Fig. 3
From: Guidelines for benchmarking of optimization-based approaches for fitting mathematical models
Ambiguous interpretation of optimization outcomes. For non-trivial optimization problems, the results of independent optimization runs are typically not the same. The upper left panel indicates an outcome for three optimization runs, e.g., generated with different starting points. If the objective function values after optimization are different (scenario A), such an outcome could be explained by local optima (explanation A1) or by convergence problems of the optimization algorithm (explanation A2). If the same values are obtained for objective function, there might be several local optima with the same value of the objective function (explanation B1), there might be a convergence problem (B2), or non-identifiabilities might exist (B3), i.e., the estimated parameters are not uniquely specified by the data and then the same value of the objective function is achieved in multi-dimensional sub-spaces