Good day! check your structure carefully! and try to use not more than 4 cores for Gaussian jobs. 

"Error in internal coordinate system." means that gaussian has problems with the geometry (likely the starting geometry, since it's after 17 minutes). It has nothing to do with the number of cores. 

See if the input geometry makes sense. Even if it does, change it slightly and rerun. The problem is that by default gaussian switches to internal coordinates (for speeds sake) and sometimes makes a bad choice which leads to convergence problems. You can also try opt(cartesian) or opt(z-matrix) which changes the default optimization route and is particuralry useful for large systems and systems with a lot of flat (close to 180 degrees) angles.

Internal coordinates are the z-matrix in Gaussian. The problem here is usually that the next minimization step on the geometry will cause collision of the atoms or bond vectors, or simply that the step cannot be performed given the bonds, angles, and dihedrals of the current z-matrix. You can switch to Cartesian as noted above, but re-ordering the z-matrix has often worked for me. I concur that the cores used for Gaussian are not the issue, more is usually better until you start distribution over multiple compute nodes. You just may need 2 gig of memory per core to avoid memory errors.

I am resubmitting the 6-31G optimized geometry to Aug-cc-pVDZ optimization at MP2 level... How can I change the internal Coordinates??

In general gaussian creates the redundant, internal coordinates on its own using the provided z-matrix or xyz input. You can add / delete / modify selected coordinates using ModRedundant keyword (see http://www.gaussian.com/g_tech/g_ur/k_opt.htm) and it may help in reaching the convergence in selected cases; you can't though create your own redundant coordinates for the entire system.

Again, fr troubling cases use opt(z-matrix) or opt(cartesian), which is slower but may help in some cases (but also make the convergence worse in other cases).

I used   keyword "otp=(loose,cartesian) "  but still I am unable to optimize geometry..!

and when I used otp(cartesian)  my optimizations goes wrong way...

Or you can just write opt=cartesian. Sometimes  opt=cartesian,noeigen helps better for this kind of errors 

Check the HDD health and availability of both memory.

"I used keyword "otp=(loose,cartesian) " but still I am unable to optimize geometry..!

and when I used otp(cartesian) my optimizations goes wrong way..."

Why can't you optimize it? Is it the same error immedaitely during the first optimization step? Same error (in internal coordinate system) later, after a number of steps? If the otpimization goes "wrong way" (i.e. the way you did not expect) it may mean the the starting structure is far from the optimal one.

Can you give us the full input and log files?

Please find attatchment....@Eric Butson 

 Dear Bartosz Trzaskowski Sir optimization goes wrong way means it gives me same energy after each 5 - 6 cycles...

Hi Rahul,

I apologize for the late reply. Have you tried using an input with z-matrix? Try the following input and see if it works for you.

Thanks, Eric. I will check that..

Dear Eric , I have submitted given Z- matrix but still it's not working. please find the attachment.

It looks like you might be getting some sort of error with the eigenvalues. Have you tried using opt=noeigentest? Sometimes this works and sometimes it doesn't. The only other suggestion I would have is to tinker with your starting structure. I will look at this more later and see if I can figure anything out. In the meantime, give noeigentest a try.

hello, Eric I tried this keyword also but it's not working....

This is odd. I tried it out last night and it was working just fine with the z-matrix input file. It was slow, but working. So I'm not really sure what else it could be. I apologize.

Thanks, Eric I will check this again.

Dear Eric, Please find attachment......