QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: zhangguangping on December 22, 2010, 06:09

Title: Question about the convergence
Post by: zhangguangping on December 22, 2010, 06:09
I have encountered a convergence problem,for my system,for other bias there is no problem in convergence.Even I have got the convergence for V=1.0, 1.5 ,2.0V, but for V=1.2Vand 1.4V,the convergence is hard to achived,I calculate V=1.4V using a initio guess from V=1.5.Even if I choose diagonal_mixing_parameter = 0.00001,my tolerance=0.0001.

The convergence process is as the following,the q varies slightly, but the Ebs and dRho....             
Who can tell me the underling reason?
                                                                                                                                               
in addition there is nothing strange for the spectrum at V=1.0V and 1.5V.
                                                                                             
Thanks in advance.
                                                                                                                                                                                                                                                                                                               
......
                     
# sc 204 : q =  585.99929 e  Ebs = -536.20515 Ry  dRho =  1.1830E-03
# sc 205 : q =  585.99917 e  Ebs = -536.76316 Ry  dRho =  8.6365E-04
# sc 206 : q =  585.99915 e  Ebs = -536.81562 Ry  dRho =  1.0491E-03
# sc 207 : q =  585.99923 e  Ebs = -536.82648 Ry  dRho =  1.0925E-03
# sc 208 : q =  585.99924 e  Ebs = -536.77731 Ry  dRho =  9.0449E-04
# sc 209 : q =  585.99924 e  Ebs = -536.78277 Ry  dRho =  9.4113E-04
# sc 210 : q =  585.99971 e  Ebs = -536.76664 Ry  dRho =  8.7112E-04
# sc 211 : q =  585.99926 e  Ebs = -536.20515 Ry  dRho =  1.1830E-03
# sc 212 : q =  585.99914 e  Ebs = -536.77953 Ry  dRho =  9.2239E-04
# sc 213 : q =  585.99914 e  Ebs = -536.82901 Ry  dRho =  1.0987E-03
# sc 214 : q =  585.99923 e  Ebs = -536.82727 Ry  dRho =  1.0910E-03
# sc 215 : q =  585.99926 e  Ebs = -536.77637 Ry  dRho =  8.9766E-04
# sc 216 : q =  585.99926 e  Ebs = -536.77149 Ry  dRho =  8.9816E-04
# sc 217 : q =  585.99971 e  Ebs = -536.76023 Ry  dRho =  8.4768E-04
# sc 218 : q =  585.99926 e  Ebs = -536.20514 Ry  dRho =  1.1829E-03
# sc 219 : q =  585.99921 e  Ebs = -536.77788 Ry  dRho =  9.1608E-04
# sc 220 : q =  585.99944 e  Ebs = -536.78886 Ry  dRho =  9.4414E-04
# sc 221 : q =  585.99913 e  Ebs = -536.34500 Ry  dRho =  6.5681E-04
# sc 222 : q =  585.99912 e  Ebs = -536.83700 Ry  dRho =  1.1288E-03
# sc 223 : q =  585.99913 e  Ebs = -536.83785 Ry  dRho =  1.1306E-03
# sc 224 : q =  585.99925 e  Ebs = -536.83654 Ry  dRho =  1.1317E-03
# sc 225 : q =  585.99919 e  Ebs = -536.76811 Ry  dRho =  8.6867E-04
# sc 226 : q =  585.99919 e  Ebs = -536.80860 Ry  dRho =  1.0286E-03
# sc 227 : q =  585.99971 e  Ebs = -536.80245 Ry  dRho =  9.9508E-04
# sc 228 : q =  585.99914 e  Ebs = -536.20515 Ry  dRho =  1.1829E-03
# sc 229 : q =  585.99916 e  Ebs = -536.82852 Ry  dRho =  1.0964E-03
Title: Re: Question about the convergence
Post by: zh on December 22, 2010, 13:13
try different mixing schemes or tune the history steps used in the mixing procedure.
Title: Re: Question about the convergence
Post by: kstokbro on December 22, 2010, 14:20
You can also try to improve the sampling of energies on the real axis contour, this will decrease the error

I would recommend that you upgrade to the latest version of ATK 2010.8.2 which has a new double contour technique which stabilize finite bias calculations and thus is easier to converge.
Title: Re: Question about the convergence
Post by: zhangguangping on December 27, 2010, 13:16
try different mixing schemes or tune the history steps used in the mixing procedure.

Thanks ,when i use 0.1 as the mixing,one of the task coverged very fast (in five steps).Thanks for your suggestion.
Title: Re: Question about the convergence
Post by: zhangguangping on December 27, 2010, 13:18
You can also try to improve the sampling of energies on the real axis contour, this will decrease the error

I would recommend that you upgrade to the latest version of ATK 2010.8.2 which has a new double contour technique which stabilize finite bias calculations and thus is easier to converge.

Thank you all the same.
Title: Re: Question about the convergence
Post by: phoenix on January 19, 2014, 13:53
try different mixing schemes or tune the history steps used in the mixing procedure.

Thanks ,when i use 0.1 as the mixing,one of the task coverged very fast (in five steps).Thanks for your suggestion.
what do you mean when you say that you have used 0.1 as the mixing, i can't seem to find such an option in atk calculator settings.
Title: Re: Question about the convergence
Post by: kstokbro on January 19, 2014, 14:43
IterationControlParameters(damping_factor=0.1)