[CPMD-list] ANDERSON vs. BROYDEN

[Eung-Gun Kim]

Dear List Subscribers,

While comparing the performance between ANDERSON MIXING and BROYDEN MIXING, I've
noticed some inconsistency in the behavior of LANCZOS DIAGO (with KPOINTS) when
BROYDEN is used.  As seen in an example below, although the calculation has only
reached DRHOMAX = 2.404E-03, FRIESNER iterations continue until B2MAX =
9.991E-08, which is below the tolerance for DRHOMAX = 5.0E-04.  However, this
does not happen with ANDERSON.  Are LANCZOS PARAMETERS expected to behave
differently with BROYDEN?  Thanks very much for reading.

Regards,

EG Kim

Portions of the output file follow @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
.
.
.
 LANCZOS DIAGONALIZATION (KRYLOV SUBSPACE)
    MAX. FRIESNER ITERATIONS                                   50
    MAX. KRYLOV SUBSPACE                                        8
    MAX. KRYLOV BLOCK SIZE                                     50
    MAX. BETA^2                                        1.0000E-04
    MAX. BETA^2 CHOICE NUMBER:                                  5
    BELOW DRHOMAX=  1.0000E-02           MAX. BETA^2=  1.0000E-05
    BELOW DRHOMAX=  1.5000E-03           MAX. BETA^2=  1.0000E-06
    BELOW DRHOMAX=  5.0000E-04           MAX. BETA^2=  1.0000E-07
    BELOW DRHOMAX=  1.2500E-04           MAX. BETA^2=  1.0000E-08
.
.
.
.

 ==------------------------------------------------------------==
 ==     NFI=     21      ETOT=   -508.768656   TCPU=    171.65 ==
 == DRHOMAX= 2.404E-03  DETOT=     1.060E-05    THL= 0.000E+00 ==
 ==------------------------------------------------------------==

 <<1:6<<<<<<<<<<<<<< LANCZOS DIAGONALIZATION <<<<<<<<<<<<<<<<<<<<
 >> TIME FOR INITIAL SUBSPACE DIAGONALIZATION:               3.93
 >> CYCLE     NCONV        B2MAX        B2MIN     #HPSI      TIME
        1       200    9.991E-08    2.287E-08      1.00       .63

 <<2:6<<<<<<<<<<<<<< LANCZOS DIAGONALIZATION <<<<<<<<<<<<<<<<<<<<