03 Recursion DAC

Computer Science and Engineering| Indian Institute of Technology Kharagpur
cse.iitkgp.ac.in
Algorithms – I (CS21203)
Autumn 2023, IIT Kharagpur
Divide and Conquer, Recursion

cse.iitkgp.ac.in
Divide and Conquer

• General techniques
• Merge sort
• Recursion Tree Method
• Master Theorem
Aug 10,11,16,17,18, 2023 2

CS21203 / Algorithms - I | Divide and Conquer
cse.iitkgp.ac.in
Divide and Conquer

• Divide and conquer involves three steps
• Divide a problem into pieces (smaller instances of same problem)

• E.g., divide into halves
• Conquer the subproblems by solving them recursively
• Can just call the same algorithm on the subproblems (recursively solving)
• Base case: when n=1 (or is small)
• Combine solutions to pieces to get answer to original problem
• Combining results from recursive calls.
• Usually the hardest part in the algorithm design
Aug 10,11,16,17,18, 2023 3

cse.iitkgp.ac.in
Merge Sort
8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
3 8 2 9 1 5 4 7
2 3 8 9 1 4 5 7
1 2 3 4 5 7 8 9
Aug 10,11,16,17,18, 2023 4
cse.iitkgp.ac.in
Merge Sort
• Dividing is easy. The key operation, though, is merge.
• How to do merging of two sorted arrays to produce a merged
sorted array?
2 5 9 3 4 8
Aug 10,11,16,17,18, 2023 5

cse.iitkgp.ac.in
Merge Sort Pseudocode
Aug 10,11,16,17,18, 2023 6

cse.iitkgp.ac.in
Merge Sort – In What Order?

8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
8 3 2 9 5 1 7 4
3 8 2 9 1 5 4 7
2 3 8 9 1 4 5 7
1 2 3 4 5 7 8 9
Aug 10,11,16,17,18, 2023 7
cse.iitkgp.ac.in
Merge Sort – Runtime Analysis

Total number of elements in two arrays is
[Constant time]
Θ ( 𝑛1 )
Θ ( 𝑛2 )
Overall runtime of “MERGE”

Θ ( 1) subroutine is (total number
of elements after merging)
Θ ( 𝑛1 +𝑛 2)
Aug 10,11,16,17,18, 2023 8

cse.iitkgp.ac.in

𝑝 𝑛 𝑟
𝑛
Θ ( 1) 𝑝 𝑛
2 𝑟𝑝 𝑛
2 𝑟
𝑛
2∗
2
Θ ( 1)
𝑝𝑛
4𝑟𝑝𝑛
4𝑟
𝑝𝑛
4𝑟𝑝𝑛
4𝑟
𝑛
4∗
4
Θ ( 𝑛)
𝑝=𝑟
𝑝=𝑟 𝑝=𝑟𝑝=𝑟
𝑝 =𝑟 𝑝=𝑟
𝑝 =𝑟
𝑝=𝑟
𝑛𝑛 𝑛𝑛 𝑛 𝑛 𝑛𝑛
44 4 4 4 4 4 4
• How do we count the recursive calls?

• Recursive calls belong to the next level
Aug 10,11,16,17,18, 2023 9

cse.iitkgp.ac.in

• What are the number of levels in mergesort?
• There are levels and at each level the number of operations is

• The runtime of mergesort is
• Next we will learn how to write an equation that represents the

running time of a divide and conquer algorithm.
Aug 10,11,16,17,18, 2023 10

cse.iitkgp.ac.in
Recurrence Relation
• Let be the running time of the
algorithm when the input size is
• Base case takes constant time
Base case
Divide time = const
Conquer time = each
Combine time =
• Recurrence relation: Represents the

running time of a recursive algorithm
Aug 10,11,16,17,18, 2023 11

cse.iitkgp.ac.in
Solving Recurrences – Recursion Tree
• We can write as some const.*

• Does the base case need to be always for ?
• General form:
Aug 10,11,16,17,18, 2023 12

cse.iitkgp.ac.in

{
𝑑 𝑛=1
𝑇 ( 𝑛 )=
• Left hand side is . This is a node
2𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1 • It expands to 2, nodes and one node
• The convention is to replace with these nodes
• And continue
𝑇𝑐𝑛
(𝑛)
𝑇 (𝑛)
𝑇 (
𝑛
2
) 𝑇 (
𝑛
2
) 𝑇 (𝑐𝑛
𝑛
2
)
𝑛 𝑛 𝑐𝑛 𝑛 𝑛 𝑐𝑛
𝑇 ( ) 𝑇 ( ) 𝑇 ( )𝑇 ( )
4 4 2 4 4 2
Aug 10,11,16,17,18, 2023 13

