betaFeedback

WorksPhysics and Astronomy

Assignment of excitonic insulators in ab initio theories: the case of NiBr2_2

Abstract

In this work we perform a detailed first-principles analysis of the electronic and optical properties of NiBr2_2 within the state-of-the-art GWGW+BSE scheme to determine whether this system displays negative excitonic energies, which would identify it as an (half) excitonic insulator. Particular attention is payed to the convergence of the GWGW band structure and to the consistency between approximations employed in the ground-state calculations and approximations employed in the linear response calculations. We show that these two issues play a crucial role in identifying the excitonic nature of NiBr2_2.

Introduction

The seminal paper [1] of Jerome and coworkers predicted the realization of the broken symmetry excitonic insulator (EI) phase in materials where the exciton binding energy overcomes the fundamental band gap. The phase transition from a symmetric state to the lower-symmetry EI would be purely electronically driven. A very rich activity followed, with various research groups trying to find signature of EI phase in different class of materials, ranging from semi-metals to small gap semiconductors and nanostructures[2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15].

The determination of the EI phase is mostly quantitative, rather then qualitative, i.e. it relies on the relative value of the excitonic binding energy versus the fundamental band gap. Therefore accurate numerical simulations are essential. Moreover, the broken symmetry phase is often associated to an instability of the crystal structure, and a distortion towards a lower symmetry phase. It then becomes a chicken and egg question whether it is the lattice distortion that drives the electronic phase transition or vice-versa, and whether or not one should speak about excitonic instability at all in many cases. The structural and the electronic phase transitions cannot be disentangled experimentally, making a direct proof of the EI existence a non trivial problem, with different experimental groups often reaching opposite conclusions. This is why ab initio numerical simulations are the key instrument to be used. Indeed, while indications of the EI phase from experimental measurements have been collected, in most works the assignment of the EI phase is based on ab initio numerical simulations, either as support of experimental data or as purely theoretical/computational works.

The state-of-the-art approach to model excitons in ab initio simulations is the so-called Bethe-Salpeter equation (BSE), computed on top of an accurate band structure including quasi-particle GWGW corrections, within the so-called GW+BSEGW+BSE scheme. Then, numerically, a material is foreseen to be an excitonic insulator if the BSE binding energy overcomes the GWGW band gap. This approach has been used in several works.[7, 8, 10, 11, 16] Recently, following this approach, a possible EI phase was identified by Jiang et al. [16] for the magnetic layered NiBr2_2. NiBr2_2 is quite unique for two reasons. First of all, it has a very large band gap (about 4 eV), compared to other materials explored so far. This would suggest that the EI phase could be ruled out from the beginning, since typical exciton binding energies range from few meV to few hundreds meV in bulk materials. Second, NiBr2_2 is magnetic, thus leading to an interplay between charge and spin degrees of freedom. As a result, the EI phase was predicted to be realized in one spin channel only. The authors referred to this case as an half-EI (HEI).

The GW+BSEGW+BSE scheme is a well established numerical scheme, and it works surprisingly well in describing the optical properties of a wide range of materials, ranging from bulk insulators and semi-conductors, to layered and 2D materials, 1D materials, such as carbon nanotubes, up to isolated systems such as clusters and molecules. In NiBr2_2, similarly to other EI candidates, the GW+BSEGW+BSE scheme results in solutions of the excitonic hamiltonian with slightly negative energies (few meV below zero). However the validity of the GW+BSEGW+BSE scheme to correctly capture poles with energy close, or even below, zero should be questioned from the beginning for various reasons. First, it is well known that the results of the GW+BSEGW+BSE scheme depend on the starting point, i.e. on the functional employed in the initial density functional theory (DFT) calculation. For example, if the initial band structure is computed starting from a DFT calculation using the PBE or the PBE0 approximation, one should refer to it as GWGW@PBE+BSE or GWGW@PBE0+BSE, respectively. Although a change of the functional might only result in small energy shifts, this could be enough to move an energy eigenvalue of the BSE across zero. Second, in this protocol approximations employed in the ground-state calculations are not consistent with the approximations used in the linear response part. As a consequence, for example, the GWGW@DFT+BSE scheme, if used to compute excitations in the spin flip channel, fails in reproducing the Goldstone sum rule for magnons, i.e. it fails to give a zero energy pole in magnetic materials [17]. This failure is independent of the DFT flavour, the corresponding error can be large, and it clearly shows that the scheme is not reliable for poles with energy close to zero.

Surprisingly, these issues have been discussed very little in the literature when addressing the existence of possible EI phases. In the present manuscript we use the quite unique case of NiBr2_2 to address this deficiency. We show that NiBr2_2 is an extreme case where one should be very careful in proposing the existence of an EI phase, since the state-of-the-art GW@GW@DFT+BSE scheme is so much affected by the starting point, that even qualitatively results can change completely. These findings are not specific to the case of NiBr2_2 but can be extended to the analysis of a wide range of materials.

The paper is organized as follows. In § 2 we define in a clear way what an EI phase (or an EI instability) is, and discuss the numerical protocols which properly address the existence of the EI phase in general. In particular we discuss why one has to be consistent in the approximations used to describe the ground state and the response of the system. In § 3 we present the results of our numerical simulations and discuss the band structure (subsection III A) as well as the absorption spectrum (subsection III B) of ferromagnetic NiBr2_2 using various approximations. We finally draw our conclusions and perspectives in Sec. IV.

Theoretical Protocol

The EI phase has two fundamental key properties: (i) it is a broken symmetry phase of electronic nature, and (ii) it is described by introducing the long range electron-hole interaction in the simulations. These are the two essential requirements that any numerical simulation predicting an EI phase should take into account.

