Diffusion in a-Fe-Si Alloys Cite as: Journal of Applied Physics 41, 5193 (1970); hittps://doi.org/10.1063/1.1658644 ‘Submitted: 10 June 1970 « Published Online: 16 December 2003 R. J. Borg and D. ¥. F. Lai ce) ARTICLES YOU MAY BE INTERESTED IN Calculation of self-diffusion coefficients in iron AIP Advances 4, 017128 (2014); hittps://doi.org/10.1063/1.4863462 The self-diffusion coefficients of liquid binary M-Si (M=Al, Fe, Mg and Au) alloy systems. by first principles molecular dynamics simulation AIP Advances 9, 035328 (2019); https://doi.org/10.1063/1.5067295 New Fe-based soft magnetic alloys composed of ultrafine grain structure Journal of Applied Physics 64, 6044 (1988); hitt/os://dol.org/10.1063/1.342149 EN eM aN ei) ETT eA y4 Read. Cite. Publish. Repeat. \\77°"PAcTctor” ‘Journal of Applied Physics 41, 5193 (1970};https://doi.org/10.1083/11658664 4a, 5195 (©1970 The American Institute of Physics.JOURNAL OF APPLIED PHYSICS VOLUME 41, NUMBER 13 DECEMBER 1970 Diffusion in a-Fe-Si Alloys* R. J. Boro axp D. ¥. F. Lat Laxeronce Radiation Laboratory, Uniersity of California, Livermore, California 94350 (Received 10 June 1970) ‘The difusion coeficients for both Fe and Si diffusing in dilute «-Fe-Si alloys have been measured. A value for the vacancy-silicon binding energy has been calculated and an explanation offered for the con- and the activation energy for each composition, If one assigns the usual uncertainty limits (j.e., +1 keal) to and as much as a factor of 2 to D*, the concentration effect could be accounted for by error alone. However, plots of D as a function of composition reveal the vari- perature for si cups with fesulls Gileulaed from Matano anal: Note the apparent equality of slope forall concentrations. 5195 ‘Tanux TI. De and Q, the baste parameters of the Arshenius equa- tion for the diffusion of Si as a function of composition. D @ ‘—(ert/sec) (eal) 0.0000 0.738 32.7 0.007 0.78 327 0.001 0.77 82.5 0.0187 0.96, 52.6 0.0281 1.05 S24 0.0374 1.8 52.6 0.0821 124 52.2 ation in D> to be indeed significant, The data as dis- played in Fig, 4 clearly refute our initial assumption; viz., that the concentration dependence of Ds: is negli- ible below 1% Si. This may in part explain the dis- crepancy between Eq. (2) and the result obtained by extrapolation for (Si)=0 as shown in Table TT. Although the respective activation energies, 54.5: and 52.731 kcal are, strictly speaking, within the combined limits of error, the strong trend of Dsi with Si concentration forces us to conclude that Eq. (1) is insalid even for relatively dilute sources. This in turn suggests that the apparent linearity of such plots as are shown in Fig, 2 does not constitute a sensitive test of the concentration dependence of D. Nevertheless, for the practical purpose of calculating Ds: in pure Fe, there is relatively little difference between the two equations, especially at the high end of the tempera- ‘ture range. “The near constant values of (as revealed in Tabl> UI clearly force the concentration dependence of Ds the pre-exponential factor of the Arrhenius relation- ship. Tt is perhaps not surprising that Q would remain unaffected by such small additions of solute (viz., 0-4 at-%), but the simple linear dependence of Ds; upon composition, as shown in Fig. 5, was unexpected. Asa result of this elementary relationship between Dat and the atom fraction of solute Xsi, we can express all the Si diffusion results with good precision by Eq. (5), where R, the gas constant, has the units of keal/°K: Dgi= 0.735 (14124X5i) exp(—52.53/RT). (5) These data are fit equally well by expressing the con- Bg 5, The conn Lion dependence of Dy a irae completely ip vihich varies leary the beponental ee nes Sept aid the daca Siding teenaivcaly the correct functional de pendence5196 RJ Pra, 6, An Arthenius ot of the data for dt fusion of Fe w-Fe-Si alloys. (The data. for Self-diftusion is) taken from Ref. 6) Aecisecal temper, 1/1 1 = 94 centration dependence as an exponential, viz,, D*(Si) = Drexp(12.4Xsi), so that the data do not distinguish between the exponential and its expansion to first order. ‘The previously reported values of the diffusivity do not agree with the present