You are on page 1of 86

Progress in Surface Science 65 (2000) 65±150

www.elsevier.com/locate/progsurf

Microkinetic simulation of catalytic


reactions
Per Stoltze
Department of Chemistry and Applied Engineering Science, Aalborg University, Niels Bohrs Vej 8, 6700,
Esbjerg, Denmark

Abstract

The study of the kinetics of heterogeneous catalyzed reactions consists of three di€erent
aspects: kinetics studies for design purposes, kinetics studies of mechanistic details and
kinetics as a consequence of a reaction mechanism. The latter aspect is clearly the least
explored of the three and it will not become routine until the development and analysis of
microkinetic models is automated. Based on a survey of existing microkinetic models of
heterogeneous catalytic reactions, three classes of Langmuir±Hinshelwood (LH)
mechanisms are shown to be suitable for microkinetic modeling.
The ®rst model consists of an LH mechanism with the quasi-equilibrium approximation
and a single, rate-limiting step. For this model detailed kinetic studies are possible and
include the analytic determination of reaction rate, surface coverages, reaction orders and
activation energy.
The second model consists of an LH mechanism with the steady-state (SS)
approximation. This model allows the treatment of some kinetic phenomena, which
cannot be treated by Model 1. However, the treatment of Model 2 is much more dicult
and the data which may be determined analytically from Model 2 are the reaction rate,
the surface coverages and the degree of rate limitation. The third model is kinetic Monte
Carlo (KMC) simulations which allow the modeling of even more mechanistic details,
such as surface di€usion and adsorbate±adsorbate interactions. Analysis of this class of
models are limited to numerical simulation. Two appendices present the implementation
of a Runge±Kutta (RK) integration of the conversion through an isothermal plug-¯ow
reactor and a KMC simulation of a reaction rate, respectively. 7 2000 Elsevier Science
Ltd. All rights reserved.

0079-6816/00/$ - see front matter 7 2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 7 9 - 6 8 1 6 ( 0 0 ) 0 0 0 1 9 - 8
66 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

Nomenclature

HP Horiuti±Polyani
IS irreversible step
KMC kinetic Monte Carlo
LH Langmuir±Hinshelwood
MARI most abundant reaction intermediate
MC Monte Carlo
ODE ordinary di€erential equation
RK Runge±Kutta
RLS rate limiting step
SS steady state

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
1.1. Caveats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
1.2. Microkinetic modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
1.3. Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

2. Applications of microkinetic modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73


2.1. Oxidation of NH3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.2. Ammonia synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.2.1. Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.2.2. Test of model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.2.3. Activation energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.2.4. Reaction orders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.2.5. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.3. CO+NO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.4. Hydrogen oxidation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5. Carbonmonoxide oxidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.6. HCN synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.7. Hydrogenation, dehydrogenation and isomerization . . . . . . . . . . . . . . . . . 83
2.8. Water±gas shift reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
2.9. Methane oxidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.10. Methanol synthesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.11. NO+NH3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

3. Statistical thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.1. Partition function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.2. Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.3. Rotation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.4. Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.5. Ground state energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.6. Thermodynamic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.7. Ideal gas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 67

3.8. Equilibrium constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91


3.9. Langmuir isotherm for non-dissociative adsorption . . . . . . . . . . . . . . . . . 91
3.10. Langmuir isotherm for dissociative adsorption. . . . . . . . . . . . . . . . . . . . . 92

4. Reaction mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.1. Elementary steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2. Consistency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.1. Stoichiometric consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.2. Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.2.3. Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2.4. Thermodynamic consistency. . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.3. Net reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4. Langmuir±Hinshelwood mechanisms. . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5. Complications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5.1. Kinetic equations are non-linear . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5.2. Inerts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5.3. Non-consecutive steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.5.4. Elusive intermediates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5.5. Undetectable steps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5.6. Dead ends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5.7. Linear dependence between reaction steps . . . . . . . . . . . . . . . . . . 99

5. Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.1. Reversibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2. Rate laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3. Rate constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5.4. Stoichiometric matrix. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.4.1. Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.4.2. Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.4.3. Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.5. Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.6. Equilibrium equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.7. Stoichiometric consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.8. Thermodynamic consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.9. Forward and backward rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.10. Kinetics versus equilibrium. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.11. Rate limiting step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.12. Net rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.13. Activation energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.14. Reaction order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.15. Plug-¯ow reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
5.15.1. Isothermal plug-¯ow reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.15.2. Adiabatic plug-¯ow reactor . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6. Approximations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
6.1. Steady-state approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.2. Quasi-equilibrium approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.3. Irreversible step approximation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.4. MARI approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
68 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

7. Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.1. Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
7.2. Types of models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

8. Model 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.1. Consistency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.2. Complications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.3. Stoichiometric number . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
8.4. Net reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
8.5. Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.6. Gas-phase equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.7. Calculation of coverages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
8.8. Activation energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
8.9. Reaction orders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

9. Model 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.1. Consistency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.2. Net reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.3. Gas-phase equilibrium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.4. Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.5. Coverages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
9.6. Rate limitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

10. Kinetic Monte Carlo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126


10.1. Phases in a simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
10.2. Basic algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
10.3. Events . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
10.4. Consistency. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
10.5. Time-scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
10.6. Advanced algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
10.7. Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

11. Software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

12. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

Appendix A . . . . . . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 132


A.1. Runge±Kutta . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.2. Calculation . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
A.3. Parameters . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
A.4. Variables . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
A.5. Calculation of equilibrium constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
A.6. surf() . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
A.7. rate() . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
A.8. driv() . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.9. rk() . . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
A.10. scrk() . . . . . . . . . . . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
A.11. Overall calculation. . . . . . ........ . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 69

Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
B.1. Event queue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
B.2. Equilibration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
B.3. Motion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
B.4. Calculation of rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
B.5. Output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

1. Introduction

From the standard textbooks on chemical engineering and catalysis [1±21] it


appears that the study of the kinetics of heterogeneous catalyzed reactions consists
of at least three rather di€erent aspects:
1. Kinetics studies for design purposes. In this ®eld, results of experimental studies
are summarized in the form of an empirical kinetic expression, which are useful
for design of chemical reactors, quality control in catalyst production,
comparison of di€erent brands of catalysts, studies of deactivation and of
poisoning of catalysts [22].
2. Kinetics studies of mechanistic details. If a reasonable and not too detailed
reaction mechanism is available, an experimental kinetic study may be used to
determine details in the mechanism [23]. However, when parameters in a
mathematical expression are ®tted to kinetic data there are limits to what
information one can determine. Qualitatively di€erent kinetic expressions may
®t the data equally well [24,25]. Still, mechanistic considerations may be very
valuable as a guidance for kinetic studies [26,27].
3. Kinetics as a consequence of reaction mechanism. A proposed reaction
mechanism often contains sweeping statements about the macroscopic kinetics
or reaction rate. The deduction of the kinetics from a proposed reaction
mechanism generally consists of a reasonably straightforward transformation,
where all the mechanistic details are eliminated until only the net gas-phase
reaction and its rate remains. This approach may be used to answer several
questions: Is the proposed mechanism consistent? Are the available data on
intermediates and on reaction steps complete and consistent? What is the net
reaction and what is the form of the macroscopic kinetics? What is the relation
between the microscopic and the macroscopic kinetics? What is the predicted
reaction rate and is it consistent with available experimental data?
For the three aspects of the study of kinetics, the optimal experimental and
70 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

theoretical approach is quite di€erent [28]. In the present review we will


concentrate on the third aspect, which is clearly the least explored of the three.
Modern approaches to the study of reaction mechanisms consist of two
approaches, experiments on well-de®ned systems and detailed calculations for
individual molecules and intermediates. The studies of well-de®ned systems consist
of spectroscopic studies of individual molecules and measurements of the rate of
catalytic reactions on single crystal surfaces [29±33], as well as structure and
reactivity of well-de®ned catalyst models [32,34,35]. The computations consist of
electron structure calculations including calculations for transition state [36±40], as
well as large Monte Carlo (MC) simulations.
For a complex system it is generally straightforward to deduce the macroscopic
behavior from a comprehensive knowledge of the details at the microscopic level.
This deduction mainly consists of averaging out irrelevant details and thus
discarding massive amounts of information. However, for a complex system the
deduction of the microscopic details purely by observations of the macroscopic
behavior is impossible. The information discarded in the transition from the
microscopic to the macroscopic level is lost and cannot be reconstructed purely
from observation at the macroscopic level.
In industry, catalytic reactions are performed at high pressures and high
temperatures with several reactants and products present in fairly high
concentrations. The success of studies of the modern approaches to the study of
catalytic reaction mechanisms is due to the fact that these approaches provide
de®nitive data for microscopic details.
The challenges in the deduction of reaction mechanisms from spectroscopic
studies are:
1. Pressure. Spectroscopic studies of molecules adsorbed on single crystal surfaces
are made in ultra-high vacuum and computations are made in the limit of zero
pressure. The pressure must be extrapolated by at least 12 orders-of-magnitude.
2. Temperature. Computations are made at zero temperature and a proper
thermodynamics must be constructed.
3. Structure. The catalyst consists of small particles stabilized by a structural
promoter. This challenge may be overcome by studies of suitable catalyst
models [32,34,35].
4. Conversion. The changes in gas phase concentrations which may be reached in
a single crystal reactor are generally very low. This dictates that measurements
are performed in the limit of zero product concentration. In this limit the
kinetics may be entirely di€erent from the kinetics at higher conversions.

1.1. Caveats

Before we proceed, it may be useful to list some of the key problems in the
study of the kinetics of catalytic reactions.
1. We cannot deduce the kinetics from the net reaction [26,28]. For the reaction
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 71

aA ‡ bBÿ
)ÿ
ÿ*
ÿcC ‡ dD

the kinetics is, in general, not


 a  b   
pA pB k pC c pD d
rˆk ÿ ,
p p K p p

although this kinetics predicts the correct equilibrium


 c  d
pC pD
p p
Kˆ  a  b ,
pA pB
p p

where p % is the reference pressure.


2. In the absence of solid evidence it is dangerous to argue by analogy [26,28]. As
an example, the reaction
ÿÿ
H 2 ‡ I2 )ÿ*
ÿ2HI

has a simple mechanism and a reaction rate of the form


    
pH2 pI 2 k pHI 2
rˆk ÿ ,
p p K p

while the reaction

H2 ‡ Br2 ÿ
)ÿ
ÿ*
ÿ2HBr

proceeds by a chain mechanism and has a complicated kinetics.


3. A reaction with a simple kinetics does not necessarily have a simple mechanism
[28]. As an example, the reaction
ÿÿ
2N2 O5 )ÿ*
ÿ4NO2 ‡ O2

has a simple kinetics of the form r ˆ kpN2 O5 , but has a rather complex
mechanism
ÿÿ
…a† N2 O5 )ÿ*
ÿNO3 ‡ NO2

…b† NO3 ‡ NO2 4NO2 ‡ O2 ‡ NO

…c† NO ‡ NO3 4 2NO2 : …1:1†

4. A simple mechanism such as

…a† A ‡ ÿ
)ÿÿ
*ÿA

ÿÿ
…b† B2 ‡ 2  )ÿ*
ÿ2B

…c† A  ‡B  4 AB ‡ 2  , …1:2†
72 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

may have a very complex kinetics [41].


5. Very di€erent reaction mechanisms may predict the same overall reaction rate.
Even if we have reliable data for the overall reaction rate over a large range of
reaction conditions we may be unable to distinguish between two di€erent
reaction mechanisms.
6. Many mechanistic details cannot be deduced from an experimental
determination of the form of the kinetic expression [24]. As an example, the
Temkin±Pyzhev rate expression for ammonia synthesis reproduces the
experimentally observed kinetics quite well. However, this rate expression was
originally derived from a proposed mechanism which had both the wrong key
intermediates and the wrong rate-limiting step.

1.2. Microkinetic modeling

The starting point for microkinetic modeling [14,42±46] is the detailed reaction
mechanism. Thus, while a conventional kinetic model is formulated as the rate for
an apparent gas-phase reaction, the surface species are explicitly included in a
microkinetic model.
Fig. 1 shows the derivation of a microkinetic model.

Fig. 1. In microkinetic models, the simplest feasible model at the molecular level is formulated based
on available data. The model is evaluated through simulation of kinetic data for the catalytic reaction
at high pressures and results of simulations are compared to existing kinetic datasets. Successful models
are useful for visualization of reactions, as well as for detailed investigation of kinetic and mechanistic
features.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 73

1.3. Software

The formulation of a microkinetic model of a catalytic reaction mechanism is


reasonably straightforward. However, the implementation of a model is very
tedious. Starting from an LH mechanism with, say, 10 elementary reactions, it can
easily take one month before a model has been implemented in an RK integration
for a plug-¯ow reactor, if the implementation is done by hand. Once the model
has been implemented various parameters can be investigated in a matter of
minutes, but implementation of the addition, deletion or change of an elementary
reaction can easily take a week. A very large investment in time is thus required
from a microkinetic model that has been formulated until data calculated from
the model become available.
It would be highly desirable if the formulation and investigation of a
microkinetic model for a proposed reaction mechanism could be automated. In
the following we show that at least for a certain class of reaction mechanisms, this
goal may be achieved.
As we proceed, we must keep in mind that if an automated analysis of reaction
mechanisms should be useful, the program must be able to analyze large reaction
mechanisms, and perform a wide range of computations, some of which require
much transformation of the input. Further, there may be more or less subtle
errors in a proposed reaction mechanism and these errors must be detected and an
intelligible diagnosis returned.

2. Applications of microkinetic modeling

In the present section, we will give a brief survey of existing microkinetic


models. The application of models range from qualitative considerations used to
guide kinetic measurements to detailed studies of a broad range of phenomena. It
is evidently a matter of personal taste which studies should be included. The
survey is meant to be representative rather than exhaustive. Still, the survey may
give an indication of the applicability of kinetic modeling and allow some
extrapolation into the future.
The formulation and analysis of each of the existing microkinetic models
summarized in the present section represent a substantial research e€ort and
emphasize the need for automation of the investigation of kinetic models.

2.1. Oxidation of NH3

The oxidation of NH3 over Pt wire at 250±15008C


5
2NH3 ‡ O2 4 2NO ‡ 3H2 O
2
74 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

3
2NH3 ‡ 2 O2 4 N2 ‡ 3H2 O

has been modeled [47] by the following mechanism


…a† NH3 ‡ ÿ
)ÿÿ
*ÿNH3 

…b† NH3  ‡O2  4NO ‡ H  ‡H2 O ‡ 

…c† NH3  ‡O  4 NO ‡ H2 ‡ H

…d† NO  ‡NH3  4 N2 ‡ H2 ‡ H  :

The reaction between NH3 and O2 goes through step (b), (c) or both. The
selectivity is controlled by step (d).

2.2. Ammonia synthesis

The synthesis of ammonia


ÿÿ
N2 ‡ 3H2 )ÿ*
ÿ2NH3
has been modeled [14,42,43,48±55] based on the mechanism
ÿÿ
…a† N2 ‡ )ÿ*
ÿN2 

ÿÿÿ
…b† N2  ‡  ) *ÿ2N

ÿÿ
…c† N  ‡H  )ÿ*
ÿNH  ‡

…d† ÿÿ
NH  ‡H  )ÿ*
ÿNH2  ‡

ÿÿ
…e† NH2  ‡H  )ÿ*
ÿNH3  ‡

ÿÿÿ
…f † NH3  ) *ÿNH3 ‡ 

ÿÿ
…g† H2 ‡ 2  )ÿ*
ÿ2H …2:1†

with step (b) is rate limiting.


These models are based on microscopic results available from single crystal
studies. The initial simulations reached quite di€erent results [50,56]. The
di€erences have been traced to di€erences in the choice of input parameters
[14,53,54].
The same mechanism has been studied using parameters obtained from studies
of the hydrogenation of chemisorbed nitrogen [57,58].
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 75

2.2.1. Coverage
When the mechanism has only one slow step, the system of equilibrium
equations and the rate equation may be solved with respect to y [43,49,51]

y ˆ
1
1=2
:
1=2 pH2

PN pNH3 p pNH3 pNH3 pNH3
1 ‡ K1 2 ‡ ‡ ‡ ‡ ‡ K 7
p0 K3 K4 K5 K6 K 3=2 3=2
7 pH2
K4 K5 K6 K7 pH2 K5 K6 K7 p 1=2 p1=2
H2
K6 p p 1=2
…2:2†

The coverages by intermediates may be expressed by y and the partial pressures


of the reactants and products.
PN2
yN2  ˆ K1 y …2:3†
p

pNH3 p
yN ˆ y …2:4†
K3 K4 K5 K6 K 3=2 3=2
7 pH 2

pNH3
yNH ˆ y …2:5†
K 4 K 5 K 6 K 7 pH 2

pNH3
yNH2  ˆ y …2:6†
K5 K6 K7 p 1=2 p1=2
H2

pNH3
yNH3  ˆ y …2:7†
K 6 p

p1=2
H2
yH ˆ K 1=2
7 y …2:8†
p 1=2

!
2
pN2 pNH p 2
r ˆ 2k2 K1 ÿ 3
y : …2:9†
p K g pH 2

Fig. 2 shows calculated TPD curves for N2 desorption from Fe.

2.2.2. Test of model


The model is tested [42,43,49,51,54] by comparison of the calculated and
measured exit concentrations in a kinetic study (Fig. 3).
Not all parameters are important. The critical parameters are the prefactor and
activation energy for dissociative adsorption of N2 and the binding energy for N
76 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

Fig. 2. Calculated TPD curves for N2 desorption from Fe [43]. The heating rate is 10 K sÿ1. Initial
coverage 0.2 (lower curve), 0.40, 0.60, 0.80 and 1.00 (upper curve). The simulation of desorption of N2
from Fe is used to determine the enthalpy of formation for N  :

[43]. The reason why these parameters are critical has been investigated in detail
[59].

2.2.3. Activation energy


The activation energy (Section 8.8) may be calculated from

2d ln r‡
H y ˆ kB T , …2:10†
dT

Fig. 3. Comparison between measured and calculated NH3 production for industrial catalyst Topsùe
KM1 [43]. Lower left cluster of points are measured at 101 kPa, N:H ratio 1:3, temperature 673 K,
NH3 concentration: inlet 0%, outlet 0.1±0.25%. Upper right cluster of points are measured at 7±20
MPa, N:H ratio 1:3±1:1, temperature 570±720 K, NH3 concentration: inlet 2±11%, outlet 6±12%.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 77

where r+ is the forward rate of NH3 synthesis


 
pN
r‡ ˆ 2k2 K1 2 y2 : …2:11†
p

The result is [43,49,51]

Hy ˆ Hy2 ‡ H1 ÿ 2H1 yN2  ‡ 2…H3 ‡ H ‡ 4 ‡ H5 ‡ H6 ‡ 3


2 H7 †yN
1 …2:12†
‡ 2…H ‡ 4 ‡ H5 ‡ H6 ‡ H7 †yNH ‡ 2…H5 ‡ H6 ‡ 2 H7 †yNH2 

‡ 2H6 yNH3  ÿ H7 yH :

The terms in the activation enthalpy can be interpreted as follows [43,49,51]

Reaction Enthalpy

N2  ÿ
)ÿ
ÿ*ÿN2 ‡  ÿH1
N  ‡3=2H2 ÿ
)ÿ*
ÿÿNH3 ‡  +(H3+H4+H5+H6+3/2H7)
NH  ‡H2 ÿ ÿ*
)ÿÿNH3 ‡  +(H4+H5+H6+H7)
ÿÿ
NH2  ‡1=2H2 ) ÿ*
ÿNH3 ‡  +(H5+H6+1/2H7)
ÿÿ
NH3  ) ÿ*
ÿNH3 ‡  +H6
H) ÿÿ
ÿ*ÿ1=2H2 ‡  ÿ1/2H7

Although the activation enthalpy depends on the reaction conditions, it is


reasonably constant at typical reaction conditions (Fig. 4).
This table allows us to interpret the activation enthalpy for the synthesis as the

Fig. 4. Calculated activation enthalpy [43] for ammonia synthesis at 10.1 MPa (solid curve) and 101
kPa (dashed curve), N:H ratio 1:3, temperature 673 K.
78 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

activation enthalpy for the step N2 ‡ 2  4 2N plus the averaged enthalpy to be
supplied to create two free sites. The average is formed by weighing the enthalpies
by the coverage for each intermediate. For typical conditions of ammonia
synthesis the contribution from the coverage by N is far larger than the
activation energy for the rate limiting step (RLS).

