

+1807n6+ 162n7+ 10n8+ n9))/(n6(2 + n)5(4 + n)2(6 + n)(8 + n)(10 + n))

a(0)

6(n) = (4(n − 1)(−115015680− 79331328n+ 74609664n2+ 96772864n3+ 44006080n4

+15702208n5+ 7513312n6+ 2862016n7+ 648560n8+ 87178n9+ 7048n10+ 364n11+ 19n12+ n13))

/(n7(2 + n)6(4 + n)3(6 + n)2(8 + n)(10 + n)(12 + n))

a(0)

7(n) = (4(n − 1)(43104337920+ 43866980352n− 5407064064n2− 15002345472n3+ 3765867520n4

+8878097920n5+ 4282305280n6+ 1196842912n7+ 326380672n8+ 97376320n9

+22123168n10+ 3228422n11+ 292472n12+ 16058n13+ 566n14+ 23n15+ n16))

/(n8(2 + n)7(4 + n)3(6 + n)3(8 + n)(10 + n)(12 + n)(14 + n))

3

Page 4

a(0)

8(n) = (4(n − 1)(−94746307461120− 126660674322432n− 5226623926272n2+ 66792515567616n3

+32795171340288n4+ 48194863104n5− 862398014464n6+ 2921457334912n7

+2239266005664n8+ 790758440112n9+ 185029551696n10+ 37818452512n11+ 7776875970n12

+1395247971n13+ 184588028n14+ 16678488n15+ 985722n16+ 36650n17+ 952n18+ 32n19+ n20))

/(n9(2 + n)8(4 + n)4(6 + n)3(8 + n)2(10 + n)(12 + n)(14 + n)(16 + n))

For n = 3 these formulae give:

m(0)(β,3) =2

3+

8

45β +208

4725β2+

10273872032

427404555703125β7+

704

70875β3+

12704

5457375β4+

934133719808

183909666174609375β8+ ...

8254816

15962821875β5

+

37545856

335219259375β6+

The HTE coefficients for m(2)(β,n), the second correlation moment, are:

a(2)

1(n) =

4(n − 1)

n2(2 + n)

a(2)

2(n) =4(n − 1)(28 + 8n + n2)

n3(2 + n)2(4 + n)

a(2)

3(n) =4(n − 1)(624 + 344n + 124n2+ 15n3+ n4)

n4(2 + n)3(4 + n)(6 + n)

a(2)

4(n) =4(n − 1)(52224+ 57856n+ 37760n2+ 13200n3+ 2844n4+ 376n5+ 25n6+ n7)

n5(2 + n)4(4 + n)2(6 + n)(8 + n)

a(2)

5(n) = (4(n − 1)(1044480+ 1553408n+ 1394176n2+ 692768n3+ 237584n4+ 50998n5

+7107n6+ 642n7+ 30n8+ n9))/(n6(2 + n)5(4 + n)2(6 + n)(8 + n)(10 + n))

a(2)

6(n) = (4(n − 1)(420249600+ 1092464640n+ 1523555328n2+ 1255855616n3+ 710497728n4

+285572064n5+ 84264528n6+ 18404144n7+ 2909092n8+ 327054n9+ 25803n10+ 1354n11+ 44n12+ n13))

/(n7(2 + n)6(4 + n)3(6 + n)2(8 + n)(10 + n)(12 + n))

a(2)

+232510854656n4+ 131053062400n5+ 54800469376n6+ 17324337248n7+ 4212618016n8

+794772080n9+ 114865432n10+ 12396858n11+ 977532n12+ 55028n13+ 2106n14+ 53n15+ n16))

/(n8(2 + n)7(4 + n)3(6 + n)3(8 + n)(10 + n)(12 + n)(14 + n))

7(n) = (4(n − 1)(83979141120+ 204309430272n+ 291728203776n2+ 300109848576n3

a(2)

8(n) = (4(n − 1)(−58788371496960+ 8364704661504n+ 280171118592000n2+ 479504520511488n3

+464202911416320n4+ 329227765829632n5+ 188047485044736n6+ 87029638424064n7

+32173730443520n8+ 9456558685824n9+ 2219800018368n10+ 419325652576n11+ 63930454192n12

+7800108776n13+ 746995212n14+ 54841620n15+ 3013992n16+ 120454n17+ 3374n18+ 67n19+ n20))

/(n9(2 + n)8(4 + n)4(6 + n)3(8 + n)2(10 + n)(12 + n)(14 + n)(16 + n))

4

Page 5

In particular, the HT expansion of m(2)(β,3) is:

m(2)(β,3) =

8

45β +

488

4725β2+

2896

70875β3+

1123712

81860625β4+

12824336768

38854959609375β7+

67018144

15962821875β5+

2023066384

1676096296875β6

18110407484144

208430954997890625β8+ ...+

A correlation length may be defined, as usual, in terms of the ratio of m(2)(β,n) and m(0)(β,n).

