WorksPhysics and Astronomy
Comparing Meta-GGAs, +U Corrections, and Hybrid Functionals for Polaronic Point Defects in Layered MnO, NiO, and KCoO
Abstract
Defects in a material can significantly tune properties and enhance utility. Hybrid functionals like HSE06 are often used to describe solids with such defects. However, geometry optimization (including accounting for effects such as Jahn-Teller distortion) using hybrid functionals is challenging for the large supercells needed for defect study. The proposed rSCAN+rVV10+U+ method, which is computationally much cheaper and faster than hybrid functionals, can successfully describe defects in materials with the proper choice of U (for the d orbitals of the host atom) and (for those of the defect atom), as shown here for small polaron defects in layered transition-metal oxides. For a range of U and around literature values (from solid-state reaction energies) for a given transition-metal ion and its oxidation state, we find that this approach predicts localized polaronic states in band gaps, as hybrid functionals do. The layered materials birnessite () and , with one K atom intercalated between layers in a supercell, are found to have one localized occupied polaronic state on the transition metal ion reduced by the insertion of the K atom, when the geometry is calculated as above using published U values. The expected Jahn-Teller distortion is not observed when U==0. Layered cobalt oxide with additional potassium ions intercalated () is different, due to a dramatic difference in electronic configuration of the defected Co(II) ion: A single extra K atom in the supercell leads to four localized electrons in the band gap, using standard U values, and even for U==0.
Introduction
This work is part of a larger project on the computational identification of alkali-atom-intercalated layered materials as promising catalysts for the oxygen evolution reaction (OER) of water splitting for clean hydrogen production. Earlier work [1, 2] showed the importance of small-polaron defects formed by transfer of an electron from an intercalated alkali atom to a neighboring manganese ion. Electron flow in catalysis seems to be facilitated by a localized polaronic state in the energy gap close to the bottom of the conduction band. Calculation of the polaron often requires a nonlocal density functional for the exchange-correlation energy, and geometry optimization in a large supercell. We found that hybrid functionals as used in Ref. [1] are too expensive for a broad materials search. In this article, we show that an alternative rSCAN+rVV10+U+U approach can predict localized polaronic states at much lower cost, and may be useful for other point defects in a range of materials. We also found that, while one intercalated K atom creates one localized electron in the energy gap of layered MnO and NiO, it can create four in layered KCoO. We further found that, while rSCAN needs a +U correction to create a polaron in MnO and NiO, it does not need one in KCoO. Our broader computational and experimental materials search is now underway.
Various types of defects exist in solids, and defects in solids can influence numerous important properties like electrical conductivity, reactivity, and magnetic or optical properties. For example, the Oxygen Evolution Reaction (OER) is frequently favored by defects like polarons in transition metal oxide (TMO)-based catalysts [1, 2]. Leveraging defects as a tool allows fine tuning of the electronic properties of materials, making the understanding of defects in materials a pivotal area of research.
The computational design and study of such materials using first-principles density functional theory (DFT) [3, 4] offer valuable early insights. However, the approach presents challenges, as density functional approximations (DFAs) introduce self-interaction error (SIE). Popular DFAs such as LDA/GGA/meta-GGA tend to underestimate the Perdew-Parr-Levy-Balduz (PPLB) [5] straight line condition, leading to inaccuracies in describing charge transfer that are reduced but not eliminated in that sequence of functionals. Given that defect studies require proper charge transfer, DFAs often fall short in accurately portraying defects in a system.
Hybrid functionals, which combine a fraction of exact exchange, such as HSE06 [6] which utilizes 25% exact exchange and 75% PBE exchange in the short range along with full PBE exchange in the long range, experience less SIE and often provide a more accurate description of the electronic structure of materials. Hybrid functionals have been successfully employed in studying defects in solids. For instance, Peng et al. [1] effectively investigated polaron-like defects in birnessite (Figure 1, KMnO, ). Nevertheless, the inclusion of exact exchange in hybrid functionals renders them computationally expensive. The structural relaxation using hybrid functionals becomes particularly costly for a reasonably sized supercell with localized defects, with computational expenses rapidly escalating as the supercell size increases toward better simulation of defects. Consequently, the structural optimization of large supercells becomes impractical. Another challenge associated with hybrid functionals is that a material-independent exact exchange mixing parameter is not determined through any exact condition, nor is the range-separation parameter in range-separated hybrids [7]. Peng et al. determined a mixing parameter of 0.22 to study defects in birnessite using HSE06 [1], deviating from the original value of 0.25. Additionally, determining a mixing parameter for semiconductors that may not be suitable to metals or insulators [8] adds to the challenge of finding an appropriate parameter for the system under study when employing hybrid functionals.

Figure 1. General structure for layered potassium metal oxides .
SCAN [9] is a recently developed meta-generalized gradient approximation that reinstates exact constraint adherence to rSCAN [10], preserving the numerical efficiency of rSCAN while simultaneously restoring the transferable accuracy of SCAN [11]. Several studies have demonstrated that SCAN predicts geometries and other properties as well as or even better than hybrid functionals. Sun et al. [7] showed that SCAN accurately predicts geometries and energies of diversely bonded materials and molecules, matching or surpassing the accuracy of computationally expensive hybrid functionals. Another study by Saynik and Cocchi [12] on cesium-based photocathode materials CsSb and CsTe reported excellent performance of SCAN and HSE06 for both structural and electronic properties. A recent paper on the arXiv [13] reports that, while SCAN may not reliably describe the properties of deep defects and small polarons in several semiconductors and insulators, it yields remarkably good agreement with experimental structural parameters for materials like ZnO, GaN, GaO and NaF. Additionally, a study by Varadwaj and Miyake [14] on the geometrical, electronic, and optical properties of vanadium dioxide found that SCAN and SCAN-rVV10 can adequately predict the most important geometrical and optoelectronic properties of VO. Numerous related studies further support the idea that SCAN successfully describes the structure and other properties of materials. Given that closely agrees with SCAN in accuracy, we expect that would exhibit similar accuracy in the aforementioned studies.
