Anal. Chem.

1984, 56, 2863-2867 2863

Determination of Arrhenius Kinetic Constants by Differential

Scanning Calorimetry
Jan C. M. Torfs,* Leen Deij, Anthonie J. Dorrepaal, and Josef C. Heijens
Dow Chemical (Nederland) E. V., Terneuzen, The Netherlands

A method to obtain Arrhenius kinetic constants of chemlcai of sample in these crucibles was generally done in a nitrogen
reactlons by differential scannlng calorimetry (DSC) has been chamber, since it was found that the presence of oxygen may affect
the degradation pattern. LVC sample pans can withstand
developed. The results of a temperature-programmed DSC
pressures up to 24 bar and temperatures up to 300 "C. Setaram
scan and an isothermal test are combined to yield the desired (France) supplies DSC crucibles which can be used up to 500 "C
kinetlc constants with a relative standard devlation of 4 %, and 100 bars. In a number of experiments cold-welded aluminum
Peroxides and an epoxide have been tested by this new crucibles supplied by Du Pont were employed. Calibration of the
method and were compared wlth a method based on variable temperature and heat flow measurements with the DSC was
heating rates. The latter procedure was found to be more carried out by using high-purity metals indium and gallium. The
time-consuming and generally less accurate than the dis- isothermal tests were usually carried out by aging test materials
in the LVC Capsules which could afterwards be run in the DSC.
closed method. The presently developed very rapid method Only glycidol, which boils at 57 "C, was aged in a steel bomb
yielded kinetic constants whlch were in excellent agreement having a larger volume. The air was removed by applying vacuum
wlth data obtained from various other tests. previous to heating the bomb to the temperature of aging.
PROCEDURES TO DERIVE KINETIC
INFORMATION FROM A DSC EXPERIMENT
It has been stated by various workers (1,2)that differential Heat Evolution Method. In the DSC the heat flowing
scanning calorimetry (DSC) offers a great potential for the from a sample to an inert reference is measured. Figure 1 is
determination of the kinetic constants of chemical reactions. a schematic representation of the DSC trace of an unstable
Well-established methods to study the kinetics of reactions substance. Baxter (14) has shown that the sensitivity of the
are based on the determination of the amount of reactant Du Pont DSC is independent of sample size and thus also
decomposed after various intervals of time. These experiments independent of heat flow. Moreover the high resolving power
must be carried out isothermally at a range of temperatures of this device was illustrated in his work. Therefote, if samples
and are therefore rather time-consuming. The DSC technique (below 20 mg) not too large and heating rates not greater than
has the advantage of being very rapid and versatile. A single, 20 OC/min are employed, the heat flow signal is as a good
temperature programmed DSC experiment can be used to approximation directly proportional to the instantaneously
compute the Arrhenius preexponential factor, activation en- evolved heat. Accordingly, the instantaneous reaction rate
ergy, and order of the reaction within a few hours (3,4). Other can be derived from the measured heat flow. Under these
thermal methods, such as isothermal and adiabatic storage conditions the reaction rate constant, k , at any temperature,
tests (5, 6) and accelerating rate calorimetry (ARC), which TI, may be calculated from eq 1 (10) which is similar to that
employs a nearly adiabatic calorimeter (7), also yield kinetic employed by Borchardt and Daniels (1, 5).
data, but they are more time-consuming than DSC.
