A bomb so sensitive a single particle of light sets it off. No way, in any classical world, to check it works without risking it. Quantum mechanics finds the live ones with a photon that never goes near them.
Here is a pile of bombs. Each has a trigger so sensitive that a single photon falling on it sets it off. Some bombs in the pile are duds — their triggers are broken, and light passes straight through as if nothing were there. You want a live one. The trouble is obvious: the only way to learn that a bomb absorbs a photon is to send a photon at it — and if it's live, it explodes. In any world built of ordinary cause and effect, to test the bomb is to risk it.
In 1993 Avshalom Elitzur and Lev Vaidman noticed that quantum mechanics breaks this deadlock. Send your photons through a Mach–Zehnder interferometer — a box where a half-silvered mirror splits each photon's path in two, the two paths are brought back together, and a second half-silvered mirror recombines them. Line it up so the two paths interfere: when nothing blocks them, every photon lands at one detector, and the other detector stays perfectly dark. Forever. Not rarely — never.
Now drop a bomb into one of the two arms. If it's a dud, nothing changes; the dark detector stays dark. But if it's live, it ruins the interference — because now the photon's path can, in principle, be known, and a path that can be known no longer interferes with itself. Sometimes (half the time) the photon goes down the bomb's arm and it explodes. But sometimes the photon goes the other way, survives — and then, with the interference broken, the once-impossible dark detector can click.
That is an interaction-free measurement: you learned that the bomb is live — that it would have absorbed a photon — without a photon ever reaching it. The instruments below build the interferometer, fire photons one at a time, count the outcomes, and show the catch (the basic scheme wastes half its bombs) and the escape (a trick that drives the waste to zero).
Switch between live and dud and watch the dark detector. With a dud, the two paths still interfere and the dark detector is silent — every photon goes to the bright one. With a live bomb, the interference is gone: half the photons explode it, and of the survivors, half reach the bright detector and half reach the dark one. Every dark click is a live bomb caught without being touched.
So you have a real measurement at a real cost. The basic scheme finds a quarter of your live bombs intact, blows up half, and leaves a quarter undecided (and a dud also gives only bright clicks, so an inconclusive run just sends another photon). It is natural to ask: how good can this get? Can you tune the interferometer to almost never explode a bomb?
Make the beam splitters unbalanced: let only a fraction R of each photon take the bomb's arm. A live bomb then explodes with probability R, certifies itself (dark click) with probability R(1−R), and is inconclusive with probability (1−R)². The efficiency η — of the runs that decide, the fraction that found the bomb intact — is (1−R)/(2−R).
The wall. Push R toward 0 and η climbs — but it never reaches ½, and the yield (the chance any given photon resolves anything) falls to nothing: you can make the test gentle only by making it almost never report. Even with infinite patience, repeating until a verdict, the best a single interferometer can do is find half its live bombs intact and explode the other half. ½ is a hard ceiling — for this scheme.
For a decade the half-ceiling stood. Then, in 1995 and 1999, Paul Kwiat and colleagues found the way past it — and it is one of the strangest tricks in physics, the quantum Zeno effect: a watched quantum kettle never boils.
Send a photon around a loop N times. Each pass, rotate its polarization by a tiny angle — a slice π/2N of a right angle — so that, left alone, after N passes it has turned a full 90°, from horizontal to vertical. But put a live bomb on the path the vertical component would take. Now each pass is a tiny measurement: with the bomb watching, the photon is repeatedly caught before it can finish turning, and snaps back to horizontal. After N gentle looks it is still horizontal — and a final check that finds it horizontal tells you a bomb is there. The bigger N, the gentler each look, the less often the bomb ever absorbs the photon at all.
With a live bomb watching the vertical path, the photon survives one pass with probability cos²(π/2N), snapping back to horizontal. After N passes it survives — and certifies the bomb intact — with probability cos²ᴺ(π/2N). Drag N up and watch the explosion probability fall toward zero.
The ceiling is gone. As N grows the detection probability climbs past ½ and on toward 1: the explosion probability falls like π²/4N. In the limit you find essentially every live bomb, intact, and almost never set one off — using, the whole time, a photon that the bomb never absorbed. Kwiat's group built it with real photons and reached efficiencies the basic scheme could never touch.