cse.iitkgp.ac.in

{
𝑑 𝑛=1
𝑇 ( 𝑛 )=
• Left hand side is . This is a node
2𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1 • It expands to 2, nodes and one node
• The convention is to replace with these nodes
• And continue
𝑐𝑛
The tree ends 𝑇 (𝑛)
when the base 𝑐𝑛 𝑐𝑛
2
2
case is reached
𝑛 𝑛 𝑛 𝑛
𝑇 ( )𝑇 ( 𝑇) ( ) 𝑇 ( )
How many ? 4 4 4 4
Base 𝑑 𝑑 𝑑 𝑑
Aug 10,11,16,17,18, 2023 14

cse.iitkgp.ac.in

𝑐𝑛
𝑐𝑛 𝑐𝑛
2 2
{
𝑑 𝑛=1
𝑇 ( 𝑛 )=
2𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1 𝑐𝑛
4
𝑐𝑛 𝑐𝑛
4 4
𝑐𝑛
4
• Total runtime comes from counting number of operations from the

internal nodes and counting them from base nodes
• Lets get introduced to trees and some related terminologies as we need
• More will come later
Aug 10,11,16,17,18, 2023 15

cse.iitkgp.ac.in

• Graphs are sets of nodes and edges
• Trees are a form of a graph with some specialties
• Trees have direction (parent/child relationships) and don't contain
cycles
• Some terminologies
A Root
B C Internal
D E F G Leaf
• In a Binary Tree, a node can have at most two children
Aug 10,11,16,17,18, 2023 16

cse.iitkgp.ac.in
Some Terminologies
A Parent
Left child Right child

B Siblings C
D E F G
Root
A
Left subtree Right subtree
B C
D E F G
Aug 10,11,16,17,18, 2023 17

cse.iitkgp.ac.in
Some Terminologies Max # of

Nodes
A 0
Level 0 1= 2
1
B C Level 1 2=2
D E F G Level 2 4=2
2
I J K L M N O 3
P Level 3 8=2
• Number of nodes at level
• Number of leaves Number of nodes at level [ is height]
• Naturally height
• Note: Number of Levels = Height (H) + 1
• Total number of nodes,
Aug 10,11,16,17,18, 2023 18

cse.iitkgp.ac.in
Some Terminologies
A
B C
D E F G For general `k’-array Tree
I J K L M N O P
• Number of nodes at level

• Number of leaves Number of nodes at level [ is height]
• Naturally height
• Total number of nodes,
Aug 10,11,16,17,18, 2023 19

cse.iitkgp.ac.in

𝑐𝑛 Level 0
𝑐𝑛 𝑐𝑛
2 2 Level 1
{
𝑑 𝑛=1
𝑇 ( 𝑛 )=
2𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1 𝑐𝑛
4
𝑐𝑛 𝑐𝑛
4 4
𝑐𝑛
4
Level
Base 𝑑 𝑑 𝑑 𝑑 Level
• To get the total runtime, first we need to get the height of the tree
• Number of leaves
• Base cost =
Aug 10,11,16,17,18, 2023 20

cse.iitkgp.ac.in

𝑐𝑛 𝑐𝑛 Level 0
𝑐𝑛 𝑐𝑛
2∗
𝑐𝑛
2
= 𝑐𝑛
2 2 Level 1
𝑐𝑛 𝑐𝑛 𝑐𝑛 𝑐𝑛 𝑐𝑛
4∗ = 𝑐𝑛
4 4 4 4 4
Level
• What about the internal costs?
• We need to sum the costs at each internal node
• Since the costs at each level is same we can multiply by height
• Internal cost =
• Total cost =
Aug 10,11,16,17,18, 2023 21
cse.iitkgp.ac.in

• Lets take another example
{
𝑑 𝑛=1
𝑇 ( 𝑛 )=
4𝑇
𝑛
2
+𝑐𝑛 𝑛>1 ( )
• Does it make sense?
• The general form of the recurrence can be
• Where, is number of subproblems (branching factor)