results. However, the com- positions differ for all three investigations. The measure- ments of Batz, Mead, and Birchenall* were made be- tween 1095°-1347°C, with couples in which half con- tained ~4.5% Si and the source half contained ~7.2%. These workers report 48.0 kcal for the activation en ergy but admit that the value may be in eror by as much as 6 kcal. The measurements of Mitani, Omishi and Shikanof were also performed at higher Si concen trations than ours, and hence are not directly compa- rable; they span but 90°C; viz., 1130°-1220°C. Our value for the activation energy of a 4.2%-Si alloy is 525 keal, which is considerably larger than the 44 kcal reported for a 77-Si alloy by Mitani et al. There is no trend in our silicon results which would allow extrapola~ tion to this value, though the activation energy and DP show a large decrease in Q within the intervening 28% Si for diffusion of Fe (see Fig. 7). As Dat and Dr. are not independent of one another but are related through the correlation factor, Dsi may in fact drasti cally increase in going from 4% to 7% silicon. We can only compare our data by extrapolating with Eq. (5) up to 7% Si. Choosing 1150°C as a con- venient common temperature at which to compare, we obtain a value of D of 4.68X10-* em*/sec from a eps nt er sterttar ro, 7, The measured activation energies Q and D* values for "Dre and Q as calculated from 2g (12). BORG AND D. Y. F. LAL ‘Taaus. TIL. Activation energy and D* for the diffusion of Fe caleulated from a least-mean-squares analysis of the data in Fig. 6. The deviation « is the average difference betmeen the measured and calculated values of D expressed in percent of the ccleulated values. Si @ D 2 {at.%) eal) em/see)(%) ° 56.5 1.39 . 47 55.5 1.68 45 63 549 1.65, 0.65 8.2 50.9, 0.30 Le 13, Sit nu 11s wT SLT 1.46 4.08 Eq, (5), 28X10 emt/see from the equation given by Mitani ef al., and 1.8710" cm?/sec from that of Bats ef al. It is obvious that we are in better agree- ment with the former, although there is gross dis- agreement among the three. The apparently low values of Q resulting from the previous two investigations may be partly the result of the extremely narrow range of temperatures which greatly curtails the accuracy with which Q can be calculated. We now turn our attention to the results of the diffusion of iron, Figure 6 compares diffusion in the alloys with that in pure bec Fe, and Table ITT lists the values of @ and D* obtained from a least-mean-squares analysis of these data. These data are also plotted in Fig. 7, from which it can be seen that a substantial decrease in the activation energy occurs in the vicinity of 1%-8% Sie A common temperature of 900°C was selected to determine the effect upon D of varying the composition. ‘This temperature was chosen because it lies above the range of the magnetic effect but yet well below the ay transition, These results are given in Fig. 8. The linear portion of the graph (ie., up to 7% Si), is well Frio. 8. Dra at_ 900°C normalized by the value for feifedifusion at this tem perature,DIFFUSION IN a-Fe-Si ALLOYS described by Eq. (6), Des =Dy(1+6Xe1), (6) where D.=3.31X10-, the diffusion coeflicient for self-diffusion at 900°C, and 6=20. This equation has the form originally proposed by Hoffman, Turnbull, and Hart® to describe diffusion in Ag-based alloys. Tt is difficult to establish the correct functional depend: ence upon composition, and in fact the data are per- hhaps even better represented on a log scale"as shown in Fig. 9. Further, these results warn us not to place too much faith in absolute numerical values, for, to be consistent with the values of D(Fe) calculated from from Fig. 7, the measured values of this ratio at 900°C ust lie below those caleulated for 1100°K, not above it. This lack of sensitivity results because of the small variation in Q and D* for Dy. within the concentration range of interest; actually this discrepancy lies well within the experimental limits of error. ‘Thus, neither the Fe nor the Si data clearly resolve the correct func- tional form, logarithmic or linear. Unfortunately, the correct theoretical equations derived from simple models, of alloy diffusion are applicable only in the dilute region where the sensitivity is inherently small. Our results are not in very good agreement with the results of Mills ef al! who obtained for the Archenius relationship, Dp= 0.440.20 exp(—52.200:43000/RT), nor do our results agree with the results of Borizov et al who obtained Drv=0.62 exp(—51 500/RT). Our result for the same composition, viz., 5.8 at.