6.

006- Introduction to Algorithms

m
co.
es
mot
cfon
psd.
stet
.njuo
wf
wpd
wst
.ijtu
wis
wv
we
or
rM

Lecture 4
Fo

Prof. Piotr Indyk

For More visit www.justpdfnotes.com

 Lecture Overview
1
5

m
co
6

.
es
• Review: Binary Search Trees

mot
7

cfon
psd.
• Importance of being balanced

stet
10

.njuo
wf
wpd
• Balanced BSTs
wst
wis
.ijtu 12
– AVL trees
wv

7
we
or

• definition
rM
Fo

5 10
• rotations, insert
1 6 12

For More visit www.justpdfnotes.com

 Binary Search Trees (BSTs)

m
co
• Each node x has: root

.
es
mot
– key[x]

cfon
psd.
– Pointers: left[x], right[x], p[x] 10

stet
.njuo
wf
wpd
5 12
• Property: for any node x:
wst
.ijtu
– For all nodes y in the left subtree of x:
wis
1 6
wv

key[y] ≤ key[x]
we
or
rM

– For all nodes y in the right subtree of x: 7

Fo

key[y] ≥ key[x]
leaf

height = 3

For More visit www.justpdfnotes.com

The importance of being balanced
for n nodes:

m
co.
es
1

mot
cfon
7

psd.
5

stet
.njuo
6

wf
wpd
5 10
wst
7 .ijtu
wis
wv

1 6 12
we

10
or
rM

12
Fo

h = Θ(log n) h = Θ(n)

For More visit www.justpdfnotes.com

Balanced BST Strategy

m
co
• Augment every node

.
es
mot
with some data

cfon
psd.
stet
• Define a local invariant

.njuo
wf
wpd
on data
wst
.ijtu
• Show (prove) that
wis
wv
we

invariant guarantees
or
rM

Θ(log n) height
Fo

• Design algorithms to
maintain data and the
invariant For More visit www.justpdfnotes.com
AVL Trees: Definition
[Adelson-Velskii and Landis’62]

m
7 32

co
• Data: for every node, maintain its

.
es
height (“augmentation”)

mot
5 21 10 1

cfon
psd.
stet
.njuo
– Leaves have height 0

wf
1 10 6 0 12 0
wpd
– NIL has “height” -1
wst
.ijtu
wis
wv

4 0
we
or
rM

x
• Invariant: for every node x, the
Fo

heights of its left child and right

child differ by at most 1
k k-1

For More visit www.justpdfnotes.com

 AVL trees have height Θ(log n)
Invariant: for every node x, the heights of its left
child and right child differ by at most 1

m
co
• Let nh be the minimum number of

.
es
nodes of an AVL tree of height h

mot
h

cfon
psd.
• We have nh ≥ 1+ nh-1 + nh-2

stet
.njuo
wf
⇒ nh > 2nh-2
wpd
h-1 h-2
wst
.ijtu
⇒ nh > 2h/2
wis
wv
we

⇒ h < 2 lg nh
or
rM

• The constant “2” can be improved

Fo

How can we maintain the invariant ?

For More visit www.justpdfnotes.com
 Rotations
B RIGHT-ROTATE(B) A

m
co
A LEFT-ROTATE(A) B

.
es
mot
cfon
γ α

psd.
stet
.njuo
α β β γ

wf
wpd
wst
Rotations maintain the inorder ordering of keys:
.ijtu
wis

• a ∈ α, b ∈ β, c ∈ γ ⇒ a ≤ A ≤ b ≤ B ≤ c.
wv
we
or
rM

1
Fo

2
LEFT-ROTATE(1) 2
1 3

For More visit www.justpdfnotes.com

Insertions

m
co
• Insert new node u as in the

.
es
mot
simple BST h+1

cfon
psd.
stet
– Can create imbalance

.njuo
wf
wpd
• Work your way up the tree, h-1 h-2
wst
.ijtu
restoring the balance
wis
wv
we
or

• Similar issue/solution when

rM

u
Fo

deleting a node

For More visit www.justpdfnotes.com

Balancing

m
• Let x be the lowest “violating”

co
x

.
node

es
mot
– We will fix the subtree of x and

cfon
y

psd.
move up

stet
A

.njuo
• Assume the right child of x is

wf
wpd
deeper than the left child of x (x is
B C
“right-heavy”) wst
.ijtu
wis

• Scenarios:
wv
we
or

– Case 1: Right child y of x is

rM

right-heavy
Fo

– Case 2: Right child y of x is

balanced
– Case 3: Right child y of x is
left-heavy
For More visit www.justpdfnotes.com
 Case 1: y is right-heavy

m
x y

co
LEFT-ROTATE(x)

.
es
x

ot
y

m
k+1

cfon
k-1 k

psd.
A C

stet
.njuo
k

wf
wpd
k-1 B C k-1 A B k-1
wst
.ijtu
wis
wv
we
or
rM
Fo

For More visit www.justpdfnotes.com

 Case 2: y is balanced

m
x y

co
LEFT-ROTATE(x)