Let us define the electronic hamiltonian HH, which also includes the potential of the atoms in a chosen crystal lattice structure, and its eigenvalues (at fixed electron number) EλE_\lambda, with E0E_0 the ground-state energy. We can define a corresponding point symmetry group SS and a group of discrete translations TT, such that [H,S]=0[H,S]=0 and [H,T]=0[H,T]=0. We call ΩST\Omega^{ST} the phase space of electronic states which respect such symmetry groups, Ω\Omega the global phase space, and ΩS\Omega^{S} (ΩT\Omega^{T}) the phase space which respects the SS (TT) group alone. We have Ω⊇ΩS⊇ΩST\Omega\supseteq\Omega^{S} \supseteq\Omega^{ST} (Ω⊇ΩT⊇ΩST\Omega\supseteq\Omega^{T} \supseteq\Omega^{ST}). The excitonic insulator phase arises from a spontaneous symmetry breaking, usually with respect to the SS group, and will be the focus of the present manuscript. Accordingly, we define E0STE_0^{ST} as the global energy minimum over ΩST\Omega^{ST}, and E0TE_0^{T} as the global energy minimum over ΩT\Omega^{T}. A broken-symmetry phase ground state exists if and only if E0T<E0STE_0^{T}<E_0^{ST}.

Standard ab initio simulations compute the ground state with the symmetry specified by the crystal lattice structure of the system, within a given approximation for the exchange and correlation (xc) energy; we indicate the corresponding total energy as E0T,xcgE_0^{T,xc_g}, with xcgxc_g referring to the xc approximation used in the ground state. There are then two options available to check if a broken-symmetry state with energy E0T,xcg<E0ST,xcgE_0^{T,xc_g}<E_0^{ST,xc_g} exists: (i.1) to perform a second ground-state simulation without symmetry or (i.2) to compute the spectrum of the neutral excitations of the material, ωλ=(Eλ−E0)\omega_\lambda=(E_\lambda-E_0), at zero transferred momentum.{^1} In the latter case there is no need to impose symmetry breaking. Indeed, neutral excitations are computed from the solution of the linear response function χ(ω)\chi(\omega). The excited-state energies, EλE_\lambda, and the corresponding wavefunctions, Ψλ\Psi_\lambda, which appear in the Lehmann representation of χ(ω)\chi(\omega), do not belong to ΩST\Omega^{ST} in general. More precisely, given the space symmetry group SS, the states Ψλ\Psi_\lambda can be classified as degenerate multiplets of the irreducible representations RSλR_S^\lambda of the space group. Symmetry operations of ΩST\Omega^{ST} can in general send a state of a given multiplet into any other state of the same multiplet. Because of that, the states of a monodimensional representation RSλR_S^\lambda are singlets, i.e. nondegenerate, and belong to ΩST\Omega^{ST}. Instead, states of a multidimensional representations RSIR_S^I do not belong to ΩST\Omega^{ST}, and they must be considered together to obtain a closed sub-space with respect to the symmetry operation of the group SS. If a multiplet of states has energy lower than the ground state, then a spontaneous symmetry breaking would be favored, with an emerging ground state composed by a random linear combination of the states in the multiplet.

If option (i.1) is chosen, the same level of approximation must be used in the ground-state calculation with and without symmetry. Comparing E0T,xcg1E_0^{T,xc_{g1}} and E0ST,xcg2E_0^{ST,xc_{g2}}, with xcg1xc_{g1} and xcg2xc_{g2} two different xc approximations used in the ground state, does not have any validity. Moreover, the long range electron-hole interaction needs to be considered. In some works the EI phase has been predicted on the basis of calculations using the local density approximation (LDA) or the generalized gradient approximation (GGA), i.e. xcg=xc_g= LDA/GGA. While also within these approximations a symmetry-broken phase could be obtained, such phases should not be labelled as EI, since LDA/GGA completely miss the long-range behaviour of the electron–hole interaction. Indeed, it is well known that time dependent DFT (TDDFT), within the LDA or GGA approximations (TDLDA or TDGGA), cannot describe excitons in the excited states, and thus these approximations cannot capture the EI phase in the ground state either.

In case option (i.2) is chosen, the total energy of possible broken-symmetry phases can be obtained by defining EλT,xc=E0ST,xcg+ωλST,xceE_\lambda^{T,xc}=E_0^{ST,xc_g}+\omega_\lambda^{ST,xc_e}, with ωλST,xce=EλT,xc−E0ST,xc\omega_\lambda^{ST,xc_e}=E_\lambda^{T,xc}-E_0^{ST,xc} a zero-momentum neutral excitation of the system, using consistent approximations between ground state (xcgxc_g) and excited state (xcexc_e) calculations. Then the criteria to establish the existence, for some λ0\lambda_0, of a broken-symmetry phase becomes ωλ0ST,xce<0\omega_{\lambda_0}^{ST,xc_e}<0. In such situation we can identify Eλ0T,xc≡E0T,xcE_{\lambda_0}^{T,xc}\equiv E_0^{T,xc} as the global ground state energy. The ab initio approach to describe excitons in the ab initio framework is BSE. ωλxce<0\omega_\lambda^{xc_e}<0, with xcexc_e referring to the xc approximation used for the BSE kernel, is then often used as the golden standard to predict the existence of the excitonic insulator phase. Within this approach the essential consistency condition at the core of option (i.2) would rely on the use of a one-body Green’s function which is a self-consistent solution of the Dyson equation for a given self-energy approximation and on the use of the same level of approximation for the self-energy and the BSE kernel [17].

However, in practical calculations, this consistency condition is violated since the state-of-the-art is the GWGW@DFT+BSE scheme. Indeed, the BSE kernel is obtained from the Hartree plus screened-exchange self-energy (HSEX) approximation, i.e. xce=HSEXx_{\mathrm{ce}} = \mathrm{HSEX}, while xcg=GWx_{\mathrm{cg}} = GW@DFT. The self-consistent DFT ground state with GWGW corrections should be replaced by a self-consistent simulation with xcg=HSEXx_{\mathrm{cg}} = \mathrm{HSEX}. Even ignoring this point and assuming that GWGW is a good starting point for a BSE calculation, at least an eigenvalue self-consistent GWGW (evGWGW) procedure should be carried out. However both SC-HSEX and evGWGW can be very demanding. An alternative path is to use DFT with hybrid functionals, which are computationally affordable. Hybrid functional simulations include (a fraction of) the long–range electron–hole interaction and can be used to capture EI phases. In this case xcg=xce=Hybridx_{\mathrm{cg}} = x_{\mathrm{ce}} = \mathrm{Hybrid} can be used, by employing TDDFT for the excited states. To underline this consistency, we will use here the notation PBE0+TDPBE0 for example for the PBE0 case. Indeed, it has been shown that hybrid functionals can capture excitonic excited states [18] and EI phases [9]. In general the exchange fraction entering hybrids is not properly screened, and this could be particularly problematic in 2D materials, since the 2D nature of the screening is completely missing. Nevertheless, as we shall see, it remains a valid approach to demonstrate the concept of consistency we discussed above.

