凸集上投影

the following content is from https://cstheory.wordpress.com/2010/08/23/projections-onto-a-convex-body/#more-203.

 

Projections onto a Convex Body

Given a point 凸集上投影 and an affine subspace 凸集上投影, the projection of 凸集上投影 onto 凸集上投影 is the point 凸集上投影 such that,

凸集上投影

One can obviously define other notions of projection but the above is probably the most commonly used in geometry. If 凸集上投影 then the projected point 凸集上投影 is the unique point in 凸集上投影 such that the vector 凸集上投影 is orthogonal to 凸集上投影. Also is well known that projecting any segment to an affine subspace can only shrink its length. The proofs of these facts are easy to see. But in fact these facts are just corollaries of the following two results.

Given any nonempty closed convex set 凸集上投影 and a point 凸集上投影, there is a unique point 凸集上投影 such that,

 

凸集上投影

and,

Let 凸集上投影 be a nonempty closed convex set. The mapping 凸集上投影 is a contraction i.e. for any 凸集上投影 we have,

 

凸集上投影

Applying these results to affine spaces (which are nonempty closed convex sets) yields the results mentioned earlier. This projection that maps 凸集上投影 to 凸集上投影 is known as the metric projection. The proofs of these facts are in order.

Proof: First we show that the minimum distance to 凸集上投影 is indeed obtained. This is easy and the details are as follows. Since 凸集上投影 is nonempty there is some point 凸集上投影. Let 凸集上投影 denote 凸集上投影. Clearly a closest point to 凸集上投影 can only lie in 凸集上投影 where 凸集上投影denotes a closed ball of radius 凸集上投影 with center 凸集上投影. But 凸集上投影 is a closed and bounded set and is therefore compact. It is also nonempty. The function 凸集上投影 defined for 凸集上投影 as 凸集上投影is a continuous function on a compact set and attains its minimum at some point 凸集上投影.

Next we prove the point 凸集上投影 where 凸集上投影 attains minimum, is unique. If 凸集上投影 is another point at which 凸集上投影 then in the triangle 凸集上投影, which has two equal sides, the perpendicular from 凸集上投影 onto segment 凸集上投影 is clearly shorter than the length 凸集上投影. However the base of this perpendicular lies on the segment itself. By convexity of 凸集上投影 the entire segment is in 凸集上投影 and thus we have found a point in 凸集上投影 closer to 凸集上投影 than 凸集上投影 which is a contradiction.

Next we prove that the metric projection is a contraction.

Proof: Let the points 凸集上投影 be projected to 凸集上投影 and 凸集上投影respectively and assume 凸集上投影. Consider the hyperplanes 凸集上投影 through 凸集上投影 that are orthogonal to the vector 凸集上投影. See figure below.

凸集上投影

The main observation is that 凸集上投影 must lie on the other side of 凸集上投影 as 凸集上投影. If 凸集上投影 lies on the same side of 凸集上投影 as 凸集上投影 then one can always find a point on the segment 凸集上投影 closer to 凸集上投影 than 凸集上投影 which would contradict that 凸集上投影 is the closest point to 凸集上投影 in 凸集上投影. Similarly 凸集上投影 cannot lie on same side of 凸集上投影 as 凸集上投影. Thus 凸集上投影 is at least the distance between 凸集上投影 and 凸集上投影 which is 凸集上投影.