Let $X$ be a random variable which follows an $n$-dimensional Gaussian distribution with mean vector $\mu\in\mathbb{R}^n$ and covariance matrix the symmetric positive definite $n\times n$ matrix $\Sigma$, i.e. $\Sigma\in\mathbb{S}_{++}^{n}$. The probability density function of $X$ is given by $f_{X}\colon\mathbb{R}^n\to\mathbb{R}^{*}_{+}$, with

$$ f_{X}(\mathbf{x};\mu,\Sigma) = \frac{1}{(2\pi)^{n/2}\lvert\Sigma\rvert^{1/2}} \operatorname{exp} \Big\{ -\frac{1}{2}(\mathbf{x}-\mu)^T\Sigma^{-1}(\mathbf{x}-\mu) \Big\}. $$

Also let $g\colon\mathbb{R}^n\to\mathbb{R}$. We would like to evaluate the following integral

$$ I = \int_{\Omega} \! \big[1-g(\mathbf{x})\big] f_{X}(\mathbf{x};\mu,\Sigma) \, \mathrm{d}\mathbf{x}, $$


$$ I = \frac{1}{(2\pi)^{n/2}\lvert\Sigma\rvert^{1/2}} \int_{\Omega} \! \big[1-g(\mathbf{x})\big] \operatorname{exp} \Big\{ -\frac{1}{2}(\mathbf{x}-\mu)^T\Sigma^{-1}(\mathbf{x}-\mu) \Big\} \, \mathrm{d}\mathbf{x}, $$

where $\Omega=\{\mathbf{x}\in\mathbb{R}^n \mid g(\mathbf{x})\leq 1\}$. In general, it would be very desired if for the function $g$ we know nothing but that it is a real-valued function defined on $\mathbb{R}^n$. If this is impossible, then we could assume additionally that $g$ can be written as

$$ g(\mathbf{x})=\sum_{j=1}^{N} \alpha_j k(\mathbf{x},\mathbf{x}_j), $$

where $N\in\mathbb{N}$, $\alpha_j\in\mathbb{R}$, $j=1,\dotsc,N$, and $k\colon\mathbb{R}^n\times\mathbb{R}^n\to\mathbb{R}_{+}^{*}$ is symmetric.

It could be found some connection with a question I asked $8$ mounths ago.

EDIT: In fact, there has been the following "replacement": $f(\mathbf{x})=\mathbf{w}\cdot\mathbf{x}$, where $\mathbf{w}\in\mathbb{R}^n$, and $f$ belongs to a Reproducing Kernel Hilbert Space (RKHS) with associated kernel $k$ (see above). This question can be connected with this one, if someone wants to go deeper.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.