1

Lecture 4.1

Spin Hamiltonians and

Exchange interactions

This version of Modern Models, Lec. 4.1, edited for P654, Spring 2008. Sections 4.1 D
ff, about the microscopic origin of spin interactions, were cut.
This lecture develops the idea of spins as a degree of freedom, with which models
are built. There is a tension in how we think of spins. On the one hand, definitions
and calculations are easiest using the z-basis description, with operators Sz , S± obeying
an operator algebra (reminscent of fermion or boson operators). On the other hand, a
semiclassical picture in which the spin is approximately a fixed-length vector is more
useful for intuitive thinking and as a starting point for calculations in ordered states
(Sec. 4.1 B).
Besides the spin degrees of freedom, a spin model needs a Hamiltonian, and the
typical terms are surveyed in Sec. 4.1 C. Most important is the dot-product spin-
spin coupling called an exchange interaction: this is the second key physical idea of
this lecture. The (omitted!) next two sections (and Sec. 4.1 X) develop the origin of
exchange interactions – both ferromagnetic and antiferromagnetic – as a consequence of
fermion statistics when we reduce the Hilbert space (by eliminating charge fluctuations
as a degree of freedom). This story (the derivation of spin Hamiltonians from a more
microscopic level) will be continued in Lec. 4.3 , which [in its present form] focuses on
“superexchange”.

4.1 A Spins as objects

A “spin” S is a discrete degree of freedom that transforms like an angular momentum
under rotations. It is a shorthand for a quantum-mechanical degree of freedom with a
discrete set of basis states |Sz i labeled by the quantum number Sz = −S, . . . , +S (the
“z basis”). Furthermore, the basis states |Sz i transform like angular momenta under
spin-space rotations.
All components of a spin S are axial vectors – i.e., they should change sign under
time reversal. Thus “time reversal symmetry” in a spin Hamiltonian means “symmetry
under reversal of all spins.”
We typically arrive at a spin Hamiltonian by eliminating portions of the Hilbert
space as first described in Lec. 1.1 E : the subspace we project onto no longer allows
variations of the number of electrons (or whatever the spin-bearing particle is), such that

Copyright c 2007 Christopher L. Henley

2 LECTURE 4.1. SPIN HAMILTONIANS AND EXCHANGE INTERACTIONS

the low-lying states form a representation of the rotation group. Thus, microscopically,
S might be (i) the spin of a single electron localized on an impurity in a semiconductor;
(ii) the combined spin of several d electrons in a transition-metal ion (commonest case);
(iii) a nuclear spin (of course this depends on the isotope) of an atom in a crystal; or
(iv) the combined spin and orbital moment of a rare-earth ion. But if the Hamiltonian
has the same form, the system has the same behavior, no matter what the spins are
built from.
Now we imagine a lattice with a “spin” on each of the N sites. This system has
(2S + 1)N basis states, which are the direct product of the basis states for each spin.
They can be labeled
|Sz1 , Sz2 , . . . , SzN i (4.1.1)
When we specialize for simplicity to the spin-1/2 case, it is convenient to rewrite the
Sz labels ±1/2 as “↑” and “↓”.
Any Hamiltonian Hspin ({Si }) in terms of spins (in a finite system) can always be
written as a polynomial in the 3N spin components. The same spin Hamiltonian could
come from diverse origins. Once we have it, it is irrelevant what the internal degrees of
freedom were that led to it – they only describe high-lying excited states. I think of the
spin as a quantum object with a finite state space.

