:orphan: Atomic shift operator ===================== DIRAC allows the definition of an atomic shift operator :ref:`HAMILTONIAN_.ASHIFT` to be added to the Hamiltonian .. math:: V = \sum_{A}\sum_{I\in A}\sum_{i\in I}|\varphi^A_i\rangle \varepsilon^A_I \langle \varphi^A_i| where :math:`A` is a sum over atomic types. For each atomic types a group :math:`I` of orbitals may be specified with an associated shift parameter :math:`\varepsilon^A_I`. In this manual we shall see an example of its use. Degeneracy-driven covalency: the case of CO ------------------------------------------- We consider carbon monoxide. We run a straightforward HF calculation using the input files *CO.xyz* .. literalinclude:: CO.xyz and *CO.inp* .. literalinclude:: CO.inp using the command:: pam --inp=CO --mol=CO.xyz Looking at the Mulliken population analysis in the output, we find for the lowest two occupied molecular orbitals:: * Electronic eigenvalue no. 1: -20.684805978278 (Occupation : f = 1.0000) m_j= 1/2 ============================================================================================ * Gross populations greater than 0.01000 Gross Total | L A1 O s -------------------------------------- alpha 0.9995 | 0.9992 beta 0.0005 | 0.0000 * Electronic eigenvalue no. 2: -11.369805624916 (Occupation : f = 1.0000) m_j= 1/2 ============================================================================================ * Gross populations greater than 0.01000 Gross Total | L A1 C s -------------------------------------- alpha 0.9997 | 0.9990 beta 0.0003 | 0.0000 We see that the first and second lowest occupied MO are :math:`O1s` and :math:`C1s`, respectively, separated by a sizable energy of :math:`9.315\ E_h`. Let us now see what happens if we bring the two atomic core orbitals into resonance. First we generate the carbon atomic orbitals, using input files *C.inp* .. literalinclude:: C.xyz and *C.inp* .. literalinclude:: C.inp and the command:: pam --inp=C --mol=C.xyz The checkpoint file *C_C.h5* contains the :math:`C_1` coefficients:: h5dump -n C_C.h5 | grep C1 dataset /result/wavefunctions/scf/mobasis/eigenvalues_C1 dataset /result/wavefunctions/scf/mobasis/orbitals_C1 We next prepare the input file *COshift.inp* .. literalinclude:: COshift.inp which notably defines an atomic shift operator for carbon which builds a projector from the :math:`C1s` orbital and shifts it down by :math:`9.315\ E_h`. We run our calculation using the command:: pam --inp=COshift --mol=CO.xyz --put "ac.C=AFCXXX" The Mulliken population analysis of the two lower orbitals now read:: * Electronic eigenvalue no. 1: -20.687541416409 (Occupation : f = 1.0000) m_j= 1/2 ========================================================================================== * Gross populations greater than 0.00010 Gross Total | L A1 C s L A1 O s A1 O _small B1 O _small B2 O _small -------------------------------------------------------------------------------------------------- alpha 0.9996 | 0.4790 0.5203 0.0001 0.0000 0.0000 beta 0.0004 | 0.0000 0.0000 0.0000 0.0001 0.0001 * Electronic eigenvalue no. 2: -20.681935936276 (Occupation : f = 1.0000) m_j= 1/2 ========================================================================================== * Gross populations greater than 0.00010 Gross Total | L A1 C s L A1 O s A1 O _small B1 O _small B2 O _small -------------------------------------------------------------------------------------------------- alpha 0.9996 | 0.5202 0.4790 0.0001 0.0000 0.0000 beta 0.0004 | 0.0000 0.0000 0.0000 0.0001 0.0001 and we indeed see that the two lowest molecular orbitals now are almost degenerate with concomitant strong mixing of the :math:`O1s` and :math:`C1s` orbitals. By scanning shift values in the resonance region we obtain the following plot of orbital energies of the two lowest occupied molecular orbitals .. image:: COshift_eig.png :width: 600px as well as the mixing of :math:`O1s` and :math:`C1s` orbitals in the LOMO. .. image:: COshift.png :width: 600px At this point it may be useful to consider the diagonalization of a :math:`2\times2` matrix .. math:: \left[\begin{array}{cc}\varepsilon_1&\delta\\\delta&\varepsilon_2\end{array}\right], where eigenvalues are given by .. math:: \lambda_\pm = \frac{1}{2}\left[(\varepsilon_1+\varepsilon_2)\pm\sqrt{(\varepsilon_1-\varepsilon_2)^2 +4\delta^2}\right];\quad\Delta\lambda=\sqrt{(\varepsilon_1-\varepsilon_2)^2 +4\delta^2} We notably see that as we bring the two starting energies into resonance, that is :math:`\varepsilon_1=\varepsilon_2`, the difference :math:`\Delta\lambda` becomes equal to the twice the interaction :math:`2|\delta|`. In this particular case :math:`2|\delta|=5.6\ mE_h`. What can learn from this ? Neidig, Clark and Martin :cite:`Neidig_CoorChemRev2013` have introduced the concept of *near-degeneracy driven covalency*, where bonding interaction arises from near-degeneracy, even in the absence of overlap. In the above example we clearly see this mechanism at work: bringing the :math:`O1s` and :math:`C1s` orbitals into resonance causes strong orbital mixing. *However*, it would be absurd to call this bonding. The two atomic core orbitals can be disentangled by localization and we also note that their interaction is extremely weak.