1: \begin{abstract}

2: 

3: Many modern datasets exhibit dependencies among observations as well as

4: variables. This gives rise to the challenging problem of analyzing

5: high-dimensional matrix-variate data with unknown dependence structures.  To

6: address this challenge, Kalaitzis et. al. (2013) proposed the Bigraphical

7: Lasso (BiGLasso), an estimator for precision matrices of matrix-normals based

8: on the Cartesian product of graphs. Subsequently, Greenewald, Zhou and Hero

9: (GZH 2019) introduced a multiway tensor generalization of the BiGLasso

10: estimator, known as  the TeraLasso estimator.  In this paper, we provide

11: sharper rates of convergence in the Frobenius and operator norm for both

12: BiGLasso and TeraLasso estimators for estimating inverse covariance matrices.

13: This improves upon the rates presented in GZH 2019. In particular, (a) we

14: strengthen the bounds for the relative errors in the operator and Frobenius

15: norm by a factor of approximately $\log p$; (b) Crucially, this improvement

16: allows for finite-sample estimation errors in both norms to be derived for

17: the two-way Kronecker sum model.  The two-way regime is important because it

18: is the setting that is the most theoretically challenging, and simultaneously

19: the most common in applications. Normality is not needed in our proofs;

20: instead, we consider sub-gaussian ensembles and derive tight concentration of

21: measure bounds, using tensor unfolding techniques. The proof techniques may

22: be of independent interest.

23: \end{abstract}

24: