This Novel Quantum Method May Help Discover New Materials for OLEDs
By Qi Gao, Senior Chief Scientist at the Materials Design Laboratory of the Mitsubishi Chemical Corporation Science & Innovation Center; Gavin Jones, Research Staff Member — Quantum Applications at IBM Quantum; and Naoki Yamamoto, Chair of the Keio University Quantum Computing Center.
Organic luminescent materials — that is, carbon-based materials that can spontaneously emit light without heat — are garnering attention from researchers across industry and academia. These materials may prove useful for a variety of applications, including OLED displays, solar energy, catalysts for certain chemical reactions, and more. However, several serious challenges stand in the way of organic luminescent materials and their commercialization.
Luminescence consists of the photonic excitation of a material’s electrons to a higher energy level followed by the relaxation of those electrons into lower energy levels. One strategy for improving materials’ luminescence is improving their emission quantum efficiency — a ratio of the desirable radiative relaxation processes to all of the relaxation processes including radiative and undesirable non-radiative relaxation. We hope to improve the emission quantum efficiency of organic luminescent materials by lowering their non-radiative relaxation rates with the help of quantum computation.
Previous studies have demonstrated that replacing hydrogen atoms in organic materials with deuterium — that is, isotopes of hydrogen with one, rather than zero neutrons in their nuclei — can help lower vibrational factors that contribute to the non-radiative processes, and improve the luminescent performance of organic materials. In order to accelerate the design of “deuterated” organic emitters with high emission quantum efficiencies, it would be useful to have a computational method that predicts the best places in the molecule to swap hydrogen atoms with deuterium in order to dampen those vibrations while still proposing synthesizable molecules. Some simulations do exist today, but these are computationally demanding.
Machine learning models could potentially design molecules with the desired properties orders of magnitude faster than other computational methods, and quantum optimization with variational algorithms such as the ubiquitous VQE or QAOA methods might further aid those machine learning models. Therefore, in our recent study, we developed a combined quantum chemistry, machine learning, and quantum optimization workflow to discover the optimal deuterated version of a well-known synthesizable OLED emitter called Alq₃.
Alq₃, short for tris(8-hydroxyquinolinato)aluminium, is a molecule consisting of aluminum bonded to three molecules called 8-hydroxyquinolines. Each of these molecules possess six hydrogen atoms to potentially replace with deuterium. We can represent this problem using a schema in which each hydrogen atom has a bit value of 0 and each replacement deuterium atom has a bit value of 1. We can then treat each of these molecules as a six-qubit bitstring and find the optimal value of this bit string with the help of a quantum computer and other classical computational chemistry methods in our arsenal. Optimizing this bitstring should let us find deuterated Alq₃ molecules with high quantum efficiencies.
Our method begins with a classical computing component: first, we perform the necessary classical quantum chemistry calculations to find the quantum efficiencies for a set of deuterated Alq₃ molecules. Specifically, we are calculating factors that describe non-radiative decay for various deuterated Alq₃ molecules using various approximation methods. We then use these results to generate two datasets: the first is a training dataset that trains a machine learning model for predicting quantum efficiency, and the second is a test dataset used to validate the performance of the trained model by comparing the model’s predicted value to the values obtained using the test data. We repeat this procedure with additional quantum chemistry results until traditional machine learning techniques can satisfactorily make predictions with a high degree of accuracy.
We then use this machine learning model to construct a Hamiltonian for the system, on which we perform an optimization algorithm using either VQE or QAOA to find the optimal deuterated Alq₃ molecules. You can read about these algorithms in the Qiskit textbook here.
We found that our calculations employing both the VQE and QAOA algorithms using Qiskit’s statevector simulator could find the optimal deuterated Alq₃ molecule with better than 0.95 probability. However, it’s important to note that today’s quantum processors are noisy, and that the noise can limit the accuracy of the values we calculate. In the presence of noise, the probability of finding the optimal deuterated bitstring decreases to 0.83 for VQE and 0.075 for QAOA, with error mitigation techniques only marginally helping to improve either result. However, we were able to improve these probabilities by employing a binary search algorithm to find the correct output binary value of each qubit even in the presence of noise.
Thinking beyond Alq₃, we think that our method is a promising proof-of-concept that could aid in the design of materials used in OLEDs, and could help find the optimal deuteration for other molecule complexes. More broadly, we think that this combination of classical quantum chemistry, machine learning, and quantum computation could provide chemists with a new tool in their arsenal for material discovery.
You can read more about our methods in our paper here, and learn more about using Qiskit for chemistry applications here.