Algebra of spin operators

Most readers should be familiar with the following algebraic relations, collected here
for reference. But keep in mind that we now picture the spin as an abstract object in
its own right, rather than an angular momentum.
The Hamiltonian for such a system is naturally built out of spin operators (Six , Siy , Siz ),
which transform as a vector and are defined to act as follows on the basis states: 1

Ŝiz |Siz ii = Siz |Siz ii (4.1.2)

p
Ŝi+ |Siz ii = S(S + 1) − Siz (Siz + 1) |Siz + 1ii (4.1.3)
p
Ŝi− |Siz ii = S(S + 1) − Siz (Siz − 1) |Siz − 1ii (4.1.4)
Here | . . . ii means the basis state for the spin on lattice site i. Also S is the total length
quantum number of the spins, a constant. which depends on the ion species(including
its ionization state); Lec. 4.2 [omitted] shows how you could figure it from the start.
When different spin sites are inequivalent, their spins might have different S values.
We’re usually interested in a lattice with a macroscopic number of spins. So the
i indices are written explicitly in (4.1.2), (4.1.3), (4.1.4), to make the point that each
spin operator acts on just one site. When the basis states are written as in (4.1.1), the
operator with index i affects only the label with index i, e.g. in a chain of five S = 1/2
spins:
S3+ |↑↓↓↑↑i = |↑↓↑↑↑i (4.1.5)
2
From (4.1.2), (4.1.3), (4.1.4), the commutation relations follow

[Siz , Sj± ] = ±δij Si± (4.1.6)

1 I’ll
usually omit the hats that indicate a quantum operator.
2 Here’s how I remember the commutators. First, to know what operator I get, I notice e.g. that
Si+ carries a net z spin of +1 so I know that either term in the first commucator, S i+ Siz − Siz Si+
carries a net z spin of +1, their sum does too, thus it can only be ∝ Si+ . To determine the coefficient
(which is independent of the spin length S), I check the S = 1/2 case where S+ | − 21 i = | + 12 i and
S− | + 12 i = | − 12 i.
4.1 B. SEMICLASSICAL VIEWPOINT 3

[Si+ , Sj− ] = δij 2Siz (4.1.7)

When we write Si± = Six ±iSiy , we find commutation relations [Six , Siy ] = iSiz (also
cyclic permutations of xyz) and indeed Ŝ is a vector. The spin’s length is S2i = S(S +1).

Comparison to fermion and boson operators

In Table 4.1.1, the spinless fermions/bosons are in discrete orbitals, with just one
orbital per lattice site. The full algebra of operators includes (i) what I called “label”
operators – diagonal operators whose eigenvalues are used to label the basis states (ii)
“ladder” operators – off-diagonal creation/annihilation or raising/lowering operators
which connect between eigenstates with adjacent values of the label operators.

What Labels states/site Ladder operators Commutators

spinless fermions ρ̂i = 0, 1 2 ci , c†i {ci , c†i } = 1
bosons; harm. osc. ρ̂i = 0, 1, . . . discrete ∞ bi , b†i [bi , b†i ] = 1
spin Siz = −S, . . . , +S 2S + 1 Si+ , Si− [Si+ , Si− ] = 2Siz

Table 4.1.1: Discrete quantum models on lattices: three kinds of operator

The mathematical structure is very similar to that of a boson or fermion operator

(see Table 4.1.1). On the one hand, spins are like fermions in that their Hilbert space
is finite-dimensional. On the other hand, they are like bosons in that operators from
different spins commute ((4.1.7)). The similarity to bosons is exploited in various exact
mappings to particle operators such as the Schwinger bosons or the Holstein-Primakoff
representations, discussed in Lec. 5.5 .
Spin operators can also be written as bilinears in electron operators:

1
Siz = (c†i↑ ci↑ − c†i↓ ci↓ )
2
Si+ = c†i↑ ci↓
Si− = c†i↓ ci↑
(4.1.8)

The Wigner-Eckart relation says that, we project the Hilbert space of some electrons
to a subspace labeled by a single spin operator, then any vector operator projects (within
that space) to a multiple of the spin operator. Sorry, I’m not prepared with a a good
explanation; it is an elementary consequence of group representation theory.