% Sis Dro=1.66 exp(—S5 100-:2000/R7). Strictly speaking our results and those of Mills et al. 10. 9. The values of the diesion coxficient of Fe in the loys, Dry(Fe), noone by the selaiuson value Dr {hie sao Togaitimialy plotted for varius tmperntur at 8 function of fompostion, The valves at 1173 age the sa those shown in Fig, 8y ibe others are obtained By interpol of the daa shown in ig 5197 Fro. 10. This drawing, taken from LaClairal! defines and labels ‘the various kinds of jumps, The Si is located at position 0, and there are ‘nite vacaney-Si binding energies associated ‘with positions 1 and 2 are well within the combined limits of error, but in actual fact one would hope for better agreement. We can offer as yet no adequate explanation for this dis- crepancy Iv. DISCUSSION We can use D(Fe) to compute the silicon-vacancy binding energy using the sort of analysis frst proposed by Lidiard!” for fce crystals and which recently has been extended by LeClaire" to include bee structures. We will use essentially the same notation as LeClaire, and for the sake of completeness and clarity repeat here the essential features of his derivation. ‘The model assumes there to be a Gibbs free energy of binding for only those vacancies which are nearest- or next-nearest neighbor to a solute atom. The free energies of formation for vacancies located at any of the eight nearest-neighbor positions (aa/2)v2 as well as those situated at any of the six next-nearest-neighbor positions a, are defined by Eqs. (7) and (8), respec- tively: eat a AG,?= AGP +AG:. (8) ‘The free energy of vacancy formation in the pure solvent is defined as AG,*, and it isfurther assumed that the solute and overall vacancy concentration are suffi- ciently low that one may neglect the possibility of clusters of either vacancies or solute atoms. The model then assigns a spectrum of jump frequencies, as defined in Fig. 10, for all solvent atoms which can exchange places with a bound vacancy, or conversely can bring vacancy into one of the two bound positions, via a single jump. All other solvent jumps are considered to lie outside the sphere of influence of the solute and are assigned a single jump frequency, wo, the same as they ‘would possess in the pure solvent. Calculating a diffu- coefficient under the foregoing set of assumptions consists merely of giving each jump frequency its properPro, 11, A plot ofthe ratio of the diffusion’ coeficient of Fe in thealloy and in pure a-Fe statistical weight and then summing over all distin- guishable frequencies in order to obtain the average value. Performing these steps LeClaire obtains Eq. (9) as his final expression which relates the average jump frequency of the solventto the soluteconcentration, Xi (Si)=8 exp(—AGi/AT) {Neil (Be! +3e1) Xesp(—AGY ET) + (Sart 3a) exp(—AGi/AT) + 3uy/ fai" + Sep 20sy] tu]. (9) Now the diffusion coefficient is related to the jump frequency by Eq. (10) Dra(Si)=tasP si fo where ay is lattice spacing and fe the correlation coefh- cient, If we assume with LeClaire that the correlation coefficient is the same for both the pure solvent and alloy we can write Eq. (11) (Dre(Si)/Dre)= (1+0X si), (10) ay where b= = 204 (1/0) (Bus! tes” + Bus) exp(—AG/ET) + (Bax+ Bus) exp(—AG2/kT )-+ dees’+05""+ 3c], By making use of the relationships of detailed balance between the various lattice sites the expressions for & may be simplified, without further assumptions heing made, to Eqs. (12) or (13): b 20+ (1/ao)[ (Ou! -+ 2us"+ 602) Xexp(—AG/AT)+6u] (12) 20+ (1/ec0)L (6us+ 6.5) exp(—AG,/kT) HOi'+2ai" (13) We have plotted Deo(Si)/Dre, and the results are shown in Fig. 11. Its clear that the slope is not strictly f constant a8 predicted by Eq. (11), but this is not too surprising, for to be consistent with the model of LeClaire, the slope need not be constant except in the region of