Monte Carlo simulation of dimensional
Wolfhard Janke
Institut fu'r Physik, Johannes Gutenberg

crossover in the XY model
Weg 7,

Un-iversitat Mainz, Staudinger

55099 Mainz, Germany

Klaus Nather
Institut fiir Theoretische Physik, Freie Universitat Berlin, Arnimallee (Received 23 July 1993)

14, 14195 Berlin, Germany

We report Monte Carlo simulations of Villain's periodic Gaussian XY model on L X N lattices of film geometry (L »N) with up to N =16 layers, employing the single-cluster update algorithm combined with improved estimators for measurements. The boundary conditions are periodic within each layer and free at the bottom and top layer. Based on data for the specific heat, the spin-spin correlation function, and the susceptibility in the high-temperature phase we study the crossover from three- to twodimensional behavior as criticality is approached. For the transition temperatures, determined from Kosterlitz-Thouless fits to the correlation length and susceptibility, we observe a pronounced scaling behavior with N. The associated critical exponent, however, deviates from theoretical expectations. More qualitatively, we further discuss the distribution and shapes of vortex loops in the crossover region.



work by Schmidt and Schneider who performed simulations for the cosine formulation.


For systems with short-range interactions the concept of universality predicts that qualitative properties of continuous phase transitions should only depend on the spatial dimension of the system and on the symmetry of the order parameter. ' Mathematically, the spatial dimension is given by the number of directions in which the system extends to infinity. While this obviously can never be realized in nature, there are still many systems where finite-size corrections are experimentally so small that this mathematical idealization works perfectly. By the same argument the two long directions of films can be considered as infinite. The film thickness, on the other hand, is definitely finite and does strongly inhuence the behavior of the system. In fact, in contrast to the threedimensional (3D) bulk behavior, we expect for films of finite thickness a phase transition (if at all) that can be classified according to the two-dimensional (2D) universality class. ' More precisely we expect a crossover from 30 to 20 behavior as soon as the bulk correlation length g of the system approaches the order of the film thickness. ' Taking the Ising model as a typical example, the crossover from 30 to 20 behavior of systems with onecomponent order parameters has been studied numericalWhile this ly some time ago by Binder and Hohenberg. has interesting applications to many magnetic materials there are also important physical systems that are described by a two-component order parameter, e.g. , liquid helium or superconductors. Here we report Monte Carlo (MC) simulations of the crossover behavior for these systems, using the XY model as the generic representative of this universality class. In our work we have chosen Villain's periodic Gaussian formulation of the XY model. Some of our results can be compared with recent

The partition function model is given by

of the periodic Gaussian XY



. I n, (x) I

— g (V, 0 —
x, i



— where V;0(x) = 0(x+ i) — 0(x) are the lattice gradients in the i direction of a cubic lattice, the integer variables ~ n,. (x) run from — to oo, and 13= Jlk~T is the (reduced) inverse temperature. Films of increasing thickness were simulated by stacking N = 1, 2, 3, 4, 6, 10, and 16 layers of size L XL with L on top of each other along the z direction. The ferromagnetic coupling was taken to be isotropic. Within each layer we took periodic boundary conditions in order to reduce finite-size effects in L as much as possible, while at the top and bottom layer we imposed free boundary conditions in the z direc-


tion. Our choice of the periodic Gaussian formulation is motivated by the fact that the partition function (l) can exactly be transformed into a gas of topological defects with long-range interactions of the Coulomb type [9, 10]. This is the main difference to the cosine formulation [with exp(Pcos(V;0)) replacing the periodic Gaussian], where the defects also interact in a complicated nonlinear Physically the deway with the spin-wave excitations. fects can be interpreted as the vortex excitations invoked in the description of liquid helium. In two dimensions the defects are pointlike objects, providing the physical



