From 4c5cca1a501638d7669e7fcdd037f1308c6bd8e4 Mon Sep 17 00:00:00 2001
From: Dmitri Naumov <github@naumov.de>
Date: Wed, 27 Jan 2021 00:19:57 +0100
Subject: [PATCH] [NL] Local Newton. Add increment tolerance.
The increment tolerance is used as an additional convergence
criterion.
Introduction of the new tolerance requires renaming of the
residuum tolerance.
---
.../nonlinear_solver/t_error_tolerance.md | 5 +++-
.../nonlinear_solver/t_increment_tolerance.md | 1 +
.../nonlinear_solver/t_residuum_tolerance.md | 1 +
.../CreateNewtonRaphsonSolverParameters.cpp | 30 +++++++++++++++----
NumLib/NewtonRaphson.h | 21 +++++++++----
...TwoPhaseFlowWithPrhoMaterialProperties.cpp | 5 ++--
6 files changed, 50 insertions(+), 13 deletions(-)
create mode 100644 Documentation/ProjectFile/nonlinear_solver/t_increment_tolerance.md
create mode 100644 Documentation/ProjectFile/nonlinear_solver/t_residuum_tolerance.md
diff --git a/Documentation/ProjectFile/nonlinear_solver/t_error_tolerance.md b/Documentation/ProjectFile/nonlinear_solver/t_error_tolerance.md
index ccae39ea3ad..d6d98d1f91f 100644
--- a/Documentation/ProjectFile/nonlinear_solver/t_error_tolerance.md
+++ b/Documentation/ProjectFile/nonlinear_solver/t_error_tolerance.md
@@ -1 +1,4 @@
-\copydoc NumLib::NewtonRaphson::_tolerance_squared
+Deprecated flag for residuum tolerance. Use `residuum_tolerance` and
+`increment_tolerance` in the future.
+
+If set, increment tolerance will be set to zero.
diff --git a/Documentation/ProjectFile/nonlinear_solver/t_increment_tolerance.md b/Documentation/ProjectFile/nonlinear_solver/t_increment_tolerance.md
new file mode 100644
index 00000000000..9a26ebd7ac4
--- /dev/null
+++ b/Documentation/ProjectFile/nonlinear_solver/t_increment_tolerance.md
@@ -0,0 +1 @@
+\copydoc NumLib::NewtonRaphson::_increment_tolerance_squared
diff --git a/Documentation/ProjectFile/nonlinear_solver/t_residuum_tolerance.md b/Documentation/ProjectFile/nonlinear_solver/t_residuum_tolerance.md
new file mode 100644
index 00000000000..80989e1f68c
--- /dev/null
+++ b/Documentation/ProjectFile/nonlinear_solver/t_residuum_tolerance.md
@@ -0,0 +1 @@
+\copydoc NumLib::NewtonRaphson::_residuum_tolerance_squared
diff --git a/MaterialLib/SolidModels/CreateNewtonRaphsonSolverParameters.cpp b/MaterialLib/SolidModels/CreateNewtonRaphsonSolverParameters.cpp
index c037578659a..35fa51b8c0c 100644
--- a/MaterialLib/SolidModels/CreateNewtonRaphsonSolverParameters.cpp
+++ b/MaterialLib/SolidModels/CreateNewtonRaphsonSolverParameters.cpp
@@ -29,10 +29,30 @@ NumLib::NewtonRaphsonSolverParameters createNewtonRaphsonSolverParameters(
auto const error_tolerance =
//! \ogs_file_param{nonlinear_solver__error_tolerance}
- config.getConfigParameter<double>("error_tolerance");
-
- DBUG("\terror_tolerance: {:g}.", error_tolerance);
-
- return {maximum_iterations, error_tolerance};
+ config.getConfigParameterOptional<double>("error_tolerance");
+ if (error_tolerance)
+ {
+ WARN(
+ "The 'error_tolerance' tag for the Newton-Raphson solver is "
+ "deprecated.\n"
+ "Use new tags 'residuum_tolerance' and 'increment_tolerance'.\n"
+ "For now we use residuum_tolerance {} and increment_tolerance 0.",
+ *error_tolerance);
+ return {maximum_iterations, *error_tolerance, 0};
+ }
+
+ auto const residuum_tolerance =
+ //! \ogs_file_param{nonlinear_solver__residuum_tolerance}
+ config.getConfigParameter<double>("residuum_tolerance");
+
