Citation |

- Permanent Link:
- http://ufdc.ufl.edu/UF00091523/00670
## Material Information- Title:
- Monte Carlo Simulation of Long-Range Self-Diffusion in Model Porous Membranes and Catalysts
- Series Title:
- Journal of Undergraduate Research
- Creator:
- DeCost, Brian
Vasenkov, Sergey - Place of Publication:
- Gainesville, Fla.
- Publisher:
- University of Florida
- Publication Date:
- 2013
- Language:
- English
## Subjects- Subjects / Keywords:
- Diffusion
Zeolites Porous media Simulation Mass transfer - Genre:
- serial ( sobekcm )
## Notes- Abstract:
- Industrial processes involving the use of micro-porous materials are very much dependent upon understanding the processes by which mass transport occurs in those materials. Porous catalysts and membranes are typically characterized by complex pore structures, involving pores of multiple length scales. Diffusion inside of systems of ordered nanoporous crystals (e.g., zeolites) has been studied extensively, both experimentally and computationally. However, transport in systems containing microporous particles that are separated by large meso- and/or macropores has not been sufficiently characterized. This study focuses on molecular diffusion in such systems by using dynamic Monte Carlo simulation. Simulations were conducted at a state of sorption equilibrium between the adsorbed molecules (molecules inside zeolite micropores) and molecules in the surrounding gas phase (molecules in zeolite bed mesopores and macropores). Long range diffusion in this study refers to the condition that the vast majority of diffusant molecules have traveled large distances in comparison to the size of a single zeolite crystal. The effect of varied potential barrier heights at the interface between micropores and larger pores was found to be significant in terms of its effects on the bulk diffusivity in the model zeolite system, and is reported in this study. The data was obtained using simulation parameters that are analogous to real systems of zeolites. Future comparison of the simulation data with experimental studies promises to enrich the interpretation of the experimental results.
## Record Information- Source Institution:
- University of Florida
- Holding Location:
- University of Florida
- Rights Management:
- All applicable rights reserved by the source institution and holding location.
- Resource Identifier:
- UF00091523_00602 ( sobekcm )
## UFDC Membership |

Downloads |

## This item is only available as the following downloads: |

Full Text |

PAGE 1 University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 1 Monte Carlo Simulation of Long Range Self Diffusion in Model Porous Membranes and Catalysts Brian DeCost and Dr. Sergey Vasenkov College of Engineering, University of Florida Industrial processes involving the use of micro porous materials are very much dependent upon understanding the processes by which mass transport occurs in those materials. Porous catalysts and membranes are typically characterized by complex pore structures, involving pores of multiple length scales. Diffusion inside of systems of ordered nanoporous crystals (e.g. zeolites) has been studied extensively, both experimentally and computationally. However, transport in systems containing microporous p articles that are separated by large meso and/or macropores has not been sufficiently characterized. This study focuses on molecular diffusion in such systems by using dynamic Monte Carlo simulation. Simulations were conducted at a state of sorption equil ibrium between the adsorbed molecules (molecules inside zeolite micropores) and molecules in the surrounding gas phase (molecules in zeolite bed mesopores and macropores). Long range diffusion in this study refers to the condition that the vast majority of diffusant molecules have traveled large distances in comparison to the size of a single zeolite crystal. The effect of varied potential barrier h eights at the interface between micropores and larger pores was found to be significant in terms of its effect s on the bulk diffusivity in the model zeolite system, and is reported in this study. The data was obtained using simulation parameters that are analogous to real systems of zeolites. Future comparison of the simulation data with experimental studies prom ises to enrich the interpretation of the experimental results. INTRODUCTION Diffusion is the dispersal of molecules due to thermal motion. The study of diffusion in porous media is important due to its relevance to industrial applications of chemical separations and heterogeneous catalysis. The d ifferent length scales and complex geometries of real porous media have a non trivial effect on the diffusion of gases in and through these media 1 Consequently, advances in industrial use