# Chapter 08

1. Inorder-tree-walk takes $T(n)$ Time

proof: substitute : $T(n)=T(k)+T(n-k-1)+d$ Supposed that $T(n)=(c+d)n+c$
$$T(n)=T(k)+T(n-k-1)+d=((c+d)k+c)+((c+d)(n-k-1)+c)+d = (c+d)n+c-(c+d)+c+d=(c+d)n+c$$

1. In a binary search tree: Operations likes: SEARCH, MINIMUM, MAXIMUM, SUCCESSSOR and PREDECESSPR cost times are $T(h)$, whree h is the height of this bst.
2. Insert operation
y <- nil
x <- root[T]
while x != nil
do y <-x
if key[z] < key[x]
then x <- left[x]
else x <= right[x]
p[z] <- y
if y=nil
then root[T] <- z
else if key[z] < key[y]
then left[y] <- z
else right[y] <- z
1. Delete operation

Search the pointer
+ if pointer has no children: delete the node safely.
+ if pointer has only one child : swap the value between pointer and its child.
+ if pointer has two children: find the node's successor. swap them. then delete the successor .

1. Height of randomly built binary search tree

Jensen's inequality: $f[E(X)] \le E[F(x)]$ for any convex function $f$ and random variable $x$. Random variable $Y_n=2^{X_n}$ where $X_n$is the random variable denoting the height of the BST.

Lemma

$$f[E(x)]=f(\sum_{k=-\infty}^{\infty}k \cdot Pr \lbrace X=k\rbrace) \le \sum_{k=-\infty}^{\infty}f(x) \times Pr \lbrace X=k \rbrace = E[f(x)]$$

let $Y_n=2^{X_n}$. $X_n=1+max\lbrace X_{k-1},X_{n-k} \rbrace$ So
$$Y_n=2^{x_n}=2^{1+max\lbrace X_{k-1},X_{n-k} \rbrace}=2 \times max \lbrace 2^{X_{k-1}}, 2^{X_{n-k}} \rbrace = 2 \cdot max \lbrace Y_{k-1},Y_{n-k} \rbrace$$

let $Z_{nk}=1$ if the root has rank $k$, otherwise $Z_{nk}=0$ $E[Z_{nk}]=\frac{1}{n}$
So.$$Y_n=\sum_{k=1}^{n}Z_{nk}(2\cdot max \lbrace Y_{k-1},Y_{n-k} \rbrace)$$

the exceptation of $Y_{n}$
$$E[Y_{n}]=E\lbrack \sum_{k=1}^{n}Z_{nk}(2 \cdot max \lbrace Y_{k-1},Y_{n-k}\rbrace )\rbrack$$
$$=2\sum_{k=1}^{n}E\lbrack Z_{nk}(max\lbrace Y_{k-1},Y_{n-k} \rbrace ) \rbrack$$
$$\le \frac{2}{n}\sum_{k=1}^{n}E\lbrack Y_{k-1},Y_{n-k} \rbrack =\frac{4}{n}\sum_{k=0}^{n-1}E[Y_k]$$
substitute : $E[Y_n]\le cn^{3}$

$$E[Y_n]=\frac{4}{n}\sum_{k=0}^{n-1}E[Y_k] \le \frac{4}{n}\sum_{k=0}^{n-1}ck^3 \le \frac{4c}{n}\int_{0}^{n}x^3dx=cn^3$$
Conclusion：
$$2^{E[X_n]} \le cn^3 \Rightarrow E[X_n] \le 3lgn$$

# Chapter 09

1. Red-Black tree properties
• Every node is either red or black
• The root and leaves(NIL's) are black
• if a node is red, then its parent is black
• all simple paths from any node x to a descendant leaf have the same number of black nodes =$black-height(x)$
2. Height of a red-black tree

a red-black tree with $n$ keys has height $$h \le 2lg(n+1)$$

merge red nodes into their black:

a tree in which each node has 2,3,or 4 children. and has uniform depth $h'$ of leaves. So we had know that $h' \ge h/2$. Each tree has $n+1$ leaves So:
$$n+1 \ge 2^{h'} \Rightarrow lg(n+1) \ge h' \ge h/2 \Rightarrow h \le 2lg(n+1)$$

1. Rotations

# Chapter 10

Dynamic order statistics

1. OS-Select(i,S): return the $i$th smallest element in the dynamic set $S$

OS-Rank(x,S)：return the rank of $x \in S$ in the sorted order of S's elements

2. Solution: Use a red-black tree for the set S, but keep subtree sizes in the nodes

$size[x]=size[left[x]]+size[right[x]]+1$

1. Algorithm
OS-Select(i,x) ==> i th smallest element in the subtree rooted at x
k <- size[left[x]] +1
if i = k then return x
if i < k
return return OS-Select(left[x],i)
else return Os-Select(right[x],i-k)
1. Data-structure augmentation

Methodology
+ choose an underlying data structure (red-black trees)
+ determine additional infomation to be stroed in the data structure
+ Verify that this infomation can be mationtained for modifying operation
+ Develop new dynamic-set operations that use this information

1. Interval tree

addtional information stroe in each node x the largest value $m[x]$ in the substree root at x, as well as the interval $int[x]$ correspoinding to the key.

Algorithm

Interval-search(i)
x <- root
while x != nil and (low[i] > high [int[x]] or low[int[x]] > high [i] )
do ==> i and int[x] don't overlap
if left[x] != nil and low[i] <= m[left[x]]
then x <- left[x]
else x <- right[x]
return x

Correctness
+ if the search goes right, then $\lbrace i' \in L : i' overlaps i\rbrace = \oslash$
+ if the search goes left, then if $\lbrace i' \in L: i' overlaps i\rbrace= \oslash \Rightarrow \lbrace i' \in R: i' overlaps i\rbrace=\oslash$

Write a Comment