While solving a model, the nonlinear solver is constantly computing a gradient. The gradient gives the rate of improvement of the objective function for small changes in the variables. If the gradient's rate of improvement computation for a given variable is less-than-or-equal-to NOPTOL, the nonlinear optimality tolerance, further adjustments to the variable's value are not considered to be beneficial. The default value for the nonlinear optimality tolerance is .0000001. Decreasing this tolerance towards a limit of 0 will tend to make the solver run longer and may lead to better solutions to poorly formulated or poorly scaled models.