Define dies with 'Too many atoms'

[Arrepiadd]

It just appeared on the Organic Letters a paper on a (finite) nanotube calculation. The biggest molecule has slightly more than 1000 atoms.
Just for kicks, to see ri and marij working, I thought of doing the same calculation. I couldn't go from the first part of the define process.

After adding the coordinate file define tells this:
"MODTRACE: no modules on stack

  too many atoms
 define ended abnormally
 define ended abnormally
program stopped."

and I get back to the command line. So clearly, define has some issues with the big molecules.

So I have a few questions:
1) What is the biggest size of a molecule for define? (There is no information on the manual, on the tutorial, online, anywhere... at least easily findable. The limit is not 1000 atoms because with 902 the define died the same way).
2) What is the alternative way to use a molecule this size? (Tmole isn't obviously the solution because it goes through the define process. Yes, I tried...)
3) Isn't this a limitation on Turbomole? We are given a myriad of possibilities for computing big stuff (RI, MA-RI, just to stay in the QM) and then, the input system doesn't allow us to "think large"...

Please don't see my last question as an attack. See it, at best, as a request for enhancement...

Best regards,
Daniel

[antti_karttunen]

Hello,

In addition to the "normal" Turbomole version, there is also a "huge" version for larger systems. The limits are as follows:

   Parameter         |     default version    |    huge version
   ----------------------------------------------------------------
    atoms            |        700             |        1400
    basis functions  |      10000             |       21000
    shells           |       6000             |       16000

So, your system would run nicely with the "huge" binaries. I guess you should ask your site administrator to contact Turbomole support to obtain the huge binaries. Unofficial binaries with even larger limits have also been available for previous versions of Turbomole, but not yet for 5.10. So to answer to your question 3): Turbomole certainly allows users to "think large"

[Arrepiadd]

So to answer to your question 3): Turbomole certainly allows users to "think large"

I knew the answer already I mean... there is a MD module in Turbomole. I didn't expect MD users to be satisfied with 50 atoms or something like that...

Thanks for the information, I'll check that.

Best regards,
Daniel

[Arnim]

Hi!

I have some small remarks to this.
If you want to play around and see capabilities of Turbomole, it can be fun to push an ab initio
program to the limits (although such a calculation can not be done on a lazy sunday afternoon.
It will consume large amounts of computational resources), but if you want to do science,
then you have to be careful. It is not assured, that the SCF cycle will converge in my opinion.
The system is probably not a well-behaving single-reference system. You should start from
smaller systems, increase the size, and look at the behaviour.

I cite from the www.turbomole.com:
"Applications to molecules with more than 100 atoms -of all kinds- are mainly carried out in
connection with experiments to check, correct and to predict properties of molecules and their existence." (Reinhart Ahlrichs)

Bottom line: You should know what you are doing otherwise you are maybe just calculating house numbers.

Cheers,
Arnim

[Arrepiadd]

Hi Arnim,

Thanks for the insightful comment.

I am not trying to do science, at least for now, with the 1000 atom nanotube thing. Quoting my original post:
"Just for kicks, to see ri and marij working, I thought of doing the same calculation."

What I really wanted to see was how would these behave. The paper I talked about had the optimized geometries (in PDB format if I recall correctly) and the only thing I wanted was to get an idea of how things would go. Not having an SCF convergence or a geometry convergence would also be acceptable answer...

Best,
Daniel