#3013Hard~50 min

Divide an Array Into Subarrays With Minimum Cost II

Time O(nlogd) · Space O(d) · Official statement on LeetCode

cpppython

Solutions

// Time:  O(nlogd)
// Space: O(d)

// fastest and less verbose
// sliding window, heap
class Solution {
public:
    long long minimumCost(vector<int>& nums, int k, int dist) {
        static const int64_t INF = numeric_limits<int64_t>::max();

        const auto& get_top = [](auto& heap, auto& total, const int d) {
            while (-heap.top().second < d) {
                heap.pop();
                --total;
            }
            return heap.top();
        };

        priority_queue<pair<int, int>> max_heap;
        priority_queue<pair<int, int>, vector<pair<int, int>>, greater<pair<int, int>>> min_heap;
        int total1 = 0, total2 = 0;
        int64_t mn = INF, curr = 0;
        for (int i = 1; i < size(nums); ++i) {
            max_heap.emplace(nums[i], -i);
            curr += nums[i];
            if (i > k - 1) {
                const auto [x, idx] = get_top(max_heap, total1, i - (1 + dist)); max_heap.pop();
                curr -= x;
                min_heap.emplace(x, idx);
            }
            if (i > 1 + dist) {
                if (get_top(min_heap, total2, i - (1 + dist)) <= pair(nums[i - (1 + dist)], -(i - (1 + dist)))) {
                    lazy_delete(min_heap, total2, i - (1 + dist));
                } else {
                    lazy_delete(max_heap, total1, i - (1 + dist));                    
                    curr -= nums[i - (1 + dist)] - min_heap.top().first;
                    max_heap.emplace(min_heap.top()); min_heap.pop();
                }
            }
            if (i >= k - 1) {
                mn = min(mn, curr);
            }
        }
        return nums[0] + mn;
    }

private:
    template<typename T>
    void lazy_delete(T& heap, int& total, const int d) {
        ++total;
        if (total <= size(heap) - total) {
            return;
        }
        T new_heap;
        while (!empty(heap)) {
            const auto x = heap.top(); heap.pop();
            if (-x.second <= d) {
                --total;
                continue;
            }
            new_heap.emplace(x);
        }
        heap = move(new_heap);
    }
};

// Time:  O(nlogd)
// Space: O(d)
// faster but more verbose
// sliding window, heap, freq table
class Solution2 {
public:
    long long minimumCost(vector<int>& nums, int k, int dist) {
        static const int64_t INF = numeric_limits<int64_t>::max();

        const auto& get_top = [](auto& heap, auto& cnt, auto& total) {
            while (cnt.count(heap.top())) {
                const int x = heap.top(); heap.pop();
                if (--cnt[x] == 0) {
                    cnt.erase(x);
                }
                --total;
            }
            return heap.top();
        };

        priority_queue<int> max_heap;
        priority_queue<int, vector<int>, greater<int>> min_heap;
        unordered_map<int, int> cnt1, cnt2;
        int total1 = 0, total2 = 0;
        int64_t mn = INF, curr = 0;
        for (int i = 1; i < size(nums); ++i) {
            max_heap.emplace(nums[i]);
            curr += nums[i];
            if (size(max_heap) - total1 > k - 1) {
                const int x = get_top(max_heap, cnt1, total1); max_heap.pop();
                curr -= x;
                min_heap.emplace(x);
                
            }
            if ((size(max_heap) - total1) + (size(min_heap) - total2) > 1 + dist) {
                if (get_top(min_heap, cnt2, total2) <= nums[i - (1 + dist)]) {
                    lazy_delete(min_heap, cnt2, total2, nums[i - (1 + dist)]);
                } else {
                    lazy_delete(max_heap, cnt1, total1, nums[i - (1 + dist)]);
                    curr -= nums[i - (1 + dist)] - min_heap.top();
                    max_heap.emplace(min_heap.top()); min_heap.pop();
                }
            }
            if (size(max_heap) - total1 == k - 1) {
                mn = min(mn, curr);
            }
        }
        return nums[0] + mn;
    }

private:
    template<typename T>
    void lazy_delete(T& heap, auto& cnt, int& total, int x) {
        ++cnt[x];
        ++total;
        if (total <= size(heap) - total) {
            return;
        }
        T new_heap;
        while (!empty(heap)) {
            const auto x = heap.top(); heap.pop();
            if (cnt.count(x)) {
                if (--cnt[x] == 0) {
                    cnt.erase(x);
                }
                continue;
            }
            new_heap.emplace(x);
        }
        total = 0;
        heap = move(new_heap);
    }
};

