Continuous Measurement-Based Holonomic QC

• The paper demonstrates that continuous measurement-based holonomic quantum computation implements logical gates by dragging the code space along closed loops in the Grassmannian, exploiting non-Abelian holonomy.
• It employs continuous monitoring of rotated stabilizer generators and the quantum Zeno effect to confine the state to instantaneous code subspaces, enabling real-time error detection and corrective steering.
• Holonomic gates are generated through geometric connections, with performance gains shown via mid-flight steering that shortens gate times and elevates fault-tolerance thresholds in stabilized architectures.

Continuous measurement-based holonomic quantum computation is a measurement-only framework for implementing logical unitaries on quantum error-correcting codes by dragging the code space around a closed loop in the Grassmannian and exploiting the resulting non-Abelian holonomy. In this setting, rotated stabilizer generators are monitored continuously or in a dense sequence of incomplete measurements, so that the quantum Zeno effect confines the state to an instantaneous code subspace while the closed path of projectors induces the target logical gate. Recent formulations extend this picture from geometric gate synthesis to fault-tolerant operation by decoding the continuous measurement record, identifying the rotated syndrome subspace populated during evolution, and steering the holonomic path in real time so that the final evolution still realizes the desired logical gate (Lanka et al., 3 Mar 2026, Lanka et al., 8 Oct 2025).

1. Formal structure of the rotating code space

At each time $t$, the $n$-qubit Hilbert space decomposes into $2^{n-k}$ simultaneous eigenspaces, or syndrome spaces, of the rotated stabilizer generators $g_j(t)$. The $+1$ eigenspace is the instantaneous code space, projected onto by

$P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$

An equivalent parametrization uses a one-parameter family of stabilizer generators

$S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$

with instantaneous projector

$$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$

so that as $\theta$ runs once from $0$ to $n$0, $n$1 traces a closed loop in the Grassmannian $n$2 (Lanka et al., 3 Mar 2026, Lanka et al., 8 Oct 2025).

The continuous implementation is formulated in terms of weak, continuous measurements of each Pauli stabilizer $n$3 at rate $n$4. The infinitesimal Kraus operators may be written as

$n$5

where $n$6 are independent Wiener increments with $n$7 and $n$8. For the pure-state ensemble, the conditioned evolution satisfies the stochastic Schrödinger equation

$n$9

In density-matrix form, without feedback and including a possibly Markovian noise Liouvillian $2^{n-k}$0,

$2^{n-k}$1

with $2^{n-k}$2 and $2^{n-k}$3 (Lanka et al., 3 Mar 2026).

This formalism makes the code space itself the moving object. Logical dynamics are not produced by a direct Hamiltonian acting on encoded qubits in the usual sense, but by the geometry of the projector family together with the measurement back-action that keeps the state aligned with that family.

2. Holonomy, connection, and the Zeno limit

A holonomic gate is the $2^{n-k}$4-holonomy acquired by horizontally lifting a loop $2^{n-k}$5 through the connection

$2^{n-k}$6

so that the final logical action is

$2^{n-k}$7

In the equivalent continuous-connection description, with a local frame $2^{n-k}$8 for the instantaneous code space, the non-Abelian connection is

$2^{n-k}$9

and the confined evolution over one loop is

$g_j(t)$0

This is the logical unitary enacted on the encoded subspace (Lanka et al., 8 Oct 2025).

The mechanism enforcing confinement is the quantum Zeno effect. In the limit $g_j(t)$1 rotation speed $g_j(t)$2 and bath-correlation time, the continuous back-action confines the state to the instantaneous code eigenspace, or Zeno subspace. In the $g_j(t)$3 limit, the conditioned state obeys

$g_j(t)$4

with $g_j(t)$5 the control Hamiltonian effecting the stabilizer rotations. The same geometric evolution can be written in the dense-measurement limit through the effective Hamiltonian

$g_j(t)$6

or, with an interleaved system Hamiltonian $g_j(t)$7, through the Zeno Hamiltonian

$g_j(t)$8

In these formulations, the confined motion is purely geometric in the moving code-space basis, and non-Markovian bath transitions are suppressed through the quantum Zeno effect (Lanka et al., 3 Mar 2026, Mommers et al., 2021, Lanka et al., 8 Oct 2025).

