Advanced Algorithms Analysis and Design

By

Dr. Nazir Ahmad Zafar

Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Lecture No. 8

Recurrence Relations
(Algorithms Design and Analysis Techniques)

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Today Covered
Solution: General Homogenous Recurrences when Roots are distinct Roots are repeated multiplicity of a root is k many roots with different multiplicities Non-homogenous Recurrence Relations Characteristics of various type of non-homogenous recurrence relations Solution to various non-homogenous recurrence relations Conclusion
Advanced Algorithms Analysis and Design

Dr Nazir A. Zafar

Solution to Generalized Recurrence with Distinct Roots

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Solution to Gen. Recurrence with Distinct Roots

Statement: Find the general solution of the kth order recurrence given below assuming that all roots are distinct a0 tn + a1 tn-1 + ..+ak tn-k = 0 Solution: Let Tn = xn, x is a constant as yet unknown. If we assume that Tn is a solution of equation a0 tn + a1 tn-1 + ..+ak tn-k = 0 Then, a0xn + a1xn-1 +.+ akxn-k = 0 Equation satisfied if x = 0 ,trivial solution, no interest. Otherwise, the equation is satisfied if and only if a0 + a1x1 +.+ akxk = 0
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solution to Recurrence with Distinct Roots

a0 + a1x1 +.+ akxk = 0
This equation of degree k in x is called the characteristic equation of above recurrence and P(x) = a0 + a1x1 +.+ akxk is called its characteristic polynomial Fundamental theorem of algebra states that polynomial P(x) of degree k has k roots, not necessarily distinct It means that it can be factorized as a product of k terms P(x) = (x-r1) (x-r2) (x-r3). . .(x-rk) where ri may be complex numbers. Moreover, these ri are only solutions of equation P(x) = 0
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solution to Recurrence with Distinct Roots

Consider any root ri of the characteristic polynomial

P(x) = a0 + a1x1 +.+ akxk ( x ri )

i 1

Since, p(r1) = 0, p(r2) = 0, p(r3) = 0, . . ., p(rk) = 0 Hence all x = ri , for i {1, 2, . . .,k} are solutions to above characteristic polynomial. Therefore, r1n, r2n, . . ., rkn are solution to our original recurrence relation. Since linear combination of solutions is also a solution to recurrence, therefore below is a solution.

T (n) ci ri
i 1
Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Solution to Recurrence with Distinct Roots

T (n) ci ri
i 1 k n

where, c1, c2, . . ., ck are all constants to be determined finding particular solution The remarkable fact is that this equation has only solutions of this form provided all ri are distinct. Constants can be determined from k initial conditions by solving system of k linear equations in k unknowns

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Solving Recurrence Problem with Distinct Roots

Problem: Consider the recurrence tn = n if n = 0, 1, 2 tn = 7.tn-2 + 6.tn-3 otherwise Find the general solution of the recurrence above.
Solution: First we rewrite the recurrence. tn - 7.tn-2 - 6tn-3 = 0 The characteristic equation become. x3 7x + 6 = (x + 1) (x + 2)(x - 3) The roots are: r1 = -1, r2 = - 2 and r3 = + 3 tn = c1 (-1)n + c2 (-2)n + c3 (3n)
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solving Recurrence Problem with Distinct Roots

tn = c1 (-1)n + c2 (-2)n + c3 (3n)
The initial conditions give c1 + c 2 + + c 3 = 0 - c1 - 2c2 + 3c3 = 1 c1 + 4c2 + 9c3 = 2 for n = 0 for n = 1 for n = 2

Solving these equations, we obtain c1 = -1/4, c2 = 0 and c3 = 1/4 Therefore, tn = c1 (-1/4)(-1)n + (1/4).(3)n
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solution to Generalized Recurrence with One Repeated Root

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Solution to Recurrence with one Repeated Root

Statement: If the characteristic polynomial P(x) = a0 + a1x1 +.+ akxk then conclude that if r is a double root then tn = rn

and tn = n rn are both solutions to recurrence.