In this work we will adopt the protocol described in option (i.2) to identify possible EI phases in NiBr2_2 and we will calculate the optical spectrum within GWGW@PBE+BSE, GWGW@PBE0+BSE and PBE0+TDPBE0. This will allow us to address both the starting-point dependence of the GWGW@DFT+BSE scheme and to consider a proper scheme with xcg=xcex_{\mathrm{cg}} = x_{\mathrm{ce}}.

Results

We studied NiBr2_2 in its ferromagnetic ground state. Density functional theory calculations were performed using the Quantum Espresso code [19] within both the Perdew-Burke-Ernzerhof (PBE) and the PBE0 exchange correlation functional [20]. Optimized norm-conserving Vanderbilt pseudopotentials have been generated using ONCVPSP [21]. The self-consistent (scf) simulations converged with an energy cutoff of 80 Ry and a k-point grid of 9×9×19 \times9 \times1 (the direct band gap at Γ\Gamma has been converged to 0.01 eV).

We then used the DFT wave-functions ψnkKS(r)\psi_{n\mathbf{k}}^{KS}(r) and energies ϵnkKS\epsilon_{n\mathbf{k}}^{KS} to perform GWGW and BSE simulations. For the PBE case we re-computed ψnkKS(r)\psi_{n\mathbf{k}}^{KS}(r) and ϵnkKS\epsilon_{n\mathbf{k}}^{KS} with a non self-consistent (nscf) calculation on a 19×19×119 \times19 \times1 k-point grid in the Brillouin zone. For the PBE0 case instead we used a 9×9×19 \times9 \times1 k-point grid. This is due both to the higher computational cost of PBE0, and to the fact that, with hybrid functionals, it is not possible, in the nscf calculations, to use a k-point grid different from the one used for the scf case. As we shall see, this restriction does not affect our conclusions. We computed the one-shot GWGW quasiparticle (QP) corrections on top of DFT using the Yambo code [22, 23]. Dynamical screening effects have been taken into account using the Godby-Needs plasmon-pole approximation (PPA) [24]. Both in DFT and, successively, in GWGW and BSE, we used a slab model with a 40 a.u. vacuum thickness. Moreover, the Coulomb cut-off technique [25] has been used to ensure no interactions between periodic images. In the GWGW step, we used a cutoff of 40 Ry for the wave-functions; an energy cut-off four times bigger, i.e. of 320 Ry, for VxcV_{\mathrm{xc}} and for the exchange part of the Self-Energy Σx\Sigma_{\mathrm{x}}; finally an energy cutoff of 10 Ry for the screening and the correlation parts of the self-energy Σc\Sigma_{\mathrm{c}} (see the Supplemental Material more for details [26]). We included a total of 200 bands in the screening and 500 in the evaluation of the self-energy. In the optical response we took into account excitonic effects via the Bethe-Salpeter Equation [27, 28]. The BSE absorption spectrum has been calculated using a static screening converged using the same parameters of the GWGW calculation. We used a cutoff of 4 Ry for the electron-hole exchange and 10 Ry for the screened interaction. We included bands from 7 to 23 to obtain a converged spectrum up to 10 eV.

In the following we demonstrate that the occurrence of negative energy solutions of the BSE strongly depends on the choice of the approximations used in the ground state and for the BSE kernel. In particular the lack of self-consistency in the GWGW band structure and of consistency between ground-state approximations and BSE kernel approximations can lead to a questionable EI phase.

Band structure

The starting point for a BSE calculation is a single-particle band structure obtained correcting the DFT energies. The approximation chosen for the DFT determines the quality of the screened interaction WW, which is computed within the RPA approximation using Kohn-Sham energies. This results in the so-called W0W_0 screened interaction. W0W_0 enters the one-shot GWGW or G0W0G_0W_0 approach, which is usually employed for the band structure. The one-shot GWGW is also employed in the present work. Even without invoking self-consistency, GWGW calculations can be problematic in 2D materials. We found, indeed, that, for the system under study, particular attention has to be paid when converging the GWGW quasiparticle corrections. See also the discussion in the Supplemental Material [26] (see, also, references [29, 30] therein).

In Fig. 1 we show PBE and PBE0 spin-resolved band structures, separated in majority (M) and minority (m) spin channels. The corresponding GWGW@PBE and GWGW@PBE0 band structures are also shown. In all cases the band gap is indirect, with 0.83 eV at the PBE level compared to 3.79 eV at the PBE0 level, and it is due to transitions from the majority valence band maximum (VBM) at Γ\Gamma towards the minority conduction band minimum (CBM), which is roughly half way along the Γ\GammaM high symmetry line. Also the direct band gaps within PBE0 (4.45 eV and 6.52 eV for m-m and M-M spin channels, respectively) are significantly larger than the ones obtained using PBE (1.33 eV and 4.09 eV for m-m and M-M spin channels, respectively). This is expected due to the fraction of HF exchange taken into account in the PBE0 functional.

Spin-resolved band structures for GW@PBE and GW@PBE0

Figure 1. Spin-resolved band structure (majority- and minority-spin channels are reported in blue and red, respectively) for GW@PBEGW@\mathrm{PBE} (left) and GW@PBE0GW@\mathrm{PBE0} (right) are reported with a continuous line. The DFT band structures used as starting point for the GWGW correction are reported with a dashed line. The VBM is set to zero.

