Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 89
5 The Potential Impact of HECC in Chemical Separations INTRODUCTION Separation processes—the production of two or more streams that are different in composition from that of the feedstock—are ubiquitous. They can operate on the smallest amounts of matter that consist of more than one atomic or molecular species or on the scale of the cosmos, where atoms and subatomic fragments are separated by the action of gravitational and other fields. Around the world, separation processes are building blocks for a wide range of industrial and environmental processes that impact society broadly and in many ways. For example, chemical separations are essential for the following purposes: • Removal of toxic substances like mercury from the flue gases of coal-fired power plants and removal of a range of organic and inorganic pollutants from wastewater streams. • Removal of the greenhouse gas carbon dioxide from power plant flue gases. • Recovery of very dilute but highly radioactive cesium-137 from nuclear-waste streams (NRC, 2000). • Separation of nitrogen, carbon dioxide, water, and other contaminants in gas from natural gas wells, coal bed methane wells, and landfills so that the methane can be added to the interstate pipeline system. • New separation applications to accommodate the commercialization of green products. • Production of potable water in many developing countries. • Purification of a growing number of new drugs from their chiral (mirror-image) compounds, which can in many instances be highly toxic. The energy requirements to achieve the separated products are substantial, however, and come at a time when we can least afford it, with sometimes negative environmental consequences that can no longer be ignored. In the mid-1990s, separation processes in the chemical industry alone consumed about 7 percent of the total energy used in the United States (NRC, 1998); separation processes used in 

OCR for page 89
0 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING other industries, while difficult to quantify from an energy-use standpoint, probably added one to several percentage points to that number. About 60 percent of the total energy requirements of the chemical and petroleum processing industries are consumed by separation processes (DOE, 2005). Capital investments in separation processes are also a very important factor, with 40-70 percent of the total investments in various separation-intensive industries being consumed by these processes (Humphrey and Keller, 1997). Given that separation processes consume so much energy, it is clear that they also contribute very significantly to the nation’s output of greenhouse gases. Thus for three reasons—energy use, investment costs, and environmental considerations—the incentives to improve these processes, as well as to invent and develop new ones, are very great. Box 5-1 portrays both the breadth of the separations field and the large number of disparate industries in which these processes are applied. Most chemical separation processes are based on thermodynamic equilibrium considerations. When, for example, a liquid stream containing two or more components is heated and forms a vapor phase in contact with the liquid, at least a partial separation of the components is possible if the resulting two phases at equilibrium have different compositions. Distillation is highly effective at separating compounds based on differences in their relative volatilities. From a design point of view, distillation-based processes are favored not only because their mechanical simplicity often leads to low investment costs but also because their design requires a much smaller set of phase-equilibrium data than all other separation options to quantify and optimize the efficiency of the separation. This fact accounts in large part for the historic preference for distillation over alternative methods. Distillation, because it requires that the mixture be repeatedly vaporized and condensed, nonethe- less consumes tremendous amounts of energy. Historically, energy consumption and its concomitant carbon dioxide release were not deemed to be of great concern, so chemical industries tended to design BOX 5-1 Major Separation Processes and Industries That Depend Heavily on Chemical Separations Separation Processes Distillation Membrane-based Filtration Solvent extraction crystallization Bubble/foam fractionation Supercritical gas extraction Ion exchange Electrodialysis Gas and liquid adsorptions Drying Liquid chromatography Gas absorption Industries Served Organic and inorganic Electronic products Industrial, municipal, and chemical production Food processing agricultural waste Polymer production Biochemical products treatment Petroleum refining Biofuels production Hospitals and other Pharmaceutical production Advanced biotech health-care entities Ore, coal, oil, and gas products Homeland security extraction and cleanup

OCR for page 89
 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS TABLE 5-1 Examples of Mass-Separating Agents and Their Applications Mass-Separating Agent Application Zeolite molecular sieve Oxygen from air, hydrogen recovery, isomer separations, glucose-fructose separation, CO2 removal from gas streams, water removal from ethanol Activated carbon Removal of trace organics from water and air, of color from petroleum fractions, and of odor and taste bodies from water Ion exchange resin Removal of specific ions from various, usually aqueous, streams Functionalized solvent Separation of derivatized organics from simpler organics Water Separation of ions and polar organics from organic phases Polymer membrane Nitrogen from air, hydrogen recovery, water removal from gases, water purification, CO2 recovery, desalination, biological materials separations Filter Removal of solids from gases and liquids Flocculating agent Concentration of fine particles and biological agents in aqueous streams separation systems based on distillation if it was a viable option, turning to other options only if it was not. This approach remains dominant, even though most of the alternatives to distillation would require less energy and produce less CO2. Given that distillation is by far the most common separation process, used in as much as 80 percent of all the chemical separations listed in Box 5.1, optimization of phase equilibria will remain an important grand challenge for the chemical separations industry. It is also true that distillation is sometimes not an effective option. Instead, mass-separating agents (MSAs)—solvents, absorbents, adsorbents, membranes, and so on—are often added to amplify the sepa- rating capability for these more intractable systems, while potentially providing for more economical, environment-friendly solutions. Some examples of MSA-based processes are given in Table 5-1, which we amplify by focusing on two examples of their use that have broad societal implications. Example 1: Pure Oxygen from Air Even though oxygen is already produced inexpensively on a massive scale, the number of uses and overall volume produced could grow substantially if its price were cut even more. Some of the existing and potential applications include the following: • Feeding oxygen instead of air to power plant furnaces to reduce the volume of flue gas produced and to increase the percentage of carbon dioxide, sulfur oxides, and nitrogen oxides in the flue gas, dramatically reducing the cost of their recovery. Whether this use comes about is highly dependent on the need to sequester the carbon dioxide. • Feeding oxygen to gasification reactions such as occur in next-generation, integrated gasification combined cycle (IGCC) coal-based power plants, which may be the wave of the future. • Feeding oxygen instead of air to aerobic waste-treatment processes, thereby reducing equipment size and costs. • Feeding oxygen to a large number of organic oxidation processes to improve selectivities and reduce energy costs. The savings would have to be larger than the capital and energy costs of producing the oxygen in order for these applications to be realized and grow. The secret to lowering oxygen costs would appear