By construction, can exhibit considerably less self-interaction error (as reflected in its smaller U value [15]) than standard GGAs. (The effective one-electron potential can still show substantial self-interaction error.) In this study, we demonstrate that the functional can predict the existence of localized polaronic states at defects, as the HSE06 hybrid functional does. Here, U represents the Hubbard U correction of Anisimov and collaborators [16, 17, 18], applied to transition metal sites other than the defect site, while U is the correction applied to the defect site. Cococcioni and de Gironcoli [19] showed that the +U correction can be regarded as a many-electron self-interaction correction that, like the PPLB condition [5], penalizes non-integer electron number on a localized orbital to which it is applied.
Additionally, vdW denotes the long-range van der Waal’s correction. In our approach, we utilize rVV10 [20, 21] to account for this interaction, which importantly reduces inter-layer spacing. In many of our calculations, including HSE06 without vdW, we use the supercell volume and shape from , although we can as a check relax the internal coordinates with HSE06.
Transition metal ions that serve as sites for defect formation in solids, are in different formal oxidation states (OS) than otherwise-identical ions. For example, in a K-intercalated MnO, the polaronic Mn site is in the +3 OS and the remaining Mn ions are in the +4 oxidation state [1, 2]. Ions in different OS have different numbers of d electrons. Ref. [22] reports that the ideal U correction decreases with increasing OS, which is attributed to a lower number of exchange interactions among fewer d electrons in a higher oxidation state. This behavior has been observed and reported [22] for vanadium ions, and suggests U > U.
However, this situation is not universal, as we can see that U values for Mn for are 1.5 eV, 2.1 eV and 1.8 eV for Mn ions with average oxidation states +3.5, +2.5 and +2.33 respectively [23]. Here we see that Mn in the +2.5 average OS needs more U correction than Mn in the +2.33 average OS, which does not follow the trend observed for vanadium, although U for Mn in the +3.5 average OS supports the trend within the same elemental series. However, it is obvious for the Mn oxide system that Mn ions in different oxidation states need different U corrections. Another example for Co is found in Ref. [22], which found U=3.0 eV as a correction to SCAN for the oxidation reaction 6CoO+O → 2CoO (in which the Co ions are in the +2 and +2.67 oxidation states), but also found that SCAN with this U wrongly predicted the meta-stable crystal O1−CoO (in which the Co oxidation state is +4) to be non-metallic, while SCAN with U=0 correctly predicted it to be a metal. Thus the U value can vary with the oxidation state of ions in the system. This indicates that if a system has transition metal ions in different OS, the proper way to describe such a system would be to apply different U corrections to the transition metal ions based on their OS. This could be because DFAs make different SIE for ions in different OS.
Here we present a study of defects in solids by the double U correction method, where U is the U correction applied to the defect site and U is the U correction applied to remaining sites as required for the system. Since U typically increases with decreasing OS, and defect sites are typically in a lower oxidation state, we expect U ≥ U, although exceptions are possible.
One can use the geometry and the HSE06 hybrid functional orbital energies to leverage the best features of the hybrid functional (e.g., band gap). It appears that employing the full rSCAN + rVV10 + U + U and/or HSE06 using rSCAN + rVV10 + U + U geometry is a state-of-the-art method for efficient study of defects in solids.
From a chemical perspective [24], for electronic configurations , the spin state of a transition metal ion M in an MO motif arises from a competition between crystal field effects which favor the low-spin state and internal effects which favor the high-spin state. The crystal field effect (which in the absence of spin removes the degeneracy of three and two orbitals) depends on the lattice geometry, including any Jahn-Teller distortion of the MO octahedron. The orbitals have lobes between the bond axes, minimizing the repulsion between the metal ion and its surrounding oxygen ions. The ions we consider here are in oxidation states Mn(IV) , Mn(III) , Ni(IV) , Ni(III) , Co(III) , and Co(II) .
Results and Discussion
We choose three layered pristine TMOs, MnO, NiO, and KCoO as the starting point for our study. Transition metal and oxygen ions in these TMOs are arranged in layers of MO6 (where M=Ni, Mn, Co) octahedra, with 32 transition metal ions per supercell. Inserting an additional potassium ion into the supercell between the layers creates a polaronic defect, specifically a Jahn-Teller electron small polaron [1]. This defect has been studied in birnessite (MnO) using the HSE06 functional [1].
Layered MnO
First, we calculated the one-electron density of states (DOS) for both the pristine(i.e., non-intercalated) and a K-intercalated birnessite (Figure 1) using the HSE06 functional with an exact exchange mixing parameter () of 0.22 to reproduce the work of Peng et al. [1]. The resulting DOS are shown in Figure 2. Comparing Figure 2A and Figure 2B in the current work with Figs. 2B and 2C of Peng et al. [1], we conclude that we have successfully reproduced the latter’s DOS results. In Figure 2B, a K-intercalated birnessite exhibits the appearance of Mn(III) -states at higher energy, a polaronic peak at the conduction band (CB) edge, and a break in spin symmetry of the total DOS in the CB. These effects were previously seen and explained by Peng et al. [1]. The spin symmetry of the total DOS in pristine MnO arises from the antiferromagnetic order of the Mn(IV) ions, and is disrupted by the defect Mn(III) ion (while the other pristine materials are non-magnetic). The appearance of the polaron peak is attributed to states in Mn(III) [1, 2].

