PHYSICAL REVIEW B VOLUME 41, NUMBER 12 15 APRIL 1990-II

Structural energy-volume relations in first-row transition metals

A. T. Paxton'
Max Pla-nck Ins-titut fur Festkorperforschung, Heisenbergstrasse I, Postfach 800665,
D 700-0 Stuttgart 80, Federal Republic of Germany

M. Methfessel and H. M. Polatoglou

Fritz Ha-ber Ins-titut, Faradayweg 4-6, D IOO-OBerlin 33, Federal Republic of Germany
(Received 29 November 1989)
The total energy of the 3d transition metals is calculated as a function of volume in each of six
different crystal structures. The calculations employ the local-density-functional scheme and the
full-potential linear muffin-tin orbitals method. Both self-consistent and non-self-consistent Harris-
Foulkes calculations are shown and the connection is made between these and simpler tight-binding
and classical models of interatomic forces. The energy-volume relations may serve as a database in
the construction of such empirical schemes.

I. INTRODUCTION neglect of the effects of the sp band. An additional com-

plication is the magnetic ground state of transition metals
In this paper, we show calculated energy-volume (E-V} near the end of the row. In a spin-polarized description,
curves for the first-row transition metals from Sc to Cu. the fcc and hcp density of states (DOS) in Ni and Co are
We not only compare the energies of the cubic (fcc) and split because their Fermi level falls in a large maximum
hexagonal (hcp) close-packed and body-centered-cubic and the Stoner parameter is large; bcc Fe is stabi. lized by
(bcc} phases, but also investigate the stability of mythical a large splitting of the DOS so that the Fermi level for
structures with coordination numbers between 4 and 8. one of the spins falls in the minimum separating the
The purpose of the exercise is threefold. (1) By showing bonding and antibonding states characteristic of the bcc
E- V curves calculated using a state-of-the-art DOS, while the opposite spins occupy an almost full d
Hohenberg-Kohn density-functional approach in the band. It is not clear whether even the local-spin-density-
Kohn-Sham local-density approximation' we are able to functional theory can correctly predict the bcc structure
provide benchmark calculations against which to com- of a-Fe (Ref. 15} or the complicated spin and crystal
pare empirically obtained interatomic force potentials structures of Cr and Mn. Furthermore, no simple model
used in materials simulation. (2) The benchmark curves can include magnetic effects easily, so we have chosen to
are used here as a test of the recently developed and make our calculations non-spin-polarized. Naturally, an
much-discussed Harris-Foulkes approximation to the empirical potential for Fe, say, must produce a bcc
self-consistent Hohenberg-Kohn energy functional. (3) ground state; however it is worthwhile emphasizing that
We are able to confirm the predictions of relative stability in the local-density approximation, its structure is hcp.
of the close-packed and bcc phases from canonical d- For calculating the structures of extended defects such
band theory and tight-binding models, discuss the' con- as surfaces, grain boundaries, dislocations, and cracks, as
nection between the Harris-Foulkes approximation and well as problems of point defect formation and migration
the tight-binding bond model, and use our results to energies, it is important that an empirical description of
assess the possible transferability of tight-binding param- bonding should not fail in situations where atoms are un-
eters for transition metals into undercoordinated defect dercoordinated. This has always been a drawback in
environments. tight-binding or classical potentials for semiconductors,
Approximate descriptions of the bonding in transition and it is not clear whether a tight-binding description for
metals have existed (to varying degrees of complexity) in transition metals would not also fail. As a rough estimate
the literature for some 30 years (see, for example, Refs. of how the energy might change with undercoordination,
10, 9, and 11}. It is now known that while pair potentials we have looked at the simple hexagonal, simple cubic,
have proved enormously useful in understanding phe- and diamond cubic structures in which atoms have, re-
nomena ranging from radiation damage' to the structure spectively, eight, six, and four neighbors.
of extended defects, ' '" a proper treatment of the prob- The organization of the paper is as follows. In Sec. II,
lem must go beyond this to include the important effect we briefly outline the method we have used to solve the
of d-band filling which dominates the trends in properties Schrodinger equation self-consistently, discuss the
across the transition series; we also know that the sim- sources of error, and estimate the accuracy of the present
plest d-band tight-binding mode1 is itself in error due to calculations. We show results for the close-packed

