betaFeedback

WorksPhysics and Astronomy

Thermodynamic accessibility of Li-Mn-Ti-O cation disordered rock-salt phases

Abstract

Disordered rock-salt with Li-excess (DRX) cathode phases within the Li-Mn-Ti-O (LMTO) composition space have recently been extensively studied, as they promise to deliver exceptional energy density at low cost in Li-ion batteries. The continued development of LMTO DRX with improved power density and cycling stability requires optimization of the composition and particle size/morphology, which are determined by synthesis conditions such as annealing temperatures and hold times. These challenges motivate our investigation of the phase diagram of the LMTO rock-salt phase space, with a focus on understanding the stability of DRX by quantifying the order-disorder transition temperature (TdisordT_\text{disord}) as a function of composition. We harness first-principles calculations and X-ray diffraction experiments to establish the LMTO phase diagram, which lies within the LiMnO2_2 – Li2_2MnO3_3 – Li2_2TiO3_3 pseudo-ternary. Our calculations predict that the LMTO phase diagram at elevated temperature (700−1300700 - 1300 C) is composed of three phases: DRX, orthorhombic LiMnO2_2, and layered Li2_2Mn1-y_\text{1-y}Tiy_\text{y}O3_3 (0<y<10 < \text{y} < 1). TdisordT_\text{disord} decreases significantly as off-stoichiometry is introduced to the end-point compositions, resulting in a eutectoid phase diagram. Importantly, a significant range of LMTO compositions containing small to moderate fractions of Li-excess and Ti doping (relative to LiMnO2_2) have TdisordT_\text{disord} spanning 700−900700 - 900 C. These temperatures are substantially lower than conventional DRX synthesis temperatures (≥1000\geq 1000 C), suggesting the promise of decreasing synthesis temperatures for specific DRX compositions. The compositions containing moderate to high fractions of Mn4+^{4+} instead have much greater TdisordT_\text{disord} and phase separation to layered Li2_2MnO3_3 becomes highly favored.

Introduction

Lithium-ion (Li) batteries have been instrumental towards the rapid growth in electric vehicles, grid-scale energy storage, and personal electronic devices [1, 2, 3]. The projected rise in Li battery demand in the next decades necessitates the continued development of battery chemistries that utilize inexpensive and earth-abundant elements [3, 4]. Cation disordered rock-salt with Li-excess (DRX) cathode materials are a particularly promising class of Li cathode materials that utilize earth abundant elements and can deliver high energy density that rivals state-of-the-art Ni-Mn-Co (NMC) cathodes [5, 6] [7, 8, 9, 10]. The Li-Mn-Ti-O (LMTO) composition space has been a fruitful source of candidate DRX compositions, such as Li1.2_{1.2}Mn0.4_{0.4}Ti0.4_{0.4}O2_2 and Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2 [5, 6] [8, 9, 11, 12, 13, 14, 15]. More recently, it has been found that Mn-rich LMTO compositions such as Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2 are especially promising, as they tend to transform to δ-DRX phases with anti-phased nanoscale spinel domains upon electrochemical cycling or chemical delithiation, leading to improved power density [5] [14, 16, 17, 18]. DRX compositions with high Mn fraction are typically prepared using conventional solid-state synthesis at relatively high temperatures ≥ 1000 °C [6] [7, 8, 15]. Recently, Patil et al. showed that LMTO DRX compositions containing relatively higher fractions of Li and Ti, such as Li1.2_{1.2}Mn0.4_{0.4}Ti0.4_{0.4}O2_2, can be prepared at lower temperatures ∼ 800 °C through sol-gel synthesis 15. Other strategies such as molten-salt 19 and microwave synthesis 20 have also been developed. The synthesis conditions have a strong impact on particle size and short-range order, which in turn affect the electrochemical kinetics of (δ-)DRX [5] [17]. For example, at high temperature the particles can grow very quickly and inhomogeneously, leading to poor electrochemical kinetics 17. Thus, it is favorable to lower DRX synthesis temperatures, as well as optimize other synthesis conditions such as the holding times and heating/cooling rates. It becomes vital then to understand the minimum temperatures at which DRX becomes thermodynamically stabilized (i.e. the order-disorder transition temperature — which we will refer to as TdisordT_{\mathrm{disord}}), and assess the possible composition dependence of TdisordT_{\mathrm{disord}}.

Since DRX is a disordered phase, it is expected to be stabilized at elevated temperature by its increased configurational entropy (SconfigS_{\mathrm{config}}). Meanwhile, the enthalpic (HH) contribution to the free energy is more favorable for the ordered ground states. The balance between SconfigS_{\mathrm{config}} and HH contributions determines the TdisordT_{\mathrm{disord}} and stability of DRX relative to the ordered phases. The SconfigS_{\mathrm{config}} can be significantly influenced by the number of distinct species and their fractions present within a given composition, as well as the tendency of certain species to form short-range order. The enthalpy of DRX can be significantly lowered by adding ions with the d0d^{0} electron configuration (such as Ti4+\mathrm{Ti}^{4+}), since the d0d^{0} electronic structure can accommodate local lattice distortions at minimal energy cost, as previously shown by Urban et al. 21. The objective of our study is to derive a phase diagram of LMTO rock-salt-type phases, using a combination of ab initio thermodynamics and experiments. As shown in Figure 1a, we span the pseudo-ternary of LiMnO2\mathrm{LiMnO}_{2} – Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} – Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} compositions. Previous experiments have shown that the ordered phases that are stable at ambient temperature are the orthorhombic LiMnO2\mathrm{LiMnO}_{2} phase 22–24, layered Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} 25,26, and layered Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} 27–29. The orthorhombic LiMnO2\mathrm{LiMnO}_{2} (Pmmn) structure exhibits corrugated layers of Mn or Li that alternate with each other, which we note is unique because other analogous LiMO2\mathrm{LiMO}_{2} compositions (where MM can be a single species or an admixture of species) typically crystallize in a layered R3ˉm\mathrm{R}\bar{3}m structure 30. The unique stability of the orthorhombic Pmmn structure has been attributed to the cooperative Jahn-Teller (JT) distortion of Mn3+\mathrm{Mn}^{3+} ions, which lowers the total energy by breaking the degeneracy of the single half-filled ege_{g} states of Mn3+\mathrm{Mn}^{3+} ions. In layered Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3}, the Mn are nominally in the +4 oxidation state with unfilled ege_{g} states, so they do not exhibit a JT distortion. The layered Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} and Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} phases are nearly isostructural in that they consist of alternating planes of Li and D2D_{2}Li (where DD can either be Mn or Ti), with the DD atoms arranged in honeycombs surrounding each Li. Layered Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} has been reported to crystallize with the C2/c space group, while Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} has been reported to have either C2/c or C2/m symmetry 25,26,31, with these differences in symmetry arising from distinct stacking patterns of the honeycomb-ordered D2D_{2}Li layers 26. Within our first-principles calculations, we find that the C2/m and C2/c stacking patterns yield nearly identical total energies

Figure 1. Composition and configuration space of the LMTO rock-salt phases. a) Pseudo-ternary LiMnO2\mathrm{LiMnO}_{2} – Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} – Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} composition space. Nominal TM oxidation states and changes in composition along the edge pseudo-binary lines are labeled. b) Unit cell of the FCC rock-salt lattice, with octahedral cation sites that can be occupied by Li+\mathrm{Li}^{+}, Mn3+\mathrm{Mn}^{3+}, Mn4+\mathrm{Mn}^{4+}, or Ti4+\mathrm{Ti}^{4+}.

(within 0.6 meV/atom) for Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} and Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3}, consistent with previous work on Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} 32.

Since the nominal oxidation states of Mn in LiMnO2\mathrm{LiMnO}_{2} and Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} are +3 and +4, respectively, the compositions along the LiMnO2\mathrm{LiMnO}_{2} – Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} pseudo-binary (Figure 1a) are expected to contain Mn3+\mathrm{Mn}^{3+}, Mn4+\mathrm{Mn}^{4+}, and excess Li (relative to LiMnO2\mathrm{LiMnO}_{2}) in ratios that preserve charge neutrality, as described by the expression Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{2x}\mathrm{O}_{2}, where xx is the fraction of excess Li (per O2\mathrm{O}_{2}) relative to LiMnO2\mathrm{LiMnO}_{2} and spans the range 0≤x≤1/30 \le x \le1/3. Compositions along the LiMnO2\mathrm{LiMnO}_{2} – Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} pseudo-binary can be described by an analogous expression Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Ti}^{4+}_{2x}\mathrm{O}_{2}. The third axis of the LMTO pseudo-ternary is the replacement of Mn4+\mathrm{Mn}^{4+} with Ti4+\mathrm{Ti}^{4+} (Li2Mn1−y4+Tiy4+O3\mathrm{Li}_{2}\mathrm{Mn}^{4+}_{1-y}\mathrm{Ti}^{4+}_{y}\mathrm{O}_{3} – Figure 1a). Considering all compositional constraints, the LMTO rock-salt compositions can be described by the expression Li1+xMn1−x−yTiyO2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-x-y}\mathrm{Ti}_{y}\mathrm{O}_{2}, where 0≤x≤130 \le x \le\frac{1}{3} and 0≤y≤2x0 \le y \le2x. All possible structures within the LMTO rock-salt space represent distinct cation arrangements (Li+\mathrm{Li}^{+}, Mn3+\mathrm{Mn}^{3+}, Mn4+\mathrm{Mn}^{4+}, or Ti4+\mathrm{Ti}^{4+}) on a face-centered-cubic (FCC) lattice of octahedral sites. The unit cell of the rock-salt framework is shown in Figure 1b.

The most important aspect of our study is to understand the stability of the DRX phase by mapping TdisordT_{\mathrm{disord}} as a function of composition. From first-principles modeling, we account for the electronic and configurational free energy contributions, by harnessing the cluster expansion (CE) and Monte Carlo (MC) sampling approach 33–35, which have been extensively applied to study the thermodynamics of a wide range of disordered materials systems spanning battery cathodes 36–38, alloys 39,40, magnetic materials 41,42, and more. We train a CE model to describe the energetics of the quaternary configuration space of cation sites (Figure 1b). We distinguish between the +3 and +4 states of Mn as it has been shown by Zhou et al. that the "configurational electronic" entropy of a species in different oxidation states can be critical for stabilizing solid-solution phases in ionic materials such as LixFe1−x2+Fex3+PO4\mathrm{Li}_{x}\mathrm{Fe}^{2+}_{1-x}\mathrm{Fe}^{3+}_{x}\mathrm{PO}_{4} 43.