4.1 B Semiclassical viewpoint

).
There is one big difference between spin and particle operators: in particles, the
occupation number basis is natural to our semiclassical thinking since we think of a
particle as being in a given place in a given time. Quantum perturbation terms can be
visualized as “virtual” proceses in which a particle hops to some other state temporarily,
and this may be given a precise meaning within a path integral formalism.
In spins, on the other hand, it is natural to think of them as having a “direction”.
But the√uncertainty relations for spin say the direction on the unit sphere is uncertain
to O(1/ S). This justifies approximately acting as though we could specify (Sx , Sy , Sz )
4 LECTURE 4.1. SPIN HAMILTONIANS AND EXCHANGE INTERACTIONS

and represent the spin classically as a point on the unit sphere, a vector of fixed length.
(This semiclassical viewpoint would be developed later in Part 4, or 5, for spin waves
in particular.) The customary |Sz i basis is handier for doing calculations; in it the z
component is perfectly definite and the others are completely uncertain.
There are systematic expansions in powers of 1/S. Rather surprisingly, they often
work – at least as a qualitative picture – even when S = 1 or S = 1/2.

Uncertainty principle for spins

Given a spin is in a state of definite and maximum Sz = S, the relative mean-squared

deviation
h|S − Sz ẑ|2 i Const
2
≈ (4.1.9)
hS i S
Thus, the angular uncertainty is of O(S −1/2 . This means that, in the limit S → ∞, we
may consider a spin as having a definite direction, just as in the limit of small ~ we may
consider an object as having both position and momentum well-defined.

Semiclassical spin dynamics

This result below can be thought of as the analog for spins of Newton’s equations; the
condition for static equilibrium is also there (that every spin be aligned with its “local
field”). They are very convenient for visualization, since we are used to a classical world.
Quite generally, if H(S) is a single-spin Hamiltonian, we have

dS/dt = γS × h(S) (4.1.10)

where
−gµB h(S) ≡ δH/δs. (4.1.11)
Here h(S) is called the “local field”.

4.1 C Spin couplings

A spin Hamiltonian (almost always) consists of a sum of one-spin and two-spin terms.
This is very analogous to the Hamiltonian of a particle system, where one has one-body
terms (an external potential) plus two-body terms (particle-particle interactions). The
terms are best visualized by pretending the spin length S is long and imagining S to be
a c-number vector.
Since we will be interested in extended arrays of spins, the (continuous or discrete)
symmetries of the spin Hamiltonian with respect to rotations in spin space are all-
important. There are two reasons for this.

(i) In an ordered state, symmetries of the Hamiltonian will be spontaneously

broken. [See Lec. 1.3 and Lec. 1.4 .] The possibilities of topological defects and
of Goldstone modes depends on this.

(ii) Invariance of the Hamiltonian under continuous rotations about axis â implies
conservation of the â spin component. Conservation laws affect the dynamics
including the nature of the spin-wave dispersion as well as transport properties.
A more current interest: if you’re doing spintronics, you care whether spin is
conserved in your material!
4.1 C. SPIN COUPLINGS 5

“Isotropic” terms of the Hamiltonian are invariant under rotations in spin space
(unaccompanied by real space). Terms which violate rotation symmetry are called
anisotropies. All of them are due to spin-orbit coupling except the dipolar. Frequently
system is rotationally symmetric at zero order, but anisotropy terms are present as
small perturbations. The form of anisotropic terms is directly dependent on the local
symmetry rotation symmetry (of the site or sites being coupled plus their neighbors).
When the crystal structure is not a Bravais lattice, the local symmetry may be lower
than the crystal’s point group, or alternatively, an approximate but practically exact
local symmetry may be higher than the point group.
A general spin Hamiltonian can be written in the following form,

Hspin = (HH + HAn ) + (Hex + HDM + Han−ex + Hdip ) (4.1.12)

I’ve grouped the single-spin and multiple-spin terms in (4.1.12); below, I’ll discuss each
term, starting with the one-spin terms.

