Bug#702801: cp2k: dependence of DFTB-results on the input-sequence of heavy atoms

[Michael Banck]

Package: cp2k
Version: 2.2.426-6
Severity: serious

A bug was found in cp2k's DFTB module where the results are dependent on
the input-ordering of heavy atoms.  The following two input files
(attached) from the current upstream regression testsuite differ only in
the order the Carbon and Oxygen atoms are defined:

--- co2_1.inp   2013-03-11 17:03:49.456297517 +0100
+++ co2_2.inp   2013-03-11 17:03:49.459607156 +0100
@@ -30,8 +30,8 @@
       PERIODIC NONE
     &END CELL
     &COORD
-      O  0.0  0.0  +1.2
       C  0.0   0.0    0.0
+      O  0.0  0.0  +1.2
       O   0.0  0.0  -1.2
     &END COORD
   &END SUBSYS

Yet, the calculated total energy are markedly different:

nighthawk~/cp2k-2.2.426/tests/DFTB/regtest-nonscc$ cp2k.popt co2_1.inp | grep 
ENERGY.*Total
 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.): -8.555862465666859
nighthawk~/cp2k-2.2.426/tests/DFTB/regtest-nonscc$ cp2k.popt co2_2.inp | grep 
ENERGY.*Total
 ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.): -8.659241457828779


Michael

co2_1.inp
Description: chemical/gamess-input

co2_2.inp
Description: chemical/gamess-input