Re: [EXTERNAL] Re: Misconduct/terminology
OK, thanks Cora. Yes, an internal standard using a simple well characterised material is a good idea. I perhaps over reacted when you listed all the parameters you refined for a multi-phase sample, including ADPs. Dr Alan Hewat, NeutronOptics Grenoble, FRANCE (from phone) alan.he...@neutronoptics.com +33.476984168 VAT:FR79499450856 http://NeutronOptics.com/hewat ___ On Tue, 16 Jan 2024, 18:05 Lind-Kovacs, Cora, wrote: > I should have been more specific – this was not with respect to QPA > specifically – it was a more general statement about the terminology of > when we should or shouldn’t use the expression “Rietveld refinement”. > Although we actually often do Rietveld refinements on 2 phase mixtures, > where one phase is an internal standard, which we use to determine sample > height offset changes with temperature. > > And no, we don’t just try to get lower R-values by randomly adding > parameters! > > Cora > > > > *From:* Alan W Hewat > *Sent:* Tuesday, January 16, 2024 11:54 AM > *To:* Lind-Kovacs, Cora > *Cc:* rietveld_l@ill.fr > *Subject:* Re: [EXTERNAL] Re: Misconduct > > > > > > "In many cases, we will indeed refine those structural parameters to get > the best possible match". > > > > The best possible match is not the objective of QPA, but rather the best > possible estimation of the phase fractions. You can refine parameters > describing particle size, strain, lattice constants, etc for each phase, > but please believe the structural coordinates obtained from well > characterised (usually mono-phase) samples. > > > > Whether it is structure refinement or phase refinement, please resist the > temptation to add parameters simply to obtain a lower R-factor. > > > > > Dr Alan Hewat, NeutronOptics > Grenoble, FRANCE (from phone) > alan.he...@neutronoptics.com > +33.476984168 VAT:FR79499450856 > http://NeutronOptics.com/hewat <http://neutronoptics.com/hewat> > ___ > > > > > On Tue, 16 Jan 2024, 17:34 Lind-Kovacs, Cora, > wrote: > > I like Stefan’s view outlined below. In many cases, we will indeed refine > those structural parameters to get the best possible match. But when I use > a structural model so that I calculate F(hkl) values, and combine that with > the ability to vary all kinds of parameters - some structure-related, some > instrument/setup related, some microstructure related – I would (and have!) > call that a Rietveld refinement. > > Because if I did not, there would be a lot of fun questions to > contemplate: What if I start doing a Rietveld refinement according to the > more restrictive definition that has been used in this discussion - and > then turn off those structural parameters for one reason or another, maybe > to look into hkl-dependent broadening or such (still constrained by the > structure in contrast to simple “profile fitting” where any peak can have > its completely independent parameters!)? Did I just quit doing a Rietveld > refinement? What if I have a problematic atom, maybe on a split site close > to an inversion center, and I turn that one off? Do I stop doing a Rietveld > refinement when I change that variable to “no”? Does any Rietveld > refinement start out as a “non-Rietveld refinement” when we adjust > background, lattice parameters, approximate peak shapes and phase > fractions, to get to a point where we can actually vary the structural > variables without disastrous consequences? And the moment we touch the > structural variables, in the same software, suddenly it becomes a Rietveld > refinement? Do I need to only turn on ADPs, or is the “threshold” that I > must allow atom positions to refine before I can call it Rietveld? If so, > do I need to turn on all atom positions or only some of them? What if I > define rigid bodies, do I still qualify for a Rietveld refinement? > > Just some fun thoughts here! > > > > Cora > > > > *From:* rietveld_l-requ...@ill.fr *On Behalf > Of *Stefan Seidlmayer > *Sent:* Tuesday, January 16, 2024 9:51 AM > *To:* rietveld_l@ill.fr > *Subject:* [EXTERNAL] Re: Misconduct > > > > Dear all, > > > > I was following the discussion also with great interest, as terminology is > important to distinguish properly between different items. > > > > To my understanding the Rietveld approach was new because it constrained > the fitting of a peak list generated "from a structure" with the refinement > of the profile of the peaks themselves. > > > > Thus I would have the impression that everytime when we use a > constrainement of peak list whi
RE: [EXTERNAL] Re: Misconduct/terminology
