The Role of Density Functional Theory Methods in the Prediction of Nanostructured Gas-Adsorbent Materials

The Role of Density Functional Theory Methods in the Prediction of Nanostructured Gas-Adsorbent Materials

Claudio Cazorla c.cazorla@unsw.edu.au School of Materials Science and Engineering, University of New South Wales, Sydney NSW 2052, Australia
Integrated Materials Design Centre, University of New South Wales, Sydney NSW 2052, Australia
Abstract

With the advent of new synthesis and large-scale production technologies, nanostructured gas-adsorbent materials (GAM) such as carbon nanocomposites and metal-organic frameworks are becoming increasingly more influential in our everyday lives. First-principles methods based on density functional theory (DFT) have been pivotal in establishing the rational design of GAM, a factor which has tremendously boosted their development. However, DFT methods are not perfect and due to the stringent accuracy thresholds demanded in modeling of GAM (e.g., exact binding energies to within  eV) these techniques may provide erroneous conclusions in some challenging situations. Examples of problematic circumstances include gas-adsorption processes in which both electronic long-range exchange and nonlocal correlations are important, and systems where many-body energy and Coulomb screening effects cannot be disregarded. In this critical review, we analyze recent efforts done in the assessment of the performance of DFT methods in the prediction and understanding of GAM. Our inquiry is constrained to the areas of hydrogen storage and carbon capture and sequestration, for which we expose a number of unresolved modeling controversies and define a set of best practice principles. Also, we identify the subtle problems found in the generalization of DFT benchmark conclusions obtained in model cluster systems to real extended materials, and discuss effective approaches to circumvent them. The increasing awareness of the strengths and imperfections of DFT methods in the simulation of gas-adsorption phenomena should lead in the medium term to more precise, and hence even more fruitful, ab initio engineering of GAM.

pacs:
31.15.-p, 71.15.Mb, 81.05.Zx, 81.05.U-, 81.05.Rm

I Introduction

Figure 1: (a) Representation of the gas species considered in this review and (b)-(d) some of the most popular families of nanostructured gas-adsorbent materials: carbon-based nanomaterials, boron-nitride nanostructures and MOF (see text).

i.1 Rational design of gas-adsorbent materials

Nanostructured gas-adsorbent materials (GAM) are the cornerstones of potentially revolutionary advancements in critical and fast growing technological fields like molecular sensing, energy storage and harvesting, and environmental and sustainability engineering. Their exceptional high surface to volume ratio, regular atomic composition, tunable reactivity, transport properties, and assembling affinity to form supramolecular systems, have permitted the realization of timely and cost-effective applications like, for instance, the detection and removal of toxic substances from water and air, dense storage of hydrogen and natural gas in solid state matrices, sequestration of carbon dioxide from flue gases generated in electricity production plants, design of high-performance photovoltaic cells, and enhanced long-lasting operation of batteries, to cite just a few examples dai12 (); bonaccorso10 (); yavari12 (); zhao12 (); sahoo12 ().

Popular families of nanostructured GAM include zeolites, metal oxides nanocrystals (e.g., CaO and AlO), carbon-based nanomaterials (CN, e.g., nanotubes, sheets, met-cars, graphite intercalation compounds, and frameworks of organic pillared graphene), metal hydrides nanoparticles (e.g., MgH and LiBH), and covalent- and metal-organic frameworks (COF and MOF). Of all these species CN and MOF (see Fig. 1) stand out as some of the most promising GAM for energy and environmmetal applications, particularly to what concerns the capture and storage of hydrogen (H), carbon dioxide (CO), and methane gases mendoza12 (); fracaroli14 (); furukawa09 (); broom13 (); urbonaite08 (); tranchemontagne12 (); yurum09 (); dietzel09 (); kumar14 (); simmons11 ().

Currently reported GAM gas-selectivity and storage capacities, however, still remain below the stringent commercial targets set by specialized government bodies and agencies. For instance, hydrogen storage systems need to achieve an overall capacity of  wt% hydrogen with a volumetric ratio of  g/L for competitive vehicle applications us11 (), and capture of the  % of the carbon dioxide produced in the generation of electricity must be reached within less than a  % of increase in the final costs rackley10 (); hester10 (). The search for optimal gas-adsorption processes and GAM, therefore, remains an area of very active scientific and technological research.

A key aspect for potential GAM to be successful is to find the optimal chemical compositions and pore topologies to work under specific thermodynamic conditions. The number of possible stoichiometric and structural GAM configurations is tremendously large, hence in practice systematic experimental searches based on trial-error strategies turn out to be cumbersome and very inefficient. Rational engineering of gas-adsorbent interactions at the atomic scale, represents a key notion to achieve success on such a design grand challenge in the short and middle term. In this context, computational simulation methods emerge as invaluable theoretical tools for the screening and rational engineering of auspicious GAM.

i.2 Computational simulation techniques for modeling of GAM

Common simulation techniques in the study of GAM can be classified into two major categories: “semi-empirical” and “first-principles”. In semi-empirical approaches, the interactions between atoms are modeled with analytical functions, known as force fields or classical potentials, which are devised to reproduce a certain amount of experimental data or the results of high-accuracy calculations. The inherent simplicity of classical potentials makes it possible to address the study of GAM and gas-adsorption processes considering realistic thermodynamic conditions and length/time scales, with well-established simulation techniques like, for instance, molecular dynamics and grand canonical Monte Carlo frenkel (). With the current computational power and algorithm development, key features in GAM which are directly comparable to observations (e.g., adsorption isotherms, elastic properties and diffusion coefficients skoulidas04 (); xiang10 (); battisti11 (); garberoglio12 (); garberoglio15 ()) can be computed routinely on a standard office computer. Also, the relevance of quantum nuclear effects in gas-adsorption phenomena can be estimated accurately with semi-empirical approaches gordillo11b (); cazorla13b (); carbonell13 (); gordillo11 (). Nevertheless, in spite of the great versatility of semi-empirical methods, classical potentials may present impeding transferability issues in certain situations. This type of drawbacks is related to the impossibility of mimicking the features of the targeted material at conditions different from those in which the setup of the corresponding force field was performed csanyi04 (); dzubak12 ().

In this context, the outputs of first-principles calculations turn out to be crucial. As the name indicates, empirical information is not contained on first-principles methods, also known as ab initio. The interactions between atoms are directly obtained from applying the principles of quantum mechanics to the electrons and nuclei. Transferability issues, therefore, are totally missing in ab initio approaches. Examples of first-principles techniques include, density functional theory and quantum Monte Carlo, to cite just a few. Although these approaches are very accurate, they can be also very demanding in terms of computational expense. This circumstance poses serious difficulties to the study of kinetic and thermodynamic effects in extended GAM with ab initio techniques. Common acceleration schemes within first-principles methods entail the use of pseudopotentials vanderbilt90 (); troullier91 (). Many materials properties can be predicted basing exclusively on the behaviour of the valence electrons, and by employing pseudopotential techniques explicit treatment of the core electrons is avoided in the simulations. Pseudopotentials can actually be the source of potential errors, however they make also the simulation of heavy atoms and medium-size systems feasible. Fortunately, some strategies can be used to minimize the impact of the approximations introduced by pseudopentials like, for instance, the projector augmented wave blochl94 () and linearized augmented plane wave methods andersen75 (). In the present critical review, we will concentrate on analyzing specific aspects of the simulation of GAM with first-principles methods, obviating the outcomes of indispensable and physically insightful semi-empirical approaches.

i.3 Capabilities of density functional theory methods

Standard density functional theory (DFT) techniques (i.e., based on local and semilocal approximations to the electronic exchange-correlation energy, typically LDA and GGA) have become the ab initio methods of choice in the study and design of nanostructured GAM. These techniques have been demonstrated to reproduce with notable accuracy the interplay of a wide range of interactions in hundreds-of-atoms systems, while keeping within reasonably affordable limits the accompanying computational expense. However, standard DFT methods present some well-known limitations in describing gas-adsorption phenomena occurring in low-coordinated atomic environments (e.g., the binding of small molecules to surfaces and cavities). For instance, due to the local nature of the employed energy functionals standard DFT methods cannot reproduce the electronic long-range correlations resulting from instantaneously induced dipole-dipole, dipole-quadrupole, quadrupole-quadrupole and so on interactions. Another common DFT fault is the presence of electronic self-interaction errors, which derive from an imperfect cancellation between the auto-correlation (i.e., spurious interaction of an electron with itself) and exchange energies. This type of errors can alter dramatically the description of charge-transfer complexes, chemical reactions, and electron affinities perdew81 (); lee88 (); patchkovskii02 (); cohen11 (). Fortunately, many developments have been realized during the last two decades which have solved part of these drawbacks (e.g., nonlocal and hybrid exchange-correlation energy functionals, see next section), permitting so to achieve remarkable agreement between theory and experiments. Nevertheless, a number of critical aspects remain yet challenging to customary DFT methods that could be hindering the rational design of GAM.

i.4 DFT challenges in GAM design

One of such challenges consists in accounting for the long-range electronic correlations while simultaneously amending the electronic self-interaction errors (see Fig. 2a). Along with the adsorption of gas molecules on surfaces, regions of significant electron depletion and accumulation may appear which induce strong electrostatic interactions and the shift of electronic energy levels. In those conditions, one could expect that by adding a portion of the exact Hartree-Fock exchange energy to the selected DFT functional, in order to lower the electron self-interaction errors to negligible levels, accurate binding energies will follow. Reported evidence, however, suggests that accounting for weak dispersion interactions in charge-transfer processes may turn out to be also decisive johnston08 (); steinmann12b (). Unfortunately the causes behind such an unexpected result are not yet totally understood, due in part to the difficulties encountered in the decomposition of DFT energies into fundamental portions. Rationalization of this complexity clearly is needed for optimizing the computational load associated to DFT modeling of GAM (i.e., accounting for dispersion interactions normally requires intensive calculations), and also for ensuring the reliability of prospective and already published simulation works on the storage and sequestration of gases.

