(539b) Improving Computational Assessment of Water Adsorption to Enable Large-Scale Screening of Porous Materials for Water Harvesting

Water harvesting using porous materials has recently drawn considerable scientific attention for its potential in producing usable water. This technology works on the principle of temperature and pressure swing adsorption (T/PSA). In this application, water is adsorbed into nanoporous materials at high relative humidity and/or low temperature while being desorbed at low relative humidity and/or high temperature. One of the most important performance metrics of a nanoporous material for such an application is its deliverable capacity—the difference between how much the material can adsorb (take in) at the adsorption condition and desorb (release) at the desorption condition. A key step to the development of such technologies is the selection of optimal adsorbent materials. In this case, for example, an adsorbent with a high deliverable capacity for the desired adsorption and desorption conditions is preferred. However, owing to the rapid advancement of materials synthesis in the last two decades, there are thousands of available materials. High-throughput computational approaches can therefore play an important role in facilitating the search for optimal adsorbents.

The widely-used approach to study water adsorption in materials is the grand canonical Monte Carlo (GCMC) method. However, a drawback of this method is that it favors short-term reduction of energy over a long-term reduction. This can be particularly problematic for water adsorption studies because water may form clusters during adsorption due to their strong hydrogen-bonding networks. This causes water adsorption simulations to converge very slowly, leading to uncertainties in the outcome of the simulations as well as long computational wall-time. In this study, we demonstrated that this drawback can be mitigated using techniques from a class of methods known as flat histogram methods. These methods force (bias) the system to sample all loading macrostates with equal probability. A variant of this technique—the NVT + W (NVT simulations and Widom sampling) has been used in this work. An important advantage of flat histogram methods is that the macrostate probability distributions can be easily reweighted to compute adsorption uptake at other temperatures and pressures. Thus, a single macrostate probability distribution can yield adsorption isotherms at a range of temperatures. We applied and validated this approach to study CO₂ adsorption in the metal-organic framework (MOF) structure IRMOF-1, as well as demonstrated its potential in studying water adsorption in MOF-806. While the probability distribution yielded by the NVT + W method is quite useful in predicting adsorption behavior, it could be computationally expensive as it requires up to hundreds of calculations per structure or tens of calculations with appropriate interpolation approaches.

Fortuitously, to screen materials for a particular application, the knowledge of the entire probability landscape (entire isotherm) is not necessary. When computing the deliverable capacity, the only properties of interest are the loading values at the adsorption and desorption conditions. Thus, we further proposed an approach for large-scale screening of materials—the so-called C-map method, which can estimate the equilibrium loading values at a specific temperature (T) and pressure (P) much more efficiently. The method exploits the fact that, for a given value of T, P, and loading (N), the ratio of the probability of deletion to that of insertion indicates the position of that loading macrostate (N) relative to the closest local probability maximum. A ratio of unity implies that the loading (N) is a local probability maximum or minimum. A ratio of < 1 indicates that the local probability maximum occurs at a larger loading than N. Conversely, a ratio > 1 indicates that the local probability maximum occurs at a smaller loading than N. Thus, for given adsorption and desorption conditions, the range of local probability maxima can be found with only a few NVT + W calculations. This knowledge leads to the prediction of a range of adsorption loadings, desorption loadings, and deliverable capacities for the simulated structures. The C-map method was subsequently applied to screen a large number of topologically diverse MOF structures for water harvesting. We identified a number of promising MOFs that can provide deliverable capacities larger than that of the state-of-the-art materials, as well as shed light on their atomic-level adsorption mechanism. Structure-property relationships for water adsorption in MOFs were also established, offering insights into the future rational design of materials with improved harvesting performance. We anticipate that this study can facilitate the search for optimal materials to harvest atmospheric water at a given geographical location.