See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/319502349

Violation of collision limit in recently published

reaction models

Article in Combustion and Flame · September 2017

CITATIONS READS

0 60

3 authors:

Dongping Chen Kun Wang

Stanford University Stanford University
14 PUBLICATIONS 100 CITATIONS 29 PUBLICATIONS 74 CITATIONS

SEE PROFILE SEE PROFILE

Hai Wang
Stanford University
209 PUBLICATIONS 8,406 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

morphology of carbon nanoparticles View project

combustion kinetic View project

All content following this page was uploaded by Dongping Chen on 06 September 2017.

The user has requested enhancement of the downloaded file.

 Combustion and Flame 186 (2017) 208–210

Contents lists available at ScienceDirect

Combustion and Flame

journal homepage: www.elsevier.com/locate/combustflame

Brief Communications

Violation of collision limit in recently published reaction models

Dongping Chen, Kun Wang, Hai Wang∗
Mechanical Engineering Department, Stanford University, Stanford, CA 94305-3032, USA

a r t i c l e i n f o a b s t r a c t

Article history: Twenty reaction models published in five consecutive issues of Combustion and Flame in 2015 and 2016
Received 12 July 2017 were screened for the occurrence of collision limit violations. It was found that among the 20 models
Revised 10 August 2017
tested, 15 of them contain either considerable numbers of rate coefficients that exceed their respective
Accepted 11 August 2017
collision limits or reactions exceeding the collision limit in a considerable manner. In the worst case, the
rate coefficient exceeds the collision limit by 73 orders of magnitude. It is proposed that computational
tools should be made available for authors to conduct the same rate coefficient screening. A standard
report generated by such screening tools should be submitted as a part of the supplementary material
for a manuscript in which one or more reaction models are proposed; reviewers should consult the report
in forming his/her opinions about the quality of the reaction model during manuscript review.
© 2017 The Combustion Institute. Published by Elsevier Inc. All rights reserved.

The combustion research community has a long-standing his-
tory in using reaction mechanisms or reaction models1 to explore
the underlying physics of complex combustion phenomena [1–3].
Since the work of Hirschfelder, Curtiss, and others [4–7] in the
1950s and by Dixon-Lewis, Gardiner and others in the 1960s
[8–11], and of Warnatz, Westbrook, Dryer and others into the
1980s (see, e.g., [12,13]), the use of reaction models has been
instrumental in advancing our basic understanding of combustion.
In the 1980s, the development of the highly-utilized Chemkin suite
of package [14–17] enabled rapid advances of kinetic modeling of
combustion processes.
Recent studies in chemical kinetic modeling typically employ
large reaction models which in some cases can have 105 reactions
and 104 species [18]. The growth in the size of the reaction mod-
els is unfortunately not matched by an equal magnitude of high-
quality measurements or theoretical calculations of fundamental
reaction rates. Today, chemical similarity or reaction class rules are
often used to estimate a vast number of rate coefficients. When
done carefully and properly, the resulting model can yield use-
ful information about a combustion process, despite the fact that
the model is largely empirical. When chemical similarity or reac-
tion rules are exercised inappropriately (e.g., mismatched rate co-
efficients and thermochemical data), non-physical predictions can
occur. Symptoms of such ill-exercised rules include a substantial
∗
Corresponding author.
E-mail address: haiwang@stanford.edu (H. Wang).
1
The use of the word “mechanism” is for historical reasons. In fact, “reaction
model” is more appropriate here because a mechanism refers to how nature works
whereas a model is what we use to mimic how nature works. A reaction model
involves assumptions which may or may not be verified by isolated experiments.
number of rate coefficients exceeding the collision limit. For exam-
ple, if a rate rule considers only the forward rate coefficient of an
endothermic reaction, it can produce an overly large back rate co-
efficient if the equilibrium constant of the reaction is not properly
taken into consideration.
To illustrate the problem, we sampled 20 reaction models
[19–35] published in five consecutive issues of Combustion and
Flame (Volume 162 Issues 11 and 12, Volume 163, Volume 164 and
Volume 165) for violation of the collision limit in bimolecular re-
action rate coefficients. In all cases, Chemkin compatible reaction
models and their thermochemical and transport data were directly
taken from the Supplementary Materials of the respective papers.
For pressure-dependent reactions, only the high-pressure limit bi-
molecular reactions were considered. The backward reaction rate
coefficient was computed via the equilibrium constant if no ex-
plicit rate expression is given. Reactions that are declared as irre-
versible reactions are not screened for the rate coefficients in the
reverse direction.
We calculate the collision rate constant kcoll using

