CS 372: Computational Geometry

Lecture 13
Arrangements and Duality

Antoine Vigneron

King Abdullah University of Science and Technology

November 28, 2012

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 1 / 41

1 Introduction

2 Point-line duality

3 Upper envelopes

4 Arrangements of lines

5 An application of arrangements and duality

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 2 / 41

Outline

We saw planar graph duality in Lecture 10.

I In this lecture, a different type of duality: point-line duality.
We introduce line arrangements.
These two notions can often be combined to solve apparently difficult
computational geometry problems.

References:
Textbook Chapter 8.
Dave Mount’s lecture notes, Lectures 8, 19 and 20.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 3 / 41

Point-Line Duality

Point-Line Duality: A transformation that exchanges points and lines.

Motivation: For the same amount of work, twice as many results.

Theorem 1 Theorem 1∗
Duality
involves points and lines involves lines and points

Program 1 Program 1∗
Duality
deals with points and lines deals with lines and points

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 4 / 41

Point-Line Duality: Example

p(2, 1)
` : y = 2x − 1

(0.5, 0)
(0, −1)

notation:
p = `∗

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 5 / 41

Dual Point

Let ` be a non-vertical line.

I ` has equation ` : y = `a x − `b .
F `a is the slope of `.
I We associate the point `∗ = (`a , `b ) to `.
I `∗ is called the dual of `.
We say that ` is a line in the primal plane.
Here, the primal plane is associated with a coordinate frame Oxy .
We say that `∗ is a point in the dual plane.
We will use a coordinate frame Uab for the dual plane.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 6 / 41

Dual Point

y b `∗ (`a , `b )

`a
1
O x U a

` : y = `a x − `b

(0, −`b )

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 7 / 41

Dual Line
Let p(px , py ) be a point in the primal plane.
Its dual is a line p ∗ in the dual plane with equation

p ∗ : b = px a − py .

y b
p(px , py )
px
1
O U a
x
p ∗ : b = px a − py
(0, −py )

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 8 / 41

Self Inverse
Property (Self inverse)
For any point p in the primal plane, (p ∗ )∗ = p.

Proof:
p(px , py ) in the primal plane.
p ∗ : b = px a − py .
(p ∗ )∗ has coordinates (px , −(−py )).

Property (Self inverse)

For any line ` of the primal plane, (`∗ )∗ = `.

Proof:
` : y = `a x − `b .
`∗ (`a , `b ).
(`∗ )∗ : y = `a x − `b .
Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 9 / 41
Point-Line Duality is Incidence Preserving

p∗
p
`∗

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 10 / 41

Point-Line Duality is Incidence Preserving

Property
p ∈ ` iff `∗ ∈ p ∗ .

Proof:
Assume p ∈ `.
I It means py = `a px − `b .
I So `b = px `a − py .
I Thus (`a , `b ) ∈ p ∗ .
Now assume that `∗ ∈ p ∗ .
I Then (p ∗ )∗ ∈ (`∗ )∗ .
I By the self-inverse property, it yields p ∈ `.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 11 / 41

Multiple Incidence
Corollary
p1 , p2 and p3 are collinear iff p1∗ , p2∗ and p3∗ intersect at a common point.

p3∗

p3
p2 p2∗
p1

p1∗

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 12 / 41

Order Reversing

Property (Order reversing)

p lies below ` iff p ∗ is above `∗ .

`
p∗
p

`∗

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 13 / 41

Example: Dual of a line segment

Let s = pq be a line segment in R2 .

How to define its dual?
Its dual s ∗ is the union of the duals of the points of s.
All the points in s are collinear, so all the lines in s ∗ pass through one
point.
So it is a double wedge. (See next slide.)

Property
A line ` intersect a segment s iff `∗ is in s ∗ .

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 14 / 41

Example: Dual of a line segment

p∗ s∗
`
q
`∗
s
p
q∗

Primal plane Dual plane

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 15 / 41

Upper Envelope of Lines

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 16 / 41

Upper Envelope of Lines

Definition (Upper envelope)

The upper envelope of a set of lines is the set of the points that are above
all lines.

How to compute the upper envelope of a set L of n lines?

Idea: use duality.
Lines configuration ⇒ points configuration.
We denote L∗ = {`∗ | ` ∈ L}.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 17 / 41

Observation

Assume ` appears as a segment pq in the upper envelope.

` p q

Interpretation in the dual space?

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 18 / 41

Observation

p and q are on or above all the lines in L.

So p ∗ and q ∗ are on or below all the points in L∗ .

L∗

p∗

q∗
`∗

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 19 / 41

Observation

So p ∗ and q ∗ are on the lower part of CH(L∗ ).

`∗ = p ∗ ∩ q ∗ is also on the lower part of CH(L∗ ).
We denote this lower part by LH(L∗ ).

CH(L∗ )

p∗

q∗
`∗

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 20 / 41

Consequences

The lines that appear in the upper envelope of L correspond to the points
that appear in the lower hull of L∗ . How to compute the upper envelope?
Compute the lower hull LH(L∗ ).
Traverse this chain from left to right, output the dual of the vertices.
This gives you a list of lines of L.
These are the lines that appear in the upper envelope.
They are in the same order as they appear in this upper envelope,
from left to right.
I Why?

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 21 / 41

Consequences

We use the algorithm from Lecture 2.

