mirror of https://gitlab.com/QEF/q-e.git
596 lines
21 KiB
Plaintext
596 lines
21 KiB
Plaintext
Fixed in version 4.0.3:
|
|
|
|
* CPV: array qv allocated in newd wasn't deallocated in all cases,
|
|
leading to either ia crash or a memory leak (Ralph)
|
|
|
|
* Task groups bug fix: array tg_rho was not cleared at every k point cycle.
|
|
This was causing problems with some combinations of "-npool" and "-ntg".
|
|
|
|
* PWCOND: a bug with some array bounds fixed (A. Smogunov)
|
|
|
|
* Problem with the generation of the atomic wavefunctions in the
|
|
projwfc code when a scalar relativistic PP is used with lspinorb=.true.
|
|
|
|
* Bug fix in symmetry analysis for the case S_6 (reported by Marino
|
|
Vetuschi Zuccolini) and also in: S_4, T_h, C_3h, C_4h, C_6h.
|
|
|
|
Fixed in version 4.0.2:
|
|
|
|
* Nuclear masses not correctly displayed for variable-cell calculations
|
|
|
|
* Probably all results for EFG (electric field gradients) were wrong,
|
|
due to an incorrect multiplication of "r" with "alat" inside a loop
|
|
(should have been outside: routine PW/ewald_dipole.f90)
|
|
|
|
* Structural optimization with external sawtooth potential was not
|
|
working correctly (electric field disappeared after first run).
|
|
All versions after october 2005 affected.
|
|
|
|
* Calculation with fixed magnetization and nspin=2 (using 2 fermi
|
|
levels) was not working in v. 4.0.1
|
|
|
|
* non linear core correction was not detected in FPMD run
|
|
|
|
* effective charges + US PP + spin-orbit not correct in noncubic cases.
|
|
|
|
* symm_type was not properly set by pw_restart (used in various
|
|
post-processing including phonons) when using free lattice
|
|
(ibrav=0) and symm_type=hexagonal.
|
|
|
|
* CP: conjugate gradient had a bug in some cases of parallel
|
|
execution. Also: default max number of iterations was not
|
|
what was promised in the documentation (100)
|
|
|
|
* phonon: alpha_pv depended on the number of unoccupied bands
|
|
in insulators (harmless).
|
|
|
|
* fpmd was using wrong forces propagate cell variables in
|
|
variable-cell calculations. Also: interpolation tables
|
|
were a little bit too small for variable cell simulation
|
|
(not really a bug but it could be annoying)
|
|
|
|
* Minor glitch in configure for pathscale compiler. Note that
|
|
in the machine that has been tested, compilation of iotk
|
|
fails for mysterious reasons if CPP = pathcc -E, while it
|
|
works with CPP = /lib/cpp -P --traditional
|
|
|
|
Fixed in version 4.0.1:
|
|
|
|
* Some scripts used in tests/ and in examples were not
|
|
posix-compliant and could fail in some cases
|
|
|
|
* In cg calculations with cp, the case of no spin multiplicity
|
|
(i.e. nspin=1) with odd number of bands was yielding an error
|
|
"c second dimension too small"
|
|
|
|
* rigid.f90: sum over G-space in long-range term, used in q2r.x
|
|
and matdyn.x, wasn't really converged for sufficiently large cells
|
|
|
|
* too many automatic arrays in "set_asr" called in matdyn.f90,
|
|
causing a mysterious crash for systems larger than a few atoms
|
|
|
|
* incorrect call to "sgama" in matdyn.f90 could lead to failures
|
|
with strange messages when calculating phonon DOS
|
|
|
|
* c_mkdir is explicitly defined as integer*4 in order to prevent
|
|
problems in 64-bit machines with default 64-bit integers
|
|
|
|
* PP/chdens.f90: incorrect orthogonality test for axis
|
|
|
|
* GIPAW: 10^3 factor missing in conversion
|
|
|
|
* GIPAW: paw_recon[]%paw_betar[] was not initialised and caused NaN's
|
|
with IBM compilers. Courtesy of Christos Gougoussis (IMPMC, Paris).
