<html>
  <head>

    <meta http-equiv="content-type" content="text/html; charset=UTF-8">
  </head>
  <body text="#000000" bgcolor="#FFFFFF">
    <div class="moz-forward-container">
      <div class="moz-forward-container">
        <p>    Thank you very much Pietro, this actually fixed the issue
          ! The frustrating stalling behaviour in the convergence has
          been suppressed, and I can now reach convergence within 1e-8Ry
          in less than 100 iterations.<br>
        </p>
        <p><br>
        </p>
        <p>    If I am not abusing of your time, can I ask you how you
          spotted that this particular PP was the issue? Gaining some
          insight on what was going wrong or how to debug those problems
          would be very benefitial for me.</p>
        <p><br>
        </p>
        <p>Thanks again for the help.</p>
        <p>Julien<br>
        </p>
        <div class="moz-cite-prefix">Le 01/03/2019 à 11:31, Pietro
          Davide Delugas a écrit :<br>
        </div>
        <blockquote type="cite"
          cite="mid:54b911fe-f433-8186-ca49-c3ef13286fa3@sissa.it">
          <meta http-equiv="Content-Type" content="text/html;
            charset=UTF-8">
          <div class="moz-cite-prefix">Hi Julien <br>
            <br>
            I checked your input, and  <a class="element_anchor"
              target="_blank"
href="http://www.quantum-espresso.org/upf_files/N.pbesol-n-kjpaw_psl.1.0.0.UPF"
              moz-do-not-send="true">N.pbesol-n-kjpaw_psl.1.0.0.UPF</a>
            seems to be the cause of your problems, you can use   this
            one instead  <br>
            <a class="moz-txt-link-freetext"
href="http://www.quantum-espresso.org/upf_files/N.pbesol-n-kjpaw_psl.0.1.UPF"
              moz-do-not-send="true">http://www.quantum-espresso.org/upf_files/N.pbesol-n-kjpaw_psl.0.1.UPF</a>
            <br>
            <br>
            and you should  have no more problems  with negative charges
            and have a more or less fast convergence. <br>
                Pietro <br>
            <br>
            On 02/28/2019 01:27 PM, Julien Barbaud wrote:<br>
          </div>
          <blockquote type="cite"
            cite="mid:82632677-2321-3397-a229-16e00eb191af@sjtu.edu.cn">
            <meta http-equiv="Content-Type" content="text/html;
              charset=UTF-8">
            <p>Thanks very much to everyone for their answers,</p>
            <p>I have tried the different fixes suggested (higher ecuts,
              trying out different smearings, raising mixing_ndim,
              changing the mixing_mode, etc...). Unfortunatley, none of
              them worked...<br>
            </p>
            <p>Following Pietro's recommandation, I also designed an
              input geometry perfectly cubic with no distortion and the
              MA+ ion oriented along the [1 1 1] direction for higher
              symmetry. I based the quantitative part of the geometry on
              other cif files available (i.e. for the lattice constant,
              bond lengths, etc...)</p>
            <p>Benefitting from the high symmetry, the algorithm worked
              faster, but was still unable to converge to a threshold of
              1e-7Ry in 500 iterations...</p>
            <p><br>
            </p>
            <p>I finally relaxed the condition to 1e-6Ry in order to
              reach a scf convergence, and see if I could get more
              informations on the problem from it. <br>
            </p>
            <p>Under those conditions I plotted the scf total energy
              (converged to 1e-6Ry) as a function of ecut. I got the
              curve attached. This seems surprising to me because the
              curve is raising past 45Ry<br>
            </p>
            <p>I thought that the total energy would only have a
              monotonic trend, decreasing with higher ecut, as a
              consequence of the variational principle (indeed, the
              energy of the ground state should be a minimum for all
              possible energies, so a truncated version of the ground
              state wavefunction should yield a higher value of energy).
