According to Radford Neal, it is possible to sum a sequence of n-bit floating point numbers exactly, i.e. rounding to the n-bit floating point value closest to

According to Radford Neal, it is possible to sum a sequence of n-bit floating point numbers exactly, i.e. rounding to the n-bit floating point value closest to the exact result. The algorithm uses 2n bits for the sum. https://radfordneal.wordpress.com/2015/05/21/exact-computation-of-sums-and-means/ https://radfordneal.wordpress.com/2015/05/21/exact-computati...