2011-05-13 01:15 xavier * [r7893] .: 1.1.0 release of libxc. 2011-05-12 16:12 xavier * [r7882] Created libxc 1.1 branch. 2011-05-12 15:57 xavier * [r7881] Libxc now defines XC_GGA_XC_LB to point to XC_GGA_X_LB. This keeps compatibility. 2011-05-11 12:46 marques * [r7863] Added the Thakkar approximation to the kinetic energy density. I believe that this concludes the GGAs for tau (at least I do not have any further on my list) 2011-05-11 10:19 marques * [r7862] The 4 versions of the PBE for kinetics by Tran and Wesolowski 2011-05-11 10:01 marques * [r7861] Small bug corrected 2011-05-11 09:57 marques * [r7860] The correlation version of the previous functional 2011-05-11 09:17 marques * [r7859] Added the APBE functional (both X and K, I will implement C now) 2011-05-01 01:17 xavier * [r7827] Missing deallocations in libxc. 2011-05-01 00:07 xavier * [r7826] The destructor of the gga_c_pbe functional was missing. 2011-04-21 13:26 marques * [r7739] An unused variable was left behind. Thanks David. 2011-04-20 20:04 marques * [r7733] Sorry for the mess. Let's see if it compiles now. 2011-04-20 14:04 marques * [r7731] Sorry forgot two "M" 2011-04-20 14:02 marques * [r7730] New functional, a modified Becke 88 for proton transfer reactions. 2011-04-18 14:18 micael * [r7720] * Two functionals had the same identifier. 2011-04-16 19:12 micael * [r7711] * Fixed AM05 correlation for very small densities. 2011-04-15 10:06 marques * [r7707] Also v2rho2 is well calculated in spin polarized mode. Now it is only missing v2sigma2 and v2rhosigma 2011-04-01 09:42 marques * [r7659] Now the VW term is defined withour the TF part as requested by Micael 2011-04-01 08:59 marques * [r7658] Sorry I had left some bugs around 2011-04-01 08:57 marques * [r7657] Added the kinetic counterpart of PW91 2011-03-31 15:22 marques * [r7653] Another bunch of functionals for the kinetic energy. These ones are all absed on a Pade approximant. 2011-03-30 13:02 marques * [r7647] Added the dePristo and Kress GGA for the kinetic energy density 2011-03-30 12:17 marques * [r7646] Bug fix in K_OL2. The formula was not correct. Added the functional X_OL2, but I have very strong reasons to believe that the formula in the paper is wrong. 2011-03-28 16:15 micael * [r7635] * The lambda parameter in the TflW functionals was not set correctly. 2011-03-28 15:30 micael * [r7633] * Another KED functional that was labeled as exchange. 2011-03-28 15:06 micael * [r7629] * A couple of KED functionals were labeled as exchange functionals. 2011-03-28 15:02 marques * [r7628] Rename the LB functional as "exchange", otherwise octopus can not sum the PW92 correlation to it. Added a modified LB potential by the Amsterdam group 2011-03-27 05:10 dstrubbe * [r7626] Fixed typos and funny characters in libxc. 2011-03-25 13:23 marques * [r7615] An assert was wrong 2011-03-25 13:23 marques * [r7614] Added a kinetic version of the PW86 exchange functional 2011-03-25 13:05 marques * [r7613] Sorry, a factor was wrong 2011-03-25 13:04 marques * [r7612] Another version of becke 88 for the kinetic energy density 2011-03-25 11:30 marques * [r7611] Bug fix: in the case of the kinetic energy density, I was using the wrong power in the definition of x. It should always be 4/3 and not 5/3 2011-03-24 14:14 marques * [r7604] Added a new LDA functional for the kinetic energy 2011-03-24 13:24 marques * [r7603] Again problems with the pre-factors. Hopefully this way of doing things is a bit more robust. 2011-03-24 11:01 marques * [r7602] Two more kinetic functionals OL1 and OL2 Fixed one reference (thanks Tobias) 2011-03-24 08:46 marques * [r7601] Yes, you were right Tobias. Thanks. 2011-03-23 14:33 marques * [r7598] Now the LLP kinetic functional shares the same routine as B88 2011-03-23 14:11 marques * [r7597] Added a bunch of functionals of the form gamma T_TF + lambda T_vW 2011-03-22 16:09 marques * [r7590] Sorry, forgot an } and had an error in a comment 2011-03-22 16:00 marques * [r7589] Corrected the value of K_FACTOR_C to the correct spin-polarized and including the factor of 2 from our definition of tau. Added the von Weiszaecker functional 2011-03-21 15:45 marques * [r7588] New kinetic energy GGA (actually the first of all), the Lee, Lee and Parr functional 2011-03-18 15:15 micael * [r7573] * Now I think the constant in the Thomas-Fermi functional is finally correct. 2011-03-18 10:24 micael * [r7572] * A constant was wrong in the Thomas-Fermi kinetic energy functional. * Definition of XC_KINETIC was missing from libxc_master.F90. 2011-03-03 13:23 marques * [r7533] Added a very simple exchange functional, the Herman X\alpha\beta functional. This is perhaps the first GGA for the exhange 2011-03-02 12:42 marques * [r7521] I rewrote lyp to the new framework. Now the file is half the size as before, and it will be easy to get the second derivatives. 2011-03-02 12:42 marques * [r7520] There was a problem in the b86 functional. First, what I called b86_r was in fact b86. Then my b86 had a wrong coefficient. So I deleted b86_r and rewrote in a clearer way b86. 2011-03-02 12:41 marques * [r7519] Some programs pass negative densities to libxc. This commit should fix many of the functionals in this case. 2011-02-28 11:11 marques * [r7507] Added a new GGA correlation functional: Wilson and Ivanov. Note that this functional only has a spin-unpolarized version 2011-02-24 15:26 marques * [r7493] *) Added the Gombas interpolation of the correlation energy *) Corrected a numerical coefficient in the RPA functional *) A factor of 1/2 was missing in the ML functional 2011-02-23 10:29 marques * [r7490] Sorry, just realized that I used // comments... 2011-02-23 10:28 marques * [r7489] Added a new GGA functional (WL) and more importantly, added a new framework for GGA correlations. Most GGA correlations will be adapted to this new framework in the future. Note: the second derivatives in the spin polarized case are still missing. 2011-02-18 11:33 xavier * [r7479] An attempt to fix the libxc problem in nowii. 2011-02-17 13:10 marques * [r7471] Patch by Janne.Blomqvist@tkk.fi: The attached patch does some minor optimizations and cleanups, mostly replacing usage of pow() with cbrt() as pow() can be somewhat expensive. - Add tests for sqrtf, cbrtf, cbrt in configure.ac - Add SQRT and CBRT macros to xc_config.h - pow => POW if argument is FLOAT - sqrt => SQRT if argument is FLOAT - cbrt => CBRT - POW(x, 1.0/3.0) => CBRT(x) - POW(x, -1.0/3.0) => 1.0/CBRT(x) - POW(x, 1.0/2.0) => SQRT(x) - POX(x, -1.0/2.0) => 1.0/SQRT(x) - POW(x, 1.0/4.0) => SQRT(SQRT(x)) - sqrt(2.0) => M_SQRT2 Some of the pow(x,y) optimizations above can be done automatically by the compiler if one enables -ffast-math (or equivalent) but due to different corner-case behavior the compiler cannot enable them otherwise (e.g. sqrt(-0.0) == -0.0 != pow(-0.0, 1.0/2.0) == +0.0). See e.g. http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25620 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43419 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42694 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=28252 I ran the testsuite, and while I did get some failures I also got exactly the same failures with the trunk, so as far as the testsuite is concerned the patch does not affect the numerics in any way. 2011-02-14 13:12 marques * [r7456] Flag XC_FLAGS_3D was missing from the hybrid functionals 2011-02-02 13:14 marques * [r7394] Some debug lines were forgotten... 2011-02-02 00:37 dstrubbe * [r7388] Corrected typos. 2011-01-25 12:59 marques * [r7329] There was still one inline left. This seems to be the source of the confusion of the xlc compiler... 2011-01-25 12:49 marques * [r7328] Tried to remove inline to see if nowii compiles libxc 2011-01-25 12:45 marques * [r7327] *) changed the order of files in makefile.am to see if it helps with nowii *) Added fxc for the BR functional 2011-01-19 23:21 xavier * [r7287] Changed the macro used to initialize libxc so it works with old libtool versions. 2011-01-18 20:46 dstrubbe * [r7285] Fixed PGI compilation error via missing prototype. PGC-S-0043-Redefinition of symbol, xc_gga_k_tflw_set_params_ (gga_k_tflw.c: 55) 2011-01-18 19:39 dstrubbe * [r7284] * Added hyphen, changed m-dash to hyphen (for page numbers) which could render strangely. * The new file was missing .c in the Makefile. 2011-01-18 15:34 marques * [r7283] Another kinetic energy functional 2011-01-18 15:13 marques * [r7282] Added a new kinetic energy functional 2011-01-16 05:52 xavier * [r7277] Some optimization for lda_xc_teter93. 2011-01-16 04:18 xavier * [r7276] Some optimizations for libxc: * Defined a faster function calculate to f(z). It calculates all the required derivatives at once, and avoids the calculation if the system is unpolarized. * The new function is used by the lda_xc_teter93 functional. 2011-01-02 04:53 xavier * [r7217] By default libxc is now compiled statically. This will probably save a lot of problems for users (and for us). 2010-12-16 14:01 marques * [r7190] Had a misplaced ';' 2010-12-16 13:47 marques * [r7189] Finished the interface for the fxcs in the meta-ggas. Moreover, I think now that the TPSS exchange fxc is correct. 2010-12-15 16:19 marques * [r7179] The rest of the derivatives are there. I will test it more thoroughly tomorrow, and finish to change the interface to include also the derivatives of the laplacian of rho. 2010-12-15 15:39 marques * [r7177] First steps at the second derivatives of the meta-GGAs. Several things are already OK for the tpss (exchange) but there are still some terms missing. 2010-12-14 09:54 marques * [r7172] BUG fixes: *) Fxc was missing from AM05 flags (thanks to Gian-Marco Rignanese) *) Vxc was wrong for x TPSS. Now I get exactly the same results as GPAW 2010-12-08 13:40 marques * [r7147] Bug fix. Thanks to Gian-Marco Rignanese for this! 2010-11-02 19:24 olivares * [r7067] Had to correct the safe_allocate_A calls that sometimes certain compilers go crazy about when commented 2010-10-29 20:55 marques * [r7056] test 2010-10-29 08:55 marques * [r7055] test 2010-10-29 08:25 * [r7054] test 2010-10-25 18:29 dstrubbe * [r7049] A correction to my r7041: now the right modules are associated in the single/double precision cases. This solves the warnings about redefining a rule for these targets. 