Topological Qubits

A topological qubit stores quantum information in a nonlocal sector of a many-body system or quantum code. Representative logical variables are the fusion channel of separated anyons, the joint fermion parity of several Majorana zero modes, or a global string degree of freedom in an actively stabilized topological code. Ideally, no sufficiently local operator can distinguish or rotate the logical states.

That definition describes an encoding principle, not a finished device. A working topological-qubit module still needs a gapped or repeatedly stabilized code space, initialization, controllable logical operations, basis-sensitive readout, cooling or reset, leakage handling, classical control, and a path to a universal gate set. Some of those operations are topologically protected; others are ordinary fault paths.

The discussion below connects nonlocal encoding to the storage, control, and measurement of a physical qubit. Topological Quantum Computation derives fusion-space gates; Surface Code explains repeated-check protection. A zero-energy spectral feature, a parity measurement, coherent logical control, and error suppression with increasing size are distinct observations. Dated experimental comparisons and open questions are discussed in Quantum Hardware Scaling and Non-Abelian Anyons.

Required background. Bits, Qubits, Qudits, and Modes separates a carrier from its encoding. Anyons and Braiding introduces fusion channels, and Topological Superconductors explains Majorana boundary modes and the bulk gap.

Local indistinguishability and protection

Let $\mathcal C$ be a logical subspace with projector $P_{\mathcal C}$. A useful idealization of local indistinguishability, discussed by Kitaev and by Nayak and colleagues, is

$$\left\|P_{\mathcal C}O_RP_{\mathcal C} -c(O_R)P_{\mathcal C}\right\| \lesssim C\|O_R\|e^{-(L-R)/\xi}.$$

Here $O_R$ is a bounded allowed local observable supported in a region of diameter $R\ll L$, $L$ is a separation or code scale, and $\xi$ is a correlation or localization length; the constant $C$ and the exponential form depend on the phase and geometry. For a Majorana device, physical isolated-system observables preserve total fermion parity. Exact commuting-projector codes can instead have an exactly vanishing left side on correctable regions. This is not a universal finite-size formula for every use of the word topological. To leading order, a local probe sees the same expectation value in every logical state. The same locality structure can suppress a residual splitting,

$$\delta E(L) \sim E_0 e^{-L/\xi},$$

up to oscillatory and device-dependent factors.

For gapped matter this describes passive suppression of selected local matrix elements. Active code protection additionally depends on the syndrome circuit and decoder. It does not protect against every mechanism:

• a quasiparticle can enter a superconducting island and change fermion parity;
• thermally created anyons can diffuse around a nontrivial cycle;
• an operation can cross the gap too quickly and leak;
• an operation can run too slowly and accumulate a dynamical phase;
• a detector can misclassify parity or disturb it;
• a local defect can close the gap or nucleate an unwanted low-energy state;
• a nonlocal environmental process can act directly as a logical operator;
• an unprotected non-Clifford resource can dominate the logical error budget.

Topological protection suppresses selected errors under these assumptions; it need not produce a particular Pauli bias and is not absolute immunity. Long computations still require active error detection, repeated measurements, decoding, and fault-tolerant composition.

Topological matter and topological codes

The word topological is used for related but operationally different resources.

Resource	Where the nonlocal sector comes from	How it is maintained
intrinsic or engineered topological matter	a gapped many-body or Bogoliubov phase and its defects	Hamiltonian gap, cooling, separation, parity control
synthesized topological state	a conventional processor prepares a code or string-net wavefunction	calibrated gates, measurements, verification
actively stabilized topological code	repeated local checks define a logical sector	syndrome cycles, decoder, feedback or Pauli frame

A surface-code qubit is genuinely topological at the code level, even though its transmons, ions, or atoms are not topological quasiparticles. Conversely, a candidate topological superconductor can possess a useful bulk invariant without yet providing a controllable qubit. Naming the layer prevents a material claim from being silently promoted into a processor claim.

Majorana Encodings

Algebra and parity constraint

Majorana operators satisfy

$$\gamma_j^\dagger=\gamma_j, \qquad \{\gamma_j,\gamma_k\}=2\delta_{jk}.$$

Two Majoranas define one ordinary fermionic mode,

$$\begin{aligned} f_j &= \frac{\gamma_{2j-1}+i\gamma_{2j}}{2}, \\ i\gamma_{2j-1}\gamma_{2j} &= 2f_j^\dagger f_j-1. \end{aligned}$$

For $2m$ independent zero modes, with other excitations excluded, there are $2^m$ occupation states before constraints. A fixed total fermion parity removes half of them. The usable Hilbert-space dimension is therefore

$$\dim \mathcal H_{\rm fixed\ parity} = 2^{m-1}.$$

Four Majoranas are the minimum for one parity-conserving qubit. A convenient choice of logical Pauli operators is

$$\begin{aligned} \bar Z &= i\gamma_1\gamma_2, \\ \bar X &= i\gamma_2\gamma_3, \\ \bar Y &= i\gamma_1\gamma_3, \end{aligned}$$

within a sector of fixed

$$P_{\rm tot} = (i\gamma_1\gamma_2) (i\gamma_3\gamma_4).$$

These bilinears square to the identity and satisfy $\bar Z\bar X=i\bar Y$. With the definition of $f_j$ above, physical fermion parity is $\prod_j(1-2f_j^\dagger f_j)=\prod_j(-i\gamma_{2j-1}\gamma_{2j})$. For four modes this equals the displayed $P_{\rm tot}$; for $2m$ modes the product of the $+i$ bilinears differs by $(-1)^m$.

