做scf计算一直不收敛

茶味果冻

模型建立的为二氧化钛锐钛矿001面，1x1的超晶胞，在QE中先做的scf计算，结果很奇怪，迭代100次都没有收敛，下面附上in文件，及结果，求大家帮看看问题出在哪:)

&CONTROL

calculation= "scf",

restart_mode='from_scratch'

prefix="tio2",

pseudo_dir= "/vol6/home/zhang00/pseudo1",

outdir= "/",

wf_collect=.TRUE

/

&SYSTEM

ibrav=0,celldm(1)=3.776000,nat=12,ntyp=2,

ecutwfc=25,ecutrho=300,

occupations ='smearing',

smearing ='gaussian',

degauss =0.002,

/

&ELECTRONS

conv_thr = 1.D-5

mixing_beta = 0.7D0

/

ATOMIC_SPECIES

O  16.00  O.pbe-n-rrkjus_psl.0.1.UPF

Ti 48.00 Ti.pbe-sp-van_ak.UPF

ATOMIC_POSITIONS crystal

O  0.0000000.000000 -0.393602

O  0.000000 0.000000-0.103370

O  0.5000000.500000 -0.145116

O  0.5000000.500000 -0.351856

O  0.0000000.500000 -0.269359

O  0.0000000.500000 -0.476099

O  0.5000000.000000 -0.020873

O  0.5000000.000000 -0.227613

Ti 0.000000 0.000000 0.000000

Ti 0.500000 0.500000 -0.248486

Ti 0.000000 0.500000 -0.372729

Ti 0.500000 0.000000 -0.124243

K_POINTS automatic

2 2 4 0 0 0

CELL_PARAMETERS angstrom

3.7760000  0.0000000   0.0000000

0.0000000  3.7760000   0.0000000

0.0000000  0.0000000  19.0875880

这是in文件，下面是计算结果：

Self-consistentCalculation

     iteration #  1    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  1.00E-02, avg # of iterations =  2.0

     negative rho (up, down):  0.163E-01 0.000E+00

     total cpu time spent up to nowis       63.4 secs

     total energy              =    -397.12256856 Ry
     Harris-Foulkes estimate   =    -736.27759390 Ry
     estimated scf accuracy    <    378.46031089 Ry

     iteration #  2    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     c_bands:  3 eigenvalues not converged
     c_bands:  5 eigenvalues not converged
     c_bands:  2 eigenvalues not converged
     c_bands:  2 eigenvalues not converged
     c_bands:  1 eigenvalues not converged
     c_bands:  4 eigenvalues not converged
     c_bands:  4 eigenvalues not converged
     ethr =  1.00E-02, avg # of iterations = 22.6

     negative rho (up, down):  0.314E+00 0.000E+00

     total cpu time spent up to nowis      228.6 secs

     total energy              =    -767.86613127 Ry
     Harris-Foulkes estimate   =  -1895.37992668 Ry
     estimated scf accuracy    <   3919.46724766 Ry

     iteration #  3    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations = 15.2

     negative rho (up, down):  0.286E+00 0.000E+00

     total cpu time spent up to nowis      347.5 secs

     total energy              =    -775.65518344 Ry
     Harris-Foulkes estimate   =  -1038.96639788 Ry
     estimated scf accuracy    <   1721.30345809 Ry

     iteration #  4    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations = 13.5

     negative rho (up, down):  0.280E+00 0.000E+00

     total cpu time spent up to nowis      418.8 secs

     total energy              =    -755.80317573 Ry
     Harris-Foulkes estimate   =   -833.23419122 Ry
     estimated scf accuracy    <   1500.67222812 Ry

     iteration #  5    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  6.3

     negative rho (up, down):  0.240E+00 0.000E+00

     total cpu time spent up to nowis      466.8 secs

     total energy              =    -707.58129894 Ry
     Harris-Foulkes estimate   =   -769.94919724 Ry
     estimated scf accuracy    <   1060.31353961 Ry

     iteration #  6    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations = 13.6

     negative rho (up, down):  0.132E+00 0.000E+00

     total cpu time spent up to nowis      540.7 secs

     total energy              =    -740.24716235 Ry
     Harris-Foulkes estimate   =   -756.86916599 Ry
     estimated scf accuracy    <     84.62104022 Ry

     iteration #  7    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization with overlap
     ethr =  1.00E-02, avg # of iterations =  3.2

     negative rho (up, down):  0.133E+00 0.000E+00

     total cpu time spent up to nowis      580.1 secs

     total energy              =    -742.97109663 Ry
     Harris-Foulkes estimate   =    -744.51709411Ry
     estimated scf accuracy    <    122.58044785 Ry

     iteration #  8    ecut=    25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  1.2

     negative rho (up, down):  0.127E+00 0.000E+00

     total cpu time spent up to nowis      613.5 secs

     total energy              =    -744.39121638 Ry
     Harris-Foulkes estimate   =   -744.38140910 Ry
     estimated scf accuracy    <    185.57825644 Ry

