← Squarepoint Interview Insights
I stared at this for a second because my instinct was to think about continuous distributions and I started rambling about uniform distributions.
Recognize that variance is maximized by placing all probability mass at the endpoints of the interval, i.e., a two-point distribution with equal probabilities at a and b. Explain that this follows from the fact that variance is a convex function of the distribution and is bounded by the Bhatia–Davis inequality, which states Var(X) ≤ (b - a)^2 / 4 for any X in [a, b].
Pro tip: Mention the Bhatia–Davis inequality by name and note that the maximum variance (b - a)^2 / 4 is achieved by any distribution with P(X = a) = P(X = b) = 1/2; this shows you know the theoretical bound and the extremal construction.
Clarify that we need a random variable X with support contained in [a, b] and we want to maximize Var(X). Note that the distribution can be discrete or continuous, but the support must lie within the interval.
Invoke the Bhatia–Davis inequality: for any random variable X with values in [a, b], Var(X) ≤ (b - a)^2 / 4. This gives an upper bound and suggests that the maximum is achieved when the distribution is concentrated at the endpoints.
Define X such that P(X = a) = 1/2 and P(X = b) = 1/2. Compute its variance: E[X] = (a+b)/2, E[X^2] = (a^2 + b^2)/2, so Var(X) = (b - a)^2 / 4, which matches the upper bound.
Explain that any other distribution on [a, b] will have variance strictly less than (b - a)^2 / 4 unless it is also a two-point distribution at a and b with equal probabilities. This follows from the equality condition in the Bhatia–Davis inequality or from convexity arguments.
Mention that in practice, if a continuous distribution is required, one can approximate the two-point distribution by a continuous distribution with mass concentrated near the endpoints (e.g., a beta distribution with parameters approaching 0 and 0, or a mixture of narrow uniforms at the endpoints).
AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.