Lecture 12 (Monday 03/06)

Link to recorded lecture: https://youtu.be/NWVn7FkHG5Y

Quantum Ising model

As a paradigmatic example for quantum phase transitions, we will study in details the spin-1/2 transverse
field Ising model, commonly known as the quantum Ising model
X X
HT F IM = − σiz σjz − g σix
⟨ij⟩ i

where we have set the Ising interaction strength to be unity, and the dimensionless parameter g above
controls the ratio of the transverse field strength to the interaction strength.
In one dimension, if the Ising interactions are nearest-neighbor only, this model can be exactly solved
through a mapping from the spins to non-interacting fermions known as the Jordan-Wigner transforma-
tion. But before we do that, let us first try to understand the ground state of this model qualitatively:

• Let us first consider the limit of g ≫ 1. In this case, the transverse field term dominates the
Hamiltonian, and to leading order in 1/g, the ground state |ψ0 ⟩ is simply
O
|ψ0 ⟩ ≈ |σix = 1⟩
i

i.e. a product state with all spins pointing in the +x direction, known as a paramagnetic state.
Obviously ⟨σix σjx ⟩ = 1 in this state for any i and j, meaning that the spins are perfectly correlated
in the x direction. However, in this state the spins are perfectly uncorrelated in the z direction,
because |σix = 1⟩ = √12 (|σiz = 1⟩ + |σiz = −1⟩) and each spin has 1/2 probability of being in the
+z or −z direction independently when measured, leading to ⟨ψ0 |σiz σjz |ψ0 ⟩ = δij .

• If we now consider corrections to the above ground state to the next order in 1/g, we expect ⟨σiz σjz ⟩
to be close to δij . In fact, we can later show that the correlations in z now decays exponentially

⟨ψ0 |σiz σjz |ψ0 ⟩ ∼ e−rij /ξ

where rij is the distance between the spins i and j on the lattice.

• Next, we consider the opposite limit of g ≪ 1. In this case, the Ising interaction dominates the
Hamiltonian. At g = 0, we have two exactly degenerate ground states

|ψ↑ ⟩ = ⊗i |σiz = 1⟩ |ψ↓ ⟩ = ⊗i |σiz = −1⟩

which are respectively a product state with all spins in the +z and in the −z directions, which is
known as ferromagnetic states.

• Now turning on a small g will mix a small fraction of spins in the opposite directions along z.
But in an infinite system the degeneracy between |ψ↑ ⟩ and |ψ↓ ⟩ will survive at any finite order
in perturbation theory. This is because only the N -body operator ⊗i σix can couple the state |ψ↑ ⟩

1
 to |ψ↓ ⟩. Therefore, the energy difference between |ψ↑ ⟩ and |ψ↓ ⟩ is on the order of g N since the
perturbation theory of less than N orders will not resolve this energy difference. For g ≪ 1 this
energy difference is exponentially small in N , and in practice, the ground state can be regarded as
degenerate and it can be in either the |ψ↑ ⟩ state or the |ψ↓ ⟩ state. This means in practice, the spins
will have nonzero magnetization in the z direction for small g. But such ground state this breaks
the Z2 symmetry of the Hamiltonian (the symmetry that if we flip each spin’s y and z components,
the Hamiltonian is invariant). This indicates a spontaneous symmetry breaking (in practice it’s due
to inevitable noise on the system because the true ground state is a superposition of |ψ↑ ⟩ and |ψ↓ ⟩,
a cat state that is very fragile to collapse to either |ψ↑ ⟩ or |ψ↓ ⟩). All non-topological continuous
phase transitions are associated with spontaneous symmetry breaking (Landau’s theory).

• ⟨ψ0 |σiz σjz |ψ0 ⟩ = 1 at g = 0 and remain finite for small g > 0. There is no way that this can smoothly
transit to ⟨ψ0 |σiz σjz |ψ0 ⟩ ∼ e−rij /ξ . Therefore we argue that there must be a phase transition at some
critical gc

• Alternatively, we can look at this phase transition using σix , the order parameter from the paramag-
netic phase.

