Theorem:

The function is Convex Function over the domain of positive definite matrices, , where is the set of symmetric positive definite matrices.

Proof Outline

Step 1: Domain and Feasibility

  • The determinant is positive for all .
  • is well-defined and finite over .
  • is thus a valid function on .

Step 2: First and Second Derivatives

To show convexity, we compute the Hessian of and verify that it is positive semidefinite.

  1. Gradient of : Using matrix calculus:

  2. Hessian of : The gradient is a matrix-valued function. Differentiating with respect to gives the Hessian:

    for any symmetric perturbation matrix .

Step 3: Positive Semidefiniteness

  • The quadratic form of the Hessian is:
  • Since is positive definite (as ) and is symmetric, the term is also symmetric. The trace of a square of a symmetric matrix is non-negative:
  • Thus, the Hessian is positive semidefinite.

Step 4: Convexity

The positive semidefiniteness of the Hessian implies that is convex over .

Intuition

  1. The determinant measures the volume of the parallelepiped defined by the rows or columns of . Minimizing encourages to “spread out” the volume.

  2. The function grows rapidly as approaches singularity (where ), discouraging near-singular configurations.