Fix #2483: Add field-based restart loading with missing field initialization #2675

guptapratykshh · 2026-01-03T14:09:56Z

Proposed Changes

Adds field-based lookup for restart files, allowing SU2 to restart when required fields are missing (e.g., enabling the energy equation or turbulence model after an initial run without them). Missing fields are initialized from config defaults (e.g., temperature from INC_TEMPERATURE_INIT), with warnings. Maintains backward compatibility with standard restart files.
Key Implementation:

Added FindFieldIndex() helper function to CSolver base class for field-based lookup
Refactored CFVMFlowSolverBase::LoadRestart() to use field-based lookup instead of fixed offsets
Updated CTurbSolver::LoadRestart() to handle missing turbulence variables
Supports both incompressible and compressible flows, as well as turbulence models

Related Work

Resolves #2483

Issue Description:

SU2 crashed when restarting a simulation with restart files that did not contain all necessary fields. For example, running an incompressible flow simulation without the energy equation, then enabling the energy equation, or enabling a turbulence model after an initial run without it.

Modified 4 core solver files to implement field-based restart loading
Added error handling and user warnings
Maintains full backward compatibility with existing restart files

PR Checklist

I am submitting my contribution to the develop branch.
My contribution generates no new compiler warnings (try with --warnlevel=3 when using meson).
My contribution is commented and consistent with SU2 style (https://su2code.github.io/docs_v7/Style-Guide/).
I used the pre-commit hook to prevent dirty commits and used pre-commit run --all to format old commits.
I have added a test case that demonstrates my contribution, if necessary.
I have updated appropriate documentation (Tutorials, Docs Page, config_template.cpp), if necessary.

pcarruscag · 2026-01-03T18:47:31Z

SU2_CFD/include/solvers/CFVMFlowSolverBase.inl

+          /*--- Load Pressure (index 0) ---*/
+          if (fieldIdx_Pressure >= 0) {
+            solutionBuffer[0] = Restart_Data[baseIndex + fieldIdx_Pressure];
+          } else {
+            /*--- Use default from config ---*/
+            solutionBuffer[0] = config->GetPressure_FreeStreamND();
+            if (rank == MASTER_NODE && counter == 0) {
+              cout << "WARNING: Pressure field not found in restart file, using config default." << endl;
+            }
+          }


Does the solver already have the default initialization when we load the restart?
If so, can we use the solution that is already in the solver nodes instead of repeating the logic for initialization?

If there is physics-specific logic required (incompressible, compressible, NEMO) it should be implemented in the respective solvers using e.g. a virtual function or static polymorphism based on the ENUM_REGIME template parameter of this class, instead of these long if/else blocks.

The names of the variables need to be in sync with those used for output. Therefore, they should be defined as constants and used in both places (output class and here).

Thanks for the review. I have updated implementation:

Defaults: Verified that nodes are pre-initialized. I removed the redundant default initialization logic, missing fields now simply preserve the existing values in the solver nodes.

Polymorphism: Refactored LoadRestartSolutionFields to use if constexpr on the ENUM_REGIME template parameter, ensuring clean static separation of physics logic.

Constants: Introduced a RestartFieldNames namespace to ensure field names are consistent and defined in one place.

…initialization

… defaults

bigfooted · 2026-01-04T12:49:33Z

Nice! Can you add a testcase to the regressions (Testcases/parallel_regression.py + new config file), for instance one of the flat plate SA testcases or some testcase that already uses a restart file, and restart with SST instead?

