Exciton crystal melting and destruction by disorder in bilayer quantum hall system with total filling factor one

Abstract

Bilayer quantum hall system with total filling factor 1 was studied in the regime of heavy layer imbalance in a recent transport experiment [1], with intriguing new findings. We demonstrate in this paper that 1) the exciton Wigner crystal in this regime can melt into a superfluid phase, giving rise to re-entrant superfluid behavior; 2) in the presence of disorder, electron and hole Wigner crystals in the two layers go through a locking/decoupling transition as layer separation increases, resulting in a sudden change in the counter flow conductance. Comparison will be made with the findings of Ref. [1].

I Introduction

Bilayer quantum hall system with total filling factor $\nu_{1}+\nu_{2}=1$ has been actively studied for several decades [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25]. The long-lasting interest in it is due to its extremely rich phase diagram and the fascinating physics associated with the novel phases and transitions among them, which is yet to be exhausted. A recent transport experiment [1] focused on a regime that is under-explored before, namely when the two layers are heavily imbalanced, such that $\Delta\nu=\nu_{1}-\nu_{2}\lesssim 1$, namely $\nu_{2}\ll 1$ is the minority layer of electrons, and the hole filling factor in the majority layer 1 is $1-\nu_{1}=\nu_{2}$. The experiment observed an exciton superfluid-insulator transition predicted more than 20 years ago [11], and revealed some new surprises. The purpose of this work is to provide theoretical understandings of two of the new findings.

We start by briefly summarising the relevant observations and basic idea/conclusion of our theoretical work. The experimentalists pass a (drive) current through one of the layers, and measure the current and/or voltage response of the same as well as opposite layer; the latter corresponds to drag response [26]. Symmetric and antisymmetric combinations of these responses form normal and counter flow response functions; the latter is usually attributed to the flow of interlayer excitons which are bound pairs of electron in one layer and hole in the other, assuming they are present and dominate the counter flow transport channel. The excitons, on the other hand, may either condense to form a superfluid (SF), or crystallize and form an insulating Wigner crystal (WC) state. We will demonstrate that under appropriate conditions an exciton Wigner crystal may melt into a superfluid state, giving rise to re-entrant superfluid behavior in the counter flow channel seen in the experiment. We further demonstrate that presence of uncorrelated disorder potential in the two layers can disrupt the formation of the interlayer excitons, driving a transition between exciton Wigner crystal and decoupled electron and hole Wigner crystals in each layer. This transition manifests itself in some transport anomalies observed in the counterflow channel.

The rest of the paper is organized as following. In Sec. II, we calculate the critical temperature of bilayer exciton superfluid using two previously established effective models[8, 11] at layer imbalance $1-|\Delta\nu|\ll 1$, and demonstrate it is often higher than the melting temperature of exciton Wigner crystal. As a result the crystal melts into a superfluid when this is the case. In Sec. III we consider the interplay of disorder and interlayer coupling and analyse the competition between them. Clearly interlayer Coulomb coupling drives formation of interlayer excitons, while uncorrelated disorder favors formation of decoupled electron and hole Wigner crystals in each layer. By comparing the energy gains from exciton formation and uncorrelated electron and hole WC distortion in the two layers, we obtain the phase diagram of the system. Some concluding remarks are provided in Sec. IV.

Unless otherwise stated, magnetic length is assumed to be the length scale, i.e. $l_{B}=1$.

II Exciton superfluid and melting of Wigner crystal

We start by discussing the phases relevant to this section. It is well-established that single layer 2-dimensional electron gas forms a Wigner crystal at zero temperature for small $\nu$ [27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43]. Putting two layers together and holding the total filling factor $\nu_{1}+\nu_{2}=1$, the electron (in the minority layer 2) and hole (on the majority layer 1) Wigner crystals with identical structure lock into an exciton crystal [11], which may melt due to either quantum or thermal fluctuations. Comparisons between drag current versus drive current, and parallel flow versus counter flow conductance, indicate that the resulting zero temperature phase is indeed correlated between the two layers [1]. Electrons in one layer and holes in the other tend to bind and condense into an exciton superfluid when $d$ is small and $1-|\Delta\nu|$ is not too close to 1, and form an exciton Wigner crystal otherwise; see orange line of Figure 2 for schematic zero temperature phase diagram near $\Delta\nu=1$. With increasing temperature the exciton Wigner crystal melts into a liquid. We find, surprisingly, that under appropriate conditions the resultant liquid state may be a superfluid.