8π kB T ∗
kcoll = σ 2 (1,1) Na (1)
μ
where T is temperature, kB is the Boltzmann constant, μ is reduced
mass, σ is the collision diameter of the Lennard–Jones (LJ) 12-6
∗
potential, Na is the Avogadro number, and (1,1 ) is the reduced
collision integral, which may be calculated in a parameterized form
[36] as
∗ =1.16145T ∗−0.14874 + 0.52487e−0.7732T ∗ + 2.16178e−2.437887T ∗
(2)

http://dx.doi.org/10.1016/j.combustflame.2017.08.005
0010-2180/© 2017 The Combustion Institute. Published by Elsevier Inc. All rights reserved.
 D. Chen et al. / Combustion and Flame 186 (2017) 208–210 209

Table 1
Number of reactions having rate coefficient violating the collision limit and the largest ratio of rate coefficient to its collision limit.a

Model no. Total number of bimolecular Applicable T Number of rate coefficients exceeding collision limit
rate coefficients range (K)b
Applicable
T range Tmin = 300 K Tmin = 500 K Tmin = 10 0 0 K

Group 1 (transport database available. kcoll (T) calculated from Eq. (1))
1 2407 298–2381 13 (18) 13 (18) 6 (18) 3 (18)
2 4234 298–1941 50 (1 × 1020 ) 56 (1 × 1020 ) 35 (4 × 104 ) 17 (130)
3 903 313–2050 10 (2 × 1014 ) 10 (2 × 1014 ) 6 (7 × 106 ) 1 (22)
4 2945 298–2200 31 (1 × 104 ) 31 (1 × 104 ) 11 (2 × 102 ) 6 (2 × 102 )
5 1716 403–2300 9 (6) 15 (11) 12 (6) 11 (4)
6 1655 30 0–180 0 39 (2 × 1019 ) 40 (2 × 1019 ) 26 (2 × 1010 ) 9 (3 × 103 )
7 13132 50 0–110 0 9 (8 × 103 ) 20 (3 × 107 ) 10 (8 × 103 ) 6 (9)
8 1178 800–1650 3 (147) 12 (7 × 102 ) 8 (170) 4 (130)
9 807 60 0–180 0 5 (5 × 108 ) 11 (3 × 1015 ) 6 (7 × 1010 ) 4 (2 × 104 )
10 4361 298–2200 60 (1 × 1020 ) 66 (1 × 1020 ) 42 (3 × 106 ) 21 (3 × 102 )
11 2497 500–2319 13 (9 × 103 ) 19 (3 × 109 ) 13 (9 × 103 ) 7 (129)
12 7374 70 0–210 0 0 (<1) 7 (15) 2 (1.5) 0 (<1)
13 3513 30 0–210 0 36 (2 × 1017 ) 38 (2 × 1017 ) 28 (3 × 109 ) 17 (4 × 103 )
Group 2 (transport database unavailable. kcoll = 2.9 × 1015 cm3 /mol s used as the limit)
14 6338 530–1365 26 (2 × 105 ) 41 (3 × 1010 ) 29 (2 × 105 ) 5 (50)
15 41 50 0–110 0 0 (<1) 0 (<1) 0 (<1) 0 (<1)
16 3749 598–1148 4 (2 × 1014 ) 13 (1 × 1031 ) 7 (5 × 1017 ) 2 (2 × 107 )
17 1417 298–1773 56 (2 × 1073 ) 58 (2 × 1073 ) 57 (1 × 1069 ) 55 (5 × 1034 )
18 378 30 0–250 0 6 (2 × 103 ) 6 (2 × 103 ) 5 (90) 4 (13)
19 4482 643–1718 5 (1 × 104 ) 19 (3 × 1010 ) 8 (2 × 105 ) 5 (50)
20 2733 450–1400 17 (5 × 105 ) 31 (2 × 108 ) 12 (1 × 106 ) 4 (1 × 106 )
a
In all cases, Tmax in Eq. (4) was set at 2500 K except for the column showing the results in the applicable range of temperature
of the model. The values in the parentheses refer to the largest ratio of rate coefficient to its collision limit.
b
The range values are extracted from the respective papers and correspond to the minimum and maximum temperature values
of the experiments for which the model was tested against.

