AVL Trees

Unit 4: Height-balanced trees

 Trees and balance
• balanced tree: One whose subtrees differ in height by at most 1
and are themselves balanced.

 A balanced BST of N nodes has a height of ~ log2 N.

 A highly unbalanced BST can have a height close to N.

root

9
 The runtime of adding to / searching a
BST is closely related to height. 6 14
height
balanced 4 8 19

7
 Some height numbers
• Observation: The shallower the BST, the better.
 Average case height is O(log N)
 Worst case height is O(N)
 Simple cases such as adding (1, 2, 3, ..., N), or the opposite order,
lead to the worst case scenario: height O(N). (as we saw last class)

root
• For binary tree of height h:
 max # of leaves: 2h-1 15
 max # of nodes: 2h - 1
 min # of leaves: 1 8 20
 min # of nodes: h
2 14 18 21
 Calculating tree height
• Height is max number of nodes in path from root to any leaf.
 height(null) = 0
 height(a leaf) = ?
 height(A) = ? A

 Hint: it's recursive!

A->left A->right
 height(a leaf) = 1
 height(A) = 1 + max(
height(A.left), height(A.right))
 AVL trees
• AVL tree: a binary search tree that uses modified add and remove
operations to stay balanced as its elements change
 one of several kinds of self-balancing trees
 invented in 1962 by two Russian mathematicians
• (Adelson-Velskii and Landis)
• A-V & L proved that an AVL tree's height is always O(log N).

 basic idea: When nodes are added to / removed from the tree,
if the tree becomes unbalanced, repair the tree until balance is
restored.
• rebalancing operations are relatively efficient (O(1))
• overall tree maintains a balanced O(log N) height, remains fast for
add/remove/search
 Balance factor
• balance factor, for a tree node T :
 = height of T's right subtree minus height of T's left subtree.
 BF(T) = Height(T->right) - Height(T->left)
-2
• (the tree at right shows BF of each node)

 an AVL tree maintains a "balance factor" 3 1

in each node of 0, 1, or -1
• i.e. for any node, its two child sub-trees
-2 0
differ in height by no more than 1

 the height of an AVL tree with N nodes -1

is O(log N)

0
 AVL tree examples
Two binary search trees:
(a) an AVL tree
(b) not an AVL tree (unbalanced nodes are darkened)
 Which are valid AVL trees?
• What is the balance factor of each tree node?
14
5 4
9 28
2 7 2 5
6 9 31
4 -1
8
4 20 44
3 -2 3
3
0 15 22
-4
1
11 33
 Tracking subtree height
• Many of the AVL tree operations depend on height.
 Height can be computed recursively by walking the tree; too slow.
 Instead, each node can keep track of its subtree height as a field:
struct tag {
<type> data;
10 data
int height; 4 4 height
struct tag * left; 10
struct tag * right; left/
} TreeNode
3 2 right
5 20

1 2 1 1
2 9 15 30

1
7
 AVL add operation
• For all AVL operations, we assume the tree was balanced before the
operation began.
 Adding a new node begins the same as with a typical BST, traversing
left and right to find the proper location and attaching the new node.
 But adding this new node may unbalance the tree by 1:

add(49); 55 55

29 87 29 87

-3 42 -3 42

49
 AVL add cases
• Consider the lowest node k2 that has now become unbalanced.
 The new offending node could be in one of the four following
grandchild subtrees, relative to k2:

1) Left-Left, 2) Left-Right, 3) Right-Left, 4) Right-Right

 Key idea: rotations
• If a node has become out of balanced in a given direction, rotate it
in the opposite direction.
 rotation: A swap between parent and left or right child,
maintaining proper BST ordering.

25 8

rotate right
8 3 25

3 11 11
 Right rotation
• right rotation (clockwise): (fixes Case 1 (LL))
 left child k1 becomes parent
 original parent k2 demoted to right
 k1's original right subtree B (if any) is attached to k2 as left subtree
 Right rotation example

43
11

11 65
8 43

8 27

3 27 65

3
 Right rotation steps
1. Detach left child (11)'s right subtree (27) (don't lose it!)
2. Consider left child (11) be the new parent.
3. Attach old parent (43) as the right child of new parent (11).
4. Attach new parent (11)'s old right subtree (27)
as left subtree of old parent (43).

