Upload
andrew-howe
View
215
Download
0
Embed Size (px)
Citation preview
Scripta Materialia 49 (2003) 619–623
www.actamat-journals.com
Discussion concerning the rationalisation of diffusion
Andrew Howe *
Department of Engineering Materials, University of Sheffield, Sheffield S1 3JD, UK
Received 6 February 2003; received in revised form 6 February 2003; accepted 11 June 2003
Abstract
Further support is given for the proposed rationalisation of interstitial diffusion, which is also extended to the
substitutional case. A discussion is presented in order to answer various issues raised by Erdelyi and Beke in their
response to the original paper.
� 2003 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
Keywords: Alloys; Diffusion; Theory; Modeling
1. Introduction
Not surprisingly, daring to query the funda-
mental formulation of solute diffusion [1] has
prompted serious argument [2], and I welcomesuch debate. In brief, the proposal is that the un-
derlying form of Fick�s first law:
J ¼ �DdCdx
ð1Þ
for isothermal diffusion, where J is the solute flux,D is the diffusion coefficient, and C the concen-
tration at a position, x, is better written as:
J ¼ � dðDCÞdx
¼ �D � dCdx
� C � dDdx
ð2Þ
It is also proposed that for simple systems
D ¼ cDI, and thus alternatively:
*Address: Corus Group plc, Swinden Technology Centre,
Rotherham S60 3AR, UK. Tel.: +44-1709-825328; fax: +44-
1709-825337.
E-mail address: [email protected] (A. Howe).
1359-6462/03/$ - see front matter � 2003 Acta Materialia Inc. Publisdoi:10.1016/S1359-6462(03)00352-X
J ¼ �DIdðcCÞdx
¼ cDI 1�
þ Cd lncdC
�� dC=dx ð3Þ
where DI is the value which the diffusion coefficientwould have were it an ideal solution, and c is theactivity coefficient, as the departure from ideality.The term DI houses the �datum� value of the acti-vation energy, which is then altered by RT ln c fordepartures from ideality. Mathematically, it is
identical to invoking c simply as a factor within D[1].
Both these aspects can be found in the literature,
and the claim that their synthesis is wrong [2]
might stem partly from a matter of definition. Theproposal is a necessary result if the diffusion co-
efficient is defined in terms of the atomistic
mechanism of diffusion, as is commonly assumed,
but the issue is confused by the treatment of Eq.
(1) as being true by definition, at the expense of a
strict, mechanistic interpretation of D. Not sur-prisingly, this loss of physical meaning of D leadsto inconsistencies, complications and errors, whicha mechanistic definition would avoid.
hed by Elsevier Ltd. All rights reserved.
Fig. 1. The partial fluxes J12 and J21 differ as the product of theconcentrations C and jump frequencies C in their source planesand thus according to the perceived extra energy, G� l1 orG� l2, in a chemical potential gradient.
620 A. Howe / Scripta Materialia 49 (2003) 619–623
2. Basis of the argument
The argument is restricted to one-dimensional
diffusion and relatively dilute solutions to avoidunnecessary complications of concept. (My ter-
minology in the original paper appears to have
caused some confusion [2], so I will now adopt that
as employed by Philibert [3].) The diffusion coef-
ficient on this basis, is simply:
D ¼ k2C ð4Þ
where k is the jump distance and C is the jumpfrequency in that direction. The rate at which sol-
ute is jumping from a given plane must be given by
the product of the jump frequency and the con-centration. The net flux between two adjacent
planes must therefore derive from the difference
between these products, which necessarily leads to
Eq. (2) rather than Eq. (1). I had only noted this
stated explicitly in the text of Christian ([4, p. 349])
although it can be inferred from Philibert [3] also.
Philibert defines the flux equation as:
J ¼ �DdCdx
þ V � C ð5Þ
where V is the drift velocity, resulting from effectsother than the concentration gradient. When
providing an atomistic interpretation of this
equation ([3, pp. 33–34]), his D is related to the
average jump frequency from plane 1 to plane 2
and from plane 2 to plane 1, and V is defined interms of the difference between these frequencies.
However, on substituting his expansions for D andV into Eq. (5), it can be shown that this generatesEq. (2) (with D, like C, being defined in relation toeach plane rather than averaged) but, curiously,
this is not mentioned in the text.
In their criticism of my original paper, Erdelyi
and Beke cite the text of Manning [5] and indeedthis is a highly relevant text. Manning invokes Eq.
