[GSPH 5/7] Solver integration #1527
Conversation
|
Thanks @Guo-astro for opening this PR! You can do multiple things directly here: Once the workflow completes a message will appear displaying informations related to the run. Also the PR gets automatically reviewed by gemini, you can: |
Summary of ChangesHello @Guo-astro, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request introduces a significant architectural change to the GSPH solver by integrating a modular Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a significant and well-executed refactoring to integrate modular physics modes into the GSPH solver using a strategy pattern. This greatly improves the solver's architecture, making it more extensible and maintainable. The changes to decouple physics-specific logic from the core solver and configuration are excellent.
I've identified a couple of critical issues, including some leftover debugging code that needs to be removed and a potential buffer overflow in the VTK dump module. I've also made several suggestions to restore documentation that was removed during the refactoring to maintain code clarity. Overall, this is a very strong contribution to the codebase.
Guo-astro
changed the title
Guo-astro
force-pushed
the
pr/gsph-solver-integration
branch
from
b781b0e to
2e61244
Compare
|
/gemini review |
There was a problem hiding this comment.
Code Review
This pull request introduces a major and well-designed refactoring of the GSPH solver to support multiple physics modes using a strategy pattern. The code is now significantly more modular and extensible. The changes successfully decouple physics-specific logic from the core solver and I/O from the physics implementation, which is a great improvement for maintainability.
My review has identified a critical issue with state management when switching between physics modes that could lead to incorrect behavior or crashes. I have also provided several medium-severity suggestions to improve robustness, maintainability, and documentation, such as using dynamic_cast for type checking instead of strings, restoring removed documentation comments, and extending the new Python data collection utility. Additionally, I've requested clarification on a significant change in tolerance values to ensure it's intentional and well-understood.
Guo-astro
force-pushed
the
pr/gsph-solver-integration
branch
from
45e892a to
e51d1f4
Compare
Core Infrastructure: - PhysicsMode base class for strategy pattern implementation - ForceKernelBase for common force computation interface - PhysicsModeFactory for creating physics mode instances - FieldNames.hpp as SSOT for field naming Modular Components: - BoundaryHandler: boundary condition processing - BuildTrees: tree construction - ComputeCFL: CFL timestep calculation - ComputeGradients: gradient computation - ComputeOmega: omega factor computation - GhostCommunicator: MPI ghost communication - IterateSmoothingLengthVolume: h-iteration - NeighbourCache: caching neighbor interactions - FunctorNode: generic functor nodes Refactors the monolithic GSPH Solver into modular components, preparing for physics mode decoupling (Newtonian vs SR).
The ForceKernelBase template class was designed as a Template Method pattern base but was never used - NewtonianForceKernel and SRForceKernel are standalone implementations with their own buffer management appropriate for their physics. The hardcoded CommonBuffers design (buf_density, buf_pressure, etc.) does not accommodate SR physics which requires distinct lab-frame vs rest-frame field naming (N_LABFRAME, LORENTZ_FACTOR, ENTHALPY, etc.).
These fields have physics-specific meanings: - Newtonian: single frame quantities - SR: lab-frame vs rest-frame distinction Each physics mode now defines its own field constants with appropriate semantic names in their respective FieldNames headers.
Headers: - NewtonianConfig: solver configuration for Newtonian hydro - NewtonianEOS: ideal gas equation of state interface - NewtonianFieldNames: field naming constants - NewtonianForceKernel: force computation interface - NewtonianMode: PhysicsMode implementation for Newtonian hydro - NewtonianTimestepper: CFL-based timestepping - forces.hpp: force math functions - ReconstructConfig: interface reconstruction settings - RiemannConfig: Riemann solver configuration Riemann Solvers: - RiemannBase: abstract Riemann solver interface - HLL: Harten-Lax-van Leer approximate Riemann solver - Iterative: exact iterative Riemann solver Sources: - Complete implementations for all Newtonian components This implements the Newtonian physics mode for GSPH using the strategy pattern for physics decoupling.
Energy fields have physics-specific meanings and are now defined directly in NewtonianFieldNames.hpp rather than imported from the common FieldNames.hpp.
Headers: - SRConfig: solver configuration for SR hydro - SREOS: relativistic equation of state interface - SRFieldNames: field naming constants for SR fields - SRForceKernel: force computation interface for SR - SRMode: PhysicsMode implementation for SR hydro - SRPrimitiveRecovery: conservative to primitive variable recovery - SRTimestepper: relativistic CFL-based timestepping - forces.hpp: SR force math functions Sources: - SREOS.cpp: relativistic EOS implementation - SRForceKernel.cpp: SR force kernel implementation - SRPrimitiveRecovery.cpp: Newton-Raphson primitive recovery - SRTimestepper.cpp: SR timestep computation This implements the core SR physics components for GSPH.
SR physics defines its own XYZ, VXYZ, AXYZ, UINT, DUINT constants directly rather than importing from the common FieldNames.hpp. This clarifies the physical meaning (lab-frame quantities for SR).
