You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It is found to be more robust to use cos.(θ)+im .*sin.(θ), which functioned correctly at point [1,0], instead of exp.(im .*θ).
I think the problem has something in common with #653.
The difference between cos.(θ)+im .*sin.(θ) and exp.(im .*θ) is printed as follows
This is probably fixed by #481, as master gives the right answer:
julia>H([1,0]) ≈H([1,1e-10])
true
(jl_JRlWH7) pkg> st ForwardDiff
Status `/private/var/folders/yq/4p2zwd614y59gszh7y9ypyhh0000gn/T/jl_JRlWH7/Project.toml`
[f6369f11] ForwardDiff v0.11.0-DEV `https://github.com/JuliaDiff/ForwardDiff.jl.git#master`
(#481 was pulled from v0.10 as it seemed that some people relied on the old behaviour, and we haven't tagged v0.11 yet... because making a breaking release of such a widely used package creates a lot of work downstream, and people have ideas for other things they would quite like to roll into a breaking 1.0 release.)
Activity
mcabbott commentedon Dec 17, 2023
This is probably fixed by #481, as master gives the right answer:
(#481 was pulled from v0.10 as it seemed that some people relied on the old behaviour, and we haven't tagged v0.11 yet... because making a breaking release of such a widely used package creates a lot of work downstream, and people have ideas for other things they would quite like to roll into a breaking 1.0 release.)
devmotion commentedon Aug 21, 2026
Fixed on master and in the latest release.