Daniel Kats

Local CC2 response method for triplet states based on Laplace transform: excitation energies and first-order properties.

A new multistate local CC2 response method for calculating excitation energies and first-order properties of excited triplet states in extended molecular systems is presented. The Laplace transform technique is employed to partition the left/right local CC2 eigenvalue problems as well as the linear equations determining the Lagrange multipliers needed for the properties. The doubles part in the equations can then be inverted on-the-fly and only effective equations for the singles part must be solved iteratively. The local approximation presented here is adaptive and state-specific. The density-fitting method is utilized to approximate the electron-repulsion integrals. The accuracy of the new method is tested by comparison to canonical reference values for a set of 12 test molecules and 62 excited triplet states. As an illustrative application example, the lowest four triplet states of 3-(5-(5-(4-(bis(4-(hexyloxy)phenyl)amino)phenyl)thiophene-2-yl)thiophene-2-yl)-2-cyanoacrylic acid, an organic sensitizer for solar-cell applications, are computed in the present work. No triplet charge-transfer states are detected among these states. This situation contrasts with the singlet states of this molecule, where the lowest singlet state has been recently found to correspond to an excited state with a pronounced charge-transfer character having a large transition strength.

Communication: The distinguishable cluster approximation

We present a method that accurately describes strongly correlated states and captures dynamical correlation. It is derived as a modification of coupled-cluster theory with single and double excitations (CCSD) through consideration of particle distinguishability between dissociated fragments, whilst retaining the key desirable properties of particle-hole symmetry, size extensivity, invariance to rotations within the occupied and virtual spaces, and exactness for two-electron subsystems. The resulting method, called the distinguishable cluster approximation, smoothly dissociates difficult cases such as the nitrogen molecule, with the modest N(6) computational cost of CCSD. Even for molecules near their equilibrium geometries, the new model outperforms CCSD. It also accurately describes the massively correlated states encountered when dissociating hydrogen lattices, a proxy for the metal-insulator transition, and the fully dissociated system is treated exactly.

Sparse tensor framework for implementation of general local correlation methods

Coupled-cluster methods offer unprecedented accuracy for a wide range of chemically important properties, but the steep scaling of computational cost with system size makes widespread use challenging. Local approximations, building on the short-range nature of electron correlation effects in insulators, help a great deal, but are much more complicated than their canonical counterparts. In this work we discuss an automated implementation scheme for local coupled-cluster methods, based on an interpreter and an underlying representation of sparse tensors. We demonstrate the efficacy of the approach through implementation of a very wide range of singles-and-doubles-based coupled-cluster schemes.

Communication: The distinguishable cluster approximation. II. The role of orbital relaxation

The distinguishable cluster approximation proposed in Paper I [D. Kats and F. R. Manby, J. Chem. Phys. 139, 021102 (2013)] has shown intriguing abilities to accurately describe potential energy surfaces in various notoriously difficult cases. The question that still remained open is to what extend the accuracy and the stability of the method is due to the special choice of orbital-relaxation treatment. In this paper we introduce orbital relaxation in terms of Brueckner orbitals, orbital optimization, and projective singles into the distinguishable cluster approximation and investigate its importance in single- and multireference cases. All three resulting methods are able to cope with many multiple-bond breaking problems, but in some difficult cases where the Hartree-Fock orbitals seem to be entirely inadequate the orbital-optimized version turns out to be superior.

Local Approximations for an Efficient and Accurate Treatment of Electron Correlation and Electron Excitations in Molecules

Local methods for the description of electron correlation in ground and electronically excited states of molecules, as implemented in the MOLPRO system of ab initio programs, are reviewed. Recent improvements in the performance of the local method resulting from an implementation of the density-fitting technique for all electron-repulsion integrals are discussed. Local fitting approximations lead to linear scaling of CPU time and disk space with molecular size, and allow for a significant increase of the size of molecules and basis sets that can be treated by the local MP2, CCSD, and CCSD(T) ab initio methods. Recent extensions of these methods to open-shell systems, as well as the inclusion of explicitly correlated terms are described. It is demonstrated that the latter lead to a drastic improvement of the accuracy of local methods. A local treatment of electron excitations within the EOM-CCSD and CC2 theories, as well as a local description of first- and second-order molecular properties are also discussed. Finally, we present some illustrative applications of the outlined methods.

