JOURNAL OF %,468-480 (1985)

The Formation of Filamentous on and Catalysts

II. Mechanism

A.J.H.M. KOCK, P. K. DE BOKX, E. BOELLAARD, W. KLOP, AND J. W. GEUS

Department of Inorganic Chemistry, State University of Utrecht, Croesestraat 77A, 3522 AD Utrecht, The Netherlands

Received May 31, 1984; revised February 18, 1985

The mechanism of filamentous carbon growth on iron and nickel catalysts has been studied using a combination of magnetic techniques and temperature-programmed hydrogenation. CO and CH4 were used as carburizing agents. It is concluded that high contents are a prerequisite for the nucleation of filamentous carbon. The presence of a substoichiometric nickel carbide during steady-state growth is established. In the case of iron the hexagonal a-Fe2C or .a’-Fe& is proposed to be the intermediate in filamentous growth. The driving force for carbon transport can be appreci- ated considering the presence of a gradient in the carbon content of nonstoichiometric car- bides. 0 1985 Academic press. inc.

INTRODUCTION to an exothermic surface reaction was the driving force for whisker growth. In this In Part I (I) we presented evidence for model carbon segregated at the cold rear of the involvement of carbide intermediates in the particle, where the of carbon filamentous carbon growth on iron and is smaller. A severe criticism of this model nickel catalysts. The characterization of is found in the occurrence of filamentous these under growth conditions as growth during endothermic decomposition well as the mechanism leading to filament reactions. Baker et al. (9) have replied that growth is the subject of the present study, saturated are known to de- A number of mechanisms explaining fila- compose via their dehydrogenated analogs. mentous carbon growth on transition Decomposition of the latter compounds is metals has been suggested (24). Carbon exothermic and, consequently, a mecha- can be transported by a surface diffusion nism involving thermal diffusion would not process or by bulk diffusion through the be contravened. This explanation, how- particle. Massaro and Petersen (5) showed ever, calls for separate sites where only the that surface diffusion of carbon on nickel endothermic dehydrogenation takes place. foils is negligibly small in the temperature Rostrup-Nielsen and Trimm (3), refer- range 620-970 K. Also the sparse data for ring to Wada et al. (IO), assumed that car- surface migration of species bon species of different origin have a differ- indicate that carbon transport does not pro- ent solubility. Thus, carbon species at the ceed through surface diffusion of hydrocar- metal-gas interface originating from hydro- bon fragments (6). At present a growth carbon decomposition reactions have a model involving bulk diffusion through the higher solubility than carbon originating metal particle is generally accepted (7). The from . The resulting concentration driving force responsible for this bulk diffu- gradient is thought to be responsible for sion, however, has not been unequivocally carbon transport. However, the physical established. basis for these different has not Baker et al. (2, 8) stated that a tempera- been clarified. ture gradient across the metal particle due Buyanov et al. (4, 11, 12) proposed a 468 0021-9517185 $3.00 Copyright 8 1985 by Academic Press, Inc. All rights of reproduction in any form reserved. FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 469

when the transport of carbon proceeds through interstitially dissolved carbon. Magnetic measurements enabled us to as- sess the presence of (intermediate) metal carbides during the different stages of car- bon deposition. The carbide content in the early stages of filamentous growth was monitored using in situ magnetization mea- surements. Carbon contents present during steady-state filamentous growth were quan- titatively determined using temperature- programmed hydrogenation (TPH). Combi- / nation with measurements of the saturation magnetization (MS) yielded information about the chemical nature of the carbon FIG. 1. Schematic diagram of apparatus for magnetic species. The identification of ferro/ferri- measurements: (1) sample holder, (2) cooler/heater, (3) magnetic phases was performed employing transportable electromagnet, (4) Helmholtz sensing thermomagnetic analysis (TMA). Implica- coils, (5) integrating device. tions for the mechanism of filamentous growth will be given. carbide-cycle mechanism: this invoked EXPERIMENTAL consecutive formation and decomposition of carbides, where carbon separated in the Apparatus latter step leads to the formation of fila- Magnetic measurements were performed ments. Thus, no longer interstitial carbon using a modified Weiss-extraction method with an extremely low (extrapolated) solu- (Fig. 1). Ferromagnetic samples could be bility was considered, but the carbon was magnetized in fields up to 0.52 MA m-l. supposed to have a carbidic nature. The magnetic field (H) was produced by a Previously (Z), we observed that during Newport Type A electromagnet with 60- carbon whisker growth on iron and nickel mm pole faces separated by a 46-mm gap. catalysts equilibria with carbides were at- Between the magnet poles two interrupted tained. Independently, we established the Helmholtz sensing coils were mounted co- presence of carbon at concentrations axially. These pick-up coils were 20 mm in greatly exceeding the solubility of intersti- diameter and each contained 18,000 turns tial carbon in iron as well as in nickel, in of insulated wire (diameter 0.1 mm). agreement with prior reports (13). Re- They were connected in series, but in an cently, for iron catalysts cementite (f3-Fe3C) opposite sense to minimize disturbances was proposed to be the catalyst for carbon caused by field fluctuations. Extraction of deposition (Sacco et al. (24)). Baker et al. the magnetized sample from the pick-up (9) observed no carbon whisker growth coil generates an induction voltage. The starting from &Fe$ at all, however. The time integral of the induction voltage corre- presence of Eckstrom-Adcock carbide sponds to the net change of magnetic flux at (Fe&) at the tip of carbon filaments has the pick-up coil. Integration was done after been demonstrated (15). amplification (1000X). Instead of displacing In this study we present experiments the specimen, in our modification of the which allow a discrimination to be made Weiss-extraction method the electromagnet between the mechanisms mentioned above. could be transported along two rails over It is very unlikely that a carbon content ap- any desired distance. The magnetization of proximating that of a carbide is attained, samples could be measured at temperatures 470 KOCK ET AL.

