To allow for an easier installation of the debian package, MoleCuilder now integrates levmar, vmg, JobMarket, CodePatterns, LinearAlgebra, and MPQC as ThirdParty packages into its distributable tarball, i.e. these dependencies no longer need to be installed extra. The only extra dependence is Qt and Qt3D (the latter being unofficially provided under DownloadRedistribution).

This makes it easy to place MoleCuilder into e.g. a ​Docker container and have its molecuilder_poolworker sweat away on all those fragment calculations up in some cloud, see the attached Dockerfile​. Moreover, JobMarket is now fully integrated, meaning that a distinct molecuilder_server process takes care of a host of poolworkers that can munch away on the calculations in parallel. These calculations naturally are provided by molecuilder acting as the controller in this server-worker-controller implementation.

Furthermore, the fitting of empirical potential has been significantly enhanced by combining each potential fit function (e.g. dihedral angle bond) with a distinct bond model. This allows to precisely match the bond model with suitable fragments and therefore obtain more general fits of the empirical potential parametrizations. In the upcoming version, this will also work with partial charges and then we will provide some examples.

We added some more complex code for saturating atoms properly. It takes fully into account already present bonds and tries to match to internally stored SphericalPointDistributions as best as possible. This gives really good initial results, i.e. building up molecules just by adding a few carbon, nitrogens and oxygens, when subsequent structure optimization relaxes everything to its proper equilibrium positions.

Moreover, we fixed the fragmentation working on just a subset of atoms, i.e. leaving out saturation water.

In the GUI, the Fragment list can now be properly sorted (and still clicked to get the fragment as set of selected atoms).

FragmentationAutomation will actually fail when the Server is unreachable and not spin forever.

ParseTremoloPotentials is gone, as it is superseded by ParseParticleParameters which is also responsible for parsing partial charges. See also SaveParticleParameters.

Note that for versions in between this and v1.5.0 we refrained from producing debian packages as these did not really bring any good change for the user. Starting from this version, it is again worth to update ... especially with what's coming up!

1.5.3? Where is 1.5.1 and 1.5.2?!

To make things clear from the start, it took quite some effort to produce a clean interface between the GUI and the rest of molecuilder. Hence, v1.5.1 and v1.5.2 are sort of versions in between.

v1.5.3 basically brings mostly internal changes in form of the QtObservedInstanceBoard, namely all of the information about atoms, molecules, ... required for visual representation are copied and updated as needed. This is to make sure that things just live long enough for the sloppy GUI.

Right now, 1% of all test cases do not work in case of using the GUI as promised with v1.5.0.

Apart from that we have:

• The FitParticleCharge action is renamed to FitPartialCharge.
• long range forces working when using vmg.
• interdistance option when using Fragmentation action allows now to combine fragments up to a given order.
• electronic charges are additionally smeared for long-range electrostatic calculations which improves accuracy.

It is done!

From this version onward we provide debian packages (Ubuntu 12.04) for convenience to the user. Please note that MoleCuilder depends on several other packages, three of which had to be manually compiled and are made available at DownloadRedistribution. Especially, Qt3D was quite hard to get to work.

Furthermore, the GUI is now thread-safe!

Qt works with threads, MoleCuilder so far did not. Sure this had to clash. Now, the interface between MoleCuilder and the GUI done with Qt has been secured with mutexes and clean structures and the GUI is safe for use. All the more because it is contained in the debian package. Have fun fooling around with it.

Note that there still might be some issues. True stable execution will probably come up with v1.5.1 but early-adopters are welcome to try it out. For the moment 1 percent of all guicheck testsuite tests (see last versions about that) are still allowed to fail.

We need to switch from ​Codepatterns v1.2.9 to CodePatterns v1.3.0. Most of the GUI tests are failing because of race conditions in the code. CodePatterns v1.3.0 brings the Observer/Observable up to multi-threaded speed by having mutexes in place.

Hence, in this version nothing is really changed but we only switch underlying library versions on a version update.

Prepare for v1.5.0 because then the GUI tests will be working.

This is a smaller update that will some time later make a difference.

First of all, a new action has been added that allows saturating any present atom, i.e. creating methane is as simple as adding a carbon atom and saturating it.

Second, we finally have something like a testsuite for the GUI which should bring this interface up to a usable level real soon. The tests are mostly failing but this should soon be corrected.

Last, some GUI stuff has been fixed and enhanced such as the info boxes that now reveal a bit more about the atom or molecule the mouse pointer is hovering over.

First of all, you can now simply flip a switch and molecules are no longer shown via their atoms and bonds but as a tesselated surface, see the image below of an amylose molecule filled with a regular grid of water molecules. This is especially useful with very large molecules (think e.g. a Tobacco Mosaic Virus or simply some large biomolecule) because a lot less triangles have to be drawn on the screen. Molecules can still be highlighted and selected by moving over or clicking onto the tesselated surface.

