Data Structures and Algorithms

CS245-2016S-03
Recursive Function Analysis
David Galles

Department of Computer Science

University of San Francisco
03-0: Algorithm Analysis
for (i=1; i<=n*n; i++)
for (j=0; j<i; j++)
sum++;
03-1: Algorithm Analysis
for (i=1; i<=n*n; i++) Executed n*n times
for (j=0; j<i; j++) Executed <= n*n times
sum++; O(1)

Running Time: O(n4 )

03-2: Algorithm Analysis
for (i=1; i<=n*n; i++)
for (j=0; j<i; j++)
sum++;

Exact # of times sum++ is executed:

2
n
! n2 (n2 + 1)
i =
i=1
2
n4 + n2
=
2
(n4 )
03-3: Recursive Functions
long power(long x, long n) {
if (n == 0)
return 1;
else
return x * power(x, n-1);
}
Ans: theta(n)
03-4: Recurrence Relations
T (n) = Time required to solve a problem of size n
Recurrence relations are used to determine the
running time of recursive programs recurrence
relations themselves are recursive
T (0) = time to solve problem of size 0
Base Case
T (n) = time to solve problem of size n
Recursive Case
03-5: Recurrence Relations
long power(long x, long n) {
if (n == 0)
return 1;
else
return x * power(x, n-1);
}
T (0) = c1 for some constant c1
T (n) = c2 + T (n 1) for some constant c2
03-6: Solving Recurrence Relations

T (0) = c1
T (n) = T (n 1) + c2
If we knew T (n 1), we could solve T (n).

T (n) = T (n 1) + c2
03-7: Solving Recurrence Relations

T (0) = c1
T (n) = T (n 1) + c2
If we knew T (n 1), we could solve T (n).
T (n) = T (n 1) + c2 T (n 1) = T (n 2) + c2
= T (n 2) + c2 + c2
= T (n 2) + 2c2
03-8: Solving Recurrence Relations

T (0) = c1
T (n) = T (n 1) + c2
If we knew T (n 1), we could solve T (n).
T (n) = T (n 1) + c2 T (n 1) = T (n 2) + c2
= T (n 2) + c2 + c2
= T (n 2) + 2c2 T (n 2) = T (n 3) + c2
= T (n 3) + c2 + 2c2
= T (n 3) + 3c2
03-9: Solving Recurrence Relations

T (0) = c1
T (n) = T (n 1) + c2
If we knew T (n 1), we could solve T (n).
T (n) = T (n 1) + c2 T (n 1) = T (n 2) + c2
= T (n 2) + c2 + c2
= T (n 2) + 2c2 T (n 2) = T (n 3) + c2
= T (n 3) + c2 + 2c2
= T (n 3) + 3c2 T (n 3) = T (n 4) + c2
= T (n 4) + 4c2
03-10: Solving Recurrence Relations

T (0) = c1
T (n) = T (n 1) + c2
If we knew T (n 1), we could solve T (n).
T (n) = T (n 1) + c2 T (n 1) = T (n 2) + c2
= T (n 2) + c2 + c2
= T (n 2) + 2c2 T (n 2) = T (n 3) + c2
= T (n 3) + c2 + 2c2
= T (n 3) + 3c2 T (n 3) = T (n 4) + c2
= T (n 4) + 4c2
= ...
= T (n k) + kc2
03-11: Solving Recurrence Relations

T (0) = c1
T (n) = T (n k) + k c2 for all k
If we set k = n, we have:
T (n) = T (n n) + nc2
= T (0) + nc2
= c1 + nc2
(n)
03-12: Building a Better Power
long power(long x, long n) {
if (n==0) return 1;
if (n==1) return x;
if ((n % 2) == 0)
return power(x*x, n/2);
else
return power(x*x, n/2) * x;
}
theta(log(n))
03-13: Building a Better Power
long power(long x, long n) {
if (n==0) return 1;
if (n==1) return x;
if ((n % 2) == 0)
return power(x*x, n/2);
else
return power(x*x, n/2) * x;
}
T (0) = c1
T (1) = c2
T (n) = T (n/2) + c3
(Assume n is a power of 2)
03-14: Solving Recurrence Relations

