Quantum dynamics can be described in two pictures. In the Schrödinger picture, states evolve and observables stay fixed. In the Heisenberg picture, observables evolve and states stay fixed. The two pictures are mathematically equivalent — they predict identical measurement outcomes for every possible experiment. This equivalence is among the first things taught in graduate quantum mechanics. It is exact, not approximate.
Settimo and colleagues (PRX Quantum, 2026) show that the equivalence breaks when you ask about memory.
A quantum process has memory when its future depends on its past in a way that can't be captured by its present state alone. The standard formalization is divisibility: a memoryless process can be split into independent steps, each depending only on the current state. A process with memory can't be split this way — the steps are correlated.
The two pictures agree on every measurable quantity. But they disagree on whether a given process is divisible. A quantum channel can be divisible in the Schrödinger picture — appearing memoryless when you track how states evolve — while being indivisible in the Heisenberg picture, where the evolution of observables reveals correlations between time steps. The same physical process, described by two provably equivalent frameworks, has memory in one and lacks it in the other.
The resolution: divisibility is a property of the description, not the system. The two pictures decompose the dynamics differently. What looks like independent steps in one decomposition looks like correlated steps in the other. Neither is wrong. The measurement outcomes are identical. But the structural claim — “this process is memoryless” — is framework-dependent.
The general pattern: equivalence between descriptions is weaker than it appears. Two frameworks can agree on every empirical prediction and still disagree about qualitative properties of the system. The disagreement isn't about what happens — it's about what the happening means. Memory, in this case, isn't a fact about the quantum system. It's a fact about which mathematical objects you chose to track.