Physics Research International

Physics Research International / 2011 / Article

Research Article | Open Access

Volume 2011 |Article ID 437093 |

Jonas Johansson, Mats-Erik Pistol, "Microcanonical Entropy of the Infinite-State Potts Model", Physics Research International, vol. 2011, Article ID 437093, 5 pages, 2011.

Microcanonical Entropy of the Infinite-State Potts Model

Academic Editor: Ashok Chatterjee
Received24 Mar 2011
Revised02 Sep 2011
Accepted07 Sep 2011
Published24 Nov 2011


In this investigation we show that the entropy of the two-dimensional infinite-state Potts model is linear in configurational energy in the thermodynamic limit. This is a direct consequence of the local convexity of the microcanonical entropy, associated with a finite system undergoing a first-order transition. For a sufficiently large number of states π‘ž, this convexity spans the entire energy range of the model. In the thermodynamic limit, the convexity becomes insignificant, and the microcanonical entropy (the logarithm of the density of states) tends to a straight line. In order to demonstrate the behaviour of the convexity, we use the Wang-Landau Monte-Carlo technique to numerically calculate the density of states for a few finite but high values of π‘ž. Finally, we calculate the free energy and discuss the generality of our results.

1. Introduction

A thermodynamic system can be considered small when the characteristic length scale for the interaction between the subunits is comparable to the system size itself. In such systems, properties that are extensive in the thermodynamic limit, such as entropy and free energy, do not scale linearly with system size [1]. In addition, a traditional viewpoint concerning small systems has been that phase transitions cannot be defined for them since their partition functions are finite and, thus, analytic. According to such a viewpoint, only systems in the thermodynamic limit can undergo phase transitions since they exhibit the necessary discontinuities in temperature derivatives of thermodynamic potentials. This classification is unsatisfactory for two reasons. First, the thermodynamic limit is a theoretical construct that does not exist in nature. Second, several investigations have been performed to understand phase transitions in macroscopic systems based on transitions in downscaled, finite systems. A modern, proposed classification scheme for phase transitions in finite systems is based on canonical partition function zeros [2]. It has also been noted that finite systems undergoing a first-order phase transition exhibit an inverted curvature of a thermodynamic potential [3, 4].

A specific case of inverted curvature is the occurrence of negative specific heat in the microcanonical ensemble. This is a peculiar effect associated with small systems undergoing a first-order phase transition [4, 5]. Mathematically, this is equivalent to a region of local convexity in an otherwise concave microcanonical entropy [6]. The significance of this convex intruder [6] vanishes in the thermodynamic limit where the thermodynamics becomes extensive and there is no difference between microcanonic and canonic properties. This has been very nicely discussed by the late Gross in a series of papers, advocating the utilization of microcanonical thermostatistics as a foundation of thermodynamics, see, for instance, [6–8].

The purpose of the current investigation is to use this finite-size property of the microcanonical entropy to express the analytical solution to a well-known lattice spin model, the infinite-state Potts model in the thermodynamic limit [9, 10].

The standard, π‘ž-state, Potts model [11] is a lattice model with nearest neighbor interaction. Each lattice site, 𝑖, is assigned a spin state, 𝑠𝑖=1,…,π‘ž, and the spins interact according to the Hamiltonian:𝐻=βˆ’π½π‘–>𝑗𝛿𝑠𝑖,𝑠𝑗,(1) where 𝐽 is the interaction strength and 𝛿 is the Kronecker delta, which has the property that 𝛿(π‘₯,𝑦)=1 if π‘₯=𝑦 and 0 otherwise. The summation is over the nearest neighbors, and each interaction is only counted once. The partition function is given by βˆ‘π‘={𝑠}exp(βˆ’π›½π») where {𝑠} indicates summation over all spin configurations and 𝛽 has its usual meaning of 1/π‘˜π΅π‘‡, where π‘˜π΅ is Boltzmann’s constant and 𝑇 is temperature. Since a vast majority of the microstates are highly degenerate, it is convenient to rewrite the partition function in the form:𝑍=𝐸Ω(𝐸)π‘’βˆ’π›½πΈ,(2) where the microcanonical partition function, Ξ©(𝐸), also known as the density of states, represents the degeneracy factor for the energy 𝐸 in the canonical partition function (2). It is easy to verify that, for a two-dimensional (2D) square lattice with 𝑁 points, the energy can take all integer multiples of 𝐽 from βˆ’2𝑁𝐽 to 0.

