Optimization Interview Questions

Optimization questions in quant interviews come in five recurring flavors: first-order-condition calculus with a convexity check, bet sizing and portfolio construction, pricing and expected-value trade-offs, dynamic programming over sequential decisions, and numerical methods like Newton's method and gradient descent.
The common thread: write down the objective, price the constraints, and let the structure of the problem tell you where the optimum must sit.

Last updated 2 July 2026 · sub-areas, difficulty mix and firm attributions on this page are compiled directly from the 119 optimization problems in the QuantVault bank; firm tags are candidate-reported, not employer-verified, and no individual author is named in our source data.

Core sub-areas
Convexity & first-order conditions · bet sizing & portfolios · pricing trade-offs · dynamic programming · numerical methods
Typical difficulty
Skews medium-to-hard — 6 easy, 70 medium and 43 hard across the 119-problem set
Who leans on it
Quant research and trading loops above all — problems here carry candidate tags from Citadel, Two Sigma, Jane Street, WorldQuant, Renaissance and others
Practice pool
119 problems in the bank · 14 free to open with the full worked solution

Where optimization shows up: mostly in quant-research and quant-trading technical rounds, where “how much would you bet?” or “what price maximizes your edge?” turns a calculus exercise into a desk decision — with a smaller software-track slice built on numerical methods. For how it sits inside specific funnels, see the Citadel interview questions guide and the Two Sigma interview questions guide.

Practice optimization questions interactively → Browse the full problem bank All interview topics

The surfaceThe optimization sub-areas quant interviews test

Five families cover essentially every optimization question in the bank. Each row pairs a sub-area with the recurring question shape and a representative type — a flavor drawn from real problems, not a leaked wording — so you can see where the difficulty actually lives.

The optimization sub-areas tested in quant interviews, with the recurring question shape and a representative question type for each.
Sub-areaRecurring shapeRepresentative type (flavor, not a real question)
Convexity & first-order conditionsDifferentiate, set to zero, then justify that the critical point is the global optimumMinimize an unusual one-variable function, show the mean minimizes squared loss, or maximize a quadratic form on the unit sphere.
Bet sizing & portfolio constructionMaximize expected log growth or a mean–variance objective over position sizesA Kelly-style sizing question for a signal with a known edge, where the answer is a fraction of bankroll and the trap is overbetting.
Pricing & expected-value trade-offsWrite revenue or fill probability as a function of your choice, then balance the marginal gain against the marginal lossChoose a tender price, a tick improvement or an inventory level where a better price wins the trade less often but earns more when it does.
Dynamic programming & sequential decisionsDefine the state, write the Bellman recursion, solve backwardA scheduling or crossing puzzle with a non-obvious greedy failure, a bandit-style learning question, or an MDP formulation for a trading policy.
Numerical methods & regularizationDerive the update rule, argue convergence, and name what breaks itImplement or analyze Newton's method or gradient descent, or explain why an L0 penalty is computationally and statistically awkward.

What's confirmed vs. what varies: the sub-areas and the difficulty mix above come straight from the problem set, so they are stable. Which firm asks which flavor is candidate-reported through problem tags — treat the firm attributions as directional, and expect research tracks to weight the convexity and portfolio families while software tracks weight the numerical-methods family.

The patternsSignature optimization question patterns

Three moves generate most correct answers in this topic. Each worked box below is a 60–90 second micro-example in the interview's actual cadence — the reasoning template, applied to a generic setup rather than any firm's wording.

First-order conditions — transform, differentiate, verify

Takeaway: before differentiating, transform the objective into something friendly — usually a log — and always close with the one-line argument for why your critical point is the global optimum.

Shape. Minimize \(f(x) = x^x\) on \((0,1]\). Where is the minimum, and how do you know it is one?

1. Take logs. Let \(g(x) = \ln f(x) = x \ln x\). Since \(\ln\) is increasing, minimizing \(g\) minimizes \(f\) — and \(g\) is far easier to differentiate.

2. First-order condition. \(g'(x) = \ln x + 1 = 0\) gives \(x = 1/e\). No boundary issue: \(g \to 0\) as \(x \to 0^+\) and \(g(1)=0\), both above the interior value.

3. Verify globally. \(g'(x) < 0\) for \(x < 1/e\) and \(g'(x) > 0\) after, so the function falls then rises — a single global minimum at \(x = 1/e\), with value \(e^{-1/e} \approx 0.69\). The move: log-transform, solve the FOC, then say out loud why it's global.

Constrained maximization — multipliers turn constraints into structure

Takeaway: a Lagrange multiplier is not bookkeeping — the stationarity condition often identifies the answer, turning a constrained search into an eigenvalue or marginal-value statement.