     iteration #  9     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  2.1

     negative rho (up, down):  0.148E+00 0.000E+00

     total cpu time spent up to nowis      647.7 secs

     total energy              =   -742.92517405 Ry
     Harris-Foulkes estimate   =   -744.57245777 Ry
     estimated scf accuracy    <    201.56723332 Ry

     iteration # 10     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  1.2

     negative rho (up, down):  0.253E+01 0.000E+00

     total cpu time spent up to nowis      681.1 secs

     total energy              =    -714.19687735 Ry
     Harris-Foulkes estimate   =   -743.17939759 Ry
     estimated scf accuracy    <    166.56686994 Ry

     iteration # 11     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations = 12.8

     negative rho (up, down):  0.119E+01 0.000E+00

     total cpu time spent up to nowis      768.0 secs

     total energy              =    -720.12214182 Ry
     Harris-Foulkes estimate   =   -771.65212427 Ry
     estimated scf accuracy    <    648.43840803 Ry

     iteration # 12     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  7.3

     negative rho (up, down):  0.623E+00 0.000E+00

     total cpu time spent up to nowis      827.8 secs

     total energy              =    -720.38448364 Ry
     Harris-Foulkes estimate   =   -747.22863447 Ry
     estimated scf accuracy    <    304.23289007 Ry

     iteration # 13     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations = 10.3

     negative rho (up, down):  0.191E+00 0.000E+00

     total cpu time spent up to nowis      893.4 secs

     total energy              =    -729.65948254 Ry
     Harris-Foulkes estimate   =   -740.42400166 Ry
     estimated scf accuracy    <     87.94588865 Ry

     iteration # 14     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  5.8

     negative rho (up, down):  0.530E-01 0.000E+00

     total cpu time spent up to nowis      939.3 secs

     total energy              =    -732.61535430 Ry
     Harris-Foulkes estimate   =   -740.72404564 Ry
     estimated scf accuracy    <     55.97266718 Ry

     iteration # 15     ecut=   25.00 Ry     beta=0.70
     Davidson diagonalization withoverlap
     ethr =  1.00E-02, avg # of iterations =  3.3

后面省略。。。。。

     End ofself-consistent calculation

     convergence NOT achieved after 100iterations: stopping

     Writing output data filetio2.save

卡开发发

PWscf我不是很会用，但是从参数上看，有下面几个问题。表面体系为啥k取2x2x4?对于1x1的表面差不多4x4x1才是合理的。晶格参数来看你的真空层也许只有10A，最好还能大一些，上下表面本身不对称，层间其实就会有微弱的作用，可以的话考虑work function的校正，截断能原则上也要进行测试。如果上述都没问题，那么应该尝试减小mixing_beta这个参数，默认是0.7还是很大的，不妨减到0.3以下试试。

茶味果冻

卡开发发 发表于 2014-11-15 12:27
PWscf我不是很会用，但是从参数上看，有下面几个问题。表面体系为啥k取2x2x4?对于1x1的表面差不多4x4x1才是 ...

刚才调整了一下参数，还是不行，

卡开发发

茶味果冻 发表于 2014-11-15 14:13
刚才调整了一下参数，还是不行，

您的in文件写的有问题，我还特地看了一下PW的资料：
&SYSTEM
ibrav=0,celldm(1)=3.776000,nat=12,ntyp=2,
上面的部分celldm(1)=3.776000修改为celldm(1)=1.89，其中1.89这个取值的由来是1/0.529。这是因为从您的Lattice vector来看您采用的单位是Angstrom，但是我按照这个输入报错了，似乎不支持在CELL_PARAMETERS后面加上单位。而默认的单位是Bohr，我们在这个位置进行换算，并且TiO2锐钛矿的001面的a和b应该是3.776Angstrom。
如果按照您原先的设置，实际计算采用的a和b会变为
a=3.776000*3.776000 b=3.776000*3.776000 c=19.0875880*3.776000，这里单位也会是Bohr，这样不收敛也非常正常。
celldm(1)这个参数相当于一个整体晶格的标度因子
celldm*CELL_PARAMETERS才是实际的晶格参数。
另外，我测试的时候我采用mixing_beta = 0.3D0算收敛的，不过0.7D0也许也能收敛。

茶味果冻

卡开发发 发表于 2014-11-16 00:35
您的in文件写的有问题，我还特地看了一下PW的资料：
&SYSTEM
ibrav=0,celldm(1)=3.776000,nat=12,ntyp= ...

非常感谢您！问题解决了，之前一直看的5.1版本的手册，而计算中心的居然是4.3版本的，现在把4.3版本的手册和in文件说明都下载下来，看了看发现一些设置都不一样，发现自己好不认真，这次在这里学到好多，真心的感谢老师您，祝好！:)