APPROXIMATE ALGORITHMS

A Project Report Submitted

for the Course

MA498 Project I

by
Pranavi Sudireddy
(Roll No. 210123045)
Adapa Jashkayth
(Roll No. 210123005)

to the

DEPARTMENT OF MATHEMATICS
INDIAN INSTITUTE OF TECHNOLOGY GUWAHATI
GUWAHATI - 781039, INDIA

November 2024
 CERTIFICATE

This is to certify that the work contained in this project report entitled “Ap-
proximate Algorithms” submitted by Pranavi Sudireddy (Roll No.: 210123045)
and Adapa Jashkayth (Roll No.: 210123005) to the Department of Mathe-
matics, Indian Institute of Technology Guwahati towards partial requirement
of Bachelor of Technology in Mathematics and Computing has been carried
out by him/her under my supervision.

It is also certified that this report is a survey work based on the references
in the bibliography.

OR

It is also certified that, along with literature survey, a few new results are es-
tablished/computational implementations have been carried out/simulation
studies have been carried out/empirical analysis has been done by the stu-
dent under the project.

Turnitin Similarity: %

Guwahati - 781 039 (Dr./Prof. Gautam Kumar Das)

November 2024 Project Supervisor

ii
 ABSTRACT

The main aim of the projeing up the courage to ask. wondering if they’re
as afraid as you, and hoping that they’ll understand you, that their fears will
fade away. it’s a song about yearning for them to accept you, because you’ll
give them heaven, you’ll treat them right and never break their heart.ct is
studying computational geometry and network design, focusing on wireless
networks and minimum spanning trees with directional constraints.
The first study : Connectivity Guarantees for Wireless Networks with Di-
rectional Antennas, addresses the challenge of ensuring network connectivity
using directional antennas. It finds that a minimum antenna angle of 60° is
sufficient for full connectivity among any arrangement of nodes in a plane.
Bounded-Angle Minimum Spanning Trees Trees study focuses on con-
structing minimum spanning trees (MSTs) with edges restricted to specific
angles, commonly 120°. Improving on prior work, the authors refine the ap-
16
proximation factor for the 120°-MST problem from 6 to 3
and introduce an
enhanced algorithm that better controls edge length and angle constraints.
This work provides insights into efficiently constructing MSTs with direc-
tional limits, advancing the theoretical understanding of MST approximation
with angle constraints.
Combined, these studies provide critical algorithms and proofs for enhanc-
ing network efficiency through optimized connectivity and low-interference
configurations in wireless networks and constrained spanning trees.

iii
Contents

List of Figures v

List of Tables vi

1 α Spanning Trees 1
1.1 60-Degree Directional Antennas . . . . . . . . . . . . . . . . . 1
1.1.1 Stage I . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Stage II . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Approximating the 120°-MST . . . . . . . . . . . . . . . . . . 6
1.2.1 6-Approximation of 120◦ -MST . . . . . . . . . . . . . . 7
1.2.2 16
3
-Approximation of 120◦ -MST . . . . . . . . . . . . . 9
1.3 Approximating 90-Degree MST
for Points on a Circle . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 2-Approximation of 90°-MST . . . . . . . . . . . . . . 18

Bibliography 22

iv
List of Figures

1.1 Example of Non-Connectivity with α < 60◦ . . . . . . . . . . . 2

1.2 Connecting all the destined points with a 60◦ wedge. . . . . . 3
1.3 Atleast one of the traingles is good . . . . . . . . . . . . . . . 5
1.4 Top: path H where H0 , H1 , H2 are colored blue, green, and red,
respectively. Bottom: spanning tree T where edges obtained
by Theorems 1.2.1 and 1.2.2 are colored purple and orange,
respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Two configurations of a forbidden pair {P1 , P2 } forming tri-
angles K1 and K2 . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Angle Subtended by Three Points on a Semicircle . . . . . . . 16
1.7 Example of Connectivity for α = 90 . . . . . . . . . . . . . . . 20

v
List of Tables

vi
Chapter 1

α Spanning Trees

1.1 60-Degree Directional Antennas

Definition 1.1.1. An α-spanning tree is a network structure that ensures
connectivity between nodes with directional antennas. In wireless networks
with such antennas, each node’s coverage area is limited to a wedge with an
adjustable direction and angle α. Two nodes can directly communicate only
if each lies within the other’s coverage wedge.
The α-spanning tree is crucial because it connects all nodes while min-
imizing communication links, interference, and energy usage. By carefully
setting the direction and angle of each node’s antenna, an α-spanning tree
helps create an efficient network, supporting reliable connectivity with opti-
mized resource use.