2010-10-25 16:52 micael * [r7048] TPSS correlation is not an exchange functional ;) 2010-10-25 13:15 marques * [r7046] Should not divide by zero in P86 2010-10-25 12:04 marques * [r7045] Becke functionals should not divide by zero when the density is zero ;) 2010-10-25 11:37 marques * [r7044] Bug fix: KT1 and KT2 were completely wrong. 2010-10-25 11:13 marques * [r7043] Bug fix: rge2 was not working. 2010-10-20 20:19 dstrubbe * [r7041] Added some missing rules and dependencies in libxc. Running a parallel make failed miserably before this, although evidently it just happened to work anyway in serial. Something is still not perfect since I get these warnings after "make -j install": Makefile:817: warning: overriding commands for target `xc_f90_lib_m.mod' Makefile:811: warning: ignoring old commands for target `xc_f90_lib_m.mod' Makefile:817: warning: overriding commands for target `xc_f90_types_m.mod' Makefile:811: warning: ignoring old commands for target `xc_f90_types_m.mod' 2010-10-18 14:40 marques * [r7034] Now the 1D CSC for beta=1 has a spin dependence 2010-10-14 15:10 marques * [r7032] Added the second-order gradient expansion of the kinetic energy functional. 2010-10-14 14:19 marques * [r7031] Updated the citations. Thanks Tobias! 2010-10-14 14:14 marques * [r7030] Added the Thomas-Fermi functional for the kinetic energy density. 2010-10-14 12:27 marques * [r7029] Added two new functionals, Keal and Tozer version 1 and 2. 2010-10-05 12:42 marques * [r7023] Added a new functional: a reparemetrization of PW86 to be used in vdW calculations. 2010-09-28 16:34 micael * [r7022] The exchange potential of the Becke-Johnson MGGA type should be zero if the density is zero in order to avoid getting some NaN. 2010-08-17 17:57 dstrubbe * [r6939] *) Made two tests run for MPI as well as serial. There doesn't seem to be any reason they shouldn't be run in parallel. *) Updated Coordinates block to only have the appropriate number of dimensions in these tests. *) Typo. 2010-08-17 07:14 juho * [r6938] Wrong sign (the equation is solvable down to -1), wrong return value. Returning zero gives us an effective lda approximation. See PRB 76, 235314, page 4. 2010-08-16 11:54 juho * [r6934] Corrected the sign of the PRP10 correction 2010-08-16 11:01 juho * [r6933] PRHG10 2D exchange functional along with optional PRP10 exchange correction. See: http://prb.aps.org/abstract/PRB/v76/i23/e235314 and http://prb.aps.org/abstract/PRB/v81/i11/e115108 2010-08-02 23:26 xavier * [r6898] * Removed some debugging output. 2010-08-02 20:04 xavier * [r6897] * The XC potential for BR89 class of functionals is not set to zero for low density and tau. This fixes the problem with the cusp. The problem for low densities remains. * The solution of the non-linear equation in the BR89 functionals works properly for small positive values. 2010-07-29 14:10 marques * [r6880] Added reference to Slater's Xalpha method. 2010-07-26 08:30 marques * [r6859] Bug fix in the Fortran interface of f90_hyb_gga_exx_coef. Thanks to Tyrel M. McQueen and John Kay Dewhurst for this one ;) 2010-07-14 19:04 dstrubbe * [r6819] autoconf on libxc is failing on some systems and demanding this line. 2010-07-09 12:49 marques * [r6804] As octopus is vulgaris, then libxc can be ordinarius 2010-07-09 12:43 marques * [r6801] Changed the version number to 1.0 2010-07-06 12:02 marques * [r6781] Added another functional to the PBE family of functionals. Another one from Scuseria, the RGE2, both the exchange and the correlation part. 2010-06-13 12:44 acastro * [r6722] Roll back to previous version (6720), since the idea of moving the obsolete variables to one place was not so good... 2010-05-27 21:58 xavier * [r6665] Improved the numerical accuracy of the non-linear solver in the BR89 class of MGGAs. This means that the equation converges for all values of the rhs and the bisection is no longer needed. 2010-05-26 08:02 marques * [r6659] Forgot an else 2010-05-26 07:58 marques * [r6658] Added the parametrization of the CSC functional for the soft Coulomb interaction and a=0.5 2010-05-05 19:09 olivares * [r6604] Returned INSTALL to its original configuration 2010-05-05 15:25 olivares * [r6601] updates on the electrostatic boundary conditions 2010-03-25 10:48 xavier * [r6390] Moved the definition of XC_FAMILY_KS_INVERSION to octopus. Since libxc is an independent library we should keep it separated from octopus specific variables. Anyway we should find a better way to deal with this situation. 2010-03-25 08:15 appel * [r6389] Adding KS inversion as a new family of functionals in a spirit similar to OEP. There is now a new module xc_ks_inversion_m that contains most of the inversion code. I have adapted the 'old' runmode invert_ks to use this new infrastructure. To deal with dependencies I had to move the eigensolver module from scf to system (as suggested by Miguel). This is analogous to the linear solver in the case of OEP. Not all things are in place yet to use the functional family in the general case, however the old runmode should work as before. 2010-03-24 08:27 marques * [r6383] I believe this is the correct way of handling the dependencies. Sorry, I had this lingering around already for a couple of weeks, but I forgot to commit it. 2010-03-23 18:34 xavier * [r6380] Another attempt to get dependencies right. 2010-03-23 18:14 xavier * [r6379] My "fix" for libxc makes things worse. Someone knows how to fix this? 2010-03-23 17:36 xavier * [r6378] A hack to fix the dependencies in libxc. 2010-03-22 16:46 xavier * [r6372] Fixed some autotools warnings in libxc. 2010-03-21 11:06 xavier * [r6362] Now libxc can deal with compilers that use uppercase filenames for modules. 2010-03-20 19:12 xavier * [r6359] * Now libxc installs the Fortran modules. * Added a macro to detect the module extension. 2010-03-08 22:28 dstrubbe * [r6347] *) Improved formatting *) Added push_sub's *) Changed single-letter variable *) Set svn:ignore on another automatically generated file in libxc 2010-03-06 03:38 dstrubbe * [r6344] *) Improved formatting and comments *) Added push_sub's, and removed one that appeared an excessive number of times in the debugging stack trace *) Set svn:ignore properties for libxc files generated by libtool 2 2010-03-02 15:55 marques * [r6334] This is an overkill hack in order for libxc to compile with different versions of libtool. Let's see what I broke now... 2010-03-02 00:21 dstrubbe * [r6329] Set svn:ignore for new files being created in libxc. 2010-02-26 16:22 marques * [r6325] Stupid hack to go around buggy libtool/automake. This works in tddft.org, let's see the others now. 2010-02-26 15:42 marques * [r6324] Let's see if this fixes the compilations issues. Needless to say that it compiles without problem in my machine... I hate the autotools! 2010-02-26 14:01 marques * [r6323] Added the option to compile libxc dynamically, and a target to create libxc.rpm I had to use libtool, which led to some problem. Let's see if all the architectures are happy with them... 2010-02-09 09:40 marques * [r6309] Renamed the PS94 functional to the name given in the original article, and added a second version of the functional 2010-02-09 09:27 marques * [r6308] Now also vxc is working for this functional. I will leave fxc and kxc for the future ;) 2010-02-08 15:12 marques * [r6307] Sorry, there was a semi-colon missing 2010-02-08 14:11 marques * [r6306] Initial implementation of the Proynov and Salahub 94 LDA. The energy is working and giving the right results. Now it is time to optimize and to compute the derivatives. Yes, I was also surprised that there are even more LDA functionals. This one is anyway different, as it is not a parametrization of the CA Monte-Carlo data, but uses a quite different approach. 2010-02-04 10:11 marques * [r6302] Small correction pointed out by Damien Caliste. 2010-02-03 15:58 marques * [r6301] I finished cleaning the interface of libxc. The release of 1.0beta will follow in the next days. 2010-02-01 09:27 micael * [r6295] *) The calculation of the laplacian of the density was wrong in the spin-polarized case. *) A variable was not initialized in libxc for the MGGAs when the density was bellow the threshold. *) Some small fixes to the etsf_io output. 2010-01-19 11:29 marques * [r6265] Several fixes for the bugs found by Xavier while setting up the xc test. Fairly trivial things, but anyway quite annoying... 2009-12-15 14:37 marques * [r6226] Bug fix: there was a "-1" missing in an index which caused some compilers to return garbage (which was the correct answer). This should fix the 1D test that is currently failing. Please 1D people check your calculations!!!! 2009-12-15 12:12 marques * [r6225] Bug fix in Pulay mixing: a range was wrong in an allocation Bug fix in 1D CSC functional: variables were not deallocated 2009-12-08 13:17 marques * [r6198] *) Bug fix when calculating v2rho2 in mixed GGAs *) The optPBE_vdW functional was wrong *) Now xc-reference.pl knows about build directories 2009-12-08 09:02 dstrubbe * [r6197] *) The kdotp run mode now prints out band velocities, which in fact needs only the perturbation Hamiltonian and not the perturbed wavefunctions. *) Broke up kdotp_output into kdotp_write_degeneracies and kdotp_write_eff_mass *) A couple typographical things in LDA C PZ. 2009-12-03 20:17 marques * [r6190] Bug fixed: this was triggered by the changing of the thresholds, but it was there for a long time. Now the tests should pass. 2009-12-02 17:16 micael * [r6185] More fixes regarding very small densities. 2009-12-02 10:59 marques * [r6184] Some cleaning regarding very small densities 2009-12-01 09:22 marques * [r6175] Sorry, this file should not have been committed: reverting it. 2009-12-01 09:10 marques * [r6174] 3 new functionals optimized to be used with the Dion et al vdW functional. See http://arxiv.org/abs/0910.0438 2009-11-24 13:16 xavier * [r6128] The macro required by C99 is not available in autoconf 2.59, that is the one in tddft. So we will have to stay with c89 for the moment. 2009-11-24 12:51 xavier * [r6127] Now Octopus and libxc set the C compiler in C99 mode. 2009-11-19 20:49 dstrubbe * [r6083] A few spelling issues. 