TECH Signal 537 2 feeds carried it
Double-double arithmetic provides 31 digits of precision at roughly 9x the cost of a standard double
↗
Double-double arithmetic pairs two standard doubles to achieve roughly 31 decimal digits of precision without heap allocation or external dependencies.
It bridges the performance gap between standard 15-digit doubles and arbitrary-precision libraries like MPFR, costing about 9x a plain double compared to MPFR's 81x. This makes it suitable for applications like deep Mandelbrot zooms where double precision fails but arbitrary-precision libraries are too slow or encumbered by LGPL licensing.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Double-double represents a number as the unevaluated sum of two doubles, maintaining an invariant where the high part equals the rounded sum of both components.
In a real kernel, double-double operations cost roughly 9x a plain double, placing performance halfway between standard doubles and MPFR.
The technique avoids the heap allocations, per-operation loops, and native binary dependencies required by arbitrary-precision libraries.
THE CLUSTER