betaFeedback

WorksPhysics and Astronomy

The Lee-Yang property of the Blume-Emery-Griffiths model

Abstract

The Lieb-Sokal theorem provides a tool for establishing the Lee-Yang property of the partition function of a spin model if the corresponding single-spin partition function has this property and the interaction term in its Hamiltonian is of the form −∑JijSiSj-\sum J_{ij} S_iS_j with Jij≥0J_{ij}\geq 0. According to the result of this paper, the spin S=1S=1 model with the Hamiltonian −∑(JijSiSj+KijSi2Sj2)+∑ΔSi2-\sum (J_{ij} S_iS_j + K_{ij}S_i^2S_j^2) + \sum ΔS_i^2 has the Lee-Yang property for all Kij∈RK_{ij}\in \mathds{R} and Δ>0Δ>0, whenever each JijJ_{ij} exceeds an explicitly computed bound C(β,Δ,Kij)>0C(β, Δ, K_{ij})>0, such that C(β,Δ,Kij)→+∞C(β, Δ, K_{ij})\to + \infty as β→0β\to 0. The latter suggests that the Lee-Yang property of this model may hold only for sufficiently low temperatures.

Introduction

The celebrated Lee-Yang theorem states that the continuation to complex values of the external magnetic field hh of the partition function ZNβZ_N^\beta of the collection of NN interacting Ising spins has purely imaginary zeros if the interaction is ferromagnetic; see [3, 11, 14, 15] and the literature quoted therein. As a Laurent polynomial in z=eβhz=e^{\beta h} (β\beta being the inverse temperature), symmetric with respect to the change z→z−1z\to z^{-1}, ZNβZ_N^\beta is an even exponential-type entire function. Combining these facts, cf. [10], one readily concludes that it can be written in the form

ZNβ(h)=Zβ(0)∏j=1∞(1+γjh2),(1)Z_N^\beta(h)=Z^\beta(0)\prod_{j=1}^{\infty}(1+\gamma_jh^2), \tag*{(1)}

where the collection {γj}\{\gamma_j\} – in this case, β\beta- and NN-dependent parameters – satisfies

0<γj+1≤γj,∑j=1∞γj<∞.(2)0<\gamma_{j+1}\le\gamma_j,\qquad\sum_{j=1}^{\infty}\gamma_j<\infty. \tag*{(2)}

Consequently, ZNβZ_N^\beta is said to have the Lee-Yang property. With increasing NN, the ‘first Lee-Yang zero’ i/γ1i/\sqrt{\gamma_1} moves towards the origin; see [7]. In the limit N→∞N\to\infty, it may reach the origin and thus produce a singularity of the free energy density at h=0h=0. In view of this, the Lee-Yang theorem contributes to the collection of powerful methods for studying phase transitions in Ising and similar spin models.

Obviously, (1.1) holds for N=1N=1, where Z1β(h)=2cosh⁡βhZ_1^\beta(h)=2\cosh\beta h. A natural generalization of the Lee-Yang theorem might be to claim the validity of (1.1) and (1.2) in the case of more general single-spin distributions that possess the Lee-Yang property, including those for ‘unbounded’ spins, which are finite measures on the real line. Newman [14], and then Lieb and Sokal [11], did this. The key aspect of the latter work is the following

@references nggun representation

ZNβ(h)=∫RNexp⁡(−βHN(S,h))∏i=1Nχ(dSi)=∫RNexp⁡(∑1≤i<j≤NβJijSiSj+∑i=1NβhSi)∏i=1Nχ(dSi)=[exp⁡(∑1≤i<j≤NβJijDiDj)∏i=1Nϕ(xi)]all xi=βh,Di=∂∂xi(3)\begin{aligned} Z_N^\beta(h)=\int_{\mathbb{R}^N}\exp\left(-\beta H_N(S,h)\right)\prod_{i=1}^{N}\chi(\mathrm{d}S_i) \\ &=\int_{\mathbb{R}^N}\exp\left(\sum_{1\le i<j\le N}\beta J_{ij}S_iS_j+\sum_{i=1}^{N}\beta hS_i\right)\prod_{i=1}^{N}\chi(\mathrm{d}S_i) \\ &=\left[\exp\left(\sum_{1\le i<j\le N}\beta J_{ij}D_iD_j\right)\prod_{i=1}^{N}\phi(x_i)\right]_{\text{all }x_i=\beta h},\qquad D_i=\frac{\partial}{\partial x_i} \tag*{(3)} \end{aligned}

where ϕ\phi is related to the aforementioned finite measure χ\chi by

ϕ(x)=∫Rexσχ(dσ).(4)\phi(x)=\int_{\mathbb{R}}e^{x\sigma}\chi(\mathrm{d}\sigma). \tag*{(4)}

The infinite-order differential operator in the last line of (3) can be defined for all Jij≥0J_{ij}\ge0 if ϕ\phi is an entire function of order less than two, or of order two and of minimal type. The Lee-Yang property of χ\chi means that ϕ\phi can be written in the form

ϕ(x)=ϕ^(x2)=ϕ(0)∏j=1∞(1+γjx2),(5)\phi(x)=\hat{\phi}(x^2)=\phi(0)\prod_{j=1}^{\infty}(1+\gamma_jx^2), \tag*{(5)}

where the collection {γj}\{\gamma_j\} satisfies (2), and hence ϕ^\hat{\phi} is a Laguerre entire function; see [2, 6] and also [8], where examples of single-spin measures possessing the Lee-Yang property are given.

