Supporting Information: Multiscale Modeling of the HKUST-1/Poly(vinyl alcohol) Interface: From an Atomistic to a Coarse Graining Approach

Transcription

1 Supporting Information: Multiscale Modeling of the HKUST-1/Poly(vinyl alcohol) Interface: From an Atomistic to a Coarse Graining Approach Rocio Semino,* Johannes P. Dürholt, Rochus Schmid,* Guillaume Maurin Institut Charles Gerhardt Montpellier UMR 5253 CNRS, Université de Montpellier, Place E. Bataillon, Montpellier Cedex 05, France Computational Materials Chemistry Group, Lehrstuhl für Anorganische Chemie 2, Ruhr-Universität Bochum, Bochum, Germany * rocio.semino@umontpellier.fr, rochus.schmid@rub.de Table of Contents I. METHODOLOGY.. S2 I.I Atomistic Interface Model... I.I.I MOF Atomistic Model.. I.I.II Polymer Atomistic Model.. I.II CG Model I.II.I MOF CG Model. I.II.II Polymer CG Model I.III Development of the Interface CG Model I.IV Interface CG Production Runs I.III Computational Details. II. RESULTS AND DISCUSSION. II.I Atomistic Interface.. II.II CG Interface. III. PARAMETER SETS III.I Atomistic Model III.II Coarse Grained Model References S2 S3 S4 S8 S8 S11 S12 S20 S20 S20 S20 S25 S26 S27 S30 S31 S1

2 I. METHODOLOGY I.I Atomistic Interface Model We started our analysis by studying the HKUST-1/PVOH interface with atomistic resolution. In order to build it, we combined previously equilibrated models for the MOF and the polymer in a series of force field-based MD simulations spanning a total time of 1.56 ns, following the methodology derived by R. Semino et al.. 1 Thermodynamic conditions for the simulations are listed in Table S1. Seven cycles of three MD simulations were performed. The first and second simulations in each cycle are run in the NVT ensemble, with constant temperatures Tmax = 600 K and Tmin = 300 K; and the last in the NPnT ensemble, to model pressure changes in the direction normal to the MOF surface slab (P = Pn). P increases up to P = Pmax = 1 kbar in the first three cycles and decreases in the last four until reaching P = kbar. This procedure was inspired by the widely used polymer equilibration protocol proposed by D. Hofmann et al, 2 and its implementation here allows the polymer to adapt its configuration to the environment offered by the morphology and chemical functions of the MOF surface slab, in order to obtain a thermodynamically sound interface model. Table S1. Details of the thermodynamic conditions for the 21 MD steps used in the interface generation protocol. Step Ensemble T (K) P(kbar) Length (ps) 1 NVT Tmax = NVT Tmin = NPnT Tmin 0.02 (0.02 Pmax) 50 4 NVT Tmax 50 5 NVT Tmin NPnT Tmin 0.6 (0.6 Pmax) 50 7 NVT Tmax 50 8 NVT Tmin NPnT Tmin 1 (Pmax) NVT Tmax 50 S2

3 11 NVT Tmin NPnT Tmin 0.5 (0.5 Pmax) 5 13 NVT Tmax 5 14 NVT Tmin NPnT Tmin 0.1 (0.1 Pmax) 5 16 NVT Tmax 5 17 NVT Tmin NPnT Tmin 0.01 (0.01 Pmax) 5 19 NVT Tmax 5 20 NVT Tmin NPnT Tmin Tmax and Pmax were set by studying the properties of the system. Tmax needs to be higher than the glass transition temperature (358 K for PVOH 3 ), and high enough to allow for the polymer chains to move. Pmax should be high enough to overcome the rigidity of the polymer and achieve a realistic density (the polymer is usually built at a much lower density than it truly has). We selected these parameters by checking that the experimental density was well reproduced in pure PVOH simulations. Note that for PVOH Pmax = 1 kbar, while for much more rigid polymers such as PIM-1 and PIM-EA-TB a much higher value of 50 kbar is needed. 1,4,5 Additional details on the interface generation methodology can be found in Ref. 1. I.I.I MOF Atomistic Model We employed MOF-FF for the atomistic simulations. MOF-FF is an ab initio parameterized force field, designed for treating MOF systems with high accuracy. It is available for a range of MOFs and will be constantly enlarged. 6 MOF-FF has been used several times to study different properties of HKUST-1 and other copper paddle-wheel based systems. 7,8 Amirjalayer and Schmid have predicted the [111] surface of HKUST-1 to be the most stable. For this reason, the [111] surface terminated by acetate groups and its corresponding parameterization was also employed in this study. 9 S3