• is the size of each subproblem (division ratio)
• total time for divide and combine operations
• What are these values for binary search?
• with base case time - constant
Aug 10,11,16,17,18, 2023 22

cse.iitkgp.ac.in

{
𝑑 𝑛=1
𝑇 ( 𝑛 )= c
4𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1
𝑐𝑛 𝑛 𝑐𝑛 𝑛 𝑐𝑛 𝑛 𝑐𝑛 𝑛
𝑇 ( ) 𝑇 ( ) 𝑇 ( ) 𝑇 ( )
22 22 22 2 2
𝑐𝑛 𝑐𝑛 𝑐𝑛 𝑐𝑛
4 4 4 4
• What will be the number of leaves? More than or less than previous?
• What about the height? More than or less than previous?
• Height
• Number of leaves
Aug 10,11,16,17,18, 2023 23

cse.iitkgp.ac.in

{
𝑑 𝑛=1
𝑇 ( 𝑛 )= c Level 0 𝑐𝑛
4𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1
2 2 2 2 Level 1 4
2
=2 𝑐𝑛
4 4 4 4
Level 2 16
4
=4 𝑐𝑛
Level
• So base cost,
• Now the internal cost
• So, number of operations at level is
• Internal cost =
Aug 10,11,16,17,18, 2023 24

cse.iitkgp.ac.in

{
𝑑 𝑛=1
𝑇 ( 𝑛 )= c Level 0 𝑐𝑛
4𝑇
𝑛
2 ( )
+𝑐𝑛 𝑛>1
2 2 2 2 Level 1 4
2
=2 𝑐𝑛
4 4 4 4
Level 2 16
4
=4 𝑐𝑛
Level
• So base cost,
• Internal cost =
• Total cost =
Aug 10,11,16,17,18, 2023 25

cse.iitkgp.ac.in
Solving Recurrences – Master Theorem

• Review of the general procedure 𝑎 𝑓 (𝑛)
• General form:
𝑛 𝑛 𝑛 𝑛
𝑓( )𝑓( )𝑓( )𝑓( )
𝑏 𝑏 𝑏 𝑏
• where is base case runtime

• is number of subproblems (branching factor) Base 𝑑 𝑑 𝑑
• is the size of each subproblem (division ratio)
• total time for divide and combine operations
• Compute Height
• Compute number of leaves
• Compute base cost (Generally - )
• Compute internal cost
• Sum base and internal cost
Aug 10,11,16,17,18, 2023 26
cse.iitkgp.ac.in

• Master theorem provides a consolidated set of rules
• It depends on the relation between the number of leaves and the
function
• Case I: If is polynomially larger i.e., where is some constant , then
• is the difference in the polynomial degree between and
• If and , then
• Similarly, if and , then 1
 0.5 and must be at
• Another way of thinking this is – the ratio between
least a polynomial
• Take and , Then and is less than polynomial
Aug 10,11,16,17,18, 2023 27

cse.iitkgp.ac.in

• Case I: If is polynomially larger i.e., where is some constant , then
• is the difference in the polynomial degree between and
• Case II: If and are asymptotically same, i.e., then
• Case III: If is polynomially larger i.e., then if for
Regularity Condition
Case II: Case I:
Aug 10,11,16,17,18, 2023 28

cse.iitkgp.ac.in
Quicksort
• Quicksort is a sorting algorithm which puts elements in the right
places one by one
• What is a “right place”?
• Is there any element which is at the right place?
3 7 9 5 15 11 4
4 7 9 5 15 11 17
4 7 5 9 15 11 12
Aug 10,11,16,17,18, 2023 29

cse.iitkgp.ac.in
Quicksort
• Lets walkthrough an example to understand how quicksort works
7 6 10 5 9 2 1 15 7
• The crucial step in quicksort is partioning (divide into two parts
such that left of pivot is less and right of pivot id more)
Aug 10,11,16,17,18, 2023 30

cse.iitkgp.ac.in
Quicksort
Aug 10,11,16,17,18, 2023 31

cse.iitkgp.ac.in
Quicksort – Runtime Analysis
Θ ( 1)
Θ ( 𝑛)
Overall runtime of “PARTITION”

subroutine is
Θ ( 1)
Aug 10,11,16,17,18, 2023 32

cse.iitkgp.ac.in
Θ ( 𝑛)
Hard to tell about these.
• Lets take two extreme cases

Partitions are perfectly balanced everytime
• Number of levels =
• Recurrence relation
• Same as mergesort –
• However, this is the best case scenario
Aug 10,11,16,17,18, 2023 33

cse.iitkgp.ac.in
Θ ( 𝑛)
Hard to tell about these.
• Lets take two extreme cases

Partitions are most unbalanced everytime
• Number of levels =
𝐶𝑜𝑛𝑠𝑡 .
• Lets try to solve this recusrsive eqn.
Aug 10,11,16,17,18, 2023 34

cse.iitkgp.ac.in

c • Base case =
• Internal cost at level is
C
• Total internal cost
C
2c • Number of levels =
Base case 𝐶𝑜𝑛𝑠𝑡 .
• Same as insertion sort –

