Conserved quantity changed and used time increase when during a time in AIMD when do restart with different EPS_SCF

[simon Louis]

Hi cp2k groups,

I'm a new cp2k user and recently i simulated the contact Angle of water on graphene using AIMD. I found that different values of EPS_SCF lead to different changes in cons qty, thus leading to different increases in used time. As shown in the following three figures, when cons qty is almost unchanged, the time taken by each MD step is the same, but when cons qty rises, usedtime rises sharply, may I ask what is the problem? How can I improve it? The following is my input file.

EPS_SCF = 1.0*10^-5, cutoff = 320 Ry

5.0*10^-6, cutoff =  380Ry

2*10-6,  cutoff = 380 Ry

I used NVT 330 K and nose-hoover chain, and here is my input file.

My questions are: 1. Does cons Qty remain constant over time once NVT reaches equilibrium? 2. Why does the cons Qty in 2*10-6 fluctuate wildly after it has remained unchanged for a long time? 3. How to accelerate AIMD. 

Best regards, 

X.J Xu

[Marcella Iannuzzi]

Dear X.J.Xu

Such total energy drifts indicate that the results of the simulation are not reliable, the errors in the integration of the equation of motion are to big. 

The conserved quantity must be conserved.

The increase in time per MD step is most probably related to an increase in the number of SCF iterations, if you are running BOMD. 

If this is the case, i.e., if it becomes more and more difficult to optimise the wave function, it might indicate that the computed forces are rather wrong and lead to the sampling of weird geometries.

By the way, have you considered that graphene is a semi-metal? This makes the convergence of the wave function even more problematic. 

Before worrying on how to speed up the sampling, it is recommendable to find settings that produce reliable results, for instance further reducing EPS_SCF, adding virtual MOS, increasing the cutoff, etc.

Kind regards

 Marcella

[simon Louis]

Dear  Marcella

Thank you very much for your reply, this problem has troubled me for a long time. When I set EPS_SCF to 1.0E-6 during NVT BOMD simulation, the conserved quantity always shows a straight line drift, while the time spent is basically stable. Is it true that the conserved quantity will increase stably in NVT simulation? By the way, can SGCPMD improve the sampling efficiency of this system?

Kind regards

X. J  Xu