Lecture Sorting

Sorting Algorithms
A Guide
Joshua Knowles
School of Computer Science
The University of Manchester
COMP26120 - Week 9 Roscoe Th. B, Nov 26 2010

The Importance of Sorting
Important because
• Fundamental to organizing data

• Good for teaching about algorithms/complexity
Every algorithms book has a large section on Sorting...

Joshua Knowles 2 Roscoe Th. B, Nov 26 2010
...On the Other Hand
• Progress in computer speed and memory has reduced the
practical importance of sorting
• quicksort() is often an adequate answer in many applications
However, you still need to know your way (a little) around the sorting
“jungle”.

Overview
What you should learn about sorting (what is examinable)
• Definition of sorting. Correctness of sorting algorithms
• How the following work: Bubble sort, Insertion sort, Selection

sort, Quicksort, Merge sort, Heap sort, Bucket sort, Radix
sort
• Main properties of those algorithms
• How to reason about complexity — worst case and special cases
Covered in: the course book; labs; this lecture; wikipedia; wider
reading

Relevant Pages of the Course Book
Selection sort: 97 (very short description only)

Insertion sort: 98 (very short)
Merge sort: 219–224 (pages on multi-way merge not needed)
Heap sort: 100–106 and 107–111
Quicksort: 234–238
Bucket sort: 241–242
Radix sort: 242–243
Lower bound on sorting 239–240
Practical issues, 244
Some of the exercise on pp. 251–252. (More advice to follow)

Sorting is Easy
Where does sorting sit in a list of growth rates?
√
O(1), O(log n), O( n), O(n), O(n log n), O(n2) ,
O(n3), ..., O(2n), O(n!)
Since the common sorting methods are upper-bounded by a
polynomial — they are all O(n2)† — the sorting problem is tractable.
This means running times should not stretch out to the age of the
Universe except for VERY large inputs.
But in practice, since n can be quite large (thousands, millions),
O(n log n) methods are valuably quicker than Ω(n2) methods. This is
why there has been research effort in devising efficient sorting
methods.
†
Remember, an algorithm with O(n log n) complexity is also in
O(n2).

The Broad View of Sorting
Algorithms
Bubblesort is terrible
Quicksort is good
More precisely, since Bubblesort is O(n2) and Quicksort is

O(n log n) (for most inputs), Quicksort will massively outperform
Bubblesort for larger input sizes.
Remember, for n = 106, n2 is 1012, but n log n is only about 2.107.
Note also: Bubblesort is poor in practice even compared with other

O(n2) algorithms like insertion sort.

Other Sorting Algorithms
...Of course, there are many other sorting algorithms. Here are the time complexities
associated with the most important ones (worst case, except for Quicksort).
Bucket sort
Counting sort O(n)
Radix sort
Mergesort
Heapsort O(n log n)
Quicksort
Tree sort
Insertion sort
Selection sort O(n2)
Bubble sort
The first three can only be used on special cases though, as we shall see.

Definition of Sorting
What is a sorting algorithm?
Input:
an (ordered) array of keys, array
All keys must be comparable. The keys must form a total order:
If X is totally ordered under ≤, then the following statements hold for all a, b and c in
X:
If a ≤ b and b ≤ a then a = b (antisymmetry);
If a ≤ b and b ≤ c then ≤ c (transitivity);
a ≤ b or b ≤ a (totality).
(Note: an array of items all with the same value is still a total order).
Output:
a sorted array sortedArray such that
1. sortedArray is a permutation of array and

2. keys are in ascending order: sortedArray[i] <= sortedArray[j] for all i, j
where i < j .

