Dynamic Programming

Introduction

Prepared by Pawan K. Mishra

 Inductive definition

Recursive program

Prepared by Pawan K. Mishra

 Inductive definition
Fact(n)
Fact(0)=1
Fact(n)= n-1 * Fact(n)

sort(A)
sort([])=[]
sort(a,b,c,d)= fit( d, sort[a,b,c])

Prepared by Pawan K. Mishra

 Recursive program

factorial(n):
if (n <= 0) return(1)
else
return (n*factorial(n-1))

Prepared by Pawan K. Mishra

 Optimal substructure property
Solution to original problem can be derived by combining
solutions to subproblems

factorial(n-1) is a subproblem of factorial(n)

So are factorial(n-2), factorial(n-3), …,
factorial(0)

sort([b,c,d]) is a subproblem of sort([a,b,c,d])

So are sort([c,d]), sort([b,c])….
Prepared by Pawan K. Mishra
 Fibonacci numbers
Inductive definition Recursive Program

fib(0) = 0
function fib(n):
if (n == 0) or (n == 1)
fib(1) = 1
value = n
fib(n) = fib(n-1) + fib(n-2) else
value = fib(n-1) + fib(n-2)

return(value)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)

fib(2) fib(1)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)

fib(2) fib(1)

fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)

fib(2)
fib(1)
1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)
1

fib(2) fib(1)
1 0
fib(1) fib(0)
fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2)
return(value) fib(3) fib(2)
1 1

fib(2) fib(1)
1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2
return(value) fib(3) fib(2)
1 1

fib(2) fib(1)
1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2
return(value) fib(3) fib(2)
1 1

fib(2) fib(1) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2
return(value) fib(3) fib(2)
1 1 1 0

fib(2) fib(1) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1
return(value) fib(3) fib(2)
1 1 1 0

fib(2) fib(1) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1
return(value) fib(3) fib(2)
1 1 1 0

fib(2) fib(1) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0

fib(2) fib(1) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
if n == 0 or n == 1 fib(5)
value = n 3 2
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
5
if n == 0 or n == 1 fib(5)
value = n 3 2
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
5
if n == 0 or n == 1 fib(5)
value = n 3 2
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0

fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)

1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Computing fib(5)
function fib(n):
5
if n == 0 or n == 1 fib(5)
value = n 3 2
else
fib(4) fib(3)
value = fib(n-1) +
fib(n-2) 2 1 1 1
return(value) fib(3) fib(2) fib(2) fib(1)
1 1 1 0 1 0
Time complexity ?? fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)
1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 OBSERVATIONS
Overlapping 5
subproblems 3
fib(5)
2

Wasteful fib(4) fib(3)

recomputation 2 1 1 1

fib(3) fib(2) fib(2) fib(1)

Computation 1 1 1 0 1 0
tree grows fib(2) fib(1) fib(1) fib(0) fib(1) fib(0)
exponentially 1 0
fib(1) fib(0)

Prepared by Pawan K. Mishra

 Never re-evaluate a subproblem
Build a table of values already computed
Memory table

Memoization

Remind yourself that this value has already been seen

before

Prepared by Pawan K. Mishra

 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
table
-Look up table Kk Fib(k)
fib(k)
before starting a
recursive
computation
-Computation
tree is linear

Prepared by Pawan K. Mishra

 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table Kk Fib(k)
fib(k)
before starting a
recursive
computation
-Computation
tree is linear

Prepared by Pawan K. Mishra

 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table Kk Fib(k)
fib(k)
before starting a fib(3) fib(2)
recursive
computation
-Computation
tree is linear

Prepared by Pawan K. Mishra

 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table Kk Fib(k)
fib(k)
before starting a fib(3) fib(2)
recursive
computation
fib(2) fib(1)
-Computation
tree is linear