Recovery Methods: - RecoveryBase: abstract interface for primitive recovery - NewtonRaphson: Newton-Raphson iterative recovery algorithm Riemann Solvers: - RiemannBase: abstract SR Riemann solver interface - Exact: exact relativistic Riemann solver (Pons 2000) Mode Implementation: - SRMode.cpp: complete SR physics mode orchestration - Ghost field setup for SR variables - Primitive recovery from conserved variables - Force computation with relativistic corrections - Integration of SR equations This completes the Special Relativistic GSPH implementation following Kitajima 2024 formulation.
Solver Changes: - Solver.hpp/cpp: Refactored to use PhysicsMode strategy pattern - SolverConfig.hpp/cpp: Updated config for physics mode selection - Model.hpp/cpp: Updated model registration Storage & IO: - SolverStorage.hpp: Updated field storage for physics modes - VTKDump.hpp/cpp: Physics-aware VTK output Python Bindings: - pyGSPHModel.cpp: Extended bindings for SR configuration - physics_mode selection (newtonian/sr) - SR-specific parameters (gamma, initial conditions) Build System: - CMakeLists.txt: Updated for new physics module structure This integrates the modular physics modes into the GSPH solver, enabling runtime selection between Newtonian and SR physics.
Newtonian Tests: - sod_tube_gsph.py: Sod shock tube validation - blast_wave_gsph.py: Extreme blast wave test SR Tests (Kitajima 2024 benchmark suite): - problem1_sod.py: Relativistic Sod shock tube - problem2_blast.py: Relativistic blast wave - problem3_strong_blast.py: Strong relativistic blast - problem4_ultra_relativistic.py: Ultra-relativistic regime - problem5_tangent_velocity.py: Tangential velocity test - problem6_2d_sod.py: 2D relativistic Sod tube - problem7_kh_instability.py: Kelvin-Helmholtz instability Common: - sr/__init__.py: SR test utilities - kitajima_plotting.py: Plotting helpers for Kitajima benchmarks All tests use: - ctx.collect_data() for direct memory access (no pyvista) - Strict tolerances (~1e-8) for regression testing - Analytic solutions for validation
Unit Tests: - GSPHForceTests.cpp: Update for new physics structure - GSPHRiemannTests.cpp: Update Riemann solver tests SPH Module Fixes: - IterateSmoothingLengthDensity: Improve h-iteration logging - BasicSPHGhosts.cpp: Fix ghost handling Math: - sphkernels.hpp: Minor kernel fixes MHD Placeholder: - MHDConfig.hpp: Placeholder for future MHD physics mode
- NewtonianMode: add compute_omega_newtonian() using standard SPH (no c_smooth) - SRMode: use SRIterateSmoothingLength with Kitajima volume-based approach - Remove shared ComputeOmega module (each mode now owns this) - Remove legacy UpdateDerivs (replaced by NewtonianForceKernel) - Move IterateSmoothingLengthVolume to physics/sr/SRIterateSmoothingLength - Update CMakeLists.txt sources This fixes the density/pressure error regression caused by c_smooth=1.2 being incorrectly applied to Newtonian mode.
…ult (1.0) SR's volume-based h-iteration (Kitajima Eq. 232-233) requires c_smooth > 1 to smooth h variation across discontinuities. The SR-specific value was defined in SRConfig but not transferred to the shared config.
The Riemann solver and force computation code has been moved to the physics/newtonian/ and physics/sr/ directories. The math/ folder contained duplicate/orphaned code that is no longer used.
Remove deprecated include of math/riemann/iterative.hpp (now deleted) and update hllc_solver call to use solve_hll from the new location in physics/newtonian/riemann/HLL.hpp.
Guo-astro
force-pushed
the
pr/gsph-solver-integration
branch
from
47b7945 to
a9a20d7
Compare
Run buildbot/update_authors.py to add --no git blame-- annotations to author headers as required by CI checks.
The kernel summation ρ = ν × ΣW gives the density based on particle positions in the lab frame. For moving particles, the true lab-frame baryon density is N = γ × ρ due to Lorentz contraction. Without this fix, pressure was wrong by factor ~1/γ for particles with non-zero tangent velocity (e.g., P=436 instead of P=1000 for v_t=0.9). Co-Authored-By: Claude <noreply@anthropic.com>
Remove SolverCallbacks struct and have PhysicsMode evaluate nodes directly via storage.solver_graph. This aligns with solvergraph design: - Branching happens at init time (node registration) - Flow is visible as graph structure - No runtime callback creation Changes: - Register Solver method nodes in init_solver_graph() - Delete SolverCallbacks struct from PhysicsMode.hpp - Update evolve_timestep() signature (remove callbacks param) - NewtonianMode/SRMode evaluate nodes directly via storage.solver_graph Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2 participants
Summary
Integrates the modular physics modes into the GSPH solver.
Solver Changes
Solver.hpp/cpp: Refactored to use PhysicsMode strategy patternSolverConfig.hpp/cpp: Updated config for physics mode selectionModel.hpp/cpp: Updated model registrationStorage & IO
SolverStorage.hpp: Updated field storage for physics modesVTKDump.hpp/cpp: Physics-aware VTK outputPython Bindings
pyGSPHModel.cpp: Extended bindings for SR configurationphysics_modeselection (newtonian/sr)Build System
CMakeLists.txt: Updated for new physics module structureUsage
Dependencies
Depends on #1524 (Newtonian physics) and #1526 (SR physics)