Skip to content

Output File

Normal Termination

                             ****ORCA TERMINATED NORMALLY****

Orbital Optimization: Macro-Iterations

CASSCF alternates between solving the CI problem within the active space and optimizing the orbitals themselves — a macro-iteration cycle (see Orbital Optimization). This calculation converged after 26 such cycles:

MACRO-ITERATION  25:
   CI-PROBLEM SOLVED
MACRO-ITERATION  26:
   CI-PROBLEM SOLVED

Initial CI Check vs. Final Converged States

ORCA first diagonalizes the CI problem using the input RHF orbitals as-is, before any orbital optimization, as a sanity check:

<<<<<<<<<<<<<<<<<<INITIAL CI STATE CHECK>>>>>>>>>>>>>>>>>>
ROOT   0:  E=   -2042.3277166206 Eh
ROOT   1:  E=   -2042.1676129466 Eh  4.357 eV  35138.7 cm**-1
ROOT   2:  E=   -2042.1140409482 Eh  5.814 eV  46896.4 cm**-1

After the full 26 macro-iterations, the final, converged CASSCF state energies are lower and shifted:

CAS-SCF STATES FOR BLOCK  0 MULT= 1 NROOTS= 3
---------------------------------------------

ROOT   0:  E=   -2042.3470259034 Eh
ROOT   1:  E=   -2042.1679107518 Eh  4.874 eV  39311.2 cm**-1
ROOT   2:  E=   -2042.1618097383 Eh  5.040 eV  40650.2 cm**-1
State Energy (Eh) ΔE vs. S₀ ΔE vs. S₀
S₀ (ROOT 0) −2042.3470259034
S₁ (ROOT 1) −2042.1679107518 4.874 eV 39311.2 cm⁻¹
S₂ (ROOT 2) −2042.1618097383 5.040 eV 40650.2 cm⁻¹

The ground state (S₀) energy drops and the excitation energies shift noticeably between the initial guess and the final result — this is exactly what orbital optimization is for: the RHF orbitals were a good starting point, but CASSCF re-optimizes them specifically for the multi-state, multi-configurational problem at hand, not just for the ground state RHF describes well.

Configuration State Function (CSF) Weights

Each state is reported as a weighted sum over configuration state functions — the coefficients showing how multi-configurational (or not) each state actually is:

ROOT   0:  E=   -2042.3470259034 Eh
      0.91933 [     0]: 2200
      0.03086 [     3]: 2020
      0.02884 [     9]: 1111
      ...
ROOT   1:  E=   -2042.1679107518 Eh  4.874 eV  39311.2 cm**-1
      0.41147 [     1]: 2110
      0.28149 [     3]: 2020
      0.09592 [     6]: 1210
      0.08801 [     2]: 2101
      ...
ROOT   2:  E=   -2042.1618097383 Eh  5.040 eV  40650.2 cm**-1
      0.52643 [     1]: 2110
      0.21929 [     3]: 2020
      0.08399 [     2]: 2101
      ...

The 2200/2020/1111-style labels are occupation strings across the 4 active orbitals (2 = doubly occupied, 1 = singly occupied, 0 = empty). Reading these:

  • S₀ is well described by a single configuration (2200, 91.9% weight) — close to what a single-determinant method would already capture, which is why RHF gives a reasonable ground-state geometry and orbitals even though it can't describe the excited states at all.
  • S₁ and S₂ are genuinely multiconfigurational — no single CSF exceeds 53% weight, and both states mix the same handful of configurations (2110, 2020, 2101, 1210) in different proportions. This is precisely the situation Why Hartree–Fock Fails describes: a single Slater determinant simply cannot represent either excited state well, which is why CASSCF — not RHF — is needed to describe them at all.

Comparing With the GAMESS CASSCF Tutorial

The GAMESS CASSCF tutorial runs the same molecule, the same CAS(4,4) active space, and the same 3 states — the only differences are the basis set (CCD there vs. def2-SVP here) and the software. That makes the two directly comparable: the state ordering and the qualitative CSF character (a well-behaved S₀ against two strongly mixed excited states) should match; the absolute energies and excitation gaps will differ somewhat with the change of basis set. Working through both tutorials side by side is a useful check that the physics — not just the software — is being understood correctly.