ranging from 100 up to 740 K using an indi- where (I). For the nickel catalyst reported rect air cooler/heater. The quartz sample on in this paper the dispersion was esti- holder was connected to a conventional Py- mated to be 0.1 [the Langevin low-field esti- rex high-vacuum system, equipped with a mate (17) for the mean particle size (assum- quadrupole mass spectrometer (Leybold- ing spherical particles) was 5.4 nm]. The Heraeus Q 200). The sample holder was dispersion of the iron catalyst was esti- also suited for atmospheric flow experi- mated to be 0.02 from X-ray line broaden- ments allowing for simultaneous mass- ing, using the Scherrer equation. spectrometric gas analysis. Temperature- programmed experiments could be carried RESULTS out at heating rates ranging from 1.7 to 300 mK s-l. Nickel Catalysts In some experiments also a vibrating- A. Transient carburization behavior. To sample magnetometer was used. The prin- investigate the transient behavior of nickel ciple of this technique has been clearly out- catalyst particles concurrent with filamen- lined in Ref. (16). This apparatus enables tous carbon growth, the magnetization of a us to perform magnetization experiments freshly reduced 50 wt% NiBi catalyst at liquid-nitrogen temperature at a field (reduction temperature 870 K) was moni- strength up to 1.12 MA m-i. After the cata- tored as a function of time. As electron mi- lyst pretreatment (reduction or carburiza- croscopy (I) revealed that nickel particles tion) 50 mg of the catalyst was brought into with a diameter larger than 10 nm were in- the sample holder under a nitrogen atmo- volved exclusively in filament growth, the sphere to prevent oxidation of the sample. measurement of the magnetization should be performed under conditions that give Procedures and Catalysts high weighting to the larger particles. The Thermomagnetic analysis was performed moderate H/T values applied guarantees at the maximum field strength (0.52 MA that nickel atoms present in larger particles m-l). These analyses were carried out in a predominantly contribute to the observed helium atmosphere or in a flow of 10 ~01% magnetization. Figure 2 shows the time de- of diluted with argon at a total pendence of the magnetization during car- flow rate of 0.14 cm3 s-i. The magnetically burization with 10 ~01% of in ni- monitored transient carburization experi- trogen at a flow rate of 0.8 cm3 s-i. The ments were also measured at the maximum magnetization was measured during 15 ks, field strength. To prevent displacement of while also after a period of 65 ks a final catalytic material during the magnetic mea- measurement was made. Methane was cho- surements the reactor bed was covered sen as the carburizing agent to prevent in- with a layer of quartz fragments. The satu- terference by Ostwald ripening of nickel ration magnetization at room temperature particles due to transport via Ni(COk. The was determined by linear extrapolation of magnetization was followed during carburi- the it4 versus l/H plot. zation at three different temperatures, viz. Temperature-programmed hydrogena- 576, 596, and 611 K. Initially a decrease of tion was performed at a heating rate of 0.1 the magnetization was observed. Having K s-l using a flow of 0.8 cm3 s-l of 10 ~01% passed through a minimum, the magnetiza- hydrogen in argon. Mass-spectrometric tion gradually increased up to typically 70% analysis was employed, scanning eight se- of the original value, as shown in Fig. 2. lected m/e values within 24 s. Attention should be paid to the relative de- Preparation and pretreatment of the cata- crease of the magnetization down to about lysts as well as the carburization proce- 25% in the case of carburizing with meth- dures have been extensively described else- ane, whereas carburizing with CO at 540 K FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 471

/IT-- 5 10 65 TIME [ksl

FIG. 2. Time dependence of the magnetization of a 50 wt% Ni/SiOz catalyst during carburization with 10 ~01% CHJNr. The carburization temperatures are indicated. Note the value of the magnetiza- tion realized after 65 ks. results in a complete loss of ferromagnetic stoichiometric carbide is required for the behavior (28). The decrease of the magneti- nucleation of graphite, however, cannot be zation cannot be associated with the forma- answered from these experiments. The ob- tion of a surface carbide only, the metal servation of the gradual increase of the dispersion being 0.1. Also the 0.02 at.% in- magnetization indicates that after the nucle- terstitially dissolved carbon predicted from ation of filamentous carbon the carbon con- solubility data (29) is not expected to de- tent is significantly lower than that realized stroy as much as 75% of the ferro- in the very early stages of carburization. magnetism. Hence, we may conclude that a Replacing the carburizing agent by a 10 bulk carbide is formed preceding the nucle- ~01% Hz/Ar mixture at 576 K results in an ation of graphite. The question whether a almost instantaneous recovery of the origi-

7 s .I

z l PARTIALLY CARBIDED k! 2 ii AFTER DECOMPOSITION FE 2 r :

8 300 LOO 500 600 700 800 T[KI

FIG. 3. Temperature-programmed hydrogenation profiles of a 50 wt% Ni/SiOz catalyst, after carburi- zation with 100% CO at 540 K during 4 ks (circles); after carburization during 20 ks (squares); after carburization during 20 ks followed by decomposition at 620 K in a He atmosphere during 8 ks (triangles). The percentages refer to the relative decrease of the magnetization observed after the carburization treatment. 472 KOCK ET AL.

nal magnetization. The rate of recovery is fed to the reactor at temperatures ranging limited by the supply of hydrogen. from 723 to 773 K, until the effluent gas To investigate the generality of the in- composition was time-invariant, indicative volvement of carbides in filament growth, of steady-state carburization. The sample irrespective of the nature of the carburizing was thereupon quenched to room tempera- agent, the transient behavior of nickel was ture, simultaneously replacing the carburiz- also studied during carburization with car- ing agent by nitrogen, and MS was again bon monoxide. Here temperature-pro- measured. In separate desorption experi- grammed hydrogenation was selected to ments it was verified that the observed de- provide information on the content of car- crease in MS was not due to the amount of bide in the sample, as the TPH profiles are hydrogen and methane adsorbed during the not significantly affected by changes in the quenching procedure. The amount of carbi- particle size distribution (as was the case die carbon was accurately determined using with the above magnetization measure- TPH with gas-chromatographic determina- ments). In Fig. 3 TPH profiles of Ni/SiOz tion of methane as described in our pre- catalysts exposed to 100% carbon monox- vious paper (I). Hydrogenation was per- ide (flow rate 0.8 cm3 s-l, temperature 540 formed at a heating rate of 11 mK s-i up to K) are collected. It can be seen that a short a temperature of 573 K, ensuring a com- exposure (4 ks) leads to a profile character- plete reduction of only the carbide phase. istic of carbidic material (I), whereas a After TPH the sample was flushed with ni- longer exposure (20 ks) leads to TPH peaks trogen at 850 K and again M, was deter- of both carbidic carbon and a small amount mined. After the first hydrogenation of a of filamentous carbon. The transient char- carburized sample typically 95% of the acter of the carbide phase is clearly illus- original MS was measured. All subsequent trated by decomposing a 100% carburized carburization-hydrogenation cycles did not sample (that is, a sample which has lost su- result in a further loss of saturation magne- perparamagnetic behavior due to decou- tization. Attention was focused on the re- pling of all nickel atoms) during 8 ks at 620 versible changes in M,. The results for five K in a flow of inert gas. The decomposition different carbon contents are collected in causes the hydrogenation behavior charac- Fig. 4. Here the bond number, i.e., the teristic of carbidic material to disappear in number of nickel atoms (ferro)magnetically favor of that of filamentous carbon accom- eliminated by one carbon atom, is plotted panied by the recovery of the original mag- netization. These TPH experiments again indicate that a carbide phase is formed pre- t ceding the nucleation of filamentous car- lo- bon. 5

