Vous êtes sur la page 1sur 13



Division of Physical Metallurgy. Royal Institute of Technology. Stockholm 70. Sweden

(Rrceired 25 Srpternbrr 19753

Abstract-A previous treatment of the solute drag effect on the movement of grain boundaries and
phase interfaces, which was based upon the evaluation of the free-energy dissipation. is developed
further. The new treatment becomes identical to the treatments by Cahn and Liicke and Stiiwe for
grain boundaries in dilute solutions but is not limited to low solute contents. to ideal solutions or
to single-phase systems.
Some numerical calculations for grain boundaries are presented and the dissipation of free energy
in different zones of the interface is discussed. The maximum solute drag. calculated with a variable
diffusivity. is always lower than the value found by using any average value of the diffusivity. The
simple expression for the estimation of the solute drag. previously given, is approximately correct
at constant diffusivity. only. A number of calcultions for diffusionless phase transformations are pre-
sented and allow some conclusions regarding the spontaneous start of such a transformation.

R&sum&-On continue de developper un traitement anterieur de I’effet du freinage par le solute du

mouvement des joints de grains et des interfaces entre phases, traitement base sur l’evaluation de
I’inergie libre dissipee. Ce nouveau traitement est identique B ceux de Cahn, et de Liicke et St&e
pour les joints de grains dans les solutions diluees. mais il n’est limit& ni aux faibles teneurs en solute.
ni aux solutions idtales. ni aux systemes monophases.
On presente des calculs numeriques pour les joints de g,rains, et on discute la dissipation de l’tnergie
libre dans diffirentes zones de l’interface. Le freinage maxtmal par le solute, calcult avec une diffusivite
variable, est toujours infirieur a la valeur que l’on trouve en utilisant n’importe quelle valeur moyenne
de la diffusivite. L’expression simple que l’on avait donnee antirieurement pour estimer le freinage
n’est & peu prts correcte que dans le cas dune diffusivitt constante. On presente pour les transforma-
tions de phases sans diffusion, des calculs qui permettent de tirer quelques conclusions concemant
le debut spontane.

Zusammenfassung-Es wird eine friihere Behandlung des Losungseinflusses auf die Bewegung van
Korngrenzen und Phasengrenzflachen, die auf der Bestimmung der Dissipation freier Energie autbaute.
weiter entwickelt. Die neue Behandlung wird identisch mit denjenigen von Cahn und Lticke und Stiitve
fur Komgrenzen in verdtinnten LGsungen. ist aber nicht auf niedrige Losungsgehalte, auf ideale
Liisungen oder auf einphasige Systeme beschriinkt.
Einige numerische Rechnungen fiir Komgrenzen werden vorgelegt; die Dissipation der freien Energie
in verschiedenen Zonen der Grenzfllche wird diskutiert. Die maximale losungsinduzierte Behinderung.
berechnet mit einer variablen Diffusivitlt. ist immer niedriger als der Wert, der mit irgendeinem mitt-
leren Wert fiir die Diffusivitlt erhalten wird. Der einfache, friiher angegebene Ausdruck zur Abschltzung
des Lijsungseinflusses ist annlhernd richtig nur bei konstanter Diffusivitlt. Vorgelegt wird eine Anzahl
van Rechnungen iiber diffusionslose Phasentransformationen, welche einige Schltisse im Hinblick auf
den spontanen Einsatz einer solchen Tranformation gestatten.

IFXRODUCTION in the interior of a crystal. Liicke and Detert [33

developed the model in more detail. They supposed
In their studies of the effect of alloy additions on that foreign atoms, segregated to a boundary. will be
the recrystallization of aluminium, Liicke et a/. [l, 21 left behind if the boundary starts to move. The
found that the rate of recrystallization could decrease boundary will then exert an attraction force on them
by many orders of magnitude without exceeding the which makes them diffuse in the same direction as
solubility limit for the alloy addition. They concluded the boundary moves. The average drift velocity of
that the effect was caused by the presence of the atoms can be obtained from
foreign atoms in solid solution and was due to a
direct interaction between moving grain boundaries
C = $.r: (1)
and the foreign atoms in solution. They suggested
that the effect was related to a tendency of the foreign
atoms to segregate to the grain boundaires and they where f‘is the average value of the attraction force.
attempted to calculate this tendency, assuming that The movement of the boundary will be held back
a foreign atom is attracted to a grain boundary by the same attraction force. The solute drag on the
because it will there cause less elastic stresses than boundary. P. can thus be evaluated as the sum of

