Yet another problem with final scf step in a vc-relax calculation

git-svn-id: http://qeforge.qe-forge.org/svn/q-e/trunk/espresso@13971 c92efa57-630b-4861-b058-cf58834340f0

Doc/release-notes

@@ -1,3 +1,11 @@
+Problems fixed by development version:
+
+  * Yet another problem with last scf step in vc-relax, present since v.6.1: 
+    if no atoms of a given kind were present, there was a division by zero
+    and a NaN in starting magnetization (reported by Malte Sachs) (r13971)
+
+  * FFT's for ARM libraries were broken (r13956,13959, courtesy Jason Wood)
+
 New in development version:
 
   * Interface with Grimme's DFT-D3, as repackaged by Bálint Aradi
@@ -11,10 +19,6 @@ Incompatible changes in development version:
     with one underscore (automatically set by all OpenMP-aware compilers)
   * "Old" xml input file deleted
 
-Problems fixed by development version:
-
-  * FFT's for ARM libraries were broken (courtesy Jason Wood)
-
 Known problems in 6.2 version:
 
   * New format for charge density no longer allows to perform a non-scf

PW/src/move_ions.f90

@@ -474,9 +474,11 @@ SUBROUTINE reset_starting_magnetization ( )
            END IF
         END IF
      END DO
-     starting_magnetization(i) = starting_magnetization(i) / REAL(nt)
-     angle1(i) = angle1(i) / REAL(nt)
-     angle2(i) = angle2(i) / REAL(nt)
+     IF ( nt > 0 ) THEN
+        starting_magnetization(i) = starting_magnetization(i) / DBLE(nt)
+        angle1(i) = angle1(i) / DBLE(nt)
+        angle2(i) = angle2(i) / DBLE(nt)
+     END IF
   END DO
   DEALLOCATE ( r_loc, m_loc )