B. Steady-state carburization behavior. E l .

In order to identify the carbon present in s . . . nickel during steady-state filament growth 4 s- and to determine the amount of this carbon, mo the following experiments were performed. Nickel catalysts were reduced at 870 K for at least 20 ks to assure a degree of reduction of 100%. The saturation magnetization of ATOM % C the freshly reduced catalyst was measured FIG. 4. Bond numbers of carbon present during after flushing with nitrogen at 850 K. Subse- steady-state filament growth as a function of the car- quently, 10 ~01% methane in nitrogen was bon content. FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 473 against the carbon content of the sample, It we conclude that the carbide present during is seen that bond numbers decrease with steady-state filament growth is almost increasing carbon contents. The carbon equally distributed over large and small contents realized under the specified condi- nickel particles. tions yield bond numbers of 6 to 7. The car- bon contents greatly exceed the reported Iron Catalysts solubility of interstitial carbon in nickel Also in the case of iron the original aim of (19). this study was to identify the intermediate The above bond numbers were deter- carbide as well as to investigate the tran- mined from changes in the saturation mag- sient behavior of iron catalyst particles con- netization. The estimation of the values of current with filamentous carbon growth. iVl, was obtained by extrapolation of M ver- The analysis of carburized iron samples is sus HIT curves, using HIT values between complicated due to the existence of at least 1.5 and 1.7 kA m-l K-‘. Although Richard- six ferromagnetic carbides under the condi- son and Desai (20) demonstrated that the tions used in our experiments: &Fe$Z, E- extrapolation procedure from measure- Fe& &‘-Fe&, x-Fe-$*, Fe,C, and Fe& ments at comparable magnetic fields (0.64 in the nomenclature of Niemantsverdriet et MA m-i) yielded similar results as obtained al. (21). It should be realized, however, from the extrapolation from ultrahigh field that in the case of iron a particular carbide measurements (up to 8.0 MA m-l), it may may not only be an intermediate in filament be questioned whether the reported extrap- growth, but also an intermediate in reac- olation procedure is correct with the cata- tions leading to more stable iron carbides. lyst used in this investigation. Therefore, Recording the magnetization as a function we collected M versus H/T curves employ- of time is thus not equivalent to monitoring ing the vibrating-sample magnetometer to the amount of the carbide intermediate in estimate M, from the region 1.5-1.7 and filament growth, as was the case with 13-14.2 kA m-i K-l. For the freshly re- nickel. The interpretation of changes in duced sample the low-region estimate was magnetization during the early stages of fil- found to be 12.5% lower than the high-re- ament growth requires TMA of the sample gion estimate. For carburized samples the at regular time intervals. Apart from the underestimation amounted to 14.6%. As continuous formation of &Fe$ the gradual the vibrating-sample magnetometer did not changes in the amounts of the other ferro- allow for in situ measurements, we per- magnetic compounds are not believed to formed the bond number determinations at have sufficient significance. In this paper the lower H/T values in the apparatus de- we therefore confine ourselves mainly to scribed extensively above. The conse- the identification of the intermediate car- quences of the underestimation of MS will bide in the early stages of filament growth. be discussed below. An iron/alumina catalyst was carburized To investigate whether the carbide in a flow of 100% (0.8 cm3 present during steady-state filament growth s-i) at 620 K. Thermomagnetic analysis of is preferentially localized in nickel particles the phases formed was performed after of a particular size range, the relative mag- quenching in an inert gas flow. A 10 ~01% netization (M/MS) versus HIT was mea- hydrogen/argon mixture was passed over sured for the freshly reduced catalyst as the catalyst during the thermomagnetic well as for the carburized catalyst. As the analysis. Figure 5 shows two thermomag- initial slope of the relative magnetization netic curves, obtained after carburizing for curve (Langevin low-field method (17)) af- 1 and 6 ks, respectively. Both curves were ter carburizing decreased by typically IO%, measured at rising and declining tempera- 474 KOCK ET AL.

M [“.U] TMA

\. .

I I I I 373 473 573 673 T I”

FIG. 5. Thermomagnetic analysis of a 50 wt% Fe/A1203 catalyst in a 10 ~01% H2/Ar atmosphere after carburization with 100% CO at 620 K, during 1 ks (circles) and during 6 ks (triangles).

ture. The analyses demonstrate the pres- reported, so the present analysis cannot ence of cementite (Curie temperature, T, = yield information on its possible involve- 485 ? 8 K) in considerable quantities espe- ment. cially after prolonged carburization, and Finally, the intermediate carbide present exclude the presence of appreciable during steady-state filament growth was in- amounts of x-Fe& and Fe& (T,, respec- vestigated. The above experiments indi- tively, 530 + 10 and 523 2 3 K (22)). cated that after prolonged carburization the After carburizing for 1 ks the TMA curve concentration of the intermediate carbide is displayed an increase of the magnetization rather low; we thus had to use our most at temperatures above 570 K due to the hy- sensitive detection technique (TPH) to drogenation of a compound of a T, higher identify the intermediate. An iron catalyst than 550 K. The presence of Fe304 (T, = was subjected to TPH after quenching in an 853 + 15 K) was excluded by performing an inert gas from a carburization experiment at identical carburization experiment followed 860 K. Methane was used as the carburiz- by TMA measured in a helium atmosphere. ing agent. X-Ray analysis as well as ther- Also in that case an increase of the magneti- momagnetic analysis demonstrated that the zation was observed, but at a higher tem- main carbide phase present was cementite. perature (about 700 K). So the observed in- Figure 6 displays the TPH characteristics. crease of the magnetization can only be From the transient experiments described attributed to the decomposition of a carbide above it was deduced that the only carbides with a T, higher than 550 K. It should be present were 8-Fe3C and E-Fe2C or E’- pointed out that in the catalyst carburized Fe& carbides. As hydrogenation of 8- for 6 ks only 8-Fe3C and cu-Fe were de- Fe$ does not proceed at temperatures be- tected. Hydrogenation for 60 ks at 740 K low 740 K, the first peak (610 K), did not result in any recovery of the magne- corresponding to 3.6 at.% of carbon in Fe, tization. Under these conditions 8-Fe3C is was tentatively attributed to the hydrogena- stable against decomposition and even tion of e-Fe& or &‘-Fe&. However, no against hydrogenation. Hence, the only direct evidence from magnetic measure- carbides responsible for the increase of the ments or from X-ray diffraction could be magnetization can be the hexagonal car- derived for this proposition. Whereas X-ray bides &-Fe$ or E’-Fez.zC (T,, respectively, patterns measured after quenching from a 653 and 723 K). For Fe,C no T, has been temperature of 650 K revealed the presence FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 475

filamentous carbon 16)

t/E’-carbide

TIKI -

FIG. 6. Temperature-programmed hydrogenation profile of a 50 wt% FelA1203 catalyst after steady- state carburization with 10 ~01% CH4/N2 at 860 K.