Theorem 1.1.2. (Paz Carmi and Matthew J. Katz) There exists an α-

spanning tree with α = 60 degrees for any set of points P in the plane.
Specifically, it is possible to position a 60-degree directional antenna at each
point p ∈ P such that the resulting communication graph is connected.

1
Proof. To support this result, let P represent a set of points in the plane. We
aim to show that it is possible to position a 60-degree directional antenna at
each point p ∈ P , resulting in a connected communication graph.
Our proof proceeds in two stages. In the first stage, we position the
antennas such that the resulting communication graph may initially consist
of multiple disconnected components. In the second stage, we adjust these
antenna placements, if necessary, to achieve a fully connected graph.

Figure 1.1: Example of Non-Connectivity with α < 60◦

Before we delve into the proof of this theorem, it is important to highlight

that 60 degrees represents the best achievable angle for ensuring connectiv-
ity in the context of directional antennas. There are configurations of points
where it becomes impossible to establish a connected communication graph
using α-degree antennas when the angle α is less than 60◦ . Figure 1.7 illus-
trates such a scenario.

1.1.1 Stage I

For each point p ∈ P , let f (p) represent the point in P that is farthest from
p. We define Fp = {q ∈ P | f (q) = p} as the set of points in P for which p is
the farthest point. In this stage, we will construct an initial communication
graph by placing a 60-degree directional antenna at each p ∈ P , ensuring

2
that all points in the set {f (p)} ∪ Fp are covered by this antenna. We will
subsequently demonstrate that this coverage is achievable.

(a) The set Fp is contained in the grey (b) ∠pa pz = 60◦

region.

Figure 1.2: Connecting all the destined points with a 60◦ wedge.

Let p ∈ P . For simplicity, we can place f (p) on the horizontal line

through p, specifically to the left of p, as depicted in Figure 2(a). For every
point q ∈ Fp , the following conditions hold: (i) the distance d(q, p) is greater
than or equal to d(q, f (p)), where d(a, b) represents the Euclidean distance
between points a and b; and (ii) the distance d(p, q) is less than or equal to
d(p, f (p)). From these observations, we conclude that Fp is situated within
the region shown in Figure 2(a).
Now, we define pa as the point in Fp located below the line segment pf (p)
that maximizes the angle ∠pa pf (p). Similarly, let pb be the point in Fp
situated above the line segment pf (p) that maximizes the angle ∠f (p)ppb .
In cases where no point in Fp exists below (or above) the line segment pf (p),
we assign pa = f (p) (or pb = f (p), respectively).

3
1.1.2 Stage II

Consider the communication graph obtained in the first stage, and

let C1 , C2 , . . . , Ck be its connected components. If k = 1, then we are done.
Assume therefore that k ≥ 2. For i = 1, . . . , k, we need to ensure that we
can readjust all these wedges so that the graph becomes connected.

Theorem 1.1.3. Consider C and C ′ as two distinct connected components

within the communication graph formed in the initial stage. Let e = (x, y)
and e′ = (u, v) represent the edges associated with these components. In this
configuration, e and e′ intersect each other.

Proof. Suppose, for example, that u lies in the upper half of lune(x, y). If
v also falls within this upper half, then at least one of the four distances
d(u, x), d(u, y), d(v, x), or d(v, y) must exceed d(u, v). However, this is not
possible, given that f (u) = v and f (v) = u.
An angle α is termed good if α ≤ 60◦ . Let C and C ′ be two connected
components, and consider their representative edges e = (x, y) and e′ =
(u, v). Assume, for example, that u is positioned above e and v below e.
Consider the quadrilateral formed by x, v, y, and u; see Figure 1.3. The
edges e and e′ divide this quadrilateral into four triangles. The following
theorem 1.1.4 establishes that at least one of these triangles is good, in that
both of its angles opposite e and e′ are good angles.