To fix logical phases explicitly, use a Jordan–Wigner representation with $f_1=|0\rangle\langle1|\otimes I$ and $f_2=Z\otimes|0\rangle\langle1|$. In the even sector a compatible basis is

$$|\bar0\rangle=|11\rangle,\qquad |\bar1\rangle=-|00\rangle.$$

The displayed bilinears then become the usual $Z,X,Y$ matrices in that order. A single pair has two formal occupations, but they differ in total parity. A usable superposition must be encoded in a larger system with fixed total parity; a reservoir is not by itself permission to ignore superselection.

Tetrons, hexons, and islands

A tetron places four Majorana modes on a superconducting island, commonly using two proximitized wire segments joined by a superconducting backbone. Charging energy

$$H_C = E_C(\hat N-n_g)^2$$

uses excess electron number $\hat N$, with $n_g$ in units of $e$ and $E_C=e^2/(2C)$ in this convention. In a Coulomb valley it penalizes changes of island charge and suppresses external parity-changing tunneling. It does not prove that the end modes are topological, nor does it eliminate poisoning from nonequilibrium quasiparticles.

A hexon uses six Majoranas. At fixed total parity it has a four-dimensional sector, which can be organized as one data qubit plus a parity ancilla or as a denser two-qubit encoding. The extra pair can simplify measurement-only Clifford operations, but it increases device area, tuning requirements, and readout connectivity (Karzig and colleagues; Plugge and colleagues).

Sparse and dense encodings trade hardware against control complexity. Dense encodings use the parity-constrained Hilbert space efficiently but make single-qubit addressing and identification of leaked states more involved. Sparse encodings reserve ancillary degrees of freedom and can simplify fault containment. A platform comparison must state which encoding and which fixed parity constraints it assumes.

What the material must supply

In an ideal one-dimensional topological superconductor, a Majorana mode is localized near each end and the overlap splitting behaves schematically as

$$\delta E(L) \sim A e^{-L/\xi} \cos(k_{\rm F}L+\phi).$$

Longer wires help only when the induced gap stays open, disorder does not introduce additional subgap states, and the coherence length $\xi$ remains short relative to $L$. Smooth confinement and partially separated Andreev states can imitate several local Majorana signatures. The device therefore needs convergent evidence from the bulk gap, nonlocal response, parity structure, perturbation tests, and ultimately fusion- or braid-sensitive operations.

The Kitaev chain gives a minimal derivation of this end-mode mechanism (Kitaev, 2001).

Quantum-dot chains tuned near a Kitaev sweet spot offer a second route. Their couplings are highly controllable, and two- or three-site chains can exhibit Majorana-like end modes. These short modes are often called poor man’s Majoranas. Exact zero modes at a finely tuned short-chain point do not establish stability throughout a gapped phase. A fixed two- or three-site device cannot by itself determine asymptotic protection as length grows.

Other Anyon Encodings

Fractional quantum Hall anyons

Fractional quantum Hall fluids provide intrinsic two-dimensional topological order. For four Ising anyons $\sigma$ with fixed total vacuum charge, the two allowed intermediate fusion channels,

$$\sigma\times\sigma = 1+\psi,$$

can encode a qubit. Braiding acts on the fusion space, and fusion or interferometric measurement reads a chosen basis. The physical module needs more than a high-quality Hall plateau:

• localized creation and positioning of individual quasiparticles;
• stable gates and interferometer area;
• control of edge reconstruction and neutral modes;
• discrimination between electromagnetic and statistical phase;
• basis-sensitive fusion readout;
• repeatable noncommuting braid words;
• a layout for many anyons without uncontrolled enclosed charge.

For fixed Abelian charges and total charge, the fusion space has dimension one: braiding supplies a scalar phase, which can be observed by comparing paths in an interferometer (Nakamura and colleagues; Bartolomei and colleagues). That achievement does not by itself provide a non-Abelian fusion-space qubit. For a proposed non-Abelian Hall implementation, the specific topological order, localized quasiparticles, and accessible fusion measurements must all be established; an even denominator alone does not determine that order.

The Fibonacci theory permits approximate universal gates by braiding (Freedman, Larsen, and Wang). Realizing that theory and its controls in a material is a separate physical task; it does not follow from the abstract braid representation.

Synthesized anyons and active topological qubits

Superconducting, trapped-ion, neutral-atom, and other processors can prepare toric-code, quantum-double, or string-net states; create encoded excitations; apply logical strings; and measure fusion or braid outcomes. These are genuine many-body quantum simulations and, when repeated checks are used, genuine topological quantum codes.

The protection mechanism is nevertheless the one implemented by the underlying hardware and circuit. A finite-depth programmed braid does not automatically inherit the passive gap protection of an intrinsic anyon worldline. The relevant questions are whether the encoded state is stabilized during the operation, whether faults remain local in the code, and whether logical error falls with code distance.

Initialization and Reset

Topological encodings are constrained sectors, so initialization means more than cooling a local two-level system.

