Machine Learning Photodynamics Unveils a Controlled H${}_2$ Loss Channel in Methaniminium Cation

Abstract

The methaniminium cation, CH${}_2$NH${}_2^{+}$, plays an important role in Titan's N${}_2$--CH${}_4$ atmospheric chemistry. As the simplest protonated Schiff base (PSB), it also serves as a model for studying the nonadiabatic dynamics of retinal PSB, the chromophore central to vertebrate vision. While previous studies have established CN bond cleavage and photoisomerization as the primary pathways in the photochemistry of CH${}_2$NH${}_2^{+}$, we now report a new UV-induced photochemical pathway to HCNH${}^{+}$, the dominant ion in Titan's upper atmosphere. Through high-level XMCQDPT2 and CASSCF(12,12) calculations, we identify a novel S${}_1$/S${}_0$ conical intersection that mediates the concerted double H-atom elimination from the carbon center of CH${}_2$NH${}_2^{+}$, yielding carbene CNH${}_2^{+}$ as a direct precursor to HCNH${}^{+}$. On-the-fly trajectory surface hopping dynamics confirm the presence of direct H${}_2$ loss following excitation to either the S${}_2$ or S${}_1$ state. Furthermore, our large-scale, machine learning-accelerated simulations reveal that mode-specific pre-excitation can selectively funnel the dynamics into this new channel via the vibronically allowed S${}_1$ state, enabling targeted control of the photochemical outcome.

13 pages, 5 figures

References in corpus (8)

• Machine learning for electronically excited states of molecules
• MACE: Higher Order Equivariant Message Passing Neural Networks for Fast and Accurate Force Fields
• Combining SchNet and SHARC: The SchNarc machine learning approach for excited-state dynamics
• Machine learning enables long time scale molecular photodynamics simulations
• Inclusion of machine learning kernel ridge regression potential energy surfaces in on-the-fly nonadiabatic molecular dynamics simulation
• The Composition and Chemistry of Titan's Atmosphere
• MLatom software ecosystem for surface hopping dynamics in Python with quantum mechanical and machine learning methods
• Physics-informed active learning for accelerating quantum chemical simulations

Cited by in corpus (1)

• XMCQDPT2-Fidelity Transfer-Learning Potentials and a Wavepacket Oscillation Model with Power-Law Decay for Ultrafast Photodynamics