arXiv subjects

Markus Meuwly

Publications and source records attributed to Markus Meuwly.

At least 91 records · Page 5Linked to original sources

arXiv2021

Impact of the characteristics of quantum chemical databases on machine learning predictions of tautomerization energies

An essential aspect for adequate predictions of chemical properties by machine learning models is the database used for training them. However, studies that analyze how the content and structure of the databases used for training impact the prediction quality are scarce. In this work, we analyze and quantify the relationships learned by a machine learning model (Neural Network) trained on five different reference databases (QM9, PC9, ANI-1E, ANI-1 and ANI-1x) to predict tautomerization energies from molecules in Tautobase. For this, characteristics such as the number of heavy atoms in a molecule, number of atoms of a given element, bond composition, or initial geometry on the quality of the predictions are considered. The results indicate that training on a chemically diverse database is crucial for obtaining good results but also that conformational sampling can partly compensate for limited coverage of chemical diversity. We explicitly demonstrate that when certain types of bonds need to be covered in the target database (Tautobase) but are undersampled in the reference databases the resulting predictions are poor. A quantitative measure for these deficiencies is the Kullback-Leibler divergence between reference and target distributions. Analysis of the results with a TreeMAP algorithm provides deeper understanding of specific deficiencies in the reference data sets. Capitalizing on this information can be used to either improve existing databases or to generate new databases of sufficient diversity for a range of ML applications in chemistry.

physics.chem-ph↗

arXiv2021

MP2 Is Not Good Enough: Transfer Learning ML Models for Accurate VPT2 Frequencies

The calculation of the anharmonic modes of small to medium sized molecules for assigning experimentally measured frequencies to the corresponding type of molecular motions is computationally challenging at sufficiently high levels of quantum chemical theory. Here, a practical and affordable way to calculate coupled-cluster quality anharmonic frequencies using second order vibrational perturbation theory (VPT2) from machine-learned models is presented. The approach, referred to as "NN + VPT2", uses a high-dimensional neural network (PhysNet) to learn potential energy surfaces (PESs) at different levels of theory from which harmonic and VPT2 frequencies can be efficiently determined. The NN + VPT2 approach is applied to eight small to medium sized molecules (H$_2$CO, trans-HONO, HCOOH, CH$_3$OH, CH$_3$CHO, CH$_3$NO$_2$, CH$_3$COOH and CH$_3$CONH$_2$) and frequencies are reported from NN-learned models at the MP2/aug-cc-pVTZ, CCSD(T)/aug-cc-pVTZ and CCSD(T)-F12/aug-cc-pVTZ-F12 levels of theory. For the largest molecules and at the highest levels of theory, transfer learning (TL) is used to determine the necessary full-dimensional, near-equilibrium PESs. Overall, NN + VPT2 yields anharmonic frequencies to within 20 cm$^{-1}$ of experimentally determined frequencies for close to 90 % of the modes for the highest quality PES available and to within 10 cm$^{-1}$ for more than 60 % of the modes. For the MP2 PESs only around 60 % of the NN + VPT2 frequencies were within 20~cm$^{-1}$ of the experiment, with outliers up to 150 cm$^{-1}$ compared with experiment. It is also demonstrated that the approach allows to provide correct assignments for strongly interacting modes such as the OH bending and the OH torsional modes in formic acid monomer and the CO-stretch and OH-bend mode in acetic acid.

physics.chem-ph↗

arXiv2021

The C($^3$P) + O$_2$($^3 Σ_g^-$) $\leftrightarrow$ CO$_2$ $\leftrightarrow$ CO($^1 Σ^+$)+ O($^1$D)/O($^3$P) Reaction: Thermal and Vibrational Relaxation Rates from 15 K to 20000 K