2009-11-18 09:53 marques * [r6077] Added a bunch of 2D GGA functionals. They are still under testing, so don't rush to do calculations with them... 2009-11-09 09:03 marques * [r6046] *) Now the relativistic corrections are also present in higher derivatives *) Corrected warning in the interface of the meta-GGAs 2009-11-09 05:21 dstrubbe * [r6045] *) Capitalization of SPARSKIT *) A u-umlaut got turned into garbage in a comment, so I replaced it with "ue" 2009-11-07 12:20 micael * [r6040] *) Put the relativistic correction to the exchange part of the LDA back into libxc. *) Replaced the Xe UPF pseudopotential with a correct one and removed all the other UPF pseudos. *) The SO coupling test is again active. 2009-11-06 09:47 marques * [r6033] Bug fix: some functionals were not working in 2D. 2009-11-06 07:52 marques * [r6032] Fixed a couple of bugs: *) LB94 was not working due to two problems with the interface *) oct-center-geom was segfaulting due to incorrect initialization of geometry 2009-11-03 16:51 micael * [r6014] The gradient and the laplacian of the density used to compute the xc potential are now computed directly from the wavefunctions. 2009-10-30 16:01 micael * [r6007] *) Bug fixed: the LB94 functional was not working for the spin-polarized case because of a couple of misplaced brackets. 2009-10-16 08:08 marques * [r5940] Cleaned up the hybrids. There is no more the specific interface hyb_gga_xxx, as the generic gga_xxx handles the requests. 2009-10-13 11:53 marques * [r5939] Added teh asymptotic expansion fo LB94 2009-10-09 08:36 micael * [r5933] *) Fixed a couple of bugs in libxc. *) Changed the initial point for the Newton-Raphton routine in the Becke & Johnson functional. 2009-10-01 08:30 marques * [r5923] This corrects the assertion failure of Fulvio 2009-09-25 18:46 dstrubbe * [r5916] *) In the file xc_oep.F90, an error message is written if in parallel saying "Full OEP is not allowed with the code parallel in states." However, the actual condition implemented in the code is whether it is parallel at all. I'm not sure if the condition, or the message, is wrong and should be changed. *) Improvements to documentation, comments, and output. *) Added and corrected push_sub's. *) Lengthened some single-letter variables. *) Alphabetized module lists. *) Made tests be called curvilinear instead of 'curvlinear' 2009-09-24 12:47 marques * [r5909] *) I started changing the interface of libxc following a suggestion of the devlopers of ELK. Hybrids should nto work, and some functionals may give segmentation faults. In any case, all tests are passed in my machine *) Added the functional of Esa and Stefano *) Corrected the builddir in mk_functionals_list.pl *) Added dummy integers to mpi_debug.F90 mpi_lib.F90 so that compilers stop complaining *) Decreased the precision of the SIC test 2009-09-09 08:32 marques * [r5886] *) Now octopus is also aware of the 1D exchange functional *) Replaced the _init_exchanged and _init_correlation by a simpler _init_functl *) Some more tests are passed in 4D. The exceptions are now the open systems tests and the periodic_systems/06-h2o_pol_lr.test *) kpoints are now properly deallocated and the volume of the unit cell is also properly calculated in 1D and 2D 2009-09-07 11:31 marques * [r5885] Now the 1D exchange functional seems to be working. It includes both spins,a nd the two forms of the interaction. I also implemented Fxc. 2009-08-28 23:24 dstrubbe * [r5870] Some changes to documentation, comments, and output. 2009-08-28 10:23 marques * [r5869] *) Added the LDA exchange in 1D. This functional is still not functional *) Added the LDA correlation in 1D of Casula for a soft-Coulomb interaction *) Corrected the segmentation fault found by Fulvio 2009-08-24 18:50 dstrubbe * [r5853] *) Removed troublesome ' in comment in libxc, in the file produced by the new script, as well as what I believe is the source from which it arises *) Made mk_functionals_list.pl refer to its actual name *) Corrected spelling and punctuation in variable documentation 2009-08-21 09:21 marques * [r5840] Added the soft-Coulomb version of the 1d_csc lda 2009-07-21 08:56 marques * [r5740] This should solve the problem of David. 2009-07-15 09:46 marques * [r5718] Some more changes related to the interface. I am sure that many of the non-standard functionals have problems, so be careful. 2009-07-15 09:23 marques * [r5717] Now the GGas are also done in blocks. The current interface is just a hack to make things work. I hope to be able to fix it soon. 2009-06-30 11:03 marques * [r5661] *) now the LDA works on vectors to a very low level *) Rearranged th exchange and xalpha interfaces to make them more standard Note that libxc at the moment has a inconsistent interface. I will try to fix it soon. 2009-06-30 09:47 marques * [r5660] *) First step in the dircetion of vector calling libxc in order to optimize performance. 2009-06-29 09:55 marques * [r5654] Some more corrections to the TP09 functional. I believe that it might be working now. 2009-06-25 15:43 marques * [r5636] One can not run the PB09 thing, but, as expected, the calculation for a finite system is kind of unstable... maybe it is better for a solid ;) 2009-06-25 14:34 marques * [r5634] Now the mggas are again implemented in the code. 2009-06-25 10:47 xavier * [r5633] Now octopus can be cross compiled. To do it, he user has to pass two variables to the configure script: FC_INTEGER_SIZE=4 CC_FORTRAN_INT=int With this change octopus compiles in a Blue Gene/P. 2009-06-24 15:19 marques * [r5631] Forgot a factor. 2009-06-24 15:18 marques * [r5630] Added the famous new functional of Tran & Blaha. This was not tested, and there is a factor of 2 that is bugging me... 2009-06-24 14:26 marques * [r5629] *) I made a mistake in the year of becke-roussel *) hopefully it now compiles 2009-06-24 14:15 marques * [r5628] *) Hartree-Fock now works with spinors (I hope) *) Added Becke-Roussel Meta-GGA. This one should be working asa it was tested against a piny_md. *) Because of that i had to change slightly the interface to libxc 2009-06-03 23:47 dstrubbe * [r5555] *) Changed svn:ignore properties to get rid of a bunch of ? from svn status. *) Added FromScratch = yes to some tests to remove warnings about being unable to read from restart directory. *) Added a Cl pseudopotential since I am using this frequently. 2009-05-27 11:05 marques * [r5508] Small changes required to compile octopus with ifort 2009-05-26 09:23 marques * [r5495] *) Added a MemoryLimit variable. Octopus will stop if this variable is set and if the total memory depasses its value *) Octopus was not compiling in 4D due to array mismatches 2009-05-18 08:25 marques * [r5417] Changes necessary to compile octopus in single precision more. Alberto, I guess I made a disaster in your opt-control code (that now will not wortk in single precision). The problem are the minimizing routines that do not have interfaces in single precision more. One should add it... 2009-05-13 10:54 marques * [r5380] Added a couple of reparametrizations of the PBE by the group of Klaus Capelle. Note that all the other combinations present in the paper can be obatined by combining the already existing exchanges and correlations of PBE and PBE_SOL. For example: GCGX = x_PBE_SOL + c_PBE 2009-04-21 07:52 marques * [r5297] I implemented the inversion of the Kohn-Sham equations (static). It seems to work in simple cases, but the iterative procedure does lack stability, and fails to converge for more complicated densities... 2009-04-17 12:56 marques * [r5257] Changed the license with the script that Tobias sent. Thanks Tobias. 2009-04-13 22:25 xavier * [r5202] This fixes the problem with make distcheck. 2009-03-31 09:29 marques * [r5133] A couple more fixes to the make environment 2009-03-31 09:09 marques * [r5131] Little corrections to some stupid files 2009-03-26 10:27 marques * [r5116] Several details necessary to package libxc. *) There is now a PACKAGING file *) FCCPP explanation now appears in ./configure --help *) Automatic generation of ChangeLog file -- maybe it would be better to separate it per years, but I don't know how to do it well... 2009-03-20 14:40 marques * [r5092] Forgot this file... sorry. 2009-03-20 12:00 marques * [r5091] Some more fixes, and now HL also has Kxc. 2009-03-20 09:48 marques * [r5090] Reorganization of the LDAs of reduced dimension. Now there is a common driver routine in work_lda that handles 1, 2, and 3 dimensions. I also took the opportunity to add fxc and kxc to the 2d ldas (both exchange and amgb). I do not know if his is useful (someone interested in hyperpolarizabilities of 2D systems?), but I added it for completeness. 2009-03-17 10:03 mjv500 * [r5086] MIGUEL CHECK THIS stray float 2009-03-16 08:33 marques * [r5085] Som cleaning in the ggas 2009-03-11 09:00 marques * [r5078] It is easier to maintain a copy of this file here, so that this library is indeed independent. 2009-03-11 08:58 marques * [r5077] Now also VWN has analytical Kxc 2009-03-09 16:52 mjv500 * [r5069] had to fix dependencies after removal of libstring_f - someone competent check this 2009-03-09 14:08 marques * [r5067] Now pw also has analytical Kxc 2009-03-09 12:30 marques * [r5065] I got rid of libstring_f. It was replaced simply by a header file (string_f.h) that is included twice in octopus (liboct_parser and libxc/src). The main difference is that TO_C_STR passed to a subroutien instead of a function. I used the opportunity to fix the strcasestr definition that was quite broken. 2009-03-09 10:16 marques * [r5064] Now Perdew-Zunger has analitic 3rd derivatives. Due to that, the values of the hyperpolarizability test changed slightly (4th digit) 2009-03-09 09:11 marques * [r5063] *) Added support for the calculation of kxc (the 3rd derivative of Exc) in libxc. I also changed the interface to something more reasonable. Now we have, when spin is polarized, kxc(4) = (uuu, uud, udd, ddd) *) Now octopus does stop if one tries to run with fxc or kxc != LDA. Up to now it was simply ignoring fxc (i.e., it was performing an RPA calculation). David and Xavier, please check this one out. 2009-03-05 13:04 marques * [r5036] *) Simplified a bit more the interface to the routines. *) Added support for spin-unpolarized Kxc, and added the third derivatives of wigner and rpa corelation functionals. 