J. Phys. F: Met. Phys. 17(1987) 2163-2178.

Printed in the UK

Prediction of ordering in binary transition-metal alloys

M Sluiter, P Turchi? and D de Fontaine

Department of Materials Science and Mineral Engineering, University of California,
Berkeley, C A 94720, USA

Received 5 January 1987, in final form 15 April 1987

Abstract. A simple model is presented which can successfully predict ordering in binary
transition-metal alloys. Rectangular electronic densities of states (DOS) for both ordered and
disordered states are assumed, along with the invariance of the first four moments of the DOS
upon ordering. The ordering energy A U can then be calculated independently of the nature of
the crystal lattice. The ordering tendency, given by the sign and magnitude of AU, is plotted
in ordering maps as a function of concentration and d-band filling. Computed ordering maps
are found to be in excellent agreement with those obtained by much more elaborate coherent
potential calculations. The overwhelming majority of the ordered structures observed
experimentally in transition-metal alloys fall within the predicted ordering regions in these
maps.

1. Introduction

In the absence of complicating effects due to elastic interactions, it is generally possible to

predict ordering tendencies in solid solutions (alloys, for example) by considering electronic
contributions alone. The procedure is then to compute the electronic density of states (DOS)
n(E)for both ordered and disordered states, to fill the DOS up to the Fermi level E , and to
determine which state has the lowest energy

U = J E FE n ( E ) d E

consistent with the band filling

&IEF n(E)dE.

The DOS is difficult to evaluate for a given binary alloy, but some general qualitative
considerations may be stated, at least in the case of well separated sub-bands (one for
element A, one for B). Ducastelle (1984) has argued that the DOS of a phase-separated
alloy must resemble the average of that of the pure elements, since, as shown schematically
in figure l(a) for the equiatomic case, minimal mixing of A and B atoms occurs. In a
completely disordered alloy, the sub-bands are expected to be somewhat narrower since
the probability of finding large pure A or pure B regions will be low, and those are primarly
the ones that contribute to the lowest and highest energy states in the sub-bands. Thus, the
expected DOS will exhibit tails in the DOS near the band edges, as shown schematically in
t Present address: Department of Nuclear Chemistry, Lawrence Livermore National Laboratory, PO Box
808, Livermore, C A 94550, USA.

0305-4608/87/112163 + 16 $02.50 @ 1987 IOP Publishing Ltd 2163

2164 M Sluiter et a1

E: EB
Energy

Figure 1. Schematic densities of state for equiatomic alloys with the following
characteristics : ( a ) segregated (phase separation), (b) disordered, (e) ordered. In the latter
case the density of states for an ordered alloy with c < 0.5 is shown (from Ducastelle 1984).

figure l(b). In the ordered state, pure A and pure B regions are forbidden, so that narrow
bands will result, as shown schematically in figure l(c).
The foregoing pertained to nearly equiatomic compositions. However, as the average
concentration becomes increasingly asymmetric, the majority sub-band (corresponding to
the most plentiful element) will tend to broaden, while the minority sub-band will tend to
sharpen (see figure I(d)). This is because, in the limit, the majority sub-band must become
indistinguishable from the broad one of the pure element, and the minority sub-band, for a
single atom, say, must reduce to a delta function.
The general features of the DOS of binary alloys have been studied more quantitatively,
for example, in the tight-binding (TB) approximation framework, with the disordered state
DOS obtained by single-site coherent potential approximation (CPA) calculations (Bieber
and Gautier 198 la, b, Bieber et a1 1983, Turchi et a1 1983) compared with TB calculations
of perfectly ordered states (Turchi 1984). The electronic energy of ordered states can also
be calculated through the so called generalised perturbation method (GPM) of Gautier and
Ducastelle (Gautier et a1 1975, Ducastelle and Gautier 1976). In that method, effective
interaction parameters V , are calculated as coefficients in an expansion in configuration
variables obtained by perturbing the C P A medium. Calculations have been performed in the
TB-CPA-GPM framework (Bieber and Gautier 1981a, b, Bieber et a1 1983, Bieber and
Gautier 1986, Turchi et a1 1983, Turchi 1984, Gautier 1985) and the ground states of
ordering have been predicted correctly for a large number of transition-element binary
systems as a function of composition, of effective number of d electrons N A and NB in A
and in B elemental solids, respectively, and of the diagonal disorder 6, given by
6, = A / W (3)
where A is related to the difference cB- c A of atomic d levels and W is the average
bandwidth
@=(l-C>W, +CW, (4)
where c is the average concentration of constituent B, and WA and WB are the bandwidths
of the pure elements.
 Ordering in binary transition-metal alloq,s 2165

