Non-quasiconvex dispersion of composite fermions and the fermionic Haffnian state in the first-excited Landau level

Abstract

It has long been puzzling that fractional quantum Hall states in the first excited Landau level (1LL) often differ significantly from their counterparts in the lowest Landau level. We show that the dispersion of composite fermions (CFs) is a deterministic factor driving the distinction. We find that CFs with two quantized vortices in the 1LL have a non-quasiconvex dispersion. Consequently, in the filling fraction $7/3$, CFs occupy the second $\Lambda$-level instead of the first. The corresponding ground state wave function, based on the CF wave function ansatz, is identified to be the fermionic Haffnian wave function rather than the Laughlin wave function. The conclusion is supported by numerical evidence from exact diagonalizations in both disk and spherical geometries. Furthermore, we show that the dispersion becomes quasiconvex in wide quantum wells or for CFs with four quantized vortices in the filling fraction $11/5$, coinciding with observations that the distinction between the Landau levels disappears under these circumstances.

Introduction.—

A two-dimensional electron gas subjected to a strong perpendicular magnetic field exhibits the fractional quantum Hall effect (FQHE), characterized by fractionally quantized Hall plateaus in specific filling fractions of Landau levels Tsui et al. (1982); Stormer (1999). The effect is also observed in topological flat bands Neupert et al. (2015) in recent experiments Cai et al. (2023); Park et al. (2023); Xu et al. (2023); Zeng et al. (2023). Theoretical understanding of the FQHE is challenging because the kinetic energies of electrons in a Landau level are completely quenched, leaving interaction to dominate. Consequently, constructing plausible ground state wave functions for the FQHE has long relied on intuition or educated guesses, from the celebrated Laughlin wave function Laughlin (1983) to those hypothesized by more elaborate approaches such as the hierarchy theory Haldane (1983); Halperin (1984), the conformal field theory Cristofano et al. (1991); Fubini (1991), and the composite fermion (CF) theory Jain (1989a, 2007). Among these, Jain’s CF theory is the most successful. The theory introduces fictitious particles called CFs, each consisting of an electron and an even number of quantized vortices. Wave functions prescribed by the CF theory yield nearly perfect overlaps with those obtained from exact diagonalizations (ED) in the lowest Landau level (LLL) Dev and Jain (1992); Wu et al. (1993); Jain and Kamilla (1997). Furthermore, it has been shown that a deductive approach for determining CF wave functions and corresponding physical wave functions can be established Shi (2024).

While the FQHE in the LLL can be well described by the CF theory, it has long been puzzling that the FQHE in the 1LL often exhibits distinct features. Most notably, a Hall plateau is developed in the even denominator fraction $\nu=5/2$ Willett et al. (1987), which has been a focus in the pursuit of topological quantum computing Nayak et al. (2008). On the other hand, ordinary fractions with odd denominators in the 1LL, such as $\nu=7/3$ and $\nu=12/5$, differ significantly from their counterparts in the LLL d’Ambrumenil and Reynolds (1988); Wójs (2001); Wójs and Quinn (2002); Töke et al. (2005). The observation casts doubt on the applicability of the CF theory in the 1LL, motivating alternative pictures such as the parafermion theory Read and Rezayi (1999) and the parton theory Jain (1989b); Balram et al. (2020).

A representative case is the filling fraction $\nu=7/3$ with an effective filling fraction $\tilde{\nu}=1/3$ in the 1LL Willett et al. (1987); d’Ambrumenil and Reynolds (1988). This state is expected to be an analog of the $1/3$ state in the LLL and described by the $1/3$ Laughlin wave function. However, it is found that the overlap between the $1/3$ and $7/3$ states is low d’Ambrumenil and Reynolds (1988); Töke et al. (2005), and their entanglement and quasi-hole excitation spectra are distinct Zaletel et al. (2015); Balram et al. (2013); Johri et al. (2014). More puzzlingly, ED shows that the $11/5$ state can nevertheless be well described by the $1/5$ Laughlin wave function d’Ambrumenil and Reynolds (1988); Töke et al. (2005). Several theories had been put forward to explain the peculiarity of the $7/3$ state. Töke et al. suggest that there exists substantial $\Lambda$-level mixing induced by residue interaction between CFs Töke et al. (2005). Balram et al. propose that the $7/3$ state hosts $\mathbb{Z}_{n}$-superconductivity of partons Balram et al. (2020). Various trial wave functions for this fraction have been numerically tested in Refs. Wójs, 2001; Wójs and Quinn, 2002; Jeong et al., 2017; Kuśmierz and Wójs, 2018.