where T∗ is reduced temperature (T∗ = kB T/ε , where ε is the well
depth). The collision diameter and well depth are calculated con-
ventionally as the arithmetic and geometric means of the self-
collision potential parameters, respectively. For reaction models
that did not include a transport database we used the colli-
sion rate coefficient of H collision with n-dodecane at 2500 K or
2.9 × 1015 cm3 /mol s as an upper limit of the collision rate constant.
For the 20 reaction models sampled, Table 1 shows the total
number of bimolecular-reaction rate coefficients, the number of
these rate coefficients that exceed their respective collision limits,
and the largest ratio of the rate coefficient to its collision limit. Ad-
ditional test results can be found in the Supplementary Materials.
The models are listed in a random order, though they are separated
into two groups: Group 1 is comprised of those with transport pa-
rameters provided as a part of the model release/publication, and
Group 2 contains those without transport parameters provided.
The highest rate ratios are defined as:
 
κ = max k∗i
i=1,I
where I is the total number of bimolecular reaction rate coeffi-
cients, and k∗i is the maximum of the ratio of the rate coeffi-
cient to its collision limit over a prescribed range of temperature
Tmin ≤ T ≤ Tmax ,
k∗i = max [ki (T )/kcoll,i (T )]
Tmin ≤T ≤Tmax
Tests were also made in which the Tmin and Tmax values were
taken from the applicable range of temperature in which a model
was tested or applied (see Table 1). Additional tests were made
with Tmax = 2500 K and Tmin = 300, 500 and 1000 K. The resulting
temperature ranges are sometimes larger than the explicit or im-
plied applicable temperature range of a reaction model. The tests
were performed considering the fact that a user may apply the
model to temperatures outside the range of applicability.
A physically sound reaction model should have no reaction with
its k∗i value substantially greater than unity over the applicable
temperature range of the model. As Table 1 reveals, this is not the
case for most of the models tested. In our assessment, five models
(1, 5, 8, 12 and 15) are acceptable as they contain no rate coeffi-
cient that exceeds the collision limit or the violations are caused
by difficulties in fitting some of the rate coefficients into the mod-
ified Arrhenius equation. Four models (4, 7, 11 and 18) are found to
be on the edge, and the remaining eleven models are quite unsat-
isfactory in two aspects: the number of rate coefficients exceeding
the collision limit and the extent to which they exceed the colli-
sion limit. The number of nonphysical rate coefficients is usually
not large as compared to the total number of rate coefficients, but
the extent to which the rate coefficients exceed the collision limit
is often disturbing. In the worst case, the ratio is of the order of
1073 due to mismatched thermochemical data assigned to species
participating in the reaction and its rate coefficient.
The problem usually worsens toward lower temperatures be-
cause of mismatched assignments of the forward rate coefficient
and species thermochemistry. The kinetic rate-thermochemistry
(3) mismatch is quite worrisome in models 2, 3, 6, 9, 10, 13, 14, 16,
17, 19 and 20. In other cases, a non-physical rate coefficient is