According to [11], ZNβZ_N^\beta given in (3) has the Lee-Yang property for all β>0\beta>0 whenever Jij≥0J_{ij}\ge0 for all i,ji,j, and ϕ\phi is as in (5). The latter is a sufficient condition; it leaves open the possibility that ZNβZ_N^\beta has the Lee-Yang property for N≥2N\ge2, at least for some β\beta, even if this condition is not satisfied. In [9], such a possibility was shown to occur for the Blume-Capel model. Here we extend this result to the Blume-Emery-Griffiths (BEG) model; see [1, 5, 12, 13, 16] for more details on this model, its versions and applications.

The prototypical model that we study corresponds to the following partition function

ZN(x)=∫RNexp⁡(∑1≤i<j≤N[JijSiSj+KijSi2Sj2]+∑i=1NxSi)∏i=1Nχθ(dSi),(6)Z_N(x)=\int_{\mathbb{R}^N}\exp\left(\sum_{1\le i<j\le N}\left[J_{ij}S_iS_j+K_{ij}S_i^2S_j^2\right]+\sum_{i=1}^{N}xS_i\right)\prod_{i=1}^{N}\chi_\theta(\mathrm{d}S_i), \tag*{(6)}

with

χθ=2θδ0+δ−1+δ1,θ>0,(7)\chi_\theta=2\theta\delta_0+\delta_{-1}+\delta_1,\qquad\theta>0, \tag*{(7)}

where δa\delta_a is Dirac’s measure centered at a∈Ra\in\mathbb{R}. This model is related to the BEG model; see the next section. Note that the function corresponding (as in (4)) to χθ\chi_\theta given in (7) is

ϕθ(x)=2(θ+cosh⁡x).(8)\phi_\theta(x)=2(\theta+\cosh x). \tag*{(8)}

For θ>1\theta>1, it does not vanish on the imaginary axis, where ∣cosh⁡x∣≤1|\cosh x|\le1. Then ∏j=1Nϕθ(xj)\prod_{j=1}^{N}\phi_\theta(x_j) does not satisfy the sufficient condition mentioned above. Another essential obstacle to proving the Lee-Yang property of (6) is the quadrupole interaction term ∑KijSi2Sj2\sum K_{ij}S_i^2S_j^2, which is incompatible with the technique of [11] or any other theory of this kind. In this paper, we prove that ZNZ_N given in (6) has the Lee-Yang property for all θ>0\theta>0 and Kij∈RK_{ij}\in\mathbb{R} if the interaction intensities Jij≥0J_{ij}\ge0 are large enough. To the best of our knowledge, this is the first result of this kind for the Blume-Emery-Griffiths model.

The Result

We begin by preparing the basic tool that we use to prove the main statement.

The Lieb-Sokal theorem

First, we formulate the relevant results of [11] in the form adapted to the context. To do this, we will need stable polynomials. For a given N∈NN \in\mathbb{N}, a complex polynomial, PP, is said to be stable if P(z1,…,zN)≠0P(z_1,\ldots,z_N) \ne0 whenever ℜzj>0\Re z_j > 0 for all j=1,…,Nj=1,\ldots,N; see [2, 15] and the literature quoted in these publications.

Let FN\mathcal{F}_N be the set of exponential-type entire functions defined on CN\mathbb{C}^N, which are: (a) not identically zero; (b) obtained as limits of the sequences of stable polynomials taken in the topology of uniform convergence on compact subsets of CN\mathbb{C}^N. Clearly, the point-wise product F1F2F_1F_2 is in FN\mathcal{F}_N if both F1F_1 and F2F_2 are there. By Hurwitz’ theorem, it follows that each F∈FNF \in\mathcal{F}_N verifies F(z1,…,zN)≠0F(z_1,\ldots,z_N) \ne0 whenever ℜzj>0\Re z_j > 0 for all j=1,…,Nj=1,\ldots,N. At the same time, by the Laguerre–Pólya theorem, an even exponential-type entire function that does not vanish at zero is in F1\mathcal{F}_1 if and only if it is as in (1.5); see [6 pages 8–23]. If F∈FNF \in\mathcal{F}_N is even, i.e., F(x1,…,xN)=F(−x1,…,−xN)F(x_1,\ldots,x_N)=F(-x_1,\ldots,-x_N), and F(0,…,0)≠0F(0,\ldots,0) \ne0, then F(x,…,x)F(x,\ldots,x) is as in (1.5).

