Session 3

• 1 - How reasonable it is to use Crysfire zero-shift ? I think it is more reasonable to crosscheck with CELREF/ERACEL. 
• 2 - How to propose a cell when you get a lot of solutions ?

• 1 - I have a problem with DMPLOT. The version that I got from the zip-file armel.zip, does not write the background in the file background.dat. The other version I have is that obtained from the file dmplt348.exe. In this version the program does not understand the correct range of the DRX pattern. For example, when I ask to read the file na5.rit, it takes the 2theta range between 10 and ~105 degrees (when it should be 10-150 dg.) I have tried every thing I could but without success. Could you help me?
• 2 - I wish to know how to decide the initial values, especially, scale, U, V, W., in structure factor extracting. I tried with the naoxa0 (Na/COO) files(copied from the tutorial) with fp90: I can not repeat the 2nd step. I got an error message "square of FWHM<0 at 2 theta/TOF: 125.0", I wish you could teach me what does this mean, and how to correct this kind of problem.

• 1 -Attached please find my answers to session 5. However, I'm wondering why the Sigma of |Fobs| are so big. Is there anything wrong ?
• 2 -The PDF card I have for CrF₃ is the one with number 16-0044 (since my PDF database is from 1993). May be you have a more recent one. If yes, could please send me a copy of it ?
• 3 -I think I have finally worked out how to extract the data about CrF₃ from the database in the correct spacegroup and I have generated a simulation in Fullprof. However, I think I must still have done something wrong as the amount of impurity in either pattern sent seems very small and so estimating the zero point is difficult/impossible and when I've tried to let Crysfire do the estimation and then find the cell, no sensible solutions seem to result.

• 1 - I've not been able to view the structures with STRUVIR. There was no warning message from STRUVIR with the .dat file (Na5test.dat). But no structure could be viewed. It seems that i'm having the same problem with the .dat file copied from the SDPD tutorial : http://pcb4122.univ-lemans.fr/iniref/tutorial/naoxa11.html

Please help me to find out what's the problem for both .dat files.
• 2 - I wish to know in extracting structure factors by Fullprof: What's ground for deciding following parameters :? Wdt, R_at, R_an, R_pr, R_gl.

1. 1 - I am not sure what kind of result you are expecting from this session's work. The solutions I am getting all seem a bit unstable and so I was wondering if you could let me know what is a usual result.

• 1 - I was wondering if you could give me some more help for this week's work. I have managed to work out the space group and cell parameters however, I couldn't, unsurprisingly, find a stable solution from SIR97 or SHELX97. I then tried Espoir using Prespoir to set up the .dat file. Again no sensible solution has been found despite numerous attempts changing different parameters. I was wondering if you could point out where I have gone wrong. I think that it is maybe in the section below:

! maximum moves for each type of atom
  1.000   1.000
! annealing law, sigma, reject
  2.0000  1.0000  0.0010
! number of events for : print, maximum, save
  20000    200000    100000
! events for restart, rmax, ichi, number of runs
  40000  0.200   2  10
The manual doesn't seem very clear on how you should determine values for these parameters so any help that you could give would be greatly appreciated. 
• 2 - I have finally managed to get a reasonable result with ESPOIR. Although the fragment was relatively easily located, the O and methyl group were less easy to sort out. I'm not sure why, but with exactly the same .dat file, I got two different results, the non windows gui giving a non-reasonable solution.
• 3 - I was wondering if you could give me any more guidance on how/why you pick the initial numbers if events, rmax, ichi and number of runs as this seems to make a difference and yet seems to be random.
• 4 -... then I cannot obtain RpF values lower than 46%. The final values are exactly the same as the input values. Can you kindly suggest me on how to improve this by ESPOIR ? Is there something wrong in my input file ?

Session 10

Final examination

General

• 1 - Redistributing the Course material would certainly be illegal. Even more illegal (if possible ;-) if you redistribute the material for some fees. I don't want to know what you will do... 
• 2 - Thank you for your interest in the SDPD Internet Course. You are not the first to ask it for free. A large part of the documentation is in free access, so that you may follow it by yourself at :

http://sdpd.univ-lemans.fr/iniref/tutorial/indexa.html
http://sdpd.univ-lemans.fr/iniref.html
http://sdpd.univ-lemans.fr/kunming/index.html
You will find there exercises with correction, lists of software, references. But you will not have access to the
exercises without correction and to the interactive help which are part of the commercial course, sorry.
• 3 - The course is not limited in time. You may go back at your convenience. Best regards, and don't kill yourself at work !
• 4 - The best is to begin immediately because administration poor efficiency will cause certainly time lost.
• 5 - I guess the name is TVA identification number here :  FR66197209166, Université du Maine.
• 6 - You will find possibilities to download Winzip at : http://www.winzip.com/ddchomea.htm
• 7 - I hope this will be possible. However, Nobody succeeded in finishing the course in 12 weeks, up to now. The knowledge to acquire is quite large so that the fastest students needed 5-6 months. Nevertheless, they were not doing only the course during that time.

