Topological Quantum Computation

Every other hardware platform in this curriculum — transmons, trapped ions, neutral atoms, spin qubits, photons — fights the same battle. The qubit lives at one physical location. A stray magnetic field, a thermal phonon, a cosmic-ray ionisation, a 50 Hz pickup from a power cable — any of these couples to that location, nudges the qubit's phase, and decoheres it. The defence is brutal: measure the qubit's neighbours constantly, detect that something has shifted, and apply a correction. This is quantum error correction, and at today's ~1% per-gate error rate it needs roughly a thousand physical qubits per one logical qubit. For a useful quantum computer — the kind that breaks 2048-bit RSA — you need millions of physical qubits in total.

Now imagine a qubit that was intrinsically blind to local noise. A qubit whose information lived not at any single spot on the chip but in a global property of a whole collection of particles — a property so non-local that a disturbance anywhere in the system literally could not see it. You would not need to measure neighbours. You would not need to apply corrections. The qubit would just stay correct, by construction.

That is the promise of topological quantum computation. And the proposal — due to Alexei Kitaev in 1997–2003 — is not a vague analogy. It is a precise mathematical recipe that requires one exotic ingredient: a two-dimensional quantum system with a particular kind of quasiparticle called a non-abelian anyon. If you have such a system, you can store one qubit in a pair of anyons, apply gates by moving the anyons around each other, and the gate fidelity is bounded not by your pulse shaping — because there are no pulses — but by the topological coherence of the anyons themselves.

You have met topological qubits already in chapter 172. This chapter is a companion: there, you built the idea of a topological qubit as a stored state; here, you build the idea of a topological computation as a braid, and you face the controversy over whether Microsoft has actually built one.

Step 1 — why 2D is special

In three spatial dimensions, identical particles come in exactly two types: bosons (exchanging two of them leaves the wavefunction unchanged) and fermions (exchange multiplies by -1). No third option exists. The reason is geometric: exchanging two particles twice brings them back to their original configuration along a loop that can be continuously shrunk to a point in 3D space. Whatever phase e^{i\theta} one exchange introduces, two exchanges must give e^{2i\theta} = 1, so \theta = 0 (boson) or \theta = \pi (fermion).

In two dimensions, this shrink-to-a-point argument collapses. Picture two particles sitting in a plane. Exchange them by moving particle A in a half-loop around particle B. Do it again. The combined two-exchange path now wraps around B twice, and in the 2D plane this double-wrap cannot be contracted to a point without passing one particle through the other. The configuration space — the set of all arrangements of two particles in 2D — is not simply connected. And when the configuration space is not simply connected, the wavefunction is allowed to pick up any phase on a single exchange.

Particles with these generalised exchange statistics were named anyons by Frank Wilczek in 1982 — the name is a pun on "any phase." In 2D, "boson or fermion" becomes one special case of a much richer structure.

• Abelian anyons pick up a phase e^{i\alpha} on exchange, for some \alpha that need not be 0 or \pi. Different exchanges all commute with each other (hence abelian). These exist in the fractional quantum Hall effect at filling \nu = 1/3 — the famous Laughlin quasiparticles, whose e^{i\pi/3} exchange phase was measured in interference experiments in 2020.
• Non-abelian anyons are stronger. If n non-abelian anyons sit in the system at fixed positions, the ground state is not unique — there is a degenerate ground-state manifold, a set of d^n (for some d > 1) linearly independent states, all at exactly the same energy. Exchanging two anyons rotates the wavefunction as a unitary matrix acting on this degenerate space, not just a scalar phase. Different exchange orderings give different matrices, and in general these matrices do not commute — hence non-abelian.

For quantum computing, non-abelian is the whole game. The degenerate ground-state manifold is a Hilbert space — it has dimension 2 for a qubit, 2^n for an n-qubit register. Braiding the anyons around each other acts as unitary gates on that Hilbert space. A quantum circuit is literally a braid.