OCR for page 89
 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING to be the discovery of new MSAs that can operate at temperatures and pressures that make it possible to use waste heat as the energy source. This strategy is already being tested in an IGCC pilot plant with a new technology—a ceramic membrane operating at about 800°C—which makes it possible to use waste heat from the furnace as the sole energy source for the air separation. The development of other MSAs that could operate in a similar fashion for other processes could usher in a revolution in oxygen use. Example 2: Separations of Chiral Compounds Chiral compounds are isomers whose structures cannot be superimposed on each other. Because they are similar in chemical composition to their mirror-image relatives, their thermodynamic properties such as vapor pressure, solubilities, and other properties are quite similar, making it often very difficult to separate them. The most common separation technique is to seed a melt of the two isomers with a crystal of one of them, causing that isomer to precipitate out of solution preferentially. Furthermore, it is almost always the case that, in nonbiological syntheses at least, chiral isomers are produced in equal amounts. Unfortunately, biological systems such as the human body do react differently to the two iso- mers, sometimes dramatically. For example, thalidomide, C13H10N2O4, a chiral isomer and a sedative, produced major birth defects when the product being sold contained more than a negligible amount of its chiral twin, which produced those defects. Nor is this an isolated problem. Naproxen, a popular pain reliever today, has a chiral twin that is a liver toxin. More and more, the fraction of new drugs coming on the market that are chiral is growing, and they must undergo precise and virtually complete separations from their chiral twins to eliminate the possibility that these twins might produce unfortunate side effects. What are needed are MSAs that can precisely separate chiral isomers and make it possible to produce drugs of the proper chiral purity much more easily and cheaply. However, even though separation systems that rely on the use of MSAs are an important area of growth in the separations industry, the design of new MSA systems is severely hindered by the lack of physical property data and novel design leads. MAJOR CHALLENGES FACING CHEMICAL SEPARATIONS There are three major challenges facing those concerned with the development of efficient chemical separations: 1. How can we predict physical properties accurately enough to set the optimal conditions for sepa- rating mixtures using distillation and MSA materials? 2. How can we design, construct, and mass produce MSAs with appropriately engineered three- dimensional structures (when appropriate) that make it easier to do difficult separations rapidly and efficiently? 3. How can we design overall separation systems that incorporate several individual separation units for economically optimal separations of complex mixtures? This list is based on several documents developed in recent years by the chemical separations commu- nity. An NRC report (1998) was used as the starting point. Reports from the Chemical Industry Vision 2020 Technology Partnership1 and another NRC report (2003), which examined the broader question of computational chemistry and materials science, provided insights into new challenges that are apparent 1See http://www.chemicalvision2020.org/library.html. Last accessed on July 25, 2008.

OCR for page 89
 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS now but were not at the forefront at the time of the 1998 NRC study. Finally, the committee benefited from presentations by Joan Brennecke (University of Notre Dame), Anne Chaka (National Institute of Standards and Technology), and Jeffery Siirola (Eastman Chemical) at a small workshop in December 2006 (see Appendix B). From these sources, the committee developed the three major challenges listed above. Dimensions, operating temperature, and operating pressure of the individual units are determined by vapor-liquid equilibrium data, the focus of Major Challenge 1, for both traditional distillation and for optimizing new MSA materials. Major Challenge 2 focuses on determining the appropriate nanoscale structures when MSAs are used and on the ability to predict interactions of the MSA components with the chemicals to be separated. Major Challenge 3 is an overall operations research problem that deals with optimizing the interplay of multiple separations processes to achieve high-performance separa- tions systems of complex mixtures. It is important to emphasize that the three major challenges pertain at very different spatial scales. Major Challenges 1 and 2 require a better understanding of molecular interactions within gases, liquids, and solids; Major Challenge 2 also deals with connecting particular nanoscale characteristics to the manufacturing of engineered separation materials with dimensions on the order of micrometers to meters. Finally, Major Challenge 3 addresses the design of systems on the scale of tens of meters. As progress is made on Major Challenges 1 to 3, we will be able to enlarge the space of options available for purification systems and make design decisions that are closer to optimal. Looking to the future, we will also create options for addressing critical separation technologies such as the following, for which no acceptable separation schemes currently exist: • Efficient recovery of highly dilute species from solutions. • Recovery of CO2 from stack gas and automobile exhaust and sequestration of this compound. • Development of less energy-intensive routes for producing oxygen. • Efficient removal of sodium and other inorganics from water. • Efficient separation of optical isomers to produce chirally pure products. The major challenges of chemical separations are driven by the demand for the capabilities they can enable. This is in contrast to the situation in astrophysics, evolutionary biology, and some aspects of atmospheric science, where the motivation for overcoming the major challenges is the gap in under- standing that must be filled to make scientific progress. The situation for chemical separations is much like that in operational meteorology because in both cases a capability already exists. Pushing the frontier amounts to improving and extending that capability, which in the case of chemical separations has staggering implications for our ability to assume prudent stewardship of our environment while maintaining our economic competitiveness. We now examine the three major challenges in chemical separations in more detail. Major Challenge 1: Accurately Predicting Physical Properties for Phase Equilibria How can we predict physical properties accurately enough to set the optimal conditions for separat- ing mixtures using distillation and MSA materials? Most current separations are equilibrium based. For example, when a feedstock undergoes a phase change, the two resulting phases typically possess different compositions, and a chemical separation has been achieved. In other cases, an MSA can be equilibrated with a single phase and the MSA phase can selectively remove certain chemicals out of the original phase. If the MSA is selective, what remains of the original phase will have a new chemical composition.

