PHYSICAL REVIEW

VOLUM E 140, NUM B ER 4A

15 NOVEM B ER 1965

Self-Consistent Equations Including Exchange and Correlation Effects*

W.
Unieersity
KOHN AND

of Ca/Bfornia, San Diego, la Jolta, California (Received 21 June 1965l

L. J.

SHAM

From a theory of Hohenberg and Kohn, approximation methods for treating an inhomogeneous system of interacting electrons are developed. These methods are exact for systems of slowly varying or high density. For the ground state, they lead to self-consistent equations analogous to the Hartree and Hartree-Fock equations, respectively. In these equations the exchange and correlation portions of the chemical potential of a uniform electron gas appear as additional effective potentials. (The exchange portion of our effective potential differs from that due to Slater by a factor of -';.) Electronic systems at finite temperatures and in magnetic lelds are also treated by similar methods. An appendix deals with a further correction for systems with short-wavelength density oscillations.

I. INTRODUCTION
'N recent years a great deal of attention has been - - given to the problem of a homogeneous gas of interacting electrons and its properties have been established with a considerable degree of confidence over a wide range of densities. Of course, such a homogeneous gas represents only a mathematical model, since in all real systeins (atoms, inolecules, solids, etc. ) the electronic density is nonuniform. It is then a matter of interest to see how properties of the homogeneous gas can be utilized in theoretical studies of inhomogeneous systems. The well-known methods of Thomas-Fermi' and the Slater' exchange hole are in this spirit. In the present paper we use the formalism of Hohenberg and Kohn' to carry this approach further and we obtain a set of self-consistent equations which include, in an approximate way, exchange and correlation effects. They' require only a knowledge of the true chemical potential, tie(e), of a homogeneous interacting electron gas as a function of the density n. sets of equations We derive two alternative [Eqs. (2.8) and (2.22)) which are analogous, respectively, to the conventional Hartree and Hartree-Fock. equations, and, although they also include correlation effects, they are no more difficult to solve. The local effective potentials in these equations are unique in a sense which is described in Sec. II. In particular, we And that the Slater exchange-hole potential, besides its omission of correlation effects, is too large by a factor of , Apart from work. on the correlation energy of the homogeneous electron gas, most theoretical many-body studies have been concerned with elementary excitations and as a result there has been little recent progress in the theory of cohesive energies, elastic constants, etc. , of real (i.e., inhomogeneous) metals and alloys. The methods proposed here offer the hope of new progress in this latter area. ~ Supported in part by the U. S. Ofhce of Naval Research. 'L. H. Thomas, Proc. Cambridge Phil. Soc. 23, 542 (1927);

In Secs. III and IV, we describe the necessary Inodidcations to deal with the finite-temperature properties and with the spin paramagnetism of an inhomogeneous electron gas. Of course, the simple methods which are here proposed in general involve errors. These are of two general origins4: a too rapid variation of density and, for 6nite systems, boundary effects. Refinements aimed at reducing the 6rst type of error are brieQy discussed in Appendix II.

II. THE

GROUND STATE

A. Local Effective Potential

It has been shown' that the ground-state energy of an interacting inhomogeneous electron gas in a static potential n(r) can be written in the form
Z=
where

tt(r)e(r) dr+

e(r)e(r')
r

dr dr'+G[e), r'[ (2.1)

i:, ,

e(r) is the density and G[e) is a universal functional of the density. This expression, furthermore, is a minimum for the correct density function e(r). In this section we propose first an approximation for G[e), which leads to a scheme analogous to Hartree's method but contains the major part of the effects of exchange and correlation. We first write

'.

G[e) = T.[e)yZ, [e),

where

(2.2)

T,[e) is the kinetic energy of a system of noninteracting electrons with density e(r) and [e) is, by our definition, the exchange and. correlation energy of an interacting system with density e(r). For an arbitrary e(r), of course, one can give no simple exact expression for [e). However, if e(r) is sufliciently slowly varying, one can show' that

F,

E,

F,[e)=
4
~

e(r)e, (e(r)) dr,

(2.3)

E. Fermi, Z. Physik 48, 73 (1928). ' J. C. Slater, Phys. Rev. 81, 385 (1951). ' P. Hohenberg and W. Kohn, Phys. Rev. 136, 3864 (1964l;
referred to hereafter as HK.

W. Kohn and L. J. Sham, Phys. Rev. 137, A1697 (1965). For such a system it follows from HK that the kinetic energy is in fact a unique functional of the density.