Accurate thermochemistry from explicitly correlated distinguishable cluster approximation

An explicitly correlated version of the distinguishable-cluster approximation is presented and extensively benchmarked. It is shown that the usual F12-type explicitly correlated approaches are applicable to distinguishable-cluster theory with single and double excitations, and the results show a significant improvement compared to coupled-cluster theory with singles and doubles for closed and open-shell systems. The resulting method can be applied in a black-box manner to systems with single- and multireference character. Most noticeably, optimized geometries are of coupled-cluster singles and doubles with perturbative triples quality or even better.

Speeding up local correlation methods

We present two techniques that can substantially speed up the local correlation methods. The first one allows one to avoid the expensive transformation of the electron-repulsion integrals from atomic orbitals to virtual space. The second one introduces an algorithm for the residual equations in the local perturbative treatment that, in contrast to the standard scheme, does not require holding the amplitudes or residuals in memory. It is shown that even an interpreter-based implementation of the proposed algorithm in the context of local MP2 method is faster and requires less memory than the highly optimized variants of conventional algorithms.

Local complete active space second-order perturbation theory using pair natural orbitals (PNO-CASPT2).

We present a CASPT2 method which exploits local approximations to achieve linear scaling of the computational effort with the molecular size, provided the active space is small and local. The inactive orbitals are localized, and the virtual space for each electron pair is spanned by a domain of pair-natural orbitals (PNOs). The configuration space is internally contracted, and the PNOs are defined for uniquely defined orthogonal pairs. Distant pair energies are obtained by multipole approximations, so that the number of configurations that are explicitly treated in the CASPT2 scales linearly with molecular size (assuming a constant active space). The PNOs are generated using approximate amplitudes obtained in a pair-specific semi-canonical basis of projected atomic orbitals (PAOs). The evaluation and transformation of the two-electron integrals use the same parallel local density fitting techniques as recently described for linear-scaling PNO-LMP2 (local second-order Møller-Plesset perturbation theory). The implementation of the amplitude equations, which are solved iteratively, employs the local integrated tensor framework. The efficiency and accuracy of the method are tested for excitation energies and correlation energies. It is demonstrated that the errors introduced by the local approximations are very small. They can be well controlled by few parameters for the distant pair approximation, initial PAO domains, and the PNO domains.

The distinguishable cluster approach from a screened Coulomb formalism

The distinguishable cluster doubles equations have been derived starting from an effective screened Coulomb formalism and a particle-hole symmetric formulation of the Fock matrix. A perturbative triples correction to the distinguishable cluster with singles and doubles (DCSD) has been introduced employing the screened integrals. It is shown that the resulting DCSD(T) method is more accurate than DCSD for reaction energies and is less sensitive to the static correlation than coupled cluster with singles and doubles with a perturbative triples correction.

Local CC2 response method based on the Laplace transform: orbital-relaxed first-order properties for excited states.

A multistate local CC2 response method for the calculation of orbital-relaxed first order properties is presented for ground and electronically excited states. It enables the treatment of excited state properties including orbital relaxation for extended molecular systems and is a major step on the way towards analytic gradients with respect to nuclear displacements. The Laplace transform method is employed to partition the eigenvalue problem and the lambda equations, i.e., the doubles parts of these equations are inverted on-the-fly, leaving only the corresponding effective singles equations to be solved iteratively. Furthermore, the state specific local approximations are adaptive. Density-fitting is utilized to decompose the electron-repulsion integrals. The accuracy of the local approximation is tested and the efficiency of the new code is demonstrated on the example of an organic sensitizer for solar-cell applications, which consists of about 100 atoms.