Downloaded from orbit.dtu.

dk on: Jun 02, 2019

Surface segregation energies in transition-metal alloys

Ruban, Andrei; Skriver, Hans Lomholt; Nørskov, Jens Kehlet

Published in:
Physical Review B Condensed Matter

Link to article, DOI:

10.1103/PhysRevB.59.15990

Publication date:
1999

Document Version
Publisher's PDF, also known as Version of record

Link back to DTU Orbit

Citation (APA):
Ruban, A., Skriver, H. L., & Nørskov, J. K. (1999). Surface segregation energies in transition-metal alloys.
Physical Review B Condensed Matter, 59(24), 15990-16000. https://doi.org/10.1103/PhysRevB.59.15990

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright
owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

 Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
 You may not further distribute the material or use it for any profit-making activity or commercial gain
 You may freely distribute the URL identifying the publication in the public portal

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
PHYSICAL REVIEW B VOLUME 59, NUMBER 24 15 JUNE 1999-II

Surface segregation energies in transition-metal alloys

A. V. Ruban, H. L. Skriver, and J. K. No” rskov
Center for Atomic-scale Materials Physics and Department of Physics, Technical University of Denmark, DK-2800 Lyngby, Denmark
~Received 27 August 1998!
We present a database of 24324 surface segregation energies of single transition metal impurities in
transition-metal hosts obtained by a Green’s-function linear-muffin-tin-orbitals method in conjunction with the
coherent potential and atomic sphere approximations including a multipole correction to the electrostatic
potential and energy. We use the database to establish the major factors which govern surface segregation in
transition metal alloys. We find that the calculated trends are well described by Friedel’s rectangular state
density model and that the few but significant deviations from the simple trends are caused by crystal structure
effects. @S0163-1829~99!05424-7#

I. INTRODUCTION eral empirical and semiempirical theories for surface segre-

gation in transition-metal alloys have been proposed on the
It is well known that the chemical composition at the basis of the Miedema theory4 and the simplest tight-binding
surface of an alloy may differ from the composition in the ~TB! approximation.5–7 However, these theories yield only
bulk; that is, one of the alloy components may enrich the the sign of the surface segregation energy and the compari-
surface region. This phenomenon, known as surface segrega- son of these limited predictions with experiment is a rather
tion, is of vital importance in all of surface chemistry as it controversial issue. In particular, the author of Ref. 8 con-
may enhance or suppress desirable and undesirable chemical cludes that ‘‘all examined theories fail to predict the correct
reactions. In spite of the obvious technological interest, how- segregation in a considerably high number of cases.’’ Today,
ever, no quantitative description based on segregation ener- one may question this conclusion as it was reached in part on
gies evaluated from first-principles have been forthcoming the basis of experimental data which for some systems, e.g.,
and the present understanding of surface segregation rests to Cu and Fe, are in conflict with the presently accepted point
a large degree on empirical theories, even in the simplest of view.
case of a single impurity in a pure host. Unfortunately, attempts to develop a quantitative, micro-
The qualitative description of surface segregation is fur- scopic theory based on more elaborate TB
ther impeded by the lack of reliable experimental data. In approximations9–11 have not been particularly successful
fact, there exists one experimental technique that yields yielding results which in some cases appear even qualita-
quantitatively reliable segregation energies, namely photo- tively incorrect. The first numerically derived surface segre-
emission spectroscopy of surface core-level shifts ~SCLS!. gation energies which appear quantitatively reasonable have
Within the so-called Z11 approximation, a SCLS corre- been obtained by Foiles et al.12 who treated the first two
sponds to the segregation energy of an atom of atomic num- layers of a fcc~100! surface of late transition and noble metal
ber Z11 in a host of atoms of atomic number Z.1 As a result, alloys by means of the embedded atom method. It turns out,
photoemission yields reliable surface segregation energies that the application of this and similar approximate total en-
but only for a very restricted set of dilute alloys. For all other ergy methods to the earlier transition metals becomes quite
alloy systems one must resort to measurements of the surface complicated because of the strong nonpairwise interatomic
composition of concentrated alloys and estimates based on interactions that exist in these metals and their alloys, and
the Langmuir-McLean relation between bulk and surface this kind of work has not been pursued further.
composition.2 To resolve the problems mentioned above and to establish
In the latter case there are two problems. First, the surface trends in the surface segregation energy for transition-metal
composition of most alloys is very sensitive to the external alloys one may turn to first-principles total-energy calcula-
conditions and to the purity of the sample. Further, elements tions based on density-functional theory. Until recently such
such as H, C, N, O, and S are, as a rule, present in all calculations have been quite rare and mainly connected with
metallic systems and may easily segregate toward the surface the determination of SCLS ~Refs. 13–21! rather than segre-
thereby changing the surface composition due to cosegrega- gation energies in general. In particular we note that the
tion effects. Secondly, the Langmuir-McLean relation is only Green’s-function linear-muffin-tin-orbitals ~GF-LMTO! cal-
approximate and, moreover, only valid if the ordering effects culations by Alden et al.18,19,21 established systematic trends
in the system may be neglected. One example of a system in the surface segregation energies of Z11 atoms in Z hosts
where this condition is not fulfilled is NiPt~110!: According for simple,19 rare-earth,21 and transition metals.18 These cal-
to the Langmuir-McLean relation one should observe strong culations revealed two important features: ~i! the surface seg-
Pt segregation towards the surface. Instead, the ordering ef- regation energy in a transition-metal alloy is essentially
fects lead to a segregation reversal.3 given by the difference in the surface energies of the pure
The dearth of experimental information leads to difficul- alloy components—a connection which has been assumed
ties in the development of qualitatively reliable models. Sev- for a long time basically from common-sense arguments ~see

0163-1829/99/59~24!/15990~11!/$15.00 PRB 59 15 990 ©1999 The American Physical Society