Thermal rates for the C($^3$P) + O$_2$($^3 Σ_g^-$) $\leftrightarrow$ CO($^1 Σ^+$)+ O($^1$D)/O($^3$P) reaction are investigated over a wide temperature range based on quasi classical trajectory (QCT) simulations on 3-dimensional, reactive potential energy surfaces (PESs) for the $^1$A$'$, $(2)^1$A$'$, $^1$A$''$, $^3$A$'$ and $^3$A$''$ states. The forward rate matches measurements at 15 K to 295 K whereas the equilibrium constant determined from the forward and reverse rates are consistent with those derived from statistical mechanics at high temperature. Vibrational relaxation, O+CO($ν=1, 2$) $\rightarrow$ O+CO$(ν=0)$, is found to involve both, non-reactive and reactive processes. The contact time required for vibrational relaxation to take place is $τ\geq 150$ fs for non-reacting and $τ\geq 330$ fs for reacting (oxygen atom exchange) trajectories and the two processes are shown to probe different parts of the global potential energy surface. In agreement with experiments, low collision energy reactions for the C($^3$P) + O$_2$($^3 Σ_g^-$, $v=0$) $\rightarrow$ CO($^1 Σ^+$)+ O($^1$D) lead to CO($^{1}Σ^{+}$, $v'=17$) with an onset at $E_{\rm c} \sim 0.15$ eV, dominated by the $^1$A$'$ surface with contributions from the $^3$A$'$ surface. Finally, the barrier for the CO$_{\rm A}$($^1Σ^+$) + O$_{\rm B}$($^3$P) $\rightarrow$ CO$_{\rm B}$($^{1}Σ^{+}$) + O$_{\rm A}$($^3$P) atom exchange reaction on the $^3$A$'$ PES yields a barrier of $\sim 7$ kcal/mol (0.300 eV), consistent with an experimentally reported value of 6.9 kcal/mol (0.299 eV).

physics.chem-ph↗

arXiv2021

Site-Selective Dynamics of Azidolysozyme

The spectroscopic response of and structural dynamics around all azido-modified alanine residues (AlaN$_3$) in Lysozyme is characterized. It is found that AlaN$_3$ is a positionally sensitive probe for the local dynamics, covering a frequency range of $\sim 15$ cm$^{-1}$ for the center frequency of the line shape. This is consistent with findings from selective replacements of amino acids in PDZ2 which reported a frequency span of $\sim 10$ cm$^{-1}$ for replacements of Val, Ala, or Glu by azidohomoalanine (AHA). For the frequency fluctuation correlation functions (FFCFs) the long-time decay constants $τ_2$ range from $\sim 1$ to $\sim 10$ ps which compares with experimentally measured correlation times of 3 ps. Attaching azide to alanine residues can yield dynamics that decays to zero on the few ps time scale (i.e. static component $Δ_0 \sim 0$ ps$^{-1}$) or to a remaining, static contribution of $\sim 0.5$ ps$^{-1}$ (corresponding to 2.5 cm$^{-1}$), depending on the local environment on the 10 ps time scale. The magnitude of the static component correlates qualitatively with the degree of hydration of the spectroscopic probe. Although attaching azide to alanine residues is found to be structurally minimally invasive with respect to the overall protein structure, analysis of the local hydrophobicity indicates that the hydration around the modification site differs for modified and unmodified alanine residues, respectively.

physics.bio-ph↗

arXiv2020

Spectroscopy, Dynamics and Hydration of S-Nitrosylated Myoglobin

S-nitrosylation, the covalent addition of NO to the thiol side chain of cysteine, is an important post-transitional modification that can alter the function of various proteins. The structural dynamics and vibrational spectroscopy of S-nitrosylation in the condensed phase is investigated for the methyl-capped cysteine model system and for myoglobin. Using conventional point charge and physically more realistic multipolar force fields for the -SNO group it is found that the SN- and NO-stretch and the SNO-bend vibrations can be located and distinguished from the other protein modes for simulations of MbSNO at 50 K. The finding of stable cis- and trans-MbSNO is consistent with experiments on other proteins as is the observation of buried -SNO. For MbSNO the observed relocation of the EF loop in the simulations by $\sim 3$ Å\/ is consistent with the available X-ray structure and the conformations adopted by the -SNO label are in good overall agreement with the X-ray structure. Despite the larger size of the -SNO group, MbSNO is found to recruit more water molecules within 10 Å\/ of the modification site than WT Mb due to the stronger electrostatics. Similarly, when comparing the hydration between the A- and H-helices they differ by up to 30 \% between WT and MbSNO. This suggests that local hydration can also be significantly modulated through nitrosylation.