41 8127 1990 The American Physical Society

8128 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU 41

phases in Sec. III which we compare with previous local- B. Sources of error and assessment of accuracy
density and tight-binding predictions and give a brief dis-
cussion of the effects of d-band magnetism. The E- V Sources of error fa11 naturally into two categories: ap-
curves are presented in Sec. IV, and results using the proximations arising in the construction of a model to
Harris-Foulkes approximation in Sec. V. The axial ratio represent the solid state, and those arising from the exe-
of the hcp metals is discussed in Sec. VI. We make con- cution of the model. The use of local-density-functional
tact again with tight-binding theory in Sec. VII, ~here all theory in calculating structural properties of transition
the E-V data points are scaled to the Rose equation of metals has been addressed in the literature '~' and
state and the analysis of Spanjaard and Desjonqueres is will be discussed in the present context briefly in Sec. III.
used to assess the transferability of tight-binding parame- Aside from the LDA, our model assumes that the full-
ters. Discussion and conclusions can be found in Sec. potential LMTO method' with a basis of 18 orbitals per
VIII. atom is suitably accurate for the present purpose. Possi-
ble sources of error are approximations made in the cal-
culation of the interstitial charge density, and exchange-
correlation energy density and potential, and have been
assessed in some detail recently. The charge density is
II. METHOD OF CALCULATION
calculated exactly in the muftin-tin spheres in angular
momentum components up to l =4. We use the same an-
A. Full-potential LMTO gular momentum cutoff in the interpolation of quantities
in the intersitial region, ' expanded in Hankel functions
of energies —1 and —3 Ry. A measure of the errors in
We solve the Schrodinger equation for each crystal this procedure is given by the dependence of the total en-
structure at a number of volumes, self-consistently within ergy on the kinetic energies (i.e., the localizations) of the
the local-density approximation (LDA). The Hamiltoni- charge-density basis set. In the close-packed metals,
an is represented in a basis of linear mufBn-tin orbitals' varying the energies from ( —1, — 3) Ry to ( —0. 1, —1)
(LMTO's) which are augmented with numerical solutions Ry leads to total energy changes of less than 0. 1 mRy per
of the radial Schrodinger equation within nonoverlapping atom. Errors arising from the size of the basis are easy to
mu5n-tin spheres. No shape approximation for the po- assess: in the present instance, the total energy is abso-
tential or charge density is used. In the interstitial re- lutely converged to within a few mRy per atom. This is a
gion, these quantities are expressed in Hankel-function far smaller error than that made by the LDA, in particu-
expansions using a recently developed approach. ' This lar in calculating the energies of free atoms. Further-
method is not much slower than standard LMTO calcu- more, this small error is known to cancel in the calcula-
lations in the atomic-spheres approximation"' (ASA). tion of energy differences between different crystal struc-
Because the augmentation is done in somewhat smaller tures.
spheres than in the LMTO-ASA method, it is necessary Convergence in the eigenvalue sums falls into the
to include two basis functions in each angular momentum second category mentioned at the beginning of this sec-
I channel. These are chosen to have kinetic energies of 0 tion, and is a consequence of the need to evaluate in-
(as do conventional LMTO's) and —1 Ry. The basis thus tegrals over the Brillouin zone. We use a uniform mesh
consists of 18 functions per atom. The core is permitted of sampling points with at least 15 divisions along each of
to relax but is taken to be spherical. the primitive vectors. In cubic lattices we shift this mesh
In the linear method, we are restricted to one principal into "special points"- positions so as to decrease the de-
quantum number in each l channel in the band calcula- generacy of each point and hence minimize the total
tion. However, in the first-row transition metals— number of sampling points. In conventional sampling,
particularly at the beginning of the series and in the a complementary error function of width W replaces the
non-close-packed structures — the 3p electrons spill out of step function at the Fermi surface so as to ensure ex-
the mu5n-tin spheres and can show non-negligible ponential convergence with the number of divisions. We
dispersion', even 3s electrons can be affected in extreme use a recently developed generalization of this, in which
cases such as the diamond cubic structure. These "sem- we approximate the step function by a complementary er-
icore" electrons from low-lying bands separated from the ror function plus N higher-order correction terms. The
valence bands by an energy gap. We have treated the width parameter 8' and the order N are determined by
semicore states in the same way as the band states, by set- stringent convergence tests in typical cases and subse-
ting up and diagonalizing the Hamiltonian a second time. quently estimated by the amount of variation in the DOS
This automatically corrects errors in the eigenvalues and near the Fermi energy. We find that in self-consistent
charge density due to the wave functions extending into calculations the total energy is less sensitive to the choice
the interstitial region and into neighboring spheres. In of 8'and N than in the problem of integrating a given set
the "second-panel" band calculation we use only an sp of bands. This is apparently a consequence of the varia-
basis, and since the bands are quite fiat and completely tional principle, and means that a choice of X and Wby
filled, only a few (special ) k points are needed in accu- inspection is appropriate here. A more careful choice of
mulating the eigenvalue sum. An alternative method of these parameters can lead to a very large saving in non-
dealing with semicore states has been discussed by self-consistent methods, such as frozen potential or
Mattheiss and Hamann. ' Harris-Foulkes calculations. We estimate that our
41 STRUCTURAL ENERGY-VOLUME RELATIONS IN FIRST-ROW. . . 8129

Brillouin-zone integrals are converged absolutely to differences in the LDA was by Skriver, who used
within 0.5 mRy. LMTO-ASA in the force theorem developed by Ander-
sen. Here, only one structure (fcc) is made self-
III. STRUCTURAL ENERGY consistent. The resulting atomic-sphere potentials are
VERSUS BANDFILLING IN THE then used to construct a trial potential for the second
CLQSK-PACKED PHASES structure. The structural energy difference is then the
difference in eigenvalue sum of the self-consistent struc-
We begin the presentation of our results with the ener- ture and that obtained from solving the Kohn-Sham
gies of the close-packed structures, fcc, hcp, and bcc. equation in the trail potential. There is a clear connec-
There is a long tradition of the calculation of the close- tion between this simplified (but in many cases accurate)
packed structural energy differences as a function of LDA approach and the association of the structural ener-
atomic number in the three transition series. This
gy difference with Friedel's bond energy. A comparison
has arisen from the curiosity evoked by the observation shows ' that the canonical moments method is in good
that the transition metals follow a consistent trend qualitative agreement with the LDA force theorem re-
hcp~bcc~hcp~fcc across each of the three series, sults. In Fig. 1, we show the structural energy differences
with the only exceptions being La, Mn and the ferromag- we have been discussing, calculated self-consistently in
netic elements Fe and Co. Friedel first suggested that the full-potential LMTO method outlined in Sec. II. We
the structural energy of the transition metals was dom- compare these with the results of Skriver for the first-
inated by the eigenvalue sum or bond energy arising from row metals.
the narrow band of d electrons. This suggestion was Although it is clear that the structure of these curves
partly motivated by the observation that the sublimation can be easily predicted from simple tight-binding models,
energies followed a parabolic development across the the fact that the actual energy differences are very small
series as did the cohesive energy in the Hartree approxi- means that great care must be taken when making these
mation, calculations from first principles. In fact, the fcc-hcp en-
E
ergy differences are on the order of the absolute accuracy
f (ed dE,
E)nd(E)—
of first-principles methods, so it is gratifying that we are
for any reasonable shape of the d-electron density of in such good agreement with the calculations of Skriver
states nz (centered at the atomic energy level ed) includ- in the ASA force theorem. Indeed, many of the
differences that remain are due to the fact that while
ing the very simple rectangular DOS. Developing n„
in its power moments, Skriver used an ideal axial ratio for all his hcp struc-
tures, we have chosen to use experimental values for
p, „=f (E —e„)"nz(E)dE, those metals (Sc and Ti) whose observed ground-state
structure is hcp. Also we have calculated the energy
Ducastelle and Cyrot-Lackmann showed the direct con- differences at our calculated atomic volumes rather than
nection between these moments and the lattice topology. at experimental ones.
Exploiting the fivefold degeneracy of the d band, and Let us comment briefly on the results. The structures
transformation rules given in the Slater-Koster tables, of the nonmagnetic elements Sc, Ti, and V are correctly
they gave the exact result that within the tight-binding reproduced, as well as that of Cr, where the half filling of
model, the fcc-hcp energy difference would go through the d band dominates over magnetic effects in stabilizing
zero at least twice as a function of the filling of the d the bcc structure. The local-density prediction of the
band. This follows from the first four moments (po —p, )
being identical in the two structures. Furthermore, using
0.25—
only the difference between the first nonidentical moment
p4 for each structure, they were able to immediately
sketch the general shape of the fcc-hcp energy-difference
curve as a function of number of d electrons. The -0.25—
F
characteristic separation of the bcc DOS into predom- O
inantly T2 bonding and predominantly E antibonding
states by a pseudogap immediately leads to the prediction )a 0.25—
that the bcc structure will be stabilized in the center of CO
UJ
the transition series, i.e. , near half band filling. Using just c3
eight exact moments of the DOS of a canonical d band
in a first-neighbor model (first and second in bcc) already -0.25—
gives enough detail to the shapes derived from the above
simple arguments to reproduce the structural trends I I l I l i 1 i
Sc Ti V Cr Mn Fe Co Ni Cu
(with the exceptions of incorrectly predicting the bcc
structure for the noble metals in which the s-electrons are
playing a major role in cohesion, and moving the second FIG. 1. hcp-fcc (circles) and bcc-fcc (squares) structural ener-
zero in the fcc-bcc curve to too large a band filling, hence gy differences in the 3d transition metals. (a) present results us-
giving the bcc structure to the hcp metals in group 8).' ing full-potential LMTO; (b) Skri ver's calculations using
The first calculation of these structural energy LMTO-ASA and the force theorem.
8130 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU 41