Prepared by Pawan K. Mishra

 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table Kk Fib(k)
fib(k)
before starting a fib(3) fib(2)
recursive
computation
fib(2) fib(1)
-Computation
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive
computation
fib(2) fib(1)
-Computation 1
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 0 0
computation
fib(2) fib(1)
-Computation 1 0
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 1 0
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 1 0
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
2
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 3 2
0
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly
computed value in a
fib(4) fib(3)
table
2 1
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 3 2
0
tree is linear
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly 3
computed value in a
fib(4) fib(3)
table
2 1
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 3 2
0
tree is linear 4 3
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization
fib(5)
-Store each newly 3 2
computed value in a
fib(4) fib(3)
table
2 1
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 3 2
0
tree is linear 4 3
fib(1) fib(0)
Prepared by Pawan K. Mishra
 Memoized fib(5)
Memoization 5
fib(5)
-Store each newly 3 2
computed value in a
fib(4) fib(3)
table
2 1
-Look up table K k Fib(k)
fib(k)
before starting a fib(3) fib(2) 1 1
recursive 1 1 0 0
computation 2 1
fib(2) fib(1)
-Computation 3 2
0
tree is linear 4 3
fib(1) fib(0) 5 5
Prepared by Pawan K. Mishra
 Memoized fibonacci
function fib(n):
if fibtable[n]
Extra
return(fibtable[n])
if n == 0 or n == 1
value = n
else
value = fib(n-1) + fib(n-2)
fibtable[n] = value Extra

return(value)
Prepared by Pawan K. Mishra
 In general
function f(x,y,z):

if ftable[x][y][z]
return(ftable[x][y][z])

value = expression in terms of subproblems

ftable[x][y][z] = value return(value)

Prepared by Pawan K. Mishra

 In general
function f(x,y,z):

if ftable[x][y][z]
return(ftable[x][y][z])

value = expression in terms of subproblems

Need to solve recursively
exploiting the inductive
ftable[x][y][z] = value return(value) definition.

Prepared by Pawan K. Mishra

 Dynamic programming

Anticipate what the

memory table looks like

Subproblems are
known from problem
structure

Dependencies form a
dag

Solve subproblems
in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like

Subproblems are
known from problem
structure

Dependencies form a
dag

Solve subproblems
in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like
fib(4)
Subproblems are
known from problem fib(3)
structure
fib(2)
Dependencies form a
dag fib(1)

Solve subproblems fib(0)

in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like
fib(4)
Subproblems are
known from problem fib(3)
structure
fib(2)
Dependencies form a
dag fib(1)

Solve subproblems fib(0)

in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like
fib(4)
Subproblems are
known from problem fib(3)
structure
fib(2)
Dependencies form a
dag fib(1)

Solve subproblems fib(0)

in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like
fib(4)
Subproblems are
known from problem fib(3)
structure
fib(2)
Dependencies form a
dag fib(1)

Solve subproblems fib(0)

in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the fib(5)

memory table looks like
fib(4)
Subproblems are
known from problem fib(3)
structure
fib(2)
Dependencies form a
dag fib(1)

Solve subproblems fib(0)

in topological order
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k)
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0 1
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0 1 1
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0 1 1 2
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0 1 1 2 3
Prepared by Pawan K. Mishra
 Dynamic programming

Anticipate what the memory table looks like fib(5)

Subproblems are known from problem fib(4)

structure
fib(3)
Dependencies form a dag
fib(2)
Solve subproblems in topological order
fib(1)

Kk 0 1 2 3 4 5 fib(0)
fib(k)
fib(k) 0 1 1 2 3 5
Prepared by Pawan K. Mishra
 Dynamic programming fibonacci

function fib(n):
fibtable[0] = 0
fibtable[1] = 1
for i = 2,3,..n
fibtable[i] = fibtable[i-1] + fibtable[i-2]
return(fibtable[n])

Prepared by Pawan K. Mishra

 Memoization

Store values of subproblems in a table

Look up the table before making a recursive
call

Dynamic programming

Solve subproblems in topological order

of dependency

Dependencies must form a dag (why?)

Iterative evaluation

Prepared by Pawan K. Mishra

 Takeaway
Overlapping sub problem
Tabular Method
Recursive formulation– Don’t solve recursively
Compute opt for smaller sub problem
Optimal substructure
Computing Optimal value vs Optimal solution

Prepared by Pawan K. Mishra

 P1. Knapsack-without profit

Input
n objects o1, o2,…, on
Size s1, s2, ..., sn
Bag with capacity W

Find a subset of objects with largest

total size ≤ W
Prepared by Pawan K. Mishra
 P1. Knapsack-without profit

Input
n objects o1, o2,…, on
Size s1, s2, ..., sn
Bag with capacity W

Find a subset of objects with largest

total size ≤ W
Prepared by Pawan K. Mishra
 Brute force

Check all subsets --- 2 n

Prepared by Pawan K. Mishra

 OPTIMAL SUBSTRUCTURE PROPERTY
Suppose object o1 is in the optimal solution, what
we need to solve now ??
Then the remaining subset of objects
{o2, o3, …, on} of largest size whose total
size is atmost W – s1.
Suppose object o1 is not in the optimal solution,
what we need to solve now ??
Then the remaining subset of objects
{o2, o3, …, on} of largest size whose total
Prepared by Pawan K. Mishra size is atmost W.
 Recusrive Formulation
OPT ( {o2 ,…, on}, W )

