We will relate and compare our method with theirs in the Results section (Sec. VI). In particular, we show that our procedure is advantageous in the nonasymptotic regime. This paper is organized as follows. We start in Sec. II by briefly reviewing classical and quantum correlations. Then we explain how to obtain the optimal Bell inequality from the observed probability distribution. We lay the framework to provide a confidence interval for the Bell expectation value in Sec. III. We provide an implementable DIQKD protocol in Sec. IV and calculate the finite-size secret key rate in Sec. V. In Sec. VI, we illustrate our method with several examples.
II. GENERAL FRAMEWORKIn this section, we review the concept of the classical correlation polytope in Sec. II A and, based on this, we explain in Sec. II B how to construct Bell inequalities that are maximally violated by the measurement data.
A. Set of correlationsConsider a set-up for two parties 1 (namely, Alice and Bob) connected by a quantum channel. The parties perform local measurements on a joint quantum state. Let us assume that Alice and Bob have m a and m b measurement settings, respectively. Alice's set of measurement settings is denoted as X = {1, • • • , m a }, and Bob's set of measurement settings as Y = {1, • • • , m b }. To estimate the probability distribution from the experimental data, we have to use the measurement device N times in succession. We assume that the devices behave independently and identically (i.i.d.) in each round, i.e. the results of the i th round are independent of the past 1 Note that our method can be extended in a straightforward way to n parties.