Finding minimum of $l-1$ norm of affine function under constrained using dual approach

296 Views Asked by At

I am trying to solve following minimization problem.

$$\,\min_{x \in R^n} \left\lVert Ax -b\right\rVert_2$$ $$s.t \left\lVert x\right\rVert_2 \le t, t\ge0$$

My procedure is as follows.

First I find it's Lagrangian

$L(x,v) = \left\lVert Ax -b\right\rVert_2 + v^T(\left\lVert x\right\rVert_2 -t)$

I can express this in $l2$ norm

$L(x,v) = \left\lVert Ax -b\right\rVert_2^2 + v^T(\left\lVert x\right\rVert_2^2 -t^2)$ (Is this correct?)

Then I write the Lagrangian dual

$g(v) = \,\inf_{x\in R^n} L(x,v)$

$g(v) = \,\inf_{x\in R^n} \left\lVert Ax -b\right\rVert_2^2 + v^T(\left\lVert x\right\rVert_2^2 -t^2)$

Considering the convexity of two terms I find the gradient of the Lagrangian and considering first order condition, find the x which minimize Lagrangian

$\nabla L(x,v) = 2A^T(Ax-b)+ 2vx^T = 0$

$x = b^TA/(v + AA^T)$

My plan is to plug this x to Lagrangian and obtain the dual of it and then maximize it. But I not sure how to plug it and proceed

  1. Am I on the correct path to solve this?
  2. Are there any mistakes in my procedure?