Accuracy electron correlation methods

Full ab initio electron-correlation methods, from MP2 to CCSD(T) (the acronyms refer to increasing complexity in the treatment of correlation, with increasing computational cost) include polarization and dispersion contributions and apply to any molecular system. Accuracy depends on the size of the basis set, but so-called complete set limit calculations can nowadays be carried out. 

Due to its excellent balance between accuracy and computational cost, Kohn-Sham density functional theory (KS-DFT) [13,14] is usually the method of choice to investigate electronic ground states and their properties in chemistry and solid-state physics [15,16]. Hartree-Fock (HF) wavefunctions, on the other hand, are the starting point for ab initio electron correlation methods [4,15] which are discussed in Section 4. 

The quantum chemist s traditional way to approximate solutions of the electronic Schrodinger equation is so-called ab initio, wave function-based electron correlation methods. These methods improve upon the HF mean-field approximation by adding many-body corrections in a systematic way [15]. As of the time of this writing, efforts to accelerate ab initio calculations with GPUs are scarce. However, it is expected that this will change in the near future because these methods are of critical importance whenever higher accuracy is required than what can be achieved by DFT or for types of interactions and properties for which DFT breaks down. 

Second-order Moller-Plesset perturbation theory (MP2) is the computationally least expensive and most popular ab initio electron correlation method [4,15]. Except for transition metal compounds, MP2 equilibrium geometries are of comparable accuracy to DFT. However, MP2 captures long-range correlation effects (like dispersion) which are lacking in present-day density functionals. The computational cost of MP2 calculations is dominated by the integral transformation from the atomic orbital (AO) to the molecular orbital (MO) basis which scales as 0(N5) with the system size. This four-index transformation can be avoided by introduction of the RI integral approximation which requires just the transformation of three-index quantities and reduces the prefactor without significant loss in accuracy [36,37]. This makes RI-MP2 the most efficient alternative for small- to medium-sized molecular systems for which DFT fails. 

In many cases electronic properties calculated at the Hartree-Fock level do not have the accuracy sufficient to make them useful in chemical predictions. For example, as revealed in a recent study,the stability of the cage isomer of the C20 carbon cluster relative to that of the cyclic isomer is underestimated at the Flartree-Fock level by as much as 200 kcal/mol. In such systems, the electron correlation effects have to be taken into account in quantum chemical calculations through application of approximate methods. One such approximate electron correlation methods that has gained a widespread popularity is the second-order Moller-Plesset perturbation theory (MP2). Until recently calculations involving the MP2 approach have used a traditional formulation in which the MP2 energy is evaluated as the sum... 

As the accuracy of electron correlation methods is Hmited by the choice of the dimension of the active orbital space, the convergence behavior of the spin density with respect to the size of the active space must be studied to ensure that accurate ah initio spin densities are obtained. Although the CASSCF spin densities were quantitatively converged for medium-sized active orbital spaces, larger active spaces (more than 13 electrons correlated in 13 orbitals) were found to be unstable, that is, active space orbitals have been rotated out of the active space during the MCSCF procedure, while the spin densities started to diverge compared to the smaller sized CASSCF results. 

Local electron-correlation methods are ab-initio wavefunction-based electronic-structure methods that exploit the short-range nature of dynamic correlation effects and in this way allow linear scaling 0 N) in the electron-correlation calculations [128,129,131-135] to be attained. 0 N) methods are applied to the treatment of extended molecular systems at a very high level of accuracy and rehabihty as CPU time, memory and disk requirements scale hneaily with increasing molecular size N. 

One of the original approximate methods is the wavefunction-theory-based Hartree-Fock (HF) method [40]. The HF method is a single determinant method that does not include any correlation interactions between the electrons, and as such has limited accuracy [41, 42]. Higher level wavefunction-based methods such as coupled cluster [43 5], configuration interaction [40,46,47], and complete active space [48-50] methods include multiple determinants to incorporate some of the electron-electron correlation. Methods based on perturbation theory, such as second order Mpller-Plesset perturbation theory [51], go beyond the HF method by perturbatively adding electron correlation. These correlated wavefunction-based methods have well-defined ways in which they approach the exact solution to the Schrodinger equation and thus have the potential to be extremely accurate, but this accuracy comes at a price [52]. 

