Long logical strings, short thermal barriers

Key claim

The two-dimensional toric code has a logical operator of weight $L$ but a local error path whose energy never rises above a constant.

Code distance measures the support of a completed error. The energy barrier measures the most expensive intermediate state along a chosen local construction of that error.

Theme

Code distance, perturbative splitting, and thermal stability are different diagnostics.

Guiding question

A noncontractible Wilson loop on an $L\times L$ torus contains at least $L$ single-edge operators.

Why can a thermal bath assemble the same loop while paying only one pair-creation energy?

Setup

Place one qubit on every edge of an $L\times L$ square lattice with periodic boundary conditions. The standard toric-code Hamiltonian is

where

All stabilizers commute. Ground states satisfy

On a torus, the ground space is four-dimensional and encodes two logical qubits. Noncontractible direct- and dual-lattice loops act as logical Pauli operators.

These are exact properties of the commuting Hamiltonian introduced by Kitaev.

We will follow the $Z$-error sector. The dual discussion exchanges

Analysis

Derivation: distance and perturbative order

A single $Z_{\ell }$ anticommutes with the two adjacent star operators. Each violated star changes its contribution to the energy from $-J_e$ to $+J_e$.

The pair-creation cost is therefore

For a connected path $\gamma$, define

At an interior vertex, two edges of $\gamma$ meet the same star. Their two anticommutations cancel. Only the endpoints violate star stabilizers:

Extending the path moves an endpoint and does not create another pair. Every nonempty open connected string has excitation energy $4J_e$, regardless of its length.

A contractible closed $Z$ loop is a product of plaquette stabilizers. It acts trivially within the ground space.

A loop that winds around the torus is not a stabilizer. It implements a logical operator $\bar{Z}$.

The shortest noncontractible loop contains $L$ edges. For the standard square toric code,

Now perturb the Hamiltonian by

Let $P_0$ project onto the ground space. A product of $n<L$ distinct edge errors cannot contain a noncontractible loop.

Consequently, terms below order $L$ may shift the common ground energy, but they cannot act as a nontrivial logical operator after projection.

The first sector-dependent term appears at order $L$. For one connected virtual process around a fixed noncontractible cycle,

This expression is the contribution of one ordering. The full coefficient also contains other paths, orderings, and normalization terms. Minimum perturbative order is the robust statement.

For sufficiently weak bounded local perturbations, a general stability theorem gives a gapped low-energy band whose width is exponentially small in $L$ under its topological-order hypotheses Bravyi, Hastings, and Michalakis.

Thus

for some perturbation-dependent $\alpha >0$ inside the stable phase. This is a zero-temperature spectral statement.

Scope and assumptions: an energy barrier is not a lifetime

Construct $\bar{Z}$ one edge at a time:

Every intermediate open string has two endpoints. The $Z$-sector energy barrier is

independent of $L$.

The dual logical process has

For an arbitrary logical failure, the bath can exploit the cheaper channel. The relevant energetic scale is therefore no larger than

This explicit path illustrates the broader constant-barrier obstruction for two-dimensional local stabilizer Hamiltonians proved by Bravyi and Terhal.

The barrier alone does not produce an exact memory lifetime. A lifetime also depends on:

• the system-bath coupling and bath spectral density;
• pair creation, hopping, and annihilation rates;
• the number and geometry of damaging paths;
• the decoder or observable used to define failure;
• the order of the limits in $L$, temperature, and observation time.

For a Markovian weak-coupling thermalization model, Alicki, Fannes, and Horodecki proved an $L$-independent upper bound on the longest relaxation time, with activated temperature dependence On thermalization in Kitaev’s 2D model.

This result uses a specified Davies-type dynamics. It should not be quoted as a theorem about every possible nonequilibrium bath.

A passive self-correcting memory requires its storage time to diverge with $L$ at fixed nonzero temperature, without repeated syndrome measurement, decoding, or engineered correction dynamics.

The standard two-dimensional toric-code Hamiltonian does not meet that criterion.

Physical interpretation: passive protection and active correction

Three protection mechanisms should remain separate:

Mechanism	Resource	Toric-code status
Ground-space insensitivity	Local indistinguishability and a gap	Stable under sufficiently weak local perturbations
Passive thermal storage	A growing free-energy obstruction	Absent in the standard two-dimensional Hamiltonian
Active error correction	Syndrome information and a decoder	Has a nonzero threshold under suitable noise assumptions

The surface-code threshold studied by Dennis, Kitaev, Landahl, and Preskill belongs to the third row. It assumes repeated recovery rather than passive equilibrium storage.

A counterfactual string tension would change the thermal argument. If separating two endpoints cost

then winding across the system could require an $O(L)$ barrier.

The standard toric code has deconfined point excitations, so it supplies no such tension. Producing confinement generally changes the phase or adds a new resource.

Weak perturbations inside the same gapped topological phase dress the anyons and logical strings. Bare Pauli weight then loses some literal meaning, while quasi-local dressed operators preserve the low-energy structure.

