MANY-BODY CENTRAL FORCE POTENTIALS

AND EMBEDDED ATOM METHOD

In pair potential schemes the potential energy of N atoms is

E p = U(!) +

1
$ "(rij)
2 i,j
i# j

where !(rij ) is the pair potential describing the interaction between the atoms i and j,
separated by the distance rij , and U() is the density dependent part of the energy in
which is the average density of the material. As explained earlier such pair
potentials belong to the class of constant volume potentials and may describe
reasonably well energy changes associated with the variation of atomic
configurations at constant volume. However, in many situations the density varies
locally quite significantly (surfaces, vacancies, interfaces, cavities) and this local
variation affects considerably the potential energy. Thus an important problem is to
take into account the local density changes in empirical schemes.
This can be done by regarding as the local density evaluated at the positions of
individual atoms, which is then a function of the positions of neighboring atoms.
This is the basis of the embedded atom method (EAM) and other many-body central
force potentials. In the framework of these schemes the potential energy of the
system of N atoms is written as

Ep =

1
2

# !(rij ) + # F($i )
i,j
i" j

(EAM1)

where F(!) is the embedding and/or many body function. It is evaluated at positions
of individual atoms and " F(!i ) replaces U() and the density !i varies from atom
i

to atom. This part of the energy is attractive, similarly as U(), and provides for the
bonding, while the pair potential part, 12 # !(rij ) , is mainly repulsive, strongly at short
i, j
i" j

separations of atoms. !i is then regarded as a linear superposition of contributions of

neighboring atoms to the density at the position of the atom i, so that

!i = $ "k (rik ) ,
k# i

(EAM2)

!k (r ik ) , which is a function of the separation of the atoms i and k, rik , is interpreted

within the EAM as the contribution of the atom k to the density (or electron density)

at the position of the atom i. In other schemes the physical interpretation may be
somewhat different but since this approach is empirical, this is not essential. In any
scheme !(r) is always a rapidly decreasing monotonous function of the separation of
particles. Since both and are only functions of the separation of pairs of
particles the scheme is a central-force scheme, albeit not a pair potential scheme. The
bonding part " F(!i ) has many-body character and it is not a simple pair potential.
i

In the scheme of many-body central-force potentials the force acting on an atom i is

#E p

d$(rik ) rik! ( dF +
F = " ! = "&
"
#ri
drik rik *) d' -, '='
k %i
i
!
i

d. k (rik ) rik!
& dr r
k %i
ik
ik

/( dF +
d. i (rik ) rik! 2
" & 1* 4
dr
r
k %i 1 ) d' , '='
43
ik
ik
0
k

(EAM3)

The force exerted on the particle i by the particle k can then be defined as

Fik!

d#(rik ) rik!
="
drik rik

% dF (
d+k (rik ) rik!
" ' *
& d$ ) $=$i drik rik

% dF (
d+ i (rik ) rik!
" ' *
(EAM4)
& d$ ) $=$k drik rik

If we assume that all the atoms are the same type of species, implying that
!i = !k = ! , the force exerted on the particle i by the particle k becomes

. d1(r ) r!
d#(rik ) rik! + % dF (
% dF (
ik
ik
F ="
" -' *
+' *
0
drik rik - & d$ ) $= $ & d$ ) $=$ 0 drik rik
,
i
k /
!
ik

(EAM5)

The atomic level stress evaluated at the position of the atom i can again be calculated
according to equation (G29) and the atomic level pressure according to equation
!
(G30) in which Fik is given by (EAM4) or (EAM5). The total stress and pressure are,
of course, obtained by summation of the atomic level stresses as in equations (G15a,
b; G18).
Embedding or Many Body Function
The contribution of the embedding function towards the potential energy is always
negative and thus it represents the cohesive part of the potential energy, similarly as
does U() in the case of pair potentials. The repulsion at short separations of atoms is
described by the pair potential. The general form of the function F in all schemes of
2

many-body central-force potentials is shown in Figure 1. This function approaches

zero quickly as the density, , becomes very small. This means that bonding is
decreasing rapidly as the particles separate. In contrast, when the particles are
compressed and the density, , is increasing, the cohesion increases because F is
negative and increases in magnitude to infinity as ! " # . The overall form of this
function has been deduced using arguments based on the density functional theory or
the tight-binding method.