the attraction force on all the foreign atoms lagging for intermediate velocities. A characteristic feature of
behind the migrating boundary. the results was that the solute drag reached a maxi-
mum at an intermediate velocity and decreased
P= n*f (2)
towards zero at high velocities.
n is the number of segregated, foreign atoms per area A different approach to the solute drag problem
of the boundary. Liicke and Detert did not estimate was taken by Hillert [7] who argued that the work
how the number of foreign atoms per area depends put into the movement of a boundary in order to
upon the velocity of the boundary but pointed out overcome the solute drag, must dissipate by the diffu-
that it should fall drastically if the velocity is larger sion of the solute taking place as a result of the
than the drift velocity of atoms obtained from equa- boundary movement. It should thus be possible to
tion (1) when the largest possible value off is inserted. evaluate P as AGAY,, where AG, is the dissipation
For lower velocities they eliminated f by combining of free energy due to diffusion when the boundary
the two equations and thus they obtained a relation passes through a volume containing one mole of
between the solute drag and the velocity for the case atoms.
where the velocity is low enough to permit the major- In principle, both treatments could apply to any
ity of the segregated foreign atoms to diffuse with type of function describing the variation of E through
the boundary, the boundary but. quite naturally, the authors chose
functions which would simplify their own calcula-
P=-. tions. As a consequence, Cahn [4] and Liicke and
Stiiwe [5] chose a wedge-shaped function which gave
Cahn [4] and Llicke and Stliwe [5,6] developed a constant value of dE/dy which changed sign in the
the model further by considering how the interaction middle of the boundary. On the other hand, Hillert
energy between the foreign atoms and the boundary, [73 chose a square well function yielding constant
E, varies with the distance y of an atom from the properties in the boundary and abrupt changes at the
middle of the boundary. The interaction force is two sides of the boundary. It is interesting that Hillert
-dE/dy and the total drag is obtained by integrating did not find a maximum of the solute drag but a
over all the atoms. Llicke and Stiiwe thus obtained monotonous increase towards an asymtotic value at
the following equation increasing velocities. He then proposed that one
should take into account the atomistic nature of the
x dE
p= - -.--‘dy (4) system and showed that one could thus obtain the
K dy same type of variation with velocity for the square
well function that Cahn and Liicke and Stiiwe had
V, is the molar volume and x is the mole fraction
found for the wedge-shaped function without consi-
of the solute.
dering the atomic nature.
Cahn chose to add a term and wrote
Recently, Liicke and Stiiwe [6] have developed an
ccx-x’ dE atomistic model of grain boundary mobility in a bi-
p= - --‘--‘dy. (5) nary ahoy, which yields results similar to their pre-
f -‘xl v, dy vious treatment.
This is permitted for a single-phase material where The purpose of the present work was to develop
Hillert’s treatment further. It will be demonstrated
that it reduces to the treatment of Cahn [4] and
dE = E(x) - E( - cx) = 0, (6) Liicke and Stiiwe [5] for dilute solutions. It will be
applied to various cases of general interest including
x0 is the constant composition of the material s&Ii- phase transformations. Although there have been
ciently far away from the boundary. several suggestions of a solute drag effect in phase
The direction of the force depends upon the sign transformations [7-lo] there have previously been no
of dE/dy. Forces acting in opposite directions will serious attempts to carry out detailed calculations.
exactly cancel in a stationary boundary. For a moving
boundary, the distribution of the solute will be
changed and there will be a net force. In order to
evaluate the net force, one must first calculate how
the solute mole fraction, X, varies across the bound-
ary. For any rate of migration, this can in principle The treatment will be limited to binary systems
be done from information on how E varies across where the two components have the same mobility.
the boundary. Using a diffusion equation, related to The driving force for interdiffusion is d(Ge - G,)/dy
equation (I), Cahn [4] and Lticke and Stiiwe [S, 6] and the fluxes can be obtained from the absolute reac-
carried out this calculation for low solute contents tion rate theory,
and Cahn was able to obtain analytical expressions
for the solute drag at very low and very high veloci- D 4% - G.,)
-JA=JB=-- . (7)
ties. He also carried out some numericaf calculations RTV, XAXB dJ

It is well known that the rate of the free energy dis- xg is an integration constant which can be identified
sipation due to diffusion is given by the product of the with the alloy composition far away from the bound-
flux and the driving force ary because the diffusion Rux must be zero there. By
using equation (16), two additional ways of expressing
dG - J .d(Gs - GJdI,,
-=- B (g) the solute drag are obtained.
dt JY d)
m xg- x: d(GB - G,)
This equation will be applied to an infinitely long
volume element with a cross section A, which is paral-
lel to a boundary, migrating with a constant velocity
p= -
--r dj
*dy (12d)

v. Let _rbe the distance of any point from the bound- (17e)
ary and let us calculate the free energy dissipation
in the volume element during a time period At when
the boundary passes 1 mole of material. We shall It is evident. particularly from equations I 12b, c and
denote this free energy by AG, e), that the integrand is everywhere positive which is
natural since it represents the dissipation of free
At = V,I’Ac energy by diffusion.
In order to carry out numerical calculations one
dV= Adq (10) must choose a thermodynamic model for the system.
*z For instance, Cahn [4] considered a dilute solution
J B.d(Gs - GJdJ
AG, = (dGjdt)-At = - v- for which he wrote, according to Henry’s law
c s --D d!: ’
(11) GB = E(y) + RTlns,. (17)
According to Hillert’s suggestion, the solute drag is The dependency on position and composition were
evaluated by dividing this AG, with V,. By the use of thus supposed to come through separate terms. We
equation (7) the result can be expressed in three should prefer to denote E(y) by ‘GB(j.) because it is
different ways. a free energy rather than an internal energy. For the
other component one obtains from Raoult’r law.
G,, = ‘GA(y) + RTlns,,. (18)
P= For a dilute or ideal solution one thus obtains

d(GB - G,) d(‘Gs - ‘GA) + RT.5. (19)

