Obtaining Knowledge and Data for Separations from Molecular Simulation J. Ilja Siepmann Nanoporous Materials Genome Center, Chemical Theory Center, Depts. of Chemistry and of Chemical Engineering & Materials Science University of Minnesota, Twin-Cities, MN Vapor–Liquid Equilibria / Distillation Liquid–Liquid Equilibria / Extraction Sorption and Transport in Porous Materials / Adsorption & Membranes Analyte Distribution / Chromatography A Research Agenda for a New Era in Separations Science Keck Center, Washington, DC August 22, 2018
Goals and Challenges: Molecular Simulation • provide atomic-level understanding of complex chemical systems and processes • investigate how changes in molecular architecture and composition influence macroscopic observables • predict accurate thermophysical properties (Chemical Industry, Vision 2020) • design improved separation processes and functional materials • precision of particle-based simulations depends solely on sampling the important regions in phase space • develop efficient Monte Carlo and molecular dynamics algorithms to sample events/processes occurring at multiple physical timescales and length scales , e.g., local displacements versus self-assembly in fluid mixtures, transfer in multi-phase systems, and chemical reactions • develop software and workflows that utilize high-performance computers with > 10 5 cores, novel memory distribution, and co-processor acceleration • accuracy depends solely on the force field or electronic structure theory used to describe inter- and intramolecular interactions • develop transferable force fields (different molecules and solid sorbents, state points, compositions, and properties) • develop Kohn-Sham density functional theory for chemisorption and reactive systems
Potentials & Forces – Goarse-Graining number of sites blobs representjng multjple molecules S trans blobs representjng many atoms S rot beads representjng 3 or 4 bonded atoms S dihedral UA with partjal charges and C6 dispersion UA with multjpoles and/or higher-order dispersion AA with partjal charges and C6 dispersion e v e AA with multjples and/or higher-order dispersion tj v c i s a s l e u off-atom sites e m v r p r r tj / e e e s r i f r e e e d t s frozen electron density f e t t n d y e s a a g d a a n l t t b n r s o s a e t a a pseudopotentjal for core electrons r b d d s e z r t - i n - e i w g y t m r t u r a n r r i a o c o o l a all electron i o a x h h r t m e g a p p c s response nuclear quantum effects classical mechanics in contjnuum space classical mechanics on lattjce particle position
Transferability, Accuracy & Efficiency Ø � Transferable � force field (FF) implies l parameters for a given interaction site should be transferable to different molecules (e.g., identical parameters should be used for the methyl group in, say, n -hexane, 1- hexene, or 1-hexanol) l a specific (set of) combining rule(s) is used consistently l parameters can be used over a wide range of temperatures and pressures; (low-temperature physics, planetary science) l parameters can be used to predict different types of properties (e.g., thermodynamic, structural, or transport) Ø � Accuracy � can only be assessed by comparison to reliable experimental data and one requires data beyond the fitting set to truly assess � transferability � (true versus effective potentials) Ø “Efficiency” is a quest for simplicity while maintaining “accuracy” by adjusting l functional form of interaction potential (square root, spherically symmetric) l number and types of interaction sites (e.g., meso-bead, united atom, all atom, nuclei & valence electron centers) l number of adjustable parameters
Parameterization and Transferability Early TraPPE-UA models : Ø Fitted to reproduce critical temperature and low- T liquid density ; i.e., the Vapor-liquid coexistence curves temperature-dependent Gibbs free energy of transfer Ø Using relatively short simulations over limited temperature range Transferability of CH x united atoms Bead-by-bead parameterization using experimental VLE data
VLE – Prediction of Critical Points Nature 1993 , 365 , 330 J. Chem. Eng. Data 2014 , 59 , 3301