Theorem 1.1.4. Refer to Figure 1.3, and let o represent the intersection
point between e and e′ . Then, at least one of the four triangles that have o
as a vertex is a good triangle.

4
 Figure 1.3: Atleast one of the traingles is good

Proof. Assume, for the sake of contradiction, that all four triangles are bad.
Consider the triangle △uxo and assume, for instance, that ∠oux (denoted as
α1 in Figure 3) is bad. Consequently, ∠ovx (denoted as β1 ) must be good.
To see why, examine the triangle △uxv. Since f (v) = u, the angle opposite
the edge uv must be greater than α1 (which we assumed to be bad), so β1
must be good.
Now, given that β1 is good, α2 must be bad (otherwise, we would be
done). This reasoning implies that β2 is good, and so forth. Ultimately, we
reach the conclusion that α1 , α2 , α3 , α4 are all bad, while β1 , β2 , β3 , β4 are all
good.
Next, we demonstrate that this scenario is impossible. Since α1 > β4 ,
we find that |ox| > |ou|. Similarly, |ou| > |oy|, |oy| > |ov|, and |ov| > |ox|.
By chaining these inequalities, we obtain |ox| > |ox|, which is clearly a
contradiction. We conclude that there must exist at least one good triangle.

Therefore, we conclude that two disconnected components can indeed be

connected by readjusting the wedges, while still meeting the angle threshold
requirement. By Theorem 1.1.4, we know that at least one of the triangles

5
will satisfy the ”good” angle condition. Following this approach, we can
iteratively adjust the initial wedge placements from Stage I, modifying them
as needed to achieve connectivity among all components in a similar manner.
Therefore, we can adjust the initial placement of the wedges to connect all
the disconnected components using a 60◦ wedge. Since we have established
that it is impossible to form a connected communication graph with α-degree
antennas when α < 60◦ , we can confidently conclude that 60◦ is the minimum
angle for which there exists a valid configuration of wedge placements that
guarantees the formation of a connected component.

1.2 Approximating the 120°-MST

Aschner and Katz demonstrated the NP-hardness of α-MST problem and
proposed a 6-approximation algorithm that constructs an α-ST with a length
no more than 6-times that of the minimum spanning tree (MST). Later using
16
similar approach this ratio is improved to 3
.
We achieve the following results for an angle α = 120◦ ; however, our
approximation ratios are valid for any angle α ≥ 120◦ .
Aschner and Katz derive their result by employing what they describe
as a “surprising theorem,” established through extensive case analysis. This
theorem ensures that it is possible to assign 120° wedges to any triplet such
that (1) the union of the three wedges fully covers the plane, (2) the resulting
graph for each triplet is connected, and (3) resulting graph on any pair of
triplets (with independently assigned wedges) also forms a connected graph.
Their method begins by constructing an MST of the points, doubling its
edges, and taking shortcuts to create a Hamiltonian path with a length no

6
more than twice that of the MST, following the standard Travelling Sales-
man Problem approximation technique. They then apply their theorem to
consecutive triplets of points along this path, which guarantees that each
pair of successive triplets remains connected. Ultimately, they achieve an
approximation factor of 6 by using the triangle inequality. However, a lim-
itation of their approach is that the two edges chosen for each triplet could
be the longest and second-longest edges within that triplet.

Theorem 1.2.1. (Aschner and Katz, 2014). Given a set P of three points
in the plane, one can place at each point of P a wedge of angle 120◦ such
that the three wedges cover the plane and the induced mutual-visibility graph
on P is connected, and hence it contains a 120◦ -ST.

Theorem 1.2.2. (Aschner and Katz, 2014). Let P1 and P2 be two disjoint
sets each containing three points in the plane. Assume that a wedge of angle
120◦ is placed at each point of P1 and at each point of P2 according to the
placement of Theorem 6. Then, the induced mutual visibility graph on P1 ∪P2
is connected, and hence it contains a 120◦ -ST.