(2) in combination with variously defined drift
velocities. However, on p. 232 he states that: ‘‘by
definition of the diffusion coefficient, when there
are no driving forces, J ¼ �D � dC=dx.’’ D is thusdefined in terms of the average energy barrier as
perceived by adjacent planes (p. 230) rather than
the actual ones, and is no longer strictly related tothe jump frequency as in Eq. (4). Manning ac-
knowledges that only his tracer diffusion coeffi-
cient is thus directly related (p. 23). After tortuous
argument, his result is the same as Philibert�s, i.e.Eq. (5), with identical definition of V :
V ¼ DFRT
where F ¼ �RT � d lncdx
ð6Þ
for this isothermal case and in the absence of other
potential driving forces, F , such as electrostaticfields or stress gradients. This V term in Eq. (5)
represents the dD=dx term in Eq. (2) as describedabove, and therefore:
dDdx
¼ Dd lncdx
ð7Þ
This can only be true if D is the product of some�ideal� value, DI, and the activity coefficient, c,which is consistent with Fig. 1 [1], although this
point does not appear in either text:
dDdx
¼ dðcDIÞdx
¼ DIdcdx
¼ cDI1
c� dcdx
¼ cDId lncdx
¼ Dd lncdx
ð8Þ
Manning describes the perceived energy barrier
from adjacent planes 1 and 2 as being G� E andGþ E respectively, such that the total energy dif-ference is 2E which is identified as kF . By simplesubstitution for F from Eq. (6), this energy dif-
ference is RT � d ln c, where d ln c is the difference inthis term evident between plane 1 and plane 2. This
0
2
4
6
8
10
12
0 1 2 3 4 5 6 7At % C
D (E
-11
m2/
s)
ig. 2. The effect of carbon concentration on its diffusivity in
cc-iron at 1000 �C, Experimental data (12); N Predictions
rom current work; j Predictions from traditional formulation.
A. Howe / Scripta Materialia 49 (2003) 619–623 621
is fully consistent with the analysis from [1], where
the resultant G�s in planes 1 and 2 are modified byRT ln c for D, and by RT ln cC for the flux, J . ThusC and hence D varies according to the gradient ofthe non-ideal part of the chemical potential, and
the flux is driven by the full chemical potential
gradient. Conventionally, no such c factor is
adopted and the flux equation is translated as:
J ¼ �DdCdx
þ VC ¼ �DdCdx
� CD � d lncdx
¼ �DhdCdx
ð9Þ
where h ¼ ð1þ C � ðd lnc=dCÞÞ as opposed to Eq.(3), which acknowledges the presence of c. Thusboth aspects of the proposed rationalisation have
been implied already in the standard literature,
though neither is now generally accepted. How-ever, it must be straightforward to test whether
the diffusivity (or mobility) houses such a factor
and whether Eqs. (3) or (9) agrees better with
experiment. Indeed, several authors have found
empirically that Eq. (3) (whether invoked in
terms of the activity gradient, dðcCÞ=dx, or themobility and thus diffusivity being proportional
to c) is successful for the bulk diffusion of simplesystems ([4, p. 366], [6, p. 463], [7–10]). Krishtal
[11] even considered Eq. (3) to be standard
theory for the �true� diffusion coefficients, corre-sponding to the DI values in this work. Fur-thermore, some proponents of this alternative
view appear to have suffered severely in discus-
sions of their work, even when clearly demon-
strating much better agreement with experiment[10].
Comparison with experiment can be aided with
Eq. (3) re-written in terms of the value of D at
infinite dilution, D0, rather than the value which
the system would have were it ideal. Thus:
D ¼ D0cHh ð10Þ
where cH is the �Henrian� activity coefficient, i.e.relative to that at infinite dilution. One of the most
thoroughly characterised interstitial systems is
probably that of carbon in iron. It is relevant
therefore to perform a direct comparison to seewhether Eqs. (9) or (10) is closer to reality, based
F
f
f
on the same value for D0 and the same, source
thermodynamic data, for this system. For this
purpose, experimental values for the diffusion co-
efficient were taken from Wells et al. [12], averag-ing their results for 999, 1000 and 1005 �C forgreater confidence in the nominal data at 1000 �C.The required thermodynamic data were taken
from the JMatPro commercial software package
[13]. In Fig. 2, the traditional expansion of the
value for infinite dilution by the thermodynamic
factor, h, can be seen to be significantly in error,whereas the results from the present analysis areseen to be substantially better, and very close to
the experimental results up to 5 at.% (�1.1 wt.%).Perhaps more complicated formulations can ex-
hibit even greater accuracy, but clearly much more
can be achieved with a simple analysis than is
commonly assumed.