Another important DFT threat consists in accounting adequately for the screening of bare interactions in periodically extended systems (see Fig. 2b). Many-body energy and Coulomb screening effects can be equally important in physisorption and chemisorption phenomena, and the issues encountered in their description are a consequence of the pairwise additivity assumed in the construction of most DFT functionals. Essentially, the interaction energy between two atoms remains unaltered no matter what medium separates them or what collective excitations happen in the material misquitta10 (); tkatchenko12 (); gobre13 (). In this context, the adiabatic connection fluctuation-dissipation (ACFD) theorem has been exploited to calculate correlation DFT energies that incorporate many-body higher-order terms. This is the case of the random phase approximation to DFT (RPA-DFT) dobson06 (); angyan11 (); hesselmann12 () and the DFT+MBD ruiz12 (); tkatchenko12b (); ambrosetti14 (); distasio14 () methods (see Sec. III), which at present are receiving the highest attention. Nonetheless, the development of many-body DFT-based methods are still on their infancy and the associated computational expenses are elevated (typically ranging from two to four orders of magnitude larger than standard DFT). The corresponding degree of applicability therefore remains still limited.

Figure 2: Schematic representation of the two main challenges of customary DFT methods in GAM design: (a) reproduction of covalent and dispersion interactions on an equal footing and (b) description of Coulomb screening and many-body energy interactions in extended systems.

i.5 Inherent complexity of DFT benchmark studies

Most of ab initio works published to date on the design of GAM are based on standard DFT calculations (see Sec. V for more details). Standard DFT methods are very effective in dealing with large atomic systems and a myriad of user-friendly DFT packages, practically available to everyone nowadays, have facilitated their widespread use. However, chemical accuracy is generally demanded on GAM design (e.g., correct binding energies to within  eV) and thus, for the reasons highlighted in the previous section, employing standard DFT methods may not always be adequate.

In fact, to carry out DFT benchmark tests on the assessment of GAM is of paramount importance for rigorously establishing acceptable balances between computational feasibility and predictive reliability. Performing computational studies of such a type, however, is not a trivial task. First, one has to be familiar with the technical and foundational aspects of highly accurate quantum chemistry methods (QCM) like Møller-Plesset perturbation theory (MP2), the coupled-cluster method with single, double and perturbative triple excitations [CCSD(T)], and quantum Monte Carlo (QMC), to cite just a few. In those approaches the Schrödinger equation of the many-electron system of interest is solved directly by handling the corresponding wavefunction. Many-electron wavefunctions generally are expressed in real space as a linear combination of atomic orbitals (LCAO). LCAO do not fulfill the conditions of Bloch functions and thus the simulation of periodic systems like crystals and surfaces, although possible, it is not straightforward with them manby10 (); pisani80 (); dovesi83 (); causa88 (). Consequently, most QCM studies are performed in supercells containing model cluster systems where periodic boundary conditions are not applied. Meanwhile, QCM are inherently complex and the associated computational load in general scales poorly with the number of electrons (namely as ).

Besides the technical intricacies, a conceptual problem also arises when performing benchmark studies in model cluster systems: on what grounds is correct to generalize the so-obtained conclusions to realistic extended materials? To help in answering this question, we show in Fig. 3 the partial density of electronic states (pDOS) calculated in graphene, i.e., an extended one-atom thick carbon surface, decorated with Ca impurities, and a coronene molecule doped with a Ca adatom (i.e., Ca@CH) using standard DFT methods (in both systems the ratio of Ca impurities to carbon atoms is the same). We note that the coronene molecule is generally considered as a large enough system in which to carry out computational accuracy tests. However, as it can be appreciated in the figure the pDOS computed in coronene differs greatly from the one obtained in the “equivalent” graphene-based system (e.g., compare the distribution of electronic states around the Fermi energy level in the two cases). Also, charge-transfer calculations based on Bader’s theory show that the Ca adatom donates about   to the coronene molecule (i.e., weak ionic interaction), whereas in Ca-decorated graphene it supplies   to the carbon surface (i.e., strong ionic interaction). In fact, quantum confinement effects can introduce important differences in the electronic structure of apparently similar extended and cluster systems, thereby leading to completely unlike gas-adsorption behaviours in the two situations ma11 (); smith13 (). In addition to this, long-range electronic exchange and nonlocal correlations, which are ubiquitous in gas-adsorption phenomena, normally turn out to be disguised in model cluster systems.

An obvious solution to overcome these likely scaling problems, affecting most benchmark studies, consists in using highly accurate quantum chemistry approaches which, alike to DFT, can handle the simulation of periodic systems. All the necessary benchmark and DFT calculations therefore can be undertaken in a same extended simulation cell. In this regard, quantum Monte Carlo (QMC) emerges as one of the most promising family of methods for benchmarking DFT. Actually, the formalism of many-electron Bloch functions has been established within QMC since long time ago rajagopal95 (); foulkes01 () and several implementations of this approach are already available in open-source packages like, for instance, CASINO needs10 (), QWALK wagner09 (), and QMCPACK qmcpack (). Also, the balance between accuracy and scalability in QMC is among the highest of all quantum chemistry methods (e.g., the accompanying computational cost climbs as with the number of electrons esler08 ()). It is worth mentioning that very recent methodological advances have permitted also the implementation of the MP2 and the CCSD approaches in a planewave framework marsman09 (); gruneis10 (); gruneis11 (). The availability of this last suite of quantum chemistry techniques however is rather limited at the moment, although it is likely to improve over the next few years booth13 ().

Figure 3: Partial density of , , and electronic states calculated in a Ca-CH molecule (Top), and in Ca-doped graphene (Bottom). The ratio of calcium to carbon atoms is the same in both systems, namely . Calculations have been performed with standard DFT methods, and the Fermi energy level in both systems has been shifted to zero. (This figure has been adapted from work. cazorla09 ().)

i.6 Focus and organization of this review

The focus of the present critical review is on the assessment of the accuracy of DFT methods in the prediction and understanding of nanostructured GAM. Despite that the number of DFT studies published so far on GAM applications is enormous, there is not yet a well-established set of best practice principles for meaningful simulation of gas-adsorption phenomena and materials. The main objective of this review is to assist in defining this, by critically discussing key contributions and unsolved controversies in the field. Also, we identify here the subtle problems found in the interpretation of complex benchmark studies and propose likely solutions for them.

Our analysis is constrained to the design of carbon-based nanomaterials (CN) and MOF for applications in H storage and CO sequestration and capture. The main reason justifying this choice is that these types of GAM and problems currently are attracting the highest attention within the community of computational and environmental materials scientists mendoza12 (); fracaroli14 (); furukawa09 (); broom13 (); urbonaite08 (); tranchemontagne12 (); yurum09 (); dietzel09 (); assfour11 (); lan10 (); fang14 (). Moreover, by understanding the processes involved in the adsorption of H and CO molecules on CN and MOF we may infer also those affecting many other substances and materials. For instance, under normal conditions H and CO molecules are both linear and consequently cannot sustain any permanent dipole moment (e.g., like N; the predominant electrostatic interactions with the solid sorbent then are of hydrogen bond and quadrupolar type, respectively. However, when CO molecules receive or donate electrons they normally get distorted and become polar. Thereby, from an electrostatic point of view CO can turn out to be similar to other important molecules like, for instance, water and amonia. On the other hand, CN and MOF are structurally and electronically alike to boron-nitride nanostructures and covalent organic frameworks (see Fig. 1), two families of GAM which at present are being investigated also very thoroughly lan10b (); feng12 (); reddy10 (); weng13 (); lin12 (). For the sake of focus, other important mechanisms exploited in gas storage and sequestration processes (e.g., gas binding in bulk chemical form and thermodynamic destabilization of nanocrystals and clusters seayad04 (); shevlin09 (); cazorla-amoros91 (); dean11 (); wang11 ()) and related families of materials (e.g., metal oxides and binary and ternary hydride compounds ramachandran07 (); shevlin12 (); shevlin13 (); sun10b (); gebhardt14 (); berube07 ()), will not be discussed exhaustively in the present work.

The organization of this review is as follows. In the next section we briefly summarize the binding energy targets pursued in the design of H storage and CO capture and sequestration GAM, together with a short description of CN and MOF. Next, we explain the generalities of DFT methods and provide essential insight into its most popular versions. In Sec. IV, we review the most recent and relevant works done in the assessment of the performance of DFT methods in GAM modeling. From them, we draw general conclusions on which DFT exchange-correlation functionals can be employed safely for the simulation of gas-adsorption processes and materials. Also, we comment on the subtle problems found in the generalization of benchmark conclusions obtained in model cluster systems to real GAM, and discuss possible solutions for them. In Sec. V, we present a discussion on the current tendencies followed in first-principles design of GAM and propose a number of new directions to explore in prospective modeling and benchmark studies. Finally, our main conclusions are outlined in Sec. VI.

Ii Nanostructured GAM

Nanostructured gas-adsorbent materials contain characteristic nanoscale motifs which are periodically repeated along one, two or three directions. Those motifs generally are composed of light atoms which congregate to form atomically sparse complexes. There are excellent review articles in the literature focusing on the physical properties and prospective implementations of GAM (see for instance Refs. hu06 (); li11 (); orimo03 (); kuppler09 (); tasis06 (); pakdel11 (); nag10 (); park12 ()), hence we highlight here only the main traits of some representative species (i.e., CN and MOF). Also, we explain the basic requirements that potential GAM must accomplish for achieving effective storage and sequestration of H and CO gases.

ii.1 GAM desiderata

The “hydrogen storage” and “carbon dioxide capture and sequestration” problems have sparked very intense research within the communities of chemists, physicists and engineers in the last past decade. The discovery of new materials with large gas uptake capacities, robust thermodynamic stability, fast adsorption-desorption kinetics, and affordable production costs, is a key notion to succeed in the encountered gas storage and sequestration challenges murray09 (); yang10 (); alessandro10 (); samanta12 (); choi09 ().

