Quantum trajectories on a short leash ✛
A quantum state under continuous measurement is not a frozen picture but a random dance. Each measurement outcome nudges the state one way or another, and to predict the future you have to track the entire trajectory. For systems with many levels, this is a computational nightmare: the density matrix grows exponentially with the system size.
But here is the good news. In many cases, the quantum state is on a short leash: noise tugs it around but does not let it stray far from a prescribed surface. This surface is a low-dimensional manifold that can be described by a few variables rather than the full density matrix. The work shows how to find such a leash and how to compute its length.
For quantum nondemolition measurements, harmonic oscillators, and composite systems, explicit formulas are derived. For example, for a qutrit with one nondemolition measurement channel, the dimension is 1: all trajectories lie on a curve. For a three-qubit repetition code with two syndrome measurements, a two-dimensional surface emerges. And for a harmonic oscillator under measurement of the annihilation operator and quadratures, the manifold dimension is 2 or 4 depending on the presence of thermal noise.
This result changes the game for quantum engineering. Previously, it was unclear when stochastic evolution could be reduced to a small number of variables and when the full density matrix was required. Now there is an exact criterion. For a quantum filter, it suffices to store just a few integrals of the measurement signal, not the entire density matrix. This opens the way to efficient controllers and real-time state estimation without an exponential explosion of complexity.
A natural next step is to study approximate confinement on the manifold, when small perturbations slowly blur the structure. Also promising is applying the criterion to Poissonian measurements, where jumps replace Wiener processes. An open question is whether one can deliberately design a system to remain on a low-dimensional manifold, and whether this yields operational advantages.
In one example, when measuring the annihilation operator in a harmonic oscillator without thermal noise, the state remains on a one-dimensional curve, even though the signal carries information about both position and momentum.