OCR for page 89
 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING Once again, a separation can be made. For systems such as these, the phase-equilibrium information is absolutely essential for determining the ease or difficulty of obtaining the desired separation and for predicting operating efficiencies and choosing among competing process designs. The field can utilize both data measured experimentally and data predicted from computational and theoretical chemistry methods that determine the energetics driving phase separations. Chemical separations is a field that benefits from combining experimental and computational approaches to the acquisition of physical property data, and this combined approach is anticipated to dominate for the foreseeable future. For easy separations of binary mixtures such as water and ethylene glycol, no great accuracy is required. But for mixtures that are not thermodynamically ideal, such as acetic acid and water or iso- propanol and water, for mixtures that have boiling point differences of 10°C or less, and for mixtures with three or more components, satisfactory definition of the phase equilibria can become an experimental nightmare, opening a primary role for computational predictions. Seldom do feed mixtures have just two components, and determining the appropriate mathematical representation of the experimental data escalates dramatically in difficulty as the number of components increases, which suggests that more research is needed for mathematical models and optimization. These predictions must be quite precise to determine the system specifications and the energy requirements for the separation. So why is industry not aggressively attacking Major Challenge 1? One reason is that quite a few of the separation processes used in the petrochemical industry (which itself is a large fraction of the chemi- cal industry) have historically been purchased from large engineering companies rather than developed in-house. Rather than pioneering a new process, the companies that operate these plants tend to think in terms of improving marketing, logistics, and supply chains as ways to differentiate themselves and increase their profitability. In addition, training and education in computational chemistry and math- ematical optimization are not well-integrated into the chemical engineering and chemistry curriculum, thus limiting the extent to which methods and algorithms can inform optimality of phase equilibria and process operation. However, it is likely that these accepted business practices will change in the near future if a green chemical revolution really takes hold. The separation requirements for a green product are normally greater than for a competing conventional product because of the greater complexity of green raw materials, the greater incidence of nonideal mixtures, and the desire to reduce the energy costs associ- ated with those nonideal mixtures. The advantage of being first in the green market will probably drive some companies to begin addressing Major Challenge 1 as a way of entering that market while control- ling costs. And once some companies have made that investment, the competitive landscape could shift quickly to favor the companies with stronger computational capabilities and resources. Major Challenge 2: Designing and Producing MSAs for Difficult Separations How can we design, construct, and mass produce MSAs with appropriately engineered three- dimensional structures (when appropriate) that make it easier to do difficult separations rapidly and effi- ciently? The chemical structure of an MSA will determine its physical properties, the nature and degree of interactions with other compounds, and, ultimately, its suitability for a given application. Accurate prediction of the properties of liquid MSAs used in extraction systems will allow engineers to design such systems. For the solid MSAs used in adsorption and membrane systems, the three-dimensional structure of the material is also of importance in determining its potential to accomplish the desired separation efficiently. We must also evaluate how amenable those potential MSA structures are to efficient and accurate production. The structures can be important not only because of their inherent thermodynamic equilibrium selectivity but also because of their various hydrodynamic and mass-transfer qualities, which

OCR for page 89
 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS can increase separation efficiencies. At present we have only a limited understanding of the relationship between structure and separation selectivity and efficiency. As an example of unexploited opportunities from new MSAs, consider metalorganic framework materials (MOFs), a class of materials that is currently attracting much attention for their selective adsorption of a vast array of solutes arising from the large number of possible MOF structures. MOFs have potential for use in many different applications, including gas storage, separations, catalysis, and sensors.2 MOFs have been shown to have some of the highest surface areas (for example, 4,900 m 2/g) reported for any material to date, making their separation capacities quite large. Experimental charac- terization of the most promising MOF materials—to understand how adsorption and diffusion of the materials to be separated correlate with structural features such as pore size, surface area, and void volume—is time consuming, and computational chemistry can suggest which novel structures are the most promising candidates for that experimental characterization. Computational chemistry in this case guides more rational design improvements for existing MSAs and de novo design of new MSAs for separations with less stringent accuracy requirements than those encompassed by Major Challenge 1. Major Challenge 3: Designing Optimal Separation Systems with Multiple Separation Units How can we design overall separation systems that incorporate several individual separation units for economically optimal separations of complex mixtures? Once a separation scheme has been proposed, determining the efficiency of the separation, the optimum operating conditions for each unit, and the sizing of the units shown and the connecting piping is rather straightforward if we have a good under- standing of the physical properties of the chemicals or mixtures and of the structures and performance of any MSAs used in the process. Yet, for any desired separation, a tremendously large number of pos- sible separation schemes exist, which might employ any combination of the processes described in Box 5-1. Determination of which processes to employ and in what order is the responsibility of the process engineer. The current state of the art relies heavily on the experience of the process engineer and on rough rules of thumb (for example, if distillation works, use it). The number of solutions to this problem is very large, with the best solution being influenced by value judgments on cost, waste produced, time required, safety, and so on. Major Challenge 3 consists of two related but different problems. In the determination of the opti- mum process system to be used to make a given product (or system of products), one must (1) determine all possible systems to be considered and then (2) evaluate which of the available solutions is optimal based on the design criteria. While both problems are difficult, the community’s progress in systematically attacking the second problem seems more advanced. Many chemical engineering groups in industry and academia are strong in computational modeling of processes to enable their design, optimization, and control. Once a pro- cess engineer has roughed out a proposed multiscale system, the tools and expertise exist to analyze its performance and then optimize the design. Unfortunately, before we can even talk about best solution for a desired chemical separation, we need a method for addressing the first problem, that of generating all the possible options to be evaluated. Currently, there is no clear algorithm for methodically surveying the space of design options, although the best process designers seem to have good intuition in this regard. For instance, an experienced process engineer can create multiple distinct processes for achieving some common separations, not all of which 2See, for example, Cho et al., 2006; Dinca et al., 2006; James, 2003; Kitagawa et al., 2004; Latroche et al., 2006; Matsuda et al., 2005; Millward and Yaghi, 2005; Mueller et al., 2006; Pan et al., 2006; Panella et al., 2006; Rosseinsky, 2004; Snurr et al., 2004; and Wang et al., 2002.