The Potts model provides an excellent model framework for the study of phase transitions (see, e.g., [6, 12, 13]) and has a lot of applications in diverse fields, for an overview, see [11]. In 2D, the Potts model undergoes a first-order phase transition for π‘ž>4 and a continuous one for π‘žβ‰€4 [11]. In the limit of large π‘ž, the model is also known as the infinite-state Potts model, and interesting applications can be found in foam coarsening [14] and number theory [15, 16]. Wu et al. [16] has shown that the infinite-state Potts model is equivalent to the problem of restricted partitioning of integers. Based on this analogy, Wu [15] has solved the infinite-state Potts model in two and three dimensions. Although ingenious, it is not straightforward to extract thermodynamic information, such as the density of states, Ξ©(𝐸), from Wu’s partition function.

In this investigation we show that the microcanonical entropy of the 2D Potts model is linear in configurational energy in the limit of large π‘ž and in the thermodynamic limit. By simulations, we demonstrate the behaviour of the microcanonical entropy as π‘ž and 𝑁 are varied. We base our proof on the asymptotic properties of the convex intruder.

2. Simulation Method

We use the Wang-Landau algorithm [17] to calculate the density of states, Ω(𝐸). This is a Monte Carlo histogram technique, which yields the density of states with arbitrary precision while it is ergodic, meaning that every point in phase space is visited an (approximately) equal amount of times.

A major advantage with this method compared to conventional Monte Carlo techniques is that repeated simulations at multiple temperatures are avoided. Instead, the entire density of states is calculated in one single simulation run per system. Knowing the density of states admits calculation of both microcanonical and canonical thermodynamic properties.

The Wang-Landau algorithm is an iterative procedure based on a random walk in energy space with the purpose of modifying the density of states and updating a histogram. When the histogram is (sufficiently) flat, the modification factor is refined, the histogram is reset, and a new iteration starts. Ultimately, the density of states converges to its true value. The algorithm is outlined in detail in [17]. In that paper, a comparison with an analytically solved system, the 2D Ising model, is made. Here it was found that the relative errors in internal energy and canonical entropy calculated from the Wang-Landau density of states as compared to the exact analytical expressions were smaller than 0.09% and 1.2%, respectively. Since the π‘ž-state Potts model has not been exactly solved, the same kind of error analysis cannot be made for that case. However, we have no reason to believe that the simulation error for the π‘ž-state Potts model should be significantly different than that for the Ising model.

In the current investigation, the microcanonical entropy π‘†πœ‡, which is given by the logarithm of the density of states,π‘†πœ‡(𝐸)=π‘˜π΅lnΞ©(𝐸),(3) is calculated for 2D Potts models on square lattices with varying lattice size, 𝑁=𝐿×𝐿, and varying π‘ž. All simulations are performed with periodic boundary conditions. We define the energy in unit 𝐽 and the temperature in 𝐽/π‘˜π΅. The units are chosen so that 𝐽=1 and π‘˜π΅=1, which enables a dimensionless notation.

3. Results and Discussion

We will demonstrate that the entropy of the 2D square lattice Potts model becomes linear in energy for large π‘ž in the thermodynamic limit and is given by the asymptotic expression for the entropy per site,ξ‚€πœ€π‘ (πœ€)∼1+2lnπ‘ž,βˆ’2β‰€πœ€β‰€0,π‘žβŸΆβˆž,(4) where πœ€ is the configurational energy per site, πœ€=𝐸/𝑁. To demonstrate this, we will consider the degeneracy factors for the ground state and the most excited state, the latent heat as π‘žβ†’βˆž, and the vanishing of the convex intruder in the thermodynamic limit.