2.2.4. Reaction orders


The reaction orders (Fig. 5) for N2, H2 and NH3 may be de®ned as
d ln…r‡ †
ai ˆ  : …2:13†
pi
d ln
p

From the forward rate of NH3


 
pN
r‡ ˆ 2k2 K1 2 y2 …2:14†
p

the reaction orders can be calculated [43,49,51]


aN2 ˆ 1 ÿ 2yN2  …2:15†

aH2 ˆ 3yN ‡ 2yNH ‡ yNH2  ÿ yH …2:16†

aNH3 ˆ ÿ2yN ÿ 2yNH ÿ 2yNH2  ÿ 2yNH3  : …2:17†

The reaction orders are not constants, but depend on the surface composition
(Fig. 5).

Fig. 5. Calculation reaction orders for N2, H2 and NH3 [43] for ammonia synthesis at 10.1 MPa, N:H
ratio 1:3, temperature 673 K.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 79

Fig. 6. Adsorption energy for N2 (left) and N (right) calculated in a Newns-Anderson model. These
data are used as input for calculations presented in Fig. 7.

2.2.5. Applications
Among other applications of the model are calculation of the lifetime for
intermediates [43] modeling of the poisoning by traces of water [43,49] and
interpretation of the promotion by potassium [49].
A simple Newns±Anderson calculation may be used to determine the trend in
enthalpy of formation for the intermediates (Fig. 6). This trend may then be used
in the microkinetic model to investigate the trend in catalytic activity across the
periodic table [44] (Fig. 7).
The model has been augmented to include a model for poisoning by water [43]
(Fig. 8), the details of the measured [59] and calculated [55] sticking dynamics for
N2, the e€ect of N±N interactions [59], adsorbate-induced reconstruction [55], as
well as the repulsion between K and NHx [59].

Fig. 7. Predicted rate of ammonia synthesis across the periodic table [44]. For metals to the left of Fe,
reactivity is limited by a strong adsorption of N  : For metals to the right of Fe, reactivity is limited by
a weak adsorption of N2  :
80 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

Fig. 8. E€ect of water on calculated activation enthalpy at 673 K, 101 kPa (dashed curve) and 10.1
MPa (solid curve). This e€ect was built into the microkinetic model by including the step H2O+ _
O+H2. Resulting complexity neatly explains the lack of consensus on the form of the macroscopic
kinetics under dynamic poisoning by water.

2.3. CO+NO

The reaction between CO and NO

CO ‡ NOÿ
)ÿ
ÿ*
ÿCO2 ‡ 1
2 N2

is catalyzed by Rh in the three-way catalyst. This reaction has been modeled [60±
64] based on the following mechanism
ÿÿ
…a† CO ‡ )ÿ*
ÿCO

…b† NO ‡ ÿ
)ÿ
ÿ*
ÿNO

…c† NO  ‡  4 N  ‡O

…d† CO  ‡O  4 CO2 ‡ 2

…e† NO  ‡N  4N2 ‡ O

…f † 2N  4 N2 ‡ 2  : …2:18†

It is a feature of this reaction that the relevant partial pressures of both CO and
NO for this reaction are rather low and only a moderate extrapolation in pressure
is required from single crystal studies to the real operating conditions.
The closely related reaction
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 81

N2 O ‡ COÿ
)ÿÿ
*ÿCO2 ‡ N2
has been modeled by the following mechanism: [65,66]
…a† CO ‡ ÿ
)ÿ
ÿ*
ÿCO

ÿÿÿ
…b† N2 O ‡ ) *ÿN2 O

…c† N2 O  4 N2 ‡ O

…d† CO  ‡O  4 CO2 ‡ 2  :

2.4. Hydrogen oxidation

The oxidation of hydrogen

H2 ‡ 1
2 O2 ÿ
)ÿÿ
*ÿH2 O

catalyzed by Pt [41,67±70] has been modeled by the following mechanism


…a† H2 ‡ 2  ÿ
)ÿ
ÿ*
ÿ2H

…b† O2 ‡ 2  ÿ
)ÿ
ÿ*
ÿ2O

…c† H  ‡O  ÿ
)ÿ
ÿ*
ÿOH  ‡

…d† H  ‡OH  ÿ
)ÿ
ÿ*
ÿH2 O  ‡

…e† H2 O  ÿ
)ÿ
ÿ*
ÿH2 O ‡ 

…f † OH  4OH ‡ 

…g† 2OH  ÿ
)ÿ
ÿ*
ÿH2 O ‡ O …2:19†

where step (c) is rate limiting for the production of H2O.


This mechanism and simpli®ed versions have been used to model the SS
reaction over Pt [71±86] and Rh [87], as well as the ignition under auto-thermal
conditions over Pt [88,89].

2.5. Carbonmonoxide oxidation

The oxidation of CO
82 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

CO ‡ 1
2 O2 ÿ
)ÿÿ
*ÿCO2

catalyzed by Pt displays the oscillating rate due to reconstructions [90]. The


oscillations have been modeled in a continuum description [91].
The simpli®ed mechanism
…a† CO ‡ ÿ
)ÿ
ÿ*
ÿCO

…b† O2 ‡ 2  4 2O

…c† CO  ‡O  4CO2 ‡ 2 …2:20†

is the archetypical mechanism displaying a kinetic phase transition due to


irreversibility [41,92,93]. Studies include the e€ect of di€usion to di€erent facets
[94] di€usion of adsorbates from support to metal particles [94] and adsorbate±
adsorbate interactions [95].
Microkinetic models have been developed for the oxidation of CO catalyzed by
Pt [96] and Rh [60,97,98]. The reaction is part of the reactions in the three-way
catalyst [60,98].
A more detailed mechanism [99]
…a† CO ‡ ÿ
)ÿ
ÿ*
ÿCO

…b† O2 ‡ 2  ÿ
)ÿ
ÿ*
ÿ2O

…c† CO  ‡O  4CO2 ‡ 2

…d† O2 ‡ ÿ
)ÿÿ
*ÿO2 

…e† O2  ‡  4 O

…f † 2CO  ‡O2  ÿ
)ÿÿ
*ÿCO2 ‡ 3

…g† CO ‡ O  ÿ
)ÿÿ
*ÿCO2 ‡ 

…h† CO ‡ O2  ÿ
)ÿ
ÿ*
ÿCO2 ‡ O …2:21†

has been used in modeling of the ignition at a Pt wire [99±101] and at a Pt(111)
[102].

2.6. HCN synthesis

The Andrussow process


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 83

NH3 ‡ CH4 ‡ 3
2 O2 ÿ
)ÿ
ÿ*
ÿHCN ‡ 3H2 O …2:22†

is catalyzed by Pt in a gauze reactor. The extreme temperature in this reaction


results in negligible coverages, this allowed the reaction to be modeled [103] by a
reaction mechanism consisting of 13 gas-phase reactions.

2.7. Hydrogenation, dehydrogenation and isomerization

The hydrogenation of ethene, C2H4, catalyzed by Pt has been modeled [14,104±


107] using a Horiuti±Polyani (HP) mechanism
1 ÿÿÿ
H2 ‡ 2_) *ÿ2H_

2 ÿÿ
H_ ‡  )ÿ*
ÿH ‡ _

3 ÿÿÿ
C2 H4 ‡ 2  ) *ÿC2 H4 2

4 ÿÿ
C2 H4 2 ‡H)ÿ*
ÿC2 H5 2 ‡

5 ÿÿ
C2 H5 2 ‡H)ÿ*
ÿC2 H6 ‡ 2  ‡

6 ÿÿÿ
H2 ‡ 2  ) *ÿ2H

7 ÿÿ
H  ‡ )ÿ*
ÿH ‡ : …2:23†

In this mechanism , _ and  are di€erent types of adsorption sites. The existence
of three types of site is an ad hoc assumption in the HP mechanism, which serves
to explain why adsorbed hydrocarbons do not block the adsorption of hydrogen.
A similar mechanism has been used to model the dehydrogenation of propane
[108] the hydrogenolysis and isomerization of n-butane [109] and the isomerization
of hexane [96].

2.8. Water±gas shift reaction

The water±gas shift reaction


CO ‡ H2 Oÿ
)ÿÿ
*ÿCO2 ‡ H2
has been modeled [110±115] using the redox mechanism
…a† H2 O ‡ ÿ
)ÿÿ
*ÿH2 O

…b† H2 O  ‡  ÿ
)ÿ
ÿ*
ÿOH  ‡H
84 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

…c† 2OH  ÿ
)ÿ
ÿ*
ÿH2 O  ‡O

…d† OH  ‡  ÿ
)ÿ
ÿ*
ÿO  ‡H

ÿÿ
…e† 2H  )ÿ*
ÿH2 ‡ 2

ÿÿ
…f † CO ‡ )ÿ*
ÿCO

ÿÿ
…g† CO  ‡O  )ÿ*
ÿCO2  ‡

ÿÿÿ
…h† CO2  ) *ÿCO2 ‡ : …2:24†

Among the applications of the model [110,112,113] are calculation of the


coverages and of the reaction rate, as well as interpretation of the activation
enthalpy and of the reactions order.

2.9. Methane oxidation

The possible products of the reaction between CH4 and O2 are HCHO, CO,
CO2 and H2O. A possible mechanism is [116]
…a† H2 ‡ ÿ
)ÿ
ÿ*
ÿH

…b† O2 ‡ ÿ
)ÿ
ÿ*
ÿO

…c† H2 O  ÿ
)ÿ
ÿ*
ÿH2 O ‡ 

…d† OH  ÿ
)ÿ
ÿ*
ÿOH ‡ 

…e† CO ‡ ÿ
)ÿ
ÿ*
ÿCO

…f † CO2 ‡ ÿ
)ÿ
ÿ*
ÿCO  ‡O

…g† CH4 ‡  4CH3  ‡H

…h† CH3  ‡5  4 CH2  ‡H

…i† CH2  ‡5  4CH  ‡H

…j† CH  ‡5  4 C  ‡H
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 85

ÿÿ
…k† H  ‡O  )ÿ*
ÿOH  ‡

…l† H  ‡OH  ÿ
)ÿ
ÿ*
ÿH2 O  ‡

…m† ÿÿ
2OH  )ÿ*
ÿH2 O  ‡O

…n† C  ‡O  ÿ
)ÿÿ
*ÿCO  : …2:25†

Kinetic modeling of this reaction has been performed for reactions over Pt using a
model consisting of four gas-phase reactions [117]. Mechanisms with more steps
have been used for reaction over Pt [71,87,118,119] and for reaction over Rh
[118,119] in a plug-¯ow reactor, as well as for studies of hysteresis phenomena
[120] for reaction over Pd.
The mechanism (2.21) has been used in simulations of the ignition of methane
Pt [116,121±123] and ethane over noble metals [124].

2.10. Methanol synthesis

The methanol synthesis has been modeled [112] by the following mechanism:
…a† H2 O ‡ ÿ
)ÿÿ
*ÿH2 O

…b† H2 O  ‡  ÿ
)ÿ
ÿ*
ÿOH  ‡H

…c† 2OH  ÿ
)ÿ
ÿ*
ÿH2 O  ‡O

…d† OH  ‡  ÿ
)ÿ
ÿ*
ÿO  ‡H

…e† 2H  ÿ
)ÿ
ÿ*
ÿH2 ‡ 2

…f † CO ‡ ÿ
)ÿ
ÿ*
ÿCO

…g† CO  ‡O  ÿ
)ÿ
ÿ*
ÿCO2 

…h† CO2  ÿ
)ÿÿ
*ÿCO2 ‡ 

…i† CO2  ‡H  ÿ
)ÿ
ÿ*
ÿHCOO  ‡

…j† HCOO  ‡H  ÿ
)ÿÿ
*ÿH2 COO  ‡

…k† H2 COO  ‡H  ÿ
)ÿ
ÿ*
ÿCH3 O  ‡O
86 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

…l† CH3 O  ‡H  ÿ
)ÿ
ÿ*
ÿCH3 OH  ‡

…m† CH3 OH  ÿ
)ÿ
ÿ*
ÿCH3 OH ‡ 

ÿÿ
…n† H2 COO  ‡  )ÿ*
ÿHCHO  ‡O

ÿÿ
…o† HCHO  )ÿ*
ÿHCHO ‡ 

ÿÿ
…p† H2 COO  ‡H  )ÿ*
ÿHCHO ‡ OH …2:26†

where steps (b), (d), (h) and (k) are slow. Data for individual adsorbates and
kinetic data measured on Cu(100) [125].
Among the applications of the model are studies of the methanol synthesis rate
for Cu(100) [112,125], the water±gas shift rate for the industrial catalyst [112,126]
and the methanol synthesis rate for the industrial catalyst [112,126], as well as
studies of the coverage by intermediates [112,126], interpretation of the activation
energy [112,126] and of the reaction orders [112,126].
The model has been augmented to include changes in crystallite shape [126,127].

2.11. NO+NH3

The selective reduction of NO by NH3 over V2O5±TiO2


4NH3 ‡ 4NO ‡ O2 4 4N2 ‡ 6H2 O

4NH3 ‡ 4NO ‡ 3O2 4 4N2 O ‡ 6H2 O …2:27†

has been modeled based on a simplistic LH mechanism [128]


…a† NH3 ‡ 4NO ‡ O2 ÿ
)ÿÿ
*ÿ4N2 O ‡ 6H2 O

…b† 4NH3 ‡ 3O2 ÿ


)ÿ
ÿ*
ÿ2N2 ‡ 6H2 O

…c† 4NH3 ‡ 6NOÿ


)ÿ
ÿ*
ÿ5N2 ‡ 6H2 O …2:28†

and a mechanism [129±131] with two types of sites  and _ analogous to an HP


mechanism
…a† NH3 ‡ ÿ
)ÿÿ
*ÿNH3

…b† NH3  ‡_ ÿ


)ÿ
ÿ*
ÿNH3 _ ‡ 

…c† NO ‡ NH3 _ ÿ
)ÿÿ
*ÿProducts ‡ _ : …2:29†
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 87

3. Statistical thermodynamics

The present section deals with the statistical mechanical treatment of chemical
reactions and in particular of chemisorption. The aim is to expose a treatment at
a level of detail sucient for the rest of the present investigation.
The distribution function to be used in the following is the classical limit of the
quantum statistics. The underlying assumptions are:
1. The treatment will be restricted to conditions where the molecules are
equilibrated at a well-de®ned temperature. While the molecules are equilibrated
at the reaction temperature, the system may or may not be in chemical
equilibrium.
2. Interactions between molecules are assumed to consist of pairwise, additive
interactions. These interactions are further assumed to be absent unless an
interaction is explicit in the treatment.
3. Molecules of the same chemical kind are assumed indistinguishable.
4. The temperature is assumed to be moderate. At very low temperatures the use
of quantum statistics would be necessary. At very high temperatures the
interaction of the internal degrees-of-freedom for each molecule must be
included.

3.1. Partition function

If we sum over states, the partition function, Q, is


X  
Ei
Qˆ exp ÿ , …3:1†
i
kB T

where Ei is the energy of the ith state. Alternatively, we may sum over the energies
X  
Ei
Qˆ ri exp ÿ , …3:2†
i
kB T

where ri is the number of states with energy Ei.

3.2. Translation

The theoretical basis for the treatment of translation is the quantum mechanics
of a particle in a box. The translational partition function, Q, for a particle with
mass m and temperature T in a box of volume V is
 3=2
2pmkB T
Qˆ V: …3:3†
h2
The internal energy, U, and the enthalpy, H, are:
88 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

3
Uˆ 2 kB T …3:4†

5
Hˆ 2 kB T: …3:5†

3.3. Rotation

For a free rotor with two axes, the energy Ei can be calculated from the
quantum mechanical treatment of the rotating dumbbell
Ej ˆ Bj… j ‡ 1†, …3:6†

where B is a constant related to the moment of inertia. For a symmetrical top


with three axes, the energy is

Ejk ˆ Bj… j ‡ 1† ‡ …A ÿ B †k 2 …3:7†

where A and B are constants related to the moments of inertia.


The symmetry number, s, enters the formula for Q because the
indistinguishable permutations created by the rotation has to be taken into
account. s is determined from the symmetry of the molecule

Molecular symmetry s

C1, Ci and Cs 1
C2, C2v and C2h 2
C3, C3v and C3h 3

For most gases the moment of inertia is suciently large to allow the calculation
of z from the high-temperature limit. For the linear molecule
kB T
Qˆ for kB T  B, …3:8†
sB

U ˆ H ˆ kB T, …3:9†

and for the symmetrical top

…kB T †3=2
Qˆ for kB T  A, B, …3:10†
sA1=2 B

3
UˆHˆ 2 kB T: …3:11†

H2( g ) is a notable exception, even at very high temperatures Q for H2( g ) must be
calculated from (3.1) and (3.6).
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 89

3.4. Vibration

The vibrational partition function is determined from the quantum mechanical


treatment of the harmonic oscillator

En ˆ hÿ o…n ‡ 12 † for n ˆ 1, 2, . . . …3:12†


1
where o is the frequency of the oscillator. The zero-point energy, 2 hÿ o, is included
in the ground state energy for the molecule.
The partition function is
 
hÿ o
exp ÿ
2k T
Qˆ  Bÿ  , …3:13†
ho
1 ÿ exp ÿ
kB T

while
 
hÿ o
h o exp
ÿ
ÿ
1 2kB T
U ˆ H ˆ hÿ o ‡   : …3:14†
2 hÿ o
1 ÿ exp ÿ
kB T

For a high-frequency vibrational degree-of-freedom, hÿ o  kB T, the partition


function is Q = 1. For a low-frequency vibrational degree-of-freedom hÿ o  kB T
the partition function is
kB T
:
hÿ o

3.5. Ground state energy

The ground state energy, E, contains the contributions from the electronic
energy of the molecule and accounts for the di€erence in chemical stability
between various chemical compounds at T = 0 K.
It is well known that energy di€erences are well-de®ned while absolute values of
the energy do not exist. Normally, this problem is solved by choosing a reference-
state, i.e. 298 K and 101 kPa, and choosing the enthalpy of the elements equal to
zero in this state.

