Philip Anderson’s Superexchange Model

Yan Ru Pei
University of California, San Diego
(Dated: November 29, 2017)
The concept of superexchange was initially introduced by Kramers in 1934[1]. However, the
original model did not mention the specific mechanism involved in the superexchange process. In
1950, Philip Anderson attempted to refine Kramer’s approach. However, his first attempt resulted
in a theory that is complex and poorly convergent[2]. In 1959, Anderson was able to improve on his
own theory, achieving a much satisfactory description of superexchange[3]. Later on, his theory was
further improved on by Goodenough and Kanamori, culminating in the GKA rules that is essential
in describing molecular bonding, Mott insulators, and superconductivity[4][5].

INTRODUCTION

The superexchange interaction is defined as the inter-

action of two cations over an intermediate anion. For
example, consider the structure Mn2+ − O2− − Mn2+ ,
the two Mn ions are able to interact with each other
despite an O2− being “in the way”. Although the defi-
nition of superexchange is straightforward, the physics Figure 1: A drawing of a hydrogen molecule. Electron 1 and
behind this process is quite complicated. It took years proton a make up one atom, and electron 2 and proton b make
of trail and error to come up with a satisfactory model up the other.[6]
for superexchange. The endeavor fortunately paid off,
as the superexchange model has proved itself to be of
fundamental importance in describing a wide range of proton b. The first two terms are the kinetic terms and
phenomenons, from Mott insulators to superconductiv- the last six terms are the interaction terms.
ity.
For the potential terms, the first two terms describe
The process of superexchange is a combination of di- the attraction between the electron and proton within
rect exchange and electron transfer. The models for both each pair of atoms, the 3rd and 4th term describe the
have been extensively developed since 30 years before the attraction between the electron and the neighboring
Anderson’s first superexchange paper was published. So proton, and the last two terms describe the Coulomb
to understand the physics behind superexchange, we first repulsion between two electrons and two atoms respec-
have to take a look at the models of direct exchange and tively.
electron transfer.
For the eight total terms, if we treat the first four terms
as the zeroth order Hamiltonian, and the last four terms
BACKGROUND as perturbative terms, then we can write H0 as:

The Heitler-London Model ~ 1 1

H0 = − (∇2 + ∇22 ) + e2 (− − )
2m 1 |r1 − ra | |r2 − rb |
First, let us begin by considering the Hamiltonian for ~ 2 e2 ~ 2 e2
= (− ∇1 − ) + (− ∇2 − )
a hydrogen molecule: 2m |r1 − ra | 2m |r2 − rb |
= H1 + H2
~
H=− (∇2 + ∇22 ) (2)
2m 1
1 1 1 1
+ e2 (− − − − so we see that the Hamiltonian is separable and we can
|r1 − ra | |r2 − rb | |r1 − rb | |r2 − ra | solve for the eigenstates of the two electrons separately.
1 1 In other words, we have:
+ + )
|r1 − r2 | |ra − rb |
(1) H1 ψa (r1 ) = E0 ψa (r1 )
(3)
H2 ψb (r2 ) = E0 ψb (r2 )
where r1 and r2 are the positions of the two electrons,
and ra and rb are the positions of the two protons. Note that the eigenstates can be written as a tensor prod-
Electron 1 pairs with proton a, and electron 2 pairs with uct of the spatial eigenstates and the spin eigenstates, or
 2

