QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: rupen_86 on September 22, 2016, 09:29

Title: speeding up my Silicene nanoribbon geometry optimization process in ATK-DFT
Post by: rupen_86 on September 22, 2016, 09:29

whenever i try to optimize SINR geometry it takes too much time compared to GNR and BNNR. what can i do to speed up this? And if i perform doping in it , bond with those doped atoms disappears!!! any suggestion from side would be highly appreciated...

thanks

Title: Re: speeding up my Silicene nanoribbon geometry optimization process in ATK-DFT
Post by: Jess Wellendorff on September 23, 2016, 08:40

Please provide scripts that you are running. Just attach then to this post. It's very hard to help without this information.

Title: Re: speeding up my Silicene nanoribbon geometry optimization process in ATK-DFT
Post by: rupen_86 on September 24, 2016, 08:05

here is my script..plz tell how can i minimize my geometry optimization run time.....

Title: Re: speeding up my Silicene nanoribbon geometry optimization process in ATK-DFT
Post by: rupen_86 on September 24, 2016, 08:56

is the reason for taking too much time while optimizing is greater bond length for SINR? If yes, what can i do to bring atoms closer?
also i went through many papers where people mentioned "region between dashed lines represent prime periodic cell under simulation". what does it signifies??? di need to perform optimization of that primitive cell only?? and then repeatind thar optimized structurein the builder??? as of now i m not doing like this...
currently i am going like
add->plugin->nanoribon->si-si(bond length=2.25)->repeating unit cell in z-direction->doping/functionalization->introduce buckling-> sen to script generator and so forth..am i doing the right way or missing something...

Title: Re: speeding up my Silicene nanoribbon geometry optimization process in ATK-DFT
Post by: Jess Wellendorff on October 10, 2016, 15:53

I have attached a script that seems to work well (with ATK 2016.2). Have decreased the number of k-points along C to 3, and tightened the SCF convergence criteria to get accurate forces. Yes, the Si-Be bond lengths increase, but do no look unreasonable.

Tips:
- You may also want to try the more accurate (and slower) SG15 pseudopoentials with 100 Ha cutoff energy.
- Your system is periodic along C, but 50 k-points is far too many (as long as this is not an electrode for a device configuration).
- 20 electronic bands for the DiagonalizationSolver may help speed up things.