I should have been more specific – this was not with respect to QPA specifically – it was a more general statement about the terminology of when we should or shouldn’t use the expression “Rietveld refinement”. Although we actually often do Rietveld refinements on 2 phase mixtures, where one phase is an internal standard, which we use to determine sample height offset changes with temperature. And no, we don’t just try to get lower R-values by randomly adding parameters! Cora From: Alan W Hewat Sent: Tuesday, January 16, 2024 11:54 AM To: Lind-Kovacs, Cora Cc: rietveld_l@ill.fr Subject: Re: [EXTERNAL] Re: Misconduct "In many cases, we will indeed refine those structural parameters to get the best possible match". The best possible match is not the objective of QPA, but rather the best possible estimation of the phase fractions. You can refine parameters describing particle size, strain, lattice constants, etc for each phase, but please believe the structural coordinates obtained from well characterised (usually mono-phase) samples. Whether it is structure refinement or phase refinement, please resist the temptation to add parameters simply to obtain a lower R-factor. Dr Alan Hewat, NeutronOptics Grenoble, FRANCE (from phone) alan.he...@neutronoptics.com<mailto:alan.he...@neutronoptics.com> +33.476984168 VAT:FR79499450856 http://NeutronOptics.com/hewat<http://neutronoptics.com/hewat> ___ On Tue, 16 Jan 2024, 17:34 Lind-Kovacs, Cora, mailto:cora.l...@utoledo.edu>> wrote: I like Stefan’s view outlined below. In many cases, we will indeed refine those structural parameters to get the best possible match. But when I use a structural model so that I calculate F(hkl) values, and combine that with the ability to vary all kinds of parameters - some structure-related, some instrument/setup related, some microstructure related – I would (and have!) call that a Rietveld refinement. Because if I did not, there would be a lot of fun questions to contemplate: What if I start doing a Rietveld refinement according to the more restrictive definition that has been used in this discussion - and then turn off those structural parameters for one reason or another, maybe to look into hkl-dependent broadening or such (still constrained by the structure in contrast to simple “profile fitting” where any peak can have its completely independent parameters!)? Did I just quit doing a Rietveld refinement? What if I have a problematic atom, maybe on a split site close to an inversion center, and I turn that one off? Do I stop doing a Rietveld refinement when I change that variable to “no”? Does any Rietveld refinement start out as a “non-Rietveld refinement” when we adjust background, lattice parameters, approximate peak shapes and phase fractions, to get to a point where we can actually vary the structural variables without disastrous consequences? And the moment we touch the structural variables, in the same software, suddenly it becomes a Rietveld refinement? Do I need to only turn on ADPs, or is the “threshold” that I must allow atom positions to refine before I can call it Rietveld? If so, do I need to turn on all atom positions or only some of them? What if I define rigid bodies, do I still qualify for a Rietveld refinement? Just some fun thoughts here! Cora From: rietveld_l-requ...@ill.fr<mailto:rietveld_l-requ...@ill.fr> mailto:rietveld_l-requ...@ill.fr>> On Behalf Of Stefan Seidlmayer Sent: Tuesday, January 16, 2024 9:51 AM To: rietveld_l@ill.fr<mailto:rietveld_l@ill.fr> Subject: [EXTERNAL] Re: Misconduct Dear all, I was following the discussion also with great interest, as terminology is important to distinguish properly between different items. To my understanding the Rietveld approach was new because it constrained the fitting of a peak list generated "from a structure" with the refinement of the profile of the peaks themselves. Thus I would have the impression that everytime when we use a constrainement of peak list which is generated from a structure and do not refine a list of "individual peaks" it is a Rietveld-type refinement. A Profile Refinement is/was in my current understanding, when the peak/reflection position is not constrained by the structure parameters, but can be refined in an arbitrary way, individual for each peak/reflection. From the positions refined in this way, one could then determine cell parameter etc. But this would require a secondary step. First refine all found peaks/reflection with a common profile. Then determine the lattice parameters from the refined peak positions etc. This is also troublesome as without prior structure "knowledge" it may and surely is that certain reflection which in fact are overlapping multiple reflections are improperly identified as "one" reflection. Thi
Re: [EXTERNAL] Re: Misconduct