For G∈FNG \in\mathcal{F}_N, define

F(x1,…,xN)=exp⁡(∑1≤i<j≤NAijDiDj)G(x1,…,xN),Di=∂∂xi,F^(x1,…,xN)=∏1≤i<j≤N(1+AijDiDj)G(x1,…,xN).(9)\begin{aligned} F(x_1,\ldots,x_N) &= \exp\left(\sum_{1\le i<j\le N} A_{ij}D_iD_j\right)G(x_1,\ldots,x_N), \qquad D_i=\frac{\partial}{\partial x_i},\\ \widehat{F}(x_1,\ldots,x_N) &= \prod_{1\le i<j\le N}(1+A_{ij}D_iD_j)G(x_1,\ldots,x_N). \tag*{(9)} \end{aligned}

Proposition 2.1. [11 Propositions 2.6 and 2.7, page 161] If Aij≥0A_{ij} \ge0 for all $1 \le i < j \le N$, then for each G∈FNG \in\mathcal{F}_N, both functions FF and F^\widehat{F} given in (2.1) are in FN\mathcal{F}_N.

For θ≤1\theta\le1, by solving ϕθ(x)=0\phi_\theta(x)=0, see (1.8), we obtain ex=θ±i1−θ2e^x=\theta\pm i\sqrt{1-\theta^2}. Therefore, ϕθ\phi_\theta is an exponential-type entire function of the form (1.5), i.e., ϕθ∈F1\phi_\theta\in\mathcal{F}_1. Then ∏j=1Nϕθ(xj)\prod_{j=1}^N\phi_\theta(x_j) is in FN\mathcal{F}_N, and the direct corollary of Proposition 2.1 is the following statement.

Proposition 2.2. [11 Corollary 3.3, page 165] Let ϕθ\phi_\theta in (1.8) correspond to θ≤1\theta\le1, Jij≥0J_{ij} \ge0 and Kij=0K_{ij}=0 for all i,j=1,…,Ni,j=1,\ldots,N. Then ZNZ_N given in (1.6) has the Lee–Yang property.

As mentioned above, we prove that ZNZ_N has the Lee–Yang property for all θ>0\theta>0 and Kij∈RK_{ij}\in\mathbb{R} whenever each positive JijJ_{ij} is greater than some explicit bound defined by θ\theta and KijK_{ij} with the same i,ji,j.

The statement

We begin by introducing the following functions

κ(v)=ln⁡(v+v2−1),v≥1,(10)\kappa(v)=\ln\left(v+\sqrt{v^2-1}\right), \qquad v\ge1, \tag*{(10)}
λ(w)=ln⁡(1+2e−w+2e−wew+1),w>0.\lambda(w)=\ln\left(1+2e^{-w}+2e^{-w}\sqrt{e^w+1}\right), \qquad w>0.

Note that κ(1)=0\kappa(1)=0 and cosh⁡κ(v)=v\cosh\kappa(v)=v. Next, we set

Λ(w)={κ(e−w),w≤0,λ(w),w>0.(11)\Lambda(w)= \begin{cases} \kappa(e^{-w}), & w\le0,\\ \lambda(w), & w>0. \end{cases} \tag*{(11)}

Now for χθ\chi_\theta as in (7), we define

FN(x1,…,xN)=∫RNexp⁡(∑1≤i<j≤N[JijSiSj+KijSi2Sj2]+∑i=1NxiSi)∏i=1Nχθ(dSi)=exp⁡(∑1≤i<j≤N[JijDiDj+KijDi2Dj2])∏i=1Nϕθ(xi),Di=∂∂xi,(12)\begin{aligned} F_N(x_1,\ldots,x_N)=\int_{\mathbb{R}^N}\exp\left(\sum_{1\leq i<j\leq N}\left[\mathcal{J}_{ij}S_iS_j+\mathcal{K}_{ij}S_i^2S_j^2\right]+\sum_{i=1}^{N}x_iS_i\right)\prod_{i=1}^{N}\chi_\theta(dS_i) \\ &=\exp\left(\sum_{1\leq i<j\leq N}\left[\mathcal{J}_{ij}D_iD_j+\mathcal{K}_{ij}D_i^2D_j^2\right]\right)\prod_{i=1}^{N}\phi_\theta(x_i),\qquad D_i=\frac{\partial}{\partial x_i}, \tag*{(12)} \end{aligned}

where the interaction matrices (Jij)(\mathcal{J}_{ij}) and (Kij)(\mathcal{K}_{ij}) are the same as in (6). The next statement is the main result of this work.

Theorem 2.3. FNF_N given in (12) is in FN\mathcal{F}_N, and thus the partition function ZN(x)=FN(x,…,x)\mathcal{Z}_N(\mathbf{x})=F_N(x,\ldots,x), has the Lee-Yang property if the following holds:

(a) For θ∈(0,1]\theta\in(0,1]: Jij≥Λ(Kij)\mathcal{J}_{ij}\geq\Lambda(\mathcal{K}_{ij}) for each pair i,ji,j.

(b) For θ>1\theta>1: N=2MN=2M, and there exists a division, {1,…,2M}=⋃k=1Mϑk\{1,\ldots,2M\}=\bigcup_{k=1}^{M}\vartheta_k, into disjoint two-element subsets (dimers) ϑk={ik,jk}\vartheta_k=\{i_k,j_k\} such that Jij≥Λ(Kij)+ϵijκ(θ2)\mathcal{J}_{ij}\geq\Lambda(\mathcal{K}_{ij})+\epsilon_{ij}\kappa(\theta^2) for each pair i,ji,j. Here κ\kappa and Λ\Lambda are as in (10) and (11), and ϵij=1\epsilon_{ij}=1 if {i,j}=ϑk\{i,j\}=\vartheta_k for some kk, and ϵij=0\epsilon_{ij}=0 otherwise.

