← Sensing 2026 hub
Sensing 2026 · Anchor 4 · The quantum ladder

More entanglement — always better?

Phase uncertainty Δφ vs number of atoms N. Climb the ladder; then turn on noise and watch which resource actually wins.

The instrument — climb the rungs, then add decoherence
Decoherence:

sets the GHZ sweet spot N* — more noise, smaller N*

the dashed vertical line — read each strategy off there

Δφ vs N (log–log). SQL 1/√N squeezed ξ/√N GHZ ideal 1/N GHZ realistic — click a rung to highlight; ◆ marks the GHZ sweet spot.

Try this Climb the rungs, then add decoherence

The question: the Heisenberg limit (1/N) beats the standard quantum limit (1/√N) — so should you always reach for the most entangled state? Set the decoherence, pick your N, and read each strategy off the dashed line.

Got a feel for it? Step to 2 · Notice

Notice this Which resource actually wins?

Explain this

At zero noise the GHZ line (1/N) is best at every N. Turn up decoherence and it grows a bathtub — a sweet spot, then it climbs. Why does piling more atoms into one entangled state eventually make the measurement worse, and why can plain atoms overtake it?

Reveal a one-paragraph answer

A GHZ state of N atoms is N times more sensitive to phase — that is the 1/N gain — but it also dephases N times faster: one stray photon scrambles the whole entangled register. So as N grows, the usable coherence time shrinks, and past a noise-set sweet spot N* the fragility outweighs the gain. Independent atoms (SQL) carry no such penalty: each fails on its own, so their 1/√N keeps improving and eventually overtakes the saturated GHZ. Spin-squeezing sits in between — a smaller, more robust gain. The fix for the fragility is not "more entanglement" but protecting it: that is level 4, quantum error correction (toggle +QEC to push N* back out).

Show the math — slopes & the schematic

SQL: independent probes give Δφ ∝ N−1/2 (this is the same 1/√N wall as the white-noise floor in the anchor-2 explorer). Squeezing lowers the prefactor: ξ/√N, ξ<1 (Caves 1981). A GHZ/NOON state reaches the Heisenberg limit Δφ ∝ N−1 (Giovannetti et al. 2011). Schematic note: the realistic curve here is a teaching cartoon — Δφ ≈ max(1/N, N/N*²), a bathtub with its minimum 1/N* at N = N*. Its asymptotic message is the real result: under uncorrelated dephasing the Heisenberg advantage is lost and one reverts toward 1/√N (Huelga et al. 1997; Demkowicz-Dobrzański et al. 2012). +QEC pushes N* up by protecting coherence.

Now connect it

  • Back to anchors 2 & 3: the blue SQL 1/√N line is the standard quantum limit (anchor 3) — the very same −½ slope you met as the white-noise floor in the Allan explorer (anchor 2). The ladder is about bending that line; decoherence is the “drift” that limits how far you can bend it — and the GHZ bathtub is the Allan bathtub's twin (a resource that stops helping past a sweet spot).
  • Forward to anchor 5 (back-action) & QEC: what keeps the 1/N Heisenberg gain in a noisy world is error correction (level 4). Back-action evasion / squeezing (anchor 5) is a different lever — it tames measurement disturbance and lowers the SQL prefactor (ξ/√N), rather than protecting Heisenberg scaling against decoherence.
Exercise check. At Realistic noise, set N to 100. Read off Δφ for the SQL line and for the realistic GHZ line, and say in one sentence which strategy you would actually build — and why "reach for the Heisenberg limit" is the wrong instinct here.
Step 1 of 4