Skip to main content

Computational Photochemistry in the Long Timescale: Sub-ns Photoprocesses in DNA

Periodic Reporting for period 2 - SubNano (Computational Photochemistry in the Long Timescale: Sub-ns Photoprocesses in DNA)

Reporting period: 2021-03-01 to 2022-08-31

Molecular excited states are at the core of a myriad of phenomena. They are crucial for Earth's biosphere, being responsible for the first step in harvesting sunlight through photosynthesis. They are equally vital for vision and bioluminescence. They play a role in mutagenesis and carcinogenesis. In technology, excited states are the basis of optoelectronics, from photovoltaics to light-emitting diodes (LEDs). In the lab, many advanced and routine analytic chemistry tools depend on them. Complementary to experiments, computational theoretical chemistry has been valuable too understand these phenomena. It allows telling details of how the electronic density is distributed in an excited molecule and how the molecule gets rid of the excess energy. Currently, available methods can successfully predict the excited-state molecular evolution within very short timescales of about a few picoseconds (1 ps corresponds to 10-12 s). Nevertheless, many crucial processes happen in scales significantly longer. SubNano aims at massively speed up dynamics simulations of excited molecules to address phenomena at the timescale of the nanosecond (1 ns = 10-9 seconds), one thousand times longer than the current simulation limits. Such methods will allow exploring processes like vibrational relaxation, fluorescence, slow internal conversion, intersystem crossing, which have been left aside by computational chemistry, too focused on ultrafast processes. The team is developing and implementing a series of methods to extend nonadiabatic dynamics simulations into the new timescale, mainly based on a novel adaptive diabatic machine learning algorithm, and a novel zero-point-corrected and vibronically-corrected mixed quantum-classical method. This methodology will be employed to investigate the long timescale nonadiabatic dynamics of photoinduced processes in nucleic acids, including DNA photostabilization via excitonic processes, biological fluorescent markers, and DNA pyrimidine-dimer repair. All novel methods will be implemented into the Newton-X software and made available for the academic community.
The Subnano project is designed to be executed in five years. The first phase of Subnano is dedicated to the development and test of the new methodology. The second phase will be dedicated to applications. We are finishing the first half, with the following achievements:

I. Method developments to speed up the dynamics and increase its reliability:
* Local-pair zero-point-energy correction (LP-ZPE), a new technique to correct zero-point energy leakage during mixed quantum-classical dynamics [1].
* Time-dependent Baeck An (TD-BA) is a method to determine nonadiabatic couplings without evaluating wavefunctions [2].
* Mixed quantum-classical dynamics with pulse ensembles (MQC-PE), a method for dynamics induced by thermal light [3].
* A new theory of the microcanonical temperature of isolated molecules [4].
* Spectrum simulations with the nuclear ensemble approach with supervised machine learning (NEA-ML) [5].

II. Implementation of scientific software.
* Newton-X NS (novel series), an entirely redesigned Newton- X version to improve performance, manage big-data, and comply with open data standards [6-7].
* Ulamdyn, a program to analyze dynamics results based on unsupervised learning techniques [8].
* New version of the PySOC program to simulate phosphorescence spectra [9].
* We have also contributed to implementing a new version of the MLatom program for atomistic machine learning [10].

III. Construction of benchmarks to critically appraise features impacting the simulations. We have analyzed:
* The conditions that should be satisfied and the algorithmic limitations of long-timescale dynamics based on ab-initio, machine-learning, and model Hamiltonian dynamics up to 1 ns [6].
* The machine-learning-potentials (MLP) accuracy, offering a guide to choosing the adequate MLP for each type of simulation[11].
* The current status and perspectives of using machine learning for excited-state simulations [12].
* The effect of the direction choice of the velocity adjustment in surface hopping [13].
* The origin of the shift between vertical excitations and band maxima to control better the comparison between theory and experiment [14].

The computer programs developed by our groups are open source and freely available for download [7-9].
All data generated by the group is available in the form of public datasets [15-20].