of these materials are contingent upon the development of theoretical knowledge of diffusion in them. S everal experimental techniques have been developed to this end. Additionally, computer modeling techniques have been used to allow researchers to explore the nature of diffusion in porous materials in ways not possible with physical experiments. The objective of this study is to extend the modeling of diffusion inside and near a single zeolite crystal to diffusion in model membranes and catalyst s consisting of many zeolite crystals separated by mesopores and macropores namely the construction of a model of long range diffusion in porous materials and zeolite beds. Of particular importance is maintaining the context of the microstructural complexities and their influence on the overall diffusivity of gases in the material. BACK GROUND Diffusion in zeolite beds and other closely related systems is a complex process involving many physical interactions, and is generally dependent on physical conditions such as pressure, temperature, the chemical species involved, and their concentr ations 2 However, diffusion in these systems can be approximated by two regions with differing properties: the space inside the microporous crystals and the space surrounding these crystals 3 Equation 1 below provides a simple way to estimate the long range diffusivity (i.e. diffusivity for displacements much larger than the size of individual crystal) as the sum of the weighted averag es of the diffusivities in the inter crystalline and intracrystalline regions. (1) In many cases only the intercrystalline fraction is considered in the long range diffusivity because the intracrystalline term is generally smaller than the intercrystalline ter m 4 This is the case because of interactions between sorbate diffusant molecules and the pore walls lower the diffusivity of these intracrystalline diffusant molecules. This adds to the effect on the diffusivity that the confinement to the interior of the pore s has In addition to this effect, there is generally a large potential barrier for desorbing molecules at the interface of the crystal surface and the gas phase, with a much lower potential energy for molecules inside the pore structures than for molec ules in the gas space between crystals. This PAGE 2 BRIAN DECOST & DR SERGEY VASENKOV University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 2 results in a large majority of molecules in the system being adsorbed on the por e surfaces in the crystals. SIMULATION METHODOLO GY Overview Two dimensional Monte Carlo simulations are used to model gas tran sport in a system of porous crystals separated by pores that have sizes comparable with the crystal size (a zeolite bed, for example). Intracrystalline diffusion is modeled as a traditional random walk algorithm, while molecular diffusion in the gas phase is modeled as a continuous motion interrupted by molecular collisions. If a diffusant molecule crosses a crystal interface, that molecule is adsorbed or desorbed with the probability P a or P d respectively. The ratio of these probabilities is analogous to the interfacial potential barrier; thus it is assumed that P d is less than P a due to the lower potential of diffusant molecules inside of the crystals. The model potential barrier in this sim ulation is smaller than what might be expected in a real system of porous crystals to avoid simulation times that are prohibitively long. The ratio of these probabilities is chosen in order to achieve a desired distribution of molecules between phases; in this study the gas fraction is set to 0.25. The initial state of the simulation is a random distribution of labeled molecules such that the concentrations in each phase match their calculated equilibrium values. The simulation then proceeds to reach a stea dy state where there is no net interfacial mass flux. Displacement data for each diffusant molecule is retained in order to calculate the overall diffusivity at selected times during the simulation. Simplifying Assumptions The simulation in this study is b ased upon previous modeling work done by Robert Mueller 5 The assumptions that form the basis for this model are laid out in a 2008 study by Krutyeva and Krger 3 Further insight into the structure of the model was given by Dr. Sergey Vasenkov and Ph.D. candidate Robert Mueller. Specifically, Krutyeva and Krger show that a simple square lattice is sufficient to model diffusion in porous systems, but is still computationally expedient. There is no significant qualitative loss of generality associated with the reduction in dimensionality because of the isotropic nature of diffusion in the system Finally, the Monte Carlo technique implicit ly assumes that particles do not interact with each