In this paper, we demonstrate that the dispersion of CFs is a key for understanding the peculiarities of the 1LL. Based on the deductive approach developed in Ref. Shi, 2024, we show that CFs with two quantized vortices (denoted as CF²) in the 1LL, unlike CFs in the LLL, have a non-quasiconvex dispersion. Consequently, in $\nu=7/3$, CFs occupy the second $\Lambda$-level Jain (2007) instead of the first. The corresponding ground state wave function is identified to be the fermionic Haffnian wave function Wen and Wu (1994); Green (2001); Yang (2021) rather than the Laughlin wave function. Using ED, we demonstrate that the unique features of the Haffnian state do manifest in exact ground state wave functions. Furthermore, we find that the dispersion becomes quasiconvex in wide quantum wells or for CF⁴ in the $11/5$ state, coinciding with observations that the distinction between the 1LL and LLL disappears under these circumstances. These findings, cumulatively, suggest that the non-quasiconvex CF dispersion is a deterministic factor behind the peculiarities of the FQHE in the 1LL.

CF dispersion.—

A CF consists of an electron and a vortex carrying an even number of quantized vortices. The Coulomb attraction between the electron and the charge void induced by the vortex gives rise to the binding energy of the CF Shi (2024). Read shows that the momentum of a CF can be defined as being proportional to the spatial separation between the electron and vortex Read (1994). Consequently, the binding energy, as a function of the spatial separation, can be interpreted as the dispersion of the CF Ji and Shi (2021). To determine the binding energy, we first calculate the electron-vortex correlation function $h\quantity(r)$, which describes the electron density profile in the vicinity of a vortex. The binding energy can then be determined by $\epsilon_{b}\quantity(r)=\quantity(\rho_{0}/2)\int\differential[2]{\boldsymbol{r}^{\prime}}v\quantity(\absolutevalue{\boldsymbol{r}-\boldsymbol{r}^{\prime}})h\quantity(r^{\prime})$, where $v\quantity(r)$ is the interaction between electrons and $\rho_{0}$ is the average electron density Shi (2024).

We can determine $h\quantity(r)$ for CF² by assuming the ground state to be the $1/3$ Laughlin state. The many-body wave function in the presence of a vortex at the origin is given by $\Psi_{0}^{\mathrm{v}}\quantity(\quantity{z_{i}})=\prod_{i}z_{i}^{2}\prod_{i<j}\quantity(z_{i}-z_{j})^{3}$, where $\quantity{z_{i}=x_{i}+y_{i}}$ is the set of complex electron coordinates Shi (2024). $h\quantity(r)$ is obtained by computing the electron density distribution of $\Psi_{0}^{\mathrm{v}}$ normalized by $\rho_{0}$. The result is shown in the inset of Fig. 1.

Different interactions $v(r)$ give rise to different binding energies of CF² for the 1LL and the LLL. For the LLL, $v(r)$ is the Coulomb interaction $v_{\mathrm{c}}\quantity(r)=e^{2}/4\pi\varepsilon r$. For the 1LL, we map the problem of interacting electrons to the mathematically equivalent problem of interacting electrons in the LLL with the effective interaction Jain (2007):

$$\tilde{v}\quantity(r)=\quantity(1+\frac{l_{B}^{2}\nabla^{2}}{2})^{2}v_{\mathrm{c}}\quantity(r),$$ (1)

where $l_{B}=\sqrt{\hbar/eB}$ is the magnetic length with $B$ being the strength of the magnetic field. The binding energy of CF² in the LLL and 1LL can then be determined using $h\quantity(r)$ and the respective interactions. The results are shown in Fig. 1. We observe that the binding energy of CF² in the 1LL, unlike that in the LLL, is a non-quasiconvex function of $r$.

Figure 1: Binding energy $\epsilon_{b}\quantity(r)$ for CF² and CF⁴ in the LLL and 1LL. The binding energy of CF² in the 1LL in a quantum well with a finite width $w=0.6l_{B}$ is also shown. The horizontal line segments indicate the energies of $\Lambda$-levels for $\nu=7/3$. The energies are in units of $\tilde{\nu}e^{2}/16\pi^{2}\epsilon l_{B}$. Inset: electron-vortex correlation function $h\quantity(r)$ for CF² and CF⁴.