Figure 2. HSE06 () spin-resolved density of states per atom in (A) pristine with Mn(IV) ions and (B) with a single intercalated ion and a defect Mn(III) cation in a supercell with 97 atoms. For comparison with Ref. [1], the interlayer spacing is set to 7.12 Å, and internal coordinates are relaxed in HSE06. In Figure 1-Figure 9 (except in Figure 5 and 8), the shaded area shows the total density of one-electron states, and the blue curve shows the transition-metal-d states projected onto the site where the small polaron forms or is expected to form. Also in Figure 1-Figure 9 (except in Figure 5 and 8), the peaks (if any) in the band gap below the conduction band minimum (CBM) are occupied, localized states of the defect ion. The energy zero is set to the energy of the lowest-energy unoccupied orbital. The total density has been scaled down to make it comparable in size to the projected Mn-d density of states.
We will see similar effects upon a K-intercalation in NiO and KCoO systems later, and these effects can be understood through similar reasoning. The -state electronic configurations of Mn(IV) in pristine MnO and Mn(III) in a single-K-intercalated MnO are as shown in Figure 5 A and B respectively.

Figure 5. d-orbital splitting (A) in Mn(IV) in pristine MnO demonstrating filled states and empty states and (B) in Mn(III) in a single K-intercalated MnO.
Swathilakshmi et al. [23] recently determined the optimal U value for rSCAN for Mn to be 1.8 eV. They utilized three oxidation reactions: MnOMnO, MnOMnO, and MnOMnO. The U values for these reactions are 2.1 eV, 1.8 eV and 1.5 eV respectively [23]. The average OS of Mn ions in these reactions are +2.5, +2.3 ,and +3.5 respectively [23]. U values for these reactions are small compared to those for the PBE GGA and do not differ significantly, suggesting that rSCAN makes a small, comparable SIE but not an equal one for Mn(II), Mn(III) and Mn(IV) ions in manganese oxide systems.
First, we performed rSCAN + rVV10 + U calculations for pristine birnessite with the optimal U of 1.8 eV and obtained the DOS as shown in Figure 3A. rSCAN + rVV10 + U predicts a band gap of 2.30 eV for the pristine structure. The underestimation of the band gap compared to HSE06 by meta-GGA/GGA is a well-known general trend [25]. Next, we performed calculations for a K-intercalated . As exhibits a small SIE, we initially set eV to see the performance of without correction. However, this method failed to resolve the Jahn-Teller defect, as depicted in Figure 3B. The corresponding electron from the K atom is delocalized, which can be seen by the extra electrons between the CB minimum and the chemical potential or Fermi level at 0 eV. This can be clearly seen in the inset of

Figure 3. rSCAN+rVV10+U+U spin-resolved density of states per atom in (A) pristine MnO with U=U=1.8 eV and in a single K-intercalated MnO with (B) U = U = 0.0 eV, (C) U = U =1.8 eV, (D) U=1.71 eV and U =1.58 eV. The geometry has been optimized in rSCAN+rVV10+U+U, which makes the interlayer spacing ≈ 5.26 Å. The insets in Figs. A and B show the DOS near the 0 eV region. The inset of Fig. B shows that for U=U=0 the extra electron from the intercalated K goes to the bottom of the conduction band, making a semi-metal.
Figure 3B. For the description of a polaron in a single K-intercalated MnO, DFA has to transfer charge from the inserted K atom to a Mn atom. The charge delocalization error of DFAs prohibits transferring a complete electron from the K atom to the Mn site. This indicates the need for +U correction, which can remove the partial occupancy and localize the electron on the defect site. Subsequently, we tested the optimal U by setting U=U=1.8 eV and the obtained DOS is shown in Figure 3C, revealing the appearance of a polaronic peak, and the appropriate distortion of Mn-O bonds at the Jahn-Teller distorted manganese atom (see Fig. S2 B). The +U correction applies an energy penalty to the partially occupied orbital. As a result the polaron is localized and appears at the CB edge. We see shifting of Mn(III) d-states to higher energy, and spin symmetry breaking in the total DOS in the CB, consistent with the HSE06 results.
As discussed in the work by Peng et al. [1] and Ding et al. [2], polaron formation can with the right distribution of intercalated potassium atoms create a potential step between layers that facilitates electron transfer between layers and enhances catalytic activity.
Motivated by studies indicating that U depends on OS, and that generally U decreases with an increase in OS, we determined U for the functional for Mn(III) and Mn(IV) by matching ’s magnetic moment for a pristine material to the HSE06+D3 magnetic moment (Supplementary Material (SM) [26]). (All magnetic moments reported in this article arise from spin density inside the Wigner-Seitz sphere.) This method gave U values of 1.71 eV and 1.58 eV for Mn(III) and Mn(IV) ions, respectively. These values are close to the values in Ref. [23].
We then obtained the DOS of a K-intercalated MnO with U=1.71 eV and U=1.58 eV which is as shown in Figure 3C. U=1.71 eV and U=1.58 eV gave similar results to U=U=1.8 eV.