2009-03-05 09:13 marques * [r5035] Cleaned a bit the LDAs. 2009-02-25 17:40 xavier * [r5001] Due to the 31 character limitation I had to rename the lda_c_2d_prm08 functional to lda_c_2d_prm. Miguel, can you check that I did it correctly? 2009-02-25 10:25 marques * [r4998] I implemented the 1D correlation LDA of Barone's group. Note that this has not been tested, and that there is the exchange missing... 2009-02-06 00:23 dstrubbe * [r4925] The macro that checks for "very long lines" used a test program with a line that was very long, but still not nearly so long as some lines generated by the ALLOCATE preprocessor macro in the code. The PGI 8.0-1 Fortran compiler passed that test in the configure script since it could handle 140 characters in a line, but could not handle 270 characters. I approximately tripled the length of that line in the test program to 433 characters, so that now PGI is considered not to accept long lines. 2009-01-05 10:41 marques * [r4848] Added a new form for the exachange enhancement factor that comes from a bayesian best fit procedure 2008-12-02 08:38 marques * [r4809] Added a new hybrid, XC_HYB_GGA_XC_mPW1K (modified Perdew-Wang for kinetics) 2008-12-01 09:58 marques * [r4808] Added the Schmider-Becke 98 Hybrid GGA. This is again a reparametrization of the Becke-97 form. There are 6 versions of it in that paper, and I included them all. Although the functional is a hybrid, I had to include also the "pure" GGA versions. The keywords are: for the "pure" GGAs XC_GGA_XC_SB98_1a, XC_GGA_XC_SB98_1b, XC_GGA_XC_SB98_1c, XC_GGA_XC_SB98_2a, XC_GGA_XC_SB98_2b, XC_GGA_XC_SB98_2c to be used as a hybrid XC_HYB_GGA_XC_SB98_1a, XC_HYB_GGA_XC_SB98_1b, XC_HYB_GGA_XC_SB98_1c XC_HYB_GGA_XC_SB98_2a, XC_HYB_GGA_XC_SB98_2b, XC_HYB_GGA_XC_SB98_2c 2008-11-28 15:40 marques * [r4803] Cleaned up a bit the hybrid GGAs, and added some more hybrids on the process. Let me see if I don't forget any: XC_HYB_GGA_XC_mPW3PW, XC_HYB_GGA_XC_B1LYP, XC_HYB_GGA_XC_B1PW91, XC_HYB_GGA_XC_mPW1PW, XC_HYB_GGA_XC_mPW3LYP 2008-11-28 10:07 marques * [r4801] *) Fixed the parameters of mPW91. I believe that this functional is correct now *) Added the correct parameters for the 2D B86 MGC functional 2008-11-27 10:39 marques * [r4795] There was a small but in b3lyp that was solved. Now libxc gives exactly the same thing as the repository, so I submit the repository data 2008-11-27 10:36 marques * [r4794] *) Rewrote the way that functionals that are linear combinations of others are handled. *) Added three new GGAs of this type that were constructed for water: XC_GGA_XC_PBE1W, XC_GGA_XC_MPWLYP1W, and XC_GGA_XC_PBELYP1W 2008-11-26 10:24 marques * [r4793] Sorry, there was a misspel that broke the compilation 2008-11-26 09:43 marques * [r4792] Added the hybrid functionals corresponding to the becke 97 GGAs. They are: XC_HYB_GGA_XC_B97, XC_HYB_GGA_XC_B97_1, XC_HYB_GGA_XC_B97_2, XC_HYB_GGA_XC_B97_K, and XC_HYB_GGA_XC_B97_3 2008-11-25 10:35 marques * [r4788] Forgot this file... 2008-11-25 10:31 marques * [r4787] Rewrote the B97 class of functionals. Now it also includes fxc and some mroe elements of this family. BTW, Xavier, included is the functional of Grimme that you sent me long ago... of course that to use it, one should also add the vdW term (trivial, actually) 2008-11-20 15:18 marques * [r4778] Added two other variants of Becke 97 2008-11-20 14:50 marques * [r4777] Added Becke 97 functional 2008-11-14 13:49 marques * [r4768] Small things I found while trying to compile in the new IBM Power 6 of CNRS 2008-10-22 09:52 marques * [r4697] Some bugs in the LTA corrected. I think one-shot calculations with LTA are fine, but self-consistent calculations give nonsense... 2008-10-22 07:06 marques * [r4695] *) Make check should work now. Can you please check, David? *) Some bug fixes in the mggas, but there are still issues 2008-10-20 14:21 xavier * [r4686] The name of a libxc function was too long for fortran. 2008-10-20 13:21 marques * [r4685] I implemented the self-consistent MGGA functionals. However, calculatiosn with them do not converge! I am not sure if it is a bug somewhere, or numerical instabilities. I am inclined to the first option, of course, but I am not really sure. 2008-10-20 09:39 marques * [r4683] Sorry for the mess: it should compile now! 2008-10-20 09:32 marques * [r4682] Cleanup the mess of 2d and 2D. Now, everything is lowercase. 2008-10-20 09:23 marques * [r4681] Added 2d to the functionals in 2d 2008-10-20 09:12 marques * [r4679] I added support for mgga calculations in the one_shot CalculationMode. Note that we can _not_ dor self-consistent mgga calculations at the moment. I also added a new 2D functional that is going to be submitted soon. 2008-10-17 16:54 marques * [r4671] Sorry, 2D GGAs will only go in the near future ;)) Now it should compile. 2008-10-17 14:55 marques * [r4670] *) Added support for the vsxc correlation functional. Again, I have to repeat that meta-GGAs have not been properly tested! 2008-10-08 13:00 marques * [r4615] Some small fixes 2008-10-07 07:45 marques * [r4598] Added the exchange part of the tau-HCTH functional 2008-10-06 14:22 marques * [r4591] Added the exchange part of VSXC (called GVT4 for whatever reason). 2008-10-06 12:46 marques * [r4590] *) Changed the copyright notice to something more "correct" 2008-10-06 12:43 marques * [r4589] *) Exchange in MGGA now is written in terms of x and t *) Added the exchange part of the M06l functional 2008-10-01 15:02 marques * [r4581] I have implemented the correlation functional of TPSS. I made the usual tests for the derivatives and they seem to work. However, I would consider this functional highly experimental ;) 2008-09-17 10:02 marques * [r4528] I believe that now the exchange in TPSS is working. It has been transformed to the new interface. It still has to be tested in real circunstances, of course. 2008-09-16 14:51 marques * [r4523] I now believe that the LTA is working. 2008-08-25 13:22 marques * [r4512] Sorry, forgot this detail 2008-08-25 13:08 marques * [r4511] Initial support for meta-GGAs. For now, I only added the local tau approximation, but without real testing. I will proceed to add also the TPSS. Note that for now, octopus is not able to use these functionals. 2008-08-25 09:19 marques * [r4510] *) added von Barth and Hedin LDA *) Now langreth & Mehl depends on the vBH LDA as it should 2008-08-21 14:12 marques * [r4504] Added Langreth & Mehl GGA correlation 2008-06-13 11:09 marques * [r4277] *) Added a 2D correlation functional by Esa, Stefano and me to the library. 2008-05-30 07:45 marques * [r4208] Some people were having problems with the definition of M_E. Apparently, it is not standard (ansi) C, so we redefine it in case we are asked strict ansi 2008-05-07 08:56 marques * [r4170] Small problem with the PBE solved. 2008-05-07 08:46 marques * [r4169] Implemented the second derivatives of the PBE xc functional. There seems to be still a small error when for v2rho2 for spin unpolarized. 2008-05-07 07:31 marques * [r4168] Added new hybrid functional (B1WC) 2008-04-30 13:18 marques * [r4128] Now also FT97 gets second derivatives. This took me almost a day to do :(( 2008-04-29 13:41 marques * [r4124] AM05 (exchange part) gets second derivatives also 2008-04-29 11:41 marques * [r4123] Secon derivatives of the WC functional 2008-04-29 10:05 marques * [r4122] Added two new functionals: mPBE and xPBE (both exchange and correlation) 2008-04-29 08:23 marques * [r4121] Now also Teter has second derivatives 2008-04-29 07:56 marques * [r4120] Now also PW92 is OK (including the second derivatives) 2008-04-28 14:29 marques * [r4117] Added the second derivatives to Perdew-Zunger. 2008-04-28 14:04 marques * [r4116] *) Now B3LYP has the correct definition of the VWN spin interpolation (correct from the point of view of Gaussian implementation) 2008-04-28 13:15 marques * [r4115] *) Added second derivatives of vwn functional 2008-04-28 11:45 marques * [r4114] *) Code should now compile in single precision mode *) I am rewritting the LDA part. I created a system similar to the exchange in GGA, where one only has to write a formula in terms of rs and zeta. *) Added the second derivatives os a number of LDA functionals. 2008-04-28 08:12 marques * [r4112] Commented all references to the PRM functional that is currently ni the drawing board. Hopefully, the problems will be solved, and I will commit it soon ;) 2008-04-28 07:53 marques * [r4111] *) Added a run mode that allows for one-shot calculations. This means that we can do the self-consistent cycle with one xc functional, and then use that density and wave-functions to calculate the energy using another functional. Note that this does not work in HF or with Hybrid functionals! This also requires some cleaning in libxc, so do not be surprised if it segfaults (see libxc/DONE file for a resume of what is working in a clean way) *) Added a lot of fxc for exchange GGA functionals. Now the next task will be the fxc for the LDA correlation functionals. 2008-04-18 09:06 xavier * [r4080] Incorporated spglib as a external library (http://spglib.wiki.sourceforge.net/), a C code to search symmetries of crystals. All source files are a clean copy of version 91 from the svn repository. 2008-04-02 16:54 marques * [r4029] Macro call missing. 2008-04-02 14:21 marques * [r4024] Added new functional Madsen 07. This was sent to me by the author. 2008-04-02 14:14 marques * [r4023] I believe that now the implementation is correct. It does give the same energy as the reference implementation I have from the authors, and xc-consistency gives me a small error, which probably means that the derivatives are correct. 2008-03-31 20:19 marques * [r3992] Added a first implementation of AM05 correlation. This is still not working. I again have some doubts concerning the spin-polarized implementation, and I am waiting for the authors to answer mye mail before finishing my implementation. 2008-03-31 19:59 marques * [r3991] I also forgot this file. Let us see if it now compiles. 