structure of Mn and Fe is hcp. In the observed ground pseudopotential calculation. These curves were original-
state Mn has the structure of the "y" phase, a body- ly intended to demonstrate the power of the LDA in pre-
centered-cubic structure with 58 atoms per cell. The dicting structural phase stability and pressure-induced
structure is almost topologically close packed. ' Of the transformations in Si. However, they also became stan-
29 atoms in the primitive cell, there are four inequivalent dard bench mark results against which classical or sem-
positions populated by 1, 4, 12, and 12 atoms, respective- iempirical tight-binding models could be tested. In
ly. Five atoms are in CN16 (i.e., with a coordination some cases they have actually been used in fitting the pa-
number of 16) coordination polyhedra, 12 in CN12 rameters of the models.
icosahedra, and 12 have 13 neighbors in their coordina- In view of this, and the increasing interest in the
tion shells. Experimentally, it is found that the Mn creation of classical and semiempirical potentials for
atoms in CN16 polyhedra have a large (spin up) local transition metals, we show in Fig. 2, E-V data points for
magnetic moment, while the remaining 24 atoms have a the elements Sc —Cu, each in six structure: fcc, hcp, bcc,
smaller (spin down) moment such that the total antiferri- simple hexagonal (sh), simple cubic (sc), and diamond cu-
magnetic moment is zero. Topological close packing is bic (dc). We have drawn curves by hand through the cal-
the most efficient way of close packing two atoms of culated data points. For those elements for whom the
different sizes. In a-Mn, the high-spin atom behaves as close-packed structural energy differences are close to
the large atom, and the low-spin atom as the small atom: zero, we show additional curves in insets with expanded
the magnetic structure seems consistent with the atomic scales. We should note, however, that on these scales the
structure of the g phase in which the higher coordinated separation of the curves approaches the accuracy of our
sites tend to be occupied by the larger atom. It is not calculations (see Sec. III). We have omitted the hcp
clear at present, though, why this structure should be curves in V, Ni, and Cu. In V, the two curves become su-
favored over simpler close-packed structures, for exam- perimposed at 0/Go&0. 9 (see also Fig. 3). hcp Ni will
ple, the antiferromagnetic bcc 5-Mn. Ferromagnetic a- be discussed in Sec. VI. The energy difference between
Fe is, of course, well known to have the bcc structure in ideal axial ratio hcp and fcc Cu is less than 0.007
the ground state. Nonmagnetic hcp Fe, however, is eV/atom in our calculations. We do, however, find fcc
thought to have a lower enthalpy than fcc below 400 K, Cu to be stable by this amount with respect to ideal axial
which is consistent with the present results. Also a-Fe ratio hcp.
begins to transform to hcp c-Fe at 130 kBar, although The energies shown in Fig. 1 are binding energies per
at this pressure the magnetic moment will still be larger atom Ez relative to the energies of spin-polarized free
than 2 Bohr magnetons ' so it cannot be the vanishing atoms. The free-atom energies are identical to those
of the moment that stabilizes the hcp phase. LMTO- given by Moruzzi, Janak, and Williams since we have
ASA calculations' in the local-spin-density approxima- used their parameterization of the von Barth —Hedin
tion with gradient corrections to the exchange- exchange-correlation energy density. The atomic
correlation energy density show a phase transformation volumes are given relative to the observed atomic
from ferromagnetic a-Fe to nonmagnetic fcc at about 80 volumes in the ground state; these volumes are given in
kBar. These authors' did not consider the hcp phase, Table I. In the same table, we show our calculated atom-
but their results must be an upper bound on the a-c
transformation pressure in their model. Our nonmagnet-
ic volumes 0;„, cohesive energies, and bulk moduli I( p
obtained from least-squares analyses of the E-V data
ic calculations incorrectly give Co the fcc structure, al- points. We will comment briefly on these results; they
though below 700 K Co is hcp. In fact, both phases are may be compared to a similar compilation by Moruzzi,
observed at room temperature, and it has been speculat- Janak, and Williams, - who pioneered these calculations
ed that fcc is in fact the stable phase. It remains to be using the Korringa-Kahn-Rostoker (KKR) method. (We
seen whether the ground-state structure of Co can be cal-
culated from first principles; we return to this point
briefly in Sec. VI. The fcc structure of Ni is correctly
have neglected corrections for zero-point motion in our
calculations. ) The cohesive energies
well known to be consistently
(E„„=
overestimated
are-
Es )
in the
predicted in the LDA. hcp Ni has been observed in thin LDA. This is most probably largely due to errors made
films subjected to heat treatment, and electron ' and in calculating the total energy of free atoms. However, it
neutron irradiation. However, there seems to be no evi- is useful to know what the LDA values of cohesive ener-
dence of the existence of hcp Ni in a bulk phase. Finally, gy are, and to compare these to more approximate calcu-
it is interesting to note that the structural energy lations. 0;„and Kp are, respectively, underestimated
differences in Cu are extremely small. Cu lies beyond the and overestimated in the LDA. In fact, we find that our
realm of d-band tight-binding theory, but it seems that in bulk moduli are also mostly higher than those found by
the LDA, the energies of hcp and fcc phases in Cu [and Moruzzi, Janak, and Williams, although both are quot-
Ag (Ref. 25)] are almost indistinguishable. ed at the calculated atomic volume. In the magnetic
metals, the magnetic moment favors larger atomic
IV. ENERGY-VOLUME RELATIONS volumes, and hence the errors in 0;„and Ep are partic-
ularly large in these elements in the present non-spin-
We now turn to our calculated E-V curves in the first- polarized calculations. A direct comparison to experi-
row transitions metals. Curves of this kind were first pro- mental quantities for magnetic materials is not very
duced for Si in an historic paper by Yin and Cohen us- relevant and is included in Table I for completeness and
ing local-density theory in a plane-wave, first-principles for comparison with previously published calculations.
41 STRUCTURAL ENERGY-VOLUME RELATIONS IN FIRST-ROW. . . 8131