The purpose of the present communication is to show that ordering tendencies in

binary non-magnetic transition-metal alloys can also be predicted, in a qualitative manner,
on the basis of a very simple model that uses rectangular d bands instead of the calculated
DOS. The characteristics of each sub-band will be fixed by a small number of pure-element
parameters, as explained below.

2. Tight-binding model

A useful model, which has been explored in detail for transition-metal alloys, is the tight-
binding model limited to d-electron contributions alone (Ducastelle 1984, Turchi 1984,
Friedel 1969). The TB scheme is particularly convenient to use if non-diagonal disorder is
not taken into account, i.e., if the hopping integrals or the matrix elements linking two
neighbouring sites are assumed to be independent of the nature of the atoms, A or B,
occupying the two sites. It is also assumed that there is but one atom per unit cell. The kth
order moment of the DOS is defined as follows :

pk= 1 Ekn(E)dE (5)

the integral being taken over all d-band states. Under the stated assumptions, it then
follows, as shown by Ducastelle (1984) and Hodges (1979), that the lowest-order moment
that depends on configuration (i.e., on the distribution of A and B atoms on lattice sites) is
the fourth one, p 4 , since it contains path contributions including two on-site steps (the only
ones depending on the site occupation) and an intersite round trip. Consequently, moments
of the order of three and lower are independent of configuration. In other words, if Apk
designates the difference in kth moment of ordered and disordered DOS, hpk vanishes for
k=0,1,2,3 for fixed concentration and value of diagonal disorder parameter 8,. A theorem
may now be invoked (Dieudonne 1968) that states that any function where the first n
moments vanish within a certain interval must vanish at least n times inside that interval
(limits excluded). It follows that the difference A n ( E ) between ordered and disordered DOS
must vanish at least four times between lowest and highest energy states.
Furthermore, if the Fermi level is considered as a variable E , taking on approximately
equal values in the ordered and disordered states, then, by equation (l), the energy
difference or ordering energy
A U ( E ) U,(E) - U d ( E ) (6)
(0, ordered; d, disordered) can be shown to have moments Amk of order k proportional to
moments of order k + 2 of the DOS difference: Apkc2(Turchi 1984, Turchi and Ducastelle
1985). Hence, Am, and Am, must be equal to zero, and therefore A U ( E ) must vanish at
least twice between lowest and highest energy states. It follows that A U i s also expected to
vanish twice as a function of band filling R,

since, by equation (2), fi is seen to be a monotonically increasing function of the energy

level E.
The stated theorem has important consequences concerning the ordering tendencies of
transition-metal alloys, as explained elsewhere (Turchi 1984, Turchi and Ducastelle 1985,
Turchi and Ducastelle 1987). Here, we shall illustrate the general concepts by adopting the
simplest possible density of states made up of rectangular sub-bands. It will be shown that,
2166 M Sluiter et a1

somewhat surprisingly, these very crude DOS nevertheless describe, at least qualitatively,
general ordering trends in real systems.

3. Rectangular band model

The parameters required to define rectangular sub-bands, assumed to be well separated,

are the band edges E , and E, of band A and E , and E, of band B (see figure 2). The sub-
band widths are
WA =E, -E,
WB =E4-E3
and the band centres are

(9)
E , = ;(E3 + E4).
By convention, the A element is taken to be that which has the greater number of d
electrons. Hence, element A will be to the right of element B in the Periodic Table. Since,
for d elctrons, the free atom d level E decreases practically linearly with the effective d-
electron number N , it follows that, with NA> N B , Q E,, hence 6, >O, without loss of
generality. Likewise, we must have EA < E , , as shown in figure 2. Care must be taken not
to identify ci and Ei (i = A, B), as a sub-band centre of gravity ( E i )need not coincide with
the free atom level E , particularly if charge transfer is taken into account. For that reason,
the diagonal disorder parameter will be defined by equation (3), with A = E, - E A , rather
than E, - E ~ .
The number of states in a sub-band is determined by the concentration, so that, for
rectangular bands, we have
nA = lo( 1 - c ) = W A h~ nB= 10c=wBhB (10)