T (n) = T (n/2) + c3
03-15: Solving Recurrence Relations

T (n) = T (n/2) + c3 T (n/2) = T (n/4) + c3

= T (n/4) + c3 + c3
= T (n/4) + 2c3
03-16: Solving Recurrence Relations

T (n) = T (n/2) + c3 T (n/2) = T (n/4) + c3

= T (n/4) + c3 + c3
= T (n/4) + 2c3 T (n/4) = T (n/8) + c3
= T (n/8) + c3 + 2c3
= T (n/8) + 3c3
03-17: Solving Recurrence Relations

T (n) = T (n/2) + c3 T (n/2) = T (n/4) + c3

= T (n/4) + c3 + c3
= T (n/4) + 2c3 T (n/4) = T (n/8) + c3
= T (n/8) + c3 + 2c3
= T (n/8) + 3c3 T (n/8) = T (n/16) + c3
= T (n/16) + c3 + 3c3
= T (n/16) + 4c3
03-18: Solving Recurrence Relations

T (n) = T (n/2) + c3 T (n/2) = T (n/4) + c3

= T (n/4) + c3 + c3
= T (n/4) + 2c3 T (n/4) = T (n/8) + c3
= T (n/8) + c3 + 2c3
= T (n/8) + 3c3 T (n/8) = T (n/16) + c3
= T (n/16) + c3 + 3c3
= T (n/16) + 4c3 T (n/16) = T (n/32) + c3
= T (n/32) + c3 + 4c3
= T (n/32) + 5c3
03-19: Solving Recurrence Relations

T (n) = T (n/2) + c3 T (n/2) = T (n/4) + c3

= T (n/4) + c3 + c3
= T (n/4) + 2c3 T (n/4) = T (n/8) + c3
= T (n/8) + c3 + 2c3
= T (n/8) + 3c3 T (n/8) = T (n/16) + c3
= T (n/16) + c3 + 3c3
= T (n/16) + 4c3 T (n/16) = T (n/32) + c3
= T (n/32) + c3 + 4c3
= T (n/32) + 5c3
= ...
= T (n/2k ) + kc3
03-20: Solving Recurrence Relations

T (0) = c1
T (1) = c2
T (n) = T (n/2) + c3
T (n) = T (n/2k ) + kc3
We want to get rid of T (n/2k ). Since we know T (1) ...

n/2k = 1
n = 2k
lg n = k
03-21: Solving Recurrence Relations

T (1) = c2
T (n) = T (n/2k ) + kc3
Set k = lg n:

lg n
T (n) = T (n/2 ) + (lg n)c3
= T (n/n) + c3 lg n
= T (1) + c3 lg n
= c2 + c3 lg n
(lg n)
03-22: Power Modifications
long power(long x, long n) {
if (n==0) return 1;
if (n==1) return x;
if ((n % 2) == 0)
return power(x*x, n/2);
else
return power(x*x, n/2) * x;
}
03-23: Power Modifications
long power(long x, long n) {
if (n==0) return 1; < stuck at this loop, always hit (n%2) == 0,
if (n==1) return x; wont reach the base case
the inner power(x,2) part
if ((n % 2) == 0)
return power(power(x,2), n/2);
else
return power(power(x,2), n/2) * x;
}
This version of power will not work. Why?
03-24: Power Modifications
long power(long x, long n) {
if (n==0) return 1; < because it stuck at this loop too, wont
if (n==1) return x; hit the base case, keep calling the
recursive function,
if ((n % 2) == 0) the inner power(x,n/2), 2) < this part

return power(power(x,n/2), 2);

else
return power(power(x,n/2), 2) * x;
}
This version of power also will not work. Why?
03-25: Power Modifications
long power(long x, long n) {
if (n==0) return 1;
if (n==1) return x;
if ((n % 2) == 0)
return power(x,n/2) * power(x,n/2);
else
return power(x,n/2) * power(x,n/2) * x;
} this version works, it will eventually hit the base case

This version of power does work.