For a Majorana island, one possible sequence is (Aasen and colleagues):

1. tune the device into the intended gapped regime;
2. isolate or Coulomb-blockade the island so total parity is defined;
3. measure a commuting set of Majorana bilinears appropriate to preparation;
4. apply outcome-dependent operations or track the corresponding logical frame;
5. bound above-gap and unintended subgap occupations through accessible checks;
6. repeat until preparation error and leakage satisfy the logical budget.

Cooling alone need not select a logical basis state when the code space is degenerate. Temporarily coupling Majoranas can lift that degeneracy and select a fusion channel, but the coupling is then an unprotected local Hamiltonian. Its turn-on, turn-off, and residual phase must be included in the preparation error.

For material anyons, initialization can create quasiparticle pairs from the vacuum, move them to separated traps, and verify total topological charge. For an active code, initialization prepares physical carriers and measures stabilizers until a consistent syndrome history is obtained. In each case, the conditional state depends on the observed outcomes; an unobserved mixture and a known state with a tracked frame are different resources.

Reset has a similarly broad boundary. It may require evacuating quasiparticles, rethermalizing an island, emptying quantum dots, reinitializing detectors, or rerunning stabilizer rounds. A fast parity measurement paired with a slow or unreliable reset does not produce a fast logical cycle.

Gates and Control

Physical braiding

Take the following sign as the definition of a positive exchange in this Majorana convention. Dropping a common Abelian phase, the state unitary is

$$U_{jk} = \exp \left( \frac{\pi}{4}\gamma_j\gamma_k \right).$$

Its action on the operators is

$$\begin{aligned} U_{jk}\gamma_j U_{jk}^\dagger &= -\gamma_k, \\ U_{jk}\gamma_k U_{jk}^\dagger &= \gamma_j. \end{aligned}$$

This is the stated $U\gamma U^\dagger$ conjugation convention; $U^\dagger\gamma U$ gives the inverse transformation. Reversing the exchange orientation replaces $U$ by $U^\dagger$. Different exchanges need not commute on the degenerate space.

For the tetron basis above, $U_{12}=e^{-i\pi\bar Z/4}$ and $U_{23}=e^{-i\pi\bar X/4}$. Starting from $|\bar0\rangle$, compare the two orders, with the rightmost operator applied first:

$$\begin{aligned} |\psi_A\rangle&=U_{12}U_{23}|\bar0\rangle,\\ |\psi_B\rangle&=U_{23}U_{12}|\bar0\rangle. \end{aligned}$$

Both give equal probabilities in a $\bar Z$ measurement. But $\langle\bar X\rangle_A=1$ and $\langle\bar X\rangle_B=0$. A complementary measurement distinguishes the order; comparing only one output population could miss it entirely.

A physical braid requires controllable worldlines in a two-dimensional network, vortices, or branched wire geometry (Alicea and colleagues). The operation must be slow relative to the inverse excitation gap but fast relative to residual splitting, poisoning, and dephasing. Writing $\tau_{\rm op}$ for its duration, a useful scale estimate, rather than an error guarantee, is

$$\frac{\hbar}{\Delta} \ll \tau_{\rm op} \ll \min \left( \frac{\hbar}{|\delta E|}, \tau_{\rm poison}, \tau_\phi \right).$$

The gap here is the smallest relevant excitation gap along the path. A local adiabatic diagnostic is

$$\frac{\hbar|\langle e(t)|\dot H(t)|g(t)\rangle|} {[E_e(t)-E_g(t)]^2}\ll1.$$

It depends on the ramp amplitude and shape, and does not by itself give a uniform final-error bound. Boundary smoothness and bath coupling matter (Knapp and colleagues). The upper splitting scale assumes an uncompensated relative dynamical phase; calibrated compensation can change that constraint. For a tight target error, the actual window can be much narrower than these inverse-energy scales, as Exercise 9 illustrates.

Measurement-only braiding

Majorana modes need not be physically moved. Sequences of projective joint parity measurements, together with an ancillary pair and feed-forward, can implement the corresponding logical Clifford transformation (Bonderson, Freedman, and Nayak). Measuring arbitrary noncommuting bilinears on a lone data tetron would instead collapse its state; the initialized ancillary space and outcome-dependent protocol are essential. Tunable quantum dots can couple selected Majorana pairs around an interferometric loop; the parity-dependent energy or quantum capacitance is then read through a charge sensor or microwave resonator.

This replaces geometric motion with a measurement graph. It also moves much of the engineering burden into:

• switchable tunnel couplings with low residual interaction;
• compatible loops for noncommuting Pauli measurements;
• single-shot detector speed and assignment fidelity;
• quantum-nondemolition repeatability;
• ancilla initialization and reset;
• low-latency outcome processing and Pauli-frame tracking;
• suppression of correlated errors when one dot or resonator touches several qubits.

Calling the resulting gate measurement-only does not mean controller-free. The measurement schedule and conditional corrections are the gate.

Universal computation

In standard Ising/Majorana qubit encodings, braids act within the Clifford group. The accessible subgroup depends on the encoding; braiding alone does not automatically supply every multi-qubit Clifford operation. Suitable ancillas and joint-parity measurements can complete a Clifford scheme, but these operations still do not form a universal gate set. A universal architecture therefore needs an additional non-Clifford resource, such as:

• a calibrated non-Clifford phase from controlled Majorana hybridization;
• preparation and injection of noisy magic states followed by distillation;
• a nontopological ancillary qubit coupled to the topological sector;
• a different anyon theory, such as an ideal Fibonacci platform.