2008-03-31 14:41 marques * [r3987] I believe that I forgot these files. 2008-03-31 14:38 marques * [r3986] This should solve the problem of Xavier with the g95 build. It is not pretty though, and I would be happy if one of you automake experts could clean it... 2008-03-28 14:16 marques * [r3961] Forgot this file 2008-03-28 14:02 marques * [r3960] *) Removed dependency on the GSL. By other words, I reimplemented the Lambert function I needed. *) I believe that the exchange part of AM05 is now working *) Some more work on fxc in the GGAs. Now xc-consistency also calculates the finite difference fxc to test against analitical values. There are still some issues with spin in this routine 2008-03-26 15:34 xavier * [r3950] Because of the linking order, sometimes gsl couldn't be found. 2008-03-22 16:48 xavier * [r3926] gsl.m4 is also required by libxc. 2008-03-18 13:55 xavier * [r3907] * Added a check for the inline keyword in libxc. 2008-03-18 12:57 marques * [r3905] Added _experimental_ and _incomplete_ support for the AM05 functional. For now, there is only exchange (correlation is actually simple), and I have no idea on how to write it for spin-polarized systems. I have contacted the authors of the functional to sort out the issues. As AM05 needs a lambert function, I had to add a dependence on the GSL. I plan to remove it further on (I will have to copy the function...) 2008-03-17 11:16 marques * [r3900] First attempt at fxcs using GGAs. For now, two functionals are implemented: B88 and LB93. There are still some issues for spin _unpolarized_, as I am not sure of some factors of 2. 2008-03-10 08:11 xavier * [r3854] This should fix the compilation problems in libxc. 2008-03-10 00:43 xavier * [r3852] Having both libxc.F90 and libxc.f90 confused automake, so I renamed libxc.F90 to libxc_master.F90 (for lack of a better name). 2008-03-09 23:24 xavier * [r3851] File libxc.F90 was also missing from Makefile.am. 2008-03-09 09:49 xavier * [r3845] One libxc file was missing from the distribution. 2008-03-07 08:24 marques * [r3835] Let's see if aramis is able to compile in single precision mode like this 2008-03-06 11:59 xavier * [r3831] Libxc still had problem with vpaths, this fixes it (I don't know if it is the correct way to do it, though). 2008-03-06 08:55 marques * [r3829] I hope this solves the compilation issue in pepita. I hate this vpath thing!!! 2008-03-05 14:42 marques * [r3822] I hope this fixes the automatic compilation 2008-03-05 08:38 marques * [r3818] Dependency missing in the makefile. Hope this solves your problem, Alberto... 2008-03-04 09:26 marques * [r3809] Now we can compile libxc in both float and double. This is accomplish by adding _s_ to the name of the single precision routines. I did not find a way of overloading the routines in Fortran in order to provide for a single interface to both single and double. P.S. Some tests are failing in my machine in single precision (apparently unrelated to libxc), and the wavepacket test is failing (in double precision) 2008-03-03 09:04 marques * [r3802] Now we can compile libxc either in single or in double precision. I did not like the previous way it was done, i.e, but having _dp and _sp in the routines and doing all the necessary conversions. I think that this is cleaner. 2008-02-20 09:31 marques * [r3728] Added the mPW functional. I still have some doubts concerning the constants in the functional (they were taken from http://scsg20.unige.ch/~dulak/files/functionals.pdf), so they will eventually have to be checked. 2008-02-16 14:30 xavier * [r3697] The declaration of "inline void" functions causes problems with the Sun C compiler. Patch submited by Marcin Dulak. 2008-02-15 13:37 marques * [r3694] *) Added support for Ilya Tokatly's pressure as a measure of localization. For this I reorganized the output layer of octopus. I hope it wortks ;) *) Added the teter 93 functionals to libxc 2008-02-12 11:24 marques * [r3680] My previous commit was not correct: xc_funcs.h should be installed, but not included in the distribution. 2008-02-12 11:18 marques * [r3679] xc_funcs.h was not being installed. 2008-01-30 10:10 marques * [r3667] Bug correction from Jussi Enkovaara: I found a small bug in the PBE of libxc. In the function "perdew_params" in the file "gga_perdew.c" the total gradient "pt->gdmt" /* get gdmt = |nabla n| */ pt->gdmt = sigma[0]; if(pt->nspin == XC_POLARIZED) pt->gdmt += 2.0*sigma[1] + sigma[2]; can become a small negative number (in my antiferromagnetic system ~ -1e-20), and the following square root produces then NaN. I fixed that with a simple if statement i.e. /* get gdmt = |nabla n| */ pt->gdmt = sigma[0]; if(pt->nspin == XC_POLARIZED) pt->gdmt += 2.0*sigma[1] + sigma[2]; if(pt->gdmt < 0.0) pt->gdmt = MIN_GRAD*MIN_GRAD; pt->gdmt = sqrt(pt->gdmt); 2008-01-23 13:53 marques * [r3660] Another memory leak fixed, this time in xc_f90_lda_end 2008-01-22 16:45 marques * [r3653] Another bug found by valgrind. Xavier, can you check this one? 2008-01-09 13:12 marques * [r3638] *) Checking of hybrids is now performed *) The guys at the functional repository now get the same PW91 functional as me, so I can commit their test file 2007-11-26 16:05 acastro * [r3607] The xlc IBM compiler complained about using global variables within "inline" functions. I just made a copy of the affected constant definitions within the function bodies. 2007-11-23 20:39 xavier * [r3597] * Now instead of using an integer of the size of a C pointer, there exists an opaque c_pointer_t type that has two methods associated, set_null and is_null. This is the same we have already for libxc. * For the parser blocks there is a different type, block_t. * The only exception is xc_functl_write_info where a pointer is used both as a pointer and an integer. The variable is defined as integer(SIZEOF_VOIDP), but we should fix it. * Removed the check for the non standard type 'long long' as a candidate for equivalence fortran integer, in theory this could be a problem, but is very likely that if a fortran integer is 8 bytes also a 'long' is 8 bytes. A better solution would be to check first if long long or int64_t are available, and then use them. * Removed the check for the size of a pointer in C, instead the size of void * is measured using the standard macro. With this changes octopus compiles with gcc -pedantic-errors. 2007-11-23 12:49 acastro * [r3593] I was trying to compile in an opteron using the Portland C compiler, and I got problems with the SSE2 extensions ("Illegal instructsion"?). I changed a bit the beak.h preprocessor test to avoid its usage. But Xavier, you may want to take a look at that change. The rest of the files are just "//" style comments, which the Portland C compiler does not accept. --This line, and those below, will be ignored-- M src/beak/beak.h M external_libs/libnbc/nbc_ialltoall.c M external_libs/libnbc/nbc.c M external_libs/libnbc/nbc_ibcast.c M external_libs/libnbc/nbc_iallreduce.c M libxc/src/test.c 2007-11-23 11:19 xavier * [r3588] * The variable ResponseMethod now validates its value. * The C minimizer code was doing some very illegal things with pointers, now it is legal C89 code (at least for gcc -pedantic). * Removed some compiler warnings. 2007-11-22 12:47 marques * [r3586] Updated TODO list of libxc 2007-11-22 12:43 marques * [r3585] Two more hybrid functionals: PBE0 and X3LYP. I have still a list of another 7 hybrids to include. 2007-11-19 16:43 marques * [r3553] Made the references more uniform. 2007-11-19 14:33 marques * [r3551] Forgot this file. 2007-11-19 14:32 marques * [r3550] Now octopus can use hybrid functionals. The exact exchange part is handled within the OEP, while the rest is taken care by libxc. I found very hard to test these functionals. Quantum chemical calculations use basis sets and full potential. The ideal would be a 3D jellium sphere, or something like that. If you know of one such calculation... I will now try to implement it using Hatree-Fock for the exact exchange part. 2007-11-16 16:43 marques * [r3528] Initial support for hybrids. I also implemented B3PW91, which historically was the first hybrid (I think). This is not at all tested, of course. 2007-11-15 15:58 marques * [r3521] The LB94 functionals is now working. There are two versions, one modified and another unmodified. The unmodified gives the same results as my 2001 calculations. I did not test the modified version, though. I made some more cosmetic arrangements, like adding the "_m" and "_t" suffixes to libxc.F90 2007-11-14 16:14 lorenzen * [r3512] Bugfix: I had to insert the $LDFLAGS at two more spots in the libxc AC macros. 2007-11-14 15:14 lorenzen * [r3510] Bugfix: the check for Fortran integer size in libxc configure forgot about $LDFLAGS during linkage. 2007-11-12 11:30 marques * [r3493] Dependecy missing: when compiling without fortran, xc_funcs.h was not generated. 2007-11-12 10:57 marques * [r3491] I built a common framework for John Perdew's functionals. This simplified a lot both PBE and PW91. Incidentally, I found why my implementation of PW91 (correlation) was giving slightly different results than the reference implementation in the Density functional repository of Daresbury. It is related to the number of digits in one constant. I believe the error is in the repository, but I mailed them to sort out this issue. 2007-10-26 16:08 lorenzen * [r3431] Bugfix: this change repairs the make distcheck target, so the distcheck Bot should not fail anymore. Some dependencies had to be made explicit because Automake's dependency tracking only has limited support for built headers. 2007-10-25 13:51 marques * [r3417] Added a TODO file 2007-10-25 13:32 marques * [r3416] Two new functionals GGA functionals: Wu & Cohen XLYP 2007-10-24 10:14 xavier * [r3394] Small fixes: * Inline functions can't use static global variables. * Removed unused multigrid code and added some debug info. * Assembler is now preprocessed using FCCPP, this is because Sun C compiler (cc -E) gets hanged with .S files. * Workaround for a bug in gfortran openmp support. * Fixed a misspelled openmp clause. 2007-10-21 17:57 acastro * [r3387] Added some new generated files to the svn:ignore lists. 