Your translation looks correct, so I'm a bit surprised by the result. I only tested it in 1-D, but the histogram looked plenty spiky. You could make it spikier by raising the random numbers used to generate @fx to some power, but it sounds like you found another solution.