In mathematics, a moment matrix is a special symmetric square matrix whose rows and columns are indexed by monomials. The entries of the matrix depend on the product of the indexing monomials only (cf. Hankel matrices.)

This article has multiple issues. Please help improve it or discuss these issues on the talk page. (Learn how and when to remove these template messages) This article may be confusing or unclear to readers. In particular, the subject of the article is not defined and the concepts linked in the introduction (e.g. monomial) seem to be not used with their common meaning. (April 2021) This article may be too technical for most readers to understand. (April 2021) This article provides insufficient context for those unfamiliar with the subject. (April 2021)

Moment matrices play an important role in polynomial fitting, polynomial optimization (since positive semidefinite moment matrices correspond to polynomials which are sums of squares)^[1] and econometrics.^[2]

A multiple linear regression model can be written as

${\displaystyle y=\beta _{0}+\beta _{1}x_{1}+\beta _{2}x_{2}+\dots \beta _{k}x_{k}+u}$

where ${\displaystyle y}$ is the explained variable, ${\displaystyle x_{1},x_{2}\dots ,x_{k}}$ are the explanatory variables, ${\displaystyle u}$ is the error, and ${\displaystyle \beta _{0},\beta _{1}\dots ,\beta _{k}}$ are unknown coefficients to be estimated. Given observations ${\displaystyle \left\{y_{i},x_{1i},x_{2i},\dots ,x_{ki}\right\}_{i=1}^{n}}$, we have a system of ${\displaystyle n}$ linear equations that can be expressed in matrix notation.^[3]

${\displaystyle {\begin{bmatrix}y_{1}\\y_{2}\\\vdots \\y_{n}\end{bmatrix}}={\begin{bmatrix}1&x_{11}&x_{12}&\dots &x_{1k}\\1&x_{21}&x_{22}&\dots &x_{2k}\\\vdots &\vdots &\vdots &\ddots &\vdots \\1&x_{n1}&x_{n2}&\dots &x_{nk}\\\end{bmatrix}}{\begin{bmatrix}\beta _{0}\\\beta _{1}\\\vdots \\\beta _{k}\end{bmatrix}}+{\begin{bmatrix}u_{1}\\u_{2}\\\vdots \\u_{n}\end{bmatrix}}}$

or

${\displaystyle \mathbf {y} =\mathbf {X} {\boldsymbol {\beta }}+\mathbf {u} }$

where ${\displaystyle \mathbf {y} }$ and ${\displaystyle \mathbf {u} }$ are each a vector of dimension ${\displaystyle n\times 1}$, ${\displaystyle \mathbf {X} }$ is the design matrix of order ${\displaystyle N\times (k+1)}$, and ${\displaystyle {\boldsymbol {\beta }}}$ is a vector of dimension ${\displaystyle (k+1)\times 1}$. Under the Gauss–Markov assumptions, the best linear unbiased estimator of ${\displaystyle {\boldsymbol {\beta }}}$ is the linear least squares estimator ${\displaystyle \mathbf {b} =\left(\mathbf {X} ^{\mathsf {T}}\mathbf {X} \right)^{-1}\mathbf {X} ^{\mathsf {T}}\mathbf {y} }$, involving the two moment matrices ${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {X} }$ and ${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {y} }$ defined as

${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {X} ={\begin{bmatrix}n&\sum x_{i1}&\sum x_{i2}&\dots &\sum x_{ik}\\\sum x_{i1}&\sum x_{i1}^{2}&\sum x_{i1}x_{i2}&\dots &\sum x_{i1}x_{ik}\\\sum x_{i2}&\sum x_{i1}x_{i2}&\sum x_{i2}^{2}&\dots &\sum x_{i2}x_{ik}\\\vdots &\vdots &\vdots &\ddots &\vdots \\\sum x_{ik}&\sum x_{i1}x_{ik}&\sum x_{i2}x_{ik}&\dots &\sum x_{ik}^{2}\end{bmatrix}}}$

and

${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {y} ={\begin{bmatrix}\sum y_{i}\\\sum x_{i1}y_{i}\\\vdots \\\sum x_{ik}y_{i}\end{bmatrix}}}$

where ${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {X} }$ is a square normal matrix of dimension ${\displaystyle (k+1)\times (k+1)}$, and ${\displaystyle \mathbf {X} ^{\mathsf {T}}\mathbf {y} }$ is a vector of dimension ${\displaystyle (k+1)\times 1}$.

• Design matrix
• Gramian matrix
• Projection matrix