• 1- OK, let us wait for your session-1 solutions when it will be possible, and start already the session-2. Only session-1 needs a commercial product (PDF-2 and a search-match software), so that you should not have other problem of this kind (but PDF-2 is certainly needed for becoming a SDPD expert). If you receive PDF-2 before the end of the course, you will try to identify the session-1 materials. No lab at your university having PDF-2 ?
• 2 - I had the same experience. It should even be possible to obtain some match without background subtraction. I have never seen any study comparing the performances with and without K_alpha2 stripping. Probably, the reason that both work well is the absence of clear separation between the two components at low angle, where matching is the most important. But it is so easy to perform the corrections that I prefer to do it every time.
• 3 - This means the you do not use the latest version of ICDD which has included powder patterns calculated from the ICSD database. In that new version (since 1998 in fact), there are >120000 patterns. Have a look at your ICDD CD-ROM. May be the data prepared for EVA were not updated ? Those data are .cat files if I remember well. They are prepared for EVA by PDFMAINT from the ICDD CD-ROM, I think (not sure).

Session 2

• 1 - It is to you to decide if you have located one or several series of harmonics from your data. For instance, if you have d = 12.6, and also d = 6.3 and d = 4.2, then you may suspect that they built a series of harmonics. Why not the 100, the 200 and the 300 reflections ? So that you have now one of the three fictitious cell parameters with a = 12.6A. Then search for b and c with 2 other series of harmonics. You will be then able to refine a zeropoint. 
• 2 - The old (DOS) version of CELREF is still distributed with the name ERACEL : http://sdpd.univ-lemans.fr/ftp/eracel.zip

and the new version with graphical user interface is at : http://www.ccp14.ac.uk/tutorial/lmgp/celref.htm
• 3 -The reduced cell is the unit cell with the smallest volume whose axes are the three shortest non-coplanar translations in the lattice (official definition).
• 4 - I always subtract the background and correct for K-alpha2 before searching for peak positions. However, the result could be very similar if you remove K-alpha2 and then search for peak positions without subtracting the background. I prefer the first option, but the second may have advantage that less errors are introduced if the background is not removed. A kind of no-yes no-no answer... 
• 5 - The fact is that untill you do not obtain an evident proposition, you may have to test with various data sets (including or not some weak reflections). The first data set to be used should not contains too weak reflections in my opinion. You should also be careful about possible artifacts near of very intense reflections which could be generated by the K-alpha2 elimination process (there is frequently a 1% false reflection after the 100% one). Indexing is more an art than a science you know. About the biesrf.rit case : the biggest experts have broken their teeth on it, it remains unsolved.
• 6 - Well, you should have it because it is the 36-1496. However, the card is a bit wrong. The 84-1651 is the good one, calculated from the ICSD database (you should update your PDF-2 database for obtaining it). But you do not need that card. All you need is the list of hkl and d(A) values. The week-2 course material gives you all you need for being able to calculate them : you have links to programs which can calculate reflexion positions, and the NAC cubic cell parameter is given together with the space group in the exercise : a=10.2533A, space group I2₁3
• 7 - It is true that data-2 is sufficient for solving the problem.

Nevertheless, it may occur that peaks from the unknown are masked by peaks from the standard. In such a case, it is
possible to extrapolate the zeropoint from the unknown + standard pattern to the pure unknown pattern, and to index from peak positions from the pure pattern. You will see that in the official correction.

• 1- You mean that crysfire has an internal possibility for zero shift estimation ? Seemed to me that the user had to give a value (from a previous estimation, involving a standard or the harmonics method). Some programs inside Crysfire may propose a post-indexing zeropoint, produced after cell parameters refinement (ITO does that). But it is certainly preferable to have a good idea of the zeropoint BEFORE to use Crysfire.

