Computational modeling from first-principles holds promise for the rational tailoring of the excited state properties of materials for frontier electronics applications, for example in organic light emitting diodes - OLEDs that have useful applications in energy efficient lighting, displays, and electronic devices. Nevertheless, the environment around active molecular units in OLEDs plays a key and often poorly understood role in determining their properties. Small changes in processing and thus the local morphology around optically active components can lead to large changes in device lifetime or performance. Treating both a molecule and its surroundings with first-principles computational modeling may be cost prohibitive with all but the lowest levels of theory (i.e., semi-empirical) because of the large system sizes. Multi-scale quantum mechanical (QM) and molecular mechanics (i.e., QM/MM) modeling holds promise but is limited by the strong sensitivity to QM region size. For excited state properties that determine OLED properties such as spectral intensity and lifetime, conventional methods, such as time-dependent density functional theory (TDDFT), may either lack predictive accuracy or be strongly sensitive to the density functional approximation (DFA) employed. A material class exemplary of the challenges and opportunities for computational tailoring of excited state properties are transition metal complex (TMC) phosphors, which are promising as phosphorescent emitters in OLEDs (PhOLEDs). The preparation of highly-efficient, blue phosphorescent complexes with long lifetimes is a formidable challenge because optimal chromophores (typically with Ir(III) or Pt(II)) must have targeted energy levels and metal–ligand bond strengths to destabilize electronic states that would lead to chemical degradation. These properties are typically characterized in organic solvents for ultimate application in thin-film OLED devices, necessitating accurate computational modeling of the solvent or organic host environment and its influence on the chromophore. The high-level theories needed to model excited state properties are too costly to also address environment effects and modeling of degradation pathways, motivating new approaches to balance this trade-off.
Our long-term goal is to advance predictive, autonomous computational modeling tools for the rational design of materials. The overall objective of the proposed work is to develop systematic methods that bring about an unprecedented level of accuracy in the study of excited state properties of molecules in the condensed phase by detecting multi-reference-derived errors at low cost, identifying the most accurate low-cost method to reproduce higher-level wavefunction theory, and developing automated approaches for adaptation to higher levels of theory when necessary. This work builds on the PI’s extensive experience in developing ML models for ground and excited states of TMCs, in devising strategies for systematic multiscale modeling, and in the development of ML models for selecting and correcting first-principles methods. The proposed work will advance modeling of excited state phenomena through:
1) Systematic multi-scale methods for detecting and adapting the QM subsystem on the fly. We will develop quantitative metrics to evaluate when spectator solvent or polymer environment components influence the excited state properties of a central chromophore to systematically adjust QM subsystems. We will develop a ML potential for high-accuracy QM(ML)/MM simulations. We will train ML models on these descriptors to predict needed changes in QM regions during excited state dynamics.
2) A recommendation system for identifying the most appropriate level of QM theory. To improve the accuracy of excited state modeling while maintaining low cost, we will develop artificial intelligence engines that detect multireference character and select active spaces when multi-reference character is high. For cases where single-reference methods are sufficient, we will develop a "recommender" of the best exchange-correlation functional to reproduce wavefunction theory or experiment.
3) Modeling of transition metal phosphorescence lifetimes and degradation pathways in the condensed phase. We will leverage the tools developed in Tasks 1 and 2 to accurately model transition metal phosphors with optimal properties for blue light emission for OLED applications to understand the mechanisms of their degradation. We will use natural language processing to extract experimental reports of phosphor properties for the benchmarking of methods developed in Tasks 1 and 2.
Expected Outcomes and Significance. The proposed work will develop new methodology and open-source software for the systematic detection of QM environmental effects on excited state properties at low cost. Beyond insights that will advance OLED design, these methods will have impact in materials modeling.