|ψi = |ψs i |χi. Note that the electrons are fermions, so der energy is simply:
they must have an anti-symmetric total wavefunction.
For the singlet state, the spin wavefunction is already 2E0 ± 2k 2 E0
E0 = = 2E0 (7)
anti-symmetric, so we know that the spatial wavefunction 1 ± k2
must be symmetric, in order to form a anti-symmetric Recall that:
product. On the other hand, the triplet state has a sym-
metric wavefunction, and therefore must have an anti- 1 1 1 1
∆H = e2 (− − + + )
symmetric spatial wavefunction. Therefore, we can write |r1 − rb | |r2 − ra | |r1 − r2 | |ra − rb |
the singlet and triplet spatial wavefunctions respectively (8)
as follow: The last term is trivial because it is just the Coulomb
repulsion between two protons. It is independent of r1
|ψ+ i = |ψa i1 |ψb i2 + |ψb i1 |ψa i2 and r2 , so we simply have:
(4)
|ψ− i = |ψa i1 |ψb i2 − |ψb i1 |ψa i2
e2 e2 e2
hab| |abi = hab|abi =
We can easily check that the two above wavefunctions Rab Rab Rab
(9)
are in fact eigenstates of the free Hamiltonian given in e2 e 2
e2
hba| |abi = hba|abi = k 2
equation 2. We can simply the expression by denoting Rab Rab Rab
|ψa i1 |ψb i2 as |abi, which is the total wavefunction for
electron 1 being on site a and electron 2 being on site where Rab = |ra − rb |. So we can absorb this term into
b. Similarly, we denote |ψb i1 |ψa i2 as |bai, which is for the zeroth order Hamiltonian and find new energy to be:
electron 1 being on site b and electron 2 being on site a. e
2E0 + (10)
Rab
Note that the wavefunctions are not yet normalized.
If we denote the overlap of the wavefunctions of the two Now, we are only left with three non-trivial terms at the
electrons as hψa |ψb i = k. For simplicity, if we assume perturbative Hamiltonian, and we can make the following
that k is real, then we can derive that hba|abi = hab|bai = denotation:
k 2 . We can normalize the two eigenstates as follow: 1 1 1
Vc = hab|e2 (− − + )|abi
|r1 − rb | |r2 − ra | |r1 − r2 |
1
|ψ+ i = p (|abi + |bai) 1 1 1
2(1 + k 2 ) Vex = hba|e2 (− − + )|abi
(5) |r1 − rb | |r2 − ra | |r1 − r2 |
1 (11)
|ψ− i = p (|abi − |bai)
2(1 − k 2 )
The first term is called the Coulomb integral, and the
We can easily verify that hψ− |ψ− i = hψ+ |ψ+ i = 1 and second term is called the exchange integral.
hψ− |ψ+ i = 0, meaning that the two eigenstates are
normalized and orthogonal, as required. We can then write the energies in the following form:

Note that the two states are only eigenstates of the ze- e2 Vc ± Vex
E± = 2E0 + + (12)
roth order Hamiltonian, but not the full Hamiltonian. In Rab 1 ± k2
fact, we can treat the last four interaction terms as per- If we make the assumption that the overlap of the two
turbations ∆H, and we can express the full Hamiltonian electron wavefunctions is zero, or hψa |ψb i = k = 0, then
as H = H0 + ∆H. We can use first order perturbation the expression of the singlet and triplet energies reduces
theory to calculate the energies of the singlet and triplet to:
states as follow:
e2
1 E± = 2E0 + + (Vc ± Vex ) (13)
E± = (hab| ± hba|)(H0 + ∆H)(|abi ± |bai) Rab
2 ± 2k 2
(6) We can prove that the exchange integral Vex is positive
hab|(H0 + ∆H)|abi ± hba|(H0 + ∆H)|abi
= if the overlap integral is zero, or Vex > 0. This is done
1 ± k2
by Fourier transforming the perturbative terms (which
where we have used the fact that hab|H|abi = hba|H|bai we will not present here).
(the Hamiltonian is symmetric under exchange of r1 and
r2 ), and also the fact that hab|H|abi = hba|H|abi (the Therefore, we see that the triplet energy E− is lower
Hamiltonian is Hermitian). than the singlet energy E+ , so the electrons will have a
“triplet ground state”, meaning that the interaction is
It is obvious that hab|H0 |abi = 2E0 hab|abi = 2E0 and ferromagnetic[7].
hba|H0 |abi = 2E0 hba|abi = 2k 2 E0 . Then the zeroth or-
 3

It is important to note that this model does not cap- number of electrons. We can then denote each atom as a
ture the full interaction between the two electrons. In “site”, and assign it with a label i. Electrons cannot be
the following section, we will present another interacting created or destroyed, but they are free to hop between
process known as charge transfer. The full interaction different atom sites. Within each atom, we can denote
between the two electrons is described by the addition of the number of up-spin electrons as ni↑ and the number
the two. of down-spin electrons as ni↓ , and we have a coulomb
potential within each site proportional to the product of
these two numbers.
Dirac’s Spin-exchange Model
The lengthy description above is called the Hubbard
The Dirac’s spin exchange operator can be written as: Model, and it can be condensed simply into a Hamilto-
nian description as[8]:
3
1 X
P12 = (1 + σ1i · σ2i ) (14) X X
2 i=1
H= (tij c+
iσ cjσ + h.c.) + U (ni↑ ni↓ ) (19)
(i,j)σ i
where σ denotes the Pauli matrix. If we denote the sin-
glet spin wavefunction as |χ+ i, and the triplet state as The first term describes the process of an electron
|χ+ i, then it can be shown that: being destroyed at site j and being created at site i.
The Hermitian conjugate describes the reverse process
P |χ+ i = −P |χ+ i (destroyed at site i and created at site j). Intuitively,
(15)
P |χ− i = +P |χ− i we can think of this as the electron “hopping” between
the two sites, and this is associated with some transition
The signs make intuitive sense since the singlet spin state amplitude tij . In addition, the hopping electron must
is anti-ferromagnetic, and the triplet spin state is ferro- preserve spin, as denoted by the σ subscript. And the
magnetic. We can then rewrite the perturbative Hamil- second term is simply the Coulomb interactions between
tonian as: the up-spin state electrons and the down-spin state
electrons within each atom site.
H = Vc − Vex P (16)
As an example, we can consider the simple case of two
then we can easily see that:
electrons and two sites. There are six possible states for
H |χ+ i = (Vc + Vex ) |χ+ i this configuration, and they are shown schematically at
(17) Fig. 2. Note that due to the Pauli Exclusion Principle,
H |χ− i = (Vc − Vex ) |χ− i
two electrons of the same spin cannot be at the same
confirming equation 13. atom site.

