r XXXX American Chemical Society A dx.doi.org/10.1021/cm200753g | Chem. Mater. XXXX, XXX, 000���000 ARTICLE pubs.acs.org/cm Evaluation of Tavorite-Structured Cathode Materials for Lithium-Ion Batteries Using High-Throughput Computing Tim Mueller, Geoffroy Hautier, Anubhav Jain, and Gerbrand Ceder* Massachusetts Institute of Technology, 77 Massachusetts Avenue 13-5051, Cambridge Massachusetts 02139, United States bSupporting S Information ��� INTRODUCTION In the search for better cathode materials for lithium-ion batteries, researchers have had considerable success devel- oping and optimizing materials with spinel,1 olivine,2 or layered3 structures. However to realize nonincremental im- provements in battery capacity, reliability, and safety it may be necessary to develop cathode materials with different crystal structures. An ideal cathode material should combine thermal stability, high voltage, and high lithium mobility and capacity, but it is di���cult to achieve these goals in one material. Materials containing polyatomic phosphate (PO4)3 anions tend to have higher thermal stability than oxides with comparable voltages,2,4 7 but these large and heavy anions adversely affect specific capacity. One way to compensate for this loss of capacity would be to develop a material that contains a polyatomic anion and is capable of reversibly inserting two lithium ions per redox-active metal ion. Recent studies indicate that materials with a struc- ture similar to LiFe(PO4)(OH) (tavorite) might be able to achieve this goal. Tavorite belongs to a class of materials with the general formula AM(TO4)X, where A is typically an alkali or alkaline- earth element, M is a metal, T is a p-block element, and X is O, OH, or F. The structure consists of vertex-linked one-dimen- sional (1D) chains of MO4X2 octahedra connected by TO4 tetrahedra. The X anions are located at the vertices shared by neighboring MO4X2 octahedra, and the A cations may be located at a number of sites throughout the framework (Figure 1). There are numerous minerals in this class, including LiAl(PO4)F (amblygonite) and CaTiO(SiO4) (titanite), but for clarity we will refer to materials in this class as tavorite- structured. Marx et al. have demonstrated reversible lithium insertion in LiFe(PO4)(OH) (tavorite) and tavorite-structured Fe(SO4)(H2O),8,9 and Reddy et al. have demonstrated rever- sible lithium insertion in Fe(SO4)(OH).10 One of the first tavorite-structured materials considered for lithium-ion bat- teries was R-LiVO(PO4), which was shown to have a capacity of 126 mAh/g at 3.8 V.11 Barker et al. demonstrated that the substitution of fluorine for oxygen to create tavorite-structured LiV(PO4)F increases the voltage to about 4.2 V and increases the rate capability of the material.12,13 Their work on a sym- metric LiV(PO4)F/LiV(PO4)F cell also demonstrated that the tavorite structure can cycle lithium between compositions M(TO4)X and Li2M(TO4)X.14 In addition, tavorite-structured LiV(PO4)F demonstrates exceptional thermal stability, exceed- ing that of olivine LiFe(PO4).15,16 Following up on a report of good lithium-ion conductivity in tavorite-structured LiMg- (SO4)F,17 Recham et al. recently demonstrated high-rate lithium insertion in tavorite-structured LiFe(SO4)F.18 In an- other paper, Recham et al. showed that it is possible to insert additional lithium into LiTi(PO4)F, taking advantage of the Ti2+/Ti3+ redox couple.19 Ramesh et al. demonstrated that LiFe(PO4)F can incorporate two lithium ions per redox-active metal, as evidenced by reversible cycling between LiFe(PO4)F and Li2Fe(PO4)F.20 These results suggest that given the right chemistry, it may be possible to find a tavorite-structured Received: March 15, 2011 Revised: August 3, 2011 ABSTRACT: Cathode materials with structure similar to the mineral tavorite have shown promise for use in lithium-ion batteries, but this class of materials is relatively unexplored. We use high-throughput density-functional-theory calculations to evaluate tavorite-structured oxyphosphates, fluorophosphates, oxysulfates, and fluorosulfates for use as cathode materials in lithium-ion batteries. For each material we consider the insertion of both one and two lithium ions per redox-active metal, calculating average voltages and stability relative to a database of nearly 100,000 previously calculated compounds. To evaluate lithium mobility, we calculate the activation energies for lithium diffusion through the known tavorite cathode materials LiVO(PO4), LiV(PO4)F, and LiFe(SO4)F. Our calculations indicate that tavorite-structured materials are capable of very high rates of one-dimensional lithium diffusion, and several tavorite-structured materials may be capable of reversibly inserting two lithium ions per redox-active metal. KEYWORDS: lithium-ion battery, cathode material, tavorite, density functional theory, high-throughput, computational

