Notes on exchange coupling parameters

Taisuke Ozaki, ISSP, Univ. of Tokyo

December 15, 2020

1 Non-collinear (NC) DFT

We first summarize the theory of non-collinear DFT (NC-DFT). A two component spinor wave function
is defined by

|ϕν ⟩ = |φαν α⟩ + |φβν β⟩, (1)

where |φαν α⟩ ≡ |φαν ⟩|α⟩ with a spatial function |φαν ⟩ and a spin function |α⟩. In the notes we consider
non-Bloch functions, but the generalization of the description to the Bloch function is straightforward.
Then, a density operator is given by
∑
n̂ = fν |ϕν ⟩⟨ϕν |,
ν
∑ ( )( )
= fν |φαν α⟩ + |φβν β⟩ ⟨φαν α| + ⟨φβν β| , (2)
ν

where fν is a step function. With the definition of density operator n̂, a non-collinear electron density
in real space is given by

nσσ′ = ⟨rσ|n̂|rσ ′ ⟩,
∑ ′
= fν φσν φσν ,∗ , (3)
ν

where σ, σ ′ = α or β, and |r⟩ is a position eigenvector. The up- and down-spin densities n′↑ , n′↓ at each
point are defined by diagonalizing a matrix consisting of a non-collinear electron densities as follows:
( )
n′↑ 0
= U n U †,
0 n′↓
( )
nαα nαβ
= U U †. (4)
nβα nββ

Based on the spinor wave function Eq. (1), the non-collinear electron density Eq. (3), and the up- and
down-spin densities, the total energy non-collinear functional [1, 2] could be written by
∑ ∑ ∑∫ 1
∫ ∫
n′ (r)n′ (r′ )
Etot = fν ⟨φσν |T̂ |φσν ⟩ + wσσ′ nσ′ σ + drdr′ + Exc {nσσ′ } , (5)
σ=α,β ν σσ ′
2 |r − r′ |

where the first term is the kinetic energy, the second the electron-core Coulomb energy, the third
term the electron-electron Coulomb energy, and the fourth term the exchange-correlation energy,
respectively. Also the total electron density n′ at each point is the sum of up- and down-spin densities

1
n′↑ , n′↓ . Alternatively, the total energy Etot can be expressed in terms of the Kohn-Sham eigenenergies
εν as follows:
∫ ∫
1
Etot = Eband − n′ vH dr − Tr(vxc n)dr + Exc , (6)
2
where vxc is a non-collinear exchange-correlation potential which will be discussed later on. Consid-
ering an orthogonality relation among spinor wave functions, let us introduce a functional F :
∑
F = Etot + ϵνν ′ (δνν ′ − ⟨ϕν |ϕν ′ ⟩) . (7)
νν ′

The variation of F with respect to the spatial wave function φ is found as:
δF ∑ ′ ∑ ′ ′ ∑
σ)∗
= T̂ φσν + wσσ′ φσν + vH φσν + vxc φν −
σσ σ
ϵνν ′ φσν′ (8)
δ(φν σ′ σ′ ν′

with
∫
d(r)
vH = dr, (9)
|r − r′ |
σσ ′ δExc
vxc = . (10)
δnσ′ σ
By setting the variation of F with respect to the spatial wave function φ to zero, and considering
a unitary transformation of φσν so that ϵνν ′ can be diagonalized, we can obtain the non-collinear
Kohn-Sham equation as follows:

δF
=0  ( )( ) ( )
δ(φα ∗  αα w
T̂ + wαα + vH + vxc αβ φαν φαν
ν) αβ + vxc
⇒ = εν . (11)
=0 
βα ββ φβν φβν
δF  wβα + vxc T̂ + wββ + vH + vxc
δ(φβ
ν)
∗

We see that the off-diagonal potentials produce explicitly a direct interaction between α and β
spin components in this α-β coupled equation. The off-diagonal potentials consist of the exchange-
correlation potential vxc and the other contributions w such as spin-orbit interactions.
The U -matrix in Eq. (4) which relates the non-collinear electron densities to the up- and down-spin
densities is expressed by a rotation operator D [3]:
( )
−iσ̂ · hϕ
D ≡ exp (12)
2
with Pauli matrices
( ) ( ) ( )
0 1 0 −i 1 0
σ1 = , σ2 = , σ3 = , (13)
1 0 i 0 0 −1

where h is a unit vector along certain direction, and ϕ a rotational angle around h. Then, consider
the following two-step rotation of a unit vector (1,0) along the z-axis:
( )
• First, rotate θ on the y-axis → exp −i σ22 θ
( )
• Second, rotate ϕ on the z-axis → exp −i σ32ϕ

