betaFeedback

WorksPhysics and Astronomy

Impact of structural distortions on the correlated electronic structure of orbital-selective Mott insulating Na3_3Co2_2SbO6_6 under strains

Abstract

Na3_{3}Co2_{2}SbO6_6 is a promising candidate to realize the Kitaev spin liquid phase since the large Kitaev spin exchange interaction is tunable via the change in electronic structure, such as the trigonal crystal field splitting (ΔTCFΔ_{TCF}). Here, we show that the uncorrelated electronic structure of Na3_{3}Co2_{2}SbO6_6 is rather insensitive to the strain effect due to the low crystal symmetry accompanied by oxygen displacements and the presence of Sb ss orbitals. This suggests that the Kitaev spin-exchange interaction obtained from perturbation theory also does not depend much on the strain effect. Using density functional theory plus dynamical mean field theory, we find that the correlated electronic structure of Na3_{3}Co2_{2}SbO6_6 is an orbital selective Mott insulating state where the trigonal a1ga_{1g} orbital is insulating due to correlation-assisted hybridization, while other dd orbitals behave as typical Mott insulators, resulting in tunability of ΔTCFΔ_{TCF} under the strain effect effectively. Our results show that the local Co-site symmetry and dynamical correlation effects will play an important role in engineering the novel magnetic phase in this and related materials.

Bond-dependent spin exchange interaction in a honeycomb structure plays a crucial role in materials for stabilizing the Kitaev quantum spin liquid (QSL) phase¹. In particular, 4d4d or 5d5d orbitals in a honeycomb structure have been attracting much interest in realizing anisotropic spin interactions due to their large spin-orbit coupling (SOC)². However, the spatially extended 4d4d or 5d5d orbitals typically generate longer-range spin interaction terms, making the magnetic phase diagram much more complex. Moreover, it is rather difficult to tune the SOC effect via structural changes such as strain or pressure. Recently, 3d3d orbitals have been suggested to stabilize the Kitaev QSL phase due to the anisotropic hopping mechanism despite their small SOC³,⁴. In this regard, Na₃Co₂SbO₆ (NCSO) has drawn much attention as a possible material candidate due to its tunability close to the QSL phase⁵.

NCSO has monoclinic structure with space group C2/mC2/m and undergoes a magnetic phase transition from a paramagnetic (PM) insulating state to an antiferromagnetic (AFM) insulating state at TN∼5T_N \sim5 K⁶–⁹. The Co²⁺ (d7d^7) ion is in the high spin (HS) state with the electronic configuration of t2g3t2g2eg2t_{2g}^{3}t_{2g}^{2}e_g^{2} (S=3/2S = 3/2, L=1L = 1), forming a spin-orbit-entangled pseudospin Jeff=1/2J_{\mathrm{eff}} = 1/2 state, from which the Kitaev interaction arises via the second or higher order perturbation³–⁵,¹⁰,¹¹. It is shown that the t2gt_{2g}-ege_g hopping channel in NCSO enhances the Kitaev interaction JKJ_K while the non-Kitaev terms (Heisenberg JHJ_H, off-diagonal Γ\Gamma, and trigonal Γ′\Gamma') are much weakened⁵. Also, the stability of the QSL phase depends sensitively on the size of the trigonal crystal field splitting (ΔTCF\Delta_{TCF}) since a large ΔTCF\Delta_{TCF} can weaken the Kitaev interaction. Therefore, tuning the ΔTCF\Delta_{TCF} or related material parameters via strain or pressure is a priority of driving the system into Kitaev QSL phase.

The experimental estimates of electronic structure in NCSO, such as the sign and magnitude of the ΔTCF\Delta_{TCF}, have been relied on fitting the crystal field multiplet calculations to the spectroscopic measurements including X-ray absorption spectroscopy (XAS), inelastic neutron scattering (INS), X-ray linear dichroism (XLD), and X-ray magnetic circular dichroism (XMCD). Kim et al found that the magnitude of the ΔTCF\Delta_{TCF} is 25.1 meV by fitting to the XAS/XLD data, and the a1ga_{1g} orbital is located below the egπe_g^\pi orbitals¹². However, Veenen­daal et al reported that the a1ga_{1g} orbital is located above the egπe_g^\pi orbitals with ΔTCF\Delta_{TCF} of 35-60 meV by fitting configuration interaction analysis to the XLD and XMCD data¹³. This sign of ΔTCF\Delta_{TCF} has been attributed to the hybridization of egπe_g^\pi orbitals with the positive Sb5+Sb^{5+} ion located in the honeycomb Co layers, and it is also consistent with the analysis using the INS data¹⁴.

Although most theoretical works of studying the QSL phase in NCSO have focused on computing the anisotropic spin exchange parameters of the localized spin Hamiltonian based on perturbation theory assuming the pure trigonal distortion, more realistic treatment of structural changes in the monoclinic NCSO under strain or pressure and its impact on the correlated electronic structure and the spin-exchange interactions have not been explored yet. Moreover, a Jahn-Teller (JT) distortion is typically allowed in the monoclinic C2/mC2/m phase such as the NCSO case, while other QSL candidate materials such as BaCo₂(AsO₄)₂ (BCAO)¹⁰,¹⁵,¹⁶, γ\gamma-BaCo₂(PO₄)₂¹⁷,¹⁸, or α\alpha-RuCl₃¹⁹, have a pure trigonal distortion associated with R3ˉR\bar{3}. A JT distortion can further split degenerate orbitals into different energy levels, which could result in some novel physical phenomena²⁰. In addition, the properties of material parameters such as ΔTCF\Delta_{TCF} depend on the orbital-dependent correlation effect of the Co²⁺ ion due to the strong on-site UU and its hybridization effect with the ligand fields of O²⁻ and Sb⁵⁺ orbitals, which can be treated accurately using first principles.

Here, we study the impact of structural distortions of NCSO under strain on its correlated electronic structure and spin-exchange interactions using first-principles. We first perform structural relaxations of strained structures using density functional theory (DFT)+UU[21] based on the projected-augmented wave (PAW) method[22] as implemented in the Vienna ab initio simulation package (VASP)[23, 24]. All atomic positions of NCSO are relaxed to obtain the strained structures, while the volume is fixed to 270.84 Å, which is the zero-strain relaxed volume. The exchange-correlation energy functional was treated using the generalized gradient approximation (GGA) adopting the Perdew-Burke-Ernzerhof (PBE) functional[25]. The cutoff energy for the plane-wave basis was used as 600 eV with the Gamma-centered $8 \times 4 \times 8 kk-point mesh. The Hellmann-Feynman force on each atom was set to be smaller than 0.01 eV/Å for convergence. We imposed the Hubbard UU and Hund coupling JJ on the Co 3d3d orbitals with U=5U=5 eV and J=0.8J=0.8 eV for DFT+UU calculations. We also impose the Néel AFM order for the relaxation since this is the ground-state spin configuration (see Appendix § 0.3).

We then performed DFT+dynamical mean field theory (DMFT) calculation[26] for each relaxed structure to study the dynamical correlation effects on NCSO. For DMFT calculations, we construct the Wannier Hamiltonian using the Wannier90 code[27] for the Co dd, O pp, and Sb ss orbitals. The Wannier Hamiltonian of this spdspd-model and its comparison to the DFT band is given in Appendix § 0.4 and Appendix C. Due to the low point group symmetry (C2hC_{2h}) of CoO6_6 octahedra, the off-diagonal elements in Wannier Hamiltonian are non-negligible. To minimize the effect of the off-diagonal elements, we diagonalize the Co dd block in the full spdspd-Hamiltonian by applying unitary rotation matrix[28], and use the diagonalized Hamiltonian for DFT+DMFT loop. Then, we solve the DMFT impurity problem using the continuous time quantum Monte Carlo method[29] with the Hubbard U=5U=5 eV and the Hund’s coupling J=0.8J=0.8 eV to treat correlations of Co $3d$ orbitals. Once the DMFT calculations converge, all final results including the spectral function, the occupancy matrix, and the self-energy Σ(ω)\Sigma(\omega) are obtained as the original trigonal basis using the unitary transform. Here, we adopt single-site DMFT to treat the local dynamical correlation effect since this approximation has been successfully applied to the 3d3d transition-metal compounds in a similar honeycomb structure$[30, 31]. The details of DFT+DMFT calculation are given in Appendix D.

Experimental studies of neutron and X-ray diffractions show that NCSO forms a\u0020centrosymmetric monoclinic cell with the space group C2/mC2/m ([No. 12])[6, 7, 8, 9]. The DFT+U$ relaxed structure with the AFM Néel configuration exhibits a slightly larger volume along with the elongated in-plane bb-axis compared to the experimental one (see Table I). While Co ions in NCSO form a layered honeycomb structure, the local Co-O octahedra follow the trigonal symmetry (Figure 1 (a,b)), similarly to those in transition-metal dihalides such as CoI2_2 with the space group P3ˉm1P\bar{3}m1 ([No. 164]). This trigonal distortion splits the t2gt_{2g} triplet of the Co 3d3d manifold into an a1ga_{1g} singlet and a degenerate egπe_g^\pi doublet, where the trigonal crystal field splitting ΔTCF\Delta_{TCF} is defined as Ea1g−EegπE_{a_{1g}}-E_{e_g^\pi} (see Figure 1c)$[20, 32, 33, 34]. However, the point group symmetry of the local CoO6_6-site in NCSO is C2hC_{2h}, which is lower than that of CoI2_2 (D3dD_{3d}). This lower local-site symmetry can play an important role in tuning the correlation effects in NCSO and related low-symmetry compounds.

Image not extracted

Figure 1. The (a) side and (b) top views of the crystal structure, Na3_3Co2_2SbO6_6. L1≠L2≠L3L_1 \ne L_2 \ne L_3 due to structural distortions. (c) A schematic diagram of orbital energy level changes due to the trigonal compression and oxygen displacements. (The cc-axis corresponds to the [111] direction along the local Co–O axis).