OPT({o1, o2 ,…, on}, W)=max

OPT( { o1, o2 ,…, on }, W – s1 ) +s1

Prepared by Pawan K. Mishra

 Recusrive Formulation
OPT ( {o2 ,…, on}, W )

OPT({o1, o2 ,…, on}, W)=max

OPT( { o1, o2 ,…, on }, W – s1 ) +s1

OPT ( i+1, w )

OPT(i, w)=max
OPT( i+1, w – si ) + si
Largest sized subset {i,….,n} with total size at most w
Prepared by Pawan K. Mishra
Prepared by Pawan K. Mishra
Prepared by Pawan K. Mishra
Prepared by Pawan K. Mishra
Prepared by Pawan K. Mishra
 This basically tells you
the direction via which
you fill the table, thus
we get the base case
too

OPT(n, w)= 0 sn >w

= sn otherwise

For all w \in {1,2,…,W}

Prepared by Pawan K. Mishra

 This basically tells you
the direction via which
you fill the table, thus
we get the base case
too

OPT(n, w)= 0 sn > w

= sn otherwise

For all w \in {1,2,…,W}

Prepared by Pawan K. Mishra

 Total Time

How many entries we are computing ?? nW

Time to compute each entry – Constant time..

By using max of two entries..

O(nW)
Prepared by Pawan K. Mishra
 Cut-and-Paste Arguments
To show optimal substructure, assume that
some piece of the optimal solution S* is not an
optimal solution to a smaller subproblem
Show that replacing that piece with the
optimal solution to the smaller subproblem
improves the allegedly optimal solution S*.
Conclude, therefore, that S* must include
an optimal solution to a smaller subproblem.
Prepared by Pawan K. Mishra
 P2. Coin Change Problem
If we want to make change for Rs. T, and we have
infinite supply of each coin in the set
Coins = {v1, v2,. . . , vn}, where vi is
the value of the i-th coin.

What is the minimum number of coins required to

reach the value S?

Prepared by Pawan K. Mishra

 Greedy Algorithm

Prepared by Pawan K. Mishra

 Greedy Algorithm

Coins = {6, 4, 1} and T = 8

Counter Example

Prepared by Pawan K. Mishra

 Greedy Algorithm

Coins = {6, 4, 1} and T = 8

Coins={1,5,10,25}-USA

Coins={1,2,5,10} -Indian
Prepared by Pawan K. Mishra
 Dynamic Programming
OPT( T, n-1) First coin with value v1 is not used

OPT(T,n)= min
OPT(T- v1, n) + 1 First coin with value v1 is used

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( T, n-1) First coin with value v1 is not used

OPT(T,n)= min
OPT(T- v1, n) + 1 First coin with value v1 is used

OPT( S, i-1) i-th coin with value vi is not used

OPT(S,i)= min
OPT(S- vi, i) + 1 i-th coin with value vi is used

the minimum number of coins required to reach sum S ≤ T with the first i coins, i.e.,
coins selected from the subset {v1, v2,. . . , vi} (where 0 ≤ i ≤ n).
Prepared by Pawan K. Mishra
 Dynamic Programming

OPT( S, i-1) i-th coin with value vi is not used

OPT(S,i)= min
OPT(S- vi, i) + 1 i-th coin with value vi is used
SEMANTICS

the minimum number of coins required to reach sum S ≤ T with the first i coins, i.e.,
coins selected from the subset {v1, v2,. . . , vi} (where 0 ≤ i ≤ n).

Prepared by Pawan K. Mishra

 Dynamic Programming

OPT( S, i-1) i-th coin with value vi is not used

OPT(S,i)= min
OPT(S- vi, i) + 1 i-th coin with value vi is used
SEMANTICS

the minimum number of coins required to reach sum S ≤ T with the first i coins, i.e.,
coins selected from the subset {v1, v2,. . . , vi} (where 0 ≤ i ≤ n).