A common source of confusion is the role of adiabaticity. The continuous-measurement formulation does not discard the standard adiabatic picture; rather, it replaces the usual adiabatic transport of a Hamiltonian eigenspace by measurement-enforced transport of a code projector. The geometry is still encoded in the connection, but the confinement is generated by strong monitoring.

3. From discrete holonomies to continuous logical gates

The immediate antecedent of continuous measurement-based holonomic computation is the discrete-holonomy construction built from sequences of incomplete projective measurements. In an $g_j(t)$9-dimensional Hilbert space with a $+1$0-dimensional computational subspace $+1$1, an incomplete projective-filtering measurement is defined by

$+1$2

and a sequence of $+1$3 such measurements with all successful outcomes transforms an initial state according to

$+1$4

The purely unitary part of $+1$5 defines the discrete holonomy $+1$6, and when the points $+1$7 become dense along a smooth closed path, the product converges to the path-ordered exponential of the adiabatic connection (Mommers et al., 2021).

This discrete framework already provides explicit universal gate constructions. For spin-coherent-state qubits, a non-trivial non-Abelian holonomy on a two-dimensional subspace requires at least $+1$8 measurement steps. Appropriate four-step loops realize

$+1$9

and Clifford$P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$0T gates on a $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$1 spin-coherent-state qubit can be realized by solving for the relevant phase parameters. A two-qubit entangling gate is obtained by extending the projector structure to a tensor-product space and generating an effective controlled phase (Mommers et al., 2021).

The continuous stabilizer-code constructions preserve this geometric logic while shifting the implementation to encoded subspaces. One analytically derives the sequence of rotated stabilizer generators that produce a desired holonomy, and the loop can be chosen so that $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$2, with $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$3 a logical operator. More generally, a universal logical-gate set is obtained by choosing stabilizer codes, such as Shor’s $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$4 code, satisfying the Knill–Laflamme rotation conditions, and then continuously measuring, decoding, and steering (Lanka et al., 8 Oct 2025, Lanka et al., 3 Mar 2026).

4. Continuous monitoring, syndrome decoding, and path steering

Fault-tolerant operation in the continuous setting depends on extracting actionable information from the measurement record. The raw measurement currents are

$P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$5

By filtering, for example by real-time stochastic-master-equation integration or exponential moving average, one retains estimates $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$6 and thus the binary syndrome $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$7. The measurement record therefore identifies which rotated syndrome subspace is populated during the evolution (Lanka et al., 3 Mar 2026).

If a Markovian error $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$8 occurs at time $P_0(t)=\prod_{j=1}^{n-k}\frac12[I+g_j(t)].$9, or if a measurement-induced jump is detected, the state is carried from $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$0 into one of the rotated syndrome spaces $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$1 or $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$2. The post-measurement state may be written as

$S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$3

or

$S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$4

with $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$5 known analytic functions of $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$6, $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$7, and $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$8. In the discrete-path treatment, a jump at $S_j(\theta)=V(\theta)\,g_j(0)\,V(\theta)^\dagger,$9 has the form

$$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$0

where $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$1 (Lanka et al., 3 Mar 2026, Lanka et al., 8 Oct 2025).

The correction mechanism is mid-flight steering of the holonomic path. By Proposition IV.3, the horizontal lift of any curve $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$2 is unique. Once an error and its syndrome are known, one chooses a corrected rotation angle $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$3 by solving for

$$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$4

so that at $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$5 the resulting lift on the error subspace reproduces $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$6. In the alternative discrete notation, one reroutes the remainder of the loop with a new path

$$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$7

choosing $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$8 so as to return either to the original code space with gate $$P(\theta)=\prod_{j=1}^{n-k}\frac12[I+S_j(\theta)] = V(\theta)\,P(0)\,V(\theta)^\dagger,$$9 or to the error space with emulated holonomy $\theta$0 (Lanka et al., 3 Mar 2026, Lanka et al., 8 Oct 2025).

The feedback here is classical in the operational sense: from the record of $\theta$1 measurements one infers errors and updates the control path $\theta$2. This is equivalent to adding a corrective Hamiltonian $\theta$3, even though the formalism does not require an explicit Wiseman–Milburn feedback Hamiltonian (Lanka et al., 3 Mar 2026).

