Supporting Information Photocatalytic Proton Reduction by a Computationally Identified, Molecular Hydrogen- Bonded Framework

Catherine M. Aitchison,ǂa Christopher M. Kane,ǂa David P. McMahon,ǂb Peter Spackman,bc Angeles Pulido, b Xiaoyan Wang,a Liam Wilbraham,d Linjiang Chen,ac Rob Clowes,a Martijn A. Zwijnenburg,d Reiner Sebastian Sprick,a Marc A. Little,a Graeme M. Day*b and Andrew I. Cooper*ac a. Department of Chemistry and Materials Innovation Factory, University of Liverpool, Liverpool L7 3NY, U.K. b. Computational Systems Chemistry, School of Chemistry, University of Southampton, Southampton SO17 1BJ, U.K c. Leverhulme Research Centre for Functional Materials Design, University of Liverpool, Liverpool, L7 3NY, U.K. d. Department of Chemistry, University College London, 20 Gordon Street, London WC1H 0AJ, U.K


Crystal Structure Prediction (CSP) Methodology
Crystal structure prediction was performed using a quasi-random sampling procedure, as implemented in our in-house Global Lattice Energy Explorer (GLEE) software (version 2). [76] Molecules were generated via a conformation search in Maestro [92] using the OPLS2005 forcefield [93] (no cutoff and a dielectric of 1.0) using a mixed torsional/low-mode search with a maximum of 10000 steps allowed. Conformers with an energy less than 35 kJ mol -1 above the lowest energy structure and with an RMSD of greater than 0.3 Å were retained. Each unique conformer was re-optimized using density functional theory (DFT) at the B3LYP GD3BJ/6-311G** level of theory using Gaussian09. [94] Redundant conformers with an RMSD < 0.3 Å were eliminated. These molecular geometries were then held rigid throughout crystal structure generation and lattice energy minimization.
Initial crystal generation involves a low-discrepancy sampling of all structural variables within each space group: unit cell lengths, angles, molecular positions and orientations within the asymmetric unit. Space-group symmetry was then applied, and a geometric test was performed for overlap between molecules. Molecular clashes were removed by lattice expansion. [76] Lattice energy calculations were performed with using DMACRYS2.2.1.0 with atom-atom repulsion and dispersion interactions modelled with the FIT intermolecular potential, [77] in conjunction with a spline. [78] Electrostatic interactions were modelled using an atomic multipole description of the molecular charge distribution (up to hexadecapole on all atoms) from the B3LYP/6-311G**-calculated charge density using a distributed multipole analysis (version 2). [95] Charge-charge, charge-dipole and dipole-dipole interactions were calculated using Ewald summation; all other intermolecular interactions were summed to a cut-off between molecular centres-of-mass (15 Å for pyrene, 20 Å for TPhP and 25 Å for all others). For each structure multiple rigid lattice energy minimizations were performed. During the initial lattice energy minimization point-charges obtained from a MULFIT [96] fit of atomic charges to the molecular electrostatic potential generated from the B3LYP/6-311G**-distributed multipole analysis atomic multipoles were used in conjunction with a pressure of 0.1 GPa and a VDW cutoff that was increased by 10 Å. For all subsequent steps, multipoles at the hexadecapole level were used without pressure. Minimization cycles were continued until at least 3 minimizations had been performed and the structure had stopped changing (as indicated by an F-value < 1).
Initial clustering within each conformer was performed within each individual spacegroup. Initially, all structures within a lattice energy window of 1.0 kJ / mol and within a density window of ±0.05 g cm -3 from the reference structure were compared. This was performed for all unique structures. The initial clustering of the data was performed using powder x-ray diffraction patterns generated by platon [69] (wavelength 0.7 Å and a two-theta range of 20°) using a constrained dynamic time-warping algorithm (with a constraint that the offset between patterns < 10 lots of 0.02 two-theta) to compare pairs of structures. [97] Structures were considered a match when the Euclidean distance between the powder patterns (normalized by area) was < 10.0 (units?). This was followed by clustering structures within 1.0 kJ / mol and ±0.01 g cm -3 with the COMPACK [98] algorithm using 30 molecule molecular clusters, a distance tolerance of 40 % and a maximum value of the RMSD30 of 0.4 Å. A final clustering across space groups was performed using the constrained dynamic time-warping algorithm powder XRD analysis using a lattice energy window of 1.0 kJ / mol and with a density window of ±0.01 g cm -3 . No clustering was performed between conformers.