0163-1829/93/48(21)/15807(5)/$06. 00

15 807


The American Physical Society

As a general rule we find that this is the case for X/2. 16. I.29 0. analyses temperature series (HTS) expansions. corrections are expected which modify the short-distance behavior of g (x). that this is a safe condition. 1. To be precise. and the spin-spin (4) G(x. thereby taking into account the periodic boundary conditions in the x direction. In two dicritical exponents 3 mensions. To reduce finite-size eA'ects in we always took care that L = 6 — Extensive tests for the pure 2D model showed 8g. THE SIMULATION In our MC simulations of the partition function (1) we applied the Z. 1. Ref. The solid cosh[(x L/2)/g— through ] cases a correlation length of /= 6. j I i I I I I I I i j I I I I j i ~ I I j I ~ I I I I ~ I i j I i I I I i I I I j i i i i I i i i i j ~ I I I I ~ i I i j I I ~ I I ~ i i I j ~ I ~ 0. e. As is demonstrated in Fig. RESULTS To determine the transition point P.32 0. ' and recent MC simulations' are all compatible with a conventional powerlaw behavior g~(l f3/P— P/P.30 0. z) = g~ i s(x. we first located the onset of the twodimensional KT behavior of g and y via goodness-of-fit analyses. however. " phase. From high-statistics MC simulations ' 140 on large lattices of sizes correlation lengths up to up to 1200. yielding in both (= III. ' In three dimensions the defects are linelike objects. the main results are based on analyses of the susceptibility. however.) '. Correlation functions for %=4 ( ) of size L XL with L =100. The results of both approaches were interpreted in favor of the KT scenario. . alternative considerations' a conventional power-law suggested behavior with nontrivial critical exponents v and y. x') = (s(x) s(x') in the high-temperature ). on the other hand. 2. 35 — with s(x. For many layers.33 0. z).28 0. y»=1. [14]. using FIG. Only for the thickest film with X =16 lay- g) y=V where s(x)— (cosO. (X) for each film thickness X. : —s x 2 (3) and V=%L. we first studied the 2D limit of this formulawith tion. we have measured the k =0 momentum projection of (4). IV.15 808 WOLFHARD JANKE AND KLAUS NATHER 48 ~ treatment of Kospicture for the renormalization-group terlitz and Thouless (KT). The data for X =30 layers in Fig. that these corrections die out quite rapidly and that there is still a significant range in x over which the simple ansatz is an extremely good approximation. . along the x direction in the innermost layers and extracted the correlation length g from fits to a hyperbolic cosine.34 0. 2 i i i I g (x) = (s(x. Since no data of comparable quality were available for the periodic Gaussian XY model. %"hile the KT theory predicts an exponentially diverging correlation length. 670 and @=1. (solid line) of the suscep- .31 0. of highproaches. using for the measurements variance reduced "cluster observables. and %=30 (+) lines are fits of the last 20 points. analyses of extended HTS expansions and MC studies ' of the cosine formulation have been performed. 316) to 2D KT behavior — tibility for X = 16 layers. approximation with n =100 to the angles 8(x). we obtained even stronger evidence for the KT predictions in this formulation. ' resummations of field-theoretic perturbation series. layers FIG.g. sin9) correlations. The phase transitions in the limiting cases N =L (3D) and K =1 (2D) have been investigated by a variety of apIn three dimensions. cosh[(x L /2)/g]. . 1 show. (x) in a lookup table.s(0. sides monitoring the peaks of the specific heat C. In this way it is then straightforward to adapt the single-cluster update Bealgorithm to the periodic Gaussian formulation. For literature on recent attempts to understand the X transition in liquid helium as proliferation of these vortex lines see. calculated numerically for each f3 once at the beginning of the run. ' ' 2 I i i I j i i I i j i I I j i ~ I I f \ I i I j i I ~ i [ i i i I j I i I ~ j i i I i j i I I I 4 ' ' —0 'I i i J i j i I I I I i I I i j i I I I I t0 20 30 distance ~ 40 g ~ exp [ b ( 1 P /f3— ].27 0. the situation has been quite controversial. This enables us to store the sums over the integer variables n. To clarify this point. z). y. ) and y~(1 — ) r with v=0. Crossover froDi 3D power-law behavior (dashed line. and susceptibility yccg " with il= —. this ansatz works perfectly in the quasi-2D case of a few layers. z) ).