+ DBUG("\tresiduum_tolerance: {:g}.", residuum_tolerance);
+
+ auto const increment_tolerance =
+ //! \ogs_file_param{nonlinear_solver__increment_tolerance}
+ config.getConfigParameter<double>("increment_tolerance");
+
+ DBUG("\tincrement_tolerance: {:g}.", increment_tolerance);
+
+ return {maximum_iterations, residuum_tolerance, increment_tolerance};
}
} // namespace MaterialLib
diff --git a/NumLib/NewtonRaphson.h b/NumLib/NewtonRaphson.h
index 633063cf2f9..b78af5dfafb 100644
--- a/NumLib/NewtonRaphson.h
+++ b/NumLib/NewtonRaphson.h
@@ -21,7 +21,8 @@ namespace NumLib
struct NewtonRaphsonSolverParameters
{
int const maximum_iterations;
- double const error_tolerance;
+ double const residuum_tolerance;
+ double const increment_tolerance;
};
/// Newton-Raphson solver for system of equations using an Eigen linear solvers
@@ -44,8 +45,10 @@ public:
_residual_update(residual_update),
_solution_update(solution_update),
_maximum_iterations(solver_parameters.maximum_iterations),
- _tolerance_squared(solver_parameters.error_tolerance *
- solver_parameters.error_tolerance)
+ _residuum_tolerance_squared(solver_parameters.residuum_tolerance *
+ solver_parameters.residuum_tolerance),
+ _increment_tolerance_squared(solver_parameters.increment_tolerance *
+ solver_parameters.increment_tolerance)
{
}
@@ -64,7 +67,7 @@ public:
_jacobian_update(jacobian);
_residual_update(residual);
- if (residual.squaredNorm() < _tolerance_squared)
+ if (residual.squaredNorm() < _residuum_tolerance_squared)
{
break; // convergence criteria fulfilled.
}
@@ -76,6 +79,11 @@ public:
_solution_update(increment);
+ if (increment.squaredNorm() < _increment_tolerance_squared)
+ {
+ break; // increment to small.
+ }
+
// DBUG("Local Newton: Iteration #{:d} |dx| = {:g}, |r| = {:g}",
// iteration, increment.norm(), residual.norm());
} while (iteration++ < _maximum_iterations);
@@ -99,6 +107,9 @@ private:
ResidualUpdate _residual_update;
SolutionUpdate _solution_update;
const int _maximum_iterations; ///< Maximum number of iterations.
- const double _tolerance_squared; ///< Error tolerance for the residual.
+ const double
+ _residuum_tolerance_squared; ///< Error tolerance for the residuum.
+ const double
+ _increment_tolerance_squared; ///< Error tolerance for the increment.
};
} // namespace NumLib
diff --git a/ProcessLib/TwoPhaseFlowWithPrho/TwoPhaseFlowWithPrhoMaterialProperties.cpp b/ProcessLib/TwoPhaseFlowWithPrho/TwoPhaseFlowWithPrhoMaterialProperties.cpp
index 126341f582a..b15d56a9cf0 100644
--- a/ProcessLib/TwoPhaseFlowWithPrho/TwoPhaseFlowWithPrhoMaterialProperties.cpp
+++ b/ProcessLib/TwoPhaseFlowWithPrho/TwoPhaseFlowWithPrhoMaterialProperties.cpp
@@ -205,14 +205,15 @@ bool TwoPhaseFlowWithPrhoMaterialProperties::computeConstitutiveRelation(
// criteria available from the input file configuration. See Ehlers
// material model implementation for the example.
const int maximum_iterations(20);
- const double tolerance(1.e-14);
+ const double residuum_tolerance(1.e-14);
+ const double increment_tolerance(0);
auto newton_solver = NumLib::NewtonRaphson<
decltype(linear_solver), LocalJacobianMatrix,
decltype(update_jacobian), LocalResidualVector,
decltype(update_residual), decltype(update_solution)>(
linear_solver, update_jacobian, update_residual, update_solution,
- {maximum_iterations, tolerance});
+ {maximum_iterations, residuum_tolerance, increment_tolerance});
auto const success_iterations = newton_solver.solve(J_loc);
--
GitLab