$\Lambda$ levels and wave functions.—

With the binding energy of a CF, we can establish its effective Hamiltonian. In the dipole picture, the electron and vortex in a CF are confined in two separate LLLs induced by the physical magnetic field and the emergent Chern-Simons magnetic field, respectively Shi (2024). As a result, the CF is described by a bi-variate wave function $\psi\quantity(z,\bar{\eta})$, which is (anti-)holomorphic in the complex electron (vortex) coordinate $z$ ($\eta\equiv\eta_{x}+\mathrm{i}\eta_{y}$). After projecting to the LLLs, $\bar{z}$ and $\eta$ become the operators $\hat{\bar{z}}=2l_{B}^{2}\partial_{z}$ and $\hat{\eta}=2l_{b}^{2}\partial_{\bar{\eta}}$, respectively, where $l_{b}=l_{B}/\sqrt{2\tilde{\nu}}$ is the magnetic length of Chern-Simons magnetic field for CF² Jain (2007); Shi (2024). We can then define the ladder operators of $\Lambda$-levels as $\hat{a}=\quantity(z-\hat{\eta})/\sqrt{\gamma}l_{B}$ and $\hat{a}^{\dagger}=\quantity(\hat{\bar{z}}-\bar{\eta})/\sqrt{\gamma}l_{B}$, with $\gamma\equiv\absolutevalue{1/\tilde{\nu}-2}$. The effective Hamiltonian operator of the CF is given by:

$$\hat{H}=:\epsilon_{b}\quantity(\hat{r}):,$$ (2)

where we treat the binding energy as a function of $r$, and map $r^{2}\equiv\absolutevalue{z-\eta}^{2}$ to the operator $\hat{r}^{2}\equiv\gamma l_{B}^{2}\hat{a}^{\dagger}\hat{a}$. The colons indicate the normal ordering of the ladder operators, which places $\hat{a}^{\dagger}$s to the left of $\hat{a}$’s. The specific ordering of the ladder operators is a result of the particular way that the binding energy is defined Shi (2024).

The eigen-energies and eigen-states of $\hat{H}$ can be determined. It is easy to see that the number operator $\hat{n}\equiv\hat{a}^{\dagger}\hat{a}$ commutes with the Hamiltonian: $[\hat{H},\hat{n}]=0$. We can thus define the index of $\Lambda$-levels as the eigenvalue of $\hat{n}$. Similar to ordinary Landau levels, wave functions for the first ($n=0$) $\Lambda$-level $\psi_{0}\quantity(z,\bar{\eta})=f(z)\exp(z\bar{\eta}/2l_{b}^{2})$, where $f(z)$ is a holomorphic function in $z$, is annihilated by the lowering operator $\hat{a}$. Wave functions for $n>0$ can be obtained by successively applying the raising operator to $\psi_{0}\quantity(z,\bar{\eta})$: $\psi_{n}\quantity(z,\bar{\eta})=(\hat{a}^{\dagger n}/\sqrt{n!})\psi_{0}\quantity(z,\bar{\eta})$. The energies of $\Lambda$-levels as a function of $n$ can be determined straightforwardly König and Hucht (2021). The result for $\nu=7/3$ is shown in Fig. 1. Remarkably, the lowest $\Lambda$-level has the index $n=1$ rather than $n=0$.

The physical wave function of the $7/3$ state can be determined by applying the wave function ansatz of the CF theory, which maps a many-body wave function of CFs to a physical wave function of electrons. From the energy spectrum of $\Lambda$-levels, we expect that CFs in the $7/3$ state fully occupy the second $\Lambda$-level. The corresponding wave function of CFs can be obtained by applying raising operators to the wave function $\Psi_{\text{CF}}^{0}\quantity(\quantity{z_{i},\bar{\eta}_{i}})$ of a fully occupied first $\Lambda$-level: $\Psi_{\text{CF}}\quantity(\quantity{z_{i},\bar{\eta}_{i}})\propto\prod_{i}\quantity(2l_{B}^{2}\partial_{z_{i}}-\bar{\eta}_{i})\Psi^{0}_{\text{CF}}\quantity(\quantity{z_{i},\bar{\eta}_{i}})$, with $\Psi_{\text{CF}}^{0}\quantity(\quantity{z_{i},\bar{\eta}_{i}})\equiv\prod_{i<j}\quantity(z_{i}-z_{j})\exp(\sum_{i}{z_{i}\bar{\eta}_{i}}/{2l_{b}^{2}})$. The electron wave function can then be obtained by overlapping $\Psi_{\text{CF}}$ with the $1/2$ Laughlin state of vortices Shi (2024). We obtain sup :

$$\Psi(\quantity{z_{i}})\propto\lim\limits_{\eta\to z}\prod_{i}(2\partial_{z_{i}}-\partial_{\eta_{i}})\prod_{i<j}\quantity(z_{i}-z_{j})\quantity(\eta_{i}-\eta_{j})^{2}.$$ (3)