|
|
|
|
* Minor glitches in PWgui
|
|
|
|
* cppp.x was not working in v.4.0
|
|
|
|
* Workaround for bluegene weirdness extended to complex hamiltonians
|
|
|
|
* PP/projwfc.f90: Problems with file names in systems > 1000 atoms
|
|
|
|
* Workaround for ATLAS bug causing random crashes
|
|
|
|
* Minor bug in helpdoc: adding syntaxFlush to linecard
|
|
|
|
* Incorrect dimensions in PW/local.f90 (courtesy of Zhiping)
|
|
|
|
Fixed in version 4.0:
|
|
|
|
* Unpredictable results when the output from a spin-polarized CP
|
|
calculation was used for post-processing. This was due to an
|
|
incorrect treatment of the case where the number of up and down
|
|
states are not the same. There was also an inconsistency in the
|
|
treatment of the number of up and down electrons, that can be in
|
|
principle real, unlike the number of states that is integer
|
|
|
|
* In MD calculations with PWscf, there was the possibility of an
|
|
out-of-bound error, with unpredictable consequences, including
|
|
in at least one case hanging of parallel jobs
|
|
|
|
* Due to a bad dimensioning of variable hubbard_l, DFT+U results could
|
|
be wrong if atomic type N with U term has N > L=maximum hubbard L
|
|
|
|
* a few symmetries were confusing the symmetry finder
|
|
|
|
* serious bugs in Berry's phase calculation. It affected only the US
|
|
case and only some terms, so the error was small but not negligible.
|
|
There were three different bugs, one introduced when the spherical
|
|
harmonics were modified in the rest of the code, two that I think
|
|
have been there from the beginning.
|
|
|
|
* various glitches with wf_collect option in the noncollinear case
|
|
|
|
* mix_rho was not working properly for lsda with data saved to file
|
|
and double grid
|
|
|
|
Fixed in version 3.2.1-3.2.3:
|
|
|
|
* CP in parallel execution had a serious bug if the third dimension
|
|
of FFT arrays (nr3x/nr3sx) was not the same as FFT order (nr3/nr3s)
|
|
|
|
* restart of pw.x in parallel could go bananas under some not-so-unusual
|
|
circumstances, due to bad setting of a variable
|
|
|
|
* various phonon glitches: pools and lsda, pools and dispersions,
|
|
option lnscf, were not working
|
|
|
|
* incorrect exchange-correlation contribution to the electro-optical
|
|
coefficient
|
|
|
|
* check for stop condition was unsafe with pools and could hang pw.x
|
|
|
|
* fixed occupations in parallel: array not allocated on all processors
|
|
|
|
* Yet another problem of poor accuracy of routines calculating spherical
|
|
bessel functions - harmless except in some cases of pseudopotential
|
|
generation
|
|
|
|
* DOS EOF characters present in some files could cause trouble
|
|
during installation
|
|
|
|
* restart in phonon calculations was not always properly working
|
|
|
|
* possible divide-by-zero error in dV_xc/dz (spin polarized case)
|
|
|
|
* gamma_gamma symmetry was not working for open-shell molecules
|
|
|
|
* T_h group not correctly identified in postprocessing
|
|
|
|
* missing initialization of rho could lead to serious trouble
|
|
if the physical and true dimensions of FFT grid did not coincide
|
|
|
|
* Ewald real-space term could have been incorrectly calculated
|
|
if an atom was far away from the unit cell
|
|
|
|
* Some variables were used before they were initialized - this could
|
|
lead to crashes or unpredictable behaviour on some machines
|
|
|
|
* lattice parameters a,b,c,cosab,cosac,cosbc were not properly
|
|
copied to the celldm in the case of triclinic lattice
|
|
|
|
Fixed in version 3.2:
|
|
|
|
* In same cases the energy under an external sawtooth potential
|
|
simulating an electric field was not correct
|
|
|
|
* Case ibrav=13 fixed for good this time!!!
|
|
|
|
* Bug in PH/clinear.f90 for cells having nr1 /= nr2 may have
|
|
affected new electron-phonon algorithm
|
|
|
|
* Poor accuracy of routines calculating spherical bessel functions
|
|
for high l and small q - harmless except in very special cases
|
|
|
|
* LDA+U with variable-cell dynamics/relaxation was wrong due to
|
|
missing rescaling of the integrals of atomic wavefunctions.