Property calculations
Accessible surface area calculations were performed using Zeo++, [99] using the high accuracy setting with the channel and probe radius set to 1.2 Å using 10000 Monte-Carlo samples.

Structure matches
RMSD30 between experimental and computed structures were calculated with Mercury. [100]

Stacking analysis
Planar stacking analysis was performed using bespoke code tailored for this purpose, using molecular planes of best fit (PBFs). The PBF of a molecule is calculated from the principal components of the molecule, taken via singular value decomposition (SVD) i.e.
where is the matrix of atomic positions in the molecule. SVD yields the principal = Σ * components of the molecule, which can be sorted by their corresponding singular values .

Σ
The least significant principal component N corresponds to the plane normal, and once atomic positions are projected onto this plane via dot product a bounding rectangle may ⋅ be evaluated using the remaining two principal components by finding the extrema of the atomic positions when projected onto this plane.
Overlap between molecular PBFs may be calculated by projecting one molecular PBF onto the plane of the other, and calculating the area of intersection of their corresponding quadrilaterals in this plane. More useful than the raw area, though, is the ratio of the area of intersection to the total area of the PBF. Thus, 'overlap' corresponds to: Two PBFs and are said to be 'stacked' when they satisfy all of the following criteria: 1 2 1. The smallest angle between their plane normals ( ) is less than 25 degrees, 1 , 2 2. The distance between the two plane centroids is less than 8.0 Å, 3. The smallest angle between the vector from centroid to centroid ( ) and normal 1 2 12 is less than 30 degrees. 4. The closest distance between the two planes (i.e. projection of in ) is less than The PBFS are then added to a 'stack' of other planes if that plane stacks with one of the existing planes (according to the above criteria) and it is facing approximately the same direction as the stack (the angle between the average of the plane normals in the stack and the candidate plane normal less than 30 degrees).
To what degree these stacks are facing the same direction (i.e. the co-directionality) of the stacks can then be measured by comparison of the average direction of the stacks, and crystal structures with no stacks, or where the maximum stack size was less than 4 were considered 'unstacked'.
For all of the stacking analysis, all molecules whose centroids were within 35 Å of the asymmetric unit centroid were considered. The code for this analysis is available upon request.

DFT potential calculations
The vertical ionization potential (IP) and electron affinity (EA) of TBAP, TPhP and TPyP were calculated using a ΔDFT approach. First the ground state geometry of each as an isolated molecule was optimized using the B97-3c approach by Grimme and co-workers. [64] Next the energy of each of the molecules in its neutral (E(N)), cation (E(N-1)) and anionic (E(N+1)) state were obtained from single-point calculations using the B3LYP functional [80][81][82][83] and the 6-31G** basis-set. [84,85] Finally, IP and EA were calculated from: Where all energies are in eV and the subtraction by 4.44 converts the calculated IP and EA from the vacuum to the standard hydrogen electrode scale. The B3LYP single-point calculations were performed using Gaussian16 [86] and employed the PCM solvation model [87] to describe the aqueous environment of the molecules near the molecular solid -solution interface. The B97-3c calculations were performed using Turbomole 7.3 [88] and employed no solvation model.
In the case of TBAP the IP and EA values were also calculated using an alternative strategy, starting from the crystal structure. Here we first the experimental crystal structure of TBAP is energy minimized in a periodic DFT calculation using the B97-3c approach as implemented in Crystal17. [89] Subsequently, three cluster models were cut out of the DFT optimized crystal structure, corresponding to one monomer, one monomer (1C) with a molecule above and below it, as well as the phenyl groups of the laterally adjacent molecules (1C+, see Fig. S49), and an analogous structure with a tetramer in the centre (4C+). The IP and EA values of the three cluster models were calculated in the same way as for the isolated molecules discussed above, other than that in the last two cases we used the ONIOM QM/MM approach [90] and described the molecule (fragments) around the monomer and tetramer using the UFF forcefield. [91]

Experimental Methods
All reagents were obtained from Sigma-Aldrich, TCI Europe, Fisher, Manchester Organics, Alfa Aesar, and Combi-Blocks and used as received. Anhydrous solvents were purchased from Acros Organics and used without further purification. All gases for sorption analysis were supplied by BOC at a purity of ≥99.999%. Reactions were carried out under nitrogen atmosphere using standard Schlenk techniques.
Nuclear Magnetic Resonance Spectroscopy NMR spectra were recorded on a Bruker 400 NMR spectrometer at 400 MHz ( 1 H) and referenced against the residual 1 H signal of the solvent.