∞ ∞∞ ∞ ∞ ∞ ∞ ∞
Prepared by Pawan K. Mishra
 P3. Set Cover Problem
Given a universe U={e1, e2, e3, …, en} of n elements,

And a family of subsets F = {S1, S2, S3,…, Sm }.

find a minimum number of subcollection C of F such that C covers all elements of U.

U = {1, 2, 3, 4, 5}

F= { S1 = {1, 2, 3}, S2 = {2,3}, S3 = {4, 5}, S4 = {1, 2, 4} }

C = {S1, S3}

Prepared by Pawan K. Mishra

 P3. Set Cover Problem
Given a universe U={e1 , e2, e3, …, en} of n elements,
Brute Force = O(2m poly(n))
And a family of subsets F = {S1, S2, S3,…, Sm }.

find a minimum number of subcollection C of F such that C covers all elements of U.

U = {1, 2, 3, 4, 5}

F= { S1 = {1, 2, 3}, S2 = {2,3}, S3 = {4, 5}, S4 = {1, 2, 4} }

C = {S1, S3}

Prepared by Pawan K. Mishra

 Greedy Algorithm ??

Prepared by Pawan K. Mishra

 Dynamic Programming
Can we improve to O*(2n) ??

Prepared by Pawan K. Mishra

 Dynamic Programming
Can we improve to O*(2n) ??

Given : F = {S1, S2, S3,…, Sm } , U

To find subproblem, but how ?--- need recurrence formulation …

Prepared by Pawan K. Mishra

 Dynamic Programming
Can we improve to O*(2n) ??

Given : F = {S1, S2, S3,…, Sm } , U

To find subproblem, but how ?--- need recurrence formulation …

How to proceed?

Is S1 in optimal solution, then what to cover and by what ?

Is S1 in not in optimal solution, then what to cover and by what ?

Prepared by Pawan K. Mishra

 Dynamic Programming
Can we improve to O*(2n) ??

Given : F = {S1, S2, S3,…, Sm } , U

To find subproblem, but how ?--- need recurrence formulation …

How to proceed?

Is S1 in optimal solution, then what to cover and by what ? U \ S1 {S2, S3,…, Sm}

Is S1 in not in optimal solution, then what to cover and by what ? U {S2, S3,…, Sm}

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( {S2, S3,…, Sn} , U) S1 is not used
OPT({S1,S2, S3,…, Sn} ,U)= min
OPT( {S2, S3,…, Sn} , U \ S1) + 1 S1 is used

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( {S2, S3,…, Sn} , U) S1 is not used
OPT({S1,S2, S3,…, Sn} ,U)= min
OPT( {S2, S3,…, Sn} , U \ S1) + 1 S1 is used

OPT( Si+1, A) Si is not used

OPT(i,A)= min
OPT(Si+1, A \ Si) + 1 Si is used

To cover a subset A of U, the minimum number of subset required of F from

Si to Sm.

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( Si+1, A) Si is not used

OPT(i,A)= min
OPT(Si+1, A \ Si) + 1 Si is used

To cover a subset A of U, the minimum number of subset required of F from

Si to Sm.

Goal : OPT(1, U)

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( Si+1, A) Si is not used

OPT(i,A)= min
OPT(Si+1, A \ Si) + 1 Si is used

To cover a subset A of U, the minimum number of subset required of F from

Si to Sm.
Base Case: We need i=1 as a goal, so basically we need in decreasing order,
so m will play the role in defining the base case.

1 If A is subset of Sm
OPT(m,A)= min
∞ Otherwise

Goal : OPT(1, U)

Prepared by Pawan K. Mishra

 Dynamic Programming
OPT( Si+1, A) Si is not used

OPT(i,A)= min
OPT(Si+1, A \ Si) + 1 Si is used

To cover a subset A of U, the minimum number of subset required of F from

Si to Sm.
Base Case: We need i=1 as a goal, so basically we need in decreasing order,
so m will play the role in defining the base case.

1 If A is subset of Sm Time Complexity=O(2nm)

OPT(m,A)= min
∞ Otherwise

Goal : OPT(1, U)

Prepared by Pawan K. Mishra

 P4. Maximum Contiguous Subsequence in Array

IP: Sequence S={e1, e2, e3, …, en} of n integers

OP: pair (i,j) ei + ei+1 + ei+2 + … + ej is maximum

Prepared by Pawan K. Mishra

 P4. Maximum Contiguous Subsequence in Array

IP: Sequence S={e1, e2, e3, …, en} of n integers Brute Force: For every
pair, find sum: O(n3) or
OP: pair (i,j) ei + ei+1 + ei+2 + … + ej is maximum O(n2)

DnC: O(nlogn), Tut-2,Q 8

Prepared by Pawan K. Mishra

 Dynamic Programming
How to proceed?
1. OPT(i-1) + ei
Optimal solution contains element ei ----
2. ei

Optimal element does not contain element ei.