Note that in items (a) and (b) of Theorem 2.3, each Kij\mathcal{K}_{ij} can be positive or negative. This theorem describes also the Blume-Capel model, studied in [9]. It corresponds to Kij=0\mathcal{K}_{ij}=0, i.e., Λ(Kij)=0\Lambda(\mathcal{K}_{ij})=0; see (11). The condition N=2MN=2M imposed in item (b) is not burdensome. In applications, it suffices to establish the Lee-Yang property of the partition function only for even NN. The assumed positivity of (Jij)(\mathcal{J}_{ij}) means that the graph on {1,…,2M}\{1,\ldots,2M\} with the edge set EN={{i,j}:Jij≥Λ(Kij)+κ(θ2)}E_N=\{\{i,j\}:\mathcal{J}_{ij}\geq\Lambda(\mathcal{K}_{ij})+\kappa(\theta^2)\} is perfectly matchable (or admits ‘dimer covering’ in the terminology of [4]). The latter property is typical for models defined on graphs, such as Zd\mathbb{Z}^d, with sufficiently strong nearest-neighbor pair interactions. Similar conditions were imposed in [9], where they are discussed in more detail.

The proof

As was already mentioned, an obstacle for us in proving the Lee-Yang property of ZN\mathcal{Z}_N is the quadrupole interaction term in the exponent in (12). We begin by deriving useful formulas that will allow us to exclude it.

For a fixed κ∈R\kappa\in\mathbb{R}, let us find the smallest τ≥0\tau\geq0 such that the following equation

eτσ+κσ2=(1+aσ)(1+bσ),σ=±1,(13)e^{\tau\sigma+\kappa\sigma^2}=(1+a\sigma)(1+b\sigma),\qquad\sigma=\pm1, \tag*{(13)}

has solutions a,b≥0a,b\geq0. Taking into account the two possibilities for σ\sigma, we transform (13) into the following system of equations

cosh⁡τeκ=1+abandsinh⁡τeκ=a+b,\cosh\tau e^\kappa=1+ab\qquad\text{and}\qquad\sinh\tau e^\kappa=a+b,

with aa and bb as unknown quantities. Its solution is

a=12[sinh⁡τeκ+(cosh⁡τeκ−2)2−e2κ],a=\frac{1}{2}\left[\sinh\tau e^\kappa+\sqrt{(\cosh\tau e^\kappa-2)^2-e^{2\kappa}}\right],
b=12[sinh⁡τeκ−(cosh⁡τeκ−2)2−e2κ].(14)b=\frac{1}{2}\left[\sinh\tau e^\kappa-\sqrt{(\cosh\tau e^\kappa-2)^2-e^{2\kappa}}\right]. \tag*{(14)}

For κ≤0\kappa\leq0, the positivity of bb implies cosh⁡τeκ≥1\cosh\tau e^\kappa\geq1. The least τ≥0\tau\geq0 that satisfies this demand, as well as the following one

∣cosh⁡τeκ−2∣≥eκ,(15)\left|\cosh\tau e^\kappa-2\right|\geq e^\kappa, \tag*{(15)}

– since aa and bb should be real – corresponds to

cosh⁡τ=e−ϰ,which yields a=1−e2ϰ, b=0.(16)\cosh\tau= e^{-\varkappa}, \qquad\text{which yields } a = \sqrt{1-e^{2\varkappa}},\ b=0. \tag*{(16)}

In particular, a=b=0a=b=0 for ϰ=0\varkappa=0. For ϰ>0\varkappa>0, the condition in (15) takes the form cosh⁡τ≥2e−ϰ+1\cosh\tau\ge2e^{-\varkappa}+1, which also guaranties b≥0b\ge0. The least such τ≥0\tau\ge0 corresponds to the equality, which yields

eτ=1+2e−ϰ+2e−ϰeϰ+1,a=b=eϰ+1.(17)e^\tau=1+2e^{-\varkappa}+2e^{-\varkappa}\sqrt{e^\varkappa+1}, \qquad a=b=\sqrt{e^\varkappa+1}. \tag*{(17)}

Both versions (16) and (17) can be summarized in the following

τ=Λ(ϰ);(18)\tau=\Lambda(\varkappa); \tag*{(18)}

see (11). Now we define, cf. (16) and (17),

Aij={1−e2Kijfor Kij≤0,1+eKijfor Kij>0,(19)A_{ij}= \begin{cases} \sqrt{1-e^{2\mathcal{K}_{ij}}} & \text{for }\mathcal{K}_{ij}\le0,\\ \sqrt{1+e^{\mathcal{K}_{ij}}} & \text{for }\mathcal{K}_{ij}>0, \end{cases} \tag*{(19)}

and Bij=0B_{ij}=0 for Kij≤0\mathcal{K}_{ij}\le0, Bij=AijB_{ij}=A_{ij} for Kij>0\mathcal{K}_{ij}>0. By means of (14), we can write