• This is the worst case scenario
Aug 10,11,16,17,18, 2023 35

cse.iitkgp.ac.in

• What should the input look like for the best case?
• The medians are chosen as pivots always
• What should the input look like for the worst case?
• The input array is already sorted
• Probability of getting an already sorted array is
• However, the problem can be tackled if instead of always taking the

first element as pivot, we take any element as pivot in random
• The average-case time complexity of the randomized version of
quicksort is
• The proof of this involves a more complicated recurrence than we
have seen so far.
Aug 10,11,16,17,18, 2023 36

cse.iitkgp.ac.in

• We will follow “Introduction to Algorithms and Data Structures” by
M. J. Dinneen, G. Gimelfarb, M. C. Wilson for the proof
• Let denote the average-case running time for size list
• In the first step, the time taken to compare all the elements with
the pivot is
• Let is the final position of the pivot. So, the two sublists are of size
and respectively
𝑖
𝑖 𝑛 −𝑖 −1
• The recurrence relation for this is
• Small detail: Last term should be as it depends on the exact implementaiton (e.g.,
whether comparison starts from pivot or one element after, etc.)
Aug 10,11,16,17,18, 2023 37

cse.iitkgp.ac.in

𝑖
𝑖 𝑛 −𝑖 −1
• The recurrence relation for this is
• The final pivot position can be any of the positions with equal
probability
• So, the value of to be (i.e., ) has probability
• Similarly, the value of to be (i.e., ) has probability and so on
• So the expected value of is given by,
Aug 10,11,16,17,18, 2023 CS21003/CS21203 / Algorithms - I | Divide and Conquer 38

cse.iitkgp.ac.in
• and so on
𝑖 𝑖
𝑖 𝑛 −𝑖 −1 𝑛 −𝑖 −1 𝑖
• Let us rewrite this as
Aug 10,11,16,17,18, 2023 39

cse.iitkgp.ac.in

• Replace by and subtract from the above
• Divide everything by
• Last term uses partial fraction decomposition
Aug 10,11,16,17,18, 2023 40

cse.iitkgp.ac.in

•
• Lets put
Aug 10,11,16,17,18, 2023 41

cse.iitkgp.ac.in
• Using a result , we can write
• -> This will shown to be true for a special case, next
Aug 10,11,16,17,18, 2023 42

cse.iitkgp.ac.in

• In case, the number of terms in the Harmonic series is a perfect
power of 2,i.e., or ,
𝑘=1
𝑘= 2
𝑘= 3
• All the bracketed sums are 1
• So, as
Aug 10,11,16,17,18, 2023 43

cse.iitkgp.ac.in

• So, … (1)
• Again, for or ,
• … (2)
• Combining (1) and (2), 1
Aug 10,11,16,17,18, 2023 44

cse.iitkgp.ac.in
Quicksort
• Quicksort was proposed by Sir Charles Antony Richard (Tony) Hoare
in 1959/60 at age 26
source: http://curation.cs.manchester.ac.uk/Turing100/www.turing100.manchester.ac.uk/speakers/invited-list/11-speakers/39-hoare.html
• He was trying to sort words for a machine translation project from
Russian to English
• In his own words – “I have been very lucky. What a wonderful way to start a
career in Computing, by discovering a new sorting algorithm! ”
• Emeritus Professor, Oxford University
• Turing award winner, 1980.
Aug 10,11,16,17,18, 2023 45

cse.iitkgp.ac.in
A Quick Animation
Click to
play
https://tin
yurl.com/3
k8pt3a3
Aug 10,11,16,17,18, 2023 46

cse.iitkgp.ac.in
Median Finding
• Given a set of numbers, a median is the “halfway point” of the set
• More formally, given a set of numbers, medians occur at (lower
median) and at (upper median)
• Following the book, by “median”, we will mean the lower median
• We would like to find the median.

