← Citadel Interview Insights

Citadel·Software Engineer·Technical Phone Screen·Senior

SeniorPrefer not to say
May 2026

Summary

One technical question, full stop. Citadel quant researcher screen that was basically a single optimization theory problem and then you're done. Sparse but not easy.

Questions Asked (1)

Q1

Consider the quadratic optimization problem: minimize x^T Q x + c^T x, where Q and c are real-valued. Under what conditions does this problem have a finite minimum?

Technical Trade-offsAlgorithms & Data Structures
Author's notes

This looks clean on paper but there's a lot packed in.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by stating the necessary and sufficient conditions for a finite minimum: Q must be positive semidefinite and c must lie in the range of Q (i.e., the linear term must be orthogonal to the null space of Q). Then explain why these conditions ensure the quadratic form is bounded below and the linear term does not cause unbounded decrease along directions in the null space.

Pro tip: Mention that if Q is positive definite, the minimum is unique and finite for any c; if Q is only positive semidefinite, the minimum may be finite but non-unique, and you must check the range condition on c. This shows depth and avoids oversimplification.

1. Identify the nature of Q

Determine whether Q is positive definite, positive semidefinite, or indefinite. This dictates whether the quadratic form is strictly convex, convex, or unbounded below.

2. Check the linear term condition

If Q is positive semidefinite but not positive definite, the linear term c must be orthogonal to the null space of Q (i.e., c ∈ range(Q)). Otherwise, the objective can be driven to -∞ along null directions.

3. State the necessary and sufficient conditions

Combine the findings: Q must be positive semidefinite, and c must lie in the range of Q. These conditions are both necessary and sufficient for a finite minimum.

4. Discuss uniqueness and solution set

If Q is positive definite, the minimizer is unique. If Q is only positive semidefinite, the set of minimizers is an affine subspace, and any solution can be found by solving the linear system Qx = -c/2 (or Qx = -c depending on convention).

5. Connect to practical implications

Relate to optimization algorithms: positive definiteness ensures convergence of gradient descent, while semidefiniteness requires careful handling of null space components.

Key Points to Mention

  • Positive semidefiniteness of Q is necessary for boundedness below.
  • If Q is indefinite, the problem is unbounded below (no finite minimum).
  • For positive semidefinite Q, the linear term c must be in the range of Q (equivalently, orthogonal to the null space of Q).
  • If Q is positive definite, the minimum is finite and unique for any c.
  • The optimality condition is Qx = -c/2 (or Qx = -c, depending on formulation), and a solution exists iff c is in the range of Q.
  • In practice, check eigenvalues of Q and project c onto the null space to verify the condition.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.