exp⁡(JijSiSj+KijSi2Sj2)=(1+AijSiSj)(1+BijSiSj)exp⁡([J~ij+ϵijκ~(θ2)]SiSj),(20)\begin{aligned} \exp\left(\mathcal{J}_{ij}S_iS_j+\mathcal{K}_{ij}S_i^2S_j^2\right) \\ &=\left(1+A_{ij}S_iS_j\right)\left(1+B_{ij}S_iS_j\right)\exp\left(\left[\widetilde{\mathcal{J}}_{ij}+\epsilon_{ij}\widetilde{\kappa}(\theta^2)\right]S_iS_j\right), \tag*{(20)} \end{aligned}
J~ij=Jij−Λ(Kij)−ϵijκ~(θ2),κ~(θ2)={κ(θ2)for θ≥1,0otherwise.\widetilde{\mathcal{J}}_{ij}=\mathcal{J}_{ij}-\Lambda(\mathcal{K}_{ij})-\epsilon_{ij}\widetilde{\kappa}(\theta^2), \qquad \widetilde{\kappa}(\theta^2)= \begin{cases} \kappa(\theta^2) & \text{for }\theta\ge1,\\ 0 & \text{otherwise}. \end{cases}

Then by (12) and (20), we can decompose

FN(x1,…,xN)=∏1≤i<j≤N(1+AijDiDj)(1+BijDiDj)GN(x1,…,xN),(21)F_N(x_1,\ldots,x_N)=\prod_{1\le i<j\le N}\left(1+A_{ij}D_iD_j\right)\left(1+B_{ij}D_iD_j\right)G_N(x_1,\ldots,x_N), \tag*{(21)}
GN(x1,…,xN)=exp⁡(∑1≤i<j≤N[J~ij+ϵijκ~(θ2)]DiDj)∏i=1Nϕθ(xi).G_N(x_1,\ldots,x_N)=\exp\left(\sum_{1\le i<j\le N}\left[\widetilde{\mathcal{J}}_{ij}+\epsilon_{ij}\widetilde{\kappa}(\theta^2)\right]D_iD_j\right)\prod_{i=1}^{N}\phi_\theta(x_i).

According to our assumptions, J~ij≥0\widetilde{\mathcal{J}}_{ij}\ge0 for all pairs i,ji,j. The same is true for AijA_{ij} and BijB_{ij}; see (19). For θ≤1\theta\le1, we have ϕθ∈F1\phi_\theta\in\mathcal{F}_1. Then the proof of claim (a) of Theorem 2.3 follows by (21) and Proposition 2.1.

In the proof of claim (b), we mostly follow the way elaborated in [9]. Recall that we assume N=2MN=2M and the existence of the decomposition {1,2,…,N}=⋃k=1Mϑk\{1,2,\ldots,N\}=\bigcup_{k=1}^{M}\vartheta_k with the properties mentioned herein. Consequently, we further decompose GNG_N in the following way

GN(x1,…,xN)=exp⁡(∑1≤i<j≤NJ~ijDiDj)HN(x1,…,xN),(22)G_N(x_1,\ldots,x_N)=\exp\left(\sum_{1\le i<j\le N}\widetilde{\mathcal{J}}_{ij}D_iD_j\right)H_N(x_1,\ldots,x_N), \tag*{(22)}
HN(x1,…,xN)=exp⁡(∑k=1Mκ(θ2)DikDjk)∏i=1Nϕθ(xi)=∏k=1MΦ(xik,xjk).H_N(x_1,\ldots,x_N)=\exp\left(\sum_{k=1}^{M}\kappa(\theta^2)D_{i_k}D_{j_k}\right)\prod_{i=1}^{N}\phi_\theta(x_i)=\prod_{k=1}^{M}\Phi(x_{i_k},x_{j_k}).

Here

Φ(x,y)=exp⁡(κ(θ2)DxDy)ϕθ(x)ϕθ(y).\Phi(x,y)=\exp\left(\kappa(\theta^2)D_xD_y\right)\phi_\theta(x)\phi_\theta(y).

To prove claim (b), we show that Φ∈F2\Phi\in\mathcal{F}_2 even though φθ\varphi_\theta is not in F1\mathcal{F}_1 for θ>1\theta> 1; see (8). To this end, we write

Φ(x,y)=4(θ2+θ(cosh⁡x+cosh⁡y)+[cosh⁡(κ(θ2)DxDy)+sinh⁡(κ(θ2)DxDy)]cosh⁡xcosh⁡y)\Phi(x,y)=4\left(\theta^2+\theta(\cosh x+\cosh y)+\left[\cosh\left(\kappa(\theta^2)D_xD_y\right)+\sinh\left(\kappa(\theta^2)D_xD_y\right)\right]\cosh x\cosh y\right)

and take into account that

cosh⁡(κ(θ2)DxDy)cosh⁡xcosh⁡y=cosh⁡(κ(θ2))cosh⁡xcosh⁡y,\cosh\left(\kappa(\theta^2)D_xD_y\right)\cosh x\cosh y=\cosh\left(\kappa(\theta^2)\right)\cosh x\cosh y,
sinh⁡(κ(θ2)DxDy)cosh⁡xcosh⁡y=sinh⁡(κ(θ2))sinh⁡xsinh⁡y,\sinh\left(\kappa(\theta^2)D_xD_y\right)\cosh x\cosh y=\sinh\left(\kappa(\theta^2)\right)\sinh x\sinh y,

as well as that θ2=cosh⁡κ(θ2)\theta^2=\cosh\kappa(\theta^2), see (10). Thereafter, we arrive at the following formula