…ssing fields

guptapratykshh · 2026-01-04T16:06:26Z

Added the requested regression test (turb_SST_flatplate_restart) to parallel_regression.py , which restarts an SA solution using the SST model. This verifies that missing turbulence fields are correctly initialized from defaults.

pcarruscag · 2026-01-04T16:23:21Z

SU2_CFD/include/solvers/CFVMFlowSolverBase.inl

-          SurfaceViscCoeff.CEff[iMarker_Monitoring] = SurfaceViscCoeff.CL[iMarker_Monitoring] / (SurfaceViscCoeff.CD[iMarker_Monitoring] + EPS);
+          SurfaceViscCoeff.CEff[iMarker_Monitoring] =
+              SurfaceViscCoeff.CL[iMarker_Monitoring] / (SurfaceViscCoeff.CD[iMarker_Monitoring] + EPS);


revert all formatting changes to this file

pcarruscag · 2026-01-04T16:24:28Z

TestCases/parallel_regression.py

+    turb_SST_flatplate_restart.cfg_dir = "rans/flatplate"
+    turb_SST_flatplate_restart.cfg_file = "turb_SST_flatplate_restart.cfg"
+    turb_SST_flatplate_restart.test_iter = 10
+    turb_SST_flatplate_restart.test_vals = [] 


This cannot possibly test correctness if it doesn't have values

…orrectness

Keep only functional changes for field-based restart loading. No formatting changes - file now has clean diff against upstream.

Remove multiple spaces before assignment operators to comply with PEP8.

bigfooted · 2026-01-04T20:07:21Z

TestCases/rans/naca0012/turb_SA_restart_test.cfg

+SOLVER= RANS
+KIND_TURB_MODEL= SA


How was the restart file made? Which missing variables are tested here?

The restart file is generated by euler_NACA0012_precursor.cfg (EULER solver) which outputs restart_flow_euler.dat containing only flow fields (Density, Momentum, Energy) but no turbulence variables.

The test then restarts using turb_SA_restart_test.cfg (RANS/SA solver) which loads this Euler restart file via SOLUTION_FILENAME= restart_flow_euler.dat.

Missing variable tested: Nu_Tilde (SA turbulence working variable) - which is absent in the Euler restart file.

The fix initializes the missing Nu_Tilde from freestream defaults (Solution_Inf), allowing successful Euler→RANS restart.

SU2_CFD/include/solvers/CFVMFlowSolverBase.inl

bigfooted · 2026-01-05T09:07:43Z

SU2_CFD/src/solvers/CTurbSolver.cpp

+    /*--- Determine number of turbulence variables in restart file ---*/
+    /*--- Total vars in restart = coordinates + flow vars + turbulence vars + grid velocities ---*/
+    unsigned short nVar_Restart = 0;
+    if (Restart_Vars[1] > skipVars) {
+      /*--- Estimate turbulence vars: total - skipVars, but cap at nVar ---*/
+      nVar_Restart = min(static_cast<unsigned short>(Restart_Vars[1] - skipVars), nVar);
+    }
+


This cannot be done in a reliable way.
The best way forward I think is to introduce for each of the transported variables fixed variable names, the names that are now introduced only in CFlowOutput.
So for SA: "Nu_Tilde"
for SST: Turb_Kin_Energy, Omega
for LM: LM_gamma, LM_Re_t

We can then use these names in CFlowOutput and search for the presence of these strings in the header of the restart file.
You can then search independent of the number of fields (missing or not) that come before or after.

Hi @bigfooted,

I have updated restart loading to use name-based field lookup instead of unreliable skipVars calculation.

Added field name constants (Nu_Tilde, Turb_Kin_Energy, Omega, LM_gamma, etc.) in CRestartFieldNames.hpp

SA/SST/LM models now use FindFieldIndex(field_name) to search for variables without relying on field position

Missing fields are detected, a warning is issued, and defaults are used smoothly

Removed unused skipVars from CFVMFlowSolverBase.inl

I tested this with turb_SA_restart_test.cfg (SA restart from Euler solution) and it correctly detected the missing Nu_Tilde, issued a warning, and ran successfully.

bigfooted · 2026-01-10T21:07:11Z

SU2_CFD/include/solvers/CRestartFieldNames.hpp

@@ -0,0 +1,101 @@
+/*!