PRB 59 SURFACE SEGREGATION ENERGIES IN TRANSITION- . . . 15 991

also Ref. 22!—and ~ii! the surface segregation energies as the trends in the surface segregation energies.
well as the surface energies depend strongly on the crystal In the analysis we find that a picture based on the differ-
structure of the alloy components—a dependence noticed al- ence in the surface energies of the alloy components predicts
ready by Mårtenson et al.23 in their analysis of experimen- a simple general behavior which is obeyed by a large pro-
tally determined SCLS. portion of the alloy systems we consider. A similar general
The GF-LMTO calculations by Alden et al. of the segre- behavior has also been found in the empirical calculations by
gation energies in the Z11 approximation were in good Chelikowsky4 and Mukherjee et al.6 However, we also find
agreement with the experimental observations as were the large deviations from the simple behavior when the host is
direct calculations of the surface segregation in CuNi and one of the earlier transition metals. In those cases it appears
NiPt alloys24,3 by means of the coherent potential approxi- that the crystal structure of the host plays a significant role
mation ~CPA!. The success of these calculations inspired a and we show how the structural contribution to the segrega-
comprehensive study by the GF-LMTO-CPA method of the tion energy may be accounted for in the virtual bond model.
segregation energies in transition-metal alloys.25 All of the
above GF-LMTO calculations employed the atomic sphere II. METHOD OF CALCULATION
approximation ~ASA! with a monopole-dipole correction to The surface segregation energy is the energy cost of trans-
the electrostatic potential and energy26 except for the work ferring an impurity atom from the interior to the surface of a
on the simple and rare earth metals where the ‘‘spherical cell host crystal and may therefore be defined as the difference in
model’’ was used.19,21 Further, in Ref. 25 all calculations the total energies of the system with the impurity in a surface
were non-spin-polarized. Although the ASA including the layer and in the bulk. An alternative, but equivalent defini-
monopole-dipole correction is quite accurate for the close- tion, is obtained from the intensive, i.e., per atom, form of
packed surfaces, i.e., fcc~111! and bcc~110!, of most metals the total and surface energies of the system. In this case the
the approximation leads to unacceptably large errors in the energy connected with the segregation of a B atom from the
anisotropies of the surface energies for the early transition interior of the host to the lth layer is given by
metals.
To obtain accurate surface segregation energies for all
combinations of transition metals we therefore in the present
work include higher multipole moments of the electron den-
B
E segr2l 5
dE sur f ~ A 12c l B c l !
dc l
U c l 50
, ~1!

sity both in the construction of the interatomic part of the

spherically symmetric one-electron potential and the electro- where E sur f (A 12c l B c l ) is the surface energy of a system
static contribution to the total energy. In addition, we allow which consists of an A 12c l B c l alloy embedded in the lth
for spin-polarized solutions in all cases where magnetic hosts layer of an otherwise pure A host and
or impurities are involved. A detailed comparison shows that
(
our approach leads to segregation energies that are in good
agreement with full-potential GF-Korringa-Kohn-Rostoker
E sur f ~ A 12c l B c l ! 5 ~ E l 8 2E bulk ! 2 m c l . ~2!
l8
calculations27 of single impurities in bcc Fe. A similar com-
parison with the measured SCLS for the 5d metals shows Here, E l 8 is the total energy per atom of the l 8 th layer,
that the present calculations leads to segregation energies E bulk the total energy per atom of the host, and m the effec-
which deviates by less than 0.1 eV from the experimental tive chemical potential of the B component in the host A,
values. which is defined as

U
Using an alternative approach Drchal and co-workers28,29
have calculated the surface segregation profiles for a number dE bulk ~ A 12c B c !
m5 . ~3!
of alloys including most recently the RhPt system.30 These dc c50
authors rely on the general perturbation method ~GPM!
which completely neglects the renormalization of the host To compute the surface segregation energies from Eqs.
effective medium and therefore, as shown in Ref. 31, may ~1!–~3! we calculated the total energies of the surface and
lead to surface concentration profiles which are not only bulk alloy using density-functional theory in the local den-
quantitatively but also qualitatively incorrect. For this reason sity approximation ~LDA! in conjunction with the CPA and
and because Drchal and co-workers do not list surface seg- the Green’s-function technique for semi-infinite
regation energies we will not consider their calculations here surfaces26,33,34 as implemented in the tight-binding
but refer to Ref. 31 where a discussion of the accuracy of the representation35–37 of the LMTO method.38,39 In all systems
GPM is presented and to Ref. 32 where a discussion of their we used an spd basis set and included the core electrons in
results for the RhPt alloy system may be found. the LDA self-consistency loop. For exchange and correlation
It is the goal of the present paper to establish the general we used the Perdew-Zunger interpolation formula40 of the
trends in the surface segregation phenomena in transition- many-body results by Ceperley-Alder41 except when the al-
metal alloys and, in particular, to discuss the reason why the loy system contained one of the magnetic 3d metals, V, Cr,
simplest models do not work in all cases. To do so in a Mn, Fe, Co, and Ni, in which case we performed spin-
meaningful way, it is important to have a consistent set of polarized calculations by means of the Vosko-Wilk-Nusair
data, experimental or theoretical, at one’s disposal. It is for parametrization.42 In the calculations for the close-packed
this reason that we devote part of the present paper, although bcc~110!, hcp~0001!, and fcc~111! surfaces we included
it is not the main subject, to discuss and establish the accu- eight atomic and two vacuum layers in the self-consistent
racy of the database which we subsequently use to establish iterations. For the more open surfaces these numbers were
15 992 ” RSKOV
A. V. RUBAN, H. L. SKRIVER, AND J. K. NO PRB 59

