Control of precision.

[gabrielfougeron]

• The user-provided quantities should scale with the precision of the result, not the time-step itself. =>
  • What is precision ?
  • How does it scale with the time step ?
  • Is there a way to define the condition number of the system or of a subsystem ?
  • Investigate splitting methods. Are they adapted to the context ?
• Find a theoretical treatment of variable time-steps ( that allows conservation of simplecticity for Gauss method in particular ). This should be independant of the procedure choosing the timestep ( investigated in the above ).
• Implement the bitches.

[gabrielfougeron]

To this point (in our work), Accuracy is mainly measured through the conservation of energy ... It is plotted at each write, but what should be plotted is the relative change in energy.