ORCA

Table of Contents

- no GPU support in ORCA 4.x
- Known Bug: Analytic Hessian with RI for (ORCA <= 4.2.1)
- VPT2
- The anharmonic treatment needs a method which is implemented with analytical hessian, otherwise the results are dominated by numerical noise.
- A note on the side and a remark for VPT2 - small vibrational frequencies are always troublesome. Small frequency means shallow potential and hence the molecule can move quite a bit in its zero-point motion, so it is likely to experience anharmonicity in the potential. VPT2, as a perturbative method, naturally has a problem when the perturbation increases, so often one finds that when you need it the most, it does not perform so well. AIMD with snapshot sampling is an -expernsive- alternative here. For temperature corrections using VPT2 I even observed that frequencies below 50-100 wavenumbers have to be excluded as the VPT2 expression becomes singular for small frequencies ...

- [2021 - JCP - DLPNO-MP2 second derivatives for the computation of polarizabilities and NMR shieldings]
- Harmonic frequency calculations are not available yet, since an efficient evaluation of Hessian is needed.

- ORCA has been optimized for best single node performance, not for multi nodes.
- External packages:
- basic linear algebra system (BLAS)
- message passing interface (MPI)
- linear algebra package (LAPACK)
- LIBINT

System | Charge | Multiplicity |
---|---|---|

neutral molecule | 0 | 1 |

protonated PAHs C16H15+ | 1 | 1 |

cation H2O+ | 1 | 2 |

radical cation PAHs C16H14+ | 1 | 2 |

- Linear molecules have to align with z-axis for vibrational analysis |

$$\text{multiplicity} = \text{number of } \alpha \text{ electrons} - \text{number of } \beta \text{ electrons} + 1$$ - When #electrons is odd, then multiplicity must be even (2, 4, 6, ...). - When #electrons is even, then multiplicity must be old (1, 3, 5, ...).

`NOPOP`

: Turns off all population analysis (Mulliken, Loewdin, Mayer)

- Approximations to the Hartree-Fock exchange term
- Two electron integrals: the first integral is performed numerically on the grid and the second is done analytically.

`M06-2x`

should use`D3zero`

instead of`D3BJ`

D3不应当理解为是用来描述色散作用的，而是对你选择的泛函基础上进行校正以提高精度，只不过是对于色散作用描述极差的诸如B3LYP，D3主要作用才是起到描述色散作用的作用。D3的参数对每种泛函是不同的。M06-2X已经能很好描述色散作用，所以加了D3只会稍微改进一点精度，完全不会double-counting。明尼苏达系列泛函不能用BJ阻尼的D3是因为BJ阻尼下必然导致double-counting。这些问题Grimme都已经充分考虑了。 - [量化理论] M06-2X加D3校正的问题

- It is recommanded to use
`Grid6`

for minnesota meta-GGA

Using MN15 functional in ORCA 4.2.1 from libxc

! RIJCOSX def2-TZVP def2/J Opt nopop %scf maxiter 600 end %method method dft exchange hyb_mgga_x_mn15 correlation mgga_c_mn15 end %maxcore 1000 %pal nprocs 16 end *xyz 0 1 C -0.047131 0.664389 0.0 O -0.047131 -0.758551 0.0 H -1.092995 0.969785 0.0 H 0.878534 -1.048458 0.0 H 0.437145 1.080376 0.891772 H 0.437145 1.080376 -0.891772 *

- https://sites.google.com/site/orcainputlibrary/dft/broken-symmetry-dft
- https://chemistry.stackexchange.com/questions/42005/initial-guess-for-unrestricted-hartree-fock-calculation
- http://wwwuser.gwdg.de/~ggroenh/exercise_H2/h2.html
- http://sobereva.com/474

Complete Active Space Self-Consistent Field (CASSCF) - CASSCF(N,M): N-electrons are active in M orbitals.

%casscf nel N norb M end

- CASSCF provides analytical gradients, so it may be used for structure optimizations and numerical frequency calculations.

ORCA 5.0.0 and 5.0.1 has a bug in interpreting the second Colvar. This can be solved by adding 3 more zeros in the end.

Metadynamics Colvar 1 Scale 1.0_A Wall Lower 0.0 50.0 Upper 10.0 50.0 Range 0.0 10.0 100 Metadynamics Colvar 2 Scale 1.0_A Wall Lower 0.0 50.0 Upper 10.0 50.0 Range 0.0 10.0 100 0 0 0

Alternatively, putting two Colvars in the same line also solves this bug

Metadynamics Colvar 1 Scale 1.0_A Wall Lower 0.0 50.0 Upper 10.0 50.0 Colvar 2 Scale 1.0_A Wall Lower 0.0 50.0 Upper 10.0 50.0 Range 0.0 10.0 100 0.0 10.0 100

- check geometry optimization convergence:
`grep 'The optimization did not converge' orca.out`

One has to create `JOBNAME.nodes`

in the running directory to run on multiple nodes.
`echo $SLURM_NODELIST >> JOBNAME.nodes`

will create `s31n[07-09]`

which cannot be recognized by ORCA.
The following script gives the correct multiple nodes setting.

#SBATCH --nodelist=s31n[07-09] #SBATCH --ntasks=36 #SBATCH --ntasks-per-node=12 #SBATCH --cpus-per-task=1 # change directory to the local scratch-directory, and run: cd /scratch/$SLURM_JOB_ID cat > orca.nodes <<! s31n07 slots=12 s31n08 slots=12 s31n09 slots=12 !

WARNING: For RIJCOSX frequencies GRIDX4 or larger grids are recommended

[file orca_main/mainfcts.cpp, line 625]: Error : multiplicity (2) is even and number of electrons (204) is even -> impossible

---------------------------------------------------------------------------- ERROR !!! The optimization did not converge but reached the maximum number of optimization cycles. As a subsequent Frequencies calculation has been requested ORCA will abort at this point of the run. Please restart the calculation with the lowest energy geometry and/or a larger maxiter for the geometry optimization. ---------------------------------------------------------------------------- [file orca_main/run.cpp, line 10680]: ORCA finished with error return - aborting the run