error for test case 201_NO_KP_DJ_CF_CS_GaAs : scalapack solver error! about abacus-develop

[caic99]

My reimplement:
Warning: Lapack routine DSTEDC failed, info= 17, Trying DSTEQR!

[dyzheng]

My reimplement:
Warning: Lapack routine DSTEDC failed, info= 17, Trying DSTEQR!

Did that warning caused this error? using genelpa?

[caic99]

Intel MKL ('17) + GCC 8.3.1

scalapack_gvx

<=2 processes: not converging (related to another issue)
>2 processes: terminate called after throwing an instance of 'std::runtime_error'
what(): info=2. /home/caichun/abacus-develop/source/src_pdiag/pdiag_double.cpp line 1171

if (MOD(INFO/2,2).NE.0),then eigenvectors corresponding to one or more clusters of eigenvalues could not be reorthogonalized because of insufficient workspace. The indices of the clusters are stored in the array ICLUSTR. more info

genelpa

>2 processes: converged.

hpseps

Converged.

Scalapack + GCC 7.5.0

genelpa

1 process (invoke w/ abacus): Converged.
multi processes (invoke w/ mpirun -n N abacus):
N=1:

Input warning : gamma_only_local algorithm is not used.
AUTO_SET lcao_ecut to 20
In SCAN_BEGIN, can't find: NUMERICAL_DESCRIPTOR block.
In SCAN_BEGIN, can't find: LATTICE_PARAMETERS block.
In SCAN_BEGIN, can't find: <PP_SPIN_ORB> block.
In SCAN_END, can't find: </PP_SPIN_ORB> block.
In SCAN_END, can't find: block.
In SCAN_BEGIN, can't find: <PP_SPIN_ORB> block.
In SCAN_END, can't find: </PP_SPIN_ORB> block.
In SCAN_END, can't find: block.
AUTO_SET NBANDS to 0
AUTO_SET NBANDS to 24
functional warning : file /home/ubuntu/abacus-develop/source/src_pw/xc_type.cpp line 297 may error
functional warning : file /home/ubuntu/abacus-develop/source/src_pw/xc_type.cpp line 297 may error
startmag_type = 1
charge from rho_at = 15
charge should be = 15
charge from rho_at = 13
charge should be = 13
Warning: negative or imaginary starting charge : neg = -0.0651382 ima = 0 SPIN = 0
SETUP ATOMIC RHO FOR SPIN 1
Electron number from rho = 28
total electron number from rho = 28
should be = 28
charge before normalized = 28
charge after normalized = 28
Processor Atom
1 23
2 22
LCAO ALGORITHM ------------- ELEC= 1--------------------------------
CHECK UPDATE INPUT INFORMATION
sum_rho=0
Charge::renormalize_rho warning : Can't find even an electron!

N>=2:

Warning: Lapack routine DSTEDC failed, info= 17, Trying DSTEQR!

scalapack_gvx

1 process: correct
2+ processes:

what(): info=2. /home/ubuntu/abacus-develop/source/src_pdiag/pdiag_double.cpp line 1169

hpseps

Correct.

[dyzheng]

another error reported for this test case :

• multi core result differ from one core's with hpseps ks_solver in docker environment.
• Only multi K points would have this difference, when gamma_only result is same between one core and multi cores.

[dyzheng]

There would be 3.0e-5 eV error in total energy between one core and 4 core results in newly version using hpseps solver.
In version e1c0d39 in 2021/05/21, there is no deviation using genelpa solver, and still have error 4.0e-5 eV using hpseps solver.

[dyzheng]

#383 solved error with ELPA, using ELPA2021 is ok for both intel oneAPI and GNU compiler

[dyzheng]

Now, only scalapack solver is error. Set parameter: "ks_solver scalapack_gvx" and use multi cores calculation, the error will occurs in dyzheng's environment and will not occurs in PeizeLin's environment.

[dyzheng]

scalapack_gvx for k is correct now, but still has problem for gamma_only