What is the recurrence relation that describes its
running time? theta(n)
03-26: Power Modifications
long power(long x, long n) {
if (n==0) return 1;
if (n==1) return x;
if ((n % 2) == 0)
return power(x,n/2) * power(x,n/2);
else
return power(x,n/2) * power(x,n/2) * x;
}
T (0) = c1
T (1) = c2
T (n) = T (n/2) + T (n/2) + c3
= 2T (n/2) + c3
(Again, assume n is a power of 2)
03-27: Solving Recurrence Relations

T (n) = 2T (n/2) + c3 T (n/2) = 2T (n/4) + c3

= 2[2T (n/4) + c3 ]c3
= 4T (n/4) + 3c3 T (n/4) = 2T (n/8) + c3
= 4[2T (n/8) + c3 ] + 3c3
= 8T (n/8) + 7c3
= 8[2T (n/16) + c3 ] + 7c3
= 16T (n/16) + 15c3
= 32T (n/32) + 31c3
...
= 2k T (n/2k ) + (2k 1)c3
03-28: Solving Recurrence Relations

T (0) = c1
T (1) = c2
T (n) = 2k T (n/2k ) + (2k 1)c3
Pick a value for k such that n/2k = 1:
n/2k = 1
n = 2k
lg n = k
T (n) = 2lg n T (n/2lg n ) + (2lg n 1)c3
= nT (n/n) + (n 1)c3
= nT (1) + (n 1)c3
= nc2 + (n 1)c3
(n)
03-29: Recursion Trees
We can also do this substitution visually, leads to
Recursion Trees
Consider:

T (n) = 2T (n/2) + Cn
T (1) = C2
T (0) = C2
03-30: Recursion Trees
Start with the recursive definition

T(n) = Cn + 2T(n/2)
03-31: Recursion Trees
Move the equation around a bit to get:

T(n) = Cn
+ +

T(n/2) T(n/2)

Repalce each occurance of T (n/2) with

T (n/4) + T (n/4) + C(n/2)
03-32: Recursion Trees
T(n) = Cn
+ +

C(n/2) C(n/2)
+ + + +

T(n/4) T(n/4) T(n/4) T(n/4)

Replace again, using T (n) = 2T (n/2) + Cn

03-33: Recursion Trees
T(n) = Cn
+ +

C(n/2) C(n/2)
+ + + +

C(n/4) C(n/4) C(n/4) C(n/4)

+ + + + + +
+ +

T(n/8) T(n/8) T(n/8) T(n/8) T(n/8) T(n/8) T(n/8) T(n/8)

If we continue replacing ...

 03-34: Recursion Trees
Totals for each
level in the tree:
T(n) = Cn Cn

C(n/2) C(n/2) Cn
lg n

C(n/4) C(n/4) C(n/4) C(n/4) Cn

C(n/8) C(n/8) C(n/8) C(n/8) C(n/8) C(n/8) C(n/8) C(n/8) Cn

...

...

...

...

...
C2 C2 C2 C2 C2 C2 C2 C2 C2 C2 C2 C2 C2 ... C2 C2 C2 C2 C2 C2 C2 C2 n
Cn( lg n-1) + C 2n
lg n
2 = n leaves (n lg n )
03-35: Recursion Trees

T (1) = C1
T (n) = T (n 1) + C2
03-36: Recursion Trees

T (0) = C1
T (1) = C1
T (n) = T (n/2) + C2
03-37: Substitution Method
We can prove that a bound is correct using
induction, this is the substituion method

T (1) = C1
T (n) = T (n 1) + C2

Show: T (n) O( ? )
03-38: Substitution Method
We can prove that a bound is correct using
induction, this is the substituion method

T (1) = C1
T (n) = T (n 1) + C2

Show: T (n) O(n), that is:

T (n) C n for all n > n0 ,
for some pair of constants C, n0
03-39: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) O(n), that is, T (n) C n

Base case: T (1) = C1 C 1 for some constant

C
This is true as long as C C1 .
03-40: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) O(n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

