[Paper Review] The power of being positive: Robust state estimation made possible by quantum mechanics

Amir Kalev, Charles H. Baldwin+1 more|arXiv (Cornell University)|Nov 4, 2015

Quantum Information and Cryptography16 citations

TL;DR

This paper proposes that rank-$r$ strictly-complete positive operator-valued measures (POVMs), derived from the positivity constraint of quantum density matrices, enable robust quantum state tomography for low-rank states—even under noise. By measuring only a few random orthonormal bases, such POVMs uniquely identify pure or bounded-rank states among all quantum states, and convex optimization yields stable, noise-resilient reconstructions, as confirmed by numerical simulations across various Hilbert space dimensions.

ABSTRACT

We study the problem of quantum-state tomography under the assumption that the state of the system is close to pure. In this context, an efficient measurements that one typically formulates uniquely identify a pure state from within the set of other pure states. In general such measurements are not robust in the presence of measurement noise and other imperfections, and therefore are less practical for tomography. We argue here that state tomography experiments should instead be done using measurements that can distinguish a pure state from {\em any} other quantum state, of any rank. We show that such nontrivial measurements follows from the physical constraint that the density matrix is positive semidefinite and prove that these measurements yield a robust estimation of the state. We assert that one can implement such tomography relatively simply by measuring only a few random orthonormal bases; our conjecture is supported by numerical evidence. These results are generalized for estimation of states close to bounded-rank.

Motivation & Objective

• To address the practical limitation of rank-$r$ complete measurements in noisy quantum state tomography (QST), which fail to distinguish low-rank states from higher-rank physical states.
• To introduce and formalize the concept of rank-$r$ strictly-complete POVMs, which uniquely identify low-rank states within the entire convex set of positive semidefinite matrices.
• To demonstrate that such strictly-complete measurements enable robust state estimation via convex optimization, even under statistical noise and experimental imperfections.
• To provide numerical evidence that measuring only a small number of random orthonormal bases yields strictly-complete POVMs for low-rank states, with weak dimension dependence.

Proposed method

• Define rank-$r$ strictly-complete POVMs as those that uniquely distinguish any rank-$\leq r$ state from all other positive semidefinite matrices, leveraging the positivity (positive semidefiniteness) of density operators.
• Prove that any strictly-complete POVM ensures robust state estimation by solving any convex optimization program over the feasible set of positive semidefinite matrices.
• Use convex relaxation techniques to estimate quantum states from measurement data, including trace minimization, least-squares, and maximum-likelihood estimation.
• Simulate realistic QST scenarios with noisy data by generating mixed states close to pure states ($q=10^{-3}$) and sampling measurement outcomes from $m=300d$ trials per basis.
• Numerically test the informational completeness of random orthonormal bases by checking whether the measurement record uniquely identifies the target state among all PSD matrices.
• Apply numerical optimization to reconstruct states and evaluate fidelity, confirming robustness across different convex programs and Hilbert space dimensions.

Experimental results

Research questions

• RQ1 Can rank-$r$ strictly-complete POVMs guarantee robust quantum state estimation in the presence of measurement noise and experimental imperfections?
• RQ2 Do measurements in a small number of random orthonormal bases yield rank-$r$ strictly-complete POVMs for low-rank quantum states?
• RQ3 How does the number of required measurement bases scale with Hilbert space dimension and rank for strict-completeness?
• RQ4 Can convex optimization methods like trace minimization, least squares, and maximum-likelihood estimation reliably reconstruct low-rank states when using strictly-complete POVMs?
• RQ5 What is the empirical threshold of bases needed to achieve strict-completeness for pure and low-rank states in both qudit and qubit systems?

Key findings

• Measuring only a few random orthonormal bases—as few as 5 for $d=21$—can yield rank-1 strictly-complete POVMs, with only one exceptional state failing to be uniquely identifiable.
• For $d=11, 21, 31$, a small number of random bases (e.g., 6–8) suffices to achieve rank-1 strict-completeness, with weak dependence on Hilbert space dimension.
• The estimation of pure or low-rank states remains robust under noise: all tested convex programs (trace minimization, least squares, maximum-likelihood) achieve low average infidelity (<10^{-3}) when the measurement POVM is strictly-complete.
• The difference in required bases between rank-1 and rank-2 strict-completeness is minimal, indicating efficient scalability for states close to pure.
• Local measurements require more bases than global ones to achieve strict-completeness, as seen in $d=16$ systems with multiple qubits.
• Numerical simulations confirm that strict-completeness enables reliable state reconstruction even when the true state is a small mixture of a pure state and a full-rank state ($q=10^{-3}$), with high fidelity to the target.