Figure 2. Illustration of the parameters pertaining to the rectangular band model used in this
paper to describe the alloy electronic properties.
 Ordering in binary transition-metal alloys 2167

where h, and h, are the band heights. The first four moments p, defined in equation (9,in
the rectangular band approximation, are given by
po = nA + n, = 10 (1 1)
pI =f[h,(E~-E~)+h,(E:-E$)] ( 12a)

In the present scheme, six parameters completely determine the DOS : the four sub-band
edges and the two sub-band heights. For a fixed concentration c, but arbitrary state of
order, the values of these six parameters can be determined by the values of the moments
po, p , , p 2 , p 3 , and by the value of nA (or ne) and of the diagonal disorder d d . Since these
moments must be identical in ordered and disordered states, and since is fixed by c
alone, the DOS for ordered and disordered states would be identical if A = E , -E, were
invariant on ordering. In order to generate different DOS for the ordered and disordered
states, it will therefore be necessary to assume a change of dd on ordering
Si = d d ( 1 - Ad) (13)
where the superscript refers to the ordered state and where Ad will be taken as a small
parameter to be varied systematically, in accordance with results from current CPA
calculations (Hodges 1979, Pettifor 1978, Turchi 1984). The quantity Ad will be regarded
as independent of concentration, which cannot be quite correct since, close to
concentrations 0 and 1, ordered and disordered states must differ insignificantly.
It will be further assumed, again in accordance with more rigorous CPA calculations,
that the centres of gravity of sub-bands do not change with concentration, but that the
disordered state sub-band widths change with concentration according to

where p is a positive constant. This ensures that the minority sub-bands be narrower than
the majority, in agreement with the qualitative considerations set forth in 5 1. In
accordance with the result obtained by Velicky et a1 (1968) for semi-elliptical sub-bands,
the exponent p has been taken as 0.5. Thus, the following expressions are obtained for the
band edges
E , = E A-$w, E,=E, +:w,
1
E3 = E , - ~ W B E~=EB+ IWB.

The determination of the DOS now proceeds as follows. Values are chosen for the band-
widths W, and W, of the pure elements, then a value for the diagonal order d d for the
disordered state is chosen. Next, the concentration c is fixed; the value of W, the average
bandwidth, is thus determined by equation (4) and hence, by equation (3), the difference A
of the centres of gravity E, and E, of the two sub-bands. Without loss of generality, we
may take E , = -EB, so that these quantities are determined as well. Having selected a
value for the exponent p , we can obtain the actual sub-band widths W, and W , by equation
(14), and all band edges by equation (15). Finally, the band heights h, and h, are
determined by equation (lo).
All characteristics of the sub-bands for the disordered state are now determined so that
the first four moments po, p , , p 2 , p 3 , can be calculated from equations ( 1 1 ) and (12). A
2168 M Sluiter et a1

- 10
Ln
c
a

L
c
n
L
2 0
>r
P
c
:-5
-10

AU ordering energy ( a r b i t r a r y units)

-6 -4 -2 0 2 4
Energy [ a r b i t r a r y units)

Figure 3. Graphical construction for calculating the ordering energy at a given concentration
(c=0.3) as a function of the band-filling A (dotted and full lines pertain to ordered and
disordered states respectively). Starting from the DOS for the ordered and disordered states
(a), fi and U are calculated as a function of the Fermi energy in both states (b) and (c)
respectively). For a given fi (here fi= 6 ) the Fermi energy and the corresponding U. and U,
are shown. The curve A U ( f i ) is obtained by plotting the difference between U, and U,,( d )
for each value of R.
 Ordering in binary transition-metal alloys 2169