caused by errors associated with extrapolating a rate expression
to low temperatures when the rate expression was obtained in
high-temperature experiments. Errors due to typos and other care-
lessness are also more common than expected. Obviously, it is not
straightforward for a reviewer to spot the rate parameter errors
(4) during the review process, especially considering the size of many
of the reaction models shown in Table 1.
Models 1, 5 and 8 have k∗ values on the order of 101 or 102 .
The problem is sometimes unavoidable, as it originates from diffi-
culties in fitting rate coefficients into a modified Arrhenius expres-
sion over an extended range of temperature. The rate coefficients
of many multi-channel, chemically activated reactions do not fol-
low the Arrhenius expression. In such cases, the range of applica-
bility of a rate expression does not extend below a cut-off temper-
ature, and the authors usually ensure that non-physical rate val-
ues do not impact model predictions. To this end, we note that
nonphysical rate coefficients do not always translate into nonphys-
210 D. Chen et al. / Combustion and Flame 186 (2017) 208–210
ical predictions. For example, for reacting flows that are close to
or governed by chemical equilibrium or chemistry is not rate lim-
iting in an overall reactive process, nonphysical rate coefficients
would not produce an appreciable effect on the prediction. Yet in
this case, the range of applicability of a model containing non-
physical rate constant must be very limited. Indeed, extending the
model to thermodynamic conditions beyond its intended use un-
der any circumstances can be a source of the problem especially
for reaction models that contain nonphysical rate coefficients. We
note that the unwritten rule or notion that an experimental kinetic
measurement must accompany itself with a detailed kinetic mod-
eling component in order for the paper to be accepted is rather
misinformed and can hinder proper interpretation of the data and
gaining useful insight from the measurement.
From these results, one may conclude that around three quar-
ters of the reaction models published recently contain non-physical
rate coefficients. To help to resolve this problem, we wish to make
the following specific recommendations.
1. It is crucial for reaction model developers to pay attention to
collision-limit violations;
2. When submitting a manuscript for publication, authors have
the responsibility to provide explanations in places when and
where collision rate violations occur;
3. Computational tools, e.g., those built for Chemkin and Cantera,
should be made available for authors to conduct the rate coef-
ficient screening. A standard report generated by such screen-
ing tools can be submitted as a supplementary material for
a manuscript in which one or more reaction models are pro-
posed;
4. Reviewers should consult the report in forming his/her opin-
ions about the quality of the reaction model during manuscript
review.
Editorial policies of a similar nature exist in other fields. Ex-
amples include the demonstration of grid-independent or grid-
convergent results in computational fluid dynamics, which has
been implemented and utilized with success for many years.
Supplementary materials
Supplementary material associated with this article can be
found, in the online version, at doi:10.1016/j.combustflame.2017.08.
