J. F. Berar and P. Lelann, J.Appl.Cryst. 24 (1991) p1-5
The full title of this is:
"ESDs and estimated probable-error obtained in rietveld refinements with local correlations" J.F. Berar and P. Lelann, Journal of Applied Crystallography, 1991, Vol.24, No.Pt1, pp.1-5.
Hugo Rietveld - Rietveld Method Web-site
|
Date: Fri, 08 Aug 1997 13:34:13 +1000 To: "Rietveld Users Mailing List" [Rietveld_l@ill.fr] From: Lachlan Cranswick [lachlan@melbpc.org.au] Subject: RIET USERS: Query on Validity of Rietveld Bragg R Hi All, Potentially silly queries. Are there any newish references around the place about the validity of R Bragg in Rietveld analysis. --- Also, when using capilliary data from such machines as an INEL PSD. You can get quite nice Bragg R's purely becuase 20% to 80% of the intensity that is being fitted relates to the background. Thus are there any background subtracted versions for Rietveld of Bragg R referenced in the literature? Thanks in advance. Lachlan. Lachlan Cranswick - Melbourne, Australia _--_|\ Phone/Fax : (613) 9455-1345 / \ E-mail : lachlan@melbpc.org.au \_.--._/ Mobile Phone/Voice Mail : 0412-1141-31 v Crystallographic WWW : http://www.unige.ch/crystal/stxnews/stx/volnteer.htm |
Date: Fri, 8 Aug 1997 09:35:35 +0200 To: RIETVELD_L Distribution List [rietveld_l@diams.ill.fr] From: "Alan Hewat, ILL Grenoble" |
Date: Fri, 8 Aug 1997 16:48:17 +0200 (MET-DST) To: rietveld_l@ill.fr CC: GREGOR@unisi.it Subject: RE: Bragg R On: Fri, 08 Aug 199 Validity of Rietveld Bragg R Lachlan Cranswick [lachlan@melbpc.org.au] wrote: >...Are there any newish references around the place about the >validity of R Bragg in Rietveld ... >You can get quite nice Bragg R's purely becuase 20% to 80% of the intensity >that is being fitted relates to the background. Thus are there any background >subtracted versions for Rietveld of Bragg R referenced in the literature? Lachlan, there was a paper in the early '90ies which treats the problem, I think it was a french author in J Appl Cryst - helas, I can't specify you more as I am on leave from my hardware (paper) file of bibliographic references! Regards Miguel Gregorkiewitz Dip Scienze della Terra, Universita Via delle Cerchia 3, I-53100 SIENA, Europe fon +39'577'298823 fax 298815 email gregor@unisi.it |
From: "Ram SESHADRI" [seshadri@Indigotrem1.chemie.uni-mainz.de] Date: Sat, 9 Aug 1997 11:06:35 -0600 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] Subject: |
Date: Sun, 10 Aug 1997 11:36:14 +0200 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: Armel Le Bail [armel@fluo.univ-lemans.fr] Subject: Standards, Rietveld Bragg R and background The Al2O3 NIST SRM 1976 is a relatively good standard for testing your diffractometer (but LaB6, SRM 660, gives lower FWHMs). Errors on intensities are mainly due to the sample itself (not intercepting the whole X-ray beam, preferred orientation, sample transparency, etc). So that I recommend the SRM 1976, if the data are treated with the Le Bail method, an option which should be included in any good Rietveld program. With this option, you will obtain a fit as if the structure had been perfect (Bragg R ~ 0.0), so that the Rp and Rwp will be as low as possible. One advantage of SRM 1976 is its compacted plate form. Another standard which has already been discussed on this list is NAC (NaCa2Al3F14), frequently used on the INEL 120 PSD for calibration. The main inconvenient is its cost. The producer (which is in my lab) sells it for a price larger than gold itself (but lower than diamond :-)... About the background influence on Rp and Rwp, it seems to me that the original Rietveld program gave exclusively the "background subtracted" and "peak only" profile R factors so that they are now called the "conventional Rietveld Rp and Rwp factors". Then, if I am not wrong, I have one question : who invented the total Rp and Rwp factors ?-) I suspect that they were invented by error or just for an endless discussion about their absence of meaning regarding the crystal structure... Armel Le Bail, back of vacations. |
