Search Blogs

Showing results for "Stock Series"

Found 5 results

LeetCode 123 — Best Time to Buy and Sell Stock III | At Most Two Transactions Explained

🚀 Try This Problem First!Before reading the solution, attempt it yourself on LeetCode — you will learn much more that way.🔗 Problem Link: https://leetcode.com/problems/best-time-to-buy-and-sell-stock-iii/What Is the Problem Asking?You have a list of stock prices, one for each day. You are allowed to buy and sell the stock, but with one important restriction — you can make at most two transactions in total. A transaction means one buy followed by one sell.You cannot hold two stocks at the same time, meaning you must sell whatever you are holding before you buy again.Your job is to find the maximum profit you can make by doing at most two transactions. If there is no way to make any profit, return 0.Example to build intuition: prices = [3, 3, 5, 0, 0, 3, 1, 4]If you buy on day 4 (price 0) and sell on day 6 (price 3), profit = 3. Then buy on day 7 (price 1) and sell on day 8 (price 4), profit = 3. Total profit = 6. That is the best you can do.Constraints:1 ≤ prices.length ≤ 10⁵0 ≤ prices[i] ≤ 10⁵Why Is This Harder Than the Previous Stock Problems?In LeetCode 121, you could only make one transaction — so you just found the best single buy-sell pair.In LeetCode 122, you could make unlimited transactions — so you greedily collected every upward price movement.Here, you are limited to exactly two transactions. Greedy does not work anymore because sometimes skipping a small profit early on allows you to make a much bigger combined profit across two trades. You need a smarter strategy.The Big Idea Behind the SolutionHere is a simple way to think about it.If you are allowed two transactions, you can imagine drawing a dividing line somewhere in the price array. The first transaction happens entirely on the left side of that line. The second transaction happens entirely on the right side.So the problem becomes:For every possible dividing position, find the best profit on the left side and the best profit on the right side.Add them together.The maximum sum across all dividing positions is your answer.The challenge is doing this efficiently without checking every position with a nested loop, which would be O(N²) and too slow.Approach 1 — Left-Right Prefix DPThe idea:Instead of trying every split point from scratch, precompute the answers in advance.Build a leftProfit array where leftProfit[i] stores the best single transaction profit you can make using prices from day 0 to day i.Build a rightProfit array where rightProfit[i] stores the best single transaction profit you can make using prices from day i to the last day.Then for every index i, the answer if you split at i is simply leftProfit[i] + rightProfit[i]. Take the maximum across all indices.How to build leftProfit:Go left to right. Keep track of the minimum price seen so far. At each day i, the best profit you could make ending at or before day i is either the same as the day before, or selling today at today's price minus the minimum price seen so far. Take whichever is larger.How to build rightProfit:Go right to left. Keep track of the maximum price seen so far from the right. At each day i, the best profit you could make starting at or after day i is either the same as the day after, or buying today at today's price and selling at the maximum price seen to the right. Take whichever is larger.Dry Run — prices = [3, 3, 5, 0, 0, 3, 1, 4]Building leftProfit left to right, tracking minimum price seen:Day 0 → min=3, leftProfit[0] = 0 (nothing to compare yet) Day 1 → min=3, best profit = 3-3 = 0, leftProfit[1] = max(0, 0) = 0 Day 2 → min=3, best profit = 5-3 = 2, leftProfit[2] = max(0, 2) = 2 Day 3 → min=0, best profit = 0-0 = 0, leftProfit[3] = max(2, 0) = 2 Day 4 → min=0, best profit = 0-0 = 0, leftProfit[4] = max(2, 0) = 2 Day 5 → min=0, best profit = 3-0 = 3, leftProfit[5] = max(2, 3) = 3 Day 6 → min=0, best profit = 1-0 = 1, leftProfit[6] = max(3, 1) = 3 Day 7 → min=0, best profit = 4-0 = 4, leftProfit[7] = max(3, 4) = 4leftProfit = [0, 0, 2, 2, 2, 3, 3, 4]Building rightProfit right to left, tracking maximum price seen:Day 7 → max=4, rightProfit[7] = 0 (nothing to compare yet) Day 6 → max=4, best profit = 4-1 = 3, rightProfit[6] = max(0, 3) = 3 Day 5 → max=4, best profit = 4-3 = 1, rightProfit[5] = max(3, 1) = 3 Day 4 → max=4, best profit = 4-0 = 4, rightProfit[4] = max(3, 4) = 4 Day 3 → max=4, best profit = 4-0 = 4, rightProfit[3] = max(4, 4) = 4 Day 2 → max=5, best profit = 5-5 = 0, rightProfit[2] = max(4, 0) = 4 Day 1 → max=5, best profit = 5-3 = 2, rightProfit[1] = max(4, 2) = 4 Day 0 → max=5, best profit = 5-3 = 2, rightProfit[0] = max(4, 2) = 4rightProfit = [4, 4, 4, 4, 4, 3, 3, 0]Combining at each index: Index 0 → 0 + 4 = 4 Index 1 → 0 + 4 = 4 Index 2 → 2 + 4 = 6 Index 3 → 2 + 4 = 6 Index 4 → 2 + 4 = 6 Index 5 → 3 + 3 = 6 Index 6 → 3 + 3 = 6 Index 7 → 4 + 0 = 4Maximum = 6 ✅Implementation code:class Solution { public int maxProfit(int[] prices) { int lefProf[] = new int[prices.length]; int rigProf[] = new int[prices.length]; int j = 1; int i = 0; int coL = 1; while (i < j && j < prices.length) { if (prices[i] > prices[j]) { i = j; if (coL - 1 != -1) { lefProf[coL] = lefProf[coL - 1]; } coL++; } else { int max = prices[j] - prices[i]; if (coL - 1 != -1) { lefProf[coL] = Math.max(lefProf[coL - 1], max); coL++; } else { lefProf[coL] = max; coL++; } } j++; } int ii = prices.length - 2; int jj = prices.length - 1; int coR = prices.length - 2; while (ii >= 0) { if (prices[ii] > prices[jj]) { jj = ii; if (coR + 1 < prices.length) { rigProf[coR] = rigProf[coR + 1]; } coR--; } else { int max = prices[jj] - prices[ii]; if (coR + 1 < prices.length) { rigProf[coR] = Math.max(rigProf[coR + 1], max); coR--; } else { rigProf[coR] = max; coR--; } } ii--; } int maxAns = 0; for (int k = 0; k < lefProf.length; k++) { maxAns = Math.max(maxAns, lefProf[k] + rigProf[k]); } return maxAns; }}The approach here is exactly right. You are building the left and right profit arrays using a two pointer scanning technique — the same idea as LeetCode 121 but applied twice, once going forward and once going backward. The manual counters coL and coR make it a bit harder to follow, but the logic underneath is solid.Here is the same approach written in a cleaner way so it is easier to read and debug:class Solution { public int maxProfit(int[] prices) { int n = prices.length; int[] leftProfit = new int[n]; int[] rightProfit = new int[n]; int minPrice = prices[0]; for (int i = 1; i < n; i++) { minPrice = Math.min(minPrice, prices[i]); leftProfit[i] = Math.max(leftProfit[i - 1], prices[i] - minPrice); } int maxPrice = prices[n - 1]; for (int i = n - 2; i >= 0; i--) { maxPrice = Math.max(maxPrice, prices[i]); rightProfit[i] = Math.max(rightProfit[i + 1], maxPrice - prices[i]); } int maxProfit = 0; for (int i = 0; i < n; i++) { maxProfit = Math.max(maxProfit, leftProfit[i] + rightProfit[i]); } return maxProfit; }}Time Complexity: O(N) — three separate linear passes through the array. Space Complexity: O(N) — two extra arrays of size N.Approach 2 — Four Variable State Machine DPThe idea:Instead of building arrays, what if you could track everything in just four variables and solve it in a single pass?Think about what is happening at any point in time as you go through the prices. You are always in one of these situations:You have bought your first stock and are currently holding it. You have sold your first stock and are sitting on that profit. You have bought your second stock (using the profit from the first sale) and are holding it. You have sold your second stock and collected your total profit.These four situations are your four states. For each one, you always want to know — what is the best possible profit I could have in this state up to today?Here is how each state updates as you look at a new price:buy1 — This represents the best outcome after buying the first stock. Buying costs money, so this is negative. As you see each new price, you ask: is it better to have bought on some earlier day, or to buy