1.2.1 6-Approximation of 120◦ -MST

Let H be a Hamiltonian path on P with length at most twice the MST length.
There exists a point set for which any Hamiltonian path has a length of at
least 2 − ϵ times the MST length. Let (p0 , . . . , pn−1 ) represent the sequence
of points in P from one endpoint of H to the other, and let (h0 , h1 , . . . , hn−2 )
denote the sequence of edges in H, where hi = (pi , pi+1 ).
Next, we partition the edges of H into three sets:

H0 = {h0 , h3 , h6 , . . . }, H1 = {h1 , h4 , h7 , . . . }, H2 = {h2 , h5 , h8 , . . . },

7
as shown in Figure 1.4. Since one of these sets, say H2 , must have a total
w(H)
length of at least 3
, it follows that w(H0 ) + w(H1 ) ≤ 32 w(H).
Now, partition P into a sequence of triplets (p0 , p1 , p2 ), (p3 , p4 , p5 ), . . . such
that the edges in H that connect consecutive triplets are contained in H2 .
For each triplet, place three wedges at the points according to Theorem 1.2.1
and let Gα be the resulting mutual-visibility graph. By Theorem 1.2.1, the
points within each triplet are connected, and by Theorem 1.2.2, any two
adjacent triplets are also connected, ensuring that Gα is fully connected.
Now, construct a spanning tree T of Gα as follows (see Figure 1.4): for
each triplet, include two edges as prescribed by Theorem 1.2.1 (these are
called inner edges), and for each pair of consecutive triplets, include one
edge as given by Theorem 1.2.2 (these are called connecting edges).
To bound the length of T , we charge edges of H for the edges in T . Every
edge of H in H2 lies between two consecutive triplets, so each of these edges
is charged only once for the corresponding connecting edge between triplets.
Each edge in H0 ∪ H1 is contained within a triplet, say t, and is therefore
charged four times: twice for the two inner edges within t and twice for the
two connecting edges linking t to adjacent triplets.
Thus, we have:

w(T ) ≤ w(H2 ) + 4(w(H0 ) + w(H1 ))

= w(H) + 3(w(H0 ) + w(H1 ))

≤ 3w(H) ≤ 6w(MST).

8
Figure 1.4: Top: path H where H0 , H1 , H2 are colored blue, green, and red,
respectively. Bottom: spanning tree T where edges obtained by Theorems
1.2.1 and 1.2.2 are colored purple and orange, respectively.

1.2.2 16
3 -Approximation of 120◦ -MST

Theorem 1.2.3. Given a set P = {p1 , p2 , p3 } of three points in the plane

and three angles α1 , α2 , α3 ≤ π where α1 + α2 + α3 = 2π, one can place at
each pi a wedge of angle αi such that the three wedges cover the plane, and
the induced mutual-visibility graph contains the shortest or the intermediate
tree on P .

Setting each αi to 120◦ in Theorem 1.2.3 yields the following corollary.

Corollary 1.2.4. Given a set P of three points in the plane, one can place
wedges of angle 120◦ at points of P such that the three wedges cover the
plane, and the induced mutual-visibility graph contains the shortest or the
intermediate tree on P .

16
The proof of the 3
-approximation relies on the observation that the con-
figurations in Figure 1.5 do not arise from triplets of a non-crossing Hamil-
tonian path. Specifically, consider two disjoint point sets P1 and P2 , each

9
containing three points in the plane. Let K1 and K2 represent the complete
graphs (triangles) on the points of P1 and P2 , respectively. The (unordered)
pair {P1 , P2 } is termed forbidden if there exist vertices p ∈ K1 and q ∈ K2
such that either (i) both edges incident to p intersect both edges incident to
q, or (ii) p lies inside the triangle K2 and the edge of K1 not incident to p
intersects both edges incident to q.

Figure 1.5: Two configurations of a forbidden pair {P1 , P2 } forming triangles

K1 and K2 .

