The mantissa (the decimal part as opposed to the exponent part) of a 64bit floating point number safely holds 21x10^14 (sats) and leaves enough overhead to add or subtract two values without overflow. This appears to be the reasoning given to by this article.
The mantissa (the decimal part as opposed to the exponent part) of a 64bit floating point number safely holds 21x10^14 (sats) and leaves enough overhead to add or subtract two values without overflow. This appears to be the reasoning given to by this article.
Log10(21x10^14)/log10(2) = 50.9 bits