A major problem with DSC is, however, that there is much b=-
dH 1
confusion in literature about the accuracy of the obtained
kinetic constants (8). Large discrepancies were found between
the results of various kinetic methods based on DSC (9). Some
authors (5, 6, 10-12) preferred the heat evolution method, In eq 1 dH/dt is the heat flow (more exactly the heat flow
which is based on the shape of a single DSC curve, whereas deviation from the base line), AHbt is the total heat of reaction
others (1,13) found an approach using a series of DSC ex- evolved, which can be derived from the integral area of the
periments a t various heating rates much more satisfactory. reaction exotherm in Figure 1, AHrestis the reaction heat
The reason for these discrepancies seems to be the ap- evolved above a temperature T, (see Figure l),and n is the
proximative character of the various calculation procedures order of the reaction. The reaction rate constant k is com-
and the presence, in some cases, of side reactions and other puted by using eq 1for n = 0, 1,2, and 3. This yields a number
thermal phenomena such as melting, evaporation, etc. oc- of Arrhenius plots of which the one corresponding to the true
curring simultaneously with the investigated reaction. In our reaction order should be linear (7,15). Figure 2 is an example
opinion the solution to these problems can be found in the of such a plot of log k vs. the reciprocal temperature for the
use of exact computer calculations and in testing a substance degradation of tert-butylperpivalate in the DSC. The most
under various experimental conditions, which generally reveals perfectly linear curve is selected. However, since none of the
the presence of interfering phenomena. Arrhenius plots are really straight, the selection of a best linear
The objective of the present study was to compare the curve can be complicated in some cases. The determination
kinetic constants obtained by using DSC with the results of of the preexponential factor, activation energy, and order of
well-established techniques, in order to assess the reliability reaction from a single DSC run can be very difficult because
of kinetic data obtained by DSC. all these constants are interrelated (8): several triplets of
EXPERIMENTAL SECTION constants can be found which describe the DSC curve rea-
DSC experiments were carried out with the Du Pont DSC sonably well. In our opinion, accurate determination of re-
controlled by the Du Pont 1090 thermal analyzer. Data are action order is best done from a series of DSC experiments
transferred to a HP 9835 A desk-top computer via the RS-232C at various concentrations. The dependence of reaction rate,
serial interface of the 1090. In most cases large volume capsules at a certain temperature, on concentration yields reaction
(LVC) supplied by Perkin-Elmer were used. Loading about 5 mg order, as is described in standard textbooks on kinetics (16).
0003-2700/84/0356-2863$01.50/00 1984 American Chemical Society
2864 0 ANALYTICAL CHEMISTRY, VOL. 56, NO. 14, DECEMBER 1984

Indeed, some authors employed more sophisticated base lines

than the straight line used here (4, 17).
Variable Heating Rate Method. The ASTM-E698 pro-
cedure was followed. This DSC method which is based on
Ozawa's work (11) measures exothermic peak maximum
temperature variations with changes in linear programmed
DSC heating rates, between 1 and 20 OC/min. The temper-
ature shift of the peak maximum is used to calculate the
kinetic constants.
Isothermal Check. In order to conduct a quick check on
the obtained constants, the isothermal test described in
ASTM-E-698 was used. This test amounts to aging a sample
in an oven for a predicted half-life time and then measuring
in the DSC whether the remaining heat of reaction, AH,is,
as expected, 50% of the original value, A",. The test result
can also be employed to compute the reaction rate constant,
k , at the temperature of aging. The equation for a first-order
~~~~ ~~

TI Temperature T
reaction is
Flgure 1. Schematic plot of a DSC curve of an unstable material. The
area between the curve and the broken base ling represents the heat k = [In (m~/AfOlt (3)
of reaction, AHtot, and the fraction of material, which has not yet
reacted at temperature T , , is related to AHrest. where t is the duration of aging.
RESULTS OF KINETIC MEASUREMENTS
tert -Butyl Perpivalate (TBPP). Heat Evolution Me-
thod. TBPP is selected because it has been thoroughly
characterized by well-established tests (5, 6). Using the heat
evolution method we calculated an activation energy of 123
kJ/mol and a frequency factor, In A, of 37.6 (min-l). This
agrees extremely well with the results of isothermal and
adiabatic tests: 123 kJ/mol and 37.7 (min-l). Also the ARC
results agree within 3%, as is shown in Table I. The kinetic
constants obtained with the heat evolution method are
practically independent of the heating rate used in the DSC,
as shown in Table 11.
Variable Beating Rate Method. Table I11 presents the
peak temperatures obtained a t various heating rates, for the
degradation of T B P P measured in cold-welded aluminum
crucibles. The repeatability of the peak temperature mea-
surement is good: the standard deviation was found to be
about 0.5 "C. The plot of log (heating rate) as a function of
the reciprocal peak maximum temperature, as required in the
ASTM procedure, was found to be in line with the anticipated
Figure 2. Arrhenius plot of 30 % TBPP in mineral oil, calculated from linearity, in the range from 0.5 to 20 OC/min. An activation
DSC data according to eq 1, for various reaction orders, indicated in energy of 114 kJ/mol and a frequency factor, In A (min-l),
the figure. Broken curve is the best linear fit to the data assuming an
order of 1. of 34.9 were obtained. From the point of view of the DSC
technique, these results are not entirely satisfactory, because
In many cases, however, the order of the reaction is known it was found that part of the sample evaporated during the
from other sources. Most diluted peroxides, as used in this DSC experiment. Weighing crucibles including the sample
work, decompose following a first-order mechanism. before and after the DSC experiment indicated a weight loss
After the order is assessed, the straight part of the selected of about 13%,for a solution of 24.3% T B P P in mineral oil.