Moving to GWGW, one would naively expect that, despite the starting-point dependence, the GW@PBEGW@\mathrm{PBE} and GW@PBE0GW@\mathrm{PBE0} gaps should be close to each other, with GW@PBEGW@\mathrm{PBE} approaching the expected self-consistent GWGW result from below, i.e. with the GWGW corrections opening the PBE gap, and with GW@PBE0GW@\mathrm{PBE0} from above, i.e. with the GWGW corrections closing the PBE0 gap. However this is clearly not the case as shown in Fig. 1. GW@PBEGW@\mathrm{PBE} opens the PBE gap from 0.83 to 2.62 eV. Direct band gaps are increased to 2.91 eV and 5.01 eV for the minority and majority spin channels, respectively. However GW@PBE0GW@\mathrm{PBE0} also opens the PBE0 gap from 3.79 to 5.75 eV. Direct band gaps are 6.56 eV and 7.41 eV for the minority and majority spin channels respectively. See also Tab. I for a comparison of all values. This clearly points to the fact that W0PBEW_0^{\mathrm{PBE}} and W0PBE0W_0^{\mathrm{PBE0}} are different and that, accordingly, the two G0W0G_0W_0 self-energies are different. One can then wonder

which result is the most reliable.

On one hand the GW@PBEGW@\mathrm{PBE} is the state-of-the-art scheme to compute quasi-particle band structure. Indeed, the PBE screening entering W0PBEW_0^{\mathrm{PBE}} is usually a reasonable approximation to the physical screening due to error cancellations between (i) the underestimated fundamental gap, and (ii) the neglected excitonic effects. On the other hand hybrid functionals like PBE0 are expected to better approximate quasi-particle wave-functions and energies. Also simulations based on the hybrid HSE0 functional [16, 31] give a band structure rather close to the PBE0 band-structure. However GWGW@hybrids is a scheme which has not been much tested yet, since the compatibility of GWGW codes with hybrid functionals is a quite recent feature, and, despite being very promising [32, 33, 34], they are not very well tested. By correcting the electronic gap, hybrid functionals tend to underestimate the electronic screening which may result in a overestimation of the GWGW@hybrids gap [32]. Indeed, in NiBr2_2 the underestimated screening leads to a very strong exchange term, even bigger than the exchange fraction entering in PBE0. As a consequence, it is more reasonable to directly compare GW@PBEGW@\mathrm{PBE} with hybrids like PBE0. The estimated experimental band gap, probed using STS on NiBr2_2 thin films, down to the monolayer, ranges between 3.4 and 4.5 eV, in good agreement with the PBE0 value [35].

Beside this evident, but yet quantitative difference, there is a more important qualitative difference between PBE and PBE0 band structures, which is also inherited by the corresponding GWGW band structure. Within PBE there exist three almost flat bands in the minority channel which are very close in energy to the top of the valence band in the majority spin channel. These bands are Nickel 3d3d states [16], in partic-

PBEPBE0GW
[30]@PBE@PBE [30]@PBE0
ind. gap M-m0.830.883.792.822.625.75
dir. gap m-m1.331.344.452.912.626.56
dir. gap M-M4.093.956.525.015.377.41
dir. gap M-m0.991.043.863.053.065.91

Table 1. Indirect and direct band gaps (in eV).

ular with t2gt_{2g} character, which are likely not well described within PBE. Indeed, they are not present within our PBE0 simulations, nor in the HSE0 and LDA+UU results in the literature [16, 31]. Moving to GWGW@PBE these same states remain there, pointing to the fact that the PBE wave-functions are not well described, and that PBE is not a good starting point. These states also show a quite peculiar behaviour in the GWGW simulations, and, as we shall see, they are responsible for the very sharp difference in the spectrum of GWGW@PBE+BSE compared to PBE0+TDPBE0 or GWGW@PBE0+BSE. In the Supplemental Material [26] we discuss the convergence of the GWGW@PBE quasi-particle corrections. We highlight the “unusual” behaviour of these flat Ni t2g↓t_{2g\downarrow} bands, and we discuss in detail the very slow convergence compared to the energy cutoff used in the description of the screened interaction. The discussion agrees with previous numerical studies [36], although this fact has been somehow overlooked in past simulations on 2D materials [16]. Indeed GWGW@PBE corrections on these occupied Ni t2g↓t_{2g\downarrow} states lead to a situation in which the majority and the minority VBM are very close in energy (during the GWGW convergence their relative position is even inverted). The direct gap switches character, from M-m to m-m, and the system almost becomes a direct gap material with an energy difference between GWGW@PBE direct and indirect gap of 9 meV (compared to 500 meV in PBE, 660 meV in PBE0 and 810 meV in GWGW@PBE0). The small m-m gap is also part of the reason why negative poles will appear in the response function calculated at the level of GWGW@PBE+BSE, as we will discuss in § 3 B.

The conclusion of this analysis is that NiBr2_2 is an indirect band gap semiconductor. The indirect gap is majority-minority, as expected in standard magnetic materials. Looking at the direct gap, this is also majority-minority, the minority-minority gap is slightly larger (less than 1 eV), while the majority-majority gap is significantly larger. This behaviour is due to the empty spin-minority Ni 3dd orbitals, with ege_g character, which lay in between the more predominantly pp-like valence and conduction bands.