Theorem 1.2.5. Let P1 and P2 be two disjoint sets, each containing three
points in the plane, such that {P1 , P2 } is not forbidden. Assume that a wedge
of angle 120◦ is placed at each point of P1 and at each point of P2 according
to the placement algorithm of Corollary 1.2.4. Then, the induced mutual-
visibility graph on P1 ∪ P2 is connected, and hence it contains a 120◦ -ST.

Observation 1. Let H be a non-crossing Hamiltonian path, and let t1

and t2 be disjoint triplets, each of which is obtained by consecutive vertices
of H. Then, {t1 , t2 } is not forbidden.
Proof. Since each of t1 and t2 contains two edges of H (due to being
formed from consecutive vertices of H), the union of these edges is non-
crossing because H is a non-crossing Hamiltonian path. However, if {t1 , t2 }
is forbidden (as shown in Figure 1.5), it is impossible to select two edges from
each of t1 and t2 such that their union remains non-crossing.

Lemma 1.2.6. Let a, b, and c be three points in the plane, and let E =

10
{ab, ac, bc} be the set of edges between them. Then the total length of the
shortest and longest edges in E is at most 1.5 times the total length of any
two edges of E.

Theorem 1.2.7. Given a set of points in the plane and an angle α = 120◦ ,
16
there is an α-spanning tree of length at most 3
times the length of the MST.
Furthermore, there is a polynomial-time algorithm to find such an α-ST, thus
16
providing a 3
-approximation algorithm for the α-MST problem.

Proof. To use this observation, we first need a non-crossing Hamiltonian

path. This can be obtained by iteratively flipping crossing edges of H. It is
known that the iterative process terminates after O(n3 ) edge flips, where n
is the number of vertices in the path. Since the edge-flip operation does not
increase the total edge length (this can be verified by the triangle inequality),
the length of the resulting non-crossing path will not exceed that of the orig-
inal path. Hence, we assume that H is non-crossing and w(H) ≤ 2w(M ST ).
With the necessary tools in place (a non-crossing path H and Theorem
1.2.5), we now proceed by orienting the wedges according to Corollary 1.2.4.
Let (p0 , p1 , . . . , pn−1 ) be the sequence of points from one endpoint of H to the
w(H)
other. We construct the edge sets H0 , H1 , H2 , and assume that w(H2 ) ≤ 3
,
which implies w(H0 ) + w(H1 ) ≤ 32 w(H). We then construct the triplets t0 =
(p0 , p1 , p2 ), t1 = (p3 , p4 , p5 ), . . ., and orient the wedges according to Corollary
1.2.4.
Since H is non-crossing, no pair of triplets is forbidden (by Observation
1). The induced mutual-visibility graph, Gα , is connected because the points
within each triplet are connected (by Corollary 1.2.4) and every two triplets
are connected (by Theorem 1.2.5). We obtain a spanning tree T from Gα as
before.
Each edge of H2 is charged only once for the connecting edge between

11
two consecutive triplets. Each pair (hi , hi+1 ) of edges in each triplet t (where
hi ∈ H0 and hi+1 ∈ H1 ) is charged 3.5 times: twice for connecting edges
between t and its adjacent triplets, and 1.5 times for the inner edges of t
(since, by Lemma 1.2.6, the length of the tree obtained by Corollary 1.2.4 is
at most 1.5 times (w(hi ) + w(hi+1 ))).
Therefore, we have:

w(T ) ≤ w(H2 ) + 3.5(w(H0 ) + w(H1 ))

= w(H) + 2.5(w(H0 ) + w(H1 ))

5
≤ w(H) + w(H)
3
8
= w(H)
3
16
≤ w(MST).
3

Christofides’ Algorithm

Christofides’ algorithm is a well-known approximation algorithm for the

Traveling Salesman Problem (TSP) in graphs with non-negative edge weights
that satisfy the triangle inequality. The algorithm constructs a tour that vis-
its each vertex once, with a total weight no more than 1.5 times the optimal
tour weight. The algorithm follows these steps:

1. Find a Minimum Spanning Tree (MST): Compute a Minimum Span-

ning Tree T of the graph G, connecting all vertices with the minimum
total edge weight.