If the two electrons are at different sites, then each

Direct Exchange with Non-zero Overlap electron can be in either the up-spin or down-spin state,
giving us 22 = 4 possible states. We denote these 4
In the case where the overlap of the electron wave- states as |ψ1 i, |ψ2 i, |ψ3 i, and |ψ6 i. The two electrons
functions is non-zero, or hψa |ψb i 6= 0. We can rewrite can also be at the same site. However, this time, the two
the energies in the following form: electrons must have opposite spins, so this give us only
2 possible states, and they are denoted by |ψ4 i and |ψ5 i.
e2 2
(Vc + R)± (Vex + k 2 eR )
E± = 2E0 + (18) Note that we can separate the Hamiltonian into a static
1 ± k2
2
component and an interacting component, or H = H0 +
The term Vc + eR is always positive, and the term Vex + ∆H, where we have:
2
k 2 eR is generally negative. So in this case, we see that the X
energy is lower for the singlet |χ+ i, so the direct exchange H0 = U (ni↑ ni↓ )
interaction is anti-ferromagnetic in this case. X
i
(20)
∆H = (−tc+
iσ cjσ + h.c.)
(i,j)σ
Hubbard Model for Charge Transfer
for simplicity sake, we have assumed that tij = −t for all
We begin by considering a simple model. Let’s i and j. It is obvious that all six states are eigenstates
imagine that there are some number of atoms and of H0 , where H0 |1i = H0 |2i = H0 |3i = H0 |6i = 0
electrons, and each atom can “hold” up to a certain (since the two electrons are not at the same site) and
 4

calculate the matrix elements and find the eigenstates of

the matrix. Note that state 1 and state 6 cannot transi-
tion to any other states, nor can they be transitioned to
from any other states, as it will violate Pauli Exclusion
Principle. States 2 and 3 can transition to states 4 and
5, and the reverse process can also happen. This gives us
8 non-zero matrix elements −t. This, in addition to the
two U elements for states 4 and 5 gives us a total of 10
non-zero elements. The matrix is as follow:
 
0 0 0 0 0 0
0 0 0 −t −t 0
 
0 0 0 −t −t 0
H=   (21)
0 −t −t U 0 0

0 −t −t 0 U 0
0 0 0 0 0 0

We can then solve for the eigenenergies and eigenstates:

E1 = 0 |Ψ1 i = |ψ1 i
E2 = 0 |Ψ2 i = |ψ6 i
1
E3 = 0 |Ψ3 i = √ (|ψ2 i − |ψ3 i)
2
1
Figure 2: The six possible states for two electrons and two E4 = U |Ψ4 i = √ (|ψ4 i − |ψ5 i)
atoms to be in.[6] 2
1 p
E5 = (U − U 2 + 16t2 )
2 √
H0 |4i = H0 |5i = U . U + U 2 + 16t2
|Ψ5 i = (|ψ2 i + |ψ3 i) + (|ψ4 i + |ψ5 i)
4t
On the other hand, note that none of the six states are 1 p
eigenstates of the interacting Hamiltonian. In fact, we E6 = (U + U 2 + 16t2 )
2 √
have hi|∆H|ii = 0, meaning that the coupling between U − U 2 + 16t2
the same states is zero. This makes intuitive sense be- |Ψ6 i = (|ψ2 i + |ψ3 i) + (|ψ4 i + |ψ5 i)
4t
cause ∆H is the “hopping” term, and an electron must (22)
move to a different site. For example, h4|∆H|2i = −t,
describing the down-spin electron hopping from site 2 Note that the last two eigenstates are not normalized.
to site 1. However, not all couplings are nonzero. For
example, h3|∆H|2i = 0, because a transition from state Let us consider two limiting cases. First, let’s consider
2 to state 3 requires two hops. First, the down-spin the case where U  t, then, let’s consider the case
electron must hop from site 2 to site 1, and then the where t  U . The first case essentially means that the
up-spin electron must hop from site 1 to site 2, making potential energy is infinitely strong, and it is impossible
this a second-order process. for the electrons to overcome it. The second case means
that there is effectively no potential energy, and the
In first order perturbation theory, the transition electrons are free to roam around.
amplitude is proportional to the coupling between the
two states and inversely proportional to the energy In the first case, the last two expressions simply reduce
difference, or hf |∆H|ii
|Ef −Ei | . It is important to note that all
to:
first order coupling is always associated with an energy 1
difference of U . This is because a first order hopping E5 = 0 |Ψ5 i = √ (|ψ2 i + |ψ3 i)
2
either brings to two electrons together or breaks them (23)
1
apart, giving us an energy change equal to the Coulomb E6 = U |Ψ6 i = √ (|ψ4 i + |ψ5 i)
2
interaction.
so we see that in this limiting case, there is no eigenstates
Fortunately, we can solve the eigenstates of the full that contain a mixture of both the “apart” state and
Hamiltonian exactly, so there is really no need to perform “together” state. This makes intuitive sense because the
perturbative calculations. All we need to do is to simply potential barrier that the electrons have to overcome
 5

