Another thing to consider is that your system is non-neutral (it has a charge of 40) and is highly inhomogeneous. The kspace error estimators are derived for neutral, homogeneous systems, so it isn’t too surprising that you need a higher accuracy.
Thanks Steve and Stan,
I increased the accuracy and it gives me the exact results of Lammps Oct 30, 2010(attached figs).
I also checked it for the system of 3 rods and got the same results.
I totally see the reason as Stan explained but it might be a good idea to add a comment to Lammps manual for people who are working with charged neutral.
Thanks again for your prompt reply.
more-accuracy.eps (14.9 KB)
3sys-more-accuracy.eps (14.3 KB)