P= dy = dy xAxg dy
Let us denote ‘Gs - ‘GA by A’G. By inserting equa-
It should be noticed that Hillert’s suggestion can hold tion (19) into (12d) the following expression is obtained
only in a stationary state. If there is a change of the for the solute drag,
concentration profile during a transient period. the
corresponding increase of the free energy of the sys-
tem must also be taken into account.
For a stationary state, one can relate the local value
p= -
s = xg - x; dA”G
--r v,’

because the contribution from the second term in

of the flux to the local composition because of the
following relation equation (19),

-= ?x, (21)
-C-;-. (13)
vanishes when the composition has the same value
If there is no side-ways diffusion, Ficks’ second law in the two limits of integration and this is true in
yields for the interdiffusion of A and B, the case under consideration. It should be noticed
--=sx, tx, gB= 3 that this term vanishes not only for a dilute or ideal
-= _v
(14) solution but for all cases where the dependency of
St 2t m ;L’ sy *
Gs - G,, on position and composition come through
We thus have separate terms.
Although the term defined by equation (21)
dJA dJB v
--_=-=_._ vanishes, its integrand is generally not zero but has
dx, dxB V,
a positive or negative value depending upon the sign
or by integration of .K~- .Y:. As a consequence. the remaining inte-
grand in equation (20) is not necessarily positive
everywhere in spite of the great similarity- with equa-
-J, = JB = ;(x, - x;) = - ; (xa - xi). (16)
tion (12d). It is thus evident that the integrand in
m m

equation (10) does not represent the dissipation of tude of the solute drag because the integrand is every-
free energy. Instead it seems to represent the local where positive. (-I) It provides information on where
contributions to the drag because it is identical to the free energy dissipates in the boundary.
equation (5) which was postulated by Cahn C-t].
except that it contains the difference A”G =
Or& - ‘G,, instead of E.
As a further demonstration of the fact that the new For the present purposes it is sufficient to apply
treatment becomes identical with the treatment by the regular solution model.
Cahn [SJ and Liicke and Stiiwe [S] for dilute solu-
tions. one may compare the expressions obtained by G_ = x:G, + $Gs + RT[x, In xA + x6 In xa]
Cahn for the limiting case of very high and very low + .YAq#L C6)
velocities with the predictions obtained for the same
and ‘G,,, ‘GB and L will be assumed to vary with
cases from the present treatment. At high velocities
the position relative to the boundary. _r. It yields the
it is convenient to apply equation (12b) and one may
following expression for the driving force for inter-
neglect the movement of the atoms when evaluating
d(Gs - GA)/dy. Since the composition is then every-
where close to .$ one finds d.xs/dy 2 0 and any d(Ge - G,,) dA”G dL
= - + (X, - x,) -d-j
thermodynamic model yields d.t d,

d(Gii - G,) dA”G +(.J&2L)2. (‘7)

=dp. (22)
By combination with equations (7 and 16) one
Equation (12b) can thus be approximated by the obtains the following expression which can be used
following equation for high velocities for calculations of the concentration profile

ha dA”G dL RTr x8 - .Y;
-= - - - @?I - .$i) -dr’ - _.D -
dv d_t X,&g

This is identical to Cahn’s equation for the same case [-&-Xl. (28)
but it is now evident that it is not limited to dilute
solutions but has a much wider application. The solute drag can then be calculated most easily
For low velocities it is convenient to apply equa- from equation (12e). In the general case, this calcula-
tion (12~).xg can be approximated by the equilibrium tion can only be solved numerically. The numerical
distribution which is governed by d(G, - G,)/dy = 0 calculation must start at the point behind the migrat-
in view of equation (7). The thermodynamic model ing boundary where G, is supposed to start to vary
defined by equations (17 and 18) yields with J and the composition must be xi up to that
point. The change in composition is then calculated
step by step in the p-direction using equation (28)
and the integrand in equation (12e) is evaluated at
the same time. This procedure was programmed for
By noting that xe - .x: is equal to xgxi - xAxg 0 one computer calculations and Fig. 1 gives a typical
can insert equation (24) into (12~). We thus obtain example. The L parameter was here assumed to be
for very low velocities zero everywhere and the diffusion coefficient D was
RTc r 1 assumed to be a constant. The top picture of Fig.
P = ----.XI:Y; 1 shows the A”G profile through the boundary. It
V” . I _,D
has a constant value in the two grains. A’-‘G’, and
-[exp A”Gi2R7 - exp - A”G/2R7]’ dy. (25) a lower constant value in the central zone 2 of the
This is identical to Cahn’s equation for the same case. boundary. A’G’. It varies linearly in the two side
Cahn pointed out that equation (25) does not depend zones of the boundary. zones 1 and 3. The middle
upon the sign of A”G and it thus predicts the same picture shows that the solute distribution at a low
drag for solutes which avoid the boundary as for rate of migration. curve 1. is close to the equilibrium
solutes which segregate to the boundary. segregation. As the rate is increased. the total amount
The present treatment has several advantages. (1) of segregation is gradually lowered, curves 2 and 3.
It can easily be applied to the whole range of compo- and a negative spike is developed in the grain in front
sitions in a binary system and it can handle any of the boundary, zone 4. This spike goes through a
model for the thermodynamic properties of the sys- maximum at some rate. curve 2. and then decreases,
tem. (2) It can be applied to phase transformations curve 3. The bottom picture shows the integrand in
as well as to the migration of a grain boundary in the calculation of the drag. dPidy. as evaluated from
a single-phase material. (3) It allows comparatively (sg - s~)~/.Y,,s~ according to equation (12e) at con-
accurate numerical calculations of the total magni- stant D. In order to compare the different curves, one

