Function-led design of new porous materials

See allHide authors and affiliations

Science  29 May 2015:
Vol. 348, Issue 6238, aaa8075
DOI: 10.1126/science.aaa8075

It's all about the holes

From kitchen sieves and strainers to coffee filters, porous materials have a wide range of uses. On an industrial scale, they are used as sorbents, filters, membranes, and catalysts. Slater and Cooper review how each application will limit the materials that can be used, and also the size and connectivity of the pores required. They go on to compare and contrast a growing range of porous materials that are finding increasing use in academic and industrial applications.

Science, this issue 10.1126/science.aaa8075

Structured Abstract


Porous materials are important in established processes such as catalysis and molecular separations and in emerging technologies for energy and health. Porous zeolites have made the largest contribution to society so far, and that field is still developing rapidly. Other porous solids have also entered the scene in the past two decades, such as metal-organic frameworks (MOFs), covalent organic frameworks (COFs), and porous organic polymers. No single class of porous material is ideal for all purposes. For example, crystallinity and long-range order might enhance selectivity for a molecular separation while also reducing mechanical stability or processability with respect to less ordered structures. To have an impact on real applications, porous materials must be scalable and must satisfy multiple functional criteria such as long-term stability, selectivity, adsorption kinetics, and processability, all within a viable cost envelope. This presents a broad design challenge, and it requires us to be able to control structure and to understand multiple structure-property relationships at a detailed level.


In addition to MOFs, COFs, and porous polymer networks, other classes of molecular porous solids have emerged in the past 10 years, such as polymers of intrinsic microporosity and porous organic cages. The range of possible functions for porous solids is thus much broader than before. For example, conjugated microporous polymers and some COFs have extended, conjugated structures that are not present in zeolites or MOFs and have led to porous organic photocatalysts and electronic materials. The crystal engineering approaches developed for zeolites, MOFs, and COFs cannot be applied directly to amorphous solids such as porous polymers, but analogous modular strategies have allowed functions such as porosity and electronic band gap to be controlled by choosing the appropriate molecular building blocks. Rapid advances in the computational prediction of structure and function offer a strategy for identifying the best porous materials for specific applications, for example, via large-scale screening of gas adsorption in hypothetical MOFs.


Advances in synthesis have produced new classes of functional porous solids as well as fundamental breakthroughs in areas such as selective carbon dioxide capture, molecular separations, and catalysis. As yet, these rapid developments in basic understanding are unmatched by large-scale commercial implementation, but enhanced functions (such as enzyme-like CO2 selectivity) and new processing options (such as soluble porous solids) present exciting opportunities. A general challenge will be to reengineer porous materials where scale-up is prohibited by cost, retaining the advanced function but using cheaper and more sustainable building blocks. It is therefore important to develop structure-property relationships to understand how promising materials work. Not all future opportunities for porous solids involve improving on existing materials or the development of more scalable preparation routes. For example, porous photocatalysts that can perform direct solar water splitting might provide a completely new platform for energy production. As we seek increasingly complex functions for porous materials, the use of in silico computational design to guide experiment will become more important.

Porous materials can be defined by type or by function, but it is function that will determine the scope for practical applications.

Our ability to design functions in porous solids has advanced markedly in the past two decades as a result of developments in modular synthesis, materials characterization, and (more recently) computational structure-property predictions. This figure is based on the pore channels, shown in yellow, for an organic cage molecule, a new type of solution-processable porous solid developed over the past 6 years.


Porous solids are important as membranes, adsorbents, catalysts, and in other chemical applications. But for these materials to find greater use at an industrial scale, it is necessary to optimize multiple functions in addition to pore structure and surface area, such as stability, sorption kinetics, processability, mechanical properties, and thermal properties. Several different classes of porous solids exist, and there is no one-size-fits-all solution; it can therefore be challenging to choose the right type of porous material for a given job. Computational prediction of structure and properties has growing potential to complement experiment to identify the best porous materials for specific applications.

Porous materials such as zeolites (1), metal-organic frameworks (MOFs) (24), covalent organic frameworks (COFs) (5), and porous polymers (6) have widespread uses in adsorption, catalysis, separation, purification, and energy storage and production. Moreover, the use of porous solids is set to grow in the future—for example, in membranes for water purification. Improving upon commercially established materials, such as zeolites or activated carbons, is a big challenge because they have been optimized over many years to fulfill multiple, combined functions; they not only act as catalysts or adsorbents but can also be processed into a usable form, such as pellets. Efficient manufacturing infrastructure with inexpensive raw materials also makes purely inorganic systems such as zeolites economically attractive. However, not all functions are satisfied by commercial materials. For example, there are as yet no commercial adsorbents that look poised to compete with aqueous amine technologies for carbon dioxide capture, despite the numerous disadvantages of liquid amines (7). Also, porous materials are not solely of interest in adsorption and catalysis: They also have potential as electronic materials, in light harvesting and energy transduction, as proton conductors, and in applications such as molecular sensing. There are hence many motivations to explore new porous materials.