|
|
This bug has been present since at least 3.0
|
|
|
|
* Parallel subspace diagonalization could occasionally fail;
|
|
replaced by a new algorithm that is much more stable
|
|
|
|
* Restart problems in parallel run for two cases:
|
|
1) with pools, 2) with local filesystems
|
|
|
|
Fixed in version 3.1.1:
|
|
|
|
* Methfessel-Paxton broadening was hardcoded in the calculation of
|
|
the electron-phonon coefficients (ngauss1=1 in PH/elphon.f90).
|
|
There is no good reason to use this instead of simple gaussian
|
|
(ngauss1=0), which, moreover, guarantees positive definite results.
|
|
|
|
Fixed in version 3.1:
|
|
|
|
* various problems in stress calculation, both in PW and in CP
|
|
|
|
* in phonon dispersion calculation, the threshold for diagonalization
|
|
was not always what was expected to be. Minor numerical differences
|
|
could result.
|
|
|
|
* the new algorithm for electron-phonon calculation removes a serious
|
|
bug in the old algorithm, present in v.2.1 to 3 included: when
|
|
electron-phonon coefficients were calculated together with the
|
|
dynamical matrix, the symmetrization of coeffcients was incorrect.
|
|
Results from separate calculations were correct.
|
|
|
|
Fixed in version 3.0:
|
|
|
|
* latgen.f90 : case ibrav=13 bad
|
|
|
|
* kpoints.f : case ibrav=5 bad
|
|
|
|
Fixed in version 3.1.1:
|
|
|
|
* Methfessel-Paxton broadening was hardcoded in the calculation of
|
|
the electron-phonon coefficients (ngauss1=1 in PH/elphon.f90).
|
|
There is no good reason to use this instead of simple gaussian
|
|
(ngauss1=0), which, moreover, guarantees positive definite results.
|
|
|
|
Fixed in version 3.1:
|
|
|
|
* various problems in stress calculation, both in PW and in CP
|
|
|
|
* in phonon dispersion calculation, the threshold for diagonalization
|
|
was not always what was expected to be. Minor numerical differences
|
|
could result.
|
|
|
|
* the new algorithm for electron-phonon calculation removes a serious
|
|
bug in the old algorithm, present in v.2.1 to 3 included: when
|
|
electron-phonon coefficients were calculated together with the
|
|
dynamical matrix, the symmetrization of coeffcients was incorrect.
|
|
Results from separate calculations were correct.
|
|
|
|
Fixed in version 3.0:
|
|
|
|
* latgen.f90 : case ibrav=13 bad
|
|
|
|
* kpoints.f : case ibrav=5 bad
|
|
|
|
Fixed in version 2.1.5:
|
|
|
|
* bad forces and stresses with LDA+U in spin-unpolarised case
|
|
|
|
* bad printout of Lowdin charges in projwfc
|
|
|
|
* FPMD had a problem with some types of UPF PPs
|
|
|
|
Fixed in version 2.1.4:
|
|
|
|
* Incorrect initial guess for occupancies in LDA+U (init_ns)
|
|
|
|
* bogus error in postprocessing with local pseudopotentials only
|
|
|
|
* several errors in third-order energy derivatives (D3/)
|
|
|
|
* checks on several unimplemented cases were missing
|
|
|
|
Fixed in version 2.1.3:
|
|
|
|
* case ibrav=0 in CPV was not properly working
|
|
|
|
* forces in CP with core corrections were wrong
|
|
(reported by Giacomo Saielli)
|
|
|
|
* damped variable-cell dynamics in PWscf was not working properly
|
|
|
|
* lambda.x could yield NaN's on negative frequencies
|
|
|
|
* option "write_save" was not working in parallel
|
|
|
|
* diagonalization of (0,0) matrix in init_paw_1
|
|
|
|
* out-of-bound error in readnewvan.f90 fixed
|
|
|
|
* FPMD: bug with UPF PP when betas are not ordered as l=0,1,2,...