To accurately compute the LMTO phase diagram, the CE model must be trained on density-functional theory (DFT)-computed total energies that realistically describe the energetics of the possible structures — and most crucially, reproduce the correct ground states. In a recent study, we have shown that commonly utilized DFT approaches such as GGA (PBEsol44) and meta-GGA (r²SCAN45) functionals, with or without empirical Hubbard UU corrections46, spuriously predict the ground state of LiMnO2_2 to be the γ\gamma-LiMnO2_2 polymorph (isostructural to γ\gamma-LiFeO2_247), which has never been reported in experiment48. Meanwhile, the HSE06 hybrid-GGA functional49 can correctly predict the orthorhombic phase to be the LiMnO2_2 ground state. Thus, we use HSE06 to calculate the total energy of each structure to train our CE model (more details in § 3.1). We note that HSE06 requires significantly more computational resources compared to GGA and meta-GGA functionals, which is the main reason why most previous studies involving DFT and CE calculations of LMTO phases have employed GGA or meta-GGA-based techniques36–38. The failures of the PBEsol and r²SCAN-based methods to predict the correct LiMnO2_2 ground state properties have been attributed to the complex interplay between JT distortions, electronic localization, and antiferromagnetism of Mn3+^{3+} ions, which the HSE06 functional can capture with better accuracy due to the more precise treatment of screened electronic exchange interactions within the hybrid-GGA formalism48. This helps reduce self-interaction and delocalization errors within DFT that can be large when treating correlated 3dd electron states46,49,50. Furthermore, accounting for AFM order was shown in previous studies to significantly lower the energy of each LiMnO2_2 phase compared to the ferromagnetic (FM) state and improve LiMnO2_2 phase stability predictions48,51, so we also search for the lowest energy AFM ordering for each ion configuration within PBEsol+UU, using a previously developed DFT workflow52,53.

By calculating the energies of each LMTO structure using the HSE06 functional (more details in § 3.1), we can reasonably capture the complex electronic structure of the LMTO phases (especially near the LiMnO2_2 composition) and curate a DFT data set that contains the physically correct ground states. A CE model is parameterized on this DFT data, which recovers the known LMTO ground states. The CE is then harnessed to perform MC sampling and thermodynamic integration to compute the free energies of the competing ordered and disordered LMTO phases, and determine the phase boundaries between them (more details in Methods).

We perform a complementary set of experiments utilizing heating and in situ X-ray diffraction (XRD) measurements within an inert environment to track the phase evolution as a function of temperature for specific LMTO compositions along the LiMnO2_2 – Li2_2TiO3_3 and LiMnO2_2 – Li2_2MnO3_3 pseudo-binary composition axes (Figure 1a). For each composition, we quantify the temperature at which DRX becomes phase-pure (which we define as TdisordT_{\mathrm{disord}}) and identify the ordered phases that form below TdisordT_{\mathrm{disord}}. The experiments are performed under an inert environment to prevent oxidation of the samples, which can lead to the formation of undesired phases such as spinel LiMn2_2O4_454,55. A similar heating and in situ XRD set-up has recently been used by Liang et al. to rationalize the reaction mechanisms leading to DRX formation in the fluorinated Li1.1_{1.1}Mn0.8_{0.8}Ti0.1_{0.1}O1.9_{1.9}F0.1_{0.1} composition56. We also perform quenching of select compositions from temperatures near TdisordT_{\mathrm{disord}}.

We identify the LMTO phase diagram to be a eutectoid-like phase diagram, in which the TdisordT_{\mathrm{disord}} decreases upon introducing off-stoichiometry to the end-point phases (orthorhombic LiMnO2_2, layered Li2_2TiO3_3, and layered Li2_2MnO3_3). Importantly, introducing small to moderate amounts of Li-excess, Ti4+^{4+}, and/or Mn4+^{4+} (relative to LiMnO2_2) can lead to TdisordT_{\mathrm{disord}} values that are significantly lower than typically used DRX synthesis temperatures (≥1000 ∘C\geq1000\,^{\circ}\mathrm{C}). In general, the results of our calculations agree well with experiments. However, in experiments on compositions with a small amount of Li excess compensated by Mn4+^{4+}, we observe a layered LiMnO2_2-like phase to form. Our calculations do not predict this phase to be thermodynamically stable, suggesting it is a metastable phase that can form more rapidly than the thermodynamically predicted phase separation upon cooling.

Results

To construct our DFT dataset of LMTO rock-salt structures, we enumerate structures across the pseudo-ternary LiMnO2_2 – Li2_2MnO3_3 – Li2_2TiO3_3 composition space, with the sampled compositions shown in Figure 2a. Many of these structures were generated by introducing off-stoichiometry and/or anti-site defects to the ordered LMTO ground states (orthorhombic LiMnO2_2, layered Li2_2MnO3_3, layered Li2_2TiO3_3), as well as other low-energy ordered LiMnO2_2 phases such as layered, spinel, and γ\gamma-LiMnO2_2. Disordered configurations are also sampled from MC simulations at elevated TT across various LMTO compositions. An internal database of LMTO rock-salt structures compiled in previous studies is also used36–38. In all, we gather 745 unique cation configurations in the LMTO rock-salt lattice and calculate their formation energies at T=0T = 0 K using the HSE06 hybrid-GGA functional. The compositions and energies are shown as color-coded squares in Figure 2a. With the exception of the end-point ground state phases, we do not identify any additional LMTO structures that are predicted to be stable at T=0T = 0 K from DFT, as all enumerated structures have positive energies above the hull (EhullE_{\mathrm{hull}}).

DFT-computed formation energies and Mn magnetic moments of LMTO rock-salt orderings

Figure 2. DFT-computed formation energies at 0 K and Mn magnetic moments of the training set of LMTO rock-salt orderings. a) Formation energies of structures within the LMTO pseudo-ternary — specifically the lowest energies of each sampled composition. Green circles represent stable phases and squares represent configurations with energy above the convex hull. b) Histogram of absolute values of Mn magnetic moments (∣mMn∣|m_{\mathrm{Mn}}|). The vertical line at ∣mMn∣=3.35 μB|m_{\mathrm{Mn}}| = 3.35\ \mu_{\mathrm{B}} denotes the threshold between Mn3+^{3+} and Mn4+^{4+}, with all Mn with ∣mMn∣>3.35 μB|m_{\mathrm{Mn}}| > 3.35\ \mu_{\mathrm{B}} assigned to be Mn3+^{3+}, while Mn with ∣mMn∣<3.35|m_{\mathrm{Mn}}| < 3.35 assigned as Mn4+^{4+}. Formation energies along the pseudo-binary c) LiMnO2_2 – Li2_2TiO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2) and d) LiMnO2_2 – Li2_2MnO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2) composition lines, where green squares are stable phases and orange lines are configurations above the convex hull.

To provide a more quantitative visualization of the energy landscape, we plot the formation energies along the pseudo-binary LiMnO2_2 – Li2_2TiO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2) and LiMnO2_2 – Li2_2MnO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2) axes in Figure 2c and Figure 2d, respectively. The Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 structures can have significantly lower values of EhullE_{\mathrm{hull}} compared to the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2 space. For example, there are many Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 structures with Ehull≤10E_{\mathrm{hull}} \leq10 meV/atom, while there are no Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2 structures with Ehull<10E_{\mathrm{hull}} < 10 meV/atom. The structures with lowest EhullE_{\mathrm{hull}} are the layered LiMnO2_2, orthorhombic LiMnO2_2, γ\gamma-LiMnO2_2, spinel LiMnO2_2, and layered Li2_2MnO3_3/Li2_2TiO3_3 phases with off-stoichiometry.

We assign oxidation states to the Mn atoms in each LMTO structure based on the absolute values of the Mn magnetic moments (∣mMn∣|m_{\mathrm{Mn}}|) computed within HSE06, a histogram of which is shown in Figure 2b. We find a clearly bimodal distribution of ∣mMn∣|m_{\mathrm{Mn}}| with peaks centered at ∣mMn∣=3.0|m_{\mathrm{Mn}}| = 3.0 and 3.7 μB3.7\ \mu_{\mathrm{B}}. By assigning all higher spin Mn with ∣mMn∣>3.35 μB|m_{\mathrm{Mn}}| > 3.35\ \mu_{\mathrm{B}} to be Mn3+^{3+} and Mn with ∣mMn∣≤3.35 μB|m_{\mathrm{Mn}}| \le3.35\ \mu_{\mathrm{B}} to be Mn4+^{4+}, the number of structures that are formally charge-neutral is maximized36. The average ∣mMn∣|m_{\mathrm{Mn}}| of Mn3+^{3+} and Mn4+^{4+} that we calculate within the HSE06 functional are in reasonable agreement with previous studies 33,36 that utilized meta-GGA (SCAN or r2^{2}SCAN) functionals, though our ∣mMn∣|m_{\mathrm{Mn}}| values computed within HSE06 are systematically slightly higher and the distribution more discrete (less continuous). These differences in our calculated ∣mMn∣|m_{\mathrm{Mn}}| compared to previous studies can be attributed to the reduced self-interaction error (SIE) of the HSE06 hybrid-GGA functional, which helps enforce greater electron localization onto the Mn-3d states compared to meta-GGA methods. The calculated magnetic moments of all Ti atoms are

near zero (within 0.01 μB0.01\ \mu_{\mathrm{B}}), which confirms that all Ti are in the +4 oxidation state.

To understand the phase stability at elevated temperature, we parametrize a CE on the DFT energies of the LMTO rock-salt structures and perform MC simulations and thermodynamic integration to map the phase diagram of this system. Our trained CE model correctly recovers the DFT-predicted ground states, as shown in Supplementary Information (SI) Figure S1. We specifically analyze the pseudo-binary spaces of LiMnO2_2 – Li2_2TiO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2), LiMnO2_2 – Li2_2MnO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2), and LiMnO2_2 – Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2). By examining these selected pseudo-binary systems, we can quantitatively analyze the influence of Li-excess (xx) and type/concentration of the +4 ion (which we will refer to as D4+D^{4+}, where DD can either be Mn or Ti) towards stabilizing the DRX phase. We calculate the temperature – composition phase diagrams of Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_{2}, Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2}, and Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{x}^{4+}\mathrm{Ti}_{x}^{4+}\mathrm{O}_{2}, and plot them along with the TdisordT_{\mathrm{disord}} determined from in situ XRD experiments (gold stars) in Figure 3. The in situ XRD heatmaps of each composition are shown in SI Figure S2.

