Application of Haldane’s statistical correlation theory in classical systems

Abstract

This letter investigates the application of Haldane’s statistical correlation theory in classical systems. A modified statistical correlation theory has been proposed by including non-linearity into the original theory of Haldane. It is shown that indistinguishability can be introduced as a form of external statistical correlation into distinguishable systems. It is proved that this modified statistical correlation theory can be used to derive classical fractional exclusion statistics (CFES) using maximum entropy methods for a self-correlating system. An extended non-linear correlation model based on power series expansion is also proposed, which can produce various intermediate statistical models.

I Introduction

Since the inception of statistical mechanics, there were several attempts Gentile j. (1940); Tsallis (1988); March et al. (1993); March (1997); Acharya and Swamy (1994) to unify all fundamental statistical theories, i.e., Fermi-Dirac (FD) statistics, Bose-Einstein (BE) statistics, and Maxwell-Boltzmann (MB) statistics. The first comprehensive approach for such a unified theory was taken by Haldane Haldane (1991) for quantum systems. In a nutshell, the postulate of Haldane states that the statistical correlation in quantum systems can be described as a reduction of the total available degenerate states ($\tilde{g_{i}}$) at an energy level ($\epsilon_{i}\in[0,\infty]$) due to the population ($n_{i}$) of the building blocks at the same energy level. The nature of the reduction was assumed to be linear, which resulted in the following equation,

$$\tilde{g_{i}}=g_{i}-\gamma n_{i}$$ (1)

where $g_{i}$ is the number of ideal degenerate states at $n_{i}=0$, and $\gamma$ is the correlation factor independent of the system size. Applying this idea in Bosonic systems, Wu Wu (1994) was able to derive an equation for intermediate statistics which interpolates between Bose-Einstein ($\gamma$ = 0) and Fermi-Dirac ($\gamma$ = 1) statistics.

Application of Haldane-Wu (HW) statistics can be found in the field of 2D-anyones Dasnières de Veigy and Ouvry (1995); de Veigy and Ouvry (1995), Laughlin liquids Camino et al. (2005); Arovas et al. (1984), adsorption of polyatomic molecules Riccardo et al. (2004, 2005); Cerofolini (2006); Dávila et al. (2009a, b); Matoz-Fernandez et al. (2011); Matoz-Fernandez and Ramirez-Pastor (2014); Riccardo et al. (2019), etc. A similar statistical theory was proposed by A. Polychronakos–namely the Polychronakos statistics (AP) Polychronakos (1996a, b); Chung and Hassanabadi (2018); Hoyuelos (2018)– which avoids the negative probability problem in HW statistics Nayak and Wilczek (1994); Polychronakos (1996a, b); Chaturvedi and Srinivasan (1997a, b); Murthy and Shankar (1999). Various other forms of intermediate statistics also exist in the literature Niven (2009); Niven and Grendar (2009); Ourabah and Tribeche (2018); Abutaleb (2014); Yan (2021). Such models can be termed quantum fractional exclusion statistics (QFES) models as they inherently assume the quantum nature of the systems.

Recently, a theory of intermediate statistics based on classical MB statistics–namely, the classical fractional exclusion statistics (CFES) Roy (2022)–was derived using the maximum entropy (MaxEnt) methods. Three major constraints were used during the derivation: (i) constant energy, (ii) constant particle number, and (iii) a special constrain, which can be written as,

$$C=\sum_{i}\Bigg{(}\frac{n_{i}^{a}}{g_{i}^{a-1}}\Bigg{)}$$ (2)

Where $a$ is an integer and $a>0$. Interestingly, using the constraints on the classical MB statistics framework, it was possible to derive FD and BE statistics at $a=2$. However, the origin of the special constraint is still not clear. A concept of varying indistinguishability was proposed in Ref. Roy (2022), which showed that this factor is essentially a bridge between the MB and BE statistics.

This letter proposes a similar derivation of CFES using a non-linear Haldane-type statistical correlation theory for indistinguishable systems. I show that using this modified statistical correlation theory in classical systems, one can also arrive at the CFES-type equation. The letter is organized as follows–in section I, I describe the non-linear modifications to Eq. 1. In section II, I describe a self-correlating system following the modified statistical correlation theory. In section III, I describe extended non-linear modifications using power series expansion. Finally, I conclude in section V with an outlook.