22. (N) =P.401( 13 ) 0. 670+0. (N) values given in Table I [labeled according to the type of KT fit used to determine p. =0.342 28(64) 0. . however. P. is much worse than in the case of P. It is tempting to identify this 3D region with = 16 layers this is. [Recalling that also in 3D the specific-heat peak is finite with a cusplike singularity (the critical exponent a3D =2 — 0.86 2. . Dv3D= — 01 is slightly negative). (N) ] are collected in Table II.21 8.397(6) 0. (6) Here coupling of the 3D bulk system and v3D is the bulk correlation length exponent. In this comparison. 3 we have adopted the quite plausible and commonly accepted hypothesis that „=p. however. since in the present fits we used only the five data points with the largest correlation length. 2.758 6(59) 0. of course. In a recent MC study of the cosine formulation Schmidt and Schneider obtained from an of Binder's parameter a similar value of analysis 1/X=0.3334(4) 0.349 74(24) 0.725(8) 0. fits to P.366 41( 30) 0. In Table I we also give the locations /3. TABLE II. (N) and. g N /10. (N) compiled in Table I.390 87(21) 0. 334. We see that the critical coUplings do indeed scale quite nicely down to remarkably small values of N.751 0(32) 0. we start seeing a region with 3D bulk behavior. According to Fisher's scaling prediction ' the P. we are.391 29(22) 0. howev- P.365 81(92) 0. Single fits to all data points gave compatible results. (N) =P. In Fig.358 29( 11) 0.04 0.02 fits P. Three-parameter 7. this time a trend to lower values of I /k and higher values of /3.420 61(64) 0. . — 334 vs N on a log-log scale.349 92( 12) 0. „ is visible. requiring layers of sizes of the order of 200 to 400 .391 30(53) 0.426 20(58) 0.706( 19 ) 0. P. which is hardly significant in view of the increasing error bars.349 85( 12) 0. from kT g(p) 2 3 y(T) 0.24 3. We see that with increasing film thickness N the absolute distance from p.339 76( 18) to the exponent (8) ers.374 43(20) 0. We have checked for a systematic trend in our data by discarding more and more points for small N in the fits. with a maximal correlation length between 26 and 46 for N ~ 2. p.28 0. Depending on N.07 0. albeit with a much larger error bar.. it should be kept in mind that (6) is only valid asymptotically for large N.3336(4) 0. fixed at its theoretical value.(N).365 98(92) 0.3343( 9 ) 0. 3 we plot =0. In these fits all points with N ~ 2 are taken into account. The values for a single layer deviate slightly from our earlier results in Ref.754 8(35) 0. The results of three-parameter fits to (6) for each of the four sets of P.484 4( 12) 0. In these fits we ' thus always kept v= —.08. (N) should scale asymptotically for large N as a straight line corresponding ( 1/X=O. : (7) This value is significantly larger than the theoretical prediction (7).48 MONTE CARLO SIMULATION OF DIMENSIONAL CROSSOVER.349 67(23 ) 0. The solid curve is = v3D =0. Inverse critical temperatures P. It is well known that in two dimensions the peak maximum stays finite in the infinite volume limit and that the peak location does not coincide with the transition point but is displayed by about 30% to higher temperatures.389(6) 0. 1 TABLE I.417 70(91) 0. fits give an even larger exponent of three-parameter I/A. As a result we observe only a slight trend to smaller values of I/A. — ( ~ ) is the critical p.341 89(25) pmax 4 6 10 16 0. For each film thickness N these fits provided us with four estimates of P. the number of data points included in the fits was 5 or 6. with p. The quality of the fits.703( 20) 0. In order to clarify the discrepancy with the theoretical prediction (7) it would be desirable to perform further simulations of much larger systems.342 25(60) 0. (N) and the location of the specific-heat maximum.345 77( 15) 0.490 3( 30) 0.487 8(27) 0. In Fig. 331 that is slightly smaller than p.538 7( 19) 0. +cN with c being a constant and I /A. p. 71+0.393 79(36) 0.341 86(25 ) x(p) 0.754 5( 54) 0. „(N) of the specific-heat peaks.721(9) . p. Assuming that also „(N) scales according to (6). by discarding data points for small N.002 . Having determined the region of 2D behavior.„( ~ )=0. =0. we then fitted the data for g and y to the KT prediction (2) and the corresponding formula for g. see Fig. As a test for systematic errors (due to omitted correction terms) we also used (2) rewritten as a function of T. 15 809 P.366 23( 30) 0.407( 15 ) 0.417 14( 89) 0. but even for N still perturbed by nonuniversal lattice corrections.391 72(54) 0. 70+0. 8 and favor a value of „==/3. (N) decreases.57 0.419 88(60) 0. respectively.482 6(11) 0. (N) p.3344(10) 0. + cN p3D gl T) g(p) y( T) y(P) 0.01 . .

