Comparing a model-generated amplitude with Parke–Taylor
Three rational momentum configurations test the compact four-gluon expression for helicities (--++).
Build momentum-conserving spinors
Take four outgoing massless momenta with cyclic color order (1, 2, 3, 4) and helicities (−, −, +, +). Each momentum is the outer product of an undotted and a dotted spinor. Use their two-by-two determinants in leg order to form angle and square brackets. Independent rational spinors provide test points in complexified kinematics without a Lorentzian reality condition.
For each point choose the four undotted spinors and the last two dotted ones. Solve the two linear equations for each dotted index of legs 1 and 2. Check that all four components of the total momentum are zero before evaluating either expression.
Compare the two expressions
The reference is the four-point Parke–Taylor expression in Eq. (1.1). The comparison uses the compact model output printed immediately after Eq. (G.2). The coupling and overall phase are stripped consistently. Both formulas are finite at the chosen points.
| Point | Parke–Taylor | Model formula | Sign flipped |
|---|---|---|---|
| 1 | 1/2 | 1/2 | -1/2 |
| 2 | 3/70 | 3/70 | -3/70 |
| 3 | 49/40 | 49/40 | -49/40 |
Rescale one undotted spinor by 3/2 and its dotted partner by 2/3. This leaves the momentum fixed. Both expressions acquire the expected factor of 9/4 on either negative-helicity leg and 4/9 on either positive-helicity leg, in all twelve individual rescalings.
What the checks detect
An overall sign error fails at all three points. Perturbing one dotted-spinor component breaks momentum conservation and the script rejects the input. The comparison tests the two compact expressions at these points. It does not evaluate the preceding 17-term expression, retrain the model, or prove equality at all kinematics.
Repeat the calculation
Save the Python script and recorded output (JSON) in the same directory. The site ran the script with CPython 3.12.14 and its standard library. The calculation uses exact arithmetic and completes in under a second.
python3 -I check.py > rerun.json
cmp result.json rerun.jsonA successful comparison exits with status zero. The output includes the inputs and individual checks.
Sources and interpretation
Learning the Simplicity of Scattering Amplitudes. arXiv:2408.04720v2. §1, Eq. (1.1), Parke–Taylor benchmark. §§4.2–4.3, grouping and sequential simplification. §4.4, Fig. 9 and Eq. (4.7), physical examples. Appendix G, four-gluon input in Eq. (G.2) and the unnumbered output immediately following.
Read the research entry for the role of AI in the original work and the range of its conclusions.