2
The unit vector (1,0) along the z-axis is then transformed as follows:
( ) ( ) ( )( )
1 σ3 ϕ σ2 θ 1
⇒ exp −i exp −i , (14)
0 2 2 0

where
( ) ( ) ( )( )
σ3 ϕ σ2 θ exp(−i ϕ2 ) 0 cos( 2θ ) − sin( 2θ )
exp −i exp −i =
2 2 0 exp(i ϕ2 ) sin( 2θ ) cos( 2θ )
( )
exp(−i ϕ2 ) cos( 2θ ) − exp(−i ϕ2 ) sin( 2θ )
= . (15)
exp(i ϕ2 ) sin( 2θ ) exp(i ϕ2 ) cos( 2θ )

Thus, if the direction of the spin is specified by the Euler angle (θ, ϕ), the U -matrix in Eq. (4) is given
by the conjugate transposed matrix of Eq. (15).
( )
exp(i ϕ2 ) cos( 2θ ) exp(−i ϕ2 ) sin( 2θ )
U = (16)
− exp(i ϕ2 ) sin( 2θ ) exp(−i ϕ2 ) cos( 2θ )

The meaning of Eq. (4) becomes more clear when it is written in a matrix form as follows:
{ ( ) }
∑ φαν ( )
†
Un U = U fν φα,∗
ν φβ,∗
ν U† (17)
ν φβν

We see that the U -matrix diagonalizes the total (average) non-collinear spin matrix rather than the
non-collinear spin matrix of each state ν. Since the exchange-correlation term is approximated by
the local density approximation (LDA) or generalized gradient approximation (GGA), once the non-
collinear spin matrix n is diagonalized, the diagonal up- and down-densities are used to evaluate the
exchange-correlation potentials v̄xc within LDA or GGA:
( )
↑
vxc 0
v̄xc = ↓
,
0 vxc
1 ↑ ↓ 1 ↑ ↓
= (vxc + vxc )σ0 + (vxc − vxc )σ3 ,
2 2
0
= vxc σ0 + ∆vxc σ3 . (18)

Then, the potential v̄xc is transformed to the non-collinear exchange-correlation potential vxc as follows:

vxc = U † v̄xc U,
0
= vxc σ0 + ∆vxc U † σ3 U,
0
= vxc σ0 + ∆vxc σ̄3 ,
( )
0 + ∆v cos(θ)
vxc ∆vxc exp(−iϕ) sin(θ)
xc
= 0 − ∆v cos(θ)
,
∆vxc exp(iϕ) sin(θ) vxc xc
0
= vxc σ0 + ∆vxc e · σ̂. (19)

where
( )
cos(θ) exp(−iϕ) sin(θ)
σ̄3 = , (20)
exp(iϕ) sin(θ) − cos(θ)

e = (sin(θ) cos(ϕ), sin(θ) sin(ϕ), cos(θ)) , (21)

3
and

σ̂ = (σ1 , σ2 , σ3 ). (22)

The Euler angle (θ, ϕ) and the up- and down-spin densities (n′↑ , n′↓ ) are determined from the non-
collinear electron densities so that the following relation can be satisfied:
( )
n′↑ 0
U nU † = . (23)
0 n′↓

After some algebra, they are given by

( )
Im nαβ
ϕ = − arctan (24)
Re nαβ
( )
2(Re nαβ cos(ϕ) − Im nαβ sin(ϕ))
θ = arctan (25)
nαα − nββ
1 1
n′↑ = (nαα + nββ ) + (nαα − nββ ) cos(θ) + (Re nαβ cos(ϕ) − Im nαβ sin(ϕ)) sin(θ) (26)
2 2
′ 1 1
n↓ = (nαα + nββ ) − (nαα − nββ ) cos(θ) − (Re nαβ cos(ϕ) − Im nαβ sin(ϕ)) sin(θ) (27)
2 2
Then, it is noted that the effective potential veff in Eq. (11) can be written in Pauli matrices as
follows:

veff = v0 σ0 + ∆vxc σ̄3 + W,

= v0 σ0 + b · σ̂ + W,
( )
vαα vαβ
= , (28)
vβα vββ

where

0 0
veff = vH + vxc , (29)

( )
wαα wαβ
W = , (30)
wβα wββ

b1 = ∆vxc sin(θ) cos(ϕ), (31)

b2 = ∆vxc sin(θ) sin(ϕ), (32)
b3 = ∆vxc cos(θ), (33)

2 Energy change by perturbation

In the section, we derive a formula of a perturbation theory which will be used to calculate exchange
coupling parameters.
The Dyson equation is generally given by

