XOR Basis Algorithm

The XOR operation between two numbers P and Q can be looked in a different way as the bitwise sum modulo 2 on the bits of P and Q. Consider P = 10010 and Q = 00110.

XOR of P and Q

Starting from leftmost, we verify the above statement :

•  (1 + 0) % 2 = 1
•  (0 + 0) % 2 = 0
•  (0 + 1) % 2 = 1
•  (1 + 1) % 2 = 0
•  (0 + 0) % 2 = 0

A binary number is in the form of a ℤ₂^d vector where d is the number of bits in the binary number and 2 represents the allowed integer values in the vector viz., {0, 1}. So, the result of a XOR operation between two numbers P and Q is the vector addition mod 2 of two ℤ₂^d vectors P and Q.

P + Q = P ⊕ Q

For subtracting Q, XOR with Q on RHS to obtain P on RHS,

P = P ⊕ Q ⊕ Q

P = P

Now, given an array A of N non-negative integers, we need to find the Basis B for the elements of the array when represented as  ℤ₂^d vectors in form of a bitmask.

Some points to note –

• The size of the basis B for a d-dimensional Vector Space can not exceed d.
• The basis vectors are independent, i.e., none of them can be expressed as a linear combination of a subset of basis vectors (other than the vector itself).

The algorithm goes as follows:

• Assume we have the basis for all the vectors till index ‘i‘ (i < N) and we need to check if A[i + 1] can be represented as a linear combination of current basis vectors.
• If not so, then add A[i + 1] to our basis, otherwise increment the index. We can effectively check this if we have all our basis vectors differ by the first set bit index (from left), let’s denote it by msb(B[j]).
• Start checking from the left bits, if index ‘s’ is set in current value of A[i + 1] and there is no basis vector with msb(B[j]) = s, then no linear combination of the existing basis vectors can represent current value of A[i + 1]. 
  • So, insert current value of A[i + 1] in the basis. 
  • Otherwise, subtract B[j] with msb = s from A[i + 1] by XORing with B[j] and continue with other bits. 
  • If at end A[i + 1] is a null vector, then it can be represented as linear combination of the current basis vectors, otherwise not and has to be inserted in the basis.

Examples:

Input: N = 5, A = {2, 5, 11, 9, 20}
Output: {5, 2, 12, 24}
Explanation: Given input = 2(00010), 5(00101), 11(01011), 9(01001), 20(10100) 
and basis = {0, 0, 0, 0, 0}, considering d = 5(for simplicity)
For i = 0, A[i] = 2(00010), first set bit is 1^st bit,  basis[1] = 0,  
So basis becomes {0, 2, 0, 0, 0}
For i = 1, A[i] = 5(00101), first set bit is 0^th bit,  basis[0] = 0,  
So basis becomes {5, 2, 0, 0, 0}
For i = 2, A[i] = 11(01011), first set bit is 0^th bit, basis[0] = 5,  
So A[i] = 11 ^ 5 = 14(01110)
Now A[i] = 14(01110), first set bit is 1^st bit, basis[1] = 2,  
So A[i] = 14 ^ 2 = 12(01100)
Now A[i] = 12(01100), first set bit is 2^nd, basis[2] = 0,  
So basis becomes {5, 2, 12, 0, 0}
For i = 3, A[i] = 9(01001), first set bit is 0^th bit, basis[0] = 5,  
So A[i] = 9 ^ 5 = 12(01100)
Now A[i] = 12(01100), first set bit is 2^nd bit, basis[2] = 12,  
So A[i] = 12 ^ 12 = 0(00000)
For i = 4, A[i] = 20(10100), first set bit is 2^nd bit, basis[2] = 12,  
So A[i] = 20 ^ 12 = 24(11000)
Now A[i] = 24(11000), first set bit is 3^rd, basis[3] = 0,  
So basis becomes {5, 2, 12, 24, 0}
So XOR Basis is {5, 2, 12, 24}

Input: N = 7, A = {5, 16, 7, 18, 34, 24, 9}
Output: {5, 2, 12, 24, 16, 32}

Below is the implementation of the above approach

C++

// C++ code to implement above algorithm

#include <bits/stdc++.h>
using namespace std;

// change d based on dimension of your vector
#define d 20

// Basis vector corresponding to each first 1-bit
vector<int> Basis(d, 0);

// Function to get the basis vectors
void add(int mask)
{
    for (int i = 0; i < d; i++) {
        if (mask & (1 << i)) {

            // If the current value of mask
            // has first set bit at i
            if (Basis[i] == 0) {

                // If no basis vector exists
                // with first set bit at i
                Basis[i] = mask;
                return;
            }

            // Subtracting Basis[i] from the mask
            mask ^= Basis[i];
        }
    }
    return;
}

// Driver code
int main()
{
    // Number of elements in the array
    int N = 5;

    // Vector containing those N elements
    vector<int> A{ 2, 5, 11, 9, 20 };

    for (int i = 0; i < N; i++)

        // Finding a Basis for
        // the first i + 1 elements
        add(A[i]);

    cout << "The basis vectors are: \n";
    for (int i = 0; i < d; i++) {
        if (Basis[i])