Sc -30—
-25- -25—

d cubic

-30— cubIc -30—

$& -35- 0C -45—

4I
IS s cubic
LU
-40~ -55 -50—

-45— hcp
-55—
bcc
fcc -61—
hcp
I I I I I I I I
-65— I I I
08
I
09 I I
10
I I I I I I I I

08 10 12 14 08 10 12 14 08 10 12 14 16
0/Ap A/Ap 0/Ap

-15
-35—
d cub
-25—
-20—
-40—
-30—
s cubic
—-30-
C
O -45-
s cubic -35-
s hex
g) -35-
LIJ
QX -50—
-43
-40-
-4p— -44
-55-
bcc
bcc
-45- hcp -45
-45— fcc
-46 ,

I
08 09
I I I I I I I I I I I I I I I I I I I I I

08 10 12 '14 16 08 10 12 14 08 10 12 14
9/Ap A/Ap 0/Ap

-35 -15
-30—
-40— -20
-35—
-45— -25
O -40—
C5

+y -50- -30
CI
4J
-45—
-55— -3.5
-5.0—

-60— -40
5.3
I I I I I
08
I I
09
I
10
I
-5. 5- I I I I I I
09
I
10
I I I I
09
I I
10
I

08 10 12 14 16 0.8 10 1 2 14 16 08 10 12 14 16
A/Ap 0/Ap 0/Ap

FIG. 2. Self-consistent, full-potential LMTO structural energy-volume curves in the 3d transition metals: binding energy (eV) vs
atomic volume 0 normalized to the experimental atomic volume Qp. Structures are fcc, hcp, bcc, simple hexagonal, simple cubic,
and diamond cubic.
 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU 41

TABLE I. Self-consistent, ground-state properties of the 3d metals. Qo is the reference atomic

volume and the experimental cohesive energy E„h and bulk modulus KT are taken from Kittel's book
[C. Kittel, Introduction to Solid State Physics, 5th ed. (Wiley, New York, 1976)].
no E„h (eV/ai, om) ET (Mbar)
Element Structure (A ) 0,„/Qo Theory Expt. Theory Expt.
Sc hcp 25.00 0.92 4.87 3.90 0.6 0.44
T1 hcp 17.65 0.95 5.98 4.85 1.2 1.05
V bcc 13.83 0.94 5.83 5.31 2.0 1.62
Cr bcc 12.01 0.93 4.58 4. 10 2. 8 1.90
Mn hcp 12.21 0.87 4.61 2.92 2.9 0.60
Fe hcp 11.80 0.87 5.90 4.28 3.0 1.68
Co fcc 11.06 0.94 5.96 4.39 2.6 1.91
Ni fcc 10.90 1.00 5.29 4.44 2.0 1.86
Cu fcc 11.81 1.01 3.89 3.49 1.6 1.37

V. THE HARRIS-FOULKES APPROXIMATION

One of the most expensive or time-consuming aspects

E[np]= P
l, OCC
e; — f 5F[np] np+F[np] .

of the density functional method is the need to proceed The Harris-Foulkes (HF) total energy is a functional only
with the calculation to self-consistency. This was the of a trial input density n;„:
motivation for Harris to develop his approximate energy
functional. Independently, Foulkes discovered the same =g — 5F[n;„]
'"
functional in a study of the relationship between density- EH„[n;„]—
I, OCC
f c, , n;„+F[n;„]
functional and tight-binding theories. The Hohenberg-
=E[np]+ , f f—z(n, „np)—
Kohn' total energy is, as usual, separated into a nonin- 5 E[np] '
teracting kinetic energy term T, and an electrostatic term 5n
F, both functionals of the self-consistent charge density
no. X(n, „, np)+—

E[np]=T, [np]+F[np], where Z; and n, „, are the eigenvalues and charge density
obtained from solution of the Kohn-Sham equations with
and if c; are the eigenvalues of the self-consistent Kohn- effective potential 5F[n;„]/5n The F functional is seen H.

Sham problem, then here to have the following properties. (i) It is identical to

-3.0— -2.5— V

-35— -3.0—
cubic Ic

-5.0 (
0
s. cubic
CLl

-5.5 -50—
LU s. hex

-6.0 -5.5—

hcp
-61 —, l l

0.8 0.9 1.0

I I I [ ] I i I l ] I

0.8 0
1. 1 2 1.4 0.8 1 0 1 2 1.4
0/00 0/Ap

FIG. 3. Structural energy-volume curves for Ti and V in the Harris-Foulkes approximation. Axes as in Fig. 2.
41 STRUCTURAL ENERGY-VOLUME RELATIONS IN FIRST-ROW. . . 8133

the Hohenberg-Kohn functional when n;„=no .(ii) For a atomic volumes are generally overestimated; the bulk
given trial density, the HF total energy is equal to the moduli are very similar to the self-consistent values. (We
self-consistent total energy to second order. As a conse- have estimated the bulk moduli at the minimum of the
quence, Enr, is stationary (but not necessarily minimal) at Harris-Foulkes E Vc-urves. ) Absolute cohesive energies
no. are also overestimated in almost all cases. We find this to
The HF approximation is only one in a class of non- be a general trend for the transition metals in the first
self-consistent density-functional methods exploiting the row. This is possible because E~„ is a stationary but not
variational principle. These include the non-self- necessarily variational estimate of the total energy as
consistent procedure based on the Hohenberg-Kohn func- mentioned above. An improvement in the total energy
tional (recently reviewed and compared with the HF can be made by renormalizing the trial atomic densi-
'
approximation ) and methods based on the force ty, ' but for simplicity we have not attempted this here.
theorem. ' The HF approximation is especially appeal-
ing in the present context because of its close connection VI. AXIAL RATIO OF hcp TRANSITION METALS
with tight-binding theory as developed by Foulkes ' and
Sutton et al. Furthermore, the HF approximation has It would have been desirable to minimize the self-
in, the recent past been shown to be remarkably successful consistent total energy in the hcp structures with respect
in reproducing self-consistent calculations of structural to the axial ratio as well as the atomic volume. It has al-
properties of elements and compounds. We have found ready been shown in the case of Si that the full-potential
similarly excellent agreement between self-consistent and LMTO method is capable of accurately reproducing the
HF calculations in the present study of transition metals. change in total energy with lattice distortions, even up to
The procedure we have followed is the same as that of third order in the strain. ' Nevertheless, we have
Polatoglou and Methfessel, except that we remove the chosen to use ideal axial ratios in all metals except Sc and
shape approximation of the ASA. Details can be found Ti, where we have used experimental values. Apart from
in Ref. 60. There is then a close connection between our the expense involved, we find that nonmagnetic Fe, Co,
implementations of the self-consistent and non-self- and Ni are mechanically unstable with respect to a rhom-
consistent problems. Our trial density is a superposition bohedral c/a distortion. This is a Peierls instability
of overlapping, self-consistent free-atom densities. (We driven by the very flat d bands at the Fermi surface
use an s d" atomic configuration. ) From this density, we characteristic of the fcc and hcp band structures. (This
construct an effective Kohn-Sham potential exactly as in effect is of course removed in the magnetic electronic
the self-consistent calculations, following the same pro- structure by shifting the nondegenerate spin bands up
cedure for the solution of the Schrodinger equation. ' and down with respect to the Fermi energy. ) No doubt,
The sum of occupied eigenvalues is added to the "double the fcc structure would also distort into a rhombohedral
counting" terms evaluated for the input density. We may lattice, but because the fcc structure is at a symmetry dic-
choose to stop here or accumulate n, „, and continue to tated energy extremum with respect to rhombohedral dis-
self-consistency to obtain the Hohenberg-Kohn total en- tortion, it is admissible to calculate its total energy. The
ergy. hcp structure with ideal axial ratio is at no such ex-
For brevity, we concentrate on results for Ti and V, tremum, and we therefore tentatively have shown E-V
and show in Fig. 3 their E-V relations, and give in Table curves for hcp Fe and Co (especially since the hcp-fcc
II a detailed tabulation of properties for all six structures. phase transformation in Co is of interest) but not in Ni,
Figure 3 and Table II may be used to compare directly where the instability is strongest. We also remark that in
the HF with the self-consistent results. The equilibrium stable hcp structures, we find the energy change involved

