I have a set of $N$ points $(x_i, y_i, z_i)$ and I need to find the best plane fit to them. So far I've been numerically minimizing the sum of the absolute valued point-plane distances:
$$f(a, b, c, d)= \sum_{i=1}^{N} \frac{|ax_i + by_i + cz_i + d|}{\sqrt{a^2 + b^2 + c^2}}$$
which works. But recently I found out about the singular value decomposition method, which as far as I understand, analytically minimizes the squared point-plane distances:
$$f(a, b, c, d)= \sum_{i=1}^{N} \frac{(ax_i + by_i + cz_i + d)^2}{a^2 + b^2 + c^2}$$
Keeping in mind that I am not a mathematician, could you explain step by step why (or how) the SVD method minimizes the squared point-plane distance?