ِ‫ﺑِﺳْ مِ ٱ ﱠٰ ِ ٱﻟرﱠ ﺣْ َٰﻣ ِن ٱﻟرﱠ ﺣِﯾم‬

In the name of Allah, Most Gracious, Most Merciful

CSE 4303
Data Structures

Topic: Sparse Table, Fenwick Tree

Asaduzzaman Herok
Lecturer | CSE | IUT
[email protected]
Problem Scenario
Think of a university of 10 thousand ex-students. Their results and other informations are sorted by their
student IDs. Now it is very frequent that authority want to know from roll x to roll y what is the maximum
CGPA.

What to do?

● Linear Search?
● Binary Search?
● Sorting the data based on CGPA?

Is there a way to get the result without changing the relative order of the data?

These types of query / questions are called Range Query.

Example: Range Maximum Query, Range Minimum Query, Range Sum Query, Range GCD / LCM Query

Asaduzzaman Herok, Lecturer

17 November,, 2023 2
Department of Computer Science and Engineering, IUT
Range Sum Problem
Given array of n elements A = { a1, a2, … … }, in each query (l, r) it is asked to give the sum of
elements from al to ar. That is return {al+al+1+al+2+ … … + ar-2+ar-1+ar}

Range Minimum Problem

Given array of n elements A = { a1, a2, … … }, in each query (l, r) it is asked to give the minimum
element among al to ar. That is return minimum {al,al+1,al+2, … … , ar-2,ar-1,ar}

Generalised Way
Given array of n elements A = { a1, a2, … … }, in each query (l, r) it is asked to give the
F({al,al+1,al+2, … … , ar-2,ar-1,ar}).

Asaduzzaman Herok, Lecturer

17 November,, 2023 3
Department of Computer Science and Engineering, IUT
Naive Solutions
● Linear Search on every query starting from al to ar.
○ Complexity on each O(r-l+1) i.e O(n)
● Precalculation for once for every pair of (l,r).
○ Need O(n2) space and time.
○ But can answer query in constant time i.e O(1)

What to do then?
● Sparse Table
● Fenwick Tree
● Segment Tree
● Square Root Decomposition
● Square Root Tree
● Mo’s Algorithm

Our Discussion will be restricted upto Segment tree

Asaduzzaman Herok, Lecturer

17 November,, 2023 4
Department of Computer Science and Engineering, IUT
Sparse Table
A sparse table is a data structure that can answer some range query problems, such as range minimum
query problem, in O(1) time. Range sum in O(log n)

Pros:
● Building time complexity O(n log n).
● Space complexity O(n log n)
● Supported query function:
○ Element in the array supports the associative property i.e x o (y o z) = (x o y) o z.
[Here o is a operator]
○ F(a, b, c, d) = F(a,b) o F(c,d), from the non-overlapping segments. Time
complexity O(log n)
○ F(a, b, c, d) = F(a,b,c) o F(b,c,d), from the overlapping segments. Time
complexity O(1) i.e constant time.

Cons:
● Can’t work with update i.e data sequence must be immutable
● Update operation would cause O(n log n) time complexity to rebuild.

Asaduzzaman Herok, Lecturer

17 November,, 2023 5
Department of Computer Science and Engineering, IUT
Sparse Table
Intuition:
● Any non-negative number can be uniquely represented as a sum of decreasing powers of two.
○ Example: 13 = (1101)2 = 8 + 4 + 1
○ Number of powers of two needed is log2(n)
● So a range of m elements can also be represented as union of continuous sub-segments of size
decreasing powers of two
○ Example: [2,14] = [2,9] U [10,13] U [14, 14]
Size: 13 8 4 1
○ If we already know these smaller ranges value then the large range can be computed.

Idea:
● Pre-compute all the answers of range with length equal to some powers of 2
● While querying break the query segments into some segments of powers of 2 and use their value
to compute for the query segment.

Asaduzzaman Herok, Lecturer

17 November,, 2023 6
Department of Computer Science and Engineering, IUT
Sparse Table Pre-Computation

● Declare a 2D array sparse[n][k+1] where k = floor(log2(n)+1)

● sparse[i][j] keeps the answer for segment / range starting at i and ending at i + 2j - 1
● The recurrence can be written as:
○ sparse[i][j] = F(sparse[i][j-1], sparse[i + 2j-1][j-1])
○ Example: sparse[4][3] = min(sparse[4][2], sparse[4+4][2])
Range: [4, 11] = min([4, 7], [8, 11])
● Base case:
○ Sparse[i][0] = F(ai)
○ Example: sparse[4][0] = min(a4) = a4
Range: [4, 4] = a4

Asaduzzaman Herok, Lecturer

17 November,, 2023 7
Department of Computer Science and Engineering, IUT
Sparse Table Pre-Computation

Asaduzzaman Herok, Lecturer

17 November,, 2023 8
Department of Computer Science and Engineering, IUT
Sparse Table Query
Range Sum Query: (l, r) Non-Overlapping Case
● Compute the power k = floor(log2 ( r-l+1))
○ Example: Range [4, 9], k = floor ( log2 (9-4+1 = 6)) = 2
● While (l<=r) do ans += sparse[l][k], l += 2k, k = floor(log2 ( r-l+1))
○ Example: Range [4, 9]
ans += sparse[4][2] // [4, 7]
l = 4 + 22 = 8
k = log2(9-8+1 = 2) = 1
ans += sparse[8][1] // [8,9]
l = 8 + 21 = 8

Range Min Query: (l, r) Overlapping Case

● Compute the power k = floor(log2 ( r-l+1))
○ Example: Range [4, 9], k = floor ( log2 (9-4+1 = 6)) = 2
● ans = min(sparse[l][k], sparse[R-2k+1][k]
○ Example: Range [4, 9]
ans = min( sparse[4][2], sparse[9-22+1 = 6][2] )
[4, 7] [6, 9]

Asaduzzaman Herok, Lecturer

17 November,, 2023 9
Department of Computer Science and Engineering, IUT
Fenwick Tree

Fenwick tree is a data structure which:

● Calculates the value of function F in the given range [l, r] (i.e. F({al,al+1, … … , ar-1, ar}) in O(l
n) time;
● Updates the value of an element of A in O(log2 n) time;
● Requires O(n) memory, or in other words, exactly the same memory required for A
● Build time is also O(n log2 n)
● Also called Binary Indexed Tree, or just BIT abbreviated.