of &Fe&, a-Fe, and graphite, X-ray analy- dicates a fraction of the nickel to be present sis performed after hydrogenation up to 900 in smaller particles. We cannot rule out that K only indicated the presence of a-Fe. This initially also carburization of these small implies that the most pronounced hydroge- particles takes place, followed by decom- nation peak (800 K) corresponds to the hy- position not leading to filament growth. Al- drogenation of both filamentous carbon and though the fraction of nickel present in 8-Fe3C. small particles contributing to the initial de- crease in magnetization is not accurately DISCUSSION known, it can be excluded that the effect is entirely due to nickel in small particles, Nickel Catalysts considering the magnitude of the Langevin A great number of publications report on low-field estimate for the mean particle the formation of nickel carbide under a va- size. riety of experimental conditions (24, 25). Let us now focus on the steady-state car- Also the decomposition of nickel carbide is burization behavior. The observation of a rather well documented (26-28). In this minimum diameter for the filaments could study we provided evidence that a high car- be explained assuming that small particles bide content is a prerequisite for nucleation become encapsulated due to decomposition of graphite. A few remarks have to be of carbides under carburizing conditions. made. First, the transient decrease in mag- The reported low activity of small nickel netization reported is not limited to the particles in the hydrogenation of carbon fraction of nickel involved in filament monoxide (29) can also be explained as- growth. As mentioned above, electron mi- suming encapsulation of small particles. croscopic analysis revealed that only car- Moreover, Kehrer and Leidheiser (30), bon filaments were grown with a diameter studying the carburization of macroscopic larger than 10 nm, while the Langevin low- nickel crystals at 823 K, observed a vehe- field estimate for the mean particle size in- ment carbon deposition at the minor faces 476 KOCK ET AL.

(high index planes). Small particles having change in the particle size distribution. Due a relatively large fraction of these planes in to decoupling by interaction with carbon their surface are thus liable to rapid encap- atoms the number of ferromagnetically sulation. coupled nickel atoms within a particle One might imagine that the decomposi- will decrease. Therefore, the degree of tion of the formed bulk carbide leads to a underestimation of MS may be different for surface carbide and free carbon. Martin and freshly reduced and carburized samples. Imelik (31) reporting on saturation magneti- Let us denote the fractional underestima- zation studies established a bond number of tion of MS for the freshly reduced and car- 3 for a surface carbide. Thus, we can elimi- burized catalyst by a and b, respectively, nate the exclusive formation of a surface while the fraction of nickel atoms being de- carbide. coupled by the incorporation of carbon at- The high bond numbers we found were oms is called c; the magnetization change also measured by Martin and Imelik (31). determined according to the extrapolation They adsorbed a great variety of hydrocar- procedure using too low values of H/T will bons a 196 K, and performed saturation be (1 - a) *MS - (1 - b) * M, . (1 - c) = MS magnetization measurements varying the * (c + (b - a) - bc) instead of M, - MS - (1 “holding temperature,” i.e., the tempera- - c) = c - Ms. ture at which the catalyst was kept prior to From the above formulas, it is evident the magnetic measurements. In these ex- that bond numbers (the fraction of decou- periments the bond numbers gradually in- pled nickel atoms times the number of creased from holding temperatures of about nickel atoms present in the sample and di- 474 K onward. The effect was attributed to vided by the number of incorporated car- dissolution of carbon in the bulk of the bon atoms) may be both over- and underes- metal. A possible rationalization of these timated, depending on the magnitude of a, high bond numbers was given by Martin et b, and c. From the fractional estimates of a al. (32), but it should be emphasized that an and b (0.125 and 0.146) and the apparent adequate theory quantitatively describing relative decrease of MS (larger than 0.1, the influence of dissolved foreign atoms on based on the extrapolation using too low Hl collective magnetism of 3d metals has not T values), it may be deduced that the rela- been developed as yet. The high bond num- tive overestimation of the bond numbers bers rule out that nickel is present partly as does not exceed 7.5%. Thus, the conclu- metallic nickel and partly as a stoichio- sion based on the high value of the bond metric carbide. We conclude that we are numbers, i.e., the substoichiometric char- dealing with a substoichiometric carbide acter of the intermediate carbide, is not af- under steady-state filament growth condi- fected by the underestimation of M,. tions. Finally, the low-field magnetization In the above paragraph the reported bond study of Kuijpers et al. (33) should be com- numbers were based on the extrapolation mented on in this context. It was concluded procedure to obtain M, using H/T values that carburizing of nickel catalysts with between 1.5 and 1.7 kA m-i K-i. Previ- methane at temperatures up to 573 K ously, we reported that the degree of under- results in the formation of a surface carbide estimation of MS for the freshly reduced and only. It should be emphasized that for- the carburized sample was found to be 12.5 mation of a bulk carbide at the mentioned and 14.6%, respectively. Let us now con- experimental conditions in the static sys- sider the effect of the underestimation of M, tem used by these authors is prohibited by on the bond number determinations. Gener- thermodynamics. Our study clearly indi- ally, incorporation of carbon atoms in cates that considerable bulk carburization nickel particles will induce an apparent does take place at 576 K. FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 477