should multiply them with their migration rates. c. RTm

As a consequence. the low-velocity curve, curve 1. dA’G!d:
represents a very small drag and it is evident that
the drag originates in the central zone of the bound- cd 1 dA”G ’ 4cb dA”G
ary. As the velocity increases. the contributions from
m= of-- RT d-_ ) --DRT d= .” 1

the two side zones of the boundary and from the

c isthe integration constant. This expression of xB
spike in the grain ahead of the boundary grow in
can be inserted into equation (12e) yielding the solute
importance. curve 2. At even higher velocities. the drag. The integration of equation (12e) must be car-
contribution from the spike decreases and only the ried out in several steps where each step is limited
two side zones are important. These results are equiv- to a region where D and dA’G;d: have constant
alent to those reported by Cahn. values. The model defined at the top of Fig. 1 has
In the following discussions diagrams will be pre- four such regions. The expression thus obtained for
sented. showing the variation of the drag with the the solute drag is
rate of migration or the alloy composition. From the
practical point of view, it is then impossible to present pv, rdRT (.~;)2
curves for dP/dy. Instead, the diagrams will show the RT=- D 1dA’G. d-1 a(a + b) .K;
total drag obtained by integration of dP/dy over the
(.$)2 ‘X*
whole boundary and in some cases the integrated
contributions from each one of the four zones defined - (1 - a)(1 - n - b)In x
at the top of Fig. 1. It should be noticed that the (a - .u$’ ln I.yB - a
-___ _
contribution from the region lying in front of the ab(1 - n) I xg - a I
boundary. zone 4, can be obtained by direct calcula-
tion of .$G,,(z ) - G,,(6,‘2)] + xg[GB(x ) - Gs(&‘7)] (a + b - xi)’
as soon as the composition at _V= 6: 2 has been found. + b(o + b)(l - a - b)
With the regular solution model one obtains the fol- 2 5
lowing expression for this contribution + xi In + + xi In -B
I.4 ‘?cB1

- -&x, - xi)*, (29) (.YY ln &

- (1 - a’)( 1 - a’ - b’) l.xA
where ‘s,,. 3.~;8denotes the composition at y = 6/2,
i.e. at the point of discontinuity on the right hand
side of the region concerned. The same procedure
(a’ -+ b’ - x;)~ 3x* - n’ - b’
could be used for any part of the boundary where In
the properties stay constant. for instance in the cen- + b’(a’ + b’)(l - a’ - b’) zxg - a’ - b II
tral zone of the boundary described in Fig. 1. This 0
is how Hillert [7] calculated the dissipation of free + x(:ln++xgln- , (33)
-*A 5g I
energy and why he chose to consider a boundary with
uniform properties. The four terms on the right-hand side originate from
The calculation can be carried out analytically un- the four regions. The first and third terms originate
der some other conditions as well. A comparatively from the two sides where dA’G/dc is not zero. The
simple case is found then D and dA’G;dy are constant quantities a, a’, b and b’ are all defined by equations
and L = 0. Equation (28) can then be rewritten in (32a and b), respectively, and have different values
the following form by introducing a dimensionless because the values of dA’G/d: have different signs.
coordinate : = y/6 where d is the total width of the ‘XB, ‘x8 and 3.~ygare the concentrations of solute at
boundary. the points of discontinuity between regions 1 and 2,
regions 2 and 3 and regions 3 and 4. respectively.
dx, 1 dA”G c6 Their values are
-= ------_ ,y - jy (xe - xi,. (30)
d: RT dz .A ’

This equation has the solution

xB = n + b[l -t cexp(4: - z~))]-*, (31) z~B = xg+ (lx* - xg)exp (34’4
‘x,-o’+b’[l +(&-I)exp(:)]-i,
a=_ -+ ----_ (32a)


-0.005 j- Migration rate, as expkg)

Fig. 2. The solute drag as function of the migration rate.

L’,calculated for three grain boundary models. The quanti-
ties h and S are defined in Fig. 1.

that the solute drag has a maximum at a velocity

which moves to the infinity if one makes the disconti-
nuity sharp by setting b = 0. It is thus demonstrated
that the absence of a maximum in Hillert’s first calcu-
Fig. 1. Top diagram: A grain boundary model with four
different zones. Middle diagram: Concentration profiles lation was due to his choice of sharp discontinuities.
calculated for three migration rates. Bottom diagram: the Hillert obtained a maximum by introducing the
integrand in the calculation of solute drag from the con- atomistic nature of the system. It may be stated that
centration profiles. the model chosen by Cahn and Liicke and Stiiwe re-
flects the atomistic nature of the system by the width
The second and fourth terms are simpler because of the discontinuity, b. It thus appears essential never
they originate from regions where dA’G/d-_ = 0. They to give b a value which is smaller than the atomic
can be compared with equation (29). dimensions. As a consequence. h will be chosen as
A similar solution can be obtained with the regular
solution model if L is so small that (1 - 2Lx,xB/
RT)- 1 can be approximated by 1 + 2Lx,xB/RT:



