Topological Quantum Computation

Topological quantum computation (TQC) assembles the previous ideas into a complete model of computation: encode qubits in anyon fusion spaces, compute by braiding, and read out by fusion. Its defining virtue is that information is processed only through topological operations, so the dominant sources of error in conventional hardware — imprecise control pulses and local decoherence — are suppressed at the physical level.

The three primitives

A topological quantum computation consists of:

Initialization. Create anyons from the vacuum in pairs. Pair-creation prepares a definite fusion state (each created pair is in the vacuum channel), fixing a known starting vector in the fusion space.
Computation (braiding). Physically move anyons around one another. Each braid applies the unitary $\rho(\sigma_i)$ determined by the model's $R$ - and $F$ -matrices. The accumulated braid word is the circuit, and the gate it implements depends only on the topology of the worldlines.
Readout (fusion). Bring pairs of anyons together and detect which anyon results. The fusion outcome is the measurement; its probabilities are the squared amplitudes of the final fusion state.

Encoding a qubit

A qubit is a two-dimensional subspace of a fusion space. With Fibonacci anyons, three $\tau$ 's with total charge $\tau$ already span a two-dimensional space (the fusion of the first two can be $1$ or $\tau$ ), giving one qubit; a register of qubits uses groups of $\tau$ 's. With Ising anyons, four $\sigma$ 's with total charge $1$ span a two-dimensional space (qubit), encoded in the fusion channel of the inner pair. The choice of anyon decides what gates braiding can produce.

Universality and its absence

The computational power of a model is exactly the image of the braid-group representation in the unitary group on the fusion space.

Ising / Majorana: braids generate the Clifford group only. By the Gottesman–Knill theorem, Clifford circuits are efficiently classically simulable, so Ising braiding is not universal. It must be supplemented (e.g. by a non-topological "magic" $T$ -gate via measurement or by injecting magic states) to become universal — at the cost of reintroducing a non-protected operation.
Fibonacci: braids are dense in the relevant unitary group, so braiding alone is universal. Any target gate $U$ can be approximated to error $\varepsilon$ by a braid of length $O\!\left(\log^{c}(1/\varepsilon)\right)$ via the Solovay–Kitaev theorem — efficient and entirely topological.

Why the errors are small

A logical error requires a process that is topologically nontrivial: a stray thermally activated anyon must wind around a computational anyon, or two computational anyons must be brought close enough to tunnel. The amplitude for such events is exponentially small in the anyon separation $L$ and in the gap, scaling like $e^{-L/\xi}$ and $e^{-\Delta/k_BT}$ , with $\xi$ the correlation length. Increasing separation or lowering temperature suppresses errors exponentially, without any active error correction.

What to take away

TQC computes by initializing anyons from the vacuum, braiding them to apply topologically determined gates, and fusing them to read out. Ising/Majorana braiding gives the Clifford group and needs supplementing for universality; Fibonacci braiding is universal on its own via Solovay–Kitaev. Errors require topologically nontrivial, exponentially suppressed processes, which is the source of the model's intrinsic fault tolerance.