Merge pull request #28923 from lynnmunday/fixTaoParamsVec

Fix use of TAO solution vector
idaholab · Oct 25, 2024 · e96644b · e96644b
2 parents 0d384a3 + e62b0fc
commit e96644b
Show file tree

Hide file tree

Showing 16 changed files with 20 additions and 202 deletions.
diff --git a/modules/optimization/examples/materialTransient/gold/optimize_nograd_out_forward0.e b/modules/optimization/examples/materialTransient/gold/optimize_nograd_out_forward0.e
diff --git a/modules/optimization/examples/materialTransient/gold/optimize_nograd_out_forward0.e b/modules/optimization/examples/materialTransient/gold/optimize_nograd_out_forward0.e
@@ -0,0 +1 @@
+optimize_grad_out_forward0.e
diff --git a/modules/optimization/examples/materialTransient/optimize_nograd.i b/modules/optimization/examples/materialTransient/optimize_nograd.i
@@ -6,7 +6,7 @@
   objective_name = objective_value
   parameter_names = 'D'
   num_values = '4'
-  initial_condition = '0.01 0.01 0.01 0.01'
+  initial_condition = '0.2 0.2 0.2 0.2'
 []
 
 [Reporters]
@@ -51,7 +51,7 @@
 [Executioner]
   type = Optimize
   tao_solver = taonm
-  petsc_options_iname = '-tao_gatol'
-  petsc_options_value = '1e-6'
+  petsc_options_iname = '-tao_gatol -tao_nm_lambda'
+  petsc_options_value = '1e-8 0.25'
   verbose = true
 []
diff --git a/modules/optimization/examples/materialTransient/tests b/modules/optimization/examples/materialTransient/tests
@@ -18,6 +18,7 @@
       type = Exodiff
       input = optimize_nograd.i
       exodiff = optimize_nograd_out_forward0.e
+      rel_err = 1e-2
       prereq = material_transient/forward
       method = '!dbg'
       valgrind = none

diff --git a/modules/optimization/examples/simpleTransient/gold/gradFree.csv b/modules/optimization/examples/simpleTransient/gold/gradFree.csv
diff --git a/modules/optimization/examples/simpleTransient/main.i b/modules/optimization/examples/simpleTransient/main.i
diff --git a/modules/optimization/examples/simpleTransient/tests b/modules/optimization/examples/simpleTransient/tests
@@ -3,17 +3,6 @@
     design = "ElementOptimizationSourceFunctionInnerProduct.md NearestReporterCoordinatesFunction.md  ReporterTimePointSource.md"
     issues = '#21885'
     requirement = "The system shall be able to perform force-inversion optimization with time-dependent parameters and outputs with"
-    [objective]
-      type = CSVDiff
-      input = main.i
-      cli_args = "-tao_gatol 90 forward:Outputs/file_base=gradFree
-                  OptimizationReporter/initial_condition='0.017248124297389537 -0.4383269038787613 -1.0120688493193575 -0.604005388481836 -1.4561640635027022 3.231744273062789 0.30497109596745786 0.44633697183648136 0.01016924485170484 2.102321678477381 -0.6633902660583816 0.8744951418636224 -1.6563188219465808 -0.7960335532123246 -0.8868197436891168 3.4789948727882036 1.221786635733348 -1.3649297188353513 0.6718141978617336 1.073784058368632 -0.20990475499085504 -0.23140796936323005 2.719922596118634 0.49032306047413676 0.5754881340320752 -1.4387408219638242 0.5868285452264156 -0.26563061173982583 0.7632228812045031 0.5847693390767825 1.2297597229219401 0.018872725738175458 1.3701093737026238 0.5124595468128872 0.388110595828123 0.11247112873355217 0.9234434408370418 0.3748330890983532 -0.1772288909293001 -0.9744206675334496 -0.3917172574619362 -0.2538603477404897 0.5517774216239373 0.07213835476751881'"
-      csvdiff = gradFree.csv
-      rel_err = 1e-4
-      heavy = true
-      method = OPT
-      detail = 'only a forward solve,'
-    []
     [gradient]
       type = CSVDiff
       input = main_gradient.i

diff --git a/modules/optimization/include/executioners/OptimizeSolve.h b/modules/optimization/include/executioners/OptimizeSolve.h
@@ -173,7 +173,7 @@ class OptimizeSolve : public SolveObject
   /// Number of parameters being optimized
   dof_id_type _ndof;
 