This means that ei

OPT(i-1) + ei
OPT(i)= max
ei

Optimal value if sub sequence ends at ei.

Prepared by Pawan K. Mishra
 P5. Longest Common Subword
Given two strings, find the (length of the) longest
common subword
“secret”, “secretary” — “secret”, length 6

“bisect”, “trisect” — “sect”, length 4

Prepared by Pawan K. Mishra

 More formally …

Let u = a0 a1…am and v = b0 b1… bn be two sequences

If we can find i, j such that

ai ai+1 …ai+k-1 = bj bj+1… bj+k-1,
u and v have a common subword of length k

Aim is to find the length of the longest common

subword of u and v

Prepared by Pawan K. Mishra

 Brute force
Let u = a0 a1… am and v = b0 b1… bn
Try every pair of starting positions i in u, j in v

Match (ai, bi), (ai+1,bi+1),… as far as possible

Keep track of the length of the longest match

Assuming m > n, this is O(mn2)

mn pairs of positions

From each starting point, scan can be O(n)

Prepared by Pawan K. Mishra

 Inductive structure
Let u = a0 a1 … am and v = b0 b1 …bn

aiai+1…ai+k-1 = bjbj+1…bj+k-1 is a common subword of length k at (i,j),

iff ai+1…ai+k-1 = bj+1…bj+k-1 is a common sub word of length k-1 at (i+1,j+1).

LCW(i,j): length of the longest common subword

starting at ai and b j

If ai ≠ b j, LCW(i,j) is 0,

otherwise 1+LCW(i+1,j+1)

Boundary condition: when we have reached the end

of one of the words
Prepared by Pawan K. Mishra
 Inductive structure
Consider positions 0 to m+1 in u, 0 to n+1 in v
To know the end of the sequence
m+1, n+1 means we have reached the end of the word

LCW(m+1,j) = 0 for all j

LCW(i,n+1) = 0 for all i

LCW(i,j) = 0, if ai ≠ b j

= 1+ LCW(i+1,j+1),

if ai= b j
Prepared by Pawan K. Mishra
 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b
Last row and column have no 1 i
dependencies
2 s
Start at bottom right 3 e
corner and fill by row or by 4 c
column 5 t
6 •

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 33 4 5 66
LCW(i,j) depends on
LCW(i+1,j+1) s e c rr e t ••
0 b
Last row and column have no 1 i
dependencies
2 s
Start at bottom right 3 e
corner and fill by row or by 4 c
column 5 t
6 •

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0
Last row and column have no 1 i 0
dependencies
2 s 0
Start at bottom right 3 e 0
corner and fill by row or by 4 c 0
column 5 t 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0 0
Last row and column have no 1 i 0 0
dependencies
2 s 0 0
Start at bottom right 3 e 0 0
corner and fill by row or by 4 c 0 0
column 5 t 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0 0 0
Last row and column have no 1 i 0 0 0
dependencies 2 s 0 0 0
Start at bottom right 3 e 1 0 0
corner and fill by row or by 4 c 0 0 0
column 5 t 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0 0 0 0
Last row and column have no 1 i 0 0 0 0
dependencies 2 s 0 0 0 0
Start at bottom right 3 e 0 1 0 0
corner and fill by row or by 4 c 0 0 0 0
column 5 t 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0 0 0 0 0
Last row and column have no 1 i 0 0 0 0 0
dependencies 2 s 0 0 0 0 0
Start at bottom right 3 e 0 0 1 0 0
corner and fill by row or by 4 c 1 0 0 0 0
column 5 t 0 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 b 0 0 0 0 0 0
Last row and column have no 1 i 0 0 0 0 0 0
dependencies 0 0 0
2 s 0 0 0
Start at bottom right 3 e 2 0 0 1 0 0
corner and fill by row or by 4 c 0 1 0 0 0 0
column 5 t 0 0 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
0 1 2 3 4 5 6
LCW(i,j) depends on
LCW(i+1,j+1) s e c r e t •
0 0 0 0
b 0 0 0 0
Last row and column have no 1 i 0 0 0 0 0 0 0
dependencies 2 s 3 0 0 0 0 0 0
Start at bottom right 3 e 0 2 0 0 1 0 0
corner and fill by row or by 4 c 0 0 1 0 0 0 0
column 5 t 0 0 0 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Reading off the solution
Find (i,j) with largest
entry 0 1 2 3 4 5 6
s e c r e t •
LCW(2,0) = 3 0 0 0 0
b 0 0 0 0
Read offthe actual
1 i 0 0 0 0 0 0 0
subword diagonally 2 s 3 0 0 0 0 0 0
3 e 0 2 0 0 1 0 0
4 c 0 0 1 0 0 0 0
5 t 0 0 0 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Complexity

