Noise in Metrology

The Heisenberg limit was derived for ideal, noiseless probes. Real sensors decohere: the delicate phase information leaks into the environment. This lesson asks what survives, and the answer is sobering — under the most common noise model the dramatic $1/N$ advantage of entanglement largely evaporates, leaving only a constant-factor improvement over the SQL.

Dephasing: the dominant enemy

The phase being measured lives in the coherence between $|0\rangle$ and $|1\rangle$ — the off-diagonal element of the density matrix. Dephasing attacks exactly this element. For a single qubit evolving for time $t$ , the coherence decays as

\rho_{01}(t) = \rho_{01}(0)\, e^{-t/T_2},

where $T_2$ is the transverse coherence time. The measurable Ramsey-fringe contrast shrinks by the same factor, so the usable phase signal is multiplied by $e^{-t/T_2}$ .

Why GHZ states lose their edge

A GHZ state's whole power came from accumulating phase $N$ times faster. The flip side is that it also dephases $N$ times faster: if any one of its $N$ qubits picks up random phase noise, the global superposition between $|0\rangle^{\otimes N}$ and $|1\rangle^{\otimes N}$ is scrambled. Under independent dephasing with rate $1/T_2$ per qubit, the GHZ coherence decays as

e^{-N t / T_2},

$N$ times faster than a single qubit. The longer interrogation time that helped the noiseless GHZ probe now hurts it.

The Huelga result: back to $1/\sqrt N$

Huelga and co-workers (1997) optimised the full noisy trade-off — interrogation time versus contrast — and found that under uncorrelated Markovian dephasing the best entangled strategy achieves

\Delta\theta \;\sim\; \frac{1}{\sqrt N}\cdot\frac{1}{\sqrt{\,T_2 / e\,}} ,

the same $1/\sqrt N$ scaling as independent probes. Entanglement buys only a bounded constant factor, not the asymptotic $\sqrt N$ gain. The Heisenberg scaling is not robust to this kind of noise.

What still helps

Not all hope is lost, and several routes recover an advantage:

Different noise geometry. If the noise is transverse to the signal generator (rather than parallel to it), partial Heisenberg scaling can survive, and quantum error correction can in principle restore $1/N$ when the noise and signal have distinguishable directions.
Shorter, smarter probes. Spin-squeezed states and intermediate-size entangled blocks often beat both the bare SQL and the fragile full-GHZ strategy at realistic noise levels.
Longer $T_2$ . Dynamical decoupling and better isolation push $T_2$ up, directly improving the constant factor for every strategy.

The bigger picture

Noisy quantum metrology is an active research frontier precisely because the clean hierarchy SQL $<$ Heisenberg becomes a subtle, noise-dependent landscape. The next lesson surveys the devices — clocks and magnetometers — where these trade-offs are engineered in practice, and where even a modest constant-factor gain translates into world-leading measurements.