JOURNAL OF CATALYSIS 86,465-472 (1984)

Selectivity Characterization of Ethylene Oxidation Reactions:

Oxygen Chemisorption

The ethylene oxidation reactions over sil- cle reactor as a continuously stirred tank
ver catalyst have been the subject of many reactor (CSTR). The maximum tempera-
studies and there are several review articles ture gradient under typical reaction condi-
on the subject (1-4). The unique selectivity tions along the axis of the reaction zone
exhibited by the silver catalyst, structure- was measured to be less than 1 K. The line
sensitive nature of the reactions (5, 6), and from the reactor to the gas chromatograph
not so simple catalysis involving oxygen ad- was at around room temperature such that
sorption are perhaps the reasons why this no post-reactor reaction (7) occurred.
reaction system has attracted so much at- An extensive investigation revealed that
tention. Well over 15 different reaction the following procedures yield consistent
mechanisms reported in the literature are and reproducible results: (1) As a standard
an indication how intriguing the reaction pretreatment, the catalyst is oxygenated at
system is. While the kinetics depend on 673 K for 45 min by feeding pure oxygen at
how the catalyst is prepared or what the 20 ml/min into the recycle reactor. (2) Im-
reaction conditions are, there must be some mediately following the pretreatment, the
underlying cause for the diverse kinetic be- reactor temperature was brought down to
havior observed. This paper presents an at- the required reaction temperature, which
tempt to describe the apparently diverse ki- took less than 15 min. (3) A mixture of eth-
netic behavior in terms of some basic ylene, oxygen, and helium (as the diluent)
measurement relevant to the catalysis. In was then introduced into the reactor
particular, an attempt is made to relate the through a multiple channel electronic mass
selectivity to oxygen chemisorption behav- flow controller at the desired rates and
ior for the catalyst prepared, heat-treated, composition. Usually, the system reached
and sintered in a variety of ways. Such a steady state within 45 min. (4) When the
characterization would enable us to isolate steady state was reached, the exit stream
a set of conditions in terms of a basic mea- was analyzed for the composition and the
surement that could lead to better perfor- analysis repeated after 15 min, concluding a
mance of the catalyst. kinetic run. The system including the gas
chromatograph was purged completely be-
EXPERIMENT tween runs. A Tracer 550 gas chromato-
A recycle reactor was used to generate graph (dual column) with a thermal conduc-
kinetic data in the temperature range of 421 tivity detector and Porapak-Q columns (3.2
to 489 K and in the ethylene concentration mm o.d. and 3.65 m long) set at 323 K was
range of 1 to 9 mole% at atmospheric pres- used for the composition analysis. Re-
sure with excess oxygen and no moderator. search grade gases used in the experiments
We found through theoretical calculations were further purified by passing them
and experiments that a pellet size of 3 mm through molecular sieves.
or less and a recycle ratio of 20 for the feed The catalyst used for the determination
rate of 100 ml/min yield intrinsic kinetics of intrinsic kinetic data was prepared using
with less than 1% error in treating the recy- silver oxide dissolved in a solution of lactic
465
0021-9517/84 $3.00
Copyright 8 1984 by Academic Press, Inc.
All rights of reproduction in any form reserved.
466 NOTES

acid and distilled water (8) to form silver we define the overall conversion of ethyl-
lactate. The solution was heated to around ene x as
368 K. Small fused alumina pellets (BET
{(&2Hq)in - (FC2H4)~~t)
area of 1 m2/g) were added to the solution. x= (1)
The impregnation of silver lactate onto the (FC2HJin