extreme dilution which is where we have in fact calculated 6 for the three temperatures shown in Fig. ®. These values are 6 28.4 TCK) 1100 ‘These individual values may vary by a factor of 2 24 1400 20.4 1700, J. BORG AND DL ¥ LAL depending upon the method of analysis, but the tem- perature dependence as given implicitly by Eq. (14) will not vary greatly. We cannot analyze the data using Eqs. (12) or (13) in their present form, since there are far too many unknowns. Tnstead, we take the derivative which gives, Eq. (4) 8 In(6+20) Gus exp(3GV/AT) |: a(a/T) [om /+( ae (ear ata) 4) We still cannot solve Eq, (14) directly but we can set reasonable limits to Aff which may be considered to be the binding energy for the nearest-neighbor position. ‘The linearity of the plot of In(b+20) vs T-!, as shown in Fig. 12, indicates that AG, is small compared to £7. Furthermore, because Si enhances the diffusion of Fe wwe can assume a negative bi say, AG1<0, Thus setting AG, upper limit AHi= —0.66 keal. For a lower limit we set AG=—=, which gives AIL=—046 kcal. The first approximation also requires setting «: which leads to the same result as assuming ws: ex’ exp(2GW/AT). ‘The values of AM; thus cal- culated are extremely insensitive to a systematic error in 6 although an etror of as little as 10% in the ration, Do(Si)/Dre leads to a considerable change in the absolute value of b as shown in Fig, 12, ‘Thus, within the limits of the aforesaid approximations, we conclude that the $ y binding energy is ‘~1 kcal. It is unfortunate that any set of approxima- tions which permits one to solve Eqs. (12) and (13) climinates their distinguishability. Using the same sim plifications as before, we calculate for the upper limit AH,=—0.81 keal; the lower limit of course remains the same. Hence, our Binal values for the binding en- ergy might apply equally well to Alf; or Alls. How- ever, one might expect that general relaxation about the vacaney in the bee structure would render identical these positions (ie., nearest and next-nearest neigh- bors). Recent isotope measurements by Gonziles and Walsie de Reca"* in Fe-3% Si appear to confirm this hypothesis. As shown in Fig. 7 the activation energy for Drs decreases very little up to approximately 7% Si at which point it abruptly decreases to less than 90% of its initial value. One can easily rationalize the initially n dependence of Q as follows: We weak composi ic, 12, plot ofthe param. cer eceuery forthe slution ot Eq ta). Note che apparent linearity. The ers ats cor Feapond to 10% erzor in the i)/ Bee (eeDIFFUSION IN a-Fe-Si ALLOYS can calculate the relative concentration of vacancy-Si complexes assuming a completely random distribution of both vacancies and Si. This is of course not strictly true because of the negative enthalpy of binding, but the approximation is in accord with other approxima- tions necessary for the calculation. We then assume that the energy of vacancy formation varies additively with the number of nearest- and next-nearest neighbor Si. Thus we may write Eq. (15): AH,(Si)—AH°(0) u (14 -2( ) ean (15) ‘Taking AH equal to 1.0 kcal, we are able to reproduce the experimentally obtained ‘values which are shown in Fig. 7 up to approximately 5% Si at which point the experimental values deviate significantly from the theoretical. Tt is quite probable that the diffusive be- havior of the Fe is in this case amenable to theories which limit the vacancy-solute interaction to nearest, and next nearest neighbors only below ~4-5% Si, Certainly, the drastic change in both Q and D* pre- vents the extension of this simpler analysis to higher solute concentrations Considering the results for the diffusion of Si, first, cone can quite easily rationalize the Tinear dependence of Dsi upon composition, i.e., Eq. (5), a8 follows: The iffusion coefficient at fixed temperature is directly proportional to the concentration of Si-v complexes which is given by Eq. (16) NXoie=ZXaiX, exp(AGi/kT)WZNX5X, exp(AG/ET), (16) where Xsio, Xsi, and Xpi° are the mole fractions of complexes, free Si and total Si, X, the mole fraction of vacancies, and AGi the free energy of binding. The approximation of Xs; by Xs is a particularly good one in view of the relatively small binding energy, viz., ~2 kcal. In terms of Ds’, the pre-exponential term at Xsi=0, we can represent the temperature and com- position dependence by Eq. (17): Dai= Dest exp— (AH g!