duanebyer / sidis Goto Github PK

The make install command doesn't provide libmstwpdf, but it should be included for linking with libsidis.

The vacuum hadron polarization calculation is a bit outdated. It should be replaced with a more modern estimate.

It's very difficult to make changes to the parameter file format. A large refactor of the code for the parameter system is underway, that should make adding/removing new parameters in the future much easier.

• Check compatibility of parameters between initialization stage with generation stage.
• Use parameters to check whether two event files can be merged.
• Add hash of generator engine into the parameter files, to ensure correctness during merging and event generation.
• Test thoroughly--this is a big change.

The num_events[4] in stats does not match with what it is supposed to be as in paper

The cross section calculation in example/process_events.cpp is weight*norm. Should be weight/norm

The exclusive contribution to the cross section is not currently included. We need support for exclusive structure functions to move forward on this.

• Kinematics for exclusive events.
• Cuts for exclusive events, and extend radiative cuts to work correctly on exclusive events.
• Structure function interface ExclusiveSfSet.
• Cross section calculations using the exclusive structure functions.
• Generators for the 8-D exclusive phase space.
• Parameters for MC generation of exclusive events.

The convention is:

• 0 for total
• 1 for non-radiative
• 2 for radiative
• 3 for exclusive

This might not be followed by all parts of the code, especially the stats output in the TTree, so it should be double-checked.

The sf_out option depends on there being no padding in the SfLP struct, which could change depending on the compiler.

The asym namespace lets UT asymmetries be calculated efficiently. What about other kinds, like LT, LL, and so on? These should also have efficient and convenient implementations.

Right now, only the TMD factorization is supported. For regions of high qT/Q, it is necessary to use the collinear factorization. It would also be useful to provide matching between TMD factorization and collinear factorization for intermediate qT/Q.

• Add new interfaces (ex. CollinearPDFSet) to represent collinear PDFs/FFs.
• Write a new implementation of SfSet (ex. CollinearSfSet) that can calculate structure functions from collinear factorizations.
• Write a new implementation of SfSet that can combine a TMD factorization (TmdSfSet) with a collinear factorization (CollinearSfSet) using a matching procedure for intermediate qT/Q.

The QED coupling alpha_QED should run with Q, right now it's always kept constant.

The user should be able to choose which kinematic variables get logged to the event generation file. Right now, only the six SIDIS phase space variables (x, y, z, pt^2, phi_h, phi) are logged. The parameter file should support a selection of kinematic variables to log. E.x.

out.file.vars = x,y,Q_sq,W_sq

The CMake configuration is broken for clang on linux, there are linking issues with libm and LTO.

The code generation with cog makes the source much harder to read. It's also an awkward part of the build.

• Remove code generation from polarized structs (#19).
• Remove code generation from version headers
• Remove code generation from number-type header.

Different structure functions (even with different polarization) will share some common TMDs between them. For Gaussian-approximated structure functions, this means that TMDs are redundantly calculated several times when computing a single cross-section.

• Refactor SfSet so that there are special virtual methods for calculating polarized groups of structure functions (#19).
• Make TMDs return array of flavours, instead of each flavour being calculated by a different call.
• Allow TMD results to be cached, by polarization.
• Cached TMD results used by GaussianSfSet to calculate groups of structure functions more efficiently.

The headers within the "sidis/extra" folder should be kept internal to the project only.

• When installing libsidis, they should not be installed.
• "sidis/extra/exception.hpp" should be moved out of the extra folder.
• Interpolation capabilities should be provided through specialized functions for creating LHAPDF, TMDLib, or CSV grids for use in user-defined TMDs/Sfs.

We should support additional targets, such as neutron and nuclear targets. Right now, users can define their own structure functions for these situations, but it would be nice if there was a simple default option.

• Neutron structure functions can be estimated from proton structure functions by interchanging valence up and down quarks.
• Nuclear structure functions can be estimated by combining proton and neutron structure functions together.
• Fermi motion and other nuclear effects should be considered.

Currently, if Gaussian approximation is not used, the TMDs are convoluted numerically in 2D to calculate the structure functions. They are integrated out to an arbitrary upper bound. This can be improved in a few ways.

• Use a 1D convolution integration when the TMDs are spherically symmetric (which is the usual case).
• Use a change of variables so that the TMDs can be integrated all the way to infinity, instead of to a fixed upper bound.
• Relatedly, use a change of variables that estimates the TMDs as Gaussian for the numerical integration. This should substantially improve the performance of the TMD integration.
• Experiment with the numerical integration method. Right now cubature is used (I think?), but other choices might work better.

We should consider whether certain libraries should be factored out of libsidis/sidisgen.

• 3,4-vectors and transformations
• Interpolation
• Parameter system
• Monte-Carlo statistics class (Stats)

Currently, a modified version of FOAM is used for the underlying Monte-Carlo generator in sidisgen. Alternate algorithms like VEGAS might perform better in certain situations. It should be possible for the user to choose which MC generator they want to use. The code almost supports this already, it just needs a little bit more work to get all the way there.

• Abstract the generator interface so that the underlying MC engine can be swapped in/out.
• Add support for three generator engines: original FOAM, modified FOAM, and VEGAS.
• Alternatively, if VEGAS just outperforms FOAM in all situations (testing needed), we'll just stick to VEGAS.

We should provide easy integration with external libraries for TMDs and PDFs, such as LHAPDF library and TMDlib.

• Should be enabled by a CMake configuration variable to avoid a required dependency.
• For libsidis, add new implementation of TmdSet that can load TMDs/PDFs from the system.
• For sidisgen, it should be easy to specify a desired set of TMDs and PDFs in the parameter file.
• Should be combine-able with WW-type and Gaussian approximations.

It would be useful for users of libsidis to be able to construct a structure function from a simple string description. The code to do this could be moved from sidisgen where it lives currently, to libsidis. Then call something like:

std::unique_ptr<sf::SfSet> sf = sf::create("leading prokudin");

The only issue is that sidisgen also supports loading ROOT libs to provide structure functions. This could be implemented as a fallback, in case sf::create were to fail.

Only ROOT file formats are supported by sidisgen currently. Other MC file formats, like HepMC, are sometimes more useful.

• Abstract over output file format.
• HepMC support.
• Plain text support.
• Make ROOT an optional dependency for sidisgen, use CMake configuration variable to enable.