Quantum metrology with full and fast quantum control

1 Introduction

• Noisy metrology motivates the analysis because typical uncorrelated noise limits quantum gain to a constant-factor improvement, although some noise perpendicular to the parameter Hamiltonian permits Heisenberg scaling with additional resources.Classical probes have 1/N standard scaling, while entangled probes can attain 1/N^2 scaling in the noiseless setting.Source: §1 · p. 1
• The framework gives analytic results for time-homogeneous master-equation noise and bounds across metrology schemes without explicitly solving system dynamics.It allows ancillary resources and full, fast control of the system and ancillae.Source: §1 · pp. 1–2
• FFQC restores Heisenberg scaling for rank-one Pauli noise not parallel to the Hamiltonian, while other noise types limit quantum gain to a constant-factor improvement over standard scaling.The paper also derives analytic bounds on the achievable improvement factor.Source: §1 · p. 2
• For standard-scaling-limited noise, FFQC can improve precision with limited resources, including examples comparing controlled schemes with schemes lacking intermediate control.The paper considers both continuous and discrete processes.Source: §1 · p. 2
• The results apply both to frequentist estimation with many repetitions and to Bayesian estimation with finite data, including a single experimental run.The general FFQC protocol also permits adaptive changes based on collected measurement records.Source: §1 · p. 2

2 Quantum metrology protocols

• Parallel sensing prepares N entangled probes, lets them sense independently for time t, and measures their final state to estimate ω.Without noise, this strategy can achieve Heisenberg scaling; uncorrelated noise can constrain precision to standard scaling asymptotically.Source: §2 · p. 3
• Ancillary particles enable a final-stage, single-step error-correction protocol that generalizes parallel sensing, while existing bounds for it are not guaranteed tight.For dephasing and particle loss, the bounds are asymptotically achievable within parallel sensing; for amplitude damping, the ancillary protocol has better scaling than parallel sensing.Source: §2 · p. 3
• For X-noise, optimizing sensing time in frequency estimation can yield 1/N^(5/3) scaling, despite fixed-time bounds imposing standard scaling.X-noise has a generator transversal to the parameter-encoding Hamiltonian; prior work also shows fast multi-step correction can restore Heisenberg scaling for this noise.Source: §2 · p. 3
• FFQC uses one sensing probe entangled with many ancillae and allows arbitrarily frequent control of their joint state, including changes to the noise process.Taking the intervals between control operations sufficiently small makes the sensing time between them infinitesimal.Source: §2 · p. 3
• Protocol (d) can match or outperform protocol (c) by sequentializing the same N probes over total sensing time t′ = Nt.The protocols are not equivalent: (d) assumes infinitesimal steps and permits sequential use of resources.Source: §2 · p. 3
• These protocols also apply to Bayesian estimation with a finite number of repetitions and a prior distribution over the parameter.Examples include reference-frame alignment, adaptive Bayesian estimation, and single-shot estimation.Source: §2 · p. 3

2.0.1 A remark on error correction

• An error-correction step needs ancilla space no larger than the system space, since relevant bipartite pure states use only a system-sized subspace of the ancillae.The passage motivates this bound through Schmidt decomposition and orthonormal ancilla states spanning a d-dimensional subspace.Source: §2.0.1 · pp. 3–4

3.1 QFI and its role in quantum metrology

• The final state of probes and ancillae encodes the parameter, and its QFI is the central quantity for assessing estimation precision.The state is measured to construct an estimate of the parameter.Source: §3.1 · p. 4, §3 · p. 4
• For repeated protocols, the Cramér-Rao bound relates mean squared error to QFI, while single-shot average error can be bounded using Bayesian Cramér-Rao or Ziv-Zakai bounds.The Bayesian Cramér-Rao bound applies for well-behaved priors.Source: §3.1 · p. 4
• For pure states under unitary parameter encoding, QFI equals four times the generator variance, and a GHZ state achieves Heisenberg scaling for a local Hamiltonian.The passage identifies the maximum- and minimum-eigenvalue eigenstates as the states used to maximize QFI.Source: §3.1 · p. 4