plausible value for the change A6 of diagonal disorder on ordering is chosen, then inserted
into equation (13) in order to obtain the ordered state So, and hence the difference between
ordered sub-band centres. For a fixed concentration, the first four moments remain
constant upon ordering, so that the sub-band edges and heights can now be calculated for
the ordered state.
Energies for ordered and disordered states are obtained from equation (1) after first
calculating the Fermi energies E$ and E; from equation (2), for the two states with (for
fixed c) constant band filling given by equation ( 7 ) . The calculations are illustrated in figure
3 for concentration c=O.7. The DOS are given in figure 3(a), with band filling and energy
plotted in 3(b) and (c), with the same energy abscissa. Fermi levels are determined from
Lv
the plot as shown, and energy values U , and U , are found in figure 3(c),as illustrated.
The energy difference A U is then replotted, on a different scale, in figure 3(d), which
shares the lq axis with figure 3(b). In accordance with the theorem quoted earlier, it is seen
that the ordered and disordered DOS share four common values, i.e. the DOS difference
passes through zero four times in the energy interval, and the A N and A U differences
vanish three times and twice within the interval, respectively. It is seen also that the AU
curve plotted against I? vanishes twice, as expected, inside the (0, 10) interval.
By means of equation ( 7 ) , the A U against 8 curves can be replotted for the whole
range of concentration c, as illustrated in figure 4(a). The graphical construction for
plotting the vanishing of A U in a (I?, c) diagram is indicated in figure 4 : the loci of the
zeros of AU separate regions where ordering is expected from those where disordering
(actually phase separation, at low temperature) is expected. Such an ordering map (see
figure 4(b)) thus indicates the sign of the ordering energy A U as a function of I? and c, The
practical procedure for using these ordering maps is indicated in figure 4(b): first the
number of d electrons in pure A and pure B is determined. Since it has been shown that the
total effective number of s-p electrons is constant and close to unity for all elements across
a transition-metal series (Pettifor 1977), the numbers N A and N e are taken to be simply one

10
I lb)

8-

2-
Ordering separation
0 , J
0
.-

-0.6 -0.4 -0.2 0 0.2 0.4 0 0.2 0.4 0.6 0.8 1.

AU ordering energy (arbitrory units) A c concentration B

Figure 4. Procedure for converting ordering energy as a function of band filling for various
concentrations into an ordering map. (a) The ordering energy is shown for two
concentrations, c=O.3 and c = 0.6 (full and broken curves respectively). ( b ) The regions
where h Ci is negative (ordering) are projected on the concentration-band-filling plane. The
use of the ordering map is illustrated for a particular alloy ( N A= 9, NB= 4); it is seen that for
concentrations less than about 40%, the alloy has a tendency to order, and that for higher
concentrations phase separation will prevail.
2170 M Sluiter et a1

less than the Periodic Table group index. By equation (7), the straight line constructed
between numbers N A and NB in figure 4(b) thus represents the band filling as a function of
concentration, i.e. the average number of d electrons per atom. Portions of that straight
line lying in the ‘ordering’ region (AV< 0) therefore indicate the range of concentrations
over which ordering is expected, disordering elsewhere. The effect of varying several DOS
parameters on the position of the AU=O loci will now be examined.

4. Results

Ordering energy values A U have been calculated for WA= W, = 5 eV, 6, = 2, p = 0.5 and

Figure 5. Ordering energy as a function of concentration and band filling (for p=O.S, Ad=
-0.001, 8 d = 2, W , = W , = 5 eV). The base plane shows the contours of this function; the
AU=O contour denotes the ordering map.
 Ordering in binary transition-metalalloys 2171

0 0 2 0.4 0.6 0.8 1.

A Concentration B

Figure 6. Ordering m a p as function of the exponent p with d d = 2 , A6=-0.001, W A =

W , = 5 eV. Full curve:p={, broken c u r v e : p = 1, dotted c u r v e : p = 2 .

A6=-0.001, and plotted as function of concentration and band filling in a three-

dimensional plot (figure 5 ) . Figure 5 includes a contour map of the ordering energy in the
concentration-band-filling plane and thus illustrates that the ordering map is simply the
contour map sectioned at A U = 0. Calculated values of the ordering energy are of the order
of a few meV. It is interesting to note that a very small change in A6 on ordering is
sufficient to produce realistic ordering energy values.
The location of AU nodes on an (fi,c) diagram has been calculated for rectangular
bands whose parameters were varied one by one in a systematic way. In figures 6 and 7,
the exponent p is varied from 0.5 to 2 and the parameter Ad is given values - 0.001 and
-0.1 respectively. In both figures it is clearly seen that uncertainties in the values of
parameters p and A6 have no significant influence on the node loci of AU. Figure 8 shows
the effect of 6, on the ordering map: a decrease of 6, (i.e. the sub-bands tend to overlap)

0 02 OL 0 6 08 1
A c concentration B

Figure 7. Ordering m a p a s function of the parameter A6 with p = 0.5, ad= 2, W A= W B=

5 eV. Broken curves : A6 = -0.1,full curves : A6 = -0.001.
2172 M Sluiter et a1

1
, I
0 0.2 0.4 0.6 0.8 1
A c concentation

Figure 8. Ordering map as function of 6 d with p = 0 . 5 , A6=-0.001, W A = W B =eV.

