* Updated packaging info.
 * Tagged version 4.1.2.

------------------------------------------------------------------------
r11532 | micael | 2013-11-18 08:59:49 +0000 (Mon, 18 Nov 2013) | 2 lines

 * Fixed bug pointed out by Xavier: vtau should be divided by 2 outside the space loop.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11528 | dstrubbe | 2013-11-15 16:32:27 -0500 (Fri, 15 Nov 2013) | 2 lines

Bugfix: gga_x_lb with max-dim > 3 caused a segmentation fault. Correcting allocation.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11526 | dstrubbe | 2013-11-14 20:02:09 -0500 (Thu, 14 Nov 2013) | 2 lines

Although current DFT was removed in r8079, the variable CurrentDFT remained around to confuse users.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11511 | dstrubbe | 2013-11-12 23:39:03 -0500 (Tue, 12 Nov 2013) | 2 lines

More space for info in error message from SAFE_ALLOCATE and SAFE_DEALLOCATE_A, SAFE_DEALLOCATE_P.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11508 | micael | 2013-11-12 05:56:46 -0500 (Tue, 12 Nov 2013) | 2 lines

 * Bug fix: the calculation of GGA and MGGA xc potentials was completely wrong due to a misplaced "end if". This bug was introduced in [9729].

------------------------------------------------------------------------

Comment: I checked that the 1D, 2D, and 3D functionals tests in trunk are passed by 4.1.x as of this change, with two exceptions below. The reference value for gga_x_pbea had been changed in trunk due to a change in libxc. The 1D full OEP test differs by only a small amount, should not be affected by GGA/mGGA, and has convergence problems anyway. So everything seems fine.

{{{
Using input file : ./03-xc.gga_x_pbea.inp 

Match gga_x_pbea Exchange                    :

   Calculated value : -0.31167167
   Reference value  : -0.31200606
   Difference       : 0.000334389999999962
   Tolerance        : 0.0001

 gga_x_pbea Exchange                    :        [  FAIL  ] 

Match gga_x_pbea Int[n*v_xc]                 :

   Calculated value : -0.39996075
   Reference value  : -0.40021553
   Difference       : 0.00025478000000001
   Tolerance        : 0.0001

 gga_x_pbea Int[n*v_xc]                 :        [  FAIL  ] 

Using input file : ./01-xc_1d.oep_exx.inp 

 OEP Exact Exchange                       :      [   OK   ]       (Calculated value = -1.07717830) 

-----------------------------------------

Match OEP EXX Int[n*v_xc]                      :

   Calculated value : -2.14114269
   Reference value  : -2.14242036
   Difference       : 0.0012776699999999
   Tolerance        : 0.0001

 OEP EXX Int[n*v_xc]                      :      [  FAIL  ] 

-----------------------------------------
}}}

------------------------------------------------------------------------
r11517 | dstrubbe | 2013-11-14 12:58:24 -0500 (Thu, 14 Nov 2013) | 2 lines

Correcting misleading output from full OEP: it always said "converged" even when it was not.

------------------------------------------------------------------------

Bugfix (regarding performance) to r8180: when I moved to use forces_derivative instead of pert_expectation_value, I accidentally left it inside a loop over modes. As a result, rather than reducing the amount of work a bit, instead it multiplied the work by the number of modes. Now this is fixed.

------------------------------------------------------------------------
r11435 | dstrubbe | 2013-10-30 18:04:17 -0400 (Wed, 30 Oct 2013) | 6 lines

Bugfixes for the three directions of the spectrum in oct-casida_spectrum; the isotropic average is not affected.
* The length units for the x, y, z dipole matrix elements were not considered, so the results were incorrect when using eV_Angstrom units.
* The spectrum for x, y, z was mistakenly divided by the dimensionality of space; this is only supposed to be done for the isotropic average.
* Corresponding updates to test values.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11422 | dstrubbe | 2013-10-29 16:12:25 -0400 (Tue, 29 Oct 2013) | 2 lines

Making an error message more informative.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11386 | dstrubbe | 2013-10-23 17:13:10 -0400 (Wed, 23 Oct 2013) | 4 lines