To understand this we go back to zero temperature, where the exciton superfluid and Wigner crystal phases compete with each other. They are (most likely) separated by a 1st order phase boundary, allowing us to consider thermal effects on them at finite temperature separately. As discussed earlier the exciton Wigner crystal melts into a liquid at some melting temperature which we estimate below. The exciton superfluid, on the other hand, goes through a Kosterlitz-Thouless (KT) transition and becomes a normal fluid. If the superfluid critical (KT) temperature is lower than the melting temperature, we expect WC melts into a normal fluid which is the usual situation. If it turns out the KT temperature is higher than the melting temperature, we conclude that the WC melts into a superfluid instead, resulting in re-entrant superfluidity. The resultant (schematic) phase diagram takes the form of Figure 1. Our results compare favorably with those of [1].

To determine the phase diagram we start by calculating the superfluid stiffness which determines the KT temperature of the superfluid phase, and then compare it with the melting temperature of the WC.

II.1 Phase stiffness and Kosterlitz-Thouless temperature of exciton superfluid

When $\Delta\nu$ is fixed, the low temperature superfluid behavior can be described by an effective XY model. In this section we calculate the phase stiffness from two different models: spin $1/2$ easy-plane ferromagnet [8] and dilute exciton [11]. Once the phase stiffness $\rho_{s}$ is obtained, critical temperature of SF is bounded by $T_{c}=\frac{\pi\rho_{s}}{2}$. It turns out in the vicinity of $\Delta\nu=1$, two models lead to the same result. Let $Q^{2}=e^{2}/(4\pi\epsilon)$ for simplicity.

II.1.1 spin-1/2 easy-plane ferromagnet

To begin with, we setup the notations here. Let $\nu_{1}=\nu_{\uparrow}=1-\delta,\nu_{2}=\nu_{\downarrow}=\delta$, we have $\Delta\nu=(1-2\delta)=\cos\theta=2(S_{\uparrow}-S_{\downarrow})=m^{z}$ and $\delta=\frac{1-\Delta\nu}{2}=\sin^{2}(\theta/2)$, density of electron in one layer $n=\delta/2\pi=\sin^{2}(\theta/2)/2\pi$.

The gradient energy density of xy components of local spin is

$$\frac{\rho_{E}}{2}[(\nabla m^{x})^{2}+(\nabla m^{y})^{2}],$$ (1)

where $\rho_{E}=-\frac{\nu}{32\pi^{2}}\int_{0}^{\infty}V_{k}^{E}h(k)k^{3}\mathrm{d}k$, and $V_{k}^{E}=V_{k}^{A}e^{-kd},V_{k}^{A}=\frac{2\pi Q^{2}}{k}$ are fourier transforms of intralayer Coulomb potential and interlayer Coulomb potential respectively [8] . $h(k)=\frac{\nu}{2\pi}\int\mathrm{d}^{2}r(g(r)-1)\exp(-i\mathbf{k}\cdot\mathbf{r})$ and $g(r)=\langle c^{{\dagger}}(\mathbf{r})c(0)\rangle$ are particle-hole correlation of Laughlin function in momentum space and real space.

For $\nu=1$, $g(r)=\exp(-r^{2})$ and $h(k)=-\exp(-\frac{|k|^{2}}{2})$, hence we have

$$\rho_{E}=-\frac{Q^{2}}{16\pi}\left[d-\sqrt{\frac{\pi}{2}}(d^{2}+1)e^{\frac{d^{2}}{2}}\operatorname{erfc}\left(d/\sqrt{2}\right)\right]\equiv\frac{Q^{2}f(d)}{16\pi},$$ (2)

where $d$ is the interlayer spacing, $f(d)=\sqrt{\frac{\pi}{2}}(d^{2}+1)e^{\frac{d^{2}}{2}}\text{erfc}\left(d/\sqrt{2}\right)-d$ and $\operatorname{erfc}(x)=1-\operatorname{erf}(x)$ is the complementary error function.

