2021 Volume 62 Issue 3 Pages 374-379
Thermodynamics on a Cantor-lattice Ising model is studied to clarify effects of fractal structure. Exact solutions based on the transfer matrix are investigated for finite size systems, and it is found that there is non-trivial relationship between entropy and fractal structure. In order to understand the nature in the thermodynamic limit, the renormalization method is applied. The results suggest a possibility of residual entropy due to the competition between non-uniformity of the fractal structure and uniform external field. These pave a simple way to approach general behaviors of non-uniform systems including fractal structures, such as quasicrystals.
Quasicrystals,1–3) experimentally observed in binary or ternary alloys [e.g., Cd6Yb and Al–Zn–Mg3)], have a self-similarity such as the Penrose lattice4–9) with an inflation rule.10) These kinds of quasicrystals often show fractal structures in their essentials. To clarify physical behaviors on fractal structures, we need to know how the structure affects such thermodynamic properties as entropy, internal energy, and so on. Traditionally, a lot of fractal structures have been investigated from theoretical points of view in statistical mechanics.11–16) Actually, as is well known, fractal structures of local orders often appear even in uniform systems at critical temperatures, where the correlation functions diverge.15,16) Furthermore, there are some previous theoretical approaches to physical properties such as magnetism on quasiperiodic structures.17–21) However, effects of fractal structures, which are constitutionally given in quasicrystals, on thermodynamic properties have still remained unclear so far.
Here we introduce a Cantor-lattice Ising model22–24) to study thermodynamic properties. This is a simple example of fractal structure where we can control its strength, called fractality. In a limiting case of this model, it corresponds to the one-dimensional Ising model, while it corresponds to an isolated Ising model in the other limiting case. We explain more details of the Cantor lattice and the controlling parameter λ in the following section. In Sec. 3, we show some exact results for finite size Cantor-lattice Ising model. The most interesting case is shown in thermodynamic limit. In the thermodynamic limit, the Cantor-lattice Ising models are analyzed by a renormalization treatment as discussed in Sec. 4. Summary and discussions are included in Sec. 5.
The Cantor-lattice Ising model22,23) is constructed by analogy with the Cantor set.24) Figure 1 shows the Cantor lattice. Introducing a concept of “generation” will help us to create the Cantor-lattice Ising model. The first generation has three spins {σ1, σ2, σ3} and two interactions with the same magnitude J as shown in the top of Fig. 1. Once we obtain the first generation, the second generation is given by connecting three first generations by the interaction λJ. Here the parameter λ is assumed as |λ| < 1. And then, the third generation is given by a group of second generations with the interaction λ2J. In the same manner, the k-th generation is obtained as three (k − 1)-th generations connected by the interaction λk−1J. The model Hamiltonian of first generation is described as
\begin{equation} \mathcal{H}_{1} = -J(\sigma_{1}\sigma_{2} + \sigma_{2}\sigma_{3}). \end{equation} | (1) |
\begin{align} \skew3\hat{Z}_{1}(\sigma_{1},\sigma_{3}) &\equiv \sum_{\sigma_{2} = \pm 1} \exp [K(\sigma_{1}\sigma_{2} + \sigma_{2}\sigma_{3})] \\ &= 2(\cosh^{2}K + \sigma_{1}\sigma_{3}\sinh^{2}K), \end{align} | (2) |
\begin{align} \skew3\hat{Z}_{k}(\sigma_{1},\sigma_{N_{k}}) &= \sum_{\tau_{1},\tau_{2},\tau_{3},\tau_{4}}\skew3\hat{Z}_{k - 1} (\sigma_{1},\tau_{1})e^{\lambda^{k - 1}K\tau_{1}\tau_{2}}\\ &\quad \times\skew3\hat{Z}_{k - 1}(\tau_{2},\tau_{3})e^{\lambda^{k - 1}K\tau_{3}\tau_{4}}\skew3\hat{Z}_{k - 1}(\tau_{4},\sigma_{N_{k}}), \end{align} | (3) |
\begin{equation} Z_{k}(T,H,\lambda) = \sum_{\sigma_{1},\sigma_{N_{k}}}\skew3\hat{Z}_{k}(\sigma_{1},\sigma_{N_{k}}). \end{equation} | (4) |
Schematic picture of the first, second, and k-th generation.
