Chapter 4 PDF

1
Data Structures and Algorithms
BITS-Pilani K. K. Birla Goa Campus
1
Material for the presentation taken from Cormen, Leiserson, Rivest and
Stein, Introduction to Algorithms, Third Edition; 1
BITS-Pilani K. K. Birla Goa Campus Data Structures and Algorithms
Ch 4: Divide-and-Conquer
Divide the problem into a number of subproblems that are

smaller instances of the same problem.
Conquer the subproblems by solving them recursively.
Combine the solutions to the subproblems into the solution
for the original problem.
2
Maximize Profit
3
Maximize Profit
I Possible approach : either buy at the lowest price or sell at the

highest price.
3
Maximize your Profit
4
Maximize your Profit
I Profit may not be maximized by either buying at the lowest

price or selling at the highest price.
4
Transformation: Maximum subarray problem
5
Transformation: Maximum subarray problem
I We want to find the contiguous subarray having the largest

sum.
5
Maximum subarray problem
6
Maximum subarray problem
I The problem is interesting only when the array contains some

negative numbers.
6
Maximum subarray problem : Divide-and-conquer approach
7
I Divide :
7
I Divide :
I Conquer : Maximum subarray of A[low , mid] and
A[mid + 1, high] can be found recursively, because they are
smaller instances of the bigger problem.
7
I Divide :
I Conquer : Maximum subarray of A[low , mid] and
A[mid + 1, high] can be found recursively, because they are
smaller instances of the bigger problem.
I Combine step: Where will maximum subarray of A[low , high]
lie?
7
Maximum subarray that lies is both the subproblems
8
Maximum subarray: Divide-and-conquer algorithm
9
Analyzing the divide-and-conquer algorithm
9
I Running time : Θ(n lg n)
9
I Running time : Θ(n lg n)

I Can we do better? o(n lg n)?
9
Matrix multiplication
I C = A · B, where A, B and C are n × n matrices
10

n
X
I cij = aik · bkj
k=1
10

n
X
I cij = aik · bkj
k=1
10
I SQUARE-MATRIX-MULTIPLY procedure runs in Θ(n3 ) time.
11

I Volker Strassen came up with an algorithm that takes O(nlg 7 )
time, which is o(n2.81 ). (lg 7 = 2.8074)
11

I Volker Strassen came up with an algorithm that takes O(nlg 7 )
time, which is o(n2.81 ). (lg 7 = 2.8074)
I Strassen’s algorithm uses a divide-and-conquer approach.
11
Matrix multiplication: divide-and-conquer
12
12
12
13
14
I T (n) = 8T (n/2) + D(n) + C (n)
14
I T (n) = 8T (n/2) + D(n) + C (n)

I T (n) = 8T (n/2) + Θ(1) + Θ(n2 )
14
Recursion tree for merge sort
15
I T (n) = 8lg n + . . .
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
I Strassen modified the divide-and-conquer approach explained
above such that T (n) = O(nlg 7 )
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
I T (n) = 7lg n + . . .
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
T (n) = Θ(nlg 7 )
16
I T (n) = 8lg n + . . .
T (n) = nlg 8 + . . .
T (n) = Θ(n3 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
T (n) = Θ(nlg 7 )
n 0 1 2 3 4 5 6
8n 1 8 64 512 4096 32768 262144
7n 1 7 49 343 2401 16807 117649
2n 1 2 4 8 16 32 64
16
Strassen’s algorithm : Divide (Step 1)
Divide step involves finding ten n/2 × n/2 matrices (Si ).
17
Strassen’s algorithm : Divide (Step 1)
Divide step involves finding ten n/2 × n/2 matrices (Si ).
17
Strassen’s algorithm : Conquer (Step 2)
18
Strassen’s algorithm : Conquer (Step 2)
18
19
Strassen’s algorithm : Combine (Step 3)
20
21
22
23
Strassen’s algorithm
I T (n) = 7T (n/2) + D(n) + C (n)
24
I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + Θ(n2 ) + Θ(n2 )
24
I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + Θ(n2 ) + Θ(n2 )
I T (n) = 7lg n + . . .
24
I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + Θ(n2 ) + Θ(n2 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
24
I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + Θ(n2 ) + Θ(n2 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
T (n) = O(nlg 7 ) = o(n2.81 )
24
I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + Θ(n2 ) + Θ(n2 )
I T (n) = 7lg n + . . .
T (n) = nlg 7 + . . .
T (n) = O(nlg 7 ) = o(n2.81 )
I Why should n be a power of 2?
24
Q. How would you modify Strassen’s algorithm to multiply n × n

matrices in which n is not an exact power of 2?
25

nlg 7 <= mlg 7 <
25

nlg 7 <= mlg 7 < (2n)lg 7
25

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7
25

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7 = 7nlg 7
25

25

The above implies that mlg 7 = Θ(nlg 7 ).
25
Substitution method : T (n) = 2T (bn/2c) + n
I Find an upper bound for T (n) = 2T (bn/2c) + n
26