3.6. Thermodynamic functions

The thermodynamic functions are all related to the partition function, Q:


90 P. Stoltze / Progress in Surface Science 65 (2000) 65±150
 
2 d ln Q
U ˆ kB T , …3:15†
dT V

 
2 d ln Q
H ˆ kB T , …3:16†
dT p

 
d ln Q
G ˆ F ‡ kB TV , …3:17†
dV T

F ˆ ÿkB T ln Q, …3:18†

 
d ln…Q†
S ˆ kB ln…Q† ‡ kB T , …3:19†
dT V

   
2
d ln Q 2 d ln Q
Cp ˆ 2kB T ‡kB T , …3:20†
dT p dT 2 p

   
d ln Q 2 d 2 ln Q
Cv ˆ 2kB T ‡kB T : …3:21†
dT V dT 2 V

The chemical potential is related to Q through F


@F
mˆ : …3:22†
@n

3.7. Ideal gas

The basis for the description of the ideal gas is a collection of n molecules in a
volume V. The partition function for the gas is
qn
Qˆ , …3:23†
n!
where q is the partition function for a molecule. From (3.22)
 
@F q
mˆ ˆ ÿkB T ln : …3:24†
@n n

By de®ning q % as the partition function at p=p %


p
q ˆ q , …3:25†
p

we obtain
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 91
 
p
m ˆ m ‡ kB T ln : …3:26†
p

As a check, we might verify the equation of state


 
@Q nkB T
p ˆ kB T ˆ : …3:27†
@T T V

3.8. Equilibrium constant

By introduction of Eq. (3.25) in Eq. (3.22) we obtain the following expression


for the equilibrium constant
Y
ni
Kˆ …q
i † …3:28†
i

 
Sni m
i
K ˆ exp ÿ …3:29†
kB T
where ni is the stoichiometric coecient1 for gas number i.

3.9. Langmuir isotherm for non-dissociative adsorption

The basis for the derivation of the Langmuir isotherm for


A ‡ ÿ
)ÿ
ÿ*
ÿA
is n molecules adsorbed on N sites. The partition function, Q, is
N!
Qˆ qn …3:30†
…N ÿ n†!n!

and the chemical potential is


@F y y
mˆ ˆ ÿkB T ln q ‡ kB ln ˆ m
a ‡ kB ln , …3:31†
@n 1ÿy 1ÿy
where y=n/N is the coverage.
If we assume that the adsorbed molecules are in equilibrium at a gas with
pressure p, we obtain the Langmuir isotherm

1
The stoichiometric coecients are conventionally called n. In the following we will discover ®ve
coecients, a, b, g, d and E, with distinct mathematical properties. As the de®nitions of n in the litera-
ture are not all consistent, we avoid using the symbol n.
92 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

p
K
p
yˆ p , …3:32†
1‡K
p

where
!
m
a ÿ mg

K ˆ exp ÿ : …3:33†
kB T

The important message in this derivation is that the Langmuir isotherm is not the
consequence of a particular kinetics, it is a consequence of the equilibrium.

3.10. Langmuir isotherm for dissociative adsorption

The basis for the derivation of the Langmuir isotherm for dissociative
adsorption
A2 ‡ 2  ÿ
)ÿ
ÿ*
ÿ2A
is 2n molecules on N adsorption sites. The partition function is
N!
Qˆ qn : …3:34†
…N ÿ 2n†!…2n†!

If we assume that the adsorbed molecules are in equilibrium with a gas with
pressure p
mA2 ˆ 2mA , …3:35†

we obtain the Langmuir isotherm


r
p
K
p
yˆ r , …3:36†
p
1‡ K
p

where
!
2m
a ÿ mg
K ˆ exp ÿ : …3:37†
kB T

4. Reaction mechanism

A net reaction, such as


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 93

A2 ‡ 2Bÿ
)ÿÿ
*ÿ2AB,
often consists of a number of steps. Short-lived intermediates may be formed by
some steps and consumed in other steps, e.g.
ÿÿ
A2 ‡ B)ÿ*
ÿA2 B,

ÿÿ
A2 B ‡ B)ÿ*
ÿ2AB:

4.1. Elementary steps

Evidently, we can always subdivide the steps further and introduce hypothetical
intermediates, e.g.
ÿÿ
A2 ‡ B)ÿ*
ÿA2 B,

ÿÿ
A2 B ‡ B)ÿ*
ÿA2 B2 ,

ÿÿ
A2 B2 )ÿ*
ÿ2AB:
This leads to the introduction of the concept of an elementary step. A step in a
reaction mechanism is elementary if it is the most detailed, sensible description of
the step. A step which consists of a sequence of two or more elementary steps is a
composite step. The description of a net reaction as a sequence of elementary
steps is the mechanism for the reaction.
The question if a step in a reaction is an elementary step obviously depends on
how detailed the available information is [46]. The reaction mechanism deduced
from a few, crude measurements of the reaction rate may consist of a small
number of elementary steps. If we then decide to investigate the reaction through
quantum chemical calculations, we will most likely ®nd that many of these steps
are in fact composite. When the level of details cross the border where molecules
are no longer equilibrating at the reaction temperature, the model will become a
reaction dynamic model, rather than a kinetic model. Kinetic methods (Section 5)
and statistical thermodynamic methods (Section 3) will then no longer be
adequate.
The key features of a mechanistic kinetic model [24,132] is that it is reasonable,
consistent with known data and amenable to analysis.

4.2. Consistency

For a proposed reaction mechanism, there must be a sequence of steps that


leads from reactants to products. If this requirement is implicit in the de®nition of
a reaction mechanism (Section 4.1), all intermediates occur as a reactant for at
least one step and as a product for at least one step. This requirement is
94 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

essentially the de®nition of an intermediate. Further, all reaction steps must have
a thermodynamics and all slow steps must have a rate. These requirements lead to
two important consistency checks for a proposed reaction mechanism,
stoichiometric consistency and thermodynamic consistency.

4.2.1. Stoichiometric consistency


In a proposed reaction mechanism, each reactant, intermediate and product
may evidently be consumed or formed in at least one way. However, if a reactant,
intermediate or product participates in two or more steps, the stoichiometry of a
molecule must be independent of the way the intermediate is formed. This is the
principle of stoichiometric consistency.
A simple example of a stoichiometric inconsistency is the following:
ÿÿ
A ‡ B)ÿ*
ÿC,

A ‡ 2Bÿ
)ÿ
ÿ*
ÿC: …4:1†

While it is trivial to decide if a simple reaction mechanism is stoichiometrically


consistent, we want to consider reaction mechanisms containing G gases, S
adsorbates and R steps for arbitrarily large values of G, S and R.
In Sections 8.1 and 9.1 we return to the problem of how to decide if a
mechanism is stoichiometrically consistent.
While most stoichiometrically consistent mechanisms will have S=R, knowledge
of S, R and G is not enough to decide if a proposed mechanism is
stoichiometrically consistent. Here we will present counterexamples to two
intuitive, but false statements.

4.2.2. Example 1
A mechanism with S=R may be stoichiometrically inconsistent. The following
mechanism for the reaction
Aÿ
)ÿ
ÿ*
ÿD
has R = 5, G = 2 (A, B ), S = 5 …, A, B, C, D† and is inconsistent
A ‡ ÿ
)ÿ
ÿ*
ÿA

A‡ÿ
)ÿ
ÿ*
ÿB  ‡C

B  ‡C  ÿ
)ÿÿ
*ÿD  ‡

ÿÿÿ
B  ‡2C  ) *ÿD  ‡2
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 95

Dÿ
)ÿ
ÿ*
ÿD ‡ :

4.2.3. Example 2
A mechanism with S > R may be stoichiometrically consistent. The following
mechanism for
ÿÿ
A)ÿ*
ÿF
has S = 4 and R = 3 and is stoichiometrically consistent
ÿÿ
A ‡ )ÿ*
ÿB  ‡C

ÿÿÿ
B  ‡C  ) *ÿD  ‡E

ÿÿÿ
D  ‡E  ) *ÿF ‡ :

4.2.4. Thermodynamic consistency


As a reaction mechanism contains a path from reactants to products, a
description of the gas-phase thermodynamics is implicit in a reaction mechanism.
If two or more di€erent sequences of steps lead from reactants to products, these
sequences must describe the same gas-phase thermodynamics. This is the principle
of thermodynamic consistency.
Stoichiometric consistency may be viewed as more fundamental than thermo-
dynamic consistency as stoichiometric inconsistency always implies thermodynamic
inconsistency. However, the following example demonstrates that a mechanism
may be stoichiometrically consistent and thermodynamically inconsistent
A ‡ Bÿ
)ÿ
ÿ*
ÿC with K ˆ 1,

A ‡ Bÿ
)ÿ
ÿ*
ÿC with K ˆ 10:

As for the question of stoichiometric consistency, the question of thermodynamic


inconsistency must be investigated systematically, for larger mechanisms (Sections
8.1 and 9.1).

4.3. Net reaction

In a reaction mechanism there may be more than one sequence of reaction steps
from one of the reactants to one of the products. Methanol synthesis (Section
2.10) is an example of such a reaction mechanism. This reaction appears to consist
of two reactions
CO2 ‡ 3H2 ÿ
)ÿ
ÿ*
ÿCH3 OH ‡ H2 O, …4:2†
96 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

CO ‡ H2 Oÿ
)ÿÿ
*ÿCO2 ‡ H2 : …4:3†

We might try to describe the mechanism using two net reactions. This is not very
useful as these rates are mutually dependent. Alternatively, we might try
describing the mechanism as an unstoichiometric reaction
ÿÿ
…1 ÿ x†CO2 ‡ …3 ÿ 2x†H2 ‡ xCO)ÿ*
ÿCH3 OH ‡ …1 ÿ x†H2 O, …4:4†

or as a stoichiometric reaction with a conversion dependent stoichiometry. Neither


of these suggestions are of any help in solving the problem.
The explanation for this diculty is the following: reactions (4.2) and (4.3)
happen to contain two stoichiometric reactions and this is a special case. For a
large mechanism it is dicult to decide how many stoichiometric net reactions are
in the mechanism and dicult to ®nd those equations.

4.4. Langmuir±Hinshelwood mechanisms

The simplest class of reaction mechanisms for catalytic reactions are LH


mechanisms [23,24,28,133±142]. These mechanisms consist of just three types of
reaction steps:
1. adsorption of molecules from the gas phase;
2. reaction between adsorbed molecules;
3. desorption of adsorbed molecules;
where the reaction proceeds on a surface consisting of a constant number of sites
and each site is occupied by zero or one molecule. The neglect of the surface
di€usion phenomena implies that surface di€usion must be fast.
While the following treatment is restricted to LH mechanisms in their simplest
form, there is no fundamental diculty in lifting one or more of the assumptions.
The treatment may be augmented to include a variable number of surface sites
due to reconstructions [93,143±145] formation of steps [146] roughening [145,147]
or reshaping of supported particles [94,148,149]. Adsorbate±adsorbate interactions
[95,150±158] formation of adsorbate islands [152,146] and adsorption of two or
more molecules at each site [150,151,158±162], as well as a ®nite rate of surface
di€usion [94,153] might be included.
The validity of LH models and the reason for their success has been the subject
of long and heated discussions [24,25,28,138,140,141].
One possibility which should not be overlooked is that the LH treatment
actually gives an adequate description of the essential physics. LH certainly
contains a description of adsorbates competing for adsorption sites based on
thermodynamics stability. Further, the limits for the reaction rate at zero and at
in®nite pressures are at least qualitatively correct.
It has been suggested [163] that LH mechanisms are quite general and that
apparent failures may be due to autocatalysis. For example, for A) ÿÿÿ
*ÿ 12 B ‡ C
with step (a) rate limiting
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 97

…a† A ‡ 2  ÿ
)ÿÿ
*ÿB  ‡C

…b† A  ‡B  ÿ
)ÿÿ
*ÿ2B  ‡C

ÿÿÿ
…c† 2B  ) *ÿB2 ‡ 2 …4:5†

Interestingly, the mechanism for the water±gas shift reaction (Section 2.8) and for
the methanol synthesis (Section 2.10) contains such an autocatalytic step.
Another explanation [141] for the success of the LH treatment is that many
catalytic reactions happen to proceed at high coverages by intermediates. At these
conditions the assumptions in the LH treatment are more or less correct.
A third possible explanation is that approximately the same quality of ®t may
be obtained using very di€erent kinetic expressions [25]. An implication of this
explanation is that the apparent success of the LH models is accidental, most
other models would perform equally well.

4.5. Complications

There are a number of features a reaction mechanism may have, which greatly
complicates the situation.
4.5.1. Kinetic equations are non-linear
For mechanisms where all steps consist of unimolecular reaction steps, the
kinetics of the reaction is available analytically for arbitrarily large mechanisms
[164,165]. However, kinetic expressions for elementary steps are not necessarily
®rst-order in the concentration of reactants. In a mechanism consisting of several
steps, steps may even have a di€erent same order.
4.5.2. Inerts
An inert adsorbate does not have a well-de®ned chemical potential and, if inert
surface species are present, the model is not soluble without additional
assumptions on the behavior of the inert.
Adsorbed inerts with constant coverage are better described by an adjustment
of the number of adsorption sites. Adsorbed inerts with variable coverage are
better described as reactants.

4.5.3. Non-consecutive steps


The reaction mechanism does not necessarily consist of a sequence of
consecutive steps. Apart from the trivial case where consecutive steps are written
in random order, some more interesting possibilities are:
. One or more steps have been written backwards, e.g. step (b) in the mechanism:
…a† A2 ‡ ÿ
)ÿÿ
*ÿA2 

…b† 2A  ÿ
)ÿÿ
*ÿA2 
98 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

…c† ÿÿ
B ‡ )ÿ*
ÿB

ÿÿ
…d † A  ‡B  )ÿ*
ÿAB ‡ : …4:6†
. The mechanism has been written such that all steps except one have the form
ni1 A1 ‡ ni2 A2 ‡    ‡  ˆ mi1 B1 ‡ mi2 B2 ‡    C  , …4:7†

where A1, A2, . . ., B1, B2, . . ., are all gases and C is an adsorbed molecule.
. Parallel steps which convert the same reactants into the same products through
di€erent routes.

4.5.4. Elusive intermediates


In a reaction mechanism, short-lived intermediates may be formed by some
steps and consumed by other steps. The mechanism may contain intermediates,
which have not been observed experimentally.
The introduction of a hypothetical intermediate in the mechanism is in many
cases a necessity to link the observed intermediates formed from the reactants
with the observed intermediates formed from the products. If the calculated
concentration of the hypothetical intermediate is too small and the lifetime too
short to allow the experimental observation, the introduction of the hypothetical
intermediate is of no consequence for the agreement between the model and
experimental results.
The introduction of hypothetical intermediates in excess of the absolutely
necessary is not sensible.

4.5.5. Undetectable steps


A mechanism may contain steps that are irrelevant as they are of no
consequence for the consistency of the mechanism and of no consequence for the
reaction rate. Some examples of undetectable steps are:
. A slow reaction step that is short circuited by a sequence of equilibrium steps.
The net rate of the slow step is then zero.
. A fast step in series with a slow step.
. A slow step in parallel to a fast step.

4.5.6. Dead ends


One or more steps may form a dead end in the form of an intermediate formed
through an elementary reaction and consumed exclusively by the reverse of this
step. Although the dead-end will not contribute to the overall reaction rate, the
step may a€ect the kinetics if the intermediate is strongly adsorbed on the surface.
The poisonous e€ect of H2O in ammonia synthesis is an example (Section 2.2).
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 99

4.5.7. Linear dependence between reaction steps


A reaction mechanism may have linearly dependent reaction steps. This may
happen for two reasons.
First, the same reaction step occurs more than once with di€erent kinetic
parameters, e.g.

A  ‡B  ÿ
)ÿÿ
*ÿAB  ‡  with A ˆ 109 and E y ˆ 6 kJ mol ÿ1 , …4:8†

ÿÿÿ
A  ‡B  ) *ÿAB  ‡  with A ˆ 1013 and E y ˆ 52 kJ mol ÿ1 , …4:9†

where (4.8) describes a low barrier, low temperature channel and (4.9) describes a
high barrier, high temperature channel.
Second, steps may be combined, such as the following steps that occur in the
water±gas shift reaction (Section 2.8)
ÿÿÿ
…a† H2 O ‡ ) *ÿH2 O

ÿÿ
…b† H2 O  ‡  )ÿ*
ÿOH  ‡H

ÿÿ
…c† OH  ‡  )ÿ*
ÿH  ‡O

…d† ÿÿ
2OH  )ÿ*
ÿH2 O  ‡O  : …4:10†

Although linear dependence among the reaction steps complicates the analysis of
a mechanism, the mechanism is physically meaningful provided the mechanism is
stoichiometrically and thermodynamically consistent.

5. Kinetics

The most important aspect of a reaction mechanism is the reaction rate. The
rate of chemical reactions can be described at two levels: dynamics and kinetics.
Dynamics [142,166,167] is the description of the rate of transformation for
individual molecules. The molecule has a well-de®ned energy, it may even start in
a well-de®ned quantum state. There is no temperature as temperature is a
property of a large number of molecules, not of individual molecules.
Kinetics [41,135,137,158,168±173] is the description of the rate of reaction for a
large number of molecules. The molecules have a temperature, although the
temperature may change in the course of the reaction. The energy is well-de®ned,
but the energy is a statistical average.

5.1. Reversibility

For a reaction, say


100 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

A2 ‡ 2B 42AB, …5:1†

the reverse reaction


2AB 4A2 ‡ 2B …5:2†

will proceed through the same mechanism, although the sequence and the
direction of each of the elementary reactions is reversed [174]. This is known as
the principle of microscopic reversibility.
The cause of this principle is that in the kinetic description, we explicitly assume
that the intermediates equilibrate at the reaction temperature. This implies that
the intermediates have no memory of how they are formed. A formed by
dissociation of AB is identical to A formed by dissociation of A2.
The principle of microscopic reversibility has to be used with some care. While
the principle states that the mechanism for reaction (5.1) is the same as the
mechanism for reaction (5.2) in a mixture of A2, B and AB at a given set of
concentrations, temperature and pressure. However, the principle is not necessarily
valid when two di€erent sets of reaction conditions are compared.
As the reaction proceeds through the same steps in the forward and in the
backward reaction, while the rate of the individual steps may di€er by many
orders-of-magnitude it is convenient to consider two classes of steps. Fast steps
have a high rate in both forward and backward directions, while slow steps have a
low rate in the forward direction, in the backward direction or both.
Although the concept of fast and slow reactions is convenient and important,
there is a hidden assumption. A step that appears fast at a timescale of seconds in
an isotopic labeling experiment may appear awfully slow at a timescale of pico-
seconds in a spectroscopic experiment.

5.2. Rate laws

In the simplest case the reaction rates are proportional to the coverages [133±
137,158,170,172], e.g. for the mechanism
A2 ‡ 2  ÿ
)ÿ
ÿ*
ÿ2A

B ‡ ÿ
)ÿ
ÿ*
ÿB

A  ‡B  ÿ
)ÿÿ
*ÿAB ‡ 2  , …5:3†

the rates are


pA 2 2 2
r1 ˆ k‡1 y ÿ kÿ1 yA , …5:4†
p

pB
r2 ˆ k‡2 y ÿ kÿ2 yB , …5:5†
p
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 101

pAB 2
r3 ˆ k‡3 yA yB ÿ k‡3 y : …5:6†
p

The assumption that rates are proportional to coverage eliminates some, but not
all, hysteresis phenomena [41].
However, there is an assumption hidden in (5.14). The equation presumes that
the occupation probabilities are the same for each site and that the occupation
probability for the sites is identical to the coverage for the surface as a whole.
This assumption will break down in the presence of signi®cant adsorbate±
adsorbate interactions.