Step 2 — a braid is a gate

Picture two non-abelian anyons sitting in a 2D system at positions labelled 1 and 2. Now move them around each other over time. If you plot their positions in (x, y, t) space — two spatial dimensions plus time — the two worldlines trace out a pair of curves that braid around each other, exactly like two strands of rope twisting in a cable.

A braid on n strands is a topological object. Two braids are the same if one can be continuously deformed into the other without passing strands through each other. Wiggling a strand sideways does not change the braid; neither does speeding it up or slowing it down; neither does tiny random noise in the path. The only way to change the braid is to change which strand crosses over which at some step — a discrete, topological event.

Non-abelian anyons attach a unitary matrix to every braid. Call the matrix U(\beta) where \beta is the braid. The map \beta \mapsto U(\beta) respects composition: if you first do braid \beta_1, then \beta_2, the resulting unitary is U(\beta_2) \cdot U(\beta_1). Two braids equivalent under continuous deformation produce the same unitary. This is the content of the phrase braid-group representation.

For quantum computing, three pieces fit together:

1. The state of the system lives in the degenerate ground-state manifold — a Hilbert space of dimension 2^k for k encoded qubits.
2. Physically moving anyons around (a braid) applies a unitary U(\beta) to that state.
3. Small perturbations to the braid's path — speed, wobble, thermal jitter — do not change U(\beta), because they do not change the topology of the braid.

Compare this to a transmon gate. A transmon X gate is a 50-nanosecond microwave pulse whose amplitude and phase must be calibrated to roughly 0.1% to achieve 99.9% fidelity. The gate is an analogue quantity; drift, noise, and miscalibration each contribute proportionally to the error. For a topological gate, drift and noise do not contribute at all, as long as they do not change the topology. The gate is essentially digital — either the right braid happened or a different one did, with nothing in between.

Step 3 — Majorana zero modes and their braid group

The physical ingredient you need is a 2D system (or an effective-2D arrangement) hosting non-abelian anyons. The most-pursued candidate is the Majorana zero mode (MZM), a localised zero-energy excitation predicted to appear at the ends of a one-dimensional topological superconductor.

Here is the physical setup in a sentence: take a semiconductor nanowire (InAs or InSb) with strong spin-orbit coupling, proximity-couple it to an s-wave superconductor (a thin aluminium film), place it in a magnetic field of about 300 mT, and cool it to 20 mK. The theory says a Majorana zero mode localises at each end of the wire. A pair of such wires arranged in a T or H geometry gives you four MZMs, which in the 2+1D "effective braid" picture — where you can permute MZMs by measurement sequences or gate-voltage cycles — behave as four non-abelian anyons.

Four MZMs encode one qubit. The degenerate ground-state manifold of four MZMs has dimension 2 in the even-total-parity sector: two states differing in the occupation of a pair of non-local fermion modes. Call them |0\rangle_L and |1\rangle_L.

Now the key fact that shapes everything else: the braid group of Majorana zero modes generates only the Clifford subgroup of single-qubit gates.

What is the Clifford group

The Clifford group on one qubit is generated by the Hadamard H and the phase gate S = \text{diag}(1, i). On multiple qubits you add the CNOT. Every Clifford gate is a product of these. Pauli gates X, Y, Z are all Cliffords (they equal HSH\,S^2\,H and similar short words).

The Clifford group has a defining feature: Clifford gates map Pauli operators to Pauli operators under conjugation. If P is a Pauli string and C is a Clifford, then CPC^\dagger is another Pauli string (possibly with a sign). This stability is what makes Cliffords easy — and also what makes them not universal.

Gottesman–Knill theorem (1998). Quantum circuits composed entirely of Clifford gates, starting from a stabilizer state and ending with a computational-basis measurement, can be simulated on a classical computer in polynomial time. So Clifford-only quantum computing is no more powerful than classical computing. You absolutely need at least one non-Clifford gate to get any quantum speedup.

The standard non-Clifford gate is the T gate, T = \text{diag}(1, e^{i\pi/4}) — a "45-degree" phase rotation, exactly half the angle of the Clifford S. The set \{H, S, T, \text{CNOT}\} is universal (you can approximate any unitary on any number of qubits to any precision using just these).