Let us notice that a few simple checks of the formulae are possible: all HTE of the connected correlations have

to vanish for n = 1 because of the triviality of the RP0model. For n = 2, the expansions should reduce to the

corresponding ones for the O(2) (or XY-) vector model. Finally, for n = 3 the HTE of C(e1;β,n) agrees with the

calculation of Ref.[ [7]].

Our HTE have been computed from the Schwinger-Dyson equations of the model, an infinite system of linear

equations among the correlation functions. The generic equation, which may be deduced following closely Ref.[ [14]],

has the structure

< φ(C) >=

1

n + g1− 2

?

β

?

µ

(< φ(C−

µ) > − < φ(C+

µ) >) + (b12− 1) < φ(C12,12) > −

n

?

j=3

b1j< φ(C2j

12,1j) >

?

(5)

Here we have assumed that the vertices x1and x2are connected by one line at least, g1is the degree of the vertex

x1, bij the number of lines connecting the vertices xiand xj, < φ(C−

from < φ(C) > by removing a factor s(x1) · s(x2) and replacing it by s(x1) · s(x1+µ)s(x2) · s(x1+µ), namely

µ) > denotes the correlation function obtained

φ(C−

µ) = φ(C)s(x1) · s(x1+µ)s(x2) · s(x1+µ)

s(x1) · s(x2)

(6)

and analogously

φ(C+

µ)= φ(C)((s(x1) · s(x1+µ))2

φ(C)

(s(x1) · s(x2))2

φ(C12,12)=

φ(C2j

12,1j)= φ(C)

s(x2) · s(xj)

s(x1) · s(x2)s(x1) · s(xj).

The HTE of the correlation < φ(C) > is obtained solving iteratively eqs.(5) by the same procedure as in the case of

the n−vector model[ [14]]. Here however, a difficulty is met: while in the case of the n−vector model a large fraction

of the graphs generated after the first few iterations can be neglected, in this case, due to the local Z2 symmetry,

all graphs contribute nontrivially to the final results and therefore must be recorded. Thus the required computer

memory rapidly becomes exceedingly large and it is difficult to push the expansion beyond the 8-th order. However

not all the blame should be laid upon the computational technique since the combinatorial complexity of the expansion

is really higher and of a faster growth with the order than in the n−vector case. It is also interesting to recall that

analogous difficulties were met when performing strong coupling expansions in the Hamiltonian formalism[ [6]].

A simple analysis of the series by ratio and Pad´ e approximants methods[ [17]] (see Fig.1 and Fig.2 ) suggests

the existence of a critical point when n ≈ 2, but, unfortunately, the series seem to be not long enough to warrant

any reasonably safe conclusion when n = 3 or greater. To be sure, for various values of n there are some Pad´ e

approximants of the susceptibility having a real positive singularity or a complex conjugate pair of singularities

nearby the real positive β axis and in the expected position. The same happens for the logarithmic derivative of the

susceptibility. These poles however, at this order of approximation, are not stable enough to enable us to exclude

the possibility of an artifact of low order approximants to mimic the steep increase of the susceptibility. Thus some

completely different scenarios are still compatible with our series, for instance:

a) in analogy with the behavior of the n−vector model[ [14]] a critical point exist for n ≈ 2. As n is increased and

varied through some ˜ n ≤ 3, the critical point might split into an unphysical pair of complex conjugate singularities

so that the model becomes asymptotically free for n ≥ 3. This conjecture might be supported both by the alternate

ratios plots of Fig. 1, which seem to show the onset of an oscillatory trend [ [17]] and by some Pad´ e approximants to

the susceptibility or its log-derivative whose nearest singularities in the right half β plane are complex.

b) a critical point exists for all n as suggested by Ref.[ [2]].