[1] Mukherjee; Barbatti. ChemRxiv. 2022. (DOI: 10.26434/chemrxiv-2022-53g43)
[2] do Casal, et al. Open Res Europe 2021, 1, 49. (DOI: 10.12688/openreseurope.13624.1)
[3] Barbatti. JCTC 2020, 16, 4849. (DOI: 10.1021/acs.jctc.0c00501)
[4] Barbatti. J Chem Phys 2022, accepted. (DOI: 10.26434/chemrxiv-2022-5kr7s)
[5] Xue; Barbatti; Dral. J Phys Chem A 2020, 124, 7199. (DOI: 10.1021/acs.jpca.0c05310)
[6] Mukherjee, et al. Philos Trans R Soc A 2022, 380, 20200382. (DOI: 10.1098/10.1098/rsta-2020-0382)
[7] Demoulin, et al. Newton-x ns: Novel series, 2022.
[8] Pinheiro Jr, et al. 2021.
[9] Bhati, et al. Pysoc 2, 2021.
[10] Dral, et al. Top Curr Chem 2021, 379, 27. (DOI: 10.1007/s41061-021-00339-5)
[11] Pinheiro, et al. Chem Sci 2021, 12, 14396. (DOI: 10.1039/d1sc03564a)
[12] Dral; Barbatti. Nat Rev Chem 2021, 5, 388. (DOI: 10.1038/s41570-021-00278-1)
[13] Barbatti. JCTC 2021, 17, 3010. (DOI: 10.1021/acs.jctc.1c00012)
[14] Bai, et al. J Mol Model 2020, 26, 107. (DOI: 10.1007/s00894-020-04355-y)
[15] Pinheiro Jr; Mukherjee; Barbatti. Dataset 2021. (DOI: 10.1098/rsta-2020-0382)
[16] Pinheiro Jr; Mukherjee; Barbatti. Dataset 2021. (DOI: 10.6084/m9.figshare.14873094.v1)
[17] Pinheiro Jr; Mukherjee; Barbatti. Dataset 2021. (DOI: 10.6084/m9.figshare.14823126.v1)
[18] Casal, et al. Dataset. 2021. (DOI: 10.6084/m9.figshare.14446998.v1)
[19] Barbatti. Dataset. 2021. (DOI: 10.6084/m9.figshare.14010311.v1)
[20] Barbatti. Dataset. 2021. (DOI: 10.6084/m9.figshare.13522856.v2)
To this point of the SubNano project, I can highlight the following main achievements:

* Development of Newton-X NS [1-2]: this new version of the program can be 1800 times faster than the classical Newton-X, overcoming all our expectations.

* Development of the LP-ZPE correction [3]. For decades, the ZPE leakage has been a problem haunting classical molecular dynamics simulations. Every previous attempt to solve it involved the evaluation of Hessian matrices during the propagation, which is unaffordable even in short timescales when using on-the-fly computation of electronic properties. Our method is a game-change: it corrects the ZPE leakage using a collisional model based on local vibrations, with minimum computational costs and interference in the dynamics. We have shown for a test case that dissociation of water dimers induced by ZPE leakage is fully inhibited in 20 ps ab-initio dynamics when using LP-ZPE corrections.

* Development of dynamics induced by thermal light with MQC-PE [4]. This methodology has two striking features. First, it is the only technique that allows such simulations with full nuclear dimensionality. Second, the post-processing strategy to include quantum effects in pre-computed simulations is extremely promising. We are exploring it now to include other types of quantum effects in dynamics.

[1] Mukherjee, et al. Philos Trans R Soc A 2022, 380, 20200382. (DOI: 10.1098/10.1098/rsta-2020-0382)
[2] Demoulin, et al. Newton-x ns: Novel series, 2022.
[3] Mukherjee; Barbatti. ChemRxiv. 2022. (DOI: 10.26434/chemrxiv-2022-53g43)
[4] Barbatti. JCTC 2020, 16, 4849. (DOI: 10.1021/acs.jctc.0c00501)
Subnano logo