CS213/293 Data Structure and Algorithms 2024

Lecture 7: Red-Black Trees

Instructor: Ashutosh Gupta

IITB India

Compile date: 2024-09-22

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 1
 Topic 7.1

Balance and rotation

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 2
Maintain balance

BST may have a large height, which is bad for the algorithms.

Height is directly related to branching. More branching implies a shorter height.

We call BST imbalanced when the difference between the left and right subtree height is large.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 3
Balancing height by rotation
max(h1 , h2 + 1, h3 + 1) + 1 max(h1 + 1, h2 + 1, h3 ) + 1
(A, B)-rotate
A B

h1 B A h3
T1 T3
(B, A)-rotate
h2 h3 h1 h2
T2 T3 T1 T2

For example, if h3 > h2 = h1 and we rotate the BST, we will get a valid and more balanced BST
with less height.

Rotation may be applied in the reverse direction, where A is the left child of B. We define the
symmetric rotation in both directions.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 4
 Example: rotation

Example 7.1
In the following BST, we can rotate 8-17 edge.

8 17
(8, 17) rotate

5 17 8 21

11 21 5 11 18 30

18 30
Commentary: This tree operation is important. Please carefully observe the destination of red,blue, and brown subtrees.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 5
When to rotate? Can only rotation fix imbalance?

Rotation is a local operation, which must be guided by non-local measure height.

We need a definition of balance such that rotations operations should be able to achieve the
balance.

Design principle:
We minimize the number of rotations while allowing some imbalance.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 6
 Topic 7.2

Red-black tree

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 7
Null leaves

To describe a red-black tree, we replace the null pointers for absent children with dummy null
nodes.
Example 7.2
The following tiny nodes are the dummy null nodes.

5 17

21

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 8
Red-black tree

Definition 7.1 Example 7.3

A red-black tree T is a binary search tree such An example of a red-black tree.
that the following holds.
▶ All nodes are colored either red or black 8
▶ Null leaves have no color
5 17
▶ Root is colored black
▶ All red nodes do not have red children 21
▶ All paths from the root to null leaves
have the same number of black nodes.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 9
Exercise: Identify red-black trees
Exercise 7.1
Which of the following are red-black trees?

17 17

5 5 17

16 16 5 17

Observations:
▶ Red nodes are not counted in the imbalance. We need them only when there is an imbalance.
▶ There cannot be too many red nodes. (Why?)
▶ Red nodes can be at every level except the root.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 10
Black height

Definition 7.2
The black height (bh) for each node is defined as follows.

0
 n is a null leaf
bh(n) = max(bh(right(n)), bh(left(n))) + 1 n is a black node

max(bh(right(n)), bh(left(n))) n is a red node


Until it is stated otherwise, all heights mentioned in this lecture are black heights.

Exercise 7.2
Prove that for each node n in a red-black tree, bh(right(n)) == bh(left(n)).

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 11
Example: black height

Example 7.4
The black height of the following red-black tree is 2.

Exercise 7.3
Can we change the color of some nodes without breaking the conditions of a red-black tree?

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 12
 Bound on the height of a red-black tree

Let h be the black height of a red-black tree containing n nodes.

▶ n is the smallest when all nodes are black. Therefore, the tree is a complete binary tree.
Therefore, n = 2h − 1.

▶ n is largest when the alternate levels of the tree are red. The height of the tree is 2h.
Therefore, n = 22h − 1.

log4 n < h < 1 + log2 n

Exercise 7.4
Define red-black tree inductively and write the proof of the above bounds using induction.

Commentary: Our definition of red-black tree is not inductive therefore the above proof does not read so formal. However, an inductive definition would make proof formal.
Can you write an inductive definition of red-black tree and prove the above statements?
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 13
Search, Maximum/Minimum, and Successor/Predecessor

We can search, maximum/minimum, and successor/predecessor on the red-black tree as usual.

Their running time will be O(log n) because h < 1 + log2 n.

How do we do insertion and deletion on a red-black tree?

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 14
 Topic 7.3

Insertion in red-black tree

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 15
 BST insertion in red-black tree

1. Follow the usual algorithm of insertion in the BST, which inserts the new node n as a leaf.
▶ Note that there are dummy nodes. n is inserted as the parent of a dummy node.

2. We color n red.

▶ Good news: No change in the black height of the tree.

▶ Bad news: n may have a red parent.

After insertion, we may have a red-red violation, where a red node has a red child.