Shape. Maximize \(x^{\mathsf T} A x\) over unit vectors \(\lVert x \rVert = 1\), for a symmetric matrix \(A\).

1. Set up the Lagrangian. \(\mathcal{L} = x^{\mathsf T} A x - \lambda (x^{\mathsf T} x - 1)\); stationarity gives \(A x = \lambda x\) — every candidate optimum is an eigenvector.

2. Read the objective at candidates. At an eigenvector, \(x^{\mathsf T} A x = \lambda\, x^{\mathsf T} x = \lambda\). The objective's value is the multiplier.

3. Pick the best candidate. The maximum is the largest eigenvalue \(\lambda_{\max}\), attained at its eigenvector. The move: stationarity first — the constraint's multiplier frequently is the answer, not a step toward it.

Bet sizing — maximize log growth, not expected value

Takeaway: expected value says bet everything on any positive edge; expected log wealth penalizes ruin and produces a finite, testable fraction — that switch of objective is the entire question.

Shape. A repeated bet wins \(b\) per unit staked with probability \(p\), loses the stake otherwise. What fraction \(f\) of bankroll do you bet?

1. Choose the right objective. Maximize \(G(f) = p \ln(1 + b f) + (1-p) \ln(1 - f)\), the expected log growth per round — not the expected profit, which is linear in \(f\) and would push you to \(f = 1\) and eventual ruin.

2. First-order condition. \(G'(f) = \frac{p b}{1 + b f} - \frac{1-p}{1-f} = 0\) solves to \(f^* = p - \frac{1-p}{b}\) — the Kelly fraction. For an even-money bet (\(b=1\)) this is \(2p - 1\): edge equals size.

3. Sanity-check the boundaries. \(G\) is concave, \(f^* < 1\) always, and \(f^* \le 0\) exactly when the edge is non-positive — no edge, no bet. The move: name the objective before touching the calculus; conditional on i.i.d. repeated bets, the FOC does the rest.

Free practiceOptimization practice questions by sub-area

All 14 free problems from the 119-problem optimization set, grouped by the sub-areas above — every link opens the full worked solution, so you can check your reasoning line by line. One honesty note: the free pool is 1 easy, 12 medium and 1 hard; the bank's 43-problem hard tail sits mostly in the paid tier, so treat a clean, fast medium here as the bar.

Convexity & first-order conditions

Bet sizing & portfolio construction

Pricing & expected-value trade-offs

Dynamic programming & sequential decisions

Numerical methods & regularization

The planHow to prepare for optimization questions

Five techniques to make reflexive, in the order they pay off. Each maps onto one of the practice groups above, so you can drill it immediately after reading.

  1. Make differentiate-and-verify one motion. The FOC finds candidates; the convexity or sign-change argument makes them answers. Interviewers listen for the second half — practice saying “and this is global because…” without being prompted.
  2. Hunt for the friendly transformation. Logs turn products into sums and powers into products; substitutions expose hidden convexity. If the raw objective looks ugly, that ugliness is usually the hint.
  3. Price constraints with multipliers. Read a Lagrange multiplier as the marginal value of relaxing its constraint, and know the pattern where stationarity itself is the result — the quadratic-form-to-eigenvalue reduction is the canonical case.
  4. Own the log-growth objective. Derive the Kelly fraction from scratch until it takes two minutes, and be ready to explain why expected log wealth is the right objective for repeated bets — the derivation is easy, the justification is the signal.
  5. Reach for the Bellman recursion in sequential setups. When decisions unfold over time, define the state, write the one-step recursion and solve backward — greedy is the trap, not the tool. Then run the free practice set and take the interactive optimization playlist on a clock.

FAQOptimization interview questions — frequently asked

How important is optimization for quant interviews?

Central for quant research and quant trading roles: sizing a bet, setting a price and fitting a model are all optimization problems, so interviewers use the topic to test whether your calculus is connected to decisions. In this bank the problems skew toward research and trading tracks, with a smaller software-engineering slice built on numerical methods.

What optimization topics should I focus on?

First-order conditions with a convexity check cover the most ground, followed by log-utility Kelly sizing, Lagrange multipliers for constrained problems, and backward-induction or Bellman reasoning for sequential decisions. Newton's method and gradient descent round out the numerical side.

How hard are optimization interview questions?

Mostly medium-to-hard: of the 119 problems in this set, 70 are medium, 43 are hard and only 6 are easy. The calculus itself is rarely the obstacle — the hard part is modeling the situation so the objective and constraints are the right ones.

Are these real quant interview questions?

They are representative, not verbatim. The problems are curated from our bank to match the optimization shapes candidates report, rewritten for clarity with worked solutions we author ourselves — we never claim any wording is a leaked question.