Near-optimal parallel M-matrix scaling

A new poly-logarithmic iteration algorithm solves the scaling of symmetric M-matrices to high accuracy.

Researchers addressed the problem of computing a positive diagonal scaling for a d x d symmetric M-matrix. This process ensures all row and column sums equal 1.

The provided poly-logarithmic iteration algorithm solves this problem to high accuracy. The cost of each iteration is dominated by solving a Laplacian system.

This method improves upon the recent O~(d1/3)\tilde{O}(d^{1/3}) iteration bound of [Vladu STOC 2025].

The work yields a randomized algorithm for the problem. This algorithm runs in near-linear time and poly-logarithmic parallel depth.

Key facts

  • The algorithm computes a positive diagonal scaling for a d x d symmetric M-matrix.
  • The iteration cost is dominated by solving a Laplacian system.
  • The randomized algorithm runs in near-linear time and poly-logarithmic parallel depth.

All sections

Edition