other which is physically acceptable for the case of low gas molecule concentrations. Simulation Geometry Figure 1 depicts the two dimensional model system. The simulation box has length L and contains four model crystals. This simulation lattice is based upon a simulation lattice used in 2009 by Sanders and Mueller, et al. 5 6 T he crystals are arranged in the simulation box in such a way that diagonal symmetry is achieved in accordance with the assumption of an isotropic medium. Periodic boundary conditions are applied at each edge of the simulation box, resulting in a large syst em size in comparison to the simulation box size, as shown in Figure 1 Figure 1 Schematic of the simulation box, including periodic boundary conditions Figure 2 shows the detailed layout of the simulation box. Each crystal is a square of characteristic length L c c where is the random walk step length for molecules inside of a crystal. Each crystal is a minimum distance of L sep c from the next nearest crystal, and the simulation box is laid out in such a way that trajectories that might bypass many crystals are minimized. PAGE 3 MONTE C ARLO SIMULATION OF LONG RANGE SELF D IFFUSION IN MODEL POROUS MEMBRAN ES AND C ATALYSTS University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 3 Figure 2 Simulation box details A flux balance at the crystal external surface is used to calculate the ratio of adsorption to desorption probabilities required to achieve a desired distribution of diffusant molecules between the adsorbed and gas phases. Because of the transition between separate time and length scales at the crystal boundaries, the concentration of the diffusant molecules near the interface is different from the concentration further away from the interface. T he concentration gradient was determined by a parametric fit o f the simulation data and used in the flux balance to determine how to calculate the probability ratio so that a desired gas fraction is attained at steady state as follows: (2) and are the probabilities for molecules encounter ing crystal gas interfaces to be adsorbed or to desorb, respectively. is the random walk step length for molecules inside of crystals and are the desired adsorbed and gas phase bulk concentrations, respectively, and is a function of which was determined using simulation data. Actual gas fractions achieved through the use of this relation vary by 10% compared to the desired gas fraction used in the calculation (see Results and Discussion Section ). Simulation Dynamics This simulation uses a random walk algorithm to model diffusion in a similar manner to previous versions of the simulation. The key difference between this model and previous versi ons is the use of separate random walk algorithms for the gas phase and the sorbate phase, as shown schematically in Figure 3. This section outlines both of these algorithms, as well as the overall dynamics of the system. Figure 3 Schematic depiction of both random walk algorithms, as well as adsorption and desorption subroutines. D iffusion in the zeolite micropores is modeled as a classical random walk; the simulation lattice represents the pore structure of the material. At each iteration of the simulation, sorbate molecules are randomly propagated along the simulation lattice in one of four directions, with a c the simulation lattice constant. This behavior is i llustrated in the bottom left part of Figure 3. In the gas phase, diffusant molecules may occupy any position, analogous to continuous movement of real gases. At each iteration of the simulation, gas phase molecules g at a randomly chosen is shown in Figure 3. This diffusion step is roughly analogous to a mean free path, and is set in order to achieve the desired gas phase diffusivity. This modified random w alk algorithm for the gas phase diffusant molecules affords a more realistic model of diffusion at the same time that it reduces the required number of computations in comparison with the use of the classical random walk algorithm. In prior studies the sa me random walk was used for gas molecules as for sorbate molecules, requiring ten calls to the random number generator per gas molecule for each iteration. The modified random walk algorithm reduces this to a single call to the random number generator per gas molecule for each iteration, allowing more efficient modeling and larger gas fractions to be studied. In the overview of the methods it is mentioned that the interfacial potential barrier is modeled by probabilities that sorbate or gas molecules that encounter a crystal external PAGE 4 BRIAN DECOST & DR SERGEY VASENKOV University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 4 surface will desorb or be adsorbed, respectively. These probabilities are set in such a way