ˆ Ĝ(0) (Z) + Ĝ(0) (Z)∆

Ĝ(Z) = Ĝ(0) (Z) + Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z) + · · · ,
ˆ Ĝ(0) (Z)∆ (34)

4
 ˆ is an operator for a certain perturbation. We extend
where Ĝ(0) is the non-perturbed Greenian, and ∆
the Dyson equation to the KS equation of Eq. (11) in the NC-DFT, and therefore Ĝ(0) is expressed
by a 2 × 2 matrix as
 
(0) (0)
Ĝαα (Z) Ĝαβ (Z) 
Ĝ(0) (Z) ≡  (0) (0) , (35)
Ĝβα (Z) Ĝββ (Z)

(0)
where each element Ĝσσ′ is defined by

(0) ∑ |φσ ⟩⟨φσ′ |

ν ν
Ĝσσ′ (Z) = . (36)
ν Z − εν

Also, the perturbation in the NC-DFT is given by a 2 × 2 matrix as

( )
ˆ αα ∆
∆ ˆ αβ
ˆ ≡
∆ . (37)
ˆ βα ∆
∆ ˆ ββ

The band energy under the perturbation can be obtained by

∫ EF ( )
1
Eband − EF Nele = − Im (E − EF ) Tr Ĝ(Z) dE,
π
∫ EF ( )
1
= − Im (E − EF ) Tr Ĝ(0) (Z) dE
π
∫ EF ( )
1
− Im (E − EF ) Tr Ĝ(0) (Z)∆
ˆ Ĝ(0) (Z) dE
π
∫ EF ( )
1
− Im (E − EF ) Tr Ĝ(0) (Z)∆
ˆ Ĝ(0) (Z)∆
ˆ Ĝ(0) (Z) dE
π
+···, (38)

where Z = E + i0+ . From the definition Ĝ(0) (Z)(Z Iˆ − Ĥ (0) ) = I,

ˆ we have

( )2 ∂ Ĝ(0) (Z)
Ĝ(0) (Z) =− . (39)
∂Z
In addition to Eq. (39), by noting the following formula:
( ) ( )
ˆ Ĝ(0) (Z) = Tr Ĝ(0) (Z)Ĝ(0) (Z)∆
Tr Ĝ(0) (Z)∆ ˆ , (40)

the first order term in Eq. (38) is simplified as follows:

∫
EF ( )
1
− Im (E − EF ) Tr Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z) dE,
π ( )
∫ EF
1 ∂ Ĝ(0) (Z) ˆ
= Im (E − EF ) Tr ∆ dE,
π ∂Z
∫ EF ( )
1
= − Im ˆ dE.
Tr Ĝ(0) (Z)∆ (41)
π
Also, by noting the following relation:
( ) 1 d ( )
Tr Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z) = −
ˆ Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z)∆
Tr Ĝ(0) (Z)∆ ˆ , (42)
2 dZ

5
the second order term in Eq. (38) is simplified as follows:
∫EF ( )
1
− Im (E − EF ) Tr Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z)∆ˆ Ĝ(0) (Z) dE,
π
∫ EF ( )
1 d
= Im (E − EF ) Tr Ĝ(0) (Z)∆ ˆ Ĝ(0) (Z)∆ˆ dE,
2π dZ
∫ EF ( )
1
= − Im ˆ Ĝ(0) (Z)∆
Tr Ĝ(0) (Z)∆ ˆ dE. (43)
2π
Thus, the change ∆Eband in Eband under the perturbation is given by
( )
(0)
∆Eband = Eband − EF Nele − Eele − EF Nele ,
∫ EF ( ) ∫ EF ( )
1 ˆ dE − 1 Im
= − Im Tr Ĝ(0) (Z)∆ ˆ dE + · · · , (44)
ˆ Ĝ(0) (Z)∆
Tr Ĝ(0) (Z)∆
π 2π

Now we consider that the perturbation ∆ ˆ locally acts on atomic sites. Letting ∆
ˆ i and ∆
ˆ j be the
perturbation on an atomic sites i and j, respectively. Then, using Eq. (44) we define an interaction
energy Vij between two atomic sites i and j by
(ij) (i) (j)
Vij = ∆Eband − ∆Eband − ∆Eband ,
∫ EF ( )
1
= − Im ˆ i Ĝ(0) (Z)∆
Tr Ĝ(0) (Z)∆ ˆ j dE (45)
π
(ij) ˆi + ∆
ˆ j ) for two atomic sites in Eq. (44).
where ∆Eele is evaluated by applying the perturbation (∆
The Liechtenstein formula [4, 5] is based on the local force theorem, and takes account of the change
in the band energy, corresponding to the sum of the KS eigenvalues for the occupied states.