5

Page 6

ACKNOWLEDGMENTS

Our thanks are due to Sergio Caracciolo for suggesting to undertake this computation and to Alan Sokal for

further encouragement and useful discussions. We also are indebted to U. Wolff for a useful discussion and for kindly

permitting us to use his unpublished MonteCarlo data in Fig.2. Finally we thank A. J. Guttmann and G. Marchesini

for carefully reading a draft of this note. Our work has been partially supported by MURST.

[1] W. Maier, A. Saupe, Z.

15A, 287 (1960); P. A. Lebwohl, G. Lasher, Phys. Rev.

A 6, 426 (1972); M. J. Stephen, J. P. Straley, Rev. Mod.

Phys. 46, 617 (1974); P. G. de Gennes, The Physics of

liquid crystals , (London, Oxford University Press 1974).

M. Plischke, M. Bergersen, Equilibrium Statistical Physics,

( Englewood Cliffs, Prentice Hall 1989).

[2] H.Kunz and G. Zumbach, J. Phys. A 22, L1043 (1989);

H.Kunz and G. Zumbach, Phys. Lett. B 257, 299 (1991).

[3] J. Kogut, M. Snow, M. Stone, Nucl. Phys. B 215[FS7],

45 (1983).

[4] R. D. Mountain, Th. W. Ruijgrok, Physica 89A, 522

(1977).

[5] S. Duane, M. B. Green, Phys. Lett. B 103, 359 (1981).

[6] S. Solomon, Phys. Lett. B 100, 492 (1981).

[7] M. Fukugita, M. Kobayashi, M. Okawa, Y.Oyanagi, A.

Ukawa, Phys. Lett. B 109, 209 (1982).

[8] D. K. Sinclair, Nucl. Phys. B 205[FS5], 173 (1982).

[9] M. Caselle, F. Gliozzi, Phys. Lett. B 147, 132 (1984); M.

Caselle, Zeit. fur Phys. C 28, 233 (1985).

[10] C. Chiccoli, P. Pasini, C. Zannoni, Physica 148A, 298

(1988).

[11] U. Wolff, unpublished work

[12] S. Caracciolo, A. Pelissetto, A. D. Sokal, work in progress.

[13] J. Goodman, A. D. Sokal, Phys. Rev. Lett. 56, 1015 (1986);

R. H. Swendsen, J. S. Wang, Phys.

(1987); U. Wolff, Phys.Rev.

Goodman, A. D. Sokal, Phys. Rev. D 40, 2035 (1989); A.

D. Sokal, Nucl. Phys. B (Proc. Suppl.) 20, 55 (1991).

[14] P. Butera, M. Comi and G. Marchesini, Nucl. Phys. B300

[FS22], 1 (1988); P. Butera, M. Comi and G. Marchesini,

Phys. Rev. B 41, 11494 (1990).

[15] N. M. Mermin, H. Wagner, Phys. Rev. Lett. 17, 1137

(1966).

[16] S. Elitzur, Phys. Rev. D 12, 3978 (1975).

[17] G. A. Baker and D. L. Hunter, Phys. Rev. B7, 3346,

3377(1973); D. S. Gaunt, A.J. Guttmann in Phase Tran-

sitions and Critical Phenomena, C. Domb and M. S.

Green Eds. Vol. 3, (London, Academic Press 1974); A. J.

Guttmann in Phase Transitions and Critical Phenomena,

C. Domb and J. Lebowitz Eds. Vol.13, ( London, Academic

Press 1989).

Naturforsch. 14A, 882 (1959);

Rev. Lett. 58, 86

Lett. 62, 361 (1989); J.

FIG. 1. Alternate ratios ¯ rs(n) = (a(0)

the expansion coefficients of the susceptibility for various val-

ues of the spin dimensionality n are plotted versus 1/s. Going

from the lower plot to the upper, we have n = 2,3,..,8.

s−1(n)/a(0)

s+1(n))1/2of

6

Page 7

FIG. 2. The susceptibility for n = 3 at order β8vs. β. The

continuous curve shows the [4/4] Pad´ e approximant (which

is singular at β ≈ 5.175 ± 0.315i). The dashed curve shows

the sum of the susceptibility series truncated at order β8.

The squares represent data from the MonteCarlo simulation

of Ref. [ [7]]. The triangles represent unpublished data from

a MonteCarlo cluster simulation performed by U. Wolff[ [11]]

on lattices of size up to 2562.

7