curve in Figure 2 is obtained. This is done by eliminating data Therefore, the experiments were repeated using pressure-tight
points a t the high- and the low-temperature side, while the crucibles.
correlation coefficient of a linear least-squares fit is followed. The peak temperatures found for the various crucibles are
This coefficient increases at first because nonlinear parts are presented in Table I11 together with some data regarding the
discarded, until the noise on the heat flow data causes a crucibles. It is clear that the peak temperatures depend on
decrease of the correlation again. The relative importance the type of crucibles employed. This dependency leads to
of this noise increases with the decreasing number of data different values for the activation energy E, which amounted
points. This procedure has the advantage that it is repro- to 114,103, and 80 kJ/mol for the various sample containers.
ducible and independent of the operator's skill. The kinetic The reason for these variations is not completely clear but
parameters of the substance, Le., the preexponential factor, may well be associated with the variation in the masses of
A , and the activation energy, E, which are related to the rate crucibles. Their weights range from 55 to 1050 mg. when the
constant, k , by the Arrhenius equation weight is increased the activation energy decreases and si-
multaheously the calibration factor for the measurement of
k = A exp(-E/RT) (2)
heat flow increases, indicating that heat losses increase from
can be derived from this plot. In eq 2 R is the gas constant. about 0% to 20%. These heat losses might cause the deviation
Figure 2 shows also the best linear fit. In the range from of the results from the expected activation energy value.
90 to 120 "C the agreement is good. This portion of the curve Nevertheless, experiments with the relatively heavy LV cru-
is employed to calculate kinetic constants. The nonlinearity cibles led to correct results when using the heat evolution
outside this range may be due to poor base line selection. principle, and the results did not depend on the heating rate
 ANALYTICAL CHEMISTRY, VOL. 56, NO. 14, DECEMBER 1984 2865

Table I. Kinetic Coostants of the Arrhenius Equation (eq 2) far the Decomposition of tert-Butyl Perpivalate (TBPP),
tert -Butyl Perbenzoate (qBPB), Di-tert -butyl Peroxide (DTBP), and 2,3-Epoxy-l-propanol (Glgcidol) as Obtained by a
Number of Techniques and Methods"
DSC, heating, E, In AI,
material technique conditions rate, %/min kJ/mol min-'
TBPP DSC 30% in mineral oil 10 123 37.6
TBPP DSC 25% in mineral oil; variable heating rates 2, 5, 10, 20 103 31.5
TBPP ARC solution in mineral oil 124.1 38.1
TBPP is0 and adiab. tests (3) pure 123 37.7

TBPB DSC 6.4% in trimethylbenzene 2 152.0 41.9

TBPB DSC 10% in mineral oil 10 171.2 48.1
TBPB ARC 10% in trimethylbenzene 149.2 40.9
TBPB is0 and adiab. tests (12) pure 144.7 39.7

DTBP DSC 30% in mineral oil 10 154.6 38.9

DTBP DSCb 15% in mineral oil 10 153.8 39.0
DTBP DSCb variable heating rates, 15% in mineral oil 2, 5, 10, 20 129.1 32.4
DTBP ARC 20% in mineral oil 166.0 42.9
DTBP ARC 20% in toluene 158.3 41.2
DTBP ARC' 10-100% in toluene (results are averaged) 158.9 41.2
DTBP chem anald various 158.3 40.5

glycidol DSC pure 5 85.8 19.5

glycidol DSC pure 10 77.1 17.6
glycidol DSC pure 20 77.5 17.6
glycidol DSC variable heating rates; pure 2, 5, 10 65.0 14.1
g1y cido1 ARC pure 84.0 18.8

E = activation energy. In A = the natural logarithm of the frequency factor. In the DSC work carried out in our lab stainless steel LV
capsules were used. bReference9; Perkin-Elmer DSC-2 apparatus was used. 'Reference 7. Reference 20; an average of a large number of
measurements was made.