today? You keep whichever gives you the least cost, which means the highest value of negative price.sell1 — This represents the best profit after completing the first transaction. As you see each new price, you ask: is it better to have sold on some earlier day, or to sell today using the best buy1 we have? Selling today means adding today's price on top of buy1.buy2 — This represents the best outcome after buying the second stock. The key insight here is that you are not starting from zero — you already have the profit from the first transaction in your pocket. So buying the second stock costs today's price but you subtract it from sell1. As you see each new price, you ask: is it better to have bought the second stock earlier, or to buy today using the profit from sell1?sell2 — This is your final answer. It represents the best total profit after completing both transactions. As you see each new price, you ask: is it better to have completed both transactions earlier, or to sell the second stock today using the best buy2 we have?The beautiful thing is that all four updates happen in order on every single day, in one pass through the array. Each variable feeds into the next — buy1 feeds sell1, sell1 feeds buy2, buy2 feeds sell2.Dry Run — prices = [3, 3, 5, 0, 0, 3, 1, 4]We start with buy1 and buy2 as very negative (not yet bought anything) and sell1 and sell2 as 0 (no profit yet).Day 0, price = 3: buy1 = max(-∞, -3) = -3. Best cost to buy first stock is spending 3. sell1 = max(0, -3+3) = 0. Selling immediately gives no profit. buy2 = max(-∞, 0-3) = -3. Best cost to buy second stock after first sale is 3. sell2 = max(0, -3+3) = 0. No total profit yet.Day 1, price = 3: Nothing changes — same price as day 0.Day 2, price = 5: buy1 = max(-3, -5) = -3. Still best to have bought at price 3. sell1 = max(0, -3+5) = 2. Selling today at 5 after buying at 3 gives profit 2. buy2 = max(-3, 2-5) = -3. Buying second stock today costs too much relative to first profit. sell2 = max(0, -3+5) = 2. Completing both trades gives 2 so far.Day 3, price = 0: buy1 = max(-3, -0) = 0. Buying today at price 0 is the best possible first buy. sell1 = max(2, 0+0) = 2. Selling today at 0 gives nothing — keep the 2 from before. buy2 = max(-3, 2-0) = 2. Using the profit of 2 and buying today at 0 gives buy2 = 2. sell2 = max(2, 2+0) = 2. Selling second stock today at 0 gives nothing extra.Day 4, price = 0: Nothing changes — same price as day 3.Day 5, price = 3: buy1 = max(0, -3) = 0. Still best to have bought at price 0. sell1 = max(2, 0+3) = 3. Selling today at 3 after buying at 0 gives profit 3. buy2 = max(2, 3-3) = 2. Buying second stock today at 3 using sell1=3 gives 0. Keep 2. sell2 = max(2, 2+3) = 5. Selling second stock today at 3 after buy2=2 gives total 5.Day 6, price = 1: buy1 = max(0, -1) = 0. Still best to have bought at 0. sell1 = max(3, 0+1) = 3. Selling today at 1 gives less than 3 — no change. buy2 = max(2, 3-1) = 2. Buying second stock at 1 using sell1=3 gives 2 — same as before. sell2 = max(5, 2+1) = 5. No improvement yet.Day 7, price = 4: buy1 = max(0, -4) = 0. No change. sell1 = max(3, 0+4) = 4. Selling today at 4 after buying at 0 gives profit 4 — new best. buy2 = max(2, 4-4) = 2. No improvement. sell2 = max(5, 2+4) = 6. Selling second stock today at 4 with buy2=2 gives total 6 — new best.Output: 6 ✅class Solution { public int maxProfit(int[] prices) { int buy1 = Integer.MIN_VALUE; int sell1 = 0; int buy2 = Integer.MIN_VALUE; int sell2 = 0; for (int price : prices) { buy1 = Math.max(buy1, -price); sell1 = Math.max(sell1, buy1 + price); buy2 = Math.max(buy2, sell1 - price); sell2 = Math.max(sell2, buy2 + price); } return sell2; }}Time Complexity: O(N) — single pass through the array. Space Complexity: O(1) — only four integer variables, no extra arrays.Approach 3 — Generalizing to At Most K TransactionsOnce you understand Approach 2, there is a natural question — what if instead of two transactions, you were allowed K transactions? That is exactly LeetCode 188.Instead of four hardcoded variables, you use two arrays of size K. buy[j] tracks the best outcome after the j-th buy and sell[j] tracks the best outcome after the j-th sell. The logic is identical to Approach 2, just in a loop.For this problem set K = 2 and it gives the same answer:class Solution { public int maxProfit(int[] prices) { int k = 2; int[] buy = new int[k]; int[] sell = new int[k]; Arrays.fill(buy, Integer.MIN_VALUE); for (int price : prices) { for (int j = 0; j < k; j++) { buy[j] = Math.max(buy[j], (j == 0 ? 0 : sell[j - 1]) - price); sell[j] = Math.max(sell[j], buy[j] + price); } } return sell[k - 1]; }}Time Complexity: O(N × K) — for K=2 this is effectively O(N). Space Complexity: O(K) — two small arrays of size K.If you understand this, LeetCode 188 becomes a five minute problem.Comparing All Three ApproachesApproach 1 — Left-Right Prefix DP (Your Approach)You build two arrays — one scanning left to right, one scanning right to left — and combine them. This is the most visual and intuitive approach. It is easy to explain because you are essentially solving LeetCode 121 twice from opposite ends and adding the results. The downside is it uses O(N) extra space for the two arrays.Approach 2 — Four Variable State MachineYou solve the entire problem in a single pass using just four variables. No arrays, no extra memory. This is the most efficient and elegant solution. Once you understand what each variable represents, the code almost writes itself. This is what most interviewers are hoping to see.Approach 3 — General K-Transaction DPThis is Approach 2 extended to handle any number of transactions using arrays instead of hardcoded variables. Solving LeetCode 123 with this approach means you have already solved LeetCode 188 as a bonus.Mistakes That Catch Most PeopleTrying to use the greedy approach from LeetCode 122: Adding every upward price movement does not work when you are limited to two transactions. You need to pick the two best non-overlapping windows, not collect everything.Buying twice without selling in between: The state machine prevents this naturally — buy2 depends on sell1, so you can never enter the second buy before completing the first sell.Initializing buy variables to 0: Both buy1 and buy2 must start at Integer.MIN_VALUE. Setting them to 0 implies you bought a stock for free, which is wrong and will give inflated profits.Returning the wrong variable: Always return sell2, not buy2 or sell1. sell2 is the only variable that represents a completed two-transaction profit.Where This Fits in the Stock SeriesLeetCode 121 — One transaction only. Find the single best buy-sell pair. [ Blog is also avaliable on this - Read Now ]LeetCode 122 — Unlimited transactions. Collect every upward price movement greedily. [ Blog is also avaliable on this - Read Now ]LeetCode 188 — At most K transactions. Direct extension of Approach 3 above.LeetCode 309 — Unlimited transactions but with a cooldown day after every sale.LeetCode 714 — Unlimited transactions but with a fee charged on every transaction.Each problem adds one new constraint on top of the previous one. If you understand LeetCode 123 deeply — especially Approach 2 — every other problem in this series becomes a small modification of the same framework.Key Takeaways✅ When you are limited to two transactions, greedy fails. You need to find two non-overlapping windows that together give maximum profit.✅ The left-right prefix DP approach is the most intuitive — precompute the best single transaction from both ends, then combine at every split point.✅ The four variable state machine solves it in one pass with O(1) space. Each variable tracks the best outcome for one state — buy1 feeds sell1, sell1 feeds buy2, buy2 feeds sell2.✅ Always initialize buy variables to Integer.MIN_VALUE to represent "not yet bought anything."✅ Understanding Approach 3 here means LeetCode 188 is already solved — just change the hardcoded 2 to K.Happy Coding! This problem is the turning point in the stock series where DP becomes unavoidable — once you crack it, the rest of the series feels manageable. 🚀