3 Exchange coupling in the collinear case

We assume that the ground state has a collinear magnetic structure where the spin quantization is along
the z-axis, and introduce a small spin rotation at an atomic site i. It is noted that the spin rotation
does not change the total charge density if the rotation is described by the unitary transformation with
Eq. (15). This is called the rigid spin approximation. Thus, within the approximation the perturbation
term related to the Hartree term does not appear.
Using Eq. (19) the perturbation ∆ ˆ i in Eq. (45) is given by

ˆ i = ∆v̂ (i) ∆ei · σ̂,

∆ (46)
xc

where
( )
(0) (0)
∆ei = ei − ei = eix , eiy , eiz − eiz . (47)

By considering Eq. (21) and cos(θ) = 1 − θ2 /2! + · · ·, for a small spin rotation ∆ei is approximated by
(0)
∆ei ≃ ei − ei = (eix , eiy , 0) . (48)

ˆ i can be rewritten as
With ∆ei given by Eq. (48) the perturbation ∆
( )
ˆ i = ∆v̂ (i) 0 eix − ieiy
∆ xc . (49)
eix + ieiy 0

6
Similary, by considering a small spin rotation ∆ej = (ejx , ejy , ejz ) = (−eix , −eiy , 0) at an atomic site
j towards the opposite direction compared to the case of the atomic site i, we have the perturbation
ˆ j as
∆
( )
ˆj = (j) 0 −eix + ieiy
∆ ∆v̂xc . (50)
−eix − ieiy 0

By inserting Eqs. (49) and (50) into Tr() of Eq. (45) and considering that Eq. (35) is diagonal for the
ground state with a collinear magnetic structure, we have
( (0)
) ( )
(0) ˆ (0) ˆ Ĝαα 0 (i) 0 eix − ieiy
Ĝ ∆i Ĝ ∆j = (0) ∆v̂xc
0 Ĝββ eix + ieiy 0
( (0)
) ( )
Ĝαα 0 0 −eix + ieiy
× (0)
(j)
∆v̂xc ,
0 Ĝββ −eix − ieiy 0
( (0)
)
0 Ĝαα (eix − ieiy ) (i)
= (0) ∆v̂xc
Ĝββ (eix + ieiy ) 0
( (0)
)
0 Ĝαα (−eix + ieiy )
× (0)
(j)
∆v̂xc ,
Ĝββ (−eix − ieiy ) 0
 
(0) (i) (0) (j)
Ĝαα ∆v̂xc Ĝββ ∆v̂xc 0
=  (0) (i) (0) (j)
 (−e2 − e2 ).
ix iy (51)
0 Ĝββ ∆v̂xc Ĝαα ∆v̂xc

Further, by inserting Eq. (51) into Eq. (45), and noting the following relations:
1
−e2ix − e2iy = (−e2ix − e2iy + e2iz − e2ix − e2iy − e2iz ),
2
1
= (ei · ej − 1) , (52)
2

( )
Tr (ABCD) = Tr (ABCD)† (53)

we obtain
∫ EF ( )
1 (0) (0)
Vij = − Im (i)
Tr Ĝ↑ (Z)∆v̂xc (j)
Ĝ↓ (Z)∆v̂xc dE × (ei · ej − 1) , (54)
π
(0) (0) (0) (0)
where Ĝ↑ ≡ Ĝαα and Ĝ↓ ≡ Ĝββ . Since the term arising from −1 in Eq. (54) is a constant term, we
define the exchange coupling parameter Jij by
∫ EF ( )
1 (0)
(i) (j) (0)
Jij = Im Tr Ĝ↑ (Z)∆v̂xc Ĝ↓ (Z)∆v̂xc dE. (55)
π
The Jij evaluated by Eq. (55) can be used in the Heisenberg and Ising models.

References
[1] U. Von Barth and L. Hedin, J. Phys. C: Solid State Phys. 5, 1629 (1972).

[2] J. Kubler, K-H. Hock, J. Sticht, and A. R. Williams, J. Phys. F: Met. Phys. 18, 469 (1988).

7
[3] J. J. Sakurai, Modern Quantum Mechanics.

[4] A.I. Liechtenstein, M.I. Katsnelson, V.P. Antropov, and V.A. Gubanov, J. Magn. Mang. Mater. 67, 65 (1987).

[5] M.J. Han, T. Ozaki, and J. Yu, Phys. Rev. B 70, 184421 (2004).

[6] P. Bruno, Phys. Rev. Lett. 90, 087205 (2003).

[7] S. Lounis and P.H. Dederichs, Phys. Rev. B 82, 180404(R) (2010).