The extra resource may be much noisier than the protected Clifford layer. Bravyi’s Ising-anyon construction makes this additional resource explicit. Resource estimates must include its factories, verification, routing, distillation, and failure probability. “Topological gates” is incomplete unless it says which gates and how universality is completed.

Measurement and Readout

Majorana parity readout

A detector should couple to a chosen Majorana product without learning an unwanted local occupation. For a two-Majorana parity $P_{jk}=i\gamma_j\gamma_k$, an ideal binary measurement has effects $\Pi_\pm$ given by

$$\Pi_\pm = \frac{\mathbb 1\pm P_{jk}}{2}.$$

These effects specify outcome probabilities, not the post-measurement state. The ideal Lüders instrument is $\mathcal I_\pm(\rho)=\Pi_\pm\rho\Pi_\pm$. Adding an outcome-dependent unitary after projection can preserve the same probabilities while changing the backaction. A noisy instrument can have several Kraus operators for each recorded outcome.

Real devices have finite contrast, relaxation during integration, leakage, drift, and backaction. At minimum, report:

• integration time and total cycle time;
• assignment errors conditioned on each parity;
• repeatability under immediate repeated measurements;
• parity-switching rate with the detector on and off;
• false correlations caused by charge-state or resonator drift;
• the operating region and tuning procedure;
• whether the signal distinguishes topology from a trivial low-energy state.

Interferometric quantum-capacitance readout can be an excellent parity sensor while remaining agnostic about whether the underlying end modes are topological Majoranas or finely tuned Andreev states. The InAs–Al parity-readout paper explicitly makes this distinction (Microsoft Azure Quantum and Aghaee and colleagues); the detector alone does not settle the bulk phase.

Anyon fusion and interferometry

In a Hall interferometer, the measured phase can contain an Aharonov–Bohm contribution, Coulomb effects, edge dynamics, and an anyonic statistical phase. Changing magnetic field, area, bulk quasiparticle number, temperature, bias, and path orientation supplies essential controls. A fusion-space qubit needs more than a periodic oscillation: the measurement must resolve fusion-channel-dependent outcomes and distinguish noncommuting operation orders.

QND is a channel property

For a desired parity $P$, nondemolition means that eigenvalue sectors are preserved by the measurement dynamics, apart from specified physical faults. The measurement is allowed to destroy coherence between different eigenvalues; that is the intended projective backaction. A repeated output alone can also arise from a stuck classifier or uncontrolled re-preparation. Use prepared eigenstates and complementary preparations to determine both assignment and the conditional quantum operation (Steiner and von Oppen).

Keep the classical outcome when composing the readout with feedback. If $\mathcal I_s$ is the measurement map, $\mathcal F_s$ the physical conditional correction, and $\mathcal U$ the preceding control, an unconditioned cycle has the form

$$\mathcal E_{\rm cycle} =\mathcal R_{\rm reset}\circ\sum_s\mathcal F_s\circ\mathcal I_s\circ\mathcal U.$$

A Pauli-frame update is classical tracking and may change later controls or the interpretation of results; it is not automatically a physical unitary applied at that instant. Include rejected outcomes as flags when modeling the full trace-preserving process. Here reset acts on auxiliary systems; unintended disturbance of the stored state must also be included.

Noise inside and outside the parity sector

Residual hybridization

Finite overlap produces a Hamiltonian term

$$H_{\rm split} = \frac{i}{2} \sum_{j<k} \varepsilon_{jk}\gamma_j\gamma_k.$$

The $\varepsilon_{jk}$ are real energies in this $j<k$ convention. A single pair term has level splitting $|\varepsilon_{jk}|$, because $i\gamma_j\gamma_k$ has eigenvalues $\pm1$. Within the code space, some coefficients act as unwanted logical fields. Slow drift dephases superpositions; a stable coefficient accumulates a coherent phase. Echoes or calibration may suppress one component, but they turn passive storage into an actively controlled experiment.

Quasiparticle poisoning

Poisoning transfers a quasiparticle into or out of the computational zero-mode sector. An electron crossing the island boundary changes island parity. An internal transfer between a zero mode and an above-gap mode can instead change computational parity while preserving total island parity. Charging energy suppresses the former; it does not by itself prevent the latter. The gap and nonequilibrium quasiparticle population matter as well.

Stability of either parity is not a logical coherence time. A superposition can dephase inside a fixed sector. Conversely, a measured bilinear such as $\bar Z=i\gamma_1\gamma_2$ can change under a parity-preserving logical $\bar X$ operation. A telegraph trace of that bilinear alone therefore does not identify every switch as quasiparticle poisoning.

Charging energy, gap engineering, quasiparticle traps, infrared shielding, filtering, and reduced nonequilibrium quasiparticle density can reduce poisoning. Their benefits must be measured during the actual control and readout sequence, because tunnel couplers and detectors can open new poisoning paths.

Thermal anyons and unintended charge

For an excitation energy $\Delta_a$, equilibrium density is suppressed schematically as

$$n_a \propto e^{-\Delta_a/(k_{\rm B}T)}.$$

Here $\Delta_a$ denotes a single-excitation energy in a dilute equilibrium description; the activation cost to create a constrained pair can differ. A created pair can separate and form a logical string before annihilating. For the two-dimensional toric-code Hamiltonian, extending that string does not raise its energy beyond the endpoint cost. More generally, the Bravyi–Terhal result bounds the energy barrier for local commuting Pauli stabilizer Hamiltonians in two dimensions. This precise restriction should not be replaced by a theorem about every possible two-dimensional phase. These codes require active correction for reliable long storage.

