Currently, it appears that round-off error is occurring when an EnergyGrid instance is constructed directly from an std::vector<double>, as it first converts all of the values to floats, and then saves them as doubles. This is a remnant of when I used to store XS data as floats. It should suffice to just pass the energy argument to the energy_values_ member in the class initialization, and remove the entire for loop at line 63.

While the installation documentation does state that GCC >= 11 is needed to building PapillonNDL, it currently still has distribution-specific installation instruction. Some of these distributions (mainly Debian / Ubuntu) do not yet have GCC 11 however. This needs to be updated.

Also, update the C++17 mention on the docs homepage to C++20.

Additionally, the new PNDL_SHARED and PNDL_INSTALL cmake options should be documented in the installation instructions.

Now that the move to C++20 has been initiated, the ACE class should be modified to return instances of std::span, and also make use of the new ranges library, when returning transformed segments of the ACE file.

Lambdas will be needed to interface with the python bindings however, as Pybind11 does not support spans. These will just have
to be converted into vectors on the C++ side, and then Pybind will be able to handle them normally.

I just noticed a problem, where when I stream a ZAID with leading zeros in the A portion (i.e. H1), I end up getting 11, instead of 1001. This is because I stream

out << zaid.Z() << zaid.A();

I should probably use fixed precision when streaming the A portion.

Support for reading and sampling the standard URR probability tables should be added. As these are temperature dependent, they should probably only be added to STNeutron and not to CENeutronBase.

While the CeCILL license is a good license in general, I have now come to realize that Article 13 may be problematic. Additionally, it allows one to pass to the EUPL, which then allows a move to non-copyleft licenses. Because of these two problems, the license should be changed from CeCILL-2.1 to GPL3+. This will require updates to the LICENSE file, and all source files.

There are currently several TSL files distributed with Lib80x which cannot be read, as PNDL throws an exception due to negative PDF values in the incoherent inelastic distributions. Looking at one from Al27 for example, I see a PDF of -1.033976E-19. It is very likely that this could just be set to zero without any problem, but I am still somewhat hesitant about doing this. I will go through all of Lib80x at some point, and see just how many files can and cannot be read. It might also be interesting to remake all of the problematic TSL files with FRENDY, to see if that has better tendencies than NJOY.

There can possibly be several different ACE tables in an ACE file. One common example of this would be a single file for a given nuclide which contains the ACE tables for all the desired temperatures. MCNP for example explicitly allows for this and has an entry in the xsdir indicating where to start reading the file for the desired table. In practice, however, I think this functionality is rarely used ? Currently, PNDL cannot treat this case, and would only read the first ACE table in the file.

I have written an implementation of the Sigma1 algorithm to perform doppler broadening, but it has not yet been added to the library. This must be added, and a new type of cross section should be created, allowing users to supply a temperature argument, and receive the cross section at the desired temperature (so long as that temperature is above the initial temperature).

Here are the general requirments to accomplish this:

• Add the Sigma1 implementation to the library
• Add new methods to EnergyGrid, defining the start point of the URR
• Create a new S1CrossSection class
• Create new Reaction template specialization Reaction<S1CrossSection>, which should be aliased as S1Reaction
• Create new CENeutron template specialization CENeutron<S1CrossSection>, which should be aliased as S1Neutron