In the past 20 years, the philosophy of molecular design has become increasingly prevalent in porous solids, driven by the framework design principles outlined for MOFs (24, 8), which in turn build on the rules developed for zeolites (1). Design-led approaches have been enabled by advances in synthetic control and by new computational techniques, as well as by developments in physical measurement. For example, structure-property relationships for multivariate MOFs (9)—that is, structural families of MOFs with more than one organic linker—were explored using both solid-state nuclear magnetic resonance measurements and molecular simulations to rationalize properties (10). The properties of multivariate MOFs arise, or “emerge,” from the combination of the constituent molecular building units; in other porous solids, such as organic cages (11), the solid-state function can be intrinsic to the molecular building unit itself. The degree to which function is emergent or intrinsic is a central design question for porous solids, and computational prediction techniques become particularly attractive when function is emergent, and hence less intuitively linked to the isolated molecular building blocks.

Here, we survey some of the advantages and disadvantages of different types of porous solids. We also take a critical look at our ability to design new porous solids at the atomistic level, as considered from the viewpoint of individual functions. An important challenge is to apply these design approaches to real, practical systems, where it is invariably necessary to optimize more than one property, and also to mitigate undesirable properties, either through materials or systems design. For some new metal-organic and organic materials, this may require engineering around their more modest thermal and chemical stability relative to porous inorganic solids.

Functional property drivers

For industrial applications, the primary concerns are functionality and operational costs rather than materials classification. However, porous solids are often grouped into, and reviewed as, separate classes of materials. These classes are themselves diverse in structure and function, but they nonetheless have some general advantages and disadvantages (Fig. 1). For example, porous metal-organic and organic solids have enormous structural diversity but do not exhibit the high-temperature stability of some purely inorganic materials, such as zeolites.

Fig. 1 Classes of porous solids and selected functions.

The relative practical advantages of these materials will depend on the specific application, and all will be compared with established materials, such as porous zeolites (first column). Activated carbon is another important porous solid, not included here. Notes: (i) Metal-organic frameworks, or MOFs, are also known as porous coordination polymers (PCPs) (2). (ii) Microporous materials have pores smaller than 2 nm; mesoporous materials have pores in the size range 2 to 50 nm. (iii) “Ultraporous” refers here to a material with high pore volume and surface area, whether comprising micropores or mesopores [e.g., (4043) and (12)]. (iv) Covalent organic frameworks, or COFs, with boronate ester linkages [e.g., (5)]. (v) COFs with imine linkages [e.g., (86)]. (vi) A zeotype is a framework that is based on a known zeolite topology. (vii) “Multivariate” MOFs comprise more than one organic linker in a single framework [e.g., (9)]. (viii) For example, the combination of three different aromatic linkers allowed band gap tuning over a wide range for porous polymeric water-splitting catalysts (25). (ix) A cocrystal is a molecular crystal containing more than one molecular building unit [see (18, 38)]. (x) PIMs are polymers of intrinsic microporosity [e.g., (14, 21)]. (xi) Most inorganic zeolites are synthesized using an organic template, often an organic amine salt. (xii) See recent review (4). (xiii) See (11, 34). (xiv) Conjugated microporous polymers (22). (xv) PIM-1 (14) is used in a vapor sensor device (

Porous materials can also be classified by their degree of long-range order and by the relative strength of the bonds between the atomic or molecular building blocks. Again, although each class is diverse, some broad groupings are possible (Fig. 2). For instance, porous polymer networks are typically amorphous but robust (12), being composed of strong covalent bonds, whereas MOFs are generally crystalline but the metal-ligand bonding is unstable to water in some cases. Long-range order is not always an advantage; for example, crystallinity is desirable for molecular sieving, where uniform pores are important, but amorphous materials with a hierarchy of pore sizes might be superior for heterogeneous catalysis (13).

Fig. 2 Functional classification of porous solids.

One way to classify porous materials is by their degree of long-range order and their intermolecular bond strengths. These parameters affect function; for example, long-range order may be useful in molecular sieving applications, and strong intermolecular bonds may promote chemical or thermal stability. Only molecular porous solids can be processed as true solutions.

Stability issues notwithstanding, weak intermolecular bonds can be useful in some cases. Three-dimensional networks and frameworks are, by definition, insoluble, whereas linear polymers of intrinsic microporosity (PIMs) (14) and porous molecular materials (15) have no formal intermolecular bonds and can therefore be processed as molecular solutions—for example, to form gas separation membranes (14, 16). In metal-organic materials, the intermolecular bond strength is tunable over a broad range, and some crystalline porous coordination polymers can be melt-processed (17).

Choosing the best porous solid for an application is challenging because success can rarely be gauged from one figure of merit alone, such as surface area or catalytic turnover number. As well as achieving its core target function, a porous material must also meet various other requirements such as stability (e.g., thermal, mechanical, hydrolytic, chemical, or photolytic stability), processability, ease of handling, adsorption/desorption kinetics, and thermal transport behavior. The use of expensive, rare, or toxic elements or reagents, either in the material itself or in its synthesis, is also to be avoided if possible. Ultimately, industry deals less in individual classes of materials and more in processes, or unit operations, that must deliver an integrated function within an acceptable life cycle and cost. As such, the relative merits of a new porous polymer, for example, should be considered alongside MOFs, COFs, porous molecular solids, and existing materials such as zeolites and activated carbon (Figs. 1 and 2).