2007-10-18 09:47 marques * [r3375] Added EDF1 functional. This is again a fit, this time by Adamson, Gill, and Pople. 2007-10-17 12:03 marques * [r3371] I changed the license to the LGPL 3.0. I will also do that to libparser. I think that the LGPL is a bit better for these licenses if one wants them to be included in other (non-GPL) programs. Note: I added only me to the copyright line, as most of the work in this library has been done by me. If, however, if you changed any of the files and want your name to appear, please add it. 2007-10-17 10:59 marques * [r3369] Added the HCTH class of functionals. These are fits to several (sometimes 407) molecules. They seem to be the most precise GGAs for a large class of molecules. 2007-10-16 14:54 xavier * [r3366] Now the libxc_funcs module is used through libxc. As libxc is a standalone we should have a simple (and constant) interface. 2007-10-16 13:53 marques * [r3364] Added pw91 correlation functional. Sometimes, I get differences in the 9th digit when compared to reference values. I am not sure if it is a real bug or not. 2007-10-16 07:08 acastro * [r3363] The package created by "make dist" did not contain the get_funcs.pl script. 2007-10-16 03:02 acastro * [r3362] Depending on which machine I built the code (from scratch, doing autoreconf + configure + make) I got an error... I think the problem is in the use of "+=" to define variables in automake, and in the resulting ordering of the compilation (the file xc_funcs.h is not created before it is needed). I think it is more robust in this way. 2007-10-15 17:41 lorenzen * [r3359] The mgga and lca files were not processed. I don't know if this was intentionally but at least Octopus wants them. 2007-10-15 16:19 lorenzen * [r3358] This fixes the build problems with the automatically generated interface files for libxc. The problem was that the interface files were built in the source tree and not in the build tree. Now, $(srcdir) and $(top_builddir) are passed to the get_funcs.pl script. Passing these two variables is nearly always necessary when writing scripts like these. 2007-10-15 15:20 marques * [r3357] I hope to have fixed the compilation issue now. 2007-10-15 14:54 marques * [r3356] I wonder if this solves the problem with the compilation now 2007-10-15 14:20 xavier * [r3355] This fixes compilation of libxc when building outside the sources directory. I don't know if it is the correct way to solve it, though. 2007-10-15 13:37 marques * [r3353] Forgot this file... 2007-10-15 13:36 marques * [r3352] Now there is a file per functional. I also added a perl script that allows to generate automatically the interfaces. Like this, in order to add a functional we do not have to touch any other part of the code. One just has to add a single file! 2007-10-15 09:48 marques * [r3351] Added two more functionals: PBE_SOL (x and c), and RPBE (x) 2007-10-08 18:25 xavier * [r3321] Bugfix: CFLAGS were not properly passed in some macros. This can cause problems if one of your flags is -m32. 2007-09-01 19:36 xavier * [r3155] More fixes to the single precision version. Now it is possible to run gs calculations without segfaults. 2007-09-01 13:16 xavier * [r3153] Added a single precision interface for the rest of libxc calls. 2007-09-01 11:17 xavier * [r3152] Fixed a bug in my previous commit. 2007-09-01 10:29 xavier * [r3151] * Added single precision interface lda functions in libxc, for the C interface functions with an extra _sp are single precision, for Fortran polymorphism is used. * Declared input arguments as const for lda functions in libxc. 2007-07-25 08:10 xavier * [r3101] Bugfix: M_SQRT2 was not available for libxc when the C compiler is in ANSI mode. Reported by GPAW developers. 2007-07-04 09:26 lorenzen * [r3064] * Added an environment variable SKIP_CHECK that can be set during a make distcheck to omit the testsuites: $ DISTCHECK_CONFIGURE_FLAGS="..." SKIP_CHECK=yes make distcheck This change is to provide the possibility to let the BuildBot run make distcheck regularly without the testsuites (just to find forgotten files in the Makefile.ams). this environment variable is honoured in testsuite/Makefile.am and libxc/testsuite/xc-run_testsuite. * Fixed a bug in the clean-local rule in share/PP. * Added *.mod and others to CLEANFILES in libxc/srx/Makefile.am 2007-06-19 08:50 acastro * [r3007] Finally, also (most of) the files generated by a typical "make" will also be ignored by the svn commands by including them in the svn:ignore lists. 2007-06-19 08:06 acastro * [r3006] Now (almost all) the files generated by configure are also included in the svn:ignore properties. 2007-06-19 07:58 acastro * [r3005] Added to the svn:ignore property of several directories the files generated by autoreconf. 2007-06-19 07:52 acastro * [r3004] Removed the .cvsignore files, useless since a lot of time ago, I guess. 2007-06-14 16:26 xavier * [r2998] Bugfix in libxc: The interface of xc_f90_lda was incorrect and the interface of xc_f90_lda_vxc was missing. 2007-06-14 11:11 xavier * [r2997] * Added some modifications to libxc configuration files suggested by Yann Poullion to ease the integration with Abinit. 2007-06-13 17:49 xavier * [r2995] Moved the fortran interface of libxc to the libxc/src/ directory. This Instead of the C_POINTER, I defined some derived types (with a pointer inside), so in theory they should have enough space to hold a pointer but I have to test it. This approach is also more elegant as the types can't be passed incorrectly and their contents are private, so the user can't modify them without libxc. 2007-06-12 11:24 xavier * [r2991] Bugfix: libxc was trying to compile the fortran interface even when --disable-fortran was specified. 2007-05-24 12:47 acastro * [r2926] Removed C++ ("//") comments in favor of old-style C comments ("/* */"), since some compilers refuse them. 2007-03-20 18:29 lorenzen * [r2771] This is a boring maintenance commit concerning the automake build and distribution system. It is likely that some of the nightly builds will fail or that someone encounters building problems. In that case, just bug me, I feel in charge to fix that. I wanted to get VPATH builds working, i. e. compiling in a directory that does not contain the source code. This makes it easily possible to configure and build a single source tree with lots of different options (avoiding tedious copying). VPATH is basically a GNU make feature but the Makfefile.ams have to follow certain restrictions to get it working: 1. Relative file names should be constructed with the $(srcdir), $(top_srcdir), $(builddir), $(top_builddir) variables. 2. Wildcards should not be used in automake variables. This means that all source files have to be mentioned in the Makefile.am. This is a bit tiresome (especially in the PP directory) but it is recommended in the GNU Coding Standard and considered good practice. So, please add any .inp and .test files you include in the testsuite to the corresponding Makefile.ams. This also holds for the .psf files (and the .hgh) if a new element should be discovered... 3. I had to change the way the variable documentation is generated a bit. The main thing was to add command-line flags to specify input and output directories to the var2html.pl and mk_varinfo.pl scripts. 4. Filename flags for the compilers (like -I) should not be added to the corresponding variable in the configure.ac but to the AM_ version in the approripate Makefile.am (or to its per target variant). This makes use of $(srcdir), etc. possible. In some Makefile.ams the compilation rule is redefined, in these cases, it has to be checked if the AM_ variant is included (e. g. src/Makefile.am). There were some more things I ran across: 1. The testsuite can only be run after a make install because the share data is not available otherwise. For that reason, I replaced the check target by an installcheck-local rule. 2. The oct-run_testsuite script used hardcoded paths, they are now patched by the configure script. So, the oct-run_testuite is replaced by oct-run_testsuite.in and the original script will disappear in my next commit. 3. The libxc testuite also needed some fixing to consider different source and build trees. 4. I completely removed the AC_CONFIG_SUBDIRS for the python tree. The conditional configuration of a subtree is a very subtle issue as automake does not support not configured subtrees. The configure.ac and Makefile.am of the python part seem to be in a very fragile state, and as it is not used anyway at the moment, I did not spend time to fix them. By the way, how is the state of the GUI? There is just few lines of code in the repository. When it will be included again, I recommend just another Makefile.am and no subpackage configure. Necessary checks should be added to our main configure.ac (and can be skipped, if no GUI is requested). The code now passes the distcheck target which is desireable. 2007-01-19 17:57 xavier * [r2656] * Added reference to mgga_c_tpss * Changed indentifier from i to u for the output of a size_t value in write_iter.c. * This fixes #8 , the rest of the bugs were already fixed. 2006-11-18 17:39 xavier * [r2590] Optimization: * Wrote a C version of dnl_operator_operate. The inner loop is unrolled by hand to depth 3 and prefetching directives are used to keep the array of weights in the cache. For a test case (benzene gs with non interacting electrons, h=2.0 r=6.0), total performance was increased by a factor of 1.47 for ifort 9.1 and by a factor of 2.2 for gfortran (p4 3.0GHz). In both cases gcc-4.1 was used. The new function is not enabled by default yet, but it is selected by the input variable OperateFunction, values can be either 'fortran' or 'c'. * Configure detects the presence of builtin functions __builtin_expect and __builtin_prefetch in the C compiler. If they are not found, alternative dummy macros are provided. Fixed a couple of bugs revealed by gcc-2.95: * Assert is not always available in C. Put a check in configure.ac. * Move a declaration to the beginning of the function. 2006-11-12 12:36 xavier * [r2569] Libxc header is now usable from C++. 2006-10-26 10:49 marques * [r2518] Added a workaround so that the functionals do not return NaN when sigma=0. 