The Signal
Huang, Yang and Zhou Prove Four Settings Needed for Two-Qubit Moments
A paper on arXiv establishes that four orthogonal projective measurement settings are necessary and sufficient to estimate nine collective first and second moments of two qubits without bias for arbitrary input states. Its numerical four-se
Quantum.world Team · Written with AI · fact-checked ·
Dimension counting stops at three
A basic count gives a lower bound of three measurement settings for estimating the collective first and second moments of two qubits. A structural argument in a paper submitted to arXiv on October 27 rules out every three-setting arrangement, including global measurement bases that mix the system’s symmetric and antisymmetric sectors. The authors then give a four-setting construction, establishing the minimum for this task.
Yibin Huang, Muchun Yang and D. L. Zhou study how many orthogonal projective measurement settings are needed to estimate multiple observables without bias for arbitrary input states. Each setting applies a unitary rotation before measurement in the computational basis; outcomes from a setting can contribute to estimates of several observables. The paper’s two-qubit example targets nine quantities: three collective first moments and six second moments.
Four bases cover all nine moments
The lower bound turns on the observables’ symmetry. They commute with the operation that swaps the two qubits, so their space separates into a three-dimensional triplet sector and a one-dimensional singlet sector. Counting dimensions alone gives a lower bound of three settings, but the paper’s structural bound shows that three bases cannot cover all nine target directions. A matching construction uses four mutually unbiased bases in the triplet sector, with the singlet added to each.
“For the collective first and second moments of two qubits, we certify that four projective settings are necessary and sufficient.”
Yibin Huang, Muchun Yang and D. L. Zhou, authors of the paper, write
The authors distinguish the setting count from statistical precision. For the nine two-qubit observables, their numerically optimized four-setting design achieves a Haar-averaged variance about 0.0341% above a lower bound that applies to any number of projective settings within their fixed-allocation, unbiased linear-estimation model. The comparison averages over uniformly sampled pure input states; an individual state can have a different variance.
Four settings without a symmetry prior
The paper contrasts its result with existing permutationally invariant tomography. A standard method using common local spin axes takes six settings, while a previous four-basis approach relies on knowing that the input state is permutation-invariant. Huang, Yang and Zhou say their four-setting construction estimates the same nine moments for arbitrary states in the full four-dimensional two-qubit space, without that prior.
The result is a statement about a defined estimation task, not a universal recipe for two-qubit measurements. It concerns orthogonal projective settings, arbitrary global basis rotations and unbiased estimates of these nine observables. The paper also evaluates precision under ideal measurements; it does not report an experimental demonstration.
Separate tests show where shared measurements help
Beyond the two-qubit proof, the authors test shared measurement bases on constructed observable sets. For one four-qubit example, three settings exactly reconstruct 30 observables. Against the strongest comparison in that example—spectral measurements that reuse data across target eigenbases—the shared design requires about 4.1 times fewer state copies for the same average variance over Haar-random pure states.
That copy-count comparison does not include the circuit cost of implementing arbitrary global transformations, the authors state. In a separate six-qubit simulation, they study 96 observables with finite-depth measurement circuits, where reconstruction error introduces bias. Shared designs perform better over a range of sample budgets, but spectral reuse becomes more efficient for stricter error targets as the shared design approaches its residual-error floor; the comparisons assume ideal gates and readout.
For the two-qubit case, the concrete conclusion is narrow and exact: three settings cannot provide the required unbiased coverage, and four can. The variance result puts the optimized four-setting design close to the paper’s model-specific bound, rather than establishing the same performance under hardware noise or circuit constraints.