Robust self-testing of nonmaximal entanglement from a reduced inner product game

Technical Summary: Robust Self-Testing of Nonmaximal Entanglement from a Reduced Inner Product Game

Problem Statement
The paper addresses a fundamental question in quantum nonlocality: whether a maximally entangled state (one with equal nonzero Schmidt coefficients) is sufficient for perfect play in all pseudo-telepathy games. While it is known that maximally entangled states do not always maximize violations of Bell inequalities, their sufficiency for perfect winning strategies in finite-dimensional games was an open problem. Lalonde previously resolved this by constructing an inner product game ($G_{4,3,6,6}$) where a specific nonmaximally entangled state wins perfectly, but no maximally entangled state of any finite dimension can. However, Lalonde's game required six dimensions. This work seeks to reduce the dimensionality of this counterexample while preserving its self-testing properties—specifically, the ability to uniquely determine the shared state and all measurement effects up to local isometries.

Methodology
The author constructs a five-dimensional reduction of Lalonde's game, denoted $G_{4,3,5,5}$ (or $G_5$). The reduction involves:

1. Dimensionality Reduction: Removing one of the two low-amplitude coordinates from the original six-dimensional strategy.
2. Outcome Compression: Combining measurement outcomes that coincide after compression, reducing the answer alphabet for both Alice and Bob from six to five.
3. Game Definition: The game retains four questions for Alice ($x \in \{0,1,2,3\}$) and three for Bob ($y \in \{0,1,2\}$). The winning condition is defined by a non-zero inner product involving a diagonal matrix $D = \text{diag}(2,2,2,2,1)$ and specific vectors derived from the compressed bases.

The proof strategy relies on analyzing the "forbidden" outcomes (those with zero probability in the ideal strategy) to reconstruct the state structure. The core technical steps include:

• Coarse-Grained Labeling: Using Bob's first measurement ($y=0$) to establish five orthogonal subspaces. Alice uses a sequence of coarse projective measurements ($x=0$ and $x=1$) to coherently record the same five labels.
• Amplitude Ratio Identification: By grouping outcomes and analyzing binary comparisons (specifically at questions $x \in \{2,3\}$ and $y \in \{1,2\}$), the author derives relations that force the fifth amplitude to be exactly half the magnitude of the other four equal amplitudes.
• Extraction Maps: The author constructs explicit local isometries (unitary transformations controlled by recorded labels) that map the players' physical states to a reference state $|\psi^\star\rangle$ and a residual system.
• Robustness Analysis:
  • State Robustness: The author estimates the distance between the extracted state and the target state directly on unnormalized branch vectors, avoiding assumptions about lower bounds on outcome probabilities.
  • Measurement Robustness: To handle measurement effects, the author extends the exact identification to commuting-operator representations on a universal algebra. By utilizing a compactness argument on the space of states with vanishing loss, they derive a dimension-independent modulus of continuity for the measurement actions.

Key Contributions and Results

1. Reduced Game Construction ($G_5$): The paper presents a game with classical value $11/12$ and quantum value $1$. The reference state is:
   $$|\psi^\star\rangle = \frac{2|00\rangle + 2|11\rangle + 2|22\rangle + 2|33\rangle + |44\rangle}{\sqrt{17}}$$
   This state has a non-uniform Schmidt spectrum (ratios of 4:1 for squared coefficients).

2. Robust State Self-Testing (Theorem 2.1): The paper proves that any strategy winning with probability $1-\epsilon$ allows for the extraction of a state with squared fidelity $\Xi$ such that the extraction distance $d_{ext} \le \min\{1, 221\sqrt{\epsilon}\}$. This provides a quantitative bound on how close the physical state must be to $|\psi^\star\rangle$.

3. Exact Self-Testing of State and Measurements (Theorem 2.2): At perfect score ($\epsilon=0$), the strategy is shown to be a full self-test. There exist local isometries such that the shared state is mapped to $|\psi^\star\rangle \otimes \chi$ (where $\chi$ is a residual vector), and every labeled measurement effect ($A_{xa}, B_{yb}$) is mapped to the ideal rank-one projections. This holds for arbitrary mixed states and POVMs.

4. Qualitative Robustness of Measurements (Theorem 2.3): The paper establishes that the measurement actions are also robust. For any $\epsilon$, there exists a function $r(\epsilon)$ (independent of local dimensions) such that the extracted measurement effects are within distance $r(\epsilon)$ of the ideal effects. While the paper does not provide an explicit rate for $r(\epsilon)$, it proves the existence of a uniform modulus of convergence.

5. Impossibility of Maximal Entanglement (Corollary 2.5): As a direct consequence of the self-test, the paper proves that no maximally entangled state of any finite dimension can win $G_5$ perfectly. This is because local isometries preserve the Schmidt spectrum, and a flat spectrum (maximal entanglement) cannot be transformed into the specific non-flat spectrum of $|\psi^\star\rangle$.

Significance and Claims
The paper claims to resolve the sufficiency question for maximal entanglement in a more compact setting than Lalonde's original work. By reducing the dimension from six to five, it demonstrates that the rigidity required to certify nonmaximal entanglement persists even with fewer resources.

The author emphasizes that their proof connects two lines of research: robust self-testing of nonmaximal entanglement (previously limited to specific Bell inequalities or states with optimal probabilities $<1$) and pseudo-telepathy games. The key novelty lies in using the zero-probability constraints of a single fixed game to determine both the unequal Schmidt coefficients and every labeled measurement effect simultaneously.

The paper remains modest regarding quantitative improvements, noting that the constant 221 in the state bound is not optimized and that the measurement robustness bound is qualitative (existence of a modulus without an explicit rate). The author identifies obtaining useful quantitative bounds for all measurement actions under the same isometries as a natural next step, rather than claiming to have solved it. They also explicitly state that no minimality claim is made regarding the number of questions, answers, or dimensions.