The two prevailing paradigms in storage are physisorption on structures such as metal-organic frameworks and chemical storage (i.e. storage in the form of chemical bonds). The first tends to require temperatures that are too low to be practical, having hydrogen that is weakly bound to the support material; The second currently involves bonding that is too strong, limiting efficient regeneration of the material. The ideal bonding of hydrogen, in fact, requires a strength that is in between these two extremes. This fact adds to the search for materials with high H gravimetric and volumetric densities an effort to tune the binding of hydrogen to a support material so as to create an efficient H release and recharge device. To assist with the design of hydrogen storage material, simulation and modeling has focused mainly on four, interconnected, areas: Analysis of decomposition reactions in known materials, search for new materials and prediction of their properties, studies of the thermodynamics of storage materials, studies of the kinetics of storage materials.
A problem common to these four areas of research is the development of potentials that can accurately and efficiently describe the materials. This issue, together with the development of powerful global optimization algorithms, is particularly relevant for the prediction of the crystal structure and the computation of the thermodynamic properties of novel materials. Electronic structure calculation methods used so far range from quantum Monte Carlo and molecular orbital based approaches (such as coupled cluster theory) to DFT with gradient corrected functionals. Molecular based methods are very accurate, but can only be applied to study the properties of small system (tenths of atoms), while DFT has the potential to describe solid state systems with strong bonding but it has problems in properly describing weakly bound solids (such as molecular crystals) or the dispersion interactions that govern H physisorption processes. This is particularly relevant, for example, for the quantitative assessment of the storage capacity of carbon based nano-structures. Systems governed by weak, Van der Waals, interactions can be described very accurately via quantum Monte Carlo, but the computational efficiency of this method must be improved before it can be routinely applied. Due to the cost of the more refined electronic structure approaches, DFT methods have been used, together with approximate methods for estimating the thermodynamics of solids, for fast screening of large classes of candidate hydrogen storage materials. The properties of interesting materials thus identified were then refined via more accurate calculations. Alternatives to these methods for computing the interactions employ tight binding approximations or reactive force fields (such as ReaxFF) which can empirically model bond forming and breaking without full electronic structure calculations. The development of these model potentials is still in its early stages, but it may provide a good compromise between accuracy and efficiency and also allow to perform dynamical calculations extending to relatively long times for these systems.
The analysis of the decomposition properties of known materials (such as sodium alanates and amino-boranes), but also of new compounds, pursues two main objectives. Both objectives aim at identifying the microscopic characteristics of the dissociation reactions so as to interpret experiments and to suggest modifications of the materials at the atomic level to improve their operational characteristics. The first objective is to identify and/or predict the decomposition pathways of solids containing hydrogen in their molecular structure. These studies usually employ methods to accelerate the dynamics of the system (the reactions are activated processes) such as Nudged Elastic Band calculations and, more recently, Temperature Accelerated Molecular Dynamics to explore the free energy surface of the system and identify both the metastable states corresponding to the reactant, the product and the intermediates of the reaction and - perhaps even more importantly - the saddle points corresponding to the transition states. Most calculations performed so far have relied on intuition to identify possible dissociation pathways thus creating the risk of biasing the analysis of the reaction. Recently, new rare event methods have appeared that allow, once a set of collective coordinate has been defined, to search in an essentially blind way for the path of maximum likelihood (the Minimum Free Energy Path) on a given free energy profile and/or to analyze an ensemble of reactive paths among the metastable states to obtain information on the mechanism (Transition Path Sampling methods). The further development of these approaches will confine the need for intuition to the choice of the collective variables and thus minimize biasing of the analysis. The second goal of these studies is to elucidate the role of catalysts (subsitutional or interstitial dopants such as Ti in alanates or Li in amino boranes, for example) in facilitating the dissociation of the materials and in improving the kinetics of the reaction. Studies of the kinetic aspects of the dissociation reactions have so far been conducted using mainly Transition State Theory estimates. Because hydrogen is a light element, quantum effects can play a role and some efforts, based on path integral methods or to quantum extensions of Transition State Theory, have been made in particular for the calculation of the rate of the dissociation reaction of H storage materials but theoretical improvements are necessary in this context and, more in general, to include quantum behavior in the accelerated dynamics techniques employed.
Finally, let us point out that the need to employ mesoscopic modeling techniques has recently emerged, in particular, to address microstructural changes, such as microfractures and plastic deformations, associated to the free volume changes related to the structural and/or phase transitions that often accompany the uptake and release of H. While this is a very interesting computational challenge, we have decided not to address it directly in this workshop for reasons of time. However, most mesoscopic techniques rely on input from microscopic simulation methods such as the ones that will be addressed in the meeting so progress fostered by the discussions during this meeting may prove useful also in this field.