A 2026 preprint studies another route: a Lindblad equation that continuously runs a cellular-automaton decoder. Its finite-size simulations report a steady-state self-correction transition Kumar and Weimer.

The model adds an engineered correction field and update rates. It does not convert the equilibrium toric-code Hamiltonian into a passive self-correcting memory.

False claim to diagnose

Every logical operator has weight at least $L$, so every local sequence implementing one must cross an energy barrier proportional to $L$.

Weight counts how many qubits appear in the finished operator. The barrier counts the greatest number and type of violated stabilizers present at one time.

For a connected growing string,

while

The valid conclusion is narrower:

A weak local perturbation needs at least $L$ single-edge factors to generate a logical term, but a thermal trajectory can accumulate those factors sequentially at constant excitation energy.

What follows — and what does not

Statement	Status
The unperturbed square toric code has $d=L$.	Exact finite-size code property
A $Z$ string has excitations only at its endpoints.	Exact stabilizer algebra
The first logical contribution from $-h\sum Z_{\ell }$ occurs at order $L$.	Perturbative selection rule
Weak local perturbations produce an exponentially narrow ground-state band.	Theorem under locality and topological-order assumptions
The $Z$-logical energy barrier is $4J_e$.	Exact for single-edge Pauli paths in the standard model
The exact lifetime equals a universal Arrhenius formula.	False without a bath and failure definition
The passive two-dimensional toric code self-corrects at finite temperature.	False for the standard local thermal setting
Engineered dissipative decoding can change the conclusion.	Possible because it adds dynamics beyond passive storage

Large distance protects against low-weight errors and suppresses virtual logical mixing. Thermal stability asks whether local errors can accumulate without crossing a growing energetic or free-energy obstruction.

Exercise

Use

and restrict attention to one shortest noncontractible cycle.

1. Show that every connected open partial string has energy $E_0+4J_e$.
2. Explain why $P_0{V}^n{P}_0$ cannot contain a logical operator for $n<L$, although it may contain a scalar ground-energy shift.
3. Estimate the contribution from one connected $L$-step virtual process.
4. Construct a real local-error sequence with maximum excitation energy $4J_e$.
5. Identify the extra information required to convert the barrier into a quantitative lifetime.

Hint 1

Count the stars that anticommute with a partial $Z$ string. Interior vertices touch two string edges.

Hint 2

For one connected ordering, all $L-1$ virtual intermediate states have the same denominator $E_0-(E_0+4J_e)$.

Oral check 1. Why does extending an open string move an excitation rather than create another pair?

Oral check 2. Which quantity enters a simple Arrhenius exponent: completed Pauli weight or maximal intermediate energy?

Solution

An interior star overlaps with two edges of the partial string, so it commutes with their product. Only the endpoint stars anticommute.

Each violated star costs $2J_e$. Therefore

$$E(\gamma )-E_0=4J_e$$Copy LaTeX

for every nonempty connected open path $\gamma$.

For $n<L$, a product of $n$ edge operators cannot wind around a shortest torus cycle. Projection either removes a state with excitations or reduces a contractible closed product to a stabilizer.

Such terms may contribute a multiple of $P_0$, but they cannot distinguish logical sectors.

At order $L$, one connected ordering gives

$$t_{\gamma }\sim (-h{)}^L\prod _{k=1}^{L-1}\frac{1}{-4J_e}=\frac{(-h{)}^L}{(-4J_e{)}^{L-1}}.$$Copy LaTeX

This calculation shows why logical mixing starts at order $L$. It is not the complete coefficient after summing all virtual processes.

For the thermal path, create a neighboring pair, move one endpoint around a noncontractible cycle, and annihilate it with the stationary endpoint.

The maximum energy is

$${\mathrm{\Delta }}_{\mathrm{b}\mathrm{a}\mathrm{r}\mathrm{r}\mathrm{i}\mathrm{e}\mathrm{r}}^{(Z)}=4J_e.$$Copy LaTeX

A lifetime calculation still needs a bath generator or transition rates, temperature, an initial-state and decoding protocol, and a precise logical-failure criterion.

Check your understanding

Explain how the same noncontractible loop can imply both

and

Your answer should name the different processes counted by perturbative order and thermal activation.

You may also reply with “deeper,” “too easy,” “too hard,” or your derivation.

Further Reading

• Fault-tolerant quantum computation by anyons
• Topological quantum memory
• Topological quantum order: stability under local perturbations
• A no-go theorem for a two-dimensional self-correcting quantum memory based on stabilizer codes
• On thermalization in Kitaev’s 2D model
• Self-correction phase transition in the dissipative toric code

Connections and next step

• Pressure point: logical support does not determine a thermal error path’s maximum energy.
• Exact layer: string endpoints, homology, and code distance.
• Spectral layer: logical mixing begins at order $L$ under weak local perturbations.
• Thermal layer: point excitations diffuse along a constant-energy path.
• Control layer: active and autonomous decoders add resources absent from passive storage.
• Next step: compare the two-dimensional model with the four-dimensional toric code and with memories whose barriers grow logarithmically.
• Revisit: connect this error-path barrier to the free-energy barriers discussed in irreversible statistical mechanics.