that the desired concentration ratio is achieved. A random number is generated s uch that and if is greater than the probability for adsorption or desorption, then the adsorption or desorption event is rejected. R ejected desorption events are handled by simply subject ing the molecule to the sorbate random walk algorithm again. R ejected adsorption events are handled by reflect ion from the surface according to a cosine squared distribution, calculated according 7 to Equation 3: (3) where is an angle w.r.t. the normal and is a randomly generated number such that Generally, gas g when they encounter a crystal surface and are reflected; they travel the g in the reflected direction described by the angle theta. Molecules that desorb are also processed according to the cosine squared distribution of g Finally, gas molecules that are adsorbed do so at th e nearest simulation lattice point to the intersection with the crystal surface interface. Molecules that adsorb are then c ) towards the interior of the crystal in accordance with the high potential barrier at the crystal surfac e. Initial Conditions The system is initialized in a state close to the expected equilibrium distribution of molecules between phases. Molecules are randomly distributed in the gas phase only until the target gas phase concentration is reached, at which po int the remaining molecules are randomly distributed in each of the four simulation box crystals. For most of the simulation runs in this study, the ratio of adsorbed molecules to gas phase molecules is 3 to 1. In Mueller's simulations, molecules that are originally assigned in crystals are labeled as such, in analogy to tracer exchange experiments 5 The particular crystal into which a termed the parent of that molecule These tracer exchange features are p reserved in this study, and in S ection 4 will be used to validate the modifications that have been made to the model. Summary of Simulation Parameters Table 1 quantifies the simulation geometry and other model parameters that were used in obtaining the results presented in the next section Table 1 Summary of Simulation Parameters Simulation Parameter Symbol Value Crystal Mean Free Path c Gas Phase Mean Free Path c Simulation Lattice Length c Crystal Side Length c Min. Inter crystal Distance c Initial Number of Gas Molecules 25000 Initial Number of Adsorbed Molecules 75000 Adsorption Probability 0.01 0.75 Desorption Probability Probability Ratio 1.10528992 PAGE 5 MONTE C ARLO SIMULATION OF LONG RANGE SELF D IFFUSION IN MODEL POROUS MEMBRAN ES AND C ATALYSTS University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 5 RESULTS AND DISCUSSI ON Overview The main focus of this study is to elucidate the manner in which the existence of a hierarchy of pore sizes in porous materials affects the diffusion of gases on a macroscopic length scale through these materials. The results show that the diffusivity in s ystems of porous crystals depends on the interfacial potential between the microporous crystals and the gas phase (i.e. large pores). The simulation results are compared with the behavior given by Eq. 1 that assumes that there are no correlations between t he diffusion in the adsorbed and gas phases. In analyzing the diffusivity values obtained through the system, the root mean square displacement (RMSD) is nondimensionalized by dividing by the diagonal size of an individual crystal and used as a metric for relative simulation time. It is important to note that all the results reported have primarily qualitative significance due to the impact of the simulation parameters. Because the simulation parameters do not necessarily reflect actual physical values, but rather relative models of physical parameters, the results are also relative in nature. Despite this limitation, much insight into the nature of diffusion in systems of porous crystals may be gained through the use of this model. Diffusivity as a Function of Potential Barrier Figure 4 shows the dependence of overall diffusivity on the nondimensionalized Root Mean Square Displacement (RMSD) for varying potential barrier heights. I n order to achieve a gas fraction of 0.25 the ratio of P a to P d is assigned the constant value of 1.10528992 A steady state long range diffusion is seen as the RMSD becomes greater than the size of a single crystal. The diffusivity varies significantly with the height of the potential barrier. For a very large potential barrier (e.g. P a = 0.01), the long range diffusivity is sm aller than that predicted by Eq.1 where it is assumed that is the unrestricted diffusivity inside the crystals far away from the crystal boundaries. For small potential barriers, the long range diffusivity approaches the value of predicted by Equation 1 as shown in Figure 4 