The binding affinities of potential GAM are determined by their interactions with the gas molecules, which can be of electrostatic, dispersion, and/or orbital types lochan06 (). In turn, the strength of the gas-GAM interactions depend on the characteristics of the materials and gas species, which in the latter case include electronic structure, atomic shape, polarizability and permanent dipole and/or quadrupole moments. Key quantities in the assessment of gas-adsorbent materials include the alignment between frontier molecular orbitals (i.e., the energy difference between the highest occupied molecular orbital -HOMO- in the adsorbate and the lowest unoccupied molecular orbital -LUMO- in the adsorbent), and the resulting charge transfers and binding energies. Let us have a closer look into the most relevant physical aspects of promising hydrogen and carbon capture GAM.

ii.1.1 H storage

For hydrogen to become the fuel of choice in future environmentally clean vehicles and electricity production plants, large amounts of H need to be stored within relatively small volumes entailing moderate weights (for a detailed list of related targets see Refs. doe1 (); doe2 ()). In this context, H adsorption on light solid sorbents through weak and mild molecule-GAM interactions emerges as one of the most promising routes. (Dissociative adsorption in solid metal hydrides where molecular hydrogen is dissociated to form a solid solution with the hydride, is also considered as propitious seayad04 (); shevlin09 (); however, we will not discuss such an interesting gas-storage mechanism or related family of materials in the present review for the sake of focus.) Current estimations of the optimal binding energy for adsorption of hydrogen at ambient temperature and considering safe delivery pressure conditions of  bar, amount to  eV/molecule garberoglio05 (); bathia06 (); basile14 (). Meanwhile, kinetic effects may be critical for practical applications (e.g., complete H storage-release cycles must be realized within seconds) and when these are taken into account favourable binding energies turn out to be  eV/molecule li03 (). Based on these assessments, the binding energy range that typically is targeted in most ab initio H-storage works is  eV/H, which spans from dispersion to moderately covalent molecule-GAM interactions (see Table I).

The strength of the H-solid sorbent interactions can be tuned by decorating the GAM surfaces with transition or alkaline-earth metal atoms, and this normally tends to increase the affinity towards gas binding bhattacharya10 (); hamaed10 (); phillips08 (). The orbital mechanism acting behind this effect is the Kubas interaction, which consists in electron donation from the H -bonding orbital to the empty metal orbitals and simultaneous electron back-donation from the filled metal orbitals to the H -antibonding orbital kubas01 (); kubas07 (). A different approach that is promising for gas storage at elevated temperatures, is the spillover of hydrogen onto inert surfaces. This mechanism implies the chemical activation of H molecules placed on top of transition metal sites through catalysis, followed by the migration of atomic hydrogen onto the surface of the receptor material (e.g., activated carbons, COF, and MOF). Record hydrogen storage capacities of  wt. % have been recently accomplished with this strategy in metal-carbon complexes and MOF tsao10 (); psofogiannakis09 (); psofogiannakis11 (); li07 (); wang12b ().

Figure 4: (a) Sketch of a carbon dioxide molecule interacting with a carbon-based surface. (b) Partial density of electronic states calculated in the system sketched in (a), expressed in arbitrary units. The electronic states localized in the carbon surface are represented in the upper part of the figure, and those corresponding to the gas molecule in the lower. (c) Sketch of a carbon dioxide molecule interacting with a Ca-decorated carbon surface. (d) Partial density of electronic states calculated in the system sketched in (c), expressed in arbitrary units. The electronic states localized in the Ca-decorated carbon surface are represented in the upper part of the figure and those corresponding to the gas molecule in the lower. HOMO bonding and LUMO anti-bonding molecular orbitals are indicated in panels (b) and (d) [see text].

ii.1.2 Carbon capture and sequestration

Carbon capture and sequestration (CCS) actions are being implemented in fossil-fuel burning power plants for cutting the amounts of green-house gases which are expelled to the atmosphere. Current CCS means mostly rely on the scrubbing of synthesis and flue gases with amine solvents. However, solvent-based CCS technologies generally are not cost-effective due to the high energy penalty involved in solvent regeneration and equipment corrosion aaron05 (). Fortunately, membranes and solid sorbents do not suffer from these important drawbacks and thus they constitute the core of next-generation CCS technologies.

There are two main types of CCS processes in which GAM are highly promising: pre-combustion and post-combustion. In pre-combustion CCS, the fuel is converted into a mixture of hydrogen and carbon dioxide gases using processes such as “gasification” or “reforming”. The usual thermodynamic conditions in pre-combustion CCS are high temperatures and high pressures (i.e., above 100 and  bar, respectively), and the concentration of CO gas is high. At those conditions, it can be shown that CO sequestration turns out to be effective when the strength of the gas-GAM interactions roughly amounts to  eV/molecule. For this, consider the inverse calcination reaction, , to be a model of a typical pre-combustion CCS process cazorla-amoros91 (); dean11 (); wang11 (). The Gibbs free-energy balance of this reaction has been approximated by the function  eV/molecule, where represents the temperature (expressed in Kelvin) gilchrist89 (). Now, by assuming that the thermal contributions to mostly arise from the entropy of the gas,  eV/molecule follows.

In post-combustion CCS, the CO gas is separated from the exhaust of a combustion process. The usual thermodynamic conditions in post-combustion CCS are low temperatures and low partial CO pressures. In that case, it can be shown that the ideal gas-binding energy scale for solid sorbents is  eV/molecule. For this, consider that a typical CCS post-combustion process can be described by the generic capture reaction , and that the corresponding Gibbs free-energy balance vanishes at temperatures close to ambient (e.g., degrees Celsius). Now, by assuming that the entropy change in the solid sorbent upon adsorption of molecules is practically null and using tabulated thermodynamic data of the CO gas (i.e., the corresponding entropy is  eV/molecule K),  eV/molecule follows.

Based on these assessments, typical gas-adsorption energies pursued in most ab initio CCS studies span from to  eV/CO, and depending on the specific aims one end or the other of this interval is targeted. As it is shown in Table I, that energy range extends from weak dispersion to strong covalent molecule-GAM interactions. In addition to these binding energy requirements, prospective CCS GAM need to display also favorable selectivity features with respect to the adsorption of N, oxygen and water, since those species are also abundant in the generated flue and synthesis gases.

It is worth mentioning that similar electronic orbital processes to the Kubas interaction kubas01 (); kubas07 () explained in the previous section, can be exploited also for enhancing the affinities of GAM towards CO binding. In Fig. 4, we represent the partial density of electronic states calculated for a carbon dioxide molecule interacting with a pristine carbon surface, and with the same surface decorated with Ca atoms. The presence of calcium dopants induces overlappings between electronic CO orbitals and metallic states in the region surrounding the Fermi level, resulting in a transfer of charge from the molecule to the metal centers. In this process the degeneracy of the HOMO bonding and LUMO anti-bonding molecular orbitals is lifted, and electronic charge is back-donated from the filled Ca orbitals to the anti-bonding CO electronic states (compare bottom panels in both Figs. 3 and 4). This orbital mechanism provokes the bending of the gas molecule and intensifies the gas-GAM interactions cazorla11 (), thus providing a route for the rational design of sorbent materials for pre-combustion CCS applications.

ii.2 Carbon nanomaterials

Carbon nanostructures (CN) include an ample range of carbon allotropes such as nanotubes, graphene, met-cars and graphite intercalation compounds (see Fig. 1). CN exhibit a large variety of electronic and transport properties resulting from the specific way in which atoms are arranged. Carbon nanotubes, for instance, may be metallic or semiconducting depending on their diameter and the degree of helicity present in their hexagonal-ring walls wilder98 (); castro09 (). Met-cars (i.e., metallocarbohedrynes) are extremely stable symmetric clusters formed by metal and carbon atoms with stoichiometric formula C, where typically stands for a transition metal atom ( Ti, V, Zr, Hf, Fe, Cr, and Mo) and (see Fig. 1rohmer93 (); rohmer95 (). Graphite intercalation compounds (not shown in Fig. 1) are composed of alternating planes of transition, alkali or alkaline-earth metal and C atoms disposed in triangular and hexagonal lattices, respectively. In these materials relative shifts between successive atomic planes along the out-of-plane direction may occur leading to the formation of stacking patterns dresselhaus02 (); emery05 (); cobian08 ().

Interestingly, the electronic and gas-adsorption properties of CN can be finely tuned by coating their surfaces with dopant species, creating defects, or imposing mechanical strains lee10 (); cazorla10 (); krasheninnikov07 (); dresselhaus10 (); zhou10 (); dutta14 (). Among these, the doping strategy with metal adatoms has attracted a lot of attention because of its technical simplicity and expected enhancement of the binding affinity towards gas molecules through the Kubas interaction kubas01 (); kubas07 () (see Sec. II.1.1shevlin08 (); yildirim05 (); ataca09 (); cazorla12 (). However, transition metal atoms in carbon surfaces exhibit a strong tendency towards clustering sun05 (); li06 (); cazorla09 () and thereby further developments in the field of synthetic chemistry addressing this problem are awaited. Alternatively, the creation of nanostructured networks of defect sites such as vacancies, nanoribbons and islands, may increase considerably the number of unsaturated carbon sites. This kind of structural modifications generally enhance the interactions with the gas molecules and therefore lead to improved gas storage capacities wang13 (); banhart10 ().

ii.3 Metal-organic frameworks

Metal-organic frameworks (MOF) are multi-dimensional nanoporous structures composed of metal ions (or clusters) coordinated to rigid organic molecules, which are called linkers. The choice of the metal ion and linker species completely determines the structure and functionality of the resulting MOF. Common organic linkers include bidentate carboxylics (e.g., HOOC-COOH), tridendate carboxylates (e.g., CHO), 1,4-benzenedicarboxyalte (BDC), and azoles (e.g., CHN) molecules. The great freedom with which the linkers and metal ions can be chosen and combined is reflected in the more than MOF species that have been reported in the last two decades (for an extensive review on this topic, see Ref. furukawa13 ()). Among these, the -MOF74 ( Mg and Ni), MOF-5 (e.g., ZnO tetrahedra linked by BDC organic molecules) and -BTT ( Ca, Fe, Mn,Cu, Co, Ni, Cr, and Cd, and BTT = 1, 2, 5-benzenetristetrazolate) families emerge as three of the most investigated compounds. The -MOF74 structure is based on coordinated carboxyl and hydroxy groups (i.e., helical -O-C rods that emanate from 6-coordinated centers) and its primitive unit cell contains atoms. The MOF-5 crystal has a cubic symmetry and its conventional and primitive cell contain eight and four OZnO(BDC) formula units, respectively (i.e., and atoms). The primary building block of X-BTT is a truncated octahedron (i.e., a six [XCl] squares and eight BTT ligands structure) which shares its square faces to form a general cubic framework, comprising a total of atoms.