~
Dotted curves: bd = 1, broken curves: bd = 1.4, full curves: dd = 2.

leads to a substantial broadening and a counter-clockwise rotation of the ordering region.

In figure 9 the influence of the non-diagonal disorder parameter and,
given by W, - W E / @ ,
is studied by Varying W,, WE, with 6, being kept constant. When a,,, increases from
negative to positive values, the same effect as expected for a decrease of 6, is observed.
The changes in the ordering maps can be grouped roughly into three categories
depending on the type of order/disorder boundary motions. The results are summarised in
figure 10. In this figure the dotted lines represent the change in the ordering region resulting
from altering the band-parameter as indicated.
Because of the quasi-invariance of the ordering maps, it is possible to define a

I
I , , I
0 0.2 0.4 0.6 0.8 1
A c concentration B

Figure 9. Ordering map as function of the A bandwidth w i t h p = 0 . 5 , 6, = 2 , Ab=-O.OOl,

WB= 5 eV. Broken curves: WA = 2.5 eV, full curves: W A= 5 eV, dotted curves:
W , = 10 eV.
 Ordering in binary transition-metal alloys 2173

\
p increasing

c
D
0

,1 I d 6 I increasing

A 8
c concentration

Figure 10. Summary of the influence of input parameters on the shape of the ordering map.
Changing parameters as described causes the ordering region to change from the shape
outlined by full lines to the one depicted by dotted lines.

‘standard’ one, such as that of figure 5 or 7, with rather fuzzy boundaries between ordered
and disordered regions. Since the ‘ordered’ region straddles the diagonal (broken curves in
figure 7), it follows that ordering is expected when the electron concentration is equal to
roughly ten times the atomic concentration of the element with the highest number of d
electrons per atom. Deviations from this ‘diagonal rule’ occur at low and high (atomic)
concentrations, as shown. It also follows that ordering is expected over a wide range of
concentrations if one element belongs to the beginning of the transition-metal series and the
other to the end, in such a way that the d band of the alloy is approximately half filled.
Conversely, disorder (or phase separation) is expected over wide concentration ranges if
the two elements both belong to the beginning of the series or both to the end. These
general rules about ordering tendencies are by no means artifacts of the present
oversimplified model : much more elaborate tight-binding CPA calculations lead to ordering
maps that display the very same features as those observed here (Turchi 1984, Bieber and
Gautier 1986). It appears, therefore, that the details of the shape of the ordered and
disordered DOS are not of major importance, for ordering-map purposes, provided that the
properties of moments are properly taken into account. In turn, this indicates that errors
2174 M Sluiter et a1

Table 1. Electron concentrations (&') at stoichiometry for L12, A15, Llo, B2, D019, D022,
D024 and C15 phases observed in transition-metal alloys, with composition A,B (or B,A)
so that N , > N B .

LIz(A3B) fi L12 (A3B) fi

CU~AU 10.0 Au~CU 10.0 Au,Ni 9.75
Cu3Pt 9.75 Au3Pt 9.7s C UPd~ 9.7s
Ag3Pt 9.75 Pd 3 Fe 8.5 Pt3Co 8.75
Pt3Fe 8.5 Pt3Mn 8.25 Ni3Pt 9.0
Ni3Fe 8.5 Ni3Mn 8.25 Rh3Sc 6.5
Pt3Lu 7.25 Pt 3Y 7.25 Pt3Cr 8.0
Pt3Ti 7.5 Pt3Sc 7.25 Ir3Ta 7.0
Ir Hf 6.75 Ir3Nb 7 .o Ir 3Zr 6.15
Ir3Cr 7.25 Ir3V 7.0 Ir 3Ti 6.75
Pd3Lu 7.25 Pd3Y 7.25 Pd3Sc 1.25
Rh3Ta 7.0 Rh3Hf 6.15 Rh3Nb 7.0