5.3. Rate constant

At equilibrium the net rate is by de®nition zero. If we compare the rate


equation for an elementary step, e.g. the rate equation
pA 2 2 2
r1 ˆ k‡1 y ÿ kÿ1 yA , …5:7†
p

and the equilibrium equation


pA2 2 2
K1 y ˆ yA …5:8†
p

for the step A2 ‡ 2  ÿ )ÿÿ


*ÿ2A, we ®nd that the forward and backward rate
constants are related [175], i.e.

kÿ ˆ : …5:9†
K
Usually both k+ and kÿ have the Arrhenius form
 
Hz
k ˆ A exp ÿ , …5:10†
RT
at least for small or moderate variations of T. In this equation, A is the pre-
exponential factor, while H % is the activation energy

2d ln k
H z ˆ RT :
dT
If k+ and kÿ have an Arrhenius form, the equilibrium constant, K

Kˆ …5:11†
kÿ
will have an Arrhenius form as will any product or quotient of rate and
equilibrium constants. The activation energies and the energy of reaction will be
related by
102 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

H zÿ ˆ DH ‡ H z‡ :

The Arrhenius form is essentially the only plausible form for the rate and
equilibrium constants which is invariant with respect to the ambiguity if a reaction
step is elementary or composite (Section 4.1).

5.4. Stoichiometric matrix

For a systematic treatment of arbitrarily large LH mechanisms, we need a


suitable mathematical device. One possibility is to use a stoichiometric matrix to
represent the mechanisms in symbolic form.
We write the reaction mechanism using a stoichiometric matrix, a. For a
mechanism consisting of G gases, S adsorbates including free sites, and R
reactions, a is an R  (G+S ) matrix. The components2 are the stoichiometric
coecients, arc, 1 R r R R, 1 R c R G+S.
It is implicit in this de®nition that we base the stoichiometry on the
stoichiometry of the reactions rather than the stoichiometry of the molecules. This
feature has profound implications for the following treatment and allows us to
describe the stability and reactions even if two or more molecules happen to have
the same stoichiometry.
Molecules with rather di€erent chemical properties may have the same
stoichiometry. For example, acetaldehyde, vinylalcohol and ethyleneoxide all have
the stoichiometry C2H4O. At least one attempt [168,169] at a systematic treatment
of chemical kinetics is fatally ¯awed by its inability to distinguish between
di€erent molecules with the same stoichiometry.
The stoichiometric coecients, arc, for the steps in the mechanism should not be
confused with the stoichiometric coecient, ac, in the net reaction or the
stoichiometric number, br (Section 8.3).

5.4.1. Conventions
We use the convention that arc < 0 if c is a reactant of step r, arc > 0 is a
product of step r and arc=0 if c does not participate in step r.
We will frequently need products or sums running over subsets of the
molecules. We will use the convention that the molecules are enumerated with
gases number 1, . . ., G, free sites is number G + 1 and adsorbates are number G
+ 2, . . ., G+S.

5.4.2. Example
As an example we will consider the mechanism (2.1) for ammonia synthesis. We

2
We write the stoichiometric number as arc, where r is reaction or row, c is compound or column. This
conception is the simplest as the matrix a then resembles a reaction scheme written with one step per
line.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 103

include Ar in the gas phase to illustrate the e€ect of inerts. This mechanism is rich
enough to illustrate most of the features discussed below.
For this mechanism R = 7 (steps 1±7), G = 4 (N2, H2, NH3 and Ar), S = 7
…, N2 , N, NH, NH2, NH3 and H).
The stoichiometric matrix is
1 2 3 4 5 6 7 8 9 10 11
1 ÿ1 0 0 0 ÿ1 1 0 0 0 0 0
2 0 0 0 0 ÿ1 ÿ1 2 0 0 0 0
3 0 0 0 0 1 0 ÿ1 1 0 0 ÿ1
…5:12†
4 0 0 0 0 1 0 0 ÿ1 1 0 ÿ1
5 0 0 0 0 1 0 0 0 ÿ1 1 ÿ1
6 0 0 1 0 1 0 0 0 0 ÿ1 0
7 0 ÿ1 0 0 ÿ2 0 0 0 0 0 2

5.4.3. Properties
arc has a number of interesting properties:
1. Surface sites are conserved
X
R
rr arc ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S: …5:13†
rˆ1

2. All elements in the stoichiometric matrix are integers. The use of non-integer
stoichiometric coecients is unnecessary and greatly complicates the treatment.
3. Gas inerts have arc=0 for r = 1, . . ., R.

5.5. Rate

The rate is calculated as a turnover frequency [173], i.e. as a number of


molecules produced3 per site per second.
The reaction rate for step r is
rr ˆ r‡r ÿ rÿr , …5:14†

G 
Y  Y
pc ÿarc G‡S
r‡r ˆ kr …yrc † ÿarc , …5:15†
cˆ1
p cˆG‡1
arc <0 arc <0

3
The distinction between reactants and products depends on which gases are present in the initial
mixture. From now on, we use the word product for all gases when information on the initial mixture
is unknown or irrelevant.
104 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

G   Y
kr Y pc arc G‡S
rÿr ˆ …yrc †arc : …5:16†
Kr cˆ1 p cˆG‡1
arc >0 arc >0

For each of the gases, we have a formation rate


X
R
rc ˆ rr arc , …5:17†
rˆ1

which is evidently negative for the reactants.

5.6. Equilibrium equation

As a consequence of this choice of sign for arc, the equilibrium constants are
G 
Y  Y
pc arc G‡S
Kr ˆ yac rc for r ˆ 1, . . . , R: …5:18†
cˆ1
p cˆG‡1

5.7. Stoichiometric consistency

As all reaction steps have been written as adsorptions, surface reactions or


desorption and as the reaction mechanism must be cyclic with respect to
adsorbates, the question if the mechanism is stoichiometrically consistent is
answered by examining if the following equation has a solution, b
X
R
arc br ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S: …5:19†
rˆ1

We have taken the liberty of using the name b. If the mechanism has a single
RLS, br is the stoichiometric number for reaction r (Section 8.3).

5.8. Thermodynamic consistency

The question if the mechanism is thermodynamically consistent is answered by


determining if the equilibrium equation
G 
Y arc G‡S
Y
ptot
x
c
yac rc ˆ Kr for r ˆ 1, . . . , R, …5:20†
cˆ1
p cˆG‡1

X
G
x c ˆ 1 with 0Rx c R1 for 1RCRG, …5:21†
cˆ1
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 105

X
G‡S
yc ˆ 1 with 0Ryc R1 for 1RcRS, …5:22†
cˆG‡1

has a solution. The requirements 0 R xc R 1 and 0 R yc R 1 greatly complicates


the computation.

5.9. Forward and backward rate

For each step, the rate


r ˆ r ‡ ÿ rÿ …5:23†

is obviously the di€erence between a forward rate, r+, and a backward rate, rÿ.
For each of the gases, we have a formation rate (Section 5.4.3). For many
applications, it is sucient to determine the rate of formation for each of the
gases. However, if we want to determine the kinetic parameters for the net
reaction, the form of the rate expression must be determined.
For an elementary reaction, the forward and backward rates are related
[175,176]
rr ˆ rr‡ …1 ÿ br †, …5:24†

where
G   Y
1 Y pc arc G‡S
br ˆ …yrc †arc …5:25†
Kr cˆ1 p cˆG‡1

is a measure of the approach to equilibrium.


If the reaction mechanism corresponds to a single net reaction with
stoichiometric coecients a1, . . ., aG, b may be de®ned for the net reaction
G  
1X pc a c
bˆ : …5:26†
K cˆ0 p

As for the relation between rate and equilibrium constants, it is important that the
reaction is elementary. If we again consider mechanism (5.3)
 
pAB 2
p
bˆ   : …5:27†
pA pB 2
K
p p

If step (a) is rate limiting


r‡ ˆ r…1 ÿ b†, …5:28†

and if step (b) is rate limiting


106 P. Stoltze / Progress in Surface Science 65 (2000) 65±150
p
r‡ ˆ r…1 ÿ b†: …5:29†

5.10. Kinetics versus equilibrium

It is important that the reaction is elementary. If we consider the mechanism


(5.3) the equilibrium constant for the net reaction A2+2B _ 2AB is obviously

K ˆ K1 K 22 K 32 : …5:30†

If step (a) is rate limiting


0   1
pAB 2
B C
B pA p C 2
B
r ˆ k1 B ÿ  2 C
p pB C y , …5:31†
@ K1 K 22 K 32 A
p

k‡ ˆ k1 , …5:32†

k1
kÿ ˆ : …5:33†
K
If step (b) is rate limiting
0 1
pAB
B pB p C
r ˆ k2 B
B p ÿ p r C y , …5:34†
@ pB C A
K1 K2 K3
p

k‡ ˆ k2 , …5:35†

k2
kÿ ˆ p : …5:36†
K
While (5.33) is valid for an elementary step, comparison of (5.33) and (5.36)
shows that, for composite steps, the relation between the rate constants and the
equilibrium constants depends on the details of the mechanism.

5.11. Rate limiting step

It is often the situation that most reaction steps in a mechanism are fast, while
a single step is much slower than any other step. In this situation, the slow step is
called the RLS or the rate controlling step as it determines the rate of the overall
reaction.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 107

While the concept of an RLS is intuitive, some of the diculties in this concept
may be illustrated by the mechanism (5.3). Under SS conditions a mass balance
for the intermediates A and B shows that steps (b) and (c) must have exactly
twice the rate of step (a). The actual value of the rate is determined by the RLS,
but the mass balance imposes this rate on the other reaction steps.
The degree of rate limitation, xr, imposed on the mechanism by a slow step, r,
can obviously be de®ned [45] as the increase in the overall rate, r, caused by a
hypothetical increase of the rate constant, kr, for the step
kr dr d ln…r†
xr ˆ ˆ : …5:37†
r dkr d ln…kr †

5.12. Net rate

In a reaction mechanism, there may be more than one sequence of reaction


steps from one of the reactants to one of the products. The mechanism evidently
has a de®nite rate of production for each of the gases. However, as we cannot
readily determine if a reaction mechanism describes a stoichiometric net reaction
or a linear combination of several stoichiometric reactions, we have a problem
de®ning the rate for the mechanism. If the mechanism does not correspond to a
stoichiometric net reaction, the rate of consumption or production for each of the
reactants and products will in most cases have no simple relation to the rate for
any other reactant or product. If the mechanism happens to describe a
stoichiometric net reaction, the situation is di€erent (Section 8.4).
Mechanisms with a single RLS are much easier to work with than more general
reaction mechanisms. For this reason the more general reaction mechanisms are
ignored without comment by many authors.

5.13. Activation energy

While the Arrhenius form of the rate constant for an elementary step is easy to
justify (Section 5.3), the situation is more complicated for composite steps or even
complete reaction mechanisms.
The complexity occurs because even if all the rate constants for the elementary
steps have an Arrhenius form, the net rate of the mechanism will in general not
have an Arrhenius form. We can force the net rate to have an Arrhenius form
only if we allow the apparent activation energy to depend on the reaction
conditions.
Only in the case that the activation energy is more or less independent of the
reaction conditions can we determine the activation energy for the mechanism and
only in this case can the activation energy be interpreted as an energy barrier.
Fortunately, experience shows that this case is occurring quite commonly. For
most reactions of practical interest, the dependence of the activation energy on the
reaction conditions is weak except at extreme reaction conditions.
108 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

The question is then how to de®ne the activation energy for a composite
reaction. The obvious answer is that as the activation energy is determined
experimentally through an Arrhenius plot, ln k versus 1/T, we should use the
Arrhenius plot to de®ne the activation energy for composite reactions.
A composite reaction usually has a kinetics of the form (5.15). From this we
can determine the activation enthalpy as

2d ln k‡ d ln r‡
H z ˆ kB T ˆ kB T 2 : …5:38†
dT dT
It is important that it is r+ and not r which enters (5.38) as r contains a
contribution from the reverse rate. If the rate has the form (5.24) the relation is

2d ln r d ln r‡ b
kB T ˆ kB T 2 ‡ DH,
dT dT 1ÿb
where DH is the enthalpy of reaction and b is de®ned by (5.26).

5.14. Reaction order

Analogous to the situation in the previous section, the net reaction will not in
general have the form (5.15). However, we can force the rate to have this form if
we allow the reaction orders, ac, to depend on the reaction conditions.
The reaction order, ac, is then related to the rate by
d ln r‡
ac ˆ  : …5:39†
pi
d ln
p

It is again (Section 5.13) important that it is r+ and not r which enters.


As the net reaction does not in general have the form (5.15) the reaction order
is unrelated to the stoichiometric coecients of the net reaction.

5.15. Plug-¯ow reactor

We want to calculate the mole fractions, x1, . . . , xG, coverages yG + 1, . . . ,


yG+S, and ¯ow F [mol sÿ1] through a catalyst bed operating at temperature T [K]
and pressure p [Pa].
Fig. 9 shows the principle in calculation of conversion in a plug-¯ow reactor.
The parameters for the catalyst is the porosity E, the skeletal density r [kg mÿ3],
the density of sites rs [mol mÿ2] and the speci®c area a [m2 kgÿ1]. The parameters
for the catalyst bed is the cross sectional area A [m2] and the length L [m].
The mass of the bed is
m ˆ LA…1 ÿ E†r:

A slice of length dl has mass


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 109

Fig. 9. Principle in calculation of conversion in a plug-¯ow reactor: ¯ow and composition of the gas-
phase changes continuously through the bed. The di€erential equation for calculation of the
concentration pro®le is derived from consideration of a slice of thickness dl.

dm ˆ A…1 ÿ E†r dl

contains a gas volume


dVg ˆ AE dl

and the amount of gas


p Ep
dn ˆ dVg ˆ dm:
RT RT…1 ÿ E†r

The contact time


dVg E
dt ˆ ˆ dm:
F Fr…1 ÿ E†

The rate of reaction [mol sÿ1] for gas g in the slice is


X
R
rs a rr arg dm:
rˆ1

5.15.1. Isothermal plug-¯ow reactor


The mass balance for the slice are

In + Produced = Out
P
1 Fx 1 dt + rs a Rrˆ1 rr ar1 dm dt = …F ‡ dF †…x 1 ‡ dx 1 †dt
 +  P = 
G Fx G dt + rs a Rrˆ1 rr arG dm dt = …F ‡ dt†…x G ‡ dx G †dt
P PR
sum F dt + rs a Gcˆ1 rˆ1 rr arc dm dt = …F ‡ dF †dt

and the energy balance is


X
G X
G X
R X
G
dQ_ dt ‡ F hc dt ‡ rs a hc rr arG dm dt ˆ …F ‡ dF † …x c ‡ dx c †hc dt:
cˆ1 cˆ1 rˆ1 cˆ1

The mole fractions add up to unity


110 P. Stoltze / Progress in Surface Science 65 (2000) 65±150
X
x i ˆ 1, …5:40†
i

X
…x i ‡ dx i † ˆ 1: …5:41†
i

From the sum of the mass balances we have

dF XG X R
ˆ rs a rr arG m: …5:42†
dm cˆ1 rˆ1

Substituting (5.42) into the mass balances for each of the compounds we ®nd
X
R X
G X
R
rr ar1 ÿ x 1 rr arc
dx 1 rˆ1 cˆ1 rˆ1
ˆ rs a …5:43†
dm F

 ˆ 

X
R X
G X
R
rr arG ÿ x G rr arc
dx G rˆ1 cˆ1 rˆ1
ˆ rs a : …5:44†
dm F
Substituting Eqs. (5.42)±(5.44) into the energy balance gives
…m X
G X
R
Q_ ˆ rs a hc rr arG : …5:45†
0 cˆ1 rˆ1

Even for simple reactions an analytical solution is impossible. Instead we use an


ordinary di€erential equation (ODE) solver to integrate the system (5.42)±(5.44).
The boundary values are the inlet ¯ow and inlet concentrations. The heat ¯ow
may be calculated from (5.45).
As (5.45) is not required for the calculation of rate, conversion or . exit
concentrations, we will simply ignore it unless we are required to calculate Q. This
explains the relation between the isothermal plug-¯ow reactor with and without
the energy balance.

5.15.2. Adiabatic plug-¯ow reactor


The molar enthalpy (J molÿ1) of compound i is hi at temperature T and hi+cpi
dT at temperature T + dT. As the temperature will now change through the
reactor, the reaction enthalpy DH is no longer constant through the reactor.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 111

The mass balances for the slice are

In + Produced = Out
P
1 Fx 1 dt + rs a… Rrˆ1 rr ar1 †dm dt = …F ‡ dF †…x 1 ‡ dx 1 †dt
 +  P = 
G Fx G dt + rs a… Rrˆ1 rr arG †dm dt = …F ‡ dt†…x G ‡ dx G †dt
PG P R
sum F dt + rs a… cˆ1 rˆ1 rr arc †dm dt = …F ‡ dF †dt

and the energy balance is


!
X
G X
G X
R X
G
F x c hc ‡ rs a hc rr arG dm ˆ …F ‡ dF † …x c ‡ dx c †…hc ‡ cpc dT †dt:
cˆ1 cˆ1 rˆ1 cˆ1

From the sum of the mass balances, we obtain

dF XG X R
ˆ rs a rr arc : …5:46†
dm cˆ1 rˆ1

Substituting this equation into the mass balances for each of the compounds
yields
X
R X
G X
R
rr ar1 ÿ x 1 rr arc
dx 1 rˆ1 cˆ1 rˆ1
ˆ rs a …5:47†
dm F

 ˆ 

X
R X
G X
R
rr arG ÿ x G rr arc
dx G rˆ1 cˆ1 rˆ1
ˆ rs a : …5:48†
dm F
Substituting (5.46)±(5.48) into the energy balance leads to
X
G X
R
hc arc
dT cˆ1 rˆ1
ˆ rs a : …5:49†
dm X
G
F x c cpc
cˆ1

The conversion through the catalyst bed is found by integrating the system (5.46)±
(5.49) using an ODE solver. The boundary values are the inlet ¯ow, inlet
temperature and inlet concentrations.
112 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

6. Approximations

We consider an LH mechanism with stoichiometric matrix arc. The kinetic


model for this mechanism consists of the forward and backward rates (5.14), i.e.

G 
Y  Y
pc ÿarc G‡S
r‡r ˆ kr …yrc † ÿarc , …6:1†
cˆ1
p cˆG‡1
arc <0 arc <0

G   Y
kr Y pc arc G‡S
rÿr ˆ …yrc †arc , …6:2†
Kr cˆ1 p cˆG‡1
arc >0 arc >0

and the balance over sites

X
G‡S
yc ˆ 1: …6:3†
cˆG‡1

Table 1 shows the performance of kinetic approximation.


The solution to this system of equations evidently gives a correct description of
both equilibrium, steady state and transient behavior.
If the reaction mechanism contains more than one or at most two steps, the full
solution becomes very complicated and we will have to solve for the rates and
coverages by numerical methods. Although the full solution contains the SS
behavior as a special case, it is not generally suitable for studies of the steady state
as the transients may make the simulation of the steady state a numerical
nightmare.

