Coherent Potential Approximation

November 29, 2009

1 Green-function matrices in the TB formalism

In the tight binding (TB) picture the matrix of a Hamiltonian H is in the form

H = H ij , where

(1)
H ij = δij εi + γ ij . (2)

Single and double underlines denote matrices in angular momentum space and site-angular
momentum space, respectively. The size of each angular momentum block is determined by
the dimension of the basis centered at each site i. In the case of 3d transition metals e.g., the
hybridized 3d-4s-4p valence band spans a 9-dimensional space (18 including spin). In many
cases the on-site energy blocks εi in Eq. (2) are themselves diagonal, but this is not necessary.
The hopping integrals γ ij are strictly site-off-diagonal.
The resolvent (or static Green-function) matrix of a given system described by the Hamil-
tonian H can be defined as

−1
G(z) := z − H (3)

for any z ∈ C (at least where the inversion can be performed). Supposing that the solutions of
the eigenvalue equation,

H |ii = εi |ii , (4)

are known, then the Hamiltonian matrix can be written as

X
H= εi |ii hi| , (5)
i

where i runs over all eigenfunctions. This implies the spectral decomposition of the matrix
G(z),
X 1
G(z) = |ii hi| . (6)
i
z − εi

The fundamental analytic property of the resolvent,

G (z ∗ ) = G(z)† (7)

1
is a corollary of this decomposition. Another fundamental identity can be derived from defini-
tion:
dG(z)
= −G(z)2 . (8)
dz
Since G(z) is undefined at real energies, εi , we have to approach the real arguments from the
imaginary direction,
G± (ε) := lim G (ε ± ıδ) (9)
δ→0
X 1
= |ii hi| , (10)
i
ε − εi ± ı0
+ −
for any ε ∈ R. Note that G (ε) 6≡ G (ε) if ε is in the spectrum of H. Equation (7) implies
 + †
G (ε) = G− (ε) . (11)
The well-known identity of generalized functions,
 
1 1
=P ∓ ıπδ (ε − εi ) , (12)
ε − εi ± ı0 ε − εi
leads to the relationship,
X 1  +
G (ε) − G− (ε)

δ (ε − εi ) |ii hi| = −
i
2πı
1 h +
† i
=− G (ε) − G+ (ε) . (13)
2πı
By composing the trace of (13), the density of states of the system, n (ε), can be expressed
from the Green function as
X 1 h
† i
n (ε) = δ (ε − εi ) = − Tr G+ (ε) − G+ (ε)
i
2πı
1 
∗ 
=− Tr G+ (ε) − Tr G+ (ε)
2πı

1 1
n (ε) = − ImTr G+ (ε) = ImTr G− (ε) . (14)
π π
The expectation value of an observable A at zero temperature can be calculated as
ZεF X
hAi = δ (ε − εi ) hi| A |ii dε (15)
εb i

ZεF X
!
= Tr δ (ε − εi ) |ii hi| A dε (16)
εb i

ZεF
1
hAi = − Im Tr A G+ (ε) dε ,
 
(17)
π
εb

so G(z) and the spectrum of H contain the same information.

2
2 Perturbations with respect to a reference system
 −1
Suppose now that H = H 0 + ∆H, and G0 = z − H0 is the resolvent of the reference
system. Then

−1
G(z) = z − H (18)
 
z − H 0 − ∆H G(z) = I
 
I − G0 (z) ∆H G(z) = G0 (z) (19)

 −1  −1
G(z) = I − G0 (z) ∆H G0 (z) = G0 (z) I − ∆H G0 (z) . (20)

On the other hand, from Eq. (19),

G(z) = G0 (z) + G0 (z) ∆H G(z) . (21)

This equation can be solved iteratively:

G(0) (z) = G0 (z)

G(1) (z) = G0 (z) + G0 (z) ∆H G0 (z)
G(2) (z) = G0 (z) + G0 (z) ∆H G0 (z) + G0 (z) ∆H G0 (z) ∆H G0 (z)
..
.
G(z) = G0 (z) + G0 (z) ∆H G0 (z) + G0 (z) ∆H G0 (z) ∆H G0 (z) + . . . (22)