Inductively Coupled Plasma Mass Spectrometry
Samples were collected after photocatalysis by filtration, dried and digested in HNO 3 with a microwave (Perkin Elmer Microwave Titan) using standard analytical procedures. ICP-MS measurements were performed on a Perkin Elmer ICP MS NexION 2000 and analyzed using external calibrations with Ge as the internal standard.
Thermogravimetric Analysis TGA measurements were carried out using a TA Q5000IR analyzer with an automated vertical overhead thermobalance. Samples were heated at a rate of 5 °C min -1 unless otherwise stated.

UV-Vis Measurements
The UV-visible absorption spectra of the materials were recorded on a Shimadzu UV-2550 UV-vis spectrometer as powders in the solid-state. The band-gap of the materials was calculated via E (eV) = 1243.125/λg (nm). The solution spectra were recorded on an Agilent Cary 5000 UV-Vis-NIR spectrometer in DMSO solution.

Water Sorption
Water vapor isotherms were determined at 293 K using an IGA gravimetric adsorption apparatus (Hiden Isochema) with an anti-condensation system carried out in an ultrahigh vacuum system equipped with a diaphragm and turbo pumps. Crystallization of Other TBAP Phases 500 mg of as-synthesized TBAP were covered by DMF, N,N-dimethylacetamide, or Nmethyl-2-pyrrolidone (40 mL) in a large vial. The mixture was sonicated for 10 min. and left overnight so all remaining undissolved material settled at the bottom of the vial. The supernatant solutions from each dipolar aprotic solvent were filtered into 5 2.5 mL vials using a 0.45 µm PTFE syringe filter to remove any particulates. The vials were closed with a plastic cap that each had a whole bore into it. These 5 vials from each solvent were then placed into larger 15 mL vials with 2 mL of each anti-solvent (CHCl 3 , acetone, THF, 1,4dioxane, ethyl acetate) and were then capped tightly. Vapor diffusion of the anti-solvents into the TBAP solutions were carried out for several days until the vials were nearly full of solvent and in most cases, crystal growth had occurred. These crystals were then analyzed by single crystal x-ray diffraction. For unit cell and full collection results, see Table S5. The solvents N-methyl-2-pyrrolidone and N,N-dimethylacetamide lead to 1:4 solvates, which featured complementary hydrogen bonding motifs between the crystallization solvent and carboxylic acid group of TBAP (SI, Fig. S16-17), rather than TBAP-α, but second TBAP polymorphs can also be isolated by heating a DMF/EtOH solution at 90 °C. [109] Time-Correlated Single Photon Counting Experiments TCSPC experiments were performed on an Edinburgh Instruments LS980-D2S2-STM spectrometer equipped with picosecond pulsed LED excitation sources and a R928 detector, with a stop count rate below 3%. An EPL-375 diode (λ = 370.5 nm, instrument response 100 ps, fwhm) was used as the light source. Suspensions were prepared by ultrasonicating the materials in water. The instrument response was measured with colloidal silica (LUDOX HS-40, Sigma-Aldrich) at the excitation wavelength. Decay times were fitted in the FAST software using suggested lifetime estimates.

Hydrogen Evolution Experiments
A flask was charged with the photocatalyst (25 mg), aqueous ascorbic acid solution (0.1 M, 25 mL), and sealed with a septum. The resulting suspension was ultrasonicated until the photocatalyst was dispersed before degassing thoroughly by N 2 bubbling for 30 minutes. The stirred reaction mixture was illuminated with a 300 W Newport Xe light source (Model: 6258, Ozone free) for the time specified at a fixed distance under atmospheric pressure. The Xe light source was cooled by water circulating through a metal jacket. Gas samples were taken with a gas-tight syringe and run on a Bruker 450-GC gas chromatograph equipped with a Molecular Sieve 13X 60-80 mesh 1.5 m × ⅛" × 2 mm ss column at 50 °C with an argon flow of 40 mL min -1 . Hydrogen was detected with a thermal conductivity detector, referencing against standard gases with known concentrations of hydrogen. Hydrogen dissolved in the reaction mixture was not measured and the pressure increase generated by the evolved hydrogen was neglected in the calculations. The rates were determined from a linear regression fit and the error is given as the standard deviation of the amount of hydrogen evolved. No hydrogen evolution was observed for aqueous ascorbic acid solutions under λ > 420 nm illumination in absence of a photocatalyst. Where stated, platinum was added to the photocatalyst by either in situ photodeposition of H 2 PtCl 6 (8 wt. % solution in water) or preloaded onto samples by sonication in aqueous suspension of Pt nanoparticles followed by evaporation of water under reduced pressure.

