Phase-space bound misses the -sum-mass spelling of a pair cut

[oliviermattelaer]

Follow-up to #110, deliberately left out of that PR because it is a behaviour change rather than a rename. #110 can merge as it is.

The gap

lepton-lepton-sum-mass and lepton-pair_mass are the same cut: both are the invariant mass of every lepton pair. As event filters they behave identically. Only the second one bounds the integration.

Measuring the floor Cuts::m_inv_min() reports for a min=500 cut on the top pair of g g > t t~ g g:

cut	floor
top-pair_mass	500.0
top-sum-mass	500.0
top-top-sum-mass	0.0

So a user who writes the cut the pre-existing way gets the forbidden region generated and thrown away, while the same cut spelled pair_mass never generates it. Nothing warns them; the cross section is right either way, only the efficiency differs.

Why

Cuts::m_inv_min() in madspace/src/phasespace/cuts.cpp:129 matches the summed spelling with

if (o.sum_momenta() && idx.size() == 1 && idx.at(0).size() == 2) {
    pairs.emplace_back(idx.at(0).at(0), idx.at(0).at(1));
}

Observable::build_indices lays out _indices as one list per selection group. A single group (top-sum-mass) gives idx.size() == 1 with both members inside idx[0], which is the shape above. Two groups (top-top-sum-mass) give idx.size() == 2, one member each, so it falls through and contributes no floor.

That second shape is exactly what the obs_pair_mass branch just below (cuts.cpp:137) already handles - it is keyed on the observable, so it never sees an obs_mass observable.

Reproducing

import madspace as ms
pids = [21, 21, 6, -6, 21, 21]
O = ms.Observable

def floor(o):
    return ms.Cuts([ms.CutItem(o, min=500.0)]).m_inv_min()[0][1]

print(floor(O(pids, 'pair_mass', [[6, -6]])))                          # 500.0
print(floor(O(pids, 'mass', [[6, -6]], sum_momenta=True)))             # 500.0
print(floor(O(pids, 'mass', [[6, -6], [6, -6]], sum_momenta=True)))    # 0.0

Fix

Extend the summed lambda to also accept idx.size() == 2 with sum_momenta(), pairing idx[0][k] with idx[1][k] the way the obs_pair_mass branch does. Worth checking that the > 2 group case (lepton-jet-top-sum-mass) stays excluded, since a triple's mass is not a floor on any pair.

The sqrt(s_hat) bound in phasespace.cpp reads the same table, so it follows for free.

🤖 Generated with Claude Code