Page **1** of **1**

### Cobaya minimizer question

Posted: **January 21 2022**

by **Helena Garcia**

I want to run minimization over the calculated chains using mcmc with cobaya and I don't know where you I write the path of the chains so that the program can use them to do a more precise minimazation? Thanks!

### Re: Cobaya minimizer question

Posted: **January 21 2022**

by **Antony Lewis**

You need to do a minimization run to find a minimum, you can't find a global minimum from chains since in high dimensions there are almost no samples near the posterior peak.

### Re: Cobaya minimizer question

Posted: **January 21 2022**

by **Helena Garcia**

Thanks for your answer!

I am running nested models, for example, LCDM and the LCDM + w free, therefore, I'm expecting a better fit for the model that has more freedom (in this case LCDM+w. Nevertheless, when I run minimizer over this two models, I get a smaller minuslogpost (in absolute magnitude) for the LCDM than for the LCDM + w free, which is mathematically inconsistent. I have ran the full chains since I thought it was necessary in order to run the minimizer.

How I set up the minimizer run is by changing the last part in the .yaml file (instead of mcmc sampler):

Code: Select all

```
sampler:
minimize:
method: bobyqa
ignore_prior: False
max_evals: 1e6d
best_of: 2
confidence_for_unbounded: 0.9999995
```

Is this the way to proceed or am I doing something wrong? How should one set this run?

Thanks in advanced!

### Re: Cobaya minimizer question

Posted: **January 21 2022**

by **Antony Lewis**

Looks OK. You can make use of your chains by setting the minimizer "covmat" to the covariance matrix from your chains. The minimizer may work much better with a sensible covmat.