// Time:  O(nlogd)
// Space: O(d)
// fast but verbose
// sliding window, bst
class Solution3 {
public:
    long long minimumCost(vector<int>& nums, int k, int dist) {
        static const int64_t INF = numeric_limits<int64_t>::max();

        multiset<int> bst1, bst2;
        int64_t mn = INF, curr = 0;
        for (int i = 1; i < size(nums); ++i) {
            bst1.emplace(nums[i]);
            curr += nums[i];
            if (size(bst1) > k - 1) {
                curr -= *rbegin(bst1);
                bst2.emplace(*rbegin(bst1));
                bst1.erase(prev(end(bst1)));
            }
            if (size(bst1) + size(bst2) > 1 + dist) {
                if (*cbegin(bst2) <= nums[i - (1 + dist)]) {
                    bst2.erase(bst2.find(nums[i - (1 + dist)]));
                } else {
                    bst1.erase(bst1.find(nums[i - (1 + dist)]));
                    curr -= nums[i - (1 + dist)] - *cbegin(bst2);
                    bst1.emplace(*cbegin(bst2));
                    bst2.erase(begin(bst2));
                }
            }
            if (size(bst1) == k - 1) {
                mn = min(mn, curr);
            }
        }
        return nums[0] + mn;
    }
};

// Time:  O(nlogd)
// Space: O(d)
#include <ext/pb_ds/assoc_container.hpp>
#include <ext/pb_ds/tree_policy.hpp>
using namespace __gnu_pbds;

// concise but slow
// sliding window, ordered set
class Solution4 {
public:
    long long minimumCost(vector<int>& nums, int k, int dist) {
        using ordered_set = tree<pair<int, int>, null_type, less<pair<int, int>>, rb_tree_tag, tree_order_statistics_node_update>;
        ordered_set os;
        for (int i = 1; i < 1 + (1 + dist); ++i) {
            os.insert({nums[i], i});
        }
        int64_t curr = 0;
        auto it = cbegin(os);
        for (int i = 0; i < k - 1; ++i, ++it) {
            curr += it->first;
        }
        int64_t mn = curr;
        for (int i = 1 + (1 + dist); i < size(nums); ++i) {
            os.insert({nums[i], i});
            curr += min(nums[i] - os.find_by_order(k - 1)->first, 0);
            curr -= min(nums[i - (1 + dist)] - os.find_by_order(k - 1)->first, 0);
            os.erase({nums[i - (1 + dist)], i - (1 + dist)});
            mn = min(mn, curr);
        }
        return nums[0] + mn;
    }
};

Beginner Explanation

What is Divide an Array Into Subarrays With Minimum Cost II?

Divide an Array Into Subarrays With Minimum Cost II (LeetCode #3013) is a Hard problem that primarily trains two pointers.

How to think about it

  1. Restate the goal in your own words before coding.
  2. Work a tiny example by hand so the invariant becomes obvious.
  3. Identify the pattern — this problem aligns with sliding window, heap, and sorted list.
  4. Only then translate the idea into code.

Why this problem matters

Hard problems force you to combine patterns and prove complexity carefully — interview gold. Official solution notes mention: Sliding Window, Heap, Freq Table, Ordered Set.

AlgoForge explanations are original teaching notes. Always open the official problem statement on LeetCode for constraints and examples.

Interview Walkthrough

Interview approach for Divide an Array Into Subarrays With Minimum Cost II

Opening (30–60 seconds)

  • Clarify inputs/outputs and edge cases (empty input, single element, duplicates, overflow).
  • State a brute force so the interviewer knows you can solve it naively.
  • Propose the optimal direction tied to sliding window, heap, and sorted list.

Core solution narrative

  1. Define the state you track (pointers, DP cell, set membership, stack top, etc.).
  2. Explain the transition when you process the next element.
  3. Call out time (O(nlogd)) and space (O(d)) before coding.
  4. Code cleanly; narrate variable names.

What interviewers listen for

  • Correctness on edge cases
  • Complexity honesty
  • Ability to discuss trade-offs (e.g., hash map space vs. sort + two pointers)

Follow-up questions they may ask

  • Can you solve it with less memory?
  • What if the input stream is infinite / doesn't fit in RAM?
  • How would tests look for adversarial inputs?

Optimized Approach

Optimized solution notes

The reference solutions on AlgoForge target O(nlogd) time and O(d) space.