Iron Catalysts paramagnetic FeO. However, it has been established (38) that already above 473 K A great deal of information on the struc- this metastable oxide disproportionates to ture of iron carbides was collected by Naga- cu-Fe and the ferrimagnetic Fe30.+ At a car- kura and Oketani (22) and Le CaEr et al. burizing temperature of 620 K the presence (34). A few relevant data are summarized of Fe0 is therefore very unlikely. As under here. Carbon atoms are localized at octahe- our experimental conditions also no forma- dral positions or in trigonal prisms. The size tion of FeS04 was observed, we conclude ratio of iron and carbon atoms causes car- that all magnetization changes observed bon to prefer a trigonal prismatic environ- can be attributed to the formation and de- ment of iron atoms. Andersson and H-yde composition of carbides. (35) have shown that twinning on (1102) We now arrive at the identification of the planes in hcp iron generates trigonal prisms carbide which is intermediate in the steady- in the twin plane, supplying thus an elegant state growth of filamentous carbon. In the description of the formation of the trigonal above it has been proposed from TPH ex- prismatic carbides &Fe& x-Fe&l, and periments that E- and &‘-carbides, albeit at Fe&. In fact, this twinning operation cor- very low concentrations, are present in responds to changing asymmetrically the samples quenched from steady-state fila- dimensions of the interstices. The size of ment growth. Higher concentrations, de- the unoccupied interstices decreases at the tectable with TMA, are present during the expense of those containing carbon, leading early stages of growth. This transient be- to an overall increase in density (36). These havior is very similar to the behavior of the trigonal prismatic carbides are character- hexagonal N&C. Also cementite was ized by low Curie temperatures and by a formed, but this carbide proved to be stable remarkable stability as compared to the against decomposition and even against hy- hexagonal carbides (&-Fe& and &‘-Fe&), drogenation at temperatures where exten- where carbon atoms occupy octahedral po- sive filament growth was observed. From sitions. The carbon arrangement and even these observations we conclude that hexag- the carbon content of these hexagonal car- onal E-carbides are the intermediates in fila- bides is still a matter of controversy. mentous growth, cementite being merely an Goodwin and Parravano (37) reported on inactive side product. This corroborates the a conversion electron Miissbauer spectro- conclusions of Baker et al. (9) that cement- scopic study on the decarburization of iron ite does not give rise to filament formation. films by hydrogen. They attributed the ob- Sacco et al. (14) appear to have been moni- served fast decarburization after carburiza- toring the formation of a side product. tion at 569 K to the presence of the less stable E-carbide, whereas the 8- or x-car- bides precipitating above 573 K exhibited a Initiation and Mechanism drastically decreased rate of decarburiza- The two major conclusions which we tion owing to their higher stability. Summa- reached above, viz. the formation of a car- rizing, we invoke the structure of the trigo- bide preceding nucleation of carbon during nal prismatic carbides to account for their the early stages of filamentous growth and thermodynamic stability, while the kinetic the presence of a substoichiometric carbide stability of these carbides has also been es- under steady-state growth conditions, have tablished experimentally. some important implications for the under- Turning to our TMA experiments, it standing of the mechanism of filamentous should be noted that an increase in magneti- carbon formation. zation at temperatures above 573 K might Let us first discuss the initiation process. also be ascribed to decomposition of the It has been reported that nucleation of car- 478 KOCK ET AL.

bon associated with filamentous growth re- can be appreciated by realizing that the quires an induction period (39). From our number of metal particles involved in frag- in situ magnetization measurements it ap- mentation increases with rising tempera- pears that high carbide contents are re- ture, due to the increase of the nucleation quired before nucleation of carbon pro- rate. ceeds. Induction periods can thus be Once a carbon phase has nucleated we interpreted as the time needed for both the enter the regime of steady-state growth. As formation of a carbide phase of high carbon has been mentioned in the Introduction, the content and the incubation time (40) associ- major point of discussion here is the nature ated with the decomposition of this phase. of the driving force for transport of carbon The details of the initial steps of carbide through the metal particle. The observed decomposition are not well known. Tem- presence of carbides during steady-state perature and pressure (41) have been found growth seems to facilitate the comprehen- to be important parameters. For nickel it sion of this growth mechanism. has been suggested that graphite nucleates The overall driving force responsible for in the bulk of the carbide phase leading to the transport of the carbon atoms is now fractionation of the metal crystallites (42). clear. The Gibbs free energy of the metal Also in the case of iron, Le Caer et al. (34) carbide is higher than that of its constituting supposed that the relative increase of the elements. The driving force leading to contribution of superparamagnetic iron car- transport of carbon atoms within the metal bide to the Mossbauer spectrum could be (carbide) particles is, however, more diffi- explained by fragmentation of particles. cult to assess. Usually, the transport is at- For supported catalysts the metal particles tributed to an activity (or concentration) seem to fragment most probably where me- gradient of carbon in the metal phase. chanical stresses are easiest to dissipate, Baker et al. (2, 8) proposed that tempera- that is, at the metal-gas interface or near ture differences are the origin of differences the metal-support interface. In the latter in activity of carbon leading to transport of case nucleation of carbon takes place carbon atoms. Rostrup-Nielsen and Trimm through precipitation of graphite layers (3) assumed differences in activity to be near the metal-support interface, thereby caused by a different solubility of carbon lifting the particle from the support. species of a different origin. However, the Such a fragmentation of nickel particles above explanations take no account of the could explain the irreversible loss of satura- fact that when the carbon is taken up into a tion magnetization, typically 5%, observed substoichiometric carbide, significant after the first carburization. The extent of changes in the original metal structure are the permanent loss of saturation magnetiza- induced. A description of the migration of tion rises with increasing temperature. carbon through an unaltered fee (Ni) or bee Even prolonged hydrogenation at 820 K did (Fe) matrix in terms of activity gradients not result in a 100% recovery of the satura- only, is therefore not appropriate. The tion magnetization. This effect could be due transport of carbon atoms through the me- to a change of the nickel particle size distri- tal (carbide) particle may be explained by bution to a higher mean dispersion. In that assuming the carbon content of the substoi- case, the extrapolation procedure required chiometric carbide to be highest at the to calculate the saturation magnetization metal-gas interface, and lowest at the can be biased, since the smallest particles metal-carbon interface, leading to trans- (with a diameter less than 1.5 nm) are only port in the direction of the metal-carbon slightly magnetized at the applied magnetic interface. The substoichiometric character field strengths. As a result this highly dis- of the intermediate carbide facilitates the persed nickel is beyond experimental ob- realization of such a gradient in carbon con- servation. The temperature dependence tent. In the opinion of the present authors a FILAMENTOUS CARBON ON Fe AND Ni CATALYSTS, II 479