physics.chem-ph↗

arXiv2020

Polarizable Multipolar Molecular Dynamics Using Distributed Point Charges

Distributed point charge models (DCM) and their minimal variants (MDCM) have been integrated with tools widely used for condensed-phase simulations, including a virial-based barostat and a slow-growth algorithm for thermodynamic integration. Minimal DCM is further developed with a systematic approach to reduce fitting errors in the electrostatic interaction energy and a new fragment-based approach offers considerable speedup of the MDCM fitting process for larger molecules with increased numbers of off-centered charged sites. Finally, polarizable (M)DCM is also introduced in the present work. The developments are used in condensed-phase simulations of popular force fields with commonly applied simulation conditions. (M)DCM equivalents for a range of widely used water force fields and for fluorobenzene (PhF) are developed and applied along with the original models to evaluate the impact of reformulating the electrostatic term. Comparisons of the molecular electrostatic potential (MEP), electrostatic interaction energies, and bulk properties from molecular dynamics simulations for a range of models from simple TIP$n$P ($n = 3-5$) to the polarizable, multipolar iAMOEBA models for water and an existing quadrupolar model for PhF confirm that DCMs retain the accuracy of the original models, providing a homogeneous, efficient, and generic point charge alternative to a multipolar electrostatic model for force field development and multilevel simulations.

physics.chem-ph↗

arXiv2020

Reactive atomistic simulations of Diels-Alder-type reactions: Conformational and dynamic effects in the polar cycloaddition of 2,3-dibromobutadiene radical ions with maleic anhydride

The kinetics, dynamics and conformational specificities for the ionic Diels-Alder reaction (polar cycloaddition) of maleic anhydride with 2,3-dibromobutadiene radical ions have been studied theoretically using multisurface adiabatic reactive molecular dynamics. A competition of concerted and stepwise reaction pathways was found and both the s-cis and s-trans conformers of of the diene are reactive. The analysis of the minimum dynamic path of the reaction indicates that both, rotations and vibrations of the reactant molecules are important for driving the system towards the transition state. The rates were computed as $k = 5.1 \times 10^{-14}$ s$^{-1}$ for the s-cis and $k = 3.8 \times 10^{-14}$ s$^{-1}$ for the s-trans conformer of 2,3-dibromobutadiene at an internal temperature of 300 K. The present results are to be contrasted with the neutral variant of the title system in which only the gauche conformer of the diene was found to undergo a considerably slower, concerted and mostly synchronous reaction driven by the excitation of rotations. The results presented here inform detailed experimental studies of the dynamics of polar cycloadditions under single-collision conditions in the gas phase.

physics.chem-ph↗

arXiv2020

The Dynamics and Infrared Spectrocopy of Monomeric and Dimeric Wild Type and Mutant Insulin

The infrared spectroscopy and dynamics of -CO labels in wild type and mutant insulin monomer and dimer are characterized from molecular dynamics simulations using validated force fields. It is found that the spectroscopy of monomeric and dimeric forms in the region of the amide-I vibration differs for residues B24-B26 and D24-D26, which are involved in dimerization of the hormone. Also, the spectroscopic signatures change for mutations at position B24 from phenylalanine - which is conserved in many organisms and known to play a central role in insulin aggregation - to alanine or glycine. Using three different methods to determine the frequency trajectories - solving the nuclear Schrödinger equation on an effective 1-dimensional potential energy curve, instantaneous normal modes, and using parametrized frequency maps - lead to the same overall conclusions. The spectroscopic response of monomeric WT and mutant insulin differs from that of their respective dimers and the spectroscopy of the two monomers in the dimer is also not identical. For the WT and F24A and F24G monomers spectroscopic shifts are found to be $\sim 20$ cm$^{-1}$ for residues (B24 to B26) located at the dimerization interface. Although the crystal structure of the dimer is that of a symmetric homodimer, dynamically the two monomers are not equivalent on the nanosecond time scale. Together with earlier work on the thermodynamic stability of the WT and the same mutants it is concluded that combining computational and experimental infrared spectroscopy provides a potentially powerful way to characterize the aggregation state and dimerization energy of modified insulins.

