To use the cheap DFT method (e.g. B3LYP) to study this problem, in order to reasonably represent open-shell singlet diradical, you should do the calculation in symmetry-broken manner, there are two common ways:
(1) # UB3LYP/6-31G* guess=mix (Namely explicitly using unrestricted formalism with symmetry-broken initial guess, the cost is identical to usual UKS calculation)
(2) # B3LYP/6-31G* stable=opt (Program will find the closed-shell wavefunction is unstable, and thus automatically optimize the wavefunction to symmetry-broken one. This way is significantly more expensive than (1), but more robust)
Spin multiplicity should be set to 1 for both cases.
In addition, you can use a special feature of Gaussian to generate initial guess combined by fragment wavefunctions with specific settings, then use guess=read to load it from .chk file as initial guess. If the fragment setting is appropriate, the SCF will eventually converge to correct symmetry-broken solution. The step is slightly more complicated, you can have a try if you find both (1) and (2) do not work as expected.
CASSCF method is also popular in studying diradical system but it is more expensive and difficult to use than DFT. Previously I wrote a blog article to show how to use Gaussian to conduct CASSCF for diradical and discuss how to quantify diradical character: http://sobereva.com/264 (written in Chinese).
For "broken symmetry" calculations of singlet states you must use the prefix U for the method such as UB3LYP (instead of B3LYP).
Moreover you can use either the restart from triplet state (keywords Guess=Restart , do not forget the .chk file) or alternatively use keywords Guess=Mix (without any restart).
To use the cheap DFT method (e.g. B3LYP) to study this problem, in order to reasonably represent open-shell singlet diradical, you should do the calculation in symmetry-broken manner, there are two common ways:
(1) # UB3LYP/6-31G* guess=mix (Namely explicitly using unrestricted formalism with symmetry-broken initial guess, the cost is identical to usual UKS calculation)
(2) # B3LYP/6-31G* stable=opt (Program will find the closed-shell wavefunction is unstable, and thus automatically optimize the wavefunction to symmetry-broken one. This way is significantly more expensive than (1), but more robust)
Spin multiplicity should be set to 1 for both cases.
In addition, you can use a special feature of Gaussian to generate initial guess combined by fragment wavefunctions with specific settings, then use guess=read to load it from .chk file as initial guess. If the fragment setting is appropriate, the SCF will eventually converge to correct symmetry-broken solution. The step is slightly more complicated, you can have a try if you find both (1) and (2) do not work as expected.
CASSCF method is also popular in studying diradical system but it is more expensive and difficult to use than DFT. Previously I wrote a blog article to show how to use Gaussian to conduct CASSCF for diradical and discuss how to quantify diradical character: http://sobereva.com/264 (written in Chinese).