I We will first guess a solution: O(n lg n) (using Recursion tree)
26

I To show : T (n) = O(n lg n)
26

T (n) ≤ dn lg n
26

T (n) ≤ dn lg n
I We will assume that T (n/2) ≤ d(n/2) lg(n/2). Then show
that T (n) ≤ dn lg n. (Inductive step)
26
Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)
27
T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)
27
T (n) = 2T (bn/2c) + n
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
= dn lg n − dn + n
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
≤ dn lg n , for d ≥ 1
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
Base case is problematic
T (1) = 1
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
T (1) = 1 d 1 lg 1
27
T (n) = 2T (bn/2c) + n
≤ 2d(n/2) lg(n/2) + n
= dn(lg n − lg 2) + n
= dn(lg n − 1) + n
T (1) = 1 d 1 lg 1 = 0
27
I T (2) = 2T (b2/2c) + 2
28
I T (2) = 2T (b2/2c) + 2 = 4
28
I T (2) = 2T (b2/2c) + 2 = 4 ≤ d2 lg 2
28
I T (2) = 2T (b2/2c) + 2 = 4 ≤ d2 lg 2
I T (3) = 2T (b3/2c) + 3
28
I T (2) = 2T (b2/2c) + 2 = 4 ≤ d2 lg 2
I T (3) = 2T (b3/2c) + 3 = 5
28
I T (2) = 2T (b2/2c) + 2 = 4 ≤ d2 lg 2
I T (3) = 2T (b3/2c) + 3 = 5 ≤ d3 lg 3
28
I T (2) = 2T (b2/2c) + 2 = 4 ≤ d2 lg 2
I T (3) = 2T (b3/2c) + 3 = 5 ≤ d3 lg 3
I We only need to show that T (n) ≤ dn lg n for n ≥ n0 , where
n0 need not be 1.
28
Substitution method : T (n) = T (bn/2c) + T (dn/2e) + 1
29
I Find a tight upper bound for the recurrence

T (n) = T (bn/2c) + T (dn/2e) + 1
29

T (n) = T (bn/2c) + T (dn/2e) + 1
I What can be an initial guess for the above question based on
recursion tree?
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?
I We guess T (n) = O(n).
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?
To prove : T (n) ≤ cn ∀ n ≥ n0
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?
Induction step assumption :
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?
T (bn/2c) ≤ c(bn/2c) , T (dn/2e) ≤ c(dn/2e)
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?

T (n) = T (bn/2c) + T (dn/2e) + 1
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?

T (n) = T (bn/2c) + T (dn/2e) + 1 ≤ cbn/2c + cdn/2e + 1
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?

T (n) ≤ cbn/2c + cdn/2e + 1
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?

T (n) ≤ cbn/2c + cdn/2e + 1
= cn + 1
29

T (n) = T (bn/2c) + T (dn/2e) + 1
recursion tree?

T (n) ≤ cbn/2c + cdn/2e + 1
= cn + 1 ( cannot proceed )
29
We need to slightly modify the initial guess for the recurrence

T (n) = T (bn/2c) + T (dn/2e) + 1.
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
To prove: T (n) ≤ cn − d ∀ n ≥ n0 .
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
T (n) = T (bn/2c) + T (dn/2e) + 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
T (n) = T (bn/2c) + T (dn/2e) + 1 ≤ cbn/2c − d + cdn/2e − d + 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
≤ cn − d , for d ≥ 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
≤ cn − d , for d ≥ 1
T (1) = 1 ≤ cn − d
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
≤ cn − d , for d ≥ 1
T (1) = 1 ≤ cn − d , for d ≥ 1 and c ≥ d + 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
≤ cn − d , for d ≥ 1
T (1) = 1 ≤ cn − d , for d ≥ 1 and c ≥ d + 1
For c = 2, d = 1
30

T (n) = T (bn/2c) + T (dn/2e) + 1.
T (bn/2c) ≤ c(bn/2c) − d
= cn − 2d + 1
= cn − d − d + 1
≤ cn − d , for d ≥ 1
T (1) = 1 ≤ cn − d , for d ≥ 1 and c ≥ d + 1
For c = 2, d = 1 T (n) ≤2n − 1 ∀ n ≥ 1
30
I Instead of showing T (n) = O(n) we have instead shown than

T (n) = O(cn − d), which implies T (n) = O(n)
31
Substitution method : T (n) = T (n − 1) + n
I Find a lower bound for the recurrence T (n) = T (n − 1) + n
32