Fig. 1. General form of the embedding function F(!) .

The main difference between various many-body central force potentials is in the
choice of the embedding function, F, evaluation of the local densities, !i , and
physical justifications of these choices.

Embedded atom method (EAM)

(Daw and Baskes 1984, Daw, Foiles and Baskes 1993, Foiles 1996).
The form of F is justified by arguments based on the density functional theory (DFT).
Specifically, since within DFT the energy of a system of electrons is a functional of
the electron density, the energy of the system of atoms bounded together by electrons,
is regarded as a function of the local electron density that can be taken as proportional
to the local density of the material. At the position of an atom i the electron density is
given by equation (EAM2) where ! k (rik ) are chosen functions representing
contributions to the electron density at the atom i from electrons associated with the
atom k. For example, !k (r ik ) can be taken as the square of the s-electron wave
3

function centered on the atom k evaluated at the position of the atom i. The
embedding function is then given numerically (in the form of a Table) and is fitted,
together with the pair-potential !(r) , so as to reproduce as closely as possible the
following equilibrium properties of the corresponding material:
Equilibrium lattice parameters
Cohesive energy
Elastic moduli
Other possible fitted quantities are:
Vacancy formation energy, phonon spectra, universal binding curve (described
below), and other data, usually obtained by ab initio DFT calculations.
In the case of alloys the alloying and ordering energies may be fitted.

Finnis-Sinclair potentials
(Finnis and Sinclair 1984, Ackland and Finnis 1986, Ackland et al.1987)
The form of F is justified in the framework of the tight-binding calculation of the
energy within the second moment approximation of the density of states together with
the condition of charge neutrality (Ackland et al. 1988). In this model F is analytic
and equal to minus square root. Hence, the potential energy of the system of N
atoms is in this case

Ep =

1
2

# !(r ) $ #
ij

i, j
i" j

%i

The physical meaning of !i is then taken to represent the sum of the squares of the
bond (hopping) integrals between the atomic sites i and k. In practice we again write
!i according to equation (EAM2) but now !k (r ik ) is a short-range pair-potential that
is fitted, together with !(r) , so as to reproduce as closely as possible the same
quantities as in the EAM method.
The force exerted on the particle i by the particle k is in this case

d#(rik ) rik! 1 '

*
F ="
+ 2 ) & $ n (rin ),
drik rik
( n %i
+

"1/2

!
ik

'
*
+ 12 ) & $ n (rnk ),
( n %k
+

"1/2

d$ i (rik ) rik!
drik rik

d$ k (rik ) rik!
drik rik

(EAM6)

Both EAM and Finnis-Sinclair potentials can be generalized to alloys (Ackland and
Vitek 1990). In the case of the Finnis-Sinclair potentials and a binary alloy the
energy of a system composed of N atoms is
4

&
)
1
(
E p = % 2 % !S iS k (rik ) " #S i + with #Si = % , SiS k (rik )
+
i=1 ( k $i
k$ i
'
*
N

(EAM7)

where suffices i and k refer to individual atoms and the suffices Si and Sk refer to the
species of the atoms involved. Both and are empirically fitted pair-potentials
and summation over k extends over all those neighbors of the atoms i for which rik is
within the cut-off radii of these potentials. Again, the first term is a sum of pair
potential interactions and the second term the many-body attractive part of the
cohesive energy.
For the binary systems there are three pair-potentials AA, BB, AB and three
potentials AA, BB and AB. Functions AA, BB, AA and BB are usually
taken to be the same as those for the corresponding pure elements, and
!AB = !AA !BB , which is congruent with its interpretation in terms of bond
integrals within the tight-binding scheme. The pair potential AB is then fitted to
alloy properties.

