Open goerz opened 3 months ago
Do we need to normalize the spellings of Chebychev
to Chebyshev in QuantumPropagators.jl
?
No, I always spell it Chebychev
, and that will be the canonical spelling inside this package.
Alrighty!
I am excited to explore Bessel.jl
for chebychev coeff
!
As you mentioned we have to useBessel.jl
instead of (besselj)
from SpecialFunctions.jl
, as it is typically 2-12x faster than specialfunction.jl
implementations. Bessel.jl also have same names etc. besselj
if I recall correctly, the regular exponential just uses the modified Bessel function (besseli instead of bessel) and you kill the i in the definition of c and in the final phase factor
It seems that regular exponential uses besselj
instead of besseli,
, so there might be no need to kill i in c and phase factor, right?
To really "explore" this would require proper benchmarking in realistic settings. I have a WIP benchmarking repository that isn't quite ready to be pushed. Since the coefficients can typically be reused a lot, they are very unlikely to be a bottleneck in any optimization. Still, at some point, it would be nice to quantify it a bit more. This issue is more of a note to myself to look at this when my benchmarking code is in better shape. I don't think looking into this would be a good use of your time.
Sounds good! I enjoyed your talk at Yao Community Call, that was comprehensive!
Looking forward to the workshop tutorial!
This might make the calculation of coefficients significantly faster:
See https://discourse.julialang.org/t/matrix-exponential-vector-for-various-scales/112022/17