Modular design of function

Some important advances in functional porous solids, selected from the past 10 years, are shown in Fig. 3. In principle, it is possible to “dial in” almost any function in a porous material by combining atomic-level control over structure and composition with an adequate understanding of structure-property relationships. In practice there are major challenges, both in attaining structural control in solid-state materials and in predicting function from structure, even for materials where the structure is well established. For this reason, some of the materials shown in Fig. 3 were “designed” whereas others were “discovered.” Effective material design becomes particularly difficult when the porous material must satisfy a diverse range of different and sometimes orthogonal functional criteria. For example, crystallinity provides order but can also lead to brittleness; high pore volumes can provide good gravimetric sorption capacity but might also reduce volumetric sorption capacity, heat transport, and mechanical strength, both for crystalline and amorphous solids.

Fig. 3 Selected advances in functional porous solids.

Both experimental and computational methods have progressed rapidly in the past decade to yield a range of materials with new or improved functions. Numbers in parentheses denote relevant references.

MOFs have an inherent design advantage because the rigid node-strut topology and coordination bonding can be used to control both structure and composition (4, 8) (Fig. 1). Likewise, it is possible to control structure and composition in crystalline porous molecular solids, such as porous organic cages (15, 18), by exploiting strong noncovalent intermolecular interactions. COFs have similar structure-property design advantages. Their long-range order has often been more restricted, although single-crystalline COFs have also been prepared (19, 20). However, programming of structure and composition is not limited to ordered, crystalline solids; this has also been achieved in a modular way for a diverse range of amorphous porous polymers (6, 14, 21, 22). In certain cases, the molecular repeat unit in the material can dominate properties, which allows function to be programmed without any long-range structural order. For example, pore volume and surface area is closely linked to the length of the rigid organic linker in conjugated microporous polymers (CMPs) (22, 23), even though they are amorphous, because short linkers avoid network interpenetration, as in some crystalline MOFs. Another well-established design principle in amorphous porous solids is the requirement for structural rigidity and sites of contortion in PIMs (14, 21) (Fig. 4A). Molecular cages can also have intrinsic porosity: The shape selectivity for aromatic hydrocarbon isomers in a porous organic cage molecule was found to be defined by the shape of the cage itself (11) (Fig. 4B), which showed the same shape selectivity both in the solid state and in molecular solutions. Here, function is intrinsic to the molecular building unit, in contrast with MOFs where the isolated organic building units typically show none of the physical properties of the self-assembled framework. This modular philosophy, where the molecular building blocks encode the target function, also applies to electronic properties. For instance, electronic structure calculations for photocatalytic CMPs (24, 25) showed that the electronic properties of the polymer can be represented by small clusters of just a few monomer units (26).

Fig. 4 Modular programming of function in porous molecular organic solids.

(A) Porosity is defined by the rigid bonding in the monomers of linear polymer. Reaction of the diamine monomer with dimethoxydimethane (DMM) and trifluoroacetic acid (TFA) yields a polymer, PIM-EA-TB, with extreme rigidity that combines excellent gas selectivity and gas permeability. (B) Molecular selectivity for hydrocarbons is defined by an isolated organic cage molecule. The triangular cavity in the cage (highlighted in green) is a good shape match for mesitylene, which can be entrapped by the cage, and also for m-xylene, but not for p-xylene. PIM-EA-TB is amorphous while the cage molecule is crystalline, but the organic building blocks encode solid-state function in both cases.

Amorphous porous materials present particular difficulties for functional design because it is laborious to create an unambiguous model for the extended three-dimensional structure. As a result, of the several hundred recent papers on porous organic polymers (27), only a handful give extended structural models. Automated computational tools that can grow amorphous structures in silico offer a way forward (28, 29). Even for crystalline porous solids, our capacity for de novo design of function is still limited. The intuitive, topological assembly rules for MOFs, where families of metals and ligands assemble to give the same general framework topology (8), are not generic: Not all metal-ligand combinations lead to broad, structurally related families, and framework interpenetration (where one framework grows within another) can be difficult to anticipate by chemical insight alone. Exploratory syntheses, rather than more linear design strategies, are still therefore dominant. One can contrast the large number of hypothetical MOFs that we can enumerate computationally (30, 31) versus our current ability to identify their relative thermodynamic stabilities, and then to synthesize specific, desirable frameworks in a targeted way. For porous organic cage molecules, it is possible to compute the most thermodynamically stable crystal packings from knowledge of the constituent molecules alone (18)—that is, assuming no underlying topology. If these methods can be made less computationally expensive, this offers a future strategy for the unconstrained in silico prediction of structure and function.

The prediction of physical properties from structure is also challenging, even when the structure of the porous material is known. The computational prediction of gas adsorption behavior is the best-developed area, and there are many examples where gas sorption isotherms have been computed for single-component gas systems (32, 33). Most studies make use of the rigid framework approximation, which becomes increasingly unreliable for flexible structures and for structures where the pores are of similar dimensions to the sorbate, although a growing number of studies have tackled adsorption selectivity in porous materials using methods that account for flexibility and molecular dynamics (11, 34, 35). Other physical properties are more difficult to predict from structure alone, and these include important functions such as catalytic activity and proton conductivity. Also, most variants of stability, such as mechanical stability (36), and even basic properties such as melting point, are relatively difficult to predict de novo from knowledge of atomic structure.