Carbon Monoxide Production Experiments
Borosilicate crimp top vials (Agilent Technologies, 10 mL, 23 × 46 mm) were charged with 5 mg of the photocatalyst and transferred to a Chemspeed Accelerator SWING platform for liquid transfer. Degassed aqueous ascorbic acid solution (0.1 mL) and degassed stock solution of H 2 PtCl 6 were loaded into the system and the whole system was flushed with nitrogen for 4 hours. Liquids were automatically dispensed into the vials and the vials were capped under inert conditions by the system. The vials were then ultrasonicated for 10 minutes before illumination with an Oriel Solar Simulator 94123A (1 Sun, classification IEC 60904-9 2007 spectral match A, uniformity classification A, temporal stability A, 1600 W Xenon lamp, 12 × 12 in. output beam, Air ass 1.5G filter, 350-1000 nm) and continuous dispersion of the photocatalyst on a Stuart roller bar SRT9.. After photocatalysis, the samples were measured on a Shimadzu GC-2010 plus equipped with a BID detector using a HS-20 headspace auto-sampler and sampling from the headspace of the vial. Helium was used as the carrier-gas and the gases were separated on a 5 Å Molseive capillary column. The gas amounts were calculated by referencing against standard gases with known concentrations of hydrogen and CO. Hydrogen and CO dissolved in the reaction mixture was not measured and the pressure increase generated by the evolved hydrogen and CO was neglected in the calculations.
Deuterium Labelling Experiments high-density acid (440 mg) was dissolved in D 2 O (25 mL). TBAP-α (25 mg) was dispersed in this solution by ultrasonication before degassing thoroughly by N 2 bubbling for 30 minutes. The mixture was placed in a quartz vessel and sealed in a reactor under nitrogen. The sample was illuminated with a 300 W Newport Xe light source (Model: 6258, Ozone free) for the time specified at a fixed distance under atmospheric pressure. The Xe light source was cooled by water circulating through a metal jacket. Gas samples from the 1.3 mL headspace of the reactor were analyzed at the time periods specified by a customized HPR-70 batch sampling system from Hiden Analytical using a HAL3F/301 triple filter Mass Spectrometer with a Faraday detector for analysis.

External Quantum Efficiencies
EQEs were measured using LEDs controlled by an IsoTech IPS303DD power supply. The platinum loaded photocatalysts (12 mg) were suspended in aqueous ascorbic acid (0.1 M, 8 mL). An area of 8 cm 2 was illuminated and the light intensity was measured with a ThorLabs S120VC photodiode power sensor controlled by a ThorLabs PM100D Power and Energy Meter Console. The external quantum efficiencies were estimated using the equation below:

