# Solving a system of quadratic vector equations

This problem arises from my research in computer vision, specifically projective homography:

I have $n$ unknown variables, represented by an $n\times 1$ vector $\mathbf{x}$. There is a system of $n$ equations of the form $\mathbf{x}^{T}\mathbf{A_i}\mathbf{x} + \mathbf{b_i}^{T}\mathbf{x} + c_i = 0$, where $\mathbf{A_i}$ is an $n\times n$ symmetric matrix, $\mathbf{b_i}$ is an $n\times 1$ vector and $c_i$ is a scalar for $i=1\ldots n$, that $\mathbf{x}$ has to satisfy.

It is also known that every entry of the solution $\mathbf{x}$ is small, i.e. comparatively smaller than their respective coefficients.

Is there a closed-form or iterative method to solve for $\mathbf{x}$?

-

## 2 Answers

Newton's method converges fast, when it converges. In Newton's method, you start from an initial estimate $\bf x$, and solve a linear equation for $\bf h$ to get the next iterate $\bf x+h$. In this case this seems to give $${\bf x}^T({\bf A}_i{\bf x}+{\bf b}_i)+(2{\bf A}_i{\bf x}+{\bf b}_i)^T{\bf h}+c_i=0$$ for $i=1$ to $n$. It is obtained by dropping the term which is quadratic in ${\bf h}$.

-

sometimes I think the most difficult part is not the method itself, but the guess of initialized value. Does anyone know how to give a better initialized value?

-
This is not an answer to the question. If you have a new question, please use the "Ask Question" link near the top right of the page. –  robjohn Jan 20 '13 at 16:43