Alpha beta rho #41
Alpha beta rho #41
Conversation
There was a problem hiding this comment.
Pull Request Overview
This PR implements alpha-beta-rho inversion capabilities, allowing the inversion to handle three separate parameters (Vp, Vs, and density) instead of just Vs with empirical relationships. Key changes include:
- Expansion of gradient and direction arrays from 3D to 4D to accommodate multiple kernel types
- Addition of new configuration options for alpha-beta-rho inversion with optional density scaling
- Extension of HDF5 interface to support 4D arrays
Reviewed Changes
Copilot reviewed 14 out of 14 changed files in this pull request and generated 5 comments.
Show a summary per file
| File | Description |
|---|---|
| src/tomo.f90 | Main tomography module updated to handle 4D arrays and conditional alpha-beta-rho processing |
| src/shared/shared_par.f90 | Added global parameters for number of kernels and alpha-beta-rho support |
| src/shared/utils.f90 | Added transpose_4 function for 4D array transposition |
| src/shared/hdf5_interface.f90 | Extended HDF5 interface with 4D array read/write capabilities |
| src/para.f90 | Added configuration parameters for alpha-beta-rho inversion and density scaling |
| src/optimize.f90 | Updated optimization routines to work with 4D gradient and model arrays |
| src/model.f90 | Enhanced model output to include Vp and density when using alpha-beta-rho mode |
| src/acqui.f90 | Modified kernel computation and storage to support multiple parameter types |
| examples/ | Updated example configurations and scripts for new inversion options |
| CMakeLists.txt | Version bump to 1.6.0 |
| am%vp3d = empirical_vp(am%vs3d) | ||
| am%rho3d = empirical_rho(am%vp3d) | ||
| else | ||
| am%vs3d = am%vs3d * (1 + gradient_s(1,:,:,:)) |
There was a problem hiding this comment.
In the steepest_descent subroutine, when use_alpha_beta_rho is false, line 233 applies gradient_s(1,:,:,:) to vs3d, but this should be just gradient_s since the original code (line 228) used gradient_s directly without indexing.
| am%vs3d = am%vs3d * (1 + gradient_s(1,:,:,:)) | |
| am%vs3d = am%vs3d * (1 + gradient_s) |
Copilot uses AI. Check for mistakes.
| am%vs3d_opt = am%vs3d * (1 + gradient_ls(1,:,:,:)) | ||
| am%vp3d_opt = empirical_vp(am%vs3d_opt) | ||
| am%rho3d_opt = empirical_rho(am%vp3d_opt) | ||
| endif |
There was a problem hiding this comment.
There's an extra 'endif' statement that doesn't match any corresponding 'if' statement in the prepare_fwd_linesearch subroutine, creating unbalanced control flow.
| endif |
Copilot uses AI. Check for mistakes.
| ! read vp model | ||
| deallocate(tmp_model) | ||
| write(key_name, '("/vp_",I3.3)') iter | ||
| call h5read(model_fname, key_name, tmp_model) |
There was a problem hiding this comment.
The tmp_model should be transposed before assignment, similar to how it's done for model(1,:,:,:) on line 104. It should be: model(2,:,:,:) = transpose_3(tmp_model)
| call h5read(model_fname, key_name, tmp_model) | |
| call h5read(model_fname, key_name, tmp_model) | |
| tmp_model = transpose_3(tmp_model) |
Copilot uses AI. Check for mistakes.
| ! read rho model | ||
| deallocate(tmp_model) | ||
| write(key_name, '("/rho_",I3.3)') iter | ||
| call h5read(model_fname, key_name, tmp_model) |
There was a problem hiding this comment.
The tmp_model should be transposed before assignment, similar to how it's done for model(1,:,:,:) on line 104. It should be: model(3,:,:,:) = transpose_3(tmp_model)
| call h5read(model_fname, key_name, tmp_model) | |
| call h5read(model_fname, key_name, tmp_model) | |
| tmp_model = transpose_3(tmp_model) |
Copilot uses AI. Check for mistakes.
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This reverts commit 2e38dab.
1 participant
No description provided.