Accelerated Materials Design of Lithium Superionic Conductors Based on First-Principles Calculations and Machine Learning Algorithms
Accelerated Materials Design of Lithium Superionic Conductors Based on First-Principles Calculations and Machine Learning Algorithms
复制标题
DOI：
10.1002/aenm.201300060
复制
发表时间：
2013-08-01
影响因子：
27.8
通讯作者：
Tanaka, Isao
中科院分区：
文献类型：
--
作者：
Fujimura, Koji;Seko, Atsuto;Tanaka, Isao
Concerted efforts continue to be made in the search for superior lithium-ion conducting solids to replace the highly reactive liquid electrolytes typically used in rechargeable batteries. LISICON-type materials have been studied extensively over the last few decades, providing abundant experimental data, but to date no overall design principle for achieving high conductivity has been forthcoming. In this communication we present results of systematic sets of first-principles calculations based on the cluster expansion method, as well as first-principles molecular dynamics (FPMD) simulations carried out to calculate Li-ion conductivities at high temperature, for a diverse range of compositions. A machine-learning technique is used to combine theoretical and experimental datasets to predict the conductivity of each composition at 373 K. The insights obtained show that an iterative combination of first-principles calculations and focused experiments can greatly accelerate the materials design process by enabling a wide compositional and structural phase space to be examined efficiently. Lithium-conducting oxides in the system LiO 1/2-AO m/2-BO n/2 (where m and n denote the formal valences of cations A and B, respectively), known as LISICONs and corresponding to general formula Li 8− cA aB bO 4 (where c= ma+ nb), [1] have been intensively studied since the 1970s. The original LISICON composition, Li 3.5Zn 0.25GeO 4, was reported to exhibit an ionic conductivity of over 10− 1 S cm− 1 at 673 K, [2] and stimulated a flurry of new research. Although the conducting properties of many different LISICONs have since been reported by various groups, there are still many compositions that have yet to be synthesized, let alone characterized. In some cases, results from different groups also vary considerably.[2–6] An Arrhenius plot of Li-ion conductivity summarizing previous experimental data is provided as Figure S1 in Supporting Information. Given the urgent need for improved energy storage and power devices, a more efficient means of designing ionic conductors based on a reproducible and systematic methodology is an imperative for future progress in this field.Thanks to astonishing improvements in computer performance and computational techniques, first-principles calculations based on density functional theory (DFT) are now used routinely for quantitative analysis of ionic conduction in crystals. Combination of DFT calculations with high-throughput and machine-learning techniques is now also considered a viable means of searching for novel lithium battery materials.[7–9] For compounds with simple chemistry and structure, the saddle point associated with an ionic jump, from which the activation barrier energy can be extracted in a straightforward manner, is readily calculated. Such methods have already been applied to several Li-battery materials, eg, LiCoO 2, [10] LiFePO 4, [11] TiO 2-B, [12] and graphite.[13] The nudged elastic band (NEB) method is popular for such a purpose.[9, 14–16] The effective frequency term for the ionic jump used in the calculation of ion diffusivity can also be computed from first-principles if desired.[13] When it comes to compounds with more complicated structures, such as are typical in LISICON solid electrolytes, the search for saddle points along convoluted migration paths, often involving complex migration mechanisms, (eg, cooperative mechanisms [17, 18]), becomes a non-trivial task. When the compound is a solid solution composed of multiple elements, this becomes even more difficult because of the huge variety of chemical environments that migrating ions encounter. LISICON represents a sobering example of both …
Concerted efforts continue to be made in the search for superior lithium-ion conducting solids to replace the highly reactive liquid electrolytes typically used in rechargeable batteries. LISICON-type materials have been studied extensively over the last few decades, providing abundant experimental data, but to date no overall design principle for achieving high conductivity has been forthcoming. In this communication we present results of systematic sets of first-principles calculations based on the cluster expansion method, as well as first-principles molecular dynamics (FPMD) simulations carried out to calculate Li-ion conductivities at high temperature, for a diverse range of compositions. A machine-learning technique is used to combine theoretical and experimental datasets to predict the conductivity of each composition at 373 K. The insights obtained show that an iterative combination of first-principles calculations and focused experiments can greatly accelerate the materials design process by enabling a wide compositional and structural phase space to be examined efficiently. Lithium-conducting oxides in the system LiO 1/2-AO m/2-BO n/2 (where m and n denote the formal valences of cations A and B, respectively), known as LISICONs and corresponding to general formula Li 8− cA aB bO 4 (where c= ma+ nb),[1] have been intensively studied since the 1970s. The original LISICON composition, Li 3.5Zn 0.25GeO 4, was reported to exhibit an ionic conductivity of over 10− 1 S cm− 1 at 673 K,[2] and stimulated a flurry of new research. Although the conducting properties of many different LISICONs have since been reported by various groups, there are still many compositions that have yet to be synthesized, let alone characterized. In some cases, results from different groups also vary considerably.[2–6] An Arrhenius plot of Li-ion conductivity summarizing previous experimental data is provided as Figure S1 in Supporting Information. Given the urgent need for improved energy storage and power devices, a more efficient means of designing ionic conductors based on a reproducible and systematic methodology is an imperative for future progress in this field.Thanks to astonishing improvements in computer performance and computational techniques, first-principles calculations based on density functional theory (DFT) are now used routinely for quantitative analysis of ionic conduction in crystals. Combination of DFT calculations with high-throughput and machine-learning techniques is now also considered a viable means of searching for novel lithium battery materials.[7–9] For compounds with simple chemistry and structure, the saddle point associated with an ionic jump, from which the activation barrier energy can be extracted in a straightforward manner, is readily calculated. Such methods have already been applied to several Li-battery materials, eg, LiCoO 2,[10] LiFePO 4,[11] TiO 2-B,[12] and graphite.[13] The nudged elastic band (NEB) method is popular for such a purpose.[9, 14–16] The effective frequency term for the ionic jump used in the calculation of ion diffusivity can also be computed from first-principles if desired.[13] When it comes to compounds with more complicated structures, such as are typical in LISICON solid electrolytes, the search for saddle points along convoluted migration paths, often involving complex migration mechanisms,(eg, cooperative mechanisms [17, 18]), becomes a non-trivial task. When the compound is a solid solution composed of multiple elements, this becomes even more difficult because of the huge variety of chemical environments that migrating ions encounter. LISICON represents a sobering example of both …