3.2 Time-homogeneous qubit evolution

• The assumed noise arises from uncorrelated, time-homogeneous Lindblad dynamics, allowing control operations without changing the environment.The model assumes the environment disturbs the system independently of system state and time instance; the text distinguishes this from time-inhomogeneous Zeno-regime dynamics.Source: §3.2 · p. 5
• The noise generator is restricted to its Pauli subspace after FFQC cancels ω-independent Hamiltonian terms arising from L0i and Li0.The generator matrix is Hermitian positive semidefinite, with entries independent of time and ω.Source: §3.2 · p. 5
• The text groups relevant noise processes into families, including dephasing and general rank-one noise.It identifies amplitude damping with spontaneous emission when the real and imaginary parts of the noise vector are orthogonal and equal in magnitude.Source: §3.2 · p. 5
• The noise parametrization includes asymmetric X-Y noise, expressed using rates γ1 = γp and γ2 = γ(1−p) when RΩ is the identity.The broader parametrization uses Ω in SO(3) and an Euler-angle representation.Source: §3.2 · p. 5
• The qubit evolution channel can be obtained by integrating the master equation and represented through a dynamical matrix or Kraus operators.The Kraus representation is not unique: a unitary transformation of a linearly independent operator set yields another valid set.Source: §3.2 · pp. 5–6

4.2 Other noise-types: unavoidable linear scaling

• The section derives the no-go result and analytical bounds for several physically relevant noise types, using channel extension to compare full fast control with protocols lacking it.The CE bounds depend on properties of the parameter-encoding channel.Source: §4 · p. 6, §4.2 · p. 6

4.1 Removing rank-one Pauli noise

• Fast error correction completely removes any rank-one Pauli noise not parallel to the encoding Hamiltonian, preserving Heisenberg scaling while rescaling the parameter by √(1 − (n·z)^2).The strategy slows the effective evolution by the same factor and can also implement unitary gates in the presence of rank-one Pauli noise.Source: §4.1 · p. 6

4.2.4 Universal asymptotic linear bound

• For every noise type except rank-one Pauli noise not parallel to the encoding Hamiltonian, the QFI is bounded by 4 α_L t′, so its scaling is linear even with fast control.Here α_L depends only on the Liouvillian, and t′ is the effective protocol time. The proof covers dephasing, non-Pauli rank-one noise, and higher-rank noise by reducing it to orthogonal components.Source: §4.2 · p. 6, §4.2.4 · p. 10

4.2.1 Upper bounds on the QFI

• Channel-extension bounds use channel properties to upper-bound QFI for parallel and sequential protocols, but the parallel and sequential bounds are not guaranteed to be tight.The standard bound determines the maximal QFI for an ancilla-assisted parallel scheme with N = 1.Source: §4.2.1 · p. 7, Parallel CE bound · pp. 7–8

4.2.2 Optimization of the CE bounds

• The channel-extension bounds can be optimized over Hermitian Kraus-gauge matrices; standard and parallel bounds are convex in this variable and admit systematic semidefinite-programming minimization.The sequential bound is not stated to be convex in the same way.Source: §4.2.2 · p. 8

4.2.3 Infinitesimal-timestep CE bound

• The infinitesimal-timestep channel-extension bound yields linear scaling when a Kraus gauge can set β(2) to zero, a condition that fails for rank-one Pauli noise.The analytic bound is derived in the dt → 0 limit without solving the master equation, and applies to any metrological scheme.Source: §4.2 · p. 7, §4.2.3 · p. 9

4.2.5 Exemplary noise-types

• Analytical bounds for dephasing, general rank-one, and rank-two Pauli noise are obtained by minimizing the infinitesimal-bound coefficient over Hermitian Kraus gauges under the stated expansion constraints.The supplied passages do not give the resulting bound values.Source: §4.2.5 · p. 10
• For general rank-one noise, the authors derive an analytical bound that is valid but not provably tightest, because optimizing the general case involves many parameters.They use the structure of analytically optimal controls for cases where one component of the noise vector vanishes as an ansatz.Source: §4.2.5 · p. 10

4.3 The gain allowed by fast control

• The fast-control comparison focuses on X-Y noise and rank-one transversal Pauli noise, using parallel CE bounds to upper-bound the parallel scheme’s QFI.The compared strategies include parallel evolution without fast control and two strategies with fast control.Source: §4.3 · p. 10

4.3.1 Transversal rank-one Pauli noise

• For transversal rank-one Pauli noise, FFQC removes the noise and attains F(d)(t′) = (t′)^2 and F(c)(t) = (Nt)^2; strategy (d) needs only one sensing qubit and one ancilla.The parallel scheme’s QFI is bounded to linear scaling in N for fixed t, and its bound eventually decreases as t grows.Source: §4.3.1 · pp. 10–11