005.
References
[1] J.A. Miller, R.J. Kee, C.K. Westbrook, Chemical kinetics and combustion model-
ing, Ann. Rev. Phys. Chem. 41 (1990) 345–387.
[2] J.A. Miller, M.J. Pilling, J. Troe, Unravelling combustion mechanisms through
a quantitative understanding of elementary reactions, Proc. Combust. Inst. 30
(2005) 43–88.
[3] J.M. Simmie, Detailed chemical kinetic models for the combustion of hydrocar-
bon fuels, Prog. Energy Combust. Sci. 29 (2003) 599–634.
[4] J. Hirschfelder, C. Curtiss, D.E. Campbell, The theory of flame propagation. IV,
J. Phys. Chem. 57 (1953) 403–414.
[5] C. Curtiss, J.O. Hirschfelder, Integration of stiff equations, Proc. Natl. Acad. Sci.
38 (1952) 235–243.
[6] R.E. Duff, Calculation of reaction profiles behind steady state shock waves. I.
Application to detonation waves, J. Chem. Phys. 28 (1958) 1193–1197.
[7] H. Glick, J. Klein, W. Squire, Single pulse shock tube studies of the kinetics of
the reaction N2 + O2  2NO between 20 0 0–30 0 0 K, J. Chem. Phys. 27 (1957)
850–857.
[8] G. Dixon-Lewis, Flame structure and flame reaction kinetics. V. Investigation of
reaction mechanism in a rich hydrogen + nitrogen + oxygen flame by solution
of conservation equations, Proc. R. Soc. A 317 (1970) 235–263.
[9] G. Dixon-Lewis, Flame structure and flame reaction kinetics. I. Solution of con-
servation equations and application to rich hydrogen–oxygen flames, Proc. R.
Soc. A 298 (1967) 495–513.
View publication stats
[10] T. Asaba, W. Gardiner, R. Stubbeman, Shock-tube study of the hydrogen–oxy-
gen reaction, Symp. (Int.) Combust. 10 (1965) 295–302.
[11] M. Clyne, B. Thrush, Rates of the reactions of nitrogen atoms with oxygen and
with nitric oxide, Nature 189 (1961) 56–57.
[12] J. Warnatz, The structure of laminar alkane-, alkene-, and acetylene flames,
Symp. (Int.) Combust. 18 (1981) 369–384.
[13] C.K. Westbrook, F.L. Dryer, Chemical kinetic modeling of hydrocarbon combus-
tion, Prog. Energy Combust. Sci. 10 (1984) 1–57.
[14] R.J. Kee, J.A. Miller, T.H. Jefferson, CHEMKIN: a general-purpose, problem-
independent, transportable, FORTRAN chemical kinetics code package, Report
No. Sandia Report SAND80-8003, Sandia National Laboratories, Albuquerque,
NM, 1980.
[15] R.J. Kee, J.F. Grcar, M. Smooke, J. Miller, E. Meeks, PREMIX: a Fortran pro-
gram for modeling steady laminar one-dimensional premixed flames, Report
No. Sandia Report SAND85-8240, Sandia National Laboratories, Albuquerque,
NM, 1985.
[16] A.E. Lutz, R.J. Kee, J.F. Grcar, F.M. Rupley, OPPDIF: a Fortran program for com-
puting opposed-flow diffusion flames, Report No. Sandia Report SAND96-8243,
Sandia National Laboratories, Albuquerque, NM, 1996.
[17] P. Glarborg, R. Kee, J. Grcar, J. Miller, PSE: a Fortran program for modeling well-
stirred reactors, Report No. Sandia Report SAND86-8209, Sandia National Lab-
oratories, Albuquerque, NM, 1986.
[18] T. Lu, C.K. Law, Toward accommodating realistic fuel chemistry in large-scale
computations, Prog. Energy Combust. Sci. 35 (2009) 192–215.
[19] A. Rodriguez, O. Herbinet, F. Battin-Leclerc, A. Frassoldati, T. Faravelli, E. Ranzi,
Experimental and modeling investigation of the effect of the unsaturation de-
gree on the gas-phase oxidation of fatty acid methyl esters found in biodiesel
fuels, Combust. Flame 164 (2016) 346–362.
[20] K.B. Brady, X. Hui, C.-J. Sung, Comparative study of the counterflow forced ig-
