Which records win
When the world takes notes on a quantum system, which facts get copied, and how many copies does it take?
Known, reproduced
A small quantum system sits in a bath of light and air. Every photon that bounces off it carries away a faint note about where it is. Quantum Darwinism (Zurek's idea) says a fact becomes "objective" when many separate scraps of the environment each hold a copy, so many observers can read it without disturbing it.
The notebook checked which facts get copied. Only the pointer fact (the one the environment couples to, here "up or down") spreads into many copies. The phase fact (the quantum relationship between up and down) stays hidden until you hold nearly the whole environment.
Watch the environment take notes
starting…
Left · one at a time, in a seeded random order, each environment qubit touches the system and tilts by +θ/2 (ember branch, system "up") or −θ/2 (moss branch, system "down"). The brass box is the smallest scrap that already knows the pointer fact to 90%. Right · ember: how much a random scrap of f qubits knows about the system, in bits. slate: how much it knows about the phase fact. Brass dashes: the system's own uncertainty HS. Each point averages over every possible scrap, computed exactly from the overlaps cos(θ)k.
In plain words
The number of copies follows a counting law. Each environment qubit tells the two branches apart a little, set by its overlap cos θ. Stack enough of them and the branches become distinguishable, like votes adding up. The run found the copy count R ≈ N·ξ/2, where ξ measures how much one qubit separates the branches (a Chernoff rate, the standard measure of how fast repeated clues pile up).
Two things break the copies. Haze (environment qubits that start out noisy) makes each note blurrier, so far fewer scraps carry the fact. Scrambling (environment qubits talking among themselves after they took notes) smears each copy across many qubits. A small scrap then knows almost nothing, while the environment as a whole still holds all of it.
Prior artRiedel and Zurek 2010 (the dust grain figure) and Unden et al. 2019 were verified; some other references are from memory.
Next clickThe energy cost per copy.
- This is a known result reproduced in simulation. The model is the textbook one: a single system qubit, environment qubits that never talk to each other (except in the scrambling test), and perfect branching.
- The demo uses 100 environment qubits. The scaling run went to 100,000 with the same analytic engine.
- The 1e8 dust-grain figure comes from the paper. The notebook did not recompute it.
- Some prior-art references were written from memory and were not re-checked.