Recall that the brute force approach was

O(mn2)

The inductive solution is O(mn) if we use

dynamic programming (or memoization)

Need to fill an O(mn) size table

Each table entry takes constant time to

compute
Prepared by Pawan K. Mishra
 P6. Longest common subsequence
Subsequence: can drop some letters in between

Given two strings, find the (length of the) longest

common subsequence
“secret”, “secretary” — “secret”, length 6

“bisect”, “trisect” — “isect”, length 5

“bisect”, “secret” — “sect”, length 4

“director”, “secretary” —“ectr”, “retr”, length 4

Prepared by Pawan K. Mishra

 Applications
Analyzing genes of two species (in bionforamtics..)

DNA is a long string over 4 proteins A,T,G,C

Two species are closer if their DNA has longer

common subsequence

Unix Command diff

Compares text files
Find longest matching subsequence of lines

Prepared by Pawan K. Mishra

 LCS
0 1 2 3 4 5 6
LCS is longest path we
s e c r e t •
can find between non-
zero LCW entries, 0 b 0 0 0 0 0 0 0
moving right and down 1 i 0 0 0 0 0 0 0
2 s 3 0 0 0 0 0 0
3 e 0 2 0 0 1 0 0
“bisect”, “secret” — “sect”, length 4
4 c 0 0 1 0 0 0 0
5 t 0 0 0 0 0 1 0
6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Inductive structure
u a0 a1 a2 … … am-1 am
v b0 b1 b2 … bn-1 bn
If a0 = b0,
LCS(a0a1…am, b0b1…bn) = 1 + LCS(a1a2…am, b1b2…bn)

Can force (a0,b0) to be part of LCS

If not, a0 and b0 cannot both be part of LCS

Not sure which one to drop

Solve both subproblems

LCS(a1 a2…am, b0 b1…bn) and LCS(a0 a1…am, b1 b2…bn), and

Prepared by Pawan K.take
Mishra the maximum
 Inductive structure
u a0 a1 a2 … … am-1 am
v b0 b1 b2 … bn-1 bn

LCS(i,j) stands for LCS(ai ai+1… am, bj bj+1… bn)

If ai = bj, LCS(i,j) = 1 + LCS(i+1,j+1)

If ai ≠ b j, LCS(i,j) = max(LCS(i+1,j), LCS(i,j+1))

As with LCW, extend positions to m+1, n+1

LCS(m+1,j) = 0 for all j

LCS(i,n+1) = 0 for all i

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b
1 i
Dependencies for 2 s
LCS(m,n) are known 3 e

Start at LCS(m,n) and 4 c

fill by row, column or 5 t
diagonal 6 •

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 33 4 5 66
LCS(i+1,j+1) as well as s e c rr e t ••
LCS(i+1,j) and
0 b
LCS(i,j+1)
1 i
Dependencies for 2 s
LCS(m,n) are known 3 e