Figure 3. Temperature – composition phase diagrams of the pseudo-binary composition spaces of (a) LiMnO2\mathrm{LiMnO}_{2} – Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3}, (b) LiMnO2\mathrm{LiMnO}_{2} – Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3}, and (c) LiMnO2\mathrm{LiMnO}_{2} – Li2Mn0.5Ti0.5O3\mathrm{Li}_{2}\mathrm{Mn}_{0.5}\mathrm{Ti}_{0.5}\mathrm{O}_{3}, calculated from MC simulations (black lines). Gold stars represent TdisordT_{\mathrm{disord}} determined from in situ XRD experiments, which were performed for select compositions along the LiMnO2\mathrm{LiMnO}_{2} – Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} and LiMnO2\mathrm{LiMnO}_{2} – Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} pseudo-binaries only.

CE-computed formation energies of LMTO rock-salt orderings

Figure S1. CE-computed formation energies at 0 K of the training set of LMTO rock-salt orderings. a) Formation energies of structures within the LMTO pseudo-ternary — specifically the lowest energies of each sampled composition. b) LiMnO2_2 – Li2_2TiO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2) and c) LiMnO2_2 – Li2_2MnO3_3 (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2) composition lines.

In-situ XRD waterfall plots for various DRX compositions, panels (j), (k), and (l)

Figure S2. In-situ XRD waterfall plots of various DRX compositions along the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Ti}^{4+}_{2x}\mathrm{O}_2 and Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{2x}\mathrm{O}_2 pseudo-binary spaces.

Within the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_{2} pseudo-binary phase diagram (Figure 3a), the calculated TdisordT_{\mathrm{disord}} of LiMnO2\mathrm{LiMnO}_{2} is 1200 °C, which is in reasonable agreement with experiment (1070 °C). As the Li and Ti concentration increases, TdisordT_{\mathrm{disord}} linearly decreases until a predicted eutectoid point at Teutec=700T_{\mathrm{eutec}} = 700 °C and xeutec=0.14x_{\mathrm{eutec}} = 0.14 (Li1.14Mn0.58Ti0.28O2\mathrm{Li}_{1.14}\mathrm{Mn}_{0.58}\mathrm{Ti}_{0.28}\mathrm{O}_{2}). The experimental DRX phase boundary in this composition range is also linearly sloped, though the decrease of TdisordT_{\mathrm{disord}} within experiment is less pronounced compared to the simulations. The experimental eutectoid point is projected to be near the temperature Teutec,exp=680T_{\mathrm{eutec, exp}} = 680 °C and composition xeutec,exp∼0.2−0.25x_{\mathrm{eutec, exp}} \sim0.2 - 0.25 (Li1.2Mn0.4Ti0.4O2\mathrm{Li}_{1.2}\mathrm{Mn}_{0.4}\mathrm{Ti}_{0.4}\mathrm{O}_{2} to Li1.25Mn0.25Ti0.5O2\mathrm{Li}_{1.25}\mathrm{Mn}_{0.25}\mathrm{Ti}_{0.5}\mathrm{O}_{2}). Thus, the experimental and calculated eutectoid temperatures are in reasonable agreement, while the experimental eutectoid composition is higher in Li and Ti concentration than the calculated value. Adding more Li and Ti past xeutecx_{\mathrm{eutec}} leads to a linear increase in the calculated TdisordT_{\mathrm{disord}} until the Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} end-point (Figure 3a). The experiments show an increase in TdisordT_{\mathrm{disord}} at x>xeutec,expx > x_{\mathrm{eutec, exp}} as well. Specifically, the experimentally measured TdisordT_{\mathrm{disord}} of Li1.3Mn0.1Ti0.6O2\mathrm{Li}_{1.3}\mathrm{Mn}_{0.1}\mathrm{Ti}_{0.6}\mathrm{O}_{2} is 60 °C higher than that of Li1.25Mn0.25Ti0.5O2\mathrm{Li}_{1.25}\mathrm{Mn}_{0.25}\mathrm{Ti}_{0.5}\mathrm{O}_{2}. The increase of TdisordT_{\mathrm{disord}} at high Ti content is consistent with the fact that the reported TdisordT_{\mathrm{disord}} values of Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} range from 1155 – 1200 °C28,29, which is higher than our experimentally obtained Teutec,expT_{\mathrm{eutec, exp}} (680 °C).

To understand the differences in phase stability when Ti4+\mathrm{Ti}^{4+} is replaced with Mn4+\mathrm{Mn}^{4+}, we examine the Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2} pseudo-binary, as shown in Figure 3b. The trends predicted by theory near the LiMnO2\mathrm{LiMnO}_{2} composition, specifically within the composition range 0<x<0.10 < x < 0.1, are similar to those observed for Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_{2} (Figure 3a) in which TdisordT_{\mathrm{disord}} decreases as a function of xx. The predicted eutectoid point of Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2} is at Teutec=970T_{\mathrm{eutec}} = 970 °C and xeutec=0.08x_{\mathrm{eutec}} = 0.08 (Li1.08Mn0.92O2\mathrm{Li}_{1.08}\mathrm{Mn}_{0.92}\mathrm{O}_{2} or Li1.08Mn0.763+Mn0.164+O2\mathrm{Li}_{1.08}\mathrm{Mn}_{0.76}^{3+}\mathrm{Mn}_{0.16}^{4+}\mathrm{O}_{2} – Figure 3b), which is at significantly higher TeutecT_{\mathrm{eutec}} and lower xeutecx_{\mathrm{eutec}} than for the Ti4+\mathrm{Ti}^{4+} system. At x>0.08x > 0.08, the TdisordT_{\mathrm{disord}} rapidly increases as a function of xx, leading to a prominent DRX – layered Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} two-phase region. The dominant factor leading to this wider two-phase region is the high predicted TdisordT_{\mathrm{disord}} of Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} (>2800> 2800 °C), which is significantly higher than the simulated TdisordT_{\mathrm{disord}} of Li2TiO3\mathrm{Li}_{2}\mathrm{TiO}_{3} (1000 °C). Thus, the layered Li2MnO3\mathrm{Li}_{2}\mathrm{MnO}_{3} phase is predicted to be very stable and competes strongly with DRX at typical DRX synthesis temperature ranges (∼1000\sim1000 °C), significantly reducing the range of Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2} DRX compositions that are accessible at reasonable temperatures.

Our experimentally determined TdisordT_{\mathrm{disord}} of Li1.05Mn0.95O2\mathrm{Li}_{1.05}\mathrm{Mn}_{0.95}\mathrm{O}_{2} and Li1.1Mn0.9O2\mathrm{Li}_{1.1}\mathrm{Mn}_{0.9}\mathrm{O}_{2} are nearly identical at 1100 °C, in excellent agreement (within ±50\pm50 °C) with the computed DRX phase boundary (Figure 3b). We have performed in situ XRD experiments at higher Li and Mn4+\mathrm{Mn}^{4+} concentration, specifically at x=0.15x = 0.15, 0.2,

0.25, and 0.3, from which we could not determine TdisordT_{\mathrm{disord}} because the temperature at which DRX becomes phase-pure could not be reached due to our furnace’s maximum temperature of 1100 °C (SI Figure S2), indicating that the TdisordT_{\mathrm{disord}} exceeds this temperature, which is consistent with Figure 3b. Thus, although we could not experimentally quantify TdisordT_{\mathrm{disord}} at these compositions, we confirm that these TdisordT_{\mathrm{disord}} are very high and thus validate our prediction of a wide DRX – layered Li2_2MnO3_3 miscibility gap.

To understand the effect of combining Mn4+^{4+} and Ti4+^{4+} on phase stability, we calculate the pseudo-binary phase diagram between LiMnO2_2 – Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3 which contains an equal fraction of Mn4+^{4+} and Ti4+^{4+} (Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2), and is shown in Figure 3c. Within Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2, there is a rapid decrease in TdisordT_{\mathrm{disord}} at low xx until the eutectoid point of Teutec∼770T_{\mathrm{eutec}} \sim770 °C and xeutec∼0.11x_{\mathrm{eutec}} \sim0.11 (Li1.11_{1.11}Mn0.78_{0.78}Ti0.11_{0.11}O2_2), which is comparable to the eutectoid point in the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 system (Figure 3a). In fact, the DRX phase boundary in the region of x<0.11x < 0.11 is nearly identical to the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 system (Figure 3a), which is consistent with a purely entropy-driven increase in the stability of the DRX phase. At higher xx, there is a prominent two-phase region between DRX and layered Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3, though this region is significantly smaller than the one between DRX and layered Li2_2MnO3_3 (shown in Figure 3b). Thus, the partial substitution of Mn4+^{4+} with Ti4+^{4+} increases the accessible composition range of DRX at 900<T<1200900 < T < 1200 °C. We predict the

TdisordT_{\mathrm{disord}} of Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3 to be quite high (∼1900\sim1900 °C), which rationalizes why the two-phase region between DRX and layered Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3 is still relatively large. An important contribution to the stability of layered Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3 is the configurational entropy (SconfigS_{\mathrm{config}}) of mixing Mn4+^{4+} and Ti4+^{4+} within this phase, which significantly lowers the free energy (SI Figure S3). Meanwhile, the SconfigS_{\mathrm{config}} in layered Li2_2MnO3_3 and Li2_2TiO3_3 is found to be negligible up to their transition temperature, since they exhibit a first-order transition to the disordered state (SI Figure S4).

