Skip to content

Numerical improvements#232

Draft
daanboer wants to merge 41 commits intomainfrom
better-integrals
Draft

Numerical improvements#232
daanboer wants to merge 41 commits intomainfrom
better-integrals

Conversation

@daanboer
Copy link
Member

@daanboer daanboer commented Aug 12, 2025 •
edited
Loading

This patch aims to implement a multitude of numerical improvements. I have opened this draft PR to track the progress, I might split the PR into multiple smaller PRs later down the line.

  • Analytical evaluation of the collision integrals
    • by assuming logarithmic decay of the EEDF between cells (similar to the strategy of BOLSIG+),
    • by assuming linear decay of the EEDF between cells (in theory less potent than the previous strategy, with the upside that it does not require iteration).
    • These strategies should be less sensitive to alignment of the threshold region of inelastic processes to the grid, but this is still to be tested thoroughly.
  • Scharfetter-Gummel discretization for the flux terms.
  • A total flux boundary condition (there is also some experimentation with a logarithmic decay boundary condition).
  • A Newton-Raphson based solving procedure for the nonlinear growth + ee terms.
  • The implementation supports nonuniform discretization.

Resolves #197
Relates to #205

@daanboer daanboer added the enhancement New feature or request label Aug 12, 2025
@daanboer daanboer mentioned this pull request Aug 12, 2025
Open
@daanboer daanboer mentioned this pull request Oct 1, 2025
Closed
@daanboer daanboer force-pushed the better-integrals branch 3 times, most recently from 9adb19d to f54e8a1 Compare December 5, 2025 12:53
janvdijk
janvdijk reviewed Feb 7, 2026
View reviewed changes
janvdijk
janvdijk reviewed Feb 7, 2026
View reviewed changes
janvdijk
janvdijk reviewed Feb 7, 2026
View reviewed changes
@daanboer daanboer force-pushed the better-integrals branch 3 times, most recently from b14e760 to 36f0f46 Compare February 16, 2026 08:12
@daanboer
Add a function to more accurately compute inelastic collision integrals
951136a 
The function computes an analytic solution to the u*sig*f collision integrals,
assuming sig to be locally linearly dependent on u, and f to the locally
exponentially dependent on u.
@daanboer
Add sink and source function for inelastic integrals
f930289 
Adds functions `collision_integral_sink` and `collision_integral_source` that
do the piecewise integration of the inelastic source terms.
@daanboer
Add temporary invertLinearMatrixNew function
4b1daba 
This function uses the new method of computing the inelastic
collision integrals. The solution now requires iteration, hence
the new function. The nonconservative terms have been temporarily
disabled in `invertLinearMatrix`, to allow for comparison between
the two functions.
@daanboer
Remove unused import
397535a
@daanboer
Initialize u_f_start to the first cell energy
46a9559
@daanboer
Add operator classes that provide separate drift and diff coeffs
50e746e
@daanboer
Add advanced inelastic sink and source functions
041dd83 
These functions directly interpolate the given cross section
from the raw data (skipping an interpolation onto the grid
faces). These functions are a bit more involved, as the
cross section grid needs to be taken into account when
splitting up the energy domain for integration.
@daanboer
Disable not yet supported sections of code
d093d87 
Done to compare the original solution to the solution
proposed on this branch. This currently includes growth
schemes, ionization, attachment and e-e collisions.
@daanboer
Use a Scharfetter-Gummel discretization for the drift-diffusion typ…
4e5a60c 
…e operators
@daanboer
Add alternative inelastic integration functions
185e2b3
@daanboer
Fix drift and diffusion coefficients of the elastic operator
9204b68
@daanboer
Correctly increment cell iterator at cell boundaries
9ee4818
@daanboer
Skip processes that have thresholds above the maximum grid energy
809927d
@daanboer
Only normalize the EEDF after the final iteration
3c595c4
@daanboer
Use a Maxwellian as the initial condition
630b795
@daanboer
Add a central difference implementation
a4c518b 
To compare to the Scharfetter-Gummel scheme.
@daanboer
Add a more realistic constant exponential decay boundary condition
9052961
@daanboer
Clip cross sections to zero when extrapolating
7568988
@daanboer
Initial implementation of superelastic collisions
219c670 
There is a bug in the source term.
@daanboer
Fix indexing error in superelastic sink term
1c10046
@daanboer
Use the mean relative error instead of the sum
98f3b30
@daanboer
Fix superelastic sink term
aeddf78 
Change perspective of u_start to upper energy.
@daanboer
Cleanup log messages
d8a333c
@daanboer
Increase tolerance to 1e-4
6554054
@daanboer
Move updated inelastic operator to its own class
14ca6c8
@daanboer
Add conservative ionization operator
ff17586
@daanboer
Implement the equalSharing ionization operator type
20be883
@daanboer
Add alternative method for integrating the mobility
d78fd4d 
This is used as a test case for integrating different quantities.
@daanboer
Add skeleton for spatial growth operator
b3f9423 
Currently it only computes the mobility in two different ways.
@daanboer
Add integration iterator and integrator machinery
445c35b 
The idea is that using this machinery we can easily build expressions that
fold the EEDF with other quantities, while also easily being able to switch
between the different interpolating expressions for the EEDF.
@daanboer
Add new collision integral integration functions
9f189cc 
These functions use the new iterator/integrator machinery.
@daanboer
Fix GridIterator not advancing when it should
2bc7b8b 
The grid iterator sometimes got stuck when integrating the superelastic
source on (very) large grids. This is fixed by adding a small epsilon
in the lhs of the predicate.
@daanboer
Use the new integration procedures in the inelastic operators
d3508eb 
The procedures now use a linear integrator.
@daanboer
Add missing break statements in switch
11f8c61
@daanboer
Use user-defined ionization operator type
cb1e3f7
@daanboer
[TMP] Patch N2 cross sections
120ae48 
The new `InterpolatingIterator` machinery only work for cross
sections that start their energy domain at zero energy for now.
@daanboer
Remove deprecated collision integration routines
46148d6
@daanboer
Compute spatial growth operator terms
60d6513
@daanboer
Add Newton-Raphson and Picard routines for spatial growth
f664242
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@janvdijk janvdijk janvdijk left review comments

Assignees

No one assigned

Labels

enhancement New feature or request

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Implement a single flux-boundary condition

2 participants

@daanboer @janvdijk