4 I.I.II Polymer Atomistic Model For PVOH, we have adapted a CHARMM force field 10 previously reported 11 so that it is compatible with MOF-FF. Since Gaussian charge distributions are used in MOF-FF, Gaussians with a very small σi of 0.01 Å were employed for PVOH in order to reproduce the original point charge behavior. The interaction energy between two point charges is: E P ij(r ij ) = 1 4πε 0 q i q j 1 r ij, whereas the interaction energy of two Gaussian type charge distributions is given by E G ij(r ij ) = 1 erf(r ij σ ij ) q 4πε i q j. 0 r ij Assuming that the minimal distance between two atoms/beads is at least 1.0 Å, for such small σ ij = σ i 2 + σ j 2 it holds with numerical accuracy, that E ij G = E ij G, since erf(1.0 Å 0.01 Å) = erf(100) 1.0. The PVOH models were compared by setting up a model system consisting of 274 monomers (2166 atoms), grouped in 31 chains between 8 and 10 monomers long in a cubic box of L = 25 Å. This model was built by an in silico polymerization procedure using the polymatic code 12 and lammps 13 for the MD simulations. In the following, the original CHARMM force field for PVOH is denoted as PVOH1 and the MOF-FF compatible version PVOH2. Non-bonded parameters for both force fields are listed in Table S2. Figure S1. Scheme of Poly(vinyl alcohol). S4

5 Table S2. Non-bonded parameters assigned to the different atom types in PVOH1 and PVOH2. The atom types are labeled as in Figure S1. Atom Type Charge (e) PVOH1 (12-6 Lennard Jones) PVOH2 (Buckingham) εii (kcal mol -1 ) rii 0 / 2 (Å) εii (kcal mol -1 ) rii 0 / 2 (Å) 1 (CT1) (CT2) ,5,6,9 (HA) (OH1) (HC) (CT3) Coulombic charges were also taken from the CHARMM force field (Appendix Figure 2, section 2, J). 10 The ends of the chains were terminated by adding CH3 groups (see Figure S1). The systems were first equilibrated following the 21 steps protocol. 2 Afterwards, 5 ns long trajectories in the NPT ensemble were produced for data analysis with the two force fields, with T = 300 K, P = 1 bar and Berendsen thermostat and barostat 14 with relaxation times of 0.1 and 0.5 ps respectively. The density and x-ray diffraction patterns were computed to validate our models with corresponding experimental data. PVOH1 model was found to slightly underestimate the density, with a value of 1.07 g cm -3, while for the PVOH2 model it was of 1.23 g cm -3, which lies in the experimental range, between 1.19 and 1.31 g cm Figure S2 shows the calculated x-ray diffraction patterns using the Forcite Analysis module in Materials Studio 15 for PVOH1 and PVOH2 in black and red dots respectively. The shape of the pattern is similar for both models, both exhibit one peak around 2θ = 20 degrees. Measured diffraction patterns depend on the kind of sample (film, gel) and the synthesis procedure, but the general shape and the peak around 20 degrees are consistent with those reported in the literature (see Figures 1 and 6b in Ref. 16 and Figure 10 in Ref. 17). S5

6 Figure S2. Calculated powder x-ray diffraction patterns for the PVOH1 model (black dots) and the PVOH2 model (red dots). The radius of gyration and some radial distribution functions were computed in order to further compare the two force field models in terms of the structural properties of the polymer. The radius of gyration represents the distribution of the monomers around the center of mass of the polymer, and it can be calculated by the following equation: r gyration = 1 n (r n i=1 i r com ) 2 where n is the number of sites, r com and r i are the position of the center of mass of the polymer and of site i respectively, and denotes an ensemble average. PVOH1 and PVOH2 models have similar average radii of gyration of 4.7 and 4.5 Å respectively (average values over the radius of gyration for each chain and over all the configurations collected along the 5 ns production runs). The small difference between the values is probably due to the fact that PVOH2 model has a higher density than PVOH1, and thus better packing and lower radius of gyration. Site-site radial distribution functions (rdf) were computed as follows: S6

7 g αβ (r) = 1 4 π r 2 ρ β δ ( r i β r α r) i where α and β are the two types of sites considered, ρ β is the density of sites of type β and r β i is the position of site i of type β. Figure S3 compares several rdfs for the two polymer models (black for PVOH1 and red for PVOH2), namely CT1-CT1, CT2-CT2 and OH1-HC. Two kinds of rdfs have been computed for each model: one considering all the atom pairs (dashed lines), and the other only considering atom pairs when they belong to different polymer chains (full lines). The rdfs for the two models are overall similar, except for a shift of 0.25 Å towards lower radius values for the peaks in the OH1-HC rdfs for the PVOH2 model, both for the all sites and the different chains sites cases (see left bottom panel). This is consistent with the higher density of PVOH2 model. In the carbon-carbon cases, the rdfs are practically identical for the two models in the different chains profiles, and slightly differ in the all atom cases, again with a shift for lower radius values. The peaks observed in the carbon-carbon rdfs correspond to bonded monomers, since they are absent in the different chains case. In contrast, for the OH1-HC rdf there is a peak for the different chains profile as well, meaning that the hydrogen bonding between the OH groups is not only present between bonded monomers, but also it is the main interaction that holds the polymer chains together. S7