From: VONDREELE@mist.lansce.lanl.gov Date: Mon, 11 Aug 1997 10:26:35 -0600 (MDT) To: rietveld_l@ill.fr Subject: RE: Standards, Rietveld Bragg R and background To all, In case any one was wondering... >Then, if I am not wrong, I have one question : who invented the >total Rp and Rwp factors ?-) I suspect that they were invented >by error or just for an endless discussion about their absence of >meaning regarding the crystal structure... These things were originally described by Hugo Rietveld. Rpw in particular is statistically "unassailable" as it is directly linked to the function that is minimized in a Rietveld Refinement. Bob Von Dreele |
Date: Tue, 12 Aug 1997 15:42:45 +1000 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: Ian Madsen [Ian.Madsen@minerals.csiro.au] Subject: Re: Standards etc Reply-To: RIETVELD_L Distribution List [rietveld_l@ill.fr] Armel Le Bail has mentioned the use of NAC for calibration of the INEL PSD - we had problems in obtaining this material - not to mention the cost. We have decided to use Y2O3 (thanks to Bob Roth of NIST for the suggestion) since it has a peak roughly every degree in the upper part of the pattern - peaks start at about 17deg, so a low angle standard may be required as well. Y2O3 is readily available (ours is Alrich 99.99% gold label) at a reasonable cost. You may need to heat the material to ensure that crystallite size is as large as possible - we heat at 1200C for a few days - this makes a small, but observable, difference to the peak widths. At 11:36 AM 10/8/97 +0200, you wrote: > >Another standard which has already been discussed on this >list is NAC (NaCa2Al3F14), frequently used on the INEL >120 PSD for calibration. The main inconvenient is its cost. >The producer (which is in my lab) sells it for a price larger >than gold itself (but lower than diamond :-)... > > >Armel Le Bail, back of vacations. > > Cheers **** WE HAVE MOVED - PLEASE NOTE CHANGE OF ADDRESS **** \ | / ___\|/___ (o o) *-------------oOO--(_)--OOo-----------------* Ian Madsen CSIRO Minerals Box 312 Clayton South 3169 Victoria AUSTRALIA Phone +61 3 9545 8785 direct +61 3 9545 8500 switch FAX +61 3 9562 8919 Email Ian.Madsen@minerals.csiro.au XRD laboratory 'tour' starts at http://www.minerals.csiro.au/tour/begin.htm http://www.minerals.csiro.au/public/work/res_equip/tour/begin.htm *-----------------/ \Oooo.-----------------* .oooO ( ) ( ) ) / \ ( (_/ \_) |
Date: Tue, 12 Aug 1997 11:48:59 +0200 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: Armel Le Bail [armel@fluo.univ-lemans.fr] Subject: RE: Standards, Rietveld Bragg R and background At 10:26 11/08/1997 -0600, Bob Vob Dreele wrote: >In case any one was wondering... >>Then, if I am not wrong, I have one question : who invented the >>total Rp and Rwp factors ?-) I suspect that they were invented >>by error or just for an endless discussion about their absence of >>meaning regarding the crystal structure... >These things were originally described by Hugo Rietveld. Rpw in particular is >statistically "unassailable" as it is directly linked to the function that is >minimized in a Rietveld Refinement. Yes Bob, sure that Rp and Rwp were originally described by Hugo Rietveld. But. Remember that the background was not originally described by functions with refinable parameters. So that the function practically minimized (i.e. in the original softwares) was not including the background. And this was logical. Have a look at fig. 4 in J. Appl. Cryst. 2 (1969) 65-71 by H.M. Rietveld and observe the zero background, a fact which is certainly not experimental for CaUO4 on any neutron diffractometer. So that, even if this is not explicitly written in the paper, the yi(obs) in the Rp formula is in fact Iobs-background. Remember the first versions were in two steps, one for preparing data and one for the refinement. The background was subtracted in the first step (or am I wrong ??). This background corresponded to values interpolated between intensities estimated manually at a few angular values. So that the "total Rp and Rwp" (i.e. including background and points where no reflections are contributing) were probably introduced by those having included the background as described by refinable parameters. May be it was for ToF data? The question remains interesting. If you remove the sample, the background is still there... This is the best way to explain why the total Rp and Rwp are not a lot related to the compound structure. The more the background is high and the