Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited., Page 1 of 18
This manuscript was accepted and published by Industrial & Engineering Chemistry
Research, a journal of the American Chemical Society.
Publication data of the final, corrected work:
Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited.
Ind. Eng. Chem. Res. 1998, 37 1267-1275. doi:
10.1021/ie970144v
Cellulose Pyrolysis Kinetics: Revisited
Michael Jerry Antal, Jr.*
Hawaii Natural Energy Institute and the Department of Mechanical Engineering,
University of Hawaii at Manoa, Honolulu, Hawaii 96822
Gábor Várhegyi and Emma Jakab
Research Laboratory for Inorganic Chemistry, Hungarian Academy of Sciences,
Pf. 132, Budapest 1518, Hungary
Abstract
In the same thermogravimetric analyzer (TGA) under identical conditions, samples of pure, ash-free
cellulose (i.e. Avicel PH-105, Whatman CF-11, Millipore ash-free filter pulp, and Whatman #42)
obtained from different manufacturers undergo pyrolysis at temperatures which differ by as much as 30
C. Thus the pyrolysis chemistry of a sample of pure cellulose is not governed by a universal rate law, as
is the case with a pure hydrocarbon gas (for example). Nevertheless, the pyrolytic weight loss of all the
samples studied in this work is well represented by a high activation energy (228 kJ/mol), first order rate
law at both low and high heating rates. These results do not corroborate the recent findings of
Milosavljevic and Suuberg (1995). For a particular cellulose sample (for example Avicel PH-105),
variations in the pre-exponential constant determined at different heating rates reflect uncontrolled,
systematic errors in the dynamic sample temperature measurement (thermal lag).
Introduction
A recent review of the literature of cellulose pyrolysis (Antal and Várhegyi, 1995) concluded that the
pyrolysis of a small sample of pure cellulose is characterized by an endothermic reaction governed by a
first order rate law with a high activation energy (ca. 238 kJ/mol). We employed the terminology “high
activation energy (ca. 238 kJ/mol)” to indicate that, even for a particular cellulose (e.g. Avicel),
considerable uncertainty ( 10 kJ/mol or more) exists in any determination of the exact value of its
activation energy (see below). Almost immediately after the review was published, these conclusions
were contradicted by the findings of Milosavljevic and Suuberg (1995), who hypothesized the role of a
high temperature, low activation energy (140 - 155 kJ/mol) step competing with a low temperature (below
Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited., Page 2 of 18
327 C), high activation energy (218 kJ/mol) reaction during the pyrolysis of a Whatman CF-11 fibrous
cellulose powder. Milosavljevic and Suuberg argued that their findings were consistent with the results of
earlier kinetic studies of many workers, which were summarized in Figures 1 and 2 of their paper. They
used the kinetic data displayed in their Figure 1 to rationalize the existence of the purported high
temperature, low activation energy (140 - 155 kJ/mol) reaction. We were surprised to find our high
activation energy (205 kJ/mol) rate measurements at 80 K/min (Várhegyi et al., 1989) displayed in that
figure, with no discussion of the fact that our data contradicted their hypothesis. Also included in their
Figure 1 were the kinetics data of Tabatabaie-Raissi et al. (1989). But his measurements were made in a
covered sample pan that greatly enhanced vapor-solid interactions; consequently, his experiments did not
offer a true measurement of the primary rate of cellulose pyrolysis and should not have been used to assert
the existence of a low activation energy reaction. Likewise, the early rate measurements of Antal et al.
(1980) were used to argue the existence of the low activation energy step, but Antal and his co-workers
later showed (Antal, 1985; Antal et al., 1985; Antal and Várhegyi, 1995) that his early measurements were
compromised by heat transfer intrusions and did not represent a true determination of the temperature
dependence of the rate law. In summary, our scrutiny of Milosavljevic and Suuberg’s Figure 1 left us
with doubts about the role of a high temperature, low activation energy reaction in the pyrolysis of
cellulose. To examine the matter further, we measured the rates of pyrolysis of the same cellulose
employed by Milosavljevic and Suuberg (1995) in our equipment. We also studied the kinetics of other
cellulose samples to learn if different pure celluloses evidence markedly different pyrolysis behavior.
This paper documents the surprising results of our investigations.