I What would be a good guess?
32

I What would be a good guess? T (n) = Ω(n2 )
32

To show :T (n) ≥ cn2 ∀ n ≥ n0
32


Induction step:
T (n − 1) ≥ c(n − 1)2
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
T (n) ≥ cn2 + c − 2cn + n
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
T (n) ≥ cn2 + c − 2cn + n ( when c=1/2 , R.H.S. ≥ cn2 )
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
≥ cn2 ( for c=1/2 and n ≥ 1)
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
≥ cn2 ( for c=1/2 and n ≥ 1)
Base case:
32


Induction step:
T (n − 1) ≥ c(n − 1)2
= c(n2 + 1 − 2n) = cn2 + c − 2cn
≥ cn2 ( for c=1/2 and n ≥ 1)
Base case:
T (1) ≥ cn2
32
Solving recurrences : Recursion tree
I Recursion tree can be used to generate a good guess for the

Substitution method.
33

I Finding a good guess for T (n) = 3T (bn/4c) + Θ(n2 )
33

I Finding a good guess for T (n) = 3T (bn/4c) + Θ(n2 )
I We will ignore the floors and ceilings, because usually they
don’t affect the asymptotic order of growth.
33
Recursion tree : T (n) = 3T (n/4) + cn2
34
35
36
I Number of levels :
n
=1
4i−1
37
n
=1
4i−1
i = log4 n + 1
37
n
=1
4i−1
i = log4 n + 1
I Running time for the non-leaf levels:

2 !
2 3 3
cn 1 + + + . . . ( log4 n terms)
16 16
37
n
=1
4i−1
i = log4 n + 1

2 !
2 3 3
cn 1 + + + . . . ( log4 n terms)
16 16
2 !
2 3 3
< cn 1 + + + . . . ( infinite series sum)
16 16
37
n
=1
4i−1
i = log4 n + 1

2 !
2 3 3
cn 1 + + + . . . ( log4 n terms)
16 16
2 !
2 3 3
16 16
16 2
= cn
13
37
n
=1
4i−1
i = log4 n + 1

2 !
2 3 3
cn 1 + + + . . . ( log4 n terms)
16 16
2 !
2 3 3
16 16
16 2
= cn = O(n2 )
13
37
I Running time for leaf level nodes :
3log4 n = nlog4 3
38
3log4 n = nlog4 3
I Total running time
38
3log4 n = nlog4 3
= O(n2 ) + Θ(nlog4 3 )
38
3log4 n = nlog4 3
= O(n2 ) + Θ(nlog4 3 )
= O(n2 )
38
39
I By looking at the recursion tree we can guess T (n) = O(n2 )
40

I Then we can use the substitution method to show that our
guess is correct.
40

I Then we can use the substitution method to show that our
guess is correct.
I To prove : T (n) ≤ dn2 ∀ n ≥ n0
40
Recursion tree : T (n) = T (n/3) + T (2n/3) + O(n)
41
n
=1
3 i−1

2
42
n
=1
3 i−1

2
i = log 3 n + 1
2
42
n
=1
3 i−1

2
i = log 3 n + 1
2
I Running time :
T (n) = O(n log 3 n)
2
42
n
=1
3 i−1

2
i = log 3 n + 1
2
I Running time :
2
I Is Θ(log10 n) = Θ(log3 n)?
42
n
=1
3 i−1

2
i = log 3 n + 1
2
I Running time :
2

I Running time :
T (n) = O(n log 3 n) = O(n lg n)
2
42
n
=1
3 i−1

2
i = log 3 n + 1
2
I Running time :
2

I Running time :
T (n) = O(n log 3 n) = O(n lg n)
2
I We can again use the substitution method to show the upper

bound. 42
Polynomially larger/smaller
I For any positive (however small), is n = ω(lg n) ?
43

I Using L’Hospital’s rule:
lg n
lim =0
n→∞ n
43

I Using L’Hospital’s rule:
lg n
lim =0
n→∞ n
I Is n1+ = ω(n lg n)?
43
I n = o(n2 )
44
I n = o(n2 ) (n is asymptotically smaller)
44

I n1+ = o(n2 )
44

I n1+ = o(n2 ) (n is polynomially smaller because we can find a
positive constant )
44