We also performed HSE06+D3 calculations on the geometries, and the resulting DOS are plotted in Figure 4. The D3 dispersion correction was used for the HSE06 functional with new D3 damping parameters a1 = 0.383, a2 = 5.685, and s8 = 2.310 generated for the HSE06 functional [27]. For HSE06+D3 calculations, we

Figure 4. HSE06(α=0.25)+D3 spin-resolved density of states per atom using rSCAN+rVV10+U+U geometry of Fig. 3 in (A) pristine MnO with U=U=1.8 eV and in a single K-intercalated MnO with (B) U=U=0.0 eV, (C) U=U=1.8 eV, (D) U=1.71 eV and U=1.58 eV. The insets in Figs. A and B show DOS near the 0 eV region. The inset in Fig. B shows that for U=U=0 the extra electron from the intercalated K goes to the bottom of the conduction band, making a semi-metal.
used an exact exchange mixing parameter () of 0.25. This is because we found that the geometry in much better agreement with the HSE06+D3 geometry for =0.25 than for 0.22. More details are in the SM [26].
Figure 4A shows the HSE06+D3 DOS of pristine MnO obtained using the (U=U=1.8 eV) geometry. HSE06+D3 increases the band gap of pristine MnO, close to the HSE06 result (Figure 2A). Figure 4B shows the HSE06+D3 DOS of a K-intercalated MnO obtained using the (U=U=0.0 eV) geometry. Like the (U=U=0.0 eV) method, HSE06+D3 also fails to localize the defect state due to the absence of the localized geometric distortion near a single Mn atom. Figure 4C shows the HSE06+D3 DOS of a K-intercalated MnO obtained using the (U=U=1.8 eV) geometry. Here, the defect state appears as in the (U=U=1.8 eV) method, but deep in the band gap. Figure 4D shows the HSE06+D3 DOS of a K-intercalated MnO obtained using the (U=1.71 eV and U=1.58 eV) geometry. Here, also, the defect state appears as in the (U=U=1.8 eV) method, but deep in the band gap.
We also calculated HSE06 and HSE06+D3 DOS using an exact exchange mixing parameter of 0.22 as determined by Peng et al. [1] on geometry with eV and obtained DOS as shown in Fig. S1 A and B, respectively. As expected, HSE06() and HSE06()+D3 DOS do not look different. Notably, the HSE06() DOS does not look different from the full HSE06()+D3 evaluated on the same geometry, which is shown in Figure 4C.
R. Ding et al. [2] have proposed a position of the polaron close to the CB in layered MnO with an alternation of polaron-rich and polaron-poor layers. This scenario matches better with DOS, where the polaron is close to the CB.
It is challenging to conclude which method provides a better description of the defect. However, we can assert that all three methods- HSE06, and HSE06+D3 using geometry-have successfully described the polaronic defect in birnessite with the proper and for calculations.
Layered NiO
Our next system under study was a layered NiO. This material has a hexagonal crystal structure with space group P6/mmc. The Materials Project website (https://nextgen.materialsproject.org/) reports that the material is synthesizable but not stable. Whether stable or not, this material is of interest for our study. Layered NiO has a similar structure to birnessite, where Ni ions are in a +4 OS with completely occupied states, and empty states [28].
We began the DOS calculation using the functional for both pristine and K-intercalated NiO systems. The value for Ni ions was recently determined by Swathilakshmi et al. [23]. For the pristine NiO, with eV, correctly predicts a non-magnetic ground state with a band gap of 1.62 eV, as shown in Figure 6A. As in the birnessite case, we expect that adding an extra K atom between layers would transfer an electron from the inserted K atom to a Ni site, forming a defect. The defected Ni site would undergo Jahn-Teller distortion, localizing the electron in the state, as shown in Figure 8B.

Figure 6. rSCAN+rVV10+U+U spin-resolved density of states per atom in (A) pristine NiO with Ni(IV) ions with U=U=2.1 eV and in a single K-intercalated NiO with (B) U=U =0.0 eV, (C) U=U=2.1 eV, (D) U= U =2.06 eV. The geometry was optimized in rSCAN+rVV10+U+U, which makes the interlayer spacing ≈ 5.42 Å. Here we can see one localized occupied state and several localized unoccupied states on the defect Ni(III) cation. The insets in Figs. A and B show the DOS near the 0 eV region. The inset in Fig. B shows that for U=U=0 the extra electron from the intercalated K goes to the bottom of the conduction band, making a semi-metal.

Figure 8. d-orbital splitting (A) in Ni(IV) in pristine NiO demonstrating filled states and empty states and (B) in Ni(III) in a single K-intercalated NiO
For a K-intercalated NiO, we initially used eV to observe how without U correction performs for this system, and we obtained the DOS as plotted in Figure 6B. Again, this choice of and failed to describe the defect state, as can be seen from the inset in Figure 6B. The extra electron goes to a delocalized state at the bottom of the conduction band, with fractional occupation on each Ni ion in the supercell, and no individual geometric distortion is observed at any nickel atom. We then used the available U value by setting eV and obtained the DOS, as plotted in Figure 6C. This choice of U values forms a defect at one of the Ni sites, evidenced by the expected geometric distortion (see Fig. S3 B). The polaron is localized and appears in the gap just below the CB minimum. The formation of the polaron at the VB edge indicates that a K-intercalated NiO could show OER catalytic activity similar to that of birnessite by lowering the overpotential. Most of the charge from the K atom is transferred to this defect site, which has a magnetic moment of 0.786 . However, we observe a Ni site in another layer picking up a small but nonzero magnetic moment of 0.137 , with the magnetic moment of all remaining Ni sites smaller than 0.03 . This suggests that we might need different values of U and for to accurately describe charge transfer to the defect site in a K-intercalated NiO system.
