lower bound, upper bound in c++, visually explained

2024-04-02

One of the most common problems programmers have to solve is retrieving specific data — finding the needle in the haystack. To make this simpler, languages provide standard tools to perform this task: in particular, this post focuses on C++’s lower_bound and upper_bound.

Personally, I find that the documentation for these functions is quite unclear and verbose. For example, cppreference.com describes lower_bound like this:

Searches for the first element in the partitioned range [first, last) which is
**not** ordered before value.

That sounds like gibberish, and is too technical to quickly understand. For that reason, I’m making this blog post to explain my own mental model of these functions.

refresher on binary search

First, it’s important to understand how lower_bound and upper_bound work under the hood.

As you know, finding words in a dictionary is relatively fast. This is possible because the words are in alphabetical order. If the words weren’t ordered, you’d have to look through every single word in the dictionary, one by one. That would be an excruciating, and much slower process. Because of the ordering, you can rapidly narrow down the word you want.

Computers can do the same with ordered data: this is called binary search, and is what powers lower_bound and upper_bound. For example, say our dictionary is 1000 pages, and the computer wants to look for the word “rabbit”. These are the steps it takes:

Start at exactly page 500.
See the word “murmur”, so go forwards to page 750.
See the word “sunny”, so go backwards to page 625.
And so on.

This is called “binary search” because we halve the region we are looking in every time (we pick either the left half, or the right half.) For step 1, the computer is halving the range 1-1000. In step 2, 500-1000. Then for step 3, 500-750. This is like the way humans look at dictionaries, but more structured.

Anyways, this is not intended to be a full explanation of binary search: refer to Tom Scott’s video about it for more information.

lower bound and upper bound

Back to the real subject of this post: lower_bound and upper_bound in C++. What I used to understand of these functions is that they use binary search to find elements in a sorted container. However, I didn’t get what differentiated them. Again, if you read solely the documentation about these functions, it’s not easily comprehensible.

First of all, say we wish to search for the integer k (k for key) in a sorted vector (array) of integers v. We can find the lower and upper bounds with these function calls:

// (you could use auto here instead of the verbose type)
std::vector<int>::iterator lb = std::lower_bound(v.begin(), v.end(), k);
std::vector<int>::iterator ub = std::upper_bound(v.begin(), v.end(), k);

Based on the documentation, we know the first two arguments specify the region of v we’re looking in. Here, it’s the entire vector (from the beginning to the end). Also, put simply, the functions return by default:

lower_bound: the first element e where k <= e;
upper_bound: the first element e where k < e.

Note: Both functions return v.end() if no valid element is found. This iterator points just after the last element of v.

This is the technical definition; it doesn’t mean much by itself. However, with a concrete example with real numbers, it clicked in my mind. For example, let k = 3. Here is an example sorted array v, with upper and lower bounds marked:

    lower   upper
      ↓       ↓
1 2 2 3 3 3 3 4 5 6
      ───────
         ↑
 matching interval

The first 3 is the lower bound: it’s the first element bigger or equal to our key. The 4 is the upper bound, the first element strictly bigger than our key.

Here, when it’s laid out visually, it’s now clear what the lower and upper bounds mean: it’s the bounds of the interval that matches our search key. This is mostly useful if the array has duplicate elements.

Notice how the upper bound is one past the end of the interval, just like how v.end() is one past the last element of the vector. This is usually how C++ iterators work, and makes some tasks more convenient. Take this regular for loop:

for (int i = 0; i < 10; i++) { ... }

This loop will iterate over the numbers 0 to 9, excluding the upper bound 10. The same logic applies to C++ iterators. If we want to iterate over all elements of a vector, we’d use:

for (auto it = v.begin(); it != v.end(); it++) { ... }

Here, we use != instead of < for iterators, but it does practically the same thing. When the iterator goes past the end of the vector, it’ll hit v.end() (which is one past the last element), and as such the loop stops.

Note: Usually, you’d do for (auto number : v) to iterate over the entire array.

So, having the upper bound be right past the end of the interval makes this possible:

for (auto it = lb; it != ub; it++) {
    // *it is like pointer dereference:
    // it gets the number pointed to by the iterator
    std::cout << *it << std::endl;
}

Anyways, I’ll repeat it again: lower_bound and upper_bound represent the interval that matches what you’re looking for.

conclusion

So, that is my “visual” explanation how lower and upper bound works in C++. In hindsight, this seems obvious, but back when I was first told about these functions, I could not understand it because of the confusing descriptions. Having this intuition for concepts is pretty helpful for truly understanding them: you don’t want to be stuck memorizing things that don’t make sense.