After applying the derivatives and taking the limit, we find that the wave function is nothing but the fermionic Haffnian wave function Greiter et al. (1992); Wen and Wu (1994); Green (2001):

$$\Psi\quantity(\quantity{z_{i}})\propto\text{Hf}\quantity[\frac{1}{\quantity(z_{i}-z_{j})^{2}}]\prod_{i<j}\quantity(z_{i}-z_{j})^{3},$$ (4)

where Hf denotes the Haffnian of a matrix with off-diagonal elements $1/\quantity(z_{i}-z_{j})^{2}$. The Haffnian can also be written as the determinant $\det[1/\quantity(z_{i}-z_{j})]$ Greiter et al. (1992). We note that the Haffnian wave function is defined only for an even number of electrons. For an odd number of electrons, the wave function Eq. (3) predicted by the CF theory vanishes identically Wu and Jain (1995). The Haffnian state was first proposed as a $d$-wave pairing state analogous to the $p$-wave Pfaffian state Wen and Wu (1994). As far as we know, this is the first time that the state is related to the CF theory, with the underlying CF state identified.

Similarly, we can determine the $\Lambda$-levels for the filling fraction $\nu=12/5$ with the effective filling fraction $\tilde{\nu}=2/5$ in the 1LL. In this case, we find that the two lowest $\Lambda$-levels have the indices $n=2,\,3$ sup . Consequently, CFs in the $12/5$ state occupy the third and fourth $\Lambda$-levels rather than the first two as in the $2/5$ state. The resulting electron state should also differ significantly from the $2/5$ state in the LLL.

Numerical verification.—

Using ED, we test our conclusion for $\nu=7/3$ by examining whether features of the Haffnian state manifest in exact ground state wave functions. Compared to the Laughlin state, the Haffnian state has a smaller $z$-component of the total angular momentum $L_{z}$ on a disk and a different topological shift $S$ on a sphere. Moreover, the pairing nature of the state suggests stability only for even numbers of electrons. We therefore solve the ground state wave functions of interacting electrons in both the disk and spherical geometries. The results corroborate our conclusion well.

Figure 2: Lowest ED energies per particle of $\nu=1/3$ and $\nu=7/3$ in different $L_{z}$ sectors. $E_{0}$ denotes the ground state energy. The energies are in units of $e^{2}/4\pi\epsilon l_{B}$. $L_{z}^{\text{Laughlin}}=3N(N-1)/2$ denotes the $z$ component of the total angular momentum of the Laughlin state. The vertical dashed lines indicate the values expected for the Haffnian state $L_{z}^{\text{Haffnian}}=L_{z}^{\text{Laughlin}}-N$.

In the disk geometry, $L_{z}$ of the Haffnian state Eq. (4) is smaller than that of the Laughlin state by $N$, where $N$ is the number of electrons. To determine $L_{z}$ of the exact ground states, we identify the lowest energies in different $L_{z}$ sectors for $\nu=1/3$ and $7/3$. The results are shown in Fig. 2. We see that the exact ground states for $\nu=7/3$ do have the $L_{z}$ values expected for the Haffnian state for $N\geq 10$. For $N=8$, we observe a deviation by one, likely due to finite size effects.

In the spherical geometry, the Haffnian state has a different topological shift $S$ compared to the Laughlin state. The topological shift $S$ is defined by the relation $2l\equiv\tilde{\nu}^{-1}N-S$, where $l$ is the angular-momentum quantum number of the physical Landau level Wen and Zee (1992). For CF², we have $l=l^{\ast}+(N-1)$, where $l^{\ast}$ is the angular-momentum quantum number of the first $\Lambda$-level Jain (2007); Haldane (1983). Fully occupying a $\Lambda$-level with the index $n$ requires $N=2\quantity(l^{\ast}+n)+1$. Combining these relations, we have $S=3$ and $S=5$ for the Laughlin state ($n=0$) and the Haffnian state ($n=1$), respectively.

While $S$ is a free parameter for ED calculations in the spherical geometry, its probable value could be identified by examining the degeneracies and stability (excitation gaps) of the ground states with respect to different values of $S$ and $N$. For the $7/3$ state, Wójs et al. investigated a few candidate values of $S$ and concluded that $S=7$ is the most probable, based on the reasoning that for the particular shift, a gapped non-degenerate ground states with a total angular momentum $L=0$ can always be found for all calculated values of $N$ Wójs and Quinn (2002); Wójs (2001). $S=5$ was ruled out because the ground states become degenerate ($L>0$) for odd numbers of electrons.