more the total Rp and Rwp are low if the background is well fitted :-). Hence the old (continuing) controversy. I systematically give the conventional Rp and Rwp in all my manuscripts, indicating that they are "background subtracted" and "peak only". I never mention the total Rp and Rwp because their meaning is null for the structure- related intensities. The problem is that some softwares give only these meaningless total Rp and Rwp so that their users have no way to access to the conventional R values (if you want a list of these softwares, let me know). So who invented first the "total Rp and Rwp" ?? This don't seems to be H.M. Rietveld, IMHO. Or maybe the old software I just had a look inside was not the original ? Armel |
Date: Tue, 12 Aug 1997 14:02:00 +0200 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: "Alan Hewat, ILL Grenoble" [hewat@ill.fr] Subject: RE: Standards, Rietveld Bragg R and background Armel Le Bail |
From: VONDREELE@mist.lansce.lanl.gov Date: Tue, 12 Aug 1997 8:38:55 -0600 (MDT) To: rietveld_l@ill.fr Subject: Re: Standards etc Ian, >You may need to heat the material to ensure that crystallite size is >as large as possible - we heat at 1200C for a few days - this makes a small, >but observable, difference to the peak widths. I suggest that you actually are removing microstrains in the material by heating; not making the particles bigger. Did the stuff sinter at all at 1200C? We use CaF2 as a neutron diffractometer calibrant after heating; there it definitely removes microstrains and leaves a little remaining peak broadening which is due to particle size. Bob Von Dreele |
From: VONDREELE@mist.lansce.lanl.gov Date: Tue, 12 Aug 1997 9:02:39 -0600 (MDT) To: rietveld_l@ill.fr Subject: RE: Standards, Rietveld Bragg R and background Armel (and others), > The problem is that some softwares give only > these meaningless total Rp and Rwp. I beg to differ. The function minimized by the least squares in a Rietveld refinement is M=sum[w(Io-Ic)^2] and Rwp is defined as sqrt{M/sum[w(Io)]}. Nothing is more clear than that. The other residuals are interesting and sometimes useful for characterizing how this residual is distributed between Bragg intensities and background but they are not directly related to the minimization function. The subtracted background R is "flawed" because it assumes that all the random error in the measurements belongs to the Bragg contribution and that the (subtracted) background contribution has none. Thus, it is a over estimate of the contribution to the residual from the Bragg intensities. I think that the real point in all this is that R-value comparisions between different classes of diffraction experiments (x-ray vs neutron TOF, etc.) is particularly meaningless because the differing background contributions, scattering factors, Lorentz factors between the experiments have such an enormous impact on the quality of the refinement results. What really matters is the quality of the structural result (atom positions, temperature factors, bond lengths, angles, etc.) and how they compare with other known data and not whether Rwp is such-and-such a value. Bob Von Dreele |
From: "Ram SESHADRI" [seshadri@Indigotrem1.chemie.uni-mainz.de] Date: Tue, 12 Aug 1997 09:16:56 -0600 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] Subject: |
Date: Tue, 12 Aug 1997 11:34:45 -0400 (EDT) From: "R. A. Young" [r.young@physics.gatech.edu] Subject: R's and background Dear Armel: You wrote: >So who invented first the "total Rp and Rwp" ?? This don't seems to >be H.M. Rietveld, IMHO. Or maybe the old software I just had a >look inside was not the original ? >Armel Well, in the late 70's we put a refinable background (polynomial) into our program which later became DBW* and, later, DBWS*. We did not forsee that this was to bring us opprobrium! -Ray R. A. Young, Professor Emeritus School of Physics Georgia Institute of Technology Atlanta, GA 30332 USA Tel: 404-894-5208 e-mail: r.young@physics.gatech.edu |