LeetCodeDynamic ProgrammingHardJavaDSAPrefix DPArraysStock Series

LeetCode 122 — Best Time to Buy and Sell Stock II | Every Approach Explained

🚀 Try This Problem First!Before reading the solution, attempt it yourself on LeetCode — you'll retain the concept far better.🔗 Problem Link: https://leetcode.com/problems/best-time-to-buy-and-sell-stock-ii/Understanding the ProblemYou are given an array prices where prices[i] is the stock price on day i. Unlike the classic version, here you can make as many transactions as you want — but you can only hold one share at a time. You may buy and sell on the same day.Goal: Return the maximum total profit achievable.Key Rules:You can buy and sell multiple times.You cannot hold more than one share at a time — you must sell before buying again.If no profit is possible, return 0.Constraints:1 ≤ prices.length ≤ 3 × 10⁴0 ≤ prices[i] ≤ 10⁴How This Differs From LeetCode 121 #In LeetCode 121, you were limited to exactly one buy-sell transaction. Here, the restriction is lifted — you can participate in as many transactions as you want. This fundamentally changes the strategy. Instead of hunting for the single best pair, you want to capture every profitable price movement in the array.The Core InsightLook at the price chart mentally. Every time the price goes up from one day to the next, that's money on the table. The question is — how do you collect all of it?The answer is surprisingly simple: add every single upward price difference to your profit. If prices go up three days in a row from 1 → 3 → 5 → 8, you collect (3-1) + (5-3) + (8-5) = 7, which is exactly the same as buying at 1 and selling at 8. You never miss a gain.This is the foundation of all approaches below.Approach 1 — Simple Greedy (Collect Every Upward Move)Intuition: Every time prices[i] > prices[i-1], add the difference to profit. You are essentially buying at every valley and selling at every peak, collecting each individual daily gain without explicitly tracking buy/sell days.Why it works: The total gain from buying at day 0 and selling at day N is mathematically equal to the sum of all positive daily differences in between. You never lose anything by collecting gains day by day.Example: prices = [1, 2, 3, 4, 5] Daily gains: (2-1) + (3-2) + (4-3) + (5-4) = 1+1+1+1 = 4 Same as buying at 1 and selling at 5 directly.class Solution {public int maxProfit(int[] prices) {int maxProfit = 0;for (int i = 1; i < prices.length; i++) {if (prices[i] > prices[i - 1]) {maxProfit += prices[i] - prices[i - 1];}}return maxProfit;}}Time Complexity: O(N) — single pass through the array. Space Complexity: O(1) — no extra space used.This is the cleanest and most recommended solution for this problem.Approach 2 — Peak Valley ApproachIntuition: Instead of collecting every daily gain, explicitly find every valley (local minimum) to buy at and every peak (local maximum) to sell at. You buy when price stops falling and sell when price stops rising.How it works: Scan through the array. When you find a valley (prices[i] ≤ prices[i+1]), that is your buy point. Then keep going until you find a peak (prices[i] ≥ prices[i+1]) — that is your sell point. Add the peak minus valley to profit. Repeat.Example: prices = [7, 1, 5, 3, 6, 4]Valley at index 1 (price = 1), Peak at index 2 (price = 5) → profit += 4 Valley at index 3 (price = 3), Peak at index 4 (price = 6) → profit += 3 Total = 7 ✅class Solution {public int maxProfit(int[] prices) {int i = 0;int maxProfit = 0;int valley, peak;while (i < prices.length - 1) {while (i < prices.length - 1 && prices[i] >= prices[i + 1]) {i++;}valley = prices[i];while (i < prices.length - 1 && prices[i] <= prices[i + 1]) {i++;}peak = prices[i];maxProfit += peak - valley;}return maxProfit;}}Time Complexity: O(N) — each element is visited at most twice. Space Complexity: O(1) — no extra space used.This approach is more explicit and easier to visualize on a graph, though the code is slightly more involved than Approach 1.Approach 3 — Two PointerIntuition: Use two pointers i (buy day) and j (sell day). Move j forward one step at a time. Whenever prices[j] > prices[i], you have a profitable window — add the profit and immediately move i to j (simulate selling and rebuying at the same price on the same day). Whenever prices[j] < prices[i], just move i to j since a cheaper buy day has been found.Why moving i to j after every profitable sale works: Selling at j and immediately rebuying at j costs nothing (profit of 0 for that rebuy). But it positions i at the latest price so you can catch the next upward movement. This correctly simulates collecting every upward segment.Example: prices = [7, 1, 5, 3, 6, 4]i=0, j=1 → 7 > 1, move i to 1. j=2. i=1, j=2 → 1 < 5, profit += 4, move i to 2. j=3. i=2, j=3 → 5 > 3, move i to 3. j=4. i=3, j=4 → 3 < 6, profit += 3, move i to 4. j=5. i=4, j=5 → 6 > 4, move i to 5. j=6. Loop ends. Total profit = 7 ✅class Solution {public int maxProfit(int[] prices) {int i = 0;int j = 1;int maxProfit = 0;while (i < j && j < prices.length) {if (prices[i] > prices[j]) {i = j;} else {maxProfit += prices[j] - prices[i];i = j;}j++;}return maxProfit;}}Time Complexity: O(N) — j traverses the array exactly once. Space Complexity: O(1) — only three integer variables.This approach is functionally identical to Approach 1 — both collect every upward daily movement. The two pointer framing makes the buy/sell simulation more explicit.Approach 4 — Dynamic ProgrammingIntuition: At any point in time, you are in one of two states — either you hold a stock or you do not hold a stock. Define two DP values updated each day:hold = maximum profit if you are currently holding a stock at the end of this day.cash = maximum profit if you are not holding any stock at the end of this day.Transitions:To hold on day i: either you already held yesterday, or you buy today. hold = max(hold, cash - prices[i])To have cash on day i: either you already had cash yesterday, or you sell today. cash = max(cash, hold + prices[i])Initialization:hold = -prices[0] (you bought on day 0)cash = 0 (you did nothing on day 0)class Solution {public int maxProfit(int[] prices) {int hold = -prices[0];int cash = 0;for (int i = 1; i < prices.length; i++) {hold = Math.max(hold, cash - prices[i]);cash = Math.max(cash, hold + prices[i]);}return cash;}}Time Complexity: O(N) — single pass. Space Complexity: O(1) — only two variables maintained at each step.This approach is the most powerful because it extends naturally to harder variants of this problem — like LeetCode 309 (with cooldown) and LeetCode 714 (with transaction fee) — where greedy no longer works and you need explicit state tracking.Dry Run — All Approaches on Example 1Input: prices = [7, 1, 5, 3, 6, 4], Expected Output: 7Approach 1 (Simple Greedy): Day 1→2: 1 - 7 = -6, skip. Day 2→3: 5 - 1 = 4, add. profit = 4. Day 3→4: 3 - 5 = -2, skip. Day 4→5: 6 - 3 = 3, add. profit = 7. Day 5→6: 4 - 6 = -2, skip. Result = 7 ✅Approach 4 (DP): Start: hold = -7, cash = 0. Day 1 (price=1): hold = max(-7, 0-1) = -1. cash = max(0, -1+1) = 0. Day 2 (price=5): hold = max(-1, 0-5) = -1. cash = max(0, -1+5) = 4. Day 3 (price=3): hold = max(-1, 4-3) = 1. cash = max(4, 1+3) = 4. Day 4 (price=6): hold = max(1, 4-6) = 1. cash = max(4, 1+6) = 7. Day 5 (price=4): hold = max(1, 7-4) = 3. cash = max(7, 3+4) = 7. Result = 7 ✅Comparison of All ApproachesApproach 1 — Simple Greedy Code simplicity: Simplest possible. Best for interviews — clean and readable. Does not extend to constrained variants.Approach 2 — Peak Valley Code simplicity: Moderate. Best for visual/conceptual understanding. Slightly verbose but maps directly to a chart.Approach 3 — Two Pointer Code simplicity: Simple. Explicit simulation of buy/sell actions. Functionally identical to Approach 1.Approach 4 — Dynamic Programming Code simplicity: Moderate. Most powerful — extends to cooldown, fee, and k-transaction variants. Worth mastering for the full stock problem series.Common Mistakes to AvoidThinking you need to find exact buy/sell days: The problem only asks for maximum profit — you do not need to output which days you traded. This frees you to use the simple greedy sum approach.Trying to find the global minimum and maximum: Unlike LeetCode 121, the single best buy-sell pair is not always optimal here. You need to capture multiple smaller movements, not one big one.Holding more than one share: You cannot buy twice in a row without selling in between. In Approach 3, moving i = j after every transaction ensures you always sell before the next buy.Not handling a flat or decreasing array: If prices never go up, all approaches correctly return 0 — the greedy sum adds nothing, peak-valley finds no valid pairs, and DP's cash stays at 0.Complexity SummaryAll four approaches run in O(N) time and O(1) space. The difference between them is conceptual clarity and extensibility, not raw performance.The Full Stock Problem Series on LeetCodeThis problem is part of a six-problem series. Understanding them in order builds intuition progressively:LeetCode 121 — One transaction only. Two pointer / min tracking greedy. [ Blog is also avaliable on this - Read Now]LeetCode 123 — At most 2 transactions. DP with explicit state for two transactions. [ Blog is also avaliable on this - Read Now]LeetCode 188 — At most k transactions. Generalized DP.LeetCode 309 — Unlimited transactions with cooldown after selling. DP with three states.LeetCode 714 — Unlimited transactions with a fee per transaction. DP with adjusted transitions.Each problem adds one constraint on top of the previous. If you understand the DP state machine from Approach 4 deeply, every problem in this series becomes a small modification of the same framework.Key Takeaways✅ When transactions are unlimited, collect every upward daily price movement — that is the global optimum.✅ The sum of all positive daily differences equals the sum of all peak-valley differences. Both are provably optimal.✅ The two pointer approach explicitly simulates buy and sell events — moving i = j after a sale means selling and immediately rebuying at the same price to stay positioned for the next gain.✅ The DP approach with hold and cash states is the most versatile — it is the foundation for every harder variant in the stock series.✅ Always initialize maxProfit = 0 so that the no-profit case (prices only falling) is handled correctly without extra conditions.Happy Coding! Once you have this problem locked down, the rest of the stock series will feel like natural extensions rather than new problems entirely. 🚀

LeetCodeGreedyTwo PointersDynamic ProgrammingMediumJavaArrays

LeetCode 121 — Best Time to Buy and Sell Stock I | Two Pointer / Sliding Window Explained