8 Figure S3. Site-site pair correlation functions for the pure polymers, modeled in an atomistic resolution, by using PVOH1 and PVOH2 models. Left upper panel: CT1-CT1; right upper panel: CT2-CT2; left bottom panel: OH1-HC. Labels for the atom types can be found in Table S2. I.II CG Model I.II.I MOF CG Model Recently, Dürholt et al. published a first CG force field for HKUST-1, 8 parameterized on the basis of the atomistic MOF-FF. In that work, a maximal coarse graining was applied, where each building block was represented by only one bead. This CG force field was able to predict local deformation energies of the building blocks as well as bulk properties like the difference between the framework polymorphs tbo and pto in terms of energies and elastic constants in a semi-quantitative way. Here, we used the same parameterization and validation methodology that was used before to derive a CG model with a less coarse mapping in order to retain more chemically detailed information. The new model consists of S8

9 six beads for the copper paddlewheel and three beads for one BTC linker (see Figure S4). This representation was inspired by the MARTINI force field. 18 Like in the MARTINI coarse grained force field, the phenyl group consists of three beads and the distance between them is 2.7 nm, the same as in the MARTINI force field. To properly model the size of the paddle wheel, two beads were kept at the copper positions, whereas the carboxylate groups were reduced to a single bead. Figure S4. CG mapping (opaque) superimposed to the atomistic model (transparent). Color code: carbon (black), oxygen (red), hydrogen (white), copper (ochre). The force field was parameterized in respect to reference information at the MOF-FF level of theory obtained for a copper paddlewheel saturated with phenyls and for a copper paddlewheel saturated with methyl groups. The latter one was only used for the parameterization concerning the acetate groups on the surface. Every surface acetate group is represented by one bead. As outlined in Ref. 8, a coarse grained Hessian and geometry S9

10 was calculated for the atomistic system and the coarse grained force field was fitted in respect to this information. Like in the atomistic MOF-FF, the fit was performed in redundant internal coordinates. In this work, the Covariance Matrix Adapted Evolution Strategy (CMA- ES) 19 developed by Hansen was used as optimizer instead of the genetic algorithm PIKAIA 20 used before, since CMA-ES features faster convergence and better fitness values. In contrast to the previous CG model of HKUST-1, in this model some beads are charged. The beads representing copper atoms were assigned exactly the same charge as the copper atoms in MOF-FF, namely 1.06e and those representing the carboxylates, a charge of which is the sum of those assigned to the atoms belonging to the carboxylate unit in MOF-FF. Since the pydlpoly code 9 is designed for using Gaussian charge distribution instead of point charge distributions, Gaussians with very small widths were employed in order to mimic a point charge-like behavior. For the van der Waals interactions, the same dispersion damped Buckingham potential as for the atomistic MOF-FF was employed. Consequently, the beads representing the copper atoms were assigned exactly the same r 0 ii and ε ii as in the atomistic case. The phenyl beads r 0 ii and ε ii values were those recommended from the MARTINI force field for ring systems, and they were also used for the surface terminations. This model was validated in analogy to the recently published maximal coarse grained FF for HKUST-1, by comparing the energy differences between the tbo and pto structures and their elastic constants. With MOF-FF, the energy difference between the tbo and pto phase is E = kcal mol -1. By increasing the chemical resolution from the maximal coarse grained model to the new one, the error is reduced from E = 3.67 kcal mol -1 to E = 0.38 kcal mol -1. Further validation results are shown in Table S3. S10

11 Table S3. Comparison of elastic constants (B for Bulk modulus, E for Young s modulus) and lattice parameters (L) of the tbo and pto phases of HKUST-1 obtained with MOF-FF (AA), the maximal coarse grained model (MAX) and the model used in this study (BB). tbo pto AA MAX BB AA MAX BB L [Å] E [GPa] B [GPa] I.II.II Polymer CG Model For the coarse grained model of PVOH, we took as a basis an existing CG model by A. Gautieri and coworkers, 11 with two modifications: (i) the monomer size was increased and (ii) the van der Waals term was changed from 12-6 Lennard Jones to the Buckingham form, as discussed previously. In this model, one bead is located at the center of mass of each monomer (reduction 7 to 1) and the beads interact through a harmonic potential when bonded, and through a van der Waals potential otherwise. The original model had been developed with the objective of studying the diffusion of small molecules within the polymer matrix, and thus, it was validated by comparison of the diffusion coefficients extracted from the simulation with those coming from experimental measurements. For our purpose, we needed a model that accurately reproduced the structural properties of the bulk polymer, such as density and radial distribution functions. This was achieved by increasing the r 0 ii 2-1 parameter of the van der Waals interaction from 2.05 to 2.64 Å. The new, modified model corresponds to a density of 1.19 g cm -3, in the experimental range of g cm -3, 3 and its bead-bead rdf matches better with the atomistic center of mass center of mass (com) for the PVOH2 model, in terms of the position of the second peak (see Figure S5). For the computed rdfs, all com-com and bead-bead pairs were considered, even those for coms/beads belonging to the same polymer chain. S11