____ ?SC Ildrldble -.earin9 P d r e I
Table 11. Influence of Heating Rate in the RSC on the CSC ' H e d t evoluilonl
Kinetic Constants of tert-Butyl Perpivalate (15% Solution A D I A B A T I C 4ND ! S O T I E R H A L TESTS
in Shell Sol1 J',' Supplied by Shell), Using the Heat = I
ARC
Evolution Method 0 DSC I Iiothermd' check I

heating rate, activation energy preexponential

K/min E, kJ/mol factor In A , min-'
2 128.0 39.27
10 122.6 37.31
20 122.5 37.31

Table 111. Peak Temperatures Found for the Degradation

of 24.3 wt % of TBPP in Mineral Oil, Using Various
Sample Crucibles"
2.4 2.8 3.2
stainless
aluminum steel stainless i(10-3K-l I
1
cold-welded LVC steel
(Du Pont), (Perkin- (Setaram), Figure 3. Arrhenius plot of TBPP showing the results of various
heating rate, OC/min OC Elmer), "C OC techniques (calculated by using eq 2).
0.5 88
1 94 90
out in our laboratory confirmed the heat evolution method
2 101 104 98 in DSC and agreed with ARC and literature values (see Figure
5 110 114 112 3) *
10 117 121 123 Since the method of variable heating rate led to erroneous
20 125 130 128 kinetic constants, one of its basic premises, viz., the assumption
50 138 that the peak maximum represents a point of constant con-
activation energy E , 114 103 80
kJ/mol version independent of heating rate, was investigated. By
wt of sample crucible, mg 55 320 1050 partial integration it was found that the conversion at the peak
DSC calibration factor (at 1.0 1.15 1.27 maximum was ranging from 56% to 61% €or six experiments
10 OC/min) in the range from 5 to 20 K/min. No systematic variation
"Also some physical data of the crucibles are given. Peak tem- was found. These variations were found to account for at most
peratures are corrected for thermal lag and heat rate as described a 1-K variation in peak temperature and cannot completely
in ASTM-E-698-79. explain the too low activation energy found with the variable
heating rate method.
(Table 11). The most clear comparison of all data results from tert-Butyl Perbenzoate (TBPB). The kinetic constants
the inspection of the Arrhenius plots of Figure 3 which are for the decomposition of TBPB, obtained by DSC, ARC, and
calculated (eq 2) by using the kinetic constants derived from a combination of isothermal and adiabatic tests (5,6,18,19),
the various methods. The data agree well over a broad tem- are given in Table I.
perature range except for those obtained by the variable Most results for the activation energy are scattered around
heating rate method (LV crucibles). Isothermal tests carried 150 kJ/mol (A5 kJ/mol). A DSC run obtained at 10 OC/min
2866 ANALYTICAL CHEMISTRY, VOL. 56, NO. 14, DECEMBER 1984

Table IV. Repeatability of Kinetic Constants, and of Reaction Rates Calculated from These Constants, for Di-tert-butyl
Peroxide (30% in Mineral Oil) Obtained by DSC Using the Heat Evolution Method"
kinetic constants rate at various temp
no. E , kJ/mol In A , min-' J/g k200, min-' k100, min k20, lo-" m i d
1 157.69 39.73 1234 0.69 1.4 1.4
2 166.85 42.12 1165 0.73 0.8 0.4
3 147.90 37.23 1309 0.68 2.8 6.3
4 154.21 38.86 1308 0.70 1.9 2.5
5 156.61 39.49 1306 0.72 1.6 1.7
6 156.49 39.46 1221 0.71 1.6 1.8
7 151.96 38.27 1205 0.69 2.1 3.4
8 152.73 38.46 1283 0.68 2.0 3.0
mean 154.6 38.9 1261 0.70 1.78 2.7
std dev 6.0 1.6 56 0.02 0.58 1.9
re1 std dev, % 4 4 4 2.6 33 68
"The obtained activation energy, E, and preexponential factor, A , are used in the Arrhenius equation (eq 2) to calculate the rate constant,
12, at various temperatures.AH is the heat of reaction, based on pure peroxide.

n
s
n
0 1

O - _
...-.
.. ___ 2
_..