Optical properties

We now turn our attention to the optical properties of NiBr2_2. In Fig. 2 the GWGW@PBE+BSE absorption spectrum is reported together with the GWGW@PBE0+BSE and the PBE0+TDPBE0 spectra. All spectra are dominated by an intense excitonic peak, which we label E0E_0. All methods describe E0E_0 as an exciton mostly located nearby Γ\Gamma in the BZ (see Fig. 3) and corresponding to a transition from pp bands ϵk↓\epsilon_{\mathbf{k}\downarrow} to d(eg)d(e_g) bands ϵdk↓\epsilon_{d\mathbf{k}\downarrow} in the spin minority channel. At energy larger then E0E_0 the spectral intensity is strongly reduced and eventually it goes up again at even larger energy. The position of the E0E_0 peak shifts from 3.3 eV (GWGW@PBE+BSE), to 3.6 eV (PBE0+TDPBE0), and to 4.1 eV (GWGW@PBE0+BSE). The E0E_0 energy is always lower compared to the (ϵdk↓−ϵpk↓)(\epsilon_{d\mathbf{k}\downarrow}-\epsilon_{p\mathbf{k}\downarrow}) energy difference. We define its binding energy bE0=min⁡k(ϵdk↓−ϵpk↓)−E0b_{E_0}=\min_{\mathbf{k}}(\epsilon_{d\mathbf{k}\downarrow}-\epsilon_{p\mathbf{k}\downarrow})-E_0. bE0≈1b_{E_0}\approx1 eV in both GWGW@PBE+BSE (1.1 eV) and PBE0+TDPBE0 (0.9 eV), while bE0=2.4b_{E_0}=2.4 eV for GWGW@PBE0+BSE. This discrepancy is again due to the underestimated RPA screening obtained using the PBE0 starting point. This leads to an overestimation of the band-gap which is here compensated by an overestimation of the binding energy and results in a peak position not very different from GWGW@PBE+BSE and PBE0+TDPBE0. As for the band structure case, the conclusion here is that it is better to directly compare GWGW@PBE+BSE to PBE0+TDPBE0.

Imaginary part of the macroscopic polarizability calculated within GW@PBE+BSE, GW@PBE0+BSE and PBE0+TDPBE0

Figure 2. Imaginary part of the macroscopic polarizability calculated within GWGW@PBE+BSE (upper panel), GWGW@PBE0+BSE (middle panel), and PBE0+TDPBE0 (lower panel). The results obtained within the corresponding independent particle approximation, i.e. ignoring electron-hole interaction, are also reported (dashed lines). Curves are vertically shifted for clarity. Bright and dark excitons energies are reported as vertical thicks.

The qualitative behaviour of the spectrum is instead significantly different at energy below E0E_0. All methods show poles at energy well below E0E_0, due to transitions between occupied and flat bands in the minority spin channel, which we identify as d(t2g)d(t_{2g}) states (see Supplemental Material [26]), and the empty spin minority d(eg)d(e_g)-states. However, within GWGW@PBE+BSE, this leads to a series of three bright excitons to appear, which we label DiD_i in Fig. 2 following the choice done in Ref. [16]. These peaks are located at 1.67 eV (D1D_1), 1.98 eV (D2D_2), 2.21 eV (D3D_3). As already pointed out, the description of the occupied t2gt_{2g} states is questionable in PBE, and so are the absorption features related to them. Indeed the series of the DD peaks is missing within GWGW@PBE0+BSE and PBE0+TD-BPE0, where all d(t2g)→d(eg)d(t_{2g}) \to d(e_g) transitions are dark. This is probably due to the fact that the PBE starting point mixes dd and pp bands character, while PBE0 does not.

Band structures and excitonic wavefunctions for three computational approaches

Figure 3. Band contribution to the main excitonic peak E0E_0: (left panel) GWGW@PBE+BSE, (middle panel) GWGW@PBE0+BSE, (right panel) PBE0+TDPBE0. Note that the plotted band structures are the DFT ones. We also report the excitonic wavefunction in reciprocal space.

Beside these aspects, we are here interested in discussing whether or not NiBr2_2 is an HEI. We thus need to turn our attention to the lowest energy solutions among the d(t2g)→d(eg)d(t_{2g}) \to d(e_g) transitions. As discussed, the occupied d(t2g)d(t_{2g}) states are very close to the valence band maximum within GWGW@PBE, while they are deeper in energy within PBE0 and GWGW@PBE0. As a consequence the GWGW@PBE+BSE approach produces negative eigenvalues. There is a doubly degenerate dark exciton at −0.17-0.17 eV (X1,1X_{1,1}) and a non-degenerate exciton at −0.16-0.16 eV (X1,2X_{1,2}). These two excitons share a similar physics and are labeled as X1X_1 in Fig. 2. Other two dark excitons located at 0.59 (X2,1X_{2,1}) and 0.68 eV (X2,2X_{2,2}) are indicated with X2X_2. Both X1X_1 and X2X_2 are associated with transitions between t2gt_{2g} and ege_g bands in the minority channel (see Supplemental Material [26]). The X1X_1 poles in particular are the ones which were used to claim the possible existence of an HEI ground state in Ref. [16]. Instead, within GWGW@PBE0+BSE and PBE0+TD-BPE0 no negative eigenvalues appear and therefore there is no EI phase. Since the qualitative behaviour of these three approaches is the same, and, as discussed in the introduction, the existence of a possible EI phase is mostly quantitative, we have now to address the question: which result is the most reliable? Both GWGW@DFT+BSE schemes are based on a one-shot GWGW calculation and plagued by inconsistencies between the approximation used for the ground state and the approximation used for the excited states. This points to the fact that, as far as this quantitative question is concerned, PBE0+TDPBE0 is the scheme to rely on. For NiBr2_2 it should be noticed that the GWGW@DFT+BSE inconsistency is not so drastic. Indeed the physics of both the ground state and the excited states is determined by the same screened interaction WW. However, the state-of-the-art GWGW@PBE+BSE is affected by a bad description of the occupied dd bands in the current study, which is here crucial to address this quantitative answer. GWGW@PBE0+BSE suffers from a PBE0 screening which is likely underestimated, although errors induced in the band structure are cancelled by the overestimation of the binding energy, thus affecting in a less severe way the final position of the BSE peaks. We also notice that in Ref. [16], the HEI phase is also supported by HSE0+BSE simulations where, however, the inconsistency between ground state and excited states approximation is more problematic.

In conclusion, our analysis of the band structure and the optical spectral features indicates that PBE0+TDPBE0 gives a good overall description of NiBr2_2. Since this scheme does not produce any negative-energy pole in the optical absorption, we conclude that NiBr2_2 is not an HEI. This is the most important result of this work.