12 Figure S5. Comparison of site-site pair correlation functions for atomistic com-com (black) and CG PVOH bead-bead (red). The atomistic curve is an average for the 10 MD trajectories. I.III Development of the Interface CG Model First, an initial configuration for the CG PVOH was generated. For this, an atomistic configuration was mapped into the CG space by assigning one bead to the center of mass of each monomer. Terminations of the atomistic chains were discarded. As a second step, this model and that for the CG HKUST-1 slab were put together in an orthorhombic simulation box, in such a way that the MOF was centered in the direction perpendicular to the surface (the z direction), and the polymer was placed above and below the MOF, cut by the periodic boundary conditions in the z direction. The third step was to equilibrate the polymer in the presence of the MOF slab. To this end, we applied a similar approach to the one we had applied to build the HKUST-1/PVOH atomistic interface, 1 a 21 steps scheme of MD simulations followed by 10 ns long NVT runs (see details in section I.I). For all the CG simulations, we used a timestep of 1 fs. The CG approach is conceived to allow for larger S12

13 timesteps, in order to obtain information at longer timescales, but since in this work we have focused in obtaining structural information and not dynamics, we have kept the same timestep as for the atomistic simulations. Note, however, that the time scale for CG simulations loses the physical meaning that it has for atomistic simulations, because of the removal of many degrees of freedom from the system. In order to equilibrate the polymer in the presence of the MOF, the force field for MOF/polymer interactions needs to be defined, and Pmax and Tmax parameters need to be chosen. As a first approximation, we chose for Pmax and Tmax the same values as for the atomistic interface: 1 kbar and 600 K. The MOF/polymer interaction potential was considered as the sum of dispersion damped Buckingham van der Waals terms only, since the polarization of the polymer is not included in this one-bead-per-monomer coarse model. For these interactions, we took as a first approximation the parameters given by applying the Lorentz-Berthelot mixing rules, 21 as for the rest of the crossed-term interactions in this study. A representative snapshot of the resulting CG interface model is presented in Figure S6. S13

14 Figure S6. Illustration of the first CG model interface developed. MOF/polymer interactions are modeled by computing crossed-terms parameters by the Lorentz-Berthelot mixing rules. The polymer is depicted in orange, and the MOF beads in cyan (Cu), blue (benzene beads), pink (COO) and white (terminations). Simple visual inspection of a typical configuration for the CG interface obtained using this model, shows that the polymer penetrates into the second pore layer of the MOF. This deviates from the results obtained by atomistic simulation, where the polymer was found to penetrate into the first pore layer only (see section II.I). To explain this behavior, we decided to first establish whether our methodology was good for the CG system. For instance, it could be that results were too sensible to the chosen Pmax and Tmax parameters. We thus explored three different possible values for each parameter: Pmax = 0.2, 0.5, 1 kbar and Tmax = 400, 600, 800 K. In all these we found the same polymer penetration behavior, with changes in the volume of the simulation boxes of less than 2%, hence proving the robustness of the S14

15 methodology for these systems. From here onwards, the chosen Pmax and Tmax were kept as 1 kbar and 600 K respectively, as for the atomistic model. The other possible cause for the excess polymer penetration would be the potential used to model MOF/polymer interactions, most probably because of the absence of coulombic interactions, that strongly bind the polymer to the pore walls in the atomistic representation. Indeed, we have conducted some tests with dimers and trimers of PVOH inside the pores of bulk HKUST-1, and we have verified that in this non-charged model, the interactions between the MOF and the dimers/trimers are not well reproduced with respect to the atomistic case. A possible way to tackle this problem would be to choose a CG polymer model with two beads per monomer or more, in order to be able to include charges. However, because our objective is to have a model as coarse as possible to model much larger systems, we decided to keep the 7-1 un-charged model, but to decrease the interaction between the polymer and the MOF by reducing the value of the ε ij parameters of all 4 crossed-terms in 90, 80, 70, 60 and 50% of that obtained by the Lorentz-Berthelot approximation. Representative snapshots of the resulting interfaces are displayed in Figure S7. S15

16 Figure S7. Illustration of different CG models for the HKUST-1/PVOH interface, where the crossed-term interactions were modeled by Buckingham van der Waals potentials, with r 0 ij computed by Lorentz-Berthelot mixing rules and ε ij as a percentage of the Lorentz- Berthelot value (100% is the same as in Figure S6). Color code is that of Figure S6. Red circles indicate polymer chains penetrating the second pore layer in the MOF slab. Figure S7 shows that when the strength of the interaction is decreased by reducing the depth of the potential well from 100 to 50 %, the polymer gradually retreats from the open pores of the MOF slab. From 100 to 80 %, some polymer chains enter the second pore layer. As opposed, for the 50% model the polymer only fills the first pore layer partially. 70 and 60 % models exhibit intermediate behaviors. From these results, it can be concluded that the model of choice needs to have HKUST-1/PVOH interaction potentials with S16