Pattern focus: sliding window, heap, and sorted list

Use the pattern as a checklist:

  • sliding window — confirm the invariant holds after each step
  • heap — confirm the invariant holds after each step
  • sorted list — confirm the invariant holds after each step

Multiple methods appear in the source solutions — compare them and explain when each is preferable.

Implementation tips

  • Prefer readable names over micro-optimizations in interviews.
  • Extract helpers only when they clarify (e.g., expand-around-center, DFS visit).
  • After AC-level logic, re-scan for off-by-one and null checks.

Complexity Analysis

Complexity

Measure Bound
Time O(nlogd)
Space O(d)

How to justify this in an interview

  • Time: count loops, map/set operations, and recursive branching; state average vs worst case if relevant.
  • Space: include hash maps, recursion stack, and output allocation when the problem asks for it.

If your implementation differs from the reference, re-derive big-O from your code — never memorize a complexity you cannot defend.

Common Mistakes

Common mistakes on Divide an Array Into Subarrays With Minimum Cost II

  1. Skipping edge cases — empty collections, single-element inputs, max constraints.
  2. Wrong invariant for sliding window, heap, and sorted list — updating state too early or too late.
  3. Mutating input unexpectedly when the problem forbids it.
  4. Off-by-one in windows, ranges, or binary search bounds.
  5. Ignoring overflow / precision for integer arithmetic problems.
  6. Overengineering — jumping to an advanced structure when a simpler approach works.

Alternative Approaches

Alternatives

The source file includes more than one method. Compare:

  1. Primary optimized path — best complexity for typical interviews.
  2. Secondary approach — often brute force, sorting-based, or space-optimized variant.

Practice articulating when you would pick each (constraints, readability, follow-ups).

Edge Cases

Edge cases checklist

  • Minimum input size
  • Maximum input size / time limits
  • Duplicates and already-sorted input
  • Negative numbers / zeros (if applicable)
  • Disconnected structures (graphs/trees)
  • Single path vs branching recursion depth

Pattern Recognition

Spotting this pattern

Signal phrases that point to sliding window, heap, and sorted list:

  • Sorted input or ability to sort without changing the answer class
  • Need for contiguous subarray / substring → consider sliding window
  • Need for O(1) membership → hash set/map
  • Optimal substructure + overlapping subproblems → DP
  • Connectivity / components → graph DFS/BFS or Union-Find

Primary topics: two pointers.

Follow-up Interview Questions

Follow-ups

  1. How does the solution change if the input is a stream?
  2. Can you solve it in-place?
  3. What if duplicates must be handled differently?
  4. How would you parallelize the approach?
  5. Design tests that would break a buggy implementation.

Practice Recommendations

What to practice next

  1. Re-solve Divide an Array Into Subarrays With Minimum Cost II in a second language (cpp, python).
  2. Drill 3–5 more problems tagged two pointers.
  3. Teach the solution out loud in under 5 minutes.
  4. Add this problem to your revision calendar in 3 days and 14 days.

Visualization

Conceptual diagram for Divide an Array Into Subarrays With Minimum Cost II: show input structure (two pointers), highlight the moving parts of the sliding window, heap, and sorted list approach, and annotate each step with the maintained invariant and complexity.

Study checklist

  • Read the official problem statement on LeetCode
  • Solve on paper / whiteboard first
  • Implement the sliding window, heap, and sorted list approach
  • Verify edge cases from the checklist
  • State time and space complexity aloud
  • Compare with the AlgoForge reference solution
  • Schedule a revision session

Revision notes

Divide an Array Into Subarrays With Minimum Cost II (#3013) — Hard. Pattern: sliding window, heap, and sorted list. Complexity: O(nlogd) time / O(d) space. Re-derive the invariant before coding.

FAQs

What is the time complexity of Divide an Array Into Subarrays With Minimum Cost II?+

The reference solutions aim for O(nlogd) time and O(d) space. Always re-derive complexity from the code you write in the interview.

What pattern does Divide an Array Into Subarrays With Minimum Cost II use?+

It primarily maps to sliding window, heap, and sorted list, within the broader topic of two pointers.

Is Divide an Array Into Subarrays With Minimum Cost II good for interviews?+

Yes — as a Hard problem it is a solid practice target. Pair it with related problems in the same pattern family for spaced repetition.

Where can I read the official statement?+

Open the official LeetCode page for constraints and examples: https://leetcode.com/problems/divide-an-array-into-subarrays-with-minimum-cost-ii/