Local Coupled Cluster

I mainly use local correlation methods in Molpro. Local correlation methods exploit that electron correlation is relatively local in nature. This property is not taken into the account in the canonical calculations. Canonical orbitals are highly delocalized over the whole system. This causes the unfortunate scaling with the number of electrons in these methods. To overcome this problem a mathematically equivalent representation of the canonical orbitals has been suggested. Both, occupied and virtual spaces are localized using different procedures. Different methods which are using the locality of electron correlation have been introduced.

Local correlation methods as implemented in Molpro are based on two approximations. The first one is the domain approximation. In the domain approximation the excitations are restricted to subspaces of projected atomic orbitals, so called domains. The second approximation is the pair approximation. This is coming from the fact that the correlation energy decreases quickly with the distance between two correlated localized orbitals. Therefore, orbital pairs very far from each other can be neglected or treated at a lower level of theory.

During my research stay in the group of Prof. Ulf Ryde in Lund we used local coupled cluster method in combination with the fragmentation scheme developed in his group to predict binding energies of host-guest complexes in the SAMPL4 blind challenge.

For more information you can see:

Chem. Phys. Chem. 15, 3270-3281 (2014)

QM/QM Hybrid Schemes 

It is the common case that one deals with system sizes too large to be investigated with conventional wavefunction QM methods. However, one is often interested in a chemical event, so the most important part of the molecule is relatively localized around a point of interest. In inorganic catalysis, for example, this would correspond to the metal center or the part where the bond formation/breaking occurs. Therefore, one can apply accurate correlation methods only to this part of the system. In the scope of local correlation methods this can be done by assigning groups of localized molecular orbitals and related domains to the specific region and then apply different methods to the different regions. This approach is called Local Molecular Orbital : Molecular Orbital (LMOMO) method.

With this approach only one calculation is necessary to obtain the result. Also, the definition of a model system which would be treated at the high level is not needed and cutting of bonds is avoided. Since the high level region is computed in the presence of the low level region the coupling between the regions is included in the calculation. All these points are an improvement over methods which are more in use, e.g. ONIOM.

During my PhD I was investigating the possibility of using the LMOMO method on systems which contain a metal center, more specifically molybdenum enzymes. We showed that the results obtained with this hybrid scheme are in good agreement with full local coupled cluster results. Furthermore, it was shown that local coupled cluster results for these systems are in perfect agreement with experimental results, therefore they can be used as a reference for the accuracy of the LMOMO scheme.

Furthermore, I was working on developing this hybrid scheme for open shell systems. This scheme was firstly tested on small systems where binding energies of a series of metal complexes were calculated. These systems were chosen so that canonical coupled cluster calculations could be carried out and also to represent all three common metal-binding atoms in proteins (O, S, N). In the next step, the hybrid scheme was tested against local coupled cluster results on the example of copper nitrite reductase, where electron and proton affinities were calculated with both methods.

For more information you can see:

J. Chem. Theory Comput. 10, 5397-5404 (2014)

Reaction pathways

I started investigating reaction mechanisms during my PhD, where we explored reaction mechanisms of two different molybdenum enzymes. My work was mainly focused on doing accurate calculations using local coupled cluster and LMOMO methods.

Furthermore, during my first postdoc I was working on reaction mechanisms and understanding of metallorganic reactions. More precisely on reactions which involve C-H activations. During this period I worked in an experimental group and used DFT methods to predict and explain different experimental results.

For more information you can see:

J. Biol. Inorg. Chem. 7, 1165-1179 (2014)

Angew. Chem. Int. Ed. 55, 7408-7412 (2016)

Chem. Eur. J. 22, 17958-17961 (2016)

Heme and Non-Heme Iron Chemistry

Nowadays I work on understanding reactions which include heme and non-heme iron centers. I am using a wide range of methods, starting with DFT, over canonical and local coupled cluster methods up to multirefenrece methods such as CASPT2.

The first project on which I worked involved the benchmarking of local and canonical coupled cluster methods with singles, doubles and perturbative triples against higher order coupled cluster methods as well as CASPT2 results. During this process we also implemented a new version of local coupled cluster methods, so called hotspots which significantly improved local coupled cluster results for these kind of the systems.