17 ε ij between 80 and 60 % of the originally computed by applying the Lorentz-Berthelot mixing rules. We first compared the atom distribution of both PVOH and HKUST-1 as a function of the z coordinate, normal to the surface slab. For this, we have normalized the particle density by a factor of 2.6 and 7 for the beads in CG HKUST-1 and PVOH respectively, in order to account for the factor of coarsening. Figure S8 depicts the resulting density profiles. S17

18 Figure S8. Atom density profiles for a typical atomistic configuration (top panel) and typical configurations for the different CG models (bottom panel). Orange: HKUST-1 atoms/beads, black: atomistic PVOH, red: PVOH CG model 60%, green: PVOH CG model 70% and blue: PVOH CG model 80%. The z axis has been centered in zero for clarity purposes. CG model 60% reproduces the overall shape of the atomistic density profile for the polymer much better than the others, which are too structured. Model 70% is the best if the penetration is taken into account, since the atom density drops to zero at values around + 7 Å, close to the Å for the atomistic case. Model 60% underestimates the penetration while model 80 % overestimates it, with zero polymer density values around + 9 Å and + 5 Å respectively. Radial distribution functions for the polymer beads versus the MOF beads were also computed and compared with the atomistic case (there, the center of mass of the corresponding moiety was taken into consideration for the comparisons). The positions of the peaks were well reproduced by all models. The stronger the interaction (higher percentage), the more artificially structured the rdf became. It is of particular interest to compute spatially restricted rdfs, in particular, those related to the penetration of the polymer, such as the monomer versus the internal Cu layer in HKUST-1, shown in Figure S9. S18

19 Figure S9. Site-site pair correlation function CuinternalHKUST-1 COM/bead PVOH (left panel). Black: average atomistic results, the other colors correspond to different CG models, 60% (red), 70% (green) and 80% (blue). In the right there is a picture of the HKUST-1 slab, where the different layers of Cu atoms (external, middle and internal) are labeled. Color codes are those of Figure S4. The general trend already described can also be seen in Figure S9: the overall shape of the atomistic rdf is best reproduced by model 60% (red versus black line). Once again, the penetration extent observed for this model underestimates the atomistic one, with a minimum CuinternalHKUST-1 - bead PVOH distance of 9.4 Å, while the atomistic curve has a small broad peak in the low radius region with a maximum around 8.2 Å. Consistently, model 70% is the best in this aspect, while model 80% overestimates the penetration, as can be seen by the much larger area under the peak for r < 10 Å. In order to further refine the MOF/polymer non-bonded terms for the CG interface model, we have studied the radius of gyration, since one of our objectives is to use this model to compute the length and extent of the variations in the structure of PVOH as a function of the distance to the HKUST slab. The average radius of gyration for the atomistic S19

20 model is (7.2 ± 0.2) Å, where the error was computed as the standard deviation of the values for the 10 MD runs. The three CG models, 60, 70 and 80%, show values of 7.3, 7.6 and 7.7 Å respectively. In this sense, model 60% is the only one whose value fits within the interval given by the error bars of the atomistic model. The average radius of gyration is not only an ensemble average, but also an average of the radius of gyration obtained for chains of different lengths. We also plotted the ensemble distribution and the distribution as a function of the number of monomers in the polymer chain for the atomistic and the 60% CG model, and found that there is an excellent agreement between model 60% and the atomistic one (see Figure 5 in the manuscript). From all the points analyzed, we conclude that the 60% CG model is the one that best reproduces the chosen set of properties of the atomistic interface relevant for this work, and thus, we choose it to continue with our analysis. I.IV Interface CG Production Runs First, we performed several cycles of annealing spanning 5 ns (1000 to 5000 K and 300 K as low T) on the polymer to generate different initial configurations. Then, we equilibrated the polymer in order to obtain its bulk properties, using the 21 steps scheme. 2 Finally, we put the different model polymers in contact with the HKUST-1 slab and subsequently followed our strategy to let them equilibrate in the presence of the external field imposed by the MOF. 5 independent MD runs of 10 ns were produced, and statistics regarding the density profile and the radius of gyration distributions were collected. I.III Computational Details MD simulations (atomistic and coarse grained) were mostly performed by pydlpoly, 9 except for the generation of the atomistic polymer models, for which we used lammps. 13 pydlpoly is a molecular mechanics code developed by R. Schmid et al. that consists of an adaptation of the freely available code DLPOLY Classic 22 for the MD core, and incorporates a python interface. This code is adapted to the use of MOF-FF potentials 9 and, for the purpose of this work, it has also been adapted to allow MD simulations in the NPnT ensemble. S20