-+ AH, AHi)/ATYA+2ZXs1). any Comparing this with the experimentally obtained Eq, (5) we set Das*= 0.735 (see Table II) and Z= 124 (see Fig. 5). Of course Z, properly speaking, must ‘equal 8 or 14, depending upon whether one chooses to count next-nearest as well as nearest neighbors. The fact that Z is experimentally much closer to 14 would seem to indicate the existence of a negative binding energy for a vacancy in either of the two sorts of posi- tions. In any case the difference between 12.4, as shown in Eq. (5), and 14 is within the limits of error. It should be noted that this result is perfectly general for solute 5199 diffusion and any deviation from this result indicates a long range interaction between solute atoms. LeClaire derived Eq. (18) for the correlation coeti- cient fc using the same model which he used to derive Eqs. (12) and (13), with the additional approxima. tion that «=== Dsu(0) oh L071 D, (0) b+ 154 LL SM+9B)/OF900))" (18) Substituting fo=0.727 which is the correlation coefli- cient for diffusion in an elemental bee structure, and taking b= 28, we calculate fi=0.71 at Xsi=0. Thus, as might be expected for a faster diffusing solute with negative vacancy binding energy, fst is less than fy Y. CONCLUSIONS Our conclusions, beginning with the experimental ones, are as follows: (1) The apparent linearity of graphs of solute con- centration vs distance on probability paper (i.., Fig. 2) does not constitute a reliable basis for concluding that Dis independent of concentration. (2) Fe-1.8% V may be used to extend the range of, the bee phase above its normal transition temperature with values of Dsi being the same as in pure Fe. Tt is, quite likely that other properties of bec Fe could be more accurately determined by extending measure ments to higher temperatures using this binary alloy. (3) A vacancy-Si binding energy of ~—0.5 kcal can be calculated based upon LeClaire’s model, and this value is at least consistent with all the experi- ‘mental results and subsequent analysis. Although the calculation is straightforward, the model is strictly valid. only for extremely dilute solid solutions. This unfortu- nately greatly restricts the range over which useful measurements can be made which in turn greatly di- minishes the accuracy of the pertinent data. Conse- quently, independent of the other approximations in- voked by the model, the accuracy of such calculations is limited by the inherently small experimental range and can never become extremely high. Nevertheless, it is within the capacity of such experiments as this to distinguish easily between the extreme values of binding energies which are sometimes reported. ACKNOWLEDGMENTS The authors gratefully acknowledge the invaluable assistance of Yuriko Hashimoto and Ralph Otto in the experimental portion of this work. Also, they wish to thank D. D. McCoy and R. G. Gutmacher for the microprobe analyses. We are also indebted to A. D. LeClaire, not only for several profitable discussions but, also for permission to use his analysis prior to its publication,5200 R. Work pevformod under the auspices of the US. Atomic Enexy Cotmiton, DEY Lat and. J. Borg, Lawrence Radiation Laboratory ep, UCL th 1 FT onan VEAL, Goliloe, BY, Ljubow, and GY. ‘ebeibedily, in Rodioucipes i Sceuic Record ited IC, Excermann (Pergamon, New Vork, 1988), Vol. 1, p. 212, “fics, Gr Re Walker and GM. Lea, Phi Mag 13, 935 0988) Oy Batt 1070 (8 14 Mian, M, Onishi, and . Sbikan, J. Jap. Tat. Metals (Seni) 90,88 (1960) DAY. tata or, Tian. IME 238 1975 (1985) TR J Borg Dc: Laland G2 Me eroran, basence Racin Lairatry Rep CORIO, 1962 IW. Mead, and C. E, Birchenall, J, Metals 4 . BORG AND D. ¥. F. LAT "R.A. Vecher, Va. I Gerasimov, and V. A. Geiderikh, Russ. J. Phys, Chem. (Leng. Tranal.