(a) Comparison between precious treatments of grain

We shall first use the new treatment for a compari-
son between the previous treatments. This can be
done by representing the properties of the grain
boundary with the mode1 described at the top of Fig.
1 and using various b values. b is the width of the
discontinuities and Cahn [4] and Liicke and Stiiwe
[S] treated the caSe b = b/2 where there is no homo-
geneous region in the middle of the boundary. Hillert
[7] treated the case b = 0 where the discontinuities
are sharp. Figure 2 shows the results obtained with
b = S/2, S/4 and 6/10000 and with the regular solu-
tion parameter L = 0. The calculations were carried
out for a dilute solution, low .x8, and the results are
then proportional to xi . x$ The results were plotted
Alloy composition, xi
vs the velocity expressed as exp( - c6/2D) in order to
cover the whole u range from zero to infinity and Fig. 3. The solute drag as function of the alloy composi-
to make the curves at low velocities approximate the tion. calculated for different migration rates. c. All the
curves obtained in a plot vs r. Figure 2 demonstrates curves are symmetric.

(a) thus appears that the total drag is independent of

the sign of A”G at all velocities. One can thus obtain
the same result with positive segregation of the minor
element to the boundary and \vith negative segrega-
tion. It should be noticed that the diagram shows
the solute drag divided by s”,.$. For high and low
velocities. the solute drag is proportional to .Y~.$.
according to equations (23 and 25). The shapes of
the curves for intermediate velocities show that the
solute drag then increases less rapidly than X:X; but
the maximum drag is still obtained at the central
Figures ?(a and b) show some results from the same
calculation. but now plotted for constant composition
and versus the velocity. One composition u-as chosen
at each side of the alloy system and the curves for
Migration rate, as log(W2D)
the total solute drag are identical. However. curves
representing the dissipation of free energy in the four
regions defined in Fig. I sho\v large differences. The
free energy dissipation in the central zone 2 of the
boundary is most important uhen the solute is
attracted to the boundary. Fig. J(a) but the dissipa-
tion in front of the boundary. zone 4. is most impor-
tant if the solute is trying to avoid the boundary.
(Fig. 4b). Equation (23) indicates that the side regions
where A”G varies. zone I and 3. should dominate
at higher velocities and, since the sign of A”G is here
of no importance. the contributions from zone I and
3 should here be equal. This is confirmed by both
the diagrams.
All the calculations presented so far were obtained
under the assumption of a constant value of the diffu-
sivity. D. Figure 5 shows the effect of a variation of
-3 -2 -1 0 1 2 3
Migration rate, as log(v5120)

Fig. 4. The solute drag contribution from different zones

in the grain boundary. The zones are defined in Fig. I.
The relative importance of the four zones varies with the
migration rate. (a) is for a case with a positive segregation
to the boundary. (b) is for a case with negative segregation.
The total solute drag is the same in the two cases but
the roles of the four zones are different.

d/4 in most of the following calculations since the

width of a boundary. 6. is usually assumed to be a
few atomic distances.

(b) Examirlation of‘ the solute drag ocer the complete

range of cornpositiotl
Figure 3 shows the size of the solute drag over
the complete range of composition, each curve repre-
senting a constant velocity. The only difference in the
calculation for solutions of B in A and of A in B
is the difference in sign between (‘GB - ‘GA) and
(‘GA - ‘GB) and it is evident from equations (23 and
25) that the sign of A”G does not affect the total mag- 1 0.5 0
nitude of the solute drag at high and low velocities. Migrotion rote, 0s rxp(-v8/20ij
One should thus expect to find symmetric curves for Fig. 5. The solute drag as function of the migrarion rate.
high and low velocities. In fact, the diagram seems calculated for a variable diffusivity. D’ holds in rh: center
to indicate that all the curves are symmetric and it of the boundary and D holds in the undisturbed crystals.

-L -3 -2 -1 0 1 2 3
-3 -2 -1 0 I 2 3 4
Migration rate. as 10g(v612Di)
Migration rate. as bg(v6nOJ

(d I I I I
am: opproxmate
I NequaGm
x *

z a010 -


-9 -9 -7 -6 -5 -4 -3 -2 -1 0 1 2 3
Migration rat., as log(v8/2oi

Fig. 6. (a-c) The solute drag contributions from different zones in the grain boundary. evaluated for
three of the curves in Fig. 5. The dashed curves are calculated from an approximate equation (Refs.
4 and 5).

D. It was assumed to have different values in the crys- In order to examine the reasons for these changes,
tals. D, and in the central region of the boundary. it is advantageous to plot the solute drag and its four
D’. and it was further assumed that log D varies components versus the velocity on a logarithmic
linearly in the intermediate regions. These calcula- scale. Figures 6(a-c) show the results for Di/Dy = 1,
tions were only carried out for a dilute solution. IO3 and 10’. It is evident that the maximum solute
The most interesting feature of Fig. 5 is that the drag occurs at a velocity which is closely related to
malimum solute drag decreases with increasing D’ but it is important to note that a considerable
values of D’iD;. Previous suggestions [3. 6, 1I] that drag is obtained much earlier for the case D’/D: = lo8
one could employ some average value of D are and there is an almost constant plateau for 4 orders
not justified. The position of the maximum is also of magnitude of the velocity.
interesting. At D’ = Dy it occurs at about Cahn as well as Liicke and Stiiwe constructed a
L’= 4O”d , = -ID;j& At increasing values of D’ly the simple interpolation formula by combining the asym-
maximum quite naturally moves to values somewhat totic expressions for low and high velocities. equa-
smaller than JD’;‘6 but increasingly larger than 40:/B. tions (23 and 25). The predictions obtained from this

because there is no tendency for segregation at that

composition. This fact is described by the term
(s,, - .u,)dL dy in equation (35).