21 DLPOLY Classic allows for the simulation of systems up to particles, the largest system size reported in this work consists of a total of atoms. II. RESULTS AND DISCUSSION II.I Atomistic Interface We start our analysis by a visual inspection of the atomistic interfaces, a representative snapshot of the system is shown in Figure S10. Interestingly, PVOH penetrates into the open pores of HKUST-1, up to the first pore layer, and it completely fills it. This picture is in contrast to what had been found for the PIMs in ZIF-8, 1,4 in that case, microvoids could be found between the MOF and the polymer phases, indicating that the compatibility between them was not optimal. Moreover, the existence of these voids might be detrimental for the use of these materials for gas separation purposes, since they could act like a trap for the gas molecules. Instead, the penetration of PVOH inside the open pores guarantees a continuity when passing from one phase to the other. The penetration can be ascribed both to the higher size of the pores of HKUST-1 with respect to ZIF-8, and also to the much higher flexibility of PVOH than that of the PIMs, 23 as well as to strong intermolecular interactions, as will be shown. S21

22 Figure S10. Snapshot of the atomistic HKUST-1/PVOH interface. The MOF atoms are color coded as in Figure S4. All polymer atoms are shown in grey, for clarity purposes. Image produced by using VMD. 24 Further inspection shows that the polymer does not reach the second pore layer due to the steric hindrance by the three organic linkers it encounters in its way (see Figure S11a, top view of the MOF surface slab). Only the terminations can in some few cases circumvent the BTC units blocking the polymer way, and enter the second pore layer (see yellow circle in Figure S11b). S22

23 Figure S11. Snapshots: a) top view of the HKUST-1 atomistic slab, b) atomistic HKUST- 1/PVOH interface, where only the polymer terminations are shown. Color codes are those of Figure S10. Images produced by using VMD. 24 Finally, the interactions between the polymer and the MOF were studied by computing several site-site pair correlation functions. First, the interactions of both the carbon chain and the OH group of PVOH with the terminations of the HKUST-1 slab were considered. Average results from the 10 MD independent runs are shown in Figure S12. The terminations of the MOF were found to strongly interact with the OH group of the polymer, with a characteristic H(Me)HKUST-1 - O(OH1)PVOH distance of 2.71 Å. S23

24 Figure S12. Site-site pair correlation functions for the H of the acetate terminations of the HKUST-1 slab and several polymers sites (left upper panel: CT1 and CT2; right upper panel: HC; left bottom panel: OH1), from the atomistic interface model. Labels for the involved polymer atom types are displayed in the scheme. As a further step, we computed radial distribution functions for the polymer sites versus the inner sites in the MOF (oxygen and copper), results are presented in Figure 3. The overall picture is of a strongly interacting MOF/polymer pair, with excellent compatibility. The high number of strong intermolecular interactions makes it improbable that the polymer could go further into the second pore layer, even for timescales larger than those explored in the atomistic molecular dynamics simulations. In our CG MD long runs, we have not seen any signs of further polymer penetration. Nevertheless, enhanced sampling techniques, such as metadynamics 25 or biased potential dynamics 26 could be applied for further confirmation. S24

25 II.II CG Interface The density profiles for HKUST-1 and PVOH were calculated for the large interface and compared to those corresponding to the atomistic and the CG model used for the development stage, results are presented in Figure 6 (see manuscript). The general trend for the atomistic model is very well reproduced by the CG model, both in the development (same size of the interface) and production stages (larger interface), as can be seen by the good agreement between the black, red and green curves in the ±(10,25) Å intervals of the density profile (region A and part of region B). Further, the end of region B, which marks the start of the bulk polymer, can be determined from the CG simulation. From an average of the 5 independent MD simulations, it was found that the length of region A is of 11.5 Å (first pore layer), and that the structure of PVOH is affected by the presence of HKUST-1 up to an average distance of 17 Å from the terminations of the surface slab. We further explored the polymer structure in the different regions, in terms of the radius of gyration. The radius of gyration of the interfacial polymer is slightly smaller than for the bulk, (7.1 ± 0.1) Å versus (7.3 ± 0.1) Å. These values are averages over the independent MD trajectories and the errors come from the standard deviation. We assessed the different regions separately, and found that it is region A where the radius of gyration is smaller, of (6.9 ± 0.1) Å whereas for region B, the value is comparable to that of the bulk, (7.4 ± 0.1) Å. We plotted the chain size distribution and the radius of gyration distribution for the different regions in Figure S13. The chains are very similarly distributed in the different regions, as can be appreciated by comparing the top panels. As a consequence, a real change in the structure must occur, probably due to the confinement inside the pores, which forces the polymer to pack more efficiently. S25

26 Figure S13. Top panels: distribution of the polymer chains as a function of the number of monomers N. Bottom panels: distribution of the radius of gyration. Results for the different regions A, B, and C are presented. III. PARAMETER SETS In this section, the final force field parameters are given both for the atomistic and for the coarse grained model in the format of a pydlpoly key-file. All reference distances are given in Å and all angles in degrees. Force constants are given in mdyn/å (bonds) or mdynå/rad 2 (angles) and the torsional barriers in kcal/mol. S26