We start our line of investigation by writing the microcanonical entropy in the general form,π‘†πœ‡(𝐸)=𝜎(𝐸)βˆ’πœ“(𝐸),(5) where 𝜎 is linear in energy and πœ“ is a correction function [18]. This expression is valid also for finite lattices.

An explicit expression for 𝜎 can be written down if at least two degeneracy factors are known. For the ferromagnetic (𝐽>0) Potts model, the ground state, where every spin is the same, is π‘ž-fold degenerate,Ξ©(βˆ’2𝑁)=π‘ž.(6) At the highest excited state, Ξ©(0), every neighboring lattice site has different spin. In fact, Ξ©(0) is equal to the chromatic polynomial, 𝑃(𝐺,π‘ž), over the graph, 𝐺 [11]. In our case, 𝐺 is the 2D square lattice with 𝑁 sites (vertices) and periodic boundary conditions. 𝑃(𝐺,π‘ž) is a polynomial in π‘ž with degree 𝑁 and leading order coefficient 1. The signs of the coefficients, which depend on 𝑁 and 𝐺, alternate, and the coefficient of the π‘žπ‘βˆ’1 term is equal to the negative of the number of edges [19], that is, βˆ’2𝑁 in our case. Since 𝑃(𝐺,π‘ž) is a polynomial in π‘ž with degree 𝑁, we have that 𝑃(𝐺,π‘ž)βˆΌπ‘žπ‘ as π‘žβ†’βˆž, for any positive 𝑁.

Let us now investigate the large 𝑁 limit. It has been shown that the limit π‘Š=limπ‘β†’βˆžπ‘ƒ(𝐺,π‘ž)1/𝑁 exists and that limπ‘žβ†’βˆžπ‘žβˆ’1π‘Š=1 [20]. This means that 𝑃(𝐺,π‘ž)βˆΌπ‘žπ‘ also when we take the limit π‘β†’βˆž before the limit π‘žβ†’βˆž. More formally, it has been shown that the two limits in the expression limπ‘žβ†’π‘žπ‘ limπ‘β†’βˆžπ‘ƒ(𝐺,π‘ž)1/𝑁 commute if π‘žπ‘ >πœ’(𝐺) [21], where πœ’(𝐺) is the chromatic number of 𝐺 [22]. The 2D square lattice is a planar graph, and, since all planar graphs are four colorable, meaning that πœ’(𝐺)≀4, the limits clearly commute for π‘žπ‘ β†’βˆž.

This means that Ξ©(0) approaches π‘žπ‘ asymptotically,Ξ©(0)βˆΌπ‘žπ‘,π‘žβŸΆβˆž,(7) for any positive 𝑁. Our specific approach is based on taking the large π‘ž limit before the thermodynamic limit. However, since the limits commute, the final expression, (4), is independent on the order of the limits.

Now we can write the expression for 𝜎, the straight line between lnΞ©(βˆ’2𝑁) and lnΞ©(0),ξ‚€πœŽ(𝐸)=𝑁+π‘βˆ’1𝐸2𝑁lnπ‘ž,βˆ’2𝑁≀𝐸≀0.(8)

For the correction function, πœ“, we cannot give an explicit expression. That would correspond to solving the π‘ž-state Potts model analytically, which clearly is beyond the scope of the current investigation. We will instead be satisfied by demonstrating that πœ“ in (5) becomes insignificant in the thermodynamic limit of the infinite-state Potts model. We start by describing how the microcanonical entropy changes as π‘ž is increasing. For small values of π‘ž, there is a local maximum in π‘†πœ‡. For π‘ž=2, the Ising model, this maximum is at the midpoint of the energy scale. As π‘ž is increased, this local maximum shifts to higher energies and finally turns into a global maximum at 𝐸=0, tending to 𝑁lnπ‘ž, in agreement with (7).