This Dyson-equation can be rearranged as

h i
G(z) =G0 (z) + G0 (z) ∆H + ∆H G0 (z) ∆H + . . . G0 (z)
=G0 (z) + G0 (z) T (z) G0 (z) , (23)

where T (z) is the so-called scattering matrix,

T (z) = ∆H + ∆H G(z) ∆H
= ∆H + ∆H G0 (z) ∆H + ∆H G0 (z) ∆H G0 (z) ∆H + . . .
= ∆H + ∆H G0 T (z) . (24)

This can be rearranged to give

h i−1 h i−1
T (z) = I − ∆H G0 (z) ∆H = ∆H I − G0 (z) ∆H . (25)

It can easily be shown that the T matrix has similar analytical properties as the resolvent,

T (z ∗ ) =T (z)† , (26)
dT (z) dG0 (z)
=T (z) T (z) , and (27)
dz dz
T ± (ε) := lim T (ε ± ıδ) (28)
δ→0

3
at real energies ε. By using equations (14) and (23), we get the density of states (DOS) of the
perturbed system with respect to the reference system,

1 h
+ + +
i
n(ε) = n0 (ε) − ImTr G0 (ε) T (ε) G0 (ε) . (29)
π

Using properties (8) and (27), then integrating with respect to energy, we arrive at the Lloyd-
formula, which gives the integrated DOS of the perturbed system,

Zε
1
N (ε) := n (ε0 ) dε0 = N0 (ε) + ImTr ln T + (ε) . (30)
π
−∞

3 On-site impurities
Case of a single on-site impurity: ∆H i = {∆H i δin δim },

T =∆H i + ∆H i G0 ∆H i + . . . (31)
= ∆H i + ∆H i Gii0 ∆H i + . . . δin δim
 

= {ti δin δim } , thus

ti =∆H i + ∆H i Gii0 ti . (32)
P
Now let ∆H be a sum of such on-site differences: ∆H = ∆H i . Then
i

! ! !
X X X
T = ∆H i + ∆H i G0 ∆H i + ...
i i i
X X X
= ∆H i + ∆H i G0 ∆H j + ∆H i G0 ∆H j G0 ∆H k + . . . (33)
i i,j i,j,k

X
T nm =∆H n δnm + ∆H n Gnm
0 ∆H m + ∆H n Gnk km
0 ∆H k G0 ∆H m + . . .
k
X
=∆H n δnm + ∆H n Gnk
0 T
km
. (34)
k

Both in operator and in matrix sense,

X X
T = ∆H i + ∆H i G0 ∆H j + . . . (35)
i i,j
!
X X
= ∆H i + ∆H i G0 ∆H j + . . .
i j
X
= Q , where (36)
i
i

4
 X
Q :=∆H n + ∆H n G0 ∆H m + . . . (37)
n
m
X
=∆H n + ∆H n G0 ∆Q
m
m
X
=∆H n + ∆H n G0 Q + ∆H n G0 ∆Q
n m
m(6=n)

⇓
  X
I − ∆H n G0 Q =∆H n + ∆H n G0 ∆Q
n m
m(6=n)
X
Q =tn + tn G0 Q , (38)
n m
m(6=n)

where tn is formally a single impurity T -matrix on the n-th site,

tn = {tn δin δjn } . (39)
Solving equation (38) iteratively,
Q(0) :=tn (40)
n
X
(1)
Q =tn + tn G0 tm
n
m(6=n)
X X
Q(2) =tn + tn G0 tm + tn G0 tm G0 tk
n
m(6=n) m(6=n)
k(6=m)
..
.
⇓
X X
Q =tn + tn G0 tm + tn G0 tm G0 tk + . . . (41)
n
m(6=n) m(6=n)
k(6=m)

Using equation (37), we arrive at the multiple scattering expansion of the T matrix,
X X X
T = tn + tn G0 tm + tn G0 tm G0 tk + . . . (42)
n n6=m n6=m6=k

