Can quantum squeezing beat the standard quantum limit in sensing?

The claim depends on how success is defined

The paper's answer is conditional, and the condition carries the argument. Success is defined as a measurement precision that cannot be achieved using the same number of particles and the same time without entanglement. Under that definition, the author concludes that squeezed states cannot succeed, and that a single non-separable squeezed state provides fundamentally no better precision per unit time than a single particle.

This is a claim about fundamental precision, not about every practical measurement. The paper separates the probabilistic noise of quantum measurement from technical noise sources, and it states that other technical noise should of course be reduced. The theorem addresses the fundamental part only, so it does not say that squeezing has no laboratory use.

Entanglement is proposed to help in two ways

The paper sets out two proposed routes past the standard quantum limit. The first uses entangled NOON, CAT or GHZ states so that the sensor's response to the parameter becomes N-fold greater than a single spin's, while the measurement uncertainty stays at the single-spin level. In effect, the device behaves like one spin with a larger magnetic moment. The author notes that when the resources needed to generate entanglement are fully accounted for, no improvement over the standard quantum limit or the single-particle bound has been demonstrated, and he does not pursue this route further.

The second route uses squeezed states. Here the response to the signal stays the same as for a single spin, while the measurement noise is said to fall by up to a factor of N. That is the claim the paper targets: precision improved by reducing noise rather than by increasing the state's response.

Reducing noise does not add information about the parameter

The argument rests on a standard bound in quantum metrology. For a single non-separable state interacting with a Hamiltonian of the form θ·H for a time t, the quantum Fisher information about θ is bounded by (t·Eδ/ℏ)², where Eδ is the maximum difference between energy eigenvalues of the states making up the superposition. The quantum Fisher information in turn bounds the classical Fisher information of any measurement made on that state, and the Fisher information bounds the mean squared error of any unbiased estimate.

The practical reading is that only the state's response to the parameter raises that bound. A squeezed state, as commonly described, does not have a larger energy gap than a single spin. If two states share the same Eδ, they obey the same information bound, even if one is described as having less noise. The author puts this bluntly: there is no noise in the quantum Fisher information, just as there is no noise in a pure quantum state.

Measurement noise is not the same as uncertainty

A central distinction in the paper is between the variance of a measured observable and the uncertainty in estimating the parameter. The author gives a projective measurement example: a data set of ten identical outcomes has a variance below 1/4, yet that alone does not mean the standard quantum limit has been surpassed, because the uncertainty in the spin direction can be the same as for a state with a more spread-out data set.

The paper also warns that a collective spin variance below |S|/4 is often presented as evidence of enhanced precision, although the uncertainty in estimating the spin direction is not the variance of an operator or of a data set. In the author's account, this conflation is one of the most common analytical mistakes in spin-squeezing work.

Why experiments appear to contradict the theorem

The paper acknowledges that many experimental papers claim a precision violating its theorem, and it attributes the gap to analysis rather than to an error in the proof. One recurring pattern is comparison with a limit that is not the standard quantum limit: an imperfect device limit, or a limit defined as the operating precision of the device as it is, with all its imperfections, where entanglement-based modifications are allowed but other improvements are not.

Other patterns concern how noise and signal are handled. The author describes experiments that reduce measured noise by subtracting one measurement from a similar one, which also removes signal, and experiments operated near a dark fringe where the probability of a meaningful outcome and its derivative with respect to the parameter both shrink. A further pattern is fair sampling: sensitivity is higher when data is observed, but the lower probability of obtaining meaningful data is not fully accounted for.

The proof rests on stated assumptions

The result is a theorem, so its assumptions matter. The paper lists the postulates of quantum mechanics, including the Born rule; the assumption that all dependence on θ is encoded in H(θ,t), with the starting state and measurement operator independent of θ; the assumption that everything outside the initial state vectors is the same for entangled and unentangled ensembles, including prior information and the ability to apply control fields; and the assumption that each copy of a state is independent and measured identically.

The author extends the argument to probability mixtures of squeezed states, to the expected mean information averaged over a prior distribution of θ, and to a modified claim for arbitrary time-dependent Hamiltonians. He also notes that the analysis concerns fundamental probabilistic noise from quantum measurements, and he raises the question of whether the information bound is achievable in practice. If it is not, he argues, that would amount to saying quantum mechanics is incorrect, since the squeezing community claims the limit can be surpassed rather than merely approached.

The decisive comparison is per particle and per unit time

The paper's conclusion is a resource comparison, not a verdict on every squeezed-state experiment. Entanglement reduces the number of independent samples an ensemble can provide, and the author argues that the only way to overcome sampling noise is to take more samples. That is why the relevant comparison is precision per unit time against the best precision achievable with the same number of independent, identical particles.

For a reader weighing a sensing proposal, the paper's priority is therefore to check the baseline before accepting a squeezing advantage: whether the comparison is to the standard quantum limit for the same particle number and time, whether noise and uncertainty have been kept distinct, and whether the probability of obtaining meaningful data has been counted. The paper's own scope is fundamental precision under the stated assumptions; technical noise reduction and practical regimes where the bound is not reached are left outside the theorem.

Sources used in this answer