The Observer Notebook · 2026-10-03

The bomb you never touch

Can you learn that something is there without a single photon touching it?

Known, reproduced

Yes. A photon that could have hit a bomb and didn't still carries the news that the bomb is there. The simple version (Elitzur and Vaidman) finds a live bomb untouched only 25% of the time. Looping the photon past the bomb many times, with a tiny turn of its polarization (the direction its wave wiggles) each lap, pushes that toward 100%. Real mirrors leak a little light every lap, and that sets the ceiling.

50/25/25boom / no info / found, one passCLOSED FORM
78% → 97.6%found untouched, N = 10 → 100 lapsCLOSED FORM
91%best with 0.1% loss per lap, N = 50EXACT SIM
73%best with 1% loss per lap, N = 16EXACT SIM
01

The Zeno loop, live

Each lap turns the photon's polarization by 90°/N toward the bomb's path. The bomb then asks "are you on my path?" with odds sin²(angle). A no snaps the angle back to zero. Each lap the photon can also be lost.

Live · one photon, N laps DEMO SCALE · THE REAL RUN USED 100,000 PHOTONS PER N + EXACT CURVES

…

left · the photon laps the loop; the arrow is its polarization (up = safe path, sideways = the bomb's path) · right · lines = ((1−L)·cos²(π/2N))N for loss L = 0, 0.1%, 1% per lap · dots = this session's photons.

02

In plain words

One pass through a two-path interferometer gives a live bomb 50% boom, 25% bright click (no information) and 25% dark click (bomb found, untouched). Firing again after every bright click raises the found share to 1/3.

The loop does better because of the Zeno effect. Each lap's turn is tiny, so the bomb almost always answers "no", and that answer resets the photon. It is the same formula as the watched qubit: 10 laps find the bomb untouched 78% of the time, 100 laps 97.6%. Losses spoil it. At 0.1% loss per lap the best plan is 50 laps and 91%; at 1% it is 16 laps and 73%.

There is a catch. A semi-transparent object also freezes the photon, so the test tells presence, not degree: a mostly see-through object still reads "there".

Prior artKwiat et al. 1999 and Mitchison and Massar verified; others cited from memory.

Next clickThe limits of safer imaging.

Known physics, reproduced. The model has perfect polarization turns, perfect bomb sensors and a single loss number per lap; real setups add mode mismatch and detector noise. The panel's tally is a few hundred photons at most and will scatter around the lines. The loss curves use the exact sim's rule, and the entry's prior art beyond Kwiat and Mitchison-Massar was cited from memory.

Nearby on the bench