TABLE II. Comparison of equilibrium volume, cohesive energy, and Sulk modulus between self-
consistent (SC) calculations and the Harris-Foulkes (HF) approximation. We use Ti and V as examples
and show results from all six phases studied.

Element Structure SC
n .
/&0
HF SC
Ecph (eV/atom)
HF SC
Ko (Mbar)
HF
fcc 0.91 0.94 5.92 5.97 1.2 1.1
hcp 0.95 0.97 5.98 6.04 1.2 1.1
bcc 0.90 0.92 5.89 5.94 1.2 1.1
sh 0.96 0.97 5.65 5.69 1.0 1.0
sc 0.98 1.00 5. 16 5. 12 0.9 0.8
dc 1.19 1.13 3.48 3.53 0.3 0.3

V fcc 0.98 1.00 5.51 5.69 1.9 1.9

hcp 0.98 0.99 5.50 5.66 1.9 1.9
bcc 0.94 0.95 5.83 6.01 2.0 2.0
sh 1.00 1.01 5.25 3.37 1.6 1.7
sc 1.03 1.03 4.87 4.92 1.4 1.5
dc 1.29 1.25 3.22 3.34 0.8 0.7
8134 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU

in minimizing the axial ratio is of the order of 1 mRy per VII. UNIVERSAL FEATURES OF THE
atom (see also Ref. 26) which is undetectable on the scale ENERGY-VOLUME RELATIONS
of Fig. 2. In noble metals, the variation is almost ten A. The zero-temperature equation of state
times smaller: such a change is unlikely to be sufficient
to stabilize hcp Cu with respect to fcc; however, we have In a recent series of papers, ' by examining a large
not attempted the calculation. database of theoretical and experimental findings, Rose
Notwithstanding the above remarks, some comments and co-workers have found that in a wide variety of in-
on the axial ratio in hcp transition metals are in order. stances, energy-length relationships have a universal

:
Ducastelle and Cyrot-Lackmann have proved (again, as shape. This has become known as the "Rose equation of
described in Sec. IV, using the fourth-moment degenerate state" and takes the form
d-band model) the following remarkable result. Transi-
tion metals for which the fcc-hcp energy difference is neg-
E(a) —E'(a')=p(a')e ' = —(1+a')e
ative will further lower their energy by a reduction in axi-
al ratio. Conversely, in those metals whose fcc phase is where p is a polynomial and
stable with respect to hcp, the hcp structure mould dis- p(a') = —1 —a' —0.05(a')
tort so as to increase its axial ratio. The same result was
found numerically by Finnis, ' ' who calculated the It is claimed that any binding energy-bond length relation
(ddcr), (ddt), and (dd5) bond orders in the canonical of the form E(a) can be cast into the universal form by
d-band model as a function of band filling. In the hcp scaling first the energy with its value at the minimum and
structure, the six bonds parallel to the close-packed then the length a by subtracting the equilibrium length
(0001) planes are inequivalent to the six out-of-plane and dividing by a scaling factor which is chosen to ensure
bonds. Finnis found that the ratio of the in-plane to out- that the second derivative E'"(a')
is equal to unity.
of-plane (ddt ) bond orders was ( 1 around band fillings Rose et al. choose the Wigner-Seitz radius rws as their
of 1 and 2 d electrons (corresponding to Sc and Ti) and 5 measure of length so that a'=(rws rwsz— )/I, where
and 6 (Mn and Fe), and ) 1 elsewhere. The result is that, rwsE is the equilibrium value of res and I turns out to be
since in the tight-binding bond model the interatomic '
1/2
force is proportional to the bond order, in the former
case the interatomic forces would act to shorten the out- 12m. r wsE+o
of-plane bonds and hence lower the axial ratio, and uice
Uersa, exactly as predicted by Ducastelle and Cyrot- Given a set of energy-length relationships, one may scale
Lackmann. the energies until all curves have a minimum in the ordi-
Considering the connection between the HF and tight- nate at —1, shift the curves so that the minimum falls in
binding bond models, it would be disappointing if this the abscissa at zero, and scale the lengths until all curves
effect were not reproduced in the HF approximation. have the same second derivative. It is obvious that in the
Table III shows that for Sc, Ti, and V the prediction is range where the variation of energy with length is quad-
indeed confirmed. Moreover the axial ratios in Sc and Ti ratic, the curves will fall on a universal relationship. It is
are in satisfactory agreement with experiment. less obvious that the universality will extend well beyond
The tight-binding picture of the axial ratio is remark- the harmonic range. Since we have amassed a fairly large
ably well supported in the axial ratios of metastable hcp database of E-V points, it seemed approprate to attempt
metals. The a-c phase transformation in Fe is rather to scale our data to the Rose equation of state, and the re-
sluggish, but on its completion at about 300 kBar sult is shown in Fig. 4. In Fig. 5 we show in several
(II/00=0. 88), the axial ratio is 1.59, having quite
strongly decreased with pressure as the hcp phase be-
comes more stable. hcp Ni is almost certainly
unstable — thin films grown electrolytically on hcp Co
-0.4-
transform to the fcc structure when removed from their
substrates — and its axial ratio ' ' is 1.65
(larger than
ideal 1.633). In Co, the axial ratio is 1.632; this implies
that hcp is marginally stable with respect to fcc, contrary -O. B-

to our LDA result and the speculation of Troiano and

Tokich.
—0.8-
TABLE III. Hexagonal axial c/a ratio for Sc, Ti, and V cal-
culated in the Harris-Foulkes approximation and compared to
the experimental values.