Commentary: We have null nodes in this setting. We need to add nulls as children to the new node.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 16
Example: insert in red-black tree

Example 7.5
Inserting 20 in the following tree.

8 8
Insert 20

5 17 5 17

21 21

20

The insertion results in violation of the condition of the red-black tree, which says red nodes can
only have black children.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 17
Iteratively remove red-red violation
A red-red violation starts at a leaf. However, an attempt to remove the violation may push up the
violation to the parent.

Commentary: We attempt to remove the violation by

local transformations on the tree. The transformation
may cause the violation to traverse up the tree until we
reach the root. In the following, we need to consider a
p general situation where a violation has occurred in the
middle of the tree.
red-red violation
n

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 18
 Commentary: Since the new node is always added at
the leaf, one may think that h must be zero, which
Pattern around red-red violation may not be the case. As noted in the previous slide,
the violation may move upwards.

If n has a red parent, we correct the error by rotation and re-coloring operations.

h+1 g Orange means that we need to consider all possible colors of

the nodes.
u p
We have three cases.
n h
h is the T4 ▶ Case 1: u is red “not red” means either
black black node or null node.
h h
height. T2 T3 ▶ Case 2: u is not red and the g to n path is not straight

▶ Case 3: u is not red and the g to n path is straight

Exercise 7.5
Why must g exist and be black? No transformation should change the black height of g .

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 19
Case 1: The uncle is red
h+1 g h+1 g
re-color

h u p h+1 u p
T1 T1
n h n h
T4 T4
h h h h
T2 T3 T2 T3

In the subtree of g , there is no change in the black height and no red-red violation.

Now g is red. We have three possibilities: the parent of g is black, the parent of g is red, and g is
the root.
Commentary: Possibility 1: Nothing. Possibility 2:
Exercise 7.6 We have a red-red violation a level up and need to
apply the transformations there. Possibility 3: turn g
What do we do in each of the possibilities? back to black.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 20
Case 2: The uncle is not red and the path to the grandparent is not straight
straight means left 2 (parent 2 (n)) = n
or right 2 (parent 2 (n)) = n

h+1 g (p,n) rotate h+1 g

h p h n
T1 T1
n h h p
T4 T2
h h h h
T2 T3 T3 T4

This transformation does not resolve the violation but converts the violation to case 3.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 21
Case 3: The uncle is not red and the path to the grandparent is straight

h+1 g h+1 p

h p g n
(g , p) rotate
T1
n swap colors
h h h h h
T2 T1 T2 T3 T4
h h
T3 T4

The transformation removes the red-red violation.

Exercise 7.7
a. Why are the roots of T2 , T3 , and T4 not red?
b. Show that if the root of T1 is red then the above operation does not work.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 22
Example: red-red correction case 1

Example 7.6
We just inserted 20 in the following tree. We need to apply case 1 to obtain a red-black tree.

8 8
re-color

5 17 5 17

12 21 12 21

20 20

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 23
Example: red-red correction case 2
Example 7.7
Consider the following example. We are attempting to insert 20. We apply case 2 to move towards
a red-black tree.

8 (20,21)-rotate 8

5 17 5 17

21 20

20 21

The above is not a red-black tree. Furthermore, we need to apply case 3 to obtain the red-black
tree.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 24
Example: red-red correction case 3 (continued)

We apply case three as follows.

8 (17,20)-rotate 8

and color swap

5 17 5 20

20 17 21

21

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 25
 Summary of insertion

Insert like BST, make the new node red, and run the following algorithm.

Algorithm 7.1: RedRedRepair(Node n)

ASSUME(there is no red-red violation at any other node);
if n is root then n.color := black; return;
p := parent(n);
if p is black then return;
g := parent(p); u := sibling(p);
switch(pattern around n):
case 1: Re-color according to case 1; RedRedRepair(g);
case 2: Rotate according to case 2; RedRedRepair(p); // p is in case 3
case 3: Re-color and Rotate according to case 3;

Commentary: What is the intuitive difference between cases 1 and 3? Case 1 reduces the total number of red nodes in a subtree. Case 3 balances if there are too many red
nodes on one side of the subtree.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 26
 Topic 7.4

Deletion in red-black tree

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 27
BST deletion in red-black tree

▶ Delete a node as if it is a binary search tree.

▶ Recall: In the BST deletion we always delete a node n with at most one non-null child.

p Remove n p

n x
T2 T2
x T1

T1

x can be either a null or non-null node.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 28
 What can go wrong with a red-black tree?