From: "Jens Trommer" [trommer@aca.fta-berlin.de] Organization: Institut f. Angew. Chemie Berlin-Adlershof To: rietveld_l@ill.fr Date: Wed, 13 Aug 1997 11:12:51 -60 Subject: RE: Rietveld Bragg R and background Dear Bob Von Dreele, Dear Armel Le Bail and all other Rietveld Users, On 12 Aug 97 at 11:48, Armel Le Bail wrote: > At 10:26 11/08/1997 -0600, Bob Vob Dreele wrote: > > >In case any one was wondering... > >>Then, if I am not wrong, I have one question : who invented the > >>total Rp and Rwp factors ?-) I suspect that they were invented > >>by error or just for an endless discussion about their absence of > >>meaning regarding the crystal structure... > > Yes Bob, sure that Rp and Rwp were originally described by Hugo > Rietveld. > > But. > > Remember that the background was not originally described by > functions with refinable parameters. So that the function practically > minimized (i.e. in the original softwares) was not including the > background. And this was logical. > > Have a look at fig. 4 in J. Appl. Cryst. 2 (1969) 65-71 by H.M. > Rietveld and observe the zero background, a fact which is > certainly not experimental for CaUO4 on any neutron diffractometer. > So that, even if this is not explicitly written in the paper, the yi(obs) > in the Rp formula is in fact Iobs-background. Remember the first > versions were in two steps, one for preparing data and one for > the refinement. The background was subtracted in the first step > (or am I wrong ??). This background corresponded to values > interpolated between intensities estimated manually at a few > angular values. > I just want to confirm Armel Le Bail's remarks about the background corrected intensities in H. M. Rietveld's original paper in J. Appl. Cryst. 2 (1969), 65-71. It is not only obvious from the plotted patterns in Fig. 4 and 5 in addition Hugo Rietveld wrote explicitly in his paper in the Section: Experimental about the background corrected intensities yi in the form: yi = Yi - Bi, where Yi is the total intensity at step i and Bi is the estimated background as described above and in the paper. The equation dealing with the calculation of Rp uses only the corrected yi values. The easiest way to proof is the original paper by Rietveld which you can read without leaving your desk by browsing through the WWW to Hugo Rietveld's own homepage at: http://home.pi.net/~rietvld/home.html or to the location of the paper at: http://chemsg7.tau.ac.il/xtal/paper2/paper2.html > So who invented first the "total Rp and Rwp" ?? This don't seems to > be H.M. Rietveld, IMHO. Or maybe the old software I just had a > look inside was not the original ? > Its an interesting question for me too. Jens Trommer |
Date: Mon, 25 Aug 1997 10:39:11 +0200 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: Armel Le Bail [armel@fluo.univ-lemans.fr] Subject: RE: Standards, Rietveld Bragg R and background Bob wrote: >What really matters is the quality of the structural result (atom positions, >temperature factors, bond lengths, angles, etc.) and how they compare with other >known data and not whether Rwp is such-and-such a value. Hum, Bob, I have just received the CPD Newsletter No18 in which I read "The commission is following up the Rietveld refinement comparison study published by Hill et al in 1994 with a set of recommendations. This is being led by Lynne McCusker and Bob von Dreele and will probably appear in one of the IUCr journals this year". So, if I understand well your point, the R values do not really matter and this could be your recommendation to let users propose any definition of R in their manuscript ? Most Journals give informations to authors. For instance, the J. Solid State Chem. ask for the "definitions and final values of R factors (weighted and unweighted) [e.g. RI (RF), Rwp, Rexp, S]". I would add Rp to the list. Only few of these were provided by the last GSAS version I used. Even RI was not available. I am sorry but my opinion is that a Rietveld software should at least provide the Rs as defined by Rietveld (background subtracted). Now there is no opprobrium to those who want to add other R definitiions (background included). I agree with Alan writing : >Hugo Rietveld didn't include the background in the definition of the >profile R-factors, and that choice IMHO appears to be the best. He did >correctly include the background in calulating the statistical error for >each point (the weighting scheme), and he did correctly calculate the >estimated errors in the parameters. That is much more important than >simply