I don't think we need this file. It should be sufficient to add the field names to the solver class, for instance vector solverFieldNames={"Turb_Kin_Energy","Omega"} to CTurbSSTSolver.cpp.

… literals for restart field names in solvers

bigfooted · 2026-01-10T21:49:20Z

I don't see why you need most of the other changes. Let's first think about the use case: A user has a result without a turbulence model and now wants to add a turbulence model. Or a user wants to use the SST model instead of the SA model. I don't think we will have use cases where a user has a restart file with a missing flow variable. You can still use the same approach of getting the index from the location of the field names in the restart file, but we do not expect missing field variables here.

The only thing you need to do for the ascii restart (let's focus on that one first) is read the first line of the restart file into a vector of strings and then check the indices of the fields in solverFieldNames for the respective solvers. So that's a function with like 5 lines of code (your lambda helper). Place this function centrally and re-use. Then just use these indices to read the data, when the index is valid. If it is not valid, you do nothing and the default data is used.

…lver::FindFieldIndex

bigfooted · 2026-01-11T08:39:45Z

SU2_CFD/include/solvers/CFVMFlowSolverBase.hpp

+  void LoadRestartSolutionFields(CGeometry **geometry, CSolver ***solver, CConfig *config,
+                                  bool update_geo, bool static_fsi, bool steady_restart,
+                                  su2double* SolutionRestart, unsigned short nVar_Restart,
+                                  const string& restart_filename);


we don't need this. remove.

bigfooted · 2026-01-11T08:40:02Z

SU2_CFD/include/solvers/CFVMFlowSolverBase.hpp

+   */
+  void LoadRestartGridData(CGeometry **geometry, long iPoint_Local, unsigned long baseIndex,
+                           bool update_geo, bool static_fsi, bool steady_restart);
+


we don't need tihs. remove.

bigfooted · 2026-01-11T08:41:26Z

SU2_CFD/src/solvers/CTurbSolver.cpp

+    struct TurbFieldInfo {
+      string name;
+      unsigned short model_var_index;
+    };
+    vector<TurbFieldInfo> target_fields;
+


use vector<string>

bigfooted · 2026-01-11T08:43:04Z

SU2_CFD/include/solvers/CSolver.hpp

+   * \param[in] fieldName - Name of the field to find (with or without quotes).
+   * \return Index of the field (0-based, excluding PointID), or -1 if not found.
+   */
+  int FindFieldIndex(const string& fieldName) const;


make input the vector of strings target_field, and output is vector of int with the indices.

bigfooted · 2026-01-11T08:44:52Z

SU2_CFD/src/solvers/CSolver.cpp

+int CSolver::FindFieldIndex(const string& fieldName) const {
+  /*--- Search for the field name in the fields vector.
+        Fields are stored with quotes, e.g., "Temperature", so we need to check
+        both with and without quotes. The first field is "PointID" which we skip. ---*/
+
+  string fieldNameWithQuotes = "\"" + fieldName + "\"";
+  string fieldNameNoQuotes = fieldName;
+
+  /*--- Search starting from index 1 (skip PointID) ---*/
+  for (size_t i = 1; i < fields.size(); i++) {
+    string field = fields[i];
+    PrintingToolbox::trim(field);
+
+    /*--- Check if field matches (with or without quotes) ---*/
+    if (field == fieldNameWithQuotes || field == fieldNameNoQuotes) {
+      /*--- Return index excluding PointID (0-based) ---*/
+      return static_cast<int>(i - 1);
+    }
+  }
+
+  /*--- Field not found ---*/
+  return -1;
+}
+


std::vector findIndices(const std::vectorstd::string& values, const std::vectorstd::string& searchvalues) {
std::vector indices;
for (const auto& search : searchvalues) {
auto it = std::find(values.begin(), values.end(), search);
indices.push_back(it != values.end() ? std::distance(values.begin(), it) : -1);
}
return indices;
}

bigfooted · 2026-01-11T08:46:44Z

SU2_CFD/src/solvers/CTurbSolver.cpp

+    /*--- Fallback variables for legacy loading ---*/
+    unsigned short skipVars = 0;
+    unsigned short nVar_Restart_Legacy = 0;


