Tracially embeddable strategies: Lifting MIP* tricks to MIPco

We prove that any two-party correlation in the commuting operator model can be approximated using a tracially embeddable strategy, a class of strategies defined on a finite tracial von Neumann algebra, which we define in this paper. Using this characterization, we show that any approximately synchronous correlation can be approximated to the average of a collection of synchronous correlations in the commuting operator model. This generalizes the result from Vidick [JMP 2022] which only applies to finite-dimensional quantum correlations. As a corollary, we show that the quantum tensor code test from Ji et al. [FOCS 2022] follows the soundness property even under the general commuting operator model. Furthermore, we extend the state-dependent norm variant of the Gowers-Hatami theorem to finite von Neumann algebras. Combined with the aforementioned characterization, this enables us to lift many known results about robust self-testing for non-local games to the commuting operator model, including a sample efficient finite-dimensional EPR testing for the commuting operator strategies. We believe that, in addition to the contribution from this paper, this class of strategies can be helpful for further understanding non-local games in the infinite-dimensional setting.