Bakowies Dirk
Laboratory of Physical Chemistry, ETH Zürich , CH 8093 Zürich, Switzerland.
J Phys Chem A. 2014 Dec 18;118(50):11811-27. doi: 10.1021/jp510249v. Epub 2014 Dec 8.
The ATOMIC protocol is a quantum-chemical thermochemistry protocol designed to obtain accurate atomization energies and derived heats of formation. It reduces errors of computationally tractable composite schemes through the use of bond separation reactions, which are implemented in a consistent ab initio framework. The present work explores possible simplification of previously introduced ATOMIC models. While coupled cluster calculations with singles and doubles excitations and perturbational treatments of connected triples excitations [CCSD(T)] are still required for high accuracy, basis-set truncations are possible in the CCSD-MP2 and CCSD(T)-CCSD components. The resulting models B4, B5, and B6 show root-mean-square (RMS) errors of only 0.21 to 0.46 kcal/mol for the AE set, which is a benchmark comprising complete-basis-set CCSD(T)(full) atomization energies of 73 neutral, closed-shell molecules composed of H, C, N, O, and F atoms. The evaluation of connected triples excitations can be avoided at medium levels of accuracy if the complete-basis-set MP2 energy is augmented with an empirically calibrated fraction of the difference between MP3 (or CCSD) and MP2 energies, calculated with small basis sets. The corresponding EMP3 and ECCSD models show RMS errors of 1.01 and 0.70 kcal/mol, respectively. Spin-component scaling is an option to rely entirely on the MP2 level of theory and still cut the RMS error of 4.38 kcal/mol by roughly a factor of 2 and achieve an accuracy comparable to accurate density functionals, such as M05-2X. The proposed new models are additionally tested with the HOF benchmark, a subset of G3/99 heats of formation that includes only neutral closed-shell molecules composed of H, C, N, O, and F atoms. The assessment shows that a number of experimental reference values are in error and should be replaced with more recent data. Results obtained with the new models are compared to original HOF (G3/99) reference data, to updated reference data, and to accurate ATOMIC/A theoretical data.
ATOMIC协议是一种量子化学热化学协议,旨在获得精确的原子化能和推导的生成热。它通过使用键分离反应减少了可计算处理的复合方案的误差,这些反应在一致的从头算框架中实施。目前的工作探索了对先前引入的ATOMIC模型进行可能简化的方法。虽然为了获得高精度仍需要进行含单双激发的耦合簇计算以及对连接三激发的微扰处理[CCSD(T)],但在CCSD-MP2和CCSD(T)-CCSD组件中可以进行基组截断。由此产生的模型B4、B5和B6对于AE集的均方根(RMS)误差仅为0.21至0.46千卡/摩尔,AE集是一个基准,包含由H、C、N、O和F原子组成的73个中性闭壳层分子的全基组CCSD(T)(完整)原子化能。如果用小基组计算的MP3(或CCSD)和MP2能量之差的经验校准分数来增加全基组MP2能量,那么在中等精度水平下可以避免对连接三激发的评估。相应的EMP3和ECCSD模型的RMS误差分别为1.01和0.70千卡/摩尔。自旋分量标度是一种完全依赖MP2理论水平的选择,它仍可将4.38千卡/摩尔的RMS误差大致降低一半,并达到与精确密度泛函(如M05-2X)相当的精度。所提出的新模型还使用HOF基准进行了测试,HOF基准是G3/99生成热的一个子集,仅包括由H、C、N、O和F原子组成的中性闭壳层分子。评估表明,一些实验参考值存在误差,应替换为更新的数据。将新模型获得的结果与原始HOF(G3/99)参考数据、更新后的参考数据以及精确的ATOMIC/A理论数据进行了比较。