2006-10-25 08:48 marques * [r2517] Slater's Xalpha functional was not working, so the test was failing. The test testsuite/finite_systems_3d/11-sic.test seems a bit weird to me. The functional used is Slater's Xalpha with alpha=2/3, which means that it only uses exchange (it would be easier to put Cfunctional=none ;) Is this really the functional used in the reference (that I corrected, BTW)? 2006-10-23 15:56 marques * [r2511] I am resyncing my tree that started diverging ;) *) Further cleanup of libxc. Now it was fxc that now has only 3 components (uu, ud, dd) for spin-polarized calculations. This amounted to changes in libxc and in octopus. Note that kxc is still not changed. It should include only the independent elements! Maybe Xavier would like to do it ;)) *) Added a Perdew Wang modified lda functional. This is the functional used by Kieron in his implementation of the PBE (that i assume is the reference), and that is different from the published functional (different number of digits in some costants, and he even has incompatible definitions for the same constants). In this way, the PBE now gives the same as the reference numbers, so I rehabilitated the test. The differences to the previous implementation are very small (4th or 5th digit, but anyway...) *) Now we can have all-electron species also in 1 and 2 dimensions. 2006-10-16 11:11 marques * [r2500] Todays functional is the exchange part of the Filatov & Thiel functional (both A and B versions). At least the B version gives the same results are a reference implementation. 2006-10-15 08:57 marques * [r2495] Hopefully, this fixes the compilation errors of the last couple of nightly builds. Trivial changes. 2006-10-13 14:13 marques * [r2491] I believe that these file names are more convinient. 2006-10-13 09:32 marques * [r2490] Today's functional is Perdew 86. For this one I have reference data, so I can be pretty sure it is well implemented... 2006-10-12 09:19 marques * [r2488] Another little exchange functional: Lacks & Gordon 1993. The list of GGA exchange functionals to go is decreasing, but I still have the GGA correlation (another 10 functionals), and the metaGGAs (probably another 15 functionals)... life is hard ;) 2006-10-11 11:00 marques * [r2485] Today's new functional is dePristo & Kress 87 in two different versions. 2006-10-10 08:15 marques * [r2479] Added the OPTX exchange functional (Handy & Cohen 2001) asked by Angelica. 2006-10-09 07:43 marques * [r2477] *) The nightly build using ifc was not working due to the infamous "static" flag, and subsequent problems with the pthread library. I could not solve cleanly the problem, so I made a workaround: the configure script doe not die, and assumes 4 byte ints. *) A couple of references added/corrected in libxc 2006-10-06 16:53 marques * [r2471] I believe that Axel copied an old version of acx.m4. This is the correct version... 2006-10-06 16:34 marques * [r2470] Sorry, there are several things in this commit: *) I finally understood the problem that Alberto was having. The variable i in xc_functl_write_info was used for two things. In the first it should be a normal integer, and in the second it should be a pointer. That is why the code was crashing. I simply solved the problem by using two variables with different names. *) Added a macro called C_POINTER that has the obvious meaning of integer(POINTER_SIZE). I also changed all occurences in the code of integer(POINTER_SIZE) to C_POINTER. I believe it is much more readable like that. *) Added two new functionals, Perdew Wang 86 and the exchange part of Perdew Wang 91. I think the introduction of new functionals will slow down now ;) 2006-10-06 16:12 athimm * [r2469] Convert symlinks to regular files to please bitten. There is a bug in trac's API concerning extracting properties of nodes containing the metadata. 2006-10-06 10:56 marques * [r2468] *) Added automatic detection of the C type that corresponds to the Fortran type "integer". xc_f.c was the changed to use this macro. This should, hopefully, solve our previous problems. Now libxc has all arguments "pure" Fortran integers. I tested this changes in i386 (ifc7.1), x64 (pathscale, nag, and gfortran), and it passed all tests (with the exception of the hyperpolarizability test that failed in ifc7.1 and *pathscale*, but Xavier alerady knows about it). *) Bug fix in xc-get_data.c. A missing "#include <stdlib.h>" was making the xc tests fail sometimes. 2006-10-05 11:14 marques * [r2462] *) changed the names of all exported routines of libxc to xc_XXX, and the F90 variants as xc-f90_XXX. Now it looks more like a library ;) This, of course, amountd to a lot of small changes in the code... *) pursuing a problem with the exchange part of the PBE (a division by zero), I decided to dump the old routine and to implement it in the more general framework of "Becke 86-like functionals". It is now 5 lines of code! 2006-10-04 14:09 marques * [r2456] *) Added a consistency cjeck on the functionals. Basic it is a script that calculates vxc using finite differences, calls libxc for the analitic functional and compares the difference. By now, three functionals give problems PBE (exchange) has some NAN lingering around; PBE (correlation) and G96 have large deviations from the fd results. *) Added a "provides" line to func_type. This is, e.g., "XC_PROVIDES_EXC | XC_PROVIDES_VXC". In this way, the upper level can know if it has to calculate something using finite differences. 2006-10-03 14:49 marques * [r2452] Fixed the problems with the LYP functional when rho[is]=0. 2006-10-03 14:48 marques * [r2451] * Disabled the test of the correlation part of the PBE. I get differences in teh 4th or 5th digit that I do not understand... * Changed the way the tests were done, so that they are now really passed. 2006-10-02 22:28 marques * [r2448] Forgot this file... 2006-10-02 22:26 marques * [r2447] Next iteration of libxc *) Merged several functionals (B86, B88, G96 and variants) in one routine as they are all of the same form *) Added a routine family_from_id that simplifies initialization of the functionals *) Added the functionals Gill 96 (anyone heard about this one?) and LYP. The LYP functional still has problems when the density is zero in one of the spin channels. So, please don't use if until I fix it. *) Made the corresponding changes on the fortran side. Octopus cannow use all the functionals in libxc. 2006-09-29 13:44 marques * [r2440] Next iteration in libxc. In this new version I have: *) Moved source files to the directory src/. It is neater that way *) Delete the LDA speedup thing that Alberto had implemented, but which he agreed to delete *) Added two more functionals: Becke 86 and Becke 86 reoptimized *) GGAs are now handled in the same way as LDAs 2006-09-28 13:58 marques * [r2434] Started a big cleaning in order to make libxc a bit more elegant.Basically, there is now a mechanism to initialize, run the functional, and terminate it. In this way, there are no more the horrendous switch statements in lda.c. The job is not finished yet. 2006-09-28 11:06 marques * [r2433] *) Added Becke 88 to libxc. However, this functional is still not called from octopus. That will come later. *) The GGAs are now also tested. The exchange PBE and B88 pass all test without problems. The correlation part of the PBE does not, however. The differences are small (5th digit or so), so you may relax ;) In any case, I believe that the problem is, in fact, the "reference" implementation that has an inconsistency in the constants it uses... 2006-09-26 14:26 marques * [r2430] The most important part of this commit is the inclusion of tests for some LDA functionals. The calculated value of these functionals is compared to reference data that can be found in http://www.cse.clrc.ac.uk/qcg/dft/. Note that the tests do give some errors. This is related to fully polarized cases, i.e., when rho_down=0. The reference data hard-wires, in this case, the value of vpot_down=0, which, in my opinion, is not correct. In any case, if the gas is fully polarized, the value of vpot_down is irrelevant in this case ;) There were some very small precision issues in a couple of functionals that I adjusted in order to match the reference data. In the process I added two new LDA "functionals": *) lda_c_pz_mod. This is the version of the Perdew Zunger functional modified by Alberto to match better the low and high rs regions of the parametrization. This is the old pz that we had in the code. The original PZ functional is called lda_c_pz. Note that the default is the corrected functional. *) lda_c_vwn_rpa. This is a parametrization of the RPA correlation energy using the VWN form, and that is also present in the original paper. Soon I will start playing with the GGAs ;) 2006-09-25 08:18 marques * [r2424] First step in Making libxc independent. Now it contains its own configuration script. There are, however, several things still to clean. 2006-06-18 11:32 athimm * [r2221] Exclude build-aux from entering svn. 2006-06-01 09:06 xavier * [r2174] Portability fix, variable declaration was not in the beginning of file. 2006-05-24 10:16 athimm * [r2149] a) python and xml - Make python conditional, on by default. Use --disable-python to turn off python (see ./configure --help) - Remove variables.xml generation from main code, make it python-private. variables.xml are only used by the GUI. Let it live under share for now, we'll see were we'll put it in the future, perhaps it will be removed altogether. b) Add some vpath fixes 2006-05-19 15:17 athimm * [r2117] Update svn:ignore recursively. 2006-05-14 10:52 athimm * [r2073] Ignore config.h stamp file. 2006-05-12 18:41 athimm * [r2063] Sanitize .cvsignore and recursively feed svn:ignore with them. The entries mirror the school of not committing generated files into the repository. Whether this is good or bad needs to be decided, and I'm probably in favour of adding them (as well as maintainer-mode protection), but until this is agreed upon the current curse is `minimal'. 2006-04-08 19:42 acastro * [r1992] Some more profiling tags, and some unimportant changes. For example, now the wavefunctions are not randomly generated if then a LCAO calculation is going to be performed. 