              Here it seems to break that rule, and to "converge" to a
              value that is not the minimum among all possible
              wavefunctions. This can not be attributed to the accuracy
              of the calculation, because the scf is converged down to
              1e-6Ry, and the augmentation observed is significantly
              above that threshold<br>
            </p>
            <p>Did I misunderstand something, or is it another sign that
              something is seriously wrong with the calculation ?</p>
            <p><br>
            </p>
            <p>(by the way, the curve stops after 65Ry, because scf
              failed to reach convergence again at ec=70Ry, even for a
              threshold at 1e-6 and 500 iterations).</p>
            <p>What other options should I try to solve this problem ?</p>
            <p><br>
            </p>
            <p>Julien<br>
            </p>
            <p><br>
            </p>
            <p><br>
            </p>
            <div class="moz-cite-prefix">Le 22/02/2019 à 16:45, Pietro
              Delugas a écrit :<br>
            </div>
            <blockquote type="cite"
              cite="mid:0244887b-b39c-55c3-471a-2a4df142d68d@sissa.it">
              <meta http-equiv="Content-Type" content="text/html;
                charset=UTF-8">
              <p><font size="-1">Dear Julien <br>
                </font></p>
              <p><font size="-1">even if the scf loop converges you have
                  still to check that the k-point sampling and the plane
                  wave basis set guarantee you an accurate result.  <br>
                </font></p>
              <p><font size="-1"> obviously before worrying about
                  accuracy you would like to have a converged density. <br>
                </font></p>
              <p><font size="-1">You could try to start with  a more 
                  symmetric  cell, use a  cubic cell without distortions
                  and   align the molecule along one of the diagonals of
                  the perovskite box.  <br>
                </font></p>
              <p><br>
              </p>
              <div class="moz-cite-prefix">On 22/02/19 08:22, Julien
                Barbaud wrote:<br>
              </div>
              <blockquote type="cite"
                cite="mid:5ed6547f-f86e-5f57-ea86-06805e1a99fd@sjtu.edu.cn">
                <meta http-equiv="content-type" content="text/html;
                  charset=UTF-8">
                <p> </p>
                <div class="moz-forward-container">
                  <p>Thank you Pietro for your experienced advices,</p>
                  <p><br>
                  </p>
                  <p>I had tried to increase the kmesh size before but
                    only up to sizes of  7x7x7.  Reading your
                    suggestions, I ran additional tests up to 10x10x10
                    but this did not show any sign of improvement on 70
                    iterations. As shown in file kmesh.png, the
                    estimated accuracy is still stagnating after a while
                    and the 10*10*10 is actually giving arguably worse
                    results than the 9*9*9 although this is most likely
                    not significant. Actually, some papers report DFT
                    simulation of MAPbI3 using 6x6x6 kmesh (<a
                      class="moz-txt-link-freetext"
                      href="https://aip.scitation.org/doi/full/10.1063/1.4864778"
                      moz-do-not-send="true">https://aip.scitation.org/doi/full/10.1063/1.4864778</a>),
                    or even single gamma-point calculation (<a
                      class="moz-txt-link-freetext"
href="http://people.bath.ac.uk/aw558/publications/2013/aplm_perovskite_13.pdf"
                      moz-do-not-send="true">http://people.bath.ac.uk/aw558/publications/2013/aplm_perovskite_13.pdf</a>),
                    so I guess this should not be the obstacle to
                    convergence here.</p>
                  <p><br>
                  </p>
                  <p>Regarding the orientation of MA, I definitely agree
                    with you, but I don't think it can prevent the
                    system from converging ? Sure enough, it can have an
                    important influence on the precision of the results
                    in later uses. But I would like to achieve
                    convergence on this simple single cell first, before
                    building up supercells to take more complex effects
                    into account. A crystal with perfectly aligned MA
                    might not reflect the true experimental system, but
                    it should still be a possible configuration that the
                    QE code should be able to compute, am I wrong ?</p>
                  <p><br>
                  </p>
                  <p>As to your suggestion on VdW corrections, I just
                    gave it a try, but unfortunately, this is
                    unconclusive too. I report the accuracy at each
                    iteration in vdw.png. Again, the accuracy stops
                    improving after a while. Plese note that I had to
                    change my pseudo-potentials to use 'xdm' correction
                    (which only supports PAW PP). the input file for
                    this test is included as attached file</p>
                  <p><br>
                  </p>
                  <p>Julien<br>
                  </p>
                  <div class="moz-cite-prefix">Le 21/02/2019 à 16:35,
                    Pietro Davide Delugas a écrit :<br>
                  </div>
                  <blockquote type="cite"
                    cite="mid:fb09ead9-e310-4142-0c01-7c8af7a8ab46@sissa.it">
                    <meta http-equiv="Content-Type" content="text/html;
                      charset=UTF-8">
                    <div class="moz-cite-prefix">Hi <br>
                      <br>
                      Have you tried to increase the k_point mesh ?  4 4
                      4 seems a little bit lax as mesh for MAPbI3. <br>
                      If I remember well I am afraid that to get
                      convergence you will need something like 10X10X10.<br>
                      As for the structure neighboring methylammoniums 
                      like to orient differently one from the other, you
                      should probably use a larger cell.   Also consider
                      to add some correction for van der Waals
                      interactions see here (  <a
                        class="moz-txt-link-freetext"
href="https://www.quantum-espresso.org/Doc/INPUT_PW.html#idm45922794348896"
                        moz-do-not-send="true">https://www.quantum-espresso.org/Doc/INPUT_PW.html#idm45922794348896</a>)<br>
                      <br>
                      hope it helps <br>
                      Pietro  <br>
                      <br>
                       <br>
                      On 02/21/2019 04:17 AM, Julien Barbaud wrote:<br>
                    </div>
                    <blockquote type="cite"