Solution: It can be found as in case of second order linear homogenous recurrence relation. Since r is a multiple root. By definition of multiple roots, there exists a polynomial q(x) of degree k-2 such that the following holds p(x) = (x r)2q(x), for every n k
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solution to Recurrence with one Repeated Root

Consider the kth degree polynomials un (x) = a0 xn + a1 xn-1 + . . . + ak xn-k and vn (x) = a0 n xn + a1 (n-1)xn-1 + . . . + ak (n-k)xn -k It is to be noted that vn(x) = x.un (x), where un (x) denotes the derivative of un (x) with respect to x . But un (x) can be written as un (x) = xn-k p(x) = xn-k (x-r)2q(x) = (x-r)2(xn-k q(x)) Using rule for computing the derivative of a product of functions, we obtain that derivative of un(x) with respect to x is given by
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Solution to Recurrence with one Repeated Root

un (x) = 2 (x-r) xn-k q(x) +(x-r)2(xn-k q(x)) therefore un (r) = 0, which implies that vn (r) = r . un (r) = 0 for all n k. It means: a0 n rn + a1 (n-1) rn-1 +. . .+ ak (n-k) rn-k = 0 Hence, tn = n rn is also solution to the recurrence. Now we conclude that if r is a double root then tn = rn and tn = n rn are both solutions to the recurrence. Rest of k-2 are all distinct roots hence general solution

tn c1r c2nr b r b2r2 ... bk 2rk 2

n n n 11 n
Dr Nazir A. Zafar

where c1, c2, b1, b2,. . ., and bk-2 are all real constants
Advanced Algorithms Analysis and Design

Higher Order Homogenous Recurrence with k-multiplicity of a Root

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Solution to Recurrence with Root of Multiplicity k

Now if we have to solve recurrence order k, then to solve polynomial, degree k, given below is sufficient Statement: If the characteristic polynomial P(x) = a0 + a1x1 +.+ akxk has r as double root then it can be written as p(x) = (x r)2q(x), for every n k and solutions are: rn and n rn has r as triple root then it can be written as p(x) = (x r)3q1(x), for every n k and solutions are: rn, n.rn and n2.rn
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

K-Multiplicity of a Root: General Result

r has multiplicity k then it can be written as p(x) = (x r)k, for every n k and solutions are: rn, n.rn, n2.rn,. . ., nk-1.rn then general solution is

t n c1r c2 nr ... ck n r
n n

k 1 n

tn c j n r
j 1

j 1 n

where b1, b2,. . ., bk are all real constants

Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Multiplicity of Roots: More General Result

If there are l roots, r1, r2, . . ., rl with multiplicities m1, m2, . . ., ml respectively, of the polynomial:
P(x) = a0 + a1x1 +. . .+ akxk s.t. m1 + m2 + . . .+ ml = k then the general solution to the recurrence is

t n c11r1 c12 nr1 c13n r1 ... c1m1 n

n n 2 n n n 2 n

m1 1 n 1 m2 1

r r2
n

c21r2 c22 nr2 c23n r2 ... c2 m2 n ... cl1rl cl 2 nrl cl 3n rl ... clml n
n n 2 n
Dr Nazir A. Zafar

