Protecting Heisenberg scaling in quantum metrology via engineered dressed states

1. Problem Statement

Quantum metrology aims to estimate physical parameters with precision surpassing classical limits, ideally achieving Heisenberg Scaling (HS) where the mean squared error scales as $1/(NT)^2$ (with $N$ probes and time $T$). However, environmental noise typically suppresses entanglement and long-time signal accumulation, reducing the scaling to the Standard Quantum Limit (SQL) of $1/(NT)$.

The central challenge is to design control strategies that suppress noise while preserving sensitivity to the target signal. Existing solutions include:

• Quantum Error Correction (QEC): Effective for Markovian noise but relies on the Hamiltonian-not-in-Lindblad-span (HNLS) criterion.
• Dynamical Decoupling: Effective for non-Markovian noise but often fails to achieve HS for qubits under dephasing.
• Limitation: Standard criteria (like HNLS) often forbid HS if the signal generator lies within the span of noise operators. Furthermore, these criteria often assume the noise operators are fixed, ignoring the possibility that static control fields can modify the system's eigenbasis, thereby altering the effective noise operators.

The authors ask: Can HS be recovered in higher-dimensional systems under general noise by engineering "dressed states" (eigenstates of a controlled Hamiltonian) to decouple the system from the environment?

2. Methodology

The authors employ a theoretical framework based on open quantum systems and quantum control theory:

• Model: A system coupled to a bath via a Hamiltonian $H = H_S + H_B + H_I$. The interaction is $H_I = \sum A_\alpha \otimes B_\alpha$.
• Dynamics: Under the Born-Markov approximation and Rotating Wave Approximation (RWA), the system evolves according to a Gorini-Kossakowski-Sudarshan-Lindblad (GKSL) master equation. Crucially, the Lindblad jump operators $L_\nu^\alpha$ depend on the eigenbasis of the system Hamiltonian $H_S$.
• Control Strategy: A constant control Hamiltonian $H_C$ is added to the free Hamiltonian. This modifies the eigenbasis of $H_S$, creating dressed states. Since the Lindblad operators are defined by transitions between these eigenstates, $H_C$ effectively reshapes the noise acting on the system.
• Optimization: The goal is to find a control $H_C$ and a code space $\mathcal{C} = \text{span}\{|\psi_0\rangle, |\psi_1\rangle\}$ such that:
  1. The code space is sensitive to the signal generator $G$ (where $H_S \approx H_0 + \delta\omega G$).
  2. The noise operators act trivially on the code space (decoherence-free) or errors are correctable (via QEC).
• Ancilla: The framework allows for coupling the sensing system to a noiseless auxiliary system (ancilla) to expand the Hilbert space, facilitating the construction of suitable subspaces.

3. Key Contributions and Theoretical Results

The paper establishes necessary and sufficient conditions for achieving HS using dressed states, categorized by noise type:

A. Dephasing and Relaxation (Low-Temperature Noise)

For environments where thermal excitation is negligible ($\gamma(\nu) \approx 0$ for $\nu < 0$), the authors prove Theorem 1:

• Condition: HS is achievable (via a decoherence-free dressed subspace) if and only if:
  $$G \notin \text{span}_{\mathbb{R}}\{ \mathbb{1}, A_\alpha \}$$
  where $G$ is the signal generator and $A_\alpha$ are the system-bath coupling operators.
• Significance: This condition is less restrictive than the standard HNLS criterion (which requires $G \notin \text{span}\{\mathbb{1}, A_\alpha, A_\alpha A_\beta\}$). By engineering the dressed basis, one can cancel noise even if the signal generator is "in the span" of the noise operators in the bare basis, provided it is not in the span of the coupling operators themselves.
• Mechanism: The control field creates a basis where the expectation values of noise operators are identical for the two code states (canceling dephasing) or zero (canceling relaxation).

B. Thermal Excitation (General Noise)

When the bath allows for thermal excitation (non-zero spectral density at negative frequencies), preventing all errors is generally impossible. The authors extend the analysis to Quantum Error Correction (QEC):

• Condition: HS is achievable (via a correctable dressed subspace) if and only if:
  $$G \notin \text{span}_{\mathbb{C}}\{ \mathbb{1}, A_\alpha, A_\alpha A_\beta \}$$
• Significance: This recovers the standard HNLS condition but applies it to the dressed operators. The quadratic term ($A_\alpha A_\beta$) arises from the Knill-Laflamme conditions required to preserve orthogonality when errors move the state out of the code space.

C. Mathematical Formulation

The optimization of finding the optimal dressed states is formulated as a Semidefinite Program (SDP), making the search for optimal control Hamiltonians computationally tractable.

4. Application: NV-Center Thermometry

The authors illustrate their theory using Nitrogen-Vacancy (NV) centers in diamond for thermometry under magnetic field fluctuations.

• Scenario: The signal is temperature-dependent zero-field splitting ($\delta\omega$), and noise arises from isotropic magnetic field fluctuations ($S_x, S_y, S_z$).
• Without Control: The signal generator is $G = S_z^2$. The noise operators are $A_\alpha = S_\alpha$. Since $S_z^2$ is in the span of $\{S_z, S_z^2\}$ (and effectively related to the noise), standard HNLS suggests HS is impossible for a qubit-like subspace.
• With Dressed States (Dephasing only): By applying a perpendicular magnetic field ($B_x$), the eigenstates are dressed into $|0\rangle$ and $|\psi_\pm\rangle = \frac{1}{\sqrt{2}}(|+1\rangle \pm |-1\rangle)$. In this basis, the expectation values of $S_x, S_y, S_z$ vanish or are equal, creating a decoherence-free subspace. Result: HS is achieved.
• With Relaxation: Relaxation requires $\langle \psi_0 | S_\alpha | \psi_1 \rangle = 0$. This cannot be satisfied within the 3D spin-1 space alone. However, by coupling to a noiseless ancilla (e.g., a nearby $^{13}$C nuclear spin), a valid code space is constructed. Result: HS is achieved.
• With Thermal Excitation: If thermal excitation is significant, the condition $G \notin \text{span}\{\mathbb{1}, A_\alpha, A_\alpha A_\beta\}$ fails because $S_z^2$ is proportional to $A_z^2$. Result: HS cannot be recovered even with QEC in this specific model.

5. Significance and Impact

1. Redefining Limits: The paper demonstrates that the "no-go" theorems for HS (like HNLS) are not absolute; they depend on the basis in which the noise is evaluated. By using static fields to engineer dressed states, one can effectively "rotate" the noise away from the signal.
2. Prevention vs. Cure: The work highlights a fundamental distinction: preventing noise accumulation via unitary dynamics (dressed states) is less restrictive than correcting errors after they occur (standard QEC). This aligns with the principle that "it is easier to prevent than to cure."
3. Experimental Relevance: The proposed method relies on static or slowly varying fields (e.g., magnetic fields for NV centers), which are experimentally feasible and do not require the high-speed, complex pulse sequences often needed for dynamical decoupling.
4. Generalizability: While illustrated with NV centers, the algebraic conditions derived are universal. They apply to trapped ions, Rydberg atoms, and cold atomic gases, providing a clear recipe for designing control Hamiltonians to achieve Heisenberg scaling in noisy environments.

In summary, this work provides a rigorous theoretical foundation for using static field engineering to create decoherence-free or error-correctable subspaces, significantly expanding the regime where Heisenberg-limited quantum metrology is possible.