 

 

Comments from Robin Shirley about that answer :
"You mean that crysfire has an internal possibility for zero shift estimation ?"

Yes, Crysfire (or at least its front-end program CRYS) does indeed have facilities for Z2Th (2Theta zero correction) estimation, primarily by self-calibration from lines and their second orders.

In CRYS v9.34h (Crysfire 2000 - i.e. the July 2000 release), this can  be used either directly for Z2th (the default option: Z) or now for specimen-displacement ratio (new option: T).

1) Option Z is appropriate for parallel beam geometry, and in 
principle for any constant residual that remains after a specimen-displacement correction has been applied correctly.

2) Option T is more appropriate for the usual flat-sample,
convergent-beam geometry (Bragg-Brentano, etc).

3) Linear recalibration is also possible as an option at the end of an obs-calc pattern comparison within CA.  However, this can only be used when a trial cell is available, and is now superseded in many cases by self-calibration for specimen displacement. 

Of course the determination of these correction parameters in the course of a Rietveld-method refinement or other full-profile process is more reliable (though this may require more information than initially is available).

Best wishes

Robin Shirley
 

• 2 - Well, the difference is made on the Figure of Merit (FoM)

which should be the highest possible, and at least > 20. Your best FoM are <13. I don't know what could be the error you made. I hope you will see where was the error by
looking later at the official corrections (sign of zeropoint ?). Another way to test a cell proposition is to extract the
structure factors by a cell constraint method. The profile 
reliability Rp factor should be as low as possible (<10%).
This is the subject of session 4.

• 1 - In fact, the background is written in the file backgrnd.xy by default, and, if you try to give another name, it should not have more than 8 characters before the . (this is old DOS program...). If all that fail, then you may use the classic DMPLOT by cutting the pattern in 2 pieces of less than 9600 points. 
• 2 - In structure factor extracting mode, there is no scale to refine. You have to fix it to an arbitrary value, say 0.01. If the extracted structure factor amplitudes in the .fou file are too small (using Fullprof), then you may consider to decrease that fixed scale factor at 0.001 or 0.0001 or less. Try to obtain the "|Fobs|" max near of 500-1000. 

About the U,V,W, they enter in the FWHM formula by U tan**2(theta) + V tan(theta) +W. So that, either you already know your diffractometer performances and have a set of starting U, V, W or not. If not, the best is to start by U=V=0 and select W as 0.01 for instance. See what happen in a few refinement cycles. Refine V and W. And finally U, V, and W, but this may be difficult if your data do nor go far beyond 90° 2theta. Anyway, the old FP90 Fullprof version is more stable than the recent WinFullprof when extracting structure factors. You should perform a few cycles at the beginning without refining nothing but the zeropoint maybe. This supposes that your cell parameters are already very good, and also that your profile parameters are not too bad. Otherwise you may encounter that problem with FWHM <0. because some parameters diverge quickly.

• 1 - The ESD for the "|Fobs|" as extracted by WinFullprof, have not a lot of meaning. The previous fp90 version of Fullprof gave even higher ESDs. Remember that they are absolutely not used by any Direct or Patterson method program, but for elimination of the too bad data. A program like OVERLAP takes these values and change them with ESD="|Fobs|"/100.
• 2 - There is not more recent one. And the data in the Wyckof book or in the ICSD database are also very old with quite bad cell parameters. This is one difficulty of the exercise... The best is that I give you the cell parameters of a CrF3 sample made at the Fluoride Lab : a = 4.977 A, c =  13.145 A , space group R-3c, hexagonal setting. isostructural with FeF3, VF3, etc. You may use HKLGEN in order to generate the list of hkl and their angular positions (or another program).
• 3 - About CrF₃, the impurity in the session 5 exercise, you

may have noticed that there were available only old published data including rather wrong cell parameters. This is probably the origin of your problems. However, I recommend you to first print a drawing of the powder patterns. If you had examined more carefully the two patterns, one with less impurity than the other, you would certainly not have concluded that both mixtures were containing very few CrF₃ impurity... Powder diffraction expertise has a visual part : looking first and carefully at the patterns allows to avoid many traps.

• 1 - I think that your problem is just that the EOF character (End of Line) is placed at the end of the O2 line. It should be at the beginning of the last line. So that you have to make a "return" at the end of the O2 line. This is the most common error with many programs reading data file. In that case, the O2 line is ignored, and the program cannot find any polyhedron.
• 2 - WDT = width (range) of calc. profile in units of Hk (typically 3.5 for Gaussian and 10 for Lorentzian, 3-3.5 for T.O.F.).