physics.bio-ph↗

arXiv2020

ML Models of Vibrating H$_2$CO: Comparing Reproducing Kernels, FCHL and PhysNet

Machine Learning (ML) has become a promising tool for improving the quality of atomistic simulations. Using formaldehyde as a benchmark system for intramolecular interactions, a comparative assessment of ML models based on state-of-the-art variants of deep neural networks (NN), reproducing kernel Hilbert space (RKHS+F), and kernel ridge regression (KRR) is presented. Learning curves for energies and atomic forces indicate rapid convergence towards excellent predictions for B3LYP, MP2, and CCSD(T)-F12 reference results for modestly sized (in the hundreds) training sets. Typically, learning curve off-sets decay as one goes from NN (PhysNet) to RKHS+F to KRR (FCHL). Conversely, the predictive power for extrapolation of energies towards new geometries increases in the same order with RKHS+F and FCHL performing almost equally. For harmonic vibrational frequencies, the picture is less clear, with PhysNet and FCHL yielding respectively flat learning at $\sim$ 1 and $\sim$ 0.2 cm$^{-1}$ no matter which reference method, while RKHS+F models level off for B3LYP, and exhibit continued improvements for MP2 and CCSD(T)-F12. Finite-temperature molecular dynamics (MD) simulations with the same initial conditions yield indistinguishable infrared spectra with good performance compared with experiment except for the high-frequency modes involving hydrogen stretch motion which is a known limitation of MD for vibrational spectroscopy. For sufficiently large training set sizes all three models can detect insufficient convergence (``noise'') of the reference electronic structure calculations in that the learning curves level off. Transfer learning (TL) from B3LYP to CCSD(T)-F12 with PhysNet indicates that additional improvements in data efficiency can be achieved.

physics.chem-ph↗

arXiv2020

Water Dynamics Around Proteins: T- and R-States of Hemoglobin and Melittin

The water dynamics, as characterized by the local hydrophobicity (LH), is investigated for tetrameric hemoglobin and dimeric melittin. For the T0 to R0 transition in Hb it is found that LH provides additional molecular-level insight into the Perutz mechanism, i.e., the breaking and formation of salt bridges at the alpha1 / beta2 and alpha2 / beta1 interface is accompanied by changes in LH. For Hb in cubic water boxes with 90 Aengstroem and 120 Aengstroem edge length it is observed that following a decrease in LH as a consequence of reduced water density or change of water orientation at the protein/water interface the alpha / beta interfaces are destabilized; this is a hallmark of the Perutz stereochemical model for the T to R transition in Hb. The present work thus provides a dynamical view of the classical structural model relevant to the molecular foundations of Hb function. For dimeric melittin, earlier results by Cheng and Rossky (Nature, 1998, 392, 696-699) are confirmed and interpreted on the basis of LH from simulations in which the protein structure is frozen. For the flexible melittin dimer the changes in the local hydration can be as much as 30 % than for the rigid dimer, reflecting the fact that protein and water dynamics are coupled.

physics.chem-ph↗

arXiv2020

Machine Learning for Observables: Reactant to Product State Distributions for Atom-Diatom Collisions

Machine learning-based models to predict product state distributions from a distribution of reactant conditions for atom-diatom collisions are presented and quantitatively tested. The models are based on function-, kernel- and grid-based representations of the reactant and product state distributions. While all three methods predict final state distributions from explicit quasi-classical trajectory simulations with R$^2$ > 0.998, the grid-based approach performs best. Although a function-based approach is found to be more than two times better in computational performance, the kernel- and grid-based approaches are preferred in terms of prediction accuracy, practicability and generality. The function-based approach also suffers from lacking a general set of model functions. Applications of the grid-based approach to nonequilibrium, multi-temperature initial state distributions are presented, a situation common to energy distributions in hypersonic flows. The role of such models in Direct Simulation Monte Carlo and computational fluid dynamics simulations is also discussed.

physics.chem-ph↗

arXiv2020

Permutationally Invariant, Reproducing Kernel-Based Potential Energy Surfaces for Polyatomic Molecules: From Formaldehyde to Acetone

