Floor and Ceil in Sorted Array¶

Question¶

Given a sorted array nums and an integer x, find the floor and ceiling of x. - Floor: The largest element in the array \(\le x\). - Ceiling: The smallest element in the array \(\ge x\). If no floor or ceiling exists, return -1 for that respective value.

Solution¶

Pattern¶

Single-Pass Boundary Search Modify a standard Binary Search to search for the first element strictly greater than \(x\). Because the array is sorted, this target index naturally gives us the Ceiling, and the index immediately preceding it naturally gives us the Floor.

How to Identify¶

The array is sorted.
You need to find the closest values to a target, bounding it from above and below.
Searching for two related boundaries simultaneously often implies they can be found using the exact same binary search state.

Description¶

Step-by-step explanation:

Handle edge cases. If the array is empty, return [-1, -1].
Initialize left = 0 and right = nums.length. We set right out of bounds to handle the case where \(x\) is larger than all elements in the array.
Loop while left < right.
Calculate mid = left + (right - left) / 2.
Optimization: If nums[mid] == x, then \(x\) exists in the array. By definition, the largest element \(\le x\) is \(x\), and the smallest element \(\ge x\) is \(x\). Return [x, x] immediately.
Predicate Evaluation: - If nums[mid] > x: The element is strictly greater. It is a candidate for the ceiling. Keep it in the search space and look left: right = mid.
If nums[mid] < x: The element is strictly less. It cannot be the ceiling. Discard it and look right: left = mid + 1.
Once the loop ends, left (and right) will point to the index of the first element strictly greater than \(x\).
Map to Results:
The Ceiling is at index right. (Ensure right < nums.length to avoid out-of-bounds, returning -1 if it is).
The Floor is at index right - 1. (Ensure right - 1 >= 0 to avoid out-of-bounds, returning -1 if it is).

The Intuition¶

Imagine finding your assigned seat in a sorted theater row. Your ticket says seat 25, but the seats jump from 20 to 30. If you walk down the row, the moment you pass a seat numbered greater than 25 (seat 30), you stop. The seat you are currently looking at (30) is your Ceiling. The seat you just walked past (20) is your Floor. You don't need to walk down the row twice to find both numbers; finding the exact gap between them reveals both boundaries simultaneously. Binary search just helps you find that gap in \(O(\log N)\) time instead of walking linearly.

Complexity¶

Label	Worst	Average
Time Complexity	\(O(\log N)\)	\(O(\log N)\)
Space Complexity	\(O(1)\)	\(O(1)\)

Time Complexity¶

The search space is halved exactly once per loop iteration. The maximum number of iterations is \(\log_2(N)\).

Space Complexity¶

We use a few primitive pointers and allocate a size-2 array to return the answer. Auxiliary space is strictly \(O(1)\).

Code¶

class Solution {
    public int[] getFloorAndCeil(int[] nums, int x) {
        if (nums == null || nums.length == 0) return new int[]{-1, -1};

        int left = 0, right = nums.length;

        while (left < right) {
            int mid = left + (right - left) / 2;

            if (nums[mid] == x) return new int[]{x, x};

            if (nums[mid] > x)  {
                right = mid;
            } else {
                left = mid + 1;
            }
        }

        int floor = (right - 1 >= 0) ? nums[right - 1] : -1;
        int ceil = (right < nums.length) ? nums[right] : -1;

        return new int[]{floor, ceil}; 
    }
}

Caveats¶

Two Separate Searches vs One: A naive (but correct) candidate might write two entirely separate binary search functions: one for getFloor and one for getCeil. While asymptotically equivalent \(O(2 \log N) \rightarrow O(\log N)\), writing it in a single pass demonstrates superior mastery of binary search state invariants.
left <= right Pitfall: If you use the left <= right template with right = mid - 1, the final states of left and right cross each other, making boundary extraction slightly more confusing (left becomes ceil, right becomes floor). The [left, right) template is much cleaner for bounding problems.

Concepts to Think About¶

Lower/Upper Bound C++ Equivalents: In C++, std::lower_bound returns an iterator to the first element \(\ge x\) (Ceiling). std::upper_bound returns an iterator to the first element \(> x\).
Database Indexing: This exact algorithm is how B-Tree database indexes rapidly locate ranges for BETWEEN SQL queries. It finds the floor of the lower bound and the ceiling of the upper bound to fetch the block.
The Neighborhood Rule: In sorted arrays, the elements satisfying nums[i]<x and nums[i]≥x are always adjacent. Finding one usually gives you the other for free.
Lower Bound as Ceil: Why is lowerBound the same as Ceil? Because both look for the first element that hasn't "failed" the ≥x condition.
Edge Cases: What happens when x is smaller than nums[0]? (Ceil is nums[0], Floor is -1). What if x is larger than nums[n-1]? (Ceil is -1, Floor is nums[n-1]).

Logical Follow-up¶

Question: What if the array is an infinite data stream (you don't know the length), and you need the floor/ceil of \(x\)?

Solution: You cannot use nums.length to initialize right. You must first find the search bounds. Initialize left = 0, right = 1. While nums[right] < x, expand the window exponentially: left = right and right = right * 2. Once nums[right] >= x, you have established a finite window [left, right]. You then run the exact same binary search within that window. Total time remains \(O(\log k)\) where \(k\) is the index of the ceiling.

Question: "Given a sorted array, find the Closest Element to a target \(x\). If two numbers are equally close, return the smaller one."

Solution:

Use the Lower Bound logic to find the ceiling.
Compare the absolute difference between \(x\) and the Ceiling (nums[idx]) and \(x\) and the Floor (nums[idx-1]).
Return the one with the smaller difference.

Question (Find K Closest Elements): "Given a sorted integer array arr, two integers k and x, return the k closest integers to x in the array. The result should also be sorted. If there is a tie, the smaller strategy is preferred."

Solution: A naive L4 solution would be to find the closest element and then expand outwards using two pointers for \(O(\log n + k)\). However, an L5 candidate might suggest a more elegant Binary Search on the Window.

Intuition: We are looking for the starting index left of a window of size k.
Search Space: The possible starting index left ranges from 0 to n - k.
Binary Search Criteria: We compare the distance of the elements at the edges of the window. For a mid index:
Is \(x\) closer to nums[mid] or nums[mid + k]?
If x - nums[mid] > nums[mid + k] - x, then mid is too far to the left, so left = mid + 1.
Else, right = mid.
Result: After \(O(\log(n-k))\) time, we have the start of the perfect window.