5. Measurement-induced jumps and relaxation of adiabaticity

A distinctive feature of continuous measurement-based holonomic computation is that non-adiabaticity manifests as a measurable syndrome event rather than only as an untracked loss of fidelity. Even in the absence of environment, if $\theta$4 is not small compared to $\theta$5, the stochastic back-action can push the state into the orthogonal rotated-$\theta$6 syndrome space $\theta$7. This is formally a jump into a measurement-induced error subspace (Lanka et al., 3 Mar 2026).

For this process, Proposition II.1 gives the jump probability

$\theta$8

Once the syndrome identifies $\theta$9, the jump is corrected by choosing an appropriate supplementary rotation $0$0 so that the final holonomy is $0$1 in the $0$2-twisted code space. The same logic applies to ordinary Markovian errors: detection via the rotated syndrome is followed by an adaptive continuation of the path (Lanka et al., 3 Mar 2026).

This correction capability changes the usual adiabatic-speed trade-off. Conventional adiabatic holonomy requires $0$3 so that jumps are negligible. In the steering-based framework, one allows $0$4 as large as $0$5, accepts a finite $0$6, and corrects each jump on the fly. The requirement $0$7 is therefore replaced by the weaker condition $0$8 bath-correlation rate together with efficient decoding and steering (Lanka et al., 3 Mar 2026).

The discretized treatment expresses the same transition in terms of success probability. For $0$9 steps with $n$00, if all measurements return $n$01, then

$n$02

Since $n$03 and $n$04, one has

$n$05

so the adiabatic-like condition $n$06 guarantees vanishing error in the unsteered limit (Lanka et al., 8 Oct 2025). The steering protocol supplements this asymptotic picture by making finite-speed operation compatible with error recovery.

6. Fault tolerance, code conditions, and demonstrated performance

The fault-tolerance claims of this framework are conditional on code-theoretic compatibility. The instantaneous code $n$07 must remain able to correct a fixed correctable set $n$08 for all $n$09, which is expressed by the Knill–Laflamme condition

$n$10

Since $n$11, this is equivalent to requiring that the rotated errors

$n$12

be correctable by the static code $n$13 (Lanka et al., 8 Oct 2025).

A sufficient set of Pauli-commutation conditions is given as follows. One requires: (1) $n$14; (2) every stabilizer generator whose weight is $n$15 (or $n$16 if $n$17 is even) anticommutes with $n$18; and (3) every logical operator of weight $n$19 (or $n$20 if $n$21 is even) commutes with $n$22 and anticommutes with $n$23. Under these hypotheses, the required projected operator products vanish or are proportional to $n$24 (Lanka et al., 8 Oct 2025).

When a bare $n$25 code does not meet these conditions, the protocol remains applicable by augmenting the code with at most two ancilla qubits in the $n$26 state. With enlarged projector $n$27, enlarged $n$28, and enlarged $n$29, one has: $n$30 if the number of distinct syndromes of all products $n$31 is $n$32; $n$33 if $n$34; and $n$35 otherwise (Lanka et al., 8 Oct 2025).

Within that code-theoretic setting, the 2026 steering framework reports explicit performance gains. In the bit-flip code example with environmental bit flips at rate $n$36, steering raises the threshold from $n$37, with no correction, to $n$38 for logical-fidelity $n$39. For a fixed target fidelity, enabling mid-flight steering shortens the gate time $n$40 by $n$41–$n$42 compared to a fixed-path measurement-based holonomic computation. The same framework states that continuous-measurement MHQC achieves holonomic gates transversally via Zeno confinement, suppresses non-Markovian noise through the quantum Zeno effect, and actively corrects measurement-induced and Markovian errors by real-time decoding of the measurement record and on-the-fly path steering (Lanka et al., 3 Mar 2026).

A common misconception is that the geometric character of holonomies alone yields fault tolerance. The cited schemes separate passive geometric robustness from active protection: the geometry supplies the logical gate, but fault tolerance depends on commuting or covariantly transforming syndrome measurements, real-time syndrome decoding, and corrective steering or recovery. This suggests that continuous measurement-based holonomic computation is best understood not as a purely geometric alternative to quantum error correction, but as a geometric control layer embedded inside stabilizer-based fault-tolerant architectures (Mommers et al., 2021, Lanka et al., 3 Mar 2026).