Diabaticity and leakage

Changing a gate voltage, moving a domain wall, or turning on a parity loop can couple the code space to above-gap or unintended subgap states. Leakage depends on the full spectral path, not only the initial gap. A slow ramp can reduce diabatic excitation while increasing exposure to splitting and poisoning, producing the operating-window tradeoff above.

Measurement and control faults

Detector noise, resonator photons, charge noise, flux noise, pulse miscalibration, crosstalk, and imperfect turn-off produce ordinary hardware errors. Topology does not suppress a controller that measures the wrong Majorana product. Shared sensors and couplers can generate correlated faults that are especially damaging to an outer code.

Fabrication and tune-up failures

A scalable array must reproducibly realize a gapped operating region, suitable end modes, working quantum dots, usable charging energies, and compatible resonators. Yield and tune-up complexity can dominate long before intrinsic logical error does. A single hand-tuned device does not establish an array architecture.

Connectivity and Scaling Architectures

The parity-measurement graph

In measurement-based Majorana designs, connectivity means which Majorana products can be measured without unacceptable crosstalk, not merely which islands are adjacent. A logical operation can require:

• an intra-island bilinear measurement;
• a joint parity involving modes on neighboring islands;
• an ancillary pair or qubit;
• repeated outcomes for fault detection;
• classical feed-forward and Pauli-frame updates.

The graph must support those operations in parallel. A tiling with many geometric neighbors can still have low usable connectivity if loops share dots, resonators, magnetic flux, or control lines.

From tetrons to an outer code

Passive Majorana protection is not expected to remove all faults. Proposed architectures therefore place tetrons or hexons beneath a qubit or Majorana-fermion code. Repeated joint-parity measurements can implement surface-code-like checks, lattice surgery, or fermionic-code operations. Examples are developed by Dennis and colleagues and by Litinski and von Oppen.

This creates two protection scales:

$$p_{\rm phys} \longrightarrow p_{\rm tetron} \longrightarrow p_{\rm L}(d).$$

The first map reflects the material, island, and parity-measurement module. The second reflects active code distance $d$, circuit faults, and decoding. A lower tetron error can reduce outer-code overhead, but only a measured $p_{\rm L}(d)$ trend demonstrates the combined architecture.

Sharing quantum dots, resonators, and controllers can reduce component count while restricting simultaneous measurements and introducing correlated faults. A useful comparison includes these controls, reset, and the non-Clifford operations alongside the number of data islands.

Metrics That Survive Comparison

Gap and temperature margin

Report the relevant induced or many-body gap $\Delta$, the electron or anyon temperature, and evidence that the gap remains open across the operating path. The ratio $\Delta/(k_{\rm B}T)$ is informative but does not capture nonequilibrium quasiparticles or accidental subgap states.

Splitting versus separation

A single small splitting is an ingredient. Protection requires a controlled trend with length or separation, including uncertainty and possible oscillations:

$$\delta E(L) \propto e^{-L/\xi} F_{\rm dev}(L).$$

Comparing different devices can confuse length scaling with fabrication variation; $F_{\rm dev}(L)$ includes oscillatory and device-specific factors. The strongest test changes the protection scale while holding the rest of the module as fixed as possible.

Parity lifetime and coherent lifetime

Measure poisoning, relaxation, Ramsey-like dephasing, and logical process decay separately. A parity telegraph trace gives the dwell time of the specified observable; interpreting it as a poisoning time requires a physical model. For a symmetric two-state Markov process with transition rate $r$ out of each state, the mean dwell time is $1/r$ but the parity correlation decays as $e^{-2rt}$. Neither quantity determines coherence in a complementary logical basis.

Measurement performance

For equal prior probabilities and conditional assignment errors $e_+$ and $e_-$,

$$e_{\rm assign} = \frac{e_++e_-}{2}.$$

Also report state preparation, measurement-induced transition, leakage, repeatability, integration, reset, and total round times. Postselection must include acceptance probability and the fate of rejected runs.

Logical operation and protection scaling

Characterize a logical channel over a state ensemble rather than quoting one basis-state success probability. Suitable quantities include process fidelity, leakage, worst-case bounds, and repeated-cycle decay. For a protected architecture, the decisive evidence is

$$\begin{gathered} p_{\rm L}(L_2)<p_{\rm L}(L_1)\\ \text{or}\\ p_{\rm L}(d+2)<p_{\rm L}(d), \end{gathered}$$

under matched operations and physical conditions, with statistical uncertainty small enough to resolve the change. The independent variable must actually increase the proposed protection. Several sizes and noise checks are needed to distinguish a sustained trend from an accidental oscillatory minimum; two finite-size points alone do not establish asymptotic fault tolerance.

Physical Routes and Their Tradeoffs

Semiconductor–superconductor Majorana devices

Hybrid nanowires, planar heterostructures, and quantum-dot chains offer electrostatic control, microwave-compatible charge sensing, and a natural path to tetron arrays. Measurement-only protocols can avoid moving Majoranas through physical T-junctions.