Start at LCS(m,n) and 4 c

fill by row, column or 5 t
diagonal 6 •

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 0
1 i 0
Dependencies for 2 s 0
LCS(m,n) are known 3 e 0
Start at LCS(m,n) and 4 c 0
fill by row, column or 5 t 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 0 0
1 i 0 0
Dependencies for 2 s 0 0
LCS(m,n) are known 3 e 0 0
Start at LCS(m,n) and 4 c 0 0
fill by row, column or 5 t 1 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 1 0 0
1 i 1 0 0
Dependencies for 2 s 1 0 0
LCS(m,n) are known 3 e 1 0 0
Start at LCS(m,n) and 4 c 1 0 0
fill by row, column or 5 t 1 1 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 1 1 0 0
1 i 1 1 0 0
Dependencies for 2 s 1 1 0 0
LCS(m,n) are known 3 e 1 1 0 0
Start at LCS(m,n) and 4 c 1 1 0 0
fill by row, column or 5 t 1 1 1 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 2 1 1 0 0
1 i 2 1 1 0 0
Dependencies for 2 s 2 1 1 0 0
LCS(m,n) are known 3 e 2 1 1 0 0
Start at LCS(m,n) and 4 c 2 1 1 0 0
fill by row, column or 5 t 1 1 1 1 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 3 2 1 1 0 0
1 i 3 2 1 1 0 0
Dependencies for 2 s 3 2 1 1 0 0
LCS(m,n) are known 3 e 3 2 1 1 0 0
Start at LCS(m,n) and 4 c 2 2 1 1 0 0
fill by row, column or 5 t 1 1 1 1 1 0
diagonal 6 • 0 0 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 4 3 2 1 1 0 0
1 i 43 2 1 1 0 0
Dependencies for 2 s 43 2 1 1 0 0
LCS(m,n) are known 3 e 33 2 1 1 0 0
Start at LCS(m,n) and 4 c 22 2 1 1 0 0
fill by row, column or 5 t 11 1 1 1 1 0
diagonal 6 • 00 0 0 0 0 0

Prepared by Pawan K. Mishra

 Subproblem dependency
LCS(i,j) depends on 0 1 2 3 4 5 6
LCS(i+1,j+1) as well as s e c r e t •
LCS(i+1,j) and
LCS(i,j+1)
0 b 4 3 2 1 1 0 0
1 i 43 2 1 1 0 0
Dependencies for 2 s 43 2 1 1 0 0
LCS(m,n) are known 3 e 33 2 1 1 0 0
Start at LCS(m,n) and 4 c 22 2 1 1 0 0
fill by row, column or 5 t 11 1 1 1 1 0
diagonal 6 • 00 0 0 0 0 0
LCS(0,0) =4 is the ans
Prepared by Pawan K. Mishra
 Recovering the sequence
Trace back the
0 1 2 3 4 5 6

path by which s e c r e t •
each entry was 0 b 4 3 2 1 1 0 0
filled 1 i 43 2 1 1 0 0
2 s 43 2 1 1 0 0
Each diagonal step
is an element of the 3 e 33 2 1 1 0 0
LCS 4 c 22 2 1 1 0 0
5 t 11 1 1 1 1 0
“sect”
6 • 00 0 0 0 0 0

Prepared by Pawan K. Mishra

 Recovering the sequence
Trace back the
0 1 2 3 4 5 6

path by which s e c r e t •
each entry was 0 b 4 3 2 1 1 0 0
filled 1 i 43 2 1 1 0 0
2 s 43 2 1 1 0 0
Each diagonal step
is an element of the 3 e 33 2 1 1 0 0
LCS 4 c 22 2 1 1 0 0
5 t 11 1 1 1 1 0
“sect”
6 • 00 0 0 0 0 0

Prepared by Pawan K. Mishra

 Complexity

Again O(mn) using dynamic programming

(or memoization)

Need to fill an O(mn) size table

Each table entry takes constant time to

compute

Prepared by Pawan K. Mishra

 P7. Multiplying matrices
To multiply matrices A and B, need compatible
dimensions
A of dimension m x n, B of dimension n x p AB

has dimension mp

Each entry in AB take O(n) steps to compute AB[i,j]

is A[i,1]B[1,j] + A[i,2]B[2,j] +…+ A[i,n]B[n,j]

Overall, computing AB is O(mnp)

Prepared by Pawan K. Mishra

 Multiplying matrices
Matrix multiplication is associative

ABC = (AB)C = A(BC)

Bracketing does not change the answer …

… but can affect the complexity of computing it!

Prepared by Pawan K. Mishra

 Multiplying matrices
Suppose dimensions are A[1,100], B[100,1], C[1,100]

Computing A(BC)

BC is [100,100], 100 x 1 x 100 = 10000 steps

A(BC) is [1,100],1 x 100 x 100 = 10000 steps

Computing (AB)C

AB is [1,1], 1 x 100 x 1 = 100 steps

(AB)C is [1,100], 1 x 1 x 100 = 100 steps

A(BC) takes 20000 steps, (AB)C takes 200 steps!

