Supplementary Information for Evaluating the. energetic driving force for co-crystal formation

Transcription

1 Supplementary Information for Evaluating the energetic driving force for co-crystal formation Christopher R. Taylor and Graeme M. Day School of Chemistry, University of Southampton, Highfield, Southampton, SO17 1BJ, United Kingdom List of Solvents The following molecules were specified as solvents or room-temperature liquid components for the purposes of excluding solvates or small-molecule clathrates from the list of multicomponent systems obtained from the CSD. Such structures were removed from the list by using this set as a filter within the ConQuest package. Water Methanol Ethanol Dimethyl ether Diethyl ether Formic acid Acetic acid 1

2 Acetone 3-pentanone Ethyl acetate Isopropanol tert-butanol Dichloromethane (DCM) Trichloromethane Tetrachloromethane 1,1,2,2-tetrachloroethane 1,1,1,2,2-pentachloroethane Hexachloroethane Acetonitrile Dimethylsulfoxide (DMSO) N,N -dimethylacetamide (DMA) Carbon disulfide Pentane Hexane Cyclohexane Benzene Toluene 2

3 p-xylene Tetrahydrofuran (THF) 1,4-dioxane Pyridine Phenol CSD Search Procedure Using the ConQuest software, we searched the CSD for crystal structures that met the following criteria: Contain at least two distinct chemical species (i.e. G 2) Contain only C, H, N, O, F, S, or Cl Present in the Best Hydrogens subset of the CSD as defined by van der Streek. 1 This resulted in 3303 co-crystal structures. After randomly selecting a set of these structures to run through our DFT calculations, we exported this set as SMILES strings and used in-house Python code to identify all the unique SMILES strings, and thus all unique molecular species present in our co-crystal set. We then built a set of all reliable single-component structures by searching the CSD using the same criteria but the appropriate number of chemical species (G = 1) and exporting a list of SMILES strings for these. Finally, we searched this second list of SMILES strings for all unique SMILES obtained from the co-crystal search this allowed us to identify all the single-component structures that contained the molecules of interest. As SMILES lack stereochemical information, it is feasible that this procedure would give erroneous matches when multiple chemical species share an ambiguous SMILES string. Such 3

4 cases were checked visually and amended manually, and in any case amounted to no more than 8% of the entire set of 350 structures. Halogen-bonded subset To augment our test set and consider greater chemical diversity in co-crystal interactions, we chose to perform a complimentary search of the database for additional co-crystals that specifically featured halogen bonding. As stated in the main article, we searched for such species in the CSD using the following criteria: Contain at least two distinct chemical species (i.e. G 2) Contain a contact D X A in which the D X distance is less than the sum of the atoms van der Waals radii (plus a tolerance of 0.1 Å), and the angle formed by D X A is greater than 160 degrees. D is one of N, O, S, or Cl, X is either Br or I, All other atoms are any of C, H, N, O, F, S, or Cl. This procedure yielded 34 binary co-crystal structures, of which only 28 successfully completed our minimisation procedure. The matching single-component structures were searched for manually in ConQuest. Co-crystals and component index Our set of co-crystals contains 350 structures. Two of these are ternary (three-component) co-crystals; all the others are binary (two-component) systems. 4

5 Our list of co-crystals accompanies this SI as three comma-separated variable (CSV) files of co-crystals (separated by category: hydrogen-bonding, halogen-bonding, or weaklybound) and the single component structures we have identified for them using our SMILES search procedure. Each row in these files is a co-crystal, followed by: its components and stoichiometry, the calculated relative stability (both per formula unit and per molecule), the calculated H-bond strength measure (both per formula unit and per molecule), the change in packing coefficient, the hydrogen bond counts of the co-crystal and its single components, the change in hydrogen bond count (both per formula unit and per molecule), the type of interaction present (H-bond, halogen bond, or no bonds ), whether proton transfer occurred in the DFT optimisation (only relevant for 6 of the H-bonded set), whether the halogen present is Br or I (only relevant for the halogen-bonded set), and halogen bond counts for each molecule in the co-crystal (both per formula unit and per molecule, again only for the halogen-bonded set). DFT optimization Procedure Our calculations originally began using the CASTEP code and the PBE+D2 functional and dispersion correction. VASP and the improved PBE+D3 level of theory became available to us at a later date. Rather than re-minimize many of our experimental structures from scratch, we chose to adopt a procedure in which each structure was first minimised in CASTEP at the PBE+D2 level of theory, then re-minimise these CASTEP-optimized structures using VASP (as our access to VASP was time-limited while CASTEP was not). In both codes we applied our procedure of first optimizing the structures with the cell parameters fixed, before then re-optimizing with variable cell parameters. Each crystal structure was therefore subjected to the following procedures, in this order, with the output of each step being used as the input for the next: 5

6 1. A CASTEP PBE+D2 fixed-cell optimisation 2. A CASTEP PBE+D2 optimisation with a flexible cell 3. A VASP PBE+D3 fixed-cell optimisation 4. A VASP PBE+D3 optimisation with a flexible cell 5. A final, higher-accuracy VASP PBE+D3 single-point calculation (using the PREC=Accurate flag and a plane-wave energy cut-off of 600 ev). All energies and structures presented are those that have been optimised through both levels of theory according to the above workflow. Several structures were separately optimized through VASP PBE+D3 alone to confirm that the above procedure did not give significantly different energies or structures from minimising with PBE+D3 in VASP alone. Hydrogen-bonding as a descriptor Cases of proton transfer As stated in the article text, during our DFT optimisations, we encountered 6 co-crystal structures that underwent proton transfer between species during the minimisation, becoming salts. The CSD refcodes of these species are: AWUDEB, DINSOK01, MIZMUE, RAWFAW, SEDJUI, and UJORAM. The effect of donor-acceptor repulsion As described in the article text, we compute a simple hydrogen bond strength measure based on the Coulombic interaction of the hydrogen and acceptor partial charges according to molecular DFT calculations. By definition, assuming the hydrogen and acceptor have partial charges of opposite sign, this interaction has no explicit description of interatomic repulsion, or the effect of the donor atom. 6

7 We attempted to rectify this by including the Coulombic interaction of the donor and acceptor charges, effectively modeling the hydrogen bond as an acceptor atom interacting with a model dipole formed by the hydrogen and donor atoms. However, this modification proved unsatisfactory for two reasons. Firstly, it offers no improvement in the correlation of co-crystal relative stability with the calculated hydrogen bond strength (see Figure 1). Eco-cryst kj mol 1 of molecules E H-bond H-A,electr kj mol 1 of molecules Figure 1: Relative stabilities of co-crystals as a function of the change in the H-bond strength measure defined in the article text. This measure considers the Coulombic attraction of the hydrogen and acceptor, plus the repulsion of the acceptor and donor. However, it reveals no significantly greater correlation with the relative stability than the H A attraction alone. More concerningly, including this donor-acceptor repulsion raises the energy of many hydrogen bonds so dramatically that they become repulsive; approximately 20% of both cocrystals and single-component structures have positive total hydrogen bond energies when the repulsion is included. (All total hydrogen bond energies are negative when repulsion is omitted.) Evidently the partial charges computed for the donor can be large enough that their interaction with the acceptor is destabilising, despite the greater D A distance. 7

8 References (1) van de Streek, J. Acta Crystallographica Section B Structural Science 2006, 62,