Fitting parameters corrleations between variables
Dear All, I am trying to fit in Pd compound and the fit that I obtained looks kinda good. I used Pd metal and Pd(OAc)2 shells for fitting. These standards were collected and fitted simultaneously and I got the SO2 value there as 0.67 with Pd foil coordination number as 12 which is correct when considered Pd foil CN. Next while fitting the actual sample (unknown) I used the formula in SO2 column as amp*N1 and put amp as 0.67 and N as 1. I kept N1 as variable (Guess). same I did for the second shell of Pd-O with N2. amp is kept on SET value of 0.67. After fitting I received the N1 and N2 value as 5.6 (0.5) and 1.6 (0.7) (roughly). however, when I checked the correlations between the variables I got the values as given in the log file attached. I saw there is some correlation between bkg01_01 and N2 and many such correlations between bkg values. I have so far never seen such correlations. What are the implications of these correlations? does this means that my fitting assumptions are incorrect. does correlations values in negative indicates something? Thank you very much for the input. By the way my RBKG is 1.2 in athena and I used this same file for fitting purpose and on artemis I used rmin 1.2 and rmax 3.4. All these information are in log file attached Any help in understanding this is appreciated. -- Best Regards, Pushkar Shejwalkar. Post-doctoral -Researcher, Tokyo Engineering University, Tokyo-to Japan
Pushkar, On Mon, Nov 23, 2015 at 8:59 PM, pushkar shejwalkar < pshejwalkar2004@gmail.com> wrote:
Dear All, I am trying to fit in Pd compound and the fit that I obtained looks kinda good. I used Pd metal and Pd(OAc)2 shells for fitting. These standards were collected and fitted simultaneously and I got the SO2 value there as 0.67 with Pd foil coordination number as 12 which is correct when considered Pd foil CN. Next while fitting the actual sample (unknown) I used the formula in SO2 column as amp*N1 and put amp as 0.67 and N as 1. I kept N1 as variable (Guess). same I did for the second shell of Pd-O with N2. amp is kept on SET value of 0.67. After fitting I received the N1 and N2 value as 5.6 (0.5) and 1.6 (0.7) (roughly). however, when I checked the correlations between the variables I got the values as given in the log file attached. I saw there is some correlation between bkg01_01 and N2 and many such correlations between bkg values. I have so far never seen such correlations. What are the implications of these correlations? does this means that my fitting assumptions are incorrect. does correlations values in negative indicates something?
Correlation between two variables A and B means that if you changed one variable (A) away from its best-fit value, you could improve that fit by changing the other variable (B). In fact the correlation tells you by how much, and (from the sign) in what direction that change would be. A correlation of 0 means that if one changed A, B would not need to change at all. A correlation of 1 would mean that a change in A by 1 standard error (the reported error), B would have to change by 1 standard error. And, again, the sign gives the direction of that change. Having correlated values does not mean that your assumptions are incorrect. Thank you very much for the input. By the way my RBKG is 1.2 in athena and
I used this same file for fitting purpose and on artemis I used rmin 1.2 and rmax 3.4. All these information are in log file attached Any help in understanding this is appreciated.
-- Best Regards, Pushkar Shejwalkar. Post-doctoral -Researcher, Tokyo Engineering University, Tokyo-to Japan
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
--Matt
Dear Matt,
Thank you very much for your response. One more question. What value of
corelation is acceptable/publishable? also having a correlation number
means the fit is not finished and I should change the values of variable
that are correlated? is that so?
Best
Pushkar
On Tue, Nov 24, 2015 at 1:48 PM, Matt Newville
Pushkar,
On Mon, Nov 23, 2015 at 8:59 PM, pushkar shejwalkar < pshejwalkar2004@gmail.com> wrote:
Dear All, I am trying to fit in Pd compound and the fit that I obtained looks kinda good. I used Pd metal and Pd(OAc)2 shells for fitting. These standards were collected and fitted simultaneously and I got the SO2 value there as 0.67 with Pd foil coordination number as 12 which is correct when considered Pd foil CN. Next while fitting the actual sample (unknown) I used the formula in SO2 column as amp*N1 and put amp as 0.67 and N as 1. I kept N1 as variable (Guess). same I did for the second shell of Pd-O with N2. amp is kept on SET value of 0.67. After fitting I received the N1 and N2 value as 5.6 (0.5) and 1.6 (0.7) (roughly). however, when I checked the correlations between the variables I got the values as given in the log file attached. I saw there is some correlation between bkg01_01 and N2 and many such correlations between bkg values. I have so far never seen such correlations. What are the implications of these correlations? does this means that my fitting assumptions are incorrect. does correlations values in negative indicates something?
Correlation between two variables A and B means that if you changed one variable (A) away from its best-fit value, you could improve that fit by changing the other variable (B). In fact the correlation tells you by how much, and (from the sign) in what direction that change would be.
A correlation of 0 means that if one changed A, B would not need to change at all. A correlation of 1 would mean that a change in A by 1 standard error (the reported error), B would have to change by 1 standard error. And, again, the sign gives the direction of that change.
Having correlated values does not mean that your assumptions are incorrect.
Thank you very much for the input. By the way my RBKG is 1.2 in athena and
I used this same file for fitting purpose and on artemis I used rmin 1.2 and rmax 3.4. All these information are in log file attached Any help in understanding this is appreciated.
-- Best Regards, Pushkar Shejwalkar. Post-doctoral -Researcher, Tokyo Engineering University, Tokyo-to Japan
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
--Matt
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
-- Best Regards, Pushkar Shejwalkar. Post-doctoral -Researcher, Tokyo Engineering University, Tokyo-to Japan
Hi Pushkar, On Mon, Nov 23, 2015 at 11:28 PM, pushkar shejwalkar < pshejwalkar2004@gmail.com> wrote:
Dear Matt, Thank you very much for your response. One more question. What value of corelation is acceptable/publishable?
A correlation > 0.999 is probably an indication of duplicate parameters in a model, but it's common to publish parameters (say, N and sigma2 or E0 and R) that are correlated by more than 0.9. The correlation simply measures the amount by which a parameters best-fit value would change if another parameter was changed away from its best-fit value. The uncertainties take these correlations into account.
also having a correlation number means the fit is not finished and I should change the values of variable that are correlated? is that so?
No. The fit is (or "can be") finished. The reported best-fit values are really the best values found, and the reported uncertainties are those that increase the best chi-square by reduced chi-square, taking into account the correlation. The correlation does not indicate a bad model or a bad fit, it's just a consequence of the complex model and limited data for XAFS. Hope that helps, --Matt
participants (2)
-
Matt Newville
-
pushkar shejwalkar