CS3230: Design and Analysis of Algorithms

Semester 2, 2020-21, School of Computing, NUS

Q1 (a) Let X denote the random variable for the number of heads you get when you toss the biased coin
k times.
Then,
k
k X
Pr(X ≥ )= Pr(X = i)
2 k
i= 2
k    i  k−i
X k 1 3
=
i 4 4
i= k
2

k    i  k−i
X k 1 3
= k
2 4 4
i= k
2

k  i  k−i
 X    
k 1 3 k k
= k [Using ≤ k , ∀k]
2 k
4 4 x 2
i= 2
k  i  k−i
 X
k 1 3
= k
2 k
4 4
i= 2
k
 k X
k
2k
 
2 3 k
≤ 3−i [Using k ≤ ]
2 4 k 2 2
i= 2
k
2k 3 3 2
= [Using GP formula]
2 4k 2
k
3 2k 3 2
=
4 22k
k
32 3
≤ k [Using ≤ 1]
2 4
k
2log2 3 2
= [Using 2log2 x = x, ∀x]
2k
1.6k
≤ 2 2 −k [Using log2 3 ≤ 1.6]
k
= 2− 5
1
= k
25

Therefore, Pr(X ≥ k2 ) ≤ 1
k .
25

1
(b) Number of elements that are near median = all the elements that are ∈ [ n4 , 3n
4 ]
3n n
So # elements that are near median = 4 − 4 +1
Let Y denote the random variable having the value of the element that we picked. Then,

3n
− n4 + 1
 
n 3n 4
Pr Y ∈ [ , ] =
4 4 n
n
+ 1
= 2
n
1 1
= +
2 n
n
(c) Let A denote the event that the median of the 5 log n elements is < 4 and let B denote the event that
the median is > 3n
4
Then,

Pr(Failure) = Pr(A ∪ B)
≤ Pr(A) + Pr(B) [Using Union Bound Inequality]
≤ 2 Pr(A) Both the events are symmetric and identical
Pr(A) = Pr(# elements < n/4 are at-least n/2)

n
Here we treat each element as biased coin toss where you pick the BAD event, i.e. < 4 with probability
≤ 14 . So using (a) we know this probability with k = 5 log n is ≤ 1k
25

n
Pr(A) = Pr(X ≥ )
2
1
≤ k
25
1
= log n [Substitute k = 5 log n]
2
1
=
n
Pr(Failure) ≤ 2 Pr(A)
2
≤
n
2
Pr(Success) ≥ 1 −
n

Q2 Throughout this, assuming each standard pop and push cost 1 unit of time.
a) MPUSH:
The
Pk worst case happen when stack 0 to stack k are full, where k is the maximum number such that
i
i=0 3 = (3k+1 − 1)/2 ≤ n. Then the time will be 3k+1 = Θ(n) (we need to pop and push each
element in each stack once, with one additional push for the new element at the end)

MPOP:

2
For any stack Si with i > 0, it can only be either empty, having 1/3 of its capacity, having 2/3 of its
capacity, or full.

Worst case happen when stack 0 to stack k are empty, also, using the observation,
Pk since stack k+1 is not
empty, then 3k+1 ×(1/3) = 3k ≤ n so k <= log3 n). Then the time will be i=0 3i ×2+1 = 3k+1 = Θ(n)
(similar to MPUSH, need to pop and and push each element in each stack once, with one additional
pop at the end)

b) Using accounting methods, for each MPUSH/MPOP, we will charge 20 log3 n = O(log n).
Now, consider an arbitrary MPUSH, let j equal to i after the loop (see the pseudocode in the question)
to find the first stack that is not full. Then the number of element in S0 , S1 , ..., Sj−1 , ... must be:
1, 31 , ..., 3j−1 , ... (correspondingly). Denote this configuration as Cj .
Pj−1
The cost to execute this MPUSH is 1 + 2 y=0 3y = 3j . After it is executed, the number of element in
S0 , S1 , ..., Sj−1 , ... must be: 1, 30 , ..., 3j−2 , ... (correspondingly).
Pj−1
To reach the configuration Cj again, we need at least y=1 (3y − 3y−1 ) = 3j−1 − 1 MPUSH operations.
Therefore, in MPUSH, i can only be equal a particular j at most once every 3j−1 MPUSH.
Pk m
So, total cost of MPUSH during any m operations can be at most j=0 3j 3j−1 = 3km = O(m log n)
(since j can be at most k = log3 n).
Similarly, consider an arbitrary MPOP, let j equal to i after the loop to find the first stack that is not
empty. Then the number of element in S0 , S1 , ..., Sj−1 , ... must be:
0, 0, ..., 0, ... (correspondingly). Denote this configuration as Dj .
Pj−1
The cost to execute this MPOP is 1 + 2 y=0 3y = 3j . After it is executed, the number of element in
S0 , S1 , ..., Sj−1 , ... must be: 0, 2(30 ), ..., 2(3j−2 ), ... (correspondingly).
Pj−1 y j−1
To reach the configuration Dj again, we need at least y=1 2(3 ) = 3 − 1 MPOP operations.
j−1
Therefore, in MPOP, i can only be equal a particularP j at most once every 3 MPOP. So, total cost
k m
of MPOP during any m operations can be at most j=0 3j 3j−1 = 3km = O(m log n) (since stack j
can not be empty, j can be at most k = log3 n = O(log n)).
Thus, during any m operation, the balance will be at least 20m log3 n − 3m log3 n − 3m log3 n ≥ 0.