increased in proportion to the decreasing interlayer distance The number of multipoles included in the L summations
in order to keep the size of the surface region constant. in Eqs. ~6! and ~7! is determined by the angular momentum
The crucial charge-transfer effects were included in the cutoff l max used in the Green’s-functions calculations. Due
single-site approximation by the screened impurity model to the properties of the Gaunt coefficients the multipole mo-
~SIM! ~Refs. 43 and 44! giving the following correction to ments of the charge density have nonzero components up to
the electrostatic potential 2l max . In the present case l max 52 but in the actual calcula-
tions we included Madelung contributions to the potential
l j 5e q l j /R 1 ,
V SIM ~4!
2
and the total energy only for angular momenta up to l53 as
where q l j is the intralayer charge transfer of the j 8 th alloy the inclusion of the next momenta did not affect the results.
component, determined as the difference between its net All calculations have been performed at the theoretically
charge, Q l j , and the average net charge of the lth layer, determined equilibrium volumes. For the hcp metals we used
Q̄ L 5 ( j c l j Q l j , i.e., q l j 5Q l j 2Q̄ l , and R 1 is the distance a single c/a equal to 1.59 as the experimental values vary
to the nearest coordination shell. The corresponding contri- only from 1.58 to 1.61. Lattice relaxation effects were ne-
bution to the electrostatic energy is glected both in the bulk and surface calculations. Although
the lattice relaxation contributions to the impurity solution
E SIM 52 b (
l, j
c l j V SIM
lj qlj , ~5! energies and thus to the effective chemical potentials may be
of the order of 0.5 eV in unfavorable cases where the alloy
components have large size difference, the effect on the sur-
where a prefactor b 50.6 gives the best overall CPA total
face segregation energies is very small due to the fact that
energies of the bulk random alloys relative to the corre-
the relaxation energies at the surface and in the bulk are
sponding total energies calculated either from a cluster ex-
almost equal thereby compensating each other in the final
pansion of the total energies of completely ordered alloys44
or by the super-cell approach.45 Preliminary results from calculations.25
super-cell calculations for surfaces of random CuPt alloys48
show that this empirical coefficient also works well at sur-
faces. We emphasize that this approach ensures a correct III. THE ACCURACY OF THE METHOD
concentration dependence of the total energies and a correct
In this section we compare our calculated surface energies
renormalization of the electronic structure of the host atoms
and surface segregation energies with the results of other
around the impurities. As a result, the chemical potentials
calculations as well as with experimental data to establish the
and segregation energies may be obtained correctly in the
single-site approximation.31 accuracy of the database to be presented in Sec. IV A. This is
The second important correction to the electrostatic po- quite a difficult task since the accuracy of many of the ap-
tential and energy in the surface calculations is the monopole proximations, e.g., the ASA, the single-site approximation,
contribution to the ASA potential from the higher multipole and the CPA, used in the present work are not well estab-
moments of the charge density and the corresponding lished for many of the systems we treat. Nevertheless, the
multipole-multipole contributions to the interatomic part of comparison of our results with the relatively few calculations
the Madelung energy. The inclusion of these terms, i.e., be- which do not rely on these approximations and with experi-
yond the monopole-dipole contributions, may be viewed as mental data shows that the above-mentioned approximations
the first step towards the full charge-density ~FCD! lead to surface segregation energies with an accuracy of 0.1
technique.46 Following Ref. 47 we call this correction ASA eV which is sufficient to establish reliable trends. We further
1M . find that the multipole correction, i.e., ASA1M , is important
In the ASA1M the monopole, i.e., L50, contribution to for obtaining quantitatively correct results for surface ener-
the effective one-electron potential is evaluated from the gies as well as for surface segregation energies.
multipole moments, Q Lli , of the valence electron charge by
the multipole expansion
A. Theoretical results
1
V 0li 5
S (
n , j,L 8
0,L 8
M li, L8
n jQn j , ~6! 1. Surface energies
In Fig. 1 we compare the surface energies for the most
where S is the average Wigner-Seitz radius, L is short hand close-packed surfaces of the 4d-transition metals calculated
L,L 8 within the ASA and the ASA1M with results obtained by
for the (l,m) quantum numbers, and M li, n j is the multipole
Madelung matrix which is equivalent to the conventional the FCD-LMTO method which has an accuracy similar to
~unscreened! LMTO structure constants. The corresponding that found in full-potential calculations.49,50 Although the
Madelung contribution to the total energy is then given by ASA, corrected only by the monopole-dipole term,26 pro-
vides a fairly good description of the trends it is seen that the
1 surface energies of the early transition metals have errors of
EM5
2S (
li,L
Q Lli (
n j,L 8
L,L 8 L 8
M li, n jQn j . ~7! up to 30%. It is further seen in Table I that the ASA fails to
yield correct surface energy anisotropies. In contrast, the
A description of the procedure including expressions for the ASA1M yields surface energies and anisotropies which
Madelung matrices and the multipole moments may be agree with the corresponding full charge-density and full-
found in Ref. 26. potential results.
PRB 59 SURFACE SEGREGATION ENERGIES IN TRANSITION- . . . 15 993

FIG. 1. The surface energies of 4d transition metals calculated

by the GF-LMTO technique with (ASA1M ) and without ~ASA!
multipole correction compared with the full charge density results
~FCD-GGA! ~Ref. 49!.

2. Surface core-level shifts

There exists a number of full-potential calculations of sur-
face core-level shifts based on an ordered alloy supercell
approach.14,15 For bcc~110! Mo, fcc~100! Rh, and fcc~100!
Pd Methfessel et al.15 calculate the SCLS to be 20.24 eV, FIG. 2. The surface segregation energies of 3d metals for the
20.65 eV, and 20.30 eV, respectively, which compares ~110! and ~100! surfaces of bcc Fe calculated with (ASA1M ) and
favorably with our values of 20.21 eV, 20.65 eV, and without ~ASA! multipole correction. The full-potential KKR results
20.37 eV obtained within the ASA1M and Z11 approxi- ~FP-KKR! are from Ref. 27.
mations.
means of the full-potential Green’s-function KKR method.
3. Surface segregation energies The results of these calculations are presumably the most
Recently, Nonas et al.27 calculated the segregation ener- accurate to date and may serve as a measure of the accuracy
gies for a series of single impurities in bcc ~100! Fe by of the present LMTO-CPA approach. Hence, we compare in
Fig. 2 our segregation energies obtained in the ASA as well
TABLE I. Surface energies ~in J/m2 ) for Nb, Mo, and Tc cal- as the ASA1M for 3d impurities in bcc ~110! and ~100! Fe
culated with and without multipole correction compared to the full with those of Nonas et al.
charge density ~FCD! ~Refs. 49 and 50! and full-potential ~FP! re- One may draw several conclusions from the result shown
sults ~Refs. 51 and 52!. in Fig. 2. First, we observe that the multipole correction to
the ASA is important for impurities at the beginning of tran-
ASA ASA1M FCD FP a
sition metal series. Second, the surface segregation energies
bcc Nb depend strongly on the surface orientation: All surface seg-
g (110) 1.79 2.53 2.69 2.36 2.9 b regation energies for the ~110! surface are approximately a
g (100) 1.73 2.88 2.86 2.86 3.1 b factor of 2 smaller than the corresponding values for the
g (100)/ g (110) 0.97 1.14 1.06 1.21 1.07 b ~100! surface. This is in accordance with the Friedel model,
bcc Mo which will be presented in Sec. IV B and which gives a re-
g (110) 3.18 3.60 3.45 3.14 duction of 2.19, and also with the broken bond model which
g (100) 2.78 3.97 3.84 3.52 was used by Nonas et al.27 to understand their first-principles
g (100)/ g (110) 0.87 1.10 1.11 1.12 results and which gives a reduction of a factor of 2. In both
f cc Tc cases the decrease in surface segregation energies is caused
g (111) 2.73 3.19 3.08 2.63 by the increase in the number of broken impurity-host bonds
g (100) 2.87 3.83 4.05 3.34 from 2 for ~110! surfaces to 4 for ~100! surfaces. As a result,
g (110) 2.74 3.61 3.40 3.00 it is not meaningful to compare surface segregation energies
g (100)/ g (111) 1.05 1.20 1.31 1.27 for different surface facets as is done in Ref. 27. Third and
g (110)/ g (111) 1.00 1.13 1.10 1.14 most important for our purpose, the general agreement be-
tween the ASA1M and the full-potential results is good,
a
Reference 51. although there are differences of order of 0.1 eV for Ti, V,
b
Reference 52. and Co and 0.2 eV for Cr.
15 994 ” RSKOV
A. V. RUBAN, H. L. SKRIVER, AND J. K. NO PRB 59