"In many cases, we will indeed refine those structural parameters to get the best possible match". The best possible match is not the objective of QPA, but rather the best possible estimation of the phase fractions. You can refine parameters describing particle size, strain, lattice constants, etc for each phase, but please believe the structural coordinates obtained from well characterised (usually mono-phase) samples. Whether it is structure refinement or phase refinement, please resist the temptation to add parameters simply to obtain a lower R-factor. Dr Alan Hewat, NeutronOptics Grenoble, FRANCE (from phone) alan.he...@neutronoptics.com +33.476984168 VAT:FR79499450856 http://NeutronOptics.com/hewat ___ On Tue, 16 Jan 2024, 17:34 Lind-Kovacs, Cora, wrote: > I like Stefan’s view outlined below. In many cases, we will indeed refine > those structural parameters to get the best possible match. But when I use > a structural model so that I calculate F(hkl) values, and combine that with > the ability to vary all kinds of parameters - some structure-related, some > instrument/setup related, some microstructure related – I would (and have!) > call that a Rietveld refinement. > > Because if I did not, there would be a lot of fun questions to > contemplate: What if I start doing a Rietveld refinement according to the > more restrictive definition that has been used in this discussion - and > then turn off those structural parameters for one reason or another, maybe > to look into hkl-dependent broadening or such (still constrained by the > structure in contrast to simple “profile fitting” where any peak can have > its completely independent parameters!)? Did I just quit doing a Rietveld > refinement? What if I have a problematic atom, maybe on a split site close > to an inversion center, and I turn that one off? Do I stop doing a Rietveld > refinement when I change that variable to “no”? Does any Rietveld > refinement start out as a “non-Rietveld refinement” when we adjust > background, lattice parameters, approximate peak shapes and phase > fractions, to get to a point where we can actually vary the structural > variables without disastrous consequences? And the moment we touch the > structural variables, in the same software, suddenly it becomes a Rietveld > refinement? Do I need to only turn on ADPs, or is the “threshold” that I > must allow atom positions to refine before I can call it Rietveld? If so, > do I need to turn on all atom positions or only some of them? What if I > define rigid bodies, do I still qualify for a Rietveld refinement? > > Just some fun thoughts here! > > > > Cora > > > > *From:* rietveld_l-requ...@ill.fr *On Behalf > Of *Stefan Seidlmayer > *Sent:* Tuesday, January 16, 2024 9:51 AM > *To:* rietveld_l@ill.fr > *Subject:* [EXTERNAL] Re: Misconduct > > > > Dear all, > > > > I was following the discussion also with great interest, as terminology is > important to distinguish properly between different items. > > > > To my understanding the Rietveld approach was new because it constrained > the fitting of a peak list generated "from a structure" with the refinement > of the profile of the peaks themselves. > > > > Thus I would have the impression that everytime when we use a > constrainement of peak list which is generated from a structure and do not > refine a list of "individual peaks" it is a Rietveld-type refinement. > > > > A Profile Refinement is/was in my current understanding, when the > peak/reflection position is not constrained by the structure parameters, > but can be refined in an arbitrary way, individual for each peak/reflection. > > From the positions refined in this way, one could then determine cell > parameter etc. But this would require a secondary step. First refine all > found peaks/reflection with a common profile. Then determine the lattice > parameters from the refined peak positions etc. > > This is also troublesome as without prior structure "knowledge" it may and > surely is that certain reflection which in fact are overlapping multiple > reflections are improperly identified as "one" reflection. > > This is very the neatness of the Rietveld approach comes into play, as it > generates even overlapping reflections which then generate the total > profile by constraining them with the profile parameters. > > > > So the main point in differentiation is then in my eyes: > > Rietveld-Refinement: The use of a structure generated peak list which is > constrained with profile parameters. > > > > Profile-Refinement: Profile refinement can work on individual peaks with > N
RE: [EXTERNAL] Re: Misconduct
