Comparison of the numerical and grid methods
It is not generally a good idea to rely on a single solution obtained from the optimiser. It is better to use information from an initial solution to narrow the range of values to be considered, so that better accuracy can be obtained in a further solution. A series of 3 or preferably more solutions is generally needed to be fairly sure that the best solution has been found.
Grid method
- Advantages
The points that will be considered are defined manually, so that the user is fully aware of the accuracy that is obtained from a particular solution. It is generally helpful to start by using fewer points in a wider range, and to use the information from each solution to narrow the range and increase the number of points (and hence accuracy) for a subsequent solution. - Disadvantages
With feeding schedules containing many feeds, with many points to be looked at per feed, this method can take a long time to find a solution.
Numerical method
- Advantages
This method is generally faster than the grid method, and can therefore be more suitable when many feeds are involved. - Disadvantages
- The numerical method is very sensitive to the starting point. This means that the point chosen by the optimiser can differ markedly depending on the starting point chosen. The numerical method is most useful for problems including a large number of feeds, where the grid solution might be too slow. In this case, it is recommended that the user attempts a numerical solution initially, and then uses the information from the results graph to narrow the Min/Max ranges for each feed in a sensible manner, so that eventually a grid solution can be applied using fewer points within a narrower range.
- Particularly in the amino acid optimiser, this method is often unable to find a solution if the problem is very complex. The optimiser will warn if NAG is unable to find a solution.
Each user will develop preferences as to which method to use, and in which way. It is however advisable to complete each series of solutions with a grid solution with high accuracy so that the user can be sure that they have found the best solution. If such a solution is restricted by a Min/Max bound in any of the feeds, then these bounds should generally be relaxed to check whether a better solution can be found.