Introduction to Algorithms(Asymptotic Analysis)-Part 2

·  ☕ 3 min read  ·  🤖 Arjit Sharma

Time complexity of Divide and Conquer Algorithms -

These algorithm divide problem in subparts recursively & each subproblems takes additional work to compute the subproblem.
Ex- T(n) = 2T(n/2) + O(n), here each problem is div in 2 parts of size n/2 each and it takes O(n) time for each subproblem.
Such algorithms can be expressed in form of recursive equation as shown above and there exists a "Master Theorem" to simplify calculating worst case time complexity for such algorithms.
These problems complexity can be found with subsitution method or recursive method too.

Master Theorem

For recurrence given in the following form $T(n) = a T(n/b) + f(n)$

Case 1: $f(n) = O(n^\left( log_ba-\epsilon\right))$ for $\epsilon>0$
$$T(n)=\Theta(n^\left( log_ba\right))$$

Case 2: $f(n) = θ(n^\left( log_ba\right))$
$$T(n)=\Theta(n^\left( log_ba\right)log n)$$

Case 3: $f(n) = Ω(n^\left( log_ba+\epsilon\right)) for \epsilon>0$
$$T(n)=\Theta(f(n))$$

Example 1: $T(n)=3T(n/2) + n^2$

$n^\left(log_23 \right)=>n^\left(1.58\right)$
$n^\left(1.58+\epsilon\right)= n^2$
Case 3 satisfies so
$T(n)=\theta(n^2)$

Example 2 : $T(n)=4T(n/2) + n^2$

$n^\left(log_24 \right)=>n^\left(2\right)$
$n^\left(2\right)= f(n)$
Case 2 satisfies so
$T(n)=\theta(n^2log n)$

Extended Master Theorem :

We can use extended Master theorem when function is in form of →
$T(n)=aT(n/b)+\theta\left(n^klog^pn\right)$ given that a≥1,b>1,k≥0,p is Real no.

Case 1: $a>b^k$
$T(n)=\theta(n^\left(log_ba\right))$

Case 2: $a=b^k$
a) p>-1    $T(n)=\theta(n^\left(log_ba\right)log^\left(p+1\right)n)$
b) p=-1    $T(n)=\theta(n^\left(log_ba\right)loglog(n))$
c) p< -1    $T(n)=\theta(n^\left(log_ba\right)$

Case 3: $a< b^k$
a) p≥0    $T(n)=\theta(n^klogn)$
b) p<0    $T(n)=\theta(f(n))$

Example 1: $T(n)=2T(n/2) +n/logn$

a=2,b=2,k=1,p=-1
$a=b^k satisfies$
and p=-1 so Case 2.b satisfies
$T(n)=\theta(n^\left(log_22\right)loglog(n))$
$T(n)=\theta(nloglog(n))$

Example 2: $T(n)=16T(n/4)-n^2logn$

Master theorem doesnt apply as function is negative

Master Theorem for Subtract and Conquer Recurrences

For recurrences of the form

$T(n) = c$ if n≤1
$T(n) = aT(n-b)+f(n)$ if n>1
if f(n) is $O(n^k)$,a>0,b>0 and k≥0

We can use:
Case 1: $a<1$    $O(n^k)$
Case 2: $a=1$    $O(n^\left(k+1\right))$
Case 3: $a>1$    $O(n^ka^\left(n/b\right))$

Example :
$T(n)=3T(n-1)$ if n>0
$T(n)=1$ otherwise

Soln→
$T(n)=3T(n-1)+O(1) ⇒ T(n)=3T(n-1)+O(n^0)$
a>1 so Case 3 satisfies
$T(n)=O(n^03^\left(n/1\right))=O(3^n)$

Subsitution Method

Condensed way of proving asymptotic bound on a recurrence by induction.
There are 2 steps to this method:
Step 1 - Guess the solution
Step 2 - Prove using induction

Example : $T(n) = T(n-1) + n$

Guessed $T(n) ≤ cn^2$
$T(n) = T(n-1) + n$
$T(n) ≤ c(n-1)^2 +n$
$T(n)=cn^2 + c - 2cn +n$
$T(n)≤ cn^2$ if n is sufficiently large and c>1/2

Recursive Tree Method

Popular technique for solving unbalanced recurrence relations. Example - Modified merge sort which divides problem in 2 subproblems $n/3$ and $2n/3$ each.

Example - $T(n) = 2T(n/2) + n$

Complexity = height of tree * Cost at each level

Finding ht of tree :
Size of subproblem at level 0 is $n/2^0$
Size of subproblem at level 1 is $n/2^1$
Size of subproblem at last level will become 1
$n/2^x=1$
$x=logn$
height of tree is $logn$

Cost of each level is n

Complexity = $O(nlogn)$

Reached end? Why not read Part 3

Referrences

Share on

WRITTEN BY
Arjit Sharma
Yo, I am a CS enthusiast or am I ? Just kidding.