wrong eom-ccsd energies for open-shell reference

[fishjojo]

sample input as follows:

from pyscf import gto,scf,cc
import numpy

mol = gto.M(atom='O 0 0 0;H 0 0.78 -0.78; H 0 0 1.1', spin=2)
mol.basis = 'ccpvdz'
mol.symmetry = False
mol.build(verbose = 5)
mf = scf.UHF(mol)
mf.scf()

mycc = cc.UCCSD(mf)
mycc.kernel()
mycc.eomee_ccsd(4)

output energies are as follows:

E(UCCSD) = -75.99324882036385
EOM-CCSD root 0 E = 0.04414031579953943 qpwt = 0.939451 conv = True
EOM-CCSD root 1 E = 0.1036046947840601 qpwt = 0.978899 conv = True
EOM-CCSD root 2 E = 0.2109223996063989 qpwt = 0.539942 conv = True
EOM-CCSD root 3 E = 0.2518781021802057 qpwt = 0.305218 conv = True

The same calculation by NWChem gives the same UCCSD energy but different excitation energies:

CCSD total energy / hartree = -75.993248837113669
Excited state root 1
Excitation energy / hartree = 0.035507922057598
Excited state root 2
Excitation energy / hartree = 0.102790937034369
Excited state root 3
Excitation energy / hartree = 0.204806698436480
Excited state root 4
Excitation energy / hartree = 0.247420373226294

QChem gives the same results as NWChem, so it's very likely that the Pyscf results are wrong.
And uccsd.eomsf_ccsd has the same problem.

The results for spin=0 case are correct.

[gkc1000]

As this bug has been fixed, I'm closing this issue.