For Rietveld application or for extracting structure factors, the
same kind of values will apply. For X-ray case, profile shapes
are generally lorentzian-like, and for neutron data, they are
Gaussian-like. You may find useful to increase the suggested
Wdt values above, if you see staircase steps on each side of
the calculated profiles. Values of Wdt up to 15 or 20 may become necessary  sometimes.
The R_at, R_an, R_pr, R_gl are the relaxation factors. They are used for damping the shift on the refined parameters. The shift is multiplied by the R_ value, so that this may reduce unstability effects on some parameters. R_at is for atomic parameters : coordinates, magnetic moments, site occupancies & isotropic displacement (temperature) factors. R_an is for anisotropic displacement (temperature) factors. R_pr is for profile parameters, asymmetry, overall displacement (temperature), cell constants, preferred orientation parameters, strains, size, propagation vectors & user-supplied parameters. R_gl is for global parameters, zero-shift T0, background, displacement and transparency.
For a very good pattern with many reflections, going to large angle, and few refined parameters, there is no reason to damp any parameter, so that all R_ values could be set to 1. But in many cases, if you identify one or several parameters fluctuating, if not diverging, then those R_ parameters will help you to smooth the fluctuations. In case of extracting structure factors, it is recommended to apply a large number of refinement cycles (10-40) when the K-alpha 1-2 is present.
So that, using R_ values as low as 0.2 may be decided, in a preventive manner (just in case some parameter is unstable).

• 1 - Structure solution does not provide always a complete model. This is why the week-7 exercises text concludes by : "Give your final best list of atomic coordinates together with the reliability factors."  Just send me the complete result and you will see probably in the official correction that your result is already a "usual result".

• 1- ! maximum moves for each type of atom

 1.000   1.000
This is certainly not enough : your molecule will not be allowed to move more than +-0.5 angstrom from its initial position. You should select values at least equal to half the largest cell parameter. This is probably the main reason for unsuccess. The molecule had not any chance to attain the right place.
! annealing law, sigma, reject
  2.0000  1.0000  0.0010
The reject term (0.001) may be too low for any event not improving the fit to be accepted. So that you may fall in a false minimum.
! number of events for : print, maximum, save
   20000    200000    100000
! events for restart, rmax, ichi, number of runs
  40000  0.200   2  10
This may be OK. Increase rmax to 0.3 or 0.35 would be better for not stopping all test too quickly (at 40000 Monte Carlo events). 
• 2 - Running 2 times or 3 times the program with the same data will give 2 or 3 sets of different results. That is Monte Carlo. If the success ratio is 1/50, you may  have to perform several runs of 50 tests for obtaining the best result. In that case, the pyrene molecule is located in every test with R ~20% (by the way, you did not gave me your best R factor). But the 2 last  atoms are more difficult to locate. Once located, they will provide R ~10%.
• 3 - May be it is not clear that nstart and rmax are parameters used to avoid wasting time in calculations that will not provide a solution. But in fact, there is no proof that this supposition of "if R is still larger than Rmax after nstart Monte Carlo events, no good solution can be expected"  is really is good supposition... You may find more satisfyng to discard none of the tests before their normal end by using nstart=n2. That should be your choice. I understand that ESPOIR may be surprising as a program for finding a solution. it needs long time, many runs and many tests in a run. It looks very inefficient but sometimes can provide results when classic methods fails. I hope you are convinced ;-). SDPD is much more difficult than structure determination from single crystal data. It needs much more efforts. ESPOIR is typical of this need for trying again and again before to give up.
• 4 - Yes, there is something wrong. But your error is understandable. When using the Molecular replacement method, the coordinates defining the molecule have to build the complete molecule : the atoms cannot belong to different molecules inside the cell. Rotations and translations are made around the gravity center of the group of atoms entered,

and will destroy the  original atom packing if the coordinates are those of atoms in different molecules. In fact, this is the fault of ESPOIR itself which produces some +1 or -1 additions to the coordinates for the final output. So, if you want to reuse those coordinates, you have first to verify if the original molecule connectivity was not broken. This is made fast by using ORTEP on the .res file, for instance. But, anyway, you will not obtain much better result than your 25%. Completing that result will be the scope of half the week-10. So, be patient. You may guess that this will need to apply some Rietveld refinement. Going further with the originally extracted "|Fobs|" is not possible, you attained the limit.