I like Stefan’s view outlined below. In many cases, we will indeed refine those structural parameters to get the best possible match. But when I use a structural model so that I calculate F(hkl) values, and combine that with the ability to vary all kinds of parameters - some structure-related, some instrument/setup related, some microstructure related – I would (and have!) call that a Rietveld refinement. Because if I did not, there would be a lot of fun questions to contemplate: What if I start doing a Rietveld refinement according to the more restrictive definition that has been used in this discussion - and then turn off those structural parameters for one reason or another, maybe to look into hkl-dependent broadening or such (still constrained by the structure in contrast to simple “profile fitting” where any peak can have its completely independent parameters!)? Did I just quit doing a Rietveld refinement? What if I have a problematic atom, maybe on a split site close to an inversion center, and I turn that one off? Do I stop doing a Rietveld refinement when I change that variable to “no”? Does any Rietveld refinement start out as a “non-Rietveld refinement” when we adjust background, lattice parameters, approximate peak shapes and phase fractions, to get to a point where we can actually vary the structural variables without disastrous consequences? And the moment we touch the structural variables, in the same software, suddenly it becomes a Rietveld refinement? Do I need to only turn on ADPs, or is the “threshold” that I must allow atom positions to refine before I can call it Rietveld? If so, do I need to turn on all atom positions or only some of them? What if I define rigid bodies, do I still qualify for a Rietveld refinement? Just some fun thoughts here! Cora From: rietveld_l-requ...@ill.fr On Behalf Of Stefan Seidlmayer Sent: Tuesday, January 16, 2024 9:51 AM To: rietveld_l@ill.fr Subject: [EXTERNAL] Re: Misconduct Dear all, I was following the discussion also with great interest, as terminology is important to distinguish properly between different items. To my understanding the Rietveld approach was new because it constrained the fitting of a peak list generated "from a structure" with the refinement of the profile of the peaks themselves. Thus I would have the impression that everytime when we use a constrainement of peak list which is generated from a structure and do not refine a list of "individual peaks" it is a Rietveld-type refinement. A Profile Refinement is/was in my current understanding, when the peak/reflection position is not constrained by the structure parameters, but can be refined in an arbitrary way, individual for each peak/reflection. From the positions refined in this way, one could then determine cell parameter etc. But this would require a secondary step. First refine all found peaks/reflection with a common profile. Then determine the lattice parameters from the refined peak positions etc. This is also troublesome as without prior structure "knowledge" it may and surely is that certain reflection which in fact are overlapping multiple reflections are improperly identified as "one" reflection. This is very the neatness of the Rietveld approach comes into play, as it generates even overlapping reflections which then generate the total profile by constraining them with the profile parameters. So the main point in differentiation is then in my eyes: Rietveld-Refinement: The use of a structure generated peak list which is constrained with profile parameters. Profile-Refinement: Profile refinement can work on individual peaks with NO structure information at all. Pawley-Refinement then is logically a little bit of a Hybrid, as the reflection list is pre-generated from the structure. But the intensities are just matched to best fit the profile. While in a true Rietveld even the reflection intensities is always generated from the underlying structure model applied in the Rietveld refinement. What do you think of this point of view? Am I overlooking something? Best regards Stefan Seidlmayer Am Di., 16. Jan. 2024 um 15:23 Uhr schrieb Alan W Hewat mailto:alan.he...@neutronoptics.com>>: Ha ! When Terry Sabine proposed to call it Rietveld Refinement, I told him that Rietveld was already "refined". Dr Alan Hewat, NeutronOptics Grenoble, FRANCE (from phone) alan.he...@neutronoptics.com<mailto:alan.he...@neutronoptics.com> +33.476984168 VAT:FR79499450856 http://NeutronOptics.com/hewat<http://neutronoptics.com/hewat> ___ On Tue, 16 Jan 2024, 15:11 , mailto:alberto.martine...@spin.cnr.it>> wrote: Dear all, I'm following this interesting discussion. It seems to me that sometimes there is an improper use of terminology, in particular when we talk about "profile refinement";