In summary, our ability to design function in porous materials a modular way has evolved markedly in the past 20 years, both for crystalline and amorphous porous solids. For example, crystalline MOFs (9), amorphous CMPs (37), and porous organic cage molecules (38) all constitute modular organic building blocks that may be combined interchangeably to produce more complex functional materials. Recently, but still uncommonly, computation has been used to guide rather than to post-rationalize experiments (39).

Porosity and surface area

Extraordinary surface area (4, 4042) and pore volume have been achieved in MOFs and COFs, including mesoporous frameworks with pores sizes up to 9.8 nm (43), but the stability of these ultralow-density structures is as yet unproven for practical applications. For applications such as gas storage, surface area per se, as calculated from nitrogen adsorption isotherms, may not be the defining property (39, 44, 45). The design principles for achieving “ultraporosity” in crystalline MOFs were reviewed quite recently (4). High levels of porosity are not, however, limited to crystalline frameworks. Amorphous polymers can have surface areas of more than 5600 m2/g (12) through the use of rigid tetrahedral organic nodes and short, rigid linkers to avoid framework interpenetration (46). These covalently bonded networks are also stable to boiling water. Extremely high surface areas do not necessarily translate into direct advantages for applications such as CO2 capture or molecular separations, but high pore volumes do provide other opportunities for diversification of function. Postsynthetic modification of robust, high–surface area polymers can introduce functionality such as acids (47) and amines (48). Even highly derivatized polymers could retain substantial surface area and adsorption capacity because the native, unfunctionalized framework started with such a high pore volume. Until recently, molecular solids have exhibited very low pore volumes and surface areas (49), but recent studies revealed molecular organic solids with surface areas in excess of 3700 m2/g (50). Here, the design principle is protected free volume within a porous organic cage; by contrast, most crystalline molecular materials with cavities are not stable to desolvation. Several design approaches have been used to preserve porosity in molecular crystals, such as strong intermolecular hydrogen bonding (51), the computational prediction of shape persistence in cage molecules (52), and the use of metal-coordination bonding to create “wall ties” (53) that stabilize an otherwise unstable, low-density crystal structure.


Porous solids are an energy-efficient alternative for molecular separation processes that are currently performed by cryogenic distillation or by selective liquid solvents. The major formats are porous adsorbents and membranes (54). MOFs can be highly effective in molecular separations because the pore channel size, shape, and functionality can be controlled, much as for crystalline zeolites. An iron MOF, Fe2(dobdc) (dobdc: 2,5-dioxido-1,4-benzenedicarboxylate) (55), selectively adsorbed alkenes because of open iron(II) sites that can coordinate unsaturated species, thus allowing the separation of alkenes from alkanes. Likewise, affinity separations using metal-organic coordination were achieved in metallated amorphous porous solids (56, 57). Hydrocarbons can also be separated by MOFs using a summation of weaker supramolecular interactions (58). A porous framework was used to separate carbon monoxide (CO) and nitrogen by exploiting a unique structural transformation that occurs upon CO binding, thus giving rise to “self-accelerating” CO sorption (59).

The separation of branched and linear C6, C7, and C8 hydrocarbons is another important process in industry, currently performed using zeolites. Here there is no metal coordination strategy because these hydrocarbons are saturated. An iron MOF, Fe2(BDP)3 (BDP: 1,4-benzenedipyrazolate), was shown to discriminate between linear and branched hexane isomers at industrially relevant temperatures (60). It was hypothesized that linear n-hexane has stronger van der Waals interactions with the triangular channel walls in the MOF.

Selectivity is a defining function in adsorbents for postcombustion carbon capture (7) where it is necessary to adsorb CO2 in preference to other gases, such as nitrogen. Selective CO2 adsorption has been demonstrated in MOFs where the CO2 inserts into the metal-ligand bonds, which suggests that the function of CO2-fixation enzymes might give clues to designing better synthetic CO2 adsorbents in the future (61). In this system, crystallinity is probably essential for the resulting function: The energy-efficient adsorption and desorption of CO2 over a very narrow pressure range occurs because of cooperative effects between neighboring functional groups in the crystalline lattice. Amorphous porous polymers, for example, might not give similar cooperative behavior, even if suitably metallated (56, 57). Excellent selectivity and CO2 capacity have also been demonstrated for water-tolerant metal-organic materials based on the hexafluorosilicate anion (62). Hydrophobicity is a useful qualitative design principle for CO2/H2O selectivity. Recent studies on hydrophobic hyper–cross-linked polymers show that these materials swell in CO2 at pressures relevant to precombustion CO2 capture and that they adsorb very little water, outperforming materials such as MOFs, zinc imidazolate frameworks (ZIFs), zeolites, and activated carbons under these conditions in terms of both CO2 capacity and CO2/H2O selectivity (63). The rational, atomistic design of more selective materials here is doubly complex, both because the materials are amorphous and because swelling is not easily captured in molecular simulations. Sorbent swelling could present an engineering problem in packed adsorbent beds, especially in postcombustion carbon capture where close-to-zero pressure drops are required. Fascinatingly, it seems that framework flexibility can be strongly influenced by crystal size (64), thus suggesting the scope to modulate such effects.