. --
...._-. L
-- -

-4-
~
100
~ -~ - -
200
-
300
Temperature ('C I
Figure 5. DSC of a 30 wt % solution of DTBP in mineral oil (heating
rate, 10 OC/min).
Figure 4. Arrhenius plot of TBPB showing the results of various
measurement techniques (calculated by using eq 2). -___ JSC r r d r l a t , e "ea1 r g "$:el
I S C ,Heel B V O . ~ L 3 ~ 1

yields a 10% higher value of 171 kJ/mol. This might be A O l t B A T ! C hN3 ISCTHLRHAL TESTS
ARC
related to the remarks of Duswalt ( l )who
, found that at high
heating rates runaway conditions may be met, leading to
erroneous results. The DSC obtained at 2 OC/min resulted
in a value of 152 kJ/mol.
Figure 4 reveals that the reaction rates obtained by ARC
and DSC (at low heating rate) correspond well with results
from extensive testing by isothermal and adiabatic methods
(5,6,18,19). An isothermal check by DSC also confirms the
results (see Figure 4).
Di-tert-butyl Peroxide (DTBP). DTBP has also been
tested because it is used as a model peroxide, and it has been ~ _ _ _ _ _
the subject of many kinetic studies, reviewed by Shaw and 2.0 2.5 3.0
Pritchard (20). 1(!0-31-11
Figure 5 is a DSC plot of the decomposition of DTBP. The
Arrhenius constants obtained by the various methods are Flgure 6. Arrhenius plot of DTBP showing the results of various
presented in Table I. DSC of a solution in mineral oil, using measurement techniques (calculated by using eq 2).
the heat evolution method, yields an activation energy of 154.6
kJ/mol. A nearly eqilal value of 153.8 has been found else- variable heating rates deviates significantly from the majority
where (9). In line with the results for TBPP, the method of of the data. This is evidently due to the low activation energy
variable heating rates yields a much lower value of 129.1 found with this method (see Table I).
k J / mol. Table IV illustrates the good repeatability of the DSC re-
Activation energies obtained by ARC are in the range sults. Relative standard deviations on activation energy, E ,
155-166 and confirm DSC results obtained by the heat evo- preexponential factor, In A , and reaction enthalpy, AH,are
lution method. Based on a wide range of measurements (20) all 4% for DTBP. For the degradation of T B P P a value of
a value of 158 was found, in good agreement with the DSC 3% was found. The rate constant of DTBP at 200 "C, as
and ARC data. This indicates that the anomaly at 150 "C calculated from the Arrhenius constants, displays a relative
(Figure 5) in the decomposition of DTBP does not lead to a standard deviation of 2.6%. Upon extrapolation to low tem-
significant inaccuracy in the kinetic data. Figure 6 shows the perature, Le., outside the temperature range of the DSC
Arrhenius plots corresponding to some of the results. Only measurement, the uncertainty naturally becomes greater. The
the slope of the curve corresponding to the DSC method of relative standard deviation amounts to 33% and 68% at 100
 ANALYTICAL CHEMISTRY, VOL. 56, NO. 14, DECEMBER 1984 2867

activation energy of glycidol is very close to 80 kJ/mol. Again

the activation energy obtained by the variable heating rate
ARC method, 65 kJ/mol, seems to be too low.
0 DSC ! !50Lhermdl Check 1 The results of this work suggest strongly that the variable
heating rate method should be used with considerable caution
since it led in all investigated cases to values of the activation
energy which were 10-25% too low. This conclusion is sup-
ported by the work of Barendregt and Verhoeff (5,6,18,19)
and by that of Grentzer et al. (21). The reason may well be
the strong dependence of the activation energy value on small
variations in measured peak temperature values. Moreover,
-3
the method is rather time-consuming, since it requires a range
of DSC experiments of which some use a low heating rate.
2.40 3-20 ACKNOWLEDGMENT
i,10 -3K -' H. Loonen and F. Pijper are acknowledged for their ex-
I
perimental assistance. R. Stehl, W. Graf, T. Klingler, J.
Figure 7. Arrhenius plot of glycidol, allowing a comparison of various Ravenstijn, E. Davies, P. Shepherd, A. Oomens, T. van Sint
measurement techniques (calculated by using eq 2). Fiet are acknowledged for critical reading and discussion of
the manuscript.
and 20 "C, respectively. When it is realized that extrapolation
to 20 "C involves a reduction of reaction rate over a factor Registry No. TBPP, 927-07-1; TBPB, 614-45-9; DTBP,
110-05-4; glycidol, 556-52-5.
of 10" the obtained uncertainty seems well acceptable.