face core-level shift, was therefore 0.08 eV lower the value

of 0.36 eV60.012 obtained by Riffe et al. Although our
calculations still predict a somewhat higher value for the
surface core-level shift, 0.47 eV, they give 0.08 eV for the
subsurface core-level shift, i.e., the segregation energy for W
into the second layer of Ta~110!, which is very close to the
experimental value.
A similarly large subsurface segregation energy for Z
11 atoms is calculated for W~110! and Mo~110! where we
find 0.08 eV and 0.12 eV, respectively. At the same time, the
experimental surface core-level shifts, which are 20.31 eV
for W~110! and 20.33 eV for Mo~110!, have been obtained
without including an additional subsurface peak in the analy-
sis of the photoemission spectra.55,61 This may, in fact, be the
reason why the experimental SCLS are lower than our theo-
retical values of 20.27 and 20.21 eV, respectively. We note
here that the only full-potential calculation of the SCLS in
Mo~110! by Methfessel et al.14 give a value of 20.24 eV
FIG. 3. The surface core-level shifts of the 5d metals for the
close-packed surfaces, bcc~110!, fcc~111!, or hcp~0001!, of the ex-
close to ours.
perimentally observed crystal structures calculated as the surface
segregation energies of Z11 elements in Z hosts and compared IV. TRENDS IN THE SURFACE SEGREGATION
with experimental data taken from Refs. 23, 55 and 56–59.
A. Calculated surface segregation energies
The origin of the large difference for the surface segrega- To help visualize the general trends we have plotted the
tion energy of Cr in bcc ~100! Fe is not clear at present. We calculated surface segregation energies in the 24324 matrix
have performed 96- and 216-atom supercell test calculations shown in Fig. 4, color-coded according to their magnitude.
for Fe50Cr50 and for dilute alloys of Cr in Fe by the LSGF In the figure, red colors correspond to negative segregation
method53,54 and they show unambiguously that both the energies and, hence, to segregation of the impurity ~solute!
single-site approximation and the CPA works well for this towards the surface of the host, and blue colors correspond to
system giving accurate total energy as well as average mag- positive segregation energies and, hence, to the situation
netic moments of the alloy components. We may therefore where the impurity prefers to remain in the interior of the
only speculate that the error of 0.2 eV for Cr is caused by the host. The underlying database may be found in Table II.
ASA which can lead to errors of this magnitude in the fcc- A database of surface segregation energies in transition-
bcc structural energy difference for transition metals. We metal alloys has already been established by Christensen
return to a discussion of the connection between the struc- et al.,25 and although the present database on account of its
tural energy difference and the segregation energies in Sec. improved accuracy should be preferred as a source of theo-
IV C. In spite of the discrepancy for Cr it appears that the retical surface segregation energies, the qualitative picture
accuracy of our ASA1M approach for surface segregation obtained from the earlier database is still correct. Further, the
energies, in general, is better than 0.1 eV relative to the true work of Christensen et al.25 includes a model of relaxation
local density result and that this accuracy is sufficient to effects which remains valid as well as a database of the cur-
establish a quantitatively correct picture of surface segrega- vature of the surface energy curves which may be used to
tion in transition-metal alloys. determine whether the surface alloy will form a solid solu-
tion or phase separate.
It is important to note that all the calculated segregation
B. Measured surface core-level shifts
energies have been obtained for single impurities at close-
In Fig. 3 we compare our calculated surface segregation packed surfaces and that the segregation energies at other
energies for Z11 impurities for the close-packed surfaces of surfaces may be quite different as demonstrated in Fig. 2. In
the 5d metals with available experimental data for the sur- the case of concentrated alloys the actual surface composi-
face core-level shifts.23,55–59 It is seen that the experimental tion depends on factors such as the tendency toward ordering
trends are correctly reproduced by the calculation and that and the relative values of the segregation energies for differ-
the agreement between theory and experiment generally is ent subsurface layers. In fact, even if the segregation energy
better than 0.1 eV. of a single impurity is negative, there may be segregation
In the above comparison one should note that most of the reversal in a concentrated alloy, as it happens, for instance, at
experimental values shown in the figure have been extracted the ~110! surface of PtNi.3 Further, due to the large differ-
from x-ray photoemission spectroscopy data neglecting the ence in the reactivity of transition metals with gases such as
core-level shifts from the subsurface layer~s!. The only ex- CO, O2 , and N2 the surface composition of a transition-metal
ception is the data for Ta~110! obtained by Riffe et al.56 who alloy is very sensitive to the external conditions of an experi-
showed that there is a pronounced subsurface core-level shift ment; see for instance Ref. 62. Hence, one should be careful
of about 0.07 eV which substantially influences the interpre- when comparing the results shown in Fig. 4 with experimen-
tation of the experimental data. The previous experimental tal data for concentrated alloys or alloys formed in deposi-
SCLS for Ta~110!,60 obtained without including the subsur- tion experiments.
PRB 59 SURFACE SEGREGATION ENERGIES IN TRANSITION- . . . 15 995

FIG. 4. ~Color! Surface segregation energies of transition-metal impurities ~solute! for the closed-packed surfaces of transition metal
hosts.

B. The Friedel model for surface segregation where z s and z b are the coordination numbers at the surface
It is found by Alden et al. 18,19,21
that the main contribu- and in the bulk, respectively, N A and N B are the
tion to the surface segregation energy is given by the differ- d-occupation numbers of the host and the impurity, respec-
ence in the surface energies of the impurity and the host. tively, W i is the d-bandwidth of metal i, and W B→A
This observation is confirmed by the present calculations and 5(W A W B ) 1/2 the d-bandwidth of a single B impurity in the A
one may therefore attempt to use Friedel’s rectangular state- host. In Eq. ~8! the square-root dependence on the coordina-
density model63,64 for surface energies to describe general tion numbers follows from tight-binding theory and the
trends in the surface segregation in transition-metal alloys. geometric-mean dependence of W B→A follows from the av-
For this purpose we write erage bond model.71
In the simple approximation W A 5W B→A the two terms in

segr 5
E B→A
1
20 F SDG
12
zs
zb
1/2
$ W A N A ~ 102N A !
the curly brackets of Eq. ~8! are represented by the same
parabola and the model may immediately be used to con-
struct a canonical segregation behavior. Such a construction
2W B→A N B ~ 102N B ! % , ~8! is shown in Fig. 5 for the segregation of 4d impurities in 4d
 15 996

TABLE II. Calculated surface segregation energies ~in eV! for impurity atoms ~columns! at the close-packed surfaces of transition-metal hosts ~rows!. The second row for ~host! Fe ~Fe*! corresponds to the segregation energies on the bcc~100!
surface. The numbers in the diagonal are the surface energies ~in eV/atom!.