The most common application of Fenwick tree is calculating the sum of a range.
F({al,al+1, … … , ar-1, ar}) = al + al+1 + … … + ar-1 + ar

F should support both F(a, b, c, d) = F(a,b) o F(c,d) and F(c,d) = F(a, b, c, d) o F(a, b),
o denotes some kind of operators.
Example:Sum(a, b, c, d)=Sum(a, b)+Sum(c, d) and Sum(c, d)=Sum(a, b, c, d)-Sum(a, b)
Xor(a, b, c, d)=Xor(a, b) xor Xor(c, d) and Xor(c, d)=Xor(a, b, c, d) xor Xor(a, b)

Asaduzzaman Herok, Lecturer

17 November,, 2023 10
Department of Computer Science and Engineering, IUT
Fenwick Tree
Basic Idea:
● Each integer can be represented as a sum of powers of two.
13 = (1101)2=23 + 22 + 20 = 8+4+1
● The answer of segment[1, 13] can be found by combining the range
[1,8] U [9,12] U [13,13]
● That is the same way, a cumulative values can be represented as a sum of
sets of sub segments value .
● A 1D array of size n is enough to store these sub-segment values.
BIT[1, 2, … , n ]
● Let’s i be an index of BIT and k be the position last non zero bit of the
binary representation of i.
○ Example: i = 13 = (1101)2 , k = 0
i = 6 = (110)2 , k = 1
i = 8 = (1000)2 , k = 3
● BIT[i] will keep the segment value of segment [i-2k+1, i]
○ Example: BIT[13] = [13, 13]
BIT[6] = [5, 6]
BIT[8] = [1, 8]
● So to get the segment value of [1, 13] we have to union
BIT[8] U BIT[10] U BIT[13]
BIT[(1000)2] U BIT[(1100)2] U BIT[(1101)2]

Asaduzzaman Herok, Lecturer

17 November,, 2023 11
Department of Computer Science and Engineering, IUT
Fenwick Tree
Basic Idea:
● We need to find a way to jump from these segments to
segments. Like from 13 to 12 to 8 by setting the least significant
bit to 0.

Isolating the last BIT:

● Let i be a1b as a representation of i in binary where a is the
prefix from the last significant bit and b is all zeros.
○ Example: 20 = (10100)2, a = 10, b = 00
a1b
-
● (a1b) is the inversion of bits.
○ (20)-=(10100)2- = (01011)2
● -num = (a1b)¯ + 1 = a¯0b¯ + 1 = a¯0(0…0)¯ + 1
= a¯0(1…1) + 1 = a¯1(0…0) = a¯1b.
● If we do bitwise and of a1b with a¯1b we get
(0..010..0)2
Example: 20 & -20 = 4
(10100)2 & (01100)2 = (100)2

Asaduzzaman Herok, Lecturer

17 November,, 2023 12
Department of Computer Science and Engineering, IUT
Fenwick Tree Read Operation on Segment Sum

int read(int idx) {

int sum = 0;
while (idx > 0) {
sum += tree[idx];
idx -= (idx & -idx);
}
return sum;
}

Asaduzzaman Herok, Lecturer

17 November,, 2023 13
Department of Computer Science and Engineering, IUT
Fenwick Tree Update Operation on Segment Sum
● The update operation is to increase or decrease the certain value at index
i of the given array A.
● Increase or decrease should be reflected in the affected range sums in
the BIT.
● To find the affected segments we have to add the last bit of i to itself;
and, repeat while i is less than or equal to n

void update(int idx, int val)

{
while (idx <= MaxIdx) {
tree[idx] += val;
idx += (idx & -idx);
}
}
Asaduzzaman Herok, Lecturer
17 November,, 2023 14
Department of Computer Science and Engineering, IUT
Fenwick Tree Range Query (l,r)
Recall the property:
F(c,d) = F(a, b, c, d) o F(a, b)

● Read(r) gives the segment value of [1, r]

Read(l-1) gives the segment value of [1, l-1]
● Query(l,r) = Read(r) o Read(l-1)
[1, r] o [1, l-1]

Range / Segment Sum Query(l,r)

● Read(r) gives the segment value of [1, r]

Read(l-1) gives the segment value of [1, l-1]
● Query(l,r) = Read(r) - Read(l-1)
[1, r] - [1, l-1]

Asaduzzaman Herok, Lecturer

17 November,, 2023 15
Department of Computer Science and Engineering, IUT
 Acknowledgements

Sparse Table - Algorithms for Competitive Programming (cp-algorithms.com)

Fenwick Tree - Algorithms for Competitive Programming (cp-algorithms.com)

Binary Indexed Trees (topcoder.com)

Asaduzzaman Herok, Lecturer

17 November, 2023 16
Department of Computer Science and Engineering, IUT