Pure-State Tomography: Sample-Optimal Methods

• The paper establishes protocols that achieve the theoretical limits of pure-state tomography by saturating the Massar–Popescu bound with collective measurements.
• It details efficient sequential weak measurement strategies and optimal POVM constructions, enabling sample-optimal state reconstruction under various measurement models.
• The work compares methodologies, including classical shadows and oracle-based techniques, demonstrating quadratic improvements and optimal scaling in sample complexity.

Sample-optimal pure-state tomography concerns achieving the minimal possible number of quantum samples (state copies or oracle calls) required to reconstruct or characterize an unknown pure quantum state to a specified accuracy, under various measurement models. This topic encompasses both the fundamental limits—defined precisely by information-theoretic bounds—and explicit, efficiently implementable protocols that saturate these limits. Theoretical and algorithmic developments in this area illuminate the optimal strategies for both full state reconstruction (in ℓ_q-norm or fidelity) and compressed representations such as classical shadows for observable estimation.

1. Theoretical Limits: Fidelity and Sample Complexity

For pure-state qubit tomography with access to $N$ identical copies of an unknown $|\psi\rangle$, the maximal achievable average fidelity of any collective measurement is given by the Massar–Popescu (MP) bound:

$F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$

where the average infidelity $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$. This bound is tight and applies to any collective, symmetric measurement on the total spin-$J=N/2$ subspace. Any sample-optimal protocol must saturate this bound (Shojaee et al., 2018).

For general $d$-dimensional systems and observables $O$ with $\mathrm{Tr}(O^2)\leq B$ and $\|O\|\leq 1$, the minimal sample complexity for estimating $\mathrm{Tr}(O\rho)$ of a pure state $|\psi\rangle$0 within error $|\psi\rangle$1 is (Grier et al., 2022):

• Joint measurement: $|\psi\rangle$2
• Independent measurement: $|\psi\rangle$3

Full state reconstruction in $|\psi\rangle$4-norm to error $|\psi\rangle$5 requires $|\psi\rangle$6 applications of a state-preparation unitary and its inverse when available (Apeldoorn et al., 2022).

2. The Spin-Coherent-State POVM and Optimal Measurements

For ensembles of qubits, the optimal POVM for pure-state tomography is the spin-coherent-state (SCS) POVM:

$|\psi\rangle$7

where $|\psi\rangle$8 is the spin-coherent state pointing in direction $|\psi\rangle$9, obtained by rotating the Dicke state $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$0. This POVM is overcomplete on the symmetric subspace and is uniquely optimal: any other POVM on the symmetric subspace achieves average fidelity at most equal to the MP bound, with equality if and only if each $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$1 (Shojaee et al., 2018).

Estimators that maximize fidelity choose the output Bloch vector as

$F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$2

where $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$3 is the total spin operator.

3. Sequential Weak Measurement Protocols

The SCS-POVM, despite its theoretical optimality, is not directly physical. However, Shojaee et al. showed that it is operationally realizable via sequential weak measurements of the collective spin in randomized directions (Shojaee et al., 2018).

Protocol Structure

• Each weak measurement in direction $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$4 applies the Kraus operator:

$F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$5

• Squeezing (due to the quadratic term) is canceled by averaging blocks of measurements over isotropically sampled directions.
• In the limit of many weak steps, the block evolution is governed by:

$F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$6

• As the total time $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$7 increases, Kraus operators concentrate the measurement outcome onto a single spin-coherent state:

$F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$8

The POVM element after time $F_N^{\mathrm{opt}} = \frac{N+1}{N+2}$9 is $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$0, which strongly projects onto $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$1 as $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$2 grows.

Scaling

The number of required weak measurement steps (or total measurement time $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$3) to concentrate onto a rank-1 SCS projector at infidelity $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$4 is $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$5, i.e., the protocol is polylogarithmic in $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$6 for sample complexity (Shojaee et al., 2018).

4. Sample-Optimal Classical Shadows for Pure States

Classical shadows are randomized measurement protocols whose classical output suffices for the estimation of arbitrary observables after state preparation. For pure states, the sample-optimal regime displays quadratic improvement over mixed states for joint measurements (Grier et al., 2022).

Joint Measurement Protocol

• Measurement: symmetric subspace POVM on $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$7 copies,

$1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$8

with $1-F_N^{\mathrm{opt}} = 1/(N+2) = O(1/N)$9.

• Estimator:

$J=N/2$0

• For observable $J=N/2$1 with $J=N/2$2, $J=N/2$3,

$J=N/2$4

Sample Complexity

The minimal sample complexity is

$J=N/2$5

for joint measurements, optimal up to logarithmic factors in $J=N/2$6 and $J=N/2$7. Independent measurement protocols achieve $J=N/2$8 (Grier et al., 2022).

Lower Bounds

Matching lower bounds are established via reductions to the Boolean Hidden Matching problem (for the $J=N/2$9 term) and two-state distinguishability (for the $d$0 term), as well as Holevo’s theorem on information capacity.

5. State-Preparation-Oracular Tomography

In scenarios where a unitary oracle $d$1 prepares the pure state $d$2 and $d$3 is available, optimal tomography can be achieved with minimal quantum queries (Apeldoorn et al., 2022).

• $d$4-norm estimation to error $d$5 is achieved with $d$6 calls to $d$7 and $d$8 for $d$9, saturating the quantum lower bound.
• The algorithm first performs an $O$0-approximation of the amplitude vector $O$1 (via multidimensional phase estimation and block encoding), then recovers phases, and finally converts the estimate to a $O$2 proxy without additional dimension-dependent factors.
• For higher precision ($O$3), the query cost is $O$4, with a lower bound also given by reduction to phase-oracle recovery problems.

6. Practical Implementation Considerations

Implementations in atomic-ensemble settings (e.g., Faraday/QND probing in high-cooperativity cavities) achieve $O$5 with negligible decoherence, supporting feasibility of the sequential weak measurement protocol (Shojaee et al., 2018). The protocol is robust to deviations from perfect isotropy, with errors only at subleading $O$6 corrections. Both discrete (randomized axis) and continuous (fixed measurement axis combined with rapid random Euler rotations) protocols realize isotropic POVMs.

For classical shadow protocols, random Clifford measurements (3-design POVMs) suffice for independent measurements, but optimal joint measurement performance requires symmetric subspace measurements. The quadratic improvement in $O$7 for pure states is not attainable by the original Huang–Kueng–Preskill approach for mixed states (Grier et al., 2022).

7. Comparative Summary of Sample Complexities

The following table summarizes key sample complexity scalings for pure-state tomography protocols.

Task/Protocol	Model	Sample Complexity	Reference
Qubit pure-state fidelity	Collective	$O$8 infidelity ($O$9 copies)	(Shojaee et al., 2018)
Observable estimation	Joint measurement	$\mathrm{Tr}(O^2)\leq B$0	(Grier et al., 2022)
Observable estimation	Independent	$\mathrm{Tr}(O^2)\leq B$1	(Grier et al., 2022)
Full state (ℓ₂-norm)	Unitary oracle	$\mathrm{Tr}(O^2)\leq B$2	(Apeldoorn et al., 2022)

These results establish the operational and information-theoretic limits for pure-state tomography across several practical settings, demonstrating that variants of the SCS-POVM, classical-shadow protocols, and oracle-based methods all admit sample-optimal strategies, with explicit constructions and matching lower bounds.