OCR for page 89
 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING would be obvious to a novice (or an algorithm). If an algorithm could be developed to methodically scan the design space—presumably drawing on a combination of expert knowledge, machine learning, experimental data, and computational simulations—an optimal design could in principle then be found through local or global optimization techniques if sound models for cost functions were defined. The number of dimensions in the design space and the complexity of some of the component simulations suggest that many such system optimization problems could require HECC. This is a long-term challenge for the industry and for computation. Little published research has appeared in this area, and the challenge remains at a very early stage of conceptualization. And there is, at present, little economic incentive to attack Major Challenge 3, because chemical companies are able to stay competitive without investing in new methods of process design. If appropriate algorithms can be developed that systematically create a library of options from which an optimum solution will be drawn, significant advances in the design and construction of efficient systems will surely be realized. POTENTIAL IMPACTS OF HECC FOR CHEMICAL SEPARATIONS There are numerous examples of computational chemistry leading to new understanding about the behavior of chemicals in separations systems. Examples are shown in Box 5-2. Computation can play the vital role of informing the experimental plan and focusing the expensive and time-consuming experi- ments on the precise set needed for a given design. Further, it has many potential advantages over the experimental evaluation of material properties, including these: • Safe determination of properties for chemical species that are highly toxic or highly reactive, or both. • Determination of properties for chemical species that have not yet been synthesized or purified. • Rapid prediction of properties for a wide range of chemical compounds. • Inexpensive determination of relative properties. • Fast screening of potential solvents or MSAs. Computational approaches have great potential for facilitating more progress on Major Challenges 1 and 2. Because simulations of molecules of industrial importance and of realistic systems are compu- tationally demanding, it is likely that HECC resources will be required. HECC enables more accurate predictions of properties, which can lead to gains in efficiency and cost, whether through more precise design of thermal-energy-based separations (for example, distillation) or the use of an appropriate MSA. It also can expand the range of chemical and parameter options being evaluated. For example, a profoundly improved MSA for the selective removal of oxygen from air is much more likely to be discovered using a computational approach rather than an experimental one. Typically, design of thermal-energy-based separations or of new MSAs is based on the characteristics of the best-known materials available to date. Modest changes to the known chemistry might bring small improvements in performance, but such incremental approaches will not bring true breakthrough technology. Break- throughs are likely to come when examining some entirely unexplored region of parameter space. HECC is ideally suited for mapping wide expanses of parameter space and highlighting potentially exciting regions. The HECC-developed map can then be used to direct the design of next-generation MSAs ripe for experimental evaluation. Major Challenge 3 would be critically dependent on HECC if the cost model and parameter space can be defined so that well-developed optimization and machine learning algorithms can be applied. But, as noted above, there is little movement to attack that challenge.

OCR for page 89
 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS BOX 5.2 Examples of Computational Chemistry Enabling Improved Understanding of the Behavior of Chemicals in Separations Systems • Phase behavior. Prediction of phase behavior of water and small organic molecules. This understanding is essential for the design of separation systems based on thermodynamic interactions, such as distillation and extraction. See, for example, Rablen et al., 1998. • Drug design. The design of new pharmaceuticals has been significantly impacted by advance- ments in computational chemistry. Today, computational chemistry can help predict the spe- cific structures of new pharmaceuticals based on the molecular properties of specific target molecules (Jorgensen, 2004). Additionally, it can be used to refine structures which have been discovered using experimental methods. This coupling of computation and experimentation can lead to new materials with superior properties (Martin et al., 1993). • Materials screening. Computational chemistry provides a method for fast and effective screen- ing methods for new drugs, catalysts, and materials. It allows for evaluation of a much broader phase-space than would be possible with experiments alone (Walters et al., 1998). • Materials design. Design of new solid structures capable of use as MSAs. See, for example, Lipkowitz (1998). • Design of microporous solids. The design of microporous solids with controlled pore size, vol- ume, and surface area is of tremendous importance in fields such as adsorption and catalysis. Férey et al. (2005) describe the use of targeted chemistry and computational design to create a crystal structure with very large pore size and surface area. • Industrial success stories. Westmoreland et al. (2002) cite the following as examples of notable industrial successes in the use of computational chemistry: — Rhône-Poulenc used quantum mechanical calculations of a Flory χ-parameter and relative reactivities in developing an antiscratch additive for polyurethane coatings. — Rhône-Poulenc used computation to determine that it would not be possible to develop a material to compete with its competitor using a nylon basis, a valuable negative result. — Lubrizol used a QSPR model for gasoline additive formulation to reduce testing costs by one-third for predicting intake valve deposits in BMW, Ford, and Honda engines. — Dow estimated that each ∆Hf calculation saved the company $50,000 in testing costs in 1996 and over $100,000 in 2000. — Mitsubishi Chemicals reports that 5 percent of the patents from its Yokohama facility involve some computational modeling. To successfully address Major Challenges 1 and 2 requires building on the capabilities of computa- tional chemistry, which now include calculations at the molecular scale with algorithms based on quan- tum chemical theories and classical analogs that evaluate the energetics of molecular conformations, as well as statistical mechanical methods that sample those conformations consistent with thermodynamic variables such as temperature and pressure. The general strategy typically employed in computational chemistry is to combine these methods based on the following diagram:

