[zgs]

I remember seeing a paper that concluded the expansion should not be more than the golden ratio (ϕ = 1.618033988749). Given how widespread the golden ratio, I'd hypothesise that it's probably optimal.

The first three fractions approximating ϕ are: 3/2, 8/5 and 21/13. The first two are quick and easy to compute.

[newaccount74]

If the multiplier is less than the golden ratio, then the new allocation fits into the memory released from the previous two allocations.

[stephencanon]

But most allocators will not actually take advantage of that, so it’s a pretty dubious argument.

[newaccount74]

Good point, I haven't actually checked if any realloc implementations do that.