Theoretical Background

In this section we provide the theoretical background and the derivation for the equations used in PyIsoP. Please refer to the original publication from Gopalan et al., [GBBS19] for more explanation.

The Local Langmuir constant (\(K_L\))

The free energy of adsorption for a spatially distinct site \(i\) inside a material is given by Snurr et al. [SBT94],

(1)\[exp\Big({\frac{-A_{i}}{k_BT}}\Big)=\frac{1}{8 \pi^2V_{uc}}\int_iexp(-\frac{U_i}{k_BT})d^3rd^3\phi\]

where k\(_B\) is the Boltzmann constant, and \(d^3r\) and \(d^3 \phi\) represent the integral over the space forming the site and the orientations of the adsorbate molecule respectively. \(V_{uc}\) simply serves as a volume scaling factor. For hydrogen represented by a single site probe, adsorbing at a point (cubelet in space with energy \(E_i\)) equation (1) reduces to,

(2)\[exp(-\beta A_i)=\frac{V_{cubelet}}{V_{uc}}exp(-\beta E_i)\]

where \(\beta = 1/k_BT\). Now, if there are \(n(E_i)\) such sites in a given volume V, the combined Henry’s constant resulting from the adsorption in all these sites, each of energy \(E_i\) will be

(3)\[K_{H}(E_i,T)=\frac{M_A}{N_A}\frac{V_{uc}}{V}\frac{1}{k_BT}n(E_i)exp(-\beta A_i) \quad [mass/(pres. vol.)]\]

substitute equation (2) in equation (3) to get

(4)\[K_{H}(E_i,T)=\frac{M_A}{N_A}\frac{1}{k_BT}.\frac{V_{cubelet}.n(E_i)}{V}exp(-\beta E_i)\]

The resultant Langmuir constant from the adsorption in a site with the energy \(E_i\) is related to \(K_H(E_i)\) as

(5)\[K_L(E_i,T)=\frac{K_H(E_i)}{Q_{sat}(E_i)} \quad [1/pres.]\]

where \(Q_{sat}(E_i)\) represents the overall saturation capacity of all sites with the energy \(E_i\)

(6)\[Q_{sat}(E_i)=Q_{sat}*f_V(E_i)=Q_{sat}*\frac{n(E_i)V_{cubelet}}{V}\]

where \(f_V(E_i)\) is the fraction of the total void volume that is formed by all sites of energy \(E_i\). Using equations (4) and (6), equation (5) can be simplified to

(7)\[K_L(E_i,T)=\frac{M_A}{RTQ_{sat}}exp(-\beta E_i)\]

The Fundamental Equation of adsorption

Adsorption of a pure component in a nanoporous material under a given set of conditions (T,P) can be considered as a summation over all of its adsorption sites. Amongst all the possible ways to characterize an adsorption site, one useful descriptor is its interaction energy with the guest molecule. Now the overall adsorption in the material at any given T,P is the sum over all the energetically different adsorption sites,

(8)\[Q(T,P)=\sum_iQ_{sat}.f_V(E_i)\frac{K_L(E_i,T).P}{K_L(E_i,T).P+1}\]

Now if we assume that there is a continuous distribution of such sites in the material, which we can think in terms of a normalized probability density \(\rho(E)\). This normalized density \(\rho(E)\) is related to \(f_V(E_i)\) (from equation (6)), the fraction of the total void volume that is formed by all sites of energy \(E\) as,

(9)\[f_V(E)=\rho(E)dE\]

where \(f_V(E)\) equals the probability of finding a site of energy between \(E\) and \(E+dE\) (once we assume a continuous distribution) inside the box. Hence the overall adsorption in the material can be expressed as a continuous integral involving the saturation loading \(Q_{sat}\), the local Langmuir constants \(K_L\) and the density of sites \(\rho(E)\).

(10)\[\frac{Q(T,P)}{Q_{sat}}=\int_{-\infty}^{\infty}\frac{K(T,E)*P}{1+K(T,E)*P}\rho(E)dE\]

To make accurate predictions at high pressures (order of 10 bar), the local Langmuir constant should be corrected for the guest-guest contribution. Including the guest-guest interactions into \(E\) creates a pressure dependency in \(K_L\). We include that explicitly to get

(11)\[\frac{Q(T,P)}{Q_{sat}}=\int_{-\infty}^{\infty}\frac{K(T,P,E)*P}{1+K(T,P,E)*P}\rho(E)dE\]

where the energy \(E\) now includes both the contributions,

(12)\[E=E^{(x,y,z)}_{guest-host}+E^{(T,P,material)}_{guest-guest}\]

Calculation of the guest-host and guest-guest contributions, is explained in sections guest-host and guest-guest respectively.

