1 1 First-principles based design of Pt- and Pd-based catalysts for benzene hydrogenation Maarten K. Sabbe, Gonzalo Canduela, Marie- Françoise Reyniers, Guy B. Marin
Introduction: benzene hydrogenation on Pt(111) Benzene hydrogenation: applications in hydrotreating, hydrocracking, cyclohexane production Current status of computational models: dominant path proposed based on Pt 22 cluster calculations (DP cluster ) Pt 22 cluster Electronic reaction barriers BP86/DZ on Pt 22 cluster of Pt(111) Saeys.M J.Phys.Chem.B, 109,2064-2063 (2005) Regressions to experimental data suggest other dominant path (Thybaut): DP regressed Experimental work: no consensus on the rate determining step Entropy contributions difficult at cluster level: include using periodic calculations 2
Aim Pt(111) Evaluate reaction barriers based on periodic calculations Calculate entropy contibutions and rate coefficients Perform reactor simulations and compare yields to experiment Pt- and Pd-based catalyst design evaluate stability and hydrogenation reactivity of Pt 3 M alloys and surface alloys (M= Ag,Au,Cu,Fe,Co,Ni,Pd) Pd: start design of Pd-based catalysts by developing a first principles kinetic model on Pd(111) 3
Computational approach Periodic structure 3 x 3 unit cell used to model the Pt(111) surface: 9 atoms/layer moderate lateral interactions: coverage degree 30% Artifical dipole layer Vacuum layer 10.6 Å Surface with unit cell indicated Unit cell Top view Unit cell Side views Relax 2 upper layers Fix 2 bottom layers Lattice constant: 4.011Å DFT (VASP) PW91 functional (GGA) plane waves; PAW; 400 ev; no spin polarization (for clean Pt) 5 x 5 x 1 k-point Monkhorst-Pack grid first order Methfessel-Paxton smearing, σ=0.20 ev TS determination: NEB, followed by DIMER calculation 4
Outline Part I: Hydrogenation of benzene on Pt(111): from molecule to reactor Reaction network: electronic barriers Entropy contributions Rate coefficients Compare reactor simulations to experiment Part II: Catalyst-descriptor based design of hydrogenation catalysts 5
Pt(111) network: electronic reaction barriers DP cluster,135thb MEP periodical,123thb Based on ΔE el : no clear dominant path dominant path on Pt 22 cluster level minimum energy path (periodical calculations) Electronic energy barriers ΔE el forward reverse 6
Entropy contributions are important for K and k 3N i 2 1 1 H q 2 i qi 2 2m H ij Immobile species: Harmonic frequency analysis vibrational Schrödinger equation Kinetic energy Hessian q i 2 Hq E q j (q) E (q) Potential energy requires knowledge of Hessian H q i = Δx, Δy, Δz around equilbrium geometry S Vibrational contribution to entropy 3N i rovib, HO R ln 1 h i i 1 kbt kbt 1 e h e h i kbt Mobile species free rotation and/or free translation Replace 2 translational and 1 rotational frequency S transl, surf R lnq trans '( A, T) 1 2 2 ln qrot, ( T) A: 10-19 m² for H*; 5 10-19 m² for hydrocarbon species identify mobility of surface species: calculate diffusion barriers S rot, Z R Z 1 2 7
E-Etop kj/mol Entropy contributions: mobile mode identification Determine transition states for diffusion (NEB+dimer) Species + motion All species immobile at 450 K except H and cyclohexane (barrier < 9 kj/mol) ΔE kj/mol Hydrogen (top to top) 9.2 Hydrogen (top to hollow) 11.6 Benzene (hollow to bridge-rotation) 21.1 135 THB (translation) 233.0 1235 THB (rotation) 99.8 Cyclohexyl (translation) 98.5 Cyclohexyl (rotation around C-Pt bond) 12.7 Cyclohexane (rotation) 5.9 10 8 6 4 2 0 H* top to top diffusion (NEB) Translational Coordinate 135-THB translation (diffusion barrier 233 kj/mol) Initial state Final state 8
Rate coefficients indicate dominant path no clear dominant path Evaluate full reaction network in simulation DP cluster MEP periodical DP periodical,k dominant path at Pt 22 cluster level minimum energy path (periodical calculations) dominant path based on rate coefficients (periodical calculations) rate coefficients k (s -1 ) forward reverse 9
