Data Structures and Algorithms

(CS210A)

Lecture
  5:
• More on Proof of correctness of an algorithm
• Design of O() time algorithm for Local Minima in a grid

1
PROOF OF CORRECTNESS

2
What does correctness of an algorithm mean ?

For every possible valid input, the algorithm must output correct answer.

Let us take a toy algorithm 

3
 Algorithm for computing
  from to
sum of numbers
•  
Sum_of_Numbers()
{ Sum;
for = 1 to
{
Sum Sum + ;
}
return Sum;
  How will you convince any person
} that Sum_of_Numbers()
indeed is correct ?
Natural responses:
• It is obvious !
• Compile it and run it for some random values of .
• Go over first few iterations explaining what happens to Sum.
4
 How will you respond
if you have to do it for the following code ?

5
Think for some time to realize
• the non-triviality
• the Importance
of proof of correctness of an iterative algorithm.

In the following slide, we present an overview of

the proof of correctness.

Interestingly, such a proof will be just

Expressing our intuition/insight of the algorithm in a formal way .

6
 Proof of correctness
For an iterative algorithm

•  
Insight of the algorithm Theorem
 Assertion  Assertion
P() P()

Start
of Loop
 1  2  3 4
  𝑖  − 1  𝑖 Iterations

Prove P() by  What would you expect

1.   ?Assuming P() at the end of th
iteration ?
2. Theorem
? Proof by induction
3. Body
? of the Loop
The most difficult/creative part of proof :   To come? up with the right assertion P()
7
 Algorithm for computing
  from to
sum of numbers
•  
{ Sum;
for = 1 to
{
Sum Sum + ;
}
return Sum;
}

Assertion P() : ?  At the end of th iteration Sum stores the sum of numbers from to .

Base case: P() holds.

Assuming P(), assertion P() also holds.
P() holds.
8
 An O() time Algorithm for Max-sum subarray
 
•  
Let S(): the sum of the maximum-sum subarray ending at index .

Theorem 1 : If S() > 0 then S() = S() + A[]

else S() = A[]

Max-sum-subarray-algo(A[])
{ S[]  A[]
for = 1 to
{ If S[] > 0 then S[]  S[] + A[]
else S[]  A[]
}
“Scan S to return the maximum entry”
}
Assertion P() :  ?S[] stores the sum of maximum sum subarray ending at A[].
Homework: Prove that P() holds for all
9
LOCAL MINIMA IN A GRID

10
 Local minima in a grid

•  
Definition: Given a × grid storing distinct numbers, an entry is local minima
if it is smaller than each of its neighbors.
 𝒋

31
 𝒊 5 33 10
99

11
 Local minima in a grid

•  
Problem: Given a × grid storing distinct numbers, output any local minima
in O() time.  𝒋

31
 𝒊 5 33 10
99

12
 Two simple principles

1. Respect every new idea which solves a problem even partially.

2. Principle of simplification:
If you find a problem difficult,
 try to solve its simpler version, and then
 extend this solution to the original (difficult) version.

13
 A new approach
Repeat : if current entry is not local minima, explore the neighbor storing smaller
value.
j

i 3

14
 A new approach
Explore()
{ Let c be any entry to start with;
While(c is not a local minima)
{
c  a neighbor of c storing smaller value
}
return c;
}

15
 A new approach
•  
Explore()
{ Let c be any entry to start with;
While(c is not a local minima)
{
c  a neighbor of c storing smaller value How to apply this
} principle ?
return c;
}
Worst case time complexity : O()

First principle: Second principle:

Do not discard Explore() Simplify the problem

16
 Local minima in an array
A local minima exists
•  
in this region. Why ?
23
17

A 9 17 23
 𝑖

Theorem: There is a local minima in A[0,…, ].

Proof: Suppose we execute Explore() from A[].
Explore(), if terminates, will return local minima.
It will terminate without ever entering A[,…, ].
Algorithmic proof
Hence there is a local minima in A[0,…, ].

17
 Local minima in an array
•  
23
17

A 9 17 23
 𝑖

Theorem: There is a local minima in A[0,…, ].

 We can confine our search for local minima to only A[0,…, ].

Our problem size has reduced.
Question: Which should we select so as to reduce problem size significantly ?
Answer: point of array A.
18
 Local minima in an array
(Similar to binary search)

•  Local-minima-in-array(A) {
int   O(log )
L  0;
How many
R; iterations ?
found  FALSE;
while( not?? found )
{
 (L + R)/2;
If ( is a local minima)
found  TRUE;  O() time
in one iteration
else if(A[] < A[]) ?? ;
  L
else ?? R
 
}
return ; }
 Running time of the algorithm = O(log ) Proof of correctness ?

19
 Local minima in an array
(Proof of correctness)

 What can you say about the

•   algorithm at the end of th
iteration.
 𝑳  𝑹

A
P() : At the end of th iteration,
“A local minima of array A exists in A[,…, ].”

20
 Local minima in an array
(Proof of correctness)

•  
 𝑳  𝑹

A
P() : At the end of th iteration,
“A local minima of array A exists in A[,…, ].”

=
“A[]< A[]” and “A[]< A[]”.
Homework:
• Make sincere attempts to prove the assertion P().
• How will you use it to prove that Local-minima-in-array(A) outputs a local
21
minima ?
 Local minima in an array
(Proof of correctness)

•  

Theorem: A local minima in an array storing distinct elements

can be found in O(log ) time.

22
 Local minima in a grid
(extending the solution from 1-D to 2-D)
Under what
•  
Search for a local minima in the column M[, ] AWhat
local minima
circumstances
inlocal
this minima
smallest
exists
if thereeven
is no this
region. Why
in the
element
?
is not
aentire
localcolumn.
minima ?
𝒎𝒊𝒅
 

Smallest element
  Execute Explore()
of the column
from M[, ]

𝒊  9 7

Homework:
Use this idea to design an O(log ) time algorithm for this problem.
… and do not forget to prove its correctness . 23
 Make sincere attempts to
answer all questions raised in this lecture.

24