Lecture 10, Oct 28 2014

Binary Search Trees

 In addition to insert/delete:
» Heaps supported min/max.
» Hashing supported search.
» What if we want both min/max/search, and also pred/succ ?

 Binary Search trees:

x
a  left tree key(a)  key( x)
a  right tree key(a)  key( x)
y z

122

Examples
 Legal Binary Search Trees: 25
8
9 30
3 25
8

1 5 9 30 5

 In-Order traversal: InOrder(left(x)) 1

print(x)
InOrder(right(x))

 Note that given Binary tree, can output sorted in O(n) time !
Gives lower bound on constructing Binary Tree.
(Compare with building a Heap)

123

1
 Searching in Binary Tree
8
 Check if “current node” is =x or =NIL.
Equal to x -- return,
Equal to NIL -- return “not found” 3 25
else:
if x  current key ---- search in the left tree
otherwise ---- search in the right tree.
1 5 9 30

 “Go down the tree, turning right/left as appropriate…”

 Running time: (h), h=height of the tree.

 Note that this was impossible to do with a heap

124

Inserting into Binary Tree

 Insertion: search for key, and put it in the first empty space.

 Insertion takes (h).

25

 Sort: 9
» Insert item-by-item,
» in-order walk. 8

» (n2)… 5

125

2
 Relation to Quicksort

 Randomly permute input.

 Consider example: 3 1 8 2 6 7 5
» Quicksort chooses 3, then compares 1,8,2,6,7,5 to 3.
Then chooses 1, compares to 2
chooses 8, compares 6,7,5 to 8.
» Binary Tree: chooses 3, places as root
Then chooses 1, compares with 3, put in place.
chooses 8, compares with 3, put in place
etc…
» Overall, same comparisons, only different order !!

126

Successor/Predecessor
 Min/max - go all the way left or all the way right.

 Successor:
If right(x) != NIL, return TreeMin(right(x)) Easy case
y=parent(x)
while y != NIL & x=right(y)
x=y
y=parent(x) Go up the tree to the LEFT
return y

3 25

1 5 9 30

4 6

 Successor of 6 is 8. Predecessor of 8 is 6.
 What if we go “up left” and reach root, i.e. cannot ever go “up right” ?
127

3
 Deletion

 3 Cases:
» No children – simple delete, replace pointer from parent with null
» 1 child – delete, redirect pointer from parent to point to child
» 2 children – cannot simply delete

 3rd case: put successor(x) instead of x.

» Binary Tree property satisfied.
» Delete “hole” using case 1 or 2.
» Why ??
» (successor does not have left child !)

128

Delete element - example

8
8

3 25
4 25

1 5 9 30
1 5 9 30

4 6
4 6
4.5
8 4.5

3 25

1 5 9 30

4.5 6

4.5 129

4
 Summary

 Data structure implements insert/delete/search/pred/successor

 Speed depends on height of the tree, in general can be large
 Need a way to “rebalance the tree”
 Possible to rebalance so that height < 2log n
 Basic operation: “Rotation”

130

Red/Black trees

 One of the ways to approximately balance, i.e. depth O(lg n).

 Red/Black property:
» Node is either black (default) or red.
» x red children(x) are black.
» Every simple root-leaf path has same # of black nodes.
» Root is black (for convenience).
» Add NIL to leaves so “real nodes” have 2 children, color leaves black.

 One of the ways to approximately balance.

» We will show that above condition implies depth O(lg n).

131

5
 Black height

 bh(x) = # of black nodes on x leaf path, not counting x.

132

Balancing

 Observation: bh(root)h/2

 THM: h  2 log(n+1)
Proof:
» Claim: #nodes in subtree rooted at x 2bh(x)-1.
» By induction on h(x). Base case is trivial. x

» if y black, bh(y)=bh(x)-1 y z
otherwise, bh(y)=bh(x)
» Thus, bh(y) bh(x)-1

» #nodes in x-tree at least (2bh(x)-1-1)2+1= 2bh(x)-1,

which proves the claim.
» But: bh(root)  h/2, QED.

133