OCR for page 89
 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING Quantum chemistry calculations ‡ Classical empirical force fields Ê Â Statistical mechanical sampling ‚ Phase diagram The primary uses of quantum chemistry calculations are to calculate for model chemical compounds the relative energies of different molecular conformations—the charge density descriptions from which one derives partial atomic charges—and the intermolecular interaction energies. All of these calculated quantities can then be used to develop the empirical force fields, which can approximate the clas- sical forces on all the atoms or molecules being studied. Because this force-field calculation is less expensive computationally than a full-fledged quantum calculation of the entire system, it is the more feasible approach for larger and more complicated systems. Furthermore, it is assumed that the quantum mechanical results for a given compound are still valid when that compound is part of a larger molecule, although it is recognized that this is not always a good approximation. This problem of transferability is most limiting when electronic structure calculations are transformed into empirical classical force fields. In either case, once the quantum or classical results have been validated against experimental data, the resulting energetic models of an MSA material or of the chemicals in a thermal-energy-based unit process feed into statistical mechanical simulations. Statistical mechanics provides a solid theoretical foundation for defining equilibrium and dynamical sampling schemes of these molecular conformations, thus allowing the generation of a global minimum structure, a phase diagram, absorption probabilities, or transport properties such as diffusion, all of which are needed by the engineer or scientist intent on developing new chemical separation schemes. A sweet spot for such methods at present is when qualitative predictions suffice for identifying phase equilibria thermodynamic parameters or promising MSAs to investigate experimentally. For example, using molecular simulations, the nitrogen adsorption preferences within selected MOF materials known as IRMOF-1 and IRMOF-16, shown in Figure 5-1, were predicted. The calculations predicted that nitrogen prefers to associate with only the corner regions of IRMOF-1, while for IRMOF-16 it associates with not only the corners but also the faces of the benzene rings. Thus, experimental efforts would be steered toward IRMOF-16 because it is predicted to have greater nitrogen adsorption rates and capaci- ties. When successfully executed, such computational modeling can direct experimental programs so that highly effective MSAs can be produced with a minimum of time-consuming experimentation. In other cases, qualitative insight is not enough and quantitative predictions are necessary. In order for computational chemistry to develop predictive capabilities good enough to overcome Major Chal- lenges 1 and 2, the following are needed: • Scalable algorithms for quantum electronic structure calculations. • Greatly improved classical force-field accuracy. • Improved statistical sampling via molecular dynamics and Monte Carlo methods. • Extensive validation studies on resulting phase equilibria and MSA structures. • Training and education of the next generation of computational chemists and chemical engineers.

OCR for page 89
 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS Fig. 5.1.eps FIGURE 5-1 Grand Canonical Monte Carlo simulation at 77 K of the adsorption of nitrogen within the unit cell of two MOF materials. IRMOF-1 (top) is represented by metal atoms at the corners of the unit cell connected by a -linker. IRMOF-16 (bottom) consists of metal atoms at the corners of the unit cell connected by a -linker. The calculations show that the different MOF materials have different capacities for ni- trogen separation depending on the linker chemistries. While nitrogen is concentrated only at the metal sites of IRMOF-1, it can absorb at both the metal sites and linker aromatic rings in IRMOF-16. These calculations suggest further MSA designs without costly experimentation. Red is higher density—that is, the molecule was found at that location a relatively large number of times over the course of the simulation—with orange denoting a lower density, yellow being lower still, and blue signifying that no molecules were predicted at those locations. CURRENT FRONTIERS OF HECC FOR CHEMICAL SEPARATIONS Algorithms for Quantum Electronic Structure Calculations The 1998 Nobel prize in chemistry went to John Pople for his development of computational methods in quantum chemistry, including the mean field approximation of Hartree-Fock (HF) methods and electron correlation methods that enable increasing levels of accuracy, and to Walter Kohn for his development of an alternative approach to electronic structure, known as density-functional theory (DFT). The Nobel prize press release emphasized that “as well as producing quantitative information