                      cite="mid:a55d199b-54b2-09be-4cc5-9826e0151a09@sjtu.edu.cn">
                      <meta http-equiv="content-type"
                        content="text/html; charset=UTF-8">
                      Dear users,
                      <div class="moz-forward-container">
                        <div class="moz-forward-container">
                          <p><br>
                          </p>
                          <p>I am new to QE, and trying to run a simple
                            scf calculation on a CH3NH3PbI3 crystal
                            (semi-conducting material). I am using
                            ultrasoft pseudopotentials based on the
                            exchange-correlation functionnal PBEsol.<br>
                          </p>
                          <p> I set up a first input, with values of
                            parameters inspired from literature on the
                            subject. However, I could not reach
                            convergence after 100 iterations. The
                            estimated error was actually "exploding" to
                            very high values, indicating a serious
                            problem. I tried several changes but was
                            unsuccessful: <br>
                          </p>
                          <ul>
                            <li> varying plane-wave cutoff energy does
                              not solve the problem (cf attached
                              ecut.png, giving the estimated error as a
                              function of the number of iterations. It
                              is shown here only on the first 15
                              iterations as the results pretty much only
                              stall from there)</li>
                            <li> varying cutoff energy for charge (cf
                              ecutrho.png)</li>
                            <li>taking larger k-point sampling (not
                              shown)</li>
                            <li> I also read that for metallic or "close
                              to metallic conductors", there might be
                              problems with the first unoccupied states
                              that can be solved by adding a few empty
                              bands. My system being a semi-conductor, I
                              tried adding additional bands using a m-p
                              smearing but no improvement was found (not
                              shown)</li>
                          </ul>
                          <p><br>
                          </p>
                          <p>The only change that I found effective was
                            to reduce the mixing_beta factor.<br>
                          </p>
                          <p><br>
                          </p>
                          <p>It effectively prevents the error from
                            diverging to very large values, but I still
                            do not reach convergence, even after longer
                            iterations. I tried much smaller values of
                            mixing beta which improves the final value
                            of the error, but I still cannot reach
                            convergence on 100 iterations. As shown in
                            the mixbeta2_zoom.png, the error reduces to
                            smaller values around ~1e-5~1e-6, but it
                            keeps stalling after a while. I do not
                            observe a well-converging behaviour for any
                            value.</p>
                          <p><br>
                          </p>
                          <p>I attached the "default version" of my
                            script on which the various modifications
                            described above have been independently
                            performed. I obtained the geometry from a
                            CIF file in literature and checked it with
                            visualization software; it seems perfectly
                            ok as far as I can tell.<br>
                          </p>
                          <p><br>
                          </p>
                          <p>Any insight on what I did wrong would be
                            really helpful. I suspect a shameful
                            beginner mistake, but can not find it out.<br>
                          </p>
                          <p><br>
                          </p>
                          <p>Thanks in advance,</p>
                          <p>Julien barbaud</p>
                          <p><br>
                          </p>
                          <p>P.S: this is my first time posting on this
                            user list. Please let me know if my question
                            is not suitable for it, or can be improved
                            either in its content or presentation. I
                            will gladly take any recommandation into
                            account in order not to negatively impact
                            the quality of this user list !<br>
                          </p>
                        </div>
                      </div>
                      <br>
                      <fieldset class="mimeAttachmentHeader"></fieldset>
                      <br>
                      <pre wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
                    </blockquote>
                    <p><br>
                    </p>
                    <br>
                    <fieldset class="mimeAttachmentHeader"></fieldset>
                    <pre class="moz-quote-pre" wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
                  </blockquote>
                </div>
                <br>
                <fieldset class="mimeAttachmentHeader"></fieldset>
                <pre class="moz-quote-pre" wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
              </blockquote>
              <br>
              <fieldset class="mimeAttachmentHeader"></fieldset>
              <pre class="moz-quote-pre" wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
            </blockquote>
            <br>
            <fieldset class="mimeAttachmentHeader"></fieldset>
            <br>
            <pre wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
          </blockquote>
          <p><br>
          </p>
          <br>
          <fieldset class="mimeAttachmentHeader"></fieldset>
          <pre class="moz-quote-pre" wrap="">_______________________________________________
users mailing list
<a class="moz-txt-link-abbreviated" href="mailto:users@lists.quantum-espresso.org" moz-do-not-send="true">users@lists.quantum-espresso.org</a>
<a class="moz-txt-link-freetext" href="https://lists.quantum-espresso.org/mailman/listinfo/users" moz-do-not-send="true">https://lists.quantum-espresso.org/mailman/listinfo/users</a></pre>
        </blockquote>
      </div>
    </div>
  </body>
</html>