1138

A &&34

AN

SH~M
it.z, ra ius d. ro a yp ical lengtth over which ' n is an appreciab l e change in denssit . In ts se as ty nd can expan the true exchange and ene gy a

where &x.(e)

electron of a uniform l matipn con apprpxima tutes ana e q uaterep . lation e cts in the
gene

istheexcha n g e and corre prrelation ene gy P el . eiectron gas f densltyn. O ur sole e ts pf assum~~g that 2 3) constiresentatipn pf exchang o . nd corres stems un e consl era tion. sy I,npwn from theories n the homo-

eiectron g ' From t h statio ary prop rty of Eq' obtain, subject to th, odition e

F,fej=

'

.(e)e. dr
c

2. 1) we now

('&(e))Vn('dr+

~,

(2.11)

be(r)
the equation

dr=0,
0 =,

(2.4)
expand

the

e and correlation p ortion o n rg ansion in powers o x pansi

T, e
,

in

be(r)
here

q(r)+
~(r) = ~(r)+

t(. n r

())
dr',

(2. )

T, Ln) =

', (3x'e)'"e dr
t('&(n)iVei'dr+

e(r')

(2 6)

. (2.12) .2

and

is the exc

correlation

'

From HK expecially Sec. III 2, we h have the following , expression for the energy. o

to th

)
o e obtain
om

+
dr+
2&'&(e) g2

e
Therefore
h&ch

po
p~

or g sat&sfies these e uat

go(e)

[Ve)'dr+

(2.13)
(2.14)

si

3m'e)'"+e

e . e) }e,

and setting

6,('& (n)+t(" (n) ) e.

e(r)=Z
iv

i=1

' '(r) I',

'g

'y
of the order
l

is the num er o

E. (2.8) as
g

an add't'

ii

ff

ou

re determines t e e of o d ivi4.

. ) Equatio ns (2.6 .9) h (2.

f

familiar from

pp

to o

and that

ne begins wi ed mr, 2.6) an (i pxc i' o (2. and fin s ane . 2. (2.8) and (2.9 . The energy is g'iven y

7,

Z=P c,
1

r' ir
n r
e (n(r)) .(e(r))g dr. e, n r t(. r r.

ter's exc ange ' ' eHartree- -Fock w writing the f equivalent po en
function

begin by o erator in the form

(2.10)
w.

,(r)=- Z
m ts

.*(r)A '(r')4' (r)A(r'

r'[ [r

The results of ou r p roce d u re are exact in two limiting ct w cases.

ized by
n r, ro~(1, where 6F r a review see D. Pines, elementaryr . in w (W. A. Benjamin, Inc. , ew
'

A*(r)4'(r)
to add corre a
s 9

. (2. 16)

'

ensi

P',

r, is the

ime is c a racteri ' ' 'n Solzds x

ter there have been

' R Schrieffer P y on Phys. Rev. 136, and C W. UGord, P h ys. R 139

SELF CONSISTENT EQUATIONS

where the symbols r and r' are understood to include electron spin coordinates and integration is understood to include summation over spin coordinates. One next assumes that the wave functions can be approximated by plane waves which results in

1135

krs(r) kr(r) wi, (r) = 1+

2kkr

s k

ln

r k+kr ()
k

(r)

kr(r)

(2.17)
v
&

where

= kr(r){3s rt(r)}Us. Finally, one averages over the occupied state k, which results in
n(r) = (3/2')

{3s'I(r) }'"

(2.18)

In our procedure (neglecting correlation) we obtain, in place of Slater's v

t

*()= /~){ ~'I()}'" (

~

(2

', . smaller by a factor of From the discussion in Appendix I, it follows that while p gives the exchange correction of the density correct to order V~', inclusive, s t as indeed any other function of e(r)] leads to errors of order V ~'. The same comment applies to any extension of Slater's exchange to include correlation in the selfconsistent potential. We may note that our result is equivalent to taking, not the average of (2.17), but rather its value at k=kr(r); i.e., the effective exchange potential for a state at the top of the Fermi distributions. This is since density adjustments physically understandable come about by redistribution of the electrons near the Fermi level. (b) High dertsity. This regime is characterized by the condition r, /a, ((1, where as is the Bohr radius. In this case, the entire exchange and correlation energy is smaller than the kinetic energy by a factor of order (r,/as) and hence our inaccuracy in representing these portions becomes negligible. The reader will have noticed that while in Eq. (2.3) we approximate the exchange and. correlation energy by the expression valid for a slowly varying density, we made no approximation for the kinetic-energy functional T,Lnj of Eq. (2.2). This procedure is responsible for the exactness of the high-density limit, even when the density is rapidly varying, such as in the vicinity of an atomic nucleus. We now Inake a few further remarks about our approxirnation. If in Eq. (2.2), we had approximated T, LN) by its form appropriate to a system of slowly varying density,
~