The polarity of the building units and spatial separation between the organic linkers, affect profoundly the binding strength of H and CO molecules and thereby constitutes a rationale for the design of MOF-based GAM. Also, decorating the MOF surfaces with alkali, alkaline-earth and transition metal atoms can enhance significantly the adsorption of gas species. Practical realizations of this last type of functionalization, however, remain yet technically difficult due to potential atomic coalescence kaye08 (); schroder08 (); dixit12 (). Alternatively, modifications of the porous frameworks based on the embedment of metal nanoparticles have been recently investigated, producing very impressive gas-capacity records cheon09 (); lim12 (); jeon11 ().

Iii Density Functional Theory Methods

Table 1: Characteristic traits of usual bonded and nonbonded atomic interactions taking place in condensed matter systems and surfaces.

iii.1 General considerations

The wave function of a -electron system, , contains all its physical information and it is determined by solving the corresponding Schrödinger equation. In real materials electrons interact through the Couloumb repulsion and are abundant, thus is a complex mathematical function that in most of the cases is unknown. In 1965, Kohn-Sham developed an ingenious theory to effectively calculate the energy and related properties of many-electron systems without the need of explicitly knowing  kohn65 (); sham66 (). The main idea underlying this theory, called density functional theory (DFT), is that the exact ground-state energy, , and electron density, , of a many-electron system can be determined by solving an effective one-electron Schrödinger equation of the form:

(1)

where index labels different one-electron orbitals and the corresponding spin state. In particular,

(2)

and

(3)

where represents an external field and is the potential function that results from deriving the exchange-correlation energy, , with respect to the electron density.

The exchange-correlation energy has a purely quantum mechanical origin and can be defined as the interaction energy difference between a many-electron quantum system and its classical counterpart. Electrons are indistinguishable particles called fermions and the wave function describing an ensemble of electrons must change its sign when two particles exchange orbital states. This quantum antisymmetry leads to an effective repulsion between electrons, called the Pauli repulsion, which helps in lowering their total Coulomb energy. Despite represents a relatively small fraction of the total energy, this is an extremely crucial quantity for all the physical aspects of materials and molecules because it acts directly on the bonding of atoms. The functional generally is unknown and in practice needs to be approximated. Actually, this is the only source of fundamental error in DFT methods and depending on the approximation that is used the resulting approach may turn out to be valid or not for describing the systems and phenomena of interest.

In standard cases is approximated with the expression

(4)

where is made to depend on , , and/or the electronic kinetic energy (see next sections). Actually, the exact form of the exchange-correlation energy is readily known and reads

(5)

where is the exchange-correlation hole density at position that surrounds an electron at position r. Some important constraints on have already been established. For instance, must be nonpositive everywhere and its space integral is equal to . Also, the space integral of the correlation hole density is zero. These constraints can be employed in the construction of approximate, but physically correct, functionals.

In the next subsections, we review the most popular approximations which are currently employed in computational studies of GAM. In Table I, we summarize the main types of bonded and nonbonded atomic interactions taking place in condensed matter systems and surfaces. In Table II, we outline the adequacy of the considered DFT methods in describing those interactions. In Table III, we list some popular computer simulation packages that can be used to perform standard and more advanced DFT calculations. Finally, the relative degree of accuracy and computational expense associated to those DFT approaches are sketched in Fig. 5. Additional details on these topics can be found in some recent specialized reviews perdew13 (); klimes12 (); scheffer12 (); dobson12 (); steinmann12 ().

Figure 5: Representation of common levels of approximation within density functional theory together with some general features. Regions coloured in red indicate “High” and in blue “Low”. The typical size of the systems that can be handled with those approaches are indicated on the right margin of the figure.

iii.2 Local and semi-Local energy functionals

In local approaches (e.g., local density approximation -LDA-), is approximated with Eq. (4) and the exchange-correlation energy is taken to the be that of an uniform electron gas with density , namely . The exact expression of the functional is known numerically from accurate quantum Monte Carlo calculations perdew81 (); ceperley80 (). In order to deal with the nonuniformity of realistic many-electron systems, these are normally partitioned into infinitesimal volume elements which are considered to be locally uniform. In semilocal approaches (e.g., generalized gradient approximation -GGA-), is approximated also with Eq. (4) and is made to depend on and its gradient  pw91 (); pbe ().

Both local and semilocal approximations satisfy certain exact constraints (e.g., some exact scaling relations and the exchange-correlation hole sum rules) and can work notably well for systems in which the electronic density varies slowly over space (e.g., bulk crystals at equilibrium conditions, see Table II). However, this is manifestly not the case of GAM which normally contain surfaces and pores where can change very abruptly. Moreover, by construction local and semilocal functionals totally neglect long-range electron correlations, otherwise known as dispersion interactions, which certainly are ubiquitous in gas-adsorption phenomena.

iii.3 Meta-GGA or highly parametrized energy functionals

Meta-GGA functionals are semilocal in nature (i.e., assume the approximate expression in Eq. (4) but contain an additional degree of elaboration with respect to LDA and GGA: the orbital kinetic energy density, . This new ingredient allows one to construct functionals which satisfy some additional exact constraints, like for instance the correct gradient expansion of the exchange energy up to fourth order. Another interesting feature of this family of functionals is that they can be trained to capture the short- and middle-range parts of electronic correlation, which in some special cases may be enough to describe correctly the binding of atoms (see Table II).

Meta-GGA functionals are versatile and in general do not entail significantly larger computational expense than standard LDA and GGA. Examples of this family include the suite of TPSS functionals due to Perdew, Scuseria, and others revtpss (); vdwrevtpss (), and the “Minnesota functionals” (M, with 05, 06, 08, 11, and 12) developed by Truhlar and his group in the University of Minnesota truhlar (). Meta-GGA functionals in general can provide accurate lattice constants in solids, surface energies, and molecular binding energies, and are already implemented in popular quantum chemistry and DFT packages like, for instance, GAUSSIAN gaussian (), NWCHEM nwchem (), SIESTA siesta (), and VASP vasp () (see Table III). Nevertheless, dispersion interactions cannot be reproduced systematically with these functionals since they lack of explicit nonlocal correlation kernels (see Sec. III.5 below).

iii.4 Hybrid exchange energy functionals

Hybrid functionals comprise a combination of nonlocal exact Hartree-Fock (HF) and local exchange energies, together with semilocal correlation energies. The proportion in which both nonlocal and local exchange densities are mixed generally relies on empirical rules. The popular B3LYP b3lyp (), for instance, takes a  % of the exact HF exchange energy and the rest from the approximate GGA and LDA functionals. Other well-known hybrid functionals are the HSE proposed by Heyd-Scuseria-Ernzerhof hse (), PBE0 pbe0 (), and the family of Minnesota meta hybrid GGA truhlar ().

In contrast to local and semilocal functionals, hybrids describe to some extent the delocalization of the exchange-correlation hole around an electron. This situation becomes of particular relevance in chemisorption and charge transfer processes, where the atomic bonds turn out to be elongated or shortened. Also, in strongly correlated systems containing and electronic orbitals life, for instance, transition metal oxides. In other words, hybrid functionals are useful in situations where the electron self-interaction errors, stemming from an imperfect cancellation between the artificial interaction of an electron with itself and the exchange energy, are potentially large. Hybrid functionals, however, do not account for the long range part of the correlation hole energy and thus cannot reproduce dispersion forces (see Table II). Efforts to effectively correct for such drawbacks have been made recently by Head-Gordon and other authors (e.g., the range separated and dispersion corrected B97X-D and M06-D3 functionals) gordon08 (); lin13 (); schwabe07 (); mardirossian14 ().

iii.5 Dispersion-corrected energy functionals

The condition that any DFT-based dispersion scheme must accomplish is to reasonably reproduce the asymptotic behaviour of the interaction between two particles separated by a distance in a gas. Local, semilocal, and hybrid energy functionals totally miss this requirement. The most straightforward way to correct for such a fault consists in adding an energy term to the exchange-correlation energy that is attractive and has the form (where indexes and label different particles, and a damping factor is introduced at short distances in order to avoid divergences). This approximation represents the core of a suite of methods termed DFT-D which, due to their simplicity and low computational cost, are being widely used at the moment. Probably, the most popular family of DFT-D methods are the dispersion corrected GGA functionals proposed by Grimme grimme (). Despite their appealing features, DFT-D methods present several shortcomings. First, many-body dispersion effects and faster decaying terms such as the and interactions are completely disregarded. Second, it is not totally clear from where one should obtain the optimal coefficients. And finally, once these parameters are determined they remain fixed during the simulations, and this can be problematic in situations where the orbital hybridization and oxidation states change as compared to the free atoms case.

Several improvements on DFT-D methods have been proposed, in which the value of the dispersion coefficients are made to depend on the specific atomic environment. Those correspond to the DFT-D3 method by Grimme grimme-d3 (), the vdW(TS) approach by Tkatchenko and Scheffler ts (), and the BJ model by Becke and Johnson bj (). In those approaches the specific variation of the parameters are taken on the atomic coordination or effective volume, and the calculation of reference dispersion coefficients and atomic polarizabilities are required.