II Concept of Indistinguishability: Zero-moment theorem

In the subsequent letter, the word particle will be used to denote an entity whose energy distribution follows the MaxEnt principle. Let’s consider that a single species of identical particles are present in our system. As described in the previous section, Haldane’s theory of statistical correlations in quantum systems proposes that the reduction of the degeneracy factor ($g_{i}$) in a quantum state $\epsilon_{i}$ is proportional to the population ($n_{i}$) at the same state. In this section, I demonstrate an ab-initio approach to understanding the concept of indistinguishability through the lens of statistical correlation theory. I consider that the reduction in $g_{i}$ of energy level $\epsilon_{i}$ is dependent on not just $n_{i}$ but also on the populations at all available energy levels $\epsilon_{j}$. Moreover, I consider the possibility that the net contribution to statistical correlation from $\epsilon_{j}$ state to $\epsilon_{i}$ state may depend on the different population distribution functions of the particles ($p(n_{jk})$) at different degenerate stats ($g_{jk}$) of energy level $\epsilon_{j}$. The overall contribution of the statistical correlation to state $\epsilon_{i}$ is weighted by correlation factors $\gamma_{jk}$, which may differ for different energy levels and degenerate states. Lastly, I impose non-linearity to the net correlation contribution from $g_{jk}$, by raising an exponent $m$ to the population factor $n_{jk}$. Then, the overall equation takes the shape,

$$\tilde{g_{i}}=g_{i}-\sum_{j}\sum_{k}\gamma_{jk}n_{jk}^{m}$$ (3)

In this model, all particles are distinguishable since the correlation contributions from each $g_{jk}$ level can be different. In the next section, I will show that $m$ and $\gamma$ factors become dependent on each other for intermediate statistics.

Next, I decompose this model by imposing indistinguishability conditions. Firstly, I assume $\gamma_{jk}$ to be a universal constant and independent of $n_{jk}$. Moreover, I assume that the correlation factor $\gamma_{jk}\equiv\gamma_{j}$; i.e., it does not depend on the properties of degenerate states. This enables us to assume that the correlation effects exerted by the particles belonging to different $g_{jk}$ are similar. Secondly, I assume that for all practical purposes, the probability of finding a particle at any of the degenerate states $k$ of energy level $\epsilon_{j}$ is equal and independent of $k$; i.e., $p(n_{jk})\equiv n_{j}/g_{j}$. Because of the uniform distribution of the particles at different degenerate states, one cannot distinguish between the individual particles present at the same energy level $\epsilon_{j}$ using their correlation contributions $\gamma_{jk}$ and populations $n_{jk}$. As a result, they become indistinguishable from each other under these approximations.

Therefore, for such indistinguishable particles, one can state that, the standard deviation of the $p(n_{jk})$ function concerning $k$ states is zero. Going one step further, I propose a theorem as described below.

Zero-moment theorem: The $m^{\text{th}}$ order moment ($\sigma_{m}(\epsilon_{j})|m\geq 2$) of $p(n_{jk})$ function at different $k$ states of an energy level $\epsilon_{j}$ is zero, i.e.,

$$\sigma_{m}(\epsilon_{j})=\frac{\sum_{k}n_{jk}^{m}}{g_{j}}-\left(\frac{\sum_{k}n_{jk}}{g_{j}}\right)^{m}=0$$ (4)

which leads to,

$$\sum_{k}n_{jk}^{m}=\frac{n_{j}^{m}}{g_{j}^{m-1}}$$ (5)

A corollary of this theorem is that $m$ has to be an integer to satisfy the condition in Eq. 5. We can now fully eliminate the $k$-summand in Eq. 3 by substituting Eq. 5 to Eq. 3 as,

$$\tilde{g_{i}}=g_{i}-\sum_{j}\gamma_{j}\bigg{(}\frac{n_{j}^{m}}{g_{j}^{m-1}}\bigg{)}$$ (6)

It is straightforward to show that if one disregards all contributions from the other states except the $i^{th}$ state; i.e., $\gamma_{j}\equiv\gamma\delta_{i}$ where $\delta_{i}$ is the Kronecker $\delta$; Eq. 1 can be recovered from Eq. 6 at the linear correlation case at $m=1$. Note that, the linear correlation case exerts a maximum correlation effect on $g_{i}$.

Eq. 5 represents the mathematical description of the concept of indistinguishability, derived in the light of statistical correlation theory. It clearly describes the origin of the constraint $C$ in Eq. 2 and how it is linked to indistinguishability. The dependency of statistical correlation on indistinguishability is an interesting aspect of this analysis. It provides an alternate path for introducing indistinguishability in a classical distinguishable system, rather than changing the microstate counting method. In the following section, I show that the application of Eq. 6 to classical MB statistics directly gives rise to the CFES equation.

