Global gradient-norm projection residual
= Global gradient-norm projection residual
{title2=$u=g-P_{A^*B_\alpha}g$}
Let $C=A^*B_\alpha$ with one global dual ball. Its <support function> is $\alpha\|Au\|_2$, and the <proximal operator of a support function> gives $u=g-P_Cg$. The removed component is the <convex projection>, while the reconstructed <image signal> is its residual. <Projected gradient descent> on $\tfrac12\|g-A^*p\|_2^2$, $p\in B_\alpha$, converges for $0<\tau<2/\|A\|^2$.