~ ~~ ~~

Pt3V 7.75 Pd3Ta 7.75 Pd3Nb 7.7s

Pd3V 7.75 Ni3V 1.15

AuCu 10.0 AuMn 8.0 AgTi 6.5

PtNi 9.0 PtCo 8.5 PtFe 8.0
PtMn 7.5 IrMn 7.0 PdFe 8.0
PdMn 7.5 RhMn 1 .o NiFe 8.0
NiMn 7.5 IrNb 6.0 PdCr 7.0
RhNb 6.0

A15(B3A) .q AIS(B3A) fi AlS(B3A)

MolPt 6.0 Cr3Pt 6.0 Mo31r 5.57

Cr31r 5.75 Cr3Rh 5.15 Ta3Au 5.5
MO3 OS 5.5 NblAu 5.5 Cr30s 5.5
Cr3Ru 5.5 V~AU 5.5 W3Re 5.25
Ta3Pt 5.25 Mo3Re 5.25 Nb3Pt 5.25
V,Pt 5.25 V3Pd 5.25 Nb3Rh 5.25
Ta3Rh 5.0 Nb31r 5.0 Nb3Rh 5.0
V31r 5.0 V3Rh 5 .O V~CO 5.0
Nb30~ 4.15 Zr3Au 4.75 V30s 4.75
Ti,Au 4.75 Ti3Pt 4.5 Ti Pd 4.5
Ti3Ir 4.25

committed in replacing the actual DOS by rectangular bands tend to cancel in taking the
(ordered-disordered) difference in the DOS and in the ordering energy A U.
Recently, Bieber and Gautier (1986) performed an extensive comparison between the
TB-CPA prediction of the ordered phases of a binary transition metal and experimental
observation. Here, only the most common ordered structures will be considered (see table
1): the L1, Ll,, DO,,, B2, DO,,, DO,,, A15 and C15. In this table, element A is taken to
be the one with highest number of d electrons, in keeping with the convention used
throughout this paper.
 Ordering in binary transition-metal alloys 2175

Table 1 (continued)
~

B2 (AB) N B2 (AB) N B2 (AB) N

CuPd 9.5 RhFe 7.5 CoFe 7.5
IrMn 7.0 RhMn 7.0 AuLu 6.0
AuY 6.0 AuSc 6.0 CULU 6.0
CUY 6.0 cusc 6.0 AgLu 6.0
AgY 6.0 AgSc 6.0 PtSc 5.5
osv 5.5 PdLu 5.5 PdSc 5.5
RuV 5.5 NiSc 5.5 CoHf 5.5
CoZr 5.5 CoTi 5.5 FeV 5.5
IrLu 5 .O IrSc 5 .O OsHf 5.0
OsZr 5 .O OsTi 5.0 RhLu 5.0
RhY 5.0 RhSc 5.0 RuHf 5.0
RuZr 5 .O RuTi 5.0 TcTa 5 .O
TcV 5.0 COSC 5 .O FeTi 5.0
MnV 5.0 ReTi 4.5 RuLu 4.5
RuSc 4.5 TcHf 4.5 TcTi 4.5

Ir,W 7.25 Ir3Mo 7.25 CO~W 7.25

Co3Mo 7.25 Rh3W 7.25

Pt3Hf 7.5 Pt3Zr 7.5 Pd3Hf 7.5

Pd3Zr 7.5 PdlTi 7.5 Ni3Hf 7.5
Ni3Ti '1.5

C15 (BA,) fi ClS(BA2) N C15(BA2)

HfW2 4.33 HfCoz 6.33 HfCr 4.33

HfV2 3.67 LuIr, 6.0 LuRh2 6.0
LuNi, 6.67 LUCO, 6.0 LuFe2 5.33
ZrIrz 6.33 ZrW2 4.33 ZrMo, 4.33
ZrCo, 6.33 ZrFe, 5.67 ZrV, 3.61
YPt, 6.67 YIr2 6.0 YRh2 6.0
YNiz 6.67 YCO, 6.0 YFe , 5.33
YMn2 4.67 ScIr, 6.0 ScNi, 6.61
scco, 6.0 Tic02 6.33

