A parameterization of positive definite matrices in terms of partial correlation vines (Q1410721): Difference between revisions

This paper extends a method of the authors for studying positive definiteness by interpreting a symmetric matrix as a potential correlation matrix, and using their theory of partial correlations with a new theory called vines. Vines are sequences of trees such that the nodes of each tree are the edges of the previous tree; their edges are associated with partial correlations. Positive definiteness can be determined by computing partial correlations on a regular vine and seeing if they are in \((-1,1)\). This also provides a method of adjusting non-positive definite matrices so that they become positive definite.