Molcas Forum

Samer

Member

Registered: 2020-09-16

Posts: 4

SHIFT vs. IMAGinary: Always use IMAG?

Dear Molcas users,

We are running some CASPT2 calculations along a set of geometries obtained by a linear interpolation. When using the SHIFT=0.2 keyword, we commonly encounter sudden spikes in the energy along the scan. From the output, it is clear that these spikes originate from the shift correction. Here is one extreme example, where the shift is over 900 Hartree:

Reference energy:        -866.9978712789
      E2 (Non-variational):      -2.6136440938
      Shift correction:        -907.5796940990
      E2 (Variational):        -910.1933381928
      Total energy:           -1777.1912094717
      Residual norm:              0.0000005181
      Reference weight:           0.00022

Removing the shift keyword introduces other spikes at different geometries, due to intruder states. Using MS-CASPT2, XMS-CASPT2, XDW-CASPT2, or changing the magnitude of the shift (e.g., to 0.3) does not solve the problem, it only affects the magnitude of the shift or moves the spike(s) to a different geometry. I've recently run into this issue multiple times (e.g., most recently, for a 6-atom organic molecule and for a 31-atom organic molecule), so it may be fairly common, and could go undetected if one does not check the Shift energy correction.

Using IMAGinary, instead of SHIFT, appears to solve the issue, at least for the cases we have tested so far.

This make me wonder, is it ever appropriate to use the SHIFT keyword? The Molcas manual currently says that use of the SHIFT keyword is well tested, but it seems to introduce just as many problems as it is meant to solve. Possibly, this could be a bug that has been introduced in recent versions of OpenMolcas (I am using version 18.09).

If IMAGinary works as well as it seems (I've been using it for some time with no issues), maybe it makes sense to retire the SHIFT keyword and replace it with IMAG.

Samer

stefabat

Member

Registered: 2019-12-02

Posts: 12

Re: SHIFT vs. IMAGinary: Always use IMAG?

Dear Samer

From the snippet of output that you are showing it seems that you are still facing the intruder states despite the real shift. In particular, the value of the reference weight is extremely small. In the output, along with the information you reported here, there should be information on the denominators, which should allow to identify how many intruder states you have and to which denominators are associated. I suspect that if you keep increasing the value of the real shift, you should, at a certain point, obtain converged results. Of course, a large value of the shift is questionable as well, such that if you obtain satisfactory results with the imaginary shift, I suggest you to stick to it.
I don't think there is a bug in the program, and I also suggest you to update your OpenMolcas version as it is not so recent anymore, we are at 21.06 right now. In particular, if you are using RASSI after a XMS-CASPT2 or XDW-CASPT2 calculation there might be a bug affecting the wave functions which was fixed after 18.09.

As a last remark, if you need a very large shift value to fix an intruder state problem, perhaps it is more convenient to include the offending orbital in the active space directly, if that is an option (computationally feasible, physically meaningful).

Cheers
Stefano

Samer

Member

Registered: 2020-09-16

Posts: 4

Re: SHIFT vs. IMAGinary: Always use IMAG?

That makes sense, thank you Stefano. What's odd is that I've been using IMAG for a long time, and never ran into these intruder states (at least, none that I noticed). Only recently, we switched to using SHIFT instead of IMAG (it kind of happened by mistake), and suddenly we noticed these issues with intruder states multiple times in a short span of time. This could be purely coincidental, since we are working with new systems and types of calculations (e.g., different excited state character and including more excited states than we usually do). However, after we saw that using IMAG solved the issues, this is what raised my question if IMAG should just be used instead of SHIFT every time. Looking through some recent examples in the Molcas forum, it seems both keywords still get used.

valera

Administrator

Registered: 2015-11-03

Posts: 129

Re: SHIFT vs. IMAGinary: Always use IMAG?

Samer,
real shift changes the energy, imaginary is not. it is a linear algebra, not quantum chemistry.