Issue when converting grids into APPLgrid

[fgiuli]

Dear all,
good morning.

I tried converting some PineAPPL grids generated by @scarlehoff for the W+jets 8 TeV measurement (arXiv:1711.03296), but I got the following error:

[giuli@pplxint13 xFitter_EL9]$ RUST_BACKTRACE=full pineappl export ATLAS_WP_JET_8TEV_PT-atlas-atlas-wjets-arxiv-1711.03296-xsec002.pineappl.lz4 Wjets_Wp_pT_8TeV.appl MSHT20nnlo_as118
thread 'main' panicked at pineappl/src/grid.rs:952:29:
not yet implemented
stack backtrace:
   0:     0x5557ff302859 - <std::sys::backtrace::BacktraceLock::print::DisplayBacktrace as core::fmt::Display>::fmt::hc4439741452ac1a2
   1:     0x5557ff0bba83 - core::fmt::write::h4392e9747776afaf
   2:     0x5557ff2cacf2 - std::io::Write::write_fmt::h308d1ad615b5d20c
   3:     0x5557ff3039e3 - std::sys::backtrace::BacktraceLock::print::h204f8c26e3c76121
   4:     0x5557ff303f2f - std::panicking::rust_panic_with_hook::hc7a13f6a77cad139
   5:     0x5557ff303a85 - std::panicking::begin_panic_handler::{{closure}}::hdb1cf457123442d6
   6:     0x5557ff303a19 - std::sys::backtrace::__rust_end_short_backtrace::h25571fe910c10d94
   7:     0x5557ff303a0c - rust_begin_unwind
   8:     0x5557ff04a80f - core::panicking::panic_fmt::h1a7ab0e9b5409cee
   9:     0x5557ff04a85b - core::panicking::panic::ha317b5b5b8afb83a
  10:     0x5557ff10b26d - pineappl::grid::Grid::optimize_using::hcc21f1bdf8fa480f
  11:     0x5557ff16b082 - pineappl_cli::export::applgrid::convert_into_applgrid::h0df33c7a4ce8860b
  12:     0x5557ff064cef - pineappl::main::h1396849747b6a72a
  13:     0x5557ff05f333 - std::sys::backtrace::__rust_begin_short_backtrace::h8c0f77d8b36f93d1
  14:     0x5557ff076e74 - main
  15:     0x7f43dd4295d0 - __libc_start_call_main
  16:     0x7f43dd429680 - __libc_start_main@GLIBC_2.2.5
  17:     0x5557ff05c025 - _start
  18:                0x0 - <unknown>

Have you ever seen a similar error?

Just for the sake of completeness, I am using the exact same setup as yesterday (master version, related to this issue here: #356 )

Best,
Fra

[Radonirinaunimi]

Hi @fgiuli, thanks a lot for this bug report. This error arises because of the merging of empty subgrids is not implemented yet (this line for reference). We'll look into it.

[cschwan]

This happens whenever there's a partonic channel without its transposed for symmetric collisions; for example, when there's u d~, but not d~ u in a proton-proton collision to W+.

This is probably caused by a Monte Carlo integrator that distributes phase-space points to partonic channels according to their importance, but one of the channels (due to the MC uncertainty) is below a certain threshold and thus is integrated with zero phase-space points. The right thing to do my opinion would be treat all channels that differ by transposition equally, but if you're only given the grids without being able to generate them yourself you can't do that, obviously.

[cschwan]

@fgiuli: this grid should now properly export into an APPLgrid. To install the fixed program, you need to select the fix-export-applgrid branch, which means installing with:

cargo install --locked --git https://github.com/NNPDF/pineappl.git --branch fix-export-applgrid --features=applgrid pineappl_cli

If it works as expected, please close this Issue. Thanks again for reporting this bug!

[fgiuli]

@cschwan the issue with converting the grids is now solved, but now I got the following error:

WARNING: the order O(as^2 a^2 lr^1 lf^0 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^2 a^2 lr^0 lf^1 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^3 a^2 lr^1 lf^0 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^3 a^2 lr^0 lf^1 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^3 a^2 lr^2 lf^0 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^3 a^2 lr^1 lf^1 la^0) isn't supported by APPLgrid and will be skipped.
WARNING: the order O(as^3 a^2 lr^0 lf^2 la^0) isn't supported by APPLgrid and will be skipped.
b APPLgrid PineAPPL rel. diff
--+------------+------------+--------------
0 1.7619987e3 1.7619987e3 5.4880323e-11
1 4.2953974e3 4.2953974e3 -2.2510771e-11
2 2.2024478e3 2.2024478e3 8.2100993e-12
3 8.1906554e2 8.1906553e2 -2.7845346e-9
4 3.4661562e2 3.4661562e2 1.5100077e-10
5 1.6256926e2 1.6256926e2 2.8130964e-10
6 8.1614008e1 8.1614008e1 -2.6925351e-10
7 4.3761197e1 4.3761197e1 1.4602430e-10
8 1.9197922e1 1.9197922e1 2.0379698e-10
9 6.8877948e0 6.8877948e0 3.5034420e-11
10 2.7737322e0 2.7737322e0 5.1123727e-10
11 1.2289505e0 1.2289505e0 1.1047829e-11
12 5.8169627e-1 5.8169627e-1 5.0375926e-10
13 2.9587883e-1 2.9587883e-1 7.3998474e-10
14 1.1459171e-1 1.1459171e-1 1.1146035e-9
15 2.4205527e-2 2.4205527e-2 4.9289139e-9
Error: grids are different

Not sure if this is an issue with PineAPPL though...

[felixhekhorn]

I think PineAPPL is just a bit aggressive here: by default it checks that the APPL and PineAPPL results match each other to $10^{-10}$ precision:

pineappl/pineappl_cli/src/export.rs

Lines 79 to 80 in 335e8c5

	#[arg(default_value = "1e-10", long)]
	accuracy: f64,

since (as far as I understand) the number of "0"s is different on both sides you may not get that precision here. If you are happy with the printed accuracy, I think you can change that required accuracy by setting the the --accuracy flag (see also the --help flag) and get a successful printout

[fgiuli]

Ok @felixhekhorn thanks a lot for the explanation.
Closing the issue now.