COL 351:
Analysis and Design of Algorithms
Lecture 1
Grading Policy
1. Quizzes - 15%
(surprised / announced)
2. Exams - 35% + 40%
3. Class participation - 5% (Lec) + 5% (Tut)
Academic Honesty
Cheating or allowing anyone to copy in quizzes, exams would lead to penalty of one grade
Course webpage: https://web.iitd.ac.in/~keerti/Courses/COL351-2025.htm
COL 106 This Course
Arrays, Link lists, Stacks • Designing algorithms
• Different algorithm paradigms
Trees, Binary trees, AVL trees, KD trees
1. Greedy algorithms
2. Dynamic programming
How to use these data structures
3. Divide & Conquer
in computational problems?
• Queues are used in BFS • Hard Problems:
Problems which are unlikely to
• Stacks are used in DFS
have an ef cient solution.
• Priority queues are used in
Dijkstra’s algorithm
• How to prove that a problem is hard?
fi
Algorithm Paradigms
1. Divide and Conquer (Eg. Merge sort, quick
— Divide the problem into smaller problems sort, binary search)
— Solve the smaller problems
— Combine
2. Dynamic Programming
(Eg. insertion sort)
Reduce the problem on an input of size n into
problems of size n − 1, n − 2, n − 3,.. etc.
3. Greedy Strategy
(Eg. Selection sort)
Build solution greedily.
Divide and Conquer
Example: Merge Sort
4 2 10 5 16 3 9
L1 4 2 10 5 16 3 9 L2
Sorted L1 2 4 5 10 3 9 16 Sorted L2
Example: Merge Sort
MergeSort(L)
n = length(L); n
T(n) = 2 T( ) + O(n)
If n = 1 then Return; 2
A = new list of size n;
Set L1 = L[0,
2]
L2 = L[ 2 + 1, n − 1];
n n
and
MergeSort(L1);
MergeSort(L2);
Set x, y, pos = 0;
While (x < length(L1) or y < length(L2))
If (L1[x] ⩽ L2[y] and x < length(L1)) then
Set A[pos] = L1[x], and increment pos and x by 1;
Else
Set A[pos] = L2[x], and increment pos and y by 1;
Example: Merge Sort
n
Let T(n) = number of steps taken by the algorithm. Then, T(n) ⩽ 2 T( ) + cn
2
cn cn
T(n)
T(n/2) T(n/2) cn/2 cn/2
Sum in each level = cn
T(n/4) T(n/4) T(n/4) T(n/4)
Height = log2 n
So, T(n) = O(n log n)
Merge Sort - Unequal balance
Let T(n) = number of steps taken by the algorithm.
What if | L1 | = n/3, and | L2 | = 2n/3?
cn cn
T(n)
T(n/3) T(2n/3) cn/3 2cn/3
T(n/9) T(2n/9) T(2n/9) T(4n/9)
Sum in each level ⩽ cn
Max height = log1.5 n
So, T(n) = O(n log n)
Merge Sort - Unequal balance
Let T(n) = number of steps taken by the algorithm.
What if | L1 | = n , and | L2 | = n − n?
Dynamic Programming
Nth Fibonacci Number
De ned by recurrence relation F0 = 0, F1 = 1, and Fn = Fn−2 + Fn−1 for n > 1.
1. If n = 0,1 then return a binary list of containing (n)2;
2. A = FibNaive(n − 1);
3. B = FibNaive(n − 2);
4. Return BinarySum(A, B);
FibNaive(n)
T(n) = T(n − 2) + T(n − 1) + Time-to-add-two-large-numbers
n
Remark: T(n) is exponential here because T(n) ⩾ Fn ⩾ 2
fi
Nth Fibonacci Number - Improved algorithm
De ned by recurrence relation F0 = 0, F1 = 1, and Fn = Fn−2 + Fn−1 for n > 1.
1. Allocate an array A of size n to store references/pointers.
2. For i = 0,1 : Let A[i] store pointer to binary list of containing (i)2;
3. For i= 2 to n:
A[i] = BinarySum(A[i − 1], A[i − 2]);
4. Return A[n];
Fibonacci-Improved( n )
T(n) = n × Time-to-add-two-large-numbers = O(n 2) (Note : Fn ⩽ 4n)
fi
Greedy Algorithms
Selection Sort
Given: An array A storing n distinct integers.
1. Allocate an array B of size n to store solution.
2. Let B[0] store minimum element of array A;
3. For i= 1 to n − 1:
B[i] = Smallest element in A greater than equal to B[i − 1];
4. Return B;
Sort( A, n)
T(n) = O(n 2)
Homework: Course-prerequisite ordering
Given : A DAG (directed acyclic graph) corresponding to course-prerequisites for courses
you are keen in taking.
Find : A linear ordering of vertices of DAG (course-prerequisite graph) such that for every
directed edge (u, v), the vertex u comes before v in the ordering.
1 4
6 1 4 2 3 5 6
2 3
