I see, thanks. It might be worth to open an issue with them and see if someone picks it up. At least, they will probably tell you if they’re already considering it.

I’m really not an very familiar with SWIG details and my implementation is a simple global mutex lock around every method, which uses the swig_type_info. Pretty sure, that someone more familiar with the internals will find a much better solution and don’t know what else I’ve missed.

Array localVol = Sqrt(2*dCdT/d2CdK2);

is the implementation of formula (15) from the original paper.

1)You Solve the Dupire PDE for for the available strikes one Expiry at a time iteratively by jumping from one expiry to the next by solving the PDE.

2) Now you have everything calibrated to compute call prices at T_i<t<T_i+1 you use the call grid prices at T_i and backwards flat constant vol(i.e coming from T_i+1) and do a single jump to t.

3) What I don’t getis in Quant lib method AndreasenHugeVolatilityInterpl::getLocalVolSlice when you want to extract the slice at T_i<t<T_i+1 you don’t seem to use the flat local vol assumption between [T_i,T_i+1] you used in step 2.

Instead you use localVol = Sqrt(2*dCdT/d2CdK2) to extact the slice at t.. Theis is “almost”the normalized Dupire PDE a the top of this page with with a derivative missing.

This give non piecewise constant local volatilities for all the expiries in [T_i,T_i+1]

What is the intuition behind this?

do you have a public repo that shows your own implementation that does not depend on Quanltib? Trying to do a rust lib with this algo.

QuantLib 1.3, which is still available on sourceforge. boost must have been 1.51 or so.

Thanks! Can you confirm which version of QuantLib and Boost did you use for this study, please?