🚀 Try This Problem First!Before reading the solution, attempt it yourself on LeetCode — you'll retain the concept far better.🔗 Problem Link: https://leetcode.com/problems/best-time-to-buy-and-sell-stock/Understanding the ProblemYou are given an array prices where prices[i] is the stock price on day i. You can buy on one day and sell on a later day. You want to maximize your profit.Goal: Return the maximum profit possible. If no profit can be made, return 0.Key Rules:You must buy before you sell — no going back in time.You can only make one transaction (one buy, one sell).If prices only go down, profit is impossible — return 0.Constraints:1 ≤ prices.length ≤ 10⁵0 ≤ prices[i] ≤ 10⁴Understanding the Problem With a Real-World AnalogyImagine you're watching a stock ticker for a week. You want to pick the single best day to buy and a later day to sell. You can't predict the future — you just have historical prices. The question is: looking back at all the prices, what was the best single buy-sell pair you could have chosen?Key Observation (Before Writing a Single Line of Code)The profit on any given pair of days is:profit = prices[sell_day] - prices[buy_day]To maximize profit, for every potential sell day, we want the lowest possible buy price seen so far to the left of it. This is the core insight that drives the efficient solution.If at any point the current price is lower than our tracked minimum, there is no point keeping the old minimum — the new one is strictly better for any future sell day.Intuition — The Two Pointer ApproachWe use two pointers i (buy pointer) and j (sell pointer), both starting at the beginning with i = 0 and j = 1.At every step we ask: is prices[j] greater than prices[i]?If yes → this is a profitable pair. Compute the profit and update the maximum if it's better.If no → prices[j] is cheaper than prices[i]. There's no reason to keep i where it is — any future sell day would be better served by buying at j instead. So we move i to j.Either way, j always moves forward by 1 until it reaches the end of the array.Why Moving i to j is CorrectThis is the most important conceptual point. When prices[j] < prices[i], you might wonder — why not just skip j and move on? Why move i?Because for any future day k > j, the profit buying at j is:prices[k] - prices[j]And the profit buying at i is:prices[k] - prices[i]Since prices[j] < prices[i], buying at j will always give a higher or equal profit for the same sell day k. So we update i = j — there is no scenario where keeping the old i is better.Dry Run — Example 1 (Step by Step)Input: prices = [7, 1, 5, 3, 6, 4]We start with i = 0, j = 1, maxP = 0.Step 1: i = 0, j = 1 → prices[i] = 7, prices[j] = 17 > 1 → prices[j] is cheaper. Move i to j. i = 1, j moves to 2. maxP = 0.Step 2: i = 1, j = 2 → prices[i] = 1, prices[j] = 51 < 5 → Profitable! profit = 5 - 1 = 4. 4 > 0 → update maxP = 4. j moves to 3.Step 3: i = 1, j = 3 → prices[i] = 1, prices[j] = 31 < 3 → Profitable! profit = 3 - 1 = 2. 2 < 4 → maxP stays 4. j moves to 4.Step 4: i = 1, j = 4 → prices[i] = 1, prices[j] = 61 < 6 → Profitable! profit = 6 - 1 = 5. 5 > 4 → update maxP = 5. j moves to 5.Step 5: i = 1, j = 5 → prices[i] = 1, prices[j] = 41 < 4 → Profitable! profit = 4 - 1 = 3. 3 < 5 → maxP stays 5. j moves to 6. j = 6 = prices.length → loop ends.Output: maxP = 5 ✅ (Buy at day 2 price=1, sell at day 5 price=6)Dry Run — Example 2 (Step by Step)Input: prices = [7, 6, 4, 3, 1]We start with i = 0, j = 1, maxP = 0.Step 1: i = 0, j = 1 → prices[i] = 7, prices[j] = 67 > 6 → Move i to j. i = 1, j moves to 2. maxP = 0.Step 2: i = 1, j = 2 → prices[i] = 6, prices[j] = 46 > 4 → Move i to j. i = 2, j moves to 3. maxP = 0.Step 3: i = 2, j = 3 → prices[i] = 4, prices[j] = 34 > 3 → Move i to j. i = 3, j moves to 4. maxP = 0.Step 4: i = 3, j = 4 → prices[i] = 3, prices[j] = 13 > 1 → Move i to j. i = 4, j moves to 5. j = 5 = prices.length → loop ends.Output: maxP = 0 ✅ (Prices only went down, no profitable transaction exists)The Code — Implementationclass Solution {public int maxProfit(int[] prices) {int i = 0; // Buy pointer — points to the current minimum price dayint j = 1; // Sell pointer — scans forward through the arrayint maxP = 0; // Tracks the maximum profit seen so far// j scans from day 1 to the end of the arraywhile (i < j && j < prices.length) {if (prices[i] > prices[j]) {// prices[j] is cheaper than current buy price// Move buy pointer to j — buying here is strictly better// for any future sell dayi = j;} else {// prices[j] > prices[i] — this is a profitable pair// Calculate profit and update maxP if it's the best so farint profit = prices[j] - prices[i];if (profit > maxP) {maxP = profit;}}// Always move the sell pointer forwardj++;}// If no profitable transaction was found, maxP remains 0return maxP;}}Code Walkthrough — Step by StepInitialization: i = 0 acts as our buy day pointer (always pointing to the lowest price seen so far). j = 1 is our sell day pointer scanning forward. maxP = 0 is our answer — defaulting to 0 handles the case where no profit is possible.The loop condition: i < j ensures we never sell before buying. j < prices.length ensures we don't go out of bounds.When prices[i] > prices[j]: The current sell day price is cheaper than our buy day. We update i = j, because buying at j dominates buying at i for all future sell days.When prices[j] >= prices[i]: We have a potential profit. We compute it and update maxP if it beats the current best.j always increments: Regardless of which branch we take, j++ moves the sell pointer forward every iteration — this is what drives the loop to completion.Common Mistakes to AvoidNot returning 0 for no-profit cases: If prices are strictly decreasing, maxP never gets updated and stays 0. The initialization maxP = 0 handles this correctly — never initialize it to a negative number.Using a nested loop (brute force): A common first instinct is two nested loops checking every pair. That is O(N²) and will TLE on large inputs. The two pointer approach solves it in O(N).Thinking you need to sort: Sorting destroys the order of days, which is fundamental to the problem. Never sort the prices array here.Moving i forward by 1 instead of to j: When prices[j] < prices[i], some people write i++ instead of i = j. This is wrong — you might miss the new minimum entirely and waste iterations.Complexity AnalysisTime Complexity: O(N) The j pointer traverses the array exactly once from index 1 to the end. The i pointer only moves forward and never exceeds j. So the entire algorithm is a single linear pass — O(N).Space Complexity: O(1) No extra arrays, maps, or stacks are used. Only three integer variables — i, j, and maxP.Alternative Way to Think About It (Min Tracking)Some people write this problem using a minPrice variable instead of two pointers. Both approaches are equivalent — the two pointer framing is slightly more intuitive visually, but here is the min-tracking version for reference:int minPrice = Integer.MAX_VALUE;int maxProfit = 0;for (int price : prices) {if (price < minPrice) {minPrice = price;} else if (price - minPrice > maxProfit) {maxProfit = price - minPrice;}}return maxProfit;The logic is the same — always track the cheapest buy day seen so far, and for each day compute what profit would look like if you sold today.Similar Problems (Build on This Foundation)LeetCode 122 — Best Time to Buy and Sell Stock II (multiple transactions allowed) [ Blog is also avaliable on this - Read Now ]LeetCode 123 — Best Time to Buy and Sell Stock III (at most 2 transactions) [ Blog is also avaliable on this - Read Now ]LeetCode 188 — Best Time to Buy and Sell Stock IV (at most k transactions)LeetCode 309 — Best Time to Buy and Sell Stock with CooldownLeetCode 714 — Best Time to Buy and Sell Stock with Transaction FeeLeetCode 121 is the foundation. Master this one deeply before moving to the others — they all build on the same core idea.Key Takeaways✅ For every potential sell day, the best buy day is always the minimum price seen to its left — this drives the whole approach.✅ When the sell pointer finds a cheaper price than the buy pointer, always move the buy pointer there — it strictly dominates the old buy day for all future sells.✅ Initialize maxP = 0 so that no-profit scenarios (prices only going down) are handled automatically.✅ The two pointer approach solves this in a single linear pass — O(N) time and O(1) space.✅ j always increments every iteration — i only moves when a cheaper price is found.Happy Coding! If this clicked for you, the entire Stock series on LeetCode will feel much more approachable.

LeetCodeTwo PointersEasyJavaDSA

Stack Problems Explained: NGR, NGL, NSR, NSL — The Four-Problem Family You Must Master

