In this problem we derive the sd-Hamiltonian from the Anderson impurity model.

The Anderson
Hamiltonian is given by
Xh † i X
H= ✏k ck ck + Vk d† ck + h.c. + ✏d nd + U nd" nd# (1)
k

where the operators d create electrons of spin at the localized impurity, and c†k creates an itinerant
†

electron of spin and energy ✏k in the metallic host. The total wavefunction of the Hamiltonian,
| i, can be written as the sum of three terms, | 0 i, | 1 i and | 2 i, where the subscript refers to the
occupancy of the impurity site. Using this decomposition, we can write the Schrödinger equation in
matrix form,
2
X
Hmn | ni =E| mi (2)
n=0

where Hmn = Pm HPn . The operators Pn are the projectors onto the subspace with n electrons on
the impurity site. They satisfy the relation P0 + P1 + P2 = 1, and are g given by
8 Q
>
< P0 = (1 nd )
P1 = 1 P0 P2 (3)
>
:
P2 = nd" nd#
Now, let’s find the matrix elements Hmn . The diagonal elements leave the occupation number
fixed, and they can be identified with the diagonal terms of the Anderson Hamiltonian,
8 P †
>
<H00 = Pk ✏k ck ck
H11 = k ✏k c†k ck + ✏d (4)
>
: P †
H22 = k ✏k ck ck + 2✏d + U
P
The off-diagonal terms are due to the hybridization term of the Hamiltonian, Hhyb = k (Vk d† ck +
h.c.). This term only contains one d operator, and therefore it can only connect states that differ by
one impurity electron. For this reason, H20 = H02 = 0. The other matrix elements can be found using
the explicit expressions of P0 and P2 ,

X
H10 = P1 Hhyb P0 = (1 P2 P0 )Hhyb P0 = Hhyb P0 = Vk d† ck (1 nd" )(1 nd# )
k
X X
= Vk d† ck (1 nd¯ ) = Vk d† (1 nd¯ )ck (5)
k k

X X
H21 = P2 Hhyb P1 = P2 Hhyb (1 P0 P2 ) = P2 Hhyb = Vk nd" nd# d† ck = Vk d† nd¯ ck (6)
k k

where we have used nd" nd# d†" = nd" d†" nd# = d†" nd# and nd" nd# d†# = d†# nd" . The other matrix elements
follow from Hmn = Hnm
†
,
( P
H01 = k Vk⇤ c†k (1 nd¯ )d
P (7)
H12 = k Vk⇤ c†k nd¯ d
We are interested in the effect of virtual excitations from the | 1 i subspace. Thus, we have to write
(2) only in terms of | 1 i. Using (2), we can formally write | 0 i and | 2 i in terms of | 1 i,

1
 1
H00 | 0i + H01 | 1i =E| 0i ! (E H00 ) H01 | 1i =| 0i

1
H21 | 1i + H22 | 2i =E| 2i ! (E H22 ) H21 | 1i =| 2i

Now we have


1 1
E| 1i = H10 | 0i + H11 | 1i + H12 | 2i = H10 H01 + H11 + H12 H21 | 1i (8)
E H00 E H22

In order to simplify the above expression we have to commute the terms (E H) 1

and H. First,
we commute only the matrix elements,

X X
H00 H01 = Vk⇤0 ✏k nk ck0 0 (1 nd¯ 0 ) d 0 = Vk⇤0 ✏k c†k ( kk0 0 c†k0 0 ck )(1 nd¯ 0 )d 0

kk0 0 kk0 0
X X
= Vk⇤0 ✏k0 c†k0 0 (1 nd¯ 0 )d 0 Vk⇤0 ✏k c†k c†k0 0 ck (1 nd¯ 0 )d 0 = H01 (✏k0 + H00 )(9)
k0 0 kk0 0

X X X
H00 H21 = Vk0 ✏k nk d† 0 nd¯ 0 ck0 0 = Vk0 ✏k d† 0 nd¯ 0 nk ck0 0 = Vk0 ✏k d† 0 nd¯ 0 ( kk0 0 ck0 0 c†k )ck
kk0 0 kk0 0 kk0 0
X
= Vk0 d† 0 nd¯ 0 ck0 0 ( ✏k0 + H00 ) = H21 ( ✏k0 + H00 ) (10)
kk0 0