Prolonged isothermal testing at lower temperatures can be LITERATURE CITED
used to decrease the uncertainty, which is merely limited by Duswalt, A. A. Thermochim. Acta 1974, 9 , 57.
the time available to the analyst. Ozawa, T. Netsu Sokutei 1977, 4, 45.
Glycidol(2,3-Epoxy-l-propanol). The kinetic constants Kah, A. F.; Koehler, M. E.; Grentzer, T. H.; Niemann, T. F.; Provder, T.
ACS Symp. Ser. 1982, No. 197, 297.
for the decomposition of glycidol are given in Table I. ARC Schlichenmaier, V.; WMmann, G. Thermochim. Acta 1977, 21, 39.
results are in line with DSC recorded a t 5 "C/min. DSC Barendregt, R. B.; Verhoeff, J.; Van den Berg, P. J. "Thermal
Analysis"; Birkhauser Verlag: Basel, Boston, 1980; Voi. 1, p 105.
recorded at 10 and 20 "C/min yields a lower activation energy Barendregt, R. B., Ph.D. Thesis, Prim Maurits Laboratory of TNO,
value. DSC using the method of variable heating rates leads RijswiJk, The Netherlands, 1981,
to an activation energy which is completely out of the range Tou, J. C.; Whitlng, L. F. "Thermal Analysis"; Birkhauser Verlag: Ba-
sei, Boston, 1980; Vol. 1, p 177.
found with ARC and DSC. Figure 7 shows the corresponding Arnold, M.; Veress, G. E.; Paullk, J.; Paulik, F. "Thermal Analysis";
Arrhenius plots. All methods agree well in the temperature Birkhauser Verlag: Basel, Boston, 1980; Vol. 1, pp 89-73.
McDonald, R. A.; Steiner, E. C., The Dow Chemical Co., Midland, MI,
range of 140-260 "C, where the measurements have been unpublished results, 1982.
carried out. Accordingly the reaction rates are most accurate Carroll, 8. Thermochim. Acta 1972, 3 , 449.
in this range, and deviations come about upon extrapolation Freeman, E. S.; Carroll, B. J. Phys. Chem. 1958, 62,396.
Borchardt, H. J.; Daniels, F. J. Am. Chem. Soc. 1957, 79, 41.
to higher or lower temperatures. If accurate kinetic constants Orawa, T. J. Therm. Anal. 1975, 7 , 601.
for this substance are required, some additional tests should Baxter, R. A. "Thermal Analysis"; Academic Press: New York, 1969.
Operating manual of: "Accelerating Rate Calorimeter"; Columbia Sci-
be carried out, preferably in a wide temperature range. A entlflc Industries Corp: Austin, TX, 1979; Vol. 1, pp 65-84.
substance can be made to decompose in the DSC a t low or Glasstone, S.; Lewis, D. "Elements of Physical Chemistry"; MacMillan:
high temperatures by using slow or fast scan speed, respec- London, 1976.
Heuvel, L. M.; Lind, K. C. J. H. Anal. Chem. 1970, 42, 1044.
tively. Verhoeff, J., Ph.D. Thesis, Prins Maurlts Laboratory of TNO, Rijswijk,
As a first attempt to improve accuracy an isothermal test The Netherlands, 1983.
Verhoeff, J., personal communication, 1983.
was carried out at 105 "C, Le., in the extrapolated low tem- Shaw, D. H.; Pritchard, H. A. Can. J. Chem. 1988, 46, 2721.
perature range. Aging for 21 h a t this temperature led to a Grentzer, T. H.;Hoisworth, R. M.; Provder, T.; Kline, S. Org. Coat.
reduction of the heat of reaction to 47%. According to eq 3 Piast. Chem. 1981, 44, 673.
this reduction corresponds to a rate constant of 6.1 X
min-'. RECEIVED for review January 17, 1984. Resubmitted and
This result, indicated in Figure 7, is reasonably in line with accepted July 30, 1984. Thanks are given to The Dow
DSC and ARC data. This allows us to conclude that the Chemical Co. for permission to publish this work.