Performance for VLE Prediction TraPPE-EH Gibbs99 Mie OPPE TraPPE-UA 800 NERD sim [K] 600 2 = 0.995 400 T c 2 = 0.999 R R MUPE = 1.2% MUPE = 0.2% 2 = 0.967 2 = 0.997 2 = 0.984 2 = 0.993 200 R R R R MUPE = 1.2% MUPE = 0.9% MUPE = 1.5% MUPE = 1.0% critical 0 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 0 200 400 600 800 temperature exp [K] exp [K] exp [K] exp [K] T c T c T c T c TraPPE-EH Mie OPPE TraPPE-UA 600 normal sim [K] boiling 400 T b point 2 = 0.995 R MUPE = 1.0% 200 2 = 0.999 2 = 0.989 2 = 0.993 R R R MUPE = 0.5% MUPE = 2.1% MUPE = 2.0% 0 0 200 400 600 0 200 400 600 0 200 400 600 exp [K] exp [K] exp [K] T b T b T b Mie OPLS TraPPE-UA 4 OPPE r sim / r exp - 1 [%] 2 ambient 0 density -2 MUPE = 0.8% -4 Siepmann, MUPE = 1.0% MUPE = 1.5% MUPE = 1.1% unpublished data 0 0.5 1 1.5 0 0.5 1 1.5 0 0.5 1 1.5 3 ] 3 ] 3 ] r sim [g/cm r sim [g/cm r sim [g/cm
Fitting to Binary VLE Data AIChE J. 2001 , 47 , 1676 Vapor-liquid coexistence curves TraPPE-CO 2 JPCB 2001 , 105 , 9840 EPM2
VLE – Azeotropes & Complex Mixtures JPCB 2001 , 105 , 3093 ACR 2007 , 40 , 1200 FPE 2004 , 220 , 211
VLE - Distillation AIChE J. 2017 , 63 , 5098 Model with distributed partial charges is significantly more accurate than point- quadrupole model Difference in separation factor for ethane-rich mixtures leads to 30% differences in number of stages in stripping section of distillation column
VLE – First Principles Simulations J. Phys. Chem. A 2006 , 110 , 640 PCCP 2013 , 15 , 13578 Siepmann, unpublished data 700 7 6 600 5 -ln r vap. T [K] rVV10 4 B97M-rV 500 revPBE-D3 BLYP-D3 3 M06-L-D3 PBE0-D3 400 2 0.0 0.2 0.4 0.6 0.8 1.0 1.50 1.75 2.00 2.25 r [g/ml] 1000/ T
(V)LLE – Miscibility Gaps FPE 2016 , 407 , 269 J. Phys. Chem. B 2005 , 109 , 2911
LLE – Extraction AIChE J. 2013 , 59 , 3065 4-decanol 1-decanol
LLE – Polymer Phase Behavior & Aggrgation Low-χ Mixtures: Polyolefins High-χ Mixtures: PEO/PEP PP hhPP PEP PEO H 3 C O O CH 3 n n n n − 1 Chen et al. , 1 Chen et al. , 500 550 Expt Sim Macromolecules Macromolecules 51 , 3774 (2018) 49 , 3975 (2016) 0.5 450 500 1/2 ] ( δ 1 - δ 2 ) sim [MPa T [K] T [K] 0 400 450 -0.5 350 400 PP+PEP 217-MD-1.01 222-MD-1.0 hhPP+PP 280-SZ-1.05 280-SZ-1.05 hhPP+PEP 500-SZ-1.10 500-SZ-1.1 -1 300 350 -1 -0.5 0 0.5 1 0 0.2 0.4 0.6 0.8 0 0.2 0.4 0.6 0.8 1 1/2 ] Weight fraction of PEP*-423 Weight fraction of PEP*-423 ( δ 1 - δ 2 ) SANS [MPa 10 χ MF ( T sim = 400 K or T expt = 300 K) mono-oligomers mono-oligomers di-oligomers Expt HO y -1 x -2 mean-field OH Extreme-χ Block Oligomers 1 di-oligomers Chen et al. , ( x -3)/2 ACS Nano 12 , 4351 (2018) 0.1 HO y -1 ( x -3)/2 OH 0 5 10 15 20 Domain period d [nm]
Developing the TraPPE-zeo Force Field J . Phys. Chem. C 2013 , 117 , 24375 • Challenge: Large discrepancies of current zeolite force fields for adsorption of polar guest molecules • Target: non-polar, polar, and H-bonding guest molecules in all-silica zeolites • LJ sites on both Si & O to achieve better balance between dispersive and H-bonding interactions • A three-step, grid-based search in 5-dimensional parameter space requiring ≈50,000 simulations ethanol CO 2 heptane
Adsorptive Separation: Gas Phase W W Angew. Chem. Int. Ed. 2016 , 55 , 5938
Adsorption – Alcohol/Water in Silicalite-1 1 2 3 4 Adsorption of H 2 O in MFI 10 10 10 10 1.0 20 alcohol(+water), sim water(+alcohol), sim 0.8 alcohol(+water), IAST 15 water(+alcohol), IAST 1.0 N [molec/uc] alcohol(pure), sim 0.6 0.9 x alcohol, ads 10 x alcohol, ads methanol 0.8 0.4 5 0.7 0.6 0.2 0 0.00 0.02 0.04 0.06 0.08 x alcohol, sol 0.0 12 4 10 N [molec/uc] methanol, 298 K ethanol, 298 K 8 methanol, 323 K ethanol, 323 K ethanol 3 10 S alcohol 4 2 IAST overestimates 10 0 1 2 3 4 10 10 10 10 separation factors for p alcohol [Pa] IAST: Myers & Prausnitz, 1 10 methanol and ethanol AIChE J ., 1965 , 11 , 121 by about 5000 0 0.2 0.4 0.6 0.8 1 x alcohol, sol Bai et al. , Langmuir 2012 , 28 , 15566; J. Phys. Chem. C 2013 , 117 , 24375
Adsorptive Separation – Adsorbent Screening Ø Screening of 402 IZA-SC structures at w = 0.12% and for top-64 structures at 5 higher concs Ø Screening yields large number of frameworks that outperform MFI (ranked 15 – 26) Ø Processes may exploit FER ’s high selectivity to extract 90% of EtOH contained in a 5 wt% fermentation broth by adsorptive separation reaching down to raffinate conc of w ≈ 0.5 wt% Ø Processes using ATN* may exploit its higher EtOH loading and reduce feed mixture from initial w ≈ 15% down to 5 wt% and recycle the raffinate back to the fermentation broth Bai, Jeon, Ren, Knight, Deem, Tsapatsis & Siepmann, Nature Commun . 2015 , 6 , 5912
Recommend
More recommend