obtaining "low R-factors". Anyway, I would like to obtain from the CPD some clear recommendations about which Rp and Rwp should be given. Hearing that the conventional Rietveld Rp and Rwp values should be given in any case would make me happy. Hearing that if some want to give additionally the DBW-Young Rp and Rwp values (background included), this is wellcome too because the ratio (Rietveld Rp) / (DBW-Young Rp) will give an idea about how high was the background. In many publications, Rp and Rwp are given without cleardefinition. You may be sure that peoples will choose to give the DBW-Young R values because they are systematically lower than the true Rietveld ones. The worst I have seen is in papers giving only the GoF Chi2 values, which were all near of 1 and no figure were given showing the observed and calculated patterns. In fact, Chi2 = 1 could mean that Rwp=35%=Rexp. In such a case, making another pattern with longer exposure would be the thing to do before submitting the publication... Obtaining "low R-factors" may not be very important however, obtaining "low R-factors" together with "low FWHMs" is significant. Has somebody an idea for a new R definition including both aspects ? Imagine a conventional Rietveld Rp = 5% together with FWHM as low as 0.01 2-theta degrees. How to distinguish this from a result from another experiment leading to the same Rp but with FWHM=0.15 ?? Last point on the Rietveld Refinement Round Robin. The questionnaire for structure refinement of PbSO4 asked for the Rp, Rwp, Rexp, GoF and RB (=RI) definitions (how many participants had indicated if the background was subtracted or not??) but the form for reporting the M-ZrO2 refinement results only asked for the R values. Few people have thought to indicate if the Rp and Rwp values they gave were either the true Rietveld ones or the DBW-Young ones. Fortunately the reanalysis provided both Rietveld and DBW-Young Rp and Rwp values. The recommendation given in the RRRR conclusion was : "it is desirable that both types of agreement indices be calculated and subsequently quoted in published work in order to permit valid comparisons to be made between instruments with different PtB ratios and using different wavelengths". My hope is that the CPD will not recommend now the contrary. Following the above recommendation, the first you have to do, Bob, I am afraid, is to include the true Rp and Rwp in GSAS together with RB :-). Armel P.S 1. In 1992, I asked for the compilation on the RRRR PbSO4 results deposited with the British Library Document Supply Center (supplementary publication No SUP 55098). The document had 108pp, most of them completely white. Probably an electronic error ?? P.S. 2. May I suggest that Rp and Rwp values given in published Rietveld refinements should be accompanied by the labels : "DBW-Young" if including the background and "Rietveld" (or nothing at all) if excluding the background with no supplementary indication should be the conventional ones (background subtracted) and that the "background included" values |
Date: Thu, 11 Sep 1997 23:45:50 +1000 To: "Rietveld Users Mailing List" [Rietveld_l@ill.fr] From: Lachlan Cranswick [lachlan@melbpc.org.au] Subject: RIET: Refining on R-Bragg rather than R-Wp??? Hi All, I just can't seem to keep away from these R-Bragg questions! :-) (An interesting finding based on the previous discussion is the R-Bragg calculation in the software I am using is affected by background (thus something dodgy must be going on) while Fullprof seems to be doing the right thing and not including background in the calculation of the R-Bragg. Not sure about other software as yet?) --- Does anyone know the pros and cons of refining of Weighted R-Bragg instead of the Weighted Profile (R-Wp)? Are there any packages out there that can be toggled to refine on a weighted R-Bragg rather than R-Wp? In the paper by H Boysen (Anharmonic Thermal Parameters, Disorder and Phase Transitions - NIST Special Publication 846. Proceedings of the international conference Accuracy in Powder Diffraction II, held at NIST, Gaithersburg, MD, May 26-29, 1992 (Issued October, 1992) pg 165-174)) it mentions how when refining these additional parameters how the conventional GOF is only slightly changed but that "a much larger effect may be obtained, if one uses a weighted Bragg R-factor." Using the LHPM/REIT7/SR5 software, occassionly when refining subtle parameters such as site occupancies or thermal parameters, the R-Wp can go down while the R-Bragg goes up. Are then any comments on this as it implies at this level - it would be better to be refining and minimising on the weighted R-Bragg, not the weighted profile(?). Thanks in advance, Lachlan. Lachlan Cranswick - Melbourne, Australia _--_|\ Phone/Fax : (613) 9455-1345 / \ E-mail : lachlan@melbpc.org.au \_.