For polymer membranes, selectivity is a complex function of both solubility and diffusion terms. Design of function in amorphous polymer materials has so far been largely empirical, but this has yielded many successes. In particular, PIMs with extreme rigidity, which are nonetheless solution-processable, can combine excellent gas selectivity and gas permeability (21). The one-step synthesis of these rigid polymers is simple and elegant (Fig. 4A).


Applications such as molecular separations, heterogeneous catalysis, and proton conductivity using porous materials are all critically dependent on diffusion kinetics (65). The actual regime of diffusion limitation is specific to a given application. For instance, the principal diffusion resistance may not be in the nanopore size range and may actually be in the mesopore or macropore region, and then the engineered form and shape of the sorbent material may become more important. For membranes, there is a compromise between selectivity and permeability (66), and there is a major drive to discover new materials that maximize both of these parameters (21, 54). Physical aging of membranes is a problem that needs to be addressed, because initial permeabilities in PIMs can decrease drastically over quite short periods of time (67). The use of mixed-matrix membranes is one possible solution (16, 68), where a second, rigid filler material is blended with the polymer to stabilize the porosity over time, although this also introduces various processing challenges.

In CO2 capture from flue gas, the amount of CO2 that can be captured at a given gas flow rate is determined by the adsorption kinetics. Materials with high equilibrium CO2 uptakes but slow kinetics may not be useful because these equilibrium values might not be achieved in a dynamic flow situation. Hence, for gas separations, the working capacity defines the quantity of gas that can be adsorbed and desorbed by the sorbent within a given pressure- or temperature-swing time cycle (7). A general challenge is that adsorbents with excellent selectivity might often have poor kinetics and hence poor working capacities because the pore channels are, by definition, close in size to the dimensions of the guests (11). One promising strategy is the hierarchical structuring of materials to contain interconnected large and small pores, again highlighting the importance of processability.


It is uncommon for porous materials to be usable in applications as synthesized; they are typically processed into a specific form, such as a pellet, a thin polymer membrane (54), or a surface-deposited MOF or zeolite film (6973). Processability is therefore an important functional property that is often ignored, at least in the early discovery phase for new materials.

Mixed-matrix membranes (MMMs) are promising composite materials for gas separations (68), but processing challenges arise where the constituent materials in the MMM are incompatible. This can lead to nonselective channels at the phase interfaces between the organic polymer and the filler, which is typically an inorganic material such as silica or zeolite. An important development here is the use of porous organic fillers, such as organic cage molecules (16) or porous polymers (74), which in principle can form a more stable interface with the surrounding organic polymer phase. A key advantage of the organic cage MMM approach is that both the cage and the membrane polymer are soluble in common organic solvents, thus avoiding potential processing issues associated with colloidal particulate fillers. Insoluble materials (Fig. 2) can nonetheless be processed, as illustrated by the development of large-scale pervaporation modules based on zeolite membranes grown on tubular supports (73).

Processability can also have a strong influence on cost, including that of the constituent feedstocks for the material and of any steps needed to render the material into a usable form. Simple, scalable processing routes with few steps are therefore desirable, irrespective of material type.

Mechanical properties

The success of porous materials in industrial processes depends on their stability to stresses such as tension, compression, shear, bending, torsion, and impact (75). There are comparatively few studies on this important subject. Nanoindentation studies on MOF-5 concluded that inert atmospheres and low applied stresses would be needed to maintain structural integrity (76). The mechanical properties of “soft porous crystals,” a subclass of MOFs (77), were found to be low, particularly with respect to shear stresses and fracture propagation, although their synthetic variability allows scope to design new materials with improved strength. To maximize the strength of MOFs, it was suggested that framework connectivity should be as high as possible, as in the case of UiO-66 with 12-coordinate Zr, while the organic linkers should be short in length. Porous amorphous organic polymers show a similar trend of decreasing mechanical stability with increasing linker length (78); boron-based COFs were calculated to have similar mechanical properties to porous aromatic frameworks (79). There are hence some general design rules that transcend material subclass, and calculations here can show good agreement with experiment (36).


Stability can mean many things in addition to mechanical stability, and it is an important, general challenge for porous materials in the future. Most applications for porous solids will require stability in air as a minimum. Other applications, such as post-combustion CO2 capture, also require stability under humid, acidic conditions (63). For photocatalysts, long-term stability under intense irradiation is a requirement (25). In terms of thermal stability, high-temperature heterogeneous catalysis is likely to remain the preserve of inorganic materials, because the operating temperatures effectively rule out most of the newer porous organic and metal-organic materials. For lower-temperature processes, there is a growing body of work that shows that new materials might have adequate thermal stability.

