1. 14 Nov, 2016 1 commit
    • Lukas Riedel's avatar
      major folder restructuring. dorie-rfg remains untouched · 6977b959
      Lukas Riedel authored
      * defined top-level headers for easy file includes
      * moved previous header files into 'solver' subfolder
      * moved Simulation headers and implementations into 'interface' subfolder
      * moved library instantiations into 'impl' subfolder
      6977b959
  2. 11 Nov, 2016 3 commits
  3. 10 Nov, 2016 1 commit
  4. 09 Nov, 2016 5 commits
  5. 08 Nov, 2016 9 commits
  6. 06 Nov, 2016 6 commits
  7. 04 Nov, 2016 4 commits
  8. 03 Nov, 2016 5 commits
    • Lukas Riedel's avatar
      started work on new Traits struct · d2b3b9ec
      Lukas Riedel authored
      d2b3b9ec
    • Lukas Riedel's avatar
    • Lukas Riedel's avatar
      1691c526
    • Lukas Riedel's avatar
      Merge branch 'debug/improve-evaporation-BC' into 'master' · 7ae6d3ef
      Lukas Riedel authored
      Improve evaporation BC switch
      
      I improved the evaporation BC by changing the condition for which the applied BC switches from Neumann to Dirichlet BC: For any quadrature point with evaporation BC, the numeric flux is calculated once per time step. If this numeric flux is smaller than the flux applied by the Neumann BC, the Dirichlet BC is applied. This smoothes the transition between the two BCs.
      
      Tested improvements:
      * The head value given in the evaporation BC is not a mere cutoff value but the target boundary condition if the Neumann flux cannot be sustained
      * Evaporation from homogeneous surfaces is very stable and reaches high time step values
      * The new condition allows for a smooth transition back from Dirichlet to Neumann if the soil regains water saturation (e.g., by a rising water table)
      
      Undecided:
      * Evaporation from a Miller regime is still not very stable. This is due to the fact that the Dirichlet BC itself is not very stable for a Miller regime.
      * Higher FEM orders can reduce the numeric stability
      * Numeric flux is not a physical flux and depends on the `dg.penaltyFactor`.
      
      Bad:
      * The code for the switch check is copypasta from the Dirichlet BC.
      
      See merge request !8
      7ae6d3ef
    • Lukas Riedel's avatar
  9. 02 Nov, 2016 3 commits
  10. 01 Nov, 2016 1 commit
  11. 29 Oct, 2016 1 commit
  12. 28 Oct, 2016 1 commit
    • Dion Häfner's avatar
      Merge branch 'debug/time-step-adjustment' into 'master' · c1ed2d7c
      Dion Häfner authored
      time steps are now adjusted properly to BC changes
      
      The time step adjustment until now has stepped to time `changeTime`+`dtmin`, with `changeTime` being the time stamp at which the BC changes. This led to an overestimation of the new BC because the time step scheme does some kind of (linear?) interpolation of the residual from start to finish of one time step.
      
      As a workaround, the time step is now adjusted to `changeTime-eps`. This way, the new BC is not considered at the end of the time step. After a successful computation, the time is adjusted to `changeTime`, i.e., Dorie advances the time by `eps` without any computation. At the beginning of the new time step only the new BC is considered.
      
      See merge request !7
      c1ed2d7c