• If a list is sorted, the median can be easily found out. But that means
at least operation
• We would like to do it faster than that
• We would like to find rank element in a sorted list
Aug 10,11,16,17,18, 2023 47

cse.iitkgp.ac.in
Median Finding
• The problem is called the selection problem
• Input: Set of (distinct) numbers and an integer , with
• Output: The element that is larger than exactly other elements of
• The idea we will be using is that after the partition operation we

know the “right” position of the pivot element. We will use it to
discard some “obvious” elements
6 2 9 1 12 5 10 3

cse.iitkgp.ac.in
Median Finding - PseudoCode
Aug 10,11,16,17,18, 2023 49

cse.iitkgp.ac.in
Median Finding – Running Time Analysis

• Best case is that the first pivot is
• What is the running time?
• [At least once you have to scan the array to get the partition]
• What is the worst case?
• (Similar to quicksort) To get the pivot always at one of the ends
• The running time is – Same as quicksort
• Another best case scenario is – perfectly balanced partitioning at all

levels
• Recurrence relation for such case is

cse.iitkgp.ac.in
Median Finding – Running Time Analysis
Case III: We need to check the regularity condition

– This is true for
• Just like quicksort, it can be proven that the average case

complexity is also of the same order as the best case complexity
i.e.,
Aug 10,11,16,17,18, 2023 51

cse.iitkgp.ac.in
Maximum Sum Subarray

• Given an array, we need to find a subarray whose sum of elements
among all possible subarray sums is the maximum
-1 3 4 -5 9 -2
• The problem will be trivial if the numbers are all positive
• By inspection the solution is the subarray from 3 to 9
• What can be a brute-force strategy to find this?
• As we are looking for start and end positions of the subarray, we can
try for all possible start indices and all possible end indices
• This means n-choose-2 i.e., algorithm
• Can we use divide-and-conquer paradigm to improve the runtime?
Aug 10,11,16,17,18, 2023 52

cse.iitkgp.ac.in

• The natural first step is to divide the array into half
-1 3 4 -5 9 -2
• Then recursively solve the same problem on the divided subarrays
• Next we need to combine. While combining, the maximum sum can
come from the two subarrays as well from an array that spans across
the two subarrays
• We need to find the maximum sum corresponding to the subarray
crossing the centerline. And we need to find it efficiently.
• If we can find the maximum crossing subarray in time, then we have
a chance to get recursion leading to overall runtime
Aug 10,11,16,17,18, 2023 53

cse.iitkgp.ac.in

6 7 4 -5 4 2
-1 3 4 -5 9 -2
• To do so, we will cleverly use the fact that we are trying to find a
subarray whose one end lies on the right of the centerline and the
other end lies on the left
• The maximum subarray crossing the centerline will be a
concatenation of 1) a subarray whose one end is the centerline and
goes to the left 2) a subarray whose one end is the centerline and
goes to the right
• So, start at the centerline, keep going to the left, keep doing the
sum and keep track of the maximum of the sum. Similarly, for the
subarray going to the right from the centerline
Aug 10,11,16,17,18, 2023 54
cse.iitkgp.ac.in

6 7 4 -5 4 2
-1 3 4 -5 9 -2
• Now we need to compare three sums – 1) one from entirely the left
subarray 2) one from entirely the right subarray 3) one from the
maximum crossing subarray
• Maximum of these three we will return
• Runtime of finding the maximum crossing subarray?
Aug 10,11,16,17,18, 2023 55

cse.iitkgp.ac.in
Maximum Sum Subarray - Pseudocode
Aug 10,11,16,17,18, 2023 56

cse.iitkgp.ac.in
Maximum Sum Subarray Runtime Analysis
Θ ( 1)
Θ ( 𝑛)
Θ ( 1)
Aug 10,11,16,17,18, 2023 57
cse.iitkgp.ac.in
Maximum Sum Subarray Runtime Analysis
Θ ( 1)
Θ ( 1)
𝑛
𝑇( )
2 𝑛
𝑇( )
2
Θ ( 𝑛)
Θ ( 1)
Aug 10,11,16,17,18, 2023 58

cse.iitkgp.ac.in
Finding Convex Hull

• We will now see two examples in computational geometry where
divide and conquer is used effectively
• The first problem is finding the convex hull of a set of points
• Formally - The convex hull of a set of points is the smallest convex
polygon containing all points in
• So, what is a convex polygon?
• A convex polygon is a polygon with all its interior angles 180°
• A line drawn through a convex polygon will intersect the polygon exactly twice
• All the diagonals of a convex polygon lie entirely inside the polygon
Image source: Math Open Reference

Aug 10,11,16,17,18, 2023 59
cse.iitkgp.ac.in
Finding Convex Hull

• Convex hull of a set of points in a plane is the smallest convex
polygon that contains all of them either inside or on its boundary
• Imagine that the points are represented by nails sticking out from a
board. Convex hull is the shape formed by a tight rubber band that
surrounds all the nails
Aug 10,11,16,17,18, 2023 60
cse.iitkgp.ac.in
Finding Convex Hull

