CosmoMC: how to set initial parameter value?
-
- Posts: 4
- Joined: November 02 2006
- Affiliation: shandong university
CosmoMC: how to set initial parameter value?
1.how does we set the new parameter center and width value for more quick convergence,if we know st.dev?
I guess center value = bestfit value ,and the width value =st.dev , are these right?
2 If the dotted line(mean likelihood) is very flat , but solid line looks like normal gaussian distribution, what does this mean to ?
thank you in advance.
I guess center value = bestfit value ,and the width value =st.dev , are these right?
2 If the dotted line(mean likelihood) is very flat , but solid line looks like normal gaussian distribution, what does this mean to ?
thank you in advance.
-
- Posts: 1941
- Joined: September 23 2004
- Affiliation: University of Sussex
- Contact:
Re: CosmoMC: how to set initial parameter value?
The propose width should be the order of the conditional width (i.e. the st.dev. of the new parameter when other parameters are fixed).lele zhang wrote:1.how does we set the new parameter center and width value for more quick convergence,if we know st.dev?
I guess center value = bestfit value ,and the width value =st.dev , are these right?
It means all values of the parameter can fit the database nearly equally well, but there are more ways of changing the other parameters near the central value. Or it means the chains have not converged (e.g. stuck near the peak).lele zhang wrote: 2 If the dotted line(mean likelihood) is very flat , but solid line looks like normal gaussian distribution, what does this mean to ?
-
- Posts: 4
- Joined: November 02 2006
- Affiliation: shandong university
CosmoMC: how to set initial parameter value?
thank you , but i am not sure if we want to run again ,which one could accelerate the convergence ,the marginlized value or best fit value ?
-
- Posts: 4
- Joined: November 02 2006
- Affiliation: shandong university
CosmoMC: how to set initial parameter value?
sorry , I mean for setting inital center value which one is bettter