Hi everbody,
When i did a supercell with carbon impurity, i found an imaginary frequency arising around -5 THZ(ediffg=-0.02, lreal= auto, ediff=1e-8, geometry optimization with isif =2). however, when i relax the structure with isif=4 to reduce ediffg (ediff=1e-8, lreal = .f., ediffg=-0.001), the phono spectrum became sp strange. Can you tell me how to solve it and the reason?
Thanks for your attention and support.
Yidan Wei
Here is the INCAR of geometry optimization.
#Name of system
System = ga2o3 C
#Start parameter
ISTART = 0 #startjob: no WAVECAR file (default)
ICHARG = 2 #charge: from atoms (default)
#Write flags
LWAVE = .F #create WAVECAR
LCHARG = .F #create CHGCAR
#Parallelization
NPAR = 8 #parallelization over bands
#Electronic Relaxation 1
ENCUT = 400 #energy cutoff in eV
EDIFF = 1E-7 #accuracy for electronic minimization
PREC = Accurate #precision
NELM = 200 #maximum of 200 electronic steps
NELMIN = 4 #minimum of two steps,P166
LMAXMIX = 4 #fast convergence to the groundstate,4 for d elements
#Electronic Relaxation 2
LREAL = Auto #evaluate projection operators in real space
#Ionic Relaxation
NSW = 200 #number of steps for IOM
IBRION = 2 #CG algorithm
ISIF = 4 #relax ions + volume
EDIFFG = -1E-3 #stopping-criterion for IOM
POTIM = 0.5 #time-step for ion-motion
ISYM = 0 #symmetry
#DOS related values
ISMEAR = 0 #use Gaussian smearing method
SIGMA = 0.1 #tiny smearing width to safely break symmetry
#Exchange-correlation functional
GGA = PE #PW-PB-LM-91-PE-RP-AM-PS
ISPIN = 2 #allow for spin polarisation
NELECT =572
ADDGRID = .TRUE.
I think that you will always get imaginary frequencies when you are out of equilibrium, so your structure is not in a minimum.
To get all the frequencies positive, you have to start the geometry optimization from different starting structure till you get it.