Figure 4 A Plot of Diffusivity vs. RMSD Scaled to the Size of a Crystal Figure 5 shows more clearly the dependence of the long range diffusivity on the height of the interfacial potential barrier. The data shown here was collected at iterations ; for reference this corresponds to an RMSD of ten times the size of an individual crystal. There is a nonlinear relationship between the diffusivity and P a PAGE 6 BRIAN DECOST & DR SERGEY VASENKOV University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 6 Figure 5 Variation in Concentration Distribution Because the flux balance was parameterized using simulation data, as discussed in the methodology section, some variation in the gas fraction is expected for different interfacial potential barrier heights. Figure 6 shows the variation in gas concentration for the same set of potential barriers as in the previous sections. Of particular importance is that the gas fraction is inversely related to P a while the long range diffusivity is directly related to P a Because molecules in the gas phase have a larger mean free path than those inside of crystals, a higher diffusivity is expected for larger gas fractions. T hat the higher gas fractions shown here correspond to lower long range diffusivities implies that these variations in the gas fraction do not account for the diffusivity trends discussed in previous sections. Figure 6. A Plot of Gas Phase Density vs. Potential Barrier Height PAGE 7 MONTE C ARLO SIMULATION OF LONG RANGE SELF D IFFUSION IN MODEL POROUS MEMBRAN ES AND C ATALYSTS University of Florida | Journal of Undergraduate Resea rch | Volume 14, Issue 3 | S ummer 2013 7 CONCLUSIONS Monte Carlo simulations were used to gain insight into the nature of macroscale diffusion in nanoporous materials. The simulation algorithm used was designed in such a way as to provide information about the effects of the microstucture on the mass transport properties of gases in these materials. Importantly, the long range diffusivity in the model nanoporous systems depends on the height of the interfacial potential barrier. For the case of a very large potential barr ier, it is shown that the long range diffusivity approaches the value typically assumed in practice: For moderate and lower potential barrier heights, it is shown that the contribution of sorbate molecules to the long range di ffusivity cannot be ignored, and the value of the diffusivity is more accurately described by the theoretically expected value: In the future, the model that was developed during this study will continue to be used to investigate how the microstructure of porous media affects macroscale diffusion in these systems. These simulations will be used in conjunction with Pulsed Field Gradient Nuclear Magnetic Resonan ce techniques to gain insight into real porous media. The algorithm has been designed in such a way that the extension to modeling Knudsen regime diffusion (very low gas phase density) will be relatively trivial. Additionally, chemical separations or other important industrial processes could feasibly be modeled with relatively little modification to the code base. ACKNOWLEDGEMENTS I thank Dr. Sergey Vasenkov and Robert Mueller for the direction and support they provided for me during the three semesters th at I worked with them. Dr. Vasenkov has played a key role in directing my research, and Robert has been an encouragement and an excellent resource and sounding board. REFERENCES 1. Geier, O.; Vasenkov, S.; Krger, J. PFG NMR study of long range diffusion in beds of NaX zeolite: Evidence for different apparent tortuosity factors in the knudsen and bulk regimes. Journal of Chemical Physics 2002, 117 (5) 2. Auerbach, S. M.; Ramanan, H. Modeling jump diffusion in zeolites: I. principles and methods. 2006 3. Krutyeva, M.; Krger, J. PFG NMR diffusivity with beds of nanoporous host particles: an assessment of mass transfer in compartmented two phase systems. Langmuir 2008, 24, 10474 10479 4. Papadopoulos, G. K.; Theodorou, D. N.; Vasenkov, S.; Krger, J. Mesoscopic simulations of the diffusivity of ethane in beds of NaX zeolite crystals: Comparison with PFG NMR measurements. Journal of Chemical Physics 2007, 126 (9). 5. Mueller, R. Dynamic monte carlo simulation of self diffusion in zeolite beds: A study of desorption and re adsorption of tracer molecules. 2009 6. Sanders, M.; Mueller, R.; Menjoge, A.; Vasenkov, S. Pulsed field gradient nuclear magnetic resonance study o f time dependent diffusion and exchange of lipids in planar supported lipid bilayers. The Journal of Physical Chemistry B 2009, 113 (43), 14355 14364. 7. Greenwood, J. The correct and incorrect generation of a cosine distribution of scattered particles for monte carlo modeling of vacuum systems. Vacuum 2002, 67, 217 222. |