III Self-correlating classical system

The non-linear correlation theory can certainly be used to modify the microstate counting formula of BE statistics ($W_{\text{BE}}$) and derive different forms of QFES Wu (1994). In this paper, I apply this theory to the classical systems and investigate the changes in classical MB statistics ($W_{\text{MB}}$) due to statistical correlations. Let’s assume that the system is self-correlating, i.e., a single species of identical particles are present in the system. Let’s denote their un-optimized population with a slightly different notation, $\nu_{i}$. As shown in the previous section, indistinguishability can be introduced in a classical system in the form of a statistical correlation. Using Eq. 6, I can easily show that $W_{\text{MB}}$ takes the form,

$$W_{\text{MB}}=W_{\text{MB}}^{\text{id}}\prod_{i}\Bigg{\{}1-\frac{1}{g_{i}}\sum_{j}\gamma_{j}\Bigg{(}\frac{\nu_{j}^{m}}{g_{j}^{m-1}}\Bigg{)}\Bigg{\}}^{\nu_{i}}$$ (7)

where $W_{\text{MB}}^{\text{id}}=g_{i}^{\nu_{i}}/\nu_{i}!$ is the ideal number of the microstates. At a dilute condition; i.e., $\nu_{i}/g_{i}\sim 0$; one can derive the corresponding particle distribution in the following manner. Using MaxEnt method, one can maximize $W_{\text{MB}}$ with respect to $\nu_{i}$ to get the optimized population, $\nu_{i,\text{max}}\equiv n_{i}$, as,

$$\frac{\partial\ln W_{\text{MB}}}{\partial\nu_{i}} =\frac{\partial\ln W_{\text{MB}}^{\text{id}}}{\partial\nu_{i}}+\ln\Bigg{\{}1-\frac{1}{g_{i}}\sum_{j}\gamma_{j}\Bigg{(}\frac{\nu_{j}^{m}}{g_{j}^{m-1}}\Bigg{)}\Bigg{\}}$$ (8)

$$-\frac{m\gamma_{i}\nu_{i}^{m}/g_{i}^{m-1}}{\{g_{i}-\sum_{j}\gamma_{j}(\nu_{j}^{m}/g_{j}^{m-1})\}}$$

$$=0$$

which leads to,

$$n_{i} =g_{i}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}1-\frac{1}{g_{i}}\sum_{j}\gamma_{j}\Bigg{(}\frac{n_{j}^{m}}{g_{j}^{m-1}}\Bigg{)}\Bigg{\}}$$ (9)

$$\exp\Bigg{\{}-\frac{m\gamma_{i}n_{i}^{m}/g_{i}^{m-1}}{g_{i}-\sum_{j}\gamma_{j}(n_{j}^{m}/g_{j}^{m-1})}\Bigg{\}}$$

$$\approx\tilde{g_{i}}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}1-m\gamma_{i}\Bigg{(}\frac{n_{i}^{m}}{\tilde{g_{i}}g_{i}^{m-1}}\Bigg{)}\Bigg{\}}\text{ ; using $g_{i}>>n_{i}$}$$

$$=g_{i}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}\frac{\tilde{g_{i}}}{g_{i}}-m\gamma_{i}\Bigg{(}\frac{n_{i}^{m}}{g_{i}^{m}}\Bigg{)}\Bigg{\}}$$

where $\alpha=\beta\mu$ and $\beta=(1/k_{\text{B}}T)$ are the Lagrange parameters related to the chemical potential ($\mu$) and temperature ($T$) of the system. One can show, if $\gamma_{i}\equiv\gamma\delta_{i}$, Eq. 9 reduces to the CFES equation Roy (2022) as,

$$n_{i}=g_{i}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}1-(m+1)\gamma\Bigg{(}\frac{n_{i}^{m}}{g_{i}^{m}}\Bigg{)}\Bigg{\}}$$ (10)

where $\gamma$ corresponds to the third Lagrange parameter and $m$ is the exponent  described in Ref. Roy (2022). Consequently, one can derive the constraint $C$ for the population distribution in Eq. 10 in a retrospective way as,

$$C=\sum_{i}\Bigg{(}\frac{n_{i}^{m+1}}{g_{i}^{m}}\Bigg{)}$$ (11)