OCR for page 89
00 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING on molecules and their interactions, [computational chemistry] also affords deeper understanding of molecular processes that cannot be obtained from experiments alone.” 3 Quantum chemistry is based on reformulating the Schrödinger eigenvalue equation into a large set of algebraic equations expanded in some convenient mathematical basis set—typically Gaussianfunctions— and the development of well-defined approximations to electron-electron interaction potentials. HF uses a mean-field approximation to treat electron interactions, such that the computational complexity scales quadratically with molecular size, although the algebraic steps of matrix diagonalization scales cubically to dominate the computational complexity for a large system. DFT offers the advantage of similar computational scaling complexity while treating electron correlation beyond the HF mean-field approximation. It is important to emphasize that because DFT captures only certain types of electron correlation, the quality of DFT calculations is still under debate. This is especially the case for weak nonbonded interactions, and the development of new DFT functionals is an active area of research. Routine calculations for these methods are now completely feasible for molecules with hundreds of atoms, and heroic calculations for ~1,000 atoms are possible on the most powerful computers and with a good deal of computing time. A feasible and often more robust alternative to post-HF methods is the Moller-Plesset Perturbation (MP2) series to describe electron correlation beyond the mean-field HF reference. MP2 refers to the mathematical model that perturbs the HF reference to include electron correlations up to second order. The MP2 method scales with the 5th power of system size because the formulation of MP2 uses delocalized molecular orbitals that arise from standard HF calculations. However, the molecular orbitals can be localized, and there has been a great deal of progress toward developing a “local-MP2” method that scales only quadratically with molecular size and comes to within a few percent of reproducing the exact MP2 energy for a given basis set, making the computation of molecules comprising hundreds of atoms completely feasible. The gold standard of quantum chemistry calculations is coupled cluster methods, a general formula- tion with high levels of electron correlation that can use any orbital reference. While these theoretical models have been formulated into algorithms, they have severe scaling requirements (scaling at least with the 7th power of system size), which have traditionally limited their applications to very small system sizes (tens of atoms). The post-HF methods provide a good to very good level of accuracy with regard to relative confor- mational stabilities and barriers, charge densities, and weak intermolecular interactions. They provide excellent input for developing empirical force fields for many classes of chemical compounds. HECC will make it less expensive to perform electronic structure calculations and will enable the calculation for much larger molecules of importance when the physics is well described by HF/DFT or MP2 levels of theory, on the condition that these algorithms can be deployed on massively parallel architectures, which is a limiting factor since the algorithms are still only weakly parallelizable. For classes of more complex materials, current capabilities of these methods may themselves inherently limit the accurate calculation of phase equilibria data, and coupled cluster methods are to be preferred. Improved Accuracy of Molecular Mechanics Force Fields Empirical force fields derived from electronic structure calculations and experimental data, coupled to classical molecular dynamics or Monte Carlo sampling schemes, are the main component of all computational studies of materials chemistry to date. Overall, they can be the weak link in accurate 3Available at http://nobelprize.org/nobel_prizes/chemistry/laureates/1998/press.html.

OCR for page 89
0 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS determination of phase equilibria data through computation if they have not been sufficiently validated against experiment. For example, the physical properties of homogeneous liquids of certain classes of molecules, such as Lennard-Jones fluids or liquid water, are qualitatively and even quantitatively well described by empirical molecular mechanics force fields based on productive collaborative work between theory and experiment over the last several decades. The challenge of using empirical force fields in the chemical separations industry is that each new thermal-based separation or MSA material requires empirical force field development based on an affordable electronic structure calculation, and experimental validation is then required in order for those force fields to be usefully deployed. The challenge in constructing a force field from quantum calculations lies in determining the form of the mathematical function. One must strike a balance between enough complexity to accurately describe the fundamental interactions of matter on the one hand and simplifications that decrease the computational complexity on the other. For instance, approximate classical models represent bonds and angles as harmonic springs, dihedral angle conformations by a truncated Fourier series, pairwise non- bonded interactions with a Lennard-Jones function, and electrostatic interactions between point charges by Coulomb’s law. There are several empirical force fields of this type in use, and they are widely used in industry and academic research settings. Beyond the less-accurate two-body potentials described above, the most recent generation of empirical energy functions incorporates the many-body effects of polarizability by modeling how the electron density responds to an electric field that is generated by the condensed phase of a material of interest. It is generally agreed that including polarizability into empirical force fields is necessary for good quantitative agreement between simulations and experiments not at ambient conditions, for representing realistic dynamics, and for simulating heterogeneous chemical systems of multiple compo- nents. These many-body functional forms are typically more computationally expensive (3 to 10 times as expensive as the simplest molecular mechanics force fields), making HECC even more necessary. While the most recent generation of polarizable empirical energy functions provides some significant improvement, important work remains in how to model the physics of charge transfer between separate molecules and how to describe polarization anisotropy in fluctuating charge models. There are also algorithmic issues to be addressed to achieve computational efficiency in Monte Carlo simulations and extensions to arbitrary molecular systems. However, the most fundamental expense in evaluating empirical force-field energies and deriva- tives is due to the long-range coulombic forces. The accounting of long-range forces is best introduced through the Ewald summation. Typical materials simulations periodically replicate the system in three spatial dimensions, and this approach divides the long-range coulombic interactions into a short-range part that is evaluated in real space (as a direct sum over atomic positions) and a long-range part evalu- ated in reciprocal space. New formulations of Ewald algorithms scale as N log N once N exceeds about 1,000, so systems with tens of thousands of atoms may reasonably be handled on the most advanced supercomputers. Looking to the future, the new ab initio molecular dynamics approaches allow calculation of elec- tronic structure on the fly, currently to an accuracy competitive with that of HF or DFT. Even though these methods are in their infancy and are not feasible for the long timescales and large molecular sizes that are needed for useful empirical force-field calculations, this capability will continue to grow over the next several decades. The bottlenecks for this area are primarily model physics (greater accuracy than that provided by HF and DFT), improved algorithms, and deployment on massively parallel architectures.

