# When Two Quantum Copies Are Worth More Than Twice One

For decades, physicists knew that joint measurements could recover quantum evidence that separate readouts miss. A new Pudim Letter identifies the universal turning point for full-rank states: two copies.

**By Correspondent v0.2**

## The long road to two copies

In 1991, Fumio Hiai and Dénes Petz proved a many-copy promise. If enough identically prepared quantum states are measured together, the available evidence approaches the full quantum-information limit.

That result established the destination, not the first useful step. It did not say how many copies must be joined before a collective readout beats the best repeated one-copy measurement.

The gap sharpened in 2017. Mario Berta, Omar Fawzi and Marco Tomamichel proved that a one-copy measurement loses information whenever two full-rank states do not commute.

The same year, Fumio Hiai and Milán Mosonyi showed that some finite batch must yield a strict gain. The batch could have been two copies, three or far more; their result did not locate the first.

By 2023, laboratories had made few-copy collective measurements tangible. Lorcán Conlon and colleagues wrote that “two-copy collective measurements can offer an advantage.” Their experiment studied minimum-error discrimination, a different score and state family from this Letter.

## The missing boundary

That history left a clean question. For measured relative entropy—the best outcome-level evidence obtainable after optimizing a measurement—does a universal advantage begin at two copies?

Place two transparent grids on top of each other. Broad bands appear in the overlap, though neither grid contains them alone. Quantum states are not grids, but the image captures how a shared readout can reveal a pattern that separate readings miss.

A separate strategy measures each copy in turn and combines the answers. The later choice may depend on earlier results, but every copy is still handled alone. A collective measurement instead treats the batch as one quantum object.

The answer is exact for finite systems and full-rank states, whose density matrices have no zero eigenvalues. The best joint score equals the repeated one-copy score exactly when the states commute.

When they do not commute, every fixed batch of at least two copies is strictly better. “Strict improvement begins at two copies,” the Pudim Letter concludes.

Commuting states share a basis in which both look like ordinary probability lists. One measurement can read that common basis directly. Joint access then has no hidden incompatibility to exploit.

## How the proof finds the gain

To find the gain, the proof starts with the best one-copy measurement and applies it to both copies. It then changes only two possible answer directions, like nudging a pair of compass needles while leaving the rest fixed.

One control chooses the direction of the nudge. Another chooses its size. For every noncommuting full-rank pair, the proof identifies a direction in which the score rises immediately.

That local rise proves strict gain at two copies. Repeating the construction inside larger batches extends the result to every fixed batch beyond one copy.

A 2022 theorem by Yonglong Li, Vincent Tan and Marco Tomamichel closes another route. Fresh-copy measurements guided only by earlier classical results cannot beat the repeated one-copy ceiling. The Letter uses this known bound; it does not claim it as new.

## A thermodynamic reading with firm walls

The theorem is general, but its thermodynamic reading has firm walls. The system must begin in thermal equilibrium, evolve in isolation and be compared with the thermal state for the final energy landscape at the same temperature.

Only in that setting does quantum relative entropy equal dissipated work measured in thermal units. This is not a universal definition of entropy production. Open systems can require accounting for exchanges and correlations with an environment.

Within the isolated model, the boundary becomes a test for energetic coherence: the driven state and final energy operator do not commute. A calibrated positive two-copy gain can reveal that coherence. A restricted null result cannot rule it out.

## A classical illustration

The release tests the proof on a declared family of two-level states. Classical calculations certify the best one-copy readout and evaluate the explicit two-copy adjustment. The example makes the mechanism concrete without becoming a hardware demonstration.

Across an archived grid of more than 450 sample points, the best recovered gain is small but positive. Exact values, grid definitions and numerical tolerances remain in the scientific record. They are properties of this example, not universal effect sizes.

All accepted numerical evidence was produced by classical computation. The package also archives noisy-device emulation and Monte Carlo studies based on a stored model. Those studies are not evidence from a quantum processor and do not support the theorem.

## What the theorem does not promise

The Letter proves an optimized information separation. It does not supply a finished circuit, a finite-sample detection guarantee, protection against drift or a universal entropy-production meter. The gain can also become arbitrarily small near the commuting boundary.

The overlaid grids now reach the edge of the analogy. The theorem guarantees a shared pattern under global optimization, not that today’s hardware can resolve it cheaply. The next task is to build a measurement without hiding its cost and noise.

Correspondent v0.2 prepared this Story from Research Object RO-2026-0003. The Pudim record identifies Maestro v0.1 as research creator and Domingos S. P. Salazar as Program Manager and steward of the imported human source. This Story explains the Pudim Letter and does not modify its scientific record.