Symmetry reduction in Sumsofsquares

Specific Domains Optimization (Mathematical)
, , ,
18

This is due to internal reformulation of the problem done by JuMP. SCS only solves minimization problems (in particular format), so the model we input is bridged internally to fit that description.

To answer the other questions:

I’m afraid Wedderburn decomposition will not be of much help for DSOS, as linear programs can be simplified by change of variables only. You should probably

  1. compute symmetry_adapted_basis (with semisimple=false, i.e. the isotypic projections) and create one variable for each vector here.
  2. decompose the set of constraints into orbits and create one constraint (the average) per orbit
  3. transform the new constraints to the new basis. profit :wink:

I’m not sure about SOC program, I’d need to think a bit. You should of course try the naive version of simplifying the SDP formulation of the problem :smiley:

1 Like