to break apart or come together is infinitely high. Another thing to note is the J is not negative in gen-
Therefore, if the electrons start out as an “apart” state, eral. In this section, all we have shown is that J is nega-
then it can never come together, and vice versa. tive for the particular model of hydrogen molecule. If we
apply the Hubbard model to calculate J for any molecule,
Now, let’s look at the other limiting case where t  U . the result of J will depend on the valence shells of the
In this case, the last two expressions then reduce to: atoms involved. The calculation is not always easy. How-
ever, as we will see later, there is a shortcut to determin-
1 ing the sign of J, known as the Hund’s rules.
E5 = 2t |Ψ5 i = (|ψ2 i + |ψ3 i + |ψ4 i + |ψ5 i)
2
1
E6 = 2t |Ψ5 i = (− |ψ2 i − |ψ3 i + |ψ4 i + |ψ5 i)
2
(24) ANDERSON’S APPROACH

In this case, the last two eigenstates are equal mixtures At this point, we have discussed the physics be-
of the “apart” states and the “together” states. This hind the direct exchange and electron transfer processes.
also makes sense, because the potential barrier is Now, we are ready to tackle the physics of superexchange.
essentially zero in this case, so the electrons are free to
roam around. This makes the “apart” states and the Superexchange is defined as the interaction between
“together” states symmetric to each other. two cations through an intermediate anion. For exam-
ple, in the simplest case of the Mn − O − Mn structure,
Lastly, let’s derive the spin exchange Hamiltonian from superexchange is the interaction between the two M n2+
the Hubbard model. To do so, let’s first find the effec- cations through the O2− anions. This interaction is
tive Hamiltonian from second order perturbation theory. really just a combination of the direct exchange and
Note that ∆H |2i = −t(|4i + |5i), ∆H |4i = −t(|2i + |3i), electron transfer processes.
and ∆H |5i = −t(|2i + |3i). Using these and the formula
for the effective Hamiltonian, we find: In fact, Philip Anderson has formalized the concept of
2 superexchange in his two attempts in 1950 and 1959, and
∆Hp∆H 2t
Hef f = |2i = − (|2i − |3i) (25) we will be taking a look at the two different approaches
U U he has taken. But before we do so, let’s first discuss some
Note that |3i is just |2i with spin exchanged. In other preliminary concepts to make our subsequent discussions
words, we have |3i = P |2i, where P is the Dirac’s spin easier.
exchange operator. Therefore, we can write the effective
Hamiltonian in terms of the spin exchange operator:
Slater Determinant
2t2
Hef f =− (1 − P ) = J − JP (26)
U
This section is going to be mainly math. We will
where J = − 2tU
2
< 0. Compare this with the last sec- discuss how to construct the total wavefunction of mul-
tion where we had Vex > 0 and got a ferromagnetic in- tiple identical fermions. For example, we already know
teraction. This time, J is negative so we have an anti- that the wavefunction of two electrons in two states can
ferromagnetic interaction. If we ignored the terms with- be written as ψ = ψ1 (1)ψ2 (2) − ψ1 (2)ψ2 (1), where the
out P , we can combine the two interactions (from direct subscript denotes the state and the number inside the
exchange and charge transfe), and write the total Hamil- bracket denotes the electron “label”. Note that this
tonian as wavefunction is antisymmetric under the exchange of
the two electrons.
Htotal = −(J + Vex )P (27)
Now, let’s consider the more complicated case of three
this gives us both ferromagnetic and anti-ferromagnetic electrons in three states. The total wavefunction can be
terms. expressed in the form of a Slater determinant, which is
just a fancy name for the determinant of a matrix with
There are two things to note here. In the case where single particle wavefunctions as elements. In this case,
the overlap of the two orbitals is zero, the magnitude we can express the Slater determinant of three electrons
of the charge transfer interaction J is also zero. This as[9]:
is proven in the later sections. In other words, charge
transfer only occurs we have a non-zero overlap of ψ1 (1) ψ1 (2) ψ1 (3)
orbitals. ψ2 (1) ψ2 (2) ψ2 (3) (28)
ψ3 (1) ψ3 (2) ψ3 (3)
 6