.
es
x

ot
y

m
k+1

cfon
k-1 k

psd.
A C

stet
.njuo
k

wf
wpd
k B C k-1 A B k
wst
.ijtu
wis
wv
we
or
rM
Fo

Same as Case 1

For More visit www.justpdfnotes.com

 Case 3: y is left-heavy

m
x y

co
LEFT-ROTATE(x)

.
es
x

ot
y

m
k+1

cfon
k-1 k-1

psd.
A C

stet
.njuo
k-1

wf
wpd
k B C k-1 A B k
wst
.ijtu
wis
wv
we
or
rM
Fo

Need to do more …

For More visit www.justpdfnotes.com

 Fo
rM
or
we
wv
wis
.ijtu
wst
wpd
wf
.njuo
stet
psd.

For More visit www.justpdfnotes.com

cfon
mot
es
. co
m
 Rotations
B RIGHT-ROTATE(B) A

m
co
A LEFT-ROTATE(A) B

.
es
mot
cfon
γ α

psd.
stet
.njuo
α β β γ

wf
wpd
wst
Rotations maintain the inorder ordering of keys:
.ijtu
wis

• a ∈ α, b ∈ β, c ∈ γ ⇒ a ≤ A ≤ b ≤ B ≤ c.
wv
we
or
rM

1
Fo

2
LEFT-ROTATE(1) 2
1 3

For More visit www.justpdfnotes.com

 Case 3: y is left-heavy
RIGHT-ROTATE (y)

m
x
LEFT-ROTATE(x)

co.
es
y

ot
k+1

m
k-1 z

cfon
psd.
A

stet
y

.njuo
k x
k z k-1

wf
C
wpd
wst
.ijtu k-1
k-1
wis
or
wv

k-1 A B D C
we

k-2
or

B D
rM

or
k-2 k-1
Fo

And we are done!

For More visit www.justpdfnotes.com

Conclusions

m
co
• Can maintain balanced BSTs in O(log n)

.
es
mot
time per insertion

cfon
psd.
stet
• Search etc take O(log n) time

.njuo
wf
wpd
wst
.ijtu
wis
wv
we
or
rM
Fo

For More visit www.justpdfnotes.com

Examples of insert/balancing
Insert(23) x = 29: left-left case

m
3 41 41

co
2 20 65 1 20 65 1

.
es
50 φ

mot
50 φ
φ 11 1 29 φ 11 2 29

cfon
psd.
26 1 26
φ

stet
φ 23

.njuo
Done Insert(55)

wf
wpd
3 41 3 41

wst
2 20 65 1
2 20 .ijtu 65 1
wis
50 φ 50 φ
φ 11
wv

φ 11 1 26 1 26
we
or

φ 23 29 φ φ 23 29 φ
rM

x=65: left-right case Done

Fo

41 3 41

2 20 2 65 2 20 1 55

50 1 φ 50 φ 65
11 1 26 11 1 26
φ φ
55 φ
φ 23 29 φ φ 23 29 φ

For More visit www.justpdfnotes.com

Balanced Search Trees …

m
co
• AVL trees (Adelson-Velsii and Landis 1962)

.
es
mot
cfon
• Red-black trees (see CLRS 13)

psd.
stet
.njuo
• Splay trees (Sleator and Tarjan 1985)

wf
wpd
• wst
Scapegoat trees (Galperin and Rivest 1993)
.ijtu
wis
wv

• Treaps (Seidel and Aragon 1996)

we
or
rM

• ….
Fo

For More visit www.justpdfnotes.com

 BST for runway reservation system
5-1

49
• R = (37, 41, 46, 49, 56) current landing times

m
3-1

co
1
37 41 46 49 56

.
es
time (mins) 41 56

mot
cfon
now x x x x 1 1

psd.
stet
37 46

.njuo
• remove t from the set when a plane lands

wf
wpd
4+1

wst
R = (41, 46, 49, 56) .ijtu
49
wis
• add new t to the set if no other landings are
wv
we

2 1+1
scheduled within < 3 minutes from t
or
rM

• 44 => reject (46 in R) 41 56

Fo

1 1
• 53 => ok
• delete, insert, conflict checking take O(h), where 46 53
h is the height of the tree
For More visit www.justpdfnotes.com
 Fo
rM
or
we
wv
wis
.ijtu
wst
wpd
wf
.njuo
stet
psd.

For More visit www.justpdfnotes.com

cfon
mot
es
.
1

co
m
1
4
25

0
6
7

0
3

110

0 12
 m
co
And some people

.
es
mot
cfon
psd.
stet
like to do nothing
.njuo
wf
wpd
wst
.ijtu
wis
wv
we
or
rM
Fo

For More visit www.justpdfnotes.com