4.3.2 X-Y noise

• For X-Y noise, FFQC’s ultimate QFI bound need not be attainable, while the particular bounds and strategies shown vary with whether the error record is kept.Figure 2 contrasts the ultimate upper bound with attainable bounds for a strategy that discards the error record and a fixed-ancilla variant.Source: §4.3.2 · p. 11, Figure 2 · p. 11
• Correcting the more probable X-Y error leaves a residual z-error when the guess is wrong; retaining the correction record yields announced dephasing, while discarding it gives dephasing of strength pγ.With fixed ancillas, the dephasing CE bound is attainable by strategy (c) using N ancillas or strategy (d) using N+1 total qubits.Source: §4.3.2 · p. 11

4.3.3 Advantages of FFQC

• FFQC can transform X-Y noise into weaker dephasing, prolong short-time dynamics by parallelizing evolution, and retain error records; it outperforms the parallel scheme by a large amount.Figure 2 compares the parallel bound with FFQC bounds for X-Y noise, including an attainable strategy that discards the error record.Source: §4.3.3 · pp. 11–12, Figure 2 · p. 11
• Whether strategies (c) and (d) have a gap as δt → 0 remains unclear, but strategy (d) can read error syndromes before choosing subsequent actions.For fixed δt, the bound also applies to strategy (c); unlike strategies (b) and (c), strategy (d) can defer some decisions until after reading the syndrome.Source: §4.3.3 · p. 12

5 Implications on attainable precision in metrology schemes

• For rank-one Pauli noise, FFQC can maintain precision scaling available in noiseless estimation, including in single-shot protocols.The text also notes potential use of this noise amelioration beyond metrology, including high-fidelity gate design.Source: §5 · p. 12, §5.2.1 · p. 13
• The paper bounds precision using either the per-run time-particle resource Nt or the number of probes N, with run time bounded in the latter setting.For both resource views, the bounds depend on the noise process.Source: §5 · p. 12

5.1.1 Time-particles

• For time-particle resources, the paper gives an MSE bound for rank-one Pauli noise and a lower bound for other noise processes.The bound for other processes is not guaranteed to be achievable as the number of repetitions tends to infinity.Source: §5.1.1 · p. 12

5.1 Precision in the presence of free repetitions

• Heisenberg scaling is achievable under rank-one Pauli noise, whereas bounds for other noise processes need not be achievable as repetitions grow.The non-rank-one bound is independent of single-run duration in frequency estimation.Source: §5.1.1 · p. 12, §5.1.2 · p. 13

5.1.2 Frequency estimation

• In frequency estimation with N as the resource, the rank-one Pauli bound permits Heisenberg scaling, while the other-noise bound is independent of single-run duration.The latter maintains the ultimate precision otherwise exhibited only at short time scales.Source: §5.1.2 · p. 13

5.2 Single-shot precision bounds

• Single-shot precision is assessed through the prior-averaged MSE, with Bayesian Cramér–Rao and Ziv–Zakai bounds used to constrain it.The framework treats repeated protocols as implementable within sufficiently long sequential FFQC strategies through measurements and state repreparations.Source: §5.2 · p. 13

5.2.1 Pauli rank-one noise

• 1/(NT)^2 precision scaling is attainable with a single-shot frequency-estimation strategy for the stated prior, compared with scaling only in N for repeated fixed-duration runs.The paper recommends the single-shot strategy when choosing between these options.Source: §5.2.1 · p. 13

5.2.2 Bayesian Cramér-Rao bound

• For sufficiently regular priors, the Bayesian Cramér–Rao bound recovers the many-repetition bound when the total resource is rescaled by the repetition count.The paper also generalizes the Ziv–Zakai bound to mixed states and reports recovery of the same many-repetition bound under mild prior regularity conditions.Source: §5.2.2 · p. 13, §5.2.3 · p. 13

6 Importance of quantum control for metrology with limited resources

• The section demonstrates control-assisted gains under noise that cannot be completely removed, using a one-pair frequency protocol and intermediate-control phase estimation.The proposed frequency scheme is described as experimentally tangible.Source: §6 · p. 14

6.1 Frequency estimation with X-Y noise

• For X–Y frequency noise, the asymptotic precision is limited to a constant improvement over standard scaling, but sufficient noise asymmetry lets sequential schemes outperform parallel schemes.The stated exception is p=0 or p=1, where the noise reduces to correctable perpendicular dephasing.Source: §6.1 · p. 14