which is just: Now, if we bring in the condition that t  ∆E, then we

can reduce the eigenenergy expressions to:
ψ1 (1)ψ2 (2)ψ3 (3) + ψ1 (2)ψ2 (3)ψ3 (1) + ψ1 (3)ψ2 (1)ψ3 (2)
t2
− ψ1 (3)ψ2 (2)ψ3 (1) + ψ1 (1)ψ2 (3)ψ3 (2) + ψ1 (2)ψ2 (1)ψ3 (3) E+ = E2 +
∆E (33)
(29) t2
E− = E1 −
∆E
Note that this wavefunction satisfies the two properties.
First, each term of wavefunction bijects to a possible and the eigenstate expressions to:
permutation of the electron label, meaning that the t
wavefunction describes the full picture of each electron ψ+ = ψ2 + ψ1
∆E (34)
occupying the three states with equal possibility. Sec- t
ondly, the wavefunction is antisymmetric under any pair ψ− = ψ1 − ψ2
∆E
of electron label exchange, which means that it fully
describes three electrons as fermions. so we see that the ground state wavefunction is essentially
the ground state of the non-interacting Hamiltonian plus
The Slater determinant can be extended to n fermions a small mixture of the excited state.
easily. First, we simply write the matrix with element
Mij corresponding to ψi (j). Then, we just have to Wavefunction overlap
take the determinant of the matrix to obtain the total
wavefunction.
We say that a wavefunction is symmetric (under
parity exchange) if ψ(r) = ψ(−r), and we say that
To show that this method in fact works, note that the
a wavefunction is anti-symmetric if ψ(r) = −ψ(−r).
determinant of any square matrix is simply all of pos-
Assume now that we have two wavefunctions, one is
sible permutations of the j, and if the permutation is
symmetric and the other is anti-symmetric. We can
even, then the term has a plus sign, and if the permu-
denote ψa (r) as the symmetric wavefunction and ψb (r)
tation is odd, then the term has a minus sign. So if we
as the anti-symmetric wavefunction. Then it is not hard
perform a pair exchange before we take the determinant,
to prove that the wavefunction overlap is zero, or hψb |ψa i.
then we are essentially performing an additional transpo-
sition, and this makes odd permutations even, and even
If we assume that the overlap is non-zero, then we can
permutations odd. Therefore, the odd permutations will
denote the overlap as:
have a minus sign, and the even permutations will have a Z
plus sign. And the sign of the total wavefunction changes,
ψb∗ (r)ψa (r)dr = k (35)
meaning that it is antisymmetric under any pair of ex-
change of electron labels.
And if we perform a parity operation, then the integral
becomes
Z Z
Two States with Weak Coupling ψb∗ (−r)ψa (−r)dr = − ψb∗ (r)ψa (r)dr = −k (36)

If we consider two states with energies E1 and E2 , and However, the overlap integral should be invariant under
the transition energy between much less than the energy parity exchange, so we have k = −k, which implies k = 0.
difference of the two states t  ∆E = E2 − E1 , then we
can write the Hamiltonian as: In short, the overlap of a symmetric and anti-
  symmetric wavefunction must be zero. It is important
E1 t to note that inverse of this statement is not true,
H= (30)
t E2 meaning that a zero overlap doesn’t necessary mean
that one of the wavefunction is symmetric and the
Then we can solve for the eigenenergies: other is anti-symmetric. However, for the purpose of
our discussion in this paper, we can assume that the
1 p inverse statement is true for now. (This is because we
E± = (E1 + E2 ± ∆E 2 + 4t2 ) (31)
2 are mainly dealing with orbital wavefunctions, and they
are either symmetric or anti-symmetric).
and the (un-normalized) eigenstates:
 √  Now, let’s discuss the process of electron transfer. This
E1 − E2 ± ∆E 2 + 4t2 is a one particle process, and we can calculate the ampli-
ψ± = (32)
2t tude of this transfer process to be hψb (r)|U (r)|ψa (r)i,
 7

