Speaker
Description
Nano-porous graphene (NPG) offers great potential across a range of applications, from electronics to photocatalysis. Its electronic properties, e.g., the band gap, can be tuned by changing structural parameters alone[1]. In order to optimize NPG for photo-physical and photo-chemical applications, their excited-state properties need to be studied. The method of choice for studying dynamic excited-state properties is often non-adiabatic molecular dynamics (NAMD). However, conventional NAMD relying on ab-inito methods to describe ground- and excited-state potential energy surfaces is computationally expensive, particularly for periodic systems. Employing machine-learning methods can significantly reduce the computational cost of NAMD without compromising accuracy [2], but may come with their own challenges for periodic systems.
In our work, we trained machine-learning interatomic potentials for the ground state and the five lowest excited singlet states for a specific NPG. We used these potentials in combination with Landau-Zener surface hopping [3] to run NAMD simulations for the NPG. Our findings show that the number of excited states included in the NAMD simulations affects the relaxation to the ground state. Furthermore, our NAMD simulations reveal the limitations of the chosen approach to NAMD simulations for periodic systems. In this contribution, we will discuss the advantages of the approach employed in this work as well as its limitations.
References:
[1] (a) F. Crasto de Lima, A. Fazzio, Phys. Chem. Chem. Phys. 23 (2021) 11501-11506; (b) D. Wang, X. Lu, Arramel, M. Yang, J. Wu, A. T. S. Wee, Small 17 (2021) 2102246; (c) B. Kretz, I. Lončarić, Inorg. Chem. 64 (2025) 11022-11031.
[2] J. Li, S. A. Lopez, Chem. Phys. Rev. 4 (2023) 031309
[3] (a) A. K. Belyaev, C. Lasser, G. Trigila, J. Chem. Phys. 140 (2014) 224108; (b) J. Suchan, J. Janoš, P. Slavíček, J. Chem. Theory Comput. 16 (2020) 5809–5820.