The Signal
Harvard researchers say quantum sensing can surpass Heisenberg limit
A preprint by three Harvard researchers says quantum computational spectroscopy can exceed the Heisenberg limit for broadband signals, even under constant-strength sensor noise. It also sets out three pulse-sequence methods for conventional spin or qubit platforms.

A proposed route beyond the usual limit
A quantum processor could allow a sensor to beat the Heisenberg sensitivity limit for broadband signals, even when the sensor experiences constant-strength noise, according to a new preprint. The claim comes from Ishaan Kannan, Aditi Venkatesh and Jordan Cotler, researchers in Harvard University’s Department of Physics. They posted “Algorithmic Quantum Sensing” to arXiv on November 19, 2026.
Although the work is a preprint, it sets out theoretical limits and algorithms rather than reporting a new hardware demonstration. The authors frame time-varying-field sensing as a statistical learning problem: estimate selected properties of a signal’s frequency spectrum to a chosen precision. Their framework covers tasks ranging from identifying a signal at one of many frequencies to estimating nonlinear properties, such as correlations among frequencies.
Three methods for conventional sensors
The paper presents three pulse-sequence approaches for conventional spin or qubit platforms: randomized, matched and pooled spectroscopy. Randomized quantum spectroscopy uses randomly chosen pulse sequences to build a classical dataset that can later be reused to estimate different signal properties. The authors say it can learn many observables simultaneously and reach the AC Heisenberg limit, the benchmark they derive for conventional sensing tasks.
Matched spectroscopy instead tailors pulse sequences to observables selected in advance, avoiding the need to collect a general-purpose dataset. The paper says that when only a few observables are wanted, this approach can reduce interrogation time polynomially compared with nonadaptive protocols. Pooled spectroscopy targets coarse-grained quantities, such as total signal power across a frequency band, by using noncommuting pulse sequences to process signal information before readout.
The authors compare these methods with dynamical decoupling, a widely used pulse-sequence approach, and report advantages across tasks including broadband search, lock-in detection and multidimensional spectroscopy. Those performance comparisons are claims made in the preprint; they are not presented as results from a new experiment.
Adding quantum computation
Quantum computational spectroscopy, or QCS, couples the sensor to a quantum processor. The paper says versions of the method can be implemented with one memory ancilla. Its quantum computational frequency comb coherently converts a classical signal into a quantum oracle; algorithms then query that oracle in superposition to extract information about the signal.
For broadband signals, the authors give a sensitivity scaling proportional to the square root of the number of frequencies and inversely proportional to the target precision. They describe this as a generalized Heisenberg limit that surpasses the conventional AC Heisenberg limit, with the improvement requiring ancillary quantum computation. Their analysis also places a bound on the gain: they say no protocol permitted by quantum mechanics can achieve better sensitivity for the tasks they study.
Noise and the stated limits
The preprint’s central noise claim is that QCS retains beyond-Heisenberg scaling under constant-strength sensor depolarization, a form of noise aligned with the signal. By contrast, the authors say the same noise forces protocols without ancillary quantum computation to standard-quantum-limit scaling. They present this as a theoretical result for broadband AC sensing, not a measured performance figure from a deployed sensor.
The paper also distinguishes the new method from improvements based on entangled sensors or coherent interrogation: its proposed advantage comes from processing signal information in superposition. The authors say larger processors could estimate more quantities at once, while some QCS variations require only one memory ancilla. The work’s conclusions concern the algorithms and limits developed in the preprint.
Sources
- Original source · Preprint Algorithmic Quantum Sensing arXiv:2611.25572
Related stories
- Researchers map energy levels in a silicon quantum-dot pair
- Diamond spin spirals sustain collective dynamics, paper reports
- Photonic experiment encodes optical phase in three-bit output
- PRX Quantum paper reports homodyne readout of spin information
- DLR researchers report shallower circuits for quantum data loading