pellets was carried out at around 368 K for the yield of ethylene oxide y as
1 hr. The excess solution was decanted and
the dried pellets were decomposed at 643 K
in helium for 16 hr for conversion to metal-
lic silver. This catalyst will be referred to as
and the selectivity of ethylene oxide S as
lactate catalyst. For comparison purposes,
another type of catalyst referred to as ni-
trate catalyst was also prepared using silver
nitrate instead of silver oxide (9).
A dynamic pulse method similar to the where Fi is the molar flow rate of ith spe-
one used by Wanke et al. (10) was used to cies. The selectivity data for various types
determine the amount of oxygen chemi- of the catalysts described in Table 1 are
sorbed as a function of adsorption tempera- summarized in Table 2 along with the corre-
ture. We found that the following proce- sponding conversions.
dures yielded consistent, reproducible Experiments with both the lactate and ni-
results: (1) Clean the silver surface off oxy- trate catalysts showed a marked difference
gen prior to chemisorption by maintaining in activity. The lactate catalyst required
the bed at 573 K and passing the purified much lower reaction temperatures (421-
hydrogen at 20 mVmin for 1 hr. (2) Bring the 489 K) than the nitrate catalyst (463-533 K)
temperature down to the desired level and for the same conversion range. These lower
purge hydrogen by switching to the carrier than usual reaction temperatures observed
gas (helium) at 60 ml/min. This purging is for the lactate catalyst are not unique. For
usually completed in 30 min. This can be instance, Hucknall (13) reported in a re-
checked by observing the thermal conduc- view that a special type of catalyst by Er-
tivity signal on the recorder. (3) Inject oxy- doelchemie GmbH, incorporating a blowing
gen pulses at approximately 5-min inter- agent such as lactic acid hydrazide needed
vals. Approximately 7-10 pulses were reaction temperatures as low as 433 K and
adequate for the completion of chemisorp- the catalyst prepared by Morikana and Kiy-
tion. Research grade gases further purified oshi needed temperatures around 423 K.
through liquid nitrogen trap were used and The reason for such a wide variation of re-
the weight of catalyst used in the experi- action temperatures with various types of
ments was approximately 10 g. The thermal silver catalyst is not exactly known. Huck-
conductivity signal was calibrated in the en- nail claims that the apparent advantage of
tire range of the oxygen concentration of using silver lactate is that it is less soluble
interest based on peak area (II). We did not than silver nitrate and therefore crystallizes
observe any oxygen adsorption on fused faster during the drying period, which helps
alumina (12) that we used as the support. to give a more even silver deposit later, by
reducing crystal growth. Our SEM micro-
EXPERIMENTAL RESULTS graphs did show that the lactate catalyst
For the ethylene oxidation reactions had smaller crystallite size than the nitrate
given by catalyst (1500 vs 2500 A> even though the
average crystallite size of the supported sil-
C Hq--A--+&H40 ver catalysts could not be determined with
-tc! CO2 + HZ0 Ad confidence based on the micrographs.
 NOTES 467