Axial ratio O.8 0 0.4

a~
Element HF Expt.
FIG. 4. Data points from Fig. 2 scaled to the Rose equation
Sc 1.57 1.594 of state; E is the scaled binding energy and a the scaled
T1 1.59 1.587 Wigner-Seitz radius. The solid curve shown is the function
V 1.65 —(1+a *)e -'*.
 STRUCTURAL ENERGY-VOLUME RELATIONS IN FIRST-ROW. . . 8135

fcc fcc fcc

hcp

gbcc Q, dc

OSc g« 0" OSc 3«

Tl
O Q Co

@Mn g Cu

O.S O.3 0. 1 0.2 0.3

FIG. 5. Data from Fig. 4 shown on expanded scales so that each point may be identified with a particular element and crystal
structure. Axes and the solid curve are as in Fig. 4. To avoid showing a lot of empty space, we have split the a axis into panels and
used an optimal E* scale for each panel. In the upper panels the points are labeled according to structure, and in the lower accord-
ing to element. This illustrates in which cases there is deviation from the Rose equation of state.

panels the data on expanded scale —each panel appears Rose et al. did not offer any explanation of their re-
twice, once with the points distinguished by structure and markable result; in particular, they did not associate their
once by element. As expected, the fit is perfect close to length scaling parameter l very convincingly with any
the minimum, and any deviation here is due to errors in physical property of the metals in question. For example,
the least-squares procedure and a small amount of noise replacing l by the Thomas-Fermi screening length led to
in the curves in Fig. 2. It is noticeable that our data devi- a number of anomalies in the universality of the scaling
ate from universality in the high-pressure regime while procedure. A very interesting analysis from the point
the fit is excellent up to very large volumes. Unfortunate- of view of tight-binding theory was made by Spanjaard
ly, the latter range is of little interest in practice. Note and Desjonqueres. They considered a tight-binding
that the deviation at high pressure is mostly confined to model for the d band with transfer integrals scaling with
the non-close-packed structures or to Ni. However, the bond length R like — e and a repulsive Born-
deivations are not such that one can even draw a smooth Mayer —type of interaction of the form e ~". In the
curve through all the points of a given metal. Our data second-moment approximation, the binding energy can
indicate that the universal scaling does not apply to all be written in closed form, having an attractive term pro-
structures of all metals, or even to all structures of a portional to — &Ze and a repulsive term proportion-
given metal; but for one structure of one metal one can al to Ze ~", where Z is the number of neighbors. (This is
find a parameterization of the Rose equation of state that also the motivation behind the Finnis-Sinclair "' intera-
will go through all the points. ' Also the universal tomic potential for transition metals. ) Spanjaard and
curve does fit over a range of a which is significantly Desjonqueres were able to scale their binding-energy ex-
beyond the harmonic range. pression Ez(R) to the form E*(R*), where
R '=(R — Ro)/lsD if Ro is the equilibrium bond length
B. Transferability of tight-binding parameters and lsD related to the Rose et al. scaling length via
lsD/l =I"wsF /Ro. They then found a number of interest-
The fit to the Rose equation of state is good enough ing consequences. (i) The curves of E*(R') were in-
that we may take the analysis a little further. At the dependent of the ratio p /q for all values between 2 and 5,
same time we will renew contact with tight-binding except for deviations at large volumes. (ii} Using
theory and make an assessment of the transferability of p/q =2. 95 the curve of E*(R*) was indistinguishable
the tight-binding model into situations where close pack- from the Rose equation of state. (iii} The scaling pro-
ing no longer pertains (e.g. , vacancies, surfaces, etc. ). cedure results in the following relation between the scal-
8136 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU

ing length and the tight-binding parameters: model for Si gave only moderate success compared to the
1 LDA results for phase stability, and especially quite poor
IsD— reproduction of the stability of close-packed phases.
&w
Although this analysis is only possible in the second- Later, however, Goodwin et aI. used the LDA curves
moment approximation, it would be encouraging if the to actually make a fit to a scaling law for tight-binding
values of lsD for a particular metal were fairly indepen- parameters, leading to a tight-binding model for Si that
dent of structure, since this would imply that a tight- reproduces the LDA results very well.
binding parameterization of p and q might be transfer- A particular criticism of the tight-binding method is
able. In Fig. 6, we show /sD from our calculated E-V that the diagonal Hamiltonian matrix elements are con-
curves, as well as the values obtained form the experi- stant and do not vary with atomic volume or crystal
mental bond lengths, cohesive energy, and bulk modulus. structure. This leads, in Si, to large errors in deformation
We find that, indeed, while lsD varies from metal to met- potentials, for example, and is probably the origin of the
failure of the Chadi-Harrison model to reproduce the
al, in many cases it has a universal value for all structures
LDA results for close-packed structures. In the case of
except the diamond cubic. This does not immediately
transition metals, we can address this question rather
lead to tight-binding parameters, since only the product
more easily. The canonical d-band model results in a
pq is obtained. However, the decay of the transfer in- Hamiltonian which is directly analogous to the first-order
tegrals q can be obtained for transition metals from
quantum-mechnical arguments ' ' ' —
one finds qR 0 LMTO Hamiltonian,
varies between 3 and 5 (its canonical value) across the H„=(C E. ++A;S;JQA, .
„),5„—
transition series. The Born-Mayer exponent is less easy
Here i and j
are composite site and orbital indices,
C E, and —
to estimate, and it is this parameter which could be ob- Vb are standard potential parameters, and S
,
tained from Fig. 6.
is the matrix of LMTO structure constants. ' ' Matrix
We have not made any tight-binding calculations of
elements in the second term, in the usual tight-binding
the structural E-V relations. Therefore we do not know
model, take the Slater-Koster form
whether the usual canonical tight-binding d-band model
is capable of reproducing the LDA results for non-close- —60
(ddo ) 5
packed structures. Having the curves in Fig. 2 available
(ddt) = +40 . Xb
S
should make the problem easier. This, at least, was the d
case in Si. The original Chadi-Harrison '
tight-binding (ddt) —10
in which the bandwidth is 255, d is the bond length, and

0"' s is the Wigner-Seitz radius. The first term C-E corre-

sponds to ed (see Sec. III) and is taken to be the fixed-
energy zero and center of the d band. Therefore, the
0. 8 Q hcp
volume and structure dependence enters only in the vari-
bcc ation of the Slater-Koster parameters. For fixed-volume
distortions, these scale like 1/d, while the volume
Q sh
dependence of 5 (Refs. 70 and 6) is canceled by the s so
0 7. sc that the Slater-Koster integrals also account correctly for
volume dilatations in the tight-binding model. We
dc show in Table IV self-consistent LMTO-ASA d-band po-
tential parameters in vanadium for the fcc, bcc, and sim-
0 6. + Exp t. .
ple cubic structures at the reference atomic volume Qo.
The fact that C E,. and &b, a— re nearly the same in a
wide range of coordination numbers is encouraging, but
the error involved in assuming a volume-independent
0. 5; band center remains to be assessed.
We believe, therefore, that it will be possible to con-
struct tight-binding models for transition metals that will
be transferable. The tight-binding method has many ad-
0 4- vantages: it is simple and transparent, and easy to carry
out in both reciprocal and direct space.