A third degree of elaboration exists in which no external input parameters are needed and the dispersion interaction is directly computed from the electron density. In this context, the exchange-correlation energy is expressed as where is the nonlocal correlation energy. Particularly, is calculated with a double space integral involving the electron density and a two-position integration kernel, , of the form

(6)

In Eq. (6), and are the values of an universal function evaluated in positions and , and is a complex function that obeys two main constraints: is strictly zero for any system with constant , and the interaction between two molecules has the correct dependence for long distances. The described approach was introduced by Dion et al. in 2004 dion04 () and it represents a key DFT development since it combines all types of interaction ranges within a same formula. Refinements on Dion’s approach have subsequently appeared where (i) the original two-position integration kernel is modified (e.g., the so-called nonlocal VV10 functional due to Vydrov and Voorhis vydrov10 (); vydrov12 (); sabatini13 ()), and (ii) the exchange term in is substituted with other more accurate functionals (e.g., the so-called vdW-DF2 lee10b (), vdW-optB88 and vdW-optPBE optex (), and vdW-C09 cooper10 () schemes). All these approaches are termed vdW-DF and, thanks to the seminal work of Román-Pérez and Soler roman09 (), their accompanying computational expense in a planewave framework is moderate in practice. Nevertheless, the way in which nonlocal correlations are calculated inherently assumes pairwise additivity and thus many-body effects are completely disregarded in vdW-DF methods (see Table II).

perdew81 (); ceperley80 (); pw91 (); pbe ()
revtpss (); vdwrevtpss (); truhlar ()
b3lyp (); hse (); pbe0 ()
grimme (); dion04 (); optex (); lee10b (); vydrov10 ()
tkatchenko12b (); ambrosetti14 (); scheffer12 (); dobson12 ()
Table 2: Description of the performance of some DFT variants in describing usual types of bonded and nonbonded interactions found in condensed matter systems and surfaces. Symbol () indicates correct (incomplete) description of the considered type of interaction by the corresponding DFT method.

iii.6 The random phase approximation and DFT+MBD

The adiabatic connection fluctuation-dissipation (ACFD) theorem provides a general and exact expression for the exchange-correlation energy of a many-electron system, thereby in principle can be calculated in a very accurate way incorporating higher-order many-body effects. In particular, the correlation energy of a system adopts the form

(7)

where is the bare density-density response function, the interacting density-density response function at Coulomb coupling strength , and denotes the six-dimensional integration over the variables and . The response function measures the electronic response of the system at a point due to a frequency-dependent electric field perturbation at a point . In the ACFD approach, the adiabatic connection between a reference non-interacting system (defined at ) and the fully interacting system (defined at ) provides the correlation energy of the latter, including many-body dispersion as well as other types of electron correlation effects. This theoretical framework has been exploited by several authors to develop novel many-body DFT-based approaches, among which we highlight the random phase approximation to DFT and the DFT+MBD method.

In the random phase approximation (RPA-DFT) scheme, the interacting response function is defined self-consistently via the equation . This approximation has been shown to work reasonably well for a number of cluster and extended systems ren11 (). Following the Adler-Wiser formalism adler62 (); wiser63 (), can be computed by using the occupied and virtual orbitals, and the corresponding energies and occupancies obtained in DFT calculations. In analogy to post-HF approaches, the computational expense associated to RPA-DFT is very large (i.e., typically scales with the fourth power of the number of particles) and the convergence with respect to the basis set generally is too slow lu09 (); eshuis10 (); eshuis11 (); eshuis12 (); umari10 (). Also, it must be noted that the short-range part of the electron correlation energy is not precisely reproduced by RPA-DFT and that this shortcoming may be problematic in studying molecular systems furche01 ().

Meanwhile, recent efforts done in the groups of Tkatckenko and Scheffler have given birth to the so-called DFT+MBD method ruiz12 (); tkatchenko12b (); ambrosetti14 (); distasio14 (), which accounts also correctly for the Coulomb screening and many-body effects in electronic systems. In the DFT+MBD approach, the Schrödinger equation of a set of fluctuating and interacting quantum harmonic oscillators is solved directly within the dipole approximation, and the resulting many-body energy is coupled to an approximate semilocal DFT functional. These approximations result in a significant reduction of computational load as compared to the RPA-DFT method, allowing one to describe larger systems containing up to few hundreds of atoms.

The most appealing features of the RPA-DFT and DFT+MBD methods is that they are very accurate and suitable for studying small-gap and metallic periodic systems. In fact, these techniques have already been applied successfully to the study of strongly correlated crystals and surfaces harl10 (); xiao12 (); schimka10 (), organic molecules adsorbed on metallic surfaces ruiz12 (), and ionic and semiconductor solids zhang11 (). Effective schemes of the RPA-DFT and DFT+MBD methods have already been implemented in commercial and open-source first-principles packages like, for instance, ABINIT abinit (), GPAW gpaw (), and VASP vasp () (see Table III).

 abinit ()
 adf ()
 castep ()
 cp2k ()
 crystal ()
 gamess ()
 gaussian ()
 gpaw ()
 molpro ()
 nwchem ()
 expresso ()
 siesta ()
 vasp ()
 wien2k ()
Table 3: Selected computer packages which can be used to perform DFT calculations. “Periodic” refers to the ability of handling the simulation of three-dimensional periodic systems. In “Basis set”, “PW” refers to plane waves, “STO” to Slater-type orbitals, “GTO” to Gaussian orbitals, “NAO” to numerical atomic orbitals, and “FP-(L)APW+lo ” to augmented plane waves and local orbitals. Symbol () indicates suitability (unsuitableness) of the package to perform a particular calculation type.

Iv Assessing the performance of DFT methods in the design of GAM

Making judicious comparisons between zero-temperature calculations and ambient-temperature observations turns out to be very complicate due to the presence of thermal excitations. Temperature profoundly affects the (free) energy and conformation of molecules adsorbed on GAM, and the differences with respect to conditions are difficult to assess with theory even when considering the simplest interaction models. In fact, free energies and entropies cannot be accessed straightforwardly during molecular dynamic simulations, in contrast to other thermodynamic quantities like, for instance, the total internal energy and pressure. In order to evaluate free energies technically and computationally involved methods like thermodynamic integration, free-energy perturbation and umbrella sampling, have to be employed kastner09 (); taioli07 (); cazorla07 (); cazorla12b (). Reasonably then, the most direct and exact way of evaluating the performance of approximate ground-state methods is to compare them with other computational approaches which are known to be more accurate. In the case of DFT methods, accuracy benchmark tests involve the application of quantum chemistry methods (QCM) like Møller-Plesset perturbation theory (MP2), the coupled-cluster method with single, double and perturbative triple excitations [CCSD(T)], and quantum Monte Carlo (e.g., diffusion Monte Carlo -DMC-).

In what follows, we review recent DFT benchmark studies done in the areas of hydrogen storage and carbon capture and sequestration problems involving CN and MOF. Based on them, we draw general conclusions on the suitability of approximate exchange-correlation functionals for the design of GAM. Regrettably, despite that the number of these benchmark studies is very modest, as it will be shown next, some interpretation and computational inconsistencies prevail yet among them. Also, we discuss the subtle problems found in the generalization of DFT performance tests done in model cluster systems to extended GAM. A few modeling strategies are proposed to overcome this class of problems, some of which are computationally very intensive and some others tentative.

iv.1 H storage

iv.1.1 Carbon-based GAM

A promising alternative to functionalizing carbon-based GAM with transition metals atoms is to use alkali metal species like lithium and calcium. Aggregation issues on alkali metal coatings are expected to be less severe than in the transition metal case (see Sec. II.2), and the corresponding H gravimetric densities predicted in DFT studies largely surpass the targets set by the U.S. Department of Energy (i.e., 5.5 wt. %) ataca09 (); zhou12 (); ao10 (). Reported experimental records, however, lie significantly below the impressive performances anticipated with theory (e.g., a modest  wt. %) chen13 (); hong12 (); huang13 (); zhou14 () and the reasons for those large discrepancies are not yet totally understood.

In 2009, Cha et al. published a controversial work in which the accuracy of standard DFT methods in the assessment of H-storage GAM was seriously questioned cha09 (). By using quantum chemistry [i.e., MP2 and CCSD(T)] and standard and hybrid DFT methods, Cha showed that the binding energies, , and interatomic distances calculated in a model system composed of four equidistant hydrogen molecules to a positively charged Ca ion [i.e., Ca(H), see Fig. 6a], were dramatically different. In particular, with standard and hybrid DFT functionals favorable molecular binding to the calcium cation (i.e., values around and  eV) was found at a H-Ca distance of  Å , whereas with MP2 and CCSD(T) methods no effective binding was determined at any separation (see Table IV). Cha et al. performed analogous gas-adsorption calculations in a larger system comprising a coronene molecule (i.e., CH) and a calcium atom, and concluded with the same level of inconsistency that found in the Ca(H) case. In words of Cha et al. cha09 (): “(these findings) indicate that previous suggestions for the Ca-based hydrogen storage system should be reinvestigated with particular care about the charge state of Ca”. The levels of alarm associated with this error class, however, were lowered shortly afterwards by Ohk et al.. In a formal comment on Cha’s work, Ohk argued that the discrepancies between MP2 and DFT methods reported in the Ca(H) and Ca-CH systems were numerical artifacts stemming from the use of small localized orbitals basis sets, i.e., , which did not contain polarization functions of high enough momenta ohk10 (). By using larger basis sets of the Dunning type and performing extrapolation to the complete basis set (CBS) limit to get rid of likely finite basis-set errors, Ohk et al. showed that the agreement between semilocal DFT and MP2 results obtained in Ca-decorated cluster models was qualitatively acceptable (although standard DFT methods exhibited a strong tendency towards molecular overbinding, see Table IV). The reliability of GGA functionals in the assessment of hydrogen storage materials apparently had it been restored (at least, at the qualitative level). However, a deep understanding about how standard DFT functionals, which definitely incur in self-interaction errors and neglect long-range dispersive interactions, could describe the adsorption of light molecules on surfaces and cavities correctly, was still missing.

Figure 6: Sketch of the (a) Ca(H) model cluster system, (b) graphite intercalation compounds dresselhaus02 (); emery05 (); cobian08 () seen from the front and top views, (c) organic connector and metal cluster model systems, and (d) MOF-5 [also known as IRMOF-1]. Cluster systems in (a) and (c) have been widely considered in benchmark studies as representative models of GAM depicted in (b) and (d), respectively.

In two subsequent works, the authors of article cha09 () reported further details on the anomalous performance of customary DFT methods in describing the fixation of H molecules on Ca centers via the Kubas interaction cha10 (); cha11 (). They showed that when a sharp orbital transition occurs during adsorption of the gas molecule on the metal center, the incompleteness of the electronic exchange, present in most DFT functionals, acts critically by providing erroneous overstabilization of the final complex. According to Cha et al., Ca(H) exemplifies the type of system where the highest occupied molecular orbital (HOMO) changes abruptly from to character upon the intake of gas, and the fixation of H molecules via the Kubas interaction becomes frustrated. Cha’s explanations are based on arguments put forward by Gunnarsson more than two decades ago, who showed that in situations where the nodal wavefunction surfaces are intricate approximate DFT exchange functionals tend to be imprecise by underestimating the energy cost associated to orbital transitions gunnarsson85 (). Cha’s arguments indeed brought new physical insight into the Ca(H) contention, however some technical aspects of works cha10 (); cha11 () could still be criticized (e.g., extrapolation to the CBS limit in the MP2 calculations was not pursued). Thus, further benchmark studies were still required for carefully judging the accuracy of customary DFT methods in the design of carbon-based GAM.