Magnetic field coupling

An external field H couples as

X
HH = −H · g i µB S i (4.1.13)
i

This term looks anisotropic in that H defines a special direction in space. 3 But the
material is isotropic in spin space, in the sense that the strength of its field coupling is
independent of the field’s direction. P
It’s convenient to rewrite HH = −H · i Si , absorbing the gµB coefficient into
the magnetic field H, which thenceforth has the units of energy. This term is small
compared to the others: recall µB = 0.0578meV/T, that fields over ∼ 20 T require
large and expensive magnets, and that a typical exchange constant is 10meV (often
more). In particular, it is rarely possible to apply a field large enough to force a system
into a nearly saturated state (all spins nearly parallel), if the spin-spin couplings favor
some other state.
Most generally, when microscopic spin-orbit scattering is important P and the local
symmetry is less than cubic, we should replace (4.1.13) by HH = −H · i gi µB Si where
gi (a 3 × 3 matrix) is the “g-tensor” of site i, which must have the same rotational
symmetries as the site. 4

Single-ion anisotropy

This term has the form X

HAn = EAn (i) (Si ) (4.1.14)
i

If a given point group operation around a site leaves all its neighbors in the same places,
then the same point operation applied to (4.1.14) should leave it invariant.
A very common form is uniaxial anisotropy
1
EAn (S) = − DSz2 (4.1.15)
2
3 Actually, if the only other terms are exchange, the only effect of (4.1.13) is to set the system into

uniform precession around the H axis at angular frequency ω = gµB H. (See Lec. 4.6 [Omitted]on spin
resonance.)
4 The anisotropic g tensor is to be explained in Lec. 4.2 [omitted] or Lec. 4.3 .
6 LECTURE 4.1. SPIN HAMILTONIANS AND EXCHANGE INTERACTIONS

or EAn (i) (S) = − 12 D(S · n̂i )2 where the unit vector n̂i is called the spin’s “easy axis”.
The mostPgeneral single-spin term, quadratic in spin components, can be written
HAn (S) = 12 αβ Kαβ Sα Sβ where Kαβ must be symmetric. If we diagonalize the matrix
{Kαβ }, then after a spin-space rotation that makes the principal axes into x0 , y 0 , z 0 , the
general form is
1 2 1 2 2
HAn = − DSz0 + D0 (Sx0 − Sy0 ) (4.1.16)
2 2
There were three independent eigenvalues of {Kαβ }, but one combination of them
corresponds to the unit-matrix component of {Kαβ } which gives a trivial constant
S2 = S(S + 1)... A corollary of the symmetry lemma is that when the local environ-
ment has mirror planes (symmetric under reflection in those planes), the principal axis
directions must lie in them or perpendicular to them.
Finally, if the environment has tetrahedral or cubic local symmetry, there is no
nontrivial quadratic term; the first anisotropic term is

EAn cubic (S) = K(Sx4 + Sy4 + Sz4 ) (4.1.17)

(there is only one such term). This cubic anisotropy favors spins along h111i axes when
K > 0 or h100i axes when K < 0.

Exchange

The exchange interaction (sometimes called Heisenberg exchange) is bilinear in spins

and isotropic under rotations:
X
Hex = − Jij Si · Sj (4.1.18)
i<j

For us, the name “exchange” does not refer to any particular mechanism but merely to
the dot-product form, which guarantees rotation symmetry. Notice The interactions in
(4.1.18) may extend beyond first neighbors. Exchange couplings (the coefficients J ij )
are called “ferromagnetic” (resp.“antiferromagnetic”), when they favor favoring parallel
(resp. antiparallel) alignment of interacting spins.
I now mention some algebraic tricks related to exchange couplings. Let P̂12 be the
operator that exchanges spin 1 and spin 2, i.e. P̂12 |σ1 σ2 i ≡ |σ2 σ1 i. Then we can show
1
P̂12 = + 2S1 · S2 . (4.1.19)
2