(a) 7-17~ t$ct of‘ the fr77dtv7cy of srywprion co chr


The mathematical treatment in the present paper

is limited to cases where there is no side-wa!s dil%-
sion. As a consequence. the equations can be rigor-
ously applied to phase transformations only if the
grooving phase /? has the same composition .Y: as the
parent phase x had initially. In its present form. the
theory can thus be used to predict when such a diffu-
sionless transformation is thermodynamically poss-
ible. We shall first examine the simple case where
D is a constant and the solute content is low. The
distribution constant for the solute between the two
phases was chosen to be Ki” = 4 which corresponds
to A’G’ - A”G” = I.4 RT The tendency of the solute
to segregate to the central zone of the phase interface
leas varied from &,I’ = 1’1’ to 2. i.e. Ki’ = I’J-I?.
This corresponds to values of A’G’ - 1”G:; from
0.1 02 a3 04 a5 as 0.7 ae 09 1.0 - 2.15 RF+ I.1 RT and a linear variation was
Alloy composition. xi

Fig. 7. The solute drag as function of allo; composition

calculated for different migration rates usmg a regular-
solution description of boundary segregation.

approximate equation are presented as dashed lines

in Figs. 6ja-c). Cahn tested the equation for constant K$i 12 or l/3
D and this test is repreated in Fig. 6(a). It shows a
very good agreement. except that the maximum is
somewhat too high. Liicke and Stiiwe [5] tested the /
equation for very high values of A”G, - A’G’ and
found that it is less satisfactory there. Figs. 6b and
c show it cannot at all be used to describe cases where
the diffusivity varies across the boundary.

(c) Thr rf/rct of a0 i77rrractio,7 poramrtrr

Most physical models of grain boundary segrega-

tion predict that a solute should segregate to the
boundary because the internal stresses will be less if
the dissimilar atoms are mixed in a region of disor-
dered atomic arrangement. Such a situation cannot
be adequately described by allowing (‘GB - ‘G,) to
vary but. at least to a first approximation, it should
be possible to describe the situation by giving the
L parameter negative values in the boundary. Figure
7 shows the result obtained with ,E = 0 in the two
grains, ,L!= -0.7 RTin the central zone of the bound- Migration rate. as cxp(-W2D)
ary and a linear variation in the two side zones. A
constant value of A”G was employed and equation Fig. S. The free energy dissipation due to diffusion in a
(28) demonstrates that the effect of A in B and of so-called diffusionless phase transformation, espressed as
a solute drag. The K values represent the equilibrium dis-
B in A should then be the same. As a consequence, tribution of the solute. Some combinations of K values
all the curves in Fig. 7 are symmetric. They all go yield a monotonously decreasing drag. other combinations
to zero at the central composition where s: = .$ yield a maximum at some migration rate.

the value of Kg6 because, at a low enough velocity,

equilibrium should be established between the two
phases. In order for the growing fi phase to obtain
the initial solute content .$. the y: phase must then
have a solute content of Kid ..K: in front of the inter-
face. There will thus be a pile-up of solute atoms
which are being pushed forward in front of the
migrating interface. The height of the starting point
for all the curves corresponds to the free energy loss
due to that diffusion, which can be directly calculated
3i -4 -3 -2 -1 0 1 2 3 from
.x;[G&;) - GA(KzB . x;)]
(a) Migration rate, as log(v6120) + x;[G&) - Gs(KS’p. x;)].

Figure 8 shows that the solute drag may increase

or decrease with increasing velocity. The largest
decrease is obtained with Ki8 = ,/KiE, i.e. when the
value of A”G in the center of the interface lies at
the average between the values of the two phases.
Figure 9(a) gives the four components of the solute
drag for this case and it is evident that the decrease
of the total drag is due to a decrease of the height
of the pile-up ahead of the migrating interface, zone
Figure 8 shows that the solute drag is affected in
exactly the same way if the value of A”G in the inter-
face is decreased or increased by a certain amount
from the average value. As mentioned in a previous
o- section, the same result was obtained for grain boun-
-L -3 -2 -1 0 1 2 3 daries where the two grains are of the same phase.
Furthermore, Fig. 8 shows that when A”G in the
(b) Migration rate. as log(vf5120)
interface differs too much from the average, there will
2.5 - I I I I
be an initial increase of the solute drag and a maxi-
p4 =L
mum will be reached at some critical velocity. Figures
.ax Total B 9(b and c) show the four components for the cases
of Kin = 6 and 213. The diagrams can be com-
pared with the corresponding diagrams for grain
boundaries (Figs. 4a and b). For Kg” = 6 Figure 9(b)
shows how a strong drag originates in the central
zone 2 of the boundary, as the drag due to the pile-up
ahead of the boundary, zone 4, decreases. The same
6 1.0 - total drag is obtained with Ki” = 2i3 and Fig. 9(c)
‘D shows that the maximum is here due to an initial
increase of the drag due to the pile-up.
It could be argued that the free energy dissipation
in the pile-up of solute atoms in the parent phase
ahead of the advancing interface is not caused by the
-4 -3 -2 -1 0 1 2 3 presence of the interface itself and should thus be
regarded as separate from the solute drag which by
Cc) Migration rate, as log(vb/2Di)
definition is caused by the interface. However, there
Fig. 9. The contributions from different zones in the phase is no practical way of keeping such a distinction as
interface. evaluated for three of the curves in Fig. 8. (a) the velocity is increased and the pile-up is affected
represents a case with monotonously decreasing drag. (b) by what goes on inside the interface.
and (c) show two cases with a maximum. The total drag From the thermodynamic point of view, the ques-
has the same shape but the maximum is due to different
zones. tion whether the diffusionless transformation can
occur or not is a question of whether the driving force
assumed at the two sides of the interface. Compare for the diffusionless transformation is large enough.
the top diagram in Fig. 1. The results are shown in This driving force can be calculated from
Fig. 8. All the curves start from a common point at
a positive value, which can be easily calculated from

