Could architectural changes enforcing dimensional analysis or symmetry constraints reduce the gap on research-scale problems?

Your question asks whether enforcing dimensional analysis or symmetry constraints at the architectural level could narrow the performance gap on research-scale problems. This is a thoughtful question about problem design, but the preprints provided are almost entirely about quantum optimization (QUBO, annealing, phase estimation) rather than classical dimensional analysis or symmetry-driven architecture for general research problems.

That said, there is one directly relevant insight:

Symmetry constraints do measurably improve performance. In quantum phase estimation for chemistry, exploiting symmetries in the Hamiltonian's block structure yields smaller spectral bounds than naive approaches [AG-2024.03-1561]. This matters because the spectral gap directly determines how many quantum operations you need; tighter bounds mean fewer gates and lower error. The authors show this advantage is numerically traceable and scales predictably in limiting regimes—evidence that symmetry awareness is not merely elegant but computationally concrete.

Similarly, in quantum annealing, the architecture itself can be reshaped around symmetry: a scalable 2-local (nearest-neighbor) hardware design based on triangular encodings of the problem graph preserves the connectivity needed for arbitrary Ising models [AG-2024.03-1561], and careful embedding of parity constraints maintains physical properties of the original problem [AG-2024.05-2370]. Both trades reduced hardware overhead for systematic constraint-respecting structure.

The gap narrowing question remains open for general research-scale classical problems. The archive's focus is quantum/optimization-specific; it doesn't address whether dimensional analysis or symmetry architectures close performance gaps in, say, numerical PDEs, machine learning, or materials simulation more broadly.