Thermochemistry is central to predicting and modeling the properties and behavior of matter. Some current challenges for accurate prediction are creating thermodynamically consistent datasets for a large number of chemical species, and incorporating sufficiently rigorous calculations for entropy to allow reliable application of thermodynamic data under a wide temperature range. In this part of the ECC project we
- Develop Active Thermochemical Tables (ATcT) into a public and interactive database while preserving a strict control over accuracy, and incorporate into automated chemical mechanism calculations
- Enable the routine inclusion of anharmonic effects for gas-phase species, crucial for accurate entropy
- Create a suite of methods implemented into Atomic Simulation Environment (ASE) to compute the partition function for the coupled, anharmonic motion of adsorbates relative to the catalyst surface
- Branko Ruscic (ANL)
- Senior researchers:
- David H. Bross (ANL)
- C. Franklin Goldsmith (Brown)
- Khachik Sargsyan (SNL)
- Judit Zádor (SNL)
- Students and postdocs:
- Katrin Blondal (Brown)
AdTherm will be available on github soon. The package will be linked through the Atomic Simulation Environment, ASE.
The current ATcT database.
Due to the central role of thermodynamics, there is a long history of efforts to compile the available thermodynamic data into tabulations for broader use by the scientific community. These efforts, the most recent occurring over 25 years ago, were largely successful compiling the thermodynamic data available at the time, but the resulting static tabulations cannot be updated with new information without introducing inconsistencies. Furthermore, the traditional sequential thermochemistry approach used to generate the tabulations usually selects a single “best” determination from the available information, while discarding all of the remaining information.
The Active Thermochemical Tables (ATcT) approach, developed at Argonne, instead uses all the available information and allows insertion of new thermochemical data from theory or experiment without loss of consistency. Our approach subsequently reevaluates and improves the underlying Thermochemical Network (TN), and produces thermochemical information consistent with all input information (data and error bars). The ATcT approach objectively evaluates the data, and validates the entire dataset against itself. The ATcT approach has been repeatedly hailed in scientific literature as the greatest advance in thermochemistry in decades, revolutionizing both the accuracy and reliability of thermochemical data, and replaced and made obsolete traditional thermochemical tabulations.135-137 This approach ensures that as more species and more determinations thereof are added, the underlying TN becomes more complete and the results improve.
Developments under ECC
Currently ATcT users are limited to viewing the 0 K and 298 K heats of formation for the available species. Under ECC we create an accessible scientific database to permit collaborators to directly access ATcT, while also greatly expanding its focus. This development will help generalize ATcT to all thermochemistry fields, including materials science, atmospheric chemistry, and condensed phase chemistry, and will subsequently improve the results of thermochemistry-dependent models in those fields. This work will improve the mechanism by which thermodynamic information is validated, used, and presented to the community at large, such that the entire community can utilize the best available information. We will host user specific TNs that allow researchers to grant controlled access to their data while also enabling eventual public release to open access databases. This will create a seamless integration of the public and private databases as articles become published and individual researcher data is contributed to the public repositories. In this way, a community database, similar to other successful open-access and community driven databases can develop for thermodynamics.
Anharmonic partition function tools for gas-phase species
The ubiquitous rigid rotor harmonic oscillator (RRHO) partition functions are not sufficiently accurate relative to the high accuracy composite thermochemistry methods that have become routinely capable for small molecules. Accurate partition functions rely upon the availability of information on the molecular states, which becomes substantially more difficult for states that are further from the equilibrium structure. Partition function inaccuracy is exacerbated by increasing temperatures, which makes the development of better partition functions especially necessary in high-temperature catalysis, gas phase dynamics, etc. We are automating construction of partition functions to desired accuracy, recognizing that corrections for anharmonicity, non-rigidity of the rotor, vibration-rotation coupling, and inclusion of relevant excited states are necessary. ATcT results were quintessential in these advances, and we will develop an open-source software that will permit routine calculation of accurate non-rigid-rotor-anharmonic-oscillator (NRRAO) partition functions accurate enough to enable reliable design for breakthrough advances in catalysis, photosynthesis, molecular assemblies, as well as other areas of chemistry.
Developments under ECC
This development is based on both reduced dimensionality methods and vibrational perturbation theory (VPT2) anharmonic constants that require derivatives at many grid points. The contributions to the partition function from soft vibrational modes (e.g. inversions, rotors) cannot simply be calculated using higher order derivatives at minimum. Instead reduced dimensionality methods that explicitly sample and solve a 1D or 2D potential energy surface will be used. This package will enable exascale automation through utilization of modern parallel computing resources by directly coupling to KinBot and the inclusion of sparse learning algorithms. In cases where the anharmonic constants calculated using VPT2 yield unphysical solutions (e.g. negative fundamental frequencies) the approach will be supplemented with reduced dimensionality (1D-2D) solutions to replace the contributions from those modes. The modes to be replaced will be identified through machine learning classification techniques using the anharmonic constants as a basis and existing NRRAO partition functions as a training set. In general, the reduced dimensionality approaches that replace contributions from internal rotation with state counts over the appropriate energy levels of the hindered rotor have the additional advantage of automatically including the contributions to the partition function from higher energy rotamers. Mutatis mutandis, the same is true of reduced dimensionality approaches for vibrational modes that connect other types of conformers (i.e. cis-trans).
The traditional approach to estimating the thermochemistry of adsorbates is to use the harmonic oscillator model. The 3N harmonic vibrational frequencies are calculated via finite differences from analytic first derivatives. Of these 3N normal modes, 3N-6 modes may be characterized as internal vibrational modes, which are similar (though not identical) to the vibrational modes in the gas-phase precursor. The remaining 6 modes may be characterized as relative vibrational modes, which correspond to the loss of free translation and rotation in the gas phase. Recent experimental work suggests that the conventional approach of treating the 3N modes as independent harmonic oscillators significantly underestimates the entropy of adsorbates. A likely explanation of this behavior is: the low-frequency modes that correspond to lost translation and rotation may be coupled and anharmonic.
Developments under ECC
We propose to extend the accuracy of anharmonic partition functions to the thermochemistry of adsorbates by providing a better description of these modes. The open-source code AdTherm provides more accurate partition functions for adsorbates. The 6 degrees of freedom (for arbitrary non-linear adsorbates) that describe the motion of the adsorbate relative to the surface are approximated by phase space representations. The six-dimensional partition functions are computed using Monte Carlo sampling methods.