6.1.1 Parallel strategy with N qubits and N ancillae

• The parallel strategy’s upper-bound QFI rate is optimized over interrogation time, whose optimum generally depends on probe number and noise form.For balanced X–Y noise, the bound can be obtained analytically.Source: §6.1.1 · pp. 14–15

6.1.2 Simple sequential strategy

• In the sequential protocol, independent probe–ancilla pairs make total QFI additive, so performance can be evaluated from one pair; retaining the error register avoids a QFI-rate loss by a factor of 1/(1−p).The register preserves information about how noisy each run was.Source: §6.1.2 · pp. 15–16

6.1.3 Comparison of the strategies

• At p=0.25, the N-pair sequential strategy is superior for N<5; at p=0.05, it remains superior for N<316, while the single-pair strategy is superior for N<4.The comparison is against an upper bound for all parallel protocols at ω=1 and γ=0.05.Source: Figure 4 · p. 16, §6.1.3 · p. 16
• As noise becomes more asymmetric, the probe-number thresholds for sequential superiority diverge; the N-pair threshold scales as p^-3/5.The single-pair threshold is N_th=1 for approximately 0.13≤p≤0.5.Source: §6.1.3 · p. 16

6.2 Phase estimation with bit- and bit-phase-flip noise

• Intermediate control gives higher QFI than parallel phase estimation for every 0<p<1, with equality only at p=0 and p=1.The asymptotic status of this advantage remains open.Source: §6.2 · p. 18

7 Summary and Outlook

• FFQC restores Heisenberg scaling for rank-one Pauli noise except noise identical to the Hamiltonian, while other noise types have QFI limited to linear scaling.The restored scaling comes at the cost of slowing evolution by a constant factor; the linear bound rules out Heisenberg scaling for other noise types.Source: §7 · p. 18
• Simple sequential protocols using one sensing and one auxiliary system can outperform parallel entanglement-based strategies with limited resources and no FFQC.The paper reports this comparison for frequency estimation and asymmetric X-Y noise.Source: §7 · p. 18
• The outlook identifies higher-dimensional probes as an open generalization, where Hamiltonian support on a subspace may permit more involved error-correcting codes.Source: §7 · p. 18

A Relation between the Kraus representations and the dynamical matrix

• Equivalent Kraus representations are related by unitary transformations, and the appendix constructs a canonical representation using singular value decomposition.The construction yields a Hermitian 4 × 4 matrix representation for the qubit channel.Source: A Relation between the Kraus… · p. 23

B Infinitesimal-timestep CE bound

• The infinitesimal-timestep channel-extension bound applies to protocols with FFQC and is derived by generalizing an earlier bound.In the dt → 0 limit, the bound is linear in total sensing time, producing SQL-like scaling.Source: B Infinitesimal-timestep CE bound · pp. 23–24

C.1 Existence of asymptotic bound for any Liouvillian but the non-parallel rank-one Pauli noise

• The infinitesimal-timestep bound gives standard scaling for every Liouvillian except rank-one Pauli noise not parallel to the evolution generator.For the excluded rank-one Pauli case, an FFQC strategy can remove the noise at the cost of slowing evolution by a constant factor.Source: §C.1.1 · p. 25
• Any higher-rank Liouvillian containing a non-Pauli rank-one component in its diagonal decomposition leads to standard scaling.Source: §C.1.2 · pp. 25–26
• The analysis shows that the bound holds for rank-two Pauli noise as well.The argument uses the spanning set formed by the two noise vectors and their cross product.Source: §C.1.3 · p. 26
• For the remaining Liouvillians, the QFI is upper-bounded by a term linear in sensing time, yielding standard scaling even with full quantum control.Source: §C.1.4 · p. 26

• The general rank-one bound may be improved when its free parameters are optimized.The stated bound is optimal in special cases where one coefficient of the noise vector vanishes.Source: §C.2 · p. 27
• Dephasing commutes with the evolution, so FFQC cannot detect it and the FFQC strategy cannot improve metrological performance.Source: §C.2.2 · p. 27
• For rank-two Pauli noise, the section sets the remaining free parameters to minimize the norm and obtain an optimal bound.Source: §C.2.3 · p. 28

• The X-Y noise evolution combines Hamiltonian encoding with Pauli errors weighted by p and 1 − p.Source: D Analysis of the X-Y noise · p. 28
• The master equation describes the probe state’s time evolution under that noise and parameter-encoding Hamiltonian.Source: D Analysis of the X-Y noise · p. 28
• The probe dynamics can be derived explicitly by analytically integrating the master equation.The passage points to methods described in.Source: D Analysis of the X-Y noise · p. 28