One of the three parts is relevant to 4.1.x:

* Bugfix: an attempt to block full OEP parallel in states was ineffective since it was placed before initialization of the OEP method.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11380 | dstrubbe | 2013-10-22 19:12:23 -0400 (Tue, 22 Oct 2013) | 2 lines

Making messages_not_implemented calls for H(F) with k-points at the point of reading TheoryLevel, and broadening the condition in X(exchange_operator): what matters is the number of k-points in the full, not symmetry-reduced, Brillouin zone.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11356 | dstrubbe | 2013-10-17 17:14:35 -0400 (Thu, 17 Oct 2013) | 2 lines

A variable was not initialized for independent particles.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11302 | dstrubbe | 2013-10-08 18:10:46 -0400 (Tue, 08 Oct 2013) | 3 lines

* Minor bugfix to r7250: the sanity check for parallelization gives misleading information when there are slaves, since real_group_sizes is compared to the total number of MPI tasks in the output, whereas the error condition is flagged when mc%group_sizes and MPI tasks do not match.
* Typo.

------------------------------------------------------------------------

URL::
------------------------------------------------------------------------
r11292 | dstrubbe | 2013-10-04 10:38:22 -0400 (Fri, 04 Oct 2013) | 2 lines

The debug output for finite-difference weights overflowed the format for derivatives 1D test. I made it much larger.

------------------------------------------------------------------------

 * Tagging version 4.1.1.

------------------------------------------------------------------------
r11225 | micael | 2013-09-02 18:29:04 +0100 (Mon, 02 Sep 2013) | 2 lines

 * Clarified supported UPF versions in the Species block documentation.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11224 | micael | 2013-09-02 18:26:34 +0100 (Mon, 02 Sep 2013) | 2 lines

 * More informative error message when not finding a tag in the UPF file format.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11212 | xavier | 2013-08-30 19:45:56 -0400 (Fri, 30 Aug 2013) | 4 lines

Bugfix: fixed out of bounds access when the number of platforms in the
system is larger than the requested number of platforms.


------------------------------------------------------------------------

------------------------------------------------------------------------
r11210 | xavier | 2013-08-30 18:36:57 -0400 (Fri, 30 Aug 2013) | 2 lines

Bugfix: A function had the wrong uppercase name.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11182 | micael | 2013-08-28 14:49:31 +0100 (Wed, 28 Aug 2013) | 2 lines

 Bug fix: in some specific situations, the total number of atomic wavefunctions for a given species could exceed the hard-wired value of 6.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11104 | dstrubbe | 2013-08-09 16:51:29 -0700 (Fri, 09 Aug 2013) | 2 lines

Sternheimer calculations should read with 'exact' since a different grid is not acceptable.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11101 | dstrubbe | 2013-08-09 15:27:26 -0700 (Fri, 09 Aug 2013) | 2 lines

Blocking Hartree-Fock calculations parallel in states, which will crash or be wrong.

------------------------------------------------------------------------

Fix to allow previous backport to compile: the structure was altered a bit since the 4.1.x branch split off.

------------------------------------------------------------------------
r11092 | dstrubbe | 2013-08-08 12:11:42 -0700 (Thu, 08 Aug 2013) | 2 lines

Bugfix: the check for more than one block in lobpcg was not very accurate. (r10774)

------------------------------------------------------------------------

------------------------------------------------------------------------
r11091 | dstrubbe | 2013-08-08 11:39:55 -0700 (Thu, 08 Aug 2013) | 2 lines

Hartree-Fock was broken for parallelization in k-points (and probably states too) due to incorrect communication in the Poisson solve. I think I fixed the same problem in the OCT exchange, and I suspect it is present in GCM and RDMFT too (which I commented on but did not change).