The first generation of porous MOFs and boronate ester COFs had reasonable thermal stability, but they were rather unstable to moisture. The hydrolytic stability of MOFs has since been greatly improved using design strategies such as hydrophobic organic linkers (80) or postsynthetic modification with diazo groups (81). UiO-66 and its derivatives were shown to retain their structural stability after immersion in water and 0.1 M HCl solutions; HKUST-1 (Cu-BTC) is water-stable under certain conditions but degrades at higher relative humidities and/or temperatures [90% relative humidity (RH) at 298 K] (82). MOFs have also been studied for their water adsorption and dehumidification properties (83, 84). To be useful in practical applications such as CO2 capture, porous materials must be stable to the components in flue gas, not least to CO2 itself and acidic components such as SO2, which is difficult to scrub. Recent studies cast doubt on the long-term stability of some ZIFs under humid conditions: A carbonate decomposition product was observed upon exposure of ZIF-8, dia-Zn(MeIm)2 (MeIm: methylimidazole), or ZIF-67 to either gaseous CO2 in 100% RH or CO2 dissolved in water (85). Porous organic polymers can have excellent hydrolytic stability as well as stability toward acids and bases, although long-term oxidative stability is less well studied. Crystalline or semicrystalline COFs that are formed from less labile covalent linkages (86, 87) should have good chemical stability. Surprisingly, porous molecular crystals can also show high stability to water (88), or even to acids and bases (pH = 2 to 12), so long as reversible imine linkages are not present (89).

Thermal properties

Heat management is a fundamental issue both in adsorption processes, such as gas storage or separation, and in heterogeneous catalysis. There are two related parameters to consider: (i) how much heat is released upon adsorption or by a catalytic reaction, and (ii) how readily that heat is dissipated in the material. Poor heat dissipation would be a problem for large adsorbent or catalyst beds. The degree of heat released is a function of the pore size and pore chemistry, and for catalysts it is also a function of the nature of the chemical reaction. Heat dissipation is a function of the thermal conductivity of the material, its specific heat capacity, and its physical form. A high specific heat capacity is not necessarily advantageous; for example, in temperature-swing adsorption and desorption, this may increase the amount of thermal energy that is required to regenerate the adsorbent. By contrast, good thermal conductivity is generally an advantage because it allows heat to be removed from or added to the porous material.

MOF-5 has poor thermal conductivity: 0.31 W m−1 K−1, which is similar to concrete (90) and around an order of magnitude lower than other porous solids such as the zeolites sodalite and faujasite. ZIF-8 was predicted to have an even lower thermal conductivity of 0.165 to 0.190 W m−1 K−1 (91). These low values are not surprising given the low atomic number density for the materials, and comparable values might be expected for many of the material types summarized in Fig. 1. Recent computational studies suggest that there is some scope for designing higher thermal conductivities in MOFs, and by implication other porous solids, by considering the framework topology (92).

Catalytic activity

Porous frameworks offer many advantages over both homogeneous and nonporous heterogeneous catalysts, for example, in terms of ease of recovery and shape and size selectivity. Zeolites have by far the most widespread use in industrial catalysis, especially in the area of size-selective heterogeneous catalysis (93). Catalytic activity has also been demonstrated in a wide range of MOFs (9496), COFs (97) and porous polymers (98), in principle addressing problems that zeolites cannot tackle. For example, one recent strategy is “defect engineering” to yield MOFs with coordinatively unsaturated metal centers, allowing unusual reactivity (99). In addition to more traditional thermal catalysis, heterogeneous photocatalysts are of interest in applications such as wastewater treatment, hydrogen production, artificial photosynthesis, and degradation of pollutants (100). In this regard, the growing number of metal-free porous polymer organocatalysts is of interest (98, 101, 102). In particular, carbon nitride (103) and carbon nitride composite materials (104) show promise as metal-free catalysts for the photochemical splitting of water, although those materials are not inherently porous. Extended conjugation is not a typical property of zeolites or MOFs, but this can be achieved in conjugated microporous polymers (22) and in some COFs (105107). The modular design of improved porous organic photocatalysts—for example, by using “band gap engineering” strategies (24) to allow more effective adsorption of the available solar spectrum—is a design strategy for the future, as exemplified recently for porous polypyrene materials for photochemical hydrogen evolution (25). The introduction of high levels of porosity should allow faster and more efficient mass transport of water to the photogenerated charges in the catalyst.

Other specific functions

Porous frameworks (108110), porous molecular solids (111), and zeolites (112) all show promise as proton conductors. Porous COFs (105107) and MOFs (113) have potential in organic electronics, and porous conjugated microporous polymers are promising supercapacitors (114). Solution-processable porous organic solids have been incorporated into sensors (115). The degree to which function can be designed into these various applications depends upon the property of interest. For example, the design of new proton conductors and supercapacitors is, at this stage, somewhat more empirical, not least because there is no single mechanism for proton transport or for charge storage.

