login
A336774
a(n) is the least number of repetitions such that the result of the repeated execution of the multiplication f <- f*n started at f=1 differs from the exact value n^a(n), when the multiplication is performed using 32-bit single precision floats according to the IEEE 754 standard.
8
128, 16, 64, 11, 16, 9, 43, 8, 11, 7, 16, 7, 9, 7, 32, 6, 8, 6, 11, 6, 7, 6, 16, 6, 7, 6, 9, 5, 7, 5, 26, 5, 6, 5, 8, 5, 6, 5, 11, 5, 6, 5, 7, 5, 6, 5, 16, 5, 6, 5, 7, 5, 6, 5, 9, 5, 5, 5, 7, 5, 5, 5, 22, 4, 5, 4, 6, 4, 5, 4, 8, 4, 5, 4, 6, 4, 5, 4, 11, 4, 5, 4
OFFSET
2,1
COMMENTS
For n a power of two the first deviation is caused by exceeding the maximum representable number of the binary32 single precision floating point format.
LINKS
Hugo Pfoertner, Comparison with exact results, range n=2..128, computed on x86-64 architecture.
Edmund Weitz, IEEE 754 Calculator, uses JavaScript (2016).
Wikipedia, IEEE 754.
CROSSREFS
KEYWORD
nonn,fini
AUTHOR
Hugo Pfoertner, Aug 04 2020
STATUS
approved