positive constant )
I n = o(n lg n)
44

positive constant )
I n = o(n lg n) (n is asymptotically smaller)
44

positive constant )
I However, n1+ 6= o(n lg n)
44

positive constant )
I However, n1+ 6= o(n lg n) (n is not polynomially smaller)
44
Master method : T (n) = 3T (n/4) + cn2
45
Solving recurrences : Master method
46
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
47
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 9T (n/3) + n
48
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 9T (n/3) + n
nlogb a = nlog3 9
48
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 9T (n/3) + n
nlogb a = nlog3 9
f (n) = n =
48
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 9T (n/3) + n
nlogb a = nlog3 9
f (n) = n = O(n2− )
48
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 9T (n/3) + n
nlogb a = nlog3 9
f (n) = n = O(n2− )
T (n) = Θ(n2 )
48
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2 = n0
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2 = n0
f (n) = Θ(1)
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2 = n0
f (n) = Θ(1) = Θ(nlogb a )
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2 = n0
f (n) = Θ(1) = Θ(nlogb a )
T (n) = Θ(nlogb a lg n)
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = T (2n/3) + 1
log 3 1
nlogb a = n 2 = n0
f (n) = Θ(1) = Θ(nlogb a )
T (n) = Θ(nlogb a lg n)
= Θ(lg n)
49
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
f (n) = n lg n
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
f (n) = n lg n = Ω(nlog4 3+ ) ?
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
f (n) = n lg n = Ω(nlog4 3+ ) ?
a(n/b) lg(n/b) ≤ cn lg n
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
f (n) = n lg n = Ω(nlog4 3+ ) ?
3(n/4) lg(n/4) ≤ (3/4)n lg n , for c = 3/4
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 3T (n/4) + n lg n
nlogb a = nlog4 3
f (n) = n lg n = Ω(nlog4 3+ ) ?
3(n/4) lg(n/4) ≤ (3/4)n lg n , for c = 3/4
T (n) = Θ(n lg n)
50
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 2T (n/2) + n lg n
nlogb a = nlog2 2 = n
51
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 2T (n/2) + n lg n
f (n) = n lg n
51
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 2T (n/2) + n lg n
f (n) = n lg n = Ω(n1+ ) ?
51
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 2T (n/2) + n lg n
f (n) = n lg n = Ω(n1+ ) ?
Master method cannot be applied.
51
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )
nlogb a = nlg 7 ≈ n2.81
52
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )

f (n) = Θ(n2 )
52
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )

f (n) = Θ(n2 ) = O(nlogb a− )
52
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )

f (n) = Θ(n2 ) = O(nlogb a− )
= O(nlg 7− ) ?
52
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )

f (n) = Θ(n2 ) = O(nlogb a− )
= O(nlg 7− ) ?
T (n) = Θ(nlogb a )
52
T (n) = aT (n/b) + f (n) , a ≥ 1, b > 1
I T (n) = 7T (n/2) + Θ(n2 )

f (n) = Θ(n2 ) = O(nlogb a− )
= O(nlg 7− ) ?
T (n) = Θ(nlogb a )
= Θ(nlg 7 )
52
Proof of the master theorem
I Section 4.6 will not be a part of our syllabus
53

Chapter 4 PDF

Uploaded by

Copyright:

Available Formats

Chapter 4 PDF

Uploaded by

Document Information

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

Chapter 4 PDF

Uploaded by

Copyright:

Available Formats

1

Data Structures and Algorithms

BITS-Pilani K. K. Birla Goa Campus

Divide the problem into a number of subproblems that are

I Possible approach : either buy at the lowest price or sell at the

I Profit may not be maximized by either buying at the lowest

I We want to find the contiguous subarray having the largest

I The problem is interesting only when the array contains some

I Running time : Θ(n lg n)

I Running time : Θ(n lg n)

I C = A · B, where A, B and C are n × n matrices

I C = A · B, where A, B and C are n × n matrices

I C = A · B, where A, B and C are n × n matrices

I SQUARE-MATRIX-MULTIPLY procedure runs in Θ(n3 ) time.

I SQUARE-MATRIX-MULTIPLY procedure runs in Θ(n3 ) time.

I SQUARE-MATRIX-MULTIPLY procedure runs in Θ(n3 ) time.

I T (n) = 8T (n/2) + D(n) + C (n)

I T (n) = 8T (n/2) + D(n) + C (n)

Divide step involves finding ten n/2 × n/2 matrices (Si ).

Divide step involves finding ten n/2 × n/2 matrices (Si ).

I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + D(n) + C (n)

I T (n) = 7T (n/2) + D(n) + C (n)

Q. How would you modify Strassen’s algorithm to multiply n × n

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 <

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 < (2n)lg 7

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7 = 7nlg 7

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7 = 7nlg 7

Q. How would you modify Strassen’s algorithm to multiply n × n

nlg 7 <= mlg 7 < (2n)lg 7 = 2lg 7 nlg 7 = 7nlg 7

The above implies that mlg 7 = Θ(nlg 7 ).

I Find an upper bound for T (n) = 2T (bn/2c) + n

I Find an upper bound for T (n) = 2T (bn/2c) + n

I Find an upper bound for T (n) = 2T (bn/2c) + n

I Find an upper bound for T (n) = 2T (bn/2c) + n

I Find an upper bound for T (n) = 2T (bn/2c) + n

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)

T (bn/2c) ≤ d(bn/2c) lg(bn/2c) ≤ d(n/2) lg(n/2)

Inductive step : Assume T (n/2) ≤ d(n/2) lg(n/2)