To experimentally investigate the miscibility regions of the LMTO phase diagram, we perform ex situ XRD characterization of several samples quenched from above and below the respective TdisordT_{\mathrm{disord}}. Through this analysis, we also assess the practical feasibility of quenching phase-pure DRX samples near the identified TdisordT_{\mathrm{disord}}. The ex situ XRD patterns of quenched Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2, Li1.1_{1.1}Mn0.7_{0.7}Ti0.2_{0.2}O2_2, Li1.25_{1.25}Mn0.25_{0.25}Ti0.5_{0.5}O2_2, and Li1.3_{1.3}Mn0.1_{0.1}Ti0.6_{0.6}O2_2 samples are shown in Figure 4. For the Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2 (Figure 4a) and Li1.1_{1.1}Mn0.7_{0.7}Ti0.2_{0.2}O2_2 (Figure 4b) compositions, the XRD patterns of samples quenched from 1000 °C and 900 °C, respectively, display only the peaks of a DRX reference, which confirms that phase-pure DRX can be isolated slightly above TdisordT_{\mathrm{disord}}. Refinement results show that the DRX lattice parameters are 4.17 and 4.16 for Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2 and Li1.1_{1.1}Mn0.7_{0.7}Ti0.2_{0.2}O2_2 (SI Figure S5), respectively, in agreement with previously reported results from typical solid-state synthesis methods [14]. When these samples are quenched from a temperature slightly below TdisordT_{\mathrm{disord}}, the XRD pattern contains small impurity peaks that can be ascribed to orthorhombic LiMnO2_2, in addition to the majority DRX peaks (Figure 4a and Figure 4b). The presence of a two-phase mixture of DRX and orthorhombic LiMnO2_2 validates that the samples quenched from below TdisordT_{\mathrm{disord}} are in the two-phase coexistence region, as predicted by our first-principles phase diagram of Li1+x_{1+x}Mn1−3x3+_{1-3x}^{3+}Ti2x4+_{2x}^{4+}O2_2 (Figure 3a). Similarly, for the more Li- and Ti-rich compositions, Li1.25_{1.25}Mn0.25_{0.25}Ti0.5_{0.5}O2_2 (Figure 4c) and Li1.3_{1.3}Mn0.1_{0.1}Ti0.6_{0.6}O2_2 (Figure 4d), which are past the eutectic point, pure DRX can be obtained above TdisordT_{\mathrm{disord}} while layered Li2_2TiO3_3 is found to coexist with DRX below TdisordT_{\mathrm{disord}}. The lattice parameters of the pure Li1.25_{1.25}Mn0.25_{0.25}Ti0.5_{0.5}O2_2 and Li1.3_{1.3}Mn0.1_{0.1}Ti0.6_{0.6}O2_2 DRX phases are refined to be 4.14 and 4.13, respectively (SI Figure S6).

Ex situ XRD patterns of samples quenched from various temperatures

Figure 4. Ex situ XRD patterns of samples quenched from various temperatures. a) Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2, b) Li1.1_{1.1}Mn0.7_{0.7}Ti0.2_{0.2}O2_2, c) Li1.25_{1.25}Mn0.25_{0.25}Ti0.5_{0.5}O2_2, and d) Li1.3_{1.3}Mn0.1_{0.1}Ti0.6_{0.6}O2_2, each quenched from temperatures above (red) and below (black) their respective TdisordT_{\mathrm{disord}}.

Replacing Ti4+^{4+} with Mn4+^{4+} leads to markedly different phase behavior. Ex situ XRD patterns of Li1.05_{1.05}Mn0.95_{0.95}O2_2, Li1.1_{1.1}Mn0.9_{0.9}O2_2, and Li1.25_{1.25}Mn0.75_{0.75}O2_2 samples, quenched from 1100 °C are shown in Figure 5. Upon quenching Li1.05_{1.05}Mn0.95_{0.95}O2_2, the resulting sample is predominantly a DRX phase with minor orthorhombic impurities (Figure 5). This observation is surprising because our in situ XRD patterns show that DRX is phase pure at ∼1100\sim1100 °C (SI Figure S7). We therefore attribute the presence of the orthorhombic impurity to the quench being too slow. As a result, the quenched samples may reflect the equilibrium phase behavior at lower temperatures, where Li1.05_{1.05}Mn0.95_{0.95}O2_2 would lie within a predicted two-phase region between orthorhombic LiMnO2_2 and DRX (Figure 3b). A reduced fraction of the orthorhombic phase is observed for Li1.1_{1.1}Mn0.9_{0.9}O2_2, with DRX remaining the dominant phase, as indicated by the red XRD pattern in Figure 5. In contrast, quenching Li1.25_{1.25}Mn0.75_{0.75}O2_2 from 1100 °C yields a mixture of layered Li2_2MnO3_3 and DRX, in agreement with the corresponding two-phase region in the computed phase diagram (Figure 3b).

Ex situ XRD patterns of Li1.05Mn0.95O2, Li1.1Mn0.9O2, and Li1.25Mn0.75O2

Figure 5. Ex situ XRD patterns of compositions Li1.05_{1.05}Mn0.95_{0.95}O2_2, Li1.1_{1.1}Mn0.9_{0.9}O2_2, and Li1.25_{1.25}Mn0.75_{0.75}O2_2, each quenched from 1100 °C.

When quenching Li1.05_{1.05}Mn0.95_{0.95}O2_2 from a slightly lower temperature of 1000 °C, we identify a third phase in addition to orthorhombic LiMnO2_2 and DRX. The corresponding XRD pattern is shown in SI Figure S8 and we assign this additional phase to be layered LiMnO2_2, with possible Li-excess. Since lithiated spinel and layered LiMnO2_2 have similar XRD patterns, we performed further first-principles calculations to analyze the stability of these phases.

In grand-canonical Monte Carlo (GCMC) simulations, we were unable to observe the formation of a layered phase with composition near Li1.05_{1.05}Mn0.95_{0.95}O2_2. To rationalize the experimentally observed (meta)stability of the Li-excess layered phase, we compare the DFT-computed energies of the layered, orthorhombic, and lithiated spinel phases at the Li1.0625_{1.0625}Mn0.9375_{0.9375}O2_2 (Li1.0625_{1.0625}Mn0.81253+_{0.8125}^{3+}Mn0.1254+_{0.125}^{4+}O2_2) composition. Because these phases need to accommodate Li-excess through partial occupancy of the Mn sites, we enumerate symmetrically distinct configurations of Li and Mn on those sites and compute their energy above the hull (EhullE_{\mathrm{hull}}), which are shown in Figure 6a. Though all structures have positive values of Ehull≥17E_{\mathrm{hull}} \geq17 meV/atom, the layered configurations are lower (Ehull≈17E_{\mathrm{hull}} \approx17 meV/atom) than the orthorhombic and spinel configurations (Ehull≥19E_{\mathrm{hull}} \geq19 meV/atom). The favorability of the Li-excess layered phase suggests that the experimentally found impurity phase in SI Figure S8 is indeed a Li-excess layered phase, which forms as a metastable phase at that composition. tion when phase separation cannot occur fast enough. This trend clearly contrasts with the stoichiometric LiMnO2\mathrm{LiMnO}_{2} composition, in which the orthorhombic phase is the ground state (SI Figure S9). We note that these sets of enumerated Li-excess structures are not exhaustive, so there may be other possible orderings of each phase that are not considered.

Energy above the hull of Li-excess orthorhombic, layered, and spinel structures

Figure 6. Energy above the hull (EhullE_{\mathrm{hull}}) of Li-excess orthorhombic (ortho), layered, and spinel structures at the a) Li1.0625Mn0.9375O2\mathrm{Li}_{1.0625}\mathrm{Mn}_{0.9375}\mathrm{O}_{2} and b) Li1.0625Mn0.8125Ti0.125O2\mathrm{Li}_{1.0625}\mathrm{Mn}_{0.8125}\mathrm{Ti}_{0.125}\mathrm{O}_{2} compositions, computed from DFT at T=0T = 0 K.

To understand the difference between using Ti4+\mathrm{Ti}^{4+} or Mn4+\mathrm{Mn}^{4+} to accommodate Li-excess, we similarly calculate EhullE_{\mathrm{hull}} of the orthorhombic, layered, and spinel phases at the Li1.0625Mn0.8125Ti0.125O2\mathrm{Li}_{1.0625}\mathrm{Mn}_{0.8125}\mathrm{Ti}_{0.125}\mathrm{O}_{2} composition, which is plotted in Figure 6b. These structures also have energies above the convex hulls with Ehull≥12E_{\mathrm{hull}} \ge12 meV/atom, though these values are on average lower than the Li1.0625Mn0.9375O2\mathrm{Li}_{1.0625}\mathrm{Mn}_{0.9375}\mathrm{O}_{2} structures (Figure 6a). For the Li1.0625Mn0.8125Ti0.125O2\mathrm{Li}_{1.0625}\mathrm{Mn}_{0.8125}\mathrm{Ti}_{0.125}\mathrm{O}_{2} composition, the orthorhombic phase is the more energetically favorable as it contains the lowest energy orderings, followed by the layered and spinel phases (Figure 6b).

These computed energies suggest that for Li-excess compositions along Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2}, the layered phase is metastable against phase separation but more energetically favorable compared to the orthorhombic phase, and as such, can be stabilized by quenching. Since these computed energies do not include entropic effects, we perform additional canonical MC simulations to calculate the free energy of the Li-excess layered phase at temperatures up to 800 °C, but find that the free energy always lies above the convex hull (SI Figure S10). The predicted metastability of the layered phase suggests that the layered phase observed in experiment (SI Figure S8) is a kinetic product that can more rapidly form than the fully phase-separated state, possibly because it requires no long-range diffusion. Upon doping with Ti4+\mathrm{Ti}^{4+}, the Li-excess layered phase is not found in our experiments on Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_{2} samples possibly because the orthorhombic phase is more energetically favorable compared to the layered phase within this composition range.

Discussion

Through a combined first-principles and experimental approach, we have mapped the phase diagram of the LMTO rock-salt phases. We have closely analyzed the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_{2} (LiMnO2_2 – Li2_2TiO3_3), Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_{2} (LiMnO2_2 – Li2_2MnO3_3), and Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{x}^{4+}\mathrm{Ti}_{x}^{4+}\mathrm{O}_{2} (LiMnO2_2 – Li2_2Mn0.5_{0.5}Ti0.5_{0.5}O3_3) pseudo-binary phase diagrams (Figure 3), each of which are eutectoid-like, in which the order-disorder transition temperature (TdisordT_{\mathrm{disord}}) decreases as off-stoichiometry is introduced to the end-point compositions (adding excess Li to LiMnO2_2 or Li deficiency to Li2D4+_2D^{4+}O3_3, where DD can be Mn, Ti, or an admixture of both). The strong dependence of TdisordT_{\mathrm{disord}} on composition reveals that the optimization of DRX composition and synthesis temperature are intrinsically coupled. Through heating and in situ XRD experiments in an inert atmosphere, we determine the TdisordT_{\mathrm{disord}} of specific LMTO compositions, which agree reasonably well with the calculated values, though they can deviate by up to ∼200\sim200 °C.