-  /// Parameters (solution)
+  /// Parameters (solution) given to TAO
   std::unique_ptr<libMesh::PetscVector<Number>> _parameters;
 
   /// Hessian (matrix) - usually a matrix-free representation

diff --git a/modules/optimization/src/executioners/OptimizeSolve.C b/modules/optimization/src/executioners/OptimizeSolve.C
@@ -70,7 +70,7 @@ OptimizeSolve::solve()
   _obj_function = &_problem.getUserObject<OptimizationReporterBase>("OptimizationReporter");
 
   // Initialize solution and matrix
-  _obj_function->setInitialCondition(*_parameters.get());
+  _obj_function->setInitialCondition(*_parameters);
   _ndof = _parameters->size();
 
   // time step defaults 1, we want to start at 0 for first iteration to be
@@ -357,9 +357,10 @@ OptimizeSolve::objectiveFunctionWrapper(Tao /*tao*/, Vec x, Real * objective, vo
   auto * solver = static_cast<OptimizeSolve *>(ctx);
 
   libMesh::PetscVector<Number> param(x, solver->_my_comm);
-  *solver->_parameters = param;
+  solver->_parameters->swap(param);
 
   (*objective) = solver->objectiveFunction();
+  solver->_parameters->swap(param);
   PetscFunctionReturn(PETSC_SUCCESS);
 }
 
@@ -371,26 +372,25 @@ OptimizeSolve::objectiveAndGradientFunctionWrapper(
   auto * solver = static_cast<OptimizeSolve *>(ctx);
 
   libMesh::PetscVector<Number> param(x, solver->_my_comm);
-  *solver->_parameters = param;
+  solver->_parameters->swap(param);
 
   (*objective) = solver->objectiveFunction();
-
   libMesh::PetscVector<Number> grad(gradient, solver->_my_comm);
-
   solver->gradientFunction(grad);
+  solver->_parameters->swap(param);
   PetscFunctionReturn(PETSC_SUCCESS);
 }
 
 PetscErrorCode
-OptimizeSolve::hessianFunctionWrapper(Tao /*tao*/, Vec x, Mat /*hessian*/, Mat /*pc*/, void * ctx)
+OptimizeSolve::hessianFunctionWrapper(
+    Tao /*tao*/, Vec /*x*/, Mat /*hessian*/, Mat /*pc*/, void * ctx)
 {
   PetscFunctionBegin;
   // Define Hessian-vector multiplication routine
   auto * solver = static_cast<OptimizeSolve *>(ctx);
-  libMesh::PetscVector<Number> param(x, solver->_my_comm);
-  *solver->_parameters = param;
   PetscErrorCode ierr = MatShellSetOperation(
       solver->_hessian, MATOP_MULT, (void (*)(void))OptimizeSolve::applyHessianWrapper);
+
   CHKERRQ(ierr);
   PetscFunctionReturn(PETSC_SUCCESS);
 }
@@ -424,7 +424,7 @@ Real
 OptimizeSolve::objectiveFunction()
 {
   TIME_SECTION("objectiveFunction", 2, "Objective forward solve");
-  _obj_function->updateParameters(*_parameters.get());
+  _obj_function->updateParameters(*_parameters);
 
   Moose::PetscSupport::petscSetOptions(_petsc_options, _solver_params);
   _problem.execute(OptimizationAppTypes::EXEC_FORWARD);
@@ -447,7 +447,7 @@ void
 OptimizeSolve::gradientFunction(libMesh::PetscVector<Number> & gradient)
 {
   TIME_SECTION("gradientFunction", 2, "Gradient adjoint solve");
-  _obj_function->updateParameters(*_parameters.get());
+  _obj_function->updateParameters(*_parameters);
 
