Selecting an appropriate energy cut off for response functions (emaxrf)

Elk Users
Rob N
2014-01-30
2014-01-30
  • Rob N
    Rob N
    2014-01-30

    I believe it was Markus who first suggested I reduce this parameter to save some computational expense.

    How do I determine what is sensible? Is there an output file I can check that can help me determine a reasonable value? The default is 10^6 Hartree (per the manual) which seems to be an extremely large energy in comparison to all other energy parameters, such as the Fermi Energy (0.233 hartree), and the energy window I am calculating (1.5 Hartree).

    I could just start submitting jobs with different values for emaxrf; however, I suspect my outputs can assist me in gauging what is reasonable.

    Best,

    RN

     
  • Markus
    Markus
    2014-01-30

    Hi Rob,

    yes, there is an output file that may help you: EIGVAL.OUT, which contains the k-resolved eigenvalues. emaxrf sets the energy window of states to be included in the response function calculation in the TDDFT code. So as long as emaxrf is larger than (the modulus of) the largest unoccupied and the largest occupied eigenvalue, it does nothing. Setting a smaller value will exclude all states larger than emaxrf, so you may remove semicore states or very high-lying empty states from the response function calculation without having to redo the ground state calculation.

    Anyway, the "best" value depends on the system and you have to reduce it step by step to see what happens if you need to save some time.

    Regards
    Markus

     
  • Rob N
    Rob N
    2014-01-30

    Markus,

    I suspected this is where I should look and had actually already looked through this file and the values were 5 orders of magnitude less than the default value of 10^6.

    I'm assuming when you say the largest occupied eigenvalue, you mean the largest with any occupation at all, not a full occupancy of two. So for this system I have..
    156 0.7190164185 0.5483233294E-21
    The largest unoccupied state is
    517 2.778499090 0.000000000

    I'm also going to assume when you say modulus you mean the absolute value of the ratio and not the modulo definition (remainder when dividing)

    that gives >> 2.778499090/0.7190164185 = 3.8643 which is suspiciously smaller than the default value of 10^6 and leads me to believe I made some poor assumptions above.

    Best,

    RN

     
  • Markus
    Markus
    2014-01-30

    Sorry, of course I meant the lowest occupied and highest unoccupied state. emaxrf is simply the energy window around the Fermi energy taken into account in the calcuation.