            // Checking for each msb, a
            // non-zero basis vector
            cout << Basis[i] << endl;
    }
    return 0;
}

Java

// Java code to implement the above approach
import java.util.*;

class GFG {

  // change d based on dimension of your vector
  static int d = 20;

  // Basis vector corresponding to each first 1-bit
  static int[] Basis = new int[d];

  // Function to get the basis vectors
  static void add(int mask)
  {
    for (int i = 0; i < d; i++) {
      if ((mask & (1 << i)) != 0) {

        // If the current value of mask
        // has first set bit at i
        if (Basis[i] == 0) {

          // If no basis vector exists
          // with first set bit at i
          Basis[i] = mask;
          return;
        }

        // Subtracting Basis[i] from the mask
        mask ^= Basis[i];
      }
    }
    return;
  }

  // Driver code
  public static void main(String[] args)
  {
    // Number of elements in the array
    int N = 5;

    // Vector containing those N elements
    int[] A = { 2, 5, 11, 9, 20 };
    for (int i = 0; i < d; i++) {
      Basis[i] = 0;
    }

    for (int i = 0; i < N; i++)

      // Finding a Basis for
      // the first i + 1 elements
      add(A[i]);

    System.out.print("The basis vectors are: \n");
    for (int i = 0; i < d; i++) {
      if (Basis[i] != 0)

        // Checking for each msb, a
        // non-zero basis vector
        System.out.println(Basis[i] );
    }
  }
}

// This code is contributed by phasing17

Python3

# Function to check if it is possible to

# change d based on dimension of your vector
d = 20

# Basis vector corresponding to each first 1-bit
Basis = [0] * d

# Function to get the basis vectors
def add(mask) :

    for i in range(d):
        if (mask & (1 << i)) :

            # If the current value of mask
            # has first set bit at i
            if (Basis[i] == 0) :

                # If no basis vector exists
                # with first set bit at i
                Basis[i] = mask
                return


            # Subtracting Basis[i] from the mask
            mask ^= Basis[i]

    return

# Driver code
if __name__ == "__main__":

    # Number of elements in the array
    N = 5

    # Vector containing those N elements
    A = [ 2, 5, 11, 9, 20 ]

    for i in range(N):

        # Finding a Basis for
        # the first i + 1 elements
        add(A[i])

    print("The basis vectors are: ")
    for i in range(d):
        if (Basis[i]) :

            # Checking for each msb, a
            # non-zero basis vector
             print(Basis[i])

              # This code is contributed by code_hunt.

C#

// C# code to implement the above approach
using System;

class GFG {

// change d based on dimension of your vector
static int d = 20;

// Basis vector corresponding to each first 1-bit
static int[] Basis = new int[d];


// Function to get the basis vectors
static void add(int mask)
{
    for (int i = 0; i < d; i++) {
        if ((mask & (1 << i)) != 0) {

            // If the current value of mask
            // has first set bit at i
            if (Basis[i] == 0) {

                // If no basis vector exists
                // with first set bit at i
                Basis[i] = mask;
                return;
            }

            // Subtracting Basis[i] from the mask
            mask ^= Basis[i];
        }
    }
    return;
}

// Driver code
public static void Main()
{
    // Number of elements in the array
    int N = 5;

    // Vector containing those N elements
    int[] A = { 2, 5, 11, 9, 20 };
    for (int i = 0; i < d; i++) {
    Basis[i] = 0;
    }

    for (int i = 0; i < N; i++)

        // Finding a Basis for
        // the first i + 1 elements
        add(A[i]);

    Console.Write("The basis vectors are: \n");
    for (int i = 0; i < d; i++) {
        if (Basis[i] != 0)

            // Checking for each msb, a
            // non-zero basis vector
            Console.WriteLine(Basis[i] );
    }
}
}

// This code is contributed by sanjoy_62.

Javascript

// JavaScript code to implement above algorithm

// change d based on dimension of your vector
const d = 20;

// Basis array corresponding to each first 1-bit
let Basis = Array(d).fill(0);

// Function to get the basis arrays
function add(mask)
{
  for (let i = 0; i < d; i++)
  {
    if (mask & (1 << i))
    {

      // If the current value of mask
      // has first set bit at i
      if (Basis[i] == 0)
      {

        // If no basis vector exists
        // with first set bit at i
        Basis[i] = mask;
        return;
      }

      // Subtracting Basis[i] from the mask
      mask ^= Basis[i];
    }
  }
  return;
}

// Driver code

// Number of elements in the array
let N = 5;

// Vector containing those N elements
let A = [2, 5, 11, 9, 20];

for (let i = 0; i < N; i++)

  // Finding a Basis for
  // the first i + 1 elements
  add(A[i]);

console.log("The basis vectors are: ");
for (let i = 0; i < d; i++) {
  if (Basis[i])

    // Checking for each msb, a
    // non-zero basis vector
    console.log(Basis[i]);
}

// This code is contributed by ishankhandelwals.

Output

The basis vectors are: 
5
2
12
24

Time Complexity: O(N * d)
Auxiliary Space: O(d)