What MZM braids do and don't give you

Label the four MZMs \gamma_1, \gamma_2, \gamma_3, \gamma_4. They are Hermitian operators with \gamma_k^\dagger = \gamma_k and \gamma_k^2 = \mathbb{1}. The bilinears Z_L = i\gamma_1\gamma_2 and X_L = i\gamma_2\gamma_3 act as the Pauli Z and Pauli X on the logical qubit (with Y_L = iX_L Z_L = i\gamma_1\gamma_3).

Exchanging \gamma_1 with \gamma_2 (a half-braid) applies the unitary

Why: a half-exchange of two identical particles in 2D is a \pi rotation in their relative-coordinate plane. For Majorana fermions this rotation acts on the operators as a Bogoliubov transformation, giving exactly the unitary above. The -\pi/4 rotation angle on Z_L is what identifies this as an R_z(\pi/2) gate on the logical Bloch sphere — a Clifford.

Similarly B_{23} = e^{-i\pi X_L/4} (an R_x(\pi/2)), and B_{34} = e^{-i\pi Z_L/4} again. Composing these three generators gives the full Clifford group on one logical qubit. With eight MZMs and similar braids between wires, you get two-qubit Cliffords including CNOT.

But no braid produces a T gate. Every Majorana braid is a \pi/4 rotation about some Pauli axis — angles that are exact multiples of \pi/2 on the Bloch sphere. The T gate requires a \pi/8 rotation. You literally cannot get it from braiding alone. You can approximate it by a long sequence of braids combined with measurement — but not topologically. The non-Clifford piece must come from outside, via magic-state injection (see chapter 127, Fault-Tolerant Gates and Magic States).

So a Majorana-based topological quantum computer is almost self-sufficient. The Cliffords are topologically protected — no calibration, no drift, exponentially suppressed error rates. The T gate is not. It requires preparing an ancilla in the magic state |T\rangle = \frac{1}{\sqrt{2}}(|0\rangle + e^{i\pi/4}|1\rangle) by some non-topological method (typically a noisy physical rotation followed by distillation), then consuming it to enact a T on the data. The topological platform is 85–90% of a fault-tolerant quantum computer; the last 10–15% is conventional.

Step 4 — two worked examples

Step 5 — Microsoft Majorana 1 and the controversy

You now have the theoretical picture. What about the hardware?

Microsoft has pursued Majorana-based topological quantum computing for more than two decades. The effort — centred on Station Q labs in Santa Barbara and Copenhagen, with collaborations at Delft, Copenhagen, UCSB, and a handful of other universities — has produced a sequence of claimed milestones, each followed by community reassessment:

• 2012 (Mourik et al., Delft). First zero-bias conductance peak in an InSb–Al nanowire — a predicted signature of Majorana zero modes, but also reproducible by non-topological Andreev bound states.
• 2018 (Zhang et al., Delft). A Nature paper claiming a quantised zero-bias peak at the value 2e^2/h — the stronger predicted signature. A flagship Microsoft–Delft result.
• 2021 retraction. After independent analysis revealed data-selection problems — the authors had excluded data inconsistent with their hypothesis — the 2018 paper was retracted. The investigation concluded that the quantised peak claim was not supported by the full dataset. The retraction was a serious blow to community confidence.
• 2022–2024 (Microsoft, Aghaee et al.). A new measurement suite called the topological gap protocol was introduced, designed to test both wire ends simultaneously and verify the expected gap-closing-and-reopening signature of the topological phase transition. Devices passing this protocol were argued to host bona fide Majorana zero modes.
• February 2025. Microsoft announced Majorana 1, described as "the world's first topological quantum processor" with eight qubits. The processor uses measurement-based braiding: instead of physically moving the MZMs, a sequence of fermion-parity measurements on selected MZM pairs has the same net unitary effect as an equivalent braid, up to classical byproduct corrections tracked in software.