Ti V Cr Mn Fe Co Ni Cu Zr Nb Mo Tc Ru Rh Pd Ag Hf Ta W Re Os Ir Pt Au

Ti 1.20 0.10 20.24 20.34 20.41 20.56 20.75 20.94 20.38 0.03 0.09 20.06 20.31 20.62 20.93 21.18 20.14 0.25 0.35 0.20 20.04 20.37 20.72 21.05
V 20.49 1.16 0.30 0.41 0.36 0.15 20.12 20.54 21.08 20.41 0.10 0.36 0.39 0.13 20.28 20.75 21.00 20.23 0.31 0.62 0.68 0.51 0.09 20.39
Cr 20.72 20.15 1.46 20.14 20.44 20.67 20.80 21.02 22.05 21.15 20.62 20.45 20.68 21.25 21.70 21.90 21.55 20.98 20.40 20.17 20.29 20.81 21.58 21.98
Mn 20.83 20.32 20.10 1.24 20.12 20.26 20.47 20.77 22.15 21.28 20.73 20.48 20.52 20.69 20.93 21.31 21.83 21.03 20.56 20.31 20.32 20.53 20.83 21.23
Fe 20.39 0.06 0.10 20.16 1.20 20.14 20.65 20.83 21.60 20.65 20.06 0.10 20.20 20.52 21.05 21.55 21.50 20.35 0.20 0.45 0.25 20.15 20.66 21.36
Fe* 20.69 0.19 0.16 20.38 1.73 20.03 20.77 21.37 22.22 20.83 0.03 0.24 0.00 20.53 21.43 22.37 22.15 20.50 0.42 0.78 0.60 0.08 20.78 21.93
Co 20.33 0.13 0.19 0.10 20.01 1.07 20.13 20.48 21.40 20.45 0.00 0.49 0.12 20.40 20.60 20.93 20.56 20.24 0.34 0.72 0.56 20.10 20.38 20.76
Ni 20.12 0.20 0.25 0.00 0.13 0.13 0.95 20.25 21.16 20.26 0.18 0.31 0.10 20.10 20.40 20.80 20.74 20.01 0.45 0.53 0.37 0.16 20.17 20.69
Cu 0.01 0.25 0.10 0.07 0.28 0.33 0.17 0.77 20.64 0.00 0.28 0.30 0.20 0.05 20.20 20.42 20.35 0.22 0.57 0.62 0.48 0.23 20.04 20.29
Zr 0.06 0.01 20.47 20.36 20.40 20.45 20.55 20.72 1.22 0.19 0.13 20.01 20.19 20.41 20.68 20.88 0.15 0.33 0.30 0.15 20.02 20.25 20.50 20.80
Nb 20.24 0.12 0.32 0.23 0.29 0.31 0.08 20.20 20.65 1.21 0.48 0.70 0.65 0.42 0.05 20.32 20.47 0.17 0.70 0.98 1.00 0.77 0.40 20.03
Mo 20.14 0.08 20.01 20.50 20.52 20.72 20.82 21.16 20.90 20.22 1.60 20.21 20.81 21.28 21.47 21.75 20.98 20.02 0.22 0.10 20.45 21.15 21.60 21.94
Tc 20.82 20.40 20.11 20.20 20.11 20.01 20.24 20.70 21.57 20.77 20.27 1.47 0.02 20.11 20.46 20.97 21.26 20.65 20.06 0.26 0.37 0.21 20.16 20.70
Ru 20.30 0.15 0.24 20.40 20.39 20.37 20.71 21.21 21.12 20.31 0.10 0.17 1.48 20.43 21.03 21.72 20.83 20.17 0.24 0.37 0.23 20.20 20.82 21.62
Rh 0.12 0.35 0.31 20.08 20.01 0.02 20.08 20.38 20.46 0.09 0.44 0.46 0.31 1.15 20.45 20.92 20.15 0.36 0.66 0.71 0.56 0.23 20.27 20.87
Pd 0.58 0.78 0.30 0.30 0.35 0.29 0.21 0.04 0.32 0.87 1.08 1.02 0.74 0.36 0.84 20.26 0.44 1.04 1.37 1.34 1.11 0.70 0.19 20.22
Ag 0.45 0.63 0.29 0.23 0.41 0.48 0.49 0.22 0.33 0.67 0.74 0.69 0.60 0.42 0.28 0.58 0.40 0.83 0.93 0.88 0.72 0.55 0.34 0.03
A. V. RUBAN, H. L. SKRIVER, AND J. K. NO

Hf 20.03 20.04 20.52 20.49 20.51 20.62 20.75 20.93 20.14 0.08 20.12 20.25 20.44 20.68 20.92 21.16 1.35 0.25 0.17 20.01 20.26 20.53 20.80 21.11
Ta 20.45 20.03 0.16 0.10 0.13 0.06 20.18 20.52 20.85 20.21 0.25 0.44 0.40 0.11 20.26 20.67 20.60 1.37 0.47 0.75 0.73 0.49 0.11 20.37
” RSKOV

W 0.02 0.04 20.14 20.25 20.35 20.45 20.42 20.75 20.81 20.31 20.24 20.55 21.07 21.22 21.27 21.56 20.78 20.13 1.87 20.27 20.85 21.34 21.66 21.85
Re 20.89 20.42 20.13 20.28 20.18 20.14 20.36 20.83 21.75 20.94 20.43 20.19 20.17 20.32 20.68 21.24 21.51 20.77 20.27 1.69 0.04 20.11 20.46 21.05
Os 20.22 0.18 0.36 20.21 20.31 20.30 20.62 21.21 21.07 20.27 0.07 0.09 20.20 20.70 21.31 22.00 21.04 20.17 0.13 0.23 1.81 20.48 21.25 22.14
Ir 0.29 0.51 0.35 0.09 0.11 0.16 0.12 20.12 20.43 0.10 0.35 0.35 0.23 20.08 20.55 21.00 20.17 0.26 0.47 0.48 0.32 1.44 20.58 21.20
Pt 0.66 0.98 0.60 0.38 0.37 0.46 0.43 0.32 0.30 0.76 0.93 0.85 0.60 0.26 0.00 20.27 0.47 0.95 1.16 1.11 0.86 0.44 1.03 20.36
Au 0.46 0.59 0.33 0.30 0.45 0.54 0.56 0.34 0.30 0.61 0.67 0.59 0.52 0.44 0.28 0.00 0.47 0.79 0.92 0.81 0.65 0.50 0.34 0.72
PRB 59
PRB 59 SURFACE SEGREGATION ENERGIES IN TRANSITION- . . . 15 997

FIG. 5. Construction of the canonical segregation matrix for the

4d34d metals. The parabola in the left-hand panel is the surface
energy in the Friedel model given by g 5w @ f ( f 21) # , where f is the
d-band filling and W56 eV is the 4d-band width. In the matrix on
the right-hand side a shaded square corresponds to a negative sur-
face segregation energy, i.e., surface segregation takes place and a
white square means that surface segregation will not occur.