Apparatus and Experimental Procedures
The Perkin Elmer TGS-2 thermobalance used in this work has been described in earlier publications
(Várhegyi et al., 1988). Linear heating rates of 1, 10 and 65ºC/min were employed. Since the highest
reaction rate of an experiment is roughly proportional to the heating rate, 0.3 mg sample masses were used
at 65ºC/min to keep the heat and mass fluxes low. But for such a small sample mass, buoyancy and other
base line shift effects distort the char yields; consequently the 0.3 mg experiments were corrected by
subtracting a TG curve measured under identical conditions with an empty sample pan. We also carried
out two test experiments at 1ºC/min with 0.35 mg CF-11 cellulose, where the char yield was not corrected
in this way. (Accordingly, these two tests were not included in the comparison of char yields discussed
below.) All the experiments described herein were conducted in high-purity argon with a flow rate of 140
mL/min.
We remark that we were able to measure the weight loss of samples as small as 0.3 mg because the
Perkin Elmer thermobalance used in this work possesses a sensitivity of 0.1 g. The Dupont 951
instrument employed by Milosavljevic and Suuberg (1995) possesses a sensitivity of only 2 g. The low
Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited., Page 3 of 18
sensitivity of the Dupont 951 instrument necessitates the use of large sample sizes, which considerably
alters the observed pyrolysis behavior (see below).
Professor Suuberg was kind enough to supply us with a sample of Whatman CF-11 fibrous cellulose
powder (0.15 wt % ash) taken from the same lot as the samples employed in his work. Thus there can be
no question concerning the influence of sample composition on results obtained in the two laboratories.
In addition, we examined the pyrolysis behavior of a Millipore ash-free cellulose filter pulp, and a
Whatman #42 cellulose filter paper (0.01 wt % ash). Finally, to facilitate comparison with our earlier
work, we executed some studies of an Avicel PH-105 microcrystalline cellulose with an ash content of
less than 10 ppm.
Results and Discussion
Systematic experimental errors. In this subsection we address the following questions. How accurate
is the rate of weight loss measurement determined by the Perkin Elmer instrument? How accurate is the
dynamic temperature measurement? Are weight loss vs. temperature measurements stable over time?
The decomposition of calcium oxalate monohydrate (CaC
2
O
4
.
H
2
O) is a benchmark measurement in
thermogravimetry. Figure 1 of Antal and Várhegyi (1995) displays a comparison of measured rates of gas
evolution during the decomposition of calcium oxalate by the Perkin Elmer thermobalance and a Balzers
QMG-511 mass spectrometer. The exact agreement of these two independent measurements indicates
that the thermogravimetric determination of the rate of weight loss is extremely accurate. This result is
consistent with the claimed high sensitivity (0.1 g) of the Perkin Elmer balance. To detect and estimate
systematic errors in temperature measurement we studied the behavior of a Nickel Curie-point calibrant
furnished by Perkin Elmer. At a heating rate of 10 C/min the measured Curie-point temperature was 358
C: 4 C above the actual Curie-point temperature of 354 C. At a heating rate of 65 C/min the
temperature measurement error increased to 11 C above the Curie-point value. These measurements
illustrate the pervasive (but usually unappreciated) presence of thermal lag during thermogravimetric
studies. We remark that heat demand during an endothermic pyrolysis event increases exponentially with
temperature according to Arrhenius kinetics. The Curie-point phase transition is effectively autothermal.
Consequently, Curie-point measurements of thermal lag only provide an accurate indication of the
temperature measurement error associated with infinitely small samples of cellulose. Finite samples
undergoing endothermic pyrolysis evoke larger thermal lags than those estimated by the Curie-point
determination. Recent careful experimental measurements by Di Blasi and Lanzetta (1997) and
Drummond and Drummond (1996), and numerical simulations by Narayan and Antal (1996) concur that
thermal lag (i.e. the difference between the true sample temperature and the thermocouple temperature)
increases dramatically with heating rate. The effects of this thermal lag on kinetic analysis will be
discussed later.
Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited., Page 4 of 18
Figure 1. Reproducibility of the Avicel experiments. Panel A: 10 experiments executed over a ten year
period at a nominal heating rate of 10 C/min. Measured, actual heating rates were 9.6 0.2 C/min. A
simultaneous kinetic evaluation of the TG curves, assuming identical kinetic parameters, resulted in E =
237 kJ/mol, log A/s
-1
=18.0, n = 1, and m
f
= 0.07. The simulated curve using these kinetic parameters at
the average, actual heating rate is denoted by □. Panel B: 3 experiments at 65 °C/min measured within
three consecutive days.
