Binary search algorithms are everywhere, and for good reason. They embody a simple yet powerful idea. In general, algorithms are named with purpose. Let’s take a closer look at this algorithm through an example problem.
Problem
We are given an ascending sequence of length N
and Q
requests.
Each request consists of a number, and the answer indicates whether that number exists in the sequence.
1 <= N, Q <= 10^5, abs(a[i]) <= 10^9
.
Analysis
For each request, check whether the number is equal to any number in the sequence. In the worst case, we will perform N
operations for each request.
This means that in total, we will perform Q*N
or 10^10
operations. We can’t wait for it to finish working, right!
So, how can we solve this quickly? Note that the sequence is in ascending order.
Suppose we want to determine whether the number X is present in a given sequence.
First, we perform a single operation to access the value of the middle element of the sequence. If the index of the middle element is mid
, it can be calculated as mid = (n + 1) / 2
.
Here are some things to watch out for:
a[mid] == X
, if the middle element is equal to X, we know that we have found a match.a[mid] > X
, or If the middle element is greater than X, then every element with an index from[mid, mid+1,..., n]
is clearly greater than the number we are looking for. Then, there is no need to check the number X with all of these. If the number X is in this sequence, its index clearly in the interval[1, 2, ..., mid]
.a[mid] < X
, then, all elements with indices[1, 2, ..., mid]
are clearly less than the number we are looking for. Then, we don’t need to check the number X with all of these. If the number X is in this sequence, its index is clearly in the interval[mid+1, mid+2, ..., n]
.
If conditions 2 and 3 hold, we still cannot determine whether X is in the sequence.
However, if we apply the same idea again, we can search in a smaller interval.
In case 2, we search in the interval [1, mid-1]
, and in case 3, we search in the interval [mid+1, N]
.
By dividing the sequence into two parts, we can perform a worst-case of log(N)
operations, resulting in a total of Q*log(N)
operations.
#include <bits/stdc++.h>
using namespace std;
int n, q, a[200000];
bool BS( int X ) {
int L = 1, R = n;
while( L <= R ) {
int mid = (L+R)/2;
if( a[mid] == X ) {
return true;
}
if( a[mid] > X ) {
R = mid-1;
} else {
L = mid+1;
}
}
return false;
}
int main() {
cin >> n >> q;
for(int i = 1; i <= n; i++) {
cin >> a[i];
}
while( q-- ) {
int x; cin >> x;
if( BS( x ) ) {
cout << "YES" << endl;
} else {
cout << "NO" << endl;
}
}
return 0;
}