03-41: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) O(n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

C(n 1) + C2 Inductive hypothesis
03-42: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) O(n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

C(n 1) + C2 Inductive hypothesis
Cn + (C2 C) Algebra
Cn If C > C2
This is true as long as C C1 .
Should be C >= C2
03-43: Substitution Method
We can prove that a bound is correct using
induction, this is the substituion method

T (1) = C1
T (n) = T (n 1) + C2

Show: T (n) (n)

T (n) C n for all n > n0 ,
for some pair of constants C, n0
03-44: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) (n), that is, T (n) C n

Base case: T (1) = C1 C 1 for some constant

C
This is true as long as C C1 .
03-45: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) (n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

03-46: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) (n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

C(n 1) + C2 Inductive hypothesis
03-47: Substitution Method
T (1) = C1
T (n) = T (n 1) + C2
Show: T (n) (n), that is, T (n) C n

Recursive case:

T (n) = T (n 1) + C2 Recurrence definition

C(n 1) + C2 Inductive hypothesis
Cn + (C2 C) Algebra
Cn If C C2
This is true as long as C C1 .
03-48: Substitution Method

T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n

Show: T (n) O(n lg n), that is, T (n) C n lg n

03-49: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
Show: T (n) O(n lg n), that is, T (n) C n lg n

Base cases:
T (0) = C1 C 0 lg 0 for some constant C
T (1) = C1 C 1 lg 1 for some constant C

Hmmm....
03-50: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
Show: T (n) O(n lg n), that is, T (n) C n lg n

Only care about n > n0 . We can pick 2, 3 as base

cases (why?)
T (2) = C1 C 2 lg 2 for some constant C
T (3) = C1 C 3 lg 3 for some constant C
03-51: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
T (n) = 2T (n/2) + C1 n Recurrence Definition
03-52: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
T (n) = 2T (n/2) + C1 n Recurrence Definition
2C(n/2) lg(n/2) + C1 n Inductive hypothesis
03-53: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
T (n) = 2T (n/2) + C1 n Recurrence Definition
2C(n/2) lg(n/2) + C1 n Inductive hypothesis
Cn lg n/2 + C1 n Algebra
Cn lg n Cn lg 2 + C1 n Algebra
Cn lg n Cn + C1 n Algebra
03-54: Substitution Method
T (0) = C2
T (1) = C2
T (n) = 2T (n/2) + C1 n
T (n) = 2T (n/2) + C1 n Recurrence Definition
2C(n/2) lg(n/2) + C1 n Inductive hypothesis
Cn lg n/2 + C1 n Algebra
Cn lg n Cn lg 2 + C1 n Algebra
Cn lg n Cn + C1 n Algebra
Cn lg n If C > C1
03-55: Substitution Method
Sometimes, the math doesnt work out in the
substitution method:

T (1) = 1
"n#
T (n) = 2T +1
2

(Work on board)
03-56: Substitution Method
Try T (n) cn:

"n#
T (n) = 2T 1
" n2#
2c +1
2
cn + 1
<= cn ? (so we know it is invalid proof to assume T(n) <= cn

We did not get back T (n) cn that extra +1 term

means the proof is not valid. We need to get back
exactly
$n what we started with (see invalid proof of
i=1 i O(n) for why this is true)
03-57: Substitution Method
Try T (n) cn b:

"n#
T (n) = 2T +1
" "2n # #
2 c b +1
2
cn 2b + 1
cn b

As long as b 1
03-58: Master Method
Recursion Tree for: T (n) = 2T (n/4) + C2
 03-59: Master Method
Totals for each
level in the tree:
T(n) = C C
2 2
log4 n levels in the tree

C C 2C
2 2 2

C C C C 4C
2 2 2 2 2

C C C C C C C C 8C
2 2 2 2 2 2 2 2 2

...
k
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
2 C
2

...
... + 2
log 4 n
C
C C C C C C C C C C C C 2
1 1 1 1 1 1 1 1 1 1 1 1

( n 1/2 )
log 4 n log 4 2 1/2
2 = n = n leaves in the tree
03-60: Master Method
Recursion Tree for: T (n) = 2T (n/2) + n2