The Saturation capacity

For hydrogen, the saturation loading can approximated as [SKL+14],

\[Q_{sat} = \rho_{L,H_2}*V_f = 70.8 (g/L) * V_f\]

where \(\rm \rho_{L,H_2}\) is the liquid density of hydrogen and \(V_f\) is the void fraction of the material.

Guest-Host interactions

The guest-host energy can be easily calculated by dividing the material into cubelets and calculating the energy of hydrogen placed at each position using a classical force field. The histogram of this energy grid provides us with the density distribution of sites \(\rho(E)\) to use with equation (11).

Guest-Guest Interactions

The guest-guest interaction energy of hydrogen inside a nanopore on the other hand depends upon the adsorption conditions (T,P) and the effects of confinement in the material. The difficulty in understanding phase of hydrogen inside the nanopores of a material, under different adsorption conditions, makes theoretical descriptions of the guest-guest interactions almost impossible. One of the most useful descriptors of the local environment of molecules is the average first-shell coordination number (\(n_1\)) [Was80]. The coordination number for hydrogen in a material at a given (T,P) can be calculated from GCMC simulation as the integral of the radial distribution function (\(g(r)\)) of hydrogen inside a nanopore.

(13)\[n^{T,P}_1=\int_{r_0}^{r_1}4\pi r^2 dr\rho_bg(r)\]

where \(r_0\) and \(r_1\) correspond to the range of the first peak in \(g(r)\) and \(\rho_b\) is the overll box number density. If we neglect all the interactions beyond the first coordination shell, the average guest-guest energy can then be approximated in terms of the first-shell coordination number and the Lennard Jones well-depth (\(\epsilon\)) as

(14)\[E^{T,P}_{guest-guest}=\frac{n_1^{T,P}*\epsilon}{2}\]

Hence, in order to make predictions with the model we need a means to predict the coordination number of hydrogen in a material at a given (T,P). For example in our recent study [GBBS19], we trained a machine learning model which uses Gaussian Process Regression (GPR) [Ras04] with {log(P), T, void fraction (\(V_f\)), the largest cavity diameter (LCD) and the pore limiting diameter (PLD)} as the feature space. The model was trained on data from the GCMC simulations performed on a set of 1000 representative MOFs from the CoRE-MOF (1.0_) [CCH+14] The parity between GCMC and GPR and the coordination number landscapes at 77 K are reproduced below. PyISoP offers machine learning with the same settings (GPR, kernel: Rational Quadratic, Length-scale = 0.5) for quick training on your data. To use a different kernel or a machine learning algorithm one could use Scikit-learn [PVG+11] and feed the coordination number array into PyIsoP’s predictor module to make isotherm predictions.

Final Form of \(K_L\)

If we put all the pieces together, the final expression of the local Langmuir constant to be used in equation (11) looks like

\[K_L(E^{bin}_{guest-host}, T, P) = exp\big({-\beta E^{bin}_{guest-host}}\big) \cdot exp({-\beta \frac{n^{T,P}_1\epsilon}{2}})\]

---

CCH+14

Yongchul G Chung, Jeffrey Camp, Maciej Haranczyk, Benjamin J Sikora, Wojciech Bury, Vaiva Krungleviciute, Taner Yildirim, Omar K Farha, David S Sholl, and Randall Q Snurr. Computation-ready, experimental metal–organic frameworks: a tool to enable high-throughput screening of nanoporous crystals. Chemistry of Materials, 26(21):6185–6192, 2014.

GBBS19(1,2)

Arun G Gopalan, Benjamin J Bucior, Scott N Bobbitt, and Randall Q Snurr. Prediction of hydrogen adsorption in nanoporous materials from the energy distribution of adsorption sites. Molecular Physics, ():, 2019. (in press).

PVG+11

Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, and others. Scikit-learn: machine learning in python. Journal of machine learning research, 12(Oct):2825–2830, 2011.

Ras04

Carl Edward Rasmussen. Gaussian processes in machine learning. In Advanced lectures on machine learning, pages 63–71. Springer, 2004.

SKL+14

Cory M Simon, Jihan Kim, Li-Chiang Lin, Richard L Martin, Maciej Haranczyk, and Berend Smit. Optimizing nanoporous materials for gas storage. Physical Chemistry Chemical Physics, 16(12):5499–5513, 2014.

SBT94

Randall Q Snurr, Alexis T Bell, and Doros N Theodorou. A hierarchical atomistic/lattice simulation approach for the prediction of adsorption thermodynamics of benzene in silicalite. The Journal of Physical Chemistry, 98(19):5111–5119, 1994.

Was80

Yoshio Waseda. The structure of non-crystalline materials. Liguids and Amorphous Solids, 1980.