The community reaction has been mixed. The case for:

• The topological gap protocol is a rigorous set of measurements; devices passing it are doing something consistent with the MZM hypothesis.
• The fabrication process — epitaxial InAs–Al heterostructures, chemical-beam-epitaxy tricks to produce clean interfaces — has genuinely advanced, and the Microsoft team's engineering is taken seriously.
• Measurement-based braiding, even though it hasn't been used to demonstrate a non-abelian statistics experiment, is a sensible architecture if MZMs exist.

The case against:

• Alternative explanations — disorder-induced "trivial" Andreev bound states that locally mimic MZM signatures — have not been definitively ruled out. The condensed-matter community has a well-documented history of false positives in this area.
• Braiding has not been demonstrated as a non-abelian statistics experiment. No one has yet measured the non-abelian unitary that two Majoranas are predicted to produce under exchange. This is the decisive test, and it has not been done.
• Majorana 1 is a prototype with 8 qubits and no published benchmarks (gate fidelities, coherence times, error rates). Without these, "the world's first topological qubit" is a marketing claim, not a physics result.

The honest 2026 assessment is that Majorana-based TQC is very plausibly on the right track but has not yet produced an unambiguous experimental demonstration. The theoretical framework is beyond dispute; the engineering is impressive; but the existence of topologically protected non-abelian statistics in any laboratory system remains, as of this writing, to be demonstrated.

Common confusions

• "Braiding is the same as physically moving the anyons." In the original theoretical proposals, yes — you move the anyon positions by adjusting gate voltages. But in Majorana 1 and most modern proposals, braids are implemented by measurement-based protocols: sequences of parity measurements on MZM pairs that produce the same net unitary up to classical corrections. The topological protection is preserved; the moving parts are not.
• "Topological qubits already work." The 2025 Majorana 1 announcement is an engineering milestone, but (a) the underlying MZM existence is contested, (b) braiding — the test that would confirm non-abelian statistics — has not been demonstrated, and (c) the device is an 8-qubit prototype with no published benchmarks.
• "Topological protection is perfect." It is exponential in the ratio L/\xi (wire length to coherence length) and in \Delta/k_B T (gap to temperature). For typical parameters the suppression factor is 10^3 to 10^9 — enormous but finite. Quasiparticle poisoning by thermal excitations above the gap and end-to-end tunnelling on finite wires remain real error sources.
• "Majorana braiding gives you a universal quantum computer." It does not. It gives only the Clifford subgroup; the T gate must come from non-topological magic-state injection.
• "Anyons are new particles physicists discovered." No — anyons are emergent excitations of specific quantum many-body systems in two dimensions. The "particle" is a collective mode of many underlying electrons, not a fundamental particle. You cannot buy anyons in a bottle.
• "The 2021 retraction means Majorana zero modes don't exist." The retraction concerned data-handling practices in one experiment, not the theoretical existence of MZMs. The theory is untouched; the question is whether any specific experiment has seen them cleanly.

Going deeper

The braid group and its Ising representation

For n anyons, the braid group B_n is generated by n-1 elementary generators \sigma_1, \ldots, \sigma_{n-1}, where \sigma_i is the half-exchange of strands i and i+1. The generators satisfy:

1. Far commutativity: \sigma_i \sigma_j = \sigma_j \sigma_i when |i - j| \geq 2 (crossings that don't share a strand commute).
2. Yang–Baxter relation: \sigma_i \sigma_{i+1} \sigma_i = \sigma_{i+1} \sigma_i \sigma_{i+1} (the "type-III Reidemeister move" — a topological identity for strands that cross adjacently).

A non-abelian anyon model is specified by a unitary representation of B_n on the degenerate ground-state Hilbert space. For Majorana zero modes, the representation is the Ising anyon representation: each \sigma_i maps to e^{i\pi/4} e^{-i\pi\gamma_i\gamma_{i+1}/4} = (1 + \gamma_i\gamma_{i+1})/\sqrt 2. This group is finite — concretely, the image of B_n in the Ising representation is a finite extension of the symmetric group, with only O(2^n) elements. A finite group of unitaries cannot be dense in SU(2^k); hence Ising braids are not universal, and magic states are required.