As you know from basic quantum mechanics, one may construct (using Clebsch-
Gordan coefficients) combined states of two spins such that Stot ≡ S1 + S2 has a
definite spin Stot .
Then
1   1
S1 · S2 = (S1 + S2 )2 − S21 S22 = [Stot (Stot + 1) − 2S(S + 1)]. (4.1.20)
2 2

It can be shown (see (Ex. 4.1.1)(b) that for spin 1/2 there are just two eigenvalues
of JS1 · S2 , corresponding to a combined singlet or triplet; the splitting is J, and the
average over the four eigenstates is zero.

Other two-spin terms

P
We could write a most general bilinear form, with terms αβ Mαβ Siα Sjβ , where α, β
label Cartesian components of the coupling matrix {Mαβ }. Any 3 × 3 matrix may be
4.1 C. SPIN COUPLINGS 7

decomposed into (i) a multiple of the identity matrix, (ii) an antisymmetric part (three
different coefficients), and (iii) a traceless symmetric part (five different coefficients).
[One could also say these correspond to the ways of combining two spherical harmonics
of angular momentum 1 (as characterizes the vector operator S, whatever the spin
quantum number S), so as to make a net angular momentum 0, 1, or 2, respectively.]
These three terms give respectively Hex , HDM , and (Han−ex + Hdip ) in (4.1.12). The
bilinear terms, besides exchange, are anisotropic exchange ( Han−ex and antisymmetric
or “Dzyaloshinskii-Moriya” term) and finally dipole-dipole.
The Dzyaloshinskii-Moriya interaction, or antisymmetric exchange, has the form
X
HDM = − Dij · Si × Sj (4.1.21)
i<j

It is also possible to have anisotropic exchange, for example

X xy
z
Han−ex = − Jij (Six Sjx + Siy Sjy ) + Jij Siz Sjz (4.1.22)
i<j

As far as its symmetry in spin space, the dipolar interaction is a special case (ex-
tending beyond nearest neighbors) of the “traceless symmetric” case above, i.e., as part
of Han−ex . However, in this case the interaction does not depend on the crystal axes,
and its microscopic origin is not in exchange, so it makes sense to treat it separately:
X (gµB )2
Hdip = 3 [3(r̂ij · Si )(r̂ij · Sj ) − Si · Sj ] (4.1.23)
ij
rij

This is long-range and responsible for the demagnetizing field, ferromagnetic domains,
etc. Dipolar interactions are important when exchange is small, and also in nuclear
magnets.
They can lead to different dynamics and to different critical exponents in phase
transitions.

Multi-spin terms?

In modeling the stability of crystal structures (see Lec. 1.2E ), three- and four-atom
effective potentials can be important. By contrast, three- or four-spin potentials rarely
appear in spin Hamiltonians. They arise (i) in solid 3 He (a nuclear spin S=1/2) arising
from a ring tunneling of atoms mentioned in Lec. 4.3 ; (ii) the effective interaction when
you have a magnetoelastic coupling and eliminate the elastic degrees of freedom. In
addition, ring-exchange interactions may appear in the cuprate antiferromagnets which
become high-Tc superconductors upon doping.

Exercises
Ex. 4.1.1 Spin 1/2 exchange interactions

(a) Here’s how a permutation can be converted into a dot product operation. Each
operator in the table annihilates 3 of the 4 possible states, and turns the other one
into its (1 ↔ 2) permutation. The sum of the terms in the left column is thus the
permutation exchange operator P̂12 ; show that it adds up to (4.1.19).
(b). Use (4.1.20) to show that, if S1 and S2 are both s=1/2 spins, then S1 ·S2 = +1/4
or −3/4 are the only eigenvalues of this operator.
(c). Express (S1 + S2 )2 in terms of S1 · S2 for the case of spin-1/2 spins.
8 LECTURE 4.1. SPIN HAMILTONIANS AND EXCHANGE INTERACTIONS