Notice also in the picture that the list of molecules is fully working (and hopefully) perfectly. Molecules are aggregated by their formula and can be individually selected or their visibility changed as well as for all of the same formula at once. Changing a molecule by adding an atom immediately changes the list. This rewrite became necessary as empty molecules are now automatically removed.

Furthermore, there's a number of smaller changes that should make the GUI use more easy going:

• the 3D view can now be changed via the keyboard (up/down, left/right, page-up/page-down and zooming with +/-)
• New RemoveMolecule action for convenience (before you had to select-molecules-atoms and remove-atoms)
• Selections can be pushed and popped which is useful inside MakroActions.
• Bonds are restored when undoing a remove action
• Failing actions no longer clean the queue and thus the destroy the stored session file
• when checking the command-line options for a specific action,

  .../molecuilder --help fragment-molecule
  

  is enough, actionname is no longer needed.

With this version we safeguard that the userguide contains at least one example for each and every Action.

The greatest change is however that ... how to put it ... convexification of tesselated envelopes is finally working. This means that volumes of envelopes and thus the space a molecule occupies can be calculated. This is because a convex envelope's volume can be easily calculated via an inner point and general tetrahedrons. If the volume changes can be traced when transforming a non-convex envelope into a convex one, then we also know the volume of the original non-convex envelope. The volume changes however consist of either removing points ("filling up craters") or adding general tetrahedrons, both volume changes are calculable.

As an example, we give two pictures to illustrate these envelopes, here for the amylose molecule, also known as potato starch.

There is a video among the files attached to this blog post where the progress is shown and the above description of removing points and adding tetrahedrons should become a little bit clearer.

This is a very important change because knowing the volume a molecule takes up is essential in calculating densities and thus knowing how many molecules to fill into a domain. See the upcoming amylose example.

Last but not least trajectories stored in the atom class has been changed. Now, we store them as maps, not as vectors. This means that steps in the trajectory do not necessarily have to be strictly consecutive. For example, when performing a structure optimization (with debugging on) the extra trajectory steps generated for a subset of atoms are no problem anymore.

Starting with version 1.4.8 we check for each version that valgrind runs through without glitches (except for some python sessions where valgrind shows too many python errors).

However, greater impact is that the GUI has been made more useful with a lot of smaller changes:

• No more clicking "ok" buttons when the dialog does not contain any queries.
• bonds are now properly updated in the 3D view.
• tooltips are shown for every query in the dialog, explaining its purpose.
• homology tab is working.
• file dialogs finally know something about the file types they are looking for.

With v1.4.7 a better hydrogen saturation procedure was implemented.

So far, we dealt with each bond to be saturated one after the other and could handle up to bond degree of three. The problem with the old code was that placement of saturation hydrogen was independent of that for another bond. This could result in hydrogens being placed on top of each other or at least very close. This occurred for example with the SLES molecule when the sulfur bonds were saturated due to its high valency.

Now, we calculate optimal places for saturation globally per molecule prior to the fragmentation procedure. After fragments have been determined and cut bonds need to be saturated, the required positions of the hydrogen are looked up from a table created by this prior calculation. This ensures that hydrogen placed by the saturation of two different bonds are optimally far away from each other.

Note that hydrogen saturation is a perturbation of the underlying Fock matrix (in the case a Hartree-Fock solver is used). Hence, resulting energies are affected and will now be slightly different.

There is a new make target extracheck that runs when mpqc (​http://www.mpqc.org/) is installed and can found in the path: It compares resulting values for some molecules against stored ones. There, relative differences (for bond order of 3) of up to 1e-5 are admissable!

Furthermore, the version now compiles on Ubuntu 14.04 systems. Some changes were necessary due to new boost and gcc versions.

Last but not least a simple Structural Optimization procedure based on the fragmentation has been implemented. This is so far just using calculated forces and some fixed stepwidth dampening scheme to look for the minimum. Enhancements are coming up but will take some more time.

With v1.4.6 the src/Actions/FragmentationAction/FragmentationAutomationAction.cpp​ can be used even without having [doxygen:install.html JobMarket]. Fragment Jobs (i.e. each fragment is stored as an MPQC file, MPQC is executed on it and the resulting energies and forces are extracted from its output) are launched serially.

Hint: You might be tempted to specify fragment-executable as mpirun -np 6 mpqc in order to have at least multiple threads working on one fragment at a time. However, this will fail cause the option's value is checked whether it exists as a file (and there is no file called "mpirun -np 6 mpqc"). One way out is to create a small script runmpqc.sh as follows

#!/bin/sh

mpirun -np 6 mpqc $@

Make it executable and use it as fragment-executable.

Another feature is that the console log is now shown inside the GUI (there is an extra widget called Log). You can click on an atom's name, e.g. C09, in the text and the atom will be (un)selected to highlight it. This comes in handy when something does not work as expected and you want to check what the code does while having the molecular system in front of you.

Last but not least: Each Action should give STATUS messages about success or failure, visible especially in the GUI (the bar at the bottom of the screen is the status bar).