The relationship between $\gamma$ and $m$ satisfy the equation,

$$n_{\text{c}}=\Bigg{\{}\frac{1}{(m+1)\gamma}\Bigg{\}}^{1/m}$$ (12)

where $n_{\text{c}}$ is the maximum occupancy numbers at all available degenerate states. An interesting question comes up about the limits of $m$. In the previous section, I have already shown that $m$ has to be an integer to satisfy the zero-moment theorem. If one only considers Eq. 10, $m=[-1,0,\infty]$ will produce traditional MB statistics, as proved in Ref. Roy (2022). However, if one considers Eq. 6 as well, $m=0$ state become unphysical as $\gamma\to\infty$ according to Eq. 12, which leads to $\tilde{g}_{i}\to-\infty$. Similarly, for $m=-1$, $\tilde{g}_{i}\to\infty$ at $n_{i}\to 0$. Therefore, under the modified statistical correlation theory, the limits of $m$ is $\infty\geq m\geq 1$. Here, $m\to\infty$ and $\gamma\to 0$ state is the only allowed state where MB statistics can be recovered. Thus, Eq. 10 is truly unifying statistical theory that encompasses properties MB ($m=\infty,\gamma=0$), FD ($m=1,\gamma=1$), and BE ($m=1,\gamma=-1$) statistics, as well as their intermediate statistics.

IV Extended non-linear models

It might be possible to use functions other than the simple exponent in Eq. 6 to introduce non-linearity in the statistical correlation. However, elimination of the $k$ summand in Eq. 3 is necessary to introduce indistinguishability in any non-linear model. The exponent model was perhaps the simplest where such elimination was easy because of the zero-moment theorem. Another way to study other forms of non-linearity, which will also ensure the elimination of the $k$ summand, is to use a power series expansion of $n_{jk}$ as,

$$\tilde{g}_{j}=g_{j}-\sum_{j}\sum_{k}\gamma_{jk}\Bigg{(}\sum_{l=0}^{\infty}b_{l}n_{jk}^{l}\Bigg{)}$$ (13)

where $b_{l}$’s are arbitrary constants chosen in a way that $\sum_{l}b_{l}n_{jk}^{l}$ is converging and finite valued function. Using zero-moment theorem and $\gamma_{jk}\equiv\gamma_{j}$, I get,

$$\tilde{g}_{j}=g_{j}-\sum_{j}\gamma_{j}\Bigg{(}\sum_{l=0}^{\infty}b_{l}\frac{n_{j}^{l}}{g_{j}^{l-1}}\Bigg{)}$$ (14)

which removes all $k$ summands in Eq. 13 and introduce indistinguishability in the model. Such elimination of $k$ summands is not possible for a similar power series expansion $n_{jk}$ with $l<0$, hence they are not allowed. Using the similar methods as described in the previous section, one can derive the optimal population distribution (for $\gamma_{i}\equiv\delta_{i}\gamma$) as,

$$n_{i}=g_{i}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}1-\gamma\Bigg{(}\sum_{l=0}^{\infty}(l+1)b_{l}\frac{n_{i}^{l}}{g_{i}^{l}}\Bigg{)}\Bigg{\}}$$ (15)

The maximum occupancy numbers at each degenerate state will have an upper bound for $\gamma>0$ as,

$$0\leq\sum_{l=0}^{\infty}(l+1)b_{l}n_{c}^{l}\leq\frac{1}{\gamma}$$ (16)

Using different values for the coefficients of $b_{l}$, one can derive different statistical models from Eq 15. Clearly, choice of $b_{l}$ has to be such that $b_{l}\to 0$ as $l\to\infty$ for $\gamma>0$.

An interesting example is $b_{l}=1/(l+1)!$, which leads to,

$$n_{i}=g_{i}e^{-(\alpha+\beta\epsilon_{i})}\Bigg{\{}1-\gamma e^{(n_{i}/g_{i})}\Bigg{\}}$$ (17)

As for dilute systems $\exp(n_{i}/g_{i})\sim(1+n_{i}/g_{i})$, Eq. 17 roughly converges to Eq. 10 with $m=1$. In this particular case, one can show that, $\gamma(1+n_{i}/g_{i})<1$ for $\gamma>0$, which leads to, $\gamma<1$.

V Conclusion and Outlook

In this work, I have given clear-cut evidence that the special constraint used in the previous derivation of the CFES statistics is indeed linked to the indistinguishability of the system. Usually, the linear form of Haldane’s theory is used in deriving QFES. The non-linear modification introduced in this letter helps us to understand the relationship between statistical correlation and indistinguishability using the zero-moment theorem. For a self-correlating system, this non-linear statistical correlation theory directly produces CFES statistics. Thus, I have shown that Haldane’s statistical correlation theory lies at the core of CFES statistics.

I have shown that one can derive many intermediate statistical models using appropriate choices for the $b_{l}$ coefficients, which will obey Eq. 16 and produce physically meaningful population distribution. Currently, I am exploring a binary inter-correlating CFES type system, which can produce interesting results. The application of CFES statistics in enhanced sampling methods in molecular dynamics simulations of classical systems is also being investigated.