Anda di halaman 1dari 3

Initial Notes: Understanding Toric Code to Vacuum Phase Transition

Saad Khalid

I. INTRODUCTION

I thought it may be useful to go step by step through the logic in some detail, so that we are all on the same page.
I must admit, there are a few steps that I haven’t fully figured out yet, but I thought it would be useful to have
everything laid out before the meeting this week.

II. RECREATING THE BASIC PICTURE

We begin with the Toric code on a cylinder, imagining a finite width and periodic boundary conditions vertically.
There are two possible boundaries, either "rough" or "smooth." A description of this is given in Quantum codes on
a lattice with boundary. Lets say we are working in the σ x basis. Lets define our Bulk hamiltonian (which is the
nonperturbed toric code Hamiltonian) as:
X X
H = −Je Av − Jm Bp
v p

where
Y Y
Av = σlx and σlz .
l∈+ lin

Then, to account for the rough boundary sides, we will add a term to this Hamiltonian:
X X X
H = −Je Av − Jm Bp − Bp0
v p p∈boundary

where Bp0 is like Bp except the σ z only act on the 3 sites in the partial plaquettes on the boundary.
A state |ψ > will contain an e-particle at vertex v if Av |ψ >= −|ψ >. Thus, there is an e-particle at any vertex
where an odd number of links in that vertex are in the |1 > state. An e-string then is the chain of links where each
link has σ z acting on it (ie each link is in the |1 > state). At the ends of e-strings, typically, are e-particles.
2

As e-particles correspond to -1 eigenvalues for the Av operator, they clearly have an energy cost in our given
Hamiltonian.
Note that we can move e-particles around by simply acting with σ z on them, as is shown in the diagram. Now
imagine an e-particle on the vertex next to the left boundary. If we act with the σ z operator on the link to the left of
that vertex (the "rough" part of the boundary) then we will have moved the e-particle off the edge into the vacuum.
This is because there are no vertex operators defined on the ends of the links hanging off of the rough boundary, and
so there can’t be an e-particle there. Thus, we say that the rough boundary condenses e-particles. This gives us
two properties. First, this gives us the ability to create single e-particles, as opposed to the toric code on the torus
which only allowed e/m-particles in pairs. Also, this tells us that we can create e-strings from boundary to boundary
without any energy cost (another way to say this is that the operator which gives an e-string from boundary to
boundary commutes with the Hamiltonian). As an aside, the smooth boundary condenses m-particles, the analysis
in that case is isomorphic.

For the toric code on the cylinder with rough boundaries, there are two ground states. The common proper-
ties of the two ground states are that they are both superpositions of all possible spin |1 > loop configurations in
the bulk, similar to the ground states on the torus. Because of the additional boundary term, our ground states will
also be superpositions of all possible applications of Bp0 on the boundary. Both groundstates have these aspects in
common. Now, notice that if we start with a ground state with no e-strings, then we can create 2 e-strings at no
energy cost by applying Bp operators from boundary to boundary. Similarly, we can make 4, or any even number.
However, we can not create an odd number of e-strings using the Bp operator. Thus, one ground state corresponds
to having the superposition of all possible even number of e-strings from boundary to boundary (in addition to the
other constraints mentioned earlier), and the other ground state is all the superposition of all combinations of odd
numbers of e-strings from boundary to boundary. These Q two ground states can be classified by the eigenvalues of the
noncontractible m-string loop operator, defined by l∈L σlx , where L is a noncontractible loop defined on the dual
lattice going around the cylinder. This is like the τ x defined in Dr. Lu’s notes.

The Effective Hamiltonian

We will now try to move from this picture of the cylinder to the picture of a toric code on a sphere with vacuum
islands punched into it. First, notice that the toric code on a sphere with 2 vacuum islands in it is equivalent to the
cylinder picture (we are again continuing with the rough boundary case, WLOG). Thus, it has the same two ground
states, and e-strings going from island to island cost no energy.

Lets intuitively analyze our dual-island picture. Notice that as the island size goes to 0, it corresponds to
the boundaries of the cylinder closing and becoming a sphere, which has only 1 ground state. On the other hand,
if we send island size to infinite, this corresponds to having a cylinder with its width shrunk as small as possible.
Effectively, we can ignore the bulk in this case and all that matters is the boundary (our sphere has become a thin
ring, all thats left is a column of placquettes and the rough boundary on both sides of it). However, this still has
two ground states, characterized again by the τ x operator, which is again just a noncontractable m-string loop going
around in the period direction. Now, this is a point of confusion for me, but don’t we transition from the 1 g.s. phase
to the 2 g.s. phase as soon as our two islands are created, regardless of their size? Or is the arguement that, when
island size is very small, it effectively behaves as if it were a smooth sphere with no punctures? I’m confused because
I thought, even if the islands were small, as soon as you have them you can identify two distinct ground states. What
am I missing?