D.1 Simple FFQC strategy

• Parity measurements distinguish the correct and error branches without demolishing the state, and detected errors are mapped back into the code space by correcting the most probable error, assumed to be σ2 ⊗1.The code space is spanned by |00⟩ and |11⟩; the error space by |01⟩ and |10⟩.Source: §D.1 · p. 29
• Control pulses need only be fast enough for the first-order-in-dt approximation to hold, with the required pulse duration depending on decoherence strength.Detected errors are tracked in a classical register, and the corrected state returns to the code space so the strategy can continue.Source: §D.1 · p. 29
• The detected error count m over time t follows a Poisson distribution with mean γt/2.The probe state is conditioned on the occurrence of m errors.Source: §D.1 · p. 29

D.2 Description of the probe channel

• For t>0, the Choi-Jamiolkowski matrix has rank four, so the probe channel has four Kraus operators.The channel's canonical Kraus representation is obtained by spectral decomposition of the Choi-Jamiolkowski matrix.Source: §D.2 · pp. 29–30
• The control steps need only be sufficiently fast for the first-order-in-dt approximation to hold.Their required duration depends on decoherence strength.Source: §D.2 · p. 29
• The probe channel is also expressed in the Pauli basis to derive its corresponding dynamical matrix.That dynamical matrix is used in the subsequent channel analysis.Source: §D.2 · p. 30
• For balanced X-Y noise, p=1/2, the channel commutes with parameter encoding and is phase-covariant under rotations generated by σ3.In this case, the stated parameters simplify to μ=γ̃=0, Ω=iω, and ω̃=−i.Source: §D.2 · p. 30

D.3 Parallel CE bound

• The bound is evaluated over different sensing times t and numerically optimized to obtain the optimal sensing time.This procedure determines the bound on the maximal QFI rate per probe.Source: §D.3 · p. 30
• For given γ, ω, N, and t, the parallel channel-extension bound is computed by minimizing over shifted Kraus derivatives with semidefinite programming.The calculation uses the channel's canonical Kraus operators.Source: §D.3 · p. 30
• For N=1 and N→∞, analytic expressions for the bound are obtained using an ansatz for the Hermitian matrix, with correctness checked numerically through the SDP formulation.The asymptotic expression is also simplified into a slightly loosened bound that was numerically verified to approximate the original with negligible precision loss.Source: §D.3 · p. 30

D.4 Parallel CE bound for p=1/2

• For p=1/2, the parallel CE bound coincides with the QFI rate attained by the proposed sequential strategy.For smaller p, corresponding to greater noise asymmetry, the sequential strategy performs better than parallel schemes; for any p, it outperforms all parallel schemes over a non-trivial range of N>1.Source: §D.4 · p. 31

D.5 Asymptotic parallel CE bound

• In the asymptotic N→∞ regime at fixed t, the parallel CE bound is optimized by setting β(t)=0 when possible, and an upper bound is derived using the M-matrix formulation.The chosen expression satisfying the asymptotic condition is described as fully analytic.Source: §D.5 · pp. 31–32

E Ziv-Zakai precision bound and the QFI

• The Ziv-Zakai bound relates average mean-square error to the minimum error probability for distinguishing ω from ω+τ.The derivation bounds that error probability using trace distance and a QFI bound, for priors satisfying the stated regularity condition.Source: E Ziv-Zakai precision bound and the QFI · p. 32

F Strategy for noiseless frequency estimation

• A symmetric eigenstate of Jz is rotated by the phase and measured with a covariant POVM whose outcomes are θ.Choosing θ as the estimator is explicit, though the passage notes it might not minimize the MSE.Source: F Strategy for noiseless frequency… · pp. 32–33
• For frequency on the real line, variance (MSE) is the appropriate cost, and the section lower-bounds it using the phase cost C.The circular cost sin^2(δ/2) is appropriate instead for phase defined on [−π, π).Source: F Strategy for noiseless frequency… · p. 33
• The bound is minimized at ropt ≈ 0.33, giving κ ≈ 6.74.The optimization is over r.Source: F Strategy for noiseless frequency… · p. 34
• When evolution is slowed, as for rank-one Pauli noise, the longer sensing time affects the frequency MSE bound.Source: F Strategy for noiseless frequency… · p. 34