2. Identify Odd-Degree Vertices: In the MST T , find all vertices with an

odd degree (i.e., vertices connected by an odd number of edges). An

12
 MST will always have an even number of such vertices.

3. Find a Minimum Weight Perfect Matching: On the subgraph con-

taining only the odd-degree vertices, find a minimum weight perfect
matching, which pairs each odd-degree vertex to minimize the total
edge weight in the matching.

4. Combine MST and Matching Edges: Add the edges from the perfect
matching to the MST T , resulting in an Eulerian graph (where every
vertex has an even degree).

5. fForm an Eulerian Tour and Shortcut: Traverse the Eulerian circuit

(visiting each edge exactly once), and shortcut repeated vertices to
create a Hamiltonian circuit (visiting each vertex exactly once). By
the triangle inequality, this does not increase the tour length.

Christofides’ algorithm guarantees a solution with a total length at most

1.5 times that of the optimal TSP tour, making it efficient for metric TSP
instances where the triangle inequality applies.

4-Approximation of 120◦ -MST

Result: Given a set of points in the plane and an angle α = 120◦ , there
exists an α-spanning tree of length at most 4 times the length of the MST by
applying a modified version of Theorem 1.2.7 with Christofides’ algorithm.
This approach yields a 4-approximation algorithm for the α-MST problem.
3
To approximate this result using 2
× MST, we aim to leverage a tighter
bound on the Hamiltonian path by ensuring that the minimum-weight perfect
matching also respects the constraints imposed by α for the spanning tree.
This brings several modifications to the original proof in Theorem 1.2.7:

13
1. Original Proof and 2×MST Approximation: In the original proof
of Theorem 1.2.7, a Hamiltonian path H was used with an approxima-
tion w(H) ≤ 2 w(MST). Here, by applying Christofides’ algorithm, we
3
can improve this bound to w(H) ≤ 2
w(MST), which is achieved by
combining a minimum spanning tree and a minimum-weight match-
ing on the odd-degree vertices. By breaking this Hamiltonian cycle
at any point, we obtain a Hamiltonian path that still respects this 23 -
approximation, allowing for a more efficient bound in the final steps of
the proof.

2. Ensuring Compliance with α-Constraints: For this tighter bound

to apply, we must ensure that the subgraph formed by the minimum-
weight matching aligns with the α constraints required by the spanning
tree. This means verifying that the matching edges respect the criteria
associated with α. We are still focusing on finding a subgraph that
not only minimizes weight but also satisfies these α-constraints. This
3
alignment is crucial to justify applying the 2
× MST approximation
within the constraints of the problem.

3. Applying the 32 -Factor Approximation: Once we have a subgraph

that satisfies both the minimum-weight and α-constraints, we can use
it to approximate the Hamiltonian path with w(H) ≤ 32 w(MST), tight-
ening our bound.

4. Revisiting Theorem 1.2.7 with the Tighter Bound: Substituting

3
this improved bound w(H) ≤ 2
w(MST) into the steps of Theorem
1.2.7, we proceed as follows:

14
 w(T ) ≤ w(H2 ) + 3.5(w(H0 ) + w(H1 ))

= w(H) + 2.5(w(H0 ) + w(H1 ))

5
≤ w(H) + w(H).
3

5. Extending to a 4-MST Approximation: By substituting w(H) ≤

3
2
w(MST) in the final step, we calculate:

8 3
w(T ) ≤ · w(MST)
3 2
= 4 w(MST).

Thus, using Christofides’ algorithm and focusing on finding a subgraph

that meets both the minimum-weight and α-spanning tree criteria, we achieve
16
an overall 4-approximation, reducing the bound from 3
to 4 in the final
approximation.

