version 1.5CVS

I General questions

Send me an e-mail at vincefn@users.sourceforge.net, I generally answer quickly ! And I like to have queries, that's how the software &amp algorithms get improved !

To be kept informed about new releases, you are encouraged to subscribe to the low-volume FOX mailing list at http://lists.sourceforge.net/mailman/listinfo/objcryst-foxx (the archives can be found at http://sourceforge.net/mailarchive/forum.php?forum_id=654).

If you intend to use the ObjCryst++ / Fox source code for your own projects, it is highly recommended to subscribe to the ObjCryst++ mailing list at http://lists.sourceforge.net/mailman/listinfo/objcryst-devel, (the archives can be found at http://sourceforge.net/mailarchive/forum.php?forum_id=1689).

Almost everything (Crystal, PowderPattern, Scatterer, ScatteringPower, OptimizationObj) is considered an object. In Fox/ObjCryst++ a generic approach is used, so that the algorithms can optimize them without actually knowing what they actually are: the idea is that the algorithm knows that these objects are "Refinable Objects", so that he can get from them (i) a list of parameters and (ii) a choice of Cost Functions. The advantage of such an approach is that if the modelization (parametrization) of some objects changes, the algorithms do not need any modification since they use a generic approach to the objects.

II Global Optimization

I know 'Simulated Annealing' (SA), why do you recommend to use "Parallel Tempering" (PT) ?

In SA, the 'temperature' of the algorithm (think: high 'temperatures' allows for more improbable configurations, allowing to get of local minima) is decreased slowly following a predetermined law. If the decrease is too fast, you can easily be trapped in a local minimum.With PT, all temperatures are optimized in parallel, so that you can get of a local minimum at all times. The algorithm is invariant with time and therefore does not require from the user to choose the number of trial and the temperature schedule.General Idea: keep the default algorithm and optimization schedule & parameters (temperature and displacement), they are made to work with any structure or data.

Sorry, there is no general rule for just numbers... Nothing (for a rough idea at least) replaces the eye comparison of calculated and observed patterns (use the zoom by left-click & drag !). A Rwp value of 0.15 can be sufficient (for the structure solution to be good) if the background is low, but it can also be crap if the background is high (with a background at 50 000 and the maximum intensity at 60 000, even Rwp=.05 can be very bad) !

"It's not a bug, it' a feature": Fox automatically fixes all profile and 2theta correction parameters, for several reasons:
• It is a bad idea to use a Global Optimization algorithm derived from Monte-Carlo to search for this kind of parameters.
• Profile parameters, 2theta shift (zero) can be determined before using Fox for the structure solution, using profile fitting (and therefore you should do it that way).
• By using the "integrated Rwp" (iRwp) cost function (rather than R or Rwp), the algorithm is almost insensitive to profile parameters. Personnally, for all structures I work on and for all examples given with Fox, I never use precise/refined profile parameters, but always adjust them "by eye" within Fox, which is sufficient thanks to the "i"Rwp approach. And generally I only use the 'W' term (constant width). You only need to refine these parameters to get a good full-profile agreement and please referees, but for the structure solution (finding approximate positions of atoms) part, only the integrated intensities matter.

You can optimize positionnal parameters (translations, rotations and conformation), Biso's, occupancies, and texture parameters.

However, you should begin to optimize positionnal parameters only (translations, rotations and conformation), after setting "expected" values for Biso, occupancies of 1, and without texture parameters.

You may activate the optimization of occupancies after a first optimization if you believe it is necessary. Optimizing Biso is probably a bad idea, since at low angle (the only part used for global optimization) data is not sensitive to Biso.

And do not activate texture parameters unless (i) you have a strong feeling it is necessary and (ii) you have not been able to obtain an un-textured data set. Having strong texture will significantly reduce the probability of finding the structure solution.

If you are optimizing a Crystal structure with a single Powder Pattern, you should add these two objects, and then add the "integrated Rwp" cost function. If you have several data sets, you must add all objects, and one cost function for each data object. How can I remove a Cost Function or an object from an Optimization ? You cannot do it from the interface. However, you can set the "weight" of the cost function to 0. It will then only be updated when a new "best configuration" (using the other cost functions) is found (i.e. it does not slow the computation).

The only way to remove an object or cost function is to edit an xml file, and look at the very end: just remove the line corresponding to the object/cost function you want to remove, and voila ! NOTE: do not remove an object if you are still using on of its cost functions !!

III Crystal Modelization

How can I take into account Special Positions ? What is the Dynamical Occupancy Correction ?

To take into account special positions, activate the "use dynamical occupancy correction" option in your Crystal structure interface. Do not change the occupancy of any atom. When atoms (of identical type) overlap, their occupancy will be automatically reduced (in the background, you won't see it) so that n atoms overlapping will all have an occupancy of 1/n. This allows to correct any type of overlapping, for an atom on a special position, or atoms being merged.

It is strongly recommended to activate the Dynamical Occupancy Correction for all inorganic structures. Generally it will be useless for molecular structures, as special positions or shared atoms between molecules are extremenly rare.

When (say) two atoms fully overlap (assuming you have activated the Dynamical Occupancy Correction), both their occupancies are set to 0.5, so that the sum of the two atoms look just like one. However both the atoms are still present in the global optimization, and they will be able to move away from each other later if it gives a better solution.

