Sum the four gluon current into the cubic one carrying the same colour factor - #57

Open

oliviermattelaer wants to merge 39 commits into

mainfrom

claude/gluon-amplitude-optimization-8706f5

Commits on Aug 4, 2026

optimise pure gluon amplitudes by unrolling the 4-gluon vertex

oliviermattelaer
and
claude
committed
fcd8218
record why the quartic currents cannot be summed as they stand

oliviermattelaer
and
claude
committed
5c9cdb3
drop the auxiliary line generation, keep the diagram set intact

oliviermattelaer
and
claude
committed
1c1722a
add reroot_diagram, turning a diagram around its final vertex

oliviermattelaer
and
claude
committed
98d288f
add the plan for the pure gluon amplitude optimisation

oliviermattelaer
and
claude
committed
3b3ed9e
enforce the seed rule inside diagram generation

oliviermattelaer
and
claude
committed
25fc6d1

Commits on Aug 5, 2026

reconstruct the full diagram set by unrolling the seed

oliviermattelaer
and
claude
committed
7da06ba
leave the seed rule off for decay chains

oliviermattelaer
and
claude
committed
0ad9327
record what steps 1-5 measured, and why the current sum is blocked

oliviermattelaer
and
claude
committed
18fa381
sum the quartic current into the cubic one instead of the amplitude

oliviermattelaer
and
claude
committed
8e634cf
record that the current sum lands at the amplitudes

oliviermattelaer
and
claude
committed
d8f21b6
measure what the second substitution would really be worth

oliviermattelaer
and
claude
committed
bfbf910
carry the current sum into madevent

oliviermattelaer
and
claude
committed
29b0c67
record the madevent port and what it does to AMP2

oliviermattelaer
and
claude
committed
06ad287
measure the integration both ways: no effect either side

oliviermattelaer
and
claude
committed
ceba0c4
carry the current sum into madmatrix

oliviermattelaer
and
claude
committed
e93dfa1
record the madmatrix port and the slot cost it exposes

oliviermattelaer
and
claude
committed
1ae5c34
recycle the slots the current sums take

oliviermattelaer
and
claude
committed
5b421bc
record the slot recycling and what is left for madmatrix

oliviermattelaer
and
claude
committed
5210b17
measure memory and speed for g g > N g, N=2..5, in both backends

oliviermattelaer
and
claude
committed
34486c0
drop the unit test output that got committed by mistake

oliviermattelaer
and
claude
committed
fefb115
measure what the diagram order is worth, and why it is not shipped yet

oliviermattelaer
and
claude
committed
7c446c2
add the madmatrix seven gluon row

oliviermattelaer
and
claude
committed
cd18597
check the colour structure when pairing a quartic current with a cubic one

oliviermattelaer
and
claude
committed
65c457f
find why the reordering breaks madmatrix: an antisymmetric current merged by equality

oliviermattelaer
and
claude
committed
2c34b7c
table the memory and speed across the three states of the branch

oliviermattelaer
and
claude
committed
d80a7a0
build the unrolled vertices with their legs in a canonical order, and reorder

oliviermattelaer
and
claude
committed
c5624c1
update the tables for the canonical leg order and the reordering

oliviermattelaer
and
claude
committed
473d04f
search the diagram order for the slot count, and keep the simple one

oliviermattelaer
and
claude
committed
4be9871
turn the four gluon switch into a set option, and add the slots mode

oliviermattelaer
and
claude
committed
5e5d7b3
let each output pick the diagram order, from the me exporter

oliviermattelaer
and
claude
committed
7f50f03
record why auto cannot be the default yet: it breaks fks

oliviermattelaer
and
claude
committed
b4aecfd
make the four gluon merging safe for NLO, and keep it opt-in

oliviermattelaer
and
claude
committed
5a2bb1d

Commits on Aug 6, 2026

record the scan for a fourth consumer, and what it bounds

oliviermattelaer
and
claude
committed
15cb14e

Commits on Aug 11, 2026

measure all three modes on both series, memory included

oliviermattelaer
and
claude
committed
f128aa6
gate the seed rule on the multiplicity, for auto

oliviermattelaer
and
claude
committed
f7e3e03
recycle the AMP array, the way the wavefunctions already are

oliviermattelaer
and
claude
committed
fbb5d2b
re-measure the sweep now that the AMP array is recycled

oliviermattelaer
and
claude
committed
e23e413
check the sweep against a process with little to merge

oliviermattelaer
and
claude
committed
78a801f