After we obtain $\rho_{E}$, phase stiffness of XY spin is $\rho_{s}=\rho_{E}\sin^{2}\theta$

$$\rho_{s}^{\operatorname{XY}}=\frac{Q^{2}f(d)}{4\pi}\frac{\sin^{2}(\theta)}{4}=\frac{Q^{2}f(d)}{8\pi}\frac{1-(\Delta\nu)^{2}}{2},$$ (3)

and the critical temperature $T_{\mathrm{KT}}\lesssim\frac{\pi}{2}\rho_{s}$.

II.1.2 Dilute dipolar exciton

From [11] inverse effective mass of exciton is

$$m(d)^{-1}=\frac{Q^{2}}{2}\int_{0}^{\infty}x^{2}\mathrm{e}^{-xd-x^{2}/2}\mathrm{d}x=\frac{Q^{2}}{2}\left(\sqrt{\frac{\pi}{2}}(d^{2}+1)e^{\frac{d^{2}}{2}}\text{erfc}\left(d/\sqrt{2}\right)-d\right)=\frac{Q^{2}}{2}f(d)$$ (4)

Boson spectrum given by Bogoliubov theory (see e.g. chap18 of [44]) is

$$E_{\mathbf{k}}=\sqrt{\epsilon_{\mathbf{k}}^{2}+2n\tilde{V}_{q=0}\epsilon_{\mathbf{k}}}\xrightarrow{k\rightarrow 0}\sqrt{2n\tilde{V}_{0}\epsilon_{\mathbf{k}}}=\hbar v_{s}k,$$ (5)

where the effective interaction $\tilde{V}_{k}=2\Delta V_{k}-\frac{2}{N}\sum_{\mathbf{q}}\Delta V_{q}e^{-q^{2}/2},\Delta V=V^{A}-V^{E}$ [11]. The Goldstone mode velocity $v_{s}=\sqrt{\frac{n\tilde{V}_{0}}{m}}$ is also reported in [11].

Thereafter superfluid phase stiffness $\rho_{s}=\frac{n}{m}$ can be obtained from $n\mathbf{v}_{s}=\rho_{s}\nabla\theta$ and $\mathbf{v}_{s}=\nabla\theta/m$.

$$\rho_{s}^{\operatorname{exciton}}=\frac{Q^{2}f(d)}{4\pi}\sin^{2}\frac{\theta}{2}=\frac{Q^{2}f(d)}{8\pi}(1-\Delta\nu),$$ (6)

This expression of superfluid density coincides with the result (3) when $\Delta\nu\rightarrow 1$ (or $\theta\rightarrow 0$) since $\frac{1-(\Delta\nu)^{2}}{2}=1-\Delta\nu-(1-\Delta\nu)^{2}/2\simeq 1-\Delta\nu$.

We will stick to Equation 6 and use $T_{\mathrm{KT}}=\pi\rho_{s}^{\operatorname{exciton}}/2$ as our estimate of KT temperature

$$t_{\mathrm{KT}}\equiv T_{\mathrm{KT}}/Q^{2}=\frac{f(d)}{16}(1-\Delta\nu),$$ (7)

II.2 Melting temperature of exciton Wigner crystal and phase diagrams

In this subsection we compare melting temperature of exciton Wigner crystal, $T_{m}$, with the KT temperature estimated above, and determine the finite temperature phase diagram of the system.

The melting temperature of classical exciton Wigner crystal was reported to be $T_{m}\approx 0.0907\frac{d^{2}Q^{2}}{a^{3}}$ [45, 46]. Relation $a=[\frac{\sqrt{3}}{8\pi}(1-\Delta\nu)]^{-1/2}$ can obtained from $\frac{1-\Delta\nu}{2}=\frac{n_{e}}{1/2\pi}$ where $n_{e}=2/(\sqrt{3}a^{2})$. We then have dimensionless temperatures

$$t_{m}=0.0907d^{2}[\frac{\sqrt{3}}{8\pi}(1-\Delta\nu)]^{3/2}$$ (8)

where $t_{m}=T_{m}/Q^{2}$. Compare Equation 8 with Equation 7, we are able to determine the finite temperature phase diagrams Figure 1 for two different situations, both of which are included in the zero temperature phase diagram Figure 2. Two situations are separated by $d_{c}\simeq 2$. When $d>d_{c}$ the Wigner crystal could melt into either superfluid or normal liquid, otherwise it only melts into superfluid. In the dilute limit $1-\Delta\nu\ll 1$, the exciton Wigner crystal always melts into a superfluid phase since $t_{m}<t_{\mathrm{KT}}$ is always true.

