Skip to main content

4. Median of Two Sorted Arrays

hardAsked at Juniper Networks

Find the median of two sorted arrays in O(log(m+n)) time using binary search on partition. Juniper asks this to probe candidates on binary search in non-obvious settings — the O(log n) constraint rules out the easy O(m+n) merge, requiring rigorous partitioning logic that appears in distributed query execution and sorted log analysis.

By Alex Chen, Founder, InterviewChamp.AI · Last verified

Source citations

Public interview reports confirming this problem appears in Juniper Networks loops.

  • Glassdoor (2025-Q4)Cited in Juniper senior SWE onsite reports as a binary search hard problem testing algorithmic rigor.
  • Blind (2025-10)Juniper threads list Median of Two Sorted Arrays as a hard problem asked in senior and staff engineering loops.

Problem

Given two sorted arrays nums1 and nums2 of size m and n respectively, return the median of the two sorted arrays. The overall run time complexity should be O(log(m+n)).

Constraints

  • nums1.length == m
  • nums2.length == n
  • 0 <= m <= 1000
  • 0 <= n <= 1000
  • 1 <= m + n <= 2000
  • −10^6 <= nums1[i], nums2[i] <= 10^6

Examples

Example 1

Input
nums1 = [1,3], nums2 = [2]
Output
2.0

Explanation: Merged: [1,2,3]. Median is 2.

Example 2

Input
nums1 = [1,2], nums2 = [3,4]
Output
2.5

Explanation: Merged: [1,2,3,4]. Median is (2+3)/2 = 2.5.

Approaches

1. Merge and find median — O(m+n)

Merge both arrays into a sorted array and return the middle element(s). Simple but does not meet the O(log(m+n)) requirement.

Time
O(m + n)
Space
O(m + n)
function findMedianSortedArrays(nums1, nums2) {
  const merged = [];
  let i = 0, j = 0;
  while (i < nums1.length && j < nums2.length) {
    if (nums1[i] <= nums2[j]) merged.push(nums1[i++]);
    else merged.push(nums2[j++]);
  }
  while (i < nums1.length) merged.push(nums1[i++]);
  while (j < nums2.length) merged.push(nums2[j++]);
  const mid = Math.floor(merged.length / 2);
  return merged.length % 2 === 1 ? merged[mid] : (merged[mid - 1] + merged[mid]) / 2;
}

Tradeoff: O(m+n) time and space. Easy to understand and verify. Mention this first to establish correctness, then optimize.

2. Binary search on partition — O(log(min(m,n)))

Binary search on the partition index in the smaller array. For a valid partition, the left half of both arrays combined must have (m+n)/2 elements, and max(left sides) <= min(right sides).

Time
O(log(min(m, n)))
Space
O(1)
function findMedianSortedArrays(nums1, nums2) {
  // Always binary search on the smaller array
  if (nums1.length > nums2.length) return findMedianSortedArrays(nums2, nums1);
  const m = nums1.length, n = nums2.length;
  let lo = 0, hi = m;
  while (lo <= hi) {
    const partX = Math.floor((lo + hi) / 2); // partition in nums1
    const partY = Math.floor((m + n + 1) / 2) - partX; // partition in nums2
    const maxLeftX = partX === 0 ? -Infinity : nums1[partX - 1];
    const minRightX = partX === m ? Infinity : nums1[partX];
    const maxLeftY = partY === 0 ? -Infinity : nums2[partY - 1];
    const minRightY = partY === n ? Infinity : nums2[partY];
    if (maxLeftX <= minRightY && maxLeftY <= minRightX) {
      // Correct partition
      if ((m + n) % 2 === 1) return Math.max(maxLeftX, maxLeftY);
      return (Math.max(maxLeftX, maxLeftY) + Math.min(minRightX, minRightY)) / 2;
    } else if (maxLeftX > minRightY) {
      hi = partX - 1; // too far right in nums1
    } else {
      lo = partX + 1; // too far left in nums1
    }
  }
}

Tradeoff: O(log(min(m,n))) time, O(1) space. The optimal solution meeting the O(log(m+n)) requirement. Complex to derive but the canonical answer for senior engineering roles.

Juniper Networks-specific tips

Acknowledge the two approaches upfront and their complexity difference. For senior Juniper roles, the binary search solution is expected — walk through the partition invariant carefully: 'I'm binary searching for a cut in nums1 such that the combined left halves are exactly half the total, and the maximum of the left sides is ≤ the minimum of the right sides.' Use Infinity and -Infinity as sentinels for edge partitions — it eliminates special cases cleanly.

Common mistakes

  • Using +Infinity for maxLeft and -Infinity for minRight — the sentinel values are flipped: an empty left side has max = -Infinity, an empty right side has min = +Infinity.
  • Not swapping to ensure binary search on the smaller array — the partition index in nums2 could go negative if m > n.
  • Off-by-one in the partY formula — use (m + n + 1) / 2 (integer divide) to handle both odd and even total lengths with the same formula.
  • Forgetting to handle the odd vs even total-length cases for the final median computation.

Follow-up questions

An interviewer at Juniper Networks may pivot to one of these next:

  • Kth Largest Element in Two Sorted Arrays — binary search for the kth element instead of the median.
  • How would you compute quantiles (p25, p50, p75) across two sorted telemetry streams efficiently?
  • What if there are k sorted arrays instead of two?

Solve it now

Free. No sign-up. Python and JavaScript run instantly in your browser.

Output

Press Run or Cmd+Enter to execute

FAQ

Why binary search on the smaller array?

partY = (m+n+1)/2 - partX. If m > n, partX can be large enough to make partY negative, which is invalid. Binary searching on the smaller array guarantees partY stays in [0, n].

Why use (m+n+1)/2 instead of (m+n)/2 in the partY formula?

The +1 ensures the left half is the same size or one larger for odd totals, giving the correct median without an extra conditional.

What are the sentinel values for?

-Infinity for an empty left side (no left elements exists, so the max is conceptually -∞) and +Infinity for an empty right side (no right element exists, so the min is conceptually +∞). This avoids special-casing boundary partitions.

Practice these live with InterviewChamp.AI

Drill Median of Two Sorted Arrays and other Juniper Networks interview questions under real-loop conditions with instant feedback on your reasoning, complexity claims, and code.

Practice these live with InterviewChamp.AI →