The determination of eV for Ni for involves Ni in +2 and +3 states [23], with an average OS of +2.5. Since the precise U values for Ni(III) and Ni(IV) states are not available, we set eV for NI(III) as the Ni (III) OS is close to +2.5 and searched for different U values for Ni(IV). We observed that increasing the U value beyond 2.3 eV transfers more and more charge to a Ni site in a layer different from that of the defect site. So, we lowered the U values (U=1.0 eV and 0.0 eV ) and found that lowering the U values slightly improves the solution by lowering the magnetic moment of a Ni site in another layer to 0.06 without significant change in magnetic moment of the defect site. The DOS plot for U=0.0 eV and U=2.1 eV do not look different from Fig. 6C. We also increased the U value, keeping U fixed at 2.1 eV, and observed an increase in the magnetic moment of the defect site, reaching 1 for U=3.8 eV. These are the magnetic moments due to the spin density inside the Wigner-Seitz sphere. It is important to note that the magnetic moment of the defect site 1.0 does not guarantee the full transfer of an electron from the intercalated K atom to the defect site, as this is a magnetic moment inside the Wigner-Seitz sphere, which does not reflect the actual magnetic moment of the defect site. However, this analysis shows that
one can adjust the U and U values for more charge transfer to the defect site where necessary.
We also determined the U value of 2.06 eV for the Ni(III) ion by equating the HSE06+D3 magnetic moment to the rSCAN+rVV10+U magnetic moment, as discussed in the SM [26]. As this value is not very different from 2.1 eV, we got a similar rSCAN+rVV10+U+U result with U=U=2.06 eV compared to U=U=2.1 eV. Figure 6D shows rSCAN+rVV10+U+U DOS with U=U=2.06 eV.
Similar to birnessite, the d-states corresponding to Ni(III) in the VB are shifted to higher energy, as shown in Figure 6C and Figure 6D, and there is spin symmetry breaking on the total DOS in the CB. Here we do not claim U=0.0 eV to be the precise U value for Ni(IV) ions, but we suspect that U could be different than U for a K-intercalated NiO system.
We also did HSE06+D3 calculations using the rSCAN+rVV10+U+U geometry, and the results are shown in Figure 7. We observe an increase in the band gap of pristine NiO, as shown in Figure 7A. For a single K-intercalated NiO, due to the absence of a distorted nickel center, we do not observe polaron formation for U=U=0.0 eV, similar to rSCAN+rVV10+U+U as shown in Figure 7B; however, we observed an increase in the band gap. We also performed HSE06+D3 calculations for U=U=2.1 eV and for U=U=2.06 eV and obtained DOS as shown in Figure 7C, and Figure 7D, respectively.

Figure 7. HSE06+D3 spin-resolved density of states per atom using rSCAN+rVV10+U+U geometry of Fig. 6 in (A) pristine NiO with U=U=2.1 eV and a single K-intercalated NiO with (B) U=U=0.0 eV and (C) U=U=2.1 eV, (D) U= U =2.06 eV. The insets in Figs. A and B show the DOS near the 0 eV region. The inset in Fig. B shows that for U=U=0 the extra electron from the intercalated K goes to the bottom of the conduction band, making a semi-metal.
We observe an increase in the band gap and a shift of the polaron peak deep into the band gap region compared to the corresponding rSCAN+rVV10+U+U DOS. We also observe other effects like shifting of Ni(III) d-states to higher energy and spin symmetry breaking on the total DOS in the CB.
Both methods, full rSCAN+rVV10+U+U and HSE06+D3 using the rSCAN+rVV10+U+U geometry with proper U and U values, successfully describe defects in a single K-intercalated NiO. Among the methods employed to study defect in a K-intercalated NiO, full rSCAN+rVV10+U+U is notably superfast.
Layered KCoO
We studied one final related layered material: KCoO, which has a hexagonal crystal structure with space group P6/mmc, similar to the above materials MnO and NiO. The Materials Project website (https://nextgen.materialsproject.org/) shows a similar material, LiCoO, with space group P6mc, to be synthesizable but not stable. The electronic configuration of cobalt d-states is distinct from that of Mn and Ni, as illustrated in Figure 1 in Ref. [2]. The optimal U value determined for Co ions for rSCAN is 1.8 eV [23].
First, we calculated the DOS of KCoO using the optimal U value of 1.8 eV, as shown in Figure 9A. We observed symmetry in the spin resolved total DOS and Co d-states projected onto a Co site. This suggests that Co(III) ions in KCoO have zero spin. The zero spin of Co(III) is also reported in the work of Chen et al. [29]. The d-state configuration of Co(III) is as shown in Figure 11A.

Figure 9. SCAN+rVV10+U spin-resolved density of states per atom in (A) pristine KCoO with Co(III) ions with U=U=1.8 eV, a single K-intercalated KCoO with (B) U=U=0.0 eV, (C) U=U=1.8 eV, (D) U=U=0.9 eV. The more complicated pattern of in-gap states could reflect a Co(II) configuration different from a single unpaired electron in an state. Here we see the appearance of several localized occupied states on the defect Co(II) cation. The geometry was optimized in SCAN+rVV10+U+U, which makes the interlayer spacing Å. In the fundamental gap of K-intercalated KCoO, we see two localized ↑ electrons and two localized ↓ electrons (Figure 11).

Figure 11. Schematic diagram of energy of d-orbitals in (A) Co(III) in pristine KCoO based on HSE06+D3 calculation using rSCAN+rVV10+U+U geometry with U=U=1.8 eV, (B) Co(II) in a single K-intercalated KCoO based on orbital energy from rSCAN+rVV10+U+U calculation with U=U=1.8 eV and (C) Co(II) in a single K-intercalated KCoO based on orbital energies from the HSE06+D3 calculation using the rSCAN+rVV10+U+U geometry with U=U=1.8 eV.