an atomic nucleus, and (2) it does not lead to quantum density oscillations, 4 such as the density fluctuations due to atomic shell structures. By not making the replacement (2.20), we avoid both of these shortcomings. Let us now qualitatively discuss the appropriateness of our procedure for various classes of electronic systems. In atoms and molecules one can distinguish three regions: (1) A region near the atomic nucleus, where the electronic density is high and therefore, in view of case (h) above, we expect our procedure to be satisfactory. (2) The main "body" of the charge distribution where the electronic density n(r) is relatively sjowly varying, so that our approximation (2.3) for e, is expected to be satisfactory as discussed in case (a) above. (3) The "surface" of atoms and the overlap regions in mole cules. Here our approximation (2.3) has no validity and therefore we expect this region to be the main source of error. We do not expect an accurate description of chemical binding. In large atoms, of course, this "surface" region becomes of less importance. (The surface is more satisfactorily handled in the nonlocal method described under 8 below. ) For metals, alloys, and small-gap insulators we have, of course, no surface problem and we expect our approximation (2.3) to give a good representation of exchange and correlation effects. In large-gap insulators, however, the actual correlation energy will be considerably reduced compared to that of a homogeneous electron gas of the same density.

B. Nonlocal

Effective Potential

Instead of the Hartree-type procedure discussed in Sec. IIA it is also possible to obtain a scheme which includes exchange effects exactly. We write in place of

Eq. (2.3)

F,PN)=E

Lrtj+

rt(r)e, (e(r)) dr

(2.21)

where Z LNj is the exchange energy of a Hartree-Fock system of density (r) and e, (m) is the correlation energy per particle of a homogeneous electron gas. Applying this ansatz in conjunction with Eq. (2.2) and the stationary property of (2.1) leads to the following system

of equations:

Ni(r, r')
where
I

T, Lm|

', (3x'I)'t'rt dr,

(2.20)

, 4'(r') r'I lr

d~= e'4'(r),

(2 22)

of the Thomas-Fermi method suggested by Lewis. This method shares with the Thomas-Fermi method two shortcomings: (1) It leads to an in6nite density near ' H. W. Lewis, Phys. Rev. 111, 1554 (1958).
we would have been led

to the generalization

p. = d (rte. )/dN,

(2.23) (2.24)

I (r r )=Z It't(r)A*(r ) j~1

and p(r), rt (r) are defined as before, Eqs. (2.6) and (2.9).

&&36

AND

L' I

gHAM

The energy is now

as of deity
0

E=P p,
1

1V'

e(r) e(r')
~here
1

(3.5)

f and
~

ei(r, r')ei (r', r

fp are the free energie P er elect. ron of an nd noninteracting gas respective 1y.. t g

)+ (bG [e7/be(r))+I)~p(e
)=q( "-" f ())/d".
Fq. (2.6) and
(3.7)

e p 1'qg p (e(r)) pp(e( )) dr. (2' )

&o

is given as before, b iven,

This procedure may beregar e as a Hartree-Foc k a eth()d ( orrected o Beets. It is no mo plicated than the nn t d Hartree-Foc&e 1 tor appearing l
described in
n. ow 1 we must expect, more accurate resu ts a from the m ctive rticular, near the surface of an atom the effec iv e

3 6) is identlca ]. to the corresPonndln eq ' electr()ns in the system o f noninteracting 1 tion is therefo W +~-' effective potent . terined by t tern of equat o

Equation t'ion for

a,

1"

""

'

i~2+ &(r)+pxo(e(r)))~'
N
j,
i

~,
1)

(3 .8)

n&=

=1

2/{

(e( p)/pe+

(3.9)

i p proaches zero muc hf correlation e ec s surface.

str . E
tl

n here, howeve, described near t e

p is determined d as usua b y th

r t talnumbero of P a,rticles
'

roxiteracting

III. FREE ENERGY; SPECIFIC

HEAT
p
g

allo s is the low-

tion o gr re ensembles by using th e 6ni temperature generalization o Eq . (2.1) i m b ' He has shown that the gran can t a tential can be wri en

n ex ansion,

in po A

ut more conad namics

5 e7=
BT

'(T) 1)c (0+ v=-

~()+

8G

be(r)

r + G[e7 e(r) dr,

)

