I am currently implementing DIIS to accelerate SCF convergence as
 per Pulay, J. Comp. Chem., V 3, #4, 556-60, 1982.  In the RHF
 case the construction of the error vectors is fairly straight forward:
     e(i) = F(i) D(i) S - S D(i) F(i)
 Where i is the step index, F is the operator matrix, D is the total
 density matrix, and S is the overlap matrix.   I am a bit unsure how to
 with the unrestricted case.  I have implemented
      e(i) = Fa(i) Da(i) S - S Da(i) Fa(i)  + Fb(i) Db(i) S - S Db(i)
 Where the a and b are the alpha and beta electron versions of the
 matrix.  This causes the generation of of a single B matrix and thus a
 single set of coefficients from the linear equation.
 Is this the right way to set up the error matrix for step i in UHF and
 UDFT?  If it isn't, what is?  Dose anyone know a reference where this
 is discussed?
 Thanks for your time.
 Ed Bothers.
 Merz Goup, Penn State
 enb108 ^at^ psu.edu