These aspects are illustrated in Figure 1, where we show π‘†πœ‡ as a function of 𝐸 for π‘ž = 10, 100, and 1000 for an 𝑁=100 lattice. For finite thermodynamic systems undergoing a first-order phase transition, a certain part of the entropy becomes convex [23]. This convex intruder can indeed be seen as a signature of a first-order phase transition in a finite system [6]. The local convexity is located between the energies, πΈβˆ’ and 𝐸+, where 𝐸+βˆ’πΈβˆ’ is equal to the latent heat. In Figure 1, the convexity is clearly visible in the π‘ž = 100 and 1000 curves but not directly visible in the π‘ž=10 curve. We also see that the width of the convex region increases with increasing π‘ž. For combinatorial reasons, a few configurations with energies close to the ground state will never be accounted for. Specifically, for periodic boundary conditions, we have Ξ©(1βˆ’2𝑁)=Ξ©(2βˆ’2𝑁)=Ξ©(3βˆ’2𝑁)=Ξ©(5βˆ’2𝑁)=0, which explains the irregular appearance of π‘†πœ‡(𝐸) at low 𝐸.

Due to computer time limitations, π‘†πœ‡ for arbitrarily large π‘ž and 𝑁 cannot be calculated, but, in Figure 2, we show the convex intruder 𝛿𝑆/𝑁=(𝑆linβˆ’π‘†πœ‡)/𝑁, where 𝑆lin is the line between πΈβˆ’ and 𝐸+ (not to be confused with 𝜎, which is the line between βˆ’2𝑁 and 0), for π‘ž = 10 and 100 for various lattice sizes. Here we see that the convex intruder flattens out as 𝑁 increases, illustrating how the microcanonical entropy responds to variations in π‘ž and 𝑁.

Baxter has derived an expression for the latent heat per site, 𝑄, of the 2D Potts model [24], and from this we extract that 𝑄→2 as π‘žβ†’βˆž. Thus, in the limit of large π‘ž, the convex intruder spans the entire energy interval. This leads us to the following large-π‘ž properties of πœ“,πœ“(0)⟢0,π‘žβŸΆβˆž,(9) at the high energy endpoint andπœ“(𝐸)>0,βˆ’2𝑁<𝐸<0,π‘žβŸΆβˆž(10) at the interior. We also note that πœ“(βˆ’2𝑁)=0 for all values of π‘ž. In other words, as π‘žβ†’βˆž, πœ“ becomes identical to the convex intruder.

In the thermodynamic limit, this intruder becomes insignificant by necessity. That is, limπ‘β†’βˆžπœ“(𝐸)/𝑁=0, which is in agreement with our simulations shown in Figure 2, since as π‘žβ†’βˆž, 𝑆linβ†’πœŽ and π›Ώπ‘†β†’πœ“ with the properties of (9) and (10). As discussed by Gross et al. [6], any convexity of the microcanonical entropy, 𝑠(πœ€), for an infinite system with short-range interaction is forbidden [6]. A concave 𝑠(πœ€) is required for an extensive thermodynamics.

With these facts established, we can calculate the entropy per site in the thermodynamic limit of the infinite-state Potts model as𝑠(πœ€)=limπ‘β†’βˆžπ‘†πœ‡(𝐸)𝑁=limπ‘β†’βˆžξ‚Έξ‚€πΈ1+2𝑁lnπ‘žβˆ’πœ“(𝐸)𝑁=ξ‚€πœ€1+2lnπ‘ž,(11) which proves (4).

