Optimal phase estimation in the presence of correlated dephasing

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Hi, I'm Kai.

Mira: And I'm Mira, here with Lev.

Kai: We're looking at the paper: "Optimal phase estimation in the presence of correlated dephasing".

Mira: Here's the gist of the paper. The gist The spin-squeezed state offers practically optimal performance in the regime where phase fluctuations are positively correlated, but can be outperformed by tensor-network optimized strategies for negatively correlated fluctuations Phase Estimation Model The paper investigates optimal metrological protocols for phase estimation in the presence of correlated dephasing noise The problem is framed as a quantum channel estimation problem where the parameter theta is encoded in a quantum channel Λθ The model starts with sensing of a time-varying magnetic field B(t) which fluctuates around some stable mean value B0, viewed as an effective frequency estimation problem ω(t) = ω + ν(t) The phase fluctuations φn are assumed to be normally distributed with a covariance matrix Σ where Σij = σ squared c i−j and c ∈ (-1, 1) is the noise correlation parameter This assumption makes the model Markovian, and the conditional probability of observing phase φn+1 given φn in the previous step reads φn+1φn ∼ cφn + N(0, σ squared 1 − c 2) Squeezed State Protocol Performance The paper analyzes the performance of a basic parallel phase estimation strategy using the axis twisting spin-squeezed (SS) state as the N qubit input state ρ(N) = SS⟩⟨SS The SS state has first and second moments that are expressed in terms of statistical properties of fluctuating phases, such as ⟨σ(n)x⟩ ≃ ⟨cos φn⟩ = e−σ 2/2 = η The estimation variance for the asymptotic N → ∞ regime is given by Var(ˆθ) ≃ 1 − η squared + 2η squared P∞ k=1 sinhσ squared c−k / Nη squared For the uncorrelated dephasing case c = 0, this recovers the well known result that Var(ˆθ) ≃ 1 − η 2/Nη squared, which saturates the fundamental bound Classical Simulation Bounds The paper derives upper bounds on QFI using the classical simulation (CS) method, where FQ(Λθ) ≤ Fcl

pθ(x): = Z dx 1/pθ(x) dθ squared For the specific model, this leads to a CS bound of F(N)Q ≤ Nσ-2 1 − c 1 + c This bound is not tight even for the uncorrelated case c = 0—comparing with (22), we see that the bound on QFI should be F(N)Q ≤ Nη squared 1 − η squared = N/(eσ squared − 1), which is strictly tighter than N/σ2 The refined CS bound (37) shows significant improvement over the standard bound (29) for weak correlations Tensor Network Optimization To perform exact optimizations in the parallel protocol, tensor-network methods are employed to model entangled states as matrix product states (MPS) The optimization functions iss tnet adaptive qfi and iss tnet parallel qfi are used for numerical optimization The results show that MPS with finite and small bond dimensions outperform spinsqueezed states when used as probe states within the protocol For both positive and negative correlations (c = ±0.4) as well as for the uncorrelated case, the difference in computed Fisher information between discretizations with M = 4 and M = 5 phases is negligible for all practical purposes The systematic analysis reveals that although the C adaptive protocol is constrained by finite ancilla dimensions, it underperforms relative to MPS-based parallel strategies with bond dimension of comparable size as the ancilla dimension in the adaptive strategy For both positive and negative correlations, MPS as probe states outperform the SS state strategy, with the advantage being particularly pronounced for c < 0 The results demonstrate that tensor network optimization of both the input probe state and the measurement strategy leads to optimal schemes that saturate these bounds Key Findings on Noise Correlation The comparison of bounds with respect to known achievable precision (21) is presented in Fig.

Kai: 2, where the refined CS bound is tight at c = 0 and significantly tighter for weakly correlated noise For strongly correlated noise, the bounds coincide The analysis shows that for negative correlations, c < 0, the gap between the spin-squeezed state protocol and the bound is larger giving hope for existence of better protocols However, for positive correlations c ≥ 0, spin squeezed states perform extremely well, and the bound indicates that there is not much spaces for further improvement The paper concludes that MPS with finite and small bond dimensions outperform spinsqueezed states when used as probe states within the protocol Discretization Effects The analysis of the Rouwenhorst process indicates that for both moderate positive and negative correlations (c = ±0.4) as well as for the uncorrelated case, the difference in computed Fisher information between discretizations with M = 4 and M = 5 phases is negligible for all practical purposes The authors adopted M = 4 throughout the numerical simulations presented in the main text to ensure computational efficiency while maintaining sufficient accuracy The discretization method ceases to be a good approximation of continuously supported process in strongly correlated regimec ≈ 1 Higher-Order Moments The analysis using higher-order moment measurements shows that for Gaussian noise, we should expect noticeable improvement only for strong noise The significant gain is obtained for σ squared ≫ 1 and c << 0 For the original Markovian Gaussian process CS bound (29) tells us that first limit should be equal to 0 Moreover, for c >> 0, increasing strength of noise decreases the estimation error Final Summary of Results The paper demonstrates that MPS with finite and small bond dimensions outperform spinsqueezed states when used as probe states within the protocol Furthermore, it is found that even the most general adaptive strategies, when constrained by finite ancilla dimensions, underperform relative to MPS-based parallel strategies with bond dimension of comparable size as the ancilla dimension in the adaptive strategy These findings collectively deepen our understanding of metrological optimization in correlated quantum channels and open pathways for more efficient implementations of frequency estimation protocols in realistic noisy environments Acknowledgments S.G.