IMPORTANT, practice this again

03-61: Master Method
Totals for each
level in the tree:
n2 n2
T(n) =
+ +

2 2 n2
log2 n levels in the tree

(n/2) (n/2)
2
+ + + +

n2
2 2 2 2
(n/4) (n/4) (n/4) (n/4) 4

+ + + + + +
+ +

2 2 2 2 2 2 2 2 n2
(n/8) (n/8) (n/8) (n/8) (n/8) (n/8) (n/8) (n/8)
8

...
n2
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
k
2

...
C1 C1 C1 C1 ... C1 C1 C1 C1
+ 2
log 4 n
C
1

log 2 n log 2 2 (n2 )

2 = n = n leaves in the tree
03-62: Master Method
Recursion Tree for: T (n) = 4T (n/2) + n2
 03-63: Master Method
Totals for each
level in the tree:
log2 n levels in the tree

T(n) = n2
n2

(n/2) 2 (n/2) 2 (n/2) 2 (n/2) 2 n2

(n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 (n/4) 2 n2

...
...
...
...
...

...
...
...
...

...
...
...
...

...
...
...
...
n2

...
1 1 1 1 1 1 1 1 + n2

log 2 n log 2 4
n 2 lg n
2
4 = n = n leaves in the tree
03-64: Master Method
Recursion Tree for: T (n) = aT (n/b) + f (n)
 03-65: Master Method
Totals for each
level in the tree:
logb n levels in the tree

f(n)
# of children = a f(n)

f (n/b)
... ... f (n/b) ... ... f (n/b) # of chldren = a a f (n/b)

2
f (n/b 2 ) f (n/b 2 ) f (n/b ) f (n/b 2 ) f (n/b 2 ) f (n/b 2 ) f (n/b 2 ) f (n/b 2 ) f (n/b 2 ) 2 2
# of chldren = a a f (n/b )

...
k k
...

...

...
...

...

...

...
...

...
a f (n/b )

...
C C C C ... C C C C
C n
log b a

log b n log b a
a = n leaves in the tree
03-66: Master Method
T (n) = aT (n/b) + f (n)

1. if f (n) O(nlogb a ) for some > 0, then

T (n) (nlogb a )
2. if f (n) (nlogb a ) then T (n) (nlogb a lg n)
3. if f (n) (nlogb a+ ) for some > 0, and if
af (n/b) cf (n) for some c < 1 and large n, then
T (n) (f (n))
03-67: Master Method
T (n) = 9T (n/3) + n
03-68: Master Method
T (n) = 9T (n/3) + n

a = 9, b = 3, f (n) = n
nlogb a = nlog3 9 = n2
n O(n2 )

T (n) = (n2 )
03-69: Master Method
T (n) = T (2n/3) + 1
03-70: Master Method
T (n) = T (2n/3) + 1 < f(n) = 1

Note, b = 3/2

a = 1, b = 3/2, f (n) = 1
n^log(3/2) 1 =

nlogb a = nlog3/2 1 = n0 = 1 logbase1 (1)/ logbase1 (3/2)

= 0/any number
1 O(1) =0
hence n^0

T (n) = (1 lg n) = (lg n)
03-71: Master Method
T (n) = 3T (n/4) + n lg n
03-72: Master Method
T (n) = 3T (n/4) + n lg n

a = 3, b = 4, f (n) = n lg n
nlogb a = nlog4 3 = n0.792
n lg n (n0.792+ )
3(n/4) lg(n/4) c n lg n

T (n) (n lg n)
03-73: Master Method
T (n) = 2T (n/2) + n lg n
03-74: Master Method
T (n) = 2T (n/2) + n lg n

a = 2, b = 2, f (n) = n lg n
nlogb a = nlog2 2 = n1
Master method does not apply!
n1+ grows faster than n lg n for any > 0
Logs grow incredibly slowly! lg n o(n ) for any > 0