 (Å)  (Å)
cha09 ()
ohk10 ()
bajdich10 ()
purwanto11 ()
Table 4: Summary of the binding energy and structural results obtained in the Ca(H) model cluster system by different authors employing a variety of DFT and quantum chemistry methods (see text).

In this regard, Bajdich et al. made an important contribution by performing for the first time quantum Monte Carlo (QMC) calculations in the Ca(H) model system, and comparing their results to those obtained with the MP2, and local, semilocal and hybrid DFT methods bajdich10 () (see Table IV). It was found that QMC calculations based on the diffusion Monte Carlo (DMC) predicted no binding at all of the four H molecules to the Ca center within the interval  Å , in agreement with previous MP2 results. In stark contrast to this conclusion, popular DFT functionals like LDA, GGA and B3LYP vaticinated effective fixation of the gas molecules at a distance of  Å . Bajdich’s results are in qualitative agreement (disagreement) with Cha’s (Ohk’s) conclusions explained above. Interestingly, the authors of work bajdich10 () carried out DFT calculations with a blended functional consisting of the full HF exchange energy and the PBE correlation functional (denoted as HFx-PBEc in Table IV). They showed that the results obtained with the HFx-PBEc functional were consistent with those obtained with the MP2 and DMC methods (see Table IV). In light of those outcomes, Bajdich et al. argued that the failure of common DFT functionals in describing the Ca(H) system had its origins on the partial or total omission of long-range exchange interactions, in coincidence with Cha’s reasonings found in works cha10 (); cha11 (). In spite of the significance of Bajdich’s work, this is neither free of some technical objections. For instance, in the atomic relaxations the H-H intermolecular distances were kept fixed to  Å, and the errors stemming from the incompleteness of the employed basis sets (i.e., triple-zeta) and the fixed-node surface approximation in DMC, were not evaluated.

In a posterior work, Purwanto et al. presented a highly accurate study on the binding of the Ca(H) complex which relied on auxiliary-field QMC calculations purwanto11 (). There, extrapolation to the complete basis set limit and a better treatment of the sign problem were accomplished. Purwanto et al. found that the potential energy curve of the four hydrogen molecules exhibits a double-well structure with almost equal binding minima of  eV at distances and  Å. These results are in good agreement with the MP2/CBS calculations performed by the same authors and the CCSD(T)/CBS results obtained by Ohk et al. ohk10 () (see Table IV). Regarding the differences with respect to Bajdich’s work, it was argued that these were likely to be originated by the fixed-node approximation employed in the DMC calculations.

Conclusions.- The main conclusion emerging from works cha09 (); ohk10 (); cha10 (); cha11 (); bajdich10 (); purwanto11 () is that standard DFT functionals tend to overestimate the cohesion of the Ca(H) complex, and that the likely reason for this flaw is a deficient treatment of the long-range exchange energy. On the other hand, highly accurate QCM results obtained by different groups on the same system are not fully consistent and therefore a conclusive verdict on the general performance of DFT methods, i.e., considering all its possible flavours, cannot be emitted. The current status of computational work on this topic is clearly unsatisfactory.

Yet, the situation becomes even more puzzling when the resemblance between the model system Ca(H) and real carbon-based H-storage GAM is brought into examination (e.g., see Fig. 6a-b). Namely, there is still the unresolved question about on which grounds the results obtained in model cluster systems can be translated (if possible at all) to realistic extended materials ma11 (); smith13 (). For instance, the likely presence of valence and electronic states, coming from the carbon atoms in the sorbent, is totally disregarded in the Ca(H) complex. Also, the role of the long-range dispersion interactions, which are ubiquitous in gas-adsorption processes, normally turns out to be underestimated in nano-sized systems. In this last regard, several analysis on the hydrogen storage properties of extended Ca-decorated carbon nanomaterials based on dispersion-corrected DFT functionals, have been reported recently.

Wang et al. have studied the H-storage properties of Ca-decorated graphene with the vdW-DF approach due to Dion dion04 () and compared their results to those obtained with standard local and semilocal methods wang12 (). Wang’s calculations show that the H binding energies obtained with the vdW-DF method lie systematically below those obtained with GGA-PBE (i.e., stronger molecule-GAM interactions by  eV/molecule), and generally above the LDA results (i.e., weaker molecule-GAM interactions by  eV/molecule). (Similar conclusions have been attained by other authors in alike Ca-decorated carbon nanomaterials hussain12 (); li11b ().) In light of these outcomes, it is argued that long-range dispersion interactions can be also important in the adsorption of hydrogen molecules on chemically modified carbon nanomaterials, even in the cases where the obtained binding energies are large (i.e.,  eV/molecule). Actually, Wang’s results appear to expose a new failure of standard DFT methods in the assessment of GAM, this time related to the description of the electronic correlation energy. Unfortunately, the authors of work wang12 () did not report any comparison with respect to hybrid functionals or QCM calculations, hence the size of the bias incurred by the vdW-DF method itself, i.e., due to the approximations performed on the exchange functional and pairwise additivity, cannot be inferred.

In work wong14 (), Wong et al. have investigated also the adsorption of H molecules on metal-decorated graphene but considering a large set of transition and alkali metal species. The employed method there is an improved version of Dion’s approach, the so-called vdW-DF2, in which the accuracy of the employed semilocal exchange functional is bettered lee10b (). In the case of Ca-based coatings, Wong’s results are in qualitative agreement with those obtained by Wang et al. wang12 () however the difference between the two reported vdW-DF and vdW-DF2 H-binding energies is not negligible (i.e., about  meV). When considering other types of dopants, vdW-DF2 calculations generally predict weaker molecular binding than obtained with semilocal GGA methods by  eV per molecule. In fact, Wong’s work comes to reinforce the idea that dispersion interactions can affect profoundly the interplay between extended materials and hydrogen molecules. Consequently, nonlocal correlations must be taken into account by any DFT functional that is intended for modeling of GAM.

Conclusions.- From all these benchmark studies, we can draw the general conclusion that both long-range exchange and dispersion electronic interactions are pivotal in describing the binding of H molecules to CN-based GAM. As for local and semilocal DFT functionals, these two elements are totally missing in them, therefore they are likely to provide unreliable results on the hydrogen storage topic. Situations in which local and semilocal GAM predictions seem to be correct normally are fortuitous and indicate the presence of large energy error cancellations hussain12 (); li11b (). Namely, the sign of the missing contributions to the exchange-correlation energy in local and semilocal approaches are opposite and therefore can compensate each to the other. These error cancellations, however, do not occur systematically and may depend on the specific details of GAM (e.g., see work wong14 () where the sign of the difference between the GGA and vdW-DF2 energies varies with the dopant species), and thus the use of standard DFT methods for modeling of carbon-based H-storage materials is not recommended. Meanwhile, new quantum Monte Carlo (QMC) and RPA-DFT simulations, this time performed in periodic systems, are highly desirable for rigorously evaluating the performance of meta, hybrid, and dispersion DFT functionals (see works schimka10 (); scheffer12 (); wu09 (); binnie10 (); wagner14 () for examples of applications of such advanced computational methods to simulation of relevant materials). QMC and RPA-DFT calculations are also necessary for determining the relevance of many-body energy and Coulomb screening effects on the present class of GAM, which so far have been systematically neglected.

iv.1.2 Mof

The interplay between H molecules and MOF (e.g., MOF-74,5) are dominated by dispersion interactions and the local environment surrounding the binding sites in the metal clusters kuc08a (); kuc08b (); sillar09 (); joo13 (). Consequently, employing DFT methods that completely neglect nonlocal correlations (i.e., standard and hybrid flavours) and/or carrying out benchmark tests in model cluster systems which are too small (see Fig. 6c-d) turns out to be inadequate in the present case. On the other side, the fact that the H-MOF interactions are weak implies that effective gas-storage can only be achieved at temperatures well below ambient conditions (i.e.,  K). Thus, a number of strategies have been proposed for increasing the affinity of MOF towards hydrogen binding rowsell05 (). These include, the design and control of porosity han09 (), functionalization of the organic linkers han08 (), hydrogen spillover li07 (); wang12b (), introduction of open metal sites in the organic linkers and metal clusters kaye08 (); lochan08 (); zhou08 (); sun07 (); kumar12 (), and decoration of MOF surfaces with alkali metal atoms han07 (); klontzas08 (); mulfort08 (). DFT methods have been intensively employed in the last two mentioned approaches hence we revise next the benchmark tests undertaken in those areas.

sun09 ()
sun10 ()
sun10 ()
sun10 ()
sun10 ()
dixit11 ()
jiang12 ()
Table 5: Results of different benchmark tests carried out in organic linker MOF models decorated with alkaline earth, alkali and transition metal atomic species. The types of localized-orbitals basis sets employed in the calculations and the cases in which convergence to the CBS limit is achieved, are indicated within parentheses.

It was first argued by Lochan et al. that H molecules are attracted by open transition metal and alkali centers in MOF through donor-acceptor interactions and electrostatics lochan08 (). Based on the outcomes of standard DFT calculations, those authors concluded that the strength of the H-MOF interactions was within the range of desirable binding for ambient gas storage applications (i.e.,  eV). Lochan et al. also claimed that semilocal DFT functionals could be safely employed in the study of H-MOF systems because the corresponding leading electronic interactions are strong and predominant over dispersion lochan08 ().

In two recent studies, Sun et al. have evaluated the accuracy of standard, hybrid and meta DFT functionals in the prediction of H adsorption on metal-doped organic linker systems sun09 (); sun10 (). They have found that whenever transition metal, alkaline-earth or alkali metal atoms are used, all DFT, MP2 and CCSD(T) methods provide quantitatively similar binding energy results, in satisfactory accordance with Lochan’s findings (see Table V). In particular, calculations performed with popular DFT functionals like PBE, PW91 and M05-2X are in excellent agreement with gold-standard benchmarks obtained with the CCSD(T) method (i.e., equal binding energies to within  eV). Meanwhile, hybrid functionals tend to underestimate slightly by  eV. Analogous conclusions have been attained also by Dixit et al. in a posterior work done in Li-decorated MOF (see Table V) dixit11 ().