Not Only Numbers
Much of the time, we want to sort numbers into numerical order.
Comparison is then straightforward.
But we also sort:
• names and words
• dates, written in different forms
• complex, multi-field data
• books, by e.g. Dewey-Decimal System
What are the rules for comparison? Do they vary? Homework: Write
the compare() function for the DDS
But all of these lists are totally ordered. So it is possible to sort them
correctly, provided we get the comparison rules right.
[Demo - See demo1.c]

Non-transitivity: Rock - Paper -
Scissors
Be careful! Sometimes, the items in an array or list are not

comparable in the right way. They cannot be sorted using a
standard sorting algorithm. [ See demo2.c ]
In the game, Rock - Paper - Scissors we have

Rock > Scissors
Scissors > Paper
Paper > Rock
It is impossible to determine what comes first. This is not a strictly
ordered set of things.
[DEMO - See demo3.c]

The problem is that we are not using the > sign in the correct way. It
must be a transitive relation. In Rock-Paper-Scissors it is
non-transitive.

A Basic Sorting Algorithm:
Insertion Sort
Why does insertion sort have time complexity in O(n2 ) ?

What is its space complexity? How much additional space does it need?
Insertion sort is in O(n) for already sorted input. Why ?

A Basic Sorting Algorithm:
Selection Sort
Selection sort uses the same number of comparisons, independently

of whether the input is sorted, reverse-sorted, or unsorted. Why?
Selection sort is the basis of Heap Sort.
Question: What are the other important properties of Selection Sort
and Insertion Sort? (See next slides for explanation of properties)
Some Properties of Sorting
Algorithms
(other than time complexity)
Space Complexity
Some sorting algorithms require that the data are copied to a new list
during the sort. This gives a space complexity of O(n).
Some other sorting algorithms require only a constant amount of
additional space, often just a single variable. These sorting
algorithms are said to be in-place. They require O(1) extra space.
Stability
If items with equal keys are guaranteed to remain in the same order
(not position!) they were in the input, then the sort is stable.
(You met stability in the Lab on Lexicographic Sorting.)
Some Properties of Sorting
Algorithms
General Purpose
If an algorithm can sort any list of items belonging to a total ordering,

without restrictions, then it is general purpose.
If, on the other hand, an algorithm requires the items to be restricted

to a finite number of keys, then it is not general purpose.
Comparison-based sorts are general purpose.

Distribution sorting algorithms (e.g. bucket sort) are not general
purpose. They rely on knowing a priori something useful (limiting)
about the universal set from which the elements to be sorted are
drawn. E.g. they range from 200–1000.

Time Complexity: Different Cases
Counting the Sorting Operations
The basic operations of sorting algorithms are comparison, and

copy or swap.
When the complexity of an algorithm is calculated, usually only one of

the basic operations is counted. It can be important to state which
one.
For some applications, comparison is expensive. For others, copying

is expensive. (For example, if the items to be sorted are large
database records, it can be expensive to copy them). Selection sort
uses only O(n) swaps, whereas Bubblesort uses O(n2).
If records are large, however, it can be better to apply a sorting

algorithm to pointers to the records. After sorting, the complete
records can be ordered in O(n) copy operations.
Unsorted/sorted Inputs
Often the complexity on worst-case inputs is given.
But some algorithms have poor worst-case performance, yet have

good performance on most inputs, e.g. quicksort. For quicksort, the
complexity usually quoted is for a typical (unsorted) input.
Quicksort has poor peformance O(n2) on sorted or reverse-sorted

lists. [Homework: Why?]
Insertion sort is linear O(n) on ‘nearly’ sorted or sorted lists.

[Homework: Why?]

Small input size
Some sorting algorithms have very low overhead, and so they tend to
be very efficient on small input sizes, even if their asymptotic
complexity is poor. Selection sort is like this. It is often used to sort
lists shorter than 16 within a quicksort routine.
Many values the same (duplicates)
Some sorting algorithms perform poorly when many keys in the input
have the same value. Which algorithms? Why? On the other hand,
Bingo sort is a variant of selection sort that is efficient if many values
are duplicates. (see wikipedia)

Performance Profiling
If the speed of sorting on a particular platform/input data is very

important, then the best way to select an algorithm is by experimental
performance profiling.