Jordan-Wigner transformation
Let us now try to quantitatively solve the ground state of the quantum Ising model using a powerful known
as Jordan-Wigner transformation, a mapping between spin-1/2 operators and fermionic operators. The
basic idea is that a spin-1/2 Hamiltonian on a lattice can be mathematically mapped to a Hamiltonian
for identical fermions on a lattice, although they are physically very different systems (one ignores the
motional degrees of freedom and the other ignores spin degrees of freedom). By choosing some direction
as the z direction, a lattice site with a spin-1/2 in the down state (i.e. σ z = −1 state) can be regarded
as a lattice site with no fermion, while a spin-1/2 in the up state can be regarded as a lattice site with
one fermion. The two-state nature of a spin-1/2 matches with the two-state nature of a lattice filled
with identical fermions. As a result, the creation operator for the fermion on lattice site i, denoted by
c†i , corresponds to the spin flip operator σi+ on the site i. Similarly, ci corresponds to σi− , and c†i ci ↔
σi+ σi− = (| ↑⟩⟨↓ |)(| ↓⟩⟨↑ |) = | ↑⟩⟨↑ | = (Iˆ + σiz )/2. In other words, σiz corresponds to 2ni − 1.
However, this simple intuition has a problem: We know that for two different sites i and j, c†i c†j =
−c†j c†i (anti-commutation relation), but σi+ σj+ = σj+ σi− . Note that here σi− is a many-body operator
defined by
σi+ = I| ⊗{z· · · I} ⊗σ + ⊗ I| ⊗{z
· · · I} (1)
i−1 N −i

This convention is implied whenever we use a spin operator on a lattice, such as the σiz and σix in the
quantum Ising model. Otherwise, we cannot multiply σi− and σj− . Using such definition, it’s easy to see
that σi+ σj+ = σj+ σi− for i ̸= j.
To conserve the fermionic anti-commutation relation, we need to have a more complicated map-
ping between ci and σi− . Recall that in Problem 2 of Homework 2, you showed that if we define
|n1 , n2 , · · · , nN ⟩ ≡ (c†N )nN · · · (c†2 )n2 (c†1 )n1 |vac⟩, then we can show that
(
0 ni = 1
c†i |n1 , n2 , · · · , nN ⟩ = PN
n
(2)
(−1) j=i+1 j |n1 , · · · ni + 1, · · · , nN ⟩ ni = 0

2
Note that the definition of the many-body Fock state |n1 , n2 , · · · , nN ⟩ ≡ (c†N )nN · · · (c†2 )n2 (c†1 )n1 |vac⟩
is not unique and one can change the order of the creation operators arbitrarily, as long as this same
order is used everywhere else. For example, some literature use the opposite order: |n1 , n2 , · · · , nN ⟩ ≡
(c†1 )n1 (c†2 )n2 · · · (c†N P
)nN |vac⟩.
N
The phase (−1) j=i+1 nj in the above equation is a direct result of c†i c†j = −c†j c†i for i ̸= j. This means
PN
we can map σi+ to c+ i (−1)
j=i+1 nj in order to obey the fermionic anti-commutation relation, given that

the fermionic state |n1 , n2 , · · · , nN ⟩ is mapped to the spin state ⊗i |σiz = 2ni − 1⟩. We have now arrived
at the formal Jordan Wigner transformation for mapping spin-1/2s to fermions:
PN
σi+ = c+
i (−1)
j=i+1 nj
PN
σi− = (−1) j=i+1 nj
ci
σiz = 2c†i ci −1
The inverse Jordan-Wigner transformation is given by
N
Y
c+ +
i = σi (−σjz ) (3)
j=i+1
N
Y
ci = σi− (−σjz ) (4)
j=i+1

ni = (1 + σiz )/2 (5)

Note that the Jordan-Wigner transformation is a mapping between N spins on a lattice of N sites and
identical fermions (the number of fermions depend on the total σ z of the spins) on a lattice of N sites.
It works for any Hamiltonian on any lattice, although a simple spin/fermionic Hamiltonian may end up
being a complicated fermionic/spin Hamiltonian after the transformation due to the complicated many-
body phase operators above. Note that the Jordan Wigner transformation also provides us a way to
numerically compute the eigenenergies and eigenstates of fermionic Hamiltonians (especially those that
cannot be diagonalized into non-interacting fermions) by first mapping the fermionic Hamiltonian to a
spin Hamiltonian, although the resulting spin Hamiltonian is a 2N × 2N matrix that is exponentially hard
to diagonalize.
The main use of Jordan-Wigner transformation is to solve interacting spin Hamiltonian if it can be
mapped to a fermionic Hamiltonian that can be diagonalized into non-interacting fermions. This is par-
ticularly the case for the 1D nearest-neighbor transverse-field Ising model:
N
X −1 N
X
H=− σix σi+1
x
−g σiz (6)
i=1 i=1

Here we swapped x and z from the H in the last lecture, which has no physical effect as it is just a
redefinition of the spatial coordinate system. We have also assumed open boundary condition for the
moment but will add periodic boundary condition later.
Using the Jordan-Wigner transformation, and the fact that σix = σi+ + σi− , we end up with a fermionic
Hamiltonian
N −1 h ih i N
X PN PN PN PN X
H =− c+
i (−1) j=i+1 nj + (−1) j=i+1 nj c
i c +
i+1 (−1) j=i+2 nj + (−1) j=i+2 nj c
i+1 − g (2c†i ci − 1)
i=1 i=1
(7)