The trigonal distortion of the Co–O octahedra can be parametrized by the L2/L1L_2/L_1 and L3/L1L_3/L_1 ratios where L1L_1 is the in-plane O–O bond lengths and L2L_2 and L3L_3 are the out-of-plane O–O bond lengths as shown in Figure 1(a,b). Here, L3L_3 is the edge that intersects the nearest-neighbor Co–Co bonding, while L2L_2 is not. The values of L1L_1, L2L_2, and L3L_3 computed for the fully relaxed NCSO structure without any strains are given and compared to known experimental structures in Table I. Under zero strain (Δa=0\Delta a=0), the fully relaxed NCSO structure shows the trigonal compression along the cc-axis, yielding the average L2/L1=0.878L_2/L_1=0.878 and L3/L1=0.937L_3/L_1=0.937, both of which are less than one. Additionally, the monoclinic distortion in NCSO results in unequal Co–O bond lengths within the Co–O octahedra, akin to the JT distortion observed in

Na3_3Co2_2SbO6_6VV [Å3^3]aa [Å]bb [Å]cc [Å]β\beta [∘^\circ]L1L_1 [Å]L2L_2 [Å]L3L_3 [Å]Co-O [Å]
DFT+UU Néel270.845.409.335.66108.53.212/3.207/3.1992.828/2.809/2.8093.019/2.999/2.9992.161/2.166
Viciu et al.⁶267.225.379.285.65108.53.172/3.168/3.1572.757/2.785/2.7852.915/2.927/2.9272.136/2.125
Wong et al.⁷266.815.369.295.65108.43.169/3.167/3.1522.758/2.784/2.7842.915/2.926/2.9262.132/2.127

Table 1. Lattice and structural parameters of Na3_3Co2_2SbO6_6 without any strains (Δa=0\Delta a = 0).

various perovskite transition-metal oxides. This is due to the anisotropic elongation of the oxygen positions, which alters the ratio between the Co–O bond lengths along the cc-axis and those in the abab-plane, i.e., (Co−O)z^/(Co−O)x^/y^(\mathrm{Co-O})_{\hat{z}}/(\mathrm{Co-O})_{\hat{x}/\hat{y}}. Although the JT distortion can be small in the unstrained state, its effects become more pronounced under applied strain, and a significant JT distortion is also evident in the experimental structure⁶,⁷.

Without the JT distortion, the trigonal a1ga_{1g} and egπe_g^\pi orbitals are the appropriate basis functions for diagonalizing the local Co dd Hamiltonian under trigonal distortion (see Appendix § 0.4). In general, even in the absence of JT distortions, unequal values of L2L_2 and L3L_3 (L2≠L3L_2 \ne L_3) are allowed under trigonal symmetry. For example, BCAO with the space group R3ˉR\bar{3} ([No. 148]) shows L2≠L3L_2 \ne L_3 with the local-site symmetry of C3C_3, but there are no JT distortions and Co-O bonds are still equivalent. In contrast, all out-of-plane O-O bond lengths are equivalent with L2=L3L_2=L_3 in CoI2_2. In NCSO, however, the local-site symmetry is further reduced by the JT distortion, which induces L2≠L3L_2 \ne L_3 and an energy splitting between the egπe_g^\pi orbitals (δJT\delta_{JT}). This results in the non-zero off-diagonal terms in the local Co dd Hamiltonian represented using the trigonal basis.

To understand the response of ΔTCF\Delta_{TCF} and δJT\delta_{JT} due to structural changes, we compute the local Co-site spdspd-Hamiltonian by projecting the DFT band structure to trigonal basis functions. In the case of both CoI2_2 and BCAO, the off-diagonal terms of the local Hamiltonian in the t2gt_{2g} manifold represented using trigonal basis functions are zeros, and ΔTCF\Delta_{TCF} can be defined as Ea1g−EegπE_{a_{1g}}-E_{e_g^\pi}. The δJT\delta_{JT} term is zero, thus the two egπe_g^\pi orbitals are degenerate. For NCSO, the low Co-site symmetry lifts the degenerate egπe_g^\pi orbital energy levels and results in non-negligible off-diagonal terms in the trigonal Hamiltonian. We define both terms as ΔTCF=Ea1g−(Eeg+π+Eeg−π)/2\Delta_{TCF}=E_{a_{1g}}-(E_{e_{g+}^\pi}+E_{e_{g-}^\pi})/2 and δJT=Eeg+π−Eeg−π\delta_{JT}=E_{e_{g+}^\pi}-E_{e_{g-}^\pi}.

Fig. 2a shows that the trigonal distortions of NCSO and CoI2_2 occur even at zero strain, resulting in a Co–O octahedron compressed along the cc-axis with L2/L1L_2/L_1 and L3/L1<1L_3/L_1<1. The corresponding structural values agree well with the experimental data (see Table I) and these distortions are further enhanced under tensile strains (Δa>0\Delta a > 0). In Figure 2c, we plot ΔTCF\Delta_{TCF} as a function of strain for each relaxed structure by extracting the on-site orbital energies from trigonal Wannier orbitals projected onto the uncorrelated DFT band structure. In both compounds, the ΔTCF\Delta_{TCF} is positive (∼21.5\sim21.5 meV for CoI2_2 and ∼12.8\sim12.8 meV for NCSO) at zero strain. This ΔTCF\Delta_{TCF} of NCSO is consistent with the experimental esti-

Image not extracted

Figure 2. The strain effect on structural parameters of Na3_3Co2_2SbO6_6 and CoI2_2: (a) the ratio L2/L1L_2/L_1 (solid line) and L3/L1L_3/L_1 (dashed line), (b) the Co-Co interlayer distance, (c) the trigonal crystal field ΔTCF\Delta_{TCF}, and (d) the energy splitting within the egπe_g^\pi orbitals (δJT\delta_{JT}).

mate obtained using INS¹⁴. Under the compression along the trigonal cc-axis, the negative ΔTCF\Delta_{TCF} is conventionally expected due to the stronger hybridization of egπe_g^\pi orbitals with the O ions when $L_2/L_1 < 1{sup}34 as the geometric lobes of the $e_g^\pi$ ($a_{1g}$) orbitals point to the octahedral faces parallel (perpendicular) to the $c$-axis{citep}³²,³³,³⁵. However, it is important to note that the Wannier orbitals are maximally localized, and the hybridization effect on the crystal field splitting is minimized. Moreover, the Sb $s$-orbitals also hybridize with O $p$ orbitals and weaken the Co-O hybridization effect. We find that the Co-Co interlayer distance does not have much effect on the crystal field splitting as it is large and does not depend much on strains ({xref}Fig. 2b`).

Although the structural changes (L2/L1L_2/L_1 and L3/L1L_3/L_1) due to strains are similar for both NCSO and CoI2_2 (Figure 2a), the response of the material parameters including ΔTCF\Delta_{TCF} and δJT\delta_{JT} of NCSO are much less sensitive to strains. The ΔTCF\Delta_{TCF} of CoI2_2 increases under tensile strains and changes sensitively about 27 meV per 1% of strain (Figure 2c). This trend of the positive ΔTCF\Delta_{TCF} is consistent with that of TiCl2_2 (the same point group as CoI2_2) when L2/L1<1L_2/L_1 < 1 as discussed in Georgescu et al.[32]. For NCSO, the change of structural parameters does not directly impact the energy levels of localized {a1g,egπ}\{a_{1g}, e_g^{\pi}\} orbitals due to the low Co-site symmetry. At zero strain, the JT distortion further splits the egπe_g^{\pi} orbitals of NCSO with the crystal splitting of δJT≃1.1\delta_{JT} \simeq1.1 meV and δJT\delta_{JT} changes about 6 meV per 1% of strain (Figure 2d).

We now study the effect of dynamical correlations on the electronic structure of strained NCSO in the paramagnetic phase. Since the five Co dd orbitals in NCSO are not equivalent, we treat the correlation effect of them separately with distinct DMFT self-energies. Without the effects of the Hubbard UU and the Hund’s coupling JJ (DFT calculation), the ground state of NCSO is the metallic state with a low-spin electronic configuration (egπ)4(a1g)2(egσ)1(e_g^{\pi})^4(a_{1g})^2(e_g^{\sigma})^1. The DFT DOS shows that the low-energy bands of a1ga_{1g} and egπe_g^{\pi} orbitals are fully filled, leaving a single electron on the doublet egσe_g^{\sigma} state (Figure 3a).

Image not extracted

Figure 3. (a) DFT DOS of NCSO at zero strain. (b) DFT+DMFT DOS of NCSO with U=5U=5 eV and J=0.8J=0.8 eV within paramagnetic phase. The imaginary part of the DMFT self-energy Σ(ω)\Sigma(\omega) obtained for (c) a1ga_{1g}, egπe_g^{\pi}, and (d) egσe_g^{\sigma} orbitals.

The paramagnetic DFT+DMFT calculation using U=5U = 5 eV and J=0.8J = 0.8 eV shows that NCSO at the zero strain becomes an orbital-selective Mott insulator with an energy gap of 1 eV. The egσe_g^{\sigma} bands under the effect of strong correlations become almost half-filled (Figure 3b) with the occupancy Negσ=2.16N_{e_g^{\sigma}} = 2.16. Figure 3d shows that the imaginary part of self-energy (ImΣ(ω)\mathit{Im}\Sigma(\omega)) of the egσe_g^{\sigma} orbitals has a strong pole (divergent) at ω∼0.8\omega\sim0.8 eV with the broad upper Hubbard bands near 3.5 eV above the Fermi energy, indicating that the egσe_g^{\sigma} orbitals exhibit the typical strongly-correlated Mott insulating behavior. We find that this Mott state is robust against the strain effect.

Unlike the half-filled egσe_g^{\sigma} orbitals, the a1ga_{1g} and egπe_g^{\pi} orbitals are partially filled with one shared hole in the t2gt_{2g} manifold (ht2g=1.0h_{t_{2g}} = 1.0), implying that the correlation effect can arise due to the multi-orbital physics in a non-trivial way. The partially filled occupancies among the a1ga_{1g} and egπe_g^{\pi} orbitals are due to the large Hund’s coupling JJ acting on the t2gt_{2g} manifold with a small ΔTCF\Delta_{TCF}. This shared hole is mostly located on the egπe_g^{\pi} orbitals where the broad upper Hubbard band is formed near 2 eV above the Fermi energy. The ImΣ(ω)\mathit{Im}\Sigma(\omega) pole of the egπe_g^{\pi} orbital is located near 1.5 eV above the Fermi energy and the pole strength is weaker than the egσe_g^{\sigma} orbital one (Figure 3c). This means that the Mott insulating state of the egπe_g^{\pi} orbital shows moderate correlations, which could be tunable by strains. The self-energies for eg+πe_{g+}^{\pi} and eg−πe_{g-}^{\pi} orbitals are similar despite the small JT splitting between two orbitals.

