Michael Goerz

For comparison, this is how [L-BFGS-B](https://github.com/Gnimuc/LBFGSB.jl) performs:  Much more well-behaved! (and no linesearch parameters to tune, which is a good thing)

> > For the default parameters, this looks as follows for the first two iterations: > > Can you show me a script so I can generate that alpha=3000 iteration?...

It’s the control values: “opt” [is the output `x`](https://github.com/JuliaQuantumControl/GRAPELinesearchAnalysis.jl/blob/9ce2830e503af1c38034e08f3b1204addebbcc77/src/GRAPELinesearchAnalysis.jl#L50) in the current iteration, and “guess” is the `x_previous`, i.e., the input control values. For this application, the control values are...

> Didn't you use `opt` in both cases? :) Yes, typo, sorry, fixed :-) The [LBFGSB paper](http://dx.doi.org/10.1137/0916069) references [J. J. Moré and D. J. Thuente, ACM Trans. Math. Softw.  20 ...

If you mean the second example from https://github.com/JuliaNLSolvers/Optim.jl/issues/953#issuecomment-951543625, where I've included the plot for iterations 47-50, that looks like it would have been ~~~julia opt_result = optimize_grape( problem, info_hook=chain_infohooks( GRAPELinesearchAnalysis.plot_linesearch(@__DIR__),...

As far as I understand, the point is to avoid duplicate work when you need both the function value and the gradient in the same iteration. In [my case](https://github.com/JuliaQuantumControl/GRAPE.jl), evaluating...

Sounds good! Those 25% add up though, since I may need 100,000 iterations or more to converge. So total runtimes can be a week, and the difference could be a...

If you change the rules for `det`, make sure to test them with matrices that have a complex determinant, e.g. a random unitary. The current rule gives wrong results, see...

Maybe a middle ground would be `#! format: skip` for a single line?

To elaborate a little on the syntax: There's two prefixes (which can also be combined), `.` and `~`. (There's also `!`, but I don't know when you'd ever use that...