We interpolate the calculated pseudo-binary phase diagrams to present a schematic pseudo-ternary phase diagram in Figure 7 at temperatures ranging from 800 to 1100 °C, which spans temperatures below and above conventional DRX synthesis temperatures (∼1000\sim1000 °C). Our simulations suggest that DRX competes with two other phases, which are orthorhombic LiMnO2_2 and layered Li2D4+_2D^{4+}O3_3, at the temperatures considered (800 – 1300 °C). The layered Li2D4+_2D^{4+}O3_3 phase (green line in Figure 7) forms a full solid-solution between Li2_2MnO3_3 and Li2_2TiO3_3 (SI Figure S11). In experiments, quenched samples of certain compositions contain layered LiMnO2_2-like impurity phases as well (SI Figure S8), which our calculations suggest to be metastable when phase separation does not occur. At 800 °C (Figure 7a), DRX is predicted to be stable only in a small range of compositions with moderate concentration of Li-excess and Ti4+^{4+} (grey shaded regions in Figure 7), while the other compositions lie within a multi-phase region (unshaded areas). By 900 °C, the composition range of DRX increases significantly, especially in the Ti4+-rich regions (near the LiMnO2 – Li2TiO3 pseudo-binary). As temperature increases to 1100 °C, DRX is stable within nearly the entire LiMnO2 – Li2TiO3 composition space, except in the regions with very little Li-excess, while it is only stable within a limited region within LiMnO2 – Li2MnO3. Thus, there is an increasing width of the miscibility gap between DRX and Li2D4+O3 as the ratio of Mn4+ to Ti4+ increases. These trends are verified by our experimental determination of TdisordT_{\mathrm{disord}} along Li1+xMn3+1-3xTi4+2xO2 and Li1+xMn3+1-3xMn4+2xO2 (Figure 3).

Pseudo-ternary phase diagrams of LiMnO2–Li2MnO3–Li2TiO3 at 800, 900, 1000, and 1100 °C

Figure 7. Pseudo-ternary phase diagram of LiMnO2 – Li2MnO3 – Li2TiO3 at 800, 900, 1000, and 1100 °C. The two-phase and three-phase regions are labeled as 2p and 3p, respectively. Black points are obtained from the calculated pseudo-binary phase diagrams (Figure 3) and the lines are interpolations between them. Gray shaded regions are DRX, green lines represent layered Li2D4+O3, and the orange line in c) denotes the Li-excess layered phase observed in experiment, which our calculations predict to be metastable.

The distinct impact of Ti4+ and Mn4+ on the thermodynamic accessibility of DRX can be attributed to the differences in electronic structure. Ti4+ has negligible filling of d-shell states (d0), while Mn4+ contains d-shells that are partially filled. It was shown by Urban et al. that, compared to TM species with d-shells that are (partially) filled, ions with the d0 electronic structure can better stabilize disordered configurations, as the local lattice distortions that are inherently present in disordered configurations can be accommodated with minimal energy cost 21. These differences in electronic structure rationalize why the TdisordT_{\mathrm{disord}} tends to increase significantly as the ratio of Mn4+ to Ti4+ increases.

Notably, the layered Li2MnO3 structure is predicted from MC simulations to have a remarkably high Tdisord>2800T_{\mathrm{disord}} > 2800 °C — which may not be realistic as this method does not account for potential melting or reduction, which would occur before reaching such a high temperature. Nevertheless, the high predicted TdisordT_{\mathrm{disord}} reflects the steep energy cost to disorder the layered Li2MnO3 structure. Thus, compositions with substantial Mn4+ content favor phase separation into layered Li2MnO3 (or Li2D4+O3 with high Mn4+ content), substantially limiting the DRX composition range that is accessible at reasonable temperatures. These results rationalize why there have been few reports of d0-free and Mn-rich DRX synthesized without the utilization of high-energy mechano-chemical milling techniques 17,57.

Meanwhile, layered Li2_2TiO3_3 has a much lower predicted TdisordT_{\mathrm{disord}} of 1000 °C, reflecting a lower energy cost to disorder. The significantly lower TdisordT_{\mathrm{disord}} of layered Li2_2TiO3_3 has pivotal implications, as it implies that the compositions with moderate to high fraction of Ti4+^{4+} can disorder at much lower temperatures. Specifically, a significant composition range of DRX is stable at temperatures of 800 − 900 °C, suggesting the potential of lowering the synthesis temperature of many DRX compositions below the conventionally used temperatures ≥ 1000 °C[6] [8]. At 900 °C, DRX within the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 space is predicted to be stable in the range of Li-excess fraction (xx) spanning 0.07 − 0.28, while in Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2 it is stable for xx spanning 0.07 − 0.14 (Figure 3). The accessible composition range of DRX shrinks significantly as the temperature decreases to 800 °C, but an appreciable portion remains (Figure 7). Lowering the synthesis temperature is instrumental towards lowering the synthesis energy cost and the ability to optimize DRX particle size, which is subject to rapid coarsening at high temperature, thereby reducing rate performance[6]. Although we predict and verify from experiments that certain compositions have reasonably low TdisordT_{\mathrm{disord}}, we find that some samples such as Li1.05_{1.05}Mn0.95_{0.95}O2_2 quenched from above the identified TdisordT_{\mathrm{disord}} are not pure DRX (Figure 5). This is likely due to the quench not being fast enough to retain the high-temperature state. Thus, it is essential to develop synthesis methods that can rapidly cool samples with precise temperature control, so samples spend minimal time within two-phase regions.

It is important to note that some compositions containing an admixture of Mn4+^{4+} and Ti4+^{4+} can also have reasonably low TdisordT_{\mathrm{disord}}. This is exemplified by our prediction of Tdisord≈800T_{\mathrm{disord}} \approx800 °C for compositions near Li1.1_{1.1}Mn0.73+^{3+}_{0.7}Mn0.14+^{4+}_{0.1}Ti0.14+^{4+}_{0.1}O2_2 (Figure 3c). In fact, the simulated phase boundary of DRX in the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2 phase diagram (Figure 3c) is nearly identical to the one in Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 (Figure 3a) at relatively small fractions of Li-excess (x≤0.1x \le0.1). To our knowledge, these compositions containing both Ti4+^{4+} and Mn4+^{4+} have not been previously reported, and thus constitute a relatively new design space in which LMTO DRX can be explored and optimized. The (partial) substitution of Ti4+^{4+} with Mn4+^{4+} is potentially important as Wu et al. have recently shown that Mn-rich DRX compositions without Ti can yield higher energy density compared to compositions containing Ti, given the same Li concentration57. These observations suggest that an optimal balance between thermodynamic accessibility and electrochemical performance can be found by tuning the fraction of Mn4+^{4+} and Ti4+^{4+}, especially near Li fractions ≈1.05\approx1.05 to 1.1. Although the full replacement of Ti4+^{4+} with Mn4+^{4+} generally yields higher TdisordT_{\mathrm{disord}}, our calculations predict that the Ti-free DRX composition ≈\approx Li1.08_{1.08}Mn0.92_{0.92}O2_2 has a relatively accessible Tdisord∼1000T_{\mathrm{disord}} \sim1000 °C, as this is the eutectoid point of Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2.

We note that many previous studies have incorporated fluorine (F) into the anion sublattice of DRX, which could impact the TdisordT_{\mathrm{disord}}7. In a recent study, Liang et al. applied in situ XRD to determine the TdisordT_{\mathrm{disord}} of F-doped Li1.1_{1.1}Mn0.8_{0.8}Ti0.1_{0.1}O1.9_{1.9}F0.1_{0.1} to be ∼1000\sim1000 °C56, which is close to our experimentally determined TdisordT_{\mathrm{disord}} of Li1.05_{1.05}Mn0.85_{0.85}Ti0.1_{0.1}O2_2 (∼960\sim960 °C), which contains the same fraction of Ti. Given this close agreement of TdisordT_{\mathrm{disord}}, we do not expect that the TdisordT_{\mathrm{disord}} of compositions containing a small fraction of F would deviate far from our results. It should also be noted that one study has questioned whether any significant F incorporation occurs when the Mn content is very high58.

In general, our first-principles and experimental phase diagram are in reasonable agreement with each other. Specifically, the predicted changes of TdisordT_{\mathrm{disord}} with composition agree qualitatively with experiment. However, the quantitative agreement is lacking in some regions of the phase diagram, as the simulated TdisordT_{\mathrm{disord}} can significantly deviate from the experimental values by up to 150 − 200 °C, for example at the LiMnO2_2 and Li2_2TiO3_3 compositions. The predicted TdisordT_{\mathrm{disord}} tends to be more accurate for intermediate compositions along Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 and Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mnx4+^{4+}_{x}Tix4+^{4+}_{x}O2_2, as they are typically within 100 °C of the experimental values. For Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2, experiments yield a eutectoid composition ≈\approx Li1.2_{1.2}Mn0.4_{0.4}Ti0.4_{0.4}O2_2, which contains higher Li concentration compared to the calculated eutectoid composition (Li1.14_{1.14}Mn0.58_{0.58}Ti0.28_{0.28}O2_2), though the predicted eutectoid temperature (700 °C) is in close agreement with experiment (∼680\sim680 °C). In experiments we also find layered LiMnO2_2-like phases, specifically along the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2 pseudobinary, that are not predicted to be thermodynamically stable within our simulations.

The methods that we employ capture only configurational excitations, while neglecting the vibrational (SvibS_{\mathrm{vib}}) and magnetic (SmagS_{\mathrm{mag}}) entropy contributions that are also present in this system, potentially rationalizing the difference between theory and experimental results. It has been previously shown that the SvibS_{\mathrm{vib}} and its coupling to configurational degrees of freedom can significantly affect predicted TdisordT_{\mathrm{disord}} and/or solubility limits within alloys59,60. While not always the case, the incorporation of vibrational entropy often shifts calculated order-disorder transitions down. The coupling between SvibS_{\mathrm{vib}} and SconfigS_{\mathrm{config}} has been modeled for simple binary systems using the CE approach59, but capturing these effects within our higher dimensional quaternary LMTO system would require far greater computational complexity. Magnetic entropy is also present within the LMTO phases, as the orthorhombic LiMnO2_2 and layered Li2_2MnO3_3 phases are antiferromagnetically ordered below their Néel temperatures (∼280\sim280 and 40 K, respectively)23,25, so they are magnetically disordered at the elevated temperatures that we analyze.

Another potential source of error in our predicted phase diagram stems from the DFT method employed to generate the dataset used to train the CE model. The LMTO system contains electronic structure complexities that can be challenging to model from DFT — namely the presence of strongly correlated 3dd electronic states and JT distortions of Mn3+^{3+}. In a recent study, we specifically highlighted the challenges of modeling the phase stability of the LiMnO2_2 polymorphs within DFT, which motivated our selection of the HSE06 hybrid-GGA functional in this study, despite its relatively high computational cost48. Specifically, HSE06 can accurately predict orthorhombic LiMnO2_2 to be the ground state, while more commonly used GGA+UU and r2^2SCAN approaches predict a spurious LiMnO2_2 ground state 48. It was also shown that the energy differences between phases is highly sensitive to the DFT method used, which indicates that the predicted TdisordT_{\mathrm{disord}} could possibly be improved by using even more accurate electronic structure methods.