Effective pair potential visualizing the many-body interactions

In general, the interaction between the atoms described by EAM or FinnisSinclair potentials cannot be visualized so easily as in the case of pair potentials since
the energy is not simply an interaction between pairs of atoms but it has a many-body
character. This means, for example, that the force acting on an atom i due to an atom
k depends also on where all the other atoms are positioned. However, if we assume
that for all atoms deviates only little from a value, !0 , attained for a reference
structure, for example the ideal lattice, the force acting on an atom i can be written
approximately as

Fi!

d#(rik ) rik!
' dF *
d-i (rik ) rik!
= "%
"2) ,
%
( d& + &= &0 k$ i drik rik
rik
k$ i dr ik

(EAM8)

The same force would be obtained if we assumed that the atomic interaction is
described by a pair potential

$ dF '
! eff = !(r) " 2 & )
*(r)
% d# ( #= #0

(EAM9)

This potential is called the effective pair potential related to a given many-body
central force potential scheme and can be used as a way of visualizing the many body
interaction. However, the forces and stresses evaluated using the full scheme are not
5

described well by this potential if the structures deviates far away from the reference
structure, which is usually the ideal lattice.
Within the Finnis-Sinclair scheme the effective pair potential is
1
(EAM10)
! eff = !(r) "
$(r) .
#0
and effective pair-potential corresponding to the interaction between the two different
species S1 and S2 can be defined as follows:

$ c
c S2 '
S1
(EAM11)
( r) = ! S1S 2 ( r ) " && o + o )) *S1S2 ( r)
#S2 (
% #S1
o
are concentrations of species S1 and S2, respectively, and !S1 and

! Seff1S 2
where c S1 and c S2

!oS2 are evaluated for a reference structure. Examples of such potentials are potentials
for the Cu-Bi system constructed in the context of the study of Bi segregation to grain
boundaries in Cu (Yan et al. 1993), shown in Fig. 2.

0.4
Bi-Bi
Cu-Cu
Cu-Bi

! eff (eV)

0.3
0.2
0.1
0.0

-0.1
-0.2
1.5

2.0

2.5

3.0 3.5
r ()

4.0

4.5

Fig. 2. Effective pair potentials for the Cu-Bi alloys.

5.0

Force constants matrix

!2 E p

- & d2 $(rik ) 1 d$(rik ) ) rik" rik#

0
1 d$(rik )
=
%
+
,
/
"
#
2
2
"# 2 , ij, k!
!rik !rj! . (' drik
rik drik +* rik
rik drik
1
& d2F )
+( 2+
' d3 *

3=3i

- d4(ri! ) ri!#
d4(rji ) rji#
0 d4(rik ) rik"
, ij +
,i! 2
/
drji rji 21 drik rik
/. dri! ri!

- d4(rk! ) rk!#
d4(r jk ) rjk#
0 d4(rik ) rik"
& d2F )
+( 2+
,
+
,
/
kj
k! 2
' d3 * 3=3 /. drk! rk!
drjk rjk
21 drik rik
k

(EAM12)

- & dF )
0 - & d 2 4(r ) 1 d4(r )) r" r#
0
& dF )
1 d4(rik )
ik
ik
ik ik
+ /( +
+( +
2 /(
%
+
,
2 , ij, k!
"#
2
2
+* r
'
d3
*
'
d3
*
dr
r
dr
r
dr
'
/.
ik
ik
ik
ik
ik
ik
3=3 i
3=3k 2
1
1.
This matrix is needed when analyzing the stability of a structure (see equation G6),
calculating phonon spectra and elastic moduli (see equation AG4).

Comments on many-body central force potentials

During the construction of the potentials, when fitting a variety of properties,
equilibrium data is used almost exclusively. However, the potentials are then
employed in calculations of structures in which the local environment deviates very
significantly from equilibrium. This raises an important question of transferability
of the potentials which is not necessarily guaranteed by reproduction of equilibrium
properties. While the extent of transferability of empirically constructed potentials
cannot be deduced from any general arguments, it is likely to be wide enough if the
potentials constructed for a given system ensure:
(a) The stability of the appropriate crystal structure relative to alternate structures
with different symmetry and possibly different chemical order.
(b) The mechanical stability of this structure relative to both small (positive elastic
moduli) and large homogeneous deformations.
(c) Correct vibration (phonon) spectra.
(d) In the case of alloys, reproducibility of the basic features of the phase diagram.
The reason is that unphysical structural or ordering instabilities are then much less
likely to occur in the cores of defects. Nevertheless, the validity of the structural
7

features of lattice defects found in calculations using such potentials is best

guaranteed if they can be related to fitted material properties and are not sensitively
dependent on the details of the potentials.