You have to put at least as many independent atoms as there are in the final result ; that is, you have to guess approximately how many atoms will be required. And thanks to the Dynamical Occupancy Correction, you can put too many atoms, and excess atoms will automatically be merged. Of course, don't exaggerate the number of excess atoms or the algorithm may converge very slowly.

To use the antibump you need:
• To select (Crystal, menu "Parameters -> Set Antibump Distance") the couple of ScatteringPower (atom types) and declare the interatomic distance below which a penalty is used. Do that for all couples of atom types you want to affect with an antibump.
• You must add the AntiBump cost function in you algorithm (Global Optimization object). Otherwise, the antibump does not work.
You cannot see the antibump parameters, unless you edit an xml save file (where each line <AntiBumpDistance ScattPow1="O" ScattPow2="Pb">1.2</AntiBumpDistance> corresponds to an antibump distance). You can change an antibump distance by re-setting the antibump distance using the Crystal "Parameters -> Set Antibump Distance" menu. To delete it simply remove the corresponding line.

For identical types of atoms, the antibump penalty vanishes to zero when the atoms get near each over (practically, the penalty goes from reaches its maximu when the distance decreases from the given dmin to dmin/2, and then decreases to 0 when the atoms fully merge). So the antibump cost function does not prevent merging.

Note that you cannot use different antibump parameters for (say) two different oxygen types: as the antibump evaluation currently uses the atomic number, all oxygens will necessary have the same antibump parameters.

No, these are "periodic" parameters, and are automatically corrected, e.g. set to 0.01 if they get to 1.01 (using limits is then a bad idea, as it restricts movements).

Yep, so far the notion of 'facets' is not stored in the xml file, so you loose the display.

Defining correctly cycles is currently limited in Fox, as it is not possible to define a completely closed cycle. It is still possible to deifine cycles, but these must be more or less rigid. You have first to define your cycle (entering the z-matrix or importing it), and then you need to make sure that all the dihedral angles that could 'break the cycle are fixed or with tight limits. You can see the cimetidine example to see how this can be done. As long as the cycle is approximately rigid, it works very well.

Modification to better handle this (flexible cycles) is high on my todo list.

IV Diffraction data

I copied the powder pattern 2theta zero (shift) from my profile fitting program, but it looks wrong ?

Try to change the sign and/or to multiply it by a factor 2, and see if it is better (after each change, right-click in the powder graph window to update the display).

Simply change the 2-column (2theta - intensity) file you used to import the points, and re-import it. It does not matter if the number of points used has changed.

Yes, in the PowderPattern Object, use the Pattern->"Add 2theta exclude region", and enter the 2theta limits. You can add several ones. NOTE: the powder patterns in these regions are still computed, so if you want to avoid computing the higher parts of the powder pattern, rather use the "Max sin(theta)/lambda" entry field.

To remove it, you must save in an xml file, and then remove the line declaring the region, which should be something like : <Exclude2Theta>0 5</Exclude2Theta>

Yes, generally speaking, you can add any number of data sets. You just have to add them all to the GlobalOptimization object, and declare a cost function (the iRwp preferably) for each data set.

First of all, a clear distinction should be made between (i) structure solution and (ii) structure refinement, which are both part of a complete structure determination:
• "Structure solution" consists in starting from relatively little information about the structure (one data set, and chemical composition), and one way or another to determine the approximate (0.1-0.5 Angstroem) position of all atoms.
• "Structure refinement" consist in starting from known approximate atoms positions, and to determine these positions, as well as temperature factors, occupancies, with the best precision possible.
Fox only targets (i), and therefore only requires enough data to find these "approximate" positions, which obviously is low-angle data (for a low-resolution structure determination). You should then use sin(theta)/lambda limits of between 0.25 (2 Angstroem) and 0.4 (1.25 Angstroem). I suggest you try the examples to convince yourself. After the optimization, you can change the sin(theta)/lambda limit to check that it is indeed the correct solution.

Using too high-angle data will slow the structure solution by two factors, (a) because the number of reflections is proportionnal to the cubic power of sin(theta)/lambda (and therefore much slower), and (b) because high-angle data will only contribute to confuse the algorithm with useless information (to make a simple analogy, our goal consists to drive by car from Paris, to find a street in Geneva ; although all we want is to find the correct street, having the "high resolution data" is equivalent to know the color of the phone booth in that street, which is perfectly useless).

Of course, the high angle data will be useful... but for the refinement stage (ii), i.e. in another program.

The linear polarization rate is around 1.0 for synchrotron data (typically 0.95-0.98), and 0 for laboratory diffractometer (unless you use a monochromator). A precise value is not critical for global optimization, as we mostly use low-angle data, and as errors can be swallowed by the overall B-factor.

The calculation of the polarization factor assumes that the polarization plane is horizontal, and the diffracting plane is vertical (i.e. standard equatorial geometry to maximize intensity on a synchrotron).

This is ignored for neutron diffraction.

No.

With preferred orientation, it is easy to find several close configuration, by inverting the March coefficient and using a perpendicular vector of preferred orientation. This cannot happen if you use the limits on the March coefficient, either to be above (needles) or below (plate) 1.0.