27 III.I Atomistic Model The atom types for HKUST-1 are assigned in the following way: copper atoms are assigned number 165, carboxylate carbon atoms 168, carboxylate oxygen atoms 167, phenyl carbon atoms 2, phenyl hydrogen atoms 5, acetate carbon atoms 1. ### atom types atom 5 h atom 1 c atom 2 c atom 168 c atom 165 cu atom 167 o ### vdw parameters first column r_min in angstrom ### second column potential depth in kcal/mol vdw vdw vdw vdw vdw vdw ### charges, first column charge ### second column width of gaussian in angstrom charge charge charge charge charge charge # modify charge parameters for atom with type 2 if it is bonded to an atom # of type 168 (alpha carbon), first column charge, second column width # of gaussian in angstrom chargemod ### parameters for bonds, first column force constants ### second column bond length ### if third column present, a morse pot is applied bond bond bond bond bond bond bond bond ### parameters for angles, first column force constants ### second column bond angle angle angle angle angle angle angle angle angle S27

28 angle angle ### fourier angle potential, first column barrier in kcal/mol ### second column reference angle, third column multiplicity anglef ### stretch bend coupling terms strbnd strbnd strbnd ### dihedral potentials with barriers up to multiplicity 4 torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion ### improper torsions, first column force constants opbend opbend opbend opbend The atom types for PVOH were assigned as follows: oxygen atoms are labeled 4 and hydrogens in the hydroxy groups 52, while all other hydrogens are 51. Three different types of carbon atoms are distinguished. Carbons carrying the hydroxy group are assigned number 21, whereas the carbon atoms of the CH2 group are 22 and the carbons of the acetate terminations are 23. ### atom types atom 21 c atom 22 c atom 23 c atom 51 h atom 52 h atom 4 o ### vdw parameters first column r_min in angstrom ### second column potential depth in kcal/mol vdw vdw vdw vdw vdw vdw S28

29 ### charges, first column charge ### second column width of gaussian in angstrom charge charge charge charge charge charge ### parameters for bonds, first column force constants ### second column bond length bond bond bond bond bond bond bond bond ### parameters for angles, first column force constants ### second column bond angle angle angle angle angle angle angle angle angle angle angle angle angle angle angle angle angle ### dihedral potentials with barriers up to multiplicity 4 torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion torsion S29

30 III.II Coarse Grained Model Bead types for the HKUST-1 slab are assigned as follows: copper beads are 165, beads representing the carboxylate groups are 102, beads representing the phenyl groups are 103 and those for the acetate terminal are 101. ### atom types atom 102 fe atom 165 cu atom 101 h atom 103 n ### vdw parameters first column r_min in angstrom ### second column potential depth in kcal/mol vdw vdw vdw vdw ### charges, first column charge ### second column width of gaussian in angstrom charge charge charge charge ### parameters for bonds, first column force constants ### second column bond length bond bond bond bond bond ### parameters for angles, first column force constants ### second column bond angle angle angle angle angle angle angle ### fourier angle potential, first column barrier in kcal/mol ### second column reference angle, third column multiplicity anglef ### dihedral potentials with barriers up to multiplicity 4 torsion torsion torsion torsion torsion torsion torsion ### improper torsions, first column force constants opbend opbend S30

31 opbend PVOH beads are assigned number 500. ### atom types atom 500 c ### vdw parameters first column r_min in angstrom ### second column potential depth in kcal/mol vdw ### charges, first column charge ### second column width of gaussian in angstrom charge ### parameters for bonds, first column force constants ### second column bond length bond ### fourier angle potential with multiplicity 4 anglef ### dihedral potentials with barriers up to multiplicity 4 torsion %): In the following, the cross term vdw parameters for the CG model are listed (model ### crossed terms, first column is d_min in angstrom ### second column potential depth in kcal/mol vdwpr vdwpr vdwpr vdwpr References (1) Semino, R.; Ramsahye, N. A.; Ghoufi, A.; Maurin, G. Microscopic Model of the Metal Organic Framework/Polymer Interface: A First Step Toward Understanding the Compatibility in Mixed Matrix Membranes. ACS Appl. Mater. Interfaces 2016, 8, (2) Hofmann D.; Fritz, L.; Ulbrich, J.; Schepers, C.; Böhning, M. Detailed Atomistic Molecular Modeling of Small Molecule Diffusion and Solution Processes in Polymeric Membrane Materials. Macromol. Theory Simul. 2000, 9, (3) Handbook of Polymers, 2012, ChemTec Publishing, G. Wypych, Toronto, Ontario, Canada, S31

