Solution: normalise the input vector so that its components sum to 1. Compensate for the space reduction by adding a constant `log(sum)* (dimension-1)`.