NELMDL: Difference between revisions
No edit summary |
No edit summary |
||
| Line 6: | Line 6: | ||
If the orbitals are initialized using a random number generator (the default in VASP), the initial orbitals are usually unreasonable and the iterative matrix diagonalization will required 5-10 steps to obtain reasonable orbitals. The charge density corresponding to the initial orbitals is also, at best, erratic. It is hence advisable to perform a few electronic steps while keeping the initial Hamiltonian fixed. This initial Hamiltonian is usually determined from a superposition of atomic charge densities (see {{TAG|ICHARG}}). | If the orbitals are initialized using a random number generator (the default in VASP), the initial orbitals are usually unreasonable and the iterative matrix diagonalization will required 5-10 steps to obtain reasonable orbitals. The charge density corresponding to the initial orbitals is also, at best, erratic. It is hence advisable to perform a few electronic steps while keeping the initial Hamiltonian fixed. This initial Hamiltonian is usually determined from a superposition of atomic charge densities (see {{TAG|ICHARG}}). | ||
Such a 'delay' is absolutely necessary. if the SCF-convergence is | Such a 'delay' is absolutely necessary. if the SCF-convergence is slow and problematic (e.g. for surfaces or metallic clusters, low dimensional system). Without a delay, VASP will most likely not converge or at least the convergence speed is slowed significantly. | ||
{{TAG|NELMDL}} might be set to a positive or negative value. A negative value | {{TAG|NELMDL}} might be set to a positive or negative value. A negative value means that the delay is only performed in the first ionic step (usually the recommended option). A positive number means that a delay is employed after each ionic movement. This can improve the convergence speed in vasp.6 (see below), but is not recommended in vasp.5. | ||
For calculations using a direct minimization of the Hamiltonian ({{TAG|ALGO}}=ALL or {{TAG|ALGO}}=DAMPED), the Davidson algorithm is used during the delay phase and the Hamiltonian is kept fixed during these steps. | For calculations using a direct minimization of the Hamiltonian ({{TAG|ALGO}}=ALL or {{TAG|ALGO}}=DAMPED), the Davidson algorithm is used during the delay phase and the Hamiltonian is kept fixed during these steps. | ||
| Line 14: | Line 14: | ||
VASP.6 special considerations: | VASP.6 special considerations: | ||
* For calculations using a direct minimization of the Hamiltonian ({{TAG|ALGO}}=ALL or {{TAG|ALGO}}=DAMPED): if {{TAG|NELMDL}} is set, the Davidson algorithm is used in the first {{TAG|NELMDL}} steps. Using a positive {{TAG|NELMDL}} (i.e. delay in every ionic step), does not work reliably though in vasp.5, due to issues in the orbital and charge density prediction. In vasp.6, using {{TAG|NELMDL}}=1 (or {{TAG|NELMDL}}=2) and direct minimization, often improves the stability and efficiency of molecular dynamics simulations simulations or relaxations | * For calculations using a direct minimization of the Hamiltonian ({{TAG|ALGO}}=ALL or {{TAG|ALGO}}=DAMPED): if {{TAG|NELMDL}} is set, the Davidson algorithm is used in the first {{TAG|NELMDL}} steps as described above. Using a positive {{TAG|NELMDL}} (i.e. delay in every ionic step), does not work reliably though in vasp.5, due to issues in the orbital and charge density prediction. In vasp.6, using {{TAG|NELMDL}}=1 (or {{TAG|NELMDL}}=2) and direct minimization, often improves the stability and efficiency of molecular dynamics simulations simulations or relaxations ({{TAG|ALGO}}=ALL or {{TAG|ALGO}}=DAMPED=. Note, however, that this might require one the prepare a reasonable {{TAG|WAVECAR}} file, since {{TAG|NELMDL}} =1/2 might not suffice to obtain a reasonable set of orbitals from the initial random numbers. | ||