where U (r) is the electron-ion interaction potential,

which is a scalar and therefore symmetric under parity
exchange. Now, if we assume that ψa is symmetric, and
ψb is anti-symmetric, then it is easy to see that the inte-
gral
Z
ψb∗ (r)U (r)ψa (r)dr = 0 (37)
Figure 3: This is a schematic drawing of the
Mn2+ − O2− − Mn2+ structure. The anion in the mid-
must be zero. Therefore, in a general sense, we see dle is O2− , and the cations at the two ends are Mn2+ . We
that the process of electron transfer cannot occur if the will label the leftmost ion as A, the intermediate ion as B,
wavefunction overlap is zero. and the rightmost ion as C. The + sign denotes an up-spin
electron and the − sign denotes the down-spin electron. [6]
This is not the case for direct exchange. The process of
direct exchange can occur even if the wavefunction over-
lap is zero. This is because the direct exchange process WLOG, let’s assume that the one of the electron at B
involves two particles instead of just on. Let’s say ini- site transfers to the C site (the scenario is symmetric if
tially, electron 1 is at site a, and electron 2 is at site b, it transfers to the A site instead). Note that the charge
and after the exchange, electron 1 is at site b, and elec- transfer interaction is anti-ferromagnetic, so only a
tron 2 is at site a, then the amplitude of this exchange down-spin electron can transfer to the C site, where the
process is hψa (2)ψb (1)|U (1, 2)|ψa (1)ψb (2)i. Again, we are electron is up-spin. Due to Pauli’s Exclusion Principle,
making the assumption that ψa is symmetric, ψb is anti- the remaining un-coupled electron at the B site must
symmetric, and U is symmetric. The amplitude of ex- have up-spin, so it can only couple to the up-spin
change can again be expressed in the following integral electron at A-site through direct exchange, which is a
form: ferromagnetic interaction[2]. Therefore, the net effect
Z is that the electron at site C couples to the electron at
site A through a combination of direct exchange and
ψa∗ (r2 )ψb∗ (r1 )U (r1 , r2 )ψa (r1 )ψb (r2 ) (38)
electron transfer, with site B being an intermediate site.

Note that in this case, the integrand is actually symmet- This intuitive picture is useful in visualizing the super-
ric under parity exchange. This is because ψb (r1 ) and exchange process, but it doesn’t describe the full picture.
ψb (r2 ) are both anti-symmetric, and they cancel out to To study this more carefully, we will have to treat this
make a symmetric term. Therefore, the direct exchange quantum mechanically, and write down the full fermionic
amplitude is non-zero. So we see that direct exchange wavefunction of the electrons involved in the interacting
can occur even if the wavefunction overlap is zero. process. This leads us to Anderson’s revised approach in
describing the superexchange process.
Anderson’s First Attempt

Anderson’s Second Attempt

We will only be going over briefly Anderson’s first
attempt at describing the super-exchange process,
since his first approach gives a complicated third-order Let’s first begin by carefully studying the d−p σ-bond.
perturbative process and the math is poorly convergent. Note that the d orbital has a much higher energy than
The technicality involved in his original paper is far the p orbital, meaning that the transition amplitude is
too complicated to be present here, so we will use an much smaller than the energy difference between the two
intuitive picture to illustrate the main concepts involved orbitals. And from the previous sections, we know that
in his paper. For this discussion, we will consider the we can express the ground state wavefunction as the p
structure Mn2+ − O2− − Mn2+ as a simple example. orbital wavefunction plus a small mixture of the d orbital
wavefunction:
Refer to Fig. 3 for a schematic representation of the
Mn2+ − O2− − Mn2+ structure. We see that the O2− ψbond = ψp + λψd (39)
anion is sandwiched between the two Mn2+ cations,
forming an 180 degree structure. The three ions are where λ  1. Now, if we assume that the overlap be-
bonded by σ bonds, so their wavefunction overlaps are tween the two orbitals is hψd |ψp i = k, then we can write
non-zero. We can label the sites from left to right as A, the antibonding wavefunction as:
B, and C.
ψanti = ψd − (λ + k)ψp (40)
 8