• We would like to have an algorithm that finds the segments/sides of
the convex hull given the set of points (their co-ordinates)
• The book has two efficient algorithms - Graham’s scan and Jarvis’s
march
• However, we would follow a divide-and-conquer algorithm,

famously called as “two finger” algorithm by Prof. Srini Devadas, MIT
in his “Design and Analysis of Algorithms” course
• Much of the material is taken from his lecture [Link].

cse.iitkgp.ac.in
Finding Convex Hull – Problem Definition

• Given points in plane i.e., set and assuming no two points having
same coordinates, no two points having same coordinates and no
three points in a line, we need to find the convex hull (CH())
containing all points in
• The assumptions are not limiting, they are just convenient
𝑝 𝑞
• We will use the following convention
𝑥 to represent the hull
𝑟
𝑣 • Sequence of points on the boundary
in clockwise manner
𝑠
𝑢 𝑡 𝑝 − 𝑞 −𝑟 − 𝑠 −𝑡 − 𝑢 −𝑝
Aug 10,11,16,17,18, 2023 62

cse.iitkgp.ac.in
Finding Convex Hull – Brute Force

• Can we think of a simple (can be brute force) algorithm to generate
the segments comprising the convex hull?
• Draw a line segment connecting a pair of points and check if all the
other points are on one side of the segment
𝑝 𝑞 • Assuming points, what is the

complexity?
𝑥
𝑟 • number of segments
𝑣 • We need to test roughly points
𝑠 against each segment
𝑢 𝑡 • So, total runtime complexity of the
brute-force approach is
Aug 10,11,16,17,18, 2023 63

cse.iitkgp.ac.in
Finding Convex Hull – Divide and Conquer

• Lets try to apply divide and conquer
• ‘Divide’ is pretty straight-forward - dividing them roughly into half
• The points are sorted in terms of x-coordinates. Everything to the
left of the median is one subproblem and everything to the right is
another subproblem
• Note that sorting is a single time operation. And it is
• Solve for convex hulls for the subproblems
• Base case is when the number of points are small – say 2 or 3
• Now merging the solutions is where things start to happen
Aug 10,11,16,17,18, 2023 64

cse.iitkgp.ac.in
Finding Convex Hull – How to Merge?

• Lets try to apply divide and conquer
• The naming convention is – for the left subproblem, the first point is
the closest to the dividing line (highest ). Similarly, for the right
subproblem, the first point is the closest to the dividing line (least )
• We have got the sub-hulls. To generate the ‘combined’ hull, we need
to generate additional segments that are not part of the sub-hulls
but are part of the overall hull. Similarly we need to remove the ones
that cease to be part of the overall hull 𝑏1
𝑎3 𝑎4
𝑎0 𝑏0
𝑎2 𝑎1
𝑏2
Aug 10,11,16,17,18, 2023 65
cse.iitkgp.ac.in

• One way can be picking one point from either side and check if all
other points are on one side of the segment joining the two
• looks good. It has the highest intercept
• The maximum intercept would make sure there is no point higher
than the segment
• Let us call it an ‘upper tangent’
• Similarly is the ‘lower tangent’ 𝑦 31
𝑏1
𝑎3 𝑎4
𝑎0 𝑏0
𝑦 41
𝑦 00
𝑎2 𝑎1 𝑏2
Aug 10,11,16,17,18, 2023 66
cse.iitkgp.ac.in

• This trivial strategy of choosing one point each from either side is
• Can we do better?
• May be – for upper tangent, can I go for connecting highest
coordinate points from either side?
• This will be constant time but not correct always
𝑏1
𝑎3 𝑎4
𝑎0 𝑏0
𝑎2 𝑎1 𝑏2
Aug 10,11,16,17,18, 2023 67
cse.iitkgp.ac.in
Finding Convex Hull – Two Finger Algorithm

𝑏1
Connecting highest 𝑏0
y-coordinate points
𝑏2
𝑦 00
𝑎3
𝑎2 𝑏3
𝑎0
Start with connecting two
closest points to the
𝑎1 midline on either side
Aug 10,11,16,17,18, 2023 68

cse.iitkgp.ac.in

𝑏1
𝑏0
𝑏2
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
Move clockwise
from to
𝑎1
Aug 10,11,16,17,18, 2023 69

cse.iitkgp.ac.in

𝑏1
𝑏0
𝑏2
𝑦 30
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
As intercept decreased, we
will go back to on R.H.S. and
𝑎1 move counterclockwise on
L.H.S to
Aug 10,11,16,17,18, 2023 70
cse.iitkgp.ac.in