Using the above expressions it’s straightforward to obtain

X
1
(E H00 ) 1
H01 = H01 (E ✏k 0 H00 ) = Vk⇤0 ck0 0 (1 nd¯ 0 )d 0 (✏k0 ✏d E + ✏d + H00 ) 1

kk0 0

X ✓ ◆ 1 X
c†k0 0 (1 nd¯ 0 )d 0 E ✏d H00 c†k0 0 (1 nd¯ 0 )d 0
= Vk⇤0 1 ' Vk⇤0 (11)
✏k 0 ✏d ✏k 0 ✏d ✏k 0 ✏d
kk0 0 kk0 0

X
1
(E H22 ) 1
H21 = H21 (E + ✏k0 H22 ) = Vk0 d† 0 nd¯ 0 ck0 0 (E U 2✏d H00 + ✏k0 ) 1

kk0 0

X ✓ ◆ 1 X
d† 0 nd ¯0 ck0 0 E ✏d H00 d† 0 nd ¯0 ck0 0
= Vk 0 1 ' Vk 0 (12)
U + ✏d ✏k 0 U + ✏d ✏k 0 U + ✏d ✏k 0
kk0 0 kk0 0

where we have expanded the expressions in parentheses to leading order in U and ✏d . Plugging these
expressions in (8) we get

" #
X d †
(1 nd¯ )ck c†k0 0 (1 nd¯ 0 )d 0 c† nd¯ d d† 0 nd¯ 0 ck0 0
(E H11 ) | 1i ' Vk0 Vk⇤ + Vk Vk⇤0 k | 1i
✏k 0 ✏d U + ✏d ✏k 0
kk0 0

(13)
The one-particle terms nd¯ can be dropped, becuase they act on a single-particle subspace. If we
swap k and k 0 in the first term, we obtain

2
 " #
X d† 0 ck0 0 c†k d c†k d d† 0 ck0 0
(E H11 ) | 1i = Vk Vk⇤0 + | 1i
✏k ✏d U + ✏d ✏k 0
kk0 0
2⇣ ⌘ ⇣ ⌘3
X 0 c†k ck0 0 d† 0 d c†k ck0 0 0 d† 0 d
= Vk Vk⇤0 4 + 5| 1i
0 0
✏k ✏d U + ✏d ✏k 0
kk
✓ ◆ !
X 1 1 X c†k ck0 nd
= Vk Vk⇤0 + c†k ck 0 0 d 0 d†
| 1i
U + ✏d ✏k 0 ✏k ✏d U + ✏d ✏k 0 ✏k ✏d
kk0 0 kk0

The last term in the second parenthesis is a constant, and can be absorbed into a shift of the
single-particle energy of the itinerant electron. This expression can be related to the spin operators by
using the Pauli identity,

↵ =2 ↵ ↵ (14)
In particular, we can rewrite

X X 1 X †
c†k ck0 0 d† 0 d = c†k↵ ck0 d† d ↵ = ck↵ ck0 d† d ( ↵ + ↵ )
0
2
↵ ↵
1 X † 1X † 1X †
= ck↵ ↵ c k 0 d† d + c c k 0 nd 0 = 2skk0 ·Sd + c ck0 n(15)
d 0
2 2 ↵ k↵ 2 0 k
↵
P P
where Sd = 2 ↵ d†↵ ↵ d is the impurity spin and skk0 =
1
↵ c†k↵ ↵ ck0 is the lattice electron
spin. Using this result, we can finally write

!
X 1X † X c†k ck0
(E H11 ) | 1i = Jkk0 2skk0 ·Sd + c c k 0 nd 0 | 1i Vk Vk⇤0 | 1i
2 0 k U + ✏d ✏k 0
kk0 kk0
!
X X
= 2Jkk0 skk0 ·Sd + Kkk0 c†k ck 0 | 1i (16)
kk0

where
8 ⇣ ⌘
<Jkk0 = Vk Vk⇤0 1
+ 1
⇣ ✏ k ✏d U +✏d ✏k0 ⌘
Vk Vk⇤0 (17)
:Kkk0 = 1 1
2 ✏k ✏d U +✏d ✏k0

The total effective Hamiltonian acting on the subspace | 1 i is

" #
X † X X †
Hef f = ✏k ck ck + 2Jkk0 skk0 ·Sd + Kkk0 ck ck 0 (18)
k kk0
The energy scale of the typical excitations is much smaller than U +✏d and ✏d . Thus, we can neglect
the energy dependence of Jkk0 and Kkk0 . In that case, the interaction Jkk0 can be treated as local,
and the scattering term Kkk0 can be absorbed into a shift of the single-particle energy of the itinerant
electron. These approximations lead to the sd-Hamiltonian,
X †
Hsd = ✏k ck ck + 2JSd ·s(r = 0) (19)
k