For stoichiometry 0.25 (A3B), L1, structures are seen from table 1 to have electron
concentrations in the range 5.5- 10. However, all phases that have electron concentrations
greater than 8 contain the elements Mn, Fe, CO,Ni or involve one of the noble metals, Cu,
Ag and Au. It is well known that the simple TB approximation, on which the present model
is based, is not valid for full d bands (noble metals), since the contributions from s bands
must be taken into account, or for magnetic alloys because of the competition between
chemical and magnetic ordering. These two cases being excluded, the electron
concentrations of actual ordered phases fall neatly within the allowed region in the
ordering map. This is true not only for L1, ordered structures but also for the other A3B
2176 M Sluiter et a1

0 02 0L 0.6 0.8 1.0

A c concentrotion 8

Figure 11. Comparison of the model ordering map with experimentally observed ordered
structures. The thick bars indicate ranges of band filling and the crosses ( * ) denote the
average band filling for a particular compound. The two model ordering maps have as
parameters : p = 0.5, A6 = -0.00 1, W A= W B= 5 eV and 6 d = 1 (dotted curves) or 6 d = 2 (full
curves).

structures: DO,,, DO,, and DO,,, as can be seen in figure 11, where the cross marks the
average electron concentration, over a group of compounds, for a particular structure.
The Llo phases neither contradict nor confirm the present model. As with the case of
the L1, phases, elements Cu, Ag, Au and Mn, Fe, C O and Ni form stable ordered phases
with electron concentrations that are way out of the range allowed by the model. The
remaining three L l 0 phases, which do not contain these exceptional elements (IrNb, PdCr
and RhNb), have electron concentrations of 6 and 7. The ordering map, however, indicates
ordering for AB phases for electron concentrations ranging from 4 to 6, approximately (see
figure 11).
The B2 phases are in excellent agreement with the ordering map. With only very few
exceptions, all B2 phases have electron concentrations between 4.5 and 6, fitting in nicely
within the theoretically predicted range of 4 to 6 electrons. The few phases that lie outside
this range (with higher electron concentrations) all contain the already mentioned
exceptional elements Cu, Ag, Au or exhibit magnetic properties for which the assumptions
on which the model is based are not valid. These alloys have been left out of the B2 phase
field in figure 1 1.
The C 15 phases, with the formula A, B, occur within the predicted range of electron
concentrations. C15 phases have an average band filling &' between 3.67 and 6.67. The
A,B phases with low I'?consist of A and B elements that have almost the same number of
d electrons. A small value of N A - N B leads to a small diagonal disorder parameter which
causes the ordering region in the ordering map to rotate so as to become almost a
horizontal band centred around fl= 5 (the effect is illustrated in figure 8).
The A15 phase occurs at a concentration of 0.75 and is thus expected to have an
average band filling &' in the range of 2 to 4.5. The experimentally observed A15 phases,
however, occur for fl in the interval of about 4.5 to 6, which barely overlaps with the
predicted range (see figure 11). An explanation is that a number of A15 phases are
 Ordering in binary transition-metal alloys 2177

characterised by a significant non-diagonal disorder parameter (Turchi er a1 1983, Turchi

1984), in which case the effect already mentioned in figure 9 will act in favour of the
stability of the A 15 phases. Furthermore, many A 15 phases are non-stoichiometric so that
the actual band filling differs from the one calculated.