The ImΣ(ω)\mathit{Im}\Sigma(\omega) for the a1ga_{1g} orbital shows much broader peaks with an almost filled a1ga_{1g} occupancy (Na1g=1.8N_{a_{1g}} = 1.8). The peak position is located near 2.0 eV above the Fermi energy, which corresponds to the Sb ss orbital energy level due to their hybridization (Figure 3c). This shows that the a1ga_{1g} orbital is close to a hybridization-induced insulating state rather than a Mott insulating state. Our result also suggests that the one-particle energy level of the a1ga_{1g} orbital can be effectively shifted lower due to the nearly full occupation, resulting in the negative trigonal crystal field splitting (ΔTCF<0\Delta_{TCF} < 0), while the a1ga_{1g} orbital energy is higher than the egπe_g^{\pi} orbital ones at the DFT level (ΔTCF>0\Delta_{TCF} > 0). We want to emphasize that the orbital-dependent correlation effect plays an important role in tuning the effective Co 3dd orbital levels and the resulting ΔTCF\Delta_{TCF} splitting under strains. This is also in sharp contrast to the CoI2_2 case where the DMFT occupancy of the a1ga_{1g} orbital is smaller than the egπe_g^{\pi} one (ΔTCF>0\Delta_{TCF} > 0) due to the absence of the Sb ss orbital.

Now, we study the strain effect on the change of correlations in NCSO and CoI2_2. Figure 4a shows that the DMFT occupancy in NCSO are rather insensitive to compressive strains as the a1ga_{1g} occupancy is close to the full occupation (Na1g∼1.8N_{a_{1g}} \sim1.8). Under tensile strain, Na1gN_{a_{1g}} gradually decreases, and the a1ga_{1g} and egπe_g^{\pi} occupancies are almost degenerate at Δa/a0∼1.2%\Delta a/a_0 \sim1.2\%. We expect that the sign change of ΔTCF\Delta_{TCF} could occur near the tensile strain Δa/a0≃1.2%\Delta a/a_0 \simeq1.2\% and possibly realize the spin liquid phase due to the strongly suppressed ΔTCF\Delta_{TCF}. The change of the DMFT occupancy under tensile strain is closely related to the orbital-dependent correlation effect since the occupancy increases (decreases) gradually as the ∣Im⁡Σ(ω)∣max|\operatorname{Im}\Sigma(\omega)|_{max} decreases (increases) (Figure 4b). For the CoI2_2 case, the trend of the occupancy change is more rapid (Figure 4c) but still consistent with the change of ΔTCF\Delta_{TCF} as the a1ga_{1g} and egπe_g^\pi occupancies are almost degenerate at Δa/a0∼−0.5%\Delta a/a_0 \sim-0.5\% where ΔTCF\Delta_{TCF} becomes nearly zero. Overall, the tensile (compressive) strain favors the stronger correlation effect of a1ga_{1g} (egπe_g^\pi) orbital as it is less hybridized with O pp orbitals. The difference between two compounds originates from the low Co-site symmetry and the presence of Sb ss orbitals in NCSO.

Image not extracted

Figure 4. DFT+DMFT results of (a) the occupancy and (b) the pole strength ∣Im⁡Σ(ω)∣max|\operatorname{Im}\Sigma(\omega)|_{max} of the a1ga_{1g}, eg+πe_{g+}^{\pi}, and eg−πe_{g-}^{\pi} orbitals in NCSO as a function of strain. (c) The occupancy and (d) the pole strength ∣Im⁡Σ(ω)∣max|\operatorname{Im}\Sigma(\omega)|_{max} of the a1ga_{1g}, eg+πe_{g+}^{\pi}, and eg−πe_{g-}^{\pi} orbitals in CoI2_2 for comparison.

In summary, we studied the strain and structural distortion effects on electronic and magnetic properties of NCSO using first-principles calculations. We found that the orbital-dependent correlation effect of Co ions will play an important role in tuning the material parameters under strains. While the a1ga_{1g} orbitals are insulating due to the nearly full occupation, other Co dd orbitals are more strongly correlated as a Mott insulating state. The electronic structure dependence on strains is much weaker for NCSO compared to the CoI2_2 case due to its low Co-site symmetry originated from the O distortion and the presence of the Sb ss orbitals. The study of spin-exchange interactions incorporating such realistic and correlated electronic structures[36] will be important for the future study of this and related Kitaev spin-liquid candidate materials.

Acknowledgement

This work was supported by the Materials Sciences and Engineering Division, Basic Energy Sciences, Office of Science, US Department of Energy. We gratefully acknowledge the computing resources provided on Bebop and Improv, high-performance computing clusters operated by the Laboratory Computing Resource Center at Argonne National Laboratory.

References

Magnetic Ground State of Na2_2Co3_3SbO6_6

Here, we briefly discuss about magnetic ground states of Na2_2Co3_3SbO6_6 using DFT+UU while varying UU and fixing J=0.8J=0.8 eV. The energy differences between magnetic structures, including FM, zigzag AFM, stripe AFM, and the Néel AFM show that the Néel AFM structure is more stable than the other magnetic structures in a strongly correlated regime (U>2U > 2 eV), while the FM is the most stable structure in a weakly correlated regime (Figure 5).

Image not extracted

Figure 5. Relative energy per formula unit of FM, zigzag AFM, stripe AFM structure to Néel AFM structure of NCSO as a function of the Hubbard UU while fixing J=0.8J=0.8 eV within DFT+UU calculation.

Wannier Hamiltonian

Here, we construct the tight-binding Hamiltonian using the maximally localized Wannier functions (MLWFs)[37]. The correlated subspace of Co 3d3d orbitals can be solved using the DMFT equations. Trigonal and Jahn-Teller distortions in Na3_3Co2_2SbO6_6 can also be studied using the on-site orbital energies of Co 3d3d orbitals. To construct Wannier functions, we first solve the non-spin-polarized Kohn-Sham equation within DFT using VASP code. Once the Kohn-Sham equation is solved, all eigenfunctions ψnk(r)\psi_{nk}(\mathbf{r}) and the energy bands are obtained: En(k)=⟨ψnk∣H^KS∣ψnk⟩E_n(\mathbf{k}) = \langle\psi_{nk}|\hat{H}^{KS}|\psi_{nk}\rangle, where H^KS\hat{H}^{KS} is the Kohn-Sham Hamiltonian. We then constructed a manifold of JJ bands with maximally localized Wannier functions (MLWFs)[37]

∣Rn⟩=V(2π)3∫dke−ik⋅R∑m=1JUmn(k)∣ψmk⟩(1)|Rn\rangle= \frac{V}{(2\pi)^3} \int d\mathbf{k}e^{-i\mathbf{k}\cdot\mathbf{R}}\sum_{m=1}^{J} U_{mn}^{(k)}|\psi_{mk}\rangle \tag*{(1)}

using VASP and WANNIER90[38] codes. In this work, the local [001][001], [010][010], and [001][001] axes for constructing Co dd orbitals are chosen be aligned close to the local bondings of the perfect octahedron such that the local [111][111] axis will point toward the cc-axis. After the Wannier procedure, the real-space Wannier Hamiltonian Hmn(R′−R)=⟨R′m∣H^KS∣Rn⟩\mathcal{H}_{mn}(\mathbf{R}'-\mathbf{R})=\langle R'm|\hat{H}^{KS}|Rn\rangle is obtained.

The local Wannier Hamiltonians (Hmn(0)\mathcal{H}_{mn}(0)) of the Co 3d3d orbitals of NCSO (within the spdspd-model), BaCo2_2(AsO4_4)2_2 with the space group R3ˉR\bar{3} [No. 148] (within the spdspd-model), and the CoI2_2 with the space group P3ˉm1P\bar{3}m1 [No. 164] (within the pdpd-model) respectively at zero strain are represented using the cubic harmonic basis ({∣z2⟩,∣x2−y2⟩,∣xz⟩,∣yz⟩,∣xy⟩}\{|z^2\rangle, |x^2-y^2\rangle, |xz\rangle, |yz\rangle, |xy\rangle\}):

N0spd=[4.7560.000−0.005−0.0050.0090.0004.756−0.0110.0110.000−0.005−0.0114.4370.0040.004−0.0050.0110.0044.4370.0040.0090.0000.0040.0044.440](2)\mathcal{N}^{spd}_0 = \begin{bmatrix} 4.756 & 0.000 & -0.005 & -0.005 & 0.009 \\ 0.000 & 4.756 & -0.011 & 0.011 & 0.000 \\ -0.005 & -0.011 & 4.437 & 0.004 & 0.004 \\ -0.005 & 0.011 & 0.004 & 4.437 & 0.004 \\ 0.009 & 0.000 & 0.004 & 0.004 & 4.440 \end{bmatrix} \tag*{(2)}
B0spd=[2.8600.000−0.0040.007−0.0020.0002.8600.0050.002−0.006−0.0040.0052.5980.0140.0140.0070.0020.0142.5970.014−0.002−0.0060.0140.0142.598](3)\mathcal{B}^{spd}_0 = \begin{bmatrix} 2.860 & 0.000 & -0.004 & 0.007 & -0.002 \\ 0.000 & 2.860 & 0.005 & 0.002 & -0.006 \\ -0.004 & 0.005 & 2.598 & 0.014 & 0.014 \\ 0.007 & 0.002 & 0.014 & 2.597 & 0.014 \\ -0.002 & -0.006 & 0.014 & 0.014 & 2.598 \end{bmatrix} \tag*{(3)}
C0pd=[4.0560.000−0.001−0.0010.0030.0004.056−0.0020.0020.000−0.001−0.0023.8870.0070.007−0.0010.0020.0073.8870.0070.003−0.0000.0070.0073.887](4)\mathcal{C}^{pd}_0 = \begin{bmatrix} 4.056 & 0.000 & -0.001 & -0.001 & 0.003 \\ 0.000 & 4.056 & -0.002 & 0.002 & 0.000 \\ -0.001 & -0.002 & 3.887 & 0.007 & 0.007 \\ -0.001 & 0.002 & 0.007 & 3.887 & 0.007 \\ 0.003 & -0.000 & 0.007 & 0.007 & 3.887 \end{bmatrix} \tag*{(4)}

By using a unitary matrix

Utrig=[10000010000013−16120013−16−120013260](5)U_{trig} = \begin{bmatrix} 1 & 0 & 0 & 0 & 0 \\ 0 & 1 & 0 & 0 & 0 \\ 0 & 0 & \frac{1}{\sqrt{3}} & -\frac{1}{\sqrt{6}} & \frac{1}{\sqrt{2}} \\ 0 & 0 & \frac{1}{\sqrt{3}} & -\frac{1}{\sqrt{6}} & -\frac{1}{\sqrt{2}} \\ 0 & 0 & \frac{1}{\sqrt{3}} & \frac{2}{\sqrt{6}} & 0 \end{bmatrix} \tag*{(5)}

that rotates the cubic harmonic basis to the trigonal basis {∣egσ⟩,∣egσ⟩,∣a1g⟩,∣egπ+⟩,∣egπ−⟩}\{|e_g^\sigma\rangle, |e_g^\sigma\rangle, |a_{1g}\rangle, |e_g^\pi{}_+\rangle, |e_g^\pi{}_-\rangle\}, the matrices (B2), (B3), and (B4) become