To this end, we repeat the calculation and extend it for larger values of $N$. In Fig. 3, we show the $N$ dependence of the excitation gaps for non-degenerate ground states at $S=3,\,5,\,7$. Our calculation confirms Wójs et al.’s observation that the excitation gap for $S=7$ diminishes rapidly with increasing $N$ for $N\leq 12$. When $N$ is further increased, we observe that the ground states become degenerate. Conversely, although the ground states for $S=5$ are non-degenerate only for even numbers of electrons, the magnitude of the excitation gap shows a trend of converging to a constant value for $N\geq 12$, suggesting robustness of the non-degenerate ground states for $S=5$.

Figure 3: Excitation gaps for non-degenerate ground states at $S=3,\,5,\,7$, in units of $e^{2}/4\pi\epsilon l_{B}$.

In retrospect, we believe that the odd-even alternation observed in the ED results for $S=5$ should be interpreted as a manifestation of the pairing nature of the Haffnian state, rather than a reason to dismiss it. In Fig. 4, we present the low-lying energy spectrum for $S=5$. We see that gapped non-degenerate ground states is stable only for even values of $N$. For odd values of $N$, ground states occur at $L=1$ or $3$, appearing to be gapless. They can be interpreted as pairing states with an unpaired electron or hole.

Figure 4: ED spectrum for even and odd numbers of electrons. $E_{0}$ denotes the ground state energy. The energies are in units of $e^{2}/4\pi\epsilon l_{B}$.

Quasiconvex dispersion.—

To further corroborate that the dispersion of CFs is a deterministic underlying factor, we investigate two scenarios under which the CF dispersion in the 1LL becomes quasiconvex. We anticipate that the distinction between the 1LL and LLL should disappear when the dispersion becomes quasi-convex. This is indeed observed.

The first scenario involves a different $h\quantity(r)$. We consider CF⁴, which carries four quantized vortices and underlies the filling fraction $11/5$. In this case, the binding energy can be determined similarly to CF², albeit using the wave function $\Psi_{0}^{\mathrm{v}}\quantity(\quantity{z_{i}})=\prod_{i}z_{i}^{4}\prod_{i<j}\quantity(z_{i}-z_{j})^{5}$. We find that the resulting binding energy is a quasiconvex function of $r$, similar to that of CF² in the LLL, as shown in Fig. 1. This is consistent with the fact that the $11/5$ state can be well described by the Laughlin wave function d’Ambrumenil and Reynolds (1988); Töke et al. (2005).

The second scenario involves the change of the electron interaction $v(r)$. We adopt the modified Coulomb interaction investigated in Ref. Papić et al., 2009: $v_{\mathrm{c}}\quantity(r)=e^{2}/4\pi\epsilon\sqrt{r^{2}+w^{2}}$, where $w$ is a parameter for characterizing the width of the quantum well confining the two-dimensional system. We find that for CF² in the 1LL, the dispersion becomes quasiconvex for $w\geqslant 0.6l_{B}$ sup , as shown in Fig. 1. This is consistent with the ED results of Ref. Papić et al., 2009, which show that the overlap between the exact ground state wave function of the $7/3$ state and the $1/3$ Laughlin wave function approaches unity for $w\gtrsim l_{B}$.

Summary and discussion.—

In summary, we show that CF dispersion is a deterministic factor behind the peculiarity of the FQHE in the 1LL. The deductive approach we employ enables us to not only explain the distinction between the $7/3$ or $12/5$ states and their counterparts in the LLL, but also correctly predict the evolution of electron states under varying conditions.

Remarkably, the Haffnian state, considered analogous to the Pfaffian pairing state, can be linked to the non-interacting CF state of a fully occupied second $\Lambda$-level. It suggests a novel possibility of pairing correlations among electrons primarily driven by the kinetic energies of CFs rather than their residual interactions.

We need to point out that the “mean-field” approach employed in this study may not be adequate in fully capturing quantitative details of the $7/3$ state. Numerically, the overlap between the Haffnian state and the exact ground state is found to be moderate Kuśmierz and Wójs (2018). Furthermore, pairing correlations in the Haffnian state could complicate the self-consistent determinations of the electron-vortex correlation function and the CF dispersion sup . Nevertheless, as pointed out in Ref. Jain, 2007, it is not uncommon for a qualitatively correct wave function to yield a low overlap with the exact wave function. It is reasonable to expect that considering the residual interaction between CFs could improve the quantitative agreement Töke et al. (2005); Trung and Yang (2021).