We have now shown that the microcanonical entropy of the infinite-state Potts model in the thermodynamic limit is given by (8). Combining this with (3) and (2), we can write down an explicit expression for the canonical partition function, 𝑍. From this we calculate the free energy per site, 𝑓, which is given byβˆ’π›½π‘“=limπ‘β†’βˆž1𝑁ln𝑍.(12) Using the elementary rule for summation of a geometrical series, we find thatξ‚»π‘“βˆΌβˆ’π›½βˆ’1√lnπ‘ž,𝛽<π‘™π‘›βˆšπ‘ž,βˆ’2,𝛽β‰₯lnπ‘ž.(13) The reader should be aware that this is an asymptotic relation, valid for π‘žβ†’βˆž. For any finite (but large) value of π‘ž, (13) can be regarded as an approximation. This result agrees with the mean field calculations by Mittag and Stephen [9], which are known to be exact in the π‘žβ†’βˆž limit [10].

Finally, it is straightforward to generalize our reasoning to any dimension, 𝑑>1. The Potts model on a 𝑑-dimensional hypercubic lattice takes energies from βˆ’π‘‘π‘ to 0. For large enough π‘ž, the phase transition is of first order, and the latent heat per site in the thermodynamic limit and the large π‘ž limit is given by 𝑄→𝑑, for a 𝑑-dimensional square lattice [9]. In other words, for any 𝑑>1 there is a convex intruder spanning the entire configurational energy interval. In the thermodynamic limit, the effect of this intruder vanishes, and the microcanonical entropy tends to its concave hull.

4. Conclusion

In summary, we have shown that the microcanonical entropy of the 2D standard Potts model approaches a straight line in the limit of large π‘ž and in the thermodynamic limit. This leads directly to an asymptotic form for the partition function, from which we calculate the free energy. We have also shown that it is straightforward to generalize this result to higher dimension since the phase transition is of first order and since the associated convex intruder takes up the entire energy range.


This work was performed within the Nanometer Structure Consortium at Lund University (nmC@LU) and supported by grants from the Swedish Research Council (VR) and the Swedish Foundation for Strategic Research (SSF). The authors acknowledge Jessica Bolinsson and Knut Deppert for their help to improve the paper.