Mira: thanks Arpan Das for fruitful discussions and Piotr Dulian for the support in running the numerics with the help of the QMetro++ package S.K.

Kai: is a recipient of the Foundation for Polish Science START 2025 scholarship This work was supported by National Science Center (Poland) grant No.

Title and authors: Kai: Now let's look at the title and who wrote this paper, "Optimal phase estimation in the presence of correlated dephasing."

Mira: It’s a very descriptive title, suggesting we are dealing with how to get the best possible phase measurement when there is some noise that isn't random.

Lev: What does that really mean in terms of what we're trying to measure? Is it about fighting the noise or just figuring out how to use the state better?

Kai: It’s about identifying the optimal estimation protocol, which involves spin-squeezed states sensing strategies as well as parallel and adaptive protocols optimized using tensor-network based numerical methods.

Mira: So we're seeing a combination of different approaches here, which is interesting because it suggests that no single state or method works best for every noise pattern.

Lev: What does that imply for our error correction work? Are we supposed to design the correction code around the noise correlation structure itself?

Kai: It implies that we might need to account for how the correlations affect the phase estimation process when designing those codes, because if you ignore them, your error budget won't be accurate.

The paper's summary: Mira: The main summary of this paper is about investigating optimal metrological protocols for phase estimation in the presence of correlated dephasing noise.

Kai: They frame it as a quantum channel estimation problem where the parameter theta is encoded in a quantum channel Lambda theta, and they model it starting with sensing a time-varying magnetic field B(t) that fluctuates around some stable mean value B0.

Lev: So this is linking the abstract phase estimation to something physical we can actually measure with magnetic fields, right?

Mira: Exactly, because they view the problem as an effective frequency estimation problem omega t equals omega plus nu t, where nu t is the time-varying part of the field.

Kai: They assume these phase fluctuations phi n are normally distributed with a covariance matrix Sigma where sigma i j equals sigma squared c times absolute difference between i and j, and c is the noise correlation parameter.

Lev: So they're setting up a specific mathematical model for the noise we're dealing with, which is critical because it determines what kind of physics we are analyzing.

Mira: And this assumption makes the model Markovian, meaning the conditional probability of observing phase phi n plus one given phi n in the previous step reads phi n plus one given phi n reads c times phi n plus a normal distribution with zero mean and variance sigma squared one minus c squared.

Kai: That relationship is what allows them to move from continuous time to this discrete-time model involving N steps, where each step represents sensing for a time delta t followed by a quantum control operation V n.

The paper's improvements: Mira: Now moving onto the paper's suggested improvements, they look at how exact optimizations in those protocols become infeasible as N grows due to exponential growth of the total Hilbert space dimension involved.

Lev: So that’s a real practical hurdle for any protocol that relies on these high-dimensional entangled states, right?

Kai: Because when you can't do exact optimizations easily, they suggest using tensor-network methods to model entangled states as matrix product states in parallel protocols sixty-four <ref:2511.07211#pg3,entangled states as matrix product states>.

Mira: And they use optimization functions iss tnet adaptive qfi and iss tnet parallel qfi for numerical optimization, which is how they find the best settings.

Lev: So that means instead of trying to solve the exponential growth problem directly, we use these tensor network methods to model the states in a way that makes it numerically tractable for optimization.

Kai: They also mention decomposing general quantum combs into elementary steps connected via ancillary systems of small dimensions, which is another way to handle the complexity.

Conclusion: Mira: To wrap up, this paper shows that matrix product states with finite and small bond dimensions can outperform spin-squeezed states when used as probe states in parallel protocols for both positive and negative correlations.

Kai: And they found that even the most general adaptive strategies under finite ancilla dimensions underperform relative to MPS-based parallel strategies with bond dimension of comparable size as the ancilla dimension in the adaptive strategy.

Lev: For negative correlations, c less than zero, there's a gap between the spin-squeezed state protocol and that bound that is larger, which suggests there might be room for better protocols out there.

Mira: And they also found that tensor network optimization of both the input probe state and the measurement strategy leads to optimal schemes that saturate these bounds.

Kai: So in short, this paper on "Optimal phase estimation in the presence of correlated dephasing" shows MPS with finite bond dimensions are better than spin-squeezed states when noise is negatively correlated.

Lev: And for real hardware, it means we need to think about using tensor network methods to model the entangled states if we want to get those optimal results.

Mira: It really gives us a clear path forward by showing that the paper moves beyond just finding a simple answer and shows how to use these optimization tools effectively.