Φ(x,y)=4(2θcosh⁡ucosh⁡v+eκ(θ2)cosh⁡2u+e−κ(θ2)cosh⁡2v)=4eκ(θ2)φ+(x,y)φ−(x,y),u=x+y2,v=x−y2,(23)\begin{aligned} \Phi(x,y)=4\left(2\theta\cosh u\cosh v+e^{\kappa(\theta^2)}\cosh^2u+e^{-\kappa(\theta^2)}\cosh^2v\right) \\ &=4e^{\kappa(\theta^2)}\varphi_+(x,y)\varphi_-(x,y),\qquad u=\frac{x+y}{2},\quad v=\frac{x-y}{2}, \tag*{(23)} \end{aligned}

where

φ±(x,y)=cosh⁡x+y2+ω±cosh⁡x−y2(24)\varphi_\pm(x,y)=\cosh\frac{x+y}{2}+\omega_\pm\cosh\frac{x-y}{2} \tag*{(24)}

and

ω±=e−κ(θ2)[θ±θ2−1]=exp⁡[−κ(θ2)±κ(θ)].\omega_\pm=e^{-\kappa(\theta^2)}\left[\theta\pm\sqrt{\theta^2-1}\right]=\exp\left[-\kappa(\theta^2)\pm\kappa(\theta)\right].

For θ>1\theta>1, one sees that κ(θ2)>κ(θ)\kappa(\theta^2)>\kappa(\theta), which means that both ω±\omega_\pm are in (0,1)(0,1). Consequently,

ν±:=1−ω±1+ω±∈(0,1),\nu_\pm:=\frac{1-\omega_\pm}{1+\omega_\pm}\in(0,1),

and hence, see (24),

φ±(x,y)=(1+ω±)[cosh⁡x2cosh⁡y2+ν±sinh⁡x2sinh⁡y2].\varphi_\pm(x,y)=(1+\omega_\pm)\left[\cosh\frac{x}{2}\cosh\frac{y}{2}+\nu_\pm\sinh\frac{x}{2}\sinh\frac{y}{2}\right].

By (23), Φ∈F2\Phi\in\mathcal{F}_2 if both φ±\varphi_\pm have this property. Clearly, both cosh⁡(x/2)\cosh(x/2) and sinh⁡(x/2)\sinh(x/2) can be approximated by the sequences of stable polynomials {cn(x)}\{c_n(x)\} and {sn(x)}\{s_n(x)\}, respectively. Then φ±∈F2\varphi_\pm\in\mathcal{F}_2 if

γn(x,y)=cn(x)cn(y)+νsn(x)sn(y),ν∈(0,1),\gamma_n(x,y)=c_n(x)c_n(y)+\nu s_n(x)s_n(y),\qquad\nu\in(0,1),

is a stable polynomial for all sufficiently large nn, i.e., it does not vanish for all ℜx>0\Re x>0, ℜy>0\Re y>0 and such nn. The latter is equivalent to

cn(x)sn(x)≠−νsn(y)cn(y),ℜx>0, ℜy>0.(25)\frac{c_n(x)}{s_n(x)}\ne-\nu\frac{s_n(y)}{c_n(y)},\qquad\Re x>0,\ \Re y>0. \tag*{(25)}

For ℜy>0\Re y>0, we set

sn(y)cn(y)=pn(y)−1pn(y)+1=ρneiαn,ρn>0.(26)\frac{s_n(y)}{c_n(y)}=\frac{p_n(y)-1}{p_n(y)+1}=\rho_ne^{i\alpha_n},\qquad\rho_n>0. \tag*{(26)}

Here pn(y)=(cn(y)+sn(y))/(cn(y)−sn(y))→eyp_n(y)=(c_n(y)+s_n(y))/(c_n(y)-s_n(y))\to e^y for ℜy>0\Re y>0. By (26), it follows that

pn(y)=1+ρneiαn1−ρneiαn,hence∣pn(y)∣2=1+ρn2+2ρncos⁡αn1+ρn2−2ρncos⁡αn.(27)p_n(y)=\frac{1+\rho_ne^{i\alpha_n}}{1-\rho_ne^{i\alpha_n}},\qquad\text{hence}\qquad|p_n(y)|^2=\frac{1+\rho_n^2+2\rho_n\cos\alpha_n}{1+\rho_n^2-2\rho_n\cos\alpha_n}. \tag*{(27)}

Assume γn(x,y)=0\gamma_n(x,y)=0, which corresponds to the equality in (25), and hence, cf. (26), to the following

cn(x)sn(x)=−νρneiαn.\frac{c_n(x)}{s_n(x)}=-\nu\rho_ne^{i\alpha_n}.