At the end of this section, we mention the meanings of the parameter λ. In the case of λ = 1 and k → ∞, the present model is equivalent to a one-dimensional Ising model. This dimension is of course one. On the other hand, in the case of λ = 0, the model corresponds to isolated three-spin clusters. The finite k cases also correspond to zero-dimensional even if λ = 1. Thus, in this case, the model can be regarded as zero-dimensional. According to these limiting cases, the Cantor lattice describes fractal systems in the case of 0 < λ < 1. We can treat the present Cantor-lattice Ising model as a toy model in order to observe the behavior of thermodynamic properties in fractal systems with controlling λ. Furthermore, as shown in the above, it is easy to obtain an exact solution of this model for finite k. This makes the model relatively easy to handle, and thus it is suitable for investigating how the fractal structure affects the thermodynamic properties of the system.
As discussed in the previous section, it is difficult to discuss the thermodynamic limit from eqs. (3) and (4). Therefore, discussing finite size systems cannot reveal the thermodynamics of the Cantor-lattice Ising model itself, but it does provide us with a clue as to how finite-size fractals approach to thermodynamic fractals. This may be similar to the relationship between approximants and quasicrystals.
In this section, we investigate finite size systems of the Cantor lattice. In this section, we treat the cases with k = 2 and 3. These cases give N2 = 32 = 9 and N3 = 33 = 27 systems. The thermodynamic parameters are easily obtained using the partition function (4). As are well-known relations,26) the free energy fk(T, H, λ) and the entropy sk(T, H, λ) per one spin are obtained as
\begin{equation} f_{k}(T,H,\lambda) = -\frac{k_{\text{B}}T}{N_{k}}\log Z_{k}(T,H,\lambda), \end{equation} | (5) |
\begin{equation} s_{k}(T,H,\lambda) = -\frac{\partial}{\partial T}f_{k}(T,H,\lambda), \end{equation} | (6) |
\begin{equation} \Lambda_{k}(T,H,\lambda) \equiv - \frac{\partial}{\partial\lambda}f_{k}(T,H,\lambda). \end{equation} | (7) |
Figure 2(a) shows the temperature dependence of the entropy on the second generation k = 2, while Fig. 2(b) shows that on the third generation k = 3. The parameter λ is assumed as λ = 0.0, ±0.25, ±0.50, ±0.75, ±1.0. As shown in Fig. 2, the entropy per one spin has an asymptotic value kB log 2 for higher temperatures. On the other hand, in lower temperatures, the temperature dependence of the entropy obviously changes with increasing the parameter λ. This is because the correlation becomes stronger with increasing λ. Note that the case of λ = 0 is different from the other cases. In this case, the system corresponds to isolated three spin clusters, and then, the entropy becomes larger than the other cases because of its extensivity. Figure 3 shows how the entropy depends on λ. Both of (a) and (b) are calculated on third generation, under zero magnetic field (a) and under a uniform finite magnetic field H/J = 0.3 (b). The temperature kBT/J is assumed as kBT/J = 0.1, 0.2, 0.3, 0.4, 0.5, 0.6 in both of (a) and (b). It may be interesting to note that the entropy is an even function for λ under zero magnetic field, namely sk(T, 0, λ) = sk(T, 0, −λ) owing to the entropy describing just fluctuations (non-directed) as shown in Fig. 3(a). On the other hand, this even symmetry is broken under a finite magnetic field as shown in Fig. 3(b).
Temperature dependence of the entropy on the second generation (a) and the third generation (b). The external field H does not applied. The parameter λ is assumed as |λ| = 0.0, 0.25, 0.50, 0.75, 1.0.
λ dependence of the entropy without external fields (a) and under a finite external field (b). The external field is assumed as H/J = 0.3 in the right panel (b). Both (a) and (b) were calculated at the temperatures kBT/J = 0.1, 0.2, 0.3, 0.5, 0.6. These results were obtained on third generation models.
Figure 4 shows the temperature dependence of the λ-conjugate Λk(T, H, λ) defined in eq. (7) where the parameter λ is assumed as |λ| = 0.25, 0.5, 0.75, 1.0, and the magnetic field H does not applied. If we assume λ = 0, the λ-conjugate Λk(T, 0, λ) vanishes for all temperatures. It is monotonically increasing for all temperatures with increasing λ, and it vanishes in higher temperatures.
Temperature dependence of the λ-conjugate on third generation models. Here we assumed |λ| = 0.25, 0.50, 0.75, 1.0.
In the previous section, behaviors of the entropy and λ-conjugate in finite generations are exactly obtained by the iterating relation (3). However, this method does not help us approach to the thermodynamic limit k → ∞. In this section, we treat the Cantor-lattice Ising model by a renormalization method27,28) in order to approach to the thermodynamic limit.
The renormalization procedure is as follows. At first, we prepare a chain of three spins {σ1, σ2, σ3} with interaction J, namely $\mathcal{H}_{0} = - J(\sigma _{1}\sigma _{2} + \sigma _{2}\sigma _{3}) - H(\sigma _{1} + \sigma _{2} + \sigma _{3})$. And then, we trace out the parameter σ2 in the partition function $Z_{0} = \sum\nolimits_{\sigma _{2}}e^{ - \beta \mathcal{H}_{0}} $ and obtain an effective two spin system $\mathcal{H}_{1}(\sigma _{1},\sigma _{3}) = - k_{\text{B}}T\log Z_{0} =- J_{\text{eff}}^{(1)}\,\sigma _{1}\sigma _{3} - H_{\text{eff}}^{(1)}(\sigma _{1} + \sigma _{3})$. Secondly, three effective two spin systems are connected as
\begin{align} \mathcal{H}^{0}{}_{2}(\sigma_{1},\sigma_{2}) &= \mathcal{H}_{1}(\sigma_{1},\tau_{1}) + \mathcal{H}_{1}(\tau_{2},\tau_{3}) + \mathcal{H}_{1}(\tau_{4},\sigma_{2}) \\ &\quad + \lambda J(\tau_{1}\tau_{2} + \tau_{3}\tau_{4}). \end{align} | (8) |
\begin{equation} \mathcal{H}_{2}(\sigma_{1},\sigma_{2}) = -J_{\text{eff}}^{(2)}\sigma_{1}\sigma_{2} - H_{\text{eff}}^{(2)}(\sigma_{1} + \sigma_{2}). \end{equation} | (9) |
\begin{align} \mathcal{H}^{0}{}_3(\sigma_{1},\sigma_{2}) &= \mathcal{H}_{2}(\sigma_{1},\tau_{1}) + \mathcal{H}_{2}(\tau_{2},\tau_{3}) + \mathcal{H}_{2}(\tau_{4},\sigma_{2}) \\ &\quad+ \lambda^{2}J(\tau_{1}\tau_{2} + \tau_{3}\tau_{4}), \end{align} | (10) |
\begin{equation} \mathcal{H}_{3}(\sigma_{1},\sigma_{2}) = -J_{\text{eff}}^{(3)}\sigma_{1}\sigma_{2} - H_{\text{eff}}^{(3)}(\sigma_{1} + \sigma_{2}), \end{equation} | (11) |
\begin{equation} K_{k + 1} = \frac{1}{4}\log\frac{f_{1}(K_{k},h_{k},\lambda^{k - 1})f_{3}(K_{k},h_{k},\lambda^{k - 1})}{f_{2}^{2}(K_{k},h_{k},\lambda^{k - 1})}, \end{equation} | (12) |
\begin{equation} h_{k + 1} = \frac{1}{4}\log\frac{f_{1}(K_{k},h_{k},\lambda^{k - 1})}{f_{3}(K_{k},h_{k},\lambda^{k - 1})} \end{equation} | (13) |
\begin{equation} K_{1} = \frac{1}{2}\log\Big[e^{2h}\cosh 2K\sqrt {1 - \tanh^{2}2K\tanh^{2}h}\Big] \end{equation} | (14) |
\begin{equation} h_{1} = \frac{1}{4}\log\left[\frac{1 + \tanh 2K\tanh h}{1 - \tanh 2K\tanh h}\right]. \end{equation} | (15) |
Finally, in the thermodynamic limit, Cantor-lattice Ising model is reduced as an effective model
\begin{equation} \mathcal{H}_{\text{eff}} = -H_{\text{eff}}(\sigma_{1} + \sigma_{2}). \end{equation} | (16) |
Let us investigate the thermodynamic parameters, entropy and λ-conjugate, on the effective model (16). Using the effective field Heff, the entropy per one spin, s(K, h, λ) = S/NkB, and the normalized λ-conjugate Λ0(K, h, λ) = Λ/NJ are described as
\begin{align} s(K,h,\lambda) &= \log 2\cosh h_{\text{eff}}(K,h,\lambda) \\ &\quad - K\frac{\partial h_{\text{eff}}(K,h,\lambda)}{\partial K}\tanh h_{\text{eff}}(K,h,\lambda) \end{align} | (17) |
\begin{equation} \Lambda_{0}(K,h,\lambda) = \frac{1}{K}\frac{\partial h_{\text{eff}}(K,h,\lambda)}{\partial \lambda}\tanh h_{\text{eff}}(K,h,\lambda), \end{equation} | (18) |
Temperature dependence of the entropy on renormalized effective models under the magnetic field H/J = 0.1. Here we assumed λ = −1.0, −0.5, 0.0, 0.5, 1.0. Note that the horizonal axis is defined by the normalized inverse temperature K = βJ = J/kBT, and the range of entropy is different from Fig. 6.