ml 1 n l

Advanced Algorithms Analysis and Design

Contd..
t n c11r1 c12 nr1 c13n r1 ... c1m1 n
n n 2 n n n 2 n m1 1 n 1 m2 1

r r2
n

c21r2 c22 nr2 c23n r2 ... c2 m2 n ... cl1rl cl 2 nrl cl 3n rl ... clml n
n n 2 n

ml 1 n l

tn c1 j n r c2 j n r2 ... clj n r
j 1 j 1 n 1 j 1 n j 1 j 1

m1

m2

ml

j 1 n l

tn cij n r
i 1 j 1
Dr Nazir A. Zafar

mi

j 1 n i

where all ci, j are constants

Advanced Algorithms Analysis and Design

Problem
Statement: Consider the recurrence tn = n if n = 0, 1, 2 tn = 5tn-1 - 8tn-2 + 4tn-3 otherwise Find general solution of the recurrence above.
Solution: First we rewrite the recurrence. tn - 5tn-1 + 8tn-2 - 4tn-3 = 0 The characteristic equation become. x3 5x2 + 8x -4 = (x-1) (x-2)2 The roots are: r1 = 1 of multiplicity m1 = 1 and r2 = 2 of multiplicity m2 = 2, and hence the general solution is tn = c 1 1 n + c 2 2 n + c 3 n 2 n
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Example
The initial conditions give c1 + c2 = 0 c1 + 2c2 + 2c3 = 1 c1 + 4c2 + 8c3 = 2

for n = 0 for n = 1 for n = 2

Solving these equations, we obtain

c1 = -2, c2 = 2 and c3 = -1/2

Therefore, tn = c 1 1 n + c 2 2 n + c 3 n 2 n = -2 + 2.2n .n.2n = 2n+1 n.2n-1 - 2

Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Non-homogeneous Recurrence of Higher Order

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Non-homogenous Recurrence
Solution of a linear recurrence with constant coefficients becomes more difficult when the recurrence is not homogeneous That is when linear combination is not equal to zero In particular, it is no longer true that any linear combination of solutions is a solution. Consider the following recurrence
a0 tn + a1 tn-1 + . . .+ ak tn-k = bn p(n) The left-hand side is the same as before, but on the right-hand side we have bnp(n) where b is a constant and p(n) is a polynomial in n of degree d.
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Generalization: Non-homogeneous Recurrences

If a recurrence is of the form

a0t n a1t n 1 . . . ak t n k b p1 (n)

n 1

Then the characteristics polynomial is

(a0 x k a1 x k 1 . . . ak )( x b1d 1 )
Which contains one factor for the left hand side And other factor corresponding to the right hand side, where d is degree of polynomial Characteristics polynomial can be used to solve the above recurrence.
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Problem 1: Non-homogeneous Recurrences

Problem: Consider the recurrence below. Find its solution tn 2tn-1 = 3n Solution: Compare the above recurrence with a0 tn + a1 tn-1 + . . .+ ak tn-k = bn p(n) Here: b = 3, p(n) = 1, a polynomial of degree 0. Reducing to homogeneous case, we are given tn 2tn-1 = 3n (1) Replace n by n - 1 and multiply by 3, we obtain tn-1 2tn-2 = 3n-1 3tn-1 6tn-2 = 3n (2)
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Problem 1: Non-homogeneous Recurrences

From (1) and (2) tn 2tn-1 = 3n (1) + 3tn-1 6tn-2 = 3n (2) Subtracting 2 from equation 1, we get tn - 5tn-1 + 6tn-2 = 0 The characteristic equation is x2 - 5x + 6 = 0 Roots of this equation are: x = 2 and x = 3 And therefore general solution is tn = c 1 2 n + c 2 3 n
Advanced Algorithms Analysis and Design

Dr Nazir A. Zafar

Problem 1: Non-homogeneous Recurrences

It is not always true that an arbitrary choice of c1 and c2 produces a solution to the recurrence even when initial conditions are not taken into account. Note: It is to be noted that solutions: tn = 2n and tn = 3 n which are solutions to reduced recurrence, are not solution to original one What is the reason?
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Problem 2: Non-homogeneous Recurrences

Problem Find general solution of the following recurrence. tn - 2tn-1 = (n + 5) 3n n 1 Solution: The manipulation needed to transform this into a homogeneous recurrence is slightly more complicated than with first example. tn - 2tn-1 = (n + 5) 3n n 1 (1) replace n by n-1, n-2, we get tn-1 - 2tn-2 = (n + 4) 3n-1 n 2 (2) tn-2 - 2tn-3 = (n + 3) 3n-2 n 3 (3)
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Problem 2: Non-homogeneous Recurrences