Similarly to (27) we then get

∣pn(x)∣2=1+ν2ρn2−2νρncos⁡αn1+ν2ρn2+2νρncos⁡αn.(28)\lvert p_n(x)\rvert^2 = \frac{1+\nu^2\rho_n^2-2\nu\rho_n\cos\alpha_n}{1+\nu^2\rho_n^2+2\nu\rho_n\cos\alpha_n}. \tag*{(28)}

By construction, ∣pn(x)∣2→e2ℜx\lvert p_n(x)\rvert^2 \to e^{2\Re x} and ∣pn(y)∣2→e2ℜy\lvert p_n(y)\rvert^2 \to e^{2\Re y}, point-wise on ℜx>0\Re x>0, ℜy>0\Re y>0. Hence, the right-hand sides of (27) and (28) are both strictly greater than one whenever n>nˉn>\bar n for some threshold nˉ\bar n dependent on xx and yy. Clearly, this is not possible for any value of cos⁡αn\cos\alpha_n. The obtained contradiction implies (25), which completes the proof.

The Blume-Emery-Griffiths Model

The Blume-Emery-Griffiths (BEG) model was introduced in [1] to describe the λ\lambda-transition and phase separation in He3-He4 mixtures. In general form, its Hamiltonian reads as follows

HN=−∑1≤i<j≤N(JijSiSj+KijSi2Sj2)+Δ∑i=1NSi2−∑i=1NhSi.(29)H_N=-\sum_{1\leq i<j\leq N}\left(J_{ij}S_iS_j+K_{ij}S_i^2S_j^2\right)+\Delta\sum_{i=1}^{N}S_i^2-\sum_{i=1}^{N}hS_i. \tag*{(29)}

Here, each spin SiS_i takes values 0,±10,\pm1, JijJ_{ij} and KijK_{ij} are bilinear and biquadratic (quadrupole) interaction intensities, and Δ\Delta is a crystal anisotropy field. Typically [1, 5, 16], the spins are indexed by the vertices of a graph, e.g., a crystal Zd\mathbb{Z}^d, in which case the (nearest-neighbor) interaction intensities JJ and KK are assigned to the graph edges.

Along with the aforementioned physical systems, this model is used in the theory of many other substances and phenomena; see [5, 16] and the publications quoted therein. It has a variety of significant properties that are essentially different from those of the classical S=±1S=\pm1 Ising model. The difference is especially spectacular if the quadrupole intensities satisfy Kij≤0K_{ij}\leq0 for all i,ji,j; see [5, 12, 16]. Its partition function ZNβZ_N^\beta, cf. (3), can be obtained from that in (6) by the formula ZNβ(h)=2−βNΔZN(βh)Z_N^\beta(h)=2^{-\beta N\Delta}Z_N(\beta h), in which the corresponding parameters and those of (29) are related to each other as follows

Jij=βJij,Kij=βKij,θ=eβΔ/2.(30)J_{ij}=\beta J_{ij},\qquad K_{ij}=\beta K_{ij},\qquad\theta=e^{\beta\Delta/2}. \tag*{(30)}

According to Theorem 2.3, in the case of nearest-neighbor interactions, the partition function ZNβ(h)Z_N^\beta(h) of the BEG model (with N=2MN=2M) has the Lee-Yang property if the model parameters and β\beta satisfy

βJ≥Λ(βK)+κ~(e2βΔ/4);(31)\beta J\geq\Lambda(\beta K)+\widetilde{\kappa}\left(e^{2\beta\Delta/4}\right); \tag*{(31)}

see (11), (20) and (30). Here, KK can be positive or negative. For K=0K=0, this condition coincides with the corresponding condition derived in [9]. For K<0K<0 and β>Δ−1ln⁡2\beta>\Delta^{-1}\ln2, (31) turns into the following bound, cf. (10) and (11),

βJ≥β(∣K∣+2Δ)−ε(β),(32)\beta J\geq\beta(|K|+2\Delta)-\varepsilon(\beta), \tag*{(32)}

where

ε(β)=ln⁡4−ln⁡(1+1−e−2β∣K∣)−ln⁡(1+1−e−4βΔ/16)=14(e−2β∣K∣+116e−4βΔ)+o(max⁡{e−2β∣K∣,e−4βΔ}),β→+∞.\begin{aligned} \varepsilon(\beta)&=\ln4-\ln\left(1+\sqrt{1-e^{-2\beta|K|}}\right)-\ln\left(1+\sqrt{1-e^{-4\beta\Delta}/16}\right)\\ &=\frac{1}{4}\left(e^{-2\beta|K|}+\frac{1}{16}e^{-4\beta\Delta}\right)+o\left(\max\left\{e^{-2\beta|K|},e^{-4\beta\Delta}\right\}\right),\qquad\beta\to+\infty. \end{aligned}

Therefore, the partition function of the BEG model ZNβZ_N^\beta has the Lee-Yang property for all β≥Δ−1ln⁡2\beta\geq\Delta^{-1}\ln2 whenever J≥∣K∣+2ΔJ\geq|K|+2\Delta. For β<Δ−1ln⁡2\beta<\Delta^{-1}\ln2, the condition in (31) takes the form

J−∣K∣≥β−1ln⁡(1+1−e−2β∣K∣).(33)J-|K|\geq\beta^{-1}\ln\left(1+\sqrt{1-e^{-2\beta|K|}}\right). \tag*{(33)}

The latter cannot be satisfied for all β∈(0,Δ−1ln⁡2)\beta\in(0,\Delta^{-1}\ln2) since the right-hand side is unbounded on this interval.