𝑏1
𝑏0
𝑏2
𝑦 30
𝑦 31
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
As intercept increased, we
will move clockwise on R.H.S
𝑎1 to
Aug 10,11,16,17,18, 2023 71

cse.iitkgp.ac.in

𝑏1
𝑏0
𝑦 20 𝑏2
𝑦 30
𝑦 31
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
L.H.S to
Aug 10,11,16,17,18, 2023 72
cse.iitkgp.ac.in

𝑏1
𝑏0
𝑦 20 𝑏2
𝑦 30
𝑦 31 𝑦 21
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
As intercept increased, we
will move clockwise on R.H.S
𝑎1 to
Aug 10,11,16,17,18, 2023 73

cse.iitkgp.ac.in

Move from to as well as 𝑏1
to , both resulted in 𝑏0
decrease of -intercept. So,
the move will stop and 𝑦 20 𝑏2
will be the segment with 𝑦 30
highest -intercept. 𝑦 31 𝑦 21
𝑦 10
𝑦 00
𝑎3 𝑦 01
𝑎2 𝑏3
𝑎0
L.H.S to
Aug 10,11,16,17,18, 2023 74
cse.iitkgp.ac.in
PseudoCode
// Initialize counters
i=j=0;
// j+1 is actually (j+1)%p
if // i-1 is actually (i-1)%q
Return
and are points on the left and right halves respectively
What about the runtime complexity of the two finger algorithm?
Aug 10,11,16,17,18, 2023 75

cse.iitkgp.ac.in
One Final Bit

• How to remove the segments that are part of the sub-hulls but not
of the overall hull?
• Lets say we have found the upper and the lower tangents as and
which for our example were and
• Our two sub-hulls are and
𝑏1
𝑏0
𝑏2
𝑎3
𝑏3
𝑎2
𝑎0
𝑎1
Aug 10,11,16,17,18, 2023 76

cse.iitkgp.ac.in
One Final Bit

• How to remove the segments that are part of the sub-hulls but not
of the overall hull?
• Lets say we have found the upper and the lower tangents as and
which for our example were and
• Our two sub-hulls are and
𝑏1
𝑏0 • Link (here )
𝑏2 • Traverse the list till is met
• Link to (here to )
𝑎3
𝑎2
𝑏3 • Traverse the list till is met
𝑎0
𝑎1 • So the hull is
Aug 10,11,16,17,18, 2023 77
cse.iitkgp.ac.in
Closest Pair of Points

• Problem definition: Given points in the plane, find the pair and
whose Euclidean distance is as small as possible
• Naïve approach: Try every pair of points. runtime
Aug 10,11,16,17,18, 2023 78

cse.iitkgp.ac.in

• Can we avoid comparing every pair of points?
• How about single dimension case?
• In one dimensional case, the structure of the problem setting helps.
we only need the distances between adjacent points. So .
• But we need to sort them first according to the -coordinate values.
So its
Aug 10,11,16,17,18, 2023 79

cse.iitkgp.ac.in
2 Dimensions – Divide and Conquer

• Split the set of points into two halves
• For 2-D points, a good option to split is using a vertical line that divides the set into two
of roughly the equal size
• For that we would need to sort the points according to -coordinates
• Recursively compute closest pairs in these halves until base case
• While combining compute closest pair across two halves
Aug 10,11,16,17,18, 2023 80

cse.iitkgp.ac.in
2 Dimensions – Divide and Conquer

• One way to compute closest pair across the midline, is to choose
one point from either side and compute the distance
• This will be
• But do we really need to consider every pair across the midline?
• We only need to consider checking a narrow band across the
midline where the band width is twice the minimum distance found
on either halves
Aug 10,11,16,17,18, 2023 81

cse.iitkgp.ac.in

• Let the closest pair distance on the left and right halves be and
respectively and
• However, the same issue of all points to be residing inside the band
can occur
• Savior: Points on each side separated by at least . Not enough room
for many of them to be packed nearby
• This means we need to check only a few pairs crossing the line
𝛿 𝛿
Aug 10,11,16,17,18, 2023 82

cse.iitkgp.ac.in

• Lets consider the point shown in the band
• The candidate points can only lie in the two neighboring boxes of
size shown to the right
• Now, how many points can be in these two neighboring boxes?
𝛿 𝛿
Aug 10,11,16,17,18, 2023 83

cse.iitkgp.ac.in

𝛿 𝛿 𝛿 • Now, how many points can be in
√2 2 2 these two neighboring boxes?
• Lets subdivide the boxes by half
𝛿 • Maximum one point can be there in
2 these boxes
𝛿
• So at most 8 points to check on the
2
right hand side
Aug 10,11,16,17,18, 2023 84