Experimental data: Berty set-up Catalyst: Pt/ZSM-22 (0.5 wt% Pt) Conversion: 9-85% Input variables (43 experiments) Berty-reactor: Gas phase CSTR (intrinsic kinetics) Benzene Feed (mol s -1 ) 17 10-6 -57 10-6 T (K) 425-500 P(atm) 10-30 p H2 /p B 5-11 W cat (g) 1.29-1.8 W/F benzene (kg cat s -1 mol -1 ) 22-74 10
Reactor simulation approach Simulations CSTR model Levenberg-Marquardt for parameter estimation Goal function=σ(simulated product yieldexp.observed) 2 K(T) and k(t) with mobile H* and cyclohexane*, other species are considered immobile catalyst model: 0.008 active sites/kg cat PSSA (reaching steady state using transient solver) Estimated parameters H2 adsorption enthalpy: strongly coverage dependent Estimation of this parameter required Podkolzin et al., JPCB, 105:8550 (2001) Transient continuity equations: dfi 0 Gas phase species: Fi Fi dt Surface species: Free sites: dc dt dc dt i* R i* * R* RW i 0 General reduction of activation energy: calculated E a larger than experiment temperature dependence too strong without reduction of E a E a,i = E a,i,abinitio + ΔE a,parameter 11
Simulated product yield (10-6 mol/s) Full network: reactor simulation results K(T) and k(t) for mobile H* and cyclohexane* (other immobile) surface coverage 1 => take ΔH ads (benzene)= -66.1 kj mol -1 (calculated value) 50 Cyclohexane yield parity plot Simulation Estimate ΔH H2 and ΔE a E a,i = E a,i,abinitio + ΔE a,parameter 40 30 20 ΔH ads,h2-46.1 ± 2.2 kj/mol ΔE a -14.6 ± 2.7 kj/mol F 428 10 0 0 10 20 30 40 50 Experimental product yield (10-6 mol/s) Estimating only ΔH H2 : yields still too low temperature dependence too strong without reduction of E a Estimate E a reduction 12
Full network: reaction path analysis 20 bar, 225 C, 1.8 g cat, 0.13 mol/h benzene, (H 2 /B) in =5 W/F B =48.4 kg cat s/mol Electronic energy barriers ΔE el forward reverse Clear pathway for step 4, 5 and 6 In step 2 and 3 equilibration between intermediates 13
Conclusions and prospects Conclusions No clear dominant path based on electronic energies for full network Activation energies need to be reduced in order to obtain quantitative agreement to experimental values With 2 parameters, a reasonable agreement to experimental yields is obtained Future work Multiscale modeling: development of first-principles based kinetic Monte Carlo simulation tools to assess the validity of the mean field approximation under industrially relevant operating conditions Introduce method for clean Pt catalysis If results differ significantly from mean-field results, apply on bimetallic catalysts as well 14
Outline Part I: Hydrogenation of benzene on Pt(111): from molecule to reactor Part II: Catalyst-descriptor based design of hydrogenation catalysts Pd catalysts Pt 3 M catalysts Conclusions & prospects 15
Pd-catalyzed hydrogenation First step in design of Pd-based catalysts: develop kinetic model on Pd(111) analogous to Pt(111) similar MEP as for Pt(111) PW91 PAW 400 ev benzene at hollow site 3x3 unit cell Electronic energy barriers ΔE el forward reverse Future work: entropy contributions, rate coefficients and multiscale modeling of the reactor 16
Pt 3 M catalysts: surface segregation Pt 3 M alloys (4x4 supercells) (M= Ag, Au, Cu, Fe, Co, Ni, Pd) evaluate stability & reactivity Pt 3 M Bulk alloy Pt 3 M/Pt Surface alloy Segregation No segregation Antisegregation Au, Ag E seg large E seg = E slab,seg E slab,non-seg Pd stays in place Most stable alloys studied Au/Pt Ag/Pt Pt 3 Ag/Pt Pt 3 Au/Pt surface alloy Pt 3 Pd/Pt Pt 3 Pd bulk alloy Fe, Co, Ni, Cu E antiseg large E antiseg = E slab,antiseg E slab,non-seg Pt/Pt 3 M/Pt surface alloys Pt/PtM/Pt 3 M bulk alloys M=Fe, Ni, Co and Cu 17
Adsorption sites Hydrogen Benzene Non-segregated Top-Pt Pt 3 -fcc Pt 2 M-hcp Top-M fcc-pt 2 -bri 30 2 M-fcc bri-ptm 30 2 M 0 Anti-segregated fcc-pt 3 -fcc 0 3 bri-pt 30 PtM-bri 30 2 Top-Pt 1 Pt 3 -fcc Pt hcp-m 3 -hcp 00 Pt hcp-pt0 2 M-hcp 0 Pt 3 -hcp Non-segregated Antisegregated Top-Pt 2 18