In the case K>0K>0, the large β\beta behavior of Λ(βK)\Lambda(\beta K) is completely different: by (10) and (11), one sees that Λ(βK)≤4e−βK/2\Lambda(\beta K) \le4e^{-\beta K/2}. Therefore, instead of (32) one gets the following condition

βJ≥2βΔ−ln⁡4+ln⁡(1+1−e−4βΔ/16)+4e−βK/2.\beta J \ge2\beta\Delta-\ln4+\ln\left(1+\sqrt{1-e^{-4\beta\Delta}/16}\right)+4e^{-\beta K/2}.

For β<Δ−1ln⁡2\beta<\Delta^{-1}\ln2, instead of (33) one obtains

J≥β−1ln⁡(1+2e−βK+2e−βK1+eβK),J\ge\beta^{-1}\ln\left(1+2e^{-\beta K}+2e^{-\beta K}\sqrt{1+e^{\beta K}}\right),

which is also unbounded on β∈(0,Δ−1ln⁡2)\beta\in(0,\Delta^{-1}\ln2).

Institut Informatyki i Matematyki, Uniwersytet Marii Curie-Skłodowskiej, plac Marii Curie-Skłodowskiej 1, 20-031 Lublin, Poland; fax +48 81 537 5471

Email address: [email protected]

References

  1. [1]Blume, M., Emery, V. J., Griffiths, R. B.: Ising model for the λ-transition and phase separation in He3-He4 mixtures, Phys. Rev. A 4, 1071 (1971), https://doi.org/10.1103/PhysRevA.4.1071
  2. [2]Fisk, S.: Polynomials, roots, and interlacing, arXiv preprint math/0612833 (2006).arxiv.org/abs/math/0612833
  3. [3]Fröhlich, J., Rodriguez, P.-F.: Some applications of the Lee-Yang theorem, J. Math. Phys. 53, 095218 (2012), https://doi.org/10.1063/1.4749391
  4. [4]Heilmann, O. J., Lieb, E. H.: Theory of monomer-dimer systems, Commun. Math. Phys. 25, 190–232 (1972), https://doi.org/10.1007/BF01877590
  5. [5]Hoston, W, Berker, A. N.: Multicritical phase diagrams of the Blume-Emery-Griffiths model with repulsive biquadratic coupling, Phys. Rev. Lett. 67, 1027–1030 (1991), DOI: 10.1103/PhysRevLett.67.1027DOI
  6. [6]Iliev, L.: Laguerre entire functions, Publishing House of the Bulgarian Academy of Sciences, 1987.
  7. [7]Jiang, J., Newman, C. M.: Thermodynamic limit of the first Lee-Yang zero, Commun. Pure Appl. Math., 77, 1224–1234 (2024), https://doi.org/10.1002/cpa.22159
  8. [8]Kozitsky, Y.: Laguerre entire functions and the Lee-Yang property, Appl. Math. Computation, 141, 103–113 (2003), https://doi.org/10.1016/S0096-3003(02)00324-7
  9. [9]Kozitsky, Y., Holovatch, Y.: Lee-Yang property of the Blume-Capel model, Phys. Rev. E 114, 044107 (2026), https://doi.org/10.1103/7pdz-klhn
  10. [10]Levin, B. Ya.: Lectures on entire functions. Vol. 150. American Mathematical Soc., 1996.DOI
  11. [11]Lieb, E.H., Sokal, A.D.: A general Lee-Yang theorem for one-component and multicomponent ferromagnets, Commun. Math. Phys., 80 153–179 (1981), https://doi.org/10.1007/BF01213009
  12. [12]Lima, P. C.: The BEG model in the disordered region and at the antiquadrupolar-disordered Line of parameters, J. Statist. Phys. 178 265–280 (2020), https://doi.org/10.1007/s10955-019-02431-z
  13. [13]Lima, P. C., Mariani, R., Procacci, A., Scoppola, B.: The Blume–Emery–Griffiths model on the FAD point and on the AD line, J. Statist. Phys., 190, 170 (2023), https://doi.org/10.1007/s10955-023-03181-9
  14. [14]Newman, C. M.: Zeros of the partition function for generalized Ising systems, Commun. Pure Appl. Math. 27, 143–159 (1974), https://doi.org/10.1002/cpa.3160270203
  15. [15]Ruelle, D.: Characterization of the Lee-Yang polynomials, Ann. of Math. 171, 589–603 (2010), https://www.jstor.org/stable/27799213arxiv.org/abs/0811.1327
  16. [16]Wu, X. N., Wu, F. Y.: Blume-Emery-Griffiths model on the honeycomb lattice, J. Statist. Phys. 50, 41–55 (1988), https://doi.org/10.1007/BF01022986

Paper details

vlnc:W_uwxqux7u
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents