← What’s Lost with a Look

Superposition and entanglement of untouched qubits during readouts

Does measuring some qubits mid-circuit cost the untouched qubits around them more superposition and more entanglement than an idle of exactly the same length? IBM 156-qubit processors ibm_fez, ibm_kingston, ibm_marrakesh, 21 October 2026.

Design

Per chip: 39 measured qubits. Among the 117 untouched qubits, 48 Bell pairs (entanglement) and 21 single qubits in superposition. Then 0, 8, 16 or 32 steps of either idle (measured qubits wait exactly the published readout length) or readout (measured qubits read out, held in 0). Entanglement read as concurrence from five two-qubit settings; superposition as coherence |⟨X⟩ + i⟨Y⟩|. Both are blind to a plain frequency shift or a fixed phase kick; a loss of coherence or a kick that varies from shot to shot moves them. 2000 shots per setting, one job per chip.

Plan and analysis hashed before submission: PREREG.md, SHA-256 17ecdc61f533… (full hash).

Planned result (plan hashed locally)

Loss after N readouts divided by loss after N equal idles; median over units, 95 % bootstrap interval. The plan named 32 readouts as the primary reading and stated the headline would hold only if at least two of three chips read above 1 for each quantity.

The primary reading (a loss ratio at 32 steps) came out consistent with 1 on all three chips for both quantities. That is a statement about the ruler: a ratio shrinks a fixed deficit toward 1 once the idle loss is large. For concurrence the primary reading cannot tell: by 32 steps the Bell pairs have lost their entanglement in both arms, and a ratio of two complete losses is 1 by construction (we chose that step count without checking the pairs would survive it). For coherence only 0.09–0.25 remains at 32 steps in the idle arm, and the plain difference below shows the deficit is still there. The secondary readings declared in the same plan show a small early excess: after 8 readouts coherence loss is above the idle on all three chips, concurrence loss above it on kingston (8 and 16). declared analysis output, all cuts

Exploratory (after the data): converting each unit’s loss into idle-equivalent time gives 1.03–1.07 for coherence and 1.03–1.08 for concurrence at 8 readouts, 1.19 for marrakesh concurrence at 16 (an unclipped witness, 37 of 48 pairs), and slightly below 1 for every cell at 32. script · output

What it says

A loss ratio shrinks toward 1 as the idle loss grows, so it hides a fixed deficit. In plain coherence (readout arm minus idle arm, median per qubit; coherence_difference.py, output) the deficit is −0.011 fez, −0.005 kingston, −0.013 marrakesh at 8 readouts (95 % intervals below 0) and about the same at 32 (−0.010, −0.006, −0.014; intervals reaching 0): readouts cost the untouched qubits a small amount of coherence at 8 readouts and about the same at 32; whether the cost compounds past 8 readouts cannot be read with these qubits (at 32 the intervals reach 0). Plain concurrence difference per Bell pair (concurrence_difference.py, output, after the data): −0.014 [−0.027, −0.004] on kingston at 8 readouts and −0.015 at 16; fez −0.006 [−0.015, +0.005], cannot tell; marrakesh cannot tell (its pairs have nearly lost their entanglement by 8 readouts in the idle arm). A fixed phase kick cannot reduce coherence; a kick that varies from shot to shot could, and these runs do not separate the two. A follow-up run examined whether measured qubits accidentally left in 1 could cause it: on ibm_marrakesh the excess is far above what random excitation gives (see that page); on ibm_kingston the excess is +0.004 [−0.003, 0.014], an interval that contains both 0 and the raw-readout run’s kingston cost. An earlier Bell-pair run on marrakesh (plan hashed locally) saturated: published here. The plan’s hash was computed locally and not externally timestamped; its declared exploratory split (units next to a measured qubit vs not) is in the excitation test.

Job records: fez · kingston · marrakesh

Files

Bones · Wayfinder · Sage · Shuttle · Nova
Heuremen LLC
“We found it.”