* For HF type calculations, if {{TAG|NELMDL}} is larger or equal 3, VASP will perform {{TAG|NELMDL}} non-selfconsistent steps using the Davidson algorithm and a local Hamiltonian calculated using the semi-local DFT functional corresponding to the chosen hybrid functional (i.e. PBE for HSE and PBE0). This is expedient, if the ions move by a large distance between the ionic steps. Setting {{TAG|NELMDL}} =3, can thus improve the stability and performance | * For HF type calculations, if {{TAG|NELMDL}} is larger or equal 3, VASP will perform {{TAG|NELMDL}} non-selfconsistent steps using the Davidson algorithm and a local Hamiltonian calculated using the semi-local DFT functional corresponding to the chosen hybrid functional (i.e. PBE for HSE and PBE0). This is expedient, if the ions move by a large distance between the ionic steps. Setting {{TAG|NELMDL}} =3, can thus improve the stability and performance during relaxations using HF type Hamiltonians. Try to use {{TAG|ALGO}}=All and {{TAG|NELMDL}}=3, if you encounter convergence issues during relaxations using HF type Hamiltonians. | ||
| Line 22: | Line 22: | ||
{{TAG|NELM}}, | {{TAG|NELM}}, | ||
{{TAG|NELMIN}} | {{TAG|NELMIN}} | ||
{{TAG|IALGO}} | |||
{{sc|NELMDL|Examples|Examples that use this tag}} | {{sc|NELMDL|Examples|Examples that use this tag}} | ||
Revision as of 12:03, 12 August 2019
NELMDL = [integer]
| Default: NELMDL | = -5 | if ISTART=0, INIWAV=1, and IALGO=8 |
| = -12 | if ISTART=0, INIWAV=1, and IALGO=48 |
Description: NELMDL specifies the number of non-selfconsistent steps at the beginning.
If the orbitals are initialized using a random number generator (the default in VASP), the initial orbitals are usually unreasonable and the iterative matrix diagonalization will required 5-10 steps to obtain reasonable orbitals. The charge density corresponding to the initial orbitals is also, at best, erratic. It is hence advisable to perform a few electronic steps while keeping the initial Hamiltonian fixed. This initial Hamiltonian is usually determined from a superposition of atomic charge densities (see ICHARG).
Such a 'delay' is absolutely necessary. if the SCF-convergence is slow and problematic (e.g. for surfaces or metallic clusters, low dimensional system). Without a delay, VASP will most likely not converge or at least the convergence speed is slowed significantly.
NELMDL might be set to a positive or negative value. A negative value means that the delay is only performed in the first ionic step (usually the recommended option). A positive number means that a delay is employed after each ionic movement. This can improve the convergence speed in vasp.6 (see below), but is not recommended in vasp.5.
For calculations using a direct minimization of the Hamiltonian (ALGO=ALL or ALGO=DAMPED), the Davidson algorithm is used during the delay phase and the Hamiltonian is kept fixed during these steps.
VASP.6 special considerations:
- For calculations using a direct minimization of the Hamiltonian (ALGO=ALL or ALGO=DAMPED): if NELMDL is set, the Davidson algorithm is used in the first NELMDL steps as described above. Using a positive NELMDL (i.e. delay in every ionic step), does not work reliably though in vasp.5, due to issues in the orbital and charge density prediction. In vasp.6, using NELMDL=1 (or NELMDL=2) and direct minimization, often improves the stability and efficiency of molecular dynamics simulations simulations or relaxations (ALGO=ALL or ALGO=DAMPED=. Note, however, that this might require one the prepare a reasonable WAVECAR file, since NELMDL =1/2 might not suffice to obtain a reasonable set of orbitals from the initial random numbers.
- For HF type calculations, if NELMDL is larger or equal 3, VASP will perform NELMDL non-selfconsistent steps using the Davidson algorithm and a local Hamiltonian calculated using the semi-local DFT functional corresponding to the chosen hybrid functional (i.e. PBE for HSE and PBE0). This is expedient, if the ions move by a large distance between the ionic steps. Setting NELMDL =3, can thus improve the stability and performance during relaxations using HF type Hamiltonians. Try to use ALGO=All and NELMDL=3, if you encounter convergence issues during relaxations using HF type Hamiltonians.