Temperature dependence of the entropy on renormalized effective models under the magnetic field H/J = 0.5. Here we assumed λ = −1.0, −0.5, 0.0, 0.5, 1.0.
Temperature dependence of the λ-conjugate on renormalized models under the magnetic field H/J = 0.1 (a), and 0.5 (b). Here we assumed λ = −1.0, −0.5, 0.0, 0.5, 1.0. There was no significant difference in λ.
Generation comparison in λ-conjugates. Here we assumed the external field h/J = 0.1. The parameter λ is assumed as 0.3 in (a) while it is assumed as −0.3 in (b).
In the present study, we have introduced a simple toy model to investigate how the fractal structure affects such thermodynamic parameters as entropy. To achieve this purpose, we have focused on the Cantor-lattice Ising model. In this toy model, the parameter λ plays an important role to describe its fractality. And then, we introduced a thermodynamic parameter conjugate to the parameter λ, namely the λ-conjugate Λ = −∂F/∂λ. On the Cantor-lattice Ising model, entropy and λ-conjugate have been obtained by two methods. One is exact approach based on the transfer matrix method. The other is the renormalization method discussed in Sec. 4. Using both methods, we have confirmed behaviors of the thermodynamic parameters, which will give a simple picture on fractal systems. However, unfortunately, the results obtained by these two methods look different from each other. One possible reason is due to the difference of appropriate parameter for each method. Especially, present renormalization method does not work well without external fields. According to this difficulty, it is still unclear how the behaviors of thermodynamic parameters on finite-size systems are associated with the thermodynamic limit. This issue is left for future works. Finally, these results may be reproduced in optical lattice,30) where we can estimate the λ-conjugate from the λ dependence of its heat capacity.
This work was mainly supported by JSPS KAKENHI Grant Number JP19H05821. And one of the authors (YH) is also supported by JSPS KAKENHI Grant Number JP19K22117.
In this appendix, we show an interesting relation between the entropy and the λ-conjugate. Figure A1 shows this relation of the third generation. The entropy sk(T, 0, λ) and the λ-conjugate Λk(T, 0, λ) are related to each other through two parameters T and λ. Then, a range of the values that they can take in the parameter space (Λk, sk) is limited even though T and λ are independently defined. As shown in Fig. A1, the entropy cannot take smaller values when the λ-conjugate takes appropriate regions (|Λk(T, 0, λ)| $ \lesssim $ 0.25) in a finite-size Cantor-lattice Ising model. We consider the reason for this as follows. The λ-conjugate is defined as eq. (7) and it means the correlation between adjacent (k − 1)-th generation clusters in the k-th generation clusters. Thus, the small λ-conjugate corresponds to the less correlated models. Especially, the completely isolated system corresponds to Λ3(T, 0, λ) = 0. In such a small Λ3(T, 0, λ) case, the minimum entropy is larger than the other cases. In other words, the minimum entropy is a decreasing function of λ-conjugate. Furthermore, it is suggested that a discontinuous change will appear on the minimum entropy at Λ3(T, 0, λ) = ±0.25.
Parametric plot of the entropy vs. λ-conjugate on third generation models.