A pleasant aspect of DFT is that, unlike the case of ab initio methods for including electron correlation, the accuracy with which various properties are predicted does not seem to improve much with increasing basis set size, once one has reached a split-valence basis that includes polarization functions (SVP basis set). To recover a significant amount of electron correlation energy in ab initio calculations, large basis sets are necessary to provide an adequate space of virtual MOs. However, in DFT electron correlation is sensitive to the V,. functional, but not so much to how well the exact electron density is reproduced. Thus, from a B3LYP/SVP calculation, one can generally expect to get better results and usually at smaller computational cost than from an MP2 calculation with the same basis set. 

Although orbital wave functions, such as Hartree-Fock, generalized valence bond, or valence-orbital complete active space self-consistent field wave functions, provide a semi-quantitative description of the electronic structure of molecules, accurate predictions of molecular properties cannot be made without explicit inclusion of the effects of dynamical electron correlation. The accuracy of correlated molecular wave functions is determined by two inter-related expansions the many-electron expansion in terms of antisymmetrized products of molecular orbitals that defines the form of the wave function, and the basis set used to expand the one-electron molecular orbitals. The error associated with the first expansion is the electronic structure method error the error associated with the second expansion is the basis set error. Only by eliminating the basis set error, i.e., by approaching the complete basis set (CBS) limit, can the intrinsic accuracy of the electronic structure method be determined. 

Accuracy of Electron Correlation Methods for Actinide Excited States WFT and DFT Methods... 

The disadvantage of ah initio methods is that they are expensive. These methods often take enormous amounts of computer CPU time, memory, and disk space. The HF method scales as N, where N is the number of basis functions. This means that a calculation twice as big takes 16 times as long (2" ) to complete. Correlated calculations often scale much worse than this. In practice, extremely accurate solutions are only obtainable when the molecule contains a dozen electrons or less. However, results with an accuracy rivaling that of many experimental techniques can be obtained for moderate-size organic molecules. The minimally correlated methods, such as MP2 and GVB, are often used when correlation is important to the description of large molecules. 

A variation on MNDO is MNDO/d. This is an equivalent formulation including d orbitals. This improves predicted geometry of hypervalent molecules. This method is sometimes used for modeling transition metal systems, but its accuracy is highly dependent on the individual system being studied. There is also a MNDOC method that includes electron correlation. 

To overcome these limitations, the hybrid QM-MM potential can employ ad initio or density function methods in the quantum region. Both of these methods can ensure a higher quantitative accuracy, and the density function methods offer a computaitonally less expensive procedure for including electron correlation [5]. Several groups have reported the development of QM-MM programs that employ ab initio [8,10,13,16] or density functional methods [10,41-43]. 

Each cell in the chart defines a model chemistry. The columns correspond to differcni theoretical methods and the rows to different basis sets. The level of correlation increases as you move to the right across any row, with the Hartree-Fock method jI the extreme left (including no correlation), and the Full Configuration Interaction method at the right (which fuUy accounts for electron correlation). In general, computational cost and accuracy increase as you move to the right as well. The relative costs of different model chemistries for various job types is discussed in... 

In the last few years, methods based on Density Functional Theory have gained steadily in popularity. The best DFT methods achieve significantly greater accuracy than Harttee-Fock theory at only a modest increase in cost (far less than MP2 for medium-size and larger molecular systems). They do so by including some of the effects of electron correlation much less expensively than traditional correlated methods. 

In order to calculate total energies with a chemical accuracy of 1 kcal/mol, it is necessary to use sophisticated methods for including electron correlation and large basis sets, which is only computationally feasible for small systems. Instead the focus is usually on calculating relative energies, trying to make the errors as constant as possible. 

The LSDA approximation in general underestimates the exchange energy by 10%, thereby creating errors which are larger tlian the whole correlation energy. Electron correlation is furthermore overestimated, often by a factor close to 2, and bond strengths are as a consequence overestimated. Despite the simplicity of the fundamental assumptions, LSDA methods are often found to provide results with an accuracy similar to that obtained by wave mechanics HE methods. 

The work described in this paper is an illustration of the potential to be derived from the availability of supercomputers for research in chemistry. The domain of application is the area of new materials which are expected to play a critical role in the future development of molecular electronic and optical devices for information storage and communication. Theoretical simulations of the type presented here lead to detailed understanding of the electronic structure and properties of these systems, information which at times is hard to extract from experimental data or from more approximate theoretical methods. It is clear that the methods of quantum chemistry have reached a point where they constitute tools of semi-quantitative accuracy and have predictive value. Further developments for quantitative accuracy are needed. They involve the application of methods describing electron correlation effects to large molecular systems. The need for supercomputer power to achieve this goal is even more acute. 

---