Conclusions and perspectives

In this work we addressed the study of excitonic insulator phases within common ab initio theoretical approaches such as time-dependent density functional theory and the Bethe-Salpeter equation of Many-Body Perturbation Theory. We discussed the fact that in order to correctly capture this phase a theoretical approach should be able to describe two important features, i.e. an electronic-driven broken symmetry and the long range electron-hole interaction. That is why TDDFT within the common LDA/GGA approximations (which miss the long range electron-hole interaction) cannot correctly describe this phase. The BSE is more appropriate, but, in order to correctly describe an electronic-driven broken symmetry, it should be used within consistent approximations. We discussed that this corresponds to use the same level of approximation for the self-energy used in a self-consistent ground-state calculation and for the kernel of the BSE.

For the specific case of ferromagnetic NiBr2_2, by comparing the optical spectrum and the energy position of the lowest energy dark excitons obtained with different schemes, we arrived at the conclusion that the PBE0+TDPBE0 scheme is the most reliable, also because it uses the same level of approximation for the ground state and for the excited states. Within PBE0+TDPBE0 all dark poles have a positive energy, which indicates that NiBr2_2 is not an HEI, contrary to previous assignments. Moreover, we find that the occurrence of negative energy peaks in the GWGW@PBE+BSE is due to the starting point dependence of the GWGW scheme, rather than to the inconsistency between ground-state and excited state approximations. Indeed, the PBE starting point gives a questionable description of the t2gt_{2g} states, which leads, within GWGW@PBE, to t2gt_{2g} bands located too high with respect to the occupied pp states. This does not happen within PBE0 and GWGW@PBE0 (and also with respect to other approximations studied in the literature such as HSE0 and PBE+UU).

More in general, our study points to the fact that using consistent approximations in ab initio theories is of crucial importance in order to have reliable results.

Acknowledgments

S. Di Sabatino and P. Romaniello would like to acknowledge financial support by the EUR grant NanoX ANR-17-EURE-0009 in the framework of the “Programme des Investissements d’Avenir” and by the ANR (project ANR-18-CE30-0025 and ANR-19-CE30-0011). A. M.-S. acknowledge financial support by Ramón y Cajal programme (grant RYC2018-024024-I; MINECO, Spain), Agencia Estatal de Investigación (AEI), through the project PID2020-112507GB-I00 (Novel quantum states in heterostructures of 2D materials), and Generalitat Valenciana, program SEJI-GENT (reference 2021/034), project Magnons in magnetic 2D materials for a novel electronics (2D MAGNONICS), and Planes complementarios de I+D+I en materiales avanzados, project SPINO2D, reference MFA/2022/009.

Supplemental Material to “Assignment of excitonic insulators in ab initio theories: the case of NiBr₂”

Stefano Di Sabatino; Alejandro Molina-Sánchez; Pina Romaniello; Davide Sangalli

GW@PBE convergence

For monolayer NiBr₂, whose structure is shown in Fig. 1, particular attention has to be paid when converging the GWGW quasiparticle corrections. In Fig. 2(a) we report the band structure within GWGW@PBE using the following parameters: a 19×19×119 \times19 \times1 k-point grid, an energy cutoff of Ecut=10E_{\mathrm{cut}}=10 Ry for the screening and the correlation parts of the self-energy Σc\Sigma_{\mathrm{c}}, and Nb=200N_{\mathrm{b}}=200 bands in the screening. In Fig. 2(b) we report the same band structure after the introduction of the correction obtained from the extrapolation process described in the following. In Fig. 3(a)-(f) we report the difference between εnk\varepsilon_{\mathrm{nk}} at Γ\Gamma (See Fig. 2(a)) and the top valence band at Γ\Gamma as function of the energy cutoff in the screening. We notice that some bands in the spin up channel converge quite slowly with respect to the energy cutoff, showing a ∼1/Ecut\sim1/E_{\mathrm{cut}} behaviour. The Ecut→∞E_{\mathrm{cut}} \rightarrow\infty limit can be extrapolated from the finite cutoff data. As can be seen from the band structure in Fig. 2(a) the VBM is in the minority channel at a point about half way between the Γ\Gamma and M symmetry points. However increasing the energy cutoff we observe a shift between the t2gt_{2g} in the minority channel and the ege_g in the majority channel. This produces an inversion of the channel of the VBM, which is now located at the Γ\Gamma point. Results qualitatively similar to our unconverged band structure are also presented in Ref. [1], and in Ref. [2] for the analogous case of NiCl₂. The minority-minority spin gap, instead, does not pose any particular problem; at Ecut=10E_{\mathrm{cut}}=10 Ry the gap is converged by less than 0.1 eV. In Fig. 3(g)-(h) we report the convergence of the M-M and m-m direct gap at Γ\Gamma with respect to the number of bands in the screening. The convergence error for both gaps is less than 0.1 eV when using Nb=200N_{\mathrm{b}}=200. In Fig. 3(i) we report the convergence of the M-M, m-m and m-M direct band gaps at Γ\Gamma with respect to the k-point grid. No significant difference is observed between the two different spin channels. When using a 19×19×119 \times19 \times1 k-point grid, the correction due to the convergence with respect to the grid corresponds to a rigid shift of the conduction bands with respect to the valence bands of about -0.24 eV. For comparison we report the band gap obtained using the stochastic integration scheme for the screened potential described in Ref. 3 in order to speed up the convergence with respect to the number of k points.

NiBr₂ monolayer structure

Figure 1 (PDF p. 8). NiBr₂ monolayer structure. Ni atoms are indicated in red, Br atoms in white.

Band structures showing GW @ PBE and GW @ PBE w corrections

Figure 2 (PDF p. 9). Convergence of the QP GWGW corrections. (a) Band structure using a 19×19×119 \times19 \times1 k-point grid, an energy cutoff of Ecut=10E_{\mathrm{cut}} = 10 Ry for the screening and the correlation parts of the self-energy, Nb=200N_b = 200 bands in the screening, and 500 bands for the evaluation of the self-energy. Symbols indicate the different bands considered. (b) Band structure after the introduction of the corrections obtained from the extrapolation procedure.

