MATH.13:11 - SoTA-Echoing

The selected approach defines the structure and input-output action before using symmetry. It supports direct algebra for a small problem and provides the premise for more specialized group, optimization or numerical constructions.

Bronstein, Bruna, Cohen and Veličković, Geometric Deep Learning, draft chapter 3, §§3.1-3.2 develops symmetries as invertible structure-preserving maps and distinguishes invariant and equivariant outputs. Adopt that explicit action and output discipline. Whether a transformation preserves a label or target still comes from the modeled task. Architecture construction and learning-performance claims need the corresponding further Methods and evidence.

Tong, Classical Dynamics, §2.4 derives conserved quantities from continuous symmetries of a Lagrangian using its equations of motion. The useful contribution is the extra argument connecting symmetry with time evolution. The simple oscillator calculation above performs that connection directly; it does not substitute for the wider Noether construction.

Hairer, Geometric Numerical Integration, lecture 2, §1 supplies the symplectic Euler formulas and their Hamiltonian conditions. The direct comparison above shows why a requested invariant must be examined under the actual numerical update. A different model or requested accuracy can favor a different scheme.

C.29.1 supplies the general result-transfer comparison. The fixed-point and selection constructions here make one specific consequence available without a full group-theory survey. Revisit the use when its structure, output meaning, uniqueness premise or transformation law changes.