Value errors in some CUTE problems

[lambe]

This issue comes from testing Auglag on CUTE using @syarra's nl files. When I run osbornea, the program raises a ValueError exception at the first new point. The source of this error is unknown: no details of the error reach the Python debugger.

When I run either vanderm2 or vanderm3, the computed f value at the first new point is nan. This does not raise a ValueError exception, but it infects many other calculations later on, and the algorithm does not converge.

The issue is that the functions in these problems seem "nice"; there are no obvious places for a division by zero or an excessively large value. This points to an error in the .nl file itself or the translation into an amplmodel.

Can someone confirm these results on another computer just to be sure?

[dpo]

Does it happen with a CUTEst model? I've observed errors in the AMPL interface before.

[lambe]

osbornea seems fine in the CUTEst interface, but vanderm2 and vanderm3 have the same nan problem reported above.

[dpo]

Could it be that the exponential overflows?