Since a child x of n takes the role of n, we need to check if x can replace n.

▶ If n was red, no violations occur. (Why?)

▶ If n was black, bh(x) = bh(n) − 1, or it is possible that both x and p are red.
| {z } | {z }
black height violation red-red violation

The leaves of the subtree rooted at

x will have one less black depth.

Commentary: The or in the second bullet point is not xor. Both violations are possible at the same time.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 29
Violation removal procedure
To remove the violation at x, we may recolor and rotate around x, which may push the violation to
the parent. Therefore, we consider that the violation is in the middle of the tree.

Both p and x may be red and

p may also have a red-red violation.
height violation
h to h − 1 x h s

Orange means that we need to consider all possible colors of the nodes.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 30
Violation pattern
After deletion, we may need to consider the following five nodes around x.

p We correct the violation either by rotation or

re-coloring.
h to h − 1 x h s
There are six cases
T1 1. x is red
c1 c2
2. x is not red and root.
T4
3. x is not red and s is red
T2 T3 4. x is not red and s is black
4.1 c2 is red
Exercise 7.8 4.2 c2 is not red and c1 is red
Show: If x is not root and not red, 4.3 c2 is not red and c1 is not red
s must exist. The goal is to restore the black height of p.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 31
Case 1: x is red

p p
re-color x

h to h − 1 x h s h x h s
T1 T1
c1 c2 c1 c2
T4 T4
T2 T3 T2 T3

Violation solved!

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 32
Case 2: x is not red and root.

Do nothing.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 33
Case 3: x is not red and the sibling of x is red
p s
(p, s) rotate
and swap colors p
h to h − 1 x h s h c2
T1 T4
c1 c2 h−1 x h c1
T4 T1
T2 T3 T2 T3

The transformation does not solve the height violation at the parent of x but changes the sibling of
x from red to black.

Exercise 7.9
Why must p, c1 , and c2 be non-null and black?
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 34
Case 4.1: x is not red, the sibling of x is black, and the right nephew is
red (Assuming x is the left child)
The color of p and c1 does not matter.
p s
(p, s) rotate
swap the color of p and s
h to h − 1 x h s and c2 is turned black h p h c2
T1 T4
h − 1 c1 h − 1 c2 h−1 x h − 1 c1
T4 T1
T2 T3 T2 T3

The above transformation solves the black height violation.

Exercise 7.10
Write the above case if x is the right child of p.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 35
Case 4.2: x is not red and the sibling is black, and the left and right
nephews are red and not red respectively (Assuming x is the left child)
p p
(s, c1 ) rotate
and swap colors
h to h − 1 x h s x h−1 h c1
T1 T1
c1 c2 s
T2
T4
c2
T2 T3 T3
T4

The above transformation does not solve the height violation. It changes the right child of the
sibling from not red to red, which is the case 4.1.
Exercise 7.11
a. Why can case 4.1 transformation not be applied to case 4.2?
b. Write the above case if x is the right child of p.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 36
Case 4.3: x is not red, the sibling of x is black, and the nephews of x are
not red
p p
color s red
color p black
h to h − 1 x h s h−1 x h−1 s
T1 T1
c1 c2 c1 c2
T4 T4
T2 T3 T2 T3

The above transformation reduces bh(p) by 1, if p was black, and there may be a potential
violation at p, which is at the lower level.

We apply the case analysis again at node p. The only case that kicks the can upstairs!! All cases
are covered.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 37
Structure among cases

▶ Cases 1, 2, and 4.1 solve the violation at the node.

▶ Case 3 turns the violation into 4.1 or 4.2.
▶ Case 4.2 turns the violation into 4.1.
▶ Case 4.3 moves the violation from x to its parent p.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 38
Summary of deletion

1. Delete like BST. There may be a black height violation at the child of the deleted node.
2. While case 4.3 occurs, re-color nodes and move up the black height violation.
3. For all the other cases, we rotate or re-color, and the violation is finished.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 39
 Topic 7.5

Tutorial problems

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 40
Exercise: validity of rotation

Exercise 7.12
Prove that after rotation the resulting tree is a binary search tree.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 41
Exercise: sorted insert

Exercise 7.13
Insert sorted numbers 1,2,3,..., 10 into an empty red-black tree. Show all intermediate red-black
trees.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 42
Insert and delete
Exercise 7.14
Consider the tree below. Can it be colored and turned into a red-black tree? If we wish to store
the set 1, . . . , 22, label each node with the correct number. Now add 23 to the set and then
delete 1. Also, do the same in the reverse order. Are the answers the same? When will the answers
be the same?
A

B C

D E F G

H I J K L M N

O P Q R S T U

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 43
Exercise: Hash table vs red-black tree

Exercise 7.15
a. Give running time complexities of delete, insert, and search in the red-black tree.
b. What are the advantages of the red-black tree compared to the Hash table, where every
operation (search, insert, delete) is almost constant time?

