Minimisation of $||Dx||^2$ with constraint $Ax = y$

82 Views Asked by At

Given a lower-dimensional vector $x$, a target vector of $y$ and constraints $A$ s.t. $Ax = y$, I'd like to minimise $||Dx||^2$, where $D$ is a matrix to calculate mixed forward and backward differences from x. For example, let $x$ be a 50-by-1 vector, and $y$ a 100-by-1 vector. Then $A$ would be a 100-by-50 matrix containing some linearly dependent constraints.

How could one go about solving this optimisation problem? In code is one thing, but even in principle, I'm a bit confused. I think the simple pseudoinverse isn't enough. Is some sort of iterative approach needed?

Even simpler, how does one minimise $||x||^2$, given these constraints? The $D$ is quite trivial to add.


I'm in a pickle here. This is an assignment, yes, but I was absent from a lecture and we haven't been provided with any additional material. Do these kind of problems have a specific name I could search with to find more information?

1

There are 1 best solutions below

2
On BEST ANSWER

Your problem falls into the class of quadratic programming.

According to the section of equality constraint at the solution method, it is equivalent to solving

$$\begin{bmatrix} D^TD & A^T \\ A & O\end{bmatrix}\begin{bmatrix} x \\ \lambda \end{bmatrix}=\begin{bmatrix} 0 \\ y\end{bmatrix}$$