Relaxation issue with vasp5.2.11 and 5.2.12 compiled with Intel v.12
Posted: Thu Dec 15, 2011 7:18 pm
Dear VASP Developers,
Using the following relaxed input POSCAR, I found that vasp4.6.36 relaxes this structure within two ionic steps. So does vasp5.2.2 compiled with PGI. However, vasp5.2.11 and 12, compiled with Intel v.12.0.0.084, diverge the volume and after 300 ionic steps, the structure is far from being converged. Similar reports are found in the forum:
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.9798
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.8875
Our ACEnet cluster IT engineers tried compiling the latest version of 5.2.12 by changing OFLAG to "-O1 ip -ftz". This compiled version works fine. Apparently, there is a section in the makefile for the Intel compilers that overrides the OFLAG default setting for 14 source files: some files get -O1, some get -O2, explicitly.
Interestingly enough, the Intel compilers Makefile seems to have the highest number of these exceptions among all the makefiles provided for different architectures and compilers. The version of the makefile for PGI does not have such exceptions at all.
HI
1.000000000000000
8.1693944531991180 -0.0196576127696663 1.5203211938183680
0.0500281776455709 8.1610426113641590 1.5487516416327660
-2.4049646297134170 -2.3572098055415000 8.5895050241929790
8 8
Direct
0.4473971756428095 0.1319070490816879 0.9872085574065846
0.6339218819191452 0.4238050804440289 0.9133673468731104
0.3309243783305581 0.6414263830806397 0.9400091528981713
0.1483389904866283 0.3475298292239793 0.0163460965382349
0.4610654055622555 0.2485326419072515 0.5196114152677388
0.7517042835719680 0.4400195090488600 0.4482308345567040
0.5345890307444717 0.7421133864506148 0.4175072921969856
0.2418991653480166 0.5521416304918950 0.4911773318074170
0.1482879614903867 0.1503843244735678 0.0368430107873043
0.6432030980958602 0.1218359217043237 0.9603055724662891
0.6323232713013834 0.6207589771071643 0.8911550959448784
0.1335298217412246 0.6482149516581082 0.9595284140727037
0.2647366517754530 0.2499979826812485 0.5433032478927331
0.7619700005687127 0.2431015907394510 0.4708721699101599
0.7323468473106278 0.7430062932746057 0.4018344155580317
0.2324720387592779 0.7496499880832368 0.4714002232229534
Thank you,
Vahid
Using the following relaxed input POSCAR, I found that vasp4.6.36 relaxes this structure within two ionic steps. So does vasp5.2.2 compiled with PGI. However, vasp5.2.11 and 12, compiled with Intel v.12.0.0.084, diverge the volume and after 300 ionic steps, the structure is far from being converged. Similar reports are found in the forum:
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.9798
http://cms.mpi.univie.ac.at/vasp-forum/ ... php?3.8875
Our ACEnet cluster IT engineers tried compiling the latest version of 5.2.12 by changing OFLAG to "-O1 ip -ftz". This compiled version works fine. Apparently, there is a section in the makefile for the Intel compilers that overrides the OFLAG default setting for 14 source files: some files get -O1, some get -O2, explicitly.
Interestingly enough, the Intel compilers Makefile seems to have the highest number of these exceptions among all the makefiles provided for different architectures and compilers. The version of the makefile for PGI does not have such exceptions at all.
HI
1.000000000000000
8.1693944531991180 -0.0196576127696663 1.5203211938183680
0.0500281776455709 8.1610426113641590 1.5487516416327660
-2.4049646297134170 -2.3572098055415000 8.5895050241929790
8 8
Direct
0.4473971756428095 0.1319070490816879 0.9872085574065846
0.6339218819191452 0.4238050804440289 0.9133673468731104
0.3309243783305581 0.6414263830806397 0.9400091528981713
0.1483389904866283 0.3475298292239793 0.0163460965382349
0.4610654055622555 0.2485326419072515 0.5196114152677388
0.7517042835719680 0.4400195090488600 0.4482308345567040
0.5345890307444717 0.7421133864506148 0.4175072921969856
0.2418991653480166 0.5521416304918950 0.4911773318074170
0.1482879614903867 0.1503843244735678 0.0368430107873043
0.6432030980958602 0.1218359217043237 0.9603055724662891
0.6323232713013834 0.6207589771071643 0.8911550959448784
0.1335298217412246 0.6482149516581082 0.9595284140727037
0.2647366517754530 0.2499979826812485 0.5433032478927331
0.7619700005687127 0.2431015907394510 0.4708721699101599
0.7323468473106278 0.7430062932746057 0.4018344155580317
0.2324720387592779 0.7496499880832368 0.4714002232229534
Thank you,
Vahid