Tags: alg, optimization
- infimal convolution: ,
- Minkowski sum: ,
- convex conjugate: ,
- epigraph: .
- https://angms.science/doc/CVX/Epigraph.pdf
- https://math.stackexchange.com/questions/1597809/inf-convolution-two-basic-questions
1 piecewise linear function
A intuitive but very complex definition is the following[1],
Definition1
Let
be a set of bounded convex polytopes in
.
A piecewise linear function
can be defined as
where
is a vector associated with polytope
.
For continuity, we require
For convexity, we further require that
(Later I find a much simpler definition in [3] example 3.5.)
- for any subset , is convex;(for polytopes)
- the restriction of to any two polytopes from P that share a facet is convex.[2] (for . In fact, 1 is already included in 2.)
Definition2piecewise linear function in
$\R^d$
It is shown in exercise 3.29 that every piecewise linear convex function
can be expressed in this form.
Proof
We have
for any
.
can locate in the same
as
since we can arbitrarily choose
.
Thus
.
Note that
is linear in each
.
Thus
.
Thus
.
Now given these definitions we are particularly interested in the
minimum number of polytopes which define the piecewise liner convex
function in high dimension.
2 properties
pwl = piecewise linear. Let be the number of hyperplanes in the definition of .2.1 convex conjugate
Observation3
Let
be the convex conjugate of a pwl convex function
.
is also pwl convex if restricted to
.
Consider the convex conjugate from a geometric view. The epigraph of our
pwl convex function
is some convex polytope in
.
The convex conjugate is
.
is a hyperplane with normal vector
and passing through the origin. Now
is the amount of space hyperplane
has to shift along the
dimension to make itself a supporting hyperplane of
.
Note that the tangent points are exactly vertices of
.
Proof
It is safe to write
since we only consider the extended domain. Thus we have
.
Let
be the number of vertices on
.
One can see that
is the maximum of
affine functions.
I believe there will be only
hyperplanes on
instead of
…
However, we know that in general
is the maximum of at least
functions since every vertex corresponds to a hyperplane in
.
2.2 pwl convex function in a linear mapping
Problem4
Let
be a pwl convex function. Does there always exist a pwl convex
and a linear mapping
such that
.
As you expected, the answer is no. Let
be the maximum of a set of 2D planes. Consider a series of points
on the 2D plane. After applying the linear mapping to
,
we will get a sequence of numbers(points in 1D)
.
We assume that
is non-decreasing. Note that the value of
on
is always unimodal since
is convex. However, the value of
on
may not be unimodal. Thus the composition of a linear mapping and a pwl
convex function in 1D is not equivalent to pwl convex functions in high
dimensions.
3 sum of pwl convex functions
I want to show that in general the number of hyperplanes in the sum of pwl convex functions can be large. It is known that for two pwl convex functions , . It is also known that (with some requirements on and ). There is a theorem in [4] section 4.3 which shows that the number of faces of the Minkowski sum of two polytopes can be huge. The bound can be reached by sums of cyclic polytopes. We can define pwl convex functions based on cyclic polytopes and we know that the Minkowski sum will have lots of faces(of different dimensions). We also know that the number of faces in is at least where is the number of vertices(faces of 1D) in . Now if the infimal convolution of two pwl convex functions also has many faces, the number of faces in the sum of pwl convex functions will be large.
Problem5
Let
be two pwl convex functions in
.
Let
be the number of hyperplanes in
respectively. What is the minimum number of hyperplanes in
?
References
[1]
A.
Toriello, J.P. Vielma, Fitting piecewise linear continuous functions,
European Journal of Operational Research. 219 (2012) 86–95 10.1016/j.ejor.2011.12.030.
[2]
J.M. Tarela, E. Alonso, M.V. Martı́nez, A
representation method for PWL functions oriented to parallel processing,
Mathematical and Computer Modelling. 13 (1990) 75–83 10.1016/0895-7177(90)90090-a.
[3]
S.P. Boyd, L. Vandenberghe, Convex
optimization, Cambridge University Press, Cambridge, UK ; New York,
2004.
[4]
T.
Mountford, T. Liebling, K. Fukuda, P. Gritzmann, G. Ziegler, Minkowski
Sums of Polytopes: Combinatorics
and Computation, PhD thesis, n.d.