We can check that the two wavefunctions are in fact or- equation in the following form[6]:
thogonal:
d↑ (1) − γp↑ (1) d↑ (2) − γp↑ (2) d↑ (3) − γp↑ (3)
hψanti |ψbond i p↑ (1) + λd↑ (1) p↑ (2) + λd↑ (2) p↑ (3) + λd↑ (3) (45)
= − (λ + k) hψp |ψp i + λ hψd |ψd i + (−λ(λ + k) + 1) hψd |ψp i p ↓ (1) + λd↓ (1) p ↓ (2) + λd↓ (2) p↓ (3) + λd↓ (3)
=−λ+k+λ−k =0 with λ = γ − k. Note that the equality between matrix
(41) 44 and matrix 45 is not trivial to show. The derivation
involves lengthy algebra and doesn’t really provide any
where we have made the assumption that λ  1 and illuminating physics insights, so it will be skipped for
k  1. And as usual, the bonding energy is lower than the time being.
the anti-bonding energy.
However, the important thing to note is that d − γp
Now, let’s return to the study Mn2+ − O2− − Mn2+ is an anti-bonding state, and p + λd is a bonding state.
structure, and extract a three electron wavefunction And from the Slater determinant, we see that two
from it. The three electrons we are interested in is the electrons are in the bonding state, and one electron is in
d electron at the C site and the two p electrons at the the anti-bonding state. The two electrons in the bonding
B site. Note that in the wavefunction, we also have states have opposite spin, so they cancel each other out.
to consider the electron transfer process, where one Therefore, only the anti-bonding state contributes to the
electron from the p orbital at B site transfer to the d spin of the system, and it extends over the anion (site B).
orbital at C site.
This spin then interacts with the electron at site A,
As shown in the previous sections, we can write with the energy of the spin coupling governed by equation
the three-electron wavefunction as a Slater determinant. 27:
First, we consider the wavefunction before the electron
transfer: Htotal = −(J + Vex )P (46)
d↑ (1) d↑ (2) d↑ (3) and this effectively completes our picture of the superex-
p↑ (1) p↑ (2) p↑ (3) (42) change interaction between site A and site C, with site
p↓ (1) p↓ (2) p↓ (3) B being the intermediate[3].
Note that the d electron has up-spin, and the two p elec-
trons have opposite spins. Now, we write the wavefunc- GKA RULES
tion after the electron transfer:
d↑ (1) d↑ (2) d↑ (3) The GKA rules are a series of semi-empirical rules de-
p↑ (1) p↑ (2) p↑ (3) (43) veloped by Goodenough and Kanamori as a refinement
d↓ (1) d↓ (2) d↓ (3) to the original model developed by Anderson. The rules
are based on the following facts[4][5]:
Note that after the charge transfer, the original down-
spin electron in the p orbital at B site has hopped over • When two orbitals overlap, both direct exchange
to the d orbital at C site, and this is reflected in the last and charge transfer can occur.
row of the matrix.
• When two orbitals have zero overlap, only direct
exchange can occur.
Recall that we can write the ground state wavefunc-
tion as the wavefunction before the transfer plus a small • The greater the wavefunction overlap, the greater
mixture of the wavefunction after the transfer. In other the interaction strength.
words, we can express the ground state wavefunction as:
For example, let’s consider boding between two p or-
d↑ (1) d↑ (2) d↑ (3) d↑ (1) d↑ (2) d↑ (3) bitals. Let’s define the separation of the two atoms to be
p↑ (1) p↑ (2) p↑ (3) + λ p↑ (1) p↑ (2) p↑ (3) in the z axis, then:
p↓ (1) p↓ (2) p↓ (3) d↓ (1) d↓ (2) d↓ (3)
• The pz − pz bond is a σ bond, meaning that there
d↑ (1) d↑ (2) d↑ (3) is a significant wavefunction overlap, giving us a
= p↑ (1) p↑ (2) p↑ (3) strong anti-ferromagnetic interaction.
p↓ (1) + λd↓ (1) p↓ (2) + λd↓ (2) p↓ (3) + λd↓ (3)
(44) • The px −px and py −py bonds are π bonds, meaning
that there is a smaller wavefunction overlap, giving
Note that p↑ + λp↑ = (1 + λ)p↑ = p↑ , and d↑ + λd↑ = us an anti-ferromagnetic interaction, though not as
(1 + λ)d↑ = d↑ . Note that we can rewrite the above strong as that of the σ bond.
 9

• The px − py , py − pz , and pz − px bonds are not then according to the GKA rules, the interaction
“bonds” in a chemical sense, since they have zero between them is anti-ferromagnetic, meaning that
wavefunction overlap. However, they do exhibit site C will have down spin as a result of coupling
weak ferromagnetic interaction through the direct with the spin up electron of site B. So in the end, we
exchange process. see that the net effect of site A having spin up is to
force site C having spin down, so the superexchange
This discussion can easily be generalized to d−p bonds. interaction is anti-ferromagnetic.
For example:
• Now let’s look at V2+ − 2p − Ni2+ , again in the
• The dγ −pσ bond is analogous to σ bonds, meaning
form dγ − pσ − dγ. The difference is that V2+ is
that it is strongly anti-ferromagnetic.
less than half-filled. In this case, if site A is spin up,
• The d − pπ bond is analogous to π bonds, meaning then it will also receive a spin up electron from site
that it is anti-ferromagnetic, though not as strong B. The remaining electron in site B will then have
as the first case. spin down, and it will interact with site C anti-
ferromagnetically, forcing site C to have spin up.
• The dγ − pπ and d − pσ are not bonds. They have So the superexchange interaction is ferromagnetic
zero wavefunction overlap and are weakly ferromag- in this case.
netic.
• Let’s look at Ni2+ − 2p − Ni2+ again. But this
An important thing to note is that although the GKA time, the two bonds form a 90 degree angle in-
rules are very general, they are only semi-empirical, stead of 180 degree. Even though site A and site B
meaning that they are not guaranteed to work for all will still have the same wavefunction overlap, the
molecular bonds, though the cases where they do not difference now is that site B and site C will have
work are very rare. zero wavefunction overlap. The reason is that after
“bending” the bonds, site B and site C will form
a π bond instead of a σ bond. If site A has spin
APPLYING THE GKA RULES
up, it will still receive an spin down electron from
site B, leaving a spin up electron behind. However,
In this section, we will use the GKA rules to predict this time, the spin up electron will couple with site
the superexchange interaction of various structures. In C ferromagnetically (since they have zero overlap),
doing so, we will evoke Hund’s rule. Hund’s rule, in its resulting in a spin up state for site C. So the su-
simplest form, states that “every orbital in a subshell is perexchange interaction is ferromagnetic.
singly occupied with one electron before any one orbital
is doubly occupied, and all electrons in singly occupied • Again, let’s look at V2+ − 2p − Ni2+ , this time in
orbitals have the same spin”. In other words, if an elec- the 90 degree form. Following a similar argument,
tron shell is less than half-filled, and it receives another we see that the superexchange interaction is anti-
electron, then the received electron will have the same ferromagnetic.
spin as the total spin of the original shell. And if the
electron shell is half-filled or more than half-filled, then In summary, we see how the superexchange interaction
the received electron will have a spin opposite to the orig- can be predicted by knowing the types of bonds formed,
inal shell. Using the GKA and Hund’s rules, we are ready the degree of the bonds, and the electron occupancy of
to look at some superexchagne examples: the shells.