hosts. According to this model no impurities from the middle

of a transition series is expected to segregate to the surface of
early or late transition metals. Such a trend has, in fact, al-
ready been found in the empirical calculations by
Chelikowsky4 and Mukherjee et al.6 which is to be expected
FIG. 6. The surface segregation energies for the close-packed
since both of these models employ the difference in surface
surfaces of the 4d metals. For Zr, Nb, Mo, Tc, and Ru triangles
or cohesive energy in some form. correspond to the hcp~0001! surface while for Rh, Pd, and Ag they
The ‘‘hourglass’’ shape of the canonical segregation ma- correspond to fcc~111! surfaces. Circles correspond in all cases to
trix shown in Fig. 5 may be found to a varying degree of bcc~110! surfaces.
accuracy in each of the nine 838 subblocks of Fig. 4 and it
appears that the Friedel model forms a meaningful starting
point for the description of surface segregation in transition- of the surface segregation energy simply because the differ-
metal alloys. However, it is also clear that there are signifi- ence in the segregation energies for different crystal struc-
cant deviations from the canonical hourglass behavior, espe- tures is of the same order of magnitude as the segregation
cially when the host is one of the first three metals in each of energies themselves. Hence, to predict surface segregation
the series. For instance, according to Fig. 4 many metals phenomena one must take proper account of the structural
should segregate towards the surfaces of Ti, Zr, and Hf, but contribution.
not towards the surfaces of V, Nb, and Ta, and again towards
the surfaces of the Cr, Mo, and W. This ‘‘oscillatory’’ be-
havior is clearly in contradiction to the Friedel model accord- C. Crystal structure contribution to the segregation energy
ing to which the segregation tendency should increase mono- The origin of the strong structural dependence of the sur-
tonically from Ti, Zr, and Hf to Cr, Mo, and W in
face segregation energy is the local character of the inter-
accordance with the increasing surface energies of the impu-
atomic bonding in transition-metal alloys attributed to the
rities. In fact, the oscillatory behavior destroys the predictive
valence d electrons. The sequence of crystal structures along
power of empirical approaches. It is worth noting that a simi-
lar oscillatory behavior is exhibited by the SCLS shown in a transition-metal series is governed by the structural energy
Fig. 3 and in the work of Refs. 23 and 18 where the behavior difference which may be considered a canonical function of
was attributed to crystal structure effects. the d occupation number.65–67 In a tight-binding picture this
To demonstrate the validity and the failures of the Friedel canonical behavior is determined by the local atomic ar-
model for surface segregation on a quantitative basis we rangement through the corresponding moments of the d state
have used Eq. ~8! with @ 12(z s /z b ) 1/2# 50.13, which is ap- density.68–70 The structural energy difference varies consid-
propriate for close-packed surfaces, and d bandwidths and erably along a transition series and may be as large as 1
occupation numbers taken from Ref. 36 to calculate the sur- eV/atom. If, therefore, the local atomic arrangement is dis-
face segregation energies of 4d impurities in 4d transition- turbed either by disorder, point or other structural defects,
metal hosts. The results are compared with the first- such as impurities, vacancies, or surfaces, the system may
principles calculations in Fig. 6 and from this comparison it lose or gain a substantial amount of energy depending on the
is obvious that although the Friedel model works amazingly crystal structure, the d occupation numbers of the alloy com-
well, in general, it cannot capture the structural dependence ponents, and the spatial structure of the defect.
15 998 ” RSKOV
A. V. RUBAN, H. L. SKRIVER, AND J. K. NO PRB 59

It has recently been demonstrated that the vacancy forma-

tion energies in transition metals47 as well as the impurity
solution energies in transition metal alloys71 have a substan-
tial structural contribution. It was further shown that this
structural effect could be described within a virtual bond
model71 in terms of the average d occupation of the local
bond. Here we show that the same local-bond model ac-
counts for the strong structural dependence of the surface
segregation energy in transition-metal alloys.
To do this, we rewrite the surface segregation energy in
the form

E Bsegr 5 m 1 2 m , ~9!
where m and m l are the effective chemical potentials of B
atoms in the bulk and in the lth layer at the surface of host
A, respectively. The bulk effective chemical potential has
already been defined in Eq. ~3! and the definition of the
surface effective chemical potential is @see Eqs. ~1! and ~2!#

m l5
d S( D
n
En
. ~10!
FIG. 7. The difference in the surface segregation energy for
hcp~0001! and bcc~110! in Nb, Mo, Tc, and Ru calculated from
dc l first-principles ~GF-LMTO! and by Eq. ~8!, and plotted as a func-
As a result, the difference in the surface segregation energies tion of the simple average of the impurity and host d occupation
of B atoms in the a and b structures of the host A may be number ~See Ref. 71!. The dotted curve is the bcc-hcp energy dif-
written ference calculated for a pure elements by LMTO-ASA. For com-
parison we show the bcc-hcp energy difference obtained in the
a2b
DE segr 5 m a1 2 m b1 2 m a 1 m b 5D m a1 2 b 2D m a 2 b . ~11! ASA, filled squares, and in the ASA with combined correction,
open squares.
We now determine the effective chemical potential of the
bulk assuming that the dominating interactions in the system Equation ~14! shows that the a 2 b structural difference in
are given by interatomic pair potentials71 whereby the total the chemical potentials is given solely by the a 2 b strucural
energy of an A 12c B c random alloy in the a structure be- energy difference curve. Moreover, the prefactor of 2 means
comes that the structural energy difference, which can be of order 1
eV, is enhanced in the crystal structure difference for the
E Aa a)
5 v (0) 1 ~ 12c ! 2 v (AA a)
12c ~ 12c !v (AB a)
1c 2 v (BB . chemical potential. As has been shown in Ref. 71 this leads
12c B c
~12! in some cases to a difference of about 2 eV for the solution
Here, v (0) is the on-site term and v (XY a)
is obtained as a sum energies in different crystal structures of the host.
In the case of the surface segregation energy, however,
over the whole lattice of pair potentials acting between X and
a) 1 there is second term D m a1 2 b which compensates the struc-
Y atoms: v (XY 5 2 ( iÞ j V XY (R i 2R j ), where V XY are the struc-
turally independent pair potentials defined, for instance, by tural difference of the bulk chemical potentials D m a 2 b . Un-
fortunately, it is difficult to deduce its value from the above
Moriarty72 and the sums run over the lattice sites in the a
considerations, although ~i! it is clear that for closed-packed
structure.
Using Eqs. ~3! and ~12! the last term in Eq. ~9!, i.e., the surfaces D m la 2 b 'D m a 2 b for l>3, and ~ii! that, in general,
difference in the bulk effective chemical potential of B atoms D m a1 2 b 5 h D m a 2 b , where 0, h ,1 due to the broken bonds
in a given host of different crystal structures becomes and the destroyed crystal structure at the surface. In fact, our
calculations show that h 50.5 is a perfect choice for most of
a) b) a) b)
D m a 2 b 52 @~v (AB 2 v (AB ! 2 ~ v (AA 2 v (AA !# . ~13! the systems considered here, and thus
It is easy to see that the last term in Eq. ~13! is the a 2 b
a2b a2b a2b 1
structural energy difference of the host A, and for the first DE segr 'DE str ~ N Ad ! 2DE str ~ 2 @ N Ad 1N Bd # ! . ~15!
term we will use the virtual bond approximation71 assuming
a)
that interatomic potentials between A and B atoms v (AB is In Fig. 7 we compare the difference in the surface segre-
(a)
equal to the interatomic potential v CC of a pure metal C with gation energies of 4d impurities on the closed-packed sur-
a d occupation number given by N Cd 5 21 (N Ad 1N Bd ). As a re- faces of bcc and hcp Nb, Mo, Tc, and Ru calculated from
sult, the difference in the chemical potentials ~13! is simply first principles, as well as estimated by Eq. ~15! from the
the difference of the a 2 b structural energy difference of a structural energy difference curve for the pure 4d metals. In
hypothetical transition metal C and the host A most cases the simple expression works surprisingly well.
However, there are deviations, in particular, in Nb, which we
a2b 1 a2b
D m a 2 b 52 $ DE str ~ 2 @ N Ad 1N Bd # ! 2DE str ~ N Ad ! % . trace to the neglect in the present analysis of multiatom in-
~14! teractions, i.e., beyond pair potentials, which at the begin-
PRB 59 SURFACE SEGREGATION ENERGIES IN TRANSITION- . . . 15 999