Photocurrent Measurements
Chronoamperometric measurements were carried out on an EC-Lab SP-200 (Bio-Logic Science Instruments SAS) in a three-electrode-cell system with a FTO coated glass slide as the working electrode, Ag/AgCl 2 electrode (-0.35 V vs standard hydrogen electrode) as the reference electrode, platinum wire as the counter electrode. The photocatalyst was dispersed in acetone (1 mg mL -1 ) and 100 µL were added to 100 µL of Nafion 117 (5 wt. % in a mixture of lower aliphatic alcohols and water). This mixture (30 µL) was drop-casted onto an FTO glass working electrode. The sample was then dried under in a fume hood overnight at room temperature. A three-electrode-cell system containing an aqueous sodium sulfate solution (0.01 M) was purged with N 2 for 10 minutes. The measurement was performed by illuminating the front of the working electrode with a solar simulator (1 Sun, Oriel Instruments LSH-7320 (IEC ABA certified).        In the crystal structure, there are large 1-D voids that contain disordered electron density that could not be accurately modelled. It was therefore necessary to use the SQUEEZE routine in Platon during the final refinement cycles. [112,113] SQUEEZE located two 649 Å 3 voids which each had a disordered electron count of 289 e-. As a result, 6 DMF and 6 CHCl 3 solvent molecules were tentatively added to the unit cell atom count. Disordered electron density that could be located was modelled as severely disordered CHCl 3 . The disordered CHCl 3 molecules were refined with bond distance restraints (DFIX, DANG and SADI in SHELX).                       Hydrogen production (2.8 ± 0.03 µmol) and a small amount of CO (4.05 ± 0.22 nmol) were observed. Hydrogen production is observed while only a very small amount of CO is produced, which strongly indicates that the hydrogen production is not due to decomposition of the photocatalyst or ascorbic acid.                  Figure S48. Comparison of the IP and EA values of TBAP, TPhP and TPyP when modelled as isolated molecules immersed in water. Solution potentials for the reduction of protons to hydrogen and oxidation of ascorbic acid at pH 2.6, the likely pH of a 0.1 M ascorbic acid solution (left), and reduction of protons and oxidation of triethylamine at pH 11.5, the likely pH of a triethylamine solution, taken from the literature. [68] Fig . S48 shows the predicted IP and EA values of TBAP and TPhP relative to those of the solution potentials relevant to the use of ascorbic acid as a hole scavenger and those of TPhP and TPyP for when using triethylamine instead. The two scenarios differ not only in scavenger but also in the pH of the resulting solution and hence the proton reduction potential is also different. All materials are predicted to have a very large driving force for proton reduction and reasonable driving force for the overall oxidation of ascorbic acid and triethylamine respectively. However, while TBAP and TPyP also have a small driving force for the intermediate oxidation step, this is predicted to be not true for TPhP. In the case of TPhP, the one-hole oxidation of ascorbic acid/triethylamine could thus act as an as an effective thermodynamic barrier to overall oxidation, which might be a contributing factor to the lower hydrogen evolution rates observed for TPhP when compared with TBAP/TPyP.  Table S8 demonstrates that our periodic DFT calculation reproduces the lattice parameters of the desolvated TBAP-α well. The slight contraction of the DFT structure relative to the experimental structure is probably in part related to the fact that the experimental structure solved from diffraction data obtained at room temperature. Figure S49. Structure of the 1C+ cluster model with the monomer described using DFT highlighted in blue. Figure S50. Comparison of the IP and EA values of TBAP based on an isolated TBAP molecule optimized in the gas-phase (I), an isolated TBAP molecule taken from the DFT optimized TBAP-α crystal structure (II, 1C), a TBAP single molecule taken from the DFT optimized TBAP-α crystal structure surrounded by a molecule above and below it, as well as the phenyl groups of the laterally adjacent molecules, described using a forcefield (III, 1C+, see Fig. S49), and the analogous tetramer case (IV, 4C+). Solution potentials for the reduction of protons and oxidation of ascorbic acid at pH 2.6, the likely pH of a 0.1 M ascorbic acid solution, taken from the literature. [68] Fig. S50 shows that the effect of packing on the predicted potentials of TBAP-α is very small, or at least when we make the assumption that the pores contain significant amounts of water. For all structural models it is predicted that there is a very large driving force for proton reduction and a reasonable driving force for the overall oxidation of ascorbic acid (A/H 2 A). The intermediate one-hole oxidation of ascorbic acid (HA*/H 2 A), which could act as an effective thermodynamic barrier to overall oxidation, is also in all cases predicted to be exergonic, though only by a small amount.

External Surface area Analysis
It is important to consider the effect of external surface area of the photocatalysts; particle size analysis by static light scattering indicated that both TPhP-α and TPyP-α had smaller surface area weighted particle sizes than TBAP-α when suspended in water (SI, Figure S63 and Table S9) and that the former was also smaller in acidic solution. In addition, light transmission measurements for suspensions of the materials (SI, Figure S65) indicate that TPhP-α and TBAP-α have comparable dispersibilities and that TPyP-α was significantly more dispersible. In addition the amorphous TBAP sample was found to have particle sizes, intermediate between the three crystalline TBAP-α batches ( Figure S62, Table S9) and to show similar dispersibility in ascorbic acid ( Figure S65). Taken together these results show no strong correlation between the materials particle size or dispersibility and photocatalytic activity, and thus we conclude external surface area is not a dominant factor in these systems.