phase. Figure S now indicates that a diffusionless


0,” 2.0 transformation is then possible only if Kk” lies
-2 between the values of approx. 1 and Kiti. The solute
rfl\. x; =a01
2 , 5 startmg pant for drag will then decrease at increasing growth rates and
all the curves the growth rate may thus attain high values where
>E .
o- a considerable part of the driving force becomes avail-
able to overcome the physical friction in the interface.
The resulting growth rate cannot be calculated with-
out information on the mobility of the interface.
For KiU values lower than about 1 or higher than
about K’,O. a higher driving force is necessary and
the initial solute content must lie within the single-
phase region for the new phase in order for a diffu-
Migration rate, as cxp(-v6/20i) sionless transformation to occur. A growth rate
limited by the solute drag is then obtained if the driv-
Fig. 10. The free energy dissipation in a phase transforma- ing force is not larger than the maximum on the
tion, calculated for a case with high diffusivity in the solute drag curve.
boundary. It should further be mentioned that it is in principle
possible for growth to occur below and to the right
It has been shown [lZ, 73 that the driving force for of the maximum point once such a situation has been
the diffusionless transformation corresponds exactly established but some special mechanism may be
to the height of the starting point in Fig. 8 if the required in order to pass the maximum. The same
initial solute content is exactly equal to the content problem has been discussed for grain boundaries by
of the new phase at equilibrium with the parent Cahn [4].

20 I I I I I I I 2.0 , , , , ,
i‘ W?,‘

-1 -6 -6 4 -3 -2 -1 0 I 2 3 -7 -6 -5 -L -3 -2 -1 0 1 2 3

Migration rate. as logfvb/20il

(a) Migration rate. as logIv6/2D') @I

_ o~Sdl~oi~2.ios
a;, .O.Ol

-7 -6 -5 -‘ -3 -2 -1 0 1 2 3
.\ -7 -6 -5 -L -3 -2 -1 0 I 2 3

Fig. 11. The contributions from different zones in the phase interface, evaluated for two of the curv-es
in Fig. 10. (a) and (b) show the same total drag, obtained with two different sets of K values. (c) and (d)
show another curve which has two maxima.

(b) The effect of a high diffusicity in the interjbce

The calculations of the preceding section were car-
ried out with a constant D value. However, it is often
supposed that the diffusivity is considerably higher
in an interface than in the crystals. The calculations
were thus repeated with D’ = 0” = D’?. lo6 and
with a linear variation of log D at the two sides of
the interface. The results are given in Fig. 10. Two
features are obvious. There is a drastic decrease of
the drag at a low velocity. The rest of the drag
remains almost constant until a velocity, related to L Total
D’/S. is reached. Figures ll(a and b) show how the
four components change with the velocity for the low- \
est curve in Fig. 10. In both cases the contribution
due to the pile-up in front of the boundary, zone 4, I I
decreases strongly above a velocity of about 40’/6. -5 4 -3 -2 -1 0 1 2 3
Instead a contribution develops in zone 3 and the
Migration rate. as loglv~/2DL)
total drag does not decrease strongly until this contri-
bution starts to decrease. The contributions from the Fig. 12. The solute drag evaluated from a model for solidi-
other zones are not very large and in Fig. 1l(b) they fication. The drag decreases strongly already at a migration
are even too small to be shown. rate. which is very low in comparison with the diffusivity
in the liquid.
Figures Il(c and d) show how the four contribu-
tions change with the velocity for the upper curve
in Fig. 10. This curve shows a maximum but Figs. appreciably below its value at low rates. A change
Il(c and d) reveal that it actually has two maxima. into a diffusionless growth may then take place spon-
One occurs at a velocity of about 800’/6 and is due taneously even if the total driving force for the diffu-
to the development of a very strong contribution in sionless transformation is less than the total drag at
zone 3 and some in zone 1 as the role of the pile-up low rates. This may provide the explanation why a
in zone -I decreases. The other maximum occurs at diffusionless solidification can occur in an alloy inside
a velocity of about 0.S Di/6 and is due to the develop- the solid + liquid two-phase field [13]. As already
ment of a contribution in the central zone 2 in the mentioned, the driving force for the diffusionless
boundary. transformation is lower than the solute drag for all
Figure I I(d) shows that the contribution from zone alloys inside the two-phase field and it is thus natural
3 decreases very slowly at the same velocities where to espect that a solid-solid diffusionless transforma-
zone 2 reaches its maximum. Not until the solute con- tion does not start until the alloy has been brought
centration in zone 2 comes close to its equilibrium inside the one-phase field for the new phase [IZ. 71.
value in the matrix. does the contribution from zone
3 decrease strongly again. SUMMARY
The further development of Hillert’s treatment of
(c) 77le effect of a high diffrrsicit_r in the initial phase the solute drag in binary alloys becomes identical to
In solidification, the diffusivity is much larger in the previous treatment by Cahn and Liicke and Stiiwe
the initial. liquid phase than in the growing, solid in the limit of dilute solutions. It has been pro-
phase and it seems reasonable to assume that the grammed for computerized calculations and a
properties vary monotonously through the interface. number of calculations have been carried out.
A calculation was carried out for such a case and For an ideal solution it is found that the total
a linear variation was assumed for A”G as well as solute drag is independent of whether the solute is
log D through the interface. A value of b = 612 was attracted or repelled by the boundary. This result was
thus chosen and there is no central zone 2 with con- obtained for all compositions and migration rates. On
stant properties. The results are shown in Fig. 12 and the other hand, the free energy dissipation takes place
it is particularly interesting to note that the contribu- in different zones of the boundary.
tions from the zones inside such a boundary are very The effect of a variable diffusivity can be very large
small. The contribution from the pile-up ahead of the and a similar result cannot be produced by using an
migrating interface. zone 4. dominates and the total average value for the diffusivity. A higher diffusivity
drag decreases when the pile-up starts to decrease at inside the interface will decrease the masimum value
a velocity of about Dl’/& On the other hand. the diffu- of the drag and make it rather constant over many
sion-controlled growth of a fi dendrite into the liquid orders of magnitude of velocity. By combining the
phase is governed by DL which is much larger than asymptotic behavior at very low and ver)- high veloci-
D”. It is thus quite possible that the growth rate of ties. Cahn and Liicke and Stiiwe constructed a rather
a dendrite becomes so large that the drag decreases simple formula covering the whole range of velocity.