OCR for page 89
0 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING Improved Statistical Sampling Phase equilibria calculations involve direct evaluations of thermodynamic properties of the indi- vidual phases at a series of state points to find where the temperature, pressure, and chemical potential of the phases are equal. Because the direct computation of conformational and energetic properties is so computationally expensive, as explained above, it is critical to be as efficient as possible in sampling the state points. Addressing that need, extended system equations of motion and associated numerical integrators have been developed that allow extensions from microcanonical ensemble dynamics to sam- pling of states in the canonical ensemble (NVT) as well as the isobaric-isothermal (NPT) ensembles. Based on a factorization of the evolution operator, a formal decomposition of the integration time step allows bonds to be updated more often than angle bends, and angle bends more often than short-range forces, and short-range forces more often than long-range forces. This formally correct multiple-time- step integration has been shown to generate about an order of magnitude improvement in computational efficiency in materials systems, although resonance artifacts can reduce this efficiency gain in practice. The decrease in computer time results from the fact that the most expensive terms, the double sum over atoms, need to be updated less often than local interactions. Calculations performed using multiple- time-step integration methods in isothermal or isobaric-isothermal ensembles are very scalable. Each time step results in a collective “move,” and parallelization can proceed using standard domain decom- position paradigms. These are important considerations for phase equilibria calculations since large systems are required to overcome finite-size effects and heterogeneous systems require significantly longer equilibration times. Probably the biggest breakthroughs in calculation of vapor-liquid and liquid-liquid phase equilibria are the formulations of grand canonical Monte Carlo methods in terms of the Gibbs ensemble and semi- grand canonical Monte Carlo methods developed in the 1980s and 1990s, allowing the determination of phase equilibria in one simulation without the interference of an imposed phase interface. Furthermore, once a single point on the coexistence curve is known, the rest of the curve can be calculated without resort to additional free-energy calculations by integrating the Clausius-Clapeyron equation, although care must be taken to avoid numerical instabilities. For solid-liquid phase equilibria, thermodynamic integration based on paths with a known free energy reference state are well developed. Enhanced sam- pling schemes and related methods are available that allow for efficient molecule exchanges between phases in order to converge the Gibbs ensemble. Progress on Major Challenge 3 involves a very different focus, namely mathematical optimiza- tion research as a general approach for obtaining solutions to large nonlinear systems with numerous local minima. Constrained optimization methods rely on the availability of sufficiently well-defined constraints (supplied by the application expert) so that the desired solution is the only available mini- mum, or one of few available minima, in the optimization phase of the algorithm. Alternatively, global optimization techniques attempt to systematically search the parameter space based on a cost function to find all low-lying minima, including the global energy minimum. The useful application of these optimization strategies is computationally intensive since they typically require hundreds or thousands of evaluations of a cost function (and of its derivative if available). These optimization approaches are useful in many contexts, including atomic-level structure optimization of molecules in thermal-based separations and MSA materials in Major Challenges 1 and 2 and designs for entire separation systems that define Major Challenge 3. What distinguishes the usefulness of mathematical optimization in Major Challenges 1 and 2 is that the cost functions and parameter space are relatively well defined in terms of an objective function, while Major Challenge 3 has greater uncertainty in the nature and dimension size of the mathematical model.

OCR for page 89
0 THE POTENTIAL IMPACT OF HECC IN CHEMICAL SEPARATIONS In summary, our current capabilities in computational chemistry for addressing Major Challenges 1 and 2 are sufficient for a class of materials and chemical systems important for some separations prob- lems, and improvements in HECC, broadly defined, will surely and steadily enable additional advances for these application areas, especially for larger molecular systems. However, advancing these methods to new classes of materials will require a combination of new model physics, better-scaling algorithms in quantum chemistry and statistical mechanical sampling, and deployment onto massively parallel architectures. Major Challenge 3 currently is more narrowly focused on formulating cost function models that can utilize the large array of mathematical optimization techniques. OTHER ISSUES THAT LIMIT THE VALUE OF HECC TO CHEMICAL SEPARATIONS Productive cooperation and dialogue between experimentalists and modelers is not as extensive as it should be in order for computational approaches to contribute optimally to progress in chemical separations. In particular, funding is lacking for experimental work to learn about phase equilibria in fundamental systems, knowledge that could be used to validate computational models. A combined computational/experimental strategy is critical. A good example of cooperation is the Industrial Fluid Simulation Challenge, sponsored by the National Institute of Standards and Technology (NIST), in which academic and industrial teams attempt to predict a range of thermodynamic and physical properties like vapor-liquid equilibria, density, viscosity, vapor pressure, heat of mixing, and so on. However, after four NIST Challenges, it is clear that there is a long way to go before computation can reliably predict various properties for a disparate set of chemical species. Incentives are needed to encourage collaboration between experimentalists and researchers perform- ing molecular simulations in order that computational models can be developed, validated, and run more efficiently. Within the research community in general, not much effort is being put into validating the results of atomistic scale models with experimental data. Part of the problem is that experimentalists have incentives to pursue experiments that are project-specific rather than those that will expand fundamental knowledge. Indeed, the measurements that would be most helpful in developing and verifying compu- tational methods are often perceived as having little practical value for the experimentalist or funding agency. Cooperation between industry, academia, and government could create the needed incentives. As noted in the preceding section, education and training are important if the chemical separations field is to profit from the potential of HECC. Developing and applying advanced simulation capabilities requires specialized cross-disciplinary skills; this topic is addressed in more detail in Chapter 7 because it affects nearly every field that relies on computational science and engineering. In the case of chemi- cal separations in particular, these broad computational skills must be supported on the foundations of theoretical chemistry and mathematics and are vital to overcoming all three challenges explored in this chapter. REFERENCES Cho, S.H., B.Q. Ma, S.T. Nguyen, J.T., Hupp, and T.E. Albrecht-Schmitt. 2006. A metal-organic framework material that func- tions as an enantioselective catalyst for olefin epoxidation. Chemical Communications 24: 2563. Department of Energy. 2005. Hybrid Separations/Distillation Technology: Research Opportunities for Energy and Emissions Reduction. DOE Industrial Technologies Program, Washington, D.C., April. Dinca M., A.F. Yu, and J.R. Long. 2006. Microporous metal-organic frameworks incorporating 1,4-benzeneditetrazolate: Syntheses, structures, and hydrogen storage properties. Journal of the American Chemical Society 128: 8904. Férey, G., C. Mellot-Draznieks, C. Serre, F. Millange, J. Dutour, S. Surblé, and I. Margiolaki. 2005. A chromium terephthalate- based solid with unusually large pore volumes and surface area. Science 309 (5743): 2040-2042.