B dx.doi.org/10.1021/cm200753g |Chem. Mater. XXXX, XXX, 000���000 Chemistry of Materials ARTICLE electrode material with high capacity, excellent stability, and the ability to insert lithium at high rates. In this paper we use high-throughput computing to search for promising tavorite-structured cathode materials. We sub- stitute seventeen redox-active metals (Table 1) into tavorite- structured oxyphosphates, fluorophosphates, oxysulfates, and fluorosulfates, and for each host material we calculate the average voltages for insertion of both one and two lithium ions per metal ion. To screen out compounds that are unlikely to be su���ciently stable for use in batteries, we evaluate the stability of each candidate material against a database of nearly 100,000 materials for which we have calculated energies.21 To better understand the rate capabilities of tavorite-structured materials, we model lithium diffusion through three well-known tavorite- structured compounds: VO(PO4), V(PO4)F, and Fe(SO4)F. Our results highlight several promising chemistries, including a number that may be capable of inserting two lithium ions per redox-active metal. ��� METHODS The average voltage relative to lithium metal for lithium insertion into a host material M is given by V��x�� �� G��M�� �� xG��Li�� G��LixM�� xz ��1�� where V is the voltage, G( ) is the Gibbs free energy per formula unit, and z is the elementary charge per lithium ion (z = 1). Because the contribution of pressure and entropy to the free-energy differences is expected to be relatively small,22 eq 1 is well-approximated by consider- ing only energies: V��x�� �� E��M�� �� xE��Li�� E��LixM�� xz ��2�� where E( ) is the energy per formula unit. To calculate the energies in eq 2, we use spin-polarized density functional theory23 (DFT) with the Perdew Burke Ernzerhof (PBE) exchange-correlation functional24 as implemented in the Vienna Ab-initio Simulation Package (VASP).25 To calculate accurate voltages for transition metal oxides,26 we use the generalized gradient approximation with Hubbard U corrections (GGA+U) in the rotationally invariant form proposed by Dudarev et al.27 The (U J) parameters, provided in Table 1, were fit to empirical oxidation energies as described in ref 28. The only exception was cobalt, for which we found that a (U J) value of 5.7, similar to the values found by Zhou et al.,26 produces better results. A k-point density of 500 k- points per reciprocal atom was used for all calculations. All cells were allowed to fully relax, and precision for all calculations was set to ���high���, which increases the plane-wave cutoff energy by 25% above the VASP default. The electronic and ionic minimization convergence criteria were set to the VASP defaults of 1 meV per unit cell and 10 meV per unit cell, respectively. Initial magnetic moments were set to high-spin (MAGMOM = 5) for Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Nb, Mo, Ag, Ta, and W, and low-spin (MAGMOM = 0.6) for the other elements. It was found that cobalt frequently does not relax to a low-spin ground state when initialized in a high-spin state, even if the low-spin state has lower energy. For this reason all Co-containing calculations were run twice, initialized in both high-spin and low-spin states, and the calculation resulting in the lowest energy was used. All voltage calculations were run both with both ferromagnetic and antiferromagnetic spin initialization, Figure 1. 2 2 2 supercell of a typical tavorite-structured material, LiFe(SO4)F. Structural data is from ref 18. Brown octahedra represent Fe, yellow tetrahedra represent S, red spheres represent O, and blue spheres represent F. The green-and-white spheres represent partially occupied lithium sites, with the occupancy given by the fraction of the sphere shaded green. The three views are (a) along the a axis, (b) along the b axis, and (c) along the c axis. Table 1. Redox-Active Metals Considered and the (U-J) Parameters Assigned to Them for GGA+U Calculations metal (U-J) value Ag 1.5 Bi 0.0 Co 5.7 Cr 3.5 Cu 4.0 Fe 4.0 Mn 3.9 Mo 3.5 Nb 1.5 Ni 6.0 Pb 0.0 Sb 0.0 Sn 0.0 Ta 2.0 Ti 0.0 V 3.1 W 4.0