32 (4) Benzaqui, M.; Semino, R.; Menguy, N.; Carn, F.; Kundu, T.; Guigner, J.-M.; McKeown, N. B.; Msayib, K. J.; Carta, M.; Malpass-Evans, R.; Le Guillouzer, C.; Clet, G.; Ramsahye, N. A.; Serre, C.; Maurin, G.; Steunou, N. Toward an Understanding of the Microstructure and Interfacial Properties of PIMs/ZIF-8 Mixed Matrix Membranes. ACS Appl. Mater. Interfaces 2016, 8, (5) Larsen, G. S.; Lin, P.; Hart, K. E.; Colina, C. M. Molecular Simulations of PIM-1-like Polymers of Instrinsic Microporosity. Macromolecules 2011, 44, (6) Bureekaew, S.; Amirjalayer, S.; Tafipolsky, M.; Spickermann, C.; Roy, T. K.; Schmid, R. Phys. Status Solidi B 2013, 250, (7) Bureekaew, S.; Balwani, V.; Amirjalayer, S.; Schmid, R. MOF FF A Flexible First Principles Derived Force Field for Metal Organic Frameworks. CrystEngComm 2015, 17, (8) Dürholt, J. P.; Galvelis, R.; Schmid, R. Coarse Graining of Force Fields for Metal Organic Frameworks. Dalton Trans. 2016, 45, (9) Amirjalayer, S.; Tafipolsky, M.; Schmid, R. Surface Termination of the Metal-Organic Framework HKUST-1: A Theoretical Investigation. J. Phys. Chem. Lett. 2014, 5, (10) MacKerell, A. D.; Bashford, D.; Bellott, M.; Dunbrack R. L.; Evanseck, J.D.; Field, M. J.; Fischer, M. J.; Gao, J.; Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F.T. K.; Mattos, C.; Michnick, S.; Ngo, T.; Nguyen, D. T.; Prodhom, B.; Reiher, W. E.; Roux, B.; Schlenkrich, M.; Smith, J. C.; Stote, R.; Straub, J.; Watanabe, M.; Wiórkiewicz-Kuczera, J.; Yin, D.; Karplus, M. All-Atom Empirical Potential for Molecular Modeling and Dynamics Studies of Proteins. J. Phys. Chem. B 1998, 102, (11) Gautieri, A.; Vesentini, S.; Redaelli, A. How to Predict Diffusion of Medium-Sized Molecules in Polymer Matrices. From Atomistic to Coarse Grain Simulations. J. Mol. Modeling 2010, 16, (12) Abbott, L. J.; Hart, K. E.; Colina, C. M. Polymatic: A Generalized Simulated Polymerization Algorithm for Amorphous Polymers. Theor. Chem. Acc. 2013, 132, (13) Plimpton, S. Fast Parallel Algorithms for Short Range Molecular Dynamics. J. Comput. Phys. 1995, 117, S32

33 (14) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; DiNola A.; Haak, J. R. Molecular Dynamics with Coupling to an External Bath. J. Chem. Phys. 1984, 81, (15) Materials Studio, Accelrys Accelrys Software Inc., (16) Ricciardi, R.; Auriemma, F.; De Rosa, C.; Lauprêtre, F. X-ray Diffraction Analysis of Poly(vinyl alcohol) Hydrogels, Obtained by Freezing and Thawing Techniques. Macromolecules 2004, 37, (17) Assender, H. E.; Windle, A. H. Crystallinity in Poly(vinyl alcohol). 1. An X-ray Diffraction Study of Atactic PVOH. Polymer 1998, 39, (18) Marrink, S. J.; Risselada, H. J.; Yefimov, S.; Tieleman, D. P.; de Vries, A. H. The MARTINI Force Field: Coarse Grained Model for Biomolecular Simulations. J. Phys. Chem. B 2007, 111, (19) Hansen, N.; Müller, S. D.; Koumoutsakos, P. Reducing the Time Complexity of the Derandomized Evolution Strategy with Covariance Matrix Adaptation (CMA-ES). Evol. Comput. 2003, 11, (20) Charbonneau, P ; Knapp, B. A User s Guide to PIKAIA 1.0, (21) Computer Simulation of Liquids; M. P. Allen and D. J. Tildesley; Oxford University Press: Oxford, United Kingdom, (22) Todorov, I. T.; Smith, W.; Trachenko, K.; Dove, M. T. DL_POLY_3: New Dimensions in Molecular Dynamics Simulations via Massive Parallelism. J. Mater. Chem. 2006, 16, (23) McKeown, N. B. Polymers of Intrinsic Microporosity. ISRN Mater. Sci. 2012, 2012, (24) Humphrey, W.; Dalke, A.; Schulten, K. VMD Visual Molecular Dynamics. J. Molec. Graphics 1996, 14, (25) Laio, A.; Parrinello, M. Escaping Free-Energy Minima. Proc. Natl. Acad. Sci. U. S. A. 2002, 99, (26) Voter, A. A Method for Accelerating the Molecular Dynamics Simulation of Infrequent Events. J. Chem. Phys. 1997, 106, S33