[Pw_forum] More pseudopotential troubles

[Miguel Martínez Canales]

Hi again,

Paolo Giannozzi wrote:
> On Oct 17, 2006, at 19:39 , Miguel Martínez Canales wrote:
> 
>>      ns=   3 l=   1 sum=    0.000034911 r(ikk)     1.899836304
>>      from gener_pseudo  : info #        -1
>>      chi too large beyond r_c
> 
> this occasionally happens for no apparent good reason. I think
> it basically means that the PP generation failed. Try different
> electronic configuration / matching radii / pseudization algorithm /
> pseudization energies / phase of the moon

OK, if I change the reference configuration to [Ar] 3d0 4s2 I will no 
longer get the "chi too large message"... but the transferability of the 
pseudopotential will be *really* awful. If I leave the config at 3d0 4s0 
but add the "unscreening test" 3d0 4s2 it will stil fail.

Let's say however that the p channel will only have one projector, 
pseudized using the RRKJ method. Now, with lloc = -1 and rcloc =1.5 the 
input will work. However, if I add an additional projector to the d 
channel, the "chi message" will reappear. Even if it is only an additional 
projector with rcutus = rcut. My questions are, then:

* Do you have those auxiliary chi functions when you are not generating an 
USPP pseudopotential?

* The chi message will disappear for the d channel at rcut~1.00. The RRKJ 
estimated cutoff is around 225 Ry. However, the estimated cutoff for rcutus 
= 1.70 is 45 Ry in the worst case. Will the high cutoff for rcut affect the 
pseudopotential and/or its convergence?

Thanks very much in advance,

Miguel

PS: No matter how I change the pseudization energy, rcutus or lloc (as long 
as it isn't 1), the p channel will still give the "chi too large" message, 
at least for rcut > 0.80. If rcut > 1.4, the RRKJ pseudization will fail. 
And I find 90 Ry a bit too high for a future phonon calculation.