Operator |↑↓i |↓↑i |↑↑i |↓↓i

S1+ S2− 0 |↑↓i 0 0
S2+ S1− |↓↑i 0 0 0
( 21 + S1z )( 12 + S2z ) 0 0 |↑↑i 0
( 21 − S1z )( 12 − S2z ) 0 0 0 |↓↓i

Table 4.1.2: Pieces of two-spin-1/2 exchange operator

Ex. 4.1.2 Spin uncertainties

(a). Confirm (4.1.9). Also, what is the ratio

h|[six , siy ]|2 i

? (4.1.24)
hs2ix ihs2iy i

What happens to (4.1.9) and (4.1.24) in the limit S → ∞?

Ex. 4.1.3 Semiclassical spin dynamics

(a). Consider
H0 = −gµB H · s (4.1.25)
Use Heisenberg’s equation of motion
dX
~i = [X, H0 ] (4.1.26)
dt
to find
ds/dt = γs × H (4.1.27)
What is γ?
(b). Now consider a generalization of (4.1.25)
X
H0 (s) = λklm skx sly sm
z . (4.1.28)
klm

Construct a semiclassical equation of motion as follows: First expand the commutator

from (4.1.26), so that each term in (4.1.28) produces many terms. Adopt the approxi-
mation that this result is made of c-numbers which commute freely: those many terms
should collapse back to one term.
Within this approximation, does it matter in which order we write the factors in
each term of (4.1.28)?
Show that the result can be expressed in the form (4.1.10) .
(c). Let s̄0 be the direction for the classical ground state, minimizing H0 (s) (with
the fixed-length constraint |s0 | = s).
What relation must hold between the directions of s and h(s) in a ground state?
Does this make sense, in the light of (4.1.10)?
(d). (OPTIONAL) What is the frequency ω of small oscillations around the ground
state direction s0 , in terms of h evaluated at the ground state? (Hint: Eqs. (4.1.10) and
(4.1.11) are rotationally invariant, as they ought to be, so you may choose your axes for
spin space such that h(s0 ) is along z 0 .) Then linearize in the components transverse to
s0 ).
4.1 C. SPIN COUPLINGS 9

Ex. 4.1.4 Ferromagnetic exchange in a ring

assigned in 2003 only

Compare the exact-diagonalization exercise in Lec. 1.1 .
Consider a ring of three sites, at position 0, 1, 2, (so lattice constant =1). Each site
has one orbital, with room for an up and a down spin. The HamiltonianPis a Hubbard
2
model, with positive hopping matrix element +t and a repulsive energy r=0 U n̂r↑ n̂r↓
when two electrons (necessarily ↑ and ↓) are on the same site. Two electrons are placed
on the ring.
(a). The noninteracting (single-particle) dispersion is (k) = 2t cos ka, where k n =
(2π/3)n. Show that the single-particle ground state is doubly degenerate. (This is due
to t > 0 and also to the ring having an odd number of sites.)
(b). Construct the noninteracting ground state for spins |↑↑i. Imagine that U is
small and evaluate its effect in first-order perturbation. Show hn̂r↑ nr↓ i = 0 for the |↑↑i
case and 1/9 for either of the two kinds of |↑↓i case. (You could place opposite spins
both in the same orbital, or in two different ones.)
(c). Consider the case |k1 ↑, k2 ↓i (recall k2 ≡ −k1 modulo 2π, so the net “wavevector”
is zero.) Expanding out its terms, and keep only those which have both electrons on
the same site. Now do the same for |k2 ↑, k1 ↓i. Verify that in either the sum or the
difference (which!?) of these two wavefunctions, all the terms with double occupancy
cancel, and therefore the noninteracting energy is obtained as for the |↑↑i wavefunction.
What is the relation of this wavefunction to the |↑↑i state?
(d). OPTIONAL!! Consider instead the case U = ∞, meaning two particles are
never allowed on the same site. Solve this exercise by writing a graph with all possible
states of the spins in a ring. (There are three states in the |↑↑i case and six of the |↑↓i
case). Find the ground state in either case, and verify that the ferromagnetic case is
lower. In this case, the exchange energy scale is necessarily O(t), and you can see the
difference comes entirely through Fermi statistics.
10 LECTURE 4.1. SPIN HAMILTONIANS AND EXCHANGE INTERACTIONS
 11