TABLE 1

Chemisorption Data

Catalyst Description Adsorption Net oxygen Metastable

temperature adsorption oxygen adsorption
(K) (/&4 (PM

Al Batch A was prepared with 100 g fused alumina 375 20.9 0.0
pellets. The silver loading on these pellets was 396 29.3 0.0
not so uniform. The pellets with apparently 423 50.5 0.0
heavier metal loading (-20 g) were hand- 445 46.3 5.2
picked and stabilized by oxygenating at 573 K 481 11.2 12.1
for 100 hr. This lot designated as Sample Al
was used for extensive kinetic study and its
average metal loading was estimated to be
-12% by weight
A2 The remaining portion of Batch A was evenly 373 21.4 0.0
divided into four parts. The average metal 423 34.7 0.0
loading of each part was -8.1% by weight. 430 25.4 7.0
Part 1 was designated as Sample A2, which 478 12.2 15.4
represented fresh and unsintered catalyst
A3 Part 2 was sintered at 648 K for 500 hr in a 372 19.3 0.0
stream of 5 mol % ethylene in air. This sin- 424 30.9 0.0
tered catalyst was designated as Sample A3 477 7.7 9.8
A4 Part 3 was sintered at 773 K for 500 hr in a 373 10.9 0.0
stream of 5 mol % ethylene in air. This sin- 424 20.4 0.0
tered catalyst was designated as Sample A4 475 9.8 9.3
A5 Part 4 was sintered at 973 K for 500 hr in a 374 9.2 0.0
stream of 5 mol % ethylene in air. This sin- 424 14.0 0.0
tered catalyst was designated as Sample A5 485 2.8 IO.6
B A small batch was prepared a few months prior 374 14.5 0.0
to Batch A, exactly the same way but with a 424 36.7 5.2
different batch of chemicals. This batch of 477 4.6 15.3
catalyst designated as Sample B was fresh and
unsintered and had an average metal loading
of -8.0% by weight
C Another small batch was prepared using the 374 16.4 0.0
used chemicals of Batch A catalyst. This 421 32.6 8.1
batch, designated as Sample C was also fresh 477 7.4 16.1
and unsintered and had an average metal
loading of 7.0% by weight
D A fresh batch of catalyst was prepared using 373 13.0 0.0
highly concentrated silver lactate solution to 423 33.5 15.1
obtain a maximum silver loading of 8.5% 473 4.0 20.6
by weight. This batch was designated as
Sample D
E Another fresh batch of catalyst was prepared 373 7.3 0.0
using medium concentrated lactate solution to 423 17.0 2.9
obtain a silver loading of 5% by weight. This 481 7.9 15.3
batch was designated as Sample E
F A last batch of catalyst was prepared using 378 10.8 0.0
weakly concentrated lactate solution to obtain 430 21.0 3.1
a silver loading of 2.9% by weight. This batch 480 12.0 10.8
was designated as Sample F

Note. All the catalysts were prepared by lactate method as described by Klugherz and Harriott (8). The kinetic
and adsorption studies were done using the same pellets from any sample. All the experiments were done with
the same experimental setup using the same batch of gases.
468 NOTES

TABLE 2
Selectivity Data

Catalyst Reaction Reaction gas composition Conversion Selectivity

sample temperature (%I (%I
(K) % CzH4 % 02

Al 421 Range: Range: 4.3 60.0

10.00 g catalyst 434 1.09-9.08 mol% 17.43-20.51 mol% 7.0 58.7
100 ml/min feed 448 11.4 57.3
459 20.3 56.2
473 36.5 48.0
489 47.2 42.0
A2 431 4.58 19.83 8.0 58.0
10.07 g catalyst 448 4.26 19.28 14.6 55.0
100 mumin feed 470 3.48 17.83 30.1 50.0
A3 431 4.94 20.27 1.9 58.0
9.88 g catalyst 450 4.76 20.02 5.5 52.0
100 ml/min feed 470 4.45 19.45 11.5 50.0
A4 428 4.92 20.06 3.2 57.0
10.06 g catalyst 450 4.74 19.77 6.7 55.0
100 ml/min feed 468 4.42 19.17 13.0 51.0
A5 433 4.89 20.22 5.0 52.0
10.98 g catalyst 449 4.68 19.87 9.1 54.0
100 ml/min feed 470 4.02 18.69 21.9 55.0
B 430 4.87 19.92 4.4 58.0
6.01 g catalyst 445 4.68 19.59 8.2 54.0
100 ml/min feed 466 4.39 19.03 13.9 50.0
C 429 4.76 20.01 3.0 59.0
5.77 g catalyst 449 4.55 19.65 7.2 55.0
100 ml/min feed 473 4.11 18.85 16.1 50.0
D 431 4.71 19.91 7.6 56.0
9.99 g catalyst 448 4.31 19.28 15.5 57.0
100 ml/min feed 468 3.58 18.25 29.8 60.0
E 432 5.00 20.86 2.5 55.0
9.99 g catalyst 450 4.66 20.27 9.1 52.0
100 mYmin feed 465 4.23 19.53 17.6 52.0
F 432 4.85 20.15 I.0 -
9.51 g catalyst 446 4.74 19.95 3.3 51.0
100 ml/min feed 466 4.40 19.38 10.2 52.0