1.3 Approximating 90-Degree MST

for Points on a Circle

1.3.1 Model Setup

In this section, we define the setup for our model. Until now, we have dis-
cussed the minimum angle required to guarantee the existence of a spanning
tree for any set of points. Additionally, we have explored how to adjust
wedges to minimize the Minimum Spanning Tree (MST) for a 120-degree
spanning tree configuration. In this new focus, we turn our attention to a
special case: points that are aligned on a circle. Our objective is to approxi-
mate the 90-degree Minimum Spanning Tree in this specific scenario.

15
Theorem 1.3.1. When three points A, B, and C lie on the circumference of
a semicircle, with the points in the order A, B, and C, the angle subtended
by the arc ABC will be greater than or equal to 90 degrees.

Figure 1.6: Angle Subtended by Three Points on a Semicircle

Proof. Consider three points A, B, and C that lie on the circumference of

a semicircle, with these points ordered as A, B, and C along the arc of the
semicircle.

Since A, B, and C are positioned on a semicircle, they subtend an angle

at the center O of the circle. Let the angle subtended by the arc ABC at
the center O be denoted by ∠AOC. Since the points lie on the arc in that
order, we are considering the larger arc that B lies on.

16
 By the properties of circles, the angle at the center O is twice the angle
subtended on the circumference by the same arc. Therefore, we have:

∠AOC = 2 · ∠ABC.

Since A, B, and C lie on a semicircle, the arc ABC spans more than half
of the circle, making ∠AOC greater than 180◦ .

Thus, we have:

∠AOC 180◦
∠ABC = ≥ ≥ 90◦ .
2 2

This shows that the angle ∠ABC, subtended by the arc ABC, is indeed
greater than 90◦ , as required.

Theorem 1.3.2. Given a set of more than three distinct points on the cir-
cumference of a circle, for any point in this set, there exists a semicircle
containing that point and at least two additional points from the set.

Proof. Let there be n ≥ 4 distinct points on the circumference of a circle.

Select any point P from this set of points.

Consider dividing the circle into two semicircles with endpoints at P . One
semicircle will contain the clockwise points starting from P , and the other
will contain the counterclockwise points from P .

Since n ≥ 4, there are at least three other points on the circle besides P .
By the pigeonhole principle, when we divide these remaining points between
the two semicircles, at least one of the semicircles must contain at least:
 
n−1
2

17
points in addition to P .

For n = 4, this calculation yields:

 
4−1
= 2.
2

Thus, at least one of the semicircles will contain P and at least two other
points. The same argument holds for any n > 4, guaranteeing that a semi-
circle with P and at least two additional points exists.

This completes the proof by showing that for each point P , there is always
a semicircle containing P and at least two other points from the set.

1.3.2 2-Approximation of 90°-MST

Theorem 1.3.3. Let S = {P1 , P2 , . . . , Pn } be a set of n ≥ 3 distinct points

arranged in clockwise order along the circumference of a circle. Define a
polygon P by joining each pair of consecutive points along the circle, specifi-
cally P1 → P2 → · · · → Pn → P1 . The optimal tour length for the Traveling
Salesman Problem (TSP) on S is given by:

n
X
L= d(Pi , Pi+1 ) − max d(Pi , Pi+1 ),
1≤i≤n
i=1

where d(Pi , Pi+1 ) denotes the Euclidean distance between consecutive points
Pi and Pi+1 (with Pn+1 ≡ P1 ), ni=1 d(Pi , Pi+1 ) is the perimeter of the polygon
P

P, and max1≤i≤n d(Pi , Pi+1 ) represents the longest side of P.

Proof. To solve the TSP for points arranged on a circle, we seek the shortest
path that connects all points and returns to the starting point.

18
 The path that minimizes the travel distance is the perimeter of the poly-
gon P formed by traversing each side d(Pi , Pi+1 ) once, but skipping the
longest side to avoid the largest detour on the circle.

Thus, the perimeter of P is:

n
X
Perimeter = d(Pi , Pi+1 ).
i=1

By removing the longest side, max1≤i≤n d(Pi , Pi+1 ), we form the shortest
path that still visits each point exactly once and returns to the starting point,
yielding the optimal TSP length:

n
X
L= d(Pi , Pi+1 ) − max d(Pi , Pi+1 ).
1≤i≤n
i=1

This completes the proof by showing that the minimal tour length is
achieved by taking the perimeter of P minus its longest side.