cse.iitkgp.ac.in

• So, for every point in the band on the left we have to scan points in
the neighboring boxes
• This requires the points to be sorted according to -coordinates also
• Time for comparison in this window is linear as it’s a fixed number (8
points).
𝛿 𝛿
Aug 10,11,16,17,18, 2023 85

cse.iitkgp.ac.in
Closest Pair of Points - Algorithm

• Input: Set of 2-D points (denoted by set )
• Assumption: No two points are same in its or coordinates. Not a
limiting assumption, just convenient
• Step 0: Sort all points on both and coordinates to get two different
lists, say and
• When we divide , we need to get corresponding lists in each half
𝒫
𝒬 ℛ
𝒬𝑦 ℛ𝑦
𝒬𝑥 ℛ𝑥
Aug 10,11,16,17,18, 2023 86
cse.iitkgp.ac.in

• Getting and are easy
• To get and , we have to scan and points for which coordinate is
less than the midline will be pushed to the list and points for which
coordinate is more than the midline will be pushed to the list
• So, in linear time we get
𝒫
𝒬 ℛ
𝒬𝑦 ℛ𝑦
𝒬𝑥 ℛ𝑥
Aug 10,11,16,17,18, 2023 87
cse.iitkgp.ac.in

lists, say and
• Step 1: Call ‘ClosestPair( )’
• Step 2: If hits base case (2 Points) return the distance between two
• Step 3: Recursively call ‘ClosestPair( )’ and ‘ClosestPair( )’
• Step 4: Get closest ‘across-midline’ pair and return minimum of this,
left closest and right closest pair
𝒬 ℛ
𝒬𝑦 ℛ𝑦
𝒬𝑥 ℛ𝑥
Aug 10,11,16,17,18, 2023 88
cse.iitkgp.ac.in
Closest Pair of Points - Analysis

lists, say and ->
• Step 1: Call ‘ClosestPair( )’ ->
• Step 2: If hits base case (2 Points) return the distance between two
->
• Step 3: Recursively call ‘ClosestPair( )’ and ‘ClosestPair( )’
->
• Step 4: Get closest ‘across-midline’ pair and return minimum of this,
left closest and right closest pair ->
• So the recurrence relation is
• So, runtime
Aug 10,11,16,17,18, 2023 89

cse.iitkgp.ac.in
Thank You!!

03 Recursion DAC

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

03 Recursion DAC

Uploaded by

Copyright:

Available Formats

Computer Science and Engineering| Indian Institute of Technology Kharagpur

Autumn 2023, IIT Kharagpur

Divide and Conquer, Recursion

Divide and Conquer

Aug 10,11,16,17,18, 2023 2

Divide and Conquer

• Divide a problem into pieces (smaller instances of same problem)

Aug 10,11,16,17,18, 2023 3

Aug 10,11,16,17,18, 2023 5

Merge Sort Pseudocode

Aug 10,11,16,17,18, 2023 6

Merge Sort – In What Order?

Merge Sort – Runtime Analysis

Overall runtime of “MERGE”

Aug 10,11,16,17,18, 2023 8

Merge Sort – Runtime Analysis

• How do we count the recursive calls?

Aug 10,11,16,17,18, 2023 9

Merge Sort – Runtime Analysis

• There are levels and at each level the number of operations is

• Next we will learn how to write an equation that represents the

Aug 10,11,16,17,18, 2023 10

Divide time = const

Conquer time = each

• Recurrence relation: Represents the

Aug 10,11,16,17,18, 2023 11

Solving Recurrences – Recursion Tree

• We can write as some const.*

Aug 10,11,16,17,18, 2023 12

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 13

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 14

Solving Recurrences – Recursion Tree

• Total runtime comes from counting number of operations from the

Aug 10,11,16,17,18, 2023 15

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 16

Left child Right child

Aug 10,11,16,17,18, 2023 17

Some Terminologies Max # of

Aug 10,11,16,17,18, 2023 18

D E F G For general `k’-array Tree

• Number of nodes at level

Aug 10,11,16,17,18, 2023 19

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 20

Solving Recurrences – Recursion Tree

Solving Recurrences – Recursion Tree

• Where, is number of subproblems (branching factor)

• What are these values for binary search?

• with base case time - constant

Aug 10,11,16,17,18, 2023 22

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 23

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 24

Solving Recurrences – Recursion Tree

Aug 10,11,16,17,18, 2023 25

Solving Recurrences – Master Theorem

• where is base case runtime

Solving Recurrences – Master Theorem