Since all tn have the structure of ∆H n ,

X
T nm =tn δnm + (1 − δnm ) tn Gnm
0 tm + tn Gnk km
0 tk G0 tm + . . . (43)
k
(k6=n)
(k6=m)
X
=tn δnm + tn Gnk
0 (1 − δnk ) T
km
. (44)
k

Defining the site-off-diagonal part of the reference system’s resolvent,

b := Gnk

G 0 0 (1 − δ nk ) , (45)
T = t + tGb T
0
⇓
h i−1
−1
T = t + G0
b (46)

5
On the other hand,
T =t + t G
b t + tG
0
b tG
0
b t . . . , so
0
(47)
G =G0 + G0 T G0 (48)
is given in terms of G0 and t.

4 Chemically disordered systems

4.1 Binary alloys
Let’s consider now a two-state disordered system, e.g. a two-component (binary) random alloy:
∆Hi = ξi ∆HiA + (1 − ξi ) ∆HiB , (49)
where ξi are independent random variables with Bernoulli distribution:

1 with probability Pi (1) := ci
ξi = .
0 with probability Pi (0) = 1 − ci
By definition the expected values are
Eξi ≡ hξi i = ci , (50)
thus the expected value of ∆Hi is
h∆Hi i = hξi i ∆HiA + h1 − ξi i ∆HiB = ci ∆HiA + (1 − ci ) ∆HiB . (51)
Independence means that the joint probability mass function of {ξ} decomposes to the product
of the individual probability mass functions:
N
Y
P ({ξ}) = Pi (ξi ) . (52)
i=1

Of course P ({ξ}) is a probability, since trivially

1
!
X Y X
P ({ξ}) = Pi (ξi ) = 1. (53)
{ξ} i ξi =0

The configurational average of some physical quantity is then defined as

X X X
hF ({ξ})i := P ({ξ}) F ({ξ}) = ... P1 (ξ1 ) . . . PN (ξN ) F (ξ1 , . . . , ξN ) . (54)
{ξ} ξ1 ξN

Since G = G ({ξ1 , ξ2 , . . . , ξN }) ≡ G ({ξ}), the mean of a physical quantity A in the TB picture

is
 Z 
1  
hAi = − Im f (ε) Tr A G ({ξ}) dε (55)
π
Z
1 

= − Im f (ε) Tr A G dε (56)
π
(where we suppressed the dependence of G on the energy ε).

6
4.2 Coherent Potential Approximation



G = G0 + G0 T G0
 −1
=: Gc = z − H c , (57)

where we defined the effective Hamiltonian H c often noted as Σc , the self-energy. This as-
sumption is the coherent potential approximation (CPA). The (57) CPA condition can only
be satisfied if H c = H c (z) is a function of the energy, but it is (by definition) configuration-
independent.
Let us now choose our reference system to be H c ,

T =t + t G
b t + +t G
c
b tG
c
b t + ...
c
(58)


∆H =H − H c = H i − H c,i δin δim , (59)

where H c,i are to be determined. A condition is given by



G =Gc = Gc + Gc T Gc (60)
⇓


T =0 (61)

 D E
t + t Gc t + . . . = 0.
b (62)

Single-site CPA:


t := 0 . (63)

Considering this,

b nm tm = htn i G b nm htm i =0
D E
tn G c c (64)
n6=m

b nk tk G
b km tm = nk km
X D E X D E
tn G c c t G
b
n c htk i G
b
c tm =0. (65)
k6=n k6=n
k6=m k6=m

Thus eq. (63) satisfies the condition set by eq. (61) up to fourth order in t. Since

ti = ξi tA B
i + (1 − ξi ) ti , (66)

eq. (63) reads as

hti i = ci tA B
i + (1 − ci ) ti = 0 . (67)

This is in fact a system of equations for H c , because

−1
tαi = I − ∆H αi Giic ∆Hiα , where (68)
∆H αi = H αi − H c,i , (α = A, B) (69)

and Giic can be determined from equation (57).