Validation of onset-of-controllability hyperparameter selection

Verify whether restricting a hyperparameter search to circuits at the onset of controllability, selecting the most performant configurations, and subsequently adding layers yields the best-performing quantum models while reducing computational cost.

Background

The experiments found a strong correlation between validation losses for circuits at the onset of controllability and those for circuits with additional layers. The paper proposes using this correlation to reduce hyperparameter-search cost, but explicitly leaves the proposed procedure as a conjecture requiring further experiments.

References

These results hint at a possible technique for reducing the computational costs of hyperparameter studies while still obtaining the best performing circuits. First, one performs the hyperparameter search fixed at the onset of controllability, 15 layers in our experiments, then restrict the hyperparameter search to the space spanned by the most performant circuits with 15 layers, and then add additional layers to these selected models before further training. Further experiments would need to be conducted to verify this conjecture.

— Subspace Controllability in Variational Quantum Circuits: Maximising Expressivity and Increasing Search Efficiency with Dynamical Lie Algebras  (2609.24475 - Barlow et al., 21 Sep 2026) in Section 4, “Near controllable circuits predict later winners”