Treating WC as classical leads to an overestimation of $T_{m}$, because quantum fluctuation tends to lower $T_{m}$ as well. Since our goal is to demonstrate the possibility of $T_{m}<T_{\mathrm{KT}}$, they are justified, and does not change the phase diagram qualitatively. A more serious issue is neglecting the effects of disorder, which are very important when $\Delta\nu\rightarrow 1$, where the excitons are destroyed. This is the focus of the next section. The resultant phase there is a single-layer integer quantum Hall state, which dominates the experimental phase diagram there. One should keep this in mind when comparing with the theoretical phase diagrams in this section obtained without taking these into account.

(a)

(b)

Figure 1: Finite temperature phase diagrams near $\Delta\nu=1$ based on Equations 7 and 8. The green dashed line is the natural extension of the zero temperature phase boundary between exciton superfluid and Wigner crystal phases. Blue line is the superfluid KT temperature. Orange line is the melting curve of exciton Wigner crystal. (a) Case with $d>d_{c}\simeq 2$ in which the exciton Wigner crystal can melt into either superfluid or normal liquid, depending on $\Delta\nu$. (b) Case with $d<d_{c}$ where the exciton Wigner crystal can only melt into a superfluid.

Figure 2: Schematic zero temperature phase diagrams near $\Delta\nu=1$. Orange region denotes the superfluid phase which, due to disorder, terminates before $\Delta\nu=1$ is reached. Orange solid line is the schematic zero temperature phase boundary between superfluid and Wigner crystal. Blue and blank region are both Wigner crystal at zero temperature, while the blue one melts into superfluid with increasing temperature and the blank one melts into normal liquid (see the arrows in the right panel). The blue dashed line is obtained by equating Equation 8 with Equation 7. The red dotted line marked by $d^{\ast}=0.6$, reported in [11], is obtained by comparing correlation energy per exciton in superfluid phase and kinetic energy in crystal phase, above which superfluid phase is unfavored.

III Locking-decoupling transition of bilayer Wigner crystal

In the previous section we discussed various phases interlayer excitons can form, and neglected the effects of disorder. Ref. [1] finds a single layer integer quantum Hall state when $\Delta\nu$ is very close to 1, in which the two layers are essentially decoupled. They also report evidence of a transition into the exciton Wigner crystal phase discussed above. We argue below the existence of the decoupled phase is stabilized by disorder, which also drives the transition. In the absence of disorder potential, the electron and hole WCs in the two layers always align themselves with each other to minimize the Coulomb energy, resulting in the exciton WC [11]. On the other hand, disorder potential, which is different in the two layers (assumed to be uncorrelated for simplicity), distorts the two WCs in uncorrelated ways, which tends to disrupt the formation of excitons and decouple the two layers. By comparing the energy gain and loss between disorder potential energy and interlayer Coulomb energy, we are able to obtain the transition line for the two layers to become locked/decoupled.

III.1 Disorder potential energy

We introduce the Gaussian white noise random potential $V_{i}(\mathbf{r})$ that is uncorrelated between the two layers:

$$\langle V_{i}(\mathbf{r})V_{j}(\mathbf{r^{\prime}})\rangle=\Delta^{2}\delta(\mathbf{r-r^{\prime}})\delta_{ij},$$ (9)

where $i,j=1,2$ are layer indices. Pinning length $R$ of 2d Wigner crystal, defined as $\langle[u(0)-u(R)]^{2}\rangle\simeq a^{2}$ where $a$ is the lattice constant and $u(R)$ is the field of lattice distortion, is given by balancing the energy gain of random potential and energy cost of lattice distortion [47, 48]

$$Rn_{e}\Delta=ca^{2},$$ (10)

where $n_{e}=1/(A_{c}a^{2}),A_{c}=\sqrt{3}/2$ is the density of electron, $c$ is the shear modulus. Left and right hand sides of this equation stand respectively for random potential energy gain and elastic energy cost due to lattice distortion. Since this amount of energy is for a region of linear size $R$, dividing by $R^{2}$ we obtain the density of random potential energy (for convenience in density comparison we keep one factor of $n_{e}$ here)

$$\varepsilon_{r}=\frac{Rn_{e}\Delta}{R^{2}}=\frac{n_{e}\Delta^{2}}{cA_{c}a^{4}}.$$ (11)

