Systematic Validation and Improvement of Quantum Chemistry Methods For The Prediction of Physical and Chemical Properties

A team of NIST researchers is working to systematically validate and improve popular quantum chemistry methods by comparing theoretical predictions to high-quality experimental data. They are developing databases to disseminate information on the accuracy, reliability, and transferability of methods like density functional theory (DFT) and semi-empirical Hamiltonians. Their work involves assessing sources of error in DFT, improving exchange functionals, and quantifying uncertainties in predicted properties like enthalpies of formation and vibrational frequencies.

Uploaded by

Electro_Lite

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

0% found this document useful (0 votes)

51 views2 pages

Systematic Validation and Improvement of Quantum Chemistry Methods For The Prediction of Physical and Chemical Properties

Uploaded by

Electro_Lite

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 2

Systematic Validation and Improvement of Quantum Chemistry Methods for the Prediction of Physical and Chemical Properties

C.A. Gonzalez, R. D. Johnson, and K.K. Irikura (838) The use of modern computational chemistry methodologies in the prediction of molecular properties has become increasingly popular mainly due to significant improvements in the algorithms, the accuracy of the methods, and the advent of powerful computer resources. This is particularly true in the area of thermochemistry, where researchers in industry and academia perform quantum chemistry calculations on a routine basis. Recently, there has been an increased interest in the use of theoretical predictions of physical and chemical properties made by state-of-the art quantum chemistry methodologies to fill gaps in the available experimental data. This approach could potentially eliminate the cost of experimental measurement and will be particularly A team of NIST researchers is useful in cases where extrapolations based on available experimental working on an integrated data are not possible or are unreliable. However, despite the aforementioned progress, the predictive power of most of the quantum approach that will lead to the chemistry methodologies has not been established on solid ground due systematic validation as well as to the lack of knowledge of their relative uncertainties. To understand improvements of popular and new these uncertainties an exhaustive and systematic validation of these theoretical methods using high methodologies involving robust strategies to compare theory and quality experimental data. experiment is necessary. In addition, the results of such validations must be widely available to the scientific community so new scientific discoveries could leverage from this knowledge. The NIST approach involves the development of state-of-the-art databases in order to disseminate the accuracy, reliability and transferability of the most popular quantum chemistry methodologies in the prediction of physical and chemical properties of a large variety of chemical systems ranging from isolated atoms and molecules to condensed phase. The theoretical methodologies being studied include semi-empirical Hamiltonians, wavefunction-based ab initio molecular orbital theory, and Density Functional Theory (DFT) while the experimental data used in the validation include measurements already available in the literature and also made by researchers at NIST. Assessing the Accuracy and Reliability of Density Functional Theory. Density Functional Theory (DFT) has become one of the most widely used quantum chemistry methods mainly due to its relatively low computational expense. The electronic energy within the DFT formalism is a functional of the electron density that consists of an exchange energy term and a correlation energy term. Although in principle there exists a universal DFT functional that can describe the electronic problem exactly, the discovery of such a functional has eluded scientists for years. In practice, arbitrary forms for the DFT functionals are parameterized in order to reproduce the experimental energetics of a relatively small set of molecular systems. CSTL researchers have performed a systematic validation of a particular family of DFT functionals, the so-called hybrid-GGA, in order to assess their accuracy and transferability to different chemical properties. The performance of these functionals in the prediction of properties such as singlet-triplet gaps, reaction barriers, dipole moments, and polarizabilities showed a marked erratic behavior leading to the conclusion that these functionals are not transferable to the computation of different chemical properties. In addition, the results of this work indicate that the major source of error can be traced back to the exchange functional, which exhibits significant spurious and uncontrollable electron correlation energy due to the empirical parameterization used in the development of these functionals.

Improvements to the Accuracy of Density Functional Theory. In order to solve the problems exhibited by the exchange functionals we proceeded to develop a rigorous and efficient method (the SC- method) for the calculation of the exact exchange functional. This method has been implemented in various popular quantum chemistry software packages and it has been tested in more than 40 molecules; the results have shown the possibility of computing the exact exchange functional in complicated polyatomic systems at the computational cost comparable to the inexpensive Hartree-Fock formalism. Encouraged by these results, we focused in procedures that could improve the exchangecorrelation functional as a whole. One of the most difficult steps in the generation of exchange-correlation functionals (XC) that do not require empirical parameters (and therefore, making it more general that the semiempirical hybrid-GGA functionals) is the development of closed forms for the derivatives of these functionals with respect to the electron density. These derivatives (or potentials as they are known) are critical in the calculation of molecular properties in a variational manner and are difficult if not impossible to implement for reliable XC functionals such as the Meta-GGA DFT due to their complicated mathematical I. Gonzalez, C. Gonzalez, V.V. Karasiev, forms. In this work, we have developed a general and simple E. Ludena, A. Hernandez, "Basis Set methodology called the Approximate Self-Consistent Potential Dependent SC Alpha Exchange-Only and (ASCP) that allows scientists to compute the corresponding Exchange-Correlation Calculations" J. derivatives without the knowledge of a closed form. The method has Chem. Phys. 118, 8161, 2003. been implemented in two of the most popular quantum chemistry packages and tested in the calculations of absolute energies and atomization energies of 20 different molecules. The results indicate that the method is robust and sufficiently general. Quantifying Uncertainty in Prediction of Enthalpies of Formation and Vibrational Frequencies. It is well recognized that a complete expression of experimental measurement includes both the central value and its associated uncertainty. The central value alone is incomplete. Throughout computational science, the results of models are usually reported without their associated uncertainties, making them incomplete. The term virtual measurement refers to a prediction from a computational model together with its associated uncertainty, and emphasizes the analogy with experimental (physical) measurement. Our investigation of the uncertainties associated with predictions from quantum chemistry models are intended to help virtual measurements supplant calculated results. The results from ab initio calculations differ from experiment by systematic biases, not random errors. For a given ab initio calculation corrections can be applied to compensate for this bias and an uncertainty can be determined from the distribution of corrections for a class of molecules. We use the experimental and ab initio data contained in the NIST Computational Chemistry Comparison and Benchmark Database (CCCBDB), which contains data for over 600 molecules, over 100 calculations for each molecule, and over 4000 vibrational frequencies. We combine this data with analysis based upon the Guide to the Expression of http://srdata.nist.gov/cccbdb/ Uncertainty in Measurement, published by the International Organization for Standardization (ISO), to determine biases and uncertainties. We have applied this to enthalpies of formation and vibrational frequencies. This K.K. Irikura, R.D. Johnson III, R.N. allows us to upgrade these calculated values into virtual Kacker, Uncertainty Associated with measurements. For the enthalpies of formation we found the class of Virtual Measurements from molecules to be critical in determining the bias and uncertainties. For Computational Quantum Chemistry the vibrational frequencies we found the uncertainties to be orders of Models, Metrologia 41, 369-375, 2004. magnitude larger than previously believed. We plan to continue extracting such information from the CCCBDB to determine the uncertainties for other calculated properties as well.