IntroductionAmong all the problems built around the Stack data structure, four stand out as a family — they appear repeatedly in coding interviews, competitive programming, and real-world software systems. These four are the Next Greater to the Right (NGR), Next Greater to the Left (NGL), Next Smaller to the Right (NSR), and Next Smaller to the Left (NSL).What makes them special is not just their individual solutions — it is the fact that all four are solved by a single elegant technique called the Monotonic Stack. Learn the pattern once, and you have all four in your toolkit permanently.This guide breaks down each problem with a full solution, step-by-step dry run, edge cases, and the exact reasoning behind every decision in the code. Whether you are preparing for a technical interview or simply want to deeply understand this pattern — you are in the right place.The Story That Makes This ClickBefore any code, let us understand this family of problems with one real-world story.Imagine you are standing in a queue at a cricket stadium. Everyone in the queue has a different height. You are standing somewhere in the middle. You look to your right and ask — who is the first person taller than me? That is your Next Greater Element to the Right (NGR).Now you look to your left — who is the first person taller than me on this side? That is your Next Greater to the Left (NGL).Now instead of taller, you ask shorter — who is the first shorter person to my right? That is Next Smaller to the Right (NSR).And shorter to your left? That is Next Smaller to the Left (NSL).Same queue. Same people. Four different questions. Four different answers. This is exactly what these four problems are about — and they all share the same solution pattern.What Is a Monotonic Stack?A monotonic stack is just a regular stack with one rule — elements inside it are always maintained in a specific order, either always increasing or always decreasing from bottom to top.You never enforce this rule explicitly. It happens naturally as you pop elements that violate the order before pushing a new one. This popping step is the key insight — the moment you pop an element, you have found its answer for the current element being processed.This one pattern solves all four problems. Only two small details change between them — the direction of traversal and the comparison condition inside the while loop.The Four Problems — Quick ReferenceProblemDirectionWhat You WantProblem LinksNGRTraverse Right to LeftFirst greater on right"Next Greater Element GFG"NGLTraverse Left to RightFirst greater on left"Previous Greater Element GFG"NSRTraverse Right to LeftFirst smaller on right"Next Smaller Element GFG"NSLTraverse Left to RightFirst smaller on left"Previous Smaller Element GFG"Problem 1 — Next Greater Element to Right (NGR)GFG Problem: Search "Next Greater Element" on GeeksForGeeks Difficulty: Medium | Accuracy: 32.95% | Submissions: 515K+The QuestionFor each element in the array, find the first element to its right that is strictly greater than it. If none exists, return -1.Input: [1, 3, 2, 4] Output: [3, 4, 4, -1]Input: [6, 8, 0, 1, 3] Output: [8, -1, 1, 3, -1]Real World ExampleThink of the stock market. You have daily closing prices: [1, 3, 2, 4]. For each day, you want to know — on which future day will the price first exceed today's price? Day 1 has price 1, first exceeded on Day 2 with price 3. Day 2 has price 3, first exceeded on Day 4 with price 4. Day 3 has price 2, also first exceeded on Day 4 with price 4. Day 4 has no future day, so -1. This is exactly NGR — and it is literally used in financial software to detect price breakout points.The IntuitionThe brute force is obvious — for every element, scan everything to its right and find the first greater one. That works but it is O(n²). For an array of 10⁶ elements that becomes 10¹² operations. It will time out on any large input.The stack insight is this — traverse right to left. As you move left, the stack always holds elements you have already seen on the right side. These are the candidates for being the next greater element. Before pushing the current element, pop all stack elements that are smaller than or equal to it. Why? Because the current element is blocking them — for any future element to the left, the current element will always be encountered first, so those smaller popped elements can never be an answer for anything. Whatever remains on top of the stack after popping is the answer for the current element.Step-by-Step Dry RunArray: [1, 3, 2, 4], traversing right to left.i=3, element is 4. Stack is empty. Answer for index 3 is -1. Push 4. Stack: [4]i=2, element is 2. Top of stack is 4, which is greater than 2. Answer for index 2 is 4. Push 2. Stack: [4, 2]i=1, element is 3. Top of stack is 2, which is not greater than 3. Pop 2. Top is now 4, which is greater than 3. Answer for index 1 is 4. Push 3. Stack: [4, 3]i=0, element is 1. Top of stack is 3, which is greater than 1. Answer for index 0 is 3. Push 1. Stack: [4, 3, 1]Answers collected right to left: [-1, 4, 4, 3] After Collections.reverse(): [3, 4, 4, -1] ✓The Code// NGR — Next Greater Element to Rightclass Solution {public ArrayList<Integer> nextLargerElement(int[] arr) {ArrayList<Integer> result = new ArrayList<>();Stack<Integer> st = new Stack<>();// Traverse from RIGHT to LEFTfor (int i = arr.length - 1; i >= 0; i--) {// Pop all elements smaller than or equal to current// They can never be the answer for any element to the leftwhile (!st.empty() && arr[i] >= st.peek()) {st.pop();}// Whatever is on top now is the next greater elementif (st.empty()) {result.add(-1);} else {result.add(st.peek());}// Push current — it is a candidate for elements to the leftst.push(arr[i]);}// Collected answers right to left, so reverse before returningCollections.reverse(result);return result;}}Edge CasesAll elements decreasing — Input: [5, 4, 3, 2, 1] Output: [-1, -1, -1, -1, -1] Every element has no greater element to its right. Traversing right to left, each new element is larger than everything already in the stack, so the stack gets cleared and the answer is always -1.All elements increasing — Input: [1, 2, 3, 4, 5] Output: [2, 3, 4, 5, -1] Each element's next greater is simply the next element in the array. The last element always gets -1 since nothing exists to its right.All elements equal — Input: [3, 3, 3, 3] Output: [-1, -1, -1, -1] Equal elements do not count as greater. The pop condition uses >= so equals get removed from the stack, ensuring duplicates never answer each other.Single element — Input: [7] Output: [-1] Nothing to the right, always -1.Why only 32.95% accuracy on GFG? Most people either forget to reverse the result at the end, use the wrong comparison in the while loop, or submit a brute force O(n²) solution that times out on large inputs.Problem 2 — Next Greater Element to Left / Previous Greater Element (NGL)GFG Problem: Search "Previous Greater Element" on GeeksForGeeks Difficulty: Medium | Accuracy: 68.93% | Submissions: 7K+The QuestionFor each element in the array, find the first element to its left that is strictly greater than it. If none exists, return -1.Input: [10, 4, 2, 20, 40, 12, 30] Output: [-1, 10, 4, -1, -1, 40, 40]Real World ExampleImagine you are a junior employee at a company. For each person in the office, you want to know — who is the first senior person sitting to their left who earns more? This is NGL. It is used in organizational hierarchy systems, salary band analysis tools, and even in database query optimizers to find the nearest dominant record on the left side.The IntuitionThis is the mirror image of NGR. Instead of traversing right to left, we traverse left to right. The stack holds elements we have already seen from the left side — these are candidates for being the previous greater element. For each new element, pop everything from the stack that is smaller than or equal to it. Whatever remains on top is the first greater element to its left. Then push the current element for future use.No reverse is needed here because we are already going left to right and building the result in order.Step-by-Step Dry RunArray: [10, 4, 2, 20, 40, 12, 30], traversing left to right.i=0, element is 10. Stack is empty. Answer is -1. Push 10. Stack: [10]i=1, element is 4. Top is 10, greater than 4. Answer is 10. Push 4. Stack: [10, 4]i=2, element is 2. Top is 4, greater than 2. Answer is 4. Push 2. Stack: [10, 4, 2]i=3, element is 20. Top is 2, not greater than 20. Pop 2. Top is 4, not greater. Pop 4. Top is 10, not greater. Pop 10. Stack is empty. Answer is -1. Push 20. Stack: [20]i=4, element is 40. Top is 20, not greater. Pop 20. Stack empty. Answer is -1. Push 40. Stack: [40]i=5, element is 12. Top is 40, greater than 12. Answer is 40. Push 12. Stack: [40, 12]i=6, element is 30. Top is 12, not greater than 30. Pop 12. Top is 40, greater than 30. Answer is 40. Push 30. Stack: [40, 30]Result: [-1, 10, 4, -1, -1, 40, 40] ✓ No reverse needed.The Code// NGL — Next Greater Element to Left (Previous Greater Element)class Solution {static ArrayList<Integer> preGreaterEle(int[] arr) {Stack<Integer> st = new Stack<>();ArrayList<Integer> result = new ArrayList<>();// Traverse LEFT to RIGHT — no reverse neededfor (int i = 0; i <= arr.length - 1; i++) {// Pop all elements smaller than or equal to currentwhile (!st.empty() && arr[i] >= st.peek()) {st.pop();}// Top of stack is the previous greater elementif (!st.empty() && st.peek() > arr[i]) {result.add(st.peek());} else {result.add(-1);}// Push current for future elementsst.push(arr[i]);}return result;}}Edge CasesStrictly increasing array — Input: [10, 20, 30, 40] Output: [-1, -1, -1, -1] Each new element is larger than everything before it, so the stack always gets fully cleared. No previous greater exists for any element.First element is always -1 — regardless of its value, the first element has nothing to its left. The stack is empty at i=0, so the answer is always -1 for index 0. This is guaranteed by the logic.Duplicate values — Input: [5, 5, 5] Output: [-1, -1, -1] Equal elements do not qualify as greater. The pop condition uses >= so duplicates get removed from the stack and never answer each other.Problem 3 — Next Smaller Element to Right (NSR)GFG Problem: Search "Next Smaller Element" on GeeksForGeeks Difficulty: Medium | Accuracy: 36.26% | Submissions: 225K+The QuestionFor each element in the array, find the first element to its right that is strictly smaller than it. If none exists, return -1.Input: [4, 8, 5, 2, 25] Output: [2, 5, 2, -1, -1]Input: [13, 7, 6, 12] Output: [7, 6, -1, -1]Real World ExampleYou work at a warehouse. Shelves have items of weights: [4, 8, 5, 2, 25] kg. For each item, the system needs to find the first lighter item sitting to its right on the shelf — this is used to optimize load balancing and shelf arrangement algorithms. Item of 4 kg — first lighter to the right is 2 kg. Item of 8 kg — first lighter is 5 kg. Item of 5 kg — first lighter is 2 kg. Items of 2 kg and 25 kg have no lighter item to their right, so -1.The IntuitionNSR is structurally identical to NGR — we traverse right to left and collect answers, then reverse. The only change is the pop condition. In NGR we popped elements smaller than or equal to current because we wanted greater. Here we want smaller, so we pop elements greater than or equal to current. After popping, whatever remains on top is the first smaller element to the right.Step-by-Step Dry RunArray: [4, 8, 5, 2, 25], traversing right to left.i=4, element is 25. Stack is empty. Answer is -1. Push 25. Stack: [25]i=3, element is 2. Top is 25, which is greater than or equal to 2. Pop 25. Stack is empty. Answer is -1. Push 2. Stack: [2]i=2, element is 5. Top is 2, which is less than 5. Answer is 2. Push 5. Stack: [2, 5]i=1, element is 8. Top is 5, which is less than 8. Answer is 5. Push 8. Stack: [2, 5, 8]i=0, element is 4. Top is 8, which is greater than or equal to 4. Pop 8. Top is 5, which is greater than or equal to 4. Pop 5. Top is 2, which is less than 4. Answer is 2. Push 4. Stack: [2, 4]Answers collected right to left: [-1, -1, 2, 5, 2] After Collections.reverse(): [2, 5, 2, -1, -1] ✓The Code// NSR — Next Smaller Element to Rightclass Solution {static ArrayList<Integer> nextSmallerEle(int[] arr) {Stack<Integer> st = new Stack<>();ArrayList<Integer> result = new ArrayList<>();// Traverse RIGHT to LEFTfor (int i = arr.length - 1; i >= 0; i--) {// Pop elements greater than or equal to current// Opposite of NGR — we want smaller, so clear the bigger oneswhile (!st.empty() && arr[i] <= st.peek()) {st.pop();}// Top is now the next smaller elementif (!st.empty() && st.peek() < arr[i]) {result.add(st.peek());} else {result.add(-1);}st.push(arr[i]);}Collections.reverse(result);return result;}}Edge CasesStrictly decreasing array — Input: [5, 4, 3, 2, 1] Output: [4, 3, 2, 1, -1] Each element's next smaller is simply the next element in the array. Last element is always -1.Strictly increasing array — Input: [1, 2, 3, 4, 5] Output: [-1, -1, -1, -1, -1] No element has a smaller element to its right since the array only grows.Last element is always -1 — nothing exists to its right regardless of its value.Single element — Input: [42] Output: [-1]Why 36.26% accuracy on GFG? The most common mistake is keeping the NGR pop condition (arr[i] >= st.peek()) and only changing the problem description in your head. The pop condition must flip to arr[i] <= st.peek() for NSR. Forgetting this gives completely wrong answers that look plausible, which makes the bug hard to spot.Problem 4 — Next Smaller Element to Left / Previous Smaller Element (NSL)GFG Problem: Search "Previous Smaller Element" on GeeksForGeeksThe QuestionFor each element in the array, find the first element to its left that is strictly smaller than it. If none exists, return -1.Input: [4, 8, 5, 2, 25] Output: [-1, 4, 4, -1, 2]Real World ExampleSame warehouse. Now the system looks left instead of right. For the item weighing 8 kg, the first lighter item to its left is 4 kg. For 25 kg, the first lighter to its left is 2 kg. For 4 kg, nothing lighter exists to its left so -1. For 2 kg, nothing lighter to its left so -1. This kind of lookback query appears in time-series analysis, price history tracking, and sensor data processing.The IntuitionNSL is the mirror of NSR, exactly as NGL was the mirror of NGR. We traverse left to right (no reverse needed). We maintain a stack of candidates from the left. For each element, pop all elements greater than or equal to it — they cannot be the answer since they are not smaller. Whatever remains on top is the first smaller element to the left. Push current and move on.Step-by-Step Dry RunArray: [4, 8, 5, 2, 25], traversing left to right.i=0, element is 4. Stack is empty. Answer is -1. Push 4. Stack: [4]i=1, element is 8. Top is 4, which is less than 8. Answer is 4. Push 8. Stack: [4, 8]i=2, element is 5. Top is 8, which is greater than or equal to 5. Pop 8. Top is 4, which is less than 5. Answer is 4. Push 5. Stack: [4, 5]i=3, element is 2. Top is 5, greater than or equal to 2. Pop 5. Top is 4, greater than or equal to 2. Pop 4. Stack is empty. Answer is -1. Push 2. Stack: [2]i=4, element is 25. Top is 2, which is less than 25. Answer is 2. Push 25. Stack: [2, 25]Result: [-1, 4, 4, -1, 2] ✓ No reverse needed.The Code// NSL — Next Smaller Element to Left (Previous Smaller Element)class Solution {static ArrayList<Integer> prevSmallerEle(int[] arr) {Stack<Integer> st = new Stack<>();ArrayList<Integer> result = new ArrayList<>();// Traverse LEFT to RIGHT — no reverse neededfor (int i = 0; i < arr.length; i++) {// Pop elements greater than or equal to currentwhile (!st.empty() && arr[i] <= st.peek()) {st.pop();}// Top is the previous smaller elementif (!st.empty() && st.peek() < arr[i]) {result.add(st.peek());} else {result.add(-1);}st.push(arr[i]);}return result;}}Edge CasesFirst element is always -1 — nothing exists to its left. Stack is empty at i=0 every time.All same elements — Input: [5, 5, 5, 5] Output: [-1, -1, -1, -1] Equal elements do not qualify as smaller. The condition arr[i] <= st.peek() ensures equals are popped and never answer each other.Single element — Input: [9] Output: [-1]The Master Cheat SheetThis is the one table to save and refer to whenever you encounter any of these four problems.VariantTraverse DirectionPop ConditionReverse Result?NGR — Next Greater RightRight to Leftarr[i] >= st.peek()YesNGL — Next Greater LeftLeft to Rightarr[i] >= st.peek()NoNSR — Next Smaller RightRight to Leftarr[i] <= st.peek()YesNSL — Next Smaller LeftLeft to Rightarr[i] <= st.peek()NoTwo rules to remember forever:Rule 1 — Direction. If you are looking to the right, traverse right to left and reverse at the end. If you are looking to the left, traverse left to right and no reverse is needed.Rule 2 — Pop Condition. If you want a greater element, pop when arr[i] >= st.peek() to clear out smaller useless candidates. If you want a smaller element, pop when arr[i] <= st.peek() to clear out bigger useless candidates.Mix these two rules and you derive all four variants instantly without memorizing anything separately.Common Mistakes to AvoidWrong pop condition — Using > instead of >= in the while loop. This causes duplicate values to wrongly answer each other. Always use >= for greater problems and <= for smaller problems inside the while loop.Forgetting to reverse — For right-to-left traversals (NGR and NSR), you collect answers from right to left. You must call Collections.reverse() before returning. Skipping this is the single most common reason for wrong answers on these problems.Not checking empty stack before peek — Always check !st.empty() before calling st.peek(). An empty stack peek throws EmptyStackException at runtime and will crash your solution.Wrong if-condition after the while loop — After the while loop, the if-condition must use strict comparison. For NGR use st.peek() > arr[i]. For NSR use st.peek() < arr[i]. These must be strict — no equals sign here.Confusing traversal direction with answer direction — You traverse right to left for NGR but the answer array is filled left to right. The reverse at the end handles this. Do not try to index directly into the result array to compensate — just use reverse.Time and Space ComplexityAll four problems run in O(n) time and use O(n) space.Even though there is a while loop nested inside the for loop, each element is pushed into the stack exactly once and popped from the stack at most once. So across the entire traversal, the total number of push and pop operations combined is at most 2n — which gives O(n) overall. This is the beauty of the monotonic stack.Why These Four Problems Matter Beyond GFGThese four patterns are not just textbook exercises. They appear as the hidden sub-problem inside some of the hardest stack questions:-Largest Rectangle in Histogram uses NSR and NSL to find the left and right boundaries of each bar.Trapping Rain Water uses NGR and NGL to determine the water level above each position.Stock Span Problem is literally NGL applied directly to stock prices.Sum of Subarray Minimums uses NSR and NSL together to count contributions of each element.Once you master these four patterns deeply, a whole family of hard problems that previously seemed unapproachable suddenly becomes a matter of recognizing the pattern and applying it.Also on This BlogIf you are building your stack foundation from scratch, check out the complete deep-dive here → Stack Data Structure in Java: The Complete Guide — covering everything from what a stack is, LIFO principle, all three implementations, every operation with code, and six practice problems.