nition of the butanol isomers at atmospheric and elevated pressures, Combust.
Flame 165 (2016) 34–49.
[21] S. Banerjee, R. Tangko, D.A. Sheen, H. Wang, C.T. Bowman, An experimental
and kinetic modeling study of n-dodecane pyrolysis and oxidation, Combust.
Flame 163 (2016) 12–30.
[22] N. Lamoureux, H. El Merhubi, L. Pillier, S. de Persis, P. Desgroux, Modeling of
NO formation in low pressure premixed flames, Combust. Flame 163 (2016)
557–575.
[23] Z. Jia, Z. Wang, Z. Cheng, W. Zhou, Experimental and modeling study on
pyrolysis of n-decane initiated by nitromethane, Combust. Flame 165 (2016)
246–258.
[24] K. Wang, S.M. Villano, A.M. Dean, Fundamentally-based kinetic model for
propene pyrolysis, Combust. Flame 162 (2015) 4456–4470.
[25] J. Bugler, B. Marks, O. Mathieu, R. Archuleta, A. Camou, C. Grégoire, K.A. Heufer,
E.L. Petersen, H.J. Curran, An ignition delay time and chemical kinetic modeling
study of the pentane isomers, Combust. Flame 163 (2016) 138–156.
[26] L.-S. Tran, R. De Bruycker, H.-H. Carstensen, P.-A. Glaude, F. Monge,
M.U. Alzueta, R.C. Martin, F. Battin-Leclerc, K.M. Van Geem, G.B. Marin, Pyroly-
sis and combustion chemistry of tetrahydropyran: experimental and modeling
study, Combust. Flame 162 (2015) 4283–4303.
[27] A. Stagni, A. Frassoldati, A. Cuoci, T. Faravelli, E. Ranzi, Skeletal mechanism
reduction through species-targeted sensitivity analysis, Combust. Flame 163
(2016) 382–393.
[28] X. Liu, H. Wang, L. Wei, J. Liu, R.D. Reitz, M. Yao, Development of a reduced
toluene reference fuel (TRF)-2, 5-dimethylfuran-polycyclic aromatic hydrocar-
bon (PAH) mechanism for engine applications, Combust. Flame 165 (2016)
453–465.
[29] C. Brackmann, V.A. Alekseev, B. Zhou, E. Nordström, P.-E. Bengtsson, Z. Li,
M. Aldén, A.A. Konnov, Structure of premixed ammonia + air flames at at-
mospheric pressure: laser diagnostics and kinetic modeling, Combust. Flame
163 (2016) 370–381.
[30] K. Zhang, C. Banyon, C. Togbé, P. Dagaut, J. Bugler, H.J. Curran, An experimental
and kinetic modeling study of n-hexane oxidation, Combust. Flame 162 (2015)
4194–4207.
[31] H. Ning, C. Gong, N. Tan, Z. Li, X. Li, Low- and intermediate-temperature ox-
idation of ethylcyclohexane: a theoretical study, Combust. Flame 162 (2015)
4167–4182.
[32] X. Liu, M. Yao, Y. Wang, Z. Wang, H. Jin, L. Wei, Experimental and kinetic mod-
eling study of a rich and a stoichiometric low-pressure premixed laminar 2,
5-dimethylfuran/oxygen/argon flames, Combust. Flame 162 (2015) 4586–4597.
[33] W. Sun, B. Yang, N. Hansen, C.K. Westbrook, F. Zhang, G. Wang, K. Moshammer,
C.K. Law, An experimental and kinetic modeling study on dimethyl carbonate
(DMC) pyrolysis and combustion, Combust. Flame 164 (2016) 224–238.
[34] K. Narayanaswamy, H. Pitsch, P. Pepiot, A component library framework for
deriving kinetic mechanisms for multi-component fuel surrogates: application
for jet fuel surrogates, Combust. Flame 165 (2016) 288–309.
[35] U. Burke, W.K. Metcalfe, S.M. Burke, K.A. Heufer, P. Dagaut, H.J. Curran, A de-
tailed chemical kinetic modeling, ignition delay time and jet-stirred reactor
study of methanol oxidation, Combust. Flame 165 (2016) 125–136.
[36] P.D. Neufeld, A. Janzen, R. Aziz, Empirical equations to calculate 16 of the
transport collision integrals  (l,s)∗ for the Lennard–Jones (12–6) potential, J.
Chem. Phys. 57 (1972) 1100–1102.