[N0spd]tri=[4.7560.0000.0010.011−0.0070.0004.756−0.002−0.01−0.0160.001−0.0024.4460.0020.0000.011−0.0100.0024.4340.000−0.007−0.0160.000.0004.433](6)[\mathcal{N}^{spd}_0]_{tri} = \begin{bmatrix} 4.756 & 0.000 & 0.001 & 0.011 & -0.007 \\ 0.000 & 4.756 & -0.002 & -0.01 & -0.016 \\ 0.001 & -0.002 & 4.446 & 0.002 & 0.000 \\ 0.011 & -0.010 & 0.002 & 4.434 & 0.000 \\ -0.007 & -0.016 & 0.00 & 0.000 & 4.433 \end{bmatrix} \tag*{(6)}
[B0spd]tri=[2.8600.0000.000−0.003−0.0080.0002.8600.000−0.0080.0020.0000.0002.6260.0000.000−0.003−0.0080.0002.5840.000−0.0080.0020.0000.0002.583](7)[\mathcal{B}^{spd}_0]_{tri} = \begin{bmatrix} 2.860 & 0.000 & 0.000 & -0.003 & -0.008 \\ 0.000 & 2.860 & 0.000 & -0.008 & 0.002 \\ 0.000 & 0.000 & 2.626 & 0.000 & 0.000 \\ -0.003 & -0.008 & 0.000 & 2.584 & 0.000 \\ -0.008 & 0.002 & 0.000 & 0.000 & 2.583 \end{bmatrix} \tag*{(7)}
[C0spd]tri=[4.0560.0000.0000.0040.0000.0004.0560.0000.000−0.0030.0000.0003.9010.0000.0000.0040.0000.0003.8800.0000.000−0.0030.0000.0003.880](8)[\mathcal{C}^{spd}_0]_{tri} = \begin{bmatrix} 4.056 & 0.000 & 0.000 & 0.004 & 0.000 \\ 0.000 & 4.056 & 0.000 & 0.000 & -0.003 \\ 0.000 & 0.000 & 3.901 & 0.000 & 0.000 \\ 0.004 & 0.000 & 0.000 & 3.880 & 0.000 \\ 0.000 & -0.003 & 0.000 & 0.000 & 3.880 \end{bmatrix} \tag*{(8)}

in the trigonal basis {∣egσ⟩,∣egσ⟩,∣a1g⟩,∣egπ+⟩,∣egπ−⟩}\{|e_g^\sigma\rangle, |e_g^\sigma\rangle, |a_{1g}\rangle, |e_g^\pi{}_+\rangle, |e_g^\pi{}_-\rangle\}. The {∣a1g⟩,∣egπ+⟩,∣egπ−⟩}\{|a_{1g}\rangle, |e_g^\pi{}_+\rangle, |e_g^\pi{}_-\rangle\} blocks of the matrices (B7) and (B8) have the off-diagonal terms of zeros, implying the trigonal basis is more suitable than the cubic one for BCAO and CoI2_2. This is consistent since both of the space groups R3ˉR\bar{3} of BaCo2_2(AsO4_4)2_2 and P3ˉm1P\bar{3}m1 of CoI2_2 represent for trigonal systems although the R3ˉR\bar{3} has a lower symmetry than the P3ˉm1P\bar{3}m1. On the other hand, the {∣a1g⟩,∣egπ+⟩,∣egπ−⟩}\{|a_{1g}\rangle, |e_g^\pi{}_+\rangle, |e_g^\pi{}_-\rangle\} block of the matrix (B6) has a small non-zero off-diagonal terms, originated from the small JT compression at zero strain due to the monoclinic space group C2/mC2/m of NCSO.

To see the effects of both trigonal and JT distortions within the space group C2/mC2/m, we consider a Hamiltonian

H=[E10.0ϵ1ϵ2ϵ30.0E1+δ1ϵ4ϵ5ϵ6ϵ1ϵ4E2Δtri3+δ3Δtri3ϵ2ϵ5Δtri3+δ3E2Δtri3ϵ3ϵ6Δtri3Δtri3E2+δ2](9)\mathcal{H} = \begin{bmatrix} E_1 & 0.0 & \epsilon_1 & \epsilon_2 & \epsilon_3 \\ 0.0 & E_1+\delta_1 & \epsilon_4 & \epsilon_5 & \epsilon_6 \\ \epsilon_1 & \epsilon_4 & E_2 & \frac{\Delta_{tri}}{3}+\delta_3 & \frac{\Delta_{tri}}{3} \\ \epsilon_2 & \epsilon_5 & \frac{\Delta_{tri}}{3}+\delta_3 & E_2 & \frac{\Delta_{tri}}{3} \\ \epsilon_3 & \epsilon_6 & \frac{\Delta_{tri}}{3} & \frac{\Delta_{tri}}{3} & E_2+\delta_2 \end{bmatrix} \tag*{(9)}

in the cubic basis {∣z2⟩,∣x2−y2⟩,∣xz⟩,∣yz⟩,∣xy⟩}\{|z^2\rangle, |x^2-y^2\rangle, |xz\rangle, |yz\rangle, |xy\rangle\} where E1=Edz2E_1=E_{d_{z^2}} and E2=Edxz=EdyzE_2=E_{d_{xz}}=E_{d_{yz}}. The parameters δ1\delta_1, δ2\delta_2, δ3\delta_3 represent the energy splitting between dz2d_{z^2} and dx2−y2d_{x^2-y^2} orbitals, the one between dxzd_{xz} and dxyd_{xy} orbitals, and the distortion between (Co−O)x^(\mathrm{Co-O})_{\hat{x}} and (Co−O)y^(\mathrm{Co-O})_{\hat{y}} away from 90∘90^\circ respectively, due to the JT effect from the monoclinic-cell type. The parameter Δtri3\frac{\Delta_{tri}}{3} represents the trigonal distortion.

By using the unitary matrix (B5), one can obtain the matrix representation of H\mathcal{H} in the trigonal basis such that

[H]trig=Utrig†[H]Utrig=[E10.0ϵ1′ϵ2′ϵ3′0.0E1+δ1ϵ4′ϵ5′ϵ6′ϵ1′ϵ4′E2+23Δtri+δ23+2δ3323(δ2−δ3)0.0ϵ2′ϵ5′23(δ2−δ3)E2−13Δtri+2δ23+δ330.0ϵ3′ϵ6′0.00.0E2−13Δtri−δ3](10)[\mathcal{H}]_{trig}=U_{trig}^{\dagger}[\mathcal{H}]U_{trig}= \begin{bmatrix} E_1 & 0.0 & \epsilon'_1 & \epsilon'_2 & \epsilon'_3\\ 0.0 & E_1+\delta_1 & \epsilon'_4 & \epsilon'_5 & \epsilon'_6\\ \epsilon'_1 & \epsilon'_4 & E_2+\frac{2}{3}\Delta_{tri}+\frac{\delta_2}{3}+\frac{2\delta_3}{3} & \frac{\sqrt{2}}{3}(\delta_2-\delta_3) & 0.0\\ \epsilon'_2 & \epsilon'_5 & \frac{\sqrt{2}}{3}(\delta_2-\delta_3) & E_2-\frac{1}{3}\Delta_{tri}+\frac{2\delta_2}{3}+\frac{\delta_3}{3} & 0.0\\ \epsilon'_3 & \epsilon'_6 & 0.0 & 0.0 & E_2-\frac{1}{3}\Delta_{tri}-\delta_3 \end{bmatrix} \tag*{(10)}

Then the trigonal crystal field ΔTCF\Delta_{TCF}, defined as the energy splitting between the singlet a1ga_{1g} and the doublet eg′e'_g in the trigonal basis, is given by

ΔTCF=Ea1g−Eeg′++Eeg′−2=Δtri+δ3(11)\Delta_{TCF}=E_{a_{1g}}-\frac{E_{e'_g+}+E_{e'_g-}}{2}=\Delta_{tri}+\delta_3 \tag*{(11)}

Basically, the sum of all the upper (lower) off-diagonal terms in the t2gt_{2g} block from the cubic Hamiltonian H\mathcal{H} (Eq. (9)) represents the trigonal crystal field ΔTCF\Delta_{TCF} in the trigonal basis. Also, the effect of JT distortion is given by the energy splitting between eg+′e'_{g+} and eg−′e'_{g-} such that

δJT=Eeg+′−Eeg−′=23(δ2+2δ3)(12)\delta_{JT}=E_{e'_{g+}}-E_{e'_{g-}}=\frac{2}{3}(\delta_2+2\delta_3) \tag*{(12)}

From Wannier calculations, we also obtained the d−dd-d hopping matrix between the nearest Co neighbors (ZZ-bond) of the zero strain structure within the dd-model

Td−dd=[−0.0550.000−0.010−0.0100.1170.000−0.0060.012−0.0120.000−0.0100.0120.039−0.0260.032−0.010−0.012−0.0260.0390.0320.1170.00.03220.032−0.162](13)\mathcal{T}^{d}_{d-d}= \begin{bmatrix} -0.055 & 0.000 & -0.010 & -0.010 & 0.117\\ 0.000 & -0.006 & 0.012 & -0.012 & 0.000\\ -0.010 & 0.012 & 0.039 & -0.026 & 0.032\\ -0.010 & -0.012 & -0.026 & 0.039 & 0.032\\ 0.117 & 0.0 & 0.0322 & 0.032 & -0.162 \end{bmatrix} \tag*{(13)}

By comparing the matrix (B17) to the hopping matrix of the Co nearest-neighbors (ZZ-bond) for an ideal edge-sharing bond with 90∘90^\circ metal–ligand–metal bond angles

Tijdd=[t4000t60t500000t1t2000t2t10t6000t3](14)\mathcal{T}^{dd}_{ij}= \begin{bmatrix} t_4 & 0 & 0 & 0 & t_6\\ 0 & t_5 & 0 & 0 & 0\\ 0 & 0 & t_1 & t_2 & 0\\ 0 & 0 & t_2 & t_1 & 0\\ t_6 & 0 & 0 & 0 & t_3 \end{bmatrix} \tag*{(14)}

in the cubic basis of {∣z2⟩,∣x2−y2⟩,∣xz⟩,∣yz⟩},∣xy⟩\{|z^2\rangle,|x^2-y^2\rangle,|xz\rangle,|yz\rangle\},|xy\rangle[10, 34], one can find the hopping integrals for each exchange channel t2gt_{2g}-t2gt_{2g} (t1t_1, t2t_2, t3t_3), ege_g-ege_g (t4t_4, t5t_5), and t2gt_{2g}-ege_g (t6t_6) at zero strain as shown in Table II. These hopping integrals are comparable with previous ab initio calculations[34, 39].