------------------------------------------------------------------------

------------------------------------------------------------------------
r11079 | dstrubbe | 2013-08-07 11:45:53 -0700 (Wed, 07 Aug 2013) | 2 lines

Ensuring initializations in restart.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11068 | dstrubbe | 2013-08-05 16:31:06 -0700 (Mon, 05 Aug 2013) | 2 lines

Initialization, for the case UnoccMaximumIter = 0.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11056 | dstrubbe | 2013-08-01 14:37:17 -0700 (Thu, 01 Aug 2013) | 2 lines

Bugfix: loop went only to 2 instead of 3 in a poisson_fft routine. Fortunately, I don't think this mattered in most casees. (I caused this in r8523.)

------------------------------------------------------------------------

------------------------------------------------------------------------
r11035 | dstrubbe | 2013-07-22 14:47:16 -0400 (Mon, 22 Jul 2013) | 2 lines

"true compiler" support for Cray compiler wrappers cc and ftn (which may or may not be in parallel).

------------------------------------------------------------------------

------------------------------------------------------------------------
r11027 | dstrubbe | 2013-07-21 20:22:42 -0400 (Sun, 21 Jul 2013) | 2 lines

This test, with task parallelization, requires MPI2.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11025 | dstrubbe | 2013-07-21 17:31:46 -0400 (Sun, 21 Jul 2013) | 6 lines

Fixing two small issues with the extraction of the "true compiler":
* echo from sh, at least on MacOS, does not support the -n option, so this was ending up in the output from true_compiler.sh. However, if we use bash everywhere, this seems portable.
* MPI compilers/wrappers usually, but not always, called mpicc and mpif90. For example, MacPorts openmpi calls them openmpicc and openmpif90; Intel MPI calls the mpiicc and mpiifort. I made true_compiler.sh be a bit more general, and try -show as long as the name contains 'mpi', which includes all these cases.

Also, for consistency and general good practice, I used /usr/bin/env bash rather than just bash, and did the same for svn_release_number.sh.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11019 | dstrubbe | 2013-07-19 11:48:10 -0400 (Fri, 19 Jul 2013) | 3 lines

* Expanding format for mesh_index since there can easily be more than a million points. More than a trillion is probably unlikely ;-)
* Documentation about display_mesh_index.gp.

------------------------------------------------------------------------

------------------------------------------------------------------------
r10779 | dstrubbe | 2013-06-07 14:12:21 -0400 (Fri, 07 Jun 2013) | 2 lines

Typos, updated link.

------------------------------------------------------------------------

Specifying --without-external_zoltan in configure would trigger an attempt to compile zoltan, even if MPI is not being used, which would inevitably result in failure to find 'mpif.h'. COMPILE_ZOLTAN should be false unless MPI is enabled.

(This problem does not exist anymore in trunk since zoltan support was decommissioned.)

------------------------------------------------------------------------
r11003 | dstrubbe | 2013-07-13 20:47:17 -0400 (Sat, 13 Jul 2013) | 2 lines

newuoa should be in the list of configuration options.

------------------------------------------------------------------------

------------------------------------------------------------------------
r11013 | acastro | 2013-07-17 13:54:31 +0200 (Wed, 17 Jul 2013) | 4 lines

Bug fix for the problem reported by several users: the laser printed in file
td.general/laser changes after performing a restart.


------------------------------------------------------------------------

------------------------------------------------------------------------
r10983 | dstrubbe | 2013-07-11 14:45:35 -0400 (Thu, 11 Jul 2013) | 3 lines

* Fixing possible loss of precision in netcdf and cube output where 'real' is used.
* Noting which options in OutputHow give only the real part.

------------------------------------------------------------------------

------------------------------------------------------------------------
r10981 | dstrubbe | 2013-07-11 13:05:22 -0400 (Thu, 11 Jul 2013) | 2 lines

It seems GOMinimumMove and GOTolerance have to be negative to be definitely not used as convergence criteria, since tolgrad or toldr exactly zero is possible. Setting them negative makes the condition impossible.

------------------------------------------------------------------------

------------------------------------------------------------------------
r10988 | acastro | 2013-07-12 12:08:07 +0200 (Fri, 12 Jul 2013) | 5 lines

The code allows the possibility of changing the mass of the particles, but
then forgets to consider this change of mass when applying the magnetic terms
of the Hamiltonian.


------------------------------------------------------------------------