Figure 1a displays all the publishable quality weight loss measurements obtained in Budapest during
the past ten years of 2 - 3 mg samples of Avicel PH-105 cellulose heated at 10 C/min. The Avicel
celluloses were taken from cans stored in Honolulu and opened in 1986, 1990, and 1996. The ten weight
loss curves agree within 5 C at a fixed value of weight loss. We call our readers’ attention to the fact that
most of the curves displayed in Figure 1a lie parallel to one another: each is systematically displaced by
only a few degrees Celsius from the mean value. Moreover, the three omitted curves were also parallel to
those displayed, but systematically shifted in temperature by an unacceptable amount. What is the cause
of these systematic displacements of the weight loss curves? Over 15 years ago Antal et al. (1980)
reported systematic temperature shifts of 20 to 25 C at all heating rates in a Dupont 951 TGA, depending
upon whether the sample thermocouple was slightly upstream or slightly downstream of the 2 mg
cellulose sample. Thus, small changes in the placement of the thermocouple relative to the sample result
in shifts of the weight loss curves identical to those shifts displayed in Figure 1a. The three curves which
were omitted from Figure 1a were shifted by an unacceptable amount due to a misalignment of the
thermocouple. These observations highlight the importance of periodic Curie-point calibrations of the
TGA during an experimental campaign.
Figure 1b displays similar, recent data obtained at 65 C/min. At this higher heating rate the data from
the three Avicel PH-105 samples agree to within about 4 C. We attribute this good agreement to the fact
that the data was acquired over a two-day period, which involved no changes of the thermocouple’s
position.
Effects of systematic and random experimental errors on kinetic analysis. In this subsection we
address the following questions. Recognizing our inherent inability to obtain perfectly reproducible
Antal, M. J., Jr.; Várhegyi, G.; Jakab, E.: Cellulose pyrolysis kinetics: Revisited., Page 5 of 18
weight loss curves, how should kinetic analysis of various weight loss curves (obtained under the same
experimental conditions) be accomplished? Likewise, recognizing the increasing presence of thermal lag
and the concomitant loss of accuracy in measuring temperature at higher heating rates; how should the
kinetic analysis of data, obtained over a range of heating rates, be accomplished and interpreted?
Consider a kinetic analysis of the differential thermogravimetric curves (DTG curves) displayed in
Figure 1a. For the sake of illustration, we employ the first-order rate equation:
d/dt = k (1 - ) (1)
where = (m(t) - m
f
) / (m
0
- m
f
), k = A exp (-E/RT), m(t) is the time dependent sample mass, m
0
is the
initial sample mass, m
f
is the amount of solid residue formed during cellulose decomposition, A is the
pre-exponential constant, E is the apparent activation energy, R is the gas constant, and T is the sample
temperature. A non-linear, least squares (NLS) fit of this rate equation to all the data displayed in Figure
1a offers the following values for the three free parameters employed in the rate equation: E = 227
kJ/mol, log A/s
-1
= 17.17, and m
f
/m
0
= 0.059. As seen in Figure 1a, simulated TG and DTG curves, that
employ these kinetic parameters with the first order rate equation, enjoy satisfying agreement with the
experimental data. Thus no special problems arise in the kinetic evaluation of weight-loss curves
obtained under the same experimental conditions. We believe that the small systematic discrepancy
between the model and the experimental data in the temperature range 300 to 325 C is due to the low
temperature char forming pathway (Arseneau, 1971; Broido, 1976; Bradbury et al., 1979; Várhegyi et al.,
1994) which is not included in the first order kinetic model. We discuss this matter further below.
Table 1: Kinetic evaluation of Avicel cellulose DTG curves at different heating rates
(See Figures 2a and 2c).
Independent evaluation
(variable E)
Simultaneous evaluation
(identical E)
dT/dt
(C/min)
E
(kJ/mol)
log A
(log s
-1
)
m
f
fit
(%)
E
(kJ/mol)
log A
(log s
-1
)
m
f
fit
(%)
1
245
19.0
0.11
1.9
236
18.2
0.10
2.2
10
235
17.8
0.06
1.8
236
17.9
0.06
1.8
65
230
17.2
0.02
1.9
236
17.7
0.03
2.1
Figure 2a displays DTG curves for Avicel PH-105 cellulose at 1, 10 and 65 C/min, together with
simulated data using the first-order rate law with best-fit parameters E and log A determined individually
at each heating rate (see Table 1). The gradual decline in the values of both E and log A with increasing
heating rate is a benchmark conundrum of the field. Similar trends were reported by Antal et al. (1980)
over fifteen years ago, and more recently by Font et al. (1991) using a Mettler TG50 thermobalance, and
Williams and Besler (1994) using a Stanton-Redcroft 280 TGA. A vexing aspect of this problem is the
fact that the values of E and log A decrease in such a way that their ratio remains constant. Chornet and
Roy (1980) called attention to this phenomenon, known as the “compensation effect”, sixteen years ago.
