Open
Conversation
added 12 commits
February 25, 2026 22:24
* Pass parameter Tg to the matrix-less overload of member 'evaluate'
of the CAR and elastic operators. At present, these are not needed
(only when g is used to assemble the convection and diffusion
coefficients these are needed); this prepares for a new discretization
strategy where the flux terms are considered as an ensemble, which is
needed to apply a single flux boundary condition.
* Pass parameter E/N to the field operator's 'evaluate' member. At present,
the operator always assumes E/N=1 [Vm2], and the factor (E/N)^2 needs to
be applied when the result is used (power evaluation, adding the matrix
to the Boltzmann matrix). This change makes this practice more explicit
(passing 1 instead of EoN where appropriate), and allows a more straighforward
implementation of code that wants to apply the factor (E/N)^2 right from the
start. As an example of the latter, test_nonuniform_field.cpp and
test_nonuniform_druyvesteyn.cpp looked like:
fieldOperator1.evaluate(grid1, fieldCrossSection, won, CIEff, M1);
M1 *= eon*eon;
fieldOperator2.evaluate(grid2, fieldCrossSection, won, CIEff, M2);
M2 *= eon*eon;
and are now simply implemented as
fieldOperator1.evaluate(grid1, fieldCrossSection, eon, won, CIEff, M1);
fieldOperator2.evaluate(grid2, fieldCrossSection, eon, won, CIEff, M2);
Comments have been added at call sites where E/N=1 is passed.
* This manages the convection and diffusion coefficients of an operator. * Derive ElasticOperator and FieldOperator from this class * Equip those classes with member updateCD, which updates the convection and diffusion coefficients. (These are related to the pre-existing members evaluate.) * Note that the coefficients have their actual (non-zero) values also at the upper grid boundary (unike member g). At u=0, the values are really zero.
This declares a class ConvectionDiffusionDiscretizer, to which ConvectionDiffusionOperator objects can be registered. The class provides members for aggregating the convection and diffusion coefficients, calculating the grid Peclet numbers and for discretizing one of the registered contributions, are the combination. Bot the central difference scheme and the Scharfetter-Gummel (exponential) scheme are supported. Note that for the latter scheme, the individual contributions are coupled via the grid Peclet number, which is calculated for the combination of all registered operators (see the cpp_notes document). The test discretizes the same elastic+field problem for all available discretization strategies and write the results to files f_xxx.dat. These files contain three columns: the energy, the EEDF and the Druyvesteyn EEDF (calculated for the mean energy that is calculated from the EEDF). Since Tgas=0 in the examples (this can be changed in the .cpp file), this is the expected solution. For the present settings, the exponential scheme is much better able to reproduce the analytical solution than the cd scheme (and the 'old' scheme, the CD scheme presently available in LoKI-B). This can be shown by running the example, starting gnuplot, then doing: > set logscale y > plot 'f_old.dat' u 1:2 w l, '' u 1:3 w l, 'f_cd.dat' u 1:2 w l, '' u 1:3 w l, 'f_sg.dat' u 1:2 w l, '' u 1:3 w l, 'f_sg_sep.dat' u 1:2 w l, '' u 1:3 w l The two overlapping 'low' curves are the calculated EEDFS for the CD scheme (old and new implementation). The others (all overlapping) are the results for the SG scheme, the sum-of-contributions version of the SG scheme, and the analytical (Druyvesteyn) results.
* add functions for evaluating the partial and complete flux * implement the usage of a ghost cell ot obtain the true fluxes (partial) at the upper grid boundary. See #define LOKI_DISCRETIZE_BOUNDARY_FLUX in the code * add more tests
Move these functions outside the Schemes namespace.
And call that for Schemes::CD and Schemes::SG.
Note: There is no test for this function yet.
janvdijk
force-pushed
the
rework_div_flux_operators
branch
from
b7c5462 to
346e855
Compare
added 8 commits
February 26, 2026 20:41
Fixed assert(coefs.A==std::numeric_limits<double>::quiet_NaN()). NaN comparisons always fail, even nan==Nan. Use std::isnan(coefs.A) instead. This fixes the debug builds.
This includes the classes ConvectionDiffusionOperator, ConvectionDiffusionTerms, the CD and SG schemes and function templates discretize_dflux_du and evaluate_flux_density.
Add explicit instantiations of function templates for Schemes::CD and Schemes::SG.
Consistently handle calc_coefs for boundary faces, use the results also for boundary faces (the contribution may simply be zero). This results in a more symmetric implementation.
* In the Operators' updateCD members, do *not* apply the scaling factor 1/gamma. * in discretize_dflux_du: do just that, do not multiply with -1. This means that the resulting discretization matrices are positive semi-definite, instead of negative semi-definite. * Comments have been added that make this choice clear.
Use camelCase for function names.
Collaborator
Author
|
@daanboer, as discussed this morning, IMHO this is now ready to be merged. This change does not change the existing behavior of the code. A review would be appreciated. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Firstly, this changes the matrix-less overloads of the evaluate members of the operators to take the parameters that are also needed by the versions that accept a matrix reference. While not used in all cases, this will simplify the implementation of new discretization strategies where we need the actual convection/diffusion coefficients, rather than scaled ones (e.g. assuming E/N=1). It adds comments to locations where the scaled field operator is calculated and it is assumed that the factor (E/N)^2 is applied afterwards.
Secondly, this adds test_cd_disc.cpp.
This declares a class ConvectionDiffusionDiscretizer, to which ConvectionDiffusionOperator objects can be registered. The class provides members for aggregating the convection and diffusion coefficients, calculating the grid Peclet numbers and for discretizing one of the registered contributions, are the combination. Both the central difference scheme and the Scharfetter-Gummel (exponential) scheme are supported. Note that for the latter scheme, the individual contributions are coupled via the grid Peclet number, which is calculated for the combination of all registered operators (see the cpp_notes document).
The test discretizes the same elastic+field problem for all available discretization strategies and write the results to files f_xxx.dat. These files contain three columns: the energy, the EEDF and the Druyvesteyn EEDF (calculated for the mean energy that is calculated from the EEDF). Since Tgas=0 in the examples (this can be changed in the .cpp file), this is the expected solution.
For the present settings, the exponential scheme is much better able to reproduce the analytical solution than the cd scheme (and the 'old' scheme, the CD scheme presently available in LoKI-B). This can be shown by running the example, starting gnuplot, then doing:
The two overlapping 'low' curves are the calculated EEDFS for the CD scheme (old and new implementation). The others (all overlapping) are the results for the SG scheme, the sum-of-contributions version of the SG scheme, and
the analytical (Druyvesteyn) results.