A further benchmark test confirming the accuracy of DFT methods in describing the interplay between H molecules and chemically functionalized organic linkers, has been recently reported by Jiang et al. jiang12 (). In particular, Jiang et al. have studied the binding of a hydrogen molecule to a small CHLi cluster using common GGA and hybrid DFT functionals, and the highly accurate DMC method. As it can be appreciated in Table V, notable agreement between all the considered approaches in Jiang’s calculations is obtained (i.e., within  eV).

Conclusions.- In light of the benchmark results reported in works sun09 (); sun10 (); dixit11 (); jiang12 (), we may conclude that customary DFT methods appear to perform appropriately in the simulation of the hydrogen-storage properties of chemically functionalized MOF. The physical reason underlying this favorable outcome is that donor-acceptor interactions and electrostatics in this family of GAM are dominated by short- and medium-range electron-electron exchange and correlations. Nevertheless, a note of caution must be added here.

In our compilation of benchmark tests we have realized a lack of studies analyzing the performance of dispersion-corrected DFT schemes in the simulation of extended hydrogen-loaded MOF containing open metal sites. Indeed, dispersion interactions appear to be secondary in the present case however, as we will show in Sec. IV.2.2, when H molecules are replaced by CO this type of interactions turns out to be crucial. The polarizability of the H molecule certainly is smaller than that of CO ( and  Å, respectively olney97 ()), however this still must have some effect. Actually, it would be very interesting to quantify in which proportion dispersion interactions tend to lower the DFT energies reported in Table V (which, we remind, have been obtained in model cluster systems). Also, it would be highly desirable to perform systematic studies on the performance of local, semilocal, and nonlocal DFT approaches in periodic simulation of chemically functionalized MOF loaded with hydrogen. In this regard, we would like to mention a recent work by Sumida et al. in which it has been shown that neither standard nor hybrid DFT functionals can reproduce with accuracy the measurements done on the binding of H molecules to metal-BTT sumida13 (). Rather, a range-separated hybrid and dispersion corrected DFT functional, i.e., the so-called B97X-D (see Sec. III.4), is found to be necessary for a correct interpretation of the experimental findings (see for instance Table II in work sumida13 ()). Also, Kong et al. have recently found that nonlocal interactions are crucial to achieve DFT consistency with respect to the H heats of adsorption measured in Zn(BDC)(CHNkong09 (). Further DFT work on the role of the dispersion interactions in hydrogen storage of chemically functionalized MOF, is urgently needed in order to avoid likely modeling inconsistencies.

iv.2 Co capture and sequestration

iv.2.1 Carbon-based GAM

The adsorption of CO molecules on carbon-based GAM can be of physisorption or chemisorption type depending on whether the material surfaces are smooth, contain defects or are chemically functionalized. In the case of nondefective surfaces, the interactions with the gas molecules are dominated by dispersion forces thereby the gas is retained on the adsorbent material very weakly mishra11a ().

In 2006, Xu et al. were the first in carrying out nonstandard DFT calculations on the binding of CO molecules to pristine graphene xu06 (). By using the hybrid ONIOM[B3LYP:DFTB-D] method, they found that the corresponding gas-adsorption energy was  eV. In Xu’s approach, an hybrid DFT evaluation of the interactions between CO and a coronene molecule is first performed, and a tight-binding dispersion correction is subsequently added in order to account for the presence of -conjugated interactions in the real material. More recently, Umadevi et al. have analyzed the same type of problem but employing meta-GGA DFT methods (i.e., M05-2X), and surprisingly they have found a large physisorption energy of  eV umadevi11 (). The reasons for the three-fold discrepancy between Xu’s and Umadevi’s results remain unclear to us since QCM benchmark calculations on the strength and nature of the CO-graphene or CO-coronene interactions are practically absent to date. We only know of a recent work lee13 () by Lee et al. in which the physisorption energy of CO is calculated with the MP2 method. In this case, turns out to amount ot and  eV, depending on the relative orientation between the gas molecule and carbon plane. Nevertheless, the basis set of localized orbitals employed in Lee’s study is relatively small (i.e., 6-31G**) and the model cluster system in which the MP2 calculations are performed contains carbon dangling bonds (in opposition to real graphene).

On the other side, the adsorption of gas molecules on carbon nanotubes (CNT) has been thoroughly analyzed by Quiñonero et al. with dispersion-corrected DFT methods (i.e., B97-D/SVP) quinonero12 (). By considering different types of CNT, diameters, and binding sites, Quiñonero et al. have concluded that CO-adsorption is energetically more favourable in the interior than in the exterior of nanotubes, in marked disagreement with previous reports bienfait04 (); zhao02 (). Also, they have found that the strongest CO-CNT interactions occur in the and systems where the computed binding energy amounts to  eV. Interestingly, based on a symmetry-adapted perturbation theory (SAPT) decomposition of the calculated DFT-D interaction energies, the authors of work quinonero12 () show that (i) dispersion interactions account for the  % of the total attractive forces, and (ii) electrostatic interactions resulting from the overlap between CO and CNT electronic orbitals, although secondary, are yet important. We realize, however, a lack of high-level ab initio investigations, e.g., employing the MP2 and CCSD(T) methods, in the effects of the CNT size and curvature on the binding of CO molecules. This type of studies would be highly desirable for a consistent evaluation of the performance of DFT methods on this topic. Similar investigations to these proposed have already been performed for methane smith13b (), a molecule that is akin to carbon dioxide in terms of the electric dipole and quadrupole moments. It is worth noticing that during the preparation of this review we became aware of the submission of an article by Smith and Patkowski which could fill the mentioned knowledge gap in CO-CNT systems smith14 ().

When the surfaces of the carbon nanomaterials contain reactive defect sites like vacancies, holes (i.e., clusters of vacancies) and edges, the interactions with CO molecules become more intense. For instance, Cabrera-Sanfelix has studied the adsorption of carbon dioxide on a defected graphene sheet with standard DFT methods (i.e., DFT-GGA) and has found that molecular physisorption and chemisorption have an energy cost of and  eV, respectively cabrera09 (). Similar results have been obtained by Liu et al. in an equivalent system using an analogous computational approach (i.e., and  eV, respectively) liu11 (). However, the authors of this last study propose an equilibrium chemisorption configuration that is more symmetric than the one obtained by Cabrera-Sanfelix. In a more recent and technically exhaustive work, Wood et al. have addressed the same kind of gas-adsorption problem by employing both dispersion corrected DFT and MP2 methods wood12 (). Essentially, they find that DFT methods incorporating van der Waals forces provide adsorption energies which are in good agreement with MP2 results, and that CO physisorption in the edges of a zigzag graphene nanoribbon occurs with an energy balance of  eV.

Conclusions.- In light of the results reported by Cabrera-Sanfelix, Liu, and Wood, we can conclude that using standard DFT methods for representing the interactions of CO molecules with defected graphene seems to be appropriate. It is not clear to us, however, whether the suitability of DFT-GGA methods in this case corresponds to a large cancellation between errors or simply to a minor role played by the electronic long-range exchange and correlations.

Meanwhile, some functionalization techniques have been proposed for increasing the affinity of carbon-based GAM towards CO binding, which is desirable for precombustion applications. Among those we highlight the decoration of carbon surfaces with nitrogen and light metal atoms, for which a number of experimental and first-principles computational works have been performed du09 (); chen14 (); mishra11 (); cazorla11 (); yang10b (); yong01 (); gao11 (); jiao14 (). Mo et al. have recently presented a computational benchmark study of the adsorption of carbon dioxide on nitrogen-containing hydrocarbon molecules mo13 (). Specifically, they have carried out extensive CCSD(T)/CBS and dispersion-corrected DFT calculations in a CO/2-methylpyridine (the last with formula CHN) system. Stabilization of this complex occurs through an electron donor (2-methylpyridine)-electron acceptor (CO) mechanism and the attractive forces between oxygen and hydrogen atoms. The authors of work mo13 () show that in order to reproduce the gold-standard obtained with the CCSD(T)/CBS method (i.e.,  eV), both electronic long-range exchange and correlations must be taken into account simultaneously. According to Mo’s calculations, BLYP-D3 is among the PBE-D3, BP86-D3, and TPSS-D3 methods (see Sec. III.5) the one peforming the best (i.e., providing a binding energy of  eV).

Conclusions.- The CO/2-methylpyridine complex is therefore a representative example of a system in which, despite of its reduced length, both van der Waals interactions and electronic long-range exchange are simultaneously important. Also, we note that “low-cost” dispersion-corrected DFT schemes like the BLYP-D3 one grimme-d3 () appear to work remarkably well in this system. Indeed, it would be extremely interesting to check whether the adsorption of CO (and H) molecules in alike cluster systems could be described correctly also with this last type of computationally cost-effective approaches.

The accuracy of standard and hybrid DFT functionals in describing the interactions of CO molecules with Ca-decorated graphene has been recently assessed by Cazorla et al. cazorla13 (). In Cazorla’s work, a comparative study between DFT and MP2 calculations is presented by following an original recipe: instead of adopting the customary strategy of steadily increasing the size of polyciclic aromatic hydrocarbon (PAH) molecules, the concentration of Ca dopants in anthracene (i.e., CH, a relatively small PAH) is tuned so as to mimic the partial density of electronic valence states in Ca-decorated graphene (see Fig. 7a). The reason for doing that is to constrain the size of the cluster system where to carry out the MP2 calculations as much as possible, while still reproducing the main electronic orbital mechanisms occurring in the targeted extended material. In this way, the appearance of artificial electronic transitions upon gas-loading are prevented in the model cluster system (e.g., see work cha11 ()). Cazorla’s results show that all considered DFT flavors predict equilibrium structures which are very similar to that obtained with the MP2 method, and energetically favourable CO binding (see Figs. 7b-c). Nevertheless, the differences between hybrid DFT and MP2 energy estimations amount to  eV, and between standard DFT and MP2 to  eV, with DFT methods providing always the strongest binding. The origins of the observed hybrid DFT and MP2 numerical discrepancies (see Fig. 7c) are rationalized in terms of residual self-interaction errors cazorla13 (). The amount of charge transferred between the gas molecules and Ca-CH turns out to be the same when computed with either hybrid DFT or MP2 methods, and long-range dispersion interactions appear to be secondary in the present system (see Table  in work cazorla13 ()). Cazorla et al. therefore conclude that the strength of the resulting electrostatic interactions, which are dominant, must be equal in the two compared cases. On the other side, LDA and PBE standard functionals overestimate the transfer of charge to the CO molecule by cazorla13 (). In light of those outcomes, the use of hybrid DFT functionals is recommended over that of local and semilocal approaches for investigation of the gas-uptake properties of AEM-decorated carbon GAM.

