Rake-Compress Riccati Recursions for Parallel Scenario-Tree Model Predictive Control

What the math gives to ML

The paper provides an exact rake–compress contraction for quadratic dynamic programs on arbitrary rooted trees, replacing the sequential tree-height dependency of Riccati/message-passing recursions with O(log N) span while preserving O(N) work. The transferable asset is not MPC-specific control, but the representation of a subtree by local quadratic value data and the use of Schur-complement eliminations as associative composition operations. A promising neural-network use is a differentiable Gaussian or quadratic latent tree layer whose upward and downward passes are contracted in parallel, enabling irregular hierarchical computation graphs without sequential depth proportional to tree height. This should be tested against ordinary sequential tree message passing for wall-clock latency, gradient correctness, and predictive quality at fixed work.