Constructing accurate, high dimensional molecular potential energy surfaces (PESs) for polyatomic molecules is challenging. Reproducing Kernel Hilbert space (RKHS) interpolation is an efficient way to construct such PESs. However, the scheme is most effective when the input energies are available on a regular grid. Thus the number of reference energies required can become very large even for penta-atomic systems making such an approach computationally prohibitive when using high-level electronic structure calculations. Here an efficient and robust scheme is presented to overcome these limitations and is applied to constructing high dimensional PESs for systems with up to 10 atoms. Using energies as well as gradients reduces the number of input data required and thus keeps the number of coefficients at a manageable size. Correct implementation of permutational symmetry in the kernel products is tested and explicitly demonstrated for the highly symmetric CH$_4$ molecule.

physics.chem-ph↗

arXiv2020

N$_3^+$: Full-Dimensional Potential Energy Surface, Vibrational Energy Levels and Ground State Dynamics

The fundamentals and higher vibrationally excited states for the N$_3^+$ ion in its electronic ground state have been determined from quantum bound state calculations on 3-dimensional potential energy surfaces (PESs) at the CCSD(T)-F12 and MRCI+Q levels of theory. The vibrational fundamentals are at 1130 cm$^{-1}$ ($ν_1$, symmetric stretch), 807 cm$^{-1}$ ($ν_3$, asymmetric stretch), and 406 cm$^{-1}$ ($ν_2$, bend) on the higher-quality CCSD(T)-F12 surface. For $ν_1$, the calculations are close to the estimated frequency from experiment (1170 cm$^{-1}$) and previous calculations\cite{rosmus.n3:1994} which find it at 1190 cm$^{-1}$. Calculations of the vibrational states on the MRCI+Q PES are in qualitative agreement with those using the CCSD(T)-F12 PES. Analysis of the reference CASSCF wave function for the MRCI+Q calculations provides further insight into the shape of the PES and lends support for the reliability of Hartree-Fock as the reference wave function for the coupled cluster calculations. According to this, N$_3^+$ has mainly single reference character in all low-energy regions of its electronic ground state ($^3$A$''$) 3d PES.

physics.chem-ph↗

arXiv2020

Isomerization and Decomposition Reactions of Acetaldehyde Relevant to Atmospheric Processes from Dynamics Simulations on Neural Network-Based Potential Energy Surfaces

Acetaldehyde (AA) isomerization (to vinylalcohol, VA) and decomposition (into either CO+CH$_4$ and H$_2$+H$_2$CCO) is studied using a fully dimensional, reactive potential energy surface represented as a neural network (NN). The NN, trained on 432'399 reference structures from MP2/aug-cc-pVTZ calculations has a MAE of 0.0453 kcal/mol and an RMSE of 1.186 kcal/mol for a test set of 27'399 structures. For the isomerization process AA $\rightarrow$ VA the minimum dynamical path implies that the C-H vibration, and the C-C-H (with H being the transferring H-atom) and the C-C-O angles are involved to surmount the 68.2 kcal/mol barrier. Using an excess energy of 93.6 kcal/mol - the energy available in the solar spectrum and sufficient to excite to the first electronically excited state - to initialize the molecular dynamics, no isomerization to VA is observed on the 500 ns time scale. Only with excess energies of $\sim$ 127.6 kcal/mol (including the zero point energy of the AA molecule), isomerization occurs on the nanosecond time scale. Given that collisional de-excitation at atmospheric conditions in the stratosphere occurs on the 100 ns time scale, it is concluded that formation of VA following photoexcitation of AA from actinic photons is unlikely. This also limits the relevance of this reaction pathway to be a source for formic acid.

physics.chem-ph↗

arXiv2020

Dynamics on Multiple Potential Energy Surfaces: Quantitative Studies of Elementary Processes Relevant to Hypersonics

The determination of thermal and vibrational relaxation rates of triatomic systems suitable for application in hypersonic model calculations is discussed. For this, potential energy surfaces for ground and electronically excited state species need to be computed and represented with high accuracy and quasiclassical or quantum nuclear dynamics simulations provide the basis for determining the relevant rates. These include thermal reaction rates, state-to-state cross-sections, or vibrational relaxation rates. For exemplary systems - [NNO], [NOO], and [CNO] - all individual steps are described and a literature overview for them is provided. Finally, as some of these quantities involve considerable computational expense, for the example of state-to-state cross sections the construction of an efficient model based on neural networks is discussed. All such data is required and being used in more coarse-grained computational fluid dynamics simulations.