TABLE IV. LMTO d-electron potential parameters for V in

fcc, bcc, and sc lattices.
0. 3-
C —E,, v'g
Sc T I V C r Mn Fe Co N t Cu Structure (Ry)' (R I/2)
FIG. 6. The Spanjaard and Desjonqueres scaling parameter
0 fcc 0. 178 0. 146
I» (A) calculated from data in Fig. 2 for each structure and for bcc 0. 184 0. 146
each element. Estimates from experimental cohesive energy,
sc 0.203 0. 171
atomic volume, and bulk modulus are shown as crosses.
41 STRUCTURAL ENERGY-VOLUME RELATIONS IN FIRST-ROW. .. 8137

VIII. DISCUSSION AND CONCLUSIONS example, Refs. 75 and 76).]

Little mention has been made of the class of empirical
We have presented a systematic study of the structural classical potentials such as the embedded-atom, Finnis-
energy-volume relations of first-row transition metals in Sinclair, and "glue" models, although we believe our
the local-density approximation. This serves as a demon- energy-volume curves will be of some use in constructing
stration of the full-potential LMTO method' ' as well as such models in the future. These are all essentially of the
presenting a useful database for the field testing of intera- same form, although their derivations are quite different.
tomic potentials. The analysis has concentrated on This has led to some confusion in evaluating their range
the connection to simpler models and non-self-consistent of applicability. One could extend the hierarchy given in
calculations; particular emphasis has been placed on the previous paragraph one stage further by adding clas-
the hierarchy self-consistent LDA~non-self-consistent sical potentials as derivatives of the tight-binding scheme.
LDA~semiempirical tight binding. This connection has This is implicit in the derivation of the Finnis-Sinclair po-
already been made in the literature, but only on a formal tentials, although the embedded-atom method is thought
level. For example, Sutton et al , wh. ile presenting the to originate directly from self-consistent density-
detailed justification of their tight-binding bond model functional theory. The "glue" model, on the contrary,
using the Harris-Foulkes stationary functional, did not was constructed on a wholly ad hoc basis, but nonetheless
show comparisons between their bond, promotion and it has been no less successful than the other two
repulsive energies as estimated from empirical or canoni- theories. ' We do not know whether these classical po-
cal models, and the same quantities calculated exactly in tentials can be given a firmer footing, nor whether the re-
the LDA. We have used the Rose equation of state sults presented here will be useful in doing so.
and the Spanjaard and Desjonqueres tight-binding
analysis to use our E- V database to address the question ACKNOWLEDGMENTS
of transferability of tight-binding parameters. We have
also argued that since the HF approximation is capable We thank O. K. Andersen, N. E. Christensen, M. W.
of reproducing very accurately the self-consistent calcula- Finnis, and A. Liechtenstein for many helpful conversa-
tions of static structural properties of transition metals, tions. H. Skriver kindly supplied details of his calcula-
on the basis of the arguments of Sutton et al. it should tions. We are grateful for financial support from the
be possible to find a tight-binding model suitable for de- Max-Planck Gesellschaft, the UK Science and Engineer-
fect calculations. [This is already known by experience; ing Research Council, and the U. S. Air Force Once of
there have been numerous successful calculations of de- Scientific Research (AFOSR) under Contract No.
fect and surface structures of transition metals (see, for F49620-85-K-0009.

'Present address: SRI International, 333 Ravenswood Avenue, '5P. Bagno, O. Jepsen, and O. Gunnarsson, Phys. Rev. B 40,
Menlo Park, California 94025. 1997 (1989).
Permanent address: Department of Physics, Aristoteles Uni- ' O. K. Andersen, Phys. Rev. B 12, 3060 (1975).
versity of Thessaloniki, GR-54006, Greece. ' M. Methfessel, Phys. Rev. 8 38, 1537 (1988).
'P. C. Hohenberg and W. Kohn, Phys. Rev. 136, B864 (1964). ' O. K. Andersen, Solid State Commun. 13, 133 (1973).
W. Kohn and L. Sham, Phys. Rev. 140, A1133 (1965). 'sO. K. Andersen, in Electronic Structure of Complex Systems,
J. Harris, Phys. Rev. B 31, 1770 (1985). edited by P. Phariseau and W. M. Temmerman (Plenum, New
4W. M. C. Foulkes, Ph. D. thesis, University of Cambridge, York, 1984), p. 11.
1987. D. J. Chadi and M. L. Cohen, Phys. Rev. B S, 5747 (1973).
5W. M. C. Foulkes and R. Haydock, Phys. Rev. B 39, 12520 L. F. Mattheiss and D. R. Hamann, Phys. Rev. B 33, 823
(1989). (1986).
D. G. Pettifor, J. Phys. F 7, 613 {1977). V. L. Moruzzi„J. F. Janak, and A. R. Williams, Calculated
7A. P. Sutton, M. W. Finnis, D. G. Pettifor, and Y. Ohta, J. Electronic Properties of Metals (Pergamon, New York, 19781.
Phys. C 21, 35 (1988)
~ Zhi-Wei Lu, D. Singh, and H. Krakauer, Phys. Rev. B 36,
M. W. Finnis, in Chemistry and Physics of Fracture, edited by 7335 (1987).
R. M. Latanision and R. H. Jones (Nijhoff, Dordrecht, 1987), 24C. -L. Fu and K.-M. Ho, Phys. Rev. B 28, 5480 (1983).
p. 177. 25N. Takeuchi, C. T. Chan, and K.-M. Ho, Phys. Rev. B 40,
M. W. Finnis, A. T. Paxton, D. G. Pettifor, A. P. Sutton, and 1565 (1989).
Y. Ohta, Philos. Mag. A 58, 143 (1988). B. J. Min and K.-M. Ho, Phys. Rev. B 40, 7532 (1989).
'oJ. W. Christian, The Theory of Transformations in Metals and M. Methfessel, C. O. Rodriguez, and O. K. Andersen, Phys.
Alloys, 2nd ed. (Pergamon, Oxford, 1975), Ch. 5. Rev. B 40, 2009 (1989).
"Many-Atom Interactions in Solids, edited by R. M. Niemenen, M. Methfessel and A. T. Paxton, Phys. Rev. B 40, 3616 {1989).
M. J. Puska, and M. Manninen (Springer, Berlin, 1990). D. G. Pettifor, J. Phys. C 3, 367 (1970).
'2M. W. Thompson, Defects and Radiation Damage in Metals 30F. Ducastelle and F. Cyrot-Lackmann, J. Phys. Chem. Solids
(Cambridge University Press, Cambridge, England, 1969). 32, 285 {1971).
' V. Vitek, Crystal Latt. Def. 5, 1 (1974). 'D. G. Pettifor, in Metallurgical Chemistry, edited by O. Ku-
A. P. Sutton and V. Vitek, Philos. Trans. R. Soc. London, Ser. bachewski (Her Majesty's Stationery Office, London, 1972), p.
A 309, 1 (1983). 191.
 A. T. PAXTON, M. METHFESSEL, AND H. M. POLATOGLOU 41

