算法三-----QUICKSORT

代码：
https://github.com/Shinered/quicksort
1.Partitioning Around a Pivot
Key idea: partition array around a pivot element
(1) pick element of array
(2) Rearrange array so that:
a. left of pivot —>less than pivot
b.right of pivot——->greater than pivot
(3) QuickSort(array A, length n)
– if n = 1, return
– p = ChoosePivot(A,n)
– Partition A around p
– recursively sort 1st part.
2.In-PLace Implementation
Assume: pivot=1st element of array
[if not, swap pivot <—-> 1st element as preprocessing step]
Pseudocode for Partition.. Pivot: 主轴
Partition(A, l, r): $[input = A[l,...r]]$
$i$指向小于pivot和大于pivot元素的分界线，$j$ 指向partitioned和unpartitioned元素的分界线。
$p=A[l]$
$i = l + 1$
for $j = l + 1$ to $r$
–if $A[j] < p:$ [$if A[j] > p,$do nothing.]
–Swap A[j] and A[i]
–$i = i + 1$
3.Correctness:
3.1. $A[l+1], ...A[i-1]$ are all less than the pivot.
3.2. $A[i], ...,A[j-1]$ are all greater than pivot.
conquence: at end of for loop, have after final swap, array partitioned around pivot.
1. Choose pivots
Quick sort Theorem: for every input array of lenght n，the average running time of Quick sort (with random pivots) is O(n logn).
4.1. Analysis about random pivots
Fix input array A of length n.
Samle space $\Omega =$all possible outcomes of random choices in QuickSort(pivot sequences).
Key random varible: for $\sigma \in \Omega,$
$C(\sigma) =$ #of comparisons between two input elements made by QuickSort (given random choices $\sigma$)
lemma: running time of QuickSort dominated by comparisons.
4.2 Building blocks:
Note: cant’t apply master method.
[A = fixed input array]
Notation: $z_i = i^{th}$ smallest element of A.
For $\sigma \in \Omega$ ,indices $i < j$,let:
$X_{ij} (\sigma) =$ #of times $z_i,z_j$ get compared in QuickSort with pivot sequence
A Decomposition Approach.
So: $C(\sigma) =$ # comparisons between input elements.
$X_{ij} =$ # comparisons between $z_i$ and $z_j$.
Thus: $\forall \sigma, C(\sigma) = \sum^{n-1}_{i=1} \sum^{n}_{j=i+1}X_{ij}(\sigma)$
By linearity of expectations:
$E[C] = \sum^{n-1} _{i=1} \sum^n_{j=i+1} E[X_{ij}]$
And: $E[X_{ij}] = 0\cdot P[X_{ij} = 0] + 1 \cdot P[X_{ij}=1] = P[X_{ij} = 1]$
Thus: $E[C] = \sum^{n-1}_{i=1} \sum^{n} _{j=i+ 1}P[z_i,z_j get complicated]$
Key claim: $\forall i < j, P[z_i,z_j get compared] = \frac{2}{ j-i + 1}$
Proof of Key Claim
Fixed $z_i,z_j$ with $i<j$.
So: $E[C] = \sum^{n-1}_{i=1}\sum^n_{j=i+1}\frac{2}{j-i+1}$
5.Concept#1-Sample Spaces
5.1 Sample space $\Omega=$ “all possible outcomes”
[in algorithms, $\Omega$ is usually finite]
Also: each outcome $i \in \Omega$ has a probablity $p(i) >=0$
Contraint: $\sum_{i\in \Omega} = 1$
Example #1: Rolling 2 dice.
36 ordered pairs:$\Omega = \{(1,1), (2,1), (3,1), ...,(5,6),(6,6)\}$ and $p(i)=\frac{1}{36}$ for all $i \in \Omega$
Example #2: Choosing a random pivot in outer QuickSort call.
$\Omega = \{ 1,2,...n \}$ and $p(i) = \frac{1}{n}$ for all $i \in \Omega$
5.2 Concept #2-Events
An event is a subset $S \subseteq \Omega$
The probablity of an event S is $\sum_{i \in S} p(i)$
5.3 Concept #5-Linearity of Expectation
Claim[LIN EXP]: let$x_1,....,x_n$ be random varibles.
defined on $\Omega$.Then:
$E[\sum^n_{i=1}x_i] = \sum^n_{i=1}E[x_i]$