This approximate equation is satisfactory at constant .~clirlo\~lrdyrmrnrs-Ths authors wish to thank professor
diffusivity. only. John Cahn for an interesting discussion and valuable com-
The new treatment can use a regular-solution
This research was supported b> the Swedish Board for
model in order to describe the case where the solute Technical Development.
atoms tend to segregate to the boundary because a
mixture of dissimilar atoms will cause less strain
energy there.
For diffusionless phase transformations it is found
that the free energy dissipation. which starts from an REFERESCES
appreciable value at low rates of transformation. in
some cases decrease monotonously as the rate is in- I. Liicke K.. Masing G. and Iriiilting P., Z. .\f<rcrl[k. 6-l
creased. In other cases the free energy dissipation goes
2. Detert K. and Liiske K.. The influence of defined small
through a maximum at an intermediate rate of trans- amounts of impurities on the recrystallization of alu-
formation. This can happen if a strong solute drag minium. Brown Univ. Report. AFOSR-TS-56-103;
develops inside the interface. when the pile-up of AD-S2016-March 1956.
solute atoms in front of the advancing interface de- Liickc K. and Dctert K.. .4cru .t[t,t. 5. 63 (1957).
Cahn J. W., rlcr~ .Ifrr. 10. 759 (1962).
creases at high rates of transformation. It can also
Lucks K. and Stiias H.. Rrcorery and Rrcr~~srJlirotiou
happen by an increase of the pile-up of solute atoms of‘ .\ft,rals. p. 131. Interscience. New York. f 1963).
in front of the interface. Liicke K. and Stiiwe H.. rlcta .\lrr. 19. IOS- (1971).
A model, used to describe a liquid-solid interface Hillcrt M.. Monograp and Rsport Series sa,. 33. p.
indicates that the free energy dissipation during diffu- 231. Inst. of Metals. (1969).
Rao M. M. and U’inchell P. G.. 7mns. .-tI.\[E 239,
sionless solidification decreases strongly already at a
956 (1967).
growth rate thich is related to the low diffusivity in Kinsman K. R. and Aaronson H. I., Transformations
the solid phase and not to the diffusivity in the liquid. and hardenability in steels. Climax Molybdenum
It is possible that such growth rates can be obtained Corp.. Ann Arbor. MI (1967).
by diffusion-controlled growth of dendrites and this IO. Aaronson H. I.. Laird C. and Kinsman K. R.. Phse
Trtl~lsfbrr,rcltiorls. .-\SM. Metals Park, OH (19-O).
result may thus explain why it is easy to obtain a 1I. Gordon P. and Vandermeer R. A.. Tmns. .-1I.\IE 224.
diffusionless solidification in an alloy inside the 917 (1962).
liquid + solid two-phase field. 12. Karlyn D.. Cahn J. W. and Cohen M.. Trans. .-l/.\IE
245. 191 (1969).
Note added in proof-Lticke et al. [I?, 151 have recently 13. Baker J. C. and Cahn J. W.. .-lcro :Lfrr. 17. 5-i (1969).
made further calculations on the atomistic model. using I-t. Rosenbaum F. W’..Doctor-Thesis. T. H. Aachen (1971).
discrete values of the properties at the atomic planes in 15. Liickc K., Rixen R. and Rosenbaum F. W.. in .Vcrrnre
the boundary and without any restriction regarding the nnd Behior qf‘ Grain Boundaries (Ed. Hsun Hu).
rate of change from one plane to the next. Plenum, New York (1972).