There is an extraordinary and telling moment in Raven Leilani's acclaimed novel Luster, about a young black woman who has an affair with a middle-aged white man and ends up living with his family. The woman, Edie, is heading back to her lover's house with his adopted black daughter, Akila, when the pair are stopped and questioned by two police officers. Although Edie is compliant, Akila – younger and much less worldly – challenges the cops and gets thrust to the ground and restrained. The confrontation is rife with fear and tension, and when it's over (diffused when Akila's white mother intervenes), the first thing Edie and Akila do is go inside, sit down and play a video game. Much of the fervid discussion around Luster has focused on Leilani's astute and witty analysis of sexual politics and racial power structures in the 21st-century US.

We present a novel nonnegative tensor decomposition method, called Legendre decomposition, which factorizes an input tensor into a multiplicative combination of parameters. Thanks to the well-developed theory of information geometry, the reconstructed tensor is unique and always minimizes the KL divergence from an input tensor. We empirically show that Legendre decomposition can more accurately reconstruct tensors than other nonnegative tensor decomposition methods.