This marked difference in activity be- catalysts was negligible for temperatures
tween the lactate and nitrate catalysts, in lower than the peak temperature, i.e., the
fact, prompted us to examine the oxygen temperature at which the amount of chemi-
chemisorption behavior of the two catalysts sorbed oxygen is at its maximum, which is
which eventually led us to the correlation approximately 423 K for the lactate catalyst
between selectivity and oxygen chemisorp- and 473 K for the nitrate catalyst. This tem-
tion behavior. As evident from Fig. 1. these perature dependence of the oxygen chemi-
two catalysts show markedly different be- sorption was first reported by Czanderna
havior. We found that the activity of these (14) for the silver powder and in fact his
 NOTES 469

I I I I I 1 I
0.0
373 393 413 433 453 473 493

Adsorption temperature - I;

FIG. 1. Effect of temperature on oxygen chemisorption.

peak temperature was almost the same as jected oxygen is purged from the system by
that shown in Fig. 1 for the nitrate catalyst. the carrier gas.
Our own experiments with silver powder (2
CHARACTERIZATION OF SELECTIVITY
to 3.5 pm) were in agreement with Czan-
dernas observations. The usual way of characterizing a reac-
The chemisorption results obtained for tion is to relate the specific rate (rate/cats-
the lactate catalysts prepared or sintered in lyst surface area) to percentage exposed
a number of different ways are summarized (dispersion). Our attempts to relate the spe-
in Table 1 as a function of temperature at cific rate to the percentage exposed based
which the chemisorption experiment was on the maximum amount of oxygen chemi-
carried out. The eluted (unadsorbed) oxy- sorbed failed. The net rate of ethylene ox-
gen of each injected oxygen pulse appeared ide formation, i.e. (Y] - rJ divided by the
in two distinctly different shapes for tem- maximum amount of chemisorbed oxygen,
peratures below and above approximately which is a measure of specific rate, goes
423 K. For temperatures below 423 K, the through a minimum for sintered catalysts
unadsorbed oxygen appeared as a single (catalysts A2 through A5 in Table 1) when
sharp peak. For temperatures above 423 K, plotted against the maximum amount of
however, this sharp peak is tailed by a rela- chemisorbed oxygen, which is a measure of
tively minor but distinct peak. This minor (oxygen-active) dispersion. In contrast, the
peak area reaches a steady state value, i.e., specific rate increases with increasing dis-
constant value. This steady state value is persion for fresh catalysts (catalysts D
reported in the last column of Table 1 as the through F in Table 1). Similar trends were
metastable oxygen. The oxygen is meta- observed for the overall rate of ethylene
stable in the sense that it remains adsorbed consumption, i.e. (r, + Q). When all results
in the presence of oxygen but desorbs in the were combined for all the catalysts in Table
absence of oxygen as evidenced by the ap- 1, the resulting plot of specific rate vs dis-
pearance of the second peak when the in- persion was so irregular that no pattern
470 NOTES

could be observed. This was also true when different types of catalysts. This observa-
the selectivity was plotted against tempera- tion suggests that the selectivity could be
ture. correlated to the oxygen chemisorption be-
In light of the above, the selectivity and havior. Further, the amount of chemi-
oxygen chemisorption behavior in Tables 1 sorbed oxygen at the reaction temperature
and 2 were compared and several observa- could be an indication of the selectivity. In
tions were made. First, the selectivity is the view of the fact that the maximum selectiv-
highest in all cases at the temperature ity is relatively independent of the maxi-
(peak temperature) at which the amount mum amount of chemisorbed oxygen, the
of chemisorbed oxygen is the highest. This amount of chemisorbed oxygen normalized
peak temperature is at around 423 K for all with respect to (OJmax should be used in-
the catalysts in Table 1. Second, the maxi- stead of the absolute amount for the corre-
mum selectivity is relatively independent of lation. This normalized amount of chemi-
the maximum amount of chemisorbed oxy- sorbed oxygen is in fact the fraction of
gen ((0~)~~~ ) at the peak temperature (in- maximum amount of oxygen still remaining
cluding the metastable oxygen) but appears at the reaction temperature and therefore is
to be somewhat higher for (OJmax > 30 @g- the fractional coverage of oxygen-active
cat. than for (02)max > 24 pug-cat. Third, sites (oxygen coverage). (Note that at
the temperature dependence of the amount the peak temperature, fractional coverage
of chemisorbed oxygen such as the one is.equal to unity.) In calculating the oxygen
shown in Fig. 1 is different for different coverage, the oxygen designated as the
types of catalysts in Table 1 but for a given metastable oxygen (Table 1) was included.
catalyst, this dependence seems to be re- In view of the second observation, the data
lated to the selectivity although the selec- in Tables 1 and 2 were grouped into two,
tivity at a given temperature is different for one for (02hax > 30 pi/g-cat. and the other