2006-04-05 13:52 acastro * [r1982] The option to interpolated the LDA functionals instead of recalculating them every time is now compiled always, although it is only used if the InterpolateLDA is set to yes. I will experiment with it for a few months to see if it is worthed to keep it as an option. 2006-04-04 19:59 acastro * [r1980] Still one bug. 2006-04-04 19:42 acastro * [r1979] The XAlpha correlation functional was badly calculated when code was compiled with the "LDA_SPEEDUP" preprocessor flag. This option is still way too experimental to be compiled by default. If you want to try it, you have to define LDA_SPEEDUP (for example, setting CPPFLAGS="-DLDA_SPEEDUP" when configuring). 2006-03-21 11:51 micael * [r1939] *) Syncing the libxc in APE with the one in Octopus. This new version includes the relativistic correction to the LDA exchange functional. This is not really important for Octopus (at least for the time being) but it is necessary for APE. 2006-03-07 18:41 acastro * [r1881] Bug fixed. The Vosko/Wilk/Nusair xc potential had a bug related to units. Every time the initialization function was called, some "constant" was divided by two, in order to go from rydbergs to atomic units. But the inialization function is called twice by octopus, so the result was a factor of four. 2006-01-31 11:11 acastro * [r1797] Bug fixed. Removed the 05-Na2_LB94_Casida bug. The LB94 is still rather under suspiscion, in any case. I don't know why the bug was there, in fact, it looked that the code that I removed was there for a reason, but I could not trace it down because it was there since the first revision... 2005-11-28 11:03 marques * [r1726] *) very tiny bug corrected in VWN (one of the constants was wrong in the 5th digit) *) fxc is now calculated numerically a bit better in case of the ferromagnetic gas. Actually I don't really know how to handle this case properly, as fxc should diverge (at least fx). 2005-11-12 19:47 acastro * [r1650] Bug fixed. One wrong character kept me busy the whole afternoon. 2005-11-09 09:13 marques * [r1642] *) Alberto, I think that the "break" in the fxc calculation should be a "continue" (i.e., he should skip the rest of the current cycle, but should get out of the cycle). Can you please check? *) Xavier, I guess that the calculation of Kxc should have a similar to the one Alberto put in for fxc. Can you please check? *) I cleaned a little bit the includes, whitespaces, and changed "(*p)." to "p->" as it is more C-like. 2005-11-08 22:01 acastro * [r1641] Bug fixed. The numerical functional derivation was not working properly in some spin-polarized cases (Perdew-Zunger parameterization, Na4(+1), it was just giving lots of NaN). Now, at least for those cases, it seems to work. I still have to do some more testing on the Casida module, though. In any case, the PZ kernel should be obtained analytically. Maybe in the future. 2005-10-18 12:51 xavier * [r1570] Added Kxc, the third derivatives of Exc, to libxc. It's calculated using a second order numerical derivative of Vxc. 2005-10-12 13:43 marques * [r1554] There was a sign wrong in the definition of pw92. Micael, can you please test this functional? 2005-10-12 09:10 marques * [r1552] Some cleaning: *) test.x is no longer generated in libxc/ *) changed the names of several single character variables. Please *do not* call a variable a, b, n, c or whatever!!!! *) I noticed that some of you had tabs of 3 and 4 spaces. Do not forget that the default we agreed upon some years ago was *2*. Please change you emacs settings! 2005-09-05 16:36 acastro * [r1365] It seems that the test_x_LINK variable needs to be defined in this way so that the executable is actually built. 2005-08-30 14:24 acastro * [r1346] Since the test.x program is all C, it is better to link with the C compiler. Otherwise some Fortran compilers (i.e. the Intel 9.0 version) seem to have problems with a duplicated "main" function. 2005-07-11 13:23 acastro * [r1255] The xyzanim utility was broken, at least in the IBM machines. I put an emergency fix. If I get time I will take a look at the utilities, see how the system labels affect them, and make up some tests for them. Also, the IBM preprocessor had problems with the // comments in C files.. 2005-06-30 09:54 marques * [r1220] There were a couple of misspells in function names 2005-06-07 11:58 acastro * [r1201] I forgot to change also this when I changed the config.h stuff. 2005-05-17 09:55 appel * [r1121] Add $Id:$ tags to most of the files (right below the GPL header). At every commit cvs is replacing these tags with lines of the following form $Id: run_regression.pl,v 1.3 2005/05/12 10:23:59 marques Exp $ In this way it is easy to see who was the last person changing the file, which cvs revision the file has and at what date the last change was made. Miguel: Is there a reason why the *.c files in libxc/ don't have a GPL header? 2005-04-06 16:33 marques * [r1081] LOL! Calculating derivatives is always a hassle, even if only of simple functions ;))) Please recheck them. 2005-04-06 14:45 micael * [r1079] *) Added the Vignale-Rasolt functional for CDFT to libxc. Two parametrizations are available. This has *not* been tested carefully yet. 2005-04-04 12:57 acastro * [r1074] [*] A small change in the "nl_operator_operate" subroutines, which, specially in the complex case, seems to speed up things significantly. But most likely it does so only in some architecutures/compilers. In the case of the Portland compiler for opterons, the speedup is rather spectacular, probably because before the type conversion was very slow (the opteron was running slower that my laptop!) [*] In mgga_c_tpss, some Portland C compilers complained about those array definitions, I think it is more correct now. 2005-03-28 10:47 acastro * [r1061] A pretty sick performance hack. The idea is the following: instead of calculating the LDA functionals everytime, one can "spline" them at the beginning, and then interpolate every time, since they are very smooth functions. Some comments: (*) The calculation of the LDA exchange and correlation functionals do not take a significant calculation time, *except* if one wants to use the SIC+LDA within the OEP/KLI approximation for a big molecule. (*) This commit is still experimental, and in fact the code is still exactly the same as it was, if you do not compile adding the LDA_SPEEDUP preprocessor variable (by doing CFLAGS="-O3 -DLDA_SPEEDUP", for example). So you do not have to worry at all about this for the moment... I will make sure that it works well and that it makes sense before making it definive. 2005-03-17 20:32 marques * [r1048] This file was missing from the previous commit. 2005-03-17 17:33 marques * [r1044] Just a small rearrangement. 2005-03-17 17:23 marques * [r1043] *) Removed "relativistic exchange", as, I am not mistaken, the potential was wrong. *) Added support for LDA fxc. For now, fxc for exchange and Perdew Wang correlation are analytical, while the rest is calculated through finite differences. Honestly, from the tests I've made, I am inclined to believe that it is actually better to get the numerical derivative. (However, I would not throw away the work of a full day, would I ;) 2005-02-22 12:03 acastro * [r1014] Bug fixed. A small bug in the definition of the Slater Xalpha potential. (Before it just added again the excahnge fucntional) Also, I have changed the definition of the alpha constant, to meet the usual definition in the literature: E_{xc}^{Xalpha} = (3/2)*alpha*E_x^{LDA} which means that the Xalpha correlation potential should be defined as: E_{c}^{Xalpha} = ( (3/2)*alpha - 1)*E_x^{LDA} I have set the default to alpha = 1. In any case, the logical thing would be to define this functional as a exchange functional, rather than as a correlation part to be added to the exchange. But it is not worthed to loose too much time with this functional anyway... 2005-02-15 13:51 acastro * [r1007] I have put the embryo of what could be the octopus developer's guide. It just contains the description of one module, which is lib_xc, the interface to Miguel's C library that contains the LDA, GGA and mGGA functionals. And in fact it is not even complete. In the way, I have decided to simplify a little bit the interface to the library, so that it is a little bit more user friendly. 2005-02-02 03:07 appel * [r996] now we are complete. 2004-12-15 15:01 marques * [r964] A couple of bugs solved, but I still get non-sense. This functiona is too sensitive to numerical erors in the calculation of derivatives, I think... 2004-12-07 20:40 marques * [r957] MGGA "runs" but gives non-sense! 2004-12-07 15:46 marques * [r956] Bug fixed in correlation PBE (yes, there was a serious bug there) TPSS correlation is apparently working for spin-unpolarized systems. I still did not check the spin-polarized functionals (neither x nor c). 2004-12-01 00:05 marques * [r946] code now compiles and runs with -ansi flag. Hope this solves your problems, Alberto. 2004-11-23 15:53 marques * [r938] I am implementing the TPSS meta-GGA. The exchange part is probably working, the correlation is still a first draft. 2004-10-28 20:34 marques * [r915] several bug fixes. 2004-10-27 18:34 marques * [r912] added lb94 functional. As we all know this functional ahs problems, and I am really not sure if these problems are handled correctly. 2004-10-27 15:03 marques * [r910] There is now a mechanism to provide informations about the functionals being used. 2004-10-26 20:12 marques * [r907] First support for GGAs. Added PBE. 2004-10-21 15:01 marques * [r904] removed an erraneous #include 2004-10-20 16:41 marques * [r900] *) Changed (again) the restart files because: 1) There was a bug reading the occupations and the eigenvalues 2) It did not work for spin-polarized calculations *) Added support for the libxc LDA functionals. Due to this reason, GGAs and the other functionals are desactivated. Now octopus can use the 13 LDAs provided by libxc. 2004-10-20 13:39 marques * [r898] The begining of our library of xc functionals. There are 13 LDA functionals implemented until now. The unpolarized versions work I think. The polarized I have some doubts, and we should do some further checking.