physics.chem-ph↗

arXiv2020

Thermal Activation of Methane by MgO$^+$: Temperature Dependent Kinetics, Reactive Molecular Dynamics Simulations and Statistical Modeling

The kinetics of MgO$^+$ + CH$_4$ was studied experimentally using the variable ion source, temperature adjustable selected ion flow tube (VISTA-SIFT) apparatus from 300 $-$ 600 K and computationally by running and analyzing reactive atomistic simulations. Rates and product branching fractions were determined as a function of temperature. The reaction proceeded with a rate of $k = 5.9 \pm 1.5 10^{-10}(T/300 $ K$)^{-0.5 \pm 0.2}$ cm$^3$ s$^{-1}$. MgOH$^+$ was the dominant product at all temperatures, but Mg$^+$, the co-product of oxygen-atom transfer to form methanol, was observed with a product branching fraction of $0.08 \pm 0.03 (T / 300 $ K$)^{-0.8 \pm 0.7}$. Reactive molecular dynamics simulations using a reactive force field, as well as a neural network yield rate coefficients about one order of magnitude lower. This underestimation of the rates is traced back to the multireference character of the transition state [MgOCH$_4$]$^+$. Statistical modeling of the temperature-dependent kinetics provides further insight into the reactive potential surface. The rate limiting step was found to be consistent with a four-centered activation of the C-H bond, consistent with previous calculations. The product branching was modeled as a competition between dissociation of an insertion intermediate directly after the rate-limiting transition state, and traversing a transition state corresponding to a methyl migration leading to a Mg-CH$_3$OH$^+$ complex, though only if this transition state is stabilized significantly relative to the dissociated MgOH$^+$ + CH$_3$ product channel. An alternative non-statistical mechanism is discussed, whereby a post-transition state bifurcation in the potential surface could allow the reaction to proceed directly from the four-centered TS to the Mg-CH$_3$OH$^+$ complex thereby allowing a more robust competition between the product channels.

physics.chem-ph↗

arXiv2020

Accurate Reproducing Kernel-Based Potential Energy Surfaces for the Triplet Ground States of N$_2$O and Dynamics for the N+NO$\leftrightarrow$O+N$_2$ and N$_2$+O$\rightarrow$2N+O Reactions

Accurate potential energy surfaces (PESs) have been determined for the $^3$A$'$ and $^3$A$''$ states of N$_2$O using electronic structure calculations at the multireference configuration interaction level with Davidson correction (MRCI+Q) and the augmented Dunning-type correlation consistent polarize triple zeta (aug-cc-pVTZ) basis set. More than 20000 MRCI+Q/aug-cc-pVTZ energies are represented using a reproducing kernel Hilbert space (RKHS) interpolation scheme. The RKHS PESs successfully describe all reactant channels with high accuracy. The analytical PESs are characterized by computing the minima and transition states on it. Quasiclassical dynamics simulations are then used to determine thermal and vibrational relaxation rates for the N+NO and O+N$_2$ collisions. The agreement between results obtained from the simulations and from available experiments is favourable for both types of observables, which provides a test for the accuracy of the PESs. The PESs can be used to calculate more detailed state-to-state observables relevant for applications to hypersonic reentry.

physics.chem-ph↗

arXiv2019

Formation and Stabilization of Ground and Excited State Singlet O$_2$ upon Recombination of $^3$P Oxygen on Amorphous Solid Water

The recombination dynamics of $^3$P oxygen atoms on cold amorphous solid water to form triplet and singlet molecular oxygen (O$_2$) is followed under conditions representative for cold clouds. It is found that both, formation of ground state ($X ^3 Σ_{g}^{-}$) O$_2$ and molecular oxygen in the two lowest singlet states ($a ^1Δ_g$ and $b ^1Σ_g^+$) is possible and that the species can stabilize. The relative proportions of the species is approximately 1:1:1. These results also agree qualitatively with a kinetic model based on simplified wavepacket simulations. As the chemical reactivity of triplet and singlet O$_2$ is different it is likely that substantial amounts of $a ^1Δ_g$ and $b ^1Σ_g^+$ oxygen influences the chemical evolution of cold clouds.

physics.chem-ph↗