Table 1
Performance of kinetic approximationsa

None SS QE IS MARI

Transient behaviour + ÿ ÿ ÿ ÿ
Steady state behaviour + + + ÿ +
Rate + + + ÿ
Coverages + + + ÿ
Gas-phase equilibrium + + + ÿ +
Change of RLS + + ÿ ÿ
Reaction order + + + ÿ
Activation energy + + + ÿ

a
Symbols for approximations are None for full solution (Section 6), SS for steady-state approxi-
mation (Section 6.1), QE for quasi-equilibrium approximation (Section 6.2), IS for irreversible step ap-
proximation (Section 6.3) and MARI for MARI approximation (Section 6.4). Symbols for performance
are + or ÿ for mostly correct or incorrect description, respectively.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 113

In the present section we will explore the hierarchy of approximations


applicable to the solution of (6.1)±(6.3) in various situations.

6.1. Steady-state approximation

In the SS approximation, the net rate of formation for all intermediates is


explicitly set to zero
X
R
rr arc ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S: …6:4†
rˆ1

Assuming that the net rate of formation for intermediates is zero does not imply
that the coverage by the intermediates is small. For a mechanism with S ÿ 1
intermediates in addition to free sites, (6.1)±(6.3) is a system of S equations, at
most (S ÿ 1) are linearly independent.
The SS approximation eliminates transient behavior in the kinetics. However, it
is only the transient behavior of the rates and coverages that has been eliminated.
The expression for the rate obtained through the SS approximation is perfectly
suitable for the simulation of, e.g. the conversion through a catalyst bed or most
aspects of the transient behavior of a reactor.

6.2. Quasi-equilibrium approximation

If all steps except one are fast, we can use the quasi-equilibrium approximation.
For the fast steps we use the corresponding equilibrium equations instead of the
kinetic equations:
G 
Y  Y
pc a1c G‡S
K1 ˆ yac 1c , …6:5†
cˆ1
p cˆG‡1

 ˆ 

G 
Y  Y G   Y
pc ÿarc G‡S kr Y pc arc G‡S
rr ˆ kr …yrc † ÿarc ÿ …yrc †arc , …6:6†
cˆ1
p cˆG‡1
Kr cˆ1 p cˆG‡1
arc <0 arc <0 arc >0 arc >0

 ˆ 

G 
Y  Y
pc aRc G‡S
KR ˆ yac Rc : …6:7†
cˆ1
p cˆG‡1

This approximation will, in most cases, provide a very signi®cant simpli®cation in


particular for large reaction mechanisms. In the quasi-equilibrium approximation
114 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

the transient behavior is eliminated. Further, the description of changes in the


RLS has been lost.

6.3. Irreversible step approximation

In the irreversible step approximation, we neglect the forward or backward rate


for one of the steps. For small mechanisms, the irreversible step approximation
may be used alone, for larger mechanisms it is usually combined with the quasi-
equilibrium approximation
G 
Y  Y
pc a1c G‡S
K1 ˆ yac 1c …6:8†
cˆ1
p cˆG‡1

 ˆ ,

G 
Y  Y
pc ÿarc G‡S
rr ˆ kr …yrc † ÿarc , …6:9†
cˆ1
p cˆG‡1
arc <0 arc <0

 ˆ 

G 
Y  Y
pc aRc G‡S
KR ˆ yac Rc : …6:10†
cˆ1
p cˆG‡1

This approximation is very crude, as we have lost the description of the approach
to equilibrium. The model will not continue to convert all reactants right across
the equilibrium, often ending in a spectacular numerical instability when the
concentration of a reactant becomes negative.
Although this approximation is useless for the quantitative modeling of
reactions, it has two important uses in the analysis of reaction mechanisms:
. If we want to determine the limiting behavior of a kinetic model very far from
equilibrium, the irreversible step approximation is the appropriate limit.
. If we have diculties making sense of a complicated reaction mechanism, the
irreversible step approximation may provide a simpli®cation, which allows us to
understand the mechanism well enough to choose a better approximation.

6.4. MARI approximation

The most-abundant reaction intermediate (MARI) approximation is a further


development of the quasi-equilibrium approximation. Often one of the
intermediates is much more abundant than all other intermediates, the coverages
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 115

by the less-abundant intermediates may then be neglected in the balance over


sites.
If we assume that A is the most stable intermediate and the balance over sites
(6.3) becomes
y ‡ yA ˆ 1: …6:11†

In the MARI approximation, we have made one of the intermediates MARI at all
reaction conditions. This has two consequences:
. We have lost the description of a change in MARI.
. We have explicitly assumed that all intermediates except the MARI are much
less abundant than free sites.
In particular, the latter consequence is troublesome as an estimate of the validity
of this aspect of the MARI approximation amounts to solving the problem
without using the MARI approximation. In other words, the validity of the
MARI approximation can be veri®ed, only if it is super¯uous.
The MARI approximation can be used for quantitative modeling, if we have
veri®ed that it is valid at the reaction conditions we are considering.
The MARI approximation is very much used for the analysis of reaction
mechanisms, both when we have diculties in formulating a kinetic model for a
complicated reaction mechanism and when we want to derive a limiting form of a
kinetic model.

7. Models

For mechanisms, there are two important aspects, stoichiometric and


thermodynamic consistency. However, we must make the description of the kinetic
model more concrete before we can implement the approximation schemes and
arrive at a soluble model.

7.1. Limitations

. We will limit ourselves to catalytic reactions. Surface sites consumed in the


adsorption of reactants must be regenerated in the desorption of products. The
net production of any surface intermediate must be zero.
. We describe the reactions at the molecular level. The mass balance for the
atoms is only implicitly described through the mass balances for the molecules.
We do not describe the structure of the molecules.
. The gas phase is assumed ideal.
. Di€usion limitations and temperature gradients are neglected.
. The reaction mechanism is an LH mechanism comprising G gases, S surface
species and R elementary steps, where G, S and R are arbitrary.
116 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

7.2. Types of models

Even with the limitations listed in the previous section, there are a very
large number of models. We will con®ne our attention to the following six
models:

1. Molecules have a name, a stoichiometry and a stability. One of the reaction


steps is assumed rate limiting, and has a forward rate constant. This model
maps onto the quasi-equilibrium approximation (Section 6.2) and is discussed
as Model 1 in Section 8. In this model the equilibrium constants for all steps
are calculated from stoichiometry and from the stability of the molecules. The
rate of the fast steps is not available.
2. Molecules have a name and a stoichiometry. Each step has a
thermodynamics. One of the reaction steps is assumed rate limiting, and has
a forward rate constant. This model is a trivial variation to the previous
model. The equilibrium constants for the steps are among the input
parameters. The stability of the intermediates and the rate of the fast steps
are not available.
3. Molecules have a name, a stoichiometry and a stability. Each reaction step has
a forward rate constant. This model maps onto the SS approximation (Section
6.1) and is discussed as Model 2 in Section 9. In this model the equilibrium
constant for the steps are calculated from the stoichiometry and from the
stability of the molecules.
4. Molecules have a name and a stoichiometry. Each step has a forward and
backward rate constant. This model is a trivial variation of the previous
model. The equilibrium constants for the steps are calculated from the
forward and backward rate constants. The stabilities of the molecules are
not available.
5. Molecules have a name, a stoichiometry and a stability. For each con®guration
of molecules at the surface, there are a number of possible events. The events
occur randomly with a characteristic rate for each type of con®guration. This
model is the kinetic MC simulation discussed in Section 10. In this model the
equilibrium constants for the steps are calculated from the stability of the
molecules. The rates are calculated by simulation.
6. Molecules have a name and a stoichiometry. For each con®guration of
molecules at the surface, there are a number of possible events. Each event has
a thermodynamics and occur randomly with a characteristic rate for each type
of con®guration. This is a trivial variation of the previous model. The
equilibrium constants are among the input parameters. The rate is calculated by
simulation. The stability of the molecules is not available.
Hybrids between these six models are possible, e.g. by specifying the
thermodynamics by the thermodynamics for some steps and the stability for some
of the molecules. However, these hybrids are much more dicult to work with
than the six models de®ned above, in particular for large mechanisms.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 117

8. Model 1

As de®ned in Section 7.2, Model 1 comprises a mechanism speci®ed as a


number of molecules each with a name, a stoichiometry and a stability. One of
the steps is rate limiting and has a rate constant, the other steps are equilibria.
This model maps onto the quasi-equilibrium approximation (Section 6.2). In this
model the net reaction is always a stoichiometric gas-phase reaction. As there is
only one rate in the model, this model cannot be used to study selectivity.

8.1. Consistency

Stoichiometric consistency is checked by verifying that the mechanism is


actually an LH mechanism (Section 4.4) followed by a check of (5.19).
Thermodynamic consistency is checked by checking that the mechanism is
stoichiometrically consistent and by checking that the mechanism has a single
equilibrium composition. This problem may be broken into two by ®rst verifying
that the mechanism corresponds exactly to a single gas-phase reaction and then
that this reaction is thermodynamically consistent.

8.2. Complications

All the complications listed in Section 4.5 are treated correctly in Model 1.
However, the occurrence of parallel steps require a comment. If parallel steps
occur in Model 1, there are two cases:
1. The same equilibrium reaction has been included in the mechanism two or
more times. This is a problem if and only if the mechanism is
thermodynamically inconsistent.
2. The RLS has also been included in the mechanism one or more times as an
equilibrium. The equilibrium steps will establish the equilibrium concentrations.
Provided the reaction mechanism is consistent, the mechanism will immediately
establish the equilibrium. The rate of the slow reaction is then zero.
In both cases the treatment will provide the correct answer.

8.3. Stoichiometric number

As the net reaction does not involve surface species, the weight of step r in this
linear combination is determined from the requirement that the net formation of
each surface species is zero:
X
R
br acr ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S: …8:1†
rˆ0

As this equation must have exactly one solution we must have R=S and the sub-
118 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

matrix arc with r = 1, . . ., R and c=G + 1, . . . , G+S must have dimension R


and rank R ÿ 1.
We choose the RLS as reference bRLS=1 and determine b1, . . . bRLSÿ1,
bRLS+1, . . ., from
0 1
b1
0 1B . . . C
a1, G‡1 . . . aRLSÿ1, G‡1 aRLS‡1, G‡1 . . . aR, G‡1 B C
B bRLSÿ1 C
@... ... A B C
B bRLS‡1 C
a1, G‡S . . . aRLSÿ1, G‡S aRLS‡1, G‡S . . . aR, G‡S @B C
... A
bR
0 1
ÿaRLS, G‡1
ˆ @... A: …8:2†
ÿaRLS, G‡S

br is the stoichiometric coecient for step r [27,173,176±179]. If and only if there


are no parallel reaction steps in the mechanism, br may be interpreted as the net
number of turnover for step r for each turnover of the RLS. However, br is
de®ned from stoichiometry and is well de®ned even in situations, such as
equilibrium, where the rates are zero and the ratio between rates do not exist.
br has a number of interesting properties:
. When br > 0, step r has been written in the forward direction.
. br=0 indicates that the step is a dead-end, which does not contribute to the net
reaction. However, dead-ends may be of great importance for the kinetics if the
intermediates are strongly adsorbed.
. br < 0 indicates that step r has been written in the backward direction.
This concept of direction is based on stoichiometry. The sign of br indicates if
the step runs in the same or in the opposite direction of the RLS. The sign of br is
independent if the rate of the step is actually positive or negative.

8.4. Net reaction

When br, r = 1, . . ., R has been determined, we can determine the


stoichiometric coecients, ac, c = 1, G
X
R
ac ˆ br arc , …8:3†
rˆ1

and the equilibrium constant


Y
R
Kˆ K br r , …8:4†
rˆ1

for the net reaction.


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 119

In the net reaction, gas-phase reactants have ac < 0, gas-phase products have
ac > 0 and gas-phase inerts and all surface species have ac=0.

8.5. Rate

The rates of consumption and production for each of the reactants or products
will be related as the quantity
1
rˆ ri
ai
is the same for all reactants and products. In this equation, ri is the rate of
production for reactant number i. r may thus be used as the rate of the reaction.

8.6. Gas-phase equilibrium

The calculation of the equilibrium composition for the gas phase is


straightforward as net reaction (8.3) and its equilibrium constant (8.4) are readily
determined.
The equilibrium composition is then determined by solving
G  
1Y pi a
ÿ1 ˆ 0, …8:5†
K iˆ1 p

1 ‡ ai u
pi ˆ pi0 , …8:6†
X G
1‡u ai
cˆ1

where K is the gas-phase equilibrium constant, pi is the equilibrium partial


pressure of i, pi0 is the initial partial pressure of i, and u is the extent of reaction.

8.7. Calculation of coverages

We can solve for the surface coverages from


G 
Y  Y
pc arc G‡S
Kr ˆ yac rc , …8:7†
cˆ1
p cˆG‡1

and
X
G‡S
yc ˆ 1: …8:8†
cˆG‡1

We choose the coverage by free sites, yG + 1, as parameter


120 P. Stoltze / Progress in Surface Science 65 (2000) 65±150
0 10 1
a1, G‡2 . . . aRLSÿ1, G‡2 aRLSÿ1, G‡2 . . . aR, G‡2 ln yG‡2
@... ... A@ . . . A
a1, G‡S . . . aRLSÿ1, G‡S aRLSÿ1, G‡S . . . aR, G‡S ln yG‡S
0   1
XG
pc
B ln K1 ÿ a1, c ln ÿ a1, G‡1 ln yG‡1 C
B p C
B c C
B... C
B C
B XG   C
B pc C
B ln KRLSÿ1 ÿ aRLSÿ1, c ln ÿ aRLSÿ1, G‡1 ln yG‡1 C
B p C
B c C
ˆB C: …8:9†
B XG   C
B pc C
B ln KRLS‡1 ÿ aRLS‡1, c ln ÿ aRLS‡1, G‡1 ln yG‡1 C
B p C
B c C
B... C
B   C
B XG
pc C
@ A
ln KR ÿ aR, c ln ÿ aR, G‡1 ln yG‡1
c
p

By multiplication from the left with the inverse of the left-hand matrix, we can
determine a formal solution
X
R XG  
pc
ln yc ˆ gcr ln Kr dcg ln ‡ Ec ln yG‡1 : …8:10†
rˆ1 cˆ1
p
r6ˆRLS

At this point it is helpful to realize that the transformation from (8.7) to (8.10) is
to transform the equilibrium equations into R ÿ 1 equilibrium equations, each
describing the reaction between a number of gases and Ec free sites to produce one
adsorbed molecule
ÿÿÿ
dc, 1 A1 ‡    ‡ dc, G AG ‡ Ec  ) *ÿAc  , …8:11†

which immediately demonstrates that Ec=1.


This solution (8.10) is formal since pc=0 is physically meaningful, but can only
be represented as a limit. However, this is a purely mathematical diculty that
can be eliminated by reformulating (8.10) into
Y
R G 
Y 
pg dcg
yc ˆ K gr cr yG‡1 : …8:12†
rˆ1 cˆ1
p
r6ˆRLS

This system can easily be solved for the coverages by solving (8.12) with (8.8), e.g.
by bisection using the start guesses yG + 1=0 and yG + 1=1.

8.8. Activation energy

From
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 121

ˆG  
kr cY Y
pc ÿarc cˆG‡S
r‡ ˆ …yc † ÿarc , …8:13†
br cˆ1 p cˆG‡1
arc <0 arc <0

we have
X
G‡S
2d ln r‡ d ln yc
H z ˆ kB T ˆ H zRLS ÿ kB T 2 aRLS, c : …8:14†
dT cˆG‡1
dT
arc <0

If we di€erentiate
X
G‡S
yj ˆ 1 …8:15†
G‡1

with respect to T, we ®nd


X
G ‡S
d ln yj
2
kB T yj ˆ 0: …8:16†
jˆG‡1
dT

From (8.7) we ®nd for i > G + 1


XR
2d ln yi d ln yG‡1
kB T ˆ gij DHj ‡ kB T 2 : …8:17†
dT jˆ1
dT
r6ˆRLS

By combination of (8.7) and (8.17), we ®nd


XR G X‡S
2d ln yG‡1
kB T ˆÿ yi gij DHj : …8:18†
dT jˆ1 iˆG‡1
j6ˆRLS

When this equation is substituted into (8.14), we ®nd


X
G X
‡S G ‡S X
G X
‡S G‡S X
R
H z ˆ H zRLS ÿ aRLS, c gcr DHr ‡ aRLS, j yj gjr DHr : …8:19†
cˆG‡1 rˆ1 cˆG‡1 rˆ1 jˆ1
arc <0 r6ˆRLS arc <0 r6ˆRLS j6ˆRLS

This equation has an obvious interpretation. The activation energy is a measure of


the change in the reaction rate when the temperature is changed. This change is
the sum of three contributions: the activation energy for the RLS ( ®rst term ), the
enthalpy of formation (second term ) for the reactants of the RLS and the change
in the concentration of surface concentrations (third term ).

8.9. Reaction orders

From
122 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

ˆG
cY  ÿarc cˆG‡S
Y
pc
r‡ ˆ k r …yc † ÿarc , …8:20†
cˆ1
p cˆG‡1
arc <0 arc <0

we ®nd the reaction order for i:

d ln r‡ X d ln yj
G‡S

pi ˆ ÿar, i ÿ arj pi : …8:21†


d ln jˆG‡1 d ln
p arc <0 p

If we di€erentiate
X
G‡S
yj ˆ 1 …8:22†
G‡1

with respect to ln( pi/p %), we ®nd


X
G ‡S
d ln yj X
G ‡S
dyj
yj pi ˆ pi ˆ 0: …8:23†
jˆG‡1 d ln G‡1 d ln
p p

From (8.7), we ®nd


d ln yj d ln yG‡1
pi ˆ di ‡ pi : …8:24†
d ln d ln
p p

The combination of (8.23) and (8.24) yields

d ln yG‡1 X
G‡S

pi ˆ ÿ yj dji yi : …8:25†
d ln jˆG‡2
p

When the coverages have been calculated, we can calculate the left-hand-side
of (8.25) followed by calculation of the left-hand-side of (8.24) to determine
d ln yj/d ln( pi/p %) for j=G + 1, . . ., G+S.

9. Model 2

In the list of types of models (Section 7.2), Model 2 corresponds to the situation
where molecules have a name, a stoichiometry and a stability. Each reaction step
has a forward rate constant. This model maps onto the SS approximation (Section
6.1).
In this model the equilibrium constant for the steps are calculated from the
stoichiometry and from the stability of the molecules.
Model 2 is somewhat more dicult to work with than Model 1. However, as
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 123

Model 1 has a ®xed RLS, Model 2 is the simplest possible model for studies of
some kinetic phenomena, such as selectivity or changes in RLS.