The onsite Hamiltonian for O 2p2p orbitals (within the spdspd-model) is

H0O−2p=[∣pz⟩∣px⟩∣py⟩⟨pz∣0.3110.1170.117⟨px∣0.1171.1390.006⟨py∣0.1170.0061.139](15)\mathcal{H}^{O-2p}_0= \begin{bmatrix} & |p_z\rangle& |p_x\rangle& |p_y\rangle\\ \langle p_z| & 0.311 & 0.117 & 0.117\\ \langle p_x| & 0.117 & 1.139 & 0.006\\ \langle p_y| & 0.117 & 0.006 & 1.139 \end{bmatrix} \tag*{(15)}

By using the matrices (B2) and (B15), one can find the charge transfer gap Δpd=E‾t2g−E‾2p=3.58 eV\Delta_{pd}=\overline{E}_{t_{2g}}-\overline{E}_{2p}=3.58\ \mathrm{eV} at zero strain

Finally, the numerical hopping between p−dp-d orbitals within the spdspd-model is given by

modelt1t_1 [eV]t2t_2 [eV]t3t_3 [eV]t4t_4 [eV]t5t_5 [eV]t6t_6 [eV]tpdσt_{pd}^{\sigma} [eV]tpdπt_{pd}^{\pi} [eV]Δpd\Delta_{pd} [eV]
dd0.039−0.026-0.026−0.162-0.162−0.055-0.055−0.006-0.0060.117N/AN/AN/A
Winter[34]∼0.036\sim0.036∼−0.03\sim-0.03∼−0.142\sim-0.142∼−0.036\sim-0.036∼−0.036\sim-0.036∼0.130\sim0.130∼1\sim1∼0.5\sim0.54.5
Kim et al.[39]N/A0.007−0.135-0.135N/AN/A0.1251.179−0.589-0.5894.21
spdspd0.054−0.040-0.040−0.176-0.176−0.019-0.019−0.055-0.055−0.035-0.0351.061−0.534-0.5343.58

Table 2. Hopping integrals and charge transfer gap Δpd\Delta_{pd} of NCSO at zero strain.

Tp−dspd=(⟨z2∣0.095−0.5370.010⟨x2−y2∣−0.1670.908−0.030⟨xz∣−0.516−0.1940.010⟨yz∣0.0010.0360.053⟨xy∣−0.009−0.021−0.559)(16)\mathcal{T}_{p-d}^{spd} = \begin{pmatrix} \langle z^2\rvert& 0.095 & -0.537 & 0.010 \\ \langle x^2-y^2\rvert& -0.167 & 0.908 & -0.030 \\ \langle xz\rvert& -0.516 & -0.194 & 0.010 \\ \langle yz\rvert& 0.001 & 0.036 & 0.053 \\ \langle xy\rvert& -0.009 & -0.021 & -0.559 \end{pmatrix} \tag*{(16)}

By comparing to the pp-dd hopping matrix for an ideal octahedron[10]

Tp−d0=(⟨z2∣0.0−12tpdσ0.0⟨x2−y2∣0.032tpdσ0.0⟨xz∣tpdπ0.00.0⟨yz∣0.00.00.0⟨xy∣0.00.0tpdπ),(17)\mathcal{T}_{p-d}^{0} = \begin{pmatrix} \langle z^2\rvert& 0.0 & -\frac{1}{2}t_{pd}^{\sigma} & 0.0 \\ \langle x^2-y^2\rvert& 0.0 & \frac{\sqrt{3}}{2}t_{pd}^{\sigma} & 0.0 \\ \langle xz\rvert& t_{pd}^{\pi} & 0.0 & 0.0 \\ \langle yz\rvert& 0.0 & 0.0 & 0.0 \\ \langle xy\rvert& 0.0 & 0.0 & t_{pd}^{\pi} \end{pmatrix}, \tag*{(17)}

one can obtain tpdσ=12(2∗0.537+23∗0.908)=1.061t_{pd}^{\sigma}=\frac{1}{2}(2 * 0.537+\frac{2}{\sqrt{3}} * 0.908)=1.061 eV and tpdπ=12(−0.516−−0.559)=−0.534t_{pd}^{\pi}=\frac{1}{2}(-0.516--0.559)=-0.534 eV. Figure 6 shows that only the direct d−dd-d hopping integral t3t_3 favoring Heisenberg interaction is sensitive to the strain effect.

Image not extracted

Figure 6. Strain effect on (a) the hopping integral between dd orbitals and (b) the p−dp-d hopping integral and the charge transfer gap Δpd\Delta_{pd}.

Energy Band of the Ambient NCSO Structure

Here, we compare the non-spin-polarized DFT and Wannier energy bands obtained from the fully relaxed NCSO structure. Figure 7(a,Figure 7b) show that the DFT and Wannier bands fit well for both dd- and spdspd-models, indicating that the eigenvalues of the Kohn-Sham equation are exactly the same with those obtained from the d/spdd/spd-model Wannier functions. Using the obtained Wannier functions, the Co 3d3d Hamiltonians can be extracted and analyzed.

Image not extracted

Figure 7. (a,b) Energy bands of NCSO calculated with DFT and Wannier functions. (c) Molecular-like a1ga_{1g} orbital in the trigonal basis under the effects of ligand fields within dd-model Wannier functions. (d) Atomic-like a1ga_{1g} orbital in trigonal basis within spdspd-model Wannier functions

DMFT method

DMFT maps the lattice many-body problem with the on-site interaction UU into a single-site problem with the same interaction UU and coupled to self-consistently determined bath Δ(iω)\Delta(i\omega)[40]. In the limit of the large lattice coordination, the self-energy Σ(k,iω)≈Σ(iω)\Sigma(\mathbf{k},i\omega)\approx\Sigma(i\omega), then the local self-energy Σ(iω)\Sigma(i\omega) can be treated as the single-site impurity problem coupled to electron-bath as in the Anderson impurity model (AIM)[41, 42]. The hybridization function Δ(iω)\Delta(i\omega) describing the ability of an electron to hop in and hop out of an atom is due to the spatial overlap between the correlated orbitals and the conduction band states.

The DFT+DMFT calculation procedure can be described as follows. First, we diagonalize the spdspd-model Wannier Hamiltonian, such as the matrix (B2), in order to obtain the Hamiltonian HDH_D where the Co dd submatrix can be diagonal with a corresponding unitary matrix UdiagU_{diag} of column eigenvectors. Since the matrix representations of the local Co 3d3d Wannier Hamiltonians in cubic and trigonal bases contain non-zero off-diagonal terms which can lead to the fermion-sign problem in Monte Carlo simulation[43, 44], these diagonalized Hamiltonians without off-diagonal terms can be beneficial. All of the final results, including density of states (DOS) (or spectral function), occupancy matrix, and the imaginary part of self-energy Σ(ω)\Sigma(\omega) will be rotated from the diagonal basis to the trigonal basis with the unitary matrix U=Udiag→trig=Udiag†⋅UtrigU = U_{\mathrm{diag}\to\mathrm{trig}} = U_{\mathrm{diag}}^{\dagger} \cdot U_{\mathrm{trig}}.

Next, the Hamiltonian HDH^{D} is then used to construct the local lattice Green’s function Gloc\mathbf{G}_{\mathrm{loc}} within the DMFT self-consistent loop. Within the DMFT self-consistent loop, starting from the local impurity self-energy Σ(iω)\Sigma(i\omega), one can obtain the matrix of Gloc\mathbf{G}_{\mathrm{loc}}:

Gloc(iω)=1VBZ∫BZdk[(iω+μ)1−HD−(Σ(iω)−VDC)]−1(18)\mathbf{G}_{\mathrm{loc}}(i\omega) = \frac{1}{V_{\mathrm{BZ}}} \int_{\mathrm{BZ}} d\mathbf{k}\left[(i\omega+\mu)\mathbf{1}-H^{D}-\left(\Sigma(i\omega)-V_{\mathrm{DC}}\right)\right]^{-1} \tag*{(18)}

where μ\mu is the chemical potential, 1\mathbf{1} is the identity matrix, VDCV_{\mathrm{DC}} is the double-counting potential, and VBZV_{\mathrm{BZ}} is the volume of the Brillouin zone. We use the fully-localized-limit double-counting potential VDCV_{\mathrm{DC}} given by

VDC=U(Nd−12)−J2(Nd−1)(19)V_{\mathrm{DC}} = U\left(N_{d}-\frac{1}{2}\right)-\frac{J}{2}\left(N_{d}-1\right) \tag*{(19)}

where NdN_{d} is the dd occupancy.

Once the lattice Green’s function Gloc(iω)\mathbf{G}_{\mathrm{loc}}(i\omega) is computed from the Eq. (18), then the effective non-interacting Green’s function G0\mathcal{G}^{0} of the Anderson model can be found by

[G0(iω)]−1=[Gloc(iω)]−1+Σ(iω).(20)\left[\mathcal{G}_{0}(i\omega)\right]^{-1}=\left[\mathbf{G}_{\mathrm{loc}}(i\omega)\right]^{-1}+\Sigma(i\omega). \tag*{(20)}

This non-interacting Green’s function G0\mathcal{G}_{0} is then used to solve the Anderson impurity model and calculate the interacting impurity Green’s function Gimp(iω)G_{\mathrm{imp}}(i\omega) using continuous time quantum Monte Carlo (CTQMC) impurity solver[29, 45, 46, 47]. Then, the new self-energy is given by

Σnew(iω)=[G0(iω)]−1−[Gimp(iω)]−1.(21)\Sigma_{\mathrm{new}}(i\omega)=\left[\mathcal{G}_{0}(i\omega)\right]^{-1}-\left[G_{\mathrm{imp}}(i\omega)\right]^{-1}. \tag*{(21)}

Then, we substitute the mixed new self-energy Σnew\Sigma_{\mathrm{new}} into the (Eq. (18)) so that the algorithm iterates until the convergence criterion is satisfied, e.g.