O. K. Andersen and O. Jepsen, Physica B+C (Amsterdam) the experimental atomic volume. This paper is primarily con-
91B, 317 (1977). cerned with models for transition metals which may be ap-
S. Glanville, A. T. Paxton, and M. W. Finnis, J. Phys. F 18, plied, for example, in moleuclar-dynamics calculations; there-
693 (1988). fore it is appropriate to discuss equilibrium features of the
34J. Friedel, Trans. AIME 230, 616 (1964). model rather than to obtain the best agreement with experi-
G. Allan, Ann. Phys. (Paris) 5, 169 (1970). ment.
J. Friedel, J. Phys. Paris 39, 651 (1978). J. R. Chelikowsky and S. G. Louie, Phys. Rev. B 29, 3470
D. G. Pettifor, in Solid State Physics, edited by H. Ehrenreich, (1984).
F. Seitz, and D. Turnbull (Academic, New York, 1987), Vol. M. W. Finnis, J. Phys. Condens. Matter 2, 331 (1990).
40, p. 43. OH. M. Polatoglou and M. Methfessel, Phys. Rev. B 41, 5898
J. C. Slater and G. F. Koster, Phys. Rev. 94, 1498 (1954). (1990).
H. Skriver, Phys. Rev. B 31, 1909 (1985). M. Methfessel and J. Kubler, J. Phys. F 12, 141 (1982).
~O. K. Andersen, O. Jepsen, and D. Glotzel, in Highlights of H. M. Polatoglou and M. Methfessel, Phys. Rev. B 37, 10403
Condensed Matter Theory, edited by F. Bassani et al. (North- (1988).
Holland, Amsterdam, 1985), p. 59. M. Weinert and R, E. Watson, Phys. Rev. B 29, 3001 (1984).
~~A. T. Paxton, in Atomistic Simulation of Materials, edited by ~J. G. Wright and J. Goddard, Philos. Mag. 11, 485 (1965).
V. Vitek and D. J. Srolovitz (Plenum, New York, 1989), p. J. H. Rose, J. Ferrante, and J. R. Smith, Phys. Rev. Lett. 47,
327. 675 (1981).
A. K. Sinha, Prog. Mater. Sci. 15, 81 (1972). J. H. Rose, J. R. Smith, F. Guinea, and J. Ferrante, Phys.
43F. C. Frank and J. S. Kaspar, Acta Crystallogr. 12, 483 (1959). Rev. B 29, 2963 (1984), and references therein.
~J. S. Kaspar and B. W. Roberts, Phys. Rev. 101, 537 (1956). D. Spanjaard and M. C. Desjonqueres, Phys. Rev. B 30, 4822
45L. D. Blackburn, L. Kaufman, and M. Cohen, Acta Metall. (1984).
13, 533 (1965). M. W. Finnis and J. E. Sinclair, Philos. Mag. A 50, 45 (1984).
R. L. Clenendon and H. G. Drickamer, J. Phys. Chem. Solids G. J. Ackland, M. W. Finnis, and V. Vitek, J. Phys. F 18,
25, 865 (1964). L153 {1988).
47C. S. Wang, B. M. Klein, and H. Krakauer, Phys. Rev. Lett. V. Heine, Phys. Rev. 153, 673 (1967).
54, 1852 (1985). 'F. Ducastelle, J. Phys. (Paris) 31, 1055 (1970).
D. C. Langreth and M. J. Mehl, Phys. Rev. Lett. 47, 446 D. J. Chadi, Phys. Rev. Lett. 41, 1062 (1978).
(1981). W. A. Harrison, Phys. Rev. B 27, 3592 (1983).
A. R. Troiano and J. L. Tokich, Trans. AIME 175, 728 (1949). In Si the bandwidth scales like 1/d' so that the Slater-Koster
C. Bonelle and F. Vergand-Jacquot, C. R. Acad. Sci. Paris parameters must have the same length dependence. Unfor-
252, 1448 (1961). tunately these are not short-ranged, as they are in metals. On
P. Hemminger and H. Weik, Acta Crystallogr. 19, 690 (1965). the other hand, short-ranged tight-binding parameters do not
I. Teodorescu and A. Glodeanu, Stud. Cere. Fiz. Romin. 112, benefit from this cancellation and require an explicit volume-
331 (1960). dependent correction (see Refs. 40, 41, and 55).
M. T. Yin and M. L. Cohen, Phys. Rev. B 26, 5668 (1982). B. Legrand, Philos. Mag. A 52, 83 (1985).
5~R. Biswas and D. R. Hamann, Phys. Rev. Lett. 55, 2001 B. Legrand, G. Treglia, M. C. Desjonqueres, and D. Span-
(1985). jaard, J Phys. C 19, 4463 (1986).
~

55A. T. Paxton, A. P. Sutton, and C. M. M. Nex, J. Phys. C 20, M. S. Daw and M. I. Baskes, Phys. Rev. B 29, 6443 (1984).
L263 (1987). 78F. Ercolessi, E. Tosatti, and M. Parrinello, Phys. Rev. Lett.
L. Goodwin, A. J. Skinner, and D. G. Pettifor, Europhys. 57, 719 (1986).
Lett. 9, 701 (1989). F. Ercolessi, M. Parrinello, and E. Tosatti, Philos. Mag. A 58,
Overestimation of Ko is a consequence of the underestimation 213 (1988).
of atomic volume. The error is lessened if Ko is calculated at