|
|
|
|
* Possible out-of-bound error with US PP in some cases
|
|
|
|
* Martins-Troullier norm-conserving PP generation had a small
|
|
error when rcloc > rcut(l)
|
|
|
|
* the default for relativistic vs nonrelativistic calculation
|
|
in the atomic code was the opposite of what was intended
|
|
|
|
* electron-phonon calculation was not working properly if called
|
|
after a restart
|
|
|
|
* Parallel execution on local filesystems (i.e. not visible to all
|
|
processors) could hang due to a bad check in charge extrapolation
|
|
|
|
* When imposing hermiticity in matdyn.x and dynmat.x codes in pwtools
|
|
routine dyndiag was actually computing the complex conjugate of
|
|
the dynamical matrix. Eigenvectors were therefore wrong, while
|
|
eigenvalues were fine. (thanks to Nicolas Mounet)
|
|
|
|
Fixed in version 2.1.2:
|
|
|
|
* The phonon code was yielding incorrect results when 4-dimensional
|
|
irreps were present (i.e. A point in graphite) and ultrasoft PP used
|
|
(reported by Nicolas Mounet)
|
|
|
|
* in some cases ld1 was writing a bad UPF file
|
|
|
|
* in some cases the charge density was not conserved during
|
|
the charge mixing
|
|
|
|
* various problems with potential extrapolation in neb and smd
|
|
|
|
* variable-cell dynamics and optimization was not working in parallel
|
|
|
|
* Berry phase calculation in parallel should have been disabled
|
|
|
|
* bug in readfile_config when restarting without a "*.save" file
|
|
|
|
* crash in pw2casino due to bad call to v_of_rho
|
|
|
|
Fixed in version 2.1.1:
|
|
|
|
* memory leak in Raman code
|
|
|
|
* disproportionate memory requirement in phonon code with USPP
|
|
|
|
* dangerous calls to read_restart_tetra and write_restart_tetra
|
|
when restarting with no allocated tetrahedra
|
|
|
|
* vc-relax was not working
|
|
|
|
* projwfc failed with lda+U
|
|
|
|
* incorrect automatic generation of k-points in the non colinear case:
|
|
inversion symmetry is not always present because of the presence of
|
|
a magnetic field in the Hamiltonian
|
|
|
|
* electron-phonon calculation was not working if called directly
|
|
after a phonon calculation
|
|
|
|
* PWCOND + FFTW + parallel execution = not good
|
|
|
|
* cell minimization with steepest descent was not working (CP/FPMD)
|
|
|
|
* various Alpha, IBM, SGI, SUN, PGI compilation problems
|
|
|
|
Fixed in version 2.1:
|
|
|
|
* various T3E compilation problems
|
|
|
|
* cpmd2upf was yielding incorrect DFT if converting BLYP PPs
|
|
|
|
* some variables not properly written and read in restart file
|
|
|
|
* The value of gamma_only was not correctly set when restarting or
|
|
reading from file with option __NEW_PUNCH enabled
|
|
|
|
* Incorrect calculation of eloc in pw2casino
|
|
|
|
* Two serious bugs in the local-TF screening :
|
|
possible occurrence of division by zero (present since v1.2),
|
|
wrong mixing of spin polarized systems
|
|
|
|
* cpmd2upf failed with some files due to bad check
|
|
|
|
* Intel compiler v.8: wavefunction files four times bigger than needed
|
|
|
|
* compilation problems on some version of SGI compiler
|
|
|
|
* non-collinear code was not working with insulators and nbnd > nelec/2
|
|
|
|
* multiple writes to file in parallel execution when calculating
|
|
electron-phonon coefficients
|
|
|
|
* various bugs in LBFGS
|
|
|
|
* NEB + LDA+U = crash
|
|
|
|
* compilation problems with __NEW_PUNCH
|
|
|
|
* planar average crashed if used with a cubic system
|
|
|
|
* Gamma-only phonon code not working for Raman calculations
|
|
in some cases
|
|
|
|
* yet another bug in phonon and k-point parallelization when
|
|
reading namelist (phq_readin)
|
|
|
|
* options startingwfc and startingpot were ignored if restarting
|
|
from a previous calculation
|
|
|
|
* pw2casino interface didn't work properly in spin-polarized case
|
|
and didn't use variable "outdir"
|
|
|
|