()e(r)

(BG[e7

3.1)

(3.10)

ni ue functiona 1 of the density at a o e ' 1 gi p erature g and p is t h e ch e rrect n this quan i y In analogy wit (2 2) e now write ith

"

Th e integral vanish because o f the stationary property ish ishes i e of 0, so that 3.11) 5[e7= (()G[e7/ ~~) n(r), V
~

here

[e7=G, [e7+F,[e7;

oninterment, applied to a system of noni

(3.2) (3.3)
o Coinbining obtai n
' '

rS, G, [e7= T,[e7 [e7,

n Ln

where,

t, energy a (1 R tion, ge an free energy. F For t approximation

are respective el y, the kinetic ~ ar of noninteracting e1 to th py

s. 3. Eqs. (3.11, 3.12), (3.2), and (3.4), we

n (n
87' 87
'

te

corre 1a tion

tib to

5 e7= 5 [e7+ e(r) (Bf

~(,), vdr. ( )
by

(3.13)

For small r it is we known that ell

5, is g'iven
(p),

~-[ 7=
where,
9

()f*.( ()) d,

(3 4)

5,[e7=

-', x'k'~g,

(3.14)

le- article density of states in the it o

n ntribuan e and correlation con ti on to the free energy perr electron o auni Uo e 1
h s. N. D. Mermin, Phys. Rev. 137, A1441 (19651.

(bf. ()/~

=-.)..= -'ir'k'r[g

. .

(d(o,

gp(((p e (e) )

(3.15)

SE LF Co N S1STENT EQ UATIONS

( where phi

) and Po(n) ' respective' y r the chemical tentials of an teracting and a n teractg h s of density ~ d and go are t- respective he
immediately

So fal every thmg

fo~al
~

We now write

s sectjolls,

G i
cond

rr';[n
pproxim ate

It follows
where

low

P erature heat

capacity is given by

C,, = yv-,

(3.16)
p(n)))
analogous
1

va ryi gg

which gives

=-'ee

Xg

Q)+ n(. )(g6

no

~(n)) ap(t
treatment,

x n x[n]=x,[n]+ [x(n(r)) r xp(n(r))]

where

dr,

(4.8) (4.9)
for

= (1/2

)'(~/~)&&g, (

),

We shall

Sec.

to

b
known divergen ence in the 1ow- e

res ective l y, the susceptibilities unl orm sy stems wii ith and wit h ou t interactions.

APPENDIX

IE I: GRADIE IENT .
'

EXPANSION OF N

IV. SPIN SU USCEPTIBILITY

s in susceptibility fHK as, we 6rstexten d th e h

of an 1 d

In thi's pp endix en
y
corre

we show that for a system of slow 1y w

ives

the

density

*()o ()-s * ~
ound-state energy can be written ln the grounh
@Vi

t e

e the self-consistent q d (2.9), for n(r), an ob 2) ' ' er i al rinciple and determine e r )di W th dm t to estimate the error o

1.

().()-~()
r'i r
universal

()&d

e From (2.5) an d the e pansion

(2.12

o, n,

d'+GL
of

() ()],
e
an
m,

functional

is the chemical po e otential

k

[cf. HK, Eq.. (68)].

orrect m(r),

For small m we expan

G

42)
[n])m(r)m(r') dr dr
o
h

G[n]+
2

r';
+

(4.3)

t er co pngoonl y the first term in ( .11), o , t' ns utlon of order To solve A1. 1), let us wri'te the externa c e e density as
(A1.2)

e
eL=

anis e nl

W lC

amagn pa rama etic system in th

h where

r ro +

~~

(slow

v spatia l variation),

'

an d

t ry the
(A1.3) (A1.4)

0f (4.2) we fin, or order and tha t

ansatz
where

n(r) = np(r)+ni(r),
no(r)

H(r)+
inver

I d G(r, r', [n])m(r')dr

=0, p

(4.4)

= fp(r/)

n is the zero-field density. . We now formally ensit W ' this equation, w h h

m r

e er exactly neu tralizes the ex ternal charge an e& is assumed nd y e lecting, for the momen, terms of order V~' in (A1.1

G r', [n])H(r')dr'. '(r,

' '
~
~

(A1.3) in'to A1. 1),

bt i

(4.5)
r'/ fr

For a uniform Geld this gives for the susceptibihty e or

)+

.().'(.)+o( ').
(A1.5)

x[n] =
'o

m(r) dr=

G r, r'; '(r,

r; [n]) dr dr

Now de6ne

4.6)
and write

R= r/rp,
ni(r)

(A1.6)
(A1.'/)

J. M. Luttinger,

h p. Rcp, phys, Rpv. 119, 1153 (1960).