3. Substitutional diffusion
Erdelyi and Beke�s paper [2] emphasises thesubstitutional case. The rationalisation is believed
relevant to substitutional diffusion also, although
the argument [1] was restricted to the interstitial,
or �independent� diffusion case, for the sake ofconceptual and mathematical simplicity, but brief
comments will be offered on it here. Of note,
Erdelyi and Beke described the case of substitu-tional diffusion by direct atom exchange, whereas I
622 A. Howe / Scripta Materialia 49 (2003) 619–623
believe it more useful to consider the usual case of
diffusion by the vacancy mechanism.
Considering the intrinsic diffusion with respect
to the lattice (i.e. specifically not the interdiffusionwith respect to the Boltzmann–Matano interface
which involves contributions to the �flux� fromatoms which are not actually moving), the picture
is still one of planes of solute atoms and their jump
frequencies. The jump frequencies are no longer
simply the product of the atomic vibration fre-
quency and the probability of a given atom having
sufficient energy to pass over an energy saddle-point on its jump path. This would apply only to
the proportion of solute atoms that happen to be
adjacent to a vacancy. This in turn will depend on
the vacancy concentration, for which it is usual to
assume that the small, equilibrium concentration
applies. This is given by an expression of the form
expð�Gv=RT Þ where Gv is the energy for vacancyformation, and thus, rather conveniently, thisprobability can be accommodated within the ac-
tivation energy expression for diffusion. A nice
account of this is presented by Shewmon [14].
Thus, although C is differently configured, the
point remains that there are planes of atoms with
associated jump frequencies. As before, the energy
difference controlling the jump probability will
depend on the solute�s chemical potentials in thepair of planes. Thus C cannot be assumed to be thesame, and Eq. (2) would result.
The extension to interdiffusion with respect to
the equal flux, Boltzmann–Matano interface, is
relatively straightforward. Traditionally, this �flux�,JAB, for a simple binary of A and B atoms with asmall, equilibrium concentration of vacancies, is
given by:
JAB ¼ �ðCADBh þ CBDAhÞ � dCAdx
ð11Þ
According to the revised analysis, the result is:
JAB ¼ �ðCADB þ CBDAÞ �dCAdx
� CACB �dDAdx
�� dDBdx
�ð12Þ
but as before, they are exactly equivalent if the
intrinsic diffusion coefficients should also house a cfactor as in Eq. (3).
4. Discussion
The proposed revision addresses the simpler,
typical cases of solute diffusion in the bulk matrixand does not purport to cover every eventuality.
As such, it should still have a very wide use, and
should be helpful in the consideration of the more
complicated cases. Certainly for the test case of
carbon steel, it appears to be sufficient for the
whole range of concentration of usual industrial
relevance, Fig. 2. Darken and Gurry [6] acknowl-
edged that the thermodynamic factor, h, would notbe sufficient for the general case, and that the
mobility (and thus diffusivity) should also include
some function of composition which would need
to be determined empirically for each case. The
present work revives earlier ideas that suggest one
can achieve considerable improvement from a
simple extension to the basic theory, which auto-
matically arises just from consideration of theatom jump process.
Erdelyi and Beke�s argument [2] centres on thenature of the jump frequencies, C, and the order ofexpansion of variables considered as a Taylor se-
ries. The jump frequencies employed here and in
standard texts (e.g. [3–5,14]) refer to the frequency
at which solute leaves the source plane. This is
conceptually simple and intuitive. Erdelyi andBeke�s frequencies are different in nature, being afrequency of atom exchange. However, for either
interstitial or vacancy diffusion the conceptual
model that I employ should be fundamentally ro-
bust. Regarding the influence of neighbouring
planes, only adjacent planes are interacting within
the timescale of the jump process: the probability
of a solute atom jumping two or more planes inthat timeframe is minute. Certainly, the sur-
rounding region exerts an influence, but that
should be adequately accounted for by the chem-
ical potential and activation energy. One would
expect that a solute atom is only �aware� of thechemical potential, l, where it is, and of the saddlepoint of its possible jump path (as is accommo-
dated by incorporating the c factor within C andthus D [1]). Provided that there is a jump path
available, the partial flux is controlled by the en-
ergy of the saddle point, and it does not matter
how much lower is the energy of the receiving site.