description of carbon transport in terms of tion experiments during steady-state fila- a substoichiometric carbide is more appro- ment growth as well as during the early priate, as it explains (i) the high carbon stages of growth we conclude the following: content encountered during steady-state (i) Nucleation of graphite on nickel cata- growth (200 times the reported solubility of lysts is preceded by the formation of a car- carbon), (ii) the observed high bond num- bide phase with a high carbon content. In bers, and (iii) high transport rates of car- the case of iron during the early stages of bon. Generally, transport rates in nonstoi- growth a considerable amount of carbide in- chiometric compounds are significantly termediate can be detected. higher than in stoichiometric compounds (ii) The nickel carbide present during (43). The driving force for carbon transport steady-state filament growth is substoi- is a gradient in the carbon content of a non- chiometric. stoichiometric carbide. Reaction to a (sub- (iii) The iron carbide intermediate is pro- stoichiometric) carbide involves both a posed to be .s-Fe2C or .$-Fe&, whereas modification of the metal structure and the the involvement of 0-Fe3C can be defini- establishment of a certain carbon concen- tively ruled out. tration. To attribute the driving force, as (iv) Carbon filaments grow by a continu- usual, to a mere difference in activity of ous decomposition of a metastable carbide dissolved carbon is therefore not suitable; intermediate. The driving force for filament the standard chemical potential is also af- growth is a gradient in the carbon content fected. of nonstoichiometric carbides, decreasing The concept proposed by Wada et al. in the direction to the metal-carbon inter- (IO), i.e., a different origin of carbon lead- face. ing to a different concentration of this car- bon when accommodated in a metal, can ACKNOWLEDGMENTS now be reformulated. The identity of the The authors are indebted to Mr. H. M. Fortuin for carbon source does not influence the solu- preparing the iron catalysts. The investigations were bility of carbon, but determines the way in supported by the VEG Gasinstituut nv. and by the which the carbon is taken up by the metal. Netherlands Foundation for Chemical Research (SON) Whereas formation of a metal carbide from with financial aid from the Netherlands Organization graphite and metal is thermodynamically for the Advancement of Pure Research (ZWO). Fi- nally, the authors wish to express their appreciation to unfavorable and only a small amount of car- Dr. 0. L. J. Gijzeman for his careful comment on this bon can be interstitially dissolved, reaction paper. with carbon-containing gases results in the formation of metastable, substoichiometric REFERENCES carbide compounds. We have shown above that both for iron 1. de Bokx, P. K., Kock, A. J. H. M., Boellaard, E., Klop, W., and Geus, J. W., J. Catal. %, 454 and nickel the intermediate carbide is of the (1985). hexagonal structure. Also in the case of co- 2. Baker, R. T. K., Barber, M., Harris, M., Feates, balt a hexagonal carbide has been identified F., and Waite, R., J. Catal. 26, 51 (1972). under growth conditions (44). It is tempting 3. Rostrup-Nielsen, J. R., and Trimm, D. L., J. Ca- to assume that these carbides, character- tal. 48, 155 (1977). 4. Buyanov, R. A., Chesnokov, V. V., Afanes’ev, ized by a relatively low density, facilitate A. D., and Babenko, V. S., Kinet. Katal. 18, 839 carbon transport, thereby explaining the (1977). high rates of filament formation. 5. Massaro, A., and Petersen, E. E., J. Appl. Phys. 42,5534(1971). CONCLUSIONS 6. Cooper, B. J., Trimm, D. L., and Wilkinson, A., in “Proceedings 4th London Int. Conf. Carbon From the combination of (saturation-) Graphite 1974,” p. 396. Sot. Chem. Ind., London, magnetization measurements and (quantita- 1976. tive) temperature-programmed hydrogena- 7. Rostrup-Nielsen, J. R., in “Catalysis, Science and 480 KOCK ET AL.