Gloc(iω)=Gimp(iω).(22)\mathbf{G}_{\mathrm{loc}}(i\omega)=G_{\mathrm{imp}}(i\omega). \tag*{(22)}

One then can obtain the analytically continued Green’s function GAG_{A} using the maximum entropy method[48]. Then, the spectral function A(ω)A(\omega) in the trigonal basis, which can be compared to photoemission in experiment, is given by

A(ω)=−1πIm⁡[U†GA(ω)U](23)A(\omega)=-\frac{1}{\pi}\operatorname{Im}\left[U^{\dagger}G_{A}(\omega)U\right] \tag*{(23)}

Also, the occupancy matrix in the trigonal basis is given by

N=∫ωA(ω)fE(ω)dω(24)N=\int_{\omega} A(\omega)f_{E}(\omega)d\omega \tag*{(24)}

where fE(ω)f_{E}(\omega) is the Fermi distribution function.

Effect of the different choice of Wannier orbitals

To study the effect of different choices of Wannier orbitals on ΔTCF\Delta_{TCF} and δJT\delta_{JT}, we plot the ΔTCF\Delta_{TCF} and the energy splitting δJT\delta_{JT} of NCSO obtained using different choices of Co dd, Sb ss, and O pp orbitals as a function of strain (Figure 8(a,Figure 8b)). We obtained dd-, sdsd-, pdpd-, and spdspd-model Wannier Hamiltonians with energy windows of [−2.0,0.8][-2.0, 0.8] eV, [−2.0,2.5][-2.0, 2.5] eV, [−8.0,2.5][-8.0, 2.5] eV, and [−10.5,2.5][-10.5, 2.5] eV respectively. Within these models, the slopes of the ΔTCF\Delta_{TCF} as a function of strain are the same to each other but shifted between the models (see Figure 8a). This implies the hybridization of Wannier orbitals and their environments can play a key role in determining the magnitude of ΔTCF\Delta_{TCF}. At zero strain, the ΔTCF\Delta_{TCF} within the dd-model is negative with ΔTCF=−25.8\Delta_{TCF}=-25.8 meV, consistent with the one measured by Kim et al. using XAS[12]. Also, since the geometric lobes of the egπe_{g}^{\pi} orbitals (egπ+,egπ−e_{g}^{\pi+},e_{g}^{\pi-}) are similar to each other, the energy splitting δJT\delta_{JT} between them does not change much across all models.

Image not extracted

Figure 8. (a,b) ΔTCF\Delta_{TCF} and δJT\delta_{JT} of NCSO as a function of strain within dd-, sdsd-, pdpd-, and spdspd-model Wannier Hamiltonians. (c) Crystal field splitting 10Dq10Dq between egσe_{g}^{\sigma} and t2gt_{2g} orbitals of NCSO within dd and spdspd models.

The choice of Wannier orbitals also has a similar effect on the crystal field splitting 10Dq10Dq between egσe_{g}^{\sigma} and t2gt_{2g} orbitals (Figure 8(c)). At zero strain the spdspd-model 10Dq10Dq is only about 0.32 eV, while the dd-model 10Dq10Dq is about 0.96 eV, consistent with the common crystal field of Co2+^{2+} in literature[13, 16, 49]. Within the dd-model, the egσe_{g}^{\sigma} orbitals whose lobes directly point to the surrounding O are more strongly repulsed by the ligand fields of these O2−^{2-} than the t2gt_{2g} orbitals whose lobes points between these O. Thus, the energy levels of the egσe_{g}^{\sigma} orbitals are much higher than those of the t2gt_{2g} orbitals. Within the spdspd-model, the egσe_g^{\sigma} orbitals are much localized and 10Dq10Dq is decreased.

Comparison of Trigonal Distortions between Na3_3Co2_2SbO6_6, BaCo2_2(AsO4_4)2_2, and CoI2_2

In the main text, we compared the trigonal distortions and the ΔTCF\Delta_{TCF} of NCSO and CoI2_2. Here, we also show the trigonal distortion and the ΔTCF\Delta_{TCF} of BaCo2_2(AsO4_4)2_2 (BCAO). Figure 9(a) shows the slopes of trigonal distortions (L2/L1L_2/L_1 and L3/L1L_3/L_1) as a function of strain in BCAO is similar to the ones of NCSO and CoI2_2. Figure 9(b) shows that the ΔTCF\Delta_{TCF} of BCAO, whose symmetry is higher (lower) than NCSO (CoI2_2), is more (less) sensitive to the strain than the one of NCSO (CoI2_2). These indicates the symmetry (space group) of a compound determines how sensitive the ΔTCF\Delta_{TCF} of the compound is under the effect of strain. Importantly, the positive slopes of the ΔTCF\Delta_{TCF} as a function of strain in NCSO, CoI2_2, and BCAO yield a unique and crucial effect of strain, in which the energy level of the a1ga_{1g} orbital increases (decreases) relative to the ones of the egπe_g^{\pi} orbitals as these compounds are more tensile (compressive).

Image not extracted

Figure 9. (a) Trigonal distortions and (b) trigonal crystal field ΔTCF\Delta_{TCF} versus strain of NCSO (spdspd-model), BCAO (spdspd-model), and CoI2_2 (pdpd-model)

DMFT Calculations for NCSO at a/a0=pm1%

Here, we show DMFT calculations for NCSO at Δa/a0=±1%\Delta a/a_0 = \pm1\%. Under the compressive strain Δa/a0=−1%\Delta a/a_0 = -1\%, the a1ga_{1g} orbital occupancy is almost fully filled (N∼1.8N \sim1.8) consistently with the DMFT DOS (Figure 10a). On the other hand, a noticeable unoccupied states of the a1ga_{1g} orbital is located at E−EF=2E-E_F = 2 eV for the tensile strain Δa/a0=1%\Delta a/a_0 = 1\% (Figure 10b), accompanied by a larger peaks of the imaginary self-energy of this orbital at ω∼2\omega\sim2 eV (Figure 10d), indicating this orbital becomes more correlated in the tensile region. However, the electron correlation of the a1ga_{1g} orbital is still slightly smaller than the ones of the egπe_g^{\pi} orbitals since the peak of the a1ga_{1g} imaginary self-energy is smaller than the ones of the egπe_g^{\pi} imaginary self-energy at ω∼1.5−2\omega\sim1.5-2 eV. Also, a slightly smaller (larger) peak of the imaginary self-energy of the eg−πe_{g-}^{\pi} orbital than the one of the eg+πe_{g+}^{\pi} orbital indicates a JT compression (elongation), consistent with the JT ratio (Co−O)z^/(Co−O)x^/y^<1(>1)(\mathrm{Co-O})_{\hat{z}}/(\mathrm{Co-O})_{\hat{x}/\hat{y}} < 1(> 1) at Δa/a0=1%(−1%)\Delta a/a_0 = 1\%(-1\%).

Image not extracted

Figure 10. DFT+DMFT DOS of NCSO with U=5U = 5 eV and J=0.8J = 0.8 eV for (a) compressive strain Δa/a0=−1%\Delta a/a_0 = -1\% and (b) tensile strain Δa/a0=1%\Delta a/a_0 = 1\%. The corresponding DMFT self-energy Im⁡Σ(ω)\operatorname{Im}\Sigma(\omega) of a1ga_{1g}, eg+πe_{g+}^{\pi}, and eg−πe_{g-}^{\pi} orbitals for (c) compressive (Δa/a0=−1%\Delta a/a_0 = -1\%) and (d) tensile (Δa/a0=1%\Delta a/a_0 = 1\%) strains.

DMFT Calculation for CoI2

Here, we show some DMFT results for CoI2 at Δa/a0=0,±0.5%\Delta a/a_0 = 0, \pm0.5\%. Figure 11(a,b,c) show that the unoccupied states of a1ga_{1g} orbital increases as the strain effect becomes more tensile, indicating the a1ga_{1g} orbital becomes more correlated as the occupancy gets closer to the half filling. For the compressive strain Δa/a0=−0.5%\Delta a/a_0 = -0.5\%, the imaginary self-energy of the egπe_g^{\pi} orbitals has a stronger peak than the one of the a1ga_{1g} orbital at ω∼1\omega\sim1 eV, implying the egπe_g^{\pi} orbitals are more correlated than the a1ga_{1g} one.

Image not extracted

Figure 11. (a,b,c) DFT+DMFT DOS of CoI2 with U=5U = 5 eV and J=0.8J = 0.8 eV for Δa/a0=0,±0.5%\Delta a/a_0 = 0, \pm0.5\%. (d,e) The corresponding self-energy Im⁡Σ(ω)\operatorname{Im}\Sigma(\omega) for a1ga_{1g}, egπ+e_g^{\pi+}, and egπ−e_g^{\pi-} orbitals of CoI2 for Δa/a0=0\Delta a/a_0 = 0 and −0.5%-0.5\%

Spin Exchange Interactions

Here, we show the spin exchange interactions obtained via the fourth-order strong coupling perturbation theory developed by Liu and Kee analysis[10]. The pseudospin-1/21/2 interactions of the Kitaev Hamiltonian on ZZ-type nearest-neighbor (NN) bond have a general form[5, 50]

Hij(z)=JKS~izS~jz+JHS~i⋅S~j+Γ(S~ixS~jy+S~iyS~jx)+Γ′(S~ixS~jz+S~izS~jx+S~iyS~jz+S~izS~jy)(25)\mathcal{H}_{ij}^{(z)} = J_K \tilde{S}_i^z \tilde{S}_j^z + J_H \tilde{\mathbf{S}}_i \cdot\tilde{\mathbf{S}}_j + \Gamma(\tilde{S}_i^x \tilde{S}_j^y + \tilde{S}_i^y \tilde{S}_j^x) + \Gamma'(\tilde{S}_i^x \tilde{S}_j^z + \tilde{S}_i^z \tilde{S}_j^x + \tilde{S}_i^y \tilde{S}_j^z + \tilde{S}_i^z \tilde{S}_j^y) \tag*{(25)}

where JKJ_K and JHJ_H are Kitaev and Heisenberg interactions respectively. The off-diagonal term Γ\Gamma arises from combined hoppings t1t3t_1t_3 and t2t3t_2t_3 where t1t_1, t2t_2, and t3t_3 are dd–dd hoppings via the t2gt_{2g}–t2gt_{2g} exchange channel[10, 34]. The Γ′\Gamma' arises due to the trigonal crystal field that modifies the pseudospin S~=1/2\tilde{S} = 1/2 wave functions (Kramer doublet).