A. Howe / Scripta Materialia 49 (2003) 619–623 623
As put by Borg and Dienes [15], the physics of the
jump process is contained in C. For the C as em-ployed in this work to be affected by other, nearby
planes, but not to be accommodated already bythe resultant l in the plane in question, one wouldassume it must therefore be affecting the value of G(or in extreme cases, perhaps the attempt fre-
quency � atom vibration frequency, m). It wouldbe relatively straightforward to apply the proposed
rationalisation to cases where G is dependent onthe local concentration.
Erdelyi and Beke [2] also criticise the originalpaper for its use of linear expansion of variables.
This might be a significant issue if using their
atom-exchange C�s as functions of concentrationin several planes, but I would argue that it is
actually incorrect to use higher order expansions in
my case. I invoke the simple but robust conceptual
model of planes of concentration and their asso-
ciated jump frequencies. The local gradients inthese variables are simply the difference between
their values in neighbouring planes divided by the
jump distance, e.g. dC=dx ¼ dC=k. Terms such asC and C do not have any meaning other than inthe atom planes (although the partial molal energy
does, e.g. to the saddle point at energy G); they arenot continuous functions. These gradient terms
will exhibit step changes either side of a given atomplane (giving rise to Fick�s second law of how theseconcentrations change with time). No function of
whatever order is being imposed on them; they are
free to be what they are. This becomes a somewhat
philosophical point, but diffusion is an inherently
discretised phenomenon. It can be envisaged as a
finite difference process operating at the inherent
grid spacing of the jump distance.
5. Conclusions
The recently proposed rationalisation of inter-
stitial diffusion combines aspects of theory and
experimental evidence already present in the liter-
ature. However, key aspects are commonly over-looked, and their synthesis into a coherent whole
appears to be new and controversial. It should
prove superior to standard formulations well
beyond the trivial case of infinite dilution as stated
by Erdelyi and Beke, and this has been demon-
strated for the case of carbon in fcc-iron. The
conceptual model upon which it is based is be-lieved to be simple but robust, and indeed it has a
strong �pedigree�, being invoked by the authors ofstandard texts. Cases of boundary diffusion, high
concentrations and very steep concentration gra-
dients are less straightforward and would require
further complications in the analysis, just as they
do for the standard formulation of Fick�s first law.However, employing the same principles as in thisanalysis that diffusion occurs by the mechanism of
atoms jumping between sites and must therefore
be fully describable in these terms, such extra
complications should also be addressable: the
complicated physics is �wrapped up� in the jumpfrequency, C, and thus in the diffusion coefficient,D. This of course assumes one uses a definition ofD based on the mechanism of diffusion, ratherthan defining it as the coefficient in the traditional
formulation of Fick�s first law.
References
[1] Howe AA. Scripta Mater 2002;47(10):663.
[2] Erdelyi Z, Beke DL. Scripta Mater, this issue.
[3] Philibert J. Atom movements: diffusion and mass transport
in solids. Les Ulis: Les Editions de Physique; 1991.
[4] Christian J. The theory of transformations in metals and
alloys. Oxford: Pergamon Press; 1965.
[5] Manning JR. Diffusion kinetics for atoms in crystals.
Princeton: Van Nostrand; 1968.
[6] Darken L, Gurry R. Physical chemistry of metals. Mc-
Graw-Hill; 1953.
[7] Birchenall CE, Mehl RF. Trans AIME 1947;171:143.
[8] LeClaire AD. Prog Met Phys 1949;1:306.
[9] Smith RP. Acta Met 1953;1:578.
[10] Guy AG. ASM 1952;44:382 (Part I) and 397 (Part II).
[11] Krishtal MA. Diffusion processes in iron alloys. Jerusalem,
Israel Program for Scientific Translations; 1970. p. 97.
[12] Wells C, Batz W, Mehl R. Trans AIME 1950;188:553.
[13] Saunders N, Li X, Miodownik AP, Schille J-P. In: Shao
J-C et al., editors. Materials design approaches and experi-
ences. Warrendale, PA: TMS; 2001. p. 185.
[14] Shewmon P. Diffusion in solids. Warrendale, PA: TMS;
1989.
[15] Borg RJ, Dienes GJ. An introduction to solid state
diffusion. London: Academic Press; 1988.