For single layer Wigner crystal of electron-type interaction and dipole-type interaction we simply the shear modulus from [49]

$$\begin{array}[]{ll}c_{1}(d\lesssim a)\approx 2.5\frac{D^{2}}{a^{5}}&\operatorname{dipole}\\ c_{2}=0.3\frac{Q^{2}}{a^{3}}&\operatorname{charge}\end{array}$$ (12)

where $Q^{2}=\frac{e^{2}}{4\pi\epsilon},D^{2}=\frac{e^{2}d^{2}}{4\pi\epsilon}$. Transition from coupled to decoupled picture lowers the disorder potential energy (density) by

$$\Delta\varepsilon_{r}=\frac{2n_{e}\Delta^{2}}{c_{1}A_{c}a^{4}}-\frac{n_{e}\left(\sqrt{2}\Delta\right)^{2}}{c_{2}A_{c}a^{4}}=\frac{2n_{e}\Delta^{2}}{A_{c}a^{4}}\left(\frac{1}{c_{1}}-\frac{1}{c_{2}}\right),$$ (13)

where $\sqrt{2}\Delta$ is the effective random potential strength seen by the bilayer (since $V(\mathbf{r})=V_{1}(\mathbf{r})+V_{2}(\mathbf{r})$ has $\langle V(\mathbf{r})V(\mathbf{r^{\prime}})\rangle=2\Delta^{2}\delta(\mathbf{r}-\mathbf{r^{\prime}})$). On the other hand, in $d\rightarrow\infty$ the interlayer Coulomb energy is diminished and what we have is merely two copies of single layer Wigner crystal. Therefore $c_{1}(\infty)=2c_{2}$. In this limit $\Delta E_{r}$ is exactly half of that for individual pinning. In practice for a specific $d$ in experiments, the effective spacing $d/a$ has an upper bound $d/\sqrt{2}$, which is generally smaller than 1 (see below). For such considerations, we will simply take the dipole approximation $c_{1}=2.5D^{2}/a^{5}$.

$$\Delta\varepsilon_{r}=\frac{2n_{e}\Delta^{2}}{A_{c}Q^{2}a}\left(\frac{1}{0.3}-\frac{1}{2.5d^{2}/a^{2}}\right)=q\frac{n_{e}Q^{2}}{a}\left(\frac{1}{0.3}-\frac{1}{2.5d^{2}/a^{2}}\right),$$ (14)

where

$$q=\frac{2\Delta^{2}}{A_{c}Q^{4}}=\frac{4\Delta^{2}}{\sqrt{3}Q^{4}}$$ (15)

is the dimensionless random potential strength.

III.2 Interlayer correlation energy cost

As we demonstrated above, the system can lower the disorder potential energy by distort the electron and hole WCs in the two layers independently, compared to that of the exciton WC. Doing that, however, decouples the two layers and destroy the excitons, resulting in an increase in the interlayer Coulomb interaction energy. In this subsection we calculate this energy cost.

In this subsection we let $\frac{Q^{2}}{a}$ be energy scale and $a$, the lattice constant of 2d triangular lattice, be length scale. We are evaluating the interlayer correlation energy difference of Wigner crystal vs. homogeneous electron gas (since random relative distribution of charges in one layer is seen on average as homogeneous gas of charge by the other layer), i.e.

$$\Delta E_{e}=\int\mathrm{d}\mathbf{r}[g_{1}(\mathbf{r})-g_{2}(\mathbf{r})]\frac{-1}{\sqrt{r^{2}+d^{2}}}=\int\mathrm{d}\mathbf{r}\left[\sum_{i}\delta(\mathbf{r}-\mathbf{R}_{i})-1/A_{c}\right]\frac{1}{\sqrt{r^{2}+d^{2}}},$$ (16)

where $g_{1}(\mathbf{r})=1/A_{c},g_{2}(\mathbf{r})=\sum_{i}\delta(\mathbf{r}-\mathbf{R}_{i})$, $A_{c}=\sqrt{3}/2$ is the area of unit cell. Compared with Equation 14, a transition between locked/decoupled phase will be determined.