a
60.0 - l Al to.0- A
l A2 A A4

0 E

56.0 - 0 F
56.0 -
l

,c 52.0 - 0
1A
.'
+I 0
x
5
* 48.0

44.0
A
I y
A' , 40..0 I I I IIll,l
40.01 I Illlll 0.1 0.2 0.5
0.1 0.2 0.5 1.0 1.0
oxygen coverage oxygen coverage

FIG. 2. Selectivity vs oxygen coverage for various catalysts.

 NOTES 471

TABLE 3 higher than that for the other catalysts at

high temperatures; this ratio for A5 catalyst
Selectivity Behavior of D and A5 Catalysts
is larger than 5 whereas it is in the range of
Catalyst Temperature Selectivity Metastable Oz/ 0.9 to 2.2 for all the others except B catalyst
(K) m Net 0, adsorption
which has a value of 3 at 473 K.
431 56 20.614.0 = 5.2 The correlation shown in Fig. 2 suggests
D 448 57 at 473 K that the oxygen loosely bound to the active
468 60
sites is responsible for the ethylene oxide
433 52 10.6/2.8 = 3.8
A5 449 54 at 485 K formation. This loosely bound oxygen
470 55 would be preferentially desorbed as the
temperature is increased, resulting in a de-
crease of selectivity. The decrease in oxy-
for (02)max < 24 pi/g-cat., and the selectiv- gen coverage with increasing temperature,
ity data were plotted for each group against therefore, can be attributed to the loss of
the oxygen coverage. The results are loosely bound oxygen. The amount of the
shown in Fig. 2a for (OJmax > 30 pllg-cat. loosely bound oxygen depends on how the
and in Fig. 2b for (02)max < 24 @g-cat. The catalyst is prepared or sintered, resulting in
upper line A-A in Fig. 2b coincides with the different selectivity for different catalyst at
lower line A-A in Fig. 2a, consistent with a given temperature. The unusual selectiv-
the second observation. The correlation ity behavior discussed above is consistent
given in Fig. 2 is remarkable in view of the with the explanation. The metastable oxy-
diversity of the catalysts not only in terms gen is that portion of chemisorbed oxygen
of preparation conditions but particularly in that remains adsorbed in the presence of
terms of the level of sintering. All points oxygen but desorbs in the absence of oxy-
except for three are within approximately gen. This indicates that the metastable oxy-
2% selectivity in absolute value as bounded gen is also a form of loosely bound oxygen.
by the solid lines for a given oxygen cover- Thus, the selectivity would be unusually
age. The variation in the experimentally de- high, according to the explanation, if the
termined selectivity at a given temperature majority of the chemisorbed oxygen is
is of the order of 1% in absolute value for loosely bound, which is the case as shown
the Al catalyst. All points in Fig. 2 except in Table 3 for the unusual selectivity behav-
those for the Al catalyst represent the se- ior. Thus, it might be postulated that the
lectivity determined only once and thus do same site which is not responsible for the
not represent average values. ethylene oxide formation at a lower temper-
Some unusual selectivity behavior was ature could become the site responsible for
observed for D and A5 catalysts, which are the ethylene oxide at a higher temperature.
not included in Fig. 2. As shown in Table 2, That is, the fraction of total sites responsi-
the selectivity of these catalysts increased ble for the ethylene oxide formation could
with increasing temperature, which is un- change with reaction temperature as the
usual. The oxygen chemisorption behavior temperature excitation causes a change in
of these catalysts was also unusual in terms of the distribution of the sites responsible for
the relative amounts of stable and metasta- the oxide formation.
ble oxygen chemisorbed as shown in Table
3, when compared with the other catalysts: IN CONCLUSION
the total amount of oxygen chemisorbed We have shown that the selectivity of sil-
(stable + metastable) for these catalysts is ver catalyst can be well characterized based
made up almost by the metastable oxygen. on the oxygen chemisorption behavior. The
In particular, the ratio of metastable to sta- fact that the results are applicable to the
ble oxygen for these catalysts is much silver catalyst prepared, heat-treated, and
472 NOTES