} 39, 650 (1965). , Hofman, . Turnbull, and E. W. Hart, Acta Met. 3, 417 (i985), B. Lidiard, Phi Mag. 5, 1171 (1960). WA’. LeClalee, Phil. Mag. 21, 819 (1970) 1a There exists soime confusion concerning the “binding energy’ of a vacaney which is often taken to be a "dssociatin energ the only diference being the sign. We believe that itis mare evh- sistent to regard the binding energy a8 an increase or decrease energy of formation rather than discociation; hence a negative binding energy implies a lower energy than otherwise for a vacaney adjacent to a solute atom. BC. Oredo de Gonzalez and N. K. Walste de Reca (private ccommiunieation) JOURNAL OF APPLIED PHYSICS vouuME 41, NuMBER 1s DECEMBER 1970 Mason Dark Space and High-Pressure Arc Re-Ignition Jos E. Ware Lighting Research Laboratory, General Electric Compuny, Cleland, Ohio #4112 (Received 26 June 1970) A solid object immersed in the plasma of «high pressure ac is surrounded by a dark sheath, The sheath is at plasina temperature along one boundary and near thermal equilibrium with the solid object at the ‘other boundary. Because of ion diffusion into the sheath, the latter easily transmits low cleetron current ‘densities, But to pass high current densities, such as those carried by the eletredes, sheath insulation rust be broken dowa. Tis breakdown or reignition process occurs at the electrodes after current zero every half-eycle of an ac are. In the mercury arcs studied, reignition from current zero involves heating ‘of a channel which extends outward more than J mm from the electrode toa temperatu of about 4400°K, which requires the ipa of about 0.64 Jem to the vapor. Addition of iodides tothe vapor may inhibit the ‘eigaition process by either eleetzade or vapor-phase processes 1. INTRODUCTION ‘The tendency of an ac arc to extinguish at current zero is useful in switches and circuit breakers, but ob- jectionable in devices like gas-discharge lamps. Lamps ‘which operate normally in a high pressure of mercury vapor are generally started at low pressure; if they extinguish at current zero after the vapor pressure has risen to the vicinity of its normal operating value, the starting circuit may be incapable of restarting at the high pressure. In this case the lamp “drops out” or extinguishes, and cannot be restarted until it cools to 1 low vapor pressure, which may take minutes. High- Pressure mercury vapor lamps with metal iodide addi- tives for color improvement have experienced trouble from such dropout, as discussed in the recent literature. Solutions have been found by proper choice of circuitry for the lamps, but there remains some disagreement as to the physical basis of the problem.!* The present paper offers evidence as to the mechanism of re-ignition of an ac arc in high-pressure mercury vapor after cur- rent zero, The effect of adding iodides is also considered briefly. Il, NATURE OF THE MASON DARK SPACE In 1924, Langmuir and Mott-Smith showed that a negative or floating probe in a low-pressure plasma is, surrounded by a nonluminous sheath or dark space. ‘The sheath is dark because electrons are electrostat- ically reflected from it, and do not penetrate sufficiently to produce excitation of the gas. A probe immersed in the plasma of a high-pressure arc also is surrounded by 1a dark space; this was early interpreted as due to the same cauise as in low pressure, and papers appeared purporting to give electron temperature in high-pres- sure discharges by the Langmuir technique. It was Mason who first pointed out? that the probe dark space in high-pressure thermally ionized plasmas is actually ‘a result of local cooling by the probe. His ingenious experiments and analysis warrant the designation of this sheath as the “Mason Dark Space.” ‘A photograph of the dark space around a S-tnil-diam tungsten probe in @ 3.2-A ac atc at about 3.5-atm mercury vapor pressure, is reproduced in Fig. 1. The picture was taken off axis so that the probe, whose end was dark, could be seen by reflected light. This gives, the dark space exagyerated ovality. Mason’s equation allows, with observed space thickness and probe tem- perature, calculation of the watts conducted thermally through the dark space to the probe. Such a calcula- tion, taking the temperature outside this dark space to be Elenbaas’ “effective” valuet of 3555°K, gives 86 Wem? conducted to the probe. Dissipation of this power density by radiation requires 2614°K, which