9.1. Consistency

As we consider a catalytic reaction, the net rate of formation must be zero for
each surface species
X
R
arc br ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S: …9:1†
rˆ1

For Model 1 we could use the RLS as a reference by assigning it br=1. For
Model 2 there is no unique choice of reference reaction. Stoichiometric
consistency is thus determined from the structure of the solution, (b1, . . . , bR), to
(9.1). The mechanism is stoichiometrically consistent if (9.1) has one or more
solutions, (b1, . . ., bR).
The same reaction step may occur two or more times in the reaction mechanism
and this will cause E (9.1) to have in®nitely many solutions.
Often Model 2 is used for mechanisms:
. which happen not to have parallel reaction steps;
or
. where one reaction path is dominating. An example would be a mechanism,
where one path is dominating while one or more trace products are formed by
alternate pathways.
We could in principle identify the dominating reaction path, but this can only be
done after it has been veri®ed that the mechanism is consistent. Instead, it may be
sensible to proceed as for Model 1 (Section 8.3). The reference reaction must be
identi®ed a priori and we can then proceed by verifying that the mechanism is
consistent and that the chosen step is in the dominating pathway. It should be
noted that if the mechanism appears to be inconsistent, the problem may be either
in the mechanism or in the choice of reference reaction.
The question if the mechanism is thermodynamically consistent is answered by
determining if the equilibrium equation has a unique solution.
G  
1Y pi ai
ÿ1 ˆ 0,
K iˆ1 p

1 ‡ ai u
pi ˆ pi0 ,
X G
1‡u ai
cˆ1
124 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

G 
Y  Y
pc arc G‡S
Kr ˆ yac rc ,
cˆ1
p cˆG‡1

X
G‡S
yc ˆ 1,
cˆG‡1

where K is the gas-phase equilibrium constant, pi is the equilibrium partial


pressure of i, pi0 is the initial partial pressure of i, and u is the extent of reaction.

9.2. Net reaction

In Model 2, each gas evidently has a net rate for formation. However, the
mechanism may describe two or more stoichiometric reactions running in parallel
and the relative rate of these reactions may depend on the speci®c reaction
conditions. For this reason, we cannot in general determine a net reaction for
Model 2. A particularly complex case is mechanisms consisting of several
stoichiometric reactions pairwise coupled by a few common reactants, products or
intermediates.
There are some obvious limiting cases:
. If each gas is only a reactant or product in a single reaction step, then it is
straightforward to check if the mechanism corresponds to a single,
stoichiometric net reaction.
. If any of the gases are a reactant or product of two or more reactions and any
pair of these reactions have di€erent stoichiometry, the mechanism describes
two or more stoichiometric reactions running in parallel.

9.3. Gas-phase equilibrium

As the mechanism in Model 2 may describe more than one net reaction, the
equilibrium composition of the gas phase is calculated by calculating the
equilibrium composition for the gas phase and for the adsorbates.

9.4. Rate

The rate of formation for each gas may be calculated from


X
R
rc ˆ rr arc , …9:2†
rˆ1

where the rate rr of step r is


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 125

G 
Y  ÿarc Y G  arc G‡S
ptot G‡S
kr Y ptot Y
rr ˆ kr xc y ÿarc ÿ x c yarc :
cˆ1,G
p0 cˆG‡1, ..., G‡S
Kr cˆ1, ..., G p0 cˆG‡1, ..., G‡S
arc <0 arc <0 arc >0 arc >0

…9:3†

9.5. Coverages

In Model 2, the coverages are often available as intermediate results in the


computation of other data. The coverages at gas-phase equilibrium is available
from the veri®cation of thermodynamic consistency (Section 9.1) and the
coverages through a plug-¯ow reactor are available as intermediate results in the
calculation of the conversion.
If the coverages yc for c=G = 1, . . ., G+S are to be calculated from the partial
pressures, this may be done by iterative solution of the SS approximation for the
adsorbates
X
R
rr arc ˆ 0 for c ˆ G ‡ 1, . . . , G ‡ S
rˆ1

X
G‡S
yc ˆ 1
cˆG‡1

where the rates, rr, are calculated from Eq. (9.3).

9.6. Rate limitation

In Model 2, all reaction steps are more or less rate limiting for the formation of
each gas and di€erent reaction steps may be limiting the rate for di€erent gases.
The key problem for Model 2 is thus for each gas to determine the degree of rate
limitation.
For each of the gases the rate of formation may be determined from (9.2) and
(9.3). From (9.3) we ®nd for both gases and adsorbates
dri rn X dri dym
ˆ ‡ : …9:4†
dkn kn m
dym dykn

For the adsorbates, j=G + 1, . . ., G+S, the rate of formation is zero:


X
ri ai, j ˆ 0 for j ˆ G ‡ 1, . . . , G ‡ S: …9:5†
iˆ1, ..., R

By di€erentiation of this equation with respect to kn, we ®nd


126 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

X dri
ai, j ˆ 0 for j ˆ G ‡ 1, . . . , G ‡ S …9:6†
iˆ1, ..., R
dkn

which with (9.4) gives us a linear system of S equations and S variables:


0 1
X drp X drp
ap, G‡1 ... ap, G‡1 0 1
B C rn 0 1
B p dyG‡1 p
dyG‡S C a n, G‡1 0
B C B k C
B... C ‡ B... n C @ A
BX X C @ rn A ˆ . . . , …9:7†
B drp drp AC an, G‡S 0
@ ap, G‡S ... ap, G‡S kn
p
dy G‡1 p
dy G‡S

for the determination of


dyn
for j ˆ G ‡ 1, . . . , G ‡ S:
dkn
The degree, dri/dkr, that reaction r is rate limiting for the formation of gas c, may
then be calculated from (9.4) by insertion of the values for dyn/dkn determined
from (9.7).

10. Kinetic Monte Carlo

In the list of types of models (Section 7.2), the KMC corresponds to the
following model:
. Molecules have a name, a stoichiometry and a stability. For each con®guration
of molecules at the surface, there are a number of possible events. The events
occur randomly with a characteristic kinetics for each type of con®guration.
The main advantages of KMC is that adsorbate±adsorbate interaction and surface
di€usion is explicitly included and that the KMC treats ®ner details than Models
1 or 2. The major disadvantage in KMC is that calculated as averages in a
stochastic simulation which greatly increases the computational e€ort and makes
interpretation of the results rather dicult.

10.1. Phases in a simulation

Simulation consists of three phases: initialization, equilibration and production


[180]. At the initialization, a con®guration is generated or read from a ®le. The
system is then equilibrated by propagation in time. When the system has
equilibrated, the production phase starts. Data are stored on disk when the
production phase starts and then after each major timestep.
The major timestep is an abstraction. The program internally builds up the
major timesteps through a series of minor timesteps.
It is a complication that in KMC the minor timesteps are of varying length. As
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 127

we want the output from the simulation program stored at constant time
intervals, special care is required. Fig. 10 shows the principle in simulation.

10.2. Basic algorithm

For a given con®guration a list of all possible events, n = 1, . . ., N, is generated


and the rate rn for each event is calculated. Exactly one event is selected, such that
any event, n, is selected with probability
rn
pn ˆ : …10:1†
X
N
ri
iˆ1

The selected event is executed and time is advanced by


ln…u†
tˆ …10:2†
XN
ri
iˆ1

where u is a random number uniformly distributed in the interval 0 R u < 1.


In the following we will need an algorithm to select a number, n, out of N with
weight Wn, or, equivalently, probability
Wn
pn ˆ :
X
N
Wi
iˆ1

Fig. 10. Principle in simulation.


128 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

Without loss of generality assume that the weights are non-zero. The selection can
be implemented eciently by storing
…s1 , s2 , . . . , sN †, …10:3†

where
X
n
sn ˆ Wi :
iˆ1

We generate a random number, u, uniformly distributed in [0, 1] and the


corresponding value of n from
snÿ1 < sN uRsn :

As the sequence (10.3) is ordered, n can be determined by search by bisection.

10.3. Events

In the mechanism the possible events are the adsorption of molecules, surface
reactions and desorption of molecules. Implicitly, all reactions will occur at pairs
of neighbor sites.
For this reason, surface di€usion may be treated as a special case of a surface
reaction
A‡ÿ
)ÿ
ÿ*
ÿ  ‡A
and we can readily interpret the event table as a stoichiometric matrix.
We choose that for each event, there is a forward rate constant and an
equilibrium constant. The alternative, where events are speci®ed only with a
forward rate constant and the reverse rate constant might optionally be included
by speci®cation of the forward rate constant for the event written in reverse, is
more general. However, the question if the model is thermodynamically consistent
becomes much too complex for the model to be useful.

10.4. Consistency

Stoichiometric and thermodynamic consistency may be checked by deriving a


stoichiometric matrix from the event table. The part of the KMC-model which is
described by the stoichiometric matrix is then checked exactly as for Model 2
(Section 9). The only events which cannot be described by a stoichiometric matrix
are the di€usion events. However, the di€usion events are stoichiometrically
consistent by construction and are easily checked for thermodynamic consistency.

10.5. Time-scales

Frequently the rates of surface reactions di€er by orders-of-magnitude. In KMC


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 129

we describe the surface di€usion explicitly. When surface di€usion is fast, the
timestep, (10.2), becomes small and the simulation will run very slowly. This is
unfortunate as LH corresponds to the limit of very fast surface di€usion.
This exposes a fundamental ¯aw in the KMC algorithm. When the KMC
algorithm is formulated as in Section 10.2, an increase in the rate of one of the
steps in the mechanism leads to a dramatic increase in the computational e€ort
rather than a fast approach to equilibrium.
As the surface di€usion is frequently comparable in rate to the surface
reactions, a simple approximation, such as treating the di€usion as either slow or
fast independent of the temperature, will not work.

10.6. Advanced algorithm

Fortunately, a smooth transition to fast di€usion can easily be implemented by


a slight modi®cation of the KMC algorithm. In the ®rst scan we determine for
each atom the set of sites that this atom can reach through fast reactions. For
each of these sites we calculate
 
En ÿ E0
wn ˆ exp ÿ ,
kB T

where En is the energy of the system when the atom is at the nth site and E0 is the
energy of the system prior to the move. The nth site is selected with probability,
pn
wn
pn ˆ :
X
N
wi
iˆ0

In some cases, an atom cannot move by any fast reaction, e.g. because all its
neighbor sites are occupied. It is not required to implement this as a special case.
If the atom cannot move by any fast reaction, the set consists of only the current
site of the atom.
This site then has p0=1 and is selected with probability 1.
The ®rst scan considers only fast reactions and contributes signi®cantly to the
approach to equilibrium, but does not propagate the system in time. The second
scan propagates the system in time in accordance with (10.3).
In the second scan, we consider only the desorption events and the slow
di€usion events and build the list of possible events as outlined above.

10.7. Rate

In a typical simulation, each type of event will be executed many times and the
information of the resulting rate is too basic to be of any use. Instead, each event
130 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

must be assigned to a class, where each class makes up a macroscopic reaction,


such as adsorption, di€usion, or reaction. The rate averaged for each class of
event is then readily interpreted.
In the simulation each major timestep is made up of a number of minor
timesteps of variable length. This complication is handled by calculation of the
average rate
1X X
rˆ ti rij …10:4†
h i j

for each class for each major timestep. In this expression rij is the rate of the ith
event in the class during the jith minor step, tj is the length of the jith minor step,
and h=Sj tj is the length of the major step.
The terms in the sum in Eq. (10.4) may be calculated on the ¯y while the table
of events is being built. All events are to be included in the sum independent if
they are to be treated as slow or fast in the simulation.

11. Software

In the present section, we will brie¯y outline one example of a general purpose
program for microkinetic simulation4.
mkm will treat an LH mechanism consisting of an arbitrary number of gas and
surface species and an arbitrary number of reaction steps.
The level of detail in the description of the mechanism is variable. The most
detailed description consists of a KMC model with tables of reaction
probabilities for each micro-con®guration. The intermediate description consists
of a statistical thermodynamic model of reach molecule in the mechanism. The
coarsest description consists of a forward and a backward rate-constant for each
step.
For each level of detail, the program reads the mechanism and the properties of
the molecules analyzes the model and performs a calculation for the model, e.g.
an integration for a plug-¯ow reactor.
The input for a calculation for an RK integration for an isothermal plug-¯ow
reactor may look like this:
calculation ipfRkMolRls (5 gases, 8 adsorbates, 8 reactions,
101 points);

The input contains the description of the molecules and their properties:
molecule N2
{
mass 0.028,

4
The example is based on the program mkm available from http://www.aue.auc.dk/~stoltze
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 131

rotation symmetry 2 frequency 23.960,


vibration frequency 28130.000,
enthalpy 0.000
};
etc. The program has no prede®ned names for gases or adsorbates except for free
sites, : Then follows the parameters for the catalyst bed
area 2700.000;
mass 1.000eÿ03;
temperature 873.0;
flow 2.778eÿ04;
pressure 1.01e+07;
sites 2.6180eÿ5;
porosity 0.340;
density 5200;

the composition of the reactant mixture


input (N2 0.250, H2 0.750, H20 0);

and ®nally the reaction mechanism


reaction 1 { [ N2+ ˆ N2 ] };
reaction 2 { [ N2  ‡ ˆ 2 N Š, actene 28500., prefac 4.3e+9};
reaction 3 { [ N  ‡H ˆ NH  ‡ ] };
reaction 4 { [ NH ˆ H ˆ NH2  ‡ ] };
reaction 5 { [NH2  ‡H ˆ NH3  ‡ ] };
reaction 6 { [NH3 ˆ NH3 ‡  ] };
reaction 7 { [ H2+2  ˆ 2 H ] };
reaction 8 { [ H2O ‡  ˆ O  ‡H2 ] };

In this example step 2 is rate limiting at all conditions.


The output consists ®rst of a diagnostic message
reaction 8 is a dead-end

followed by a table of temperature, ¯ow, gas composition, surface coverages and


reaction rates through the catalyst bed:
0 873.0 2.77800eÿ04 2.500eÿ01 7.500eÿ01 0.00e+00
0.000e+00 0.000e+00 . . .
1 873.0 2.71257eÿ04 2.323eÿ01 6.970eÿ01 2.355eÿ02
0.000e+00 0.000e+00 . . .
2 873.0 2.69906eÿ04 2.287eÿ01 6.861eÿ01 2.842eÿ02
0.000e+00 0.000e+00 . . .
3 873.0 2.69097eÿ04 2.265eÿ01 6.795eÿ01 3.133eÿ02
0.000e+00 0.000e+00 . . .
...
97 873.0 2.66338eÿ04 2.191eÿ01 6.572eÿ01 4.126eÿ02
132 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

0.000e+00 0.000e+00 . . .
98 873.0 2.66338eÿ04 2.191eÿ01 6.572eÿ01 4.126eÿ02
0.000e+00 0.000e+00 . . .
99 873.0 2.66338eÿ04 2.191eÿ01 6.572eÿ01 4.126eÿ02
0.000e+00 0.000e+00 . . .

12. Conclusions

Microkinetic modeling is an established technique for catalyst research.


However, the computations remain tedious and must be automated before
microkinetic modeling may become routine. An important prerequisite is a
systematic and coherent framework for the analysis of the kinetics of arbitrary,
proposed reaction mechanisms. The existing microkinetic studies show that LH
mechanisms form a useful basis for microkinetic modeling.
As mechanisms with any number of gases and adsorbates and any number of
reaction steps should be amenable to microkinetic analysis, three models for LH
mechanisms prove to be particularly useful. The ®rst of these consists of an LH
mechanism with the quasi-equilibrium approximation, and a single, RLS. For this
model detailed kinetic studies are possible and include the analytic determination
of reaction rate, surface coverages, activation energy, and reaction orders. The
second model consists of an LH mechanism with the SS approximation. This
model allows the treatment of some kinetic phenomena, which cannot be treated
by Model 1. However, the treatment of Model 2 is much more dicult and the
data which may be determined analytically from Model 2 are the reaction rate,
the surface coverages, and the degree of rate limitation. Finally, KMC simulations
alow the modeling of even more mechanistic details, such as surface di€usion and
adsorbate±adsorbate interactions. Analysis of this class of models are limited to
numerical simulation.
Although the present treatment is limited to the most basic features, it does
demonstrate that the development of a comprehensive and coherent framework
for microkinetic modeling is possible.

Acknowledgements

The ®rst draft of this manuscript was prepared while I was at the Center for
Atomic-scale Materials Physics (CAMP). I acknowledge the ideas and inspiration
I have received from countless sources. I am, in particular, grateful to Jens K.
Nùrskov and Bengt Kasemo, who have contributed de®nite ideas incorporated in
the manuscript.

Appendix A. Implementation of RK integration

In this section, we will demonstrate the implementation of the calculation in an


P. Stoltze / Progress in Surface Science 65 (2000) 65±150 133

ideal plug-¯ow reactor using an RK algorithm. The implementation we will


discuss here is limited to a ®xed mechanism. The implementation of a more
general program, such as the program mentioned in Section 11 is beyond the
scope of the present manuscript.
Reaction rates vary over many orders-of-magnitude and it is imperative that the
integration is implemented as a self-adjusting variable-stepsize algorithm.
The mechanism we will consider is mechanism (2.1) from Section 2.2. The
coverages may be determined from (2.2) and (2.4) and the rate from (2.5).

A.1. Runge±Kutta

The numerical solution to a di€erential equation


df
ˆ f 0 …t, x† …A1†
dt
is a series of points (tn, xn) which approximates
x n ˆ f…nh, x n †: …A2†

The fourth-order RK algorithm is


k1 ˆ hf 0 …tn , x n †, …A3†

 
0 h k1
k2 ˆ hf tn ‡ , x n ‡ , …A4†
2 2
 
0 h k2
k3 ˆ hf tn ‡ , x n ‡ , …A5†
2 2

k4 ˆ hf 0 …tn ‡ h, x n ‡ k3 †, …A6†

k1 k2 k3 k4
x n‡1 ˆ x n ‡ ‡ ‡ ‡ : …A7†
6 3 3 6

A.2. Calculation

In the calculation, the partition functions may be calculated from the


temperature and the properties of the molecules, while the equilibrium constants
may be calculated from the partition functions and the surface coverages may be
calculated from the equilibrium constants and the partial pressures, etc. The
calculated data form a hierarchy and this may be used to split the calculation into
a number of functions.
134 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

The arrows in the ®gure indicate function calls made repeatedly. Dashed arrows
indicate function calls made only as part of the startup of the integration.