128 256 512 1024 O1024 R1024 2048
Bubblesort 54 221 881 3621 1285 5627 14497
Insertion Sort 15 69 276 1137 6 2200 4536
Selection Sort 12 45 164 634 643 833 2497
Quicksort 12 27 55 112 1131 1200 230
Quicksort1 6 12 24 57 1115 1191 134
Mergesort 18 36 88 188 166 170 409
Mergesort1 6 22 48 112 94 93 254
Table 1: Time in microseconds of some sorting algorithms. O1024
means the input of size 1024 was in sorted order. R1024 means it
was in reverse sorted order. Quicksort1 and Mergesort1 use Selection
Sort to sort small subarrays (≤ 16 elements)
NB: These results are reproduced from IT Adamson (1996), page 155.

Time Complexity Bounds for
Sorting
Upper Bound Any existing sorting algorithm provides an upper
bound for the task of sorting.
E.g. Mergesort has worst-case time complexity of O(n log n). In

other words, from the existence of Mergesort, we know it is possible
to sort n elements in at most c.n log n comparisons for some
sufficiently large constant c and all n > no, where no is a constant.
(In practice, c = 2 and no = 10 would be sufficient)
Lower Bound What is the minimum amount of work needed to sort

an unsorted array? We must at least have to read each value in the
array, so a lower bound is O(n).
Bucket sort and Radix sort achieve this lower bound, but only on
restricted inputs.
Lower Bound for
Comparison-Based Sorting
No algorithm based on comparison of keys can sort a worst-case

input sequence in fewer than dlog2(n!)e comparisons. This is lower
bounded by Ω(n log2 n).

The input sequence is a, b, c, and a 6= b, b 6= c, a 6= c. We can see
there are n! leaves in the decision tree, one for each possible
permutation of the input. In order to be able to locate the correct
order (for the particular values of a,b,c), the decision tree must have a
depth of dlog2(n!)e.
And
dlog2 n!e ≥ log2 n! (1)
X n
≥ log2 i (2)
i=1
n/2
X
≥ log2 n/2 (3)
i=1
≥ n/2 log2 n/2 (4)
is Ω(n log n). (5)
Mergesort, Heapsort are asymptotically optimal sorting algorithms.

Quicksort is asymptotically optimal for most inputs.
O(n log n) Sorting
• Merge sort
• Quicksort
• Heap sort

Merge Sort
Principle
MergeSort is a Divide-and-Conquer approach. The input array is
divided recursively until there are n lists of length 1. All the sorting
‘work’ is done in the merge step.
Observation: given two sorted lists of total length n, it is possible to

merge them into a single sorted list in O(n) time.
Properties
• MergeSort has worst-case time complexity of O(n log n) in the

number of comparisons.
• There is no implementation as an in-place sort.

• It is stable. *Depends on implementation

Merge Sort: Merging Operation
↓
List1: 12 16 19
↓
List2: 11 12 12 35
↓
Output: 11
Compare the two pointed-at values. Copy the smaller one into the
pointed-at place in the output. If the two values compared are equal,
copy the one from List1. Move the pointer of the output, and the input
list we copied from, one place to the right.

MergeSort: Merging Operation
↓
List1: 12 16 19
↓
List2: 11 12 12 35
↓
Output: 11 12 12 12 16 19 35

Further Merge Sort Properties
Merge sort’s running time is almost unaffected by the ordering of the

input (see the table of running times on a previous slide). Why?
How can the merge operation be optimized slightly?
How is Merge sort implemented when input size n is not a power of

2?
What steps in Merge sort ensure it is stable?

Heap sort
See: pp. 99–111 of the course textbook for full details.
Principle: Store the items one by one in a heap - a priority queue

data structure. Then remove the items again in sorted order.
You will use (and possibly implement) a priority queue in the lab on
knapsack problems.
You have already stored dictionary words in a tree to sort them. The
principle is the same.
Properties:
Time complexity is O(n log n) on worst-case inputs
Heap sort can be implemented in-place by implementing the heap
within a portion of the input array
Heap sort is not stable. Why not?

QuickSort
Invented by ‘Tony’ Hoare in 1960.