--._/ Mobile Phone/Voice Mail : 0412-1141-31 v Crystallographic WWW : http://www.unige.ch/crystal/stxnews/stx/volnteer.htm |
Date: Thu, 11 Sep 1997 16:54:01 +0200 To: RIETVELD_L Distribution List [rietveld_l@diams.ill.fr] From: "Alan Hewat, ILL Grenoble" [hewat@ill.fr] Subject: Re: RIET: Refining on R-Bragg rather than R-Wp??? Lachlan Cranswick |
Date: Thu, 11 Sep 1997 17:07:47 +0200 To: RIETVELD_L Distribution List [rietveld_l@ill.fr] From: Armel Le Bail [armel@fluo.univ-lemans.fr] Subject: Re: RIET: Refining on R-Bragg rather than R-Wp??? Lachlan wrote : >Using the LHPM/REIT7/SR5 software, occassionly when refining subtle >parameters such as site occupancies or thermal parameters, the R-Wp can >go down while the R-Bragg goes up. Are then any comments on this as it >implies at this level - it would be better to be refining and minimising >on the weighted R-Bragg, not the weighted profile(?). If minimizing Rwp, then Icalc and true Iobs are involved. Trying to minimize R-Bragg, squared Fcalc but not true squared Fobs will be available (only biased "|Fobs|" depending on Fcalc when there is overlapping). Consequently I can't see how a refinement based on minimising R-Bragg could be credible. You should minimize an expression depending on the really observed data.Armel |
Date: Thu, 11 Sep 1997 11:31:20 -0400 From: "Peter Y. Zavalij" [zavalij@binghamton.edu] Organization: Materials Center, SUNY at Binghamton To: RIETVELD_L Distribution List [rietveld_l@ill.fr] Subject: Re: RIET: Refining on R-Bragg rather than R-Wp??? > Lachlan Cranswick [lachlan@melbpc.org.au] wrote 11 Sept 1997: > > >Does anyone know the pros and cons of refining of Weighted R-Bragg > >instead of the Weighted Profile (R-Wp)? > Alan Hewat, ILL Grenoble wrote: > Yeah, that's what people used to do before Hugo Rietveld invented > profile refinement. But if your resolution is so good that you can > actually MEASURE all the R-Braggs, why not :-) Otherwise the > extracted integrated intensities (R-Bragg) are biased by the calculated > structure factors, since the latter are used to obtain the former. > > Alan Hewat, ILL Grenoble, FRANCE |
From: VONDREELE@mist.lansce.lanl.gov Date: Thu, 11 Sep 1997 9:40:56 -0600 (MDT) To: rietveld_l@ill.fr Subject: Re: RIET: Refining on R-Bragg rather than R-Wp??? Lachlan wrote : >Using the LHPM/REIT7/SR5 software, occassionly when refining subtle >parameters such as site occupancies or thermal parameters, the R-Wp can >go down while the R-Bragg goes up. Are then any comments on this as it >implies at this level - it would be better to be refining and minimising >on the weighted R-Bragg, not the weighted profile(?). As others have said already, this is a somewhat confused question. As a reminder, least squares refinement is about matching a set of observations to a comparable set of computed values created from some mathematical model. Expressed in this way it is obvious that the Bragg reflections are not observations in a powder diffraction pattern. The observations are the suite of individual profile intensities obtained as a function of an independent "scanning" variable (2-theta, TOF or whatever). The Bragg peaks, on the other hand, are our method of modeling the peaky features of a powder diffraction pattern. They have no independent existence as observations. Thus, minimization on Fo-Fc makes no sense because there are no real Fo's. Those things that are obtained via some extraction procedure during a Rietveld refinement are developed by assuming that the Fc's are in the correct ratio; thus the "Fo's" are biased by the Fc ratios. And things like R(Bragg) are only useful as probably biased diagnostic tools and, most assuredly, not real refinement residuals. Bob Von Dreele |