Another potential contribution to the free energy that we have not yet considered is the presence of Mn2+^{2+}. It is possible for Mn2+^{2+} to be present within octahedral environments in the rock-salt framework, given that MnO is a known stable phase 61. Mn2+^{2+} can form in the LMTO system from the following disproportionation reaction (Equation 1).

2Mn3+⇌Mn2++Mn4+(1)2\mathrm{Mn}^{3+} \rightleftharpoons\mathrm{Mn}^{2+} + \mathrm{Mn}^{4+} \tag*{(1)}

To investigate the potential influence of Mn2+^{2+}, we train an additional 5-component CE model that includes Mn2+^{2+} as a possible species. To train this model, we augment our DFT training structure set to span the MnO – Li2_2MnO3_3 – Li2_2TiO3_3 space (more details in Methods). We calculate the free energy curves of DRX LiMnO2_2 from MC simulations within the canonical ensemble containing Mn only in +3 state (LiMn3+^{3+}O2_2) and charge-neutral GC ensemble (at fixed chemical potentials) that allows for Mn3+^{3+} disproportionation to form Mn2+^{2+} and Mn4+^{4+} as well. These calculated free energies of DRX relative to the orthorhombic LiMnO2_2 phase are shown in Figure 8a. Within these simulations, the predicted TdisordT_{\mathrm{disord}} of LiMnO2_2 is predicted to be 1280 °C, regardless whether Mn disproportionation is accounted for. We track the simulated concentration of Mn2+^{2+} as a function of temperature in Figure 8b, and find that a relatively small fraction of Mn are disproportionated (< 2%) at temperatures below 1600 °C. The low fraction of disproportionation events at these elevated temperatures suggests that the energy cost for Mn disproportionation is substantial. Thus, the additional SconfigS_{\mathrm{config}} of accounting for Mn2+^{2+}, in addition to Mn3+^{3+} and Mn4+^{4+}, is predicted to be minimal at common DRX synthesis temperatures, so we do not expect it to significantly affect the phase boundaries at typical synthesis temperatures.

Figure 8. Free energies (FF) and Mn disproportionation of DRX LiMnO2_2 as a function of temperature in MC simulations using a 5-component CE model containing Li+^{+}, Ti4+^{4+}, Mn2+^{2+}, Mn3+^{3+}, and Mn4+^{4+}. a) FF relative to ortho-LiMnO2_2 (F−ForthoF-F_{\mathrm{ortho}}) computed within the grand-canonical (GC) and canonical ensembles. GC allows for Mn disproportionation, while canonical maintains Mn in the +3 state only. Vertical line at 1280 °C marks where DRX intersects with ortho-LiMnO2_2, thus the TdisordT_{\mathrm{disord}}. b) Fraction of Mn disproportionated in DRX LiMnO2_2 within GCMC simulations.

Conclusions

A phase diagram of the LMTO rock-salt (LiMnO2_2 – Li2_2MnO3_3 – Li2_2TiO3_3) phase space has been derived from first-principles statistical mechanics calculations and XRD experiments. The LMTO phase diagram is identified to be a eutectoid phase diagram, in which the TdisordT_{\mathrm{disord}} decreases as off-stoichiometry is introduced to the end-point compositions. Importantly, this leads to a significant range of compositions with TdisordT_{\mathrm{disord}} ranging from 680 – 900 °C, which is much lower than conventional DRX synthesis temperatures (≥ 1000 °C). This finding suggests the potential for lowering synthesis temperature for specific DRX compositions, which can aid in the optimization of particle size/morphology, and in turn the electrochemical performance of LMTO DRX.

Experimental

Theory

All structural enumerations and analysis were performed using the pymatgen package 53,62. To compute the total energy of each enumerated LMTO rock-salt structure, we perform density-functional theory (DFT) calculations using a plane-wave basis set and the projector augmented wave (PAW) method, as implemented in the Vienna ab initio Simulation Package (VASP) 63–65. All reported total energies are computed using the HSE06 hybrid generalized gradient approximation (GGA) 49. Since HSE06 requires significantly higher computational cost than the more commonly utilized GGA and meta-GGA functionals, we calculate the total energy of each structure using HSE06 at the structural geometries optimized at the meta-GGA level (r2^{2}SCAN) 45. Specifically, we harness the atomate2 package 66 to perform a three-step workflow involving the following sequential steps: 1) structural relaxation within GGA (PBEsol+UU 44, where U=3.9U = 3.9 eV) functional, 2) structural relaxation within r2^{2}SCAN+UU (U=1.8U = 1.8 eV), and 3) static calculation of the total energy using HSE06, where UU are rotationally averaged Hubbard UU corrections 46 on Mn−3dd electron states that were empirically fitted in previous studies to optimize prediction of Mn oxidation reaction enthalpies 67,68. The initial relaxation using PBEsol+UU effectively speeds up the r2^{2}SCAN+UU relaxation. Although this approach does not calculate each structure’s true ground state energy within HSE06, we find that the energy differences between select LiMnO2_2 phases are reasonably consistent with fully self-consistent HSE06 calculations, with a root mean squared error (RMSE) ∼1\sim1 meV/atom, which is within the error range of the cluster expansion (SI Figure S6). We estimate that this workflow reduces the computation time by at least a factor of 5 compared to full relaxations using HSE06. More details of the DFT calculations are described in SI Note 1.

We use these DFT-computed structures and energies to parametrize a quaternary CE to describe the energies of possible cation orderings of the cubic rock-salt lattice (Fm3ˉ\bar{3}m), with a primitive cell lattice parameter a=2.97a = 2.97 Å. The CE represents the total energy of any configuration as a summation over energy contributions from local cluster interactions 33,35. We construct our CE model with pair, triplet, and quadruplet cluster interactions up to cluster distances of 8.91, 5.94, and 5.15 Å, respectively, using the smol python package, which is also used for all lattice Monte Carlo (MC) simulations 34. Mixed ℓ1\ell_1 and ℓ2\ell_2 norm penalized linear regression (Elastic Net 69), as implemented in the scikit-learn package 70, is used to fit the CE on the DFT energies of 745 unique structures, where the ℓ1\ell_1 and ℓ2\ell_2 regularization terms promote sparsity and smaller magnitudes of the interactions, respectively. The fit is performed with sample weights applied on the ground state structures and select low-energy structures, a procedure that enables the trained model to correctly preserve the DFT ground states. The 5-fold cross-validation (CV) and training errors of the parametrized CE model are 9.5 meV/atom and 8.5 meV/atom, respectively. The training curves for regularization hyperparameter optimization and further details of CE fitting are shown in SI Figure S14 and SI Note 2. The CE predicts the correct energy ordering among the orthorhombic, layered, spinel, and γ\gamma LiMnO2_2 phases (SI Figure S9), which further helps validate the model’s accuracy. To understand the influence of Mn2+^{2+}, we train an additional CE model that describes a quinary cation site space containing Li+^{+}, Mn2+^{2+}, Mn3+^{3+}, Mn4+^{4+}, and Ti4+^{4+} as possible species. This model was trained in a similar fashion as the quaternary model, with more details described in SI Note 2.

Using the quaternary CE model, we perform MC simulations and thermodynamic integration to calculate the free energies of each phase and determine the phase boundaries between them. For the Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Ti2x4+^{4+}_{2x}O2_2 and Li1+x_{1+x}Mn1−3x3+^{3+}_{1-3x}Mn2x4+^{4+}_{2x}O2_2 pseudo-binary spaces, we perform charge-neutral grand-canonical (GC)-MC simulations 71 over a range of temperature (TT) and chemical potentials (μ\mu), the details of which are described in SI Note 2. From these MC trajectories, we perform thermodynamic integration over ensemble-averaged energies and compositions to determine the GC free energy (Ω\Omega) of the ortho LiMnO2_2, DRX, layered Li2_2TiO3_3, and layered Li2_2MnO3_3 phases. The TT and μ\mu values where two Ω\Omega curves intersect are identified to determine the boundary between the two phases. To calculate Ω\Omega, the average GC energy (⟨ε−μN⟩\langle\varepsilon- \mu N \rangle) is integrated over inverse thermal energy (β≡1kBT\beta\equiv\frac{1}{k_{\mathrm{B}}T}) at a fixed reference μ0\mu_0, as described by Equation 2 72. The μ\mu dependence of Ω\Omega is obtained by integrating the average composition (⟨N⟩\langle N\rangle) over μ\mu at fixed β\beta (Equation 3).

βΩ(β,μ0)−β0Ω(β0,μ0)=∫β0β⟨ε−μ0N⟩ dβ, at fixed μ0(2)\beta\Omega(\beta,\mu_0)-\beta_0\Omega(\beta_0,\mu_0)=\int_{\beta_0}^{\beta}\langle\varepsilon-\mu_0N\rangle\,\mathrm{d}\beta,\ \text{at fixed }\mu_0 \tag*{(2)}
Ω(β,μ)−Ω(β,μ0)=−∫μ0μ⟨N⟩ dμ, at fixed β(3)\Omega(\beta,\mu)-\Omega(\beta,\mu_0)=-\int_{\mu_0}^{\mu}\langle N\rangle\,\mathrm{d}\mu,\ \text{at fixed }\beta \tag*{(3)}

The reference values of μ0\mu_0 and β0\beta_0 are chosen such that the GC-MC simulations at μ0\mu_0 and β0\beta_0 equilibrate to the phase of interest. The reference values of β0\beta_0 for the ordered phases (ortho LiMnO2_2, layered Li2_2TiO3_3, and layered Li2_2MnO3_3) are chosen at corresponding T0T_0 below their respective TdisT_{\mathrm{dis}}. Meanwhile, β0\beta_0 of DRX is chosen to be the T→∞T \to\infty limit (equivalently the β→0\beta\to0 limit), where the free energy can be analytically determined through the expression β0Ω(β0,μ0)=−ln⁡(σ)\beta_0\Omega(\beta_0,\mu_0)=-\ln(\sigma) 72, where σ\sigma is the number of unique arrangements of the LMTO lattice (i.e. −Sconfig/kB-S_{\mathrm{config}}/k_{\mathrm{B}} in the random limit). For a quaternary alloy, this can be trivially evaluated as ln⁡(σ)=ln⁡4\ln(\sigma)=\ln4. However, this expression is not valid for the ionic LMTO system, as it accounts for configurations with compositions that are not charge neutral. σ\sigma can also be evaluated analytically for a finite supercell of NN cation sites and composition defined by nLin_{\mathrm{Li}}, nMn3n_{\mathrm{Mn3}}, nMn4n_{\mathrm{Mn4}}, and nTi4n_{\mathrm{Ti4}} (where nLin_{\mathrm{Li}} is the number of Li, etc) by the expression shown in Equation 4a. To facilitate evaluating ln⁡(σ)\ln(\sigma), we apply Stirling’s approximation, as shown in Equation 4b, which converges quickly with NN.