Lets go back to our toric code Hamiltonian, and add a perturbation term to cause our phase transition. Our
new microscopic hamiltonian is:
X X X X
H = −Je Av − Jm Bp − Bp0 − hz σiz
v p p∈boundary i

Now, if hz >> Je , Jm , then this corresponds to all the spins being alligned with the field, giving us 1 ground state.
On the other hand, when hz << Je , Jm , then we have two ground states like we have been talking about, and the
two ground states are differentiated by the parity of the number of e-strings from boundary to boundary. This gives
us our minimals in order to construct an effective hamiltonian: it must transition from having 1 g.s. to having 2 g.s.
and when it has 2 g.s., the two of them should be differentiated by the parity of the number of e-strings. Up to this
point, it appears quite clear that the Transverse field ising model can model this transition, at the very least because
it obviously satisfies the first of the two conditions. What is important is to pick the terms in our ising model-like
3

effective hamiltonian that we sum over. Actually, this is where I differ slightly from what is proposed in Dr. Lu’s
notes, and instead offer a different Heff that makes a bit more sense to me; hopefully we can compare the two and see
where the confusion lies, but this is what my understanding led me to. I was thinking that a effective hamiltonian
would be:
X X
Heff = −t τix τjx − h σlz
<i,j> l

Here the τix operator is how Dr. Lu defined it in his notes, in eq 1, as the loop of σ x enclosing the island i. Also,
the sum over < i, j > is over the islands, whereas the sum over l is over every link. I couldn’t think of a way to
write the perturbation term in terms of islands, so this certainly doesn’t feel as clean as the Heff Dr. Lu proposed,
especially when we are talking about having many islands. This also differs from the Heff in his notes as, in his eq
3, the first term of Heff (which corresponds to the ground state degeneracy) is composed of e-string operators from
island i to j. I couldn’t understand how this reflected the ground state degeneracy caused by the parity in the number
of e-strings going from boundary to boundary. This is why I used τix τjx in my first term, so the ground state with
t >> h would be all islands being connected be either an even number of e-strings or an odd number of e-strings, thus
reflecting the ground state degenerate phase of the microscopic model. Hopefully I can think of a more elegant way
to write the second term. I admit that I havne’t done the perturbation analysis by hand; I wanted to get my current
understanding out before the meeting, but learning and applying the perturbation theory was going to be my next step.

Despite the confusion over the proper terms to be summing over in our effective Hamiltonian, we should still
be able to understand the intuition behind the island picture and how it drives the phase transition. I will reiterate
this in summary. We want to describe a phase transition from 2 ground states to 1 ground state (assuming we are
working in the simple cylinder case, so just two islands WLOG), as is seen in the toric code with perturbation. If we
have a sphere with the toric code on it, this has just one ground state. If we add two vacuum islands to this sphere
(topologically the same as a cylinder) we gain a twofold ground state degeneracy.
Question: It’s a bit strange to me that we are talking about perturbing our toric code and how this corresponds to
changes in island size on our sphere, when our perturbed toric code itself is on this very sphere, is it not? Meaning,
when we talk about a sphere with vacuums punched into it and a background of toric code, this background toric
code is the perturbed microscopic toric code hamiltonian from above, right? It just seemed strange to me to think
of the background toric code somehow changing the topology of the surface it is sitting on. So essentially we are
saying that if we’ve got our perturbed toric code on a sphere with 2 islands, if we increase the field (perturbation)
enough, it is as if the islands aren’t even there as far as the ground state is concerned. On the other hand, if there
is no perturbation, then the ground state should just care about the boundary, and we can sort of minimize the
bulk (ie make the islands bigger). This is speculation on my part, in trying to put the pieces together. Of course, I
may be clinging on to some incorrect assumption that makes this wrong, so please correct me if this doesn’t make sense.

Question: Is there any use in talking about what happens when we send the length of the periodic direction
to infinite? I have been assuming finite in both width and height in this discussion. The corresponding system when
we send the periodic direction to infinite seems simple enough but I wasn’t sure if there was anything interesting to
say about it using the Ising model idea.

Lastly, I came across this paper that also mapped the toric code phase transition to the Ising model, though
they don’t mention any percolation type picture. However, they do come up with some effective Hamiltonians that
are like the transverse field Ising model. Here is the link: Robustness of Topological Order in the Toric Code with
Open Boundaries

Anda mungkin juga menyukai