QuickSort
Principle: Quicksort is a divide-and-conquer algorithm. All the
sorting work is done in the divide (partitioning) step. Merging back the
sublists is trivial.
Partitioning Step: A ‘pivot’ element of the input array is chosen.

After the partitioning,
• The pivot will be in the correct place in the array

• All items to pivot’s left are less than or equal to it
• All items to pivot’s right are greater than or equal to it
After the partitioning step, partitioning is applied recursively to the left

and right ‘half’ of the array.

QuickSort
qs(int *a, int l, int r)
{
int v, i, j, t;
if(r>l)
{
v=a[r]; i=l-1; j=r; // set v (pivot) to rightmost element
for(;;)
{
while(a[++i]<v); // move left pointer
while(a[--j]>v); // move right pointer
if(i>=j)break; // stop if pointers meet or cross
t=a[i]; a[i]=a[j]; a[j]=t; // swap elements
}
t=a[i]; a[i]=a[r]; a[r]=t; // swap elements
qs(a, l, i-1); // recursive call on left half
qs(a, i+1, r); // recursive call on right half
}
from R. Sedgewick “Algorithms in C”. Comments added.

QuickSort - Random Input
The above code has been modified to print out the sub-array it is
sorting and the pivot value. Every swap operation also causes ‘swap’
to be printed. To fully understand quicksort, try this yourself!
0 7 1 8 2 9 3 10 4 11
piv=11
swap
0 7 1 8 2 9 3 10 4 11
piv=4
swap swap swap
0 3 1 2 4 9 7 10 8 __
piv=2
swap swap
0 1 2 3 __ __ __ __ __ __
piv=1
swap
0 1 __ __ __ __ __ __ __ __

piv=8
swap swap
__ __ __ __ __ 7 8 10 9 __
piv=9
swap
__ __ __ __ __ __ __ 9 10 __
0 1 2 3 4 7 8 9 10 11

QuickSort - Sorted Input
1 2 3 4 5 6 7 8 9 10
piv=10
1 2 3 4 5 6 7 8 9 __
piv=9
1 2 3 4 5 6 7 8 __ __
piv=8
1 2 3 4 5 6 7 __ __ __
piv=7
1 2 3 4 5 6 __ __ __ __
piv=6
1 2 3 4 5 __ __ __ __ __
piv=5
1 2 3 4 __ __ __ __ __ __
piv=4
1 2 3 __ __ __ __ __ __ __
piv=3
1 2 __ __ __ __ __ __ __ __
piv=2

1 __ __ __ __ __ __ __ __ __
1 2 3 4 5 6 7 8 9 10

O(n) sorting
O(n) sorting is only possible on special inputs.
You met bucket sort (or counting sort) in the lab on complexity.

Bucket Sort
Principle: Given keys are in some finite range, j to j + k , the sort

proceeds as follows:
• initialize a bucket array b[0] = 0 to b[k − 1] = 0
• increment b[v − j] for each key v in the input
• write out i + j , b[j] times, for each value i in 0 to k

The algorithm, as stated, is in O(n + k). Why?
Buckets need not be of size 1. If larger buckets are used (as above),
then an extra sorting procedure can be used to sort the contents of
each bucket.
Bucket sort can be implemented to be stable.

Radix Sort
(See pp. 242–243 of Goodrich & Tamassia)
Principle: To sort keys consisting of a sequence of symbols (e.g.

words, n-digit numbers), we can apply a bucket sort to each symbol in
turn, i.e. do multiple passes of the bucket sort.
Time Complexity: Radix sort is an O(n) sorting algorithm, if the

number of symbols in each sorting key is considered to be a
constant. Why?
Advantages: This extends the range of applications for which a

bucket sort is suitable. It would be impractical to sort (dictionary)
words by a bucket sort because it would be difficult to index words
into buckets. Using radix sort, we just need 26 buckets (1 per letter).