5 . Discussion

A simple rectangular band model has been used to derive two-dimensional ordering maps
with d-electron concentration &’ and atomic concentration c as coordinate axes. It was
found that binary compounds of non-magnetic transition metals fell nicely within the
region of the ordering maps where order, as opposed to disorder, or phase separation, was
expected. Unlike the Miedema rules (Boom et a1 1976 and references therein), the maps
predict not only whether or not ordering will take place, for a given binary alloy, but also
predict over what ranges of concentrations the formation of ordered phases is expected.
Thus, it is found that ordering is likely to take place over wide concentration ranges when
the d-electron concentration is about ten times the atomic concentration of the element
with the highest number of d electrons per atom. In that case, the alloy d band is
approximately half filled.
The location of the order/disorder boundary in the ordering maps was found to be
relatively insensitive to moderate changes in the values of the parameters of the model,
such as the bandwidths of the pure elements. For larger changes, corrections to the
position of the order/disorder boundaries can be made according to the qualitative rules
given in figure 13. In particular, when the diagonal disorder 8, becomes very small the
ordering region tends to take the form of a horizontal band across the ordering map.
Although the rectangular band approximation has been used successfully previously,
notably by Pettifor (1978, 1979, 1984) and Hodges (1979) in their analysis of the Miedema
scheme (Boom et a1 1976) it is surprising that such a simple model should work so well. A
possible explanation is that the ordering energy depends, for its electronic contribution, on
the difference between the quantities integrated over the DOS. It is probable that, say,
electron double counting or the neglect of s-p contributions affect the DOS of ordered and
disordered configurations in roughly the same way, and therefore tend to cancel out when
taking differences. Furthermore, the details of the DOS fine structure will tend to become
‘ironed out’ by the integration process.
What is most important, however, is the overall change of shape of the DOS sub-bands
on ordering. That characteristic change is brought about as a consequence of the theorem
relative to moments of the DOS:it turns out that the invariance, at fixed c, of the first four
moments severely restricts the shape modification of the DOS upon ordering. In particular,
the twice vanishing of A U in the 0 < N < 10 interval determines the general features of the
ordering maps.
The exact vanishing of the moment differences Ap,, to Ap3 can only be proved
rigorously for the case of the simple tight-binding approximation, but more elaborate
computational methods, such as the KKR-CPA, should produce DOS which, for present
purposes, will exhibit the same general features. Hence, the rules about the zeros of A U are
expected to hold quite generally.
The actual types of ordered structures expected cannot be predicted from the simple
model proposed here, or course. In fact, the determination of ground states of order require
the precise calculation of at least the first few effective pair interactions obtained, for
example, by the generalised perturbation method (GPM) (Bieber and Gautier 1981a, b,
2178 M Sluiter et a1

Bieber et a1 1983, Bieber and Gautier 1986, Gautier 1985, Turchi 1984). For higher-
neighbour pairs, especially, very accurate CPA computations are required; KKR-CPA-GPM
calculations, supplementing existing TB results, are currently in progress (Turchi et a1
1987).

Acknowledgments

This work was partially supported by a grant from the National Science Foundation. One
of the authors (PT) was partially supported by the Department of Nuclear Chemistry,
Lawrence Livermore National Laboratory, under Contract No W-7405-eng-48, Dr M
Fluss, principal investigator.

References

Bieber A, Ducastelle F, Gautier F, Treglia G and Turchi P 1983 Solid State Commun. 45 585
Bieber A and Gautier F 1981a Solid State Commun. 38 1219
-1981b Solid State Commun. 39 149
- 1986 Acta Metall. 34 229 1
Boom R, De Boer F R and Miedema A R 1976 J . Less-Common Met. 46 27 1
Dieudonne J 1968 Calcul Infinitesimal (Paris: Hermann) p 162
Ducastelle F 1984 Mater. Res. Soc. Symp. Proc. vol. 21 Phase Transformations in Solids ed. T Tsakalakos
(Amsterdam: North-Holland) p 375
Ducastelle F and Gautier F 1976 J . Phys. F : Met. Phys. 6 2039
Friedel J 1969 The Physics of Metals ed. J M Ziman (Cambridge: CUP) p 340
Gautier F 1985 High Temperature Alloys: Theory and Design ed. J Stiegler (New York: AIME) p 264
Gautier F, Ducastelle F and Giner J 1975 Phil. Mag. 31 1373
Hodges C H 1979 J. Phys. F : Met. Phys. 9 L89
Pettifor D G 1977 J . Phys. F: Met. Phys. 7 613
-1978 Solid State Commun. 28 621
- 1979 Phys. Rev. Lett. 42 846
- 1984 PhysicalMetallurgy Pt 1 ed. R W Cahn and P Haasen (Amsterdam: North-Holland) p 74
Turchi P 1984 These de doctoral d 'Etat University of Paris VI, unpublished
Turchi P and Ducastelle F 1985 Springer Series in Solid State Sciences vol. 58 ed. D G Pettifor and D L
Weaire (Berlin: Springer) p 104
- 1987 to be published
Turchi P, Gonis A and Stocks G M 1987 to be published
Turchi P, Treglia G and Ducastelle F 1983 J . Phys. F : Met. Phys. 13 2543
Velicky B, Kirkpatrick S and Ehrenreich H 1968 Phys. Rev. 175 747