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 44
 Topic 7.6

Problems

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 45
 Exercise: consistent hashing** (midsem 2024)
Exercise 7.16
Let us suppose we want to store n keys on m servers. The servers have IDs. We may use a hash function h
to map the keys to servers. Both keys and IDs of the servers are hashed. The computed hash values are
placed on a ring, like a clock. Some of the points on the ring are the servers and some are keys. A key is
stored on the closest server in the clockwise direction. Each time a server is added/removed the keys are
moved according to the above rule.
8 9
Example: Let h(x) = 1 + x%11. We want to store keys 21, 60, and 90 on 7
10 141
three server with IDs 141, 201, and 738. The hash values of the keys are 60 6
11, 6, 3 and the hash values of the servers are 10, 4, 2. The keys 21, 60, 11 21
and 90 will be stored in servers 738, 141, and 201 respectively. 5
1
201 4
3 2
a. Give an efficient algorithm for adding/removing keys on the servers. 738
90
b. Give an efficient algorithm for implementing add/remove of servers.
Commentary: Solution: To implement the above scheme. We store the server hashes in a red black tree R.
findNext(k,s,s’){ if( s=Null) return s’;if(key(s) < k) return findNext(k,key(s),s’); else return findNext(k,left(s),s);}
findServer(k){s = findNext(k,R.root,Null) if(s == Null) s = minimum(R.root); if(s == Null ) Return Null; Return value(s)}
AddKey(k){ s = findServer(h(k));storeInServer(k,s)} RemoveKey(k){ s = findServer(h(k),R.root); removeInServer(k,s)}
AddServer(s){ s’ := h(s) R.Insert( (s’,s) ); t = nextServer(s’); for k in GreaterKeysOnServer(s’,t) {removeInServer(k,t);storeInServer(k,s)} }
RemoveServer(s){ s’ := h(s) R.Delete(s’); t = nextServer(s’); for k in KeysOnServer(s) { removeInserver(k,s);storeInServer(k,t); } }
Assume red-black Insert takes (key,value) pair as input and no two servers have collision, which can be easily avoided by renaming the IDs. Keys are again stored as
Red-black tree on the servers such that we can efficiently implement GreaterKeysOnServer. removeInServer are storeInServer are trivial calls to insert/delete of RB Tree.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 46
 Excercise: Is BST red-black colorable?**

Exercise 7.17
Given a BST, can we check if there exists a colouring which makes it a valid RB-Tree?

Commentary: Solution: https://cs.stackexchange.com/questions/10990/colour-a-binary-tree-to-be-a-red-black-tree

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 47
 Topic 7.7

Extra slides: AVL trees (In GATE/GRE syllabus)

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 48
AVL (Adelson, Velsky, and Landis) tree
Definition 7.3
An AVL tree is a binary search tree such that for each node n

|height(right(n)) − height(left(n))| ≤ 1.

Example 7.8 8
An example of an AVL tree.

5 17

7 11 21

10 14
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 49
Exercise: Identify the AVL trees

Exercise 7.18
Which of the following are AVL trees?

17

17 21 17 17

5 20 5 21 5 21

3 18 11 5 11

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 50
 Topic 7.8

Height of AVL tree

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 51
 AVL tree height

Theorem 7.1
The height of an AVL tree T having n nodes is O(log n).

Proof.
Let n(h) be the minimum number of nodes for height h.

Base case:
n(1) = 1 and n(2) = 2.

Induction step:
Consider an AVL tree with height h ≥ 3. In the minimum case, one child will have a height of
h − 1 and the other child will have a height of h − 2. (Why?)

Therefore, n(h) = 1 + n(h − 1) + n(h − 2). ...

Commentary: We need to show that n(h) > n(h − 1) is monotonous. Ideally, n(h) = 1 + n(h − 1) + min(n(h − 2), n(h − 1)). This proves that n(h) > n(h − 1).

