6.3.2. Tpetra Solvers

6.3.2.1. Klu2 Solver

Scope: Tpetra Equation Solver
Summary: Use a KLU2 direct solver.

begin Klu2 Solver

end Klu2 Solver

6.3.2.2. Superlu Solver

Scope: Tpetra Equation Solver
Summary: Use a SUPERLU direct solver.

begin Superlu Solver

end Superlu Solver

6.3.2.3. Umfpack Solver

Scope: Tpetra Equation Solver
Summary: Use a UMFPACK direct solver.

begin Umfpack Solver

end Umfpack Solver

6.3.2.4. Cg Solver

Scope: Tpetra Equation Solver
Summary: Use a conjugate gradient solver.

begin Cg Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Cg Solver

6.3.2.4.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

6.3.2.5. Bicgstab Solver

Scope: Tpetra Equation Solver
Summary: Use a bicgstab solver.

begin Bicgstab Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Bicgstab Solver

6.3.2.5.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

6.3.2.6. Gmres Solver

Scope: Tpetra Equation Solver
Summary: Use a GMRES solver.

begin Gmres Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   Restart Iterations {=} value

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Gmres Solver

6.3.2.6.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

Restart Iterations

Syntax: Restart Iterations {=} value
Summary: Set the number of iterations between GMRES restarts.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	30

6.3.2.7. Flexible Gmres Solver

Scope: Tpetra Equation Solver
Summary: Use a flexible GMRES solver. This is required whenever some component of the preconditioner may change on a per-iteration basis. For example, if the preconditioner relies on an inner GMRES solver.

begin Flexible Gmres Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   Restart Iterations {=} value

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Flexible Gmres Solver

6.3.2.7.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

Restart Iterations

Syntax: Restart Iterations {=} value
Summary: Set the number of iterations between GMRES restarts.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	30

6.3.2.8. Poly Solver

Scope: Tpetra Equation Solver
Summary: Use a poly solver.

begin Poly Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Polynomial Order {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   Restart Iterations {=} value

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Poly Solver

6.3.2.8.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Polynomial Order

Syntax: Polynomial Order {=} value
Summary: Set the polynomial order of polynomial preconditioner.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	5

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

Restart Iterations

Syntax: Restart Iterations {=} value
Summary: Set the number of iterations between GMRES restarts.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	30

6.3.2.9. Gcrodr Solver

Scope: Tpetra Equation Solver
Summary: Use a GCRODR solver.

begin Gcrodr Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Recycled Blocks {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   Restart Iterations {=} value

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Gcrodr Solver

6.3.2.9.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Recycled Blocks

Syntax: Recycled Blocks {=} value
Summary: Set the number of recycle blocks in GCRODR.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	5

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

Restart Iterations

Syntax: Restart Iterations {=} value
Summary: Set the number of iterations between GCRODR restarts.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	30

6.3.2.10. Expert Belos Solver

Scope: Tpetra Equation Solver
Summary: Use a Belos solver with parameters specified in an XML file.

begin Expert Belos Solver

   Xml File {=} value

   begin Dd-Ilu Preconditioner
   end

   begin Dd-Ilut Preconditioner
   end

   begin Dd-Parilut Preconditioner
   end

   begin Expert Ifpack2 Preconditioner
   end

   begin Expert Teko Preconditioner
   end

   begin Fastilu Preconditioner
   end

   begin Jacobi Preconditioner
   end

   begin Muelu Preconditioner
   end

   begin Sgs Preconditioner
   end

   begin Sgs2 Preconditioner
   end

end Expert Belos Solver

6.3.2.10.1. Line Commands

Xml File

Syntax: Xml File {=} value
Summary: Supply the name of an XML file containing a Belos parameter list.
Description: The supplied XML file must contain a ParameterList with a “solution method” string parameter corresponding to the desired type of Belos solver. Any additional parameters that should be passed on to Belos should be part of a nested ParameterList with name=”Belos-Tpetra”.

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

6.3.2.11. Preset Solver

Scope: Tpetra Equation Solver
Summary: Use a preset solver for a specific linear system type.
Description: The preset solver for a given linear system type is intended to provide reasonable behavior for many problems of that type. If this proves to be insufficient, the user must provide more fine-grained control over solver settings. The equivalent command which produces the solver will be output to the log file, and this may be used as a starting point from which changes for robustness may be made (i.e. increasing GMRES subspace size or changing the preconditioner).

begin Preset Solver

   Convergence Tolerance {=} value [ Minimum = tolerance_floor  ]

   Maximum Iterations {=} value

   Residual Output {=} [ {disabled | enabled}  ]

   Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}

   Solver Type {=} {continuity | high_aspect_continuity | multiphysics | scalar_transport | thermal | thermal_bicgstab | thermal_multigrid | thermal_symmetric}

end Preset Solver

6.3.2.11.1. Line Commands

Convergence Tolerance

Syntax: Convergence Tolerance {=} value [ Minimum = tolerance_floor ]
Summary: Set the convergence tolerance for the linear solver. Setting CONVERGENCE TOLERANCE = AUTOMATIC enables an autonomous linear convergence tolerance calculation, which is computed from a combination of the initial linear residual and user-specified nonlinear residual tolerance. An additional floor value on the linear convergence tolerance can be declared with CONVERGENCE TOLERANCE = AUTOMATIC MINIMUM = <val>, where the default floor value is 1.0e-6

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	string	–

Maximum Iterations

Syntax: Maximum Iterations {=} value
Summary: Set the maximum number of iterations taken by the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	integer	300

Residual Output

Syntax: Residual Output {=} [ {disabled | enabled} ]
Summary: Output the linear residual history to files in the residual_output/ directory, which is created automatically. For timestep NNN and linear system NAME, this will output a .csv file named residual_output/linear_solves_NAME.stepNNN. This file includes information about the linear residual at every iteration across all nonlinear iterations.

Parameter	Value	Default
{=}	{= \| are \| is}	–

Residual Scaling

Syntax: Residual Scaling {=} {none | preconditioned_r0 | r0 | rhs}
Summary: Set the residual scaling for the linear solver

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{none \| preconditioned_r0 \| r0 \| rhs}	r0

Solver Type

Syntax: Solver Type {=} {continuity | high_aspect_continuity | multiphysics | scalar_transport | thermal | thermal_bicgstab | thermal_multigrid | thermal_symmetric}
Summary: Set the preset solver type

Parameter	Value	Default
{=}	{= \| are \| is}	–
value	{continuity \| high_aspect_continuity \| multiphysics \| scalar_transport \| thermal \| thermal_bicgstab \| thermal_multigrid \| thermal_symmetric}	–