Convergence plots for band gaps at the Γ point

Figure 3 (PDF p. 10). (c)-(h) Convergence of the difference ϵn−ϵVBM\epsilon_{\mathrm{n}}-\epsilon_{\mathrm{VBM}} at the Γ point with respect to the energy cutoff EcutE_{\mathrm{cut}} in the screening and the correlation part of the self-energy. The convergence is studied with a 7×7×17 \times7 \times1 k-point grid. Some bands in the majority spin channel show a slower convergence, with a 1/Ecut1/E_{\mathrm{cut}} behaviour. For these bands an extrapolation procedure can be used. The effect is an inversion of the top valence band at Γ. (i)-(j) Convergence of the m-m, M-M gaps at Γ with respect to the number of bands NbN_b in the screening. (k) Convergence of the M-M, m-m and m-M gaps at Γ with respect to the number of k points. Results obtained with the random integration scheme proposed in Ref. 3 are also presented with a dashed line for comparison.

Excitons Analysis

We present here a detailed analysis of the excitonic wave-functions of the dark energy states for GW@PBE+BSE, GW@PBE0+BSE, and PBE0+TD-PBE0.

The atomic lattice used for NiBr2_2 respects the D3dD_{3d} space group with 12 symmetry operation divided into 6 irreducible representations. Four representations of dimension 1 (A1gA_{1g}, A2gA_{2g}, A1uA_{1u}, and A2uA_{2u}) and two representations of dimension 2 (EgE_g and EuE_u). Bright excitons belong either to the EuE_u (in plane bright) or the A2uA_{2u} (out of plane bright). We remark that only negative poles belonging to the EgE_g (dark excitons) or EuE_u (bright excitons) groups can lead to an excitonic insulator phase according to the discussion in the main text.

Within GW@PBE+BSE the dark excitons are located at -0.17 eV (X1,1X_{1,1}, doubly degenerate) -0.16 eV (X1,2X_{1,2}, non degenerate) 0.59 eV (X2,1X_{2,1}, doubly degenerate) 0.68 eV (X2,2X_{2,2}, non degenerate) plus three all doubly degenerate bright excitons at 1.67 eV

(D1D_{1}), 1.98 eV (D2D_{2}), and 2.21 eV (D3D_{3}). In Figure 4 we present the wave function of the excitons here considered with the band transitions involved. Both the dark excitons X1X_{1}, X2X_{2} and the bright excitons DiD_{i} are related to t2g→egt_{2g} \to e_{g}.

GW@PBE+BSE excitonic wavefunctions represented on the PBE band structure and in reciprocal space

Figure 4. GW@PBE+BSE excitonic wavefunction represented on the PBE band structure (upper panels), and in reciprocal space (lower panels). We used the PBE bands for simplicity, since GW only introduces an energy shift without changing the band character. First 4 dark excitons (X1,1X_{1,1}, X1,2X_{1,2}, X2,1X_{2,1}, X2,2X_{2,2}) and 3 bright excitons (D1D_1, D2D_2, D3D_3).

Within GW@PBE0+BSE the dark excitons are located at 1.45 eV (X1,1X_{1,1}, doubly degenerate), 1.51 eV (X1,2X_{1,2}, non-degenerate), 2.00 eV (X2,1X_{2,1}, non-degenerate), 2.05 eV (X2,2X_{2,2}, doubly degenerate), and 3.99 eV (X3X_{3}, doubly degenerate). Excitonic wave functions and band transition involved are reported in Figure 5

GW@PBE0+BSE excitonic wavefunction represented on the PBE0 band structure and in reciprocal space

Figure 5. GW@PBE0+BSE excitonic wavefunction represented on the PBE0 band structure (upper panels), and in reciprocal space (lower panels). We used the PBE0 bands for simplicity, since GW only introduces an energy shift without changing the band character. First 5 dark excitons (X1,1X_{1,1}, X1,2X_{1,2}, X2,1X_{2,1}, X2,2X_{2,2}, X3X_{3}).

Within TD-PBE0 we obtain dark excitons at 1.07 eV (X1X_{1}, triply degenerate), 1.31 eV (X2,1X_{2,1}, doubly degenerate), 1.34 eV (X2,2X_{2,2}, nondegenerate), and 3.55 eV (X3X_{3}, doubly degenerate). Excitonic wave functions and band transitions involved are reported in Figure 6.

PBE0+TDPBE0 excitonic wavefunction represented on the PBE0 band structure and in reciprocal space

Figure 6. PBE0+TDPBE0 excitonic wavefunction represented on the PBE0 band structure (upper panels), and in reciprocal space (lower panels). First 4 dark excitons (X1X_{1}, X2,1X_{2,1}, X2,2X_{2,2}, X3X_{3}).