* minor bug in pwtools/pwo2xsf.sh
|
|
|
|
* serious bug in the path interpolator
|
|
|
|
* phonon, post_processing, various other auxiliary codes were
|
|
not working with k-point parallelization (pools) due to
|
|
double call to init_pool
|
|
|
|
Fixed in version 2.0 :
|
|
|
|
* wrong results when running Berry-phase calculation in parallel execution:
|
|
it was not implemented but no warning was issued
|
|
|
|
* variable-cell code was subject to overflow and floating-point errors
|
|
|
|
* phonon + nosym=.true. was not properly done
|
|
|
|
* out-of-bound error in Berry Phase calculation
|
|
|
|
* out-of-bound errors in phonon if 4-dimensional irreps were present
|
|
(also d3.x was not working properly in this case)
|
|
|
|
* Berry-phase calculation had problems in low-symmetry cases
|
|
|
|
* phonon with k-point parallelization (pools) was yielding wrong
|
|
results in some cases (since v. 1.2 included)
|
|
|
|
* upftools/cpmd2upf.f90: wrong conversion due to Rydberg-Hartree mess
|
|
|
|
* PW/input.f90: lattice parameter a converted to wrong units if input
|
|
is given as a,b,c,cos(ab),cos(ac),cos(bc) instead of celldm(:)
|
|
|
|
* Wrong coordinates written if atomic_positions='crystal'
|
|
(thanks to Francois Willaime)
|
|
|
|
Fixed in version 1.3.0 :
|
|
|
|
* PH/elphon.f90 : el-ph calculation in the US case was not correctly
|
|
working in v.1.2.0 (it was not implemented in previous versions).
|
|
An US term in the calculation of deltaV * psi_v was missing.
|
|
Fixed by M. Wierzbowska and SdG
|
|
|
|
* various problems caused by too short file names fixed:
|
|
file and directory names up to 80 characters are allowed
|
|
(thanks to Serguei Patchkovskii and others)
|
|
|
|
* LAPACK routines DSYTRF and DYSTRI require some character arguments
|
|
(like 'U', 'L'). While most LAPACK implementations accept both
|
|
lowercase and uppercase arguments, the standard is uppercase only.
|
|
Various anomalies in self-consistency were caused by lowercase
|
|
arguments.
|
|
|
|
* Incorrect Make.pc_abs fixed
|
|
|
|
* PGI compiler v.3.3-2 on Linux: PP/chdens.x coredump fixed
|
|
|
|
* various T3E glitches in v.1.2.0 fixed
|
|
|
|
* PP/work_functions.f90 : STM maps did not work in version 1.2.0
|
|
(undefined variable lscf was used, call to sum_band no longer needed)
|
|
|
|
* PP/projwave.f90: symmetrization of projected dos was incorrectly
|
|
performed using d1,d2,or d3 instead of their transponse.
|
|
(affects all previous versions)
|
|
|
|
* PW/new_ns.f90: symmetrization of occupation matrix ns needed for LDA+U
|
|
calculations used incorrectly d2 matrices instead of their transponse.
|
|
Thanks to Lixin He for finding out the problem and the solution.
|
|
(affects all previous versions)
|
|
|
|
Fixed in version 1.2.0 (f90) :
|
|
|
|
* dynmat.f90: out-of-bound error fixed
|
|
|
|
* pplib/chdens.F90, pplib/projwave.F90 : compilation problems
|
|
for alpha (found by Giovanni Cantele)
|
|
|
|
* postprocessing routines: problems with unallocate pointers
|
|
passed to subroutine plot_io fixed (found by various people)
|
|
|
|
* postprocessing with ibrav=0 was not working properly
|
|
|
|
* rather serious bug in cinitcgg (used by conjugate-gradient
|
|
diagonalization) could produce mysterious crashes. The bug
|
|
appeared in version 1.1.1.
|
|
|
|
* pplib/dos.f90 was not plotting the expected energy window
|
|
|
|
* pplib/chdens.F90, pplib/average.F90 : wrong call to setv
|
|
could cause an out-of-bound error
|
|
|
|
Fixed in version 1.1.2 (f90) :
|
|
|
|
* a check on the number of arguments to command line in parallel
|
|
execution was added - Intel compiler crashes if attempting to
|
|
read a nonexistent argument
|
|
|
|
* tmp_dir was incorrectly truncated to 35 characters in
|
|
parallel execution
|
|
|
|
* variable "kfac" was not deallocated in stres_knl. A crash in
|
|
variable-cell MD could result.
