Abstract
We study the relation between semi and fully device independent protocols. As a tool, we use the correspondence between Bell inequalities and dimension witnesses. We present a method for converting the former into the latter and vice versa. This relation provides us with interesting results for both scenarios. First, we find new random number generation protocols with higher bit rates for both the semi and fully device independent cases. As a byproduct, we obtain whole new classes of Bell inequalities and dimension witnesses. Then, we show how optimization methods used in studies on Bell inequalities can be adopted for dimension witnesses.
Introduction: The Trust Spectrum in Quantum Communication
In quantum communication, security is often defined by the level of trust placed in the hardware. This exists on a spectrum ranging from Device-Independent (DI) to Semi-Device-Independent (SDI) protocols. DI protocols require no trust in the internal workings of devices, relying instead on the violation of Bell inequalities. However, they are notoriously difficult to implement and often yield low bit rates. SDI protocols offer a middle ground; they assume a physical constraint, such as a limited dimension of the communicated quantum system (e.g., a single qubit), while remaining agnostic about the device's internal mechanisms.
Historically, these two frameworks developed in parallel without a clear mathematical bridge. This paper addresses that gap by establishing a formal correspondence between Bell inequalities and dimension witnesses, allowing breakthroughs in one regime to be translated into the other.
Methods: Bridging the Gap via Mathematical Mapping
The authors introduce a mathematical mapping that treats Alice’s measurement outcomes in a DI setup as part of her state-preparation settings in an SDI setup. This duality allows a Bell inequality to be transformed into a "dimension witness."
This conversion was applied to SDI protocols based on Quantum Random Access Codes (QRACs), deriving a new class of Bell inequalities ($I_n$) that scale with code complexity. Furthermore, the authors introduced a "lifting" technique to model SDI state preparation as a measurement on a maximally entangled state. This allows researchers to apply the NPA hierarchy—a powerful tool from the DI world—to find global bounds for SDI entropy using Semidefinite Programming (SDP) optimization.
Key Results and Numerical Findings
The established correspondence yielded significant improvements in randomness generation rates. The new DI-derived protocols allow the randomness rate to grow with complexity $n$, theoretically approaching a limit of 2 bits per round as $n \to \infty$. Numerical simulations using the SeDuMi and Yalmip MATLAB toolboxes provided concrete evidence of these gains.
As shown in the results, at $n=5$, the authors achieved a DI min-entropy of 1.4652, a substantial increase over previous standards. Beyond higher bit rates, the SDP method provided more rigorous and "honest" lower bounds for existing protocols, such as the $2 \to 1$ QRAC, compared to previous heuristic local search methods.
Limitations and Future Outlook
While the conversion method is highly effective, it remains heuristic; there is no universal guarantee that every Bell inequality will produce a useful dimension witness. Additionally, for complexities where $n > 2$, the second level of the SDP hierarchy did not reach the theoretical maximum, leaving the convergence of higher levels as an open question. Future research is required to evaluate the performance of these protocols under real-world conditions involving noise and detector inefficiencies.
Summary updated