To sort a sequence of integers of arbitrary length, first we left-fill each
integer with zeros so that all of them have the same length. E.g.
1, 100, 33 becomes 001, 100, 033
Then we use bucket sort to sort by the least significant digit.

100, 001, 033
Next, we sort by the next more significant digit.

100, 001, 033
And finally by the most significant digit.

001, 033, 100.
The bucket sort must be stable. Why?

Lecture Sorting

Uploaded by

Copyright:

Available Formats

Lecture Sorting

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Sorting

Uploaded by

Copyright:

Available Formats

Sorting Algorithms

COMP26120 - Week 9 Roscoe Th. B, Nov 26 2010

• Fundamental to organizing data

Every algorithms book has a large section on Sorting...

• quicksort() is often an adequate answer in many applications

Joshua Knowles 3 Roscoe Th. B, Nov 26 2010

• Definition of sorting. Correctness of sorting algorithms

• How the following work: Bubble sort, Insertion sort, Selection

• Main properties of those algorithms

• How to reason about complexity — worst case and special cases

Joshua Knowles 4 Roscoe Th. B, Nov 26 2010

Selection sort: 97 (very short description only)

Some of the exercise on pp. 251–252. (More advice to follow)

Joshua Knowles 5 Roscoe Th. B, Nov 26 2010

Joshua Knowles 6 Roscoe Th. B, Nov 26 2010

More precisely, since Bubblesort is O(n2) and Quicksort is

Remember, for n = 106, n2 is 1012, but n log n is only about 2.107.

Note also: Bubblesort is poor in practice even compared with other

Joshua Knowles 7 Roscoe Th. B, Nov 26 2010

Joshua Knowles 8 Roscoe Th. B, Nov 26 2010

1. sortedArray is a permutation of array and

Joshua Knowles 9 Roscoe Th. B, Nov 26 2010

Joshua Knowles 10 Roscoe Th. B, Nov 26 2010

Be careful! Sometimes, the items in an array or list are not

In the game, Rock - Paper - Scissors we have

[DEMO - See demo3.c]

Joshua Knowles 11 Roscoe Th. B, Nov 26 2010

Joshua Knowles 12 Roscoe Th. B, Nov 26 2010

Why does insertion sort have time complexity in O(n2 ) ?

Insertion sort is in O(n) for already sorted input. Why ?

Joshua Knowles 13 Roscoe Th. B, Nov 26 2010

Selection sort uses the same number of comparisons, independently

If an algorithm can sort any list of items belonging to a total ordering,

If, on the other hand, an algorithm requires the items to be restricted

Comparison-based sorts are general purpose.

Joshua Knowles 16 Roscoe Th. B, Nov 26 2010

The basic operations of sorting algorithms are comparison, and

When the complexity of an algorithm is calculated, usually only one of

For some applications, comparison is expensive. For others, copying

If records are large, however, it can be better to apply a sorting

Often the complexity on worst-case inputs is given.

But some algorithms have poor worst-case performance, yet have

Quicksort has poor peformance O(n2) on sorted or reverse-sorted

Insertion sort is linear O(n) on ‘nearly’ sorted or sorted lists.

Joshua Knowles 18 Roscoe Th. B, Nov 26 2010

Small input size

Many values the same (duplicates)

Joshua Knowles 19 Roscoe Th. B, Nov 26 2010

If the speed of sorting on a particular platform/input data is very

Joshua Knowles 20 Roscoe Th. B, Nov 26 2010

Joshua Knowles 21 Roscoe Th. B, Nov 26 2010

E.g. Mergesort has worst-case time complexity of O(n log n). In

Lower Bound What is the minimum amount of work needed to sort

No algorithm based on comparison of keys can sort a worst-case

Joshua Knowles 23 Roscoe Th. B, Nov 26 2010

Mergesort, Heapsort are asymptotically optimal sorting algorithms.

Joshua Knowles 25 Roscoe Th. B, Nov 26 2010

Observation: given two sorted lists of total length n, it is possible to

• MergeSort has worst-case time complexity of O(n log n) in the

• There is no implementation as an in-place sort.