7.2 Errors and exceptions

Response codes

The function mosek returns a response code (and its human-readable description), informing if optimization was performed correctly, and if not, what error occurred. The expected response, indicating successful execution, is always "MSK_RES_OK". Typical errors include:

  • referencing a nonexisting variable (for example with too large index),

  • incompatible dimensions of input data matrices,

  • NaN in the input data,

  • duplicate conic variable,

  • error in the optimizer.

A full list of response codes, error, warning and termination codes can be found in the API reference. For example, if the objective vector contains a NaN then

r <- mosek(prob)

will produce as output:

[1] 1470

[1] "MSK_RES_ERR_NAN_IN_C: c contains an invalid floating point value, i.e. a NaN."

Optimizer errors and warnings

The optimizer may also produce warning messages. They indicate non-critical but important events, that will not prevent solver execution, but may be an indication that something in the optimization problem might be improved. Warning messages are normally printed to a log stream (see Sec. 7.3 (Input/Output)). A typical warning is, for example:

MOSEK warning 53: A numerically large upper bound value  6.6e+09 is specified for constraint 'C69200' (46020).

Warnings can also be suppressed by setting the MSK_IPAR_MAX_NUM_WARNINGS parameter to zero, if they are well-understood.

Error and solution status handling example

Below is a source code example with a simple framework for handling major errors when assessing and retrieving the solution to a conic optimization problem.

Listing 7.2 Sample framework for checking optimization result. Click here to download.
response <- function()
    # In this example we set up a simple problem 
    prob <- setupProblem()

    # (Optionally) Uncomment the next line to get solution status Unknown
    # prob$iparam <- list(INTPNT_MAX_ITERATIONS=1)

    # Perform the optimization.
    r <- mosek(prob, list(verbose=0))
    # Use the line below instead to get more log output
    #r <- mosek(prob, list(verbose=10))

    # Expected result: The solution status of the interior-point solution is optimal.

    # Check if there was a fatal error
    if (r$response$code != 0 && startsWith(r$response$msg, "MSK_RES_ERR"))
        print(sprintf("Optimization error: %s (%d),", r$response$msg, r$response$code))
        if (r$sol$itr$solsta == 'OPTIMAL')
            print('An optimal interior-point solution is located:')
        else if (r$sol$itr$solsta == 'DUAL_INFEASIBLE_CER')
            print('Dual infeasibility certificate found.')
        else if (r$sol$itr$solsta == 'PRIMAL_INFEASIBLE_CER')
            print('Primal infeasibility certificate found.')
        else if (r$sol$itr$solsta == 'UNKNOWN')
            # The solutions status is unknown. The termination code 
            # indicates why the optimizer terminated prematurely. 
            print('The solution status is unknown.')
            print(sprintf('Termination code: %s (%d).', r$response$msg, r$response$code))
            printf('An unexpected solution status is obtained.')