Their central challenge is evidentiary as well as engineering. Trivial Andreev states, disorder, soft gaps, multiband structure, and inhomogeneity can mimic local signatures. Large magnetic fields constrain materials, resonators, and array orientation. Many gates and dots must be tuned while preserving the same topological regime.

Vortices, magnetic chains, and two-dimensional candidates

Majorana modes may occur in vortices of suitable two-dimensional superconductors, magnetic-atom structures, and topological-insulator or Josephson-junction hybrids. Two-dimensional motion is attractive for physical braiding. Vortex pinning, low-lying core states, microscopic placement, fusion-channel readout, and scalable control remain formidable.

Fractional quantum Hall devices

Hall fluids supply intrinsic many-body topological order and established Abelian anyons. Chiral edges offer interferometric access, while gates can shape paths and islands. The operating temperatures, small gaps, edge decoherence, electrostatic drift, and difficulty of controlling individual bulk quasiparticles limit processor-scale architectures.

Programmable topological states

Conventional processors provide flexible preparation and tomography of non-Abelian models. They allow controlled tests of braid and fusion algebra with explicitly programmed states and measurements. Their logical protection, however, comes from active stabilization and the underlying processor’s error model. They benchmark topological protocols, not intrinsic material topological qubits.

Practical limitations

Nonlocal encoding can suppress matrix elements of local perturbations, but longer structures also change capacitance, signal strength, fabrication uniformity, and readout geometry. Increasing the intended protection scale is useful only if the gap, initialization, and measurement remain reliable.

The same distinction applies to computation: a stable ground sector can still accumulate logical phase errors, a good detector can read a trivial state, and a Clifford braid can still need an unprotected non-Clifford resource. Metrics for Quantum Hardware explains how to compare leakage, coherent errors, and conditional versus unconditional performance. Materials and Fabrication connects the operating regime to disorder and device variation.

Exercises

1. Count a fixed-parity Majorana space

Show that $2m$ Majoranas span a $2^{m-1}$-dimensional sector when total fermion parity is fixed. How many qubits can four and six Majoranas encode mathematically?

Solution

Pair the Majoranas into $m$ ordinary fermions,

$$f_j = \frac{\gamma_{2j-1}+i\gamma_{2j}}{2}.$$

Their occupations $n_j=0,1$ give $2^m$ states. Total parity fixes

$$(-1)^{\sum_j n_j} = \pm1,$$

so only half of the occupation strings remain. Thus

$$\dim\mathcal H_{\rm fixed\ parity} = 2^{m-1}.$$

Four Majoranas have $m=2$ and encode one qubit. Six have $m=3$ and span two qubits mathematically. A hardware hexon may reserve one of those degrees of freedom as an ancilla, so logical use depends on the architecture rather than dimension alone.

2. Verify the Majorana Pauli algebra

Using

$$\begin{aligned} \bar Z &= i\gamma_1\gamma_2, \\ \bar X &= i\gamma_2\gamma_3, \\ \bar Y &= i\gamma_1\gamma_3, \end{aligned}$$

show that $\bar X^2=\bar Y^2=\bar Z^2=\mathbb 1$ and $\bar Z\bar X=i\bar Y$.

Solution

For $j\ne k$, $\gamma_j\gamma_k=-\gamma_k\gamma_j$ and $\gamma_j^2=\mathbb 1$. Hence

$$(i\gamma_j\gamma_k)^2 = -\gamma_j\gamma_k\gamma_j\gamma_k = \mathbb 1.$$

Also,

$$\begin{aligned} \bar Z\bar X &= (i\gamma_1\gamma_2) (i\gamma_2\gamma_3) \\ &= -\gamma_1\gamma_3 \\ &= i(i\gamma_1\gamma_3) = i\bar Y. \end{aligned}$$

Reversing the order gives $-i\bar Y$, so $\bar Z$ and $\bar X$ anticommute. The algebra is valid after restricting to the fixed-parity code space.

3. Conjugate a Majorana by a braid

For

$$U_{12} = \exp \left( \frac{\pi}{4}\gamma_1\gamma_2 \right),$$

derive the action of $U_{12}$ on $\gamma_1$ and $\gamma_2$. Explain why a choice of braid orientation matters.

Solution

The commutators are

$$\begin{aligned} [\gamma_1\gamma_2,\gamma_1] &= -2\gamma_2, \\ [\gamma_1\gamma_2,\gamma_2] &= 2\gamma_1. \end{aligned}$$

Exponentiating this rotation gives

$$\begin{aligned} U_{12}\gamma_1U_{12}^\dagger &= -\gamma_2, \\ U_{12}\gamma_2U_{12}^\dagger &= \gamma_1. \end{aligned}$$

Replacing $U_{12}$ by $U_{12}^\dagger$ reverses the signs and corresponds to the opposite exchange orientation. Physical braid claims must therefore specify worldline orientation and operator convention, not only say that two modes were exchanged.

4. Check an adiabatic operating window

A candidate device has $\Delta=40\ \mu{\rm eV}$, $\delta E=2\ {\rm neV}$, and $\tau_{\rm poison}=2\ {\rm ms}$. Estimate $\hbar/\Delta$ and $\hbar/\delta E$. Which upper bound controls the operating window?

Solution

Using $\hbar=6.582\times10^{-16}\ {\rm eV\,s}$,

$$\frac{\hbar}{\Delta} = \frac{6.582\times10^{-16}} {40\times10^{-6}} \ {\rm s} \approx 1.65\times10^{-11}\ {\rm s},$$

or about $16\ {\rm ps}$. Similarly,

$$\frac{\hbar}{\delta E} = \frac{6.582\times10^{-16}} {2\times10^{-9}} \ {\rm s} \approx 3.29\times10^{-7}\ {\rm s},$$

or about $0.33\ \mu{\rm s}$. The splitting bound is much shorter than the $2\ {\rm ms}$ poisoning time, so the schematic window is

$$16\ {\rm ps} \ll \tau_{\rm op} \ll 0.33\ \mu{\rm s}.$$

Whether a practical pulse fits comfortably inside it requires the actual spectral path, control bandwidth, and error target. Merely satisfying the inequalities by a small numerical factor is not enough.

5. Majority vote for a parity detector

An ideal QND parity detector has independent assignment error $e=0.02$ per shot. Compute the error of a three-shot majority vote. Why is this calculation optimistic for a real device?

Solution

The vote is wrong if exactly two or all three assignments are wrong:

$$\begin{aligned} e_3 &= \binom{3}{2}e^2(1-e)+e^3 \\ &= 3(0.02)^2(0.98)+(0.02)^3 \\ &= 0.001184. \end{aligned}$$

Thus the idealized vote reduces assignment error from $2\%$ to about $0.118\%$. Real repetitions take time. Parity can switch between shots, measurement can cause transitions, errors can be correlated, and reset or threshold drift can invalidate the independent-identical-error model. A majority-vote number is meaningful only with those effects measured.

6. Separate parity stability from coherence

Construct a Hamiltonian that preserves total parity but dephases a Majorana qubit superposition. What experiment would distinguish the two lifetimes?

Solution

Residual overlap can act within the fixed-parity code space as

$$H_{\rm eff} = \frac{\varepsilon(t)}{2}\bar Z.$$

It commutes with total parity, so a parity telegraph record can remain stable. Nevertheless, fluctuations of $\varepsilon(t)$ randomize the relative phase of

$$\frac{\lvert\bar0\rangle+\lvert\bar1\rangle}{\sqrt2}.$$

For quasistatic Gaussian $\varepsilon$ of zero mean and variance $\sigma_\varepsilon^2$, averaging the relative phase gives

$$\rho_{\bar0\bar1}(t) =e^{-\sigma_\varepsilon^2t^2/(2\hbar^2)}\rho_{\bar0\bar1}(0).$$

Both total parity and the $\bar Z$ populations remain fixed. A Ramsey experiment in complementary logical bases detects this lost coherence; a $\bar Z$ dwell-time measurement does not. Relating a parity switch to poisoning additionally requires knowing which parity the detector measures.

7. Extract a localization length

Two otherwise matched devices have measured splittings $\delta E(2\ \mu{\rm m})=20\ {\rm neV}$ and $\delta E(4\ \mu{\rm m})=2.7\ {\rm neV}$. Ignoring oscillations, estimate $\xi$ from $\delta E(L)\propto e^{-L/\xi}$. Give two reasons not to call this alone proof of topology.

Solution

The ratio gives

$$\frac{20}{2.7} = \exp \left( \frac{4-2}{\xi} \right).$$

Therefore

$$\xi = \frac{2\ \mu{\rm m}} {\ln(20/2.7)} \approx 1.00\ \mu{\rm m}.$$

The inference ignores the oscillatory factor $\cos(k_{\rm F}L+\phi)$ and assumes the devices differ only in length. Fabrication disorder, confinement, tunnel broadening, or a trivial partially separated state can also change the splitting. A topology claim needs bulk and nonlocal diagnostics plus operation-level evidence.

8. Complete the gate set

Suppose perfect Majorana braids provide every logical Clifford gate. Why is the processor still not universal, and what resource must be added?

Solution

Clifford circuits acting on stabilizer states and followed by Pauli measurements are not universal for quantum computation. A non-Clifford operation, commonly represented by

$$T = \begin{pmatrix} 1&0\\ 0&e^{i\pi/4} \end{pmatrix},$$

must be supplied. A Majorana architecture can create an approximate magic state using an unprotected dynamical operation, verify or distill many noisy copies, and inject the resulting state using protected Clifford operations and measurement. The physical preparation, factories, distillation failures, and routing belong in the architecture cost.

9. Put a tolerance on a ramp

Use a two-level model for leakage out of an instantaneous ground state:

$$\begin{aligned} H(t)&=\frac{\Delta}{2}\left[ \cos(\Omega t)Z+\sin(\Omega t)X\right],\\ \Omega&=\frac{\theta_0}{\tau}. \end{aligned}$$

Start in $|1\rangle$, the ground state at $t=0$, and set $\theta_0=\pi/2$. In a frame rotating about $Y$, derive the excitation probability at $\tau$. For $\Delta=40\,\mu\mathrm{eV}$, bound it by $10^{-4}$ without relying on a finely tuned interference zero. Separately, require the worst-input infidelity from an uncompensated constant logical splitting $\delta E=2\,\mathrm{neV}$ to remain below $10^{-4}$ on its small-phase branch. Compare with Exercise 4.