MonotonicStackNextGreaterElementStackProblemsJavaGeeksForGeeksStackPattern

What Is Dynamic Programming? Origin Story, Real-Life Uses, LeetCode Problems & Complete Beginner Guide

Introduction — Why Dynamic Programming Feels Hard (And Why It Isn't)If you've ever stared at a LeetCode problem, read the solution, understood every single line, and still had absolutely no idea how someone arrived at it — welcome. You've just experienced the classic Dynamic Programming (DP) confusion.DP has a reputation. People treat it like some dark art reserved for competitive programmers or Google engineers. The truth? Dynamic Programming is one of the most logical, learnable, and satisfying techniques in all of computer science. Once it clicks, it really clicks.This guide will take you from zero to genuinely confident. We'll cover where DP came from, how it works, what patterns to learn, how to recognize DP problems, real-world places it shows up, LeetCode problems to practice, time complexity analysis, and the mistakes that trip up even experienced developers.Let's go.The Origin Story — Who Invented Dynamic Programming and Why?The term "Dynamic Programming" was coined by Richard Bellman in the early 1950s while working at RAND Corporation. Here's the funny part: the name was deliberately chosen to sound impressive and vague.Bellman was doing mathematical research that his employer — the US Secretary of Defense, Charles Wilson — would have found difficult to fund if described accurately. Wilson had a well-known distaste for the word "research." So Bellman invented a name that sounded suitably grand and mathematical: Dynamic Programming.In his autobiography, Bellman wrote that he picked the word "dynamic" because it had a precise technical meaning and was also impossible to use negatively. "Programming" referred to the mathematical sense — planning and decision-making — not computer programming.The underlying idea? Break a complex problem into overlapping subproblems, solve each subproblem once, and store the result so you never solve it twice.Bellman's foundational contribution was the Bellman Equation, which underpins not just algorithms but also economics, operations research, and modern reinforcement learning.So the next time DP feels frustrating, remember — even its inventor named it specifically to confuse people. You're in good company.What Is Dynamic Programming? (Simple Definition)Dynamic Programming is an algorithmic technique used to solve problems by:Breaking them down into smaller overlapping subproblemsSolving each subproblem only onceStoring the result (memoization or tabulation)Building up the final solution from those stored resultsThe key insight is overlapping subproblems + optimal substructure.Overlapping subproblems means the same smaller problems come up again and again. Instead of solving them every time (like plain recursion does), DP solves them once and caches the answer.Optimal substructure means the optimal solution to the whole problem can be built from optimal solutions to its subproblems.If a problem has both these properties — it's a DP problem.The Two Approaches to Dynamic Programming1. Top-Down with Memoization (Recursive + Cache)You write a recursive solution exactly as you would naturally, but add a cache (usually a dictionary or array) to store results you've already computed.fib(n):if n in cache: return cache[n]if n <= 1: return ncache[n] = fib(n-1) + fib(n-2)return cache[n]This is called memoization — remember what you computed so you don't repeat yourself.Pros: Natural to write, mirrors the recursive thinking, easy to reason about. Cons: Stack overhead from recursion, risk of stack overflow on large inputs.2. Bottom-Up with Tabulation (Iterative)You figure out the order in which subproblems need to be solved, then solve them iteratively from the smallest up, filling a table.fib(n):dp = [0, 1]for i from 2 to n:dp[i] = dp[i-1] + dp[i-2]return dp[n]This is called tabulation — fill a table, cell by cell, bottom to top.Pros: No recursion overhead, usually faster in practice, easier to optimize space. Cons: Requires thinking about the order of computation upfront.🧩 Dynamic Programming Template CodeBefore diving into how to recognize DP problems, here are ready-to-use Java templates for every major DP pattern. Think of these as your reusable blueprints — every DP problem you ever solve will fit into one of these structures. Just define your state, plug in your recurrence relation, and you are good to go.Template 1 — Top-Down (Memoization)import java.util.HashMap;import java.util.Map;public class TopDownDP {Map<Integer, Integer> memo = new HashMap<>();public int solve(int n) {// Base caseif (n <= 1) return n;// Check cacheif (memo.containsKey(n)) return memo.get(n);// Recurrence relation — change this part for your problemint result = solve(n - 1) + solve(n - 2);// Store in cachememo.put(n, result);return result;}}Template 2 — Bottom-Up (Tabulation)public class BottomUpDP {public int solve(int n) {// Create DP tableint[] dp = new int[n + 1];// Base casesdp[0] = 0;dp[1] = 1;// Fill the table bottom-upfor (int i = 2; i <= n; i++) {// Recurrence relation — change this part for your problemdp[i] = dp[i - 1] + dp[i - 2];}return dp[n];}}Template 3 — Bottom-Up with Space Optimizationpublic class SpaceOptimizedDP {public int solve(int n) {// Only keep last two values instead of full tableint prev2 = 0;int prev1 = 1;for (int i = 2; i <= n; i++) {// Recurrence relation — change this part for your problemint curr = prev1 + prev2;prev2 = prev1;prev1 = curr;}return prev1;}}Template 4 — 2D DP (Two Sequences or Grid)public class TwoDimensionalDP {public int solve(String s1, String s2) {int m = s1.length();int n = s2.length();// Create 2D DP tableint[][] dp = new int[m + 1][n + 1];// Base cases — first row and columnfor (int i = 0; i <= m; i++) dp[i][0] = i;for (int j = 0; j <= n; j++) dp[0][j] = j;// Fill table cell by cellfor (int i = 1; i <= m; i++) {for (int j = 1; j <= n; j++) {// Recurrence relation — change this part for your problemif (s1.charAt(i - 1) == s2.charAt(j - 1)) {dp[i][j] = dp[i - 1][j - 1];} else {dp[i][j] = 1 + Math.min(dp[i - 1][j],Math.min(dp[i][j - 1], dp[i - 1][j - 1]));}}}return dp[m][n];}}Template 5 — Knapsack Patternpublic class KnapsackDP {public int solve(int[] weights, int[] values, int capacity) {int n = weights.length;// dp[i][w] = max value using first i items with capacity wint[][] dp = new int[n + 1][capacity + 1];for (int i = 1; i <= n; i++) {for (int w = 0; w <= capacity; w++) {// Don't take item idp[i][w] = dp[i - 1][w];// Take item i if it fitsif (weights[i - 1] <= w) {dp[i][w] = Math.max(dp[i][w],values[i - 1] + dp[i - 1][w - weights[i - 1]]);}}}return dp[n][capacity];}}💡 How to use these templates:Step 1 — Identify which pattern your problem fits into. Step 2 — Define what dp[i] or dp[i][j] means in plain English before writing any code. Step 3 — Write your recurrence relation on paper first. Step 4 — Plug it into the matching template above. Step 5 — Handle your specific base cases carefully.🎥 Visual Learning Resource — Watch This Before Moving ForwardIf you prefer learning by watching before reading, this free full-length course by freeCodeCamp is one of the best Dynamic Programming resources on the internet. Watch it alongside this guide for maximum understanding.Credit: freeCodeCamp — a free, nonprofit coding education platform.How to Recognize a Dynamic Programming ProblemAsk yourself these four questions:1. Can I define the problem in terms of smaller versions of itself? If you can write a recursive formula (recurrence relation), DP might apply.2. Do the subproblems overlap? If a naive recursive solution would recompute the same thing many times, DP is the right tool.3. Is there an optimal substructure? Is the best answer to the big problem made up of best answers to smaller problems?4. Are you looking for a count, minimum, maximum, or yes/no answer? DP problems often ask: "What is the minimum cost?", "How many ways?", "Can we achieve X?"Red flag words in problem statements: minimum, maximum, shortest, longest, count the number of ways, can we reach, is it possible, fewest steps.The Core DP Patterns You Must LearnMastering DP is really about recognizing patterns. Here are the most important ones:Pattern 1 — 1D DP (Linear) Problems where the state depends on previous elements in a single sequence. Examples: Fibonacci, Climbing Stairs, House Robber.Pattern 2 — 2D DP (Grid / Two-sequence) Problems with two dimensions of state, often grids or two strings. Examples: Longest Common Subsequence, Edit Distance, Unique Paths.Pattern 3 — Interval DP You consider all possible intervals or subarrays and build solutions from them. Examples: Matrix Chain Multiplication, Burst Balloons, Palindrome Partitioning.Pattern 4 — Knapsack DP (0/1 and Unbounded) You decide whether to include or exclude items under a capacity constraint. Examples: 0/1 Knapsack, Coin Change, Partition Equal Subset Sum.Pattern 5 — DP on Trees State is defined per node; you combine results from children. Examples: Diameter of Binary Tree, House Robber III, Maximum Path Sum.Pattern 6 — DP on Subsets / Bitmask DP State includes a bitmask representing which elements have been chosen. Examples: Travelling Salesman Problem, Shortest Superstring.Pattern 7 — DP on Strings Matching, editing, or counting arrangements within strings. Examples: Longest Palindromic Subsequence, Regular Expression Matching, Wildcard Matching.Top LeetCode Problems to Practice Dynamic Programming (With Links)Here are the essential problems, organized by difficulty and pattern. Solve them in this order.Beginner — Warm UpProblemPatternLinkClimbing Stairs1D DPhttps://leetcode.com/problems/climbing-stairs/Fibonacci Number1D DPhttps://leetcode.com/problems/fibonacci-number/House Robber1D DPhttps://leetcode.com/problems/house-robber/Min Cost Climbing Stairs1D DPhttps://leetcode.com/problems/min-cost-climbing-stairs/Best Time to Buy and Sell Stock1D DPhttps://leetcode.com/problems/best-time-to-buy-and-sell-stock/Intermediate — Core PatternsProblemPatternLinkCoin ChangeKnapsackhttps://leetcode.com/problems/coin-change/Longest Increasing Subsequence1D DPhttps://leetcode.com/problems/longest-increasing-subsequence/Longest Common Subsequence2D DPhttps://leetcode.com/problems/longest-common-subsequence/0/1 Knapsack (via Subset Sum)Knapsackhttps://leetcode.com/problems/partition-equal-subset-sum/Unique Paths2D Grid DPhttps://leetcode.com/problems/unique-paths/Jump Game1D DP / Greedyhttps://leetcode.com/problems/jump-game/Word BreakString