Theorem 1.3.4. For any three points A, B, and C lying on the circumfer-
ence of a semicircle in order, we can form a connected tree with an angle
bound α = 90◦ by joining A to B and then B to C.

Proof. From Theorem 1.3.1, we know that for any three points A, B, and
C lying in order on the circumference of a semicircle, the angle ∠ABC is
always greater than or equal to 90◦ .
−→ −−→
If we join the points by vectors AB and BC, the angle between these
vectors is 180◦ − ∠ABC. Since ∠ABC ≥ 90◦ , it follows that

180◦ − ∠ABC ≤ 90◦ .

19
 Figure 1.7: Example of Connectivity for α = 90

−→ −−→
Thus, the angle between AB and BC is less than or equal to 90◦ , satisfying
the condition for a connected tree with α = 90◦ .

Theorem 1.3.5. Let S = {P1 , P2 , . . . , Pn } be a set of n ≥ 3 distinct points

arranged in clockwise order along the circumference of a circle. It is possible
to construct a minimum 90◦ -spanning tree for S with a total weight that is
at most twice the weight of the minimum spanning tree (MST) for S.

Proof. By Theorem 1.3.2, we know that for any point Pi in the set S =
{P1 , P2 , . . . , Pn } arranged in clockwise order along the circle, we can form a
90◦ -spanning tree by joining each point to its neighboring points. Specifically,
for each Pi , we can either connect it to its preceding points Pi−1 and Pi−2
or to its succeeding points Pi+1 and Pi+2 , thereby ensuring that the angle
constraint of 90◦ is satisfied according to Theorem 1.3.4.
To approximate the minimum α-spanning tree with an angle constraint
α = 90◦ , we consider constructing a spanning tree by joining consecutive
points P1 , P2 , . . . , Pn along the polygon P. Since we are joining adjacent
points that lie on a semicircle, this construction is somewhat analogous to
the approach in the Travelling Salesman Problem discussed in Theorem 1.3.3.
However, we must account for the possibility that we do not always exclude

20
the longest edge of the polygon, which would increase the tree’s total length.
In the worst case, this 90◦ -spanning tree might include every edge of the
polygon except the shortest one. Thus, the length of this spanning tree can
be expressed as:

n
X
Spanning tree length = d(Pi , Pi+1 ) − min d(Pi , Pi+1 ),
1≤i≤n
i=1

where d(Pi , Pi+1 ) represents the distance between consecutive points.

On the other hand, by Theorem 1.3.3, the length of the minimum span-
ning tree (MST) for S would be:

n
X
L= d(Pi , Pi+1 ) − max d(Pi , Pi+1 ).
1≤i≤n
i=1

Since the smallest side length in the polygon is strictly positive, we ob-
serve that:
min d(Pi , Pi+1 ) > 0.
1≤i≤n

Additionally, the longest possible side in a simple polygon is at most half the
perimeter, so:
n
1X
max d(Pi , Pi+1 ) ≤ d(Pi , Pi+1 ).
1≤i≤n 2 i=1

Therefore, in the worst case, the 90◦ -spanning tree length is at most twice
the length of the MST. This gives us:

Spanning tree length ≤ 2 · L.

Hence, the length of the 90◦ -spanning tree has an upper bound of 2 × MST,
as required.

21
Bibliography

[1] K. Andrews and B. Rajiv. On some applications of eigenvalues of toeplitz

matrices. Journal of Mathematical Analysis and Applications, 56(2):237–
239, 2007.

[2] C. C. Chang. Algebraic analysis of many valued logics. Transactions of

American Mathematical Society, 88:467–490, 1958.

[3] Brunella Gerla. Automata over MV-algebras. In ISMVL ’04: Proceedings

of the 34th International Symposium on Multiple-Valued Logic, pages 49–
54, Washington, DC, USA, 2004. IEEE Computer Society.

[4] G.H. Golub and C.F. Van Loan. Matrix Computations. Second Edition.
The John Kopkins University Press, 1989.

22