Conclusions.- Cazorla’s prescription for choosing a reduced cluster system in which to undertake quantum chemistry calculations may be useful for optimizing the computational expense associated to DFT benchmark tests. Also, for justifying the subsequent generalization of the attained conclusions to extended materials cazorla13 (). Nevertheless, only few selected electronic features of the targeted system (e.g., partial density of electronic states around the Fermi energy level) can be reproduced at a time by playing with doping strategies, and probably only at a qualitative level. In this context, recent methodological progress achieved in the resolution of the so-called “inverse band structure-problem of finding an atomic configuration with given electronic properties” problem franceschetti99 (); dudly06 () (e.g., genetic algorithm searches), could be useful. In particular, those techniques could be applied to the design of model cluster systems that were replicas of the targeted extended GAM in terms of the electronic structure. In that case, besides the chemical stoichiometry, the atomic structure and shape of the simulation cell could be varied until finding a suitable system in which to perform all the calculations. (Of course, electronic long-range exchange and correlation corrections should be added somehow afterwards; this could be achieved, for instance, by carrying out additional calculations in both periodic and cluster systems with an adequate first-principles method.) We speculate that with such an original effective approach it could be possible to address important benchmark controversies affecting the performance of DFT methods in GAM modeling (e.g., those explained in the present and previous sections), while avoiding the computational burden and bias introduced by the finite size of model cluster systems.

poloni12 ()
dietzel09 ()
rana12 ()
valenzano11 ()
rana12 ()
rana12 ()
valenzano11 ()
rana12 ()
rana12 ()
rana12 ()
rana12 ()
rana12 ()
Table 6: CO-MOF binding energies, , and heats of adsorption ( K), , calculated with different DFT exchange-correlation functionals. vdW-optB86b, vdW-optB88, and vdW-optPBE represent variants of the original vdW-DF and vdW-DF2 nonlocal functionals dion04 (); lee10b () due to Klimeš and co-workers optex (). The vdW-C09 variant is due to Cooper cooper10 (), and the vdW-PBE one is based on the PBE exchange functional pbe () (see Sec. III.5). Experimental values, , correspond to measured heats of adsorption, and all energies are given in units of eV per molecule.

Figure 7: (a) Partial density of electronic states calculated in Ca-anthracene, 3Ca-anthracene, and Ca-decorated graphene using DFT-PBE. (b) Equilibrium structures obtained for the adsorption of CO in 3Ca-anthracene using DFT and MP2 methods. (c) Adsorption energies for CO in 3Ca-anthracene obtained with different geometry optimization and total energy methods. (Figure adapted from work cazorla13 ()).

iv.2.2 Mof

The interplay between open-metal site MOF and CO molecules is dominated by dispersion interactions, which represent about the  % of the total gas adsorption energy ji14 (); valenzano10 (). Besides dispersion, electrostatic and orbital interactions are also important in shaping the gas affinity of MOF. In particular, puntual charges localized in the unsaturated metal sites polarize the CO molecules inducing an electric dipole and a bend distorsion in them ji14 (); poloni12 (); park12b (); koh13 (). When low-energy empty -levels are present a forward donation of electrons from CO to the metal atoms occurs, which tends to increase the electrostatic contribution to the gas binding park12b (); ji14 (); poloni14 (). This last effect is particularly important in MOF-74 and BTT species (see Sec. II.3) containing Ti and V open-metal sites, where the antibonding state that results from the hybridization of oxygen CO lone pairs and metal orbitals remains unoccupied poloni14 (). In the case of heavily loaded MOF, quadrupolar CO-CO interactions turn out to be also very important.

Figure 8: Binding energy of a CO molecule on Ca-BTT computed with the PBE+D2 approach, expressed as a function of the cutoff distance for the pairwise corrections (adapted from Ref. poloni12 ()). The CO-MOF system in which calculations are performed is represented also in the figure.

Next, we review recent DFT benchmarking works done on the screening of MOF for applications in CO capture and sequestration. Due to the crucial relevance of long-range interactions in CO-MOF systems, we disregard here computational studies considering small organic molecules as model GAM (see, for instance, articles vogiatzis09 (); witte14 () where the CO-pyridine and CO-benzene interactions are investigated) since proper convergence of long-range dispersion forces is achieved within distances of at least  Å . This fact is illustrated in Fig. 8, which has been adapted from work poloni12 (), where the DFT binding energy of a CO molecule on a MOF containing Ca open metal sites is represented as a function of the cutoff distance that is employed in the calculation of the dispersion interactions.

In Table VI, we enclose the results of gas binding energy and enthalpy of adsorption, , calculations performed by Poloni et al. poloni12 () and Rana et al. rana12 () in different MOF, using standard and dispersion corrected DFT functionals. For present benchmarking purposes, comparisons with respect to quantum chemistry calculations performed in cluster-size model systems may result not meaningful because of the convergence reasons explained above, and thus the DFT outcomes are compared with experimental heats of adsorption. Enthalpies of adsorption at ambient temperature can be estimated with the formula . Here, and represent the zero-point energy (i.e., , being the corresponding vibrational phonon frequencies) and the thermal energy (i.e., for the CO gas phase -where accounts for the energy of an ideal gas-, and for the framework with and without the adsorbate) differences, respectively, between the joint and disjoint CO-MOF systems.

Upon comparison of the computed and measured values shown in Table VI, we can draw the following two conclusions: first, standard DFT methods are far from reproducing the experimental heat of adsorption trends, with LDA (PBE) presenting large underestimation (overestimation) bias; and second, among the two families of considered dispersion corrected approaches, i.e., semiempirical Grimme’s and vdW- (where indicates the choice of the functional -see Sec. III.5-), the latter always provides the better agreement with respect to experiments (in particular, the vdW-DF and vdW-DF2 variants show null discrepancies within the numerical uncertainties in most of the cases). It is worth noticing that the value of the and differences generally amount to less than  meV, hence the major contribution to comes from the term. Also we note that the structural traits predicted in CO-MOF systems with vdW- methods, not shown here, present overall good agreement with the experiments.

In Table VI, we also include the results obtained by Valenzano et al. valenzano11 () with an hybrid MP2:B3LYP-D approach svelle09 (); tosoni10 (). In that scheme, MP2 calculations are carried out first in a cluster model system representing the adsorption site, and a long-range correction is added afterwards. The long-range correction is defined as the B3LYP-D energy difference between the extended and model cluster systems. (The MP2:B3LYP-D method can be likewise understood as starting from a B3LYP-D calculation for the full periodic structure and adding a high-level correction for the adsorption site.) Valenzano’s MP2:B3LYP-D results obtained in the Mg- and Ni-DOBDC systems actually are in the same excellent agreement with experimental data than found with the vdW-DF and vdW-DF2 methods (i.e., null discrepancies within the numerical uncertainties). Also, they represent an improvement with respect to primary B3LYP-D calculations valenzano11 (). On one hand, Valenzano’s MP2:B3LYP-D work comes to reinforce the accuracy of vdW-DF methods in describing the gas uptake properties of MOF. On the other hand, it demonstrates the reliability and efficiency of the intuitive MP2:B3LYP-D approach for undertaking benchmark and GAM design studies.

Conclusions.- In light of the compiled results and analysis, it can be concluded that the use of standard DFT functionals must be avoided in the study of the CO adsorption features of MOF. Meanwhile, the suite of vdW-DF methods currently represent the best choice for carrying out computational studies of such a type, both in terms of computational expense and reliability. In this regard, we would like to mention that in order to complete our knowledge of the general performance of DFT methods in the simulation of CO-MOF systems it will be highly desirable to consider meta-GGA functionals (e.g., M05-2X and M06-2X) in prospective studies. Also, it would be very interesting to perform advanced RPA-DFT and QMC benchmark calculations in those same systems for complementing the comparisons reported against experiments, and determining more precisely the role of the different energy contributions to the binding of molecules (e.g., many-body dispersion effects). Meanwhile, merging quantum chemistry methods (e.g., MP2) with periodic DFT simulations that incorporate long-range dispersion interactions, appears to be an effective strategy for calculating accurate binding energies in situations where the selected cluster and extended systems are physically alike.

V Discussion and Prospective Work

From all the results and explanations presented in Sec. IV it can be concluded that standard DFT methods (i.e., LDA and GGA) generally provide results on the fixation of H and CO molecules in CN and MOF that are correct only at the qualitative level. Actually, in the sole analyzed case of hydrogen binding to chemically functionalized MOF coherent agreement between DFT-GGA results and highly accurate quantum chemistry calculations has been reported by several authors (see Sec. IV.1.2). However, in that particular case full consistency between DFT results obtained in cluster size and periodic H-MOF systems is still lacking, and further investigations are required for determining the precise role of dispersion interactions. Therefore, our general recommendation is to avoid using standard DFT methods in first-principles modeling of H-storage and CCS materials when pursuing accurate binding energies and geometries. In stark contrast to this our advice, most of the computational studies published to date on gas-adsorption topics have relied heavily on the outcomes of DFT-LDA and DFT-GGA simulations (i.e., about the  % of them in the last years). In Figs. 9 and 10, we show an estimation of the total number of DFT works reported in the areas of H storage and CCS GAM modeling, classified according to the employed DFT functional and publication date.

Figure 9: Number of DFT-based “H storage” articles published from 2000 onwards, classified according to the employed approximation (our estimation relies on data extracted from the “Web of Knowledge”, December 2014). The solid dots represent the total number of works published in different years intervals, and the boxes the percentage corresponding to each DFT variant.

In the case of hydrogen storage (see Fig. 9), we appreciate an important recent decline in the relative number of modeling studies performed with LDA and GGA functionals, as compared to the trend observed during the first decade of the present century (i.e., a decrease of the  %). On the other side, the relative number of DFT works based on dispersion corrected schemes has increased significantly over the same period of time (i.e., from  % to