DPhttps://leetcode.com/problems/word-break/Decode Ways1D DPhttps://leetcode.com/problems/decode-ways/Edit Distance2D String DPhttps://leetcode.com/problems/edit-distance/Triangle2D DPhttps://leetcode.com/problems/triangle/Advanced — Interview LevelProblemPatternLinkBurst BalloonsInterval DPhttps://leetcode.com/problems/burst-balloons/Regular Expression MatchingString DPhttps://leetcode.com/problems/regular-expression-matching/Wildcard MatchingString DPhttps://leetcode.com/problems/wildcard-matching/Palindrome Partitioning IIInterval DPhttps://leetcode.com/problems/palindrome-partitioning-ii/Maximum Profit in Job SchedulingDP + Binary Searchhttps://leetcode.com/problems/maximum-profit-in-job-scheduling/Distinct Subsequences2D DPhttps://leetcode.com/problems/distinct-subsequences/Cherry Pickup3D DPhttps://leetcode.com/problems/cherry-pickup/Real-World Use Cases of Dynamic ProgrammingDP is not just for coding interviews. It is deeply embedded in the technology you use every day.1. Google Maps & Navigation (Shortest Path) The routing engines behind GPS apps use DP-based algorithms like Dijkstra and Bellman-Ford to find the shortest or fastest path between two points across millions of nodes.2. Spell Checkers & Autocorrect (Edit Distance) When your phone corrects "teh" to "the," it is computing Edit Distance — a classic DP problem — between what you typed and every word in the dictionary.3. DNA Sequence Alignment (Bioinformatics) Researchers use the Needleman-Wunsch and Smith-Waterman algorithms — both DP — to align DNA and protein sequences and find similarities between species or identify mutations.4. Video Compression (MPEG, H.264) Modern video codecs use DP to determine the most efficient way to encode video frames, deciding which frames to store as full images and which to store as differences from the previous frame.5. Financial Portfolio Optimization Investment algorithms use DP to find the optimal allocation of assets under risk constraints — essentially a variant of the knapsack problem.6. Natural Language Processing (NLP) The Viterbi algorithm — used in speech recognition, part-of-speech tagging, and machine translation — is a DP algorithm. Every time Siri or Google Assistant understands your sentence, DP played a role.7. Game AI (Chess, Checkers) Game trees and minimax algorithms with memoization use DP to evaluate board positions and find the best move without recomputing already-seen positions.8. Compiler Optimization Compilers use DP to decide the optimal order of operations and instruction scheduling to generate the most efficient machine code.9. Text Justification (Word Processors) Microsoft Word and LaTeX use DP to optimally break paragraphs into lines — minimizing raggedness and maximizing visual appeal.10. Resource Scheduling in Cloud Computing AWS, Google Cloud, and Azure use DP-based scheduling to assign computational tasks to servers in the most cost-efficient way possible.Time Complexity Analysis of Common DP ProblemsUnderstanding the time complexity of DP is critical for interviews and for building scalable systems.ProblemTime ComplexitySpace ComplexityNotesFibonacci (naive recursion)O(2ⁿ)O(n)Exponential — terribleFibonacci (DP)O(n)O(1) with optimizationLinear — excellentLongest Common SubsequenceO(m × n)O(m × n)m, n = lengths of two stringsEdit DistanceO(m × n)O(m × n)Can optimize space to O(n)0/1 KnapsackO(n × W)O(n × W)n = items, W = capacityCoin ChangeO(n × amount)O(amount)Classic tabulationLongest Increasing SubsequenceO(n²) or O(n log n)O(n)Binary search version is fasterMatrix Chain MultiplicationO(n³)O(n²)Interval DPTravelling Salesman (bitmask)O(2ⁿ × n²)O(2ⁿ × n)Still exponential but manageable for small nThe general rule: DP trades time for space. You use memory to avoid recomputation. The time complexity equals the number of unique states multiplied by the work done per state.How to Learn and Master Dynamic Programming — Step by StepHere is an honest, structured path to mastery:Step 1 — Get recursion absolutely solid first. DP is memoized recursion at its core. If you cannot write clean recursive solutions confidently, DP will remain confusing. Practice at least 20 pure recursion problems first.Step 2 — Start with the classics. Fibonacci → Climbing Stairs → House Robber → Coin Change. These teach you the core pattern of defining state and transition without overwhelming you.Step 3 — Learn to define state explicitly. Before writing any code, ask: "What does dp[i] represent?" Write it in plain English. "dp[i] = the minimum cost to reach step i." This single habit separates good DP thinkers from struggling ones.Step 4 — Write the recurrence relation before coding. On paper or in a comment. Example: dp[i] = min(dp[i-1] + cost[i-1], dp[i-2] + cost[i-2]). If you can write the recurrence, the code writes itself.Step 5 — Master one pattern at a time. Don't jump between knapsack and interval DP in the same week. Spend a few days on each pattern until it feels intuitive.Step 6 — Solve the same problem both ways. Top-down and bottom-up. This builds deep understanding of what DP is actually doing.Step 7 — Optimize space after getting correctness. Many 2D DP solutions can use a single row instead of a full matrix. Learn this optimization after you understand the full solution.Step 8 — Do timed practice under interview conditions. Give yourself 35 minutes per problem. Review what you got wrong. DP is a muscle — it builds with reps.Common Mistakes in Dynamic Programming (And How to Avoid Them)Mistake 1 — Jumping to code before defining state. The most common DP error. Always define what dp[i] or dp[i][j] means before writing a single line of code.Mistake 2 — Wrong base cases. A single wrong base case corrupts every answer built on top of it. Trace through your base cases manually on a tiny example before running code.Mistake 3 — Off-by-one errors in indexing. Whether your dp array is 0-indexed or 1-indexed must be 100% consistent throughout. This causes more bugs in DP than almost anything else.Mistake 4 — Confusing top-down with bottom-up state order. In bottom-up DP, you must ensure that when you compute dp[i], all values it depends on are already filled. If you compute in the wrong order, you get garbage answers.Mistake 5 — Memoizing in the wrong dimension. In 2D problems, some people cache only one dimension when the state actually requires two. Always identify all variables that affect the outcome.Mistake 6 — Using global mutable state in recursion. If you use a shared array and don't clear it between test cases, you'll get wrong answers on subsequent inputs. Always scope your cache correctly.Mistake 7 — Not considering the full state space. In problems like Knapsack, forgetting that the state is (item index, remaining capacity) — not just item index — leads to fundamentally wrong solutions.Mistake 8 — Giving up after not recognizing the pattern immediately. DP problems don't announce themselves. The skill is learning to ask "is there overlapping subproblems here?" on every problem. This takes time. Don't mistake unfamiliarity for inability.Frequently Asked Questions About Dynamic ProgrammingQ: Is Dynamic Programming the same as recursion? Not exactly. Recursion is a technique for breaking problems into smaller pieces. DP is recursion plus memoization — or iterative tabulation. All DP can be written recursively, but not all recursion is DP.Q: What is the difference between DP and Divide and Conquer? Divide and Conquer (like Merge Sort) breaks problems into non-overlapping subproblems. DP is used when subproblems overlap — meaning the same subproblem is solved multiple times in a naive approach.Q: How do I know when NOT to use DP? If the subproblems don't overlap (no repeated computation), greedy or divide-and-conquer may be better. If the problem has no optimal substructure, DP won't give a correct answer.Q: Do I need to memorize DP solutions for interviews? No. You need to recognize patterns and be able to derive the recurrence relation. Memorizing solutions without understanding them will fail you in interviews. Focus on the thinking process.Q: How long does it take to get good at DP? Most people start to feel genuinely comfortable after solving 40–60 varied DP problems with deliberate practice. The first 10 feel impossible. The next 20 feel hard. After 50, patterns start feeling obvious.Q: What programming language is best for DP? Any language works. Python is often used for learning because its dictionaries make memoization trivial. C++ is preferred in competitive programming for its speed. For interviews, use whatever language you're most comfortable in.Q: What is space optimization in DP? Many DP problems only look back one or two rows to compute the current row. In those cases, you can replace an n×m table with just two arrays (or even one), reducing space complexity from O(n×m) to O(m). This is called space optimization or rolling array technique.Q: Can DP be applied to graph problems? Absolutely. Shortest path algorithms like Bellman-Ford are DP. Longest path in a DAG is DP. DP on trees is a rich subfield. Anywhere you have states and transitions, DP can potentially apply.Q: Is Greedy a type of Dynamic Programming? Greedy is related but distinct. Greedy makes locally optimal choices without reconsidering. DP considers all choices and picks the globally optimal one. Some DP solutions reduce to greedy when the structure allows, but they are different techniques.Q: What resources should I use to learn DP? For structured learning: Neetcode.io (organized problem list), Striver's DP Series on YouTube, and the book "Introduction to Algorithms" (CLRS) for theoretical depth. For practice: LeetCode's Dynamic Programming study plan and Codeforces for competitive DP.Final Thoughts — Dynamic Programming Is a SuperpowerDynamic Programming is genuinely one of the most powerful ideas in computer science. It shows up in your GPS, your autocorrect, your streaming video, your bank's risk models, and the AI assistants you talk to daily.The path to mastering it is not memorization. It is developing the habit of asking: can I break this into smaller problems that overlap? And then learning to define state clearly, write the recurrence, and trust the process.Start with Climbing Stairs. Write dp[i] in plain English before every problem. Solve everything twice — top-down and bottom-up. Do 50 problems with genuine reflection, not just accepted solutions.The click moment will come. And when it does, you'll wonder why it ever felt hard.

Dynamic ProgrammingMemoizationTabulationJavaOrigin StoryRichard Bellman