The pristine CoO has cobalt d states in a configuration, and KCoO has cobalt in a configuration. A K-intercalated KCoO would have a defect cobalt site in a configuration, with the remaining Co ions in a configuration. It is interesting to note that the Co(III) ion has no unpaired electron, but Co(II) has three unpaired electrons due to a low-spin to high-spin crossover [29, 30, 31, 32]. So, unlike one electron in , the state of defected Ni(III) and Mn(III), defected Co(II) has two electrons in the states, with d-state configuration (Figure 11 B, C) [30].
We performed calculations to study the defects in a K-intercalated KCoO. First, we tried eV to see how performs without correction, and obtained a DOS as shown in Figure 9B. The blue plot, which is the projected density of states (PDOS) of d-states of the Co(II) ion, shows a significant change in the d-state of Co(II) compared to the d-state of Co(III) ion. Our calculation shows the magnetic moment of the Co(II) ion around 3, indicating three unpaired electrons in d-states. The presence of three unpaired electrons was previously seen in Refs. [29, 30, 31, 32]. We observed Co(II) d-state splitting, as shown in Figure 11B. This explains the appearance of Co(II) d-states just above the VB maximum in both channels, absence of Co(II) d-states in the CB in the up channel and presence in the down channel. There is a breaking of spin symmetry in the DOS, due to the appearance of a spin moment on the defected Co ion. The phenomena like shifting of d-states to higher energy and spin symmetry breaking in CB are similar to those observed and explained before in Peng et al.’s study [1] of birnessite.
Then we used eV in and obtained a DOS as shown in Figure 9C. We do not see much difference between the DOS for eV and eV. This indicates that the occupied Kohn-Sham orbitals are well localized and there is no fractional occupancy in calculations with eV. However, the larger appears to bring peaks of the defected Co(II) d-states closer together in the band gap region just above the VB maximum.
As discussed in the SM [26], we found the value for Co(II) to be 0.91 eV. We used eV and obtained a DOS as shown in Figure 9D, which looks similar to Figure 9B and Figure 9C.
We tried to compare a full HSE06+D3 DOS with for pristine KCoO and a K-intercalated KCoO. We obtained the KCoO supercell by inserting a K layer for each CoO layer. Compared to the pristine CoO supercell, which contains 96 ions, the KCoO supercell has 32 extra K ions, and the K-intercalated KCoO supercell has 33 extra K ions. Therefore, one would expect a different lattice parameter for KCoO and a K-intercalated KCoO, which requires the full relaxation of the supercell. We attempted to optimize the structure using the HSE06+D3 method, but it is very challenging for these systems in terms of time and resource. After a month, we chose not to proceed further. This challenge highlights one of the important reasons that motivated us to explore the method, highlighting the significance of the meta-GGA for defect studies.
However, we calculated the HSE06+D3 DOS using geometry. As shown in Figure 10A, HSE06+D3 predicts KCoO to be non-magnetic with an increased band gap of 4.17 eV. For a K-intercalated system with geometry, the HSE06+D3 DOS is shown in Figure 10 B, C, and D for eV, 1.8 eV, and 0.91 eV, respectively. Here we see that HSE06+D3 not only opens the gap between VB and CB, but it also brings the peaks of the defected

Figure 10. HSE06+D3 spin-resolved density of states per atom using SCAN+rVV10+U+U geometry of Figure 9 in (A) pristine KCoO with U=U=1.8 eV, and in a single K-intercalated KCoO with (B) U=U=0.0 eV, (C) U=U=1.8 eV, and (D) U=U=0.9 eV.
Co(II) d-states closer together in the band gap region just above the VB maximum. We observed Co(II) d-states splitting as shown in Figure 11C for the HSE06+D3 calculation using the SCAN+rVV10+U+U geometry with U=U=1.8 eV.
Figure 11 shows our electronic configurations for Co(III) and Co(II) in octahedral or nearly octahedral coordination with oxygen. For Co(II), we have seven d electrons. There are three orbitals of each spin, and two orbitals of each spin, and we can put no more than one electron into each spin orbital.
Our output tells us, from the relative energy order of each occupied d state of each spin, that the net spin is nearly that of three spin-up electrons, and that the lowest-energy orbitals are spin-up while the highest-occupied orbital is spin-down, but not whether a given spin orbital is or . So we begin by putting three spin-up electrons into orbitals, to minimize the electronic repulsion. We assume that the electrons prefer to be spin-up, to take advantage of an exchange interaction with the three up-spin t orbitals, and that spin-up e levels will compete in energy with the spin-down t levels. Then the configuration will be the one shown in Figure 11 B and C, (t ↑)(t ↓)(e ↑). This result agrees with the (t)(e) configuration found for Co(II) in LiCoO in Ref. [30]. This electronic configuration is also supported by the total charge density plots (see Fig. S5) of the four localized orbitals that appear in the band gap region of KCoO.
Regardless of coordination, Co(II) has been found to have a net spin nearly equal to that of three spin-up electrons in Refs. [29, 30, 31, 32], but the distribution over t and e of course depends upon coordination and other details of the nearby environment.
Fig. S4 B of the SM [26] shows the Jahn-Teller distorted tetrahedron around a defect Co(II) ion, with two of the Co(II)-O bonds about 10% longer than the remaining four. The angles are also distorted.
Conclusions
Defects in solids can be utilized to tune the electronic structure of materials, impacting applications in various fields. Defects can be investigated using first-principles calculations with DFT. However, DFAs introduce SIE that could vary depending on the element, system, site, and oxidation state, hindering the performance of DFAs.
Hybrid functionals mitigate SIE by mixing exact exchange, but at the expense of computational cost, successfully describing defects in many materials. The computational cost of hybrid functionals increases significantly for a SCF (self-consistent field) calculation compared to meta-GGAs like rSCAN. Defect studies require large supercells, making SCF calculations very expensive. Ionic relaxation calculations for defect studies involve many SCF cycles, making the process cumbersome for a hybrid functional to optimize the structure.