De novo computational design

Although we are still some way from the routine de novo design of even individual functions for porous solids, there have been major recent advances in this direction. Large-scale computational screening was applied to MOFs, where 102 building blocks were used to generate 137,953 hypothetical frameworks (31). This initial example was limited to rigid frameworks, and candidate structures were restricted to no more than four unique building blocks per MOF. Monte Carlo simulations suggested a large number of hypothetical materials with methane adsorption capacities that were greater than the record material at the time (230 volSTP vol–1), and one of these predicted materials was found to be related to an experimental MOF that did indeed have excellent methane capacity that was close to the predicted value. In a more recent study, the same team explored the limits of practical methane storage in MOFs, and in particular the feasibility of reaching the Advanced Research Projects Agency–Energy target methane delivery capacity of 315 cm3(STP)/cm3 (39). A computational survey of 122,835 hypothetical frameworks considered how properties such as void fraction, volumetric surface area, and heat of adsorption affect the deliverable methane capacity (Fig. 5A). The results indicated that the best delivery capacity for any of the hypothetical frameworks was substantially lower than the 315 cm3(STP)/cm3 target (Fig. 5B). Only an artificial factor of 4 increase in the Lennard-Jones ε parameters in the simulations, combined with an increase in the delivery temperature to 398 K, allowed this target to be attained (Fig. 5C). This illustrates the difficulty in designing adsorption sites in frameworks that interact strongly enough with methane, and it is an excellent example of the use of computation to guide rather than simply rationalize the design of porous solids. Recently, a “computation ready” (116) database of MOF structures has been made publicly available, thus allowing a broader computational community to participate in such property predictions for large libraries of materials (116, 117). Computational screening has also been applied to identify zeolites for ethanol/water separations (118) and for natural gas purification (119); in the latter case, a hierarchical approach was applied that included process cost analysis as the final ranking step.

Fig. 5 Exploring the limits of methane storage and delivery using high-throughput computation.

(A) Molecular simulations for 122,835 hypothetical MOFs and 39 idealized carbon-based porous materials yield a map of methane deliverable capacity for the materials, as relevant to natural gas storage applications. The relative thermodynamic stabilities of the various hypothetical frameworks were not considered. (B) Structures of HKUST-1, NU-125, and NU-111, the three best MOFs in terms of deliverable capacity between 65 and 5 bar, and Ni-MOF-74, the best MOF for methane storage at 35 bar. (C) Methane delivery capacity (between 65 bar/298 K and 5.8 bar/398 K) for 48,000 hypothetical MOFs where the Lennard-Jones interaction parameter, which describes the interaction between methane and the MOF, is normal (purple), artificially doubled (green), and quadrupled (blue). A quadrupling of this parameter is needed to approach the ARPA-E target of 315 cm3(STP)/cm3 deliverable methane capacity, illustrating that even a wide range of hypothetical MOF structures do not interact strongly enough with methane to be practical, even though there is, in principle, sufficient surface area available in some of these materials.

These high-throughput property predictions for libraries of hypothetical frameworks are a potentially enabling development for experimentalists, but there are also limitations. The frameworks were treated as rigid to reduce computational cost, although computational methods do exist that can treat framework flexibility for smaller numbers of frameworks (35, 120, 121). These high-throughput methods also do not provide a synthesis protocol for the computer-generated MOFs, nor do they ensure their thermodynamic stability. Indeed, such methods will tend to generate a large number of structures that are insufficiently thermodynamically stable to be accessed by experiment. Calculations have been used to assess the relative stabilities of different framework topologies, both for crystalline MOFs (122, 123) and for amorphous porous polymers (46), but these methods are not yet computationally affordable or sufficiently generalized to be used for the routine selection of promising synthetic targets, even when composition is fixed or constrained. The unconstrained prediction of the most thermodynamically stable compositions from a range of possible hypothetical phases is extremely challenging.

Is it possible, then, to predict structure, thermodynamic stability, and functional properties for porous materials from their constituent molecular building blocks without any input from experiment? There are no single studies that demonstrate such a de novo computational approach in its entirety, but there is potential to do this in the future—for example, by introducing energy ranking steps into the aforementioned large-scale computational screening of hypothetical MOFs. The strategy can also be illustrated for molecular organic crystals. The most stable organic cage compound for a given combination of organic linkers can be computed (Fig. 6, step 1) (124), along with the potential of the isolated cage to be shape-persistent and porous (52). Using the most stable computed cage structure, the lowest-energy crystal packing can then be computed using crystal structure prediction (step 2) (18). This computed crystal structure then feeds molecular dynamics simulations to probe gas diffusion and the dynamic pore size envelope (step 3) (125) and Monte Carlo simulations to evaluate gas sorption properties and gas selectivities (step 4) (34), much as for large-scale computational MOF screening. Hence we could compute, without any physical experiments, that reaction of 1,3,5-triformylbenzene and (R, R)-cyclohexanediamine should form a flexible, porous molecular solid with a remarkably high selectivity for xenon adsorption, as found by experiment (34).

We stress that this is a retrospective example; in reality, steps 1 to 4 were carried out in a series of studies over the course of about 5 years (18, 34, 52, 124), with experimental validations interspersed between the steps. However, with suitable advances in methodology and hardware, virtual screening methods could become a competitive reality. The largest single challenge here is perhaps the structure prediction and lattice energy calculation (step 2), which are not yet generic, especially if less rigid building blocks or the de novo prediction of the most likely compositions for multicomponent solids are considered.

The cages illustrated in Fig. 5 are quite large, prefabricated modules. By contrast, most MOFs, COFs, and polymers are prepared from smaller building blocks, which increases computational expense for unconstrained structure-composition searches. It is possible that a strategy of “extended modules,” as used to predict function for inorganic oxides (126), might also be adapted to porous framework solids so that the structure and function of complex materials could be predicted in a more granular way.


