Force constant could not be determined due to cutoff radii

[zohansyahfatomi]

Dear Admin,

Good afternoon.
I have calculated the SnSe bulk system.
How to determine the correct cutoff radii.

  WARNING: Rank deficient                                                 

  Force constants could not be determined uniquely because                 
  the sensing matrix is not full rank.                                     
  You may need to reduce the cutoff radii and/or increase the number of    
  training datasets.   

Do yo have any suggestions?

[RicardoCecconello]

You can find the optimal cutoff via convergence tests (i.e., evaluate the fitting error for different cutoff values).
In your case, you don't have enough training data for that given cutoff/nbody configuration. For that, check this note on the ALM tutorial.
If you don't want to generate more training data, you should reduce cutoff or nbody.

[zohansyahfatomi]

@RicardoCecconello
Thank you for the answer.
"Generate more training data," does that mean that I should increase the atoms in my supercell?

[ttadano]

It means you will need to increase the size of the training dataset by adding more displacement patterns in the supercell. The number of necessary displacement patterns is estimated by ALM with MODE=suggest, but this number changes if you change the cutoff radius. So, please be careful.

[zohansyahfatomi]

@ttadano
Thank you for the answer.
Is it safe to use

&cutoff
*-* None

Or should I assign some cutoff radii value manually?
Thank you very much.

[ttadano]

For harmonic IFC calculations (NORDER = 1), setting

&cutoff
*-* None
/

should work well and is recommended.

[zohansyahfatomi]

@ttadano
Dear Tadano Sensei,

Thank you, for harmonic case is OK.
I wonder if there are two atoms, for example, SnSe.
How do we set up the cut-off radi for cubic cases?
The example is only for Silicon ( a single atom).