In the small $d$ limit, apart from a divergent $1/d$ term, this energy difference is the classic problem of static energy of 2d Wigner crystal. That is (see e.g. [50, 51])

$$\lim_{d\rightarrow 0}[\Delta E_{e}(d)-1/d]=-4.213423,$$ (17)

We now calculate this energy difference for general $d$. Let $\Delta E_{e}=E_{0}+E_{1}+E_{2}$, where $E_{0}=1/d$ and

$$\begin{array}[]{ccl}E_{1}&=&\frac{1}{\sqrt{\pi}}\left(\int_{0}^{\pi}+\int_{\pi}^{\infty}\right)\mathrm{d}tt^{-1/2}\mathrm{e}^{-td}\sum^{\prime}\mathrm{e}^{-tR_{i}^{2}}\equiv E_{11}+E_{12}\\ E_{2}&=&-\frac{1}{A_{c}}\int\mathrm{d}\mathbf{r}\frac{e^{2}}{\sqrt{r^{2}+d^{2}}}=-\frac{1}{\sqrt{\pi}A_{c}}\int_{0}^{\infty}\mathrm{d}t\int\mathrm{d}\mathbf{r}\mathrm{e}^{-tr^{2}}\mathrm{e}^{-td^{2}}t^{-1/2}=-\frac{\sqrt{\pi}}{A_{c}}\int_{0}^{\infty}\mathrm{d}tt^{-3/2}\mathrm{e}^{-td^{2}}\\ &=&-\frac{\sqrt{\pi}}{A_{c}}\int_{0}^{\pi}\mathrm{d}tt^{-3/2}\mathrm{e}^{-td^{2}}-\frac{2}{A_{c}}\left(e^{-\pi d^{2}}-\pi d\text{erfc}\left(\sqrt{\pi}d\right)\right)\end{array}$$ (18)

where $\Gamma(n)z^{-n}=\int_{0}^{\infty}t^{n-1}\mathrm{e}^{-zt}\mathrm{d}t$ is used in rewriting $1/\sqrt{d^{2}+r^{2}}=\frac{1}{\sqrt{\pi}}\int_{0}^{\infty}t^{-1/2}\mathrm{e}^{-t(d^{2}+r^{2})}\mathrm{d}t$, and $\sqrt{\pi}\int_{\pi}^{\infty}\mathrm{d}tt^{-3/2}\mathrm{e}^{-td^{2}}=\frac{2}{\sqrt{\pi}}\left(e^{-\pi d^{2}}-\pi d\text{erfc}\left(\sqrt{\pi}d\right)\right)$. $\sum^{\prime}$ stands for the summation excluding $R_{i}=0$.

Let $t=\pi x$, we have

$$E_{12}=\int_{1}^{\infty}\mathrm{d}xx^{-1/2}\sum\nolimits^{\prime}\mathrm{e}^{-\pi x(d^{2}+R_{i}^{2})}=\sum\nolimits^{\prime}\operatorname{erfc}\left[\sqrt{\pi(d^{2}+R_{i}^{2})}\right]/\sqrt{(d^{2}+R_{i}^{2})},$$ (19)

where $\int_{1}^{\infty}x^{-1/2}\mathrm{e}^{-\pi xa^{2}}\mathrm{d}x=\operatorname{erfc}\left(\sqrt{\pi}a\right)/a$ is utilized. To calculate $E_{11}$ we first complete it with a $R_{i}=0$ term