Lecture 5.0

Overview of spin order

This version is edited for Physics 654. (Lec. 5.0A and 5.0B were an overview of the
Hubbard model.)

5.0 A Semiclassical treatment of spin system

In this section, we consider a local-moment model. The degrees of freedom are N spins
si , each of length S. I’ll explain how, in the limit S  1, the spins behave as (nearly)
classical objects. It’s useful to have such a limit, as a means to visualizing spin behavior
with our (classical) minds, or to approximate them using classical statistical mechanics.
Furthermore, as with particle systems, any quantum result must pass the test of Bohr’s
correspondence principle – but the classical limit of a spin is not as obvious as Newtonian
mechanics. In actuality, the spin of an ion is at most S = 7/2, but it turns out that
S & 3/2 seems to be sufficient that the large-S limit is qualitatively correct.

Spin coherent states

Consider, for now, just one spin S. Let |ψẑ i be the state at the top of the ladder of
Sz states : its wavefunction is entirely on the state Sz = +S; equivalently, it satisfies

Ŝ z |ψẑ i = S|ψẑ i. (5.0.1)

Next, for any unit vector n̂, find a rotation matrix R such that Rẑ = n̂, and let RR be
a unitary operator that applies the same rotation to the spin. 1 Obviously,

Ŝ|ψn̂ i = S n̂|ψn̂ i; (5.0.2)

this state is pointing in the n̂ direction as strongly as it can. It turns out these states
are the analog of minimum-uncertainty wavepackets.
Notice that coherent states in different directions aren’t exactly orthogonal (only
when in opposite directions). Namely
S
1 + n̂ · n̂0

1 S
|hψn̂ |ψn̂0 i| = = | cos θ(n̂, n̂0 )|2S ≈ exp[− θ(n̂, n̂0 )2 ] (5.0.3)
2 2 4
1 There is more than one rotation matrix satisfying Rẑ = n̂; it turns out that you get the same state

RR |ψẑ i, except there is a phase factor; this is exactly like the gauge freedom in defining the phase of
the real-space state ri, in the case of a particle. Thus, for some calculations one must specify a gauge
choice, or else work with manifestly gauge-invariant quantities; but none of that matters in this lecture.

Copyright c 2007 Christopher L. Henley

12 LECTURE 5.0. OVERVIEW OF SPIN ORDER

where θ(n̂, n̂0 ) is the angle between their directions, and this is obviously like the overlap
between two offset Gaussian wavepackets.

Semiclassical limit notion

A spin vector has a well-defined direction in space in the limit S → ∞ (this is

the analog of ~ → 0 in a particle system.) For example, let’s take a coherent state,
and compute the spin’s quantum fluctuations about its expectation. It turns out (see
homework [?]) you get hδS2 i = S, i.e. the relative angle fluctuation is

hδS2 i1/2
∼ S −1/2 → 0 (5.0.4)
S
as S → ∞.
This makes common sense: the Hilbert space of a spin has dimension 2S + 1, with
which it has to represent all possible directions. So, if we want basis states that are
maximally localized in direction, each one
pmust cover 4π/(2S + 1) of solid angle, which
would be (for large S) a circle of radius 2/S radians.