sintered in a variety of ways indicates that REFERENCES

it is quite possible to predict the selectivity 1. Kilty, P. A., and Sachtler, W. M. H., Catal. Rev.
based on the oxygen chemisorption. The 10(l), 1 (1974).
results show that the selectivity depends 2. Spath, H. T., and Handel, K. D., Aduan. Chem.
mainly on the oxygen coverage with slight Ser. 133, 395 (1974).
3. Carra, S., and Forzatti, P, Catal. Rev. 15, 1
dependence of the maximum selectivity on (1977).
the maximum amount of chemisorbed oxy- 4. Dettwiler, H. R., Baiker, A., and Richarz, W.,
gen. The selectivity dependence on oxygen Helv. Chem. Acta 62, 1689 (1979).
coverage and the role of metastable oxygen 5. Wu, J. C., and Harriott, P., J. Catal. 39, 395
can be explained in terms of loosely bound (1975).
6. Verykios, X. E., Stein, F. P., and Coughlin, R.
oxygen, which is postulated to be responsi- W., J. Cutal. 66, 368 (1980).
ble for the ethylene oxide formation. 7. Wachs, I. E., and Chersich, C. C., J. Cutul. 72,
The selectivity correlation presented 160 (1981).
here should not be confused with the tem- 8. Klugherz, P. D., and Harriott, P., AIChE J. 17,
perature effect on selectivity. As indicated 856 (1971).
9. Kenson, R. E., and Lapkin, M., J. Phys. Chem.
earlier, the selectivity could not be well 74, 1493 (1970).
correlated to temperature. Further, the 10. Wanke, S. E., Lotochinski, B. K., and Sidwell, H.
temperature dependence of oxygen cover- C., Cunud. J. Chem. Eng. 59, 357 (1981).
age was different for different catalyst. Ex- II. Sarkany, J., and Gonzalez, R. D., J. Cutul. 76, 75
perimental verification of the correlation at (1982).
temperatures lower than the peak tempera- 12. Geenen, P. V., Boss, H. J., and Pott, G. T., J.
ture was not possible because the catalytic Cutal. 77, 499 (1982).
13. Hucknall, D. J., Selective Oxidation of Hydro-
activity was so low at these temperatures
carbons, pp. 6-22. Academic Press, New York,
that the selectivity could not be determined 1974.
with any confidence. 14. Czandema, A. W., J. Phys. Chem. 68, 2765
The dynamic method for the oxygen che- (1964).
misorption yielded information that cannot
be obtained from a static method, in partic-
ular the information on the metastable oxy- LAKS M. AKELLA
gen. This could be attributed to the fact that HONG H. LEE
the dynamic method represents more
closely the oxygen chemisorption behavior
Department of Chemical Engineering
during reaction than the static method can. University of Florida
Gainesville, Florida 32611
ACKNOWLEDGMENT
This is based upon work supported by National Sci-
ence Foundation under Grant CPE-8 1047 13. Received July 12, 1983; revised September 27, 1983