ning of a transition-metal series give substantial contribu- electron potential and energy evaluated in the ASA. Our
tions to the structural energy difference.73 surface segregation energies are in good quantitative agree-
In the application of Eq. ~15! we made use of a structural ment with available density-functional calculations as well as
energy difference curve calculated within the ASA. As experimental data. We show that a simple model based on
shown in Fig. 7 this may have errors which for the bcc met- Friedel’s rectangular state-density approximation for surface
als Nb and Mo are about 0.1 eV compared to the more ac- energies describes the calculated trends in the surface segre-
curate calculations67 employing the combined correction gation in transition metals quite well. This is in accord with
term.38 Later transition metals, such as Tc and Ru, do not the simple understanding that the alloy component which has
suffer from this kind of error. It follows from the direct con- the lowest surface energy will segregate towards the surface
nection between the surface segregation and the structural of the alloy. In addition, we find structural contributions as
energies that the present calculations, which do not include large as 1 eV to the surface segregation energies which com-
the combined correction, for the early bcc transition metals pletely invalidates the predictions of the simple model when
may suffer a similar 0.1 eV error. This may also explain the one of the alloy components is taken from the beginning of a
discrepances between the present and the full-potential re- transition series. We finally show that the so-called virtual
sults for spin-polarized bcc Fe in Fig. 2. bond model71 accounts for these important structural effects.

V. SUMMARY ACKNOWLEDGMENTS

We present a database of the surface segregation energies The Center for Atomic-scale Materials Physics is spon-
in transition-metal alloys which in contrast to a previous GF- sored by the Danish National Research Foundation. The
LMTO-CPA calculation include the effect of spin- present work was in part financed by The Danish Research
polarization as well as a multipole correction to the one- Councils through The Center for Surface Reactivity.

1
B. Johansson and N. Mårtensson, Phys. Rev. B 21, 4427 ~1980!. 23
N. Mårtensson, H. B. Saafeld, H. Kuhlenbeck, and M. Neumann,
2
J. M. Howe, Interfaces in Materials ~John Wiley & Sons, New Phys. Rev. B 39, 8181 ~1989!.
York, 1997!. 24
A. V. Ruban, I. A. Abrikosov, D. Ya. Kats, D. Gorelikov, K. W.
3
I. A. Abrikosov, A. V. Ruban, H. L. Skriver, and B. Johansson, Jacobsen, and H. L. Skriver, Phys. Rev. B 49, 11 383 ~1994!.
Phys. Rev. B 50, 2039 ~1994!. 25
A. Christensen, A. V. Ruban, P. Stolze, K. W. Jacobsen, H. L.
4
J. R. Chelikowsky, Surf. Sci. 139, L197 ~1984!.
Skriver, and J. K. No” rskov, Phys. Rev. B 56, 5822 ~1997!.
5
P. Lambin and J. P. Gaspard, J. Phys. F 10, 2413 ~1980!. 26
6 H. L. Skriver and N. M. Rosengaard, Phys. Rev. B 43, 9538
S. Mukherjee and J. L. Moran-Lopez, Surf. Sci. 188, L742
~1987!. ~1991!.
27
7
M. Brejnak and P. Modrak, Surf. Sci. 247, 215 ~1991!. B. Nonas, K. Wildberger, R. Zeller, and P. H. Dederichs, Phys.
8
P. M. Ossi, Surf. Sci. 201, L519 ~1988!. Rev. Lett. 80, 4574 ~1998!.
28
9
R. Riedinger and H. Dreysse, Phys. Rev. B 27, 2073 ~1983!. A. Pasturel, V. Drchal, J. Kudrnovsky, and P. Weinberger, Phys.
10
H. Dreysse and R. Riedinge, Phys. Rev. B 28, 5669 ~1983!. Rev. B 48, 2704 ~1993!.
11 29
M. C. Desjonqueres and D. Spanjaard, Phys. Rev. B 35, 952 V. Drchal, J. Kudrnovsky, A. Pasturel, I. Turek, and P. Wein-
~1987!. berger, Phys. Rev. B 54, 8202 ~1996!.
12 30
S. M. Foiles, M. I. Baskes, and M. S. Daw, Phys. Rev. B 33, 7983 V. Drchal, J. Kudrnovsky, A. Pasturel, I. Turek, P. Weinberger,
~1986!. A. Gonis, and P. E. A. Turchi, in Tight-Binding Approach to
13
P. Feibelman, Phys. Rev. B 39, 4866 ~1989!. Computatinal Materials Science, edited by P. E. A. Turchi, A.
14
M. Methfessel, D. Hennig, and M. Scheffler, Surf. Sci. 287/288, Gonis, and L. Colombo, Mater. Res. Soc. Symp. Proc. 491 ~Ma-
785 ~1993!. terials Research Society, Pittsburgh, 1998!, p. 45.
15
M. Methfessel, D. Hennig, and M. Scheffler, Surf. Rev. Lett. 2, 31
A. V. Ruban and H. L. Skriver, Phys. Rev. B 55, 8801 ~1997!.
197 ~1995!. 32
A. V. Ruban and H. L. Skriver, Comput. Mater. Sci. ~to be pub-
16
M. Alden, H. L. Skriver, and B. Johansson, Phys. Rev. Lett. 71, lished!.
2449 ~1993!. 33
O. Gunnarsson, O. Jepsen, and O. K. Andersen, Phys. Rev. B 27,
17
M. Alden, H. L. Skriver, and B. Johansson, Phys. Rev. Lett. 71, 7144 ~1983!.
2457 ~1993!. 34
I. A. Abrikosov and H. L. Skriver, Phys. Rev. B 47, 16 532
18
M. Alden, I. A. Abrikosov, B. Johansson, N. M. Rosengaard, and ~1993!.
H. L. Skriver, Phys. Rev. B 50, 5131 ~1994!. 35
O. K. Andersen and O. Jepsen, Phys. Rev. Lett. 53, 2571 ~1984!.
19 36
M. Alden, H. L. Skriver, and B. Johansson, Phys. Rev. B 50, O. K. Andersen, O. Jepsen, and D. Glötzel, in Highlights of
12 118 ~1994!. Condensed-Matter Theory, edited by F. Bassani, F. Fumi, and
20
C. Engdahl, P. Stolze, K. W. Jacobsen, J. K. No” rskov, and H. L. M. P. Tosi ~North-Holland, New York, 1985!.
Skriver, J. Vac. Sci. Technol. A 12, 1787 ~1994!. 37
O. K. Andersen, Z. Pawlowska, and O. Jepsen, Phys. Rev. B 34,
21
M. Alden, B. Johansson, and H. L. Skriver, Phys. Rev. B 51, 5253 ~1986!.
5386 ~1995!. 38
O. K. Andersen, Phys. Rev. B 12, 3060 ~1975!.
22
G. Treglia, B. Legrand, and F. Ducastelle, Europhys. Lett. 7, 575 39
H. L. Skriver, The LMTO Method ~Springer-Verlag, Berlin,
~1988!. 1984!.
16 000 ” RSKOV
A. V. RUBAN, H. L. SKRIVER, AND J. K. NO PRB 59