References

  1. [1]D. Jérôme, T. M. Rice, and W. Kohn, Phys. Rev. 158, 462 (1967).
  2. [2]B. Bucher, P. Steiner, and P. Wachter, Phys. Rev. Lett. 67, 2717 (1991).
  3. [3]F. X. Bronold and H. Fehske, Phys. Rev. B 74, 165107 (2006).
  4. [4]H. Cercellier, C. Monney, F. Clerc, C. Battaglia, L. Despont, M. G. Garnier, H. Beck, P. Aebi, L. Patthey, H. Berger, and L. Forró, Phys. Rev. Lett. 99, 146403 (2007).
  5. [5]J.-J. Su and A. H. MacDonald, Nature Physics 4, 799 (2008).
  6. [6]Y. Wakisaka, T. Sudayama, K. Takubo, T. Mizokawa, M. Arita, H. Namatame, M. Taniguchi, N. Katayama, M. Nohara, and H. Takagi, Phys. Rev. Lett. 103, 026402 (2009).
  7. [7]S. S. Ataei, D. Varsano, E. Molinari, and M. Rontani, Proceedings of the National Academy of Sciences 118, 10.1073/pnas.2010110118 (2021), https://www.pnas.org/content/118/13/e2010110118.full.pdf.pnas.org/content/118/13/e2010110118.full.pdf
  8. [8]D. Varsano, S. Sorella, D. Sangalli, M. Barborini, S. Corni, E. Molinari, and M. Rontani, Nature Communications 8, 1461 (2017).
  9. [9]M. Hellgren, J. Baima, and A. Acheche, Phys. Rev. B 98, 201103 (2018).
  10. [10]S. Dong and Y. Li, Phys. Rev. B 102, 155119 (2020).
  11. [11]Z. Jiang, W. Lou, Y. Liu, Y. Li, H. Song, K. Chang, W. Duan, and S. Zhang, Phys. Rev. Lett. 124, 166401 (2020).
  12. [12]G. Mazza, M. Rosner, L. Windgätter, S. Latini, H. Hübener, A. J. Millis, A. Rubio, and A. Georges, Phys. Rev. Lett. 124, 197601 (2020).
  13. [13]E. Baldini, A. Zong, D. Choi, C. Lee, M. H. Michael, L. Windgaetter, I. I. Mazin, S. Latini, D. Azoury, B. Lv, A. Kogar, Y. Wang, Y. Lu, T. Takayama, H. Takagi, A. J. Millis, A. Rubio, E. Demler, and N. Gedik, The spontaneous symmetry breaking in ta2nise5 is structural in nature (2020).DOI
  14. [14]B. Sun, W. Zhao, T. Palomaki, Z. Fei, E. Runburg, P. Malinowski, X. Huang, J. Cenker, Y.-T. Cui, J.-H. Chu, X. Xu, S. S. Ataei, D. Varsano, M. Palummo, E. Molinari, M. Rontani, and D. H. Cobden, Nature Physics 18, 94 (2022).
  15. [15]D. Chen, Z. Lian, X. Huang, Y. Su, M. Rashetnia, L. Ma, L. Yan, M. Blei, L. Xiang, T. Taniguchi, K. Watanabe, S. Tongay, D. Smirnov, Z. Wang, C. Zhang, Y.-T. Cui, and S.-F. Shi, Nature Physics 18, 1171 (2022).
  16. [16]Z. Jiang, Y. Li, W. Duan, and S. Zhang, Phys. Rev. Lett. 122, 236402 (2019).
  17. [17]Müller, M. C. T. D. (2017). Spin-wave excitations and electron-magnon scattering in elementary ferromagnets from ab initio many-body perturbation theory.
  18. [18]J. Paier, M. Marsman, and G. Kresse, Phys. Rev. B 78, 121201 (2008).
  19. [19]P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, C. Cavazzoni, C. Ceresoli, G. L. Chiarotti, M. Cococcioni, I. Dabo, A. D. Corso, S. de Gironcoli, S. Fabris, G. Fratesi, R. Gebauer, U. Gerstmann, C. Gougoussis, A. Kokalj, M. Lazzeri, L. Martin-Samos, F. Marzari, F. Mauri, R. Mazzarello, S. Paolini, A. Pasquarello, L. Paulatto, C. Sbraccia, S. Scandalo, G. Schlauzero, A. P. Seit­sonen, A. Smogunov, P. Umari, and R. M. Wentzcovitch, Journal of Physics: Condensed Matter 21, 395502 (2009).
  20. [20]J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).
  21. [21]D. R. Haman, Phys. Rev. B 88, 085117 (2013).
  22. [22]A. Marini, C. Hogan, M. Grüning, and D. Varsano, Computer Physics Communications 180, 1392 (2009).
  23. [23]D. Sangalli, A. Ferretti, H. Miranda, C. Attaccalite, I. Marri, E. Cannuccia, P. Melo, M. Marsili, F. Paleari, A. Marrazzo, G. Prandini, P. Bonfà, M. O. Atambo, F. Affinito, M. Palummo, A. Molina-Sánchez, C. Hogan, M. Grüning, D. Varsano, and A. Marini, Journal of Physics: Condensed Matter 31, 325902 (2019).
  24. [24]R. W. Godby and R. J. Needs, Phys. Rev. Lett. 62, 1169 (1989).
  25. [25]C. A. Rozzi, D. Varsano, A. Marini, E. K. U. Gross, and A. Rubio, Phys. Rev. B 73, 205119 (2006).
  26. [26]See Supplemental Material at [URL will be inserted by publisher].
  27. [27]R. M. Martin, L. Reining, and D. M. Ceperley, Interacting electrons: Theory and computational approaches (Cambridge University Press, 2016).DOI
  28. [28]M. Rohlfing and S. G. Louie, Phys. Rev. B 62, 4927 (2000).
  29. [29]A. Guandalini, P. D’Amico, A. Ferretti, and D. Varsano, arXiv preprint arXiv:2210.05146 (2022).
  30. [30]Computational 2D materials database (C2DB).DOI
  31. [31]M. Mushtaq, Y. Zhou, and X. Xiang, RSC Adv. 7, 22541 (2017).
  32. [32]F. Fuchs, J. Furthmüller, F. Bechstedt, M. Shishkin, and G. Kresse, Phys. Rev. B 76, 115109 (2007).
  33. [33]L. Leppert, T. Rangel, and J. B. Neaton, Phys. Rev. Mater. 3, 103803 (2019).
  34. [34]S. E. Gant, J. B. Haber, M. R. Filip, F. Sagredo, D. Wing, G. Ohad, L. Kronik, and J. B. Neaton, Phys. Rev. Mater. 6, 053802 (2022).
  35. [35]D. Bikaljević, C. González-Orellana, M. Peña-Diaz, D. Steiner, J. Dreiser, P. Gargiani, M. Foerster, M. A. Niño, L. Aballe, S. Ruiz-Gomez, N. Friedrich, J. Huelle, L. Jingcheng, M. Ilyn, C. Rogero, and J. I. Pascual, ACS Nano 15, 14985 (2021), pMID: 34491033, https://doi.org/10.1021/acsnano.1c05221.
  36. [36]F. Huser, T. Olsen, and K. S. Thygesen, Phys. Rev. B 88, 245309 (2013).

Paper details

vlnc:W_paqn5ney
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