43 43 11
11 65 11
65
8 43
8 27 8 27
3 27 65
3 3
 Right rotation code
TreeNode * rightRotate(TreeNode * oldParent) {
// 1. detach left child's right subtree
TreeNode * orphan = oldParent->left->right;

// 2. consider left child to be the new parent

TreeNode * newParent = oldParent->left;

// 3. attach old parent onto right of new parent

newParent->right = oldParent;

// 4. attach new parent's old right subtree as

// left subtree of old parent
oldParent->left = orphan;

oldParent->height = height(oldParent); // update nodes'

newParent->height = height(newParent); // height values

return newParent;
}
 Right rotation code
int height(TreeNode node) {
if (node == null) {
return 0;
}
int left = (node->left == null) ? 0 : (node->left)->height;
int right = (node->right == null) ? 0 : (node->right)->height;

if (left < right) return (right + 1);

else return (left + 1);
}
 Right rotation example
• After rotation?
• What is the balance factor of k2 before and after rotating?
 Right rotation example
• What is the balance factor of k2 before and after rotating?
 Left rotation
• left rotation (counter-clockwise): (fixes Case 4 (RR))
 right child k2 becomes parent
 original parent k1 demoted to left
 k2's original left subtree B (if any) is attached to k1 as left subtree
 Left rotation steps
1. Detach right child (65)'s left subtree (51) (don't lose it!)
2. Consider right child (65) be the new parent.
3. Attach old parent (43) as left child of new parent (65).
4. Attach new parent (65)'s old left subtree (51)
as right subtree of old parent (43).

43 43 65
21 65 65
21 43 87
51 87 51 87
21 51 73
73 73
 Left rotation code
TreeNode * leftRotate(TreeNode * oldParent) {
// 1. detach right child's left subtree
TreeNode * orphan = oldParent->right->left;

// 2. consider right child to be the new parent

TreeNode * newParent = oldParent->right;

// 3. attach old parent onto left of new parent

newParent->left = oldParent;

// 4. attach new parent's old left subtree as

// right subtree of old parent
oldParent->right = orphan;

oldParent->height = height(oldParent); // update nodes'

newParent->height = height(newParent); // height values

return newParent;
}
 Problem cases
• A single right rotation does not fix Case 2 (LR).

 (Similarly, a single left rotation does not fix Case 3 (RL).)

 Left-right double rotation
• left-right double rotation: (fixes Case 2 (LR))
 1) left-rotate k3's left child ... reduces Case 2 into Case 1
 2) right-rotate k3 to fix Case 1
 Left-right rotation steps
1. Left-rotate the overall parent's left child (11).
• This reduces Case 2 (LR) to Case 1 (LL).
2. Right-rotate the overall parent (43).
• This repairs Case 1 to be balanced.

43 43 27

11 65 27 65
11 43

8 27 11 31
8 31 65

31 8
 Left-right rotation example
• What is the balance factor of k1, k2, k3 before and after rotating?
 Right-left double rotation
• right-left double rotation: (fixes Case 3 (RL))
 1) right-rotate k1's right child ... reduces Case 3 into Case 4
 2) left-rotate k1 to fix Case 4
 Right-left rotation steps
1. Right-rotate the overall parent's right child (11).
• This reduces Case 3 (RL) to Case 4 (RR).
2. Right-rotate the overall parent (43).
• This repairs Case 4 to be balanced.

24 24 31

11 55 11 31 24 55

31 68 55 11 31 47 68

47 47 68
 AVL add example
• Draw the AVL tree that would result if the following numbers were
added in this order to an initially empty tree:
 20, 45, 90, 70, 10, 40, 35, 30, 99, 60, 50, 80

45

20 70

10 35 60 90

30 40 50 80 99
 Add implementation
• Perform normal BST add. But as recursive calls return, update each
node's height from new leaf back up to root.
 If a node's balance factor becomes +/- 2, rotate to rebalance it.

• How do you know which of the four Cases you are in?
 Current node BF < -1  Case 1 (LL) or 2 (LR).
• look at current node's left child BF.
 left child BF < 0  Case 1 (fix with R rotation)
 left child BF > 0  Case 2 (fix with LR rotations)
 Current node BF > 1  Case 3 (RL) or 4 (RR).
• look at current node's right child BF.
 right child BF < 0  Case 3 (fix with RL rotations)
 right child BF > 0  Case 4 (fix with L rotation)
 AVL remove
• Removing from an AVL tree can also unbalance the tree.
 Similar cases as with adding: LL, LR, RL, RR
 Can be handled with the same remedies: rotate R, LR, RL, L

remove(8);

11 11

8 43 43

27 65 27 65
 Efficiency of AVL
• An AVL tree has the same general operations as a BST, with
rebalancing added to the add/remove operations.
 add: O(log N) to walk down the tree and add the element;
O(log N) number of nodes' heights to update;
O(1) single node may need to be rotated. O(log N) overall.
 search: O(log N).
 remove: O(log N) to walk down the tree and remove the element;
O(log N) number of nodes' heights to update;
O(1) single node may need to be rotated. O(log N) overall.