σ=N!nLi!nMn3!nMn4!nTi4!(4)\sigma= \frac{N!}{n_{\mathrm{Li}}! n_{\mathrm{Mn3}}! n_{\mathrm{Mn4}}! n_{\mathrm{Ti4}}!} \tag*{(4)}
ln⁡(σ)≈Nln⁡(N)−nLiln⁡(nLi)−nMn3ln⁡(nMn3)−nMn4ln⁡(nMn4)−nTi4ln⁡(nTi4)(5)\ln(\sigma) \approx N\ln(N)-n_{\mathrm{Li}}\ln(n_{\mathrm{Li}})-n_{\mathrm{Mn3}}\ln(n_{\mathrm{Mn3}})-n_{\mathrm{Mn4}}\ln(n_{\mathrm{Mn4}})-n_{\mathrm{Ti4}}\ln(n_{\mathrm{Ti4}}) \tag*{(5)}

σ\sigma is thus a function of composition, but in the limits of infinite TT and NN, the composition will converge to the composition that maximizes SconfigS_{\mathrm{config}} and in turn σ\sigma. Therefore, we perform a GCMC simulation at T=58000T = 58000 K and N=720N = 720, which converges to an average composition ≈Li1.2Mn0.43+Mn0.44+Ti0.24+O2\approx\mathrm{Li}_{1.2}\mathrm{Mn}^{3+}_{0.4}\mathrm{Mn}^{4+}_{0.4}\mathrm{Ti}^{4+}_{0.2}\mathrm{O}_2. At this composition, we evaluate using Equation 4b ln⁡(σ)≈1.089\ln(\sigma) \approx1.089 kBk_B/cation, which importantly is ≈21%\approx21\% lower than the corresponding value for a quaternary alloy (ln⁡4=1.386\ln4 = 1.386 kBk_B/cation), with the reduction in entropy arising from charge neutrality constraints.

We calculate the Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{x}\mathrm{Ti}^{4+}_{x}\mathrm{O}_2 pseudo-binary phase diagram using canonical MC simulations and the common tangent construction. The canonical free energy of DRX (FDRXF_{\mathrm{DRX}}) is computed at compositions xx ranging from 0 to 0.3 in increments of 0.025 along Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{x}\mathrm{Ti}^{4+}_{x}\mathrm{O}_2. The free energy integration scheme to calculate FDRXF_{\mathrm{DRX}} is identical to the calculation of ΩDRX\Omega_{\mathrm{DRX}} in the limit of μ0=0\mu_0 = 0 and fixed composition (Equation (2)). The details of the free energy profiles and common tangent construction are shown in SI Figure S12.

Figure S12. Common tangent construction used to determine the Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{x}\mathrm{Ti}^{4+}_{x}\mathrm{O}_{2} (LiMnO2\mathrm{LiMnO}_{2} – Li2Mn0.5Ti0.5O3\mathrm{Li}_{2}\mathrm{Mn}_{0.5}\mathrm{Ti}_{0.5}\mathrm{O}_{3}) phase diagram. Formation energy landscape of DRX along Li1+xMn1−3x3+Mnx4+Tix4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{x}\mathrm{Ti}^{4+}_{x}\mathrm{O}_{2} at varying TT, computed from canonical MC simulations. Gold stars represent unstable phases while green squares are stable.

Experiment

To investigate the disorder temperatures of Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Mn}^{4+}_{2x}\mathrm{O}_2 (x=0.05, 0.1, 0.15, 0.2, 0.25, 0.3), Li2_2CO3_3 (Sigma, 99%), Mn2_2O3_3 (Sigma, 99%) and MnO2_2 (Alfa Aesar, 99.9%) were used as precursors. For the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}^{3+}_{1-3x}\mathrm{Ti}^{4+}_{2x}\mathrm{O}_2 series (x=0.05, 0.1, 0.15, 0.2, 0.25, 0.3), Li2_2CO3_3 (Sigma, 99%), Mn2_2O3_3 (Sigma, 99%) and TiO2_2 (Sigma, 99.5%) were used as precursors. For each composition, a total of 4 g of precursors was stoichiometrically mixed in ethanol using a Retsch PM200 planetary ball mill at 250 rpm for 12 h. A 10% excess of Li2_2CO3_3 was used to compensate for possible lithium loss during high-temperature treatment. The resulting slurry was dried at 70 °C in a convection oven overnight and subsequently ground. Approximately 300 mg of the mixed precursors were pressed into pellets with a diameter of 13 mm.

For in situ heating experiments, the pallets were heated to 1100 °C, held at this temperature for 20 min, and then cooled to room temperature using an Anton Paar HTK1200N furnace under a constant nitrogen flow. X-ray diffraction (XRD) data were collected every 10 °C during heating and cooling and continuously collected during holding, using a Bruker D8 Advance diffractometer. All scans were conducted over a 2θ2\theta range of 10-70°.

For ex situ quenching experiments, pellets were heated to temperatures below and above the respective disorder temperatures and held for 20 min in a tube furnace. The samples were then rapidly removed from the furnace and cooled to room temperature using a fan. A constant argon flow was maintained throughout the heating process. XRD patterns of the quenched samples were collected using an Aeris Minerals diffractometer (Malvern Panalytical) utilizing Cu Kα\alpha radiation. Scans were performed over a 2θ2\theta range of 10-100°.

Conflicts of interest

There are no conflicts to declare.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Acknowledgements

We thank Dr. Aaron Kaplan for assisting in building the DFT workflow within atome2 and Tucker Holstun for insightful discussion about LMTO rock-salt phases. This work was supported by the Assistant Secretary of Energy Efficiency and Renewable Energy, Vehicle Technologies Office of the US Department of Energy (DOE), under contract no. DE-AC02-05CH11231 under the Advanced Battery Materials Research (BMR) Program. S.W. acknowledges partial support from the Jane Lewis Fellowship at UC Berkeley. This research used computational resources of the National Laboratory of the Rockies (NLR) and the National Energy Research Scientific Computing Center (NERSC).

Supplementary Information

Ronald L. Kam; Shilong Wang; Gerbrand Ceder

si.pdf

Three temperature–time and intensity maps for lithium manganese titanium oxides

Figure 1 (PDF p. 18). (a) Li1.05Mn0.85Ti0.1O2\mathrm{Li}_{1.05}\mathrm{Mn}_{0.85}\mathrm{Ti}_{0.1}\mathrm{O}_{2}; (b) Li1.1Mn0.7Ti0.2O2\mathrm{Li}_{1.1}\mathrm{Mn}_{0.7}\mathrm{Ti}_{0.2}\mathrm{O}_{2}; (c) Li1.15Mn0.55Ti0.3O2\mathrm{Li}_{1.15}\mathrm{Mn}_{0.55}\mathrm{Ti}_{0.3}\mathrm{O}_{2}.

Temperature and X-ray diffraction intensity map for Li1.2Mn0.4Ti0.4O2

Figure d. Li1.2Mn0.4Ti0.4O2\mathrm{Li}_{1.2}\mathrm{Mn}_{0.4}\mathrm{Ti}_{0.4}\mathrm{O}_{2}

Temperature and X-ray diffraction intensity map for Li1.25Mn0.25Ti0.5O2

Figure e. Li1.25Mn0.25Ti0.5O2\mathrm{Li}_{1.25}\mathrm{Mn}_{0.25}\mathrm{Ti}_{0.5}\mathrm{O}_{2}

Temperature and X-ray diffraction intensity map for Li1.3Mn0.1Ti0.6O2

Figure f. Li1.3Mn0.1Ti0.6O2\mathrm{Li}_{1.3}\mathrm{Mn}_{0.1}\mathrm{Ti}_{0.6}\mathrm{O}_{2}

Panel (g): temperature-dependent diffraction pattern for Li$_{1.05}$Mn$_{0.95}$O$_2$

Figure 9 (PDF p. 20). (g) Li1.05_{1.05}Mn0.95_{0.95}O2_2

Panel (h): temperature-dependent diffraction pattern for Li$_{1.1}$Mn$_{0.9}$O$_2$

Figure 10 (PDF p. 20). (h) Li1.1_{1.1}Mn0.9_{0.9}O2_2

Panel (i): temperature-dependent diffraction pattern for Li$_{1.15}$Mn$_{0.85}$O$_2$

Figure 11 (PDF p. 20). (i) Li1.15_{1.15}Mn0.85_{0.85}O2_2

Free energy and internal energy of layered Li$_{2}$Mn$_{0.5}^{4+}$Ti$_{0.5}^{4+}$O$_3$ relative to its minimum energy

Figure S3. Free energy and internal energy of layered Li2_2Mn0.54+_{0.5}^{4+}Ti0.54+_{0.5}^{4+}O3_3 relative to its minimum energy, computed from canonical MC simulations.

Figure S4. Average energies relative to the ground state energy of the ortho LiMnO2_2, layered Li2_2TiO3_3, and layered Li2_2MnO3_3 phases, calculated using canonical MC heating simulations.

XRD patterns of compositions Li1.05Mn0.85Ti0.1O2 and Li1.1Mn0.7Ti0.2O2

Figure S5. XRD patterns of compositions a) Li1.05Mn0.85Ti0.1O2\mathrm{Li}_{1.05}\mathrm{Mn}_{0.85}\mathrm{Ti}_{0.1}\mathrm{O}_2, and b) Li1.1Mn0.7Ti0.2O2\mathrm{Li}_{1.1}\mathrm{Mn}_{0.7}\mathrm{Ti}_{0.2}\mathrm{O}_2, quenched from 1000 °C and 900 °C, respectively. They are refined based on the rock salt structure. The lattice parameter (aa) and weighted profile R factor (RwpR_{\mathrm{wp}}) are shown in the figures.