OCR for page 89
0 THE POTENTIAL IMPACT OF HIGH-END CAPABILITY COMPUTING Humphrey, J.L., and G.E. Keller. 1997. Separation Process Technology, New York, N.Y.: McGraw-Hill. James, S.L. 2003. Metal-organic frameworks. Chemical Society Reiews 32: 276. Jorgensen, W.L. 2004. The many roles of computation in drug discovery. Science 303 (5665): 1813-1818. Kitagawa, S., R. Kitaura, and S.Noro. 2004. Functional porous coordination polymers. Angewandte Chemie International Edition 43: 2334. Latroche, M., S. Surble, C. Serre, C. Mellot-Draznieks, P.L. Llewellyn, J.H. Lee, J.S. Chang, S.H. Jhung, and G. Ferey. 2006. Hydrogen storage in the giant-pore metal-organic frameworks MIL-100 and MIL-101. Angewandte Chemie International Edition 45: 8227. Lipkowitz, K.B. 1998. Applications of computational chemistry to the study of cyclodextrins. Chemical Reiews 98: 1829- 1873. Martin, Y.C., M.G. Bures, E.A. Danaher, J. Delazzer, I. Lico, and P.A. Pavlik. 1993. A fast new approach to pharmacophore mapping and its application to dopaminergic and benzodiazepine agonists. Journal of Computer-Aided Molecular Design 7 (1) :83-102. Matsuda, R., R. Kitaura, S. Kitagawa, Y. Kubota, R.V. Belosludov, T.C. Kobayashi, H. Sakamoto, T. Chiba, M. Takata, Y. Kawazoe, and Y. Mita. 2005. Highly controlled acetylene accommodation in a metal-organic microporous material. Nature 436 :238. Millward, A.R., and O.M. Yaghi. 2005. Metal-organic frameworks with exceptionally high capacity for storage of carbon dioxide at room temperature. Journal of the American Chemical Society 127: 17998. Mueller, U., M. Schubert, F. Teich, H. Puetter, K. Schierle-Arndt, and J. Pastré. 2006. Metal-organic frameworks—prospective industrial applications. Journal of Materials Chemistry 16: 626. NRC (National Research Council). 1998. Separation Technologies for the Industries of the Future. Washington, D.C.: National Academy Press. NRC. 2000. Alternaties for High-Leel Waste Salt Processing and the Saannah Rier Site. Washington, D.C.: National Academy Press. Pan, L., D.H. Olson, L.R. Ciemnolonski, R. Heddy, and J. Li. 2006. Separation of hydrocarbons with a microporous metal- organic framework. Angewandte Chemie International Edition 45: 616. Panella, B., M. Hirscher, H. Pütter, and U. Müller. 2006. Hydrogen adsorption in metal-organic frameworks: Cu-MOFs and Zn-MOFs compared. Adanced Functional Materials 16: 520. Rablen, P.R., J.W. Lockman, and W.L. Jorgensen. 1998. Ab initio study of hydrogen-bonded complexes of small organic molecules with water. Journal of Physical Chemistry A 102: 3782-3797. Rosseinsky, M.J. 2004. Recent developments in metal-organic framework chemistry: Design, discovery, permanent porosity and flexibility. Microporous Mesoporous Materials 73: 15. Snurr, R.Q., J.T. Hupp, and S.T. Nguyen. 2004. Prospects for nanoporous metal-organic materials in advanced separations processes. American Institute of Chemical Engineers Journal 50: 1090. Walters, W.P., M.T. Stahl, and M.A. Murcko. 1998. Virtual screening—An overview. Drug Discoery Today 3(4): 160-178. Wang, Q.M., D. Shen, M. Bülow, M.L. Lau, S. Deng, F.R. Fitch, N.O. Lemcoff, and J. Semanscin. 2002. Metallo-organic molecular sieve for gas separation and purification. Microporous Mesoporous Materials 55: 217. Westmoreland, P.R., P.A. Kollman, A.M. Chaka, P.T. Cummings, K. Morokuma, M. Neurock, E.B. Stechel, and P. Vashishta. 2002. Applications of Molecular and Materials Modeling. World Technology Evaluation Center Panel Report. Available at http://www.wtec.org/loyola/molmodel/mm_final.pdf.