Skip to main content

An Optimised Genetic Algorithm to Computationally Predict a Metal-Organic Framework to Separate Helium from Methane

Final Report Summary - GA MOF (An Optimised Genetic Algorithm to Computationally Predict a Metal-Organic Framework to Separate Helium from Methane)

Extension of the Universal Force Field for Metal-Organic Frameworks: UFF4MOF:
After it was determined that Density Functional Tight Binding (DFTB) was to expensive to use as the underlying quantum chemical method for calculating arbitrary, hypothetical MOF structures, work began on extending the Universal Force Field of Rappé et al. to MOF structures. While UFF is, in principle, universal, and contains parameters for elements across the entire periodic table, Rappé and co-developed parameters for only the most common combinations of element and coordination environment. In particular, UFF requires copper atoms to maintain tetrahedral coordination. UFF is therefore unable to treat one of the most common MOF building blocks, the copper paddlewheel, where the copper atoms are octahedrally coordinated. Our first extension to UFF, UFF4MOF, (J. Chem. Theory Comput., 2014, 10, pp 880–891) focussed primarily on these paddlewheel species and contains parameters for paddlewheels of all the 4th row transition metals (Ti – Zn), as well as several other popular secondary building units (SBUs): the Zn4O SBU (as in MOF-5), the M3O trimeric SBU, the MFU4 SBU and the 1-D MIL53 SBU.
UFF4MOF is available in several computational chemistry programs, including the General Utility Lattice Program (GULP), Amsterdam Density Functional (ADF) and deMonNano.
Since the publication of this first paper, which has already been cited 5 times, work has continued toward a MOF-optimised, full periodic table force field.
UFF4MOF is now an integral part of our framework generator, AuToGraFS, where it is used to rapidly optimise the generated structures.


Software to generate the coordinates of arbitrary, hypothetical framework structures: AuToGraFS:

AuToGraFS, the Automatic Topological Generator for Framework Structures, was published in September 2014 (J. Phys. Chem. A, 2014, 118, pp 9607–9614). This software has been released as open source (LGPL license) on github and can be downloaded at: https://github.com/maddicoat/AuToGraFS. In addition to this free release, we are currently negotiating with an industrial partner, in order to provide a supported software release with improved features. Several research groups have already downloaded the software and we anticipate that a supported release will increase the impact of AuToGraFS, by allowing non-computational researchers to conveniently build and calculate framework structures of direct interest to their research.
While AuToGraFS is designed to build framework structures from arbitrary building blocks, which may be built and specified by the user, an inherent part of AuToGraFS is the internal databases of topologies and building blocks (separated into linkers, connectors and functional groups). Currently these databases contain 68 topologies, 28 connectors and 83 linkers. Topologies are sourced from the Reticular Chemistry Structure Resource (RSCR, http://rcsr.net). Molecular building blocks were extracted from known MOF and COF structures, symmetrised and optimised to ensure they may be employed as components in arbitrary framework structures, regardless of the topology or the identity of other building blocks. With these modest databases, over 10,000 distinct framework structures may be constructed.


Implementation and validation of fitness function:

Initial testing of the several of the proposed fitness functions showed that they were not sufficiently reliable to guide a genetic algorithm. In particular, the mechanical properties of AuToGraFS-generated framework structures were shown to be very sensitive to minor perturbations in the atom positions, thus requiring a more expensive optimisation. The current UFF4MOF force field was also inadequate to produce reliable binding energies. Work continues on developing these fitness functions.


Detailed study of MOF candidates:

The topological nature of AuToGraFS opens up a new avenue of research, namely, determining which, of several possible topologies, will a given set of building blocks form. In collaboration with two experimental groups we have used AuToGraFS, followed by higher level ab initio calculations to determine and rationalise the preferred topology for both 2-D and 3-D framework structures. Two manuscripts on these systems are currently under preparation.