6

v
3 8
z
4

AVL Trees

Elementary Maths for GMT 1

AVL Tree: Balanced Binary Tree
 Definition: An AVL Tree is a binary search tree where for
every internal node v, the heights of the children of v can differ
by at most 1
4
44
2 3
17 78
1 2 1
32 50 88
An example of an 1 1
AVL tree where the 48 62
heights are shown
next to the nodes.
AVL Trees 2
 Height of an AVL Tree
 Property: The height of an AVL tree storing n keys is O(log n)
 Proof: Let us bound N(h): the minimum number of internal
nodes of an AVL tree of height h
 N(1) = 1 and N(2) = 2
 For h > 2, an AVL tree of height h contains at least a root
node, one AVL subtree of height h-1, and one AVL subtree
of height h-2, so N(h) = 1 + N(h-1) + N(h-2)
 Since N(h-1) > N(h-2), we have N(h) > 2 N(h-2), and so

N(h) > 2 N(h-2), N(h) > 4 N(h-4),

N(h) > 8 N(h-6), …, etc. N(2) 3

N(h) > 2i N(h-2i) 4 N(1)

AVL Trees 3
 Height of an AVL Tree
 Property: The height of an AVL tree storing n keys is O(log n)
 Proof: Let us bound N(h): the minimum number of internal
nodes of an AVL tree of height h
 N(1) = 1 and N(2) = 2
 N(h) > 2i N(h-2i)
 Choose i = h/2 -1:
N(h) > 2 h/2-1 N(h-2(h/2 -1)) = 2 h/2-1 N(2) = 2 h/2-1 2

h < 2 log N(h)  2 n

N(2) 3
 So the height of an AVL tree is O(log n)
4 N(1)

AVL Trees 4
 Insertion in an AVL Tree
 Insertion is as in a binary search tree: always done by expanding
an external node
imbalance

44 Example: insert 54 44

17 78
17 78

32 50 88
32 50 88

48 62
48 62
54

AVL Trees 5
Imbalance after Insertion

44
 Let w be the inserted node

17 78
z  z be the first unbalanced ancestor
y of w
 y be the child of z with higher
32 50 88
height (must be an ancestor of w)
 x be the child of y with higher
48 62
x height (must be an ancestor of w;
or w itself)
54
w

AVL Trees 6
 Trinode Restructuring
 Perform the rotations needed to make y the topmost node of the
three.

z
case 1: single rotation y

y
z x
T0
x
T1 T0 T1 T2 T3

T2 T3

AVL Trees 7
Trinode Restructuring

symmetric case y
y
T3
x z
x
T2

T0 T1 T0 T1 T2 T3

AVL Trees 8
Trinode Restructuring

z
x
case 2: double rotation
y
T0 z y
x
T3

T0 T1 T2 T3
T1 T2

AVL Trees 9
Trinode Restructuring

z x

symmetric case
y y z
T3

x
T0
T0 T1 T2 T3

T1 T2

AVL Trees 10
Removal in an AVL Tree
 Removal begins as in a binary search tree, which means the
node removed will become an empty external node
imbalance

44 Example: delete 32 44

17 62 17 62

32 50 78
50 78

48 54 88 48 54 88

AVL Trees 11
 Imbalance after Removal
z 44  Let w be the parent of the
removed node
w y  Let z be the first unbalanced
17 62
ancestor of w
 Let y be the child of z with higher
50 78 x height (is now not an ancestor of
w)
 Let x be
48 54 88
 the child of y with higher height,
if heights are different, or
 the child of y on the same side
as y, if heights are equal

AVL Trees 12
Rebalancing after a Removal
 Perform rotations to make y the topmost of the three.
 As this restructuring may upset the balance of another node
higher in the tree, we must continue checking for balance until
the root of T is reached
z 44 62

w 17 62 y 44 78

50 78 x 17 50 88

48 54 88 48 54

AVL Trees 13
Repeated Rebalancing

44 44

17 78 17 w=z 78

y
14 32 50 88 32 50 88

25 48 62 93 25 x 48 62 93

54 54
Example: delete 14

AVL Trees 14
Repeated Rebalancing

44

25 78

17
25 32 50 88

48 62 93

54

AVL Trees 15
Running Times for AVL Trees
 Finding a value takes O(log n) time
 because height of tree is always O(log n)

 Traversal of the whole set takes O(n) time

 Insertion takes O(log n) time

 initial find takes O(log n) time
 0 or 1 rebalancing in the tree, maintaining height takes O(log n) time

 Removal takes O(log n) time

 initial find takes O(log n) time
 0 or more rebalancing in the tree, maintaining height takes O(log n) time

AVL Trees 16
AVL trees vs. Hash tables
 In an AVL tree, insert/delete/search in O(log n) time worst case,
in a hash table, these operations take O(1) time in practice

 In an AVL tree, searching with x for the smallest value x also

takes O(log n) time, in a hash table it takes linear time

 Enumerating the set in order takes O(n) time in an AVL tree, in

a hash table it cannot be done (well, in O(n log n) time)

 Finding the number of values between x and y takes O(log n)

time with a simple variation of an AVL tree, in a hash table it
takes linear time
 an AVL tree is more versatile than a hash table
AVL Trees 17
Other trees
 BB[α]-trees are not height-balanced but weight-balanced
Height is also O(log n)
 Red-black trees are balanced with a different scheme and
also have height O(log n)

 For background storage, B-trees exist

These have a degree higher than two (more than 2 children)
 For 2- and higher-dimensional data, various trees exist:
 Kd-trees
 Quadtrees, Octrees
 BSP-trees
 Range trees
 R-trees

AVL Trees 18