A team is designing a loss function of one parameter . They have narrowed it to this family, where is a constant they are free to choose:
They want the loss to be convex over the whole real line — a single bowl, no side dips — so that any flat point their search reaches is guaranteed to be the global minimum rather than a trap.
For a twice-differentiable function, convexity is the condition
not merely at the point they happen to start from. A large tilts the cubic term hard enough to punch a second valley into the curve; a small enough leaves the bowl intact.
What is the largest value of for which is convex everywhere?
Round your answer to 2 decimal places.