The power of CGLMP inequalities

Cite this problem as Problem 27.


In the setting of [[Bell inequalities holding for all quantum states|Problem 26]], consider especially the case (N,M,K)=(2,2,d).

Problem 27. A

Show that every face of the local polytope C, which is not already contained in a face of the no-signalling polytope P is of CGLMP type, i.e., an inequality of the form first written out in [1], but possibly lifted from lower dimensions by fusing together some outcomes.

Problem 27. B

Numerically, the observables maximally violating the CGLMP inequality on a maximally entangled state are of a very specific form [2], involving measurements in computational basis, transformed by only discrete Fourier transformation and diagonal unitaries [1]. Show that this is necessarily the case. Show also that these measurements realize the highest resistance of violation to noise, and the best discrimination against classical realism in the sense of Kullback-Leibler divergence [3].


According to the setting (N,M,K)=(2,2,d), the CGLMP inequality features two parties, X and Y, with two observables each: X_1, X_2 and Y_1, Y_2, respectively. Each observable has d possible outcomes. In order to simplify notation, we use the function m(x) = x \mod d where m(x)\in \{0,1,...,d-1\} for integer x and we denote expectation values by \mathsf{E}. The inequality can then be written [4]

\mathsf{E}(m(X_1-Y_1)) + \mathsf{E}(m(Y_1-X_2)) + \mathsf{E}(m(X_2-Y_2)) + \mathsf{E}(m(Y_2-X_1-1)) \ge d-1.

This statement also suggests a very elegant proof of the inequality [4]:

Note that  (X_1-Y_1) + (Y_1-X_2) + (X_2-Y_2) + (Y_2-X_1-1) = -1. Apply the function m to both sides, and use m(a)+m(b)+m(c)+m(d) \ge m(a+b+c+d).

A more detailed discussion of the problem can be found in [5].

Partial Results

The CGLMP inequality is indeed a facet of the local polytope for the case (N,M,K) = (2,2,d), see [6].

In [7], using a semidefinite programming relaxation, it is shown that the maximum violation of the CGLMP inequality for d=3 on maximally entangled states of arbitrarily high dimensions differs at most by 10^{-10} from the local maximum found in [2] for maximally entangled states of dimension 3. This proves, up to computer precision, the optimality of the observables identified in [2].

It has been noticed numerically that, for d=3,4,5,6,7,8, one can achieve higher violations of the CGLMP inequality using the same measurements and the eigenvector corresponding with minimum eigenvalue of the corresponding Bell operator [8]. In [9], it was shown that the maximum violation of CGLMP achievable by quantum systems coincides (up to computer precision) with the quantum values identified in [8].


[1] D. Collins, N. Gisin, N. Linden, S. Massar, and S. Popescu, Phys. Rev. Lett, 88, 040404 (2002).

[2] T. Durt, D. Kaszlikowski, and M. Zukowski, Phys. Rev. A, 64, 024101 (2001).

[3] W. van Dam, P. Grunwald, and R. Gill, quant-ph/0307125 (2003).

[4] R. Gill, private communication.

[5] A. Acin, R. Gill, and N. Gisin, Phys. Rev. Lett. 95, 210402 (2005).

[6] Ll. Masanes, Quant. Inf. Comp. 3, 345 (2002).

[7] B. Lang, T. Vértesi and M. Navascués, J. Phys. A 47, 424029 (2014).

[8] A. Acín, T. Durt, N. Gisin and J. I. Latorre, Phys. Rev. A 65, 052325 (2002).

[9] M. Navascués, S. Pironio and A. Acín, Phys. Rev. Lett. 98, 010401 (2007).