The field of porous materials is at an exciting stage in its evolution. Compared to 20 years ago, there are many more types of functional porous solids to choose from, and zeolites, while mature and commercially established, are still developing rapidly. For example, two-dimensional zeolites (72) open up new possibilities for what were hitherto thought of as three-dimensional materials. This new diversity can, however, make it hard to select the best material for a specific purpose. The qualitative criteria presented in Figs. 1 and 2 offer some guidance, but we suggest that the long-term solution is to develop computational structure/property prediction tools to augment experiment, ultimately to allow the de novo design of new porous solids. This would have strong translational benefit because the same prediction challenges exist for a wide range of nonporous solid-state materials; steps 1 to 3 in Fig. 6, for example, could be applied to any crystalline organic solid, whether porous or not.

Fig. 6 In silico prediction of function from molecular building blocks.

As computational methods evolve, it will become increasingly possible to predict structure and function for porous materials from knowledge of the primary building blocks alone. In this example, the most stable organic cage structure can be predicted from the aldehyde and amine building blocks (step 1) and the crystal structure can then be predicted from the structure of the cage module (step 2). Molecular dynamics simulations (step 3) give information about gas diffusion rates, and grand canonical Monte Carlo simulations (step 4) predict both gas capacity and gas selectivity. This workflow could be applied in the future to hypothetical molecules, which would then allow the true in silico design of function for porous solids. Numbers in parentheses denote references relevant to the individual steps.

Pending development and generalization of these predictive tools, materials design will proceed mainly via more qualitative, empirical rules, iterated with synthesis and property measurements. In this respect, some general opportunities and challenges can be identified. Carbon dioxide capture is an important unsolved problem where porous materials could be the solution, but the barriers to commercialization are daunting, not least because of the enormous scale: A single power plant can produce more than 20 billion kg of CO2 annually. Candidate materials must therefore have low cost and long service lifetimes (i.e., good stability). Although substantial progress has been made on selectivity, both for adsorbents (6163) and for membranes (21), scale-up potential will equally depend on kinetics. For membranes, selectivity and permeability are usually studied together, but for adsorbents, kinetics are less commonly investigated, and this is an area that requires more attention. In order for PIM membranes (14, 21) to displace commercial high–free volume polymers in large scale separations, or to establish entirely new separation processes, it will be important to solve the physical aging problems associated with these materials (67). Improved water purification and water recovery (83, 84) technologies are also of growing importance worldwide. In principle, many of the material classes outlined in Fig. 1 could translate into water purification technologies, and this will likely be a future growth area, although again the scale of these processes necessitates improved performance, low cost, and high durability to compete, for example, with commercial desalination membranes. Heterogeneous catalysis is a mature field, and new porous solids must show specific advantages over established catalysts and catalyst supports such as zeolites, alumina, silica, or simple polymer resin technologies. So far, many heterogeneous MOF catalysts are simply variants of these more classical materials. However, a range of new opportunities exists; for example, the rise of asymmetric molecular organocatalysis (127) predates the rapid expansion in porous organic polymers (6, 22, 27) by just a few years. It is likely, therefore, that a range of metal-free porous heterogeneous polymer organocatalysts will soon follow (98, 101, 102), especially given the enormous synthetic versatility in porous polymers, which is perhaps their key advantage (Fig. 1). More specifically, CMP photocatalysts (25) have extended, conjugated chromophores that are not present in more traditional heterogeneous catalysts. The recent discovery of an organic, metal-free photocatalyst that performs photochemical water splitting for 200 days without a reduction in activity (104) suggests that for at least some organic materials, long-term photostability can be achieved that matches or exceeds that of traditional inorganic semiconductors. More active polymer photocatalysts for water splitting could in turn stimulate the search for improved porous separation media for the hydrogen and oxygen gas streams that would be produced. Alternatively, it might be possible to design a porous membrane material that not only produces hydrogen and oxygen gas but also separates them—for example, by tailoring the pore size.

Separation, adsorption, and heterogeneous catalysis are what might be called “traditional” areas for porous materials. There are also recent studies that demonstrate new possibilities for porous solids (105115) where the advantages over materials such zeolites are perhaps more clear-cut. For example, in battery and supercapacitor technologies, new porous solids (114) will be competing instead with materials such as graphite and graphene. New porous electronic materials have developed rapidly in recent years, where crystalline porosity in COFs, for example, has been used to position functional units with respect to one another in space (105107) more than to achieve porosity per se. Dopable, solution-processable porous molecular solids (15), perhaps designed using crystal structure prediction methods (18), would be a logical future development here.

References and Notes

  1. Acknowledgments: We thank R. L. Bedard for his advice and for discussions that led to the original concept of this review and T. Hasell, K. E. Jelfs, M. A. Little, L. Chen, and M. Zwijnenburg for input. Supported by the Engineering and Physical Sciences Research Council (EP/H000925/1) and European Research Council under the European Union’s Seventh Framework Programme/ERC Grant Agreement no. 321156.
View Abstract

Stay Connected to Science

Navigate This Article