Here we demonstrate that the rSCAN+rVV10+U+U method can successfully predict the existence of localized polaronic states near defects, with literature values of U and U. The method is much faster than hybrid functionals but but still predicts localized polaronic states in band gaps. Using the rSCAN+rVV10+U+U geometry enables the completion of hybrid functional (or even G0W0) calculations in a reasonable time and with fewer resources. This approach makes hybrid functional calculations feasible for larger systems with transition-metal ions.
For fixed geometries, we have found the same number of occupied states localized on the defect transition-metal ion in the energy gap below the bottom of the conduction band, whether we use HSE06 ( or 0.25) or rSCAN+rVV10+U+U (in a range of U and U around the literature values [23]), for layered MnO (Figure 3–Figure 4), NiO (Figure 6–Figure 7), or KCoO (Figure 9–Figure 10). We have found the same whether we use the HSE06 or the rSCAN+rVV10+U+U geometries, for MnO (Figure 2 and Figure 4). (When HSE06 is used without a dispersion correction, the interlayer spacing must be constrained to a reasonable value.) Figure 11 shows the same density-of-states orbital ladder from both functionals in KCoO. This level of agreement is found, even though the fundamental band gaps are significantly smaller for rSCAN+rVV10+U+U than for HSE06. In the SM [26], we also find good agreement with HSE06+D3 for the magnetic moments of the transition-metal ions and for the lattice constants of the pristine materials. Jahn-Teller distortions of the metal-oxygen bond lengths at the defect are also shown in the SM [26]. The feasibility of other quantitative comparisons, including comparisons with experiment, will be considered in our future work.
This method can expedite the study of defects in materials. The rSCAN+rVV10+U+U method can also be employed in systems without defects where ions of the same species are in different OS.
We have studied the layered oxides MnO, NiO, and KCoO, both in the pristine state and with one additional K atom intercalated between layers in a supercell. Inexpensive rSCAN+rVV10+U+U equilibrium geometries have been used for electronic structure calculations with rSCAN+rVV10+U+U and with the expensive HSE06 hybrid functional. For K-intercalated MnO and NiO, we find no localized e state on the defected transition metal ion for U=U=0, but we find one such state for standard positive U values. This state, in the gap between conduction and valence bands, accepts the electron donated by the intercalated K. (For U=U=0, the extra electron from the intercalated K goes into the bottom of the conduction band, making the intercalated material a semi-metal. The extra electron is then delocalized over the supercell, with fractional occupation on each transition-metal ion.)
For K-intercalated KCoO, both for U=U=0 (standard semilocal rSCAN without a nonlocal +U self-interaction correction) and for standard positive U values, we seem to find two majority-spin and two minority-spin occupied localized e states in the gap between valence and conduction bands, and three empty localized minority-spin t states in the gap above the conduction band. This surprising result is consistent with a dramatic change in the electronic configuration (Figure 11) from the undefected Co(III) ions to the defected Co(II) ion, as reported previously in Ref. [29].
The rSCAN+rVV10+U+U approach is successful, but it has limitations. It requires material-dependent parameters U, and those are typically available for only a few oxidation states of each transition metal. For cases where a literature value of U is unavailable, we have suggested (in the SM [26]) a way to find a value by matching the HSE06 magnetic moment of the transition-metal ion (when it is non-zero). The HSE06 calculation could be made for a pristine material, avoiding large supercells. An interesting approach to U estimation, based on an ensemble of low- and high-spin states for a given transition metal ion and oxidation state, was suggested by Ref. [33].
Computational Details
First-principles calculations were performed with the projector-augmented wave method [34], implemented in the VASP code [35, 36], version 5.4.4.. A 4 x 4 x 1 supercell was used to simulate defects in layered TMOs. For all supercell calculations, a 2 x 2 x 2 -centered Monkhorst-Pack k mesh [37] was used.
For rSCAN+rVV10+U+U calculations, a 500 eV cutoff for the plane waves was used. In all rSCAN+rVV10+U+U calculations, the cell volume was relaxed with the ISIF=3 setting until forces converged to less than 0.03 eV/Å and energy converged to less than eV. To conduct rSCAN+rVV10+U+ calculations, we employed the simplified rotationally invariant framework developed by Dudarev et al. [38].
For Figure 2, we used an exact mixing parameter of 0.22 in the hybrid functional HSE06 [1, 6], as determined from the Generalized Koopman’s Condition (GKC) method [39, 40, 41, 42]. For all the remaining HSE06+D3 calculations, we used the standard exact exchange mixing parameter of 0.25. A plane-wave basis with an energy cutoff of 400 eV was employed.
Acknowledgments
This work was supported by the U.S. Department of Energy, Office of Science Basic Energy Sciences under Award No. DE-SC0023356. We thank NERSC for computational resources.
References
References
- [1]H. Peng, I. G. McKendry, R. Ding, A. C. Thenuwara, Q. Kang, S. L. Shumlas, D. R. Strongin, M. J. Zdilla, and J. P. Perdew, PNAS 114, 9523 (2017).
- [2]R. Ding, Y. Yasini, H. P. Peng, J. P. Perdew, E. Borguet, and M. J. Zdilla, Adv. Energy Mater. 11, 2101636 (2021).
- [3]P. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964).
- [4]W. Kohn and L. J. Sham, Phys. Rev. 140, A1133 (1965).
- [5]J. P. Perdew, R. G. Parr, M. Levy, and J. L. Balduz, Jr., Phys. Rev. Lett. 49, 1691 (1982).