|
|
|
|
* an inconsistent check between the calling program (gen_us_dj)
|
|
and the routine calculating j_l(r) (sph_bes) could result in
|
|
error stop when calculating stress or dielectric properties
|
|
|
|
* errors at file close in pw.x and phonon.x in some cases
|
|
|
|
* tetrahedra work for parallel execution
|
|
(ltetra is now distributed in bcast_input)
|
|
|
|
* fixed some problems in automatic dependencies (Giovanni Cantele)
|
|
|
|
Fixed in version 1.1.1 (f90) and 1.0.3 (f77) :
|
|
|
|
* LSDA calculations need either gaussian broadening or tetrahedra
|
|
but no input check was performed
|
|
|
|
* restarting from a run interrupted at the end of self-consistency
|
|
yielded wrong forces
|
|
|
|
* projwave.F (projection over atomic functions) was not working
|
|
with atoms having semicore states (found by Seungwu Han)
|
|
|
|
* stm.F : option stm_wfc_matching was not working properly
|
|
if symmetry was present (no symmetrization was performed)
|
|
|
|
* dynmat.x : displacement patterns in "molden" format were
|
|
incorrectly divided by the square root of atomic masses
|
|
|
|
* d3: misc. problems in parallel execution fixed
|
|
|
|
Fixed in version 1.1.0 (f90) and 1.0.2 (f77) :
|
|
|
|
* an inconsistency in the indexing of pseudopotential arrays could
|
|
yield bad dielectric tensors and effective charges if atoms where
|
|
not listed as first all atoms of type 1, then all atoms of type 2,
|
|
and so on (found by Nathalie Vast)
|
|
|
|
* phonon with ibrav=0 was not working (info on symm_type was lost:
|
|
found by Michele Lazzeri)
|
|
|
|
* the generation of the two random matrices needed in the calculation
|
|
of third order derivatives was incorrect because the random seed
|
|
was not reset. This produced crazy results for q<>0 calculations.
|
|
|
|
* the check on existence of tmp_dir did not work properly on
|
|
Compaq (formerly Dec) alphas (thanks to Guido Roma and Alberto
|
|
Debernardi).
|
|
|
|
* a system containing local pseudopotentials only (i.e. H)
|
|
produced a segmentation fault error
|
|
|
|
* getenv was incorrectly called on PC's using Absoft compiler:
|
|
the default pseudopotential directory was incorrect
|
|
|
|
* out-of-bound bug in pplib/dosg.f fixed. It could have caused
|
|
mysterious crashes or weird results in DOS calculations using
|
|
gaussian broadening. Thanks to Gun-Do Lee for fixing the bug.
|
|
|
|
* a missing initialization to zero in gen_us_dy.F could have
|
|
yielded a wrong stress in some cases
|
|
|
|
* phonons in an insulator did not work if more bands (nbnd)
|
|
were specified than filled valence band only
|
|
|
|
* electron-phonon calculation was incorrect if nonlocal PPs
|
|
were used (that is, almost always)
|
|
|
|
* Real space term in third order derivative of ewald energy
|
|
was missing (not exactly a bug, but introduced a small error
|
|
that could be not negligible in some cases)
|
|
|
|
* bad call in dynmat.f corrected
|
|
|
|
* compilation problems for PC clusters fixed (thanks to Nicola Marzari)
|
|
|
|
Fixed in version 1.0.1:
|
|
|
|
* recovering from a previous run in pw.x did not work on PC's
|
|
|
|
* recovering from a previous run in pw.x did not work for stress
|
|
calculation
|
|
|
|
* poolrecover did not compile on some machines (thanks to Eric Wu)
|
|
|
|
* PC with absoft compiler (and maybe other cases as well):
|
|
bad type conversions for REAL and CMPLX resulted in poor
|
|
convergence in some test cases. DCMPLX, DREAL used instead.
|
|
|
|
* Asymptotic high- and low-density formulae used in PW91 and PBE
|
|
unpolarized functionals gave a small but not negligible error,
|
|
leading to bad convergence of structural optimization
|
|
|