A qubit that can’t move while you watch it
A 2015 news story says that watching atoms freezes them in place, like the statues in Doctor Who that can only move when nobody is looking. We ran the cleanest version of that on a real quantum chip: turn one qubit from H towards V in small steps, and check it after every step. Checked 16 times, it stayed put three runs in four. Left alone, the same 16 turns carried it over almost every time.
The verdict
The claim holds
Checking a qubit over and over does freeze it: on a real IBM quantum chip, a qubit watched 16 times on its way to flipping stayed put 76.7 % of the time, against 0.5 % unwatched. Every look leaks about 0.7 % of runs, so the freeze tops out near 77 %, and the “watcher” is a physical measurement, not a pair of eyes.
Each look finds the qubit almost exactly where it started and puts it back there. Sixteen looks, sixteen resets: 76.7 % of runs never left H.
The same 16 turns with no looks carried the qubit to V in 99.5 % of runs. The turning is real; only the looks stop it.
Each look loses about 0.7 % of runs on this chip. Past 16 looks the losses outweigh the gain, so the freeze tops out near 77 %.
The “look” is a second qubit that copies the V part and is read out mid-circuit. Nobody watches anything. The measurement does the freezing.
The claim, as presented. The article presents a Cornell experiment in which ultracold atoms, measured in rapid succession, were effectively frozen in place, and compares this to the Doctor Who monsters that cannot move while they are watched.
Gizmodo (Jennifer Ouellette), “Quantum "Weeping Angel" Effect Freezes Atoms in Place”, September 2015 · the episode
The registry’s note on sources. Read from the article itself on 2026-11-08. The article notes that the analogy is not exact and that the “watching” is the light used to image the atoms. The Cornell paper behind it, Patil, Chakram and Vengalattore, Phys. Rev. Lett. 115, 140402 (2015), was not opened.
The experiment behind the story is Patil, Chakram and Vengalattore at Cornell, Phys. Rev. Lett. 115, 140402 (2015). This episode tests the effect the story describes on a different system, one qubit, where every look can be counted.
Gates locked before the job ran: the Quantum Wizard round-1 predictions file (sha256 e7e2aef7…6b, committed before submission), adopted unchanged. This episode was framed after the run, and its claim mapping and verdict rule were written afterwards. Self-graded, with no outside review.
The trick, in four steps
- Turn a littleRotate the qubit 90/N degrees from H towards V. After N turns it would point straight up.
- LookAsk “has it left H?” After a tiny turn the answer is almost always no: the chance of yes is sin² of the angle.
- Snap backA “no” does more than report. It puts the qubit back exactly at H, so the next turn starts from zero again.
- RepeatN small chances of being caught add up to far less than one big one. More looks, smaller turns, a firmer freeze.
Watch it, or don’t
Runs liveThe arrow is one run. The grid is 400 runs of the same experiment, drawn from a fixed seed. Pick how many looks, then press Play or go one Step at a time. Switch to Unwatched to see the same turns with nobody looking, or to Leaky chip to add the 0.7 % per look that ibm_fez charged.
Try 2 looks, then 32. With 2, a look comes after a 45° turn and catches half the runs, so a quarter survive both. With 32, each turn is under 3° and almost nothing is caught. The chip column is the measured result for the full N, and it stays fixed while the demo runs.
The chip, look by look
One job on IBM’s ibm_fez, 2026-10-24, qubits 71 and 58, 4000 runs at each N. The bars are the measured share of watched runs that stayed at H.
The checks
The four lines below were written and locked before the job was submitted. Each one was then turned red on purpose with a planted fault.
| Check | Result | Numbers |
|---|---|---|
| G1 · watched, it stays put more than one pass allows (25 %) by 3 SE | pass | 76.7 % at N = 16, SE 0.67 %: 77 SE over |
| … with looks that record nothing | red, as it should be | 0.8 % at best |
| G2 · unwatched, it turns over in more than 80 % of runs at every N up to 16 | pass | 99.2 to 99.5 % |
| … with the turns removed | red, as it should be | 0.5 to 0.8 % |
| G3 · the peak sits within one step of where the measured leak puts it | pass | peak at 16; the leak puts it at 20, read off the grid as 24 |
| … with a leak of 5 % per look | red, as it should be | best N moves to 8 |
| G4 · the peak sits within 0.70× to 1.05× the noise model | pass | 76.7 % against 77.2 % |
Framed after the run. The job ran as round 1 of the Quantum Wizard lab, where these lines were locked as Z1 to Z4. This episode adopts them unchanged. The mapping from the news story to these checks, and the rule that G1 and G2 together decide the verdict, were written on 2026-11-08, after the numbers were known. They cannot move a number, and they are labelled for that reason. The same job also ran a weak-value test whose first line missed; that one belongs to a different claim and is in the full run log.
Who pays for the freeze
The looks are physical. Each one is a two-qubit gate followed by a mid-circuit readout and a reset. On this chip that costs about 0.7 % of runs per look, nearly the same at every N. That one number predicts the whole curve, including where it peaks.
The freeze is bought with the motion. A frozen qubit does nothing. The runs that stayed put are the runs where the rotation was thrown away, look after look. There is no energy to collect and no signal sent.
“Watching” means interacting. The news story says as much: the “watcher” in the Cornell experiment is the imaging light. Here it is a CNOT gate. In quantum theory anything that records whether the qubit has left H does the freezing; this run used a readout that was read.
- The Cornell system itself: atoms hopping in an optical lattice, frozen by imaging light.
- The anti-Zeno effect, where looks at the wrong rate speed a change up instead.
- A second chip, a second day or a second pair of qubits. This is one job.
Read: the Gizmodo article · Patil, Chakram & Vengalattore, arXiv:1411.2678. Prior art: Misra & Sudarshan (1977) proposed the effect; Itano et al. saw it in trapped ions (1990); Elitzur & Vaidman posed the bomb tester (1993); Kwiat et al. ran the many-look version with photons (Phys. Rev. Lett. 74, 4763, 1995). No novelty claimed.