References
Ackland, G. J. and Finnis, M. W., 1986, Philos. Mag. A 54, 301.
Ackland, G. J., Tichy, G., Vitek, V. and Finnis, M. W., 1987, Philos. Mag. A 56, 735.
Ackland, G, J, Finnis, M. W. and Vitek, V., 1988, J. Phys.: Metal Physics 18, L153.
Ackland, G. J. and Vitek, V., 1990, Phys. Rev. B 41, 10324.
Daw, M. S. and Baskes, M. I., 1984, Phys. Rev. B 29, 6443.
Daw, M. S., Foiles, S. M. and Baskes, M. I., 1993, Mater Sci Rep 9, 251.
Finnis, M. W. and Sinclair, J. E., 1984, Philos. Mag. A 50, 45.
Foiles, S. M., 1996, MRS Bull 21, 24.
Yan, M., Sob, M., Luzzi, D. E., Vitek, V., Ackland, G. J., Methfessel, M. and
Rodriguez, C. O., 1993, Phys. Rev. B 47, 5571.

APPENDIX
UNIVERSAL EQUATION OF STATE
In a fluid or gas the equation of state is a relationship between pressure, volume and
temperature. In the solid the full equation of state is a relation between the stress tensor
! "# , strain tensor ! "# and temperature. For example, the Hooke's law is an equation of
state for small elastic strains at constant temperature. Nevertheless, we can still ask the
question what is the relationship between the hydrostatic pressure, p, volume, V and
temperature T. The relationship p = p(V, T) is frequently called the equation of state of
the material although it is not a complete equation of state.
An interesting finding, although not theoretically unequivocally proved, is that the
equation of state p = p(V, T), when suitably scaled, is the same for many different
materials.
The same applies to the dependence of the binding energy on volume (p ! binding
energy) or to the dependence of adhesion on separation of the two materials
(p ! adhesive strength, V ! separation of the two materials), chemisorption etc.
Scaling
Wigner-Seitz radius, rws
Wigner-Seitz (W-S) radius is the radius of the sphere the volume of which is equal to
the atomic volume ! a
4! 3
r = "a
(U1)
3 ws
e
The equilibrium W-S radius, rws
, corresponds to the equilibrium atomic volume ! ea .
All the dependencies of physical quantities on ! a or, equivalently, on rws , will be
e
shown as functions of rws ! rws
.

Binding energy,
The binding energy, E b , is defined as the total energy of the system minus the energy of
the same number of isolated atoms. The equilibrium binding energy, E eb , corresponds
to the minimum of the total energy. The binding energy will be scaled by E eb .

10

Scaling factor for lengths in the E b vs rws dependence is

!
$ 1/2
E eb
&
! = ## 2
e
2 &
" d E b (r ws ) drws %

(U2)

It is related to the bulk modulus, B, since

B=
and, therefore,

"

Eb %
1 "d 2E b %
' =
$ 2 ''
2 '
e $
# d!a & e 12(rws # drws & e

d
!ea $$

" E e %1/2
b
'
! = $$
'
# 12!rws B &

Definition of the scaled separation:

(U3)

r !r
a = ws ws
!
e

(U4)

Universal dependence of the binding energy on volume per atom

Eb
vs a is the same function g(a) for a very wide variety of materials.
E eb
This is demonstrated in Fig. A1. Hence, we can write for many different materials
It appears that

E b (rws ) = E eb g(a) .

(U5)

It was shown, again empirically, that g(a) can be well approximated as

g(a) = !(1+ a + 0.05a 3 )e !a

(U6)

The equation of state at 0K is p = ! "E b "V and using the above relations we obtain

p
=
B

1/3
#
&
3%(! a !ea ) "1(
$
'

( !a

2/3
! ea

(1 " 0.15a + 0.05 a 2 )e" a

(U7)

This equation was again shown to be in a very good agreement with experiments for
many materials. Examples are shown in Fig. A2.
10

Owing to the generality of the

11

dependencies

Eb
vs a
E eb

and

p
vs a
B

these

dependencies,

when

determined

experimentally, represent a good database for fitting of empirical schemes of the

description of atomic interactions, in particular the many-body central-force potentials.

Fig. A1 Scaled binding energy vs a for six different materials

Fig. A2. Dependence of p vs volume calculated according to (U7) compared with

experimental data for several materials
11