Solution

Write $R(t)=e^{-i\Omega tY/2}$, so $H(t)=R(t)(\Delta Z/2)R^\dagger(t)$. The rotating-frame Hamiltonian is

$$H_{\rm rot}=\frac12(\Delta Z-\hbar\Omega Y).$$

It is constant, giving

$$\begin{aligned} P_{\rm ex}(\tau) &=\frac{q^2}{1+q^2} \sin^2\left[\frac{\tau\Delta}{2\hbar}\sqrt{1+q^2}\right],\\ q&=\frac{\hbar\theta_0}{\tau\Delta}. \end{aligned}$$

The envelope is below $10^{-4}$ for $q\le\sqrt{10^{-4}/(1-10^{-4})}$. Hence

$$\tau\gtrsim157\frac{\hbar}{\Delta}\simeq2.58\,\mathrm{ns}.$$

The logical splitting induces $e^{-i\delta E\tau Z/(2\hbar)}$. Its worst pure-input infidelity is $\sin^2(\delta E\tau/(2\hbar))$, so the small-phase requirement gives

$$\tau\le\frac{2\hbar}{|\delta E|}\arcsin(0.01) \simeq6.58\,\mathrm{ns}.$$

The two separately allocated tolerances leave roughly $2.6$–$6.6$ ns, far narrower than the inverse-energy estimates. This is an illustrative constant-speed ramp, not a braid-error theorem. Smooth pulses, additional levels, correlated faults, and phase compensation can change the result; the two individual error allowances also do not constitute a total-error certificate for a complete gate.

References

• D. Aasen et al., “Milestones Toward Majorana-Based Quantum Computing,” Physical Review X 6, 031016 (2016), doi:10.1103/PhysRevX.6.031016.
• J. Alicea, Y. Oreg, G. Refael, F. von Oppen, and M. P. A. Fisher, “Non-Abelian Statistics and Topological Quantum Information Processing in 1D Wire Networks,” Nature Physics 7, 412–417 (2011), doi:10.1038/nphys1915.
• H. Bartolomei et al., “Fractional Statistics in Anyon Collisions,” Science 368, 173–177 (2020), doi:10.1126/science.aaz5601.
• P. Bonderson, M. Freedman, and C. Nayak, “Measurement-Only Topological Quantum Computation,” Physical Review Letters 101, 010501 (2008), doi:10.1103/PhysRevLett.101.010501.
• S. Bravyi, “Universal Quantum Computation with the $\nu=5/2$ Fractional Quantum Hall State,” Physical Review A 73, 042313 (2006), doi:10.1103/PhysRevA.73.042313.
• S. Bravyi and B. Terhal, “A no-go theorem for a two-dimensional self-correcting quantum memory based on stabilizer codes,” New Journal of Physics 11, 043029 (2009), doi:10.1088/1367-2630/11/4/043029.
• E. Dennis, A. Kitaev, A. Landahl, and J. Preskill, “Topological Quantum Memory,” Journal of Mathematical Physics 43, 4452–4505 (2002), doi:10.1063/1.1499754.
• M. H. Freedman, M. Larsen, and Z. Wang, “A Modular Functor Which Is Universal for Quantum Computation,” Communications in Mathematical Physics 227, 605–622 (2002), doi:10.1007/s002200200645.
• T. Karzig et al., “Scalable Designs for Quasiparticle-Poisoning-Protected Topological Quantum Computation with Majorana Zero Modes,” Physical Review B 95, 235305 (2017), doi:10.1103/PhysRevB.95.235305.
• A. Y. Kitaev, “Fault-Tolerant Quantum Computation by Anyons,” Annals of Physics 303, 2–30 (2003), doi:10.1016/S0003-4916(02)00018-0.
• A. Y. Kitaev, “Unpaired Majorana Fermions in Quantum Wires,” Physics-Uspekhi 44, 131–136 (2001), doi:10.1070/1063-7869/44/10S/S29.
• C. Knapp, M. Zaletel, D. E. Liu, M. Cheng, P. Bonderson, and C. Nayak, “The nature and correction of diabatic errors in anyon braiding,” Physical Review X 6, 041003 (2016), doi:10.1103/PhysRevX.6.041003.
• D. Litinski and F. von Oppen, “Quantum Computing with Majorana Fermion Codes,” Physical Review B 97, 205404 (2018), doi:10.1103/PhysRevB.97.205404.
• Microsoft Azure Quantum, M. Aghaee et al., “Interferometric Single-Shot Parity Measurement in InAs–Al Hybrid Devices,” Nature 638, 651–655 (2025), doi:10.1038/s41586-024-08445-2.
• J. Nakamura, S. Liang, G. C. Gardner, and M. J. Manfra, “Direct Observation of Anyonic Braiding Statistics,” Nature Physics 16, 931–936 (2020), doi:10.1038/s41567-020-1019-1.
• C. Nayak, S. H. Simon, A. Stern, M. Freedman, and S. Das Sarma, “Non-Abelian Anyons and Topological Quantum Computation,” Reviews of Modern Physics 80, 1083–1159 (2008), doi:10.1103/RevModPhys.80.1083.
• S. Plugge, A. Rasmussen, R. Egger, and K. Flensberg, “Majorana Box Qubits,” New Journal of Physics 19, 012001 (2017), doi:10.1088/1367-2630/aa54e1.
• J. F. Steiner and F. von Oppen, “Readout of Majorana Qubits,” Physical Review Research 2, 033255 (2020), doi:10.1103/PhysRevResearch.2.033255.