   Moose::PetscSupport::petscSetOptions(_petsc_options, _solver_params);
   _problem.execute(OptimizationAppTypes::EXEC_ADJOINT);

diff --git a/modules/optimization/test/tests/executioners/basic_optimize/tests b/modules/optimization/test/tests/executioners/basic_optimize/tests
@@ -68,23 +68,5 @@
       max_threads = 1 # Optimize executioner does not support multiple threads
       detail = 'testing hand-coded gradient;'
     []
-    [failed_ls]
-      type = RunApp
-      input = quadratic_minimize.i
-      cli_args = '-tao_fd_gradient true'
-      allow_test_objects = True
-      expect_out = 'Solver terminated: -6   Line Search Failure'
-      max_threads = 1 # Optimize executioner does not support multiple threads
-      detail = 'indicating a failed line search;'
-    []
-    [failed_it]
-      type = RunApp
-      input = quadratic_minimize.i
-      cli_args = '-tao_fd_gradient true -tao_ls_type unit -tao_max_it 1'
-      allow_test_objects = True
-      expect_out = 'Solver terminated: -2   Maximum Iterations'
-      max_threads = 1 # Optimize executioner does not support multiple threads
-      detail = 'indicating maximum iteration reached;'
-    []
   []
 []
diff --git a/...ptimization/test/tests/optimizationreporter/mesh_source/gold/main_constant_out_forward0.e b/...ptimization/test/tests/optimizationreporter/mesh_source/gold/main_constant_out_forward0.e
diff --git a/modules/optimization/test/tests/optimizationreporter/mesh_source/gold/main_out_forward0.e b/modules/optimization/test/tests/optimizationreporter/mesh_source/gold/main_out_forward0.e
diff --git a/.../optimization/test/tests/optimizationreporter/mesh_source/gold/main_second_out_forward0.e b/.../optimization/test/tests/optimizationreporter/mesh_source/gold/main_second_out_forward0.e
diff --git a/modules/optimization/test/tests/outputs/exodus_optimization_steady/gold/adjoint.e b/modules/optimization/test/tests/outputs/exodus_optimization_steady/gold/adjoint.e
diff --git a/modules/optimization/test/tests/outputs/exodus_optimization_steady/main.i b/modules/optimization/test/tests/outputs/exodus_optimization_steady/main.i
@@ -35,7 +35,7 @@
   type = Optimize
   tao_solver = taoblmvm
   petsc_options_iname = '-tao_gatol'
-  petsc_options_value = '0.0001'
+  petsc_options_value = '1e-4'
   verbose = true
 []
 
@@ -48,7 +48,7 @@
   []
   [adjoint]
     type = FullSolveMultiApp
-    input_files = adjoint.i
+    input_files = adjoint_iteration_output.i
     execute_on = "ADJOINT"
     clone_parent_mesh = true
   []

diff --git a/modules/optimization/test/tests/outputs/exodus_optimization_steady/main_auto_adjoint.i b/modules/optimization/test/tests/outputs/exodus_optimization_steady/main_auto_adjoint.i
@@ -30,7 +30,7 @@
 [MultiApps]
   [forward]
     type = FullSolveMultiApp
-    input_files = forward_and_adjoint.i
+    input_files = forward_and_adjoint_iteration_output.i
     execute_on = "FORWARD"
   []
 []

diff --git a/modules/optimization/test/tests/outputs/exodus_optimization_steady/tests b/modules/optimization/test/tests/outputs/exodus_optimization_steady/tests
@@ -11,24 +11,20 @@
     abs_zero = 1.0e-5
     # steady solve
     recover = false
-    issues = '#25009'
-    design = 'ExodusOptimizationSteady.md'
-    cli_args = "MultiApps/adjoint/input_files='adjoint_iteration_output.i'"
     requirement = 'The system shall be able to perform gradient based material parameter inversion '
-                  'for a single material property and output the iteration-wise output for the '
+                  'for a single material property and output the iteration-wise exodus output for the '
                   'adjoint problem.'
   []
   [auto_adjoint_iteration_output]
     type = Exodiff
     input = main_auto_adjoint.i
-    cli_args = 'MultiApps/forward/input_files=forward_and_adjoint_iteration_output.i'
     exodiff = 'main_auto_adjoint_out_forward0_exodus.e'
     # Optimize executioner does not support multiple threads
     max_threads = 1
     # steady solve
     recover = false
     requirement = 'The system shall be able to invert for point loads using gradient-based '
                   'optimization with an automatically computed adjoint and output the exodus output '
-                  'per iteration.'
+                  'per iteration for the combined forward and adjoint problem variables.'
   []
 []