Blocked-Davidson algorithm

From VASP Wiki

The workflow of the blocked-Davidson iterative matrix diagonalization scheme implemented in VASP is as follows:[1][2]

  • Take a subset (block) of [math]\displaystyle{ n_1 }[/math] orbitals out of the total set of NBANDS orbitals:
[math]\displaystyle{ \{ \psi_n| n=1,..,N_{\rm bands}\}\Rightarrow \{ \psi^1_k| k=1,..,n_1\} }[/math].
  • Extend the subspace spanned by [math]\displaystyle{ \{\psi^1\} }[/math] by adding the preconditioned residual vectors of [math]\displaystyle{ \{\psi^1\} }[/math]:
[math]\displaystyle{ \left \{ \psi^1_k \, / \, g^1_k = \left (1- \sum_{n=1}^{N_{\rm bands}} | \psi_n \rangle \langle\psi_n | {\bf S} \right) {\bf K} \left ({\bf H} - \epsilon_{\rm app} {\bf S} \right ) \psi^1_k \, | \, k=1,..,n_1 \right \}. }[/math]
  • Rayleigh-Ritz optimization ("subspace rotation") within the [math]\displaystyle{ 2n_1 }[/math]-dimensional space spanned by [math]\displaystyle{ \{\psi^1/g^1\} }[/math], to determine the [math]\displaystyle{ n_1 }[/math] lowest eigenvectors:
[math]\displaystyle{ {\rm diag}\{\psi^1/g^1\} \Rightarrow \{ \psi^2_k| k=1,..,n_1\} }[/math]
  • Extend the subspace with the residuals of [math]\displaystyle{ \{\psi^2\} }[/math]:
[math]\displaystyle{ \left \{ \psi^2_k \,/ \, g^1_k \, / \, g^2_k = \left (1- \sum_{n=1}^{N_{\rm bands}} | \psi_n \rangle \langle\psi_n | {\bf S} \right ) {\bf K} \left ({\bf H} - \epsilon_{\rm app} {\bf S} \right) \psi^2_k \, | \, k=1,..,n_1 \right \}. }[/math]
  • Rayleigh-Ritz optimization ("subspace rotation") within the [math]\displaystyle{ 3n_1 }[/math]-dimensional space spanned by [math]\displaystyle{ \{\psi^1/g^1/g^2\} }[/math]:
[math]\displaystyle{ {\rm diag}\{\psi^1/g^1/g^2\} \Rightarrow \{ \psi^3_k| k=1,..,n_1\} }[/math]
  • If need be the subspace may be extended by repetition of this cycle of adding residual vectors and Rayleigh-Ritz optimization of the resulting subspace:
[math]\displaystyle{ {\rm diag}\{\psi^1/g^1/g^2/../g^{d-1}\}\Rightarrow \{ \psi^d_k| k=1,..,n_1\} }[/math]
Per default VASP will not iterate deeper than [math]\displaystyle{ d=4 }[/math], though it may break off even sooner when certain criteria that measure the convergence of the orbitals have been met.
  • When the iteration is finished, store the optimized block of orbitals back into the set:
[math]\displaystyle{ \{ \psi^d_k| k=1,..,n_1\} \Rightarrow \{ \psi_k| k=1,..,N_{\rm bands}\} }[/math].
  • Move on to the next block [math]\displaystyle{ \{ \psi^1_k| k=n_1+1,..,2 n_1\} }[/math].
  • When LDIAG=.TRUE. (default), a Rayleigh-Ritz optimization in the complete subspace [math]\displaystyle{ \{ \psi_k| k=1,..,N_{\rm bands}\} }[/math] is performed after all orbitals have been optimized.


The blocksize [math]\displaystyle{ n_1 }[/math] used in the blocked-Davidson algorithm can be set by means of the NSIM tag. In principle [math]\displaystyle{ n_1= 2\times }[/math] NSIM, but for technical reasons it needs to be dividable by an integer N:

[math]\displaystyle{ n_1 = {\rm int}\left(\frac{2*{\rm NSIM} + N - 1}{N}\right) N }[/math]

where [math]\displaystyle{ N }[/math] is the "number of band groups per k-point group":

[math]\displaystyle{ N = \frac{{\rm \#\; of\; MPI\; ranks}}{{\rm IMAGES}*{\rm KPAR}*{\rm NCORE}} }[/math]

(see the section on parallelization basics).

As mentioned before, the optimization of a block of orbitals is stopped when either the maximum iteration depth (NRMM), or a certain convergence threshold has been reached. The latter may be fine-tuned by means of the EBREAK, DEPER, and WEIMIN tags. Note: we do not recommend you to do so! Rather rely on the defaults instead.

The blocked-Davidson algorithm is approximately a factor of 1.5-2 slower than the RMM-DIIS, but more robust.

References