Although a finite ΔTCF\Delta_{TCF} under the strain effect can affect JKJ_K, JHJ_H, Γ\Gamma, and Γ′\Gamma' in general[5], for simplicity we consider the ideal octahedron limit (no trigonal and JT distortions). The total Heisenberg JHJ_H and Kitaev JKJ_K interactions can be investigated via each anisotropic hopping channel:

JH=Jt2g−t2g+Jt2g−eg+Jeg−eg,(26)J_H = J_{t_{2g}-t_{2g}} + J_{t_{2g}-e_g} + J_{e_g-e_g}, \tag*{(26)}
JK=Kt2g−t2g+Kt2g−eg.(27)J_K = K_{t_{2g}-t_{2g}} + K_{t_{2g}-e_g}. \tag*{(27)}

Within the intersite UU, two-hole 2h2h, and cyclic processes, each exchange interaction from the Eqs. (26) and (I3) is given by

Jt2g−t2g=Jt2g−t2gU+Jt2g−t2g2h+Jt2g−t2gcyclic(28)J_{t_{2g}-t_{2g}} = J_{t_{2g}-t_{2g}}^U + J_{t_{2g}-t_{2g}}^{2h} + J_{t_{2g}-t_{2g}}^{\mathrm{cyclic}} \tag*{(28)}
Jt2g−eg=Jt2g−egU+Jt2g−eg2h+Jt2g−egcyclic(29)J_{t_{2g}-e_g} = J_{t_{2g}-e_g}^U + J_{t_{2g}-e_g}^{2h} + J_{t_{2g}-e_g}^{\mathrm{cyclic}} \tag*{(29)}
Jeg−eg=Jeg−egU+Jeg−eg2h(30)J_{e_g-e_g} = J_{e_g-e_g}^U + J_{e_g-e_g}^{2h} \tag*{(30)}
Kt2g−t2g=Kt2g−t2gU+Kt2g−t2g2h+Kt2g−t2gcyclic(31)K_{t_{2g}-t_{2g}} = K_{t_{2g}-t_{2g}}^U + K_{t_{2g}-t_{2g}}^{2h} + K_{t_{2g}-t_{2g}}^{\mathrm{cyclic}} \tag*{(31)}
Kt2g−eg=Kt2g−egU+Kt2g−eg2h+Kt2g−egcyclic(32)K_{t_{2g}-e_g} = K_{t_{2g}-e_g}^U + K_{t_{2g}-e_g}^{2h} + K_{t_{2g}-e_g}^{\mathrm{cyclic}} \tag*{(32)}

Following the analysis of Liu and Kee[10], the analytic expressions of these terms are given by

Jt2g−t2gU=1486(−171U−3J+259U+J+44U+4J)t12+154(−21U−3J+29U+J+4U+4J)t22+2243(−27U−3J+43U+J+8U+4J)t32+2243(18U−3J−8U+J+5U+4J)t1t3(33)J_{t_{2g}-t_{2g}}^{U}=\frac{1}{486}\left(-\frac{171}{U-3J}+\frac{259}{U+J}+\frac{44}{U+4J}\right)t_1^2+\frac{1}{54}\left(-\frac{21}{U-3J}+\frac{29}{U+J}+\frac{4}{U+4J}\right)t_2^2 +\frac{2}{243}\left(-\frac{27}{U-3J}+\frac{43}{U+J}+\frac{8}{U+4J}\right)t_3^2+\frac{2}{243}\left(\frac{18}{U-3J}-\frac{8}{U+J}+\frac{5}{U+4J}\right)t_1t_3 \tag*{(33)}
Jt2g−t2g2h=[−808112Δpd+Up−3Jp+30424312Δpd+Up−Jp+3224312Δpd+Up+2Jp](tpdπ)4(Δpd)2(34)J_{t_{2g}-t_{2g}}^{2h}=\left[-\frac{80}{81}\frac{1}{2\Delta_{pd}+U_p-3J_p}+\frac{304}{243}\frac{1}{2\Delta_{pd}+U_p-J_p}+\frac{32}{243}\frac{1}{2\Delta_{pd}+U_p+2J_p}\right]\frac{(t_{pd}^{\pi})^4}{(\Delta_{pd})^2} \tag*{(34)}
Jt2g−t2gcyclic=281Δpd(tpdπ)4(Δpd)2(35)J_{t_{2g}-t_{2g}}^{cyclic}=\frac{2}{81\Delta_{pd}}\frac{(t_{pd}^{\pi})^4}{(\Delta_{pd})^2} \tag*{(35)}
Jt2g−egU=5243(−27U−3J+10Dq+43U+J+10Dq+8U+4J+10Dq+24U+2J−10Dq)t62(36)J_{t_{2g}-e_g}^{U}=\frac{5}{243}\left(-\frac{27}{U-3J+10Dq}+\frac{43}{U+J+10Dq}+\frac{8}{U+4J+10Dq}+\frac{24}{U+2J-10Dq}\right)t_6^2 \tag*{(36)}
Jt2g−eg2h=[−102712Δpd+Up−3Jp+10Dq+25024312Δpd+Up−Jp+10Dq+8024312Δpd+Up+2Jp+10Dq](tpdπ)2(tpdσ)2(1Δpd+1Δpd+10Dq)2(37)J_{t_{2g}-e_g}^{2h}=\left[-\frac{10}{27}\frac{1}{2\Delta_{pd}+U_p-3J_p+10Dq}+\frac{250}{243}\frac{1}{2\Delta_{pd}+U_p-J_p+10Dq}+\frac{80}{243}\frac{1}{2\Delta_{pd}+U_p+2J_p+10Dq}\right](t_{pd}^{\pi})^2(t_{pd}^{\sigma})^2\left(\frac{1}{\Delta_{pd}}+\frac{1}{\Delta_{pd}+10Dq}\right)^2 \tag*{(37)}
Jt2g−egcyclic=−40811Δpd+10Dq(tpdπ)2(tpdσ)2(1Δpd+1Δpd+10Dq)2(38)J_{t_{2g}-e_g}^{cyclic}=-\frac{40}{81}\frac{1}{\Delta_{pd}+10Dq}(t_{pd}^{\pi})^2(t_{pd}^{\sigma})^2\left(\frac{1}{\Delta_{pd}}+\frac{1}{\Delta_{pd}+10Dq}\right)^2 \tag*{(38)}
Jeg−egU=100811U+2J(t42+t52)(39)J_{e_g-e_g}^{U}=\frac{100}{81}\frac{1}{U+2J}(t_4^2+t_5^2) \tag*{(39)}
Jeg−eg2h=[−2008112Δpd+Up−3Jp+2(10Dq)+2008112Δpd+Up−Jp+2(10Dq)](tpdσ)4(Δpd+10Dq)2(40)J_{e_g-e_g}^{2h}=\left[-\frac{200}{81}\frac{1}{2\Delta_{pd}+U_p-3J_p+2(10Dq)}+\frac{200}{81}\frac{1}{2\Delta_{pd}+U_p-J_p+2(10Dq)}\right]\frac{(t_{pd}^{\sigma})^4}{(\Delta_{pd}+10Dq)^2} \tag*{(40)}
Kt2g−t2gU=1243(45U−3J+11U+J+28U+4J)t12+1243(−81U−3J+73U+J−4U+4J)t22+281(3U−3J−7U+J−2U+4J)t32+1243(−63U−3J+31U+J−16U+4J)t1t3(41)K_{t_{2g}-t_{2g}}^{U}=\frac{1}{243}\left(\frac{45}{U-3J}+\frac{11}{U+J}+\frac{28}{U+4J}\right)t_1^2+\frac{1}{243}\left(-\frac{81}{U-3J}+\frac{73}{U+J}-\frac{4}{U+4J}\right)t_2^2 +\frac{2}{81}\left(\frac{3}{U-3J}-\frac{7}{U+J}-\frac{2}{U+4J}\right)t_3^2+\frac{1}{243}\left(-\frac{63}{U-3J}+\frac{31}{U+J}-\frac{16}{U+4J}\right)t_1t_3 \tag*{(41)}
Kt2g−t2g2h=[408112Δpd+Up−3Jp−5624312Δpd+Up−Jp+3224312Δpd+Up+2Jp](tpdπ)4(Δpd)2(42)K_{t_{2g}-t_{2g}}^{2h}=\left[\frac{40}{81}\frac{1}{2\Delta_{pd}+U_p-3J_p}-\frac{56}{243}\frac{1}{2\Delta_{pd}+U_p-J_p}+\frac{32}{243}\frac{1}{2\Delta_{pd}+U_p+2J_p}\right]\frac{(t_{pd}^{\pi})^4}{(\Delta_{pd})^2} \tag*{(42)}
Kt2g−t2gcyclic=−2081Δpd(tpdπ)4(Δpd)2(43)K_{t_{2g}-t_{2g}}^{cyclic}=-\frac{20}{81\Delta_{pd}}\frac{(t_{pd}^{\pi})^4}{(\Delta_{pd})^2} \tag*{(43)}
Kt2g−egU=5243(−9U−3J+10Dq+1U+J+10Dq−4U+4J+10Dq+12U+2J−10Dq)t62(44)K_{t_{2g}-e_g}^{U}=\frac{5}{243}\left(-\frac{9}{U-3J+10Dq}+\frac{1}{U+J+10Dq}-\frac{4}{U+4J+10Dq}+\frac{12}{U+2J-10Dq}\right)t_6^2 \tag*{(44)}
Kt2g−eg2h=[−108112Δpd+Up−3Jp+10Dq−5024312Δpd+Up−Jp+10Dq−4024312Δpd+Up+2Jp+10Dq](tpdπ)2(tpdσ)2(1Δpd+1Δpd+10Dq)2(45)K_{t_{2g}-e_g}^{2h}=\left[-\frac{10}{81}\frac{1}{2\Delta_{pd}+U_p-3J_p+10Dq}-\frac{50}{243}\frac{1}{2\Delta_{pd}+U_p-J_p+10Dq}-\frac{40}{243}\frac{1}{2\Delta_{pd}+U_p+2J_p+10Dq}\right](t_{pd}^{\pi})^2(t_{pd}^{\sigma})^2\left(\frac{1}{\Delta_{pd}}+\frac{1}{\Delta_{pd}+10Dq}\right)^2 \tag*{(45)}
Kt2g−egcyclic=20811Δpd+10Dq(tpdπ)2(tpdσ)2(1Δpd+1Δpd+10Dq)2.(46)K_{t_{2g}-e_g}^{cyclic}=\frac{20}{81}\frac{1}{\Delta_{pd}+10Dq}(t_{pd}^{\pi})^2(t_{pd}^{\sigma})^2\left(\frac{1}{\Delta_{pd}}+\frac{1}{\Delta_{pd}+10Dq}\right)^2. \tag*{(46)}