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 52
AVL tree height(2)
Proof(continued.)
Since n(h − 1) > n(h − 2),
n(h) > 2n(h − 2).

Therefore,
n(h) > 2i n(h − 2i).

For i = h/2 − 1(Why?), Commentary: Here is the explanation of the last step.
Consider an AVL tree with m nodes and h height. By
h/2−1 h/2
n(h) > 2 n(2) = 2 . definition, h(n) ≤ m. Since h < 2 log n(h), h <
2 log m. Therefore, h is O(log m).

Therefore,
h < 2 log n(h).

Therefore, the height of an AVL tree is O(log n). (Why?)

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 53
Closest leaf

Theorem 7.2
Let T be an AVL tree. Let the level of the closest leaf to the root of T is k.

height(T ) ≤ 2k − 1
Proof.
.
Let D be the closest leaf of the tree.
2k − 1 r
▶ The height of right(C ) cannot be more than 2. (Why?)

▶ Therefore, the maximum height of C is 3. 5 B

▶ Therefore, the maximum height of right(B) is 4.
▶ Therefore, the maximum height of B is 5. 3 C
2k − 2 4
▶ Continuing the argument, the maximum height of
root r is 2k − 1. 1 D
2

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 54
A part of AVL is a complete tree

Theorem 7.3
Let T be an AVL tree. Let the level of the closest leaf to the root of T is k. Upto level k − 2 all
nodes have two children.

Proof.
A node at level k − 2 − i cannot be a leaf. (Why?)

Let us assume that a node n at level k − 2 − i has a single child n′ .

The height of n′ cannot be more than 1. (Why?)

Therefore, n′ is a leaf. Contradiction.

Exercise 7.19
Show T has at least 2k−1 nodes.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 55
Another proof of tree height bound

Let T have n nodes and the height of T be h.

We know the following from the previous theorems.

▶ n ≥ 2k−1 , and
▶ 2k − 1 ≥ h.

Therefore,
n ≥ 2k−1 ≥ 2(h−1)/2

Exercise 7.20
What is the maximum number of nodes given height h?

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 56
Problem: A sharper bound for the AVL tree

Exercise 7.21
a. Find largest c such that c k−2 + c k−1 ≥ c k
b. Recall n(h) = 1 + n(h − 1) + n(h − 2). Let c0 be the largest c. Show that n(h) ≥ c h−1 .
c. Prove that the above bound is a sharper bound than our earlier proof.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 57
 Topic 7.9

Insertion and deletion in AVL trees

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 58
Insert and delete

Insert and delete like BST.

At most a path to one node may have height imbalances of 2. (Why?)

We have to repair height imbalances by rotations around the deepest imbalanced node.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 59
Rebalancing AVL trees
Let x be the deepest imbalanced node. Let t be the taller child. Let g be the grandchild via t that
is not on the straight path from x.

h+3 x

h h+2 t
T1
g
T4

T2 T3

Three cases:
1. Case 1: Height of g is h + 1 and T4 is h + 1.
2. Case 2: Height of g is h and T4 is h + 1.
3. Case 3: Height of g is h + 1 and T4 is h.
cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 60
Case 1: Both grandchildren via t have height h + 1

h+3 x rotate(x,t) h+3 t

h h+2 t h+2 x h+1

T4
T1
h+1 g h+1 h h+1 g
T4 T1

T2 T3 T2 T3

The imbalance in the subtree is repaired. We check the parent of t.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 61
Case 2: Right-left grandchild has height h

h+3 x rotate(x,t) h+2 t

h h+2 t h+1 x h+1

T4
T1
h g h+1 h h g
T4 T1

T2 T3 T2 T3

Imbalance is repaired. But, the parent of t may need repair.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 62
Case 3: Right right grandchild has height h

h+3 x rotate(t,g) h+2 g

h h+2 t rotate(x,g) h+1 x h+1 t

T1
h+1 g h h h
T4 T1 T2 T3 T4
{h, h − 1} ∋ ∈ {h, h − 1}
T2 T3

Imbalance is repaired. But, the parent may need repair.

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 63
Complexity of insertion/deletion

Exercise 7.22
a. What is the bound on the number of rotations for a single insert/delete?
b. Compare the bounds with RB trees insertion/deletion.
c. Which definition is more strict RB or AVL? Or, are they incomparable?

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 64
 End of Lecture 7

cbna CS213/293 Data Structure and Algorithms 2024 Instructor: Ashutosh Gupta IITB India 65