Above equations can be written as tn - 2tn-1 = 9(n + 5) 3n-2 n 1 (4) tn-1 - 2tn-2 = 3(n + 4) 3n-2 n 2 (5) tn-2 - 2tn-3 = (n + 3) 3n-2 n 3 (6) Our objective is to eliminate the right hand side of the above equations to make it homogenous. Multiply (5) by -6, and (6) by 9 we get tn - 2tn-1 - 6tn-1 + 12tn-2 + 9tn-2 - 18tn-3
Dr Nazir A. Zafar

= 9(n + 5) 3n-2 = -18(n + 4) 3n-2 = 9(n + 3) 3n-2

Advanced Algorithms Analysis and Design

Problem 2: Non-homogeneous Recurrences

After simplification, the above equations can be written as tn - 2tn-1 - 6tn-1 + 12tn-2 + 9tn-2 - 18tn-3 = (9n + 45) 3n-2 = (-18n 72) 3n-2 = (9n + 27) 3n-2

Adding these equation, we get homogenous equation, which can be solved easily tn - 8tn-1 + 21tn-2 - 18tn-3 = 0
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design

Problem 2: Non-homogeneous Recurrences

tn 8tn-1 + 21tn-2 - 18tn-3 = 0

The characteristics equation of the above homogenous equation is: x3 8x2 +21x -18 = 0 ( x-2) (x-3)2 = 0 and hence, x = 2, 3, 3 General solution is: tn = c1 2n + c2 3n + c3 n 3n For n = 0, 1, 2 We can find values of c1, c2, c3 and then tn = (t0 - 9) 2n + (n + 3)3n+1
Advanced Algorithms Analysis and Design

Dr Nazir A. Zafar

Tower of Hanoi: Problem 3

Tower of Hanoi is a mathematical game or puzzle. It consists of three towers, a number of disks of different sizes which can slide onto any tower. The puzzle starts with disks stacked in order of size on one tower, smallest at top, making a conical shape. Objective is to move entire stack to another tower, obeying following rules: Only one disk may be moved at a time. Each move consists of taking upper disk from one of towers and sliding it onto another tower You can put on top of other disks already present No disk may be placed on top of a smaller disk.
Advanced Algorithms Analysis and Design

Dr Nazir A. Zafar

More Generalized Non-homogeneous Recurrences

If a recurrence is of the form

a0t n a1t n 1 . . . ak t n k b p1 (n) b p2 (n) ...

n 1 n 2

Then the characteristics polynomial is

d (a0 x k a1 x k 1 . . . ak )( x b1d1 1 )( x b2 2 1 )...,

Which contains one factor for the left hand side And other factor corresponding to the each term on right hand side. Once the characteristics polynomial is obtained the recurrence can be solved as before.

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Problem 6 : Non-homogeneous Recurrences

Consider the recurrence tn = 2tn-1 + n + 2n otherwise
Solution: Compare the recurrence: tn - 2tn-1 = n + 2n with

a0t n a1t n 1 . . . ak t n k b p1 (n) b p2 (n) ...

n 1 n 2

Here, b1 = 1, p1(n) = n, b2 = 2, and p2(n) = 1. Degree of p1(n) = d1 = 1, Degree of p2(n) = d2 = 0.

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Problem 6: Non-homogeneous Recurrences

The characteristic polynomial: (x-2) (x-1)2 (x-2) The roots are, x = 1, 2, both of multiplicity 2. All solutions of recurrence therefore have form tn = c1 1n + c2 n1n + c3 2n + c4 n 2n n + 2n = (2c2 - c1 ) c2 n + c4 2n For n = 0, 1, 2, 3 c1, c2, c3 and c4 can be solved and hence solution is tn= n.2n +2n+1 n -2

Dr Nazir A. Zafar

Advanced Algorithms Analysis and Design

Conclusion

Recursive relations are important because used in divide and conquer, dynamic programming and in many other e.g. optimization problems Analysis of such algorithms is required to compute complexity Recursive algorithms may not follow homogenous behavior, it means there must be some cost in addition to recursive cost Solution to homogenous recursive relations is comparatively easy, but not so easy in case of non-homogenous recursive relations
Dr Nazir A. Zafar Advanced Algorithms Analysis and Design