Also, the off-diagonal term Γ\Gamma is given by

Γ=481(3U−3J−7U+J−2U+4J)t1t2+1243(−63U−3J+31U+J−16U+4J)t2t3.(47)\Gamma=\frac{4}{81}\left(\frac{3}{U-3J}-\frac{7}{U+J}-\frac{2}{U+4J}\right)t_1t_2+\frac{1}{243}\left(-\frac{63}{U-3J}+\frac{31}{U+J}-\frac{16}{U+4J}\right)t_2t_3. \tag*{(47)}

Using the above equations, the magnetic exchange interactions JHJ_H, JKJ_K, and Γ\Gamma under strain effect can be obtained with our material parameters obtained from DFT. For U=5U=5 eV and J=0.8J=0.8 eV, we found that the Kitaev ∣JK∣|J_K| term is the largest while the Heisenberg ∣JH∣|J_H| and the off-diagonal ∣Γ∣|\Gamma| terms are still finite at zero strain. Within the limit of perfect octahedron, ∣JK∣|J_K| and ∣Γ∣|\Gamma| are not sensitive to strain, but ∣JH∣|J_H| is reduced under the tensile strain (Figure 12a) due to the reducing of ∣t3∣|t_3| hopping (Figure 6a). Our JKJ_K and JHJ_H results at zero strain are consistent with those obtained from the DFT spin-exchange calculation[14] and the trends of these values under strains are also similar to those in Cu3_3Co2_2SbO6_6[12].

Image not extracted

Figure 12. (a) Heisenberg JHJ_H, Kitaev JKJ_K, and off-diagonal Γ\Gamma interactions versus strain with U=5U = 5 eV and J=0.8J = 0.8 eV, and (b) Heisenberg (solid) and Kitaev (dashed) interactions versus strain with J=0.8J = 0.8 eV and various values of UU. The exchange interactions are calculated using Liu and Kee analysis[10] with Up=0.7UU_p=0.7U[3] and Jp=0.3UpJ_p=0.3U_p[51].

We also found that the Hubbard UU and Hund’s coupling JJ values can also affect the exchange interactions. While fixing J=0.8J=0.8 eV and varying UU, we found that the system is pushed closest to the QSL (JH∼0J_H\sim0) at U=5U=5 eV, but the JHJ_H dependence is not much sensitive to the strain, compared to the one at U=4U=4 eV (Figure 12b). On the other hand, while fixing U=5U=5 eV and varying JJ, we found that the overall Heisenberg interaction JHJ_H is very sensitive to JJ (Figure 13).

Image not extracted

Figure 13. Heisenberg and Kitaev interactions from each exchange channel and their total versus strain calculated using Liu and Kee analysis[10] with U=5U=5 eV, Up=0.7UU_p=0.7U[3], and Jp=0.3UpJ_p=0.3U_p[51] while varying the Hund’s coupling JJ.

For J=0.6−0.7J=0.6-0.7 eV (U=5U=5 eV) the overall Heisenberg interaction JHJ_H is small, while the FM Kitaev one is large due to the dominant contribution from the t2gt_{2g}-ege_g channel, consistent with Liu et al analysis for d7d^7 Co compounds[5]. We found that the JHJ_H generated from the ege_g-ege_g channel is not very sensitive to the strain and also remains unchanged as JJ varies, consistent with Liu and Khaliullin analysis[3]. However, the AFM Heisenberg interaction from the t2gt_{2g}-ege_g channel weakens the combined t2gt_{2g}-t2gt_{2g} and ege_g-ege_g FM Heisenberg interactions, thus reducing the overall Heisenberg interaction JHJ_H. As the Hund’s coupling JJ gets larger (J≥0.8J\geq0.8 eV), the contribution from the t2gt_{2g}-t2gt_{2g} channel gets larger, therefore, the total JHJ_H is enhanced while the total JKJ_K is weakened.

References

  1. [3]H. Liu and G. Khaliullin, Phys. Rev. B 97, 014407 (2018).
  2. [5]H. Liu, J. c. v. Chaloupka, and G. Khaliullin, Phys. Rev. Lett. 125, 024701 (2020).
  3. [6]L. Viciu, Q. Huang, E. Morosan, H. Zandbergen, N. Greenbaum, T. McQueen, and R. Cava, Journal of Solid State Chemistry 180, 1060 (2007).
  4. [7]C. Wong, M. Avdeev, and C. D. Ling, Journal of Solid State Chemistry 243, 18 (2016).
  5. [8]V. Politaev, V. Nalbandyan, A. Petrenko, I. Shukaev, V. Volotchaev, and B. Medvedev, Journal of Solid State Chemistry 183, 684 (2010).
  6. [9]J.-Q. Qin, S. Okamoto, Y. Wu, Q. Zheng, H. D. Zhou, H. B. Cao, and M. A. McGuire, Phys. Rev. Mater. 3, 074405 (2019).
  7. [10]X. Liu and H.-Y. Kee, Phys. Rev. B 107, 054420 (2023).
  8. [12]G.-H. Kim, M. Park, S. Samanta, U. Choi, B. Kang, U. Seo, G. Ji, S. Noh, D.-Y. Cho, J.-W. Yoo, J. M. Ok, H.-S. Kim, and C. Sohn, Science Advances 10, eadn8694 (2024).
  9. [13]M. van Veenendaal, E. H. T. Poldi, L. S. I. Veiga, P. Ben-cok, G. Fabbris, R. Tartaglia, J. L. McChesney, J. W. Free-land, R. J. Hemley, H. Zheng, J. F. Mitchell, J.-Q. Yan, and D. Haskel, Phys. Rev. B 107, 214443 (2023).
  10. [14]C. Kim, J. Jeong, G. Lin, P. Park, T. Masuda, S. Asai, S. Itoh, H.-S. Kim, H. Zhou, J. Ma, and J.-G. Park, Journal of Physics: Condensed Matter 34, 045802 (2021).
  11. [16]B. S. Mou, X. Zhang, L. Xiang, Y. Xu, R. Zhong, R. J. Cava, H. Zhou, Z. Jiang, D. Simony, N. Drichko, and S. M. Winter, Phys. Rev. Mater. 8, 084408 (2024).
  12. [20]K. Mitsumoto, C. Hotta, and H. Yoshino, Phys. Rev. Res. 4, 033157 (2022).
  13. [21]A. I. Lichtenstein, V. I. Anisimov, and J. Zaanen, Phys. Rev. B 52, R5467 (1995).
  14. [22]P. E. Blöchl, Phys. Rev. B 50, 17953 (1994).
  15. [23]G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
  16. [24]G. Kresse and D. Joubert, Phys. Rev. B 59, 1758 (1999).
  17. [25]J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 78, 1396 (1997).
  18. [26]V. Singh, U. Herath, B. Wah, X. Liao, A. H. Romero, and H. Park, Computer Physics Communications 261, 107778 (2021).
  19. [27]A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari, Computer Physics Communications 178, 685–699 (2008).
  20. [28]A. T. Lee, H. Park, and A. T. Ngo, Phys. Rev. B 108, 205146 (2023).
  21. [29]K. Haule, Phys. Rev. B 75, 155113 (2007).
  22. [30]H.-S. Kim, K. Haule, and D. Vanderbilt, Phys. Rev. Lett. 123, 236401 (2019).
  23. [31]S. Samanta, F. Cossu, and H.-S. Kim, Npj Quantum Matter 9 (2024).
  24. [32]A. B. Georges, A. J. Millis, and J. M. Rondinelli, Phys. Rev. B 105, 245153 (2022).
  25. [33]F. Rodolakis, J.-P. Rueff, M. Sikora, I. Alliot, J.-P. Itié, F. Baudelet, S. Ravy, P. Wzietek, P. Hansmann, A. Toschi, M. W. Haverkort, G. Sangiovanni, K. Held, P. Metcalf, and M. Marsi, Phys. Rev. B 84, 245113 (2011).
  26. [34]S. M. Winter, Journal of Physics: Materials 5, 045003 (2022).
  27. [36]M. I. Katsnelson and A. I. Lichtenstein, Phys. Rev. B 61, 8906 (2000).
  28. [37]N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Reviews of Modern Physics 84, 1419–1475 (2012).
  29. [38]A. A. Mostofi, J. R. Yates, G. Pizzi, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari, Computer Physics Communications 185, 2309 (2014).
  30. [39]C. Kim, H.-S. Kim, and J.-G. Park, Journal of Physics: Condensed Matter 34, 023001 (2021).
  31. [40]K. Held, Advances in Physics 56, 829 (2007).
  32. [41]P. W. Anderson, Phys. Rev. 124, 41 (1961).
  33. [42]G. Kotliar and D. Vollhardt, Physics Today 57, 53 (2004).
  34. [43]F. i. c. v. IV and M. Ferrero, Phys. Rev. B 105, 125104 (2022).
  35. [44]X.-Z. Li and H. Yao, Ann. Rev. Condensed Matter Phys. 10, 337 (2019), arXiv:1805.08219 [cond-mat.str-el].
  36. [45]E. Sheridan, C. Weber, E. Plekhanov, and C. Rhodes, Phys. Rev. B 99, 205156 (2019).
  37. [46]E. Gull, A. J. Millis, A. I. Lichtenstein, A. N. Rubtsov, M. Troyer, and P. Werner, Rev. Mod. Phys. 83, 349 (2011).
  38. [47]P. Werner and A. J. Millis, Phys. Rev. B 74, 155107 (2006).
  39. [48]M. Jarrell and J. Gubernatis, Physics Reports 269, 133 (1996).
  40. [49]B. Kang, M. Park, S. Song, S. Noh, D. Choe, M. Kong, M. Kim, C. Seo, E. K. Ko, G. Yi, J.-W. Yoo, S. Park, J. M. Ok, and C. Sohn, Phys. Rev. B 107, 075103 (2023).
  41. [50]T. Takayama, J. Chaloupka, A. Smerald, G. Khaliullin, and H. Takagi, Journal of the Physical Society of Japan 90, 062001 (2021), https://doi.org/10.7566/JPSJ.90.062001.
  42. [51]K. F. Foyevtsova, H. O. Jeschke, I. I. Mazin, D. J. Khomskii, and R. Valentí, Phys. Rev. B 88, 035107 (2013).

Paper details

vlnc:W_fq9aaj3g
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

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