A.3. Parameters

We ®rst de®ne some parameters


#define NMOL 10
#define NRCT 7
#define NPNT 101
where NMOL is the number of molecules, NRCT is the number of elementary
steps, and NPNT is the number of points in the calculation.
Symbolic names for each of the gases and surface intermediates are
#define N2GAS 0
#define H2GAS 1
#define NH3GAS 2
#define SITE 3
#define N2SITE 4
#define NSITE 5
#define NHSITE 6
#define NH2SITE 7
#define NH3SITE 8
#define HSITE 9
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 135

A.4. Variables

We want to calculate mole fractions in the gas phase, coverages at the surface,
the reaction rate and the ¯ow. We de®ne a struct for storage of these data:
struct POINT
{
double conc[NMOL];
double tof‰2  NRCT Š;
double flow;
};
The data through the catalyst bed are then the array point in addition to the
constant temperature, temp, and the pressure, pressure
double temp;
struct POINT point[NPNT];
double pressure;

A.5. Calculation of equilibrium constants

tdyn() calculates the equilibrium constants, eqcns, and the rate constant, rtcns,
at a given temperature, temp. tdyn() uses prtfct() to calculate the partition
function for the molecules.
static void tdyn(double temp, double eqcns[],
double rtcns)
{
int ia;
double z[NMOL];

for (ia=0; ia < idxMol2; ia++)


z[ia]=prtfct(&mol[ia], temp, ia < SITE);

eqcns[0]=z[N2SITE] / z[N2GAS];
eqcns[1]=sq(z[NSITE]) / z[N2SITE];
eqcns[2]=z[NHSITE] /z[NSITE] /z[HSITE];
eqcns[3]=z[NH2SITE] / z[NHSITE] / z[HSITE];
eqcns[4]=z[NH3SITE] / z[NH2SITE] / z[HSITE];
eqcns[5]=z[NH3GAS] / z[NH3SITE];
eqcns[6]=sq(z[HSITE]) / z[H2GAS];
eqcns[7]=z[OSITE]  sq(z[HSITE]) / z[H2OGAS];
eqcns[8]=sq(z[NH3GAS]) / z[N2GAS] / tr(z[H2GAS]);

rtcns=prefac  exp(ÿactene / gascon / temp);


}
136 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

A.6. surf()

surf() calculates concentrations and coverages, conc, from the pressure and the
equilibrium constants eqcns.
voidsurf(double eqcns[], double conc[])
{
double p1, p2, p3, p4;

p1=pÐ> pressure  conc[N2GAS] / refpres;


p2=pÐ> pressure  conc[H2GAS] / refpres;
p3=pÐ> pressure  conc[NH3GAS] / refpres;
p4=pÐ> pressure  conc[H2OGAS] / refpres;

conc[SITE]=1. / (1.+p1  eqcns[0]+


p3 / eqcns[2] / eqcns[3] / eqcns[4] / eqcns[5] /
sqrt(tr(eqcns[6]  p2))+
p3 / eqcns[3] / eqcns[4] / eqcns[5] / eqcns[6] / p2+
p3 / eqcns[4] / eqcns[5] / sqrt(eqcns[6]  p2)+
p3 / eqcns[5]+
sqrt(p2  eqcns[6])+
p4  eqcns[7] / eqcns[6] / p2);

conc[N2SITE]=p1  eqcns[0]  conc[SITE];


conc[NSITE]=p3 / eqcns[2] / eqcns[3] / eqcns[4] /
eqcns[5] /sqrt(tr(eqcns[6]  p2))  conc[SITE];
conc[NSITE]=p3 / eqcns[3] / eqcns[4] / eqcns[5] /
eqcns[6] /p2  conc[SITE];
conc[NH2SITE]=p3 / eqcns[4] / eqcns[5] / sqrt(eqcns[6] 
p2) conc[SITE];
conc[NH3SITE]=p3 /eqcns[5]  conc[SITE];
conc[HSITE]=sqrt(p2  eqcns[6])  conc[SITE];
}

A.7. rate()

rate() calculates the turnover frequency, tof, from the concentrations, conc, rate
constant rtcns and equilibrium constants eqcns.
void rate(double eqcns[], double rtcns,
double conc[], double tof[])
{
tof[2]=rtcns  conc[N2SITE]  conc[SITE];
tof[3]=rtcns  conc[NSITE]  conc[NSITE] / eqcns[1];
}
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 137

A.8. driv()

driv() converts the turnover frequencies, tof, to the derivatives of the


concentrations, crte, and ¯ow, frte.
void driv(double conc[], double flow, double tof[],
double crte[], double frte)
{
double r=p Ð> densit  pÐ> area  (tof[2]ÿtof[3]);
frte=ÿ2.  r;
crte[N2GAS]=(ÿ1.0+2.  conc[N2GAS])  r / flow;
crte[H2GAS]=(ÿ3.0+2.  conc[H2GAS])  r / flow;
crte[NH3GAS]=(2.0+2.  conc[NH3GAS])  r / flow;
}

A.9. rk()

rk() performs the RK step from one point, p1, to the next, p2, separated by a
slice with mass dm. If dm is too large, the integration becomes inaccurate or
unstable. rk() checks for this and returns 0 if the step succeeds and 1 if it fails.
int rk(struct POINT p1,
double dm, double eqcns[], double rtcns,
struct POINT p2)
{
double d;
double crte[NMOL];
double frte;
double ctmp[NMOL];
double ftmp;
double dc[4][NMOL];
double df[4];
int i;
double tof[2  NRCT];

/ Stage 1  /

for (i=0; i < NMOL; i++)


ctmp[i]=p1Ð> conc[i];
ftmp=p1 Ð> flow;
surf(eqcns, ctmp);
rate(eqcns, rtcns, ctmp, tof);
driv(ctmp, ftmp, tof, crte, &frte);
for (i=0; i < NMOL; i++)
{
138 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

dc[0][i]=dm  crte[i];
if (fabs(dc[0][i]) > 0.01
vv p1Ð> conc[i]+0.5  dc[0][i] < 0.
vv p1Ð> conc[i]+0.5  dc[0][i] > 1.)
return 1;
}
df[0]=dm  frte;

/ Stage 2  /

for (i=0; i < NMOL; i++)


ctmp[i]=p1Ð> conc[i]+0.5  dc[0][i];
ftmp=p1 Ð> flow+0.5  df[0];
surf(eqcns, ctmp);
rate(eqcns, rtcns, ctmp, tof);
if (rc[0] < 0 && rc[1] > 0 k rc[0] > 0 && rc[1] < 0)
return 1;
driv(p, ctmp, ftmp, tof, crte, &frte);
for (i=idxGas1; i <=idxGas2; i++)
{
dc[1][i]=dm  crte[i];
if (fabs(dc[1][i]) > 0.01
vv p1Ð> conc[i]+0.5  dc[1][i] < 0.
vv p1Ð> conc[i]+0.5  dc[1][i] > 1.)
return 1;
}
df[1]=dm  frte;

/ Stage 3  /

for (i=0; i < NMOL; i++)


ctmp[i]=p1Ð> conc[i]+0.5  dc[1][i];
ftmp=p1 Ð> flow+0.5  df[1];
surf(eqcns, ctmp);
rate(eqcns, rtcns, ctmp, tof);
driv(p, ctmp, ftmp, tof, crte, &frte);
for (i=0; i < NMOL; i++)
{
dc[2][i]=dm  crte[i];
if (fabs(dc[2][i]) > 0.01
vv p1Ð> conc[i]+dc[2][i] < 0.
vv p1Ð> conc[i]+dc[2][i] > 1.)
return 1;
}
df[2]=dm  frte;
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 139

/ Stage 4  /

for (i=0; i < NMOL; i++)


ctmp[i]=p1Ð> conc[i]+dc[2][i];
ftmp=p1 Ð>flow +df[2];
surf(eqcns, ctmp);
rate(eqcns, rtcns, ctmp, tof);
driv(p, ctmp, ftmp, tof, crte, &frte);
for (i=0; i < NMOL; i++)
{
dc[3][i]=dm  crte[i];
if (fabs(dc[3][i]) > 0.01)
return 1;
}
df[3]=dm  frte;

/ end  /

for (i=0; i < NMOL; i++)


{
p2Ð> conc[i]=p1 Ð> conc[i]+
(dc[0][i]+2.  dc[1][i]+2.  dc[2][i]+dc[3][i]) / 6.;
if (p2 Ð> conc[i]< 0. k p2Ð> conc[i] > 1.)
return 1;
}
p2Ð> flow=p1 Ð> flow+(df[0]+2. 
df[1]+2.  df[2]+df[3]) / 6.;
surf(eqcns, p2 Ð> conc);
rate(eqcns, rtcns, p2Ð> conc, tof);
return 0;
}
p1 and p2 are pointers to the start and end point, respectively. dm is the mass in
the slice. eqcns[] and rtcns are the equilibrium and rate constants, respectively.
ctmp and ftmp are used to store the input concentrations and the input ¯ow,
respectively, to each stage. crte, frte, dc[i][], and df[i] are used to store the
derivatives of the concentrations, the derivatives of the ¯ow, the increment in
concentration, and the increment in ¯ow, respectively, in each stage.

A.10. scrk()

scrk() uses rk() to perform a step. If rk() indicates that the step failed, scrk()
recursively divides the step into half.
void scrk(struct POINT  p1,double dt,
double eqcs[], double rtcns,
140 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

struct POINT  p2)


{
struct POINT pn;

if (rk(p1, dt, eqcns, rtcns, p2) !=0)


{
scrk(p, p1, 0.5  dt, eqcns, rtcns, &pn);
scrk(p, &pn, 0.5  dt, eqcns, rtcns, p2);
}
}

A.11. Overall calculation

synt() performs the calculation for the catalyst bed. The integration is
initialized by calls to temp(), surf() and rate() for the ®rst point. The integration
then performs repeated calls to synt(), surf() and rate().
void synt()
{
double rtcns;
int im;
double eqcns[NRCT];

tdyn(p Ð> temp, eqcns, &rtcns);


surf(eqcns, pÐ> point[0].conc);

rate(eqcns, rtcns, pÐ> point[0].conc,


pÐ> point[0].tof);
for (im=1; im < points; im++)
{
if (scsynt(&p Ð> point[imÿ1]), p Ð> mass / points,
eqcns, rtcns, &p Ð> point[im] !=0)
return 1;
surf(eqcns, pÐ> point[im].conc);
rate(eqcns, rtcns, pÐ> point[im].conc,
p Ð> point[im].tof);
}
}

Appendix B. Implementation of KMC simulations

The basis for the implementation is a model where the atoms occupy sites in a
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 141

grid and the dynamics consist of atoms moving from one site to another. To
represent the grid we de®ne an abstract data type in the form of a struct:
#define NBRNUM 4
struct SITE
{
struct SITE nbr[NBRNUM];
int id;
int seq;
int flag;
} 
site;

In this struct the nbr array points to the neighbor sites. NBRNUM is the number
of neighbor sites, four for a quadratic lattice and six for a hexagonal lattice. As
the sites do not move, the values stored in the nbr never change.
id is the identity of the atom occupying the site with the convention that empty
sites have id=0.
¯ag is used in the treatment of fast reactions, we return to this below.
In a grid-model the atoms move from site to site. If we loop over the sites in
the output function, the atoms would be printed in a more or less random
sequence. However, most ®le-formats require that the atoms move in a ®xed
sequence. The variable seq is initialized at startup and updated as the atoms move.
seq is used to print the data for the atoms in a ®xed sequence. When the sites are
allocated, we initialize seq where size is the size of the grid.
for (iy=0; iy < size; iy++)
for (ix=0; ix < size; ix++)
{
site[ix+sizeiy].seq=ix+sizeiy;
site[ix+sizeiy].x=. . .
site[ix+sizeiy].y=. . .
...
}
When an atom moves from one site, c, to another, n, we swap both id and seq:
void swapAtm(struct SITE  c, struct SITE  n)
{
int t;
...
t=cÐ> id;
Ð> id=nÐ> id;
nÐ> id;
t=cÐ> id=n Ð> seq;
cÐ> seq=n Ð> seq;
nÐ> seq=t;

}
142 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

B.1. Event queue

In each minor step (Section 10.1) we need to generate a list of all possible
events. For this we de®ne an abstract data type:
struct QUEUE
{
double rate;
struct SITE 
target;
int action;

} queue;
int last;

target points to a site, action is a code for what happens to the site and rate is the
rate of the event.
At the startup the queue-array is allocated large enough to guarantee that it
cannot over¯ow. The current number of entries in queue is stored in the variable
last.

B.2. Equilibration

Two subroutines are used to implement the advanced algorithm. walk()


recursively walks over the grid and marks sites that can be reached from a given
start site through fast di€usion steps.
void walk(struct SITE  s, double t)
{
int i;

sÐ> flag=sÐ> env;


if (difEq(s, t))
for (i=0; i < NBRNUM; i++)
if (sÐ> nbr[i] Ð> id==0 && s Ð> nbr[i] Ð> flag==ÿ1)
{
swapNbr(s, i);
walk(s Ð> nbr[i], t);
swapNbr(s, i);
}
}
jump() uses the path generated by walk() to calculate the statistical weight for
each of the sites and then selects the new position for the atom.
static void jump(struct SITE  s, double t)

{
int i;
int last;
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 143

double rate;
double sum;
double wght[NBRNUM];

for (i=0; i < size  size; i++)


site[i].flag=ÿ1;

walk(s, f);
for (i=0; i < NBRNUM; i++)
wght[i]=exp(ÿ(desene+2.  i  omega) / gascon / t);
last=0;
for (i=0; i < size  size; i++)
if (site[i].flag !=ÿ1)
{
queue[last].target=s;
queue[last].action=i;
queue[last].rate=wght[site[i].flag];
last++;
}
if (last==0)
return;
for (i=1; i < last; i++)
queue[i].rate+=queue[iÿ1].rate;
sum=queue[lastÿ1].rate;
level=sum  ran2(0);
for (i=0; i < last; i++)
if (level <= queue[i].rate)
break;
swapAtm(s, queue[i].action);

B.3. Motion

The variable tsum accumulates the length of the minor timesteps and tstep is the
desired length of the major timestep. The simulation performs minor timesteps
until tsum just exceeds tstep.
Before the list is built, we calculate the rates and store them in the arrays
difRate for di€usion and desRate for desorption. As each site can have 0, 1, . . . ,
NBRNUM occupied neighbors, difEq and difRate are of length NBRNUM + 1.
For more complicated encodings, longer arrays are necessary.
The only slow di€usion events, if any, are then considered in the loop over
di€usion and desorption events.
last=0;
for(is=0; is < sizesize; is++)
144 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

if (site[is].id > 0)
{
if(!difEq[site[is].env])
for (i=0; i < NBRNUM; i++)
if (site[is].nbr[i] Ð> id==0)
{
queue[last].action=i;
queue[last].target=&site[is];
queue[last].rate=difRate[site[is].env]
last++;
}
for (i=0; i < NBRNUM/2; i++)
if (site[is].nbr[i] Ð> id !=0)
{
queue[last].action=i+NBRNUM;
queue[last].target=&site[is];
queue[last].rate=desRate[site[is].env]
last++;
}
}
If we scanned all sites and found no possible events, nothing can happen. This
does happen and must be handled correctly:
if (last==0)
return;

The importance of handling this situation correctly may be illustrated by


simulation of a desorption, 2A _ A2+2 at low coverage. It is quite likely that
nothing can happen during one or another of the timesteps simply because no A-
atoms happen to be present at neighbor sites. However, this would be unwise and
unacceptable if the simulation crashed in this situation. Surface di€usion may very
well position A-atoms in a suitable con®guration for desorption if the simulation
continues.
As a preparation for the selection of the event we sum the values in the rate
array:
for (is=1; is < last; is++)
queue[is].rate+=queue[isÿ1].rate;
sum=queue[lastÿ1].rate;
dt=ÿlog(ran(0)) / sum;

sum is the sum in (10.1), queue[n].rate is Sniˆ0 ri , and dt is the timestep (10.2).
We select a random number, level, between 0 and sum, search for the
corresponding entry in queue, and decode and execute the action represented by
this entry:
if (tsum+dt  ran(0) < =tstep)
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 145

{
level=sum  ran(0);
for (is=0; is < last; is++)
if (level < =queue[is].rate)
break;

if (queue[is].action < NBRNUM)


swapAtm(queue[is].target, queue[is].action);
else
desTwo(queue[is].target, queue[is].actionÿNBRNUM);
}

B.4. Calculation of rate

One might collect information on the rate from the events actually selected for
execution. However, much better estimates may be made with only a little more
e€ort.
When the timestep, dt, has been determined, inspection of the event queue will
provide a list of all events and their rate during this timestep. Only one of these
events will be selected for execution, while the event queue may easily contain 103
or more events.
Fast di€usion events will not be included in the event queue if the advanced
algorithm is used. For the advanced algorithm an alternative to the scan of the
event queue is to collect the data on the rates when the event queue is built. The
data required for estimation of the rate are available when the decision is made if
the event is slow or fast.

B.5. Output

In the output routine, the sites are ®rst loaded into an array based seq for each
site and then printed in the original sequence of the atoms
void output()
{
struct SITE list;
int is;

list=(struct SITE ) malloc(size  size 


sizeof(struct SITE ));

for(ix=0; ix < size; ix++)


for(iy=0; iy < size; iy++)
list[site[ix+site  iy].seq]=&site[ix+size  iy];
146 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

for(is=0; is < sizesize; is++)


printf(``%d %.3f %.3f\n'', list[is]Ð> id;
list[is] Ð> x, list[is]Ð y);

free(list);
}
The program thus prints the atoms in a ®xed sequence, although the program
internally uses a ®xed sequence of sites.

References

[1] M. Boudart, Kinetics of Chemical Reactions, Prentice-Hall, Englewood Cli€s, NJ, 1968.
[2] J.J. Carberry, Chemical and Catalytic Reaction Engineering, McGraw-Hill, New York, 1976.
[3] B.C. Gates, J.R. Katzer, G.C.A. Schuit, Chemistry of Catalytic Processes, McGraw-Hill, New
York, 1979.
[4] M. Boudart, G. Djega-Mariadassou, Kinetics of Heterogeneous Catalytic Reactions, Princeton
University Press, Princeton, NJ, 1984.
[5] D.P. Woodru€, T.A. Delchar, Modern Techniques of Surface Science, 1986.
[6] A. Zangwill, Physics at Surfaces, Cambridge University Press, Cambridge, 1988.
[7] G.F. Froment, K.B. Bischo€, Chemical Reactor Analysis and Design, 2nd ed., Wiley, New
York, 1990.
[8] V.P. Zhdanov. Elementary Physicochemical Processes on Solid Surfaces, Plenum.
[9] C.N. Satter®eld, Heterogeneous Catalysis in Practice, 2nd ed., Krieger, 1991.
[10] R.A. van Santen, Theoretical Heterogeneous Catalysis, World Scienti®c, Singapore, 1991.
[11] O. Levenspiel, Chemical Reaction Engineering, 2nd ed., Wiley, New York, 1992.
[12] H. Scott Fogler, Elements of Chemical Reaction Engineering, 2nd ed., Prentice-Hall, Englewood
Cli€s, NJ, 1992.
[13] B.C. Gates, Catalytic Chemistry, 2nd ed., Wiley, New York, 1992.
[14] J.A. Dumesic, D.F. Rudd, L.M. Aparicio, J.E. Rekoske, A.A. Trevino, The Microkinetics of
Heterogenous Catalysis, American Chemical Society, Washington, DC, 1993.
[15] J.A. Moulijn, P.W.N.M. van Leeuwen, R.A. van Santen, Stud. Surf. Sci. Catal. 79 (79) 1.
[16] G.A. Somorjai, Introduction to Surface Chemistry and Catalysis, Wiley, New York, 1994.
[17] R.A. van Santen, J.W. Niemantsverdriet, Chemical Kinetics and Catalysis, Plenum, New York,
1995.
[18] R.I. Masel, Principle of Adsorption and Reaction on Solid Surfaces, Wiley, New York, 1996.
[19] J.M. Thomas, W.J. Thomas, Principles and Practice of Heterogeneous Catalysis, VCH,
Weinheim, 1997.
[20] L.D. Schmidt, The Engineering of Chemical Reactions, Oxford University Press, New York,
1998.
[21] M. Bowker, in: The Basis and Applications of Heterogeneous Catalysis, vol. 53, Oxford
Chemistry Primers, New York, 1998.
[22] D.T. Lynch, G. Emig, Chem. Eng. Sci. 44 (1989) 1225.
[23] K.H. Yang, O.A. Hougen, Chem. Eng. Prog. 46 (1950) 146.
[24] S. Weller, AIChE J. 2 (1956) 59.
[25] A. Corman, F. Llopis, J.B. Monton, S.W. Weller, Chem. Eng. Sci. 43 (1988) 785.
[26] C. Wagner, Adv. Catal. 21 (1970) 323.
[27] J. Happel, Catal. Rev. 6 (1972) 221.
[28] S.W. Weller, Catal. Rev. Sci. Eng. 34 (1992) 227.
[29] G. Ertl, in: J.R. Anderson, M. Boudart (Eds.), Catal. Sci. Technol., vol. 4, 1983, p. 209.
[30] J.A. Rodrigues, D.W. Goodman, Surf. Sci. Rep. 14 (1991) 1.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 147

[31] D.W. Goodman, Surf. Sci. 299/300 (1994) 837.


