Mod:Hunt Research Group/scf convergence

is your system not spin paired?

• try an unrestricted calculation: UB3LYP or ROB3LYP
• for triplet states try turning off the DIIS: SCF(NODIIS)

more general suggestions

• if you can get a single SCF cycle to run:

test the stability of your wave function with STABLE GUESS=READ

you can then try STABLE=OPT

• turn on extra printing so you can see what is going on: #p

then to observe the scf convergence use: grep -i 'RMSDP=' filename.log | cat -n

it is fairly clear that RMSDP is RMS density, MaxDP is MAX density and DE is energy difference

• try a mixed first guess: GUESS=MIX
• try damping in the early scf iterations: SCF=(DAMP)
• try quadratic convergence for the SCF: SCF=(QC) (you cannot use this for restricted open shell calculations)
• try shifting the LUMO and above levels up SCF=(VSHIFT)
• try reducing the SCF convergence criteria: SCF=(CONVER=6)

the default is 10*-8 on RMS density and 10*-6 on MAX density and energy

this value will reduce to 10*-6 on RMS density and 10*-4 on MAX density and energy

• you can increase the number of cycles, this may not help: SCF=(MAXCYC=200)

the default is 128 for DIIS and 512 for QC

examples

• a very difficult convergence case

#p rob3lyp/6-31g(d) geom=connectivity nosymm scf(damp,vshift,conver=5,maxcyc=200)

#p rob3lyp/6-31g(d) geom=connectivity nosymm scf(damp,vshift,conver=6,maxcyc=300)