The Kitaev chain — a toy model of MZMs

Alexei Kitaev introduced the simplest model hosting Majorana zero modes in 2001: a 1D chain of spinless fermions with p-wave pairing. The Hamiltonian is

with chemical potential \mu, hopping t, and p-wave pairing \Delta. In the topological phase |\mu| < 2t and \Delta \neq 0, rewriting each c_j in terms of two Majorana operators \gamma_{2j-1} = c_j + c_j^\dagger and \gamma_{2j} = -i(c_j - c_j^\dagger) reveals that the Hamiltonian pairs every Majorana with its neighbour except \gamma_1 (left end) and \gamma_{2N} (right end). These two are unpaired, at zero energy, and spatially separated by the full wire — the textbook MZMs. The chapter on topological qubits works this out in detail.

Why Fibonacci would be different

A different non-abelian anyon model — Fibonacci anyons, built around a single non-trivial particle type \tau with fusion rule \tau \times \tau = 1 + \tau — has a braid-group representation whose image is a dense subgroup of SU(2). Dense means every SU(2) element can be approximated arbitrarily well by a sufficiently long product of braid generators. This is the Solovay–Kitaev theorem applied to the Fibonacci braid: braiding alone is universal. No magic state needed. No conventional error correction needed beyond the topological protection itself.

The catch is physical realisation: no system has yet been confirmed to host Fibonacci anyons. The leading candidate is the \nu = 12/5 fractional quantum Hall state (see chapter 186). If Fibonacci anyons were built, topological quantum computing would leap from "almost self-sufficient" to "fully self-sufficient."

Measurement-based braiding

In Majorana 1 and its predecessors, braids are implemented by sequences of parity measurements rather than by physical motion. The theorem (first stated in this form by Bonderson, Freedman, and Nayak, 2008) is: for any braid \beta on n MZMs, there exists a sequence of fermion-parity measurements on specific pairs of MZMs such that the measurement outcomes determine a random product of Pauli byproducts, but the combined effect of the measurements and the classical corrections is exactly the braid unitary U(\beta).

This is the topological analogue of measurement-based quantum computing (MBQC) on regular qubits, and it has the huge practical advantage of not requiring any moving parts: all of the "braiding" happens through electronic measurements of fermion parity on a fixed device geometry.

Comparison to surface codes

If Majorana-based TQC works, how does it compare to the surface-code approach that IBM, Google, and AWS pursue?

Realistic roadmaps that take both seriously assume a hybrid: Majorana qubits as high-quality physical qubits, surface-code error correction as outer protection. The topological piece would cut the surface-code overhead by a factor of 10 rather than eliminating it entirely.

Where this leads next

• Topological Qubits — how the information is stored. Read this for the static picture before you build circuits.
• Fibonacci Anyons and Universal TQC — the anyon model that would make TQC universal without magic states. The holy grail that no one has yet built.
• Fault-Tolerant Gates and Magic States — where the non-Clifford gates come from. The missing piece for Majorana-based computation.
• Toric / Surface Codes — the leading alternative path. Topological codes in software rather than hardware.

References

1. Alexei Kitaev, Fault-tolerant quantum computation by anyons (2003), Annals of Physics 303 — arXiv:quant-ph/9707021.
2. Chetan Nayak, Steven Simon, Ady Stern, Michael Freedman, Sankar Das Sarma, Non-abelian anyons and topological quantum computation (2008), Rev. Mod. Phys. 80, 1083 — arXiv:0707.1889.
3. Jason Alicea, New directions in the pursuit of Majorana fermions in solid state systems (2012), Rep. Prog. Phys. 75, 076501 — arXiv:1202.1293.
4. Microsoft Azure Quantum (Aghaee et al.), InAs-Al hybrid devices passing the topological gap protocol (2023) — arXiv:2207.02472.
5. Parsa Bonderson, Michael Freedman, Chetan Nayak, Measurement-only topological quantum computation (2008) — arXiv:0802.0279.
6. Wikipedia, Topological quantum computer.