  1. T. L. Hill, β€œThermodynamics of small systems,” The Journal of Chemical Physics, vol. 36, no. 12, pp. 3182–3197, 1962. View at: Google Scholar
  2. P. Borrmann, O. Mülken, and J. Harting, β€œClassification of phase transitions in small systems,” Physical Review Letters, vol. 84, no. 16, pp. 3511–3514, 2000. View at: Google Scholar
  3. F. Gulminelli and P. Chomaz, β€œCritical behavior in the coexistence region of finite systems,” Physical Review Letters, vol. 82, no. 7, pp. 1402–1405, 1999. View at: Google Scholar
  4. P. Chomaz and F. Gulminelli, β€œFirst-order phase transitions: equivalence between bimodalities and the Yang-Lee theorem,” Physica A, vol. 330, no. 3-4, pp. 451–458, 2003. View at: Publisher Site | Google Scholar | MathSciNet
  5. P. Chomaz and F. Gulminelli, β€œThe challenges of finite-system statistical mechanics,” European Physical Journal A, vol. 30, no. 1, pp. 317–331, 2006. View at: Publisher Site | Google Scholar
  6. D. H. E. Gross, A. Ecker, and X. Z. Zhang, β€œMicrocanonical thermodynamics of first order phase transitions studied in the Potts model,” Annalen der Physik, vol. 5, no. 5, pp. 446–452, 1996. View at: Google Scholar
  7. D. H. E. Gross, β€œA new thermodynamics from nuclei to stars,” Entropy, vol. 6, no. 1, pp. 158–179, 2004. View at: Google Scholar
  8. D. H. E. Gross and J. F. Kenney, β€œThe microcanonical thermodynamics of finite systems: the microscopic origin of condensation and phase separations, and the conditions for heat flow from lower to higher temperatures,” Journal of Chemical Physics, vol. 122, no. 22, Article ID 224111, pp. 1–8, 2005. View at: Publisher Site | Google Scholar
  9. L. Mittag and M. J. Stephen, β€œMean-field theory of the many component Potts model,” Journal of Physics A, vol. 7, no. 9, pp. L109–L112, 1974. View at: Publisher Site | Google Scholar
  10. P. A. Pearce and R. B. Griffiths, β€œPotts model in the many-component limit,” Journal of Physics A, vol. 13, no. 6, pp. 2143–2148, 1980. View at: Publisher Site | Google Scholar
  11. F. Y. Wu, β€œThe Potts model,” Reviews of Modern Physics, vol. 54, no. 1, pp. 235–268, 1982. View at: Publisher Site | Google Scholar
  12. M. S. S. Challa, D. P. Landau, and K. Binder, β€œFinite-size effects at temperature-driven first-order transitions,” Physical Review B, vol. 34, no. 3, pp. 1841–1852, 1986. View at: Publisher Site | Google Scholar
  13. J. Johansson, β€œMonte Carlo investigation of the phase transition in the 2D Potts model with open boundary conditions,” Physics Letters Section A, vol. 372, no. 42, pp. 6301–6304, 2008. View at: Publisher Site | Google Scholar
  14. Y. Jiang and J. A. Glazier, β€œExtended large-Q Potts model simulation of foam drainage,” Philosophical Magazine Letters, vol. 74, no. 2, pp. 119–128, 1996. View at: Google Scholar
  15. F. Y. Wu, β€œThe infinite-state potts model and restricted multidimensional partitions of an integer,” Mathematical and Computer Modelling, vol. 26, no. 8–10, pp. 269–274, 1997. View at: Publisher Site | Google Scholar | MathSciNet
  16. F. Y. Wu, G. Rollet, H. Y. Huang, J. M. Maillard, C. K. Hu, and C. N. Chen, β€œDirected compact lattice animals, restricted partitions of an integer, and the infinite-state potts model,” Physical Review Letters, vol. 76, no. 2, pp. 173–176, 1996. View at: Google Scholar
  17. F. Wang and D. P. Landau, β€œDetermining the density of states for classical statistical models: a random walk algorithm to produce a flat histogram,” Physical Review E, vol. 64, no. 5, Article ID 056101, pp. 1–16, 2001. View at: Google Scholar
  18. Sμ, σ, and ψ all depend on the three variables E, N, and q. For a compact notation we only write the E dependence explicitly.
  19. R. C. Read, β€œAn introduction to chromatic polynomials,” Journal of Combinatorial Theory, vol. 4, no. 1, pp. 52–71, 1968. View at: Google Scholar
  20. R. Shrock, β€œChromatic polynomials and their zeros and asymptotic limits for families of graphs,” Discrete Mathematics, vol. 231, no. 1–3, pp. 421–446, 2001. View at: Publisher Site | Google Scholar | MathSciNet
  21. R. Shrock and S. H. Tsai, β€œAsymptotic limits and zeros of chromatic polynomials and ground-state entropy of Potts antiferromagnets,” Physical Review E, vol. 55, no. 5 A, pp. 5165–5178, 1997. View at: Google Scholar
  22. R. Diestel, Graph Theory, Springer, Heidelberg, German, 2005.
  23. With convex we mean strictly convex. A function, f, is strictly convex if for any two points, x and y, and any 0t1, we have f(tx+(1t)y)<tf(x)+(1t)f(y). If f is convex, then f is concave. A linear function is concave (and convex but neither strictly convex, nor strictly concave).
  24. R. J. Baxter, β€œPotts model at the critical temperature,” Journal of Physics C, vol. 6, no. 23, pp. L445–L448, 1973. View at: Publisher Site | Google Scholar

Copyright © 2011 Jonas Johansson and Mats-Erik Pistol. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

More related articles

 PDF Download Citation Citation
 Download other formatsMore
 Order printed copiesOrder

Related articles

Article of the Year Award: Outstanding research contributions of 2020, as selected by our Chief Editors. Read the winning articles.