• First, let’s look at the Ni2+ − 2p − Ni2+ structure.

This is in the form of dγ − pσ − dγ, meaning that EXPERIMENTAL SUPPORT
the two bonds are both σ bonds. Furthermore, the
valence shells of the Ni2+ cations are both more All of the structures we used in the previous section
than half filled. For simplicity sake, let’s label the as examples for superexchange interaction are all real
leftmost site as A, the middle site as B, and the molecular structures. The theoretical predictions of the
rightmost site as C. WLOG, let’s assume that A interaction within the molecules are all verified exper-
and B is bonded by electron transfer. imentally through measurements of interaction strengths.

If we assume that site A has spin up, then according In fact, the superexchange model has successfully pre-
to Hund’s rule, it must receive a spin down electron dicted the interaction for a wide range of molecules, rang-
from site B. And if site B donates a spin down ing from the simplest hydrogen molecule to the more
electron, then it is left with a spin up electron. And complicated structures such as the La2 CuO4 Mott in-
since site B and site C has wavefunction overlap, sulator.
 10

CONCLUSION we do not make any erroneous predictions.

The GKA rules are very important in the sense that

are very simple, yet they are able to successfully predict
the superexchange interaction successfully for a wide
[1] H.A. Kramers. L’interaction Entre les Atomes Magntognes
range of molecules. It also provides a possible explana- dans un Cristal Paramagntique. Physica 1, 182 (1934).
tion to the high-T superconductive property of certain [2] P.W. Anderson. Antiferromagnetism. Theory of Superex-
copper oxide. The superconductivity is due to the change Interaction. Phys. Rev. 79, 350 (1950).
condensation of electron pairs via the anti-ferromagnetic [3] P.W. Anderson. New Approach to the Theory of Superex-
superexchange between the Cu2+ ions, though this change Interactions. Phys. Rev. 115, 2 (1959).
is still a speculation, and whether the superexchange [4] J.B. Goodenough. Theory of the Role of Covalence in the
Perovskite-Type Manganites. Phys. Rev. 100, 564 (1955).
model is fundamental to the superconductivity process
[5] J. Kanamori. Superexchange interaction and symmetry
is still not established. properties of electron orbitals. J. Phys. Chem. Solid 10,
87 (1959).
Although the GKA rules are simple, we have to remind [6] M. Suzuki and I. Suzuki. Lecture note on solid state
ourselves that they are established upon the mathemat- physics Superexchange interaction.
ically rigorous model of superexchange. In deriving the [7] W. Heitler and F. London. Wechselwirkung neutraler
GKA rules, certain assumptions and approximations are Atome und homopolare Bindung nach der Quanten-
mechanik. Z. Phys. 44, 455 (1927).
made. Therefore, the GKA rules should only be viewed
[8] J. Hubbard, J Electron Correlations in Narrow Energy
as a rule of thumb rather than rules set in stone. For Bands Proceedings of the Royal Society of London. 276
example, the rules break down when the bonding angle (1365), 238257 (1963).
deviates from 180 degree, when the electron occupancy is [9] J. Slater and HC Verma. The Theory of Complex Spectra
dynamic, or when the spin-orbit coupling becomes non- Phys. Rev. 34 (2), 12931322 (1929).
negligible. Therefore, it is important for us to understand
thorough the full picture of superexchange, to ensure that