366=P. t e o h ld 1 an t near critica 1' th ity ' k h i e behave effectively like the vortex porn s ossible scenario a dvanced in the literava KT p icture. Hal p erin. 355 (/=19) an (h) P=0. 678 (19 S. Teor. e im er. K. y(T) (A).15 810 WOLFHARD JANKE AN KLAUS NATHER 48 Ct C9 cQ- -3 Z'. Joyce. Also shown are ' the peak locations P~. Ambegaokar. 7 g- . 49. ys. T. is er. D. In fact. Th o t li 1 tri oy h 1 tions displayed in Fi . . ... d' Con densed Matter I'hy sics V. Zh. 24. Capehart and M. . an ' and H. 8A. useful discussionn and the Deutsc h e ' ev. Thin So i y. Jan ke and K. -B. . is a difficult prob em w tions..o. 1/A.. . o ( E im h (1965) [Sov. Fiz. and y(P)) ('7). Schmi t an T. JETP 22. 4. (Fr ance) 36. rres ork. Dohm for a Fo h sc a ort.. I. Phys. 13 d B. W . h nd 2) J. J. 't ]. pur ' ' n p otentia lb e w ot 1 S 1S layers. p. This work was suppor ted in e G d part b y Deutsche Forsc hungsgemeinsc h a f No.. 1479 (1970). V. easi y isotro ic wit exp ic' the entropic con ri terms. S. Z. Phy . Phys. cQO -4 -5 -6 0 ln N ) ( 'th N. 975 eo 5 b M. 581 ' 7W. 2194 (1974). suc h 'hat 'h uld look like a gas of vortex p oints onto ' W1 og arithmic long-range ' r ation over t e r1ve d from a summati atment of paraents. Proceedings o the Workshop. W. Lett.. 4 clearly s h ow h ig. ' . 1993). Green ca 1971). E. nd ldf V. similar to t h e t rea ). of e. of a 6X 200 lattice X at (a) P=0. „(X)of the specific heat ( o ). corresp onding plots for t e k uite simi ar. V'llain. ations are vortex I. Eksp. t e m ines.e v cou ld also not confir m the theoretica 1 pr di tio R ther. 140. Larkin. I. One po ' ' ds de enerate near criticality to rodt a like objects in the z direction. J. Vortex lines in a 6X10 section.b--"in the periodic d f ~ ~ a roe asent to see the true asy m p totic sca 1ng e ies of liqui . 01. th li Ill ints k last three points looks reasona bl comp ' 1 t t ss Let us finally discuss the topologica ince eometnca 11y the films are 3D... =0. Schneider.„. calculated from KT fits to ) ( ). . Binder. e (X) t e spec' recise mathematical argum ent enforc' ll N a curvature is obvifittho t X). Schuttler. 52+0. Spring er Procee ingss rin (p p g Berlin.29 ACKNOWLEDGMENTS t ~ ~ Do to thank Professor V. of edited by D. K1256. CONCLUDING REMARKS . go f the critica coup i p ': t to N=161 --. Nat er in Computer S' ulation Studies ln W. w hile in early worrk a value o u i gave a muc h smaller value quo t d more recent s tudies iu. 'ther "n"d which requires fur b er. p or films of XF spins lation with up itions in the z iree io..

