Teuchos - Trilinos Tools Package  Version of the Day
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
List of all members
Trilinos::LinearSolverSetupFailure Class Reference

Exception reporting failure in setting up a linear solver. More...

#include <Trilinos_LinearSolverSetupFailure.hpp>

Inherits runtime_error.

Detailed Description

Exception reporting failure in setting up a linear solver.

This class exists so that NOX and Stratimikos can share a common exception type, without introducing a dependency between the packages. See Trilinos GitHub Issue #1608.

"Failure to set up a linear solver" includes preconditioners and sparse direct solvers. It refers to the setup phase, not the actual solve, for solvers that separate setup and solve into two separate interfaces or function calls. Setup does not refer to a "failure to converge when performing an iterative linear solve."

Here are some examples of legitimate "failures to set up a linear solver":

Any code that throws this exception is responsible for ensuring consistent behavior across multiple MPI processes. Code that catches this exception may assume that if any process in the relevant communicator throws this exception, then all processes in the relevant communicator will throw this exception, in such a way as to ensure correct, deadlock-free MPI behavior. For example, suppose that a preconditioner implements additive Schwarz domain decomposition, with one subdomain per MPI process. Suppose that the preconditioner uses a sparse direct solver on each subdomain. If the sparse direct solver discovers that one subdomain has a singular matrix, then the preconditioner is responsible for ensuring that all processes in the matrix's communicator make a consistent choice to throw the exception.

One would expect to catch and recover from this exception in a nonlinear solver or time integrator. A nonlinear solver might recover by reducing the nonlinear step size. A time integrator might recover by reducing the time step size. In general, recovery from this exception would likely involve slowing down and making choices in favor of stability.

In general, it's suboptimal coding practice to use exceptions for control flow with nonexceptional conditions. Better practice is to return with a useful error code. If a solver can detect conditions like "the matrix is singular," I consider that a nonexceptional condition. However, it looks like NOX and Stratimikos don't have a better way to communicate, so we'll do our best for now by including this exception.

Definition at line 109 of file Trilinos_LinearSolverSetupFailure.hpp.


The documentation for this class was generated from the following file: