> For the complete documentation index, see [llms.txt](https://pointsource.gitbook.io/crowwrite/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://pointsource.gitbook.io/crowwrite/2024/0918-add-prior-to-chi-square-fitting.md).

# 0918-add prior to chi square fitting

#### 1. Gaussian Prior

A Gaussian (normal) prior on a parameter $\theta$ with mean $\mu$ and standard deviation $\sigma$ is expressed as:

$$
P(\theta) = \frac{1}{\sqrt{2\pi\sigma^2}} \exp\left( -\frac{(\theta - \mu)^2}{2\sigma^2} \right).
$$

This function indicates that the parameter $\theta$ is more likely to be close to the mean $\mu$, with a spread governed by $\sigma$.

#### 2. Log-Likelihood of the Gaussian Prior

In Bayesian analysis, the prior is incorporated by adding the log of the prior probability to the log-likelihood of the data. Since the goal is to minimize a function (as in chi-square fitting), we use the negative log of the prior:

$$
-\log P(\theta) = \frac{(\theta - \mu)^2}{2\sigma^2} + \frac{1}{2}\log(2\pi\sigma^2).
$$

#### 3. Simplifying the Prior Contribution

For the purpose of adding this term to the chi-square, we can ignore the constant part $\frac{1}{2}\log(2\pi\sigma^2)$ since it does not depend on the parameter $\theta$. We focus on the part that depends on $\theta$:

$$
-\log P(\theta) \propto \frac{(\theta - \mu)^2}{2\sigma^2}.
$$

#### 4. Incorporating into Chi-Square

To incorporate this prior into the chi-square fitting procedure, we need to add a term to the chi-square statistic that represents the prior's penalty on deviations of $\theta$ from $\mu$. The conventional chi-square statistic has a factor of 1/2 in the exponent:

$$
\chi^2\_{\text{prior}} = \sum\_j \frac{(\theta\_j - \mu\_j)^2}{\sigma\_j^2}.
$$

#### 5. Full Modified Chi-Square

The total chi-square to minimize becomes:

$$
\chi^2\_{\text{total}} = \chi^2\_{\text{data}} + \chi^2\_{\text{prior}},
$$

where:

* $\chi^2\_{\text{data}}$ is the standard chi-square term representing the fit to the data:

  $$
  \chi^2\_{\text{data}} = \sum\_{i=1}^{N} \frac{(y\_i - f(x\_i; \theta))^2}{\sigma\_i^2}.
  $$
* $\chi^2\_{\text{prior}}$ is the term derived from the Gaussian prior:

  $$
  \chi^2\_{\text{prior}} = \sum\_j \frac{(\theta\_j - \mu\_j)^2}{\sigma\_{\theta\_j}^2}.
  $$

These expressions now use LaTeX formatting for both inline and standalone equations.