Phys. 15 811 Phys. G. Brewer Amsterdam. Fiz. 6519 (1988). 1. Wolff. J. in Progress in I. Wolff. 11 969 (1993). Lett. ibid. D. and W. Hulsebos. Phys. Kleinert. Eksp. Lett. Meyer. ~ W. Butera and M. U. Phys. Rev. F. Soc. Phys. Phys. 4692 {1986). and F. 238 (1990). B 241. F. C. Rev. B 45. R. Goodman. 1181 (1973). Gasparini. XIII. J. Rev. Gupta and C. 11 (1991). Stamatescu. 10769 (1990). DeLapp. Vol. B 8. 61. Phys. Wortis. C. Rev. M. Patrascioiu. Holm. Comi. Nucl. Phys. Temperature Physics. p. . B 21. J. I4W. Kosterlitz and D. Vol. Janke and K. Rev. 610 (1972)]. L. Ferer. Kleinert. B 33. 52. L137 {1985). Savit. Lett. J. S. 4725 (1986). I. Teor. Gauge Fields in Condensed Matter (World Scientific. 1992). Zinn-Justin. Apostolakis. Smit. N. B 39. 1251 (1990). 581 (1990). (Paris) 46. Theor. Rev. Hasenbusch and S. LeGuillou and J. B 33. 8W. B. J. Phys. V. Rev. Berezinskii. M. Nather. Comi. 40. Phys. C. Mod. Lett. H. Linke. and J. and A. Lett. G. Ami and H. M. 623 (1988). J. 29. 60. B 305. Hallock. . Janke. 354. Phys. Gasparini and I. Nucl. ~4U. P. J. ibid. . Phys. A. 306 (1990). Rhee. M. Gupta. Phys. A. Teor. Lett.ow (North-Holland. Phys. Ferer and Z. 775 (1991). Phys. Phys. ~ K. 7419 (1993). 875 (1988). and V. J. J. 1046 (1974). Battrouni. Patrascioiu and E. 1996 (1988). Wolff. M. B 48. 453 (1980). 42. A. B 334. Rev. Phys. 759 (1989). C 6. Rev. Kosterlitz. Sokal. Phys. Phys. and J. 361 (1989). 1991. Phys. Int. Vink. 1144 (1971) [Sov.48 MONTE CARLO SIMULATION OF DIMENSIONAL CROSSOVER. 61. ibid. 341 (1973)]. M. Jpn. J. 47. R. 1533 (1981). R. FU Berlin. Baillie. U. Mo. ibid. Singapore. J. 61. ~ Y. For a recent review. E. Zh. M. Adler. 2169 (1992). Fiz. M. Rev. Janke. P. edited by D. Zh. A 148. Lett. 356. 62. J. Rev. Baillie. ~sV. Butera. Marchesini. Edwards. Seiler. Physica A (to be published). Thouless. 64. 3976 (1980). 534 (1989). 2883 (1992). Maps and R. I. D. 672 (1973) [Sov. Y. 5205 (1973). 289 (1991). Rev. Phys. O. JETP 34. A 157. Finotello. Moore. G. C. C 7. and G. G. Popov. J. see F. C. Diplom thesis. Seiler. Nather. Nucl. 47. JETP 37. 1989). Eksp. Fox. M. Yu. A. B 322. Phys. Phys. R. Janke. Phys. Phys. and M. Lett.

Sign up to vote on this title
UsefulNot useful