$$\begin{array}[]{ccl}E_{11}&=&\frac{1}{\sqrt{\pi}}\int_{0}^{\pi}\mathrm{d}tt^{-1/2}\mathrm{e}^{-td^{2}}\Theta_{\Gamma}(t/\pi)-\frac{1}{\sqrt{\pi}}\int_{0}^{\pi}t^{-1/2}\mathrm{e}^{-td^{2}}\mathrm{d}t\\ &=&\frac{\sqrt{\pi}}{A_{c}}\int_{0}^{\pi}\mathrm{d}tt^{-3/2}\mathrm{e}^{-td^{2}}\Theta_{\Gamma^{\prime}}(\pi/t)-\operatorname{erf}\left(\sqrt{\pi}d\right)/d\\ &=&\frac{1}{A_{c}}\int_{1}^{\infty}\mathrm{d}xx^{-1/2}\mathrm{e}^{-\pi d^{2}/x}\sum^{\prime}\mathrm{e}^{-\pi xK_{i}^{2}}-\operatorname{erf}\left(\sqrt{\pi}d\right)/d+\frac{\sqrt{\pi}}{A_{c}}\int_{0}^{\pi}\mathrm{d}tt^{-3/2}\mathrm{e}^{-td^{2}}\end{array}$$ (20)

with $\Theta_{\Gamma}(t)\equiv\sum_{\mathbf{R}_{i}\in\Gamma}\mathrm{e}^{-\pi tR_{i}^{2}},\Gamma$ being a lattice. From first line to second line we used $\int_{0}^{1}t^{-1/2}\mathrm{e}^{-\pi td^{2}}\mathrm{d}t=\operatorname{erf}\left(a\sqrt{\pi}\right)/a$ and $\Theta_{\Gamma}(t)=t^{-n/2}v(\Gamma)^{-1}\Theta_{\Gamma^{\prime}}(1/t)$, where $\Gamma^{\prime}$ is the dual of lattice $\Gamma$, $v(\Gamma)$ is the measure of unit cell of $\Gamma$ and $n$ is dimension of the lattice $\Gamma$ (see e.g. pg. 115 of [52]); from second line to third line, points of dual lattice are denoted as $\mathbf{K}_{i}$ and we let $t=\pi/x$ for all $K_{i}\equiv|\mathbf{K}_{i}|\neq 0$ terms. Note that the very last divergent term in $E_{11}$ cancel the divergent part of $E_{2}$.

Since

$$\begin{array}[]{ccl}\int_{1}^{\infty}\mathrm{d}xx^{-1/2}\mathrm{e}^{-\pi(d^{2}/x+K_{i}^{2}x)}&=&\frac{\mathrm{e}^{-2\pi dK_{i}}\left(1+\operatorname{erf}\left[\sqrt{\pi}(d-K_{i})\right]\right)+\mathrm{e}^{2\pi dK_{i}}\left(1-\operatorname{erf}\left[\sqrt{\pi}(d+K_{i})\right]\right)}{2K_{i}}\\ &\equiv&\phi_{-1/2}(d,K_{i})\end{array}$$ (21)

we have

$$\begin{array}[]{ccl}E_{1}+E_{2}&=&-\frac{\operatorname{erf}\left(\sqrt{\pi}d\right)}{d}-\frac{2}{A_{c}}\left(e^{-\pi d^{2}}-\pi d\text{erfc}\left(\sqrt{\pi}d\right)\right)\\ &&+\sum^{\prime}\frac{\operatorname{erfc}\left[\sqrt{\pi(d^{2}+R_{i}^{2})}\right]}{\sqrt{d^{2}+R_{i}^{2}}}+\frac{1}{A_{c}}\sum^{\prime}\phi_{-1/2}(d,K_{i})\end{array}$$ (22)

For a sanity check, let $d\rightarrow 0$ we have

$$\begin{array}[]{ccl}E_{1}+E_{2}&=&-2\left(1+\frac{1}{A_{c}}\right)+\sum^{\prime}\operatorname{erfc}\left(\sqrt{\pi}R_{i}\right)/R_{i}+\frac{1}{A_{c}}\sum^{\prime}\operatorname{erfc}\left(\sqrt{\pi}K_{i}\right)/K_{i}\\ &\cong&-2\left(1+\frac{1}{A_{c}}\right)+6\operatorname{erfc}\left(\sqrt{\pi}\right)+6\operatorname{erfc}\left(\sqrt{\pi}/A_{c}\right)\\ &=&-4.213475\end{array}$$ (23)

where we took nearest lattice point approximation, i.e. only six terms with smallest $R_{i},K_{i}$ in the those lattice summations are kept. Nevertheless the result match the known static energy for 2d Wigner crystal up to fourth digit.