Pt 3 M: Benzene adsorption energy Adsorption Energy (kj mol -1 ) 40 20 0 Pt 3 M Bulk alloys Adsorption of benzene Hollow hcp -20 Bridge -40 Pt 3 M/Pt Surface alloys -60-80 -100 Hollow hcp Bridge up to 50 kj/mol weaker 60 to 90 kj/mol weaker than Pt(111) bridge -120-140 Au Ag Fe Co Ni Cu Pd Segregation Antisegregation No segregation Pt(111) (bridge) -119 kj/mol 4x4 unit cell 19
Pt 3 M: Hydrogen adsorption energy Adsorption Energy (kj/mol) 60 40 Top Pt 3 M Bulk alloys Adsorption of hydrogen 0.5 H2 + * H* 20 0 Hollow fcc Pt/PtM/Pt 3 M -20-40 -60-80 Pt 3 M/Pt Surface alloys Top Hollow fcc Pt/Pt 3 M/Pt Pt 3 M/Pt Ag Au Cu Co Ni Fe Pd Segregation up to 15 kj/mol weaker Antisegregation up to 30 kj/mol weaker No segregation Pt(111) fcc site -47kJ/mol 2x2 unit cell 20
Pt 3 M: activation energies first step Step 1 Electronic barrier E el = E TS + E Pt - E Bads - E Hads Pt+B+H TS BH Electronic Barrier (kj/mol) 160 140 try to add correlation with Eads 120 100 80 60 40 20 Pt 3 M/Pt Surface alloys Pt 3 M Bulk alloys 92 kj/mol Pt(111) Activation energies are lower on Pt 3 Co, Pt 3 Ni, Pt 3 Fe, Pt 3 Cu and Pt 3 Fe/Pt than on pure Pt(111) 0 Co Ni Fe Cu Pd Au Ag Antisegregation No segregation Segregation 21
Activation energies correlate well with E ads Ea (kj/mol) 150 120 90 60 30 0 Ea (kj/mol) 150 120 90-150 -100-50 0 E ads benzene (kj/mol) Bulk Pt 3 M alloys Surface Pt 3 M/Pt alloys Pt (111) Electronic barriers are well correlated to the adsorption energies of the reactants E ads as descriptor of reactivity Can activation energy however be directly linked to electronic catalyst properties? 60 30 0-60 -40-20 0 E ads hydrogen (kj/mol) 22
Ea (kj/mol) Eads (kj/mol) d-band descriptors as catalyst descriptor density of states projected on d-band of surface atoms of clean slab DOS-based descriptors Best correlation with occupied d-band center Density of states (ev -1 ) E fermi center of occupied d-band DOS-based descriptors E fermi Work function d-band center DOS at Fermi Work function Ф=E f E vacuum DOS at Fermi Energy (E-E f ) 20 0-20 -40-60 -80-100 -120-140 140 120 100 80 60 40 Pt 3 Au/Pt Pt 3 Ag/Pt -2.80-2.60-2.40-2.20 ε d - E f Pt 3 Ag/Pt Pt 3 Au/Pt : bulk alloys : surface alloys Pt -2.80-2.60-2.40-2.20 ε d - E f Pt 23
Conclusions & prospects Benzene hydrogenation on Pt(111): Succesful reaction simulation using only 2 optimized parameters Benzene hydrogenation on Pt 3 M bimetallic alloys Adsorption energies of benzene and hydrogen of the Pt 3 M alloys are, compared to pure Pt(111), weaker when alloying with Au, Ag, Fe, Co, Ni and Cu On the bulk alloys Pt 3 Co, Pt 3 Ni, Pt 3 Fe, Pt 3 Cu and the Pt 3 Fe/Pt surface alloy the activation energies are lower than on pure Pt(111) the d-band center correlates well with benzene adsorption energies and hydrogenation barriers for the studied alloys. Prospects Development of first-principles based kinetic Monte Carlo simulation tools to assess the validity of the mean field approximation under industrially relevant operating conditions Further evaluate the d-band center as useful catalyst descriptors relating the variation in activity and selectivity in going from Pt(111) to other metal catalysts, and screen the d-band center of other promising alloys Definition of optimal catalyst properties: simultaneous optimization of catalyst properties, industrial process conditions and reactor configuration 24
Acknowledgements Lucía Laín Amador Joris Thybaut Fund for scientific research - Flanders Long Term Structural Methusalem Funding by the Flemish Government grant number BOF09/01M00409 Questions? 25
Glossary DFT: Density Functional Theory Dimer method: force-based transition state search algorithm GGA: generalized gradient approximation (within DFT theory) MEP: Minimum Energy Path NEB: Nudged Elastic Band method for the calculation of MEPs PAW: Plane Augmented Waves (periodic calculation technique) PW91: Perdew-Wang type of DFT functional VASP: Vienna Ab initio Simulation Package 26