[32] P.L.J. Gunter, J.W. Niemantsverdriet, F.H. Ribeiro, G.A. Somorjai, Catal. Rev. Sci. Eng. 39
(1997) 77.
[33] J.H. Larsen, Ib Chrokendor€, Surf. Sci. Rep. 35 (1999) 163.
[34] C.T. Campbell, Surf. Sci. Rep. 27 (1997) 1.
[35] C.R. Henry, Surf. Sci. Rep. 31 (1998) 231.
[36] R. Ho€mann, Solids and Surfaces. A Chemist's View of Bonding in Extended Structures, VCH,
Weinheim, 1988.
[37] R.A. van Santen, M. Neurock, Catal. Rev. Sci. Eng. 37 (1995) 557.
[38] B. Hammer, J.K. Nùrskov, in: R.M. Lambers, G. Pacchioni (Eds.), Theory of Adsorption and
Surface Reactions, Kluwer, 1997, p. 285.
[39] R.A. van Santen, M. Neurock, Chemisorption theory, in: G. Ertl, H. KnoÈzinger, J. Weitkamp
(Eds.), Handbook of Heterogeneous Catalysis, vol. 3, Wiley-VCH, Weinheim, 1997, p. 942.
[40] R.A. van Santen, M. Neurock, Theory of surface±chemical reactivity, in: G. Ertl, H. KnoÈzinger,
J. Weitkamp (Eds.), Handbook of Heterogeneous Catalysis, vol. 3, Wiley-VCH, Weinheim,
1997, p. 991.
[41] V.P. Zhdanov, B. Kasemo, Surf. Sci. Rep. 20 (1994) 111.
[42] P. Stoltze, J.K. Nùrskov, Phys. Rev. Lett. 55 (1985) 2502.
[43] P. Stoltze, Phys. Scripta 86 (1987) 824.
[44] J.K. Nùrskov, P. Stoltze, Surf. Sci. 189/190 (1987) 91.
[45] C.T. Campbell, Topics Catal. 1 (1994) 353.
[46] J.K. Nùrskov, P. Stoltze, Theoretical modeling of catalytic reactions, in: G. Ertl, H. KnoÈzinger,
J. Weitkamp (Eds.), Handbook of Heterogeneous Catalysis, vol. 3, Wiley-VCH, Weinheim,
1997, p. 984.
[47] T. Pignet, L.D. Schmidt, Chem. Eng. Sci. 29 (1974) 1123.
[48] M. Bowker, I. Parker, K.C. Waugh, Appl. Catal. 14 (1985) 101.
[49] P. Stoltze, J.K. Nùrskov, J. Vac. Sci. Technol. A 5 (1987) 581.
[50] M. Bowker, I. Parker, K.C. Waugh, Surf. Sci. 197 (1988) L223.
[51] P. Stoltze, J.K. Nùrskov, J. Catal. 110 (1988) 1.
[52] M. Bowker, I.B. Parker, K.C. Waugh, J. Catal. 114 (1988) 457.
[53] J.A. Dumesic, A.A. Trevino, J. Catal. 116 (1989) 119.
[54] L.M. Aparicio, J.A. Dumesic, Topics Catal. 1 (1994) 233.
[55] E. Christo€ersen, J.J. Mortensen, P. Stoltze, J.K. Nùrskov, Israel J. Chem. 38 (1998) 273.
[56] P. Stoltze, J.K. Nùrskov, Surf. Sci. 197 (1988) L230.
[57] B. Fastrup, M. Muhler, H. NygaÊrd Nielsen, L. Pleth Nielsen, J. Catal. 142 (1993) 135.
[58] B. Fastrup, Topics Catal. 1 (1994) 273.
[59] P. Stoltze, J.K. Nùrskov, Topics Catal. 1 (1994) 253.
[60] S.H. Oh, G.B. Fisher, J.E. Carpenter, D.W. Goodman, J. Catal. 100 (1986) 360.
[61] G.B. Fisher, S.H. Oh, J.E. Carpenter, C.L. DiMaggio, S.J. Schmieg, D.W. Goodman, T.W.
Root, S.B. Schwarts, L.D. Schmidt, Stud. Surf. Sci. Catal. 30 (1987) 215.
[62] K.Y.S. Ng, D.N. Belton, S.J. Schmieg, G.B. Fisher, J. Catal. 146 (1994) 394.
[63] V.P. Zhdanov, Surf. Sci. 296 (1994) 261.
[64] V.P. Zhdanov, B. Kasemo, Catal. Lett. 40 (1996) 197.
[65] D.N. Belton, S.J. Schmieg, J. Catal. 138 (1992) 70.
[66] D.N. Belton, C.L. DiMaggio, K.Y.S. Ng, J. Catal. 144 (1993) 273.
[67] P.R. Norton, in: D.A. King, D.P. Woodru€ (Eds.), The Chemical Physics of Solid Surfaces and
Heterogeneous Catalysis, vol. 4, Elsevier, Amsterdam, 1982, p. 70.
[68] J.L. Gland, G.B. Fisher, E.B. Kollin, J. Catal. 77 (1982) 263.
[69] T. Engel, H. Kuipers, Surf. Sci. 90 (1979) 181.
[70] K.M. Ogle, J.M. White, Surf. Sci. 139 (1984) 43.
[71] R.W. Williams, C.M. Mark, L.D. Schmidt, J. Phys. Chem. 96 (1992) 5922.
[72] V.P. Zhdanov, Surf. Sci. 169 (1986) 1.
[73] B. Hellsing, B. Kasemo, A. Rosen, S. LjungstroÈm, T. WahnstroÈm, Surf. Sci. 189/190 (1987) 851.
[74] B. Ljungstrùm, B. Kasemo, A. Rosen, T. WahnstroÈm, E. Fridell, Surf. Sci. 216 (1989) 63.
148 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

[75] L.K. Verheij, M.B. Hugenschmidt, A.B. Anton, B. Poelsema, G. Comsa, Surf. Sci. 210 (1989) 1.
[76] T.A. Germer, W. Ho, Chem. Phys. Lett. 163 (1989) 1890.
[77] L.K. Verheij, M.B. Hugenschmidt, B. Poelsema, G. Comsa, Chem. Phys. Lett. 174 (1990) 449.
[78] L.K. Verheij, M.B. Hugenschmidt, B. Poelsema, G. Comsa, Surf. Sci. 233 (1990) 209.
[79] A.B. Anton, D.C. Codogan, J. Vac. Sci. Technol. A 9 (1991) 1890.
[80] B. Hellsing, B. Kasemo, V.P. Zhdanov, J. Catal. 132 (1991) 210.
[81] E. Fridell, B. Hellsing, B. Kasemo, S. LjungstroÈm, A. Rosen, T. WahnstroÈm, J. Vac. Sci.
Technol. A 9 (1991) 2322.
[82] V.J. Kwasniewski, L.D. Schmidt, J. Phys. Chem. 96 (1992) 5931.
[83] L.K. Verheij, M. Freitag, M.B. Hugenschmidt, I. Kempf, B. Poelsema, G. Comsa, Surf. Sci. 272
(1992) 276.
[84] F. Gudmundson, E. Fridell, A. Rosen, B. Kasemo, J. Phys. Chem. 97 (1993) 12,828.
[85] L.K. Verheij, Surf. Sci. 371 (1997) 100.
[86] F. Gudmundson, J.L. Persson, M. Forsth, F. Behrendt, B. Kasemo, A. Rosen, J. Catal. 179
(1998) 420.
[87] Mallen M.P. Zum, W.R. Williams, L.D. Schmidt, J. Phys. Chem. 97 (1993) 625.
[88] M. Fassihi, V.P. Zhdanov, M. Rinnemo, K.-E. Keck, B. Kasemo, J. Catal. 141 (1993) 438.
[89] M. Rinnemo, O. Deutschmann, F. Behrendt, B. Kasemo, Combustion and Flame 111 (1997)
312.
[90] G. Ertl, P.R. Norton, J. RuÈstig, Phys. Rev. Lett. 49 (1982) 177.
[91] R. Imbihl, M.P. Cox, G. Ertl, M. MuÈller, W. Brenig, J. Chem. Phys. 83 (1985) 1578.
[92] R.M. Zi€, E. Gularu, Y. Barshad, Phys. Rev. Lett. 56 (1986) 2553.
[93] V.P. Zhdanov, B. Kasemo, J. Stat. Phys. 90 (1998) 79.
[94] V.P. Zhdanov, B. Kasemo, J. Catal. 170 (1997) 377.
[95] V.P. Zhdanov, B. Kasemo, Physica D 70 (1994) 383.
[96] R.A. van Santen, Avd Runstraat, R.J. Gelten, Stud. Surf. Sci. Catal. 109 (1997) 61.
[97] S.B. Schwartz, L.D. Schmidt, G.B. Fisher, J. Phys. Chem. 90 (1986) 6194.
[98] G.B. Fisher, S.H. Oh, J.E. Carpenter, C.L. DiMaggio, S.J. Schmieg, in: A. Crucq, A. Frenet
(Eds.), Catalysis and Automotive Pollution Control, Elsevier, Amsterdam, 1987.
[99] M.P. Harold, M.E. Garske, J. Catal. 127 (1991) 553.
[100] M.P. Harold, M.E. Garske, J. Catal. 127 (1991) 524.
[101] M.E. Garske, M.P. Harold, Chem. Eng. Sci. 47 (1992) 623.
[102] M. Rinnemo, D. Kulginov, S. Johansson, K.L. Wong, V.P. Zhdanov, B. Kasemo, Surf. Sci. 276
(1997) 297.
[103] N. Waletzko, L.D. Schmidt, AIChE J. 34 (1988) 1146.
[104] R.D. Cortright, S.A. Goddard, J.E. Rekotske, J.A. Dumesic, J. Catal. 127 (1991) 342.
[105] J.E. Rekoske, R.D. Cortright, S.A. Goddard, S.B. Sharma, J.A. Dumesic, J. Phys. Chem. 96
(1992) 1880.
[106] S.A. Goddard, R.D. Cortright, J.A. Dumesic, J. Catal. 137 (1992) 186.
[107] R.D. Cortright, J.A. Dumesic, J. Catal. 148 (1994) 771.
[108] S. Pietrzyk, Mahmuud M.L. Ould Mohamed, T. Rembeczky, R. Bechara, M. Czernicki, N.
Fatah, Stud. Surf. Sci. Catal. 109 (1997) 263.
[109] I. Machin, T. Romero, M.M. Ramirez de Agudelo, Stud. Surf. Sci. Catal. 109 (1997) 517.
[110] C.V. Ovesen, P. Stoltze, J.K. Nùrskov, C.T. Campbell, J. Catal. 134 (1992) 445.
[111] K.-H. Ernst, C.T. Campbell, G. Mortelli, J. Catal. 134 (1992) 66.
[112] T.S. Askgaard, J.K. Nùrskov, C.V. Ovesen, P. Stoltze, J. Catal. 156 (1995) 229.
[113] C.V. Ovesen, B.S. Clausen, B.S. Hammershùi, G. Ste€ensen, T. Askgaard, I. Chorkendor€, J.K.
Nùrskov, P.B. Rasmussen, P. Stoltze, P. Taylor, J. Catal. 158 (1996) 170.
[114] K.C. Waugh, Chem. Eng. Sci. 51 (1996) 1533.
[115] E. Tserpe, K.C. Waugh, Stud. Surf. Sci. Catal. 109 (1997) 401.
[116] O. Deutschmann, F. Behrendt, J. Warnatz, Catal. Today 21 (1994) 461.
[117] N.D. Spencer, C.J. Pereira, R.K. Grasselli, J. Catal. 126 (1990) 546.
[118] D.A. Hickman, L.D. Schmidt, AIChE J. 39 (1993) 1164.
[119] D.A. Hickman, E.A. Hauptfear, L.D. Schmidt, Catal. Lett. 17 (1993) 223.
P. Stoltze / Progress in Surface Science 65 (2000) 65±150 149

[120] P. Salomonsson, S. Johansson, B. Kasemo, Catal. Lett. 33 (1995) 1.


[121] D.A. Hickman, L.D. Schmidt, J. Catal. 136 (1992) 300.
[122] F. Behnrendt, O. Deutschmann, U. Maas, J. Warnatz, J. Vac. Sci. Catal. A 13 (1995) 1373.
[123] G. Veser, J. Frauhammer, L.D. Schmidt, G. Eigengerger, Stud. Surf. Sci. Catal. 109 (1997) 273.
[124] M. Ziauddin, G. Veser, L.D. Schmidt, Catal. Lett. 46 (1997) 159.
[125] P.B. Rasmussen, P.M. Holmblad, T. Askgaard, C.V. Ovesen, P. Stoltze, J.K. Nùrskov, I.
Chorkendor€, Catal. Lett. 26 (1994) 373.
[126] C.V. Ovesen, B.S. Clausen, J. Schiùtz, P. Stoltze, H. Topsùe, J.K. Nùrskov, J. Catal. 168 (1997)
7133.
[127] H. Topsùe, C.V. Ovesen, B.S. Clausen, N.-Y. Topsùe, Stud. Surf. Sci. Catal. 109 (1997) 121.
[128] J.-P. Stringaro, G. Gut, Chemia 35 (1991) 231.
[129] N.-Y. Topsùe, H. Topsùe, J.A. Dumesic, J. Catal. 151 (1995) 226.
[130] N.-Y. Topsùe, H. Topsùe, J.A. Dumesic, J. Catal. 151 (1995) 241.
[131] J.A. Dumesic, N.-Y. Topsùe, H. Topsùe, Y. Chen, T. Slabiak, J. Catal. 163 (1996) 409.
[132] S.W. Weller, ACS Adv. Chem. Ser. 148 (1975) 26.
[133] I. Langmuir, J. Am. Chem. Soc. 34 (1912) 1310.
[134] I. Langmuir, J. Am. Chem. Soc. 40 (1918) 1361.
[135] S. Glasstone, K.J. Laidler, H. Eyring, The Theory of Rate Processes, McGraw-Hill, New York,
1941.
[136] O.A. Hougen, K.M. Watson, Ind. Eng. Chem. 35 (1943) 227.
[137] C.N. Hinshelwood, The Kinetics of Chemical Change, Clarendon Press, Oxford, 1945.
[138] M. Boudart, AIChE J. 2 (1956) 62.
[139] J.H. Happel, J. Res. Inst. Catal. Hokkaido Univ. 16 (1968) 305.
[140] J.L. Falkoner, J.A. Schwarts, Catal. Rev. Sci. Eng. 25 (1983) 141.
[141] M. Boudart, Ind. Eng. Chem. Fund. 25 (1986) 656.
[142] C.B. Mullins, W.H. Weinberg, Dynamics of surface reactions, in: G. Ertl, H. KnoÈzinger, J.
Weitkamp (Eds.), Handbook of Heterogeneous Catalysis, vol. 3, Wiley-VCH, Weinheim, 1997,
p. 972.
[143] S. Pick, Surf. Sci. Rep. 12 (1990) 99.
[144] V.P. Zhdanov, Surf. Sci. 219 (1989) L571.
[145] M. Bernasconi, E. Tosatti, Surf. Sci. Rep. 17 (1993) 363.
[146] V.P. Zhdanov, B. Kasemo, Surf. Sci. 412/413 (1998) 527.
[147] V.P. Zhdanov, B. Kasemo, J. Chem. Phys. 109 (1998) 4582.
[148] V.P. Zhdanov, B. Kasemo, Phys. Rev. B 55 (1997) 4105.
[149] V.P. Zhdanov, B. Kasemo, Phys. Rev. Lett. 81 (1998) 2482.
[150] A. Cassuto, D.A. King, Surf. Sci. 102 (1981) 388.
[151] S.J. Lombardo, A.T. Bell, Surf. Sci. Rep. 13 (1991) 1.
[152] G. Zgrablich, J.L. Sales, R. Unac, V.P. Zhdanov, Surf. Sci. 290 (1993) 163.
[153] V.P. Zhdanov, Surf. Rev. Lett. 5 (1998) 977.
[154] M. Silverberg, A. Ben-Shaul, F. Rebentrost, J. Chem. Phys. 83 (1985) 6501.
[155] B. Hellsing, V.P. Zhdanov, Chem. Phys. Lett. 147 (1988) 136.
[156] M. Silverberg, A. Ben-Shaul, J. Chem. Phys. 87 (1987) 3178.
[157] M. Silverberg, A. Ben-Shaul, Surf. Sci. 214 (1989) 17.
[158] H.C. Kang, W.J. Weinberg, Surf. Sci. 299/300 (1994) 755.
[159] P. Kisliuk, J. Phys. Chem. Solids 3 (1957) 95.
[160] P. Kisliuk, J. Phys. Chem. Solids 5 (1958) 78.
[161] R. Gorte, L.D. Schmidt, Surf. Sci. 76 (1978) 559.
[162] V.P. Zhdanov, B. Kasemo, Chem. Phys. 177 (1993) 519.
[163] M.W. Leslety, L.D. Schmidt, Chem. Phys. Lett. 102 (1983) 459.
[164] J. Wei, C.D. Prater, Adv. Catal. 13 (1962) 203.
[165] R. Aris, AIChE J. 35 (1989) 539.
[166] S. Holloway, Surf. Sci. 299/300 (1994) 656.
[167] J.C. Tully, Surf. Sci. 299/300 (1994) 667.
[168] R. Aris, Arch. Rational Mech. Anal. 19 (1965) 81.
150 P. Stoltze / Progress in Surface Science 65 (2000) 65±150

[169] R. Aris, Arch. Rational Mech. Anal. 27 (1968) 35.


[170] O.A. Hougen, K.M. Watson, Chemical Process Principles, Wiley, New York, 1947.
[171] M.I. Temkin, Kinet. Catal. 8 (1967) 865.
[172] G.L. Haller, G.W. Coulston, Dynamics of heterogeneous catalyzed reactions, in: J.R. Anderson,
M. Boudart (Eds.), Catal. Sci. Technol., vol. 9, Springer, 1991, p. 131.
[173] M. Boudart, Rate of catalytic reactions, in: G. Ertl, H. KnoÈzinger, J. Weitkamp (Eds.),
Handbook of Heterogeneous Catalysis, vol. 3, Wiley-VCH, Weinheim, 1997, p. 958.
[174] G. Iche, Ph Nozieres, J. de Physique 37 (1976) 1313.
[175] J. Horiuti, Adv. Catal. 9 (1957) 359.
[176] J.A. Dumesic, J. Catal. 185 (1999) 496.
[177] J. Horiuti, J. Res. Inst. Catal. Hokkaido Univ. 5 (1957) 1.
[178] J. Horiuti, T. Nakamura, Adv. Catal. 17 (1967) 1.
[179] J. Horiuti, J. Res. Inst. Catal. Hokkaido Univ. 16 (1968) 501.
[180] P. Stoltze, Computer Simulation Methods in Atomic-Scale Materials Physics, Polyteknisk
Forlag, Lyngby, 1997.

You might also like