XRD patterns of compositions Li1.25Mn0.25Ti0.5O2 and Li1.3Mn0.1Ti0.6O2

Figure S6. XRD patterns of compositions a) Li1.25Mn0.25Ti0.5O2\mathrm{Li}_{1.25}\mathrm{Mn}_{0.25}\mathrm{Ti}_{0.5}\mathrm{O}_2, and b) Li1.3Mn0.1Ti0.6O2\mathrm{Li}_{1.3}\mathrm{Mn}_{0.1}\mathrm{Ti}_{0.6}\mathrm{O}_2, quenched from 700 °C and 750 °C, respectively. They are refined based on the rock salt structure. The lattice parameter (aa) and weighted profile R factor (RwpR_{\mathrm{wp}}) are shown in the figures.

XRD pattern of the Li$_{1.05}$Mn$_{0.95}$O$_2$ sample collected at 1100 °C

Figure S7. XRD pattern of the Li1.05_{1.05}Mn0.95_{0.95}O2_2 sample collected at 1100 °C during the in situ experiment. No orthorhombic/monoclinic LiMnO2_2 or layered Li2_2MnO3_3 phases are detected, confirming that the material is phase-pure DRX at this temperature. The additional peaks arise from the alumina sample holder and from reactions between excess Li2_2CO3_3 and the holder at high temperature.

Ex situ XRD pattern of the Li$_{1.05}$Mn$_{0.95}$O$_2$ sample quenched from 1000 °C

Figure S8. Ex situ XRD pattern of Li1.05_{1.05}Mn0.95_{0.95}O2_2 sample quenched from 1000 °C. Layered LiMnO2_2, orthorhombic LiMnO2_2 and DRX phases are observed.

Figure S9. Comparison of CE and DFT energies of LiMnO2_2 phases relative to ortho LiMnO2_2. These phases were previously analyzed in Ref.[1].

Figure S10. Free energy of Li1.05_{1.05}Mn0.853+^{3+}_{0.85}Mn0.14+^{4+}_{0.1}O2_2 relative to the convex hull, computed from canonical MC simulations.

Figure S11. Formation energy landscape of the layered Li2Ti1−xMnxO3\mathrm{Li}_{2}\mathrm{Ti}_{1-x}\mathrm{Mn}_{x}\mathrm{O}_{3} (Li2D4+O3\mathrm{Li}_{2}D^{4+}\mathrm{O}_{3}) space at varying TT, computed from canonical MC simulations.

Details of DFT calculations

For each enumerated LMTO structure, we perform a workflow using atomate2 to optimize the structure and evaluate the total energy. The workflow involves 1) relaxation using PBEsol+UU (U=3.9U = 3.9 eV), 2) relaxation using r2^{2}SCAN+UU (U=1.8U = 1.8 eV), then 3) static calculation using HSE06. Steps 2 and 3 use the wave functions and charge density obtained from the previous step as a starting point. The PBEsol+UU relaxations use Li_sv, Mn_pv, Ti_pv, and O PBE PAW pseudopotentials, while the r2^{2}SCAN+UU and HSE06 calculations employ an analogous set from the PBE_54 library of PAW within VASP. The energy and force convergence criteria are shown in Table 1, which are consistent with the PBE-based MPRelaxSet and r2^{2}SCAN-based MPScan-RelaxSet settings within pymatgen. All calculations use a reciprocal space discretization of 0.25 Å−1^{-1}.

MethodEnergy conv. (eV)Force conv. (eV/Å)Energy cutoff (eV)
PBEsol+UU1e1e−062e2e−02520
r2^{2}SCAN+UU1e1e−052e2e−02680
HSE061e1e−05N/A680

Table 1. DFT convergence criteria and plane wave energy cutoffs.

Details of CE training and MC simulations

The CE construction and MC simulations are performed using the smol package. [2] The CE model is based on a rock-salt lattice with lattice parameter a=2.97a = 2.97 Å. We use an orthonormal sinusoid basis set to represent the pair, triplet, and quadruplet cluster correlation functions up to cluster distances of 8.91, 5.94, and 5.15 Å, respectively. To describe long-range Coulomb interactions, we parametrize an effective dielectric constant that screens the bare electrostatic interaction of idealized Li+^{+}, Mn3+^{3+}, Mn4+^{4+}, and Ti4+^{4+} point charges, [3] which were calculated using the Ewald summation method. [4] The resulting CE model contains 1012 correlation functions and corresponding interactions. Mixed ℓ1−ℓ2\ell_{1}-\ell_{2} penalized linear regression (Elastic Net [5]) is used to fit the CE on the DFT energies of 745 unique structures, where the ℓ1\ell_{1} and ℓ2\ell_{2} regularization terms promote sparsity and smaller magnitudes of the interactions, respectively. The training curves for ℓ1\ell_{1} and ℓ2\ell_{2} hyperparameter optimization are shown in SI Figure S14. The parametrized CE interactions are shown in Figure S13.

Figure S13. Effective cluster interactions (ECI) of the quaternary CE model. Black vertical lines denote the division between pair, triplet, and quadruplet interactions (from left to right).

Figure S14. Optimizing the ℓ1\ell_1 (α\alpha) and ℓ2\ell_2 (β\beta) regularization hyperparameters within the Elastic Net regression approach. For each α\alpha, the cross-validation (CV) and training set RMSE is plotted as a function of β\beta. The black line denotes the chosen model, with α=1e−4\alpha= 1\mathrm{e}{-4} and β=3.55e−3\beta= 3.55\mathrm{e}{-3}. Blue shaded regions denote the standard deviation of the 5-fold CV trials.

To calculate the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_2 and Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_2 pseudo-binary phase diagrams, we use the CE to perform charge-neutral grand-canonical (GC)-MC simulations[6] across ranges of TT and μ\mu. For the Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_2 space, we scan μLi\mu_{\mathrm{Li}} from 9.0 − 11.0 eV in increments of 0.025 eV, starting from the ortho LiMnO2\mathrm{LiMnO}_2 structure, while μ\mu for all other species are zero. We perform the same scan in the reverse direction starting from layered Li2TiO3\mathrm{Li}_2\mathrm{TiO}_3. At specific μLi\mu_{\mathrm{Li}} and TT, the ordered phases transform to DRX, which are confirmed from simulating the XRD pattern of a structure averaged over at least 500 sampled configurations in the MC trajectory. Starting from the DRX, we perform additional scans of μ\mu to obtain a more complete Ω\Omega profile of DRX at each TT. For the Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_2 space, we set μMn4+=2.0\mu_{\mathrm{Mn}^{4+}} = 2.0 eV, which drives the replacement of Ti4+\mathrm{Ti}^{4+} with Mn4+\mathrm{Mn}^{4+}, and scan μLi\mu_{\mathrm{Li}} from 7.0 − 8.6 eV starting from ortho LiMnO2\mathrm{LiMnO}_2, while the reverse scan is performed starting from layered Li2MnO3\mathrm{Li}_2\mathrm{MnO}_3. For each set of μ\mu and TT, at 10000+ MC passes were proposed on supercells containing 700+ cation sites. Among the proposed MC steps, 40% involve changes in composition, while 60% do not change the composition (canonical swaps), a procedure that facilitates phase transitions.

SI Note 3: Experimentally determined TdisordT_{\mathrm{disord}}

For each composition of Li1+xMn1−3x3+Mn2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Mn}_{2x}^{4+}\mathrm{O}_2 (x=0.05, 0.1, 0.15, 0.2, 0.25, 0.3) and Li1+xMn1−3x3+Ti2x4+O2\mathrm{Li}_{1+x}\mathrm{Mn}_{1-3x}^{3+}\mathrm{Ti}_{2x}^{4+}\mathrm{O}_2 (x=0.05, 0.1, 0.15, 0.2, 0.25, 0.3), the corresponding precursor mixtures were heated from room temperature to 1100 °C under N2\mathrm{N}_2. In situ XRD patterns were collected every 10 °C, with each scan taking approximately 3 min. Given the small temperature interval between scans, we assume that the samples reach their equilibrium phase assemblages at each measured temperature. The experimental disorder temperatures TdisordT_{\mathrm{disord}} were identified as the points at which all competing ordered phases (e.g. orthorhombic LiMnO2\mathrm{LiMnO}_2, layered Li2MnO3\mathrm{Li}_2\mathrm{MnO}_3, layered Li2TiO3\mathrm{Li}_2\mathrm{TiO}_3) disappear. At high temperatures, additional phases corresponding to Li2O\mathrm{Li}_2\mathrm{O}, Al2O3\mathrm{Al}_2\mathrm{O}_3, or LiAlO2\mathrm{LiAlO}_2 were occasionally observed, arising from decomposition of excess Li2CO3\mathrm{Li}_2\mathrm{CO}_3, alumina sample holder, or reactions between Li2O\mathrm{Li}_2\mathrm{O} and Al2O3\mathrm{Al}_2\mathrm{O}_3. These impurity phases were not considered when determining TdisordT_{\mathrm{disord}} from the in situ XRD data.

References

  1. [1]Kam, R. L.; Binci, L.; Kaplan, A. D.; Persson, K.; Marzari, N.; Ceder, G. Interplay between electron localization, magnetic order, and Jahn-Teller distortion dictates LiMnO₂ phase stability. Physical Review B 2025, 111.arxiv.org/abs/2412.16816
  2. [2]Barroso-Luque, L.; Yang, J. H.; Xie, F.; Chen, T.; Kam, R. L.; Jadidi, Z.; Zhong, P.; Ceder, G. smol: A Python package for cluster expansions and beyond. The Journal of Open Source Software 2022,
  3. [3]Barroso-Luque, L.; Zhong, P.; Yang, J. H.; Xie, F.; Chen, T.; Ouyang, B.; Ceder, G. Cluster expansions of multicomponent ionic materials: Formalism and methodology. Physical Review B 2022, 106, 144202.
  4. [4]Ong, S. P.; Richards, W. D.; Jain, A.; Hautier, G.; Kocher, M.; Cholia, S.; Gunter, D.; Chevrier, V. L.; Persson, K. A.; Ceder, G. Python Materials Genomics (pymatgen): A robust, open-source python library for materials analysis. Computational Materials Science 2013, 68, 314–319.DOI
  5. [5]Zou, H.; Hastie, T. Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B (Statistical Methodology) 2005, 67, 301–320.DOI
  6. [6]Xie, F.; Zhong, P.; Barroso-Luque, L.; Ouyang, B.; Ceder, G. Semigrand-canonical Monte-Carlo simulation methods for charge-decorated cluster expansions. Computational Materials Science 2023, 218, 112000.DOI

Paper details

vlnc:W_xghq4zst
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents