A better picture, not time travel
Scientists can use later measurements to learn more about an earlier moment in a quantum experiment. They are improving their picture of the past, not changing what happened.
Shota Yokoyama and his co-authors tested this idea with light in a laboratory. Their paper was published in npj Quantum Information on October 21, 2026. The method is called quantum state smoothing.
A quantum state describes what measurements of a system might show. It includes uncertainty. Smoothing helps scientists estimate that state using a longer measurement record.
How a later clue helps
Imagine watching a ball roll behind a fence. Seeing where it comes out can help you work out where it probably was while hidden. The later sighting does not move the ball backward.
That comparison explains how extra clues help. It does not mean quantum light follows a hidden ball-like path. Measuring a quantum system can also affect it, so the calculations must follow quantum rules.
An estimate made as an experiment runs uses only measurements available so far. Scientists call this filtering. Smoothing also uses measurements made after the moment being studied. Those later clues must be collected before they can help.
Testing the idea with light
The team studied light inside an optical cavity, a space where mirrors make light travel around repeatedly. They split the light leaving the cavity between two detectors. Detectors are instruments that measure light.
The researchers gave the two sets of measurements the names Alice and Bob. Alice's estimates used only Alice's records. Filtering used the earlier part. Smoothing added the later part.
Bob's records stayed hidden from Alice's calculations. Instead, the team combined both detectors' records to make a more informed reference for comparison. That reference was still an estimate, not a perfect view of the light.
In the tested conditions, smoothing gave estimates closer to the reference on average than filtering did. This was a laboratory experiment with recorded measurements, not just a computer simulation.
What this can—and cannot—do
The result shows how scientists can learn more from data after an experiment. It does not physically repair the light or send messages into the past.
The calculations need an accurate mathematical description of the equipment. They also need to know the detectors' settings, including Bob's. The amount of improvement depends on the setup; smoothing did not give a large extra benefit in every case.
Later clues can be useful. But they cannot help with a decision that must be made before those clues arrive.
Read the research
- Early journal version of the research paper — October 21, 2026; accepted by the journal but still subject to editing.
- Authors' earlier paper — September 5, 2025; shared before journal publication, not the 2026 journal version.