Technology” (J. R. Anderson and M. Boudart, 25. Kleefeld, J., and Levenson, L. L., Thin Solid Eds.), Vol. 5, p. 77. Springer, Berlin, 1984. Films 64, 389 (1979). 8. Baker, R. T. K., and Harris, P. S., in “Chemistry 26. Smith, M. A., Sinharoy, L., and Levenson, L. L., and Physics of Carbon” (P. L. Walker, Jr. and P. J. Vat. Sci. Technol. 16, 462 (1979). A. Thrower, Eds.), Vol. 14, p. 83. Dekker, New 27. Sinharoy, S., and Levenson, L. L., Thin Solid York, 1978. Films 53, 31 (1978). 9. Baker, R. T. K., Alonzo, J. R., Dumesic, J. A., 28. Tot&up, P. B., J. Catal. 42, 29 (1978). and Yates, D. J. C., J. Catal. 77, 74 (1982). 29. van Meerten, R. Z. C., Beaumont, A. H. G. M., 10. Wada, T., Wada, H., Elliott, J. F., and Chipman, van Nisselrooij, P. F. M. T., and Coenen, J. W. J., Metall. Trans. 2, 2199 (1971). E., Surf. Sci. I.%, 565 (1983). 11. Buyanov, R. A., Kinet. Kutal. 21, 189 (1980). 30. Kehrer, V. J., Jr., and Leidheiser, H., Jr., J. 12. Buyanov, R. A., Chesnokov, V. V., and Phys. Chem. 58, 550 (1954). Afanas’ev, A. D., Kinet. Katal. 20, 166 (1979). 31. Martin, G., and Imelik, B., Surf. Sci. 42, 157 Z3. Figueiredo, J. L., and Trimm, D. L., J. Catal. 40, (1974). 150 (1975). 32. Martin, G., Dalmai-Imelik, G., and Imelik, B., in 14. Sacco, A., Jr., Thacker, P., Chang, T. N., and “Adsorption-Desorption Phenomena” (F. Ricca, Chiang, A. T. S., J. Catal. 85, 224 (1984). Ed.), p. 433. Academic Press, New York, 1972. 15. Audier, M., Bowen, W., and Jones, W., J. Cryst. 33. Kuijpers, E. G. M., Breedijk, A. K., van der Wal, Growth 63, 125 (1983). W. J. J., and Geus, J. W., J. Catal. 72,210 (1981). 16. Zijlstra, H., in “Series of Monographs on Selected 34. Le Caer, G., Dubois, J. M., pijolat, M., Perri- Topics in Solid State Physics” (E. P. Wohlfarth, chon, V., and Bussibre, P., J. Phys. Chem. 86, Ed.), Vol. 9, p. 16. North-Holland, Amsterdam, 4799 (1982). 1%7. 35. Andersson, S., and Hyde, B. G., J. Solid State 17. Selwood, P. W., in “Chemisorption and Magneti- Chem. 9, 92 (1974). zation,” p. 46. Academic Press, New York, 1975. 36. Fasiska, E. J., and Jeffrey, G. A., Acfa Crystnl- 18. Kuijpers, E. G. M., Ph.D. thesis. State University logr. 19, 463 (1%5). of Utrecht, Utrecht, 1982. 37. Goodwin, J. G., and Parravano, G., J. Phys. 19. Lander, J. J., Kern, H. E., and Beach, A. L., J. Chem. 82, 1040 (1978). Appl. Phys. 23, 1305 (1952). 38. Schieber, M. M., in “Series of Monographs on 20. Richardson, J. T., and Desai, P., J. Caral. 42,294 Selected Topics in Solid State Physics” (E. P. (1976). Wohlfarth, Ed.), Vol. 8, p. 159. North-Holland, 21. Niemantsverdriet, J. W., van der Kraan, A. M., Amsterdam, 1967. van Dijk, W. L., and van der Baan, H. S., J. Phys. 39. Renshaw, G. D., Roscoe, C., and Walker, P. L., Chem. 84, 3363 (1980). Jr., J. Cafal. 18, 164(1970). 22. Nagakura, S., and Oketani, S., Trans. Iron Steel 40. Hofer, L. J. E., Cohn, E. M., and Peebles, W. C., Inst. Jpn. 8, 265 (1968). J. Phys. Colloid Chem. 54, 1161 (1950). 23. McCarty, J. G., Hou, P. Y., Sheridan, D., and 41. Escoubes, M., and Eyraud, C., Bull. Sot. Chim. Wise, H., in “ Formation on Metal Sur- Fr., 1374 (1966). faces” (L. F. Albright and R. T. K. Baker, Eds.), 42. Nagakura, S., J. Phys. Sot. Jpn. 12, 482 (1957). ACS Symposium Series 202, p. 253. Amer. Chem. 43. Kingery, W. D., “Introduction to ,” p. Sot., Washington, D.C., 1982. 236. Wiley, New York, 1967. 24. Palczewska, W., “Advances in Catalysis,” Vol. 44. Renshaw, G. D., Roscoe, C., and Walker, P. L., 24, p. 245. Academic Press, New York, 1975. Jr., J. Catnl. 22, 394 (1971).