- [6]A. V. Krukau, O. A. Vydrov, A. F. Izmaylov, and G. E. Scuseria, J Chem Phys 125, 224106 (2006).
- [7]J. Sun, R. C. Remsing, Y. Zhang, Z. Sun, A. Ruzsinszky, H. Peng, Z. Yang, A. Paul, U. Waghmare, X. Wu, M. L. Klein, and J. P. Perdew, Nature Chemistry 08, 831 (2016).
- [8]T. M. Henderson, J. Paier, , and G. E. Scuseria, Phys. Status Solidi B 248, 767 (2011).
- [9]J. W. Furness, A. D. Kaplan, J. Ning, J. P. Perdew, and J. Sun, J. Phys. Chem. Lett. 11, 8208 (2020).
- [10]A. P. Bartók and J. R. Yates, J. Chem. Phys. 150, 161101 (2019).
- [11]J. Sun, A. Ruzsinszky, and J. P. Perdew, Phys. Rev. Lett. 115, 036402 (2015).
- [12]H.-D. Sayníck and C. Cocchi, Electron. Struct. 3, 027001 (2021).
- [13]D. Wickramaratne and J. L. Lyons, ArXiv:2311.03634v1.arxiv.org/abs/2311.03634
- [14]A. Varadwaj and T. Miyake, ChemistrySelect 7, e202200171 (2022).
- [15]G. S. Gautam and E. A. Carter, Phys. Rev. Materials 2, 095401 (2018).
- [16]V. I. Anisimov, J. Zaanen, , and O. K. Andersen, Phys. Rev. B 44, 943 (1991).
- [17]V. I. Anisimov, I. V. Solovyev, M. A. Korotin, M. T. Czyzyk, , and G. A. Sawatzky, Phys. Rev. B 48, 16929 (1993).
- [18]I. V. Solovyev, P. H. Dedderichs, , and V. I. Anisimov, Phys. Rev. B 50, 16861 (1994).
- [19]M. Cococcioni and S. de Gironcoli, Phys. Rev. B 71, 035105 (2005).
- [20]O. A. Vydrov and T. V. Voorhis, J. Chem. Phys. 133, 244103 (2010).
- [21]R. Sabatini, T. Gorni, and S. de Gironcoli, Physical Review B 87, 041108 (2013).
- [22]O. Y. Long, G. S. Gautam, and E. A. Carter, Physical Review Material 4, 045401 (2020).
- [23]S. Swathilakshmi, R. Devi, and G. S. Gautam, J. Chem. Theory Comput. 19, 4202 (2023).
- [24]P. Atkins, T. Overton, J. Rourke, M. Weller, F. Armstrong, and S. . A. E. N. Brothers, A. F. Izmaylov, J. O. Normand, V. Barone, and G. E. Scuseria, THE JOURNAL OF CHEMICAL PHYSICS 129, 011102 (2008).
- [26]“See supplementary material for the following: determination of a U parameter from the HSE06 magnetic moment; effect of the D3 dispersion correction on lattice constants and density of states; jahn-teller symmetry breaking of the MO6 tetrahedron in a single k-intercalated MO2 (M=Mn, Ni) and in a single k-intercalated KMO2 (M=Co); and total charge density plots of four localized orbitals in the band gap region of K1.03CoO2.”
- [27]J. Jonas and S. Grimme, J. Phys. Chem. C 118, 7615 (2014).
- [28]J. M. P. Martirez and E. A. Carter, J. Am. Chem. Soc. 141, 693 (2019).
- [29]J. Chen, X. Wu, and A. Selloni, Phy. Rev. B 83, 245204 (2011).
- [30]G. Sun, F.-D. Yu, M. Lu, Q. Zhu, Y. Jiang, Y. Mao, J. A. McLeod, J. Maley, J. Wang, J. Zhou, and Z. Wang, Nature Communications 13, 6464 (2022).
- [31]P. C. Bunting, M. Atanasov, E. Damgaard-Møller, M. Perfetti, I. Crassee, M. Orlita, J. Overgaard, J. V. Slageren, F. Neese, and J. R. Long, Science 362, 1378 (2018).
- [32]W. J. Kim, M. A. Smeaton, C. Jia, B. H. Goodge, B. G. Cho, K. Lee, M. Osada, D. Jost, A. V. Levlev, B. Moritz, L. F. Kourkoutis, T. P. Devereaux, and H. Y. Hwang, Nature 615, 237 (2023).
- [33]A. Albavera-Mata, S. B. Trickey, and R. H. Hennig, J. Phys. Chem. Lett. 13, 12049 (2022).
- [34]P. E. Blöchl, Phys. Rev. B 50, 17953 (1994).
- [35]G. Kresse and J. Hafner, Phys. Rev. B Condens Matter 49, 14251–14269 (1994).
- [36]G. Kresse and D. Joubert, Phys. Rev. B 59, 1758–1775 (1999).
- [37]H. Monkhorst and J. Pack, Phys. Rev. B 13, 5188–5192 (1976).
- [38]S. L. Dudarev, G. A. Botton, S. Y. Savrasov, C. J. Humphreys, and A. P. Sutton, Phys. Rev. B 57, 1505 (1998).
- [39]S. Lany and A. Zunger, Phys. Rev. B 80, 085202 (2009).
- [40]S. Lany, Phys. Status Solidi B 248, 1052–1060 (2011).
- [41]H. Peng and S. Lany, Phys. Rev. B 85, 201202 (2012).
- [42]H. Peng, P. Ndione, D. Ginley, A. Zakutayev, and S. Lany, Phys. Rev. X 5, 021016 (2015).