Conjugate variables

Take a coherent state along ẑ. If S  1, the quantum fluctuations are small and
Sz can be approximated as a c-number. Consider the commutator [δSx , δSy ] = −iSz ≈
−iS. So long as we can treat it as a c-number, this is the exact analog of [X̂, P̂ ] = −i~
for a particle system, and hence the different components are the canonical conjugate
variables. This is the basis (see below) for semiclassical spin wave and spin precession
dynamics; also of spin tunneling.
But there’s a difference. In a particle system, we think of position and momentum
as different quantities, and they come in different units, so it’s natural to write wave-
functions in terms of one of them, e.g. ψ(x), or to define a Green’s function (e.g. in
Lec. 2.2) as the amplitude of a particle to get from position r at time t to position r 0 at
time t0 : we’re not bothered that specifying the position exactly means that momentum
is indefinite (hence our real-space basis function contains all possible wavevectors).
On the other hand, the different components of a spin are manifestly equivalent
quantities, often related by symmetries; and our intuition leads us to think of a well-
defined direction in spin space. (The analog of ψ(x) would be a wavefunction in terms
of say Sz , so the azimuthal angle would be totally indefinite.) However, the uncertainty
relation (5.0.4) means a spin state with a truly definite direction is impossible: the
coherent state basis is the best we can do (even though it’s not an orthogonal basis)

Multi-spin states

In general, the wavefunction is Ψ(S1z , S2z , S3z , . . .); it depends on the discrete [(2S +
N
1) -dimensional] Hilbert space. We can write a product wavefunction
Y
Ψcoh ({Siz }) = ψn̂i (Siz ) (5.0.5)
i

thus

hSi iΨcoh =S n̂i ; (5.0.6)

2
hSi · Si iΨcoh =S n̂i · n̂j . (5.0.7)
5.0 A. SEMICLASSICAL TREATMENT OF SPIN SYSTEM 13

(This follows since the wavefunction Ψcoh incorporated no quantum correlations between
sites; it’s simply a direct product.)
Think of Ψcoh as a variational wavefunction, specified by the set of classical directions
{n̂i } as parameters. It plays the same role as the classical ground state of a many-
particle system, the minimum of potential energy U ({xi }) as a function of posititions.
spin
Minimizing hĤhop iΨcoh is equivalent to minimizing a classical function H spin ({n̂i }).
This is what we mean by a classical spin Hamiltonian.
A next step would be to incorporate the zero-point fluctuations about this minimum,
as done for phonons in Lec. 1.5 . 2
Alternatively, at T > 0, we could replace each coherent state by a thermal distri-
bution: technically, a density matrix. The whole system’s density matrix is the direct
product of the density matrices for the respective spins. This yields a form of mean
field theory, which somewhat takes into account the quantum nature of the spins, but
also describes the high-temperature paramagnetic state and ordering transitions

Equations of motion

This subsection was missing in the previous notes I’m typing up... it may have been
postponed to Lec. 5.5 in a previous version. So this is just a sketch, and in particular
the signs and coefficients are not right.
The “local field” is
δH
hi ≡ (5.0.8)
δsi
and thus is exactly the analog of a force (except, since Si has fixed length, we can’t
actually displace in that direction). In particular say the Hamiltonian is pure exchange,
1X
Hex = − Jij si · sj ; (5.0.9)
2 ij

then X
hi = Jij sj . (5.0.10)
j
P
This is called “local field” because hi = H if the Hamiltonian were just −H · i si , an
external field.
The condition for a local minimum is si = S n̂i , with

n̂i = hi /|hi |. (5.0.11)

The dynamics which follow (e.g.) from the commutators are simple: each spin
precesses around its local field. It’s like the motion of a gyroscope, or a particle in a
strong magnetic field (see Part 9 on quantized Hall effect): it moves (along its unit
sphere) perpendicular to the direction it’s pushed:

d 1
~ n̂i = n̂i × hi . (5.0.12)
dt S

2 In a ferromagnet, Ψcoh is actually the exact ground state so there are no fluctuations. The fluctu-

ations are a consequence of the fact that the Hamiltonian doesn’t commute with the order parameter.