Improving Computational Assessment of Porous Materials for Water Adsorption Applications via Flat Histogram Methods

Abstract

Water adsorption in porous materials has recently drawn considerable attention for its tremendous potential in environmental applications such as water harvesting from air. A key step to the deployment of such technologies is the selection of optimal adsorbent materials as material performance directly impacts their efficiency. Owing to the vast materials space, computational studies may play a critically important role in facilitating the selection of the best material. The current state-ofthe-art method to predict adsorption behavior of porous materials, the grand canonical Monte Carlo (GCMC) simulations, however, can converge very slowly due to its inefficient sampling of the phase space accessible to the system and may yield unreliable results. To this end, we have demonstrated a method from a class of techniques known as flat histogram methods, which can sample the accessible states of the system much more efficiently. Further, a so-called C-map method is proposed herein to efficiently determine the applicability of a material in water adsorption applications. These methods can offer unprecedented insights into the behaviors of the GCMC-predicted isotherms and, more importantly, are promising for large-scale computational screening of porous materials for water adsorption.

Monte Carlo simulations for water adsorption in porous materials: Best practices and new insights

“…36,41,75,76 However, this approach might not be well-suited to study water adsorption because the cluster formation traps the system in local free energy minima, resulting in inadequate sampling of the phase space and an inability to reach the true equilibrium loading values. This was discussed in a previous study, 53 and the slow convergence of the GCMC simulations for water adsorption has also been reported in several prior studies. 36,41 In this study, we further probe the suitability of the GCMC method by applying it to simulate the condensation of the bulk water phase (i.e., reproduce the vapor liquid equilibrium of water).…”

“…Indeed, we found in our previous study that water adsorption simulations in MOF-806 using GCMC also showed a large hysteresis loop which reduced in size as the number of cycles increased, but even after running several million cycles requiring months of computational time, the loop still existed. 53 By contrast, the NVT + W method is able to provide a single vapor-liquid equilibrium curve and shows the step to be at pressure ~4500 Pa which resembles the reported benchmark saturation pressure for the TIP4P model, 72 indicating much improved sampling (Figure 1A). Moreover, it converges within a few days of computational time, indicating more effective sampling.…”

“…While running the simulation longer may possibly shrink the hysteresis loop, it may require an extremely large number of cycles with a computational time of months or even longer. Indeed, we found in our previous study that water adsorption simulations in MOF‐806 using GCMC also showed a large hysteresis loop which reduced in size as the number of cycles increased, but even after running several million cycles requiring months of computational time, the loop still existed 53 . By contrast, the NVT + W method is able to provide a single vapor–liquid equilibrium curve and shows the step to be at pressure ~4500 Pa which resembles the reported benchmark saturation pressure for the TIP4P model, 72 indicating much improved sampling (Figure 1A).…”

“…These probabilities were reweighted to compute the adsorption isotherm at the temperature at which these calculations were performed as well as at other temperatures. The details of this method were discussed in our previous study as well as other works in literature 51,53 . In this work, NVT calculations were performed at a temperature of 298 K while the acceptance probabilities for the Widom moves were evaluated at 20% of the saturation loading for the particular water model at T = 298 K. More details are provided in Supporting Information Section S2.…”

“…Traditional techniques such as grand canonical Monte Carlo (GCMC) simulations are known to be computationally expensive and may require several months of computational wall clock time on a single core to yield results. 36,41 While other methods such as flat histogram methods [51][52][53] can be computationally more efficient, studying tens of thousands of MOFs remains a great challenge. Thus, utilizing simple geometric properties such as largest cavity diameter (LCD), pore volume, void fraction, accessible surface area as well as energetic parameters such as heats of adsorption and Henry's constants to aid the selection of MOFs can lower computational costs.…”