= fi(R) .

1138
(A1.5) becomes

W. KOHN

AND

L.

J.

SHAM

With this notation,

fi(R')
R'[ [R

dR'+I s(fs(R) )

+fr(R)s.'(fs(R))+o(fr')
We may now write

(A1 8)

such as a metal or an alloy, the second term on the right-hand side of (A2. 2) accounts adequately for the eGect of rapid density change on exchange and correlation. This fng again leads to a set of Hartree-type equations like Eq. (2.8), with an addition to the effective potential given by

F,

fi(R)=(1/ro')fi"'(R)+(1/re')fi"&(R)+,
p

(A1.9)

(&If
2

y.
r', E,(r rr(r) )(e(r) dr'. rr(r'))

= p"'+ (1/ro') p"'+

(A1.10)

X {e(r+sr') }s dr' N(r rsr')

The 6rst term of Eq. (A1.9) is correctly determined by Eq. (A1.8) and. not affected either by the inclusion of terms of order V' in (A1.5) or by the terms of order fir in (A1.8). Hence, in spite of the errors of order 7' in (A1.1), the density given by our procedure is correct to order 1/rs' or ~V'~', inclusive. Equation (A1.8) shows that this curious result stems from the infinite range of the Coulomb interaction.

(A2. 3)

APPENDIX II: EFFECT OF RAPID DENSITY OSCILLATION ON EXCHANGE AND CORRELATION

In Eq. (2.3), we approximated E, ej by the erst L term in the gradient expansion. In actual physical systems, there are quantum density oscillations' whose effects on exchange and correlation are not included in the approximation (2.3). Now we put forward a correction to (2.3) to include such effects. In HK, the gradient expression for the energy functional is partially summed such that it is also correct for a system of almost constant density' even when the density Quctuations are of short wavelength":

Note that in the random-phase approximation E, vanishes. Hence, in a calculation which includes the eRective potential (A2. 3), we need reliable estimates of Ecalculated beyond the random-phase approximation, which are not available at present. The addition of (A2. 3) to the effective potential obviously makes the solution of the self-consistent equations much more di%cult. However, assuming that the modification of m(r) produced by this term is small, one may calculate n (r) and Z first without including it, and then, because of the stationary property, Eq. (2.5), one can obtain the correction to the energy by evaluating the second term in (A2. 2) with the unmodified density.
cVo&e added irr Proof We shou. ld like to point out that it is possible, formally, to replace the many-electron

l
GLriJ=
gs(N(r)) dr
2

E n(r)) (r r';
(A2. 1)

e(r'))'dr dr', X(rs(r)

where

problem by an exactly equivalent set of self-consistent one-electron equations. This is accomplished quite simply by using the expression (2.2) Lwithout the approximation (2.3)$ in the energy variational principle. This leads to a set of equations, analogous to Eqs. (2.4) 9), but with (2. (ri) replaced by an effective one-particle potential v dined formally as

p,

is determined by the polarizability of a homogeneous electron gas at density n, and r= r(r+r'). To the same aPProximation, s
m)

r', &(r

= 5E, v,(r) Leg/Bn (r) .

Of course, an explicit form of v, can be obtained only if the functional E, N which includes all many-body L effects, is known. This effective potential will reproduce the exact density and the exact total energy is then given by

j,

p, Lrig=

m(r)e, (N(r)) dr
2

r', Z,(r rs(r))

dr'

e(r')) s dr X (e(r)

(A2. 2)

where K,is the di6'erence between K of the interacting homogeneous gas and that of the noninteracting gas at the same density. We believe that for an infinite system,

N(r) e(r')

dr

dr'+E, tier j t(r) N(r)

dr.

"The second
@he kernel

term of HK, Eq. (83) is in error; it should be

E(r'; e r))(e r+

x') e(r ~~r'))'dx'.

E has

the same meaning as in HK.

Of course, if we make the approximation (2.3) for E, the above exact formulation reverts to the approximate theory of Sec. II.