Prepared by Pawan K. Mishra

 Multiplying matrices
Given matrices M1, M2,…, Mn of dimensions [r1,c1], [r2,c2], …, [rn,cn]

Dimensions match, so M1 x M2 x …x Mn can be computed

ci= ri+1 for 1 ≤ i < n

Find an optimal order to compute the product

That is, bracket the expression optimally

Prepared by Pawan K. Mishra

 Inductive structure
Product to be computed: M1 x M2 x …x Mn

Final step would have combined two subproducts

(M1 x M2 x …x Mk) x (Mk+1 x Mk+2 x …x Mn), for some 1 ≤ k < n

First factor has dimension (r1,ck), second (rk+1,cn)

Final multiplication step costs =r1ckcn

Add cost of computing the two factors

Prepared by Pawan K. Mishra

 Subproblems
Final step is
(M1 x M2 x …x Mk) x (Mk+1 x Mk+2 x …x Mn)

Subproblems are (M1 x M2 x …x Mk) and (Mk+1 x Mk+2x …x Mn)

Total cost is Cost(M1 x M2x …x Mk) + Cost(Mk+1 x Mk+2x …x Mn)

+ r1ckcn

Which k should we choose?

No idea! Try them all and choose the minimum!

Prepared by Pawan K. Mishra

 Inductive formulation
Cost(M1 x M2 x …x Mn) =
minimum value, for 1 ≤ k < n, of
Cost(M1 x M2 x …x Mk) +
Cost(Mk+1 x Mk+2 x …x Mn) +
r1ckcn

When we compute Cost(M1 x M2 x …x Mk) we will

get subproblems of the form Mj x Mj+1x …x Mk

Prepared by Pawan K. Mishra

 In general …
Cost(Mi x Mi+1 x …x Mj) =
minimum value, for i ≤ k < j, of
Cost(Mi x Mi+1 x …x Mk) + Cost(Mk+1x Mk+2x …x Mj) + rickcj

Write Cost(i,j) to denote Cost(Mi x Mi+1x …x Mj)

Prepared by Pawan K. Mishra

 Final equation

Cost(i,i) = 0 — No multiplication to be done

Cost(i,j) = min over i ≤ k < j

[ Cost(i,k) + Cost(k+1,j) + rickcj ]

Note that we only require Cost(i,j) when i ≤ j

Prepared by Pawan K. Mishra

 Subproblem dependency
Cost(i,j) depends on 1 … i … j … n
Cost(i,k), Cost(k+1,j) for all i ≤ k < j
1
Can have O(n) dependent values, …
unlike LCS, LCW, ED
i
Start with main diagonal and fill …
matrix by columns, bottom to top,
left to right j
…
n

Prepared by Pawan K. Mishra

 Subproblem dependency
Cost(i,j) depends on 1 … i … j … n
Cost(i,k), Cost(k+1,j) for all i ≤ k < j
1
Can have O(n) dependent values, …
unlike LCS, LCW, ED
i
Start with main diagonal and fill …
matrix by columns, bottom to top,
left to right j
…
n

Prepared by Pawan K. Mishra

 Subproblem dependency
Cost(i,j) depends on 1 … i … j … n
Cost(i,k), Cost(k+1,j) for all i ≤ k < j
1
Can have O(n) dependent values, …
unlike LCS, LCW, ED
i
Start with main diagonal and fill …
matrix by columns, bottom to top,
left to right j
…
n

Prepared by Pawan K. Mishra

 Subproblem dependency
Cost(i,j) depends on 1 … i … j … n
Cost(i,k), Cost(k+1,j) for all i ≤ k < j
1
Can have O(n) dependent values, …
unlike LCS, LCW, ED
i
Start with main diagonal and fill …
matrix by columns, bottom to top,
left to right j
…
n

Prepared by Pawan K. Mishra

 Complexity

As with LCS, we to fill an O(n2) size table

However, filling MMC[i][j] could require

examining O(n) intermediate values

Hence, overall complexity is O(n3)

Prepared by Pawan K. Mishra

 P8. Maximum Independent Set on
tree

Prepared by Pawan K. Mishra

 P8. Minimum Vertex Cover on tree

Exercise : Based on the previous idea

Prepared by Pawan K. Mishra

 P9. Diameter of a rooted tree
Excercise

Prepared by Pawan K. Mishra

 P10. Weighted tree, find a maximum
sum path between any two nodes
Excercise

Prepared by Pawan K. Mishra

1
2
3
4
1
2
1
1
2
3
4
5