MCSCF EXCITA problem

[frj]

Using Dalton2015 to run N2 in D2h symmetry with the aug-cc-pVQZ basis and a full valence CASSCF, a calculation asking for the 30 lowest excited states in symmetry 5 (b1u/sigma-u) works fine, but when asking for 40 states, the solver produces errors of the type:

**WARNING** EIGENVALUES NOT PAIRED
*** WARNING Complex eigenvalue in reduced response eq. of symmetry 5
*** WARNING real and imaginary part : 240 39 1.656212D+01 9.735621D-01

The new trial vectors for the linear equation iterations end up being linear dependent and eventually report large valued negative excitation energies.
Increasing the size of the initial guess MAXPHP to e.g. 1000 makes no difference. A corresponding HF calculation does not have the problem.

Yes, I am fully aware that 40 excited states have little connection with reality, but unfortunately I need to ask for at least 40 in order to capture the third lowest state at convergence, which may have some physical reality.

Dalton 2011 and dalton2013 do not have the problem.