It takes O(n log n) time.
So we can compute an upper envelope of lines in O(n log n) time.
So we can compute the intersection of n halfplanes in O(n log n)
time.
I Split into two sets of halfplanes: those that go upward and those that
go downward.
I The intersection of the upward halfplanes is an upper envelope of lines.
I The other subset is a lower envelope. (Similar idea.)
I Intersect these two chains. (Plane sweep for instance.)
I Overall, it takes O(n log n) time.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 22 / 41

Remarks

We have just seen that, in the plane, the following three problems are
equivalent:
I Convex hull of a point set.
I Upper (lower) envelope of lines.
I Halfspace intersection.
In higher dimension, it is similar.
I But the intersection of n half-spaces is a polytope that can have
Ω(nbd/2c ) vertices.
I Voronoi diagrams and Delaunay triangulations can be seen as upper
envelopes in one dimension higher.
These problems, that are all related, are fundamental problems in
computational geometry.
RIC works well for these problems.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 23 / 41

Arrangements of Lines

L: Set of n lines

A cell

A Vertex
An edge

Arrangement A(L)

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 24 / 41

Definition

Definition (Arrangement of lines)

Let L be a set of n lines in R2 . These lines subdivide R2 into several
regions, called cells. The edges of this subdivision are line segments or
half-lines. The vertices are intersection points between two lines of L. This
subdivision, with adjacency relation between vertices, edges and cells, is
called the arrangement A(L) of L.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 25 / 41

Bounding Box

We restrict our attention to a bounding box B that contains all the

vertices of A(L).
B

Now all edges and faces are bounded.

How fast can we compute such a bounding box?

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 26 / 41

General Position Assumptions

No two lines are parallel.

No three lines intersect at one point.

A degenerate case:
- Two lines are parallel.
- Three lines intersect at one point.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 27 / 41

Combinatorial Complexity

Definition (Combinatorial complexity)

The combinatorial complexity of an arrangement A(L) is the total number
of vertices, edges, and faces in A(L).

This quantity is Θ(n2 ) for an arrangement of n lines.

More precisely, if L is in general position.
A(L) has n2 vertices.


I
I A(L) has n2
edges.
I A(L) has n2 + n + 1 faces.
Proof?

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 28 / 41

Zone
Definition (Zone)
The zone of a line ` ∈
/ L is the set of cells in A(L) that intersect `.

zone of `

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 29 / 41

Zone Theorem

Theorem (Zone Theorem)

The total number of edges of all the cells in the zone of a line ` ∈
/ L in
A(L) is O(n).

In other words: the combinatorial complexity of the zone of a line is

linear.
Why is it not obvious?
Some lines of L appear in several cells of the zone; See previous slide.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 30 / 41

Proof
Definition
Left-bounding and right-bounding edges are edges that bound a cell of the
zone of L from the left and from the right, respectively.

Left-bounding Right-bounding
edge edge

left and right

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 31 / 41

Proof

We insert lines of L from left to right along `.

Inserting a line increases by at most 3 the number of left bounding
edges.
It follows that the zone has at most 3n left bounding edges.
So the zone of ` has at most 6n edges.
See picture next slide, and detailed proof in D. Mount’s notes.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 32 / 41

Proof

split

new

split

`n

Inserting `n : Two left bounding edges are split and one is created.
Total: +3 left bounding edges.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 33 / 41

Constructing an Arrangement

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 34 / 41

Constructing an arrangement

Incremental algorithm.
We insert the lines one by one and update the arrangement.
Arrangement maintained in a Doubly Connected Edge List.

Insertion of a new line `.

Find the leftmost point of ` in the bounding box and the cell that
contains it.
It takes O(n) time.
Traverse the zone of ` from left to right and update the arrangement
accordingly.
By the Zone Theorem, can be done in O(n) time.

Overall, we compute a DCEL of A(L) in O(n2 ) time.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 35 / 41

An application of arrangements and duality

Problem (Smallest triangle)

Given a set P of n points in R2 , find the triangle with smallest area whose
vertices are in P.

We assume general position: No three points are collinear.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 36 / 41

Characterization

Let (a, b) ∈ P 2
How can we find c ∈ P such that area of a, b, c is minimized?
Find the largest empty corridor along line ab.
`c

b
c

c lies on its boundary.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 37 / 41

Characterization

c lies on a line `c such that:

`c is parallel to ab,
and there is no other line with same slope between ab and `c .

What does it mean in the dual plane?

We call (ab)∗ the dual of line ab.
(`c )∗ is on c ∗ .
(`c )∗ and (ab)∗ have same abscissa.
No line p ∗ where p ∈ P crosses the line segment with endpoints (ab)∗
and (`c )∗ .

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 38 / 41

Characterization
c∗
a∗

(ab)∗

b∗

(`c )∗

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 39 / 41

Consequences

(`c )∗ is in the same cell of A(P ∗ ) as (ab)∗ .

(`c )∗ is vertically above or below (ab)∗ .
Once A(P ∗ ) is computed, only two candidates involving a and b.
We compute A(P ∗ ) in O(n2 ) time.
For all cell of this arrangement, we compute by plane sweep the point
of the boundary that is vertically above or below every vertex of the
cell.
It takes time linear in the number of edges of the cell as they are
given in counterclockwise order.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 40 / 41

Algorithm

For each triple a∗ , b ∗ , c ∗ we found, we can compute the area of

triangle a, b, c in O(1) time.
We maintain the minimum value found so far in O(1) time by triangle
we consider.
Overall, the running time of our algorithm is proportional to the
combinatorial complexity of A(P ∗ ).
So it runs in O(n2 ) time.

Antoine Vigneron (KAUST) CS 372 Lecture 13 November 28, 2012 41 / 41