40
J. Perdew and A. Zunger, Phys. Rev. B 23, 5048 ~1981!. 57
R. C. Baetzold, G. Apai, and E. Shustorovich, Phys. Rev. B 26,
41
D. M. Ceperley and B. J. Alder, Phys. Rev. Lett. 45, 566 ~1980!. 4022 ~1982!.
42 58
S. H. Vosko, L. Wilk, and M. Nusair, Can. J. Phys. 58, 1200 H. Tillborg, Ph. D. thesis, Uppsala University, 1993.
~1980!. 59
A. Flodstrom, R. Nyholm, and B. Johansson, in Advances in Sur-
43
I. A. Abrikosov, Yu. H. Vekilov, P. A. Korzhavyi, A. V. Ruban, face and Interface Science, Volume 1: Techniques, edited by R.
and L. E. Shilkrot, Solid State Commun. 83, 867 ~1992!. Z. Bachrach ~Plenum Press, New York, 1992!.
44 60
A. V. Ruban, I. A. Abrikosov, and H. L. Skriver, Phys. Rev. B C. Guillot, D. Chauveau, P. Roubin, J. Lecante, and M. C. Des-
51, 12 958 ~1995!. jonqueres, Surf. Sci. 162, 46 ~1985!.
45 61
I. A. Abrikosov and B. Johansson, Phys. Rev. B 57, 14 164 E. Lundgren, U. Johansson, R. Nyholm, and J. N. Andersen,
~1998!. Phys. Rev. B 48, 5525 ~1993!.
46 62
L. Vitos, J. Kollàr, and H. L. Skriver, Phys. Rev. B 55, 13 521 D. D. Beck, C. L. DiMaggio, and G. B. Fisher, Surf. Sci. 297, 293
~1997!. ~1993!.
47
P. A. Korzhavyi, I. A. Abrikosov, B. Johansson, A. V. Ruban, 63
J. Friedel, Ann. Phys. ~N.Y.! 1, 257 ~1976!.
and H. L. Skriver, Phys. Rev. B 57, 11 693 ~1999!. 64
F. Cyrot-Lackmann, Adv. Phys. 16, 393 ~1967!.
48
A. V. Ruban and H. L. Skriver ~unpublished!. 65
D. G. Pettifor, in Metallurgical Chemistry, edited by O. Kuba-
49
L. Vitos, J. Kollár, A. V. Ruban, and H. L. Skriver, Surf. Sci. 411, shevski ~HMSO, London, 1972!.
186 ~1998!. 66
O. K. Andersen, J. Madsen, U. K. Poulsen, O. Jepsen, and J.
50
L. Vitos ~private communication!. Kollar, Physica B & C 86-88B, 249 ~1977!.
51
M. Methfessel, D. Henning, and M. Scheffler, Phys. Rev. B 46, 67
H. L. Skriver, Phys. Rev. B 31, 1909 ~1985!.
4816 ~1992!. 68
F. Ducastelle and F. Cyrot-Lackmann, J. Chem. Phys. Solids 32,
52
M. Weinert, R. E. Watson, J. W. Davenport, and G. W. Fernando, 285 ~1971!.
Phys. Rev. B 39, 12 585 ~1989!. 69
P. Turchi and F. Ducastelle, in Recursion Method and Its Appli-
53
I. A. Abrikosov, A. M. N. Niklasson, S. I. Simak, B. Johansson, cations, edited by D. G. Pettifor and D. L. Weaire ~Springer,
A. V. Ruban, and H. L. Skriver, Phys. Rev. Lett. 76, 4203 New York, 1985!.
~1996!. 70
D. G. Pettifor, Bonding and Structure of Molecules and Solids
54
I. A. Abrikosov, S. I. Simak, B. Johansson, A. V. Ruban, and H. ~Clarendon Press, Oxford, 1995!.
L. Skriver, Phys. Rev. B 56, 9319 ~1996!. 71
A. V. Ruban, H. L. Skriver, and J. K. No” rskov, Phys. Rev. Lett.
55
D. M. Riffe, G. K. Wertheim, and P. H. Citrin, Phys. Rev. Lett. 80, 1240 ~1998!.
63, 1976 ~1989!. 72
J. A. Moriarty, Phys. Rev. B 38, 3199 ~1988!.
56
D. M. Riffe and G. K. Wertheim, Phys. Rev. B 47, 6672 ~1993!. 73
J. A. Moriarty, Phys. Rev. B 42, 1609 ~1990!.