For general $d$, let $\delta E(d)$ be the nearest lattice point approximation of $E_{1}+E_{2}$ in Equation 22

$$\begin{array}[]{ccl}\delta E(d)&=&-\frac{\operatorname{erf}\left(\sqrt{\pi}d\right)}{d}-\frac{2\left(e^{-\pi d^{2}}-\pi d\text{erfc}\left(\sqrt{\pi}d\right)\right)}{A_{c}}+\frac{6\operatorname{erfc}\left[\sqrt{\pi(d^{2}+1)}\right]}{\sqrt{d^{2}+1}}\\ &&+3\left\{\mathrm{e}^{-2\pi d/A_{c}}\left(1+\operatorname{erf}\left[\sqrt{\pi}\left(d-\frac{1}{A_{c}}\right)\right]\right)+\mathrm{e}^{2\pi d/A_{c}}\operatorname{erfc}\left[\sqrt{\pi}\left(d+\frac{1}{A_{c}}\right)\right]\right\}\end{array}$$ (24)

$1/A_{c}=2/\sqrt{3}$ comes from lattice constant of the dual lattice. It behaves asymptotically in the $d\rightarrow\infty$ limit as $\delta E(d)+1/d\sim 6\mathrm{e}^{-4\pi d/\sqrt{3}}$. Also for $d\rightarrow\infty$, $\operatorname{erfc}\left[\sqrt{\pi(d^{2}+R_{i}^{2})}\right]/\sqrt{d^{2}+R_{i}^{2}}\sim\mathrm{e}^{-\pi(d^{2}+R_{i}^{2})}/(\pi(d^{2}+R_{i}^{2}))$ and $\operatorname{erfc}\left(\sqrt{\pi}x\right)\sim\mathrm{e}^{-\pi x^{2}}/(\pi x)$ results in

$$\phi_{-1/2}(d,K_{i})\leqslant[2\mathrm{e}^{-2\pi dK_{i}}+\mathrm{e}^{-\pi(d^{2}+K_{i}^{2})}/(\pi(d+K_{i}))]/(2K_{i}),$$ (25)

All terms generated from farther lattice points are dominated by $6\mathrm{e}^{-4\pi d/\sqrt{3}}$. In the sense that $\delta E(d)$ is a good approximation to $E_{1}+E_{2}$ for both $d\rightarrow 0$ and $d\rightarrow\infty$, we could safely take

$$\Delta E_{e}\cong 1/d+\delta E(d)$$ (26)

Putting back dimensions, the Coulomb energy density difference is, with $\delta E$ defined in Equation 24,

$$\Delta\varepsilon_{e}\cong n_{e}\frac{Q^{2}}{a}(a/d+\delta E(d/a))$$ (27)

III.3 Phase Diagram

Comparing (14) with (27) we can immediately see that the transition between coupled/decoupled phases is determined by the root of the dimensionless equation

$$\frac{q}{0.3}x^{2}-x-q/2.5-x^{2}\delta E(x)=0,\quad x=d/a=d\sqrt{\frac{\sqrt{3}}{8\pi}(1-\Delta\nu)}$$ (28)

where $q$, defined in Equation 15, is, up to a constant, the energy scale of random potential comparing with Coulomb energy. Putting together, we can draw a phase diagram Figure 3 for the decoupled electron-hole Wigner crystal and exciton Wigner crystal.

Figure 3: Phase diagram of coupled/decoupled Wigner crystal plotted from Equation 28. $q=\frac{4\Delta^{2}}{\sqrt{3}Q^{4}}$ characterizes the random potential strength, where $\Delta$ is defined in Equation 9 and $Q^{2}=e^{2}/(4\pi\epsilon)$. Region under the surface is decoupled electron-hole Wigner crystal while region above it is exciton Wigner crystal.

IV Concluding Remarks

In this paper, we analysed the competition between different phases in a bilayer quantum hall system with total filling factor 1 driven by temperature and/or disorder. Our results compare favorably with a recent experiment [1]. Particularly interesting (and surprising) among our findings is that the exciton superfluid can (often) result from melting an exciton WC. This bears remarkable similarity to the observation [53] that melting of electron WC at low filling factor results in fractional quantum Hall liquids. Similar phenomena was observed very recently in systems supporting (fractional) anomalous quantum Hall states [54]. We speculate that melting of electron or hole WC in these systems resulted in the formation of fractional anomalous quantum Hall states.