Remove legacy loading.

bigfooted · 2026-01-11T08:48:38Z

SU2_CFD/include/solvers/CFVMFlowSolverBase.inl

+#include "../../Common/include/CConfig.hpp"
+#include "../../Common/include/geometry/CGeometry.hpp"


revert all changes to CFVMFlowSolverBase.inl. First implement only the missing restarts for turbulence solvers.

pcarruscag changed the base branch from master to develop January 3, 2026 18:35

pcarruscag reviewed Jan 3, 2026

View reviewed changes

guptapratykshh added 2 commits January 4, 2026 13:43

Fix su2code#2483: Add field-based restart loading with missing field …

eb7e707

…initialization

Refactor restart loading to use static polymorphism and existing node…

6f1f9ab

… defaults

guptapratykshh force-pushed the feature/restart-missing-fields-2483 branch from 8488024 to 6f1f9ab Compare January 4, 2026 08:16

guptapratykshh requested a review from pcarruscag January 4, 2026 08:38

guptapratykshh added 2 commits January 4, 2026 19:48

Fix post-rebase conflicts and add regression test for restart with mi…

c3e05eb

…ssing fields

Fix python lint errors in parallel_regression.py

d8ea001

pcarruscag reviewed Jan 4, 2026

View reviewed changes

guptapratykshh added 3 commits January 4, 2026 22:25

Fix formatting in CFVMFlowSolverBase.inl and verify regression test c…

e4a55af

…orrectness

Revert formatting changes in CFVMFlowSolverBase.inl

afae201

Keep only functional changes for field-based restart loading. No formatting changes - file now has clean diff against upstream.

Fix Python linting errors (E221) in parallel_regression.py

83d334c

Remove multiple spaces before assignment operators to comply with PEP8.

guptapratykshh requested a review from pcarruscag January 4, 2026 17:38

Merge branch 'develop' into feature/restart-missing-fields-2483

0f864f9

bigfooted reviewed Jan 4, 2026

View reviewed changes

Merge branch 'develop' into feature/restart-missing-fields-2483

aa82242

github-advanced-security bot found potential problems Jan 5, 2026

View reviewed changes

SU2_CFD/include/solvers/CFVMFlowSolverBase.inl Fixed Show fixed Hide fixed

bigfooted reviewed Jan 5, 2026

View reviewed changes

guptapratykshh added 3 commits January 6, 2026 22:02

Merge branch 'develop' into feature/restart-missing-fields-2483

fb3eef1

Merge branch 'develop' into feature/restart-missing-fields-2483

7bda309

Use name-based field lookup for turbulence restart variables

23ae35a

bigfooted reviewed Jan 10, 2026

View reviewed changes

Address review feedback: remove CRestartFieldNames.hpp and use string…

4ff1633

… literals for restart field names in solvers

Simplify restart logic: remove local helper and warnings, rely on CSo…

9f519cd

…lver::FindFieldIndex

bigfooted reviewed Jan 11, 2026

View reviewed changes

		#include "../../Common/include/CConfig.hpp"
		#include "../../Common/include/geometry/CGeometry.hpp"

		SOLVER= RANS
		KIND_TURB_MODEL= SA

Fix #2483: Add field-based restart loading with missing field initialization #2675

Are you sure you want to change the base?

Fix #2483: Add field-based restart loading with missing field initialization #2675

Conversation

guptapratykshh commented Jan 3, 2026

Proposed Changes

Related Work

Issue Description:

PR Checklist

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bigfooted commented Jan 4, 2026

Uh oh!

guptapratykshh commented Jan 4, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

guptapratykshh Jan 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bigfooted commented Jan 10, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bigfooted Jan 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

guptapratykshh Jan 4, 2026 •

edited

Loading

bigfooted Jan 11, 2026 •

edited

Loading