handle reverse parallel arc

[m-bossart]

No description provided.

[github-actions]

Performance Results

Precompile Time

Main	This Branch	Delta
2.243 s	2.249 s	+0.3%

Execution Time

Test	Main	This Branch	Delta
matpower_ACTIVSg2000_sys-Build PTDF First	2.249 s	2.366 s	+5.2%
matpower_ACTIVSg2000_sys-Build PTDF Second	93.7 ms	77.5 ms	-17.3%
matpower_ACTIVSg2000_sys-Build Ybus First	11.7 ms	11.7 ms	+0.1%
matpower_ACTIVSg2000_sys-Build Ybus Second	10.0 ms	11.2 ms	+12.2%
matpower_ACTIVSg2000_sys-Build LODF First	141.3 ms	154.5 ms	+9.3%
matpower_ACTIVSg2000_sys-Build LODF Second	620.7 ms	212.8 ms	-65.7%
matpower_ACTIVSg2000_sys-Build VirtualMODF First	2.162 s	4.089 s	+89.1%
matpower_ACTIVSg2000_sys-Build VirtualMODF Second	101.0 ms	1.071 s	+959.7%
matpower_ACTIVSg2000_sys-VirtualMODF Query 10 rows	512.1 ms	516.6 ms	+0.9%
matpower_ACTIVSg2000_sys-Radial network reduction First	546.9 ms	543.9 ms	-0.6%
matpower_ACTIVSg2000_sys-Radial network reduction Second	0.8 ms	0.7 ms	-17.4%
matpower_ACTIVSg2000_sys-Degree two network reduction First	1.871 s	1.936 s	+3.4%
matpower_ACTIVSg2000_sys-Degree two network reduction Second	1.3 ms	1.1 ms	-12.7%
Base_Eastern_Interconnect_515GW-Build Ybus First	834.5 ms	817.6 ms	-2.0%
Base_Eastern_Interconnect_515GW-Build Ybus Second	630.1 ms	933.4 ms	+48.1%
Base_Eastern_Interconnect_515GW-Radial network reduction First	39.2 ms	44.8 ms	+14.3%
Base_Eastern_Interconnect_515GW-Radial network reduction Second	45.9 ms	45.5 ms	-0.8%
Base_Eastern_Interconnect_515GW-Degree two network reduction First	385.1 ms	401.1 ms	+4.2%
Base_Eastern_Interconnect_515GW-Degree two network reduction Second	48.3 ms	41.4 ms	-14.2%

[jd-lara]

Code review

Found 1 issue:

1. Anti-parallel branches are merged into one BranchesParallel group keyed by reverse_new_arc, but the group's orientation invariant is not preserved. The incoming branch is stored under the reverse direction without flipping its admittance block. This is correct for the DC path (BA/ABA/PTDF/LODF use the orientation-independent scalar get_series_susceptance, which is why the fix works there), but the AC equivalent path sums each member's 2x2 Y-block positionally and explicitly assumes all members share orientation:

PowerNetworkMatrices.jl/src/Ybus.jl

Lines 471 to 486 in 59faae9

	function ybus_branch_entries(
	parallel_br::AbstractBranchesParallel;
	min_x_eps::Float64 = ZERO_IMPEDANCE_X_EPSILON,
	)
	arc = get_arc_tuple(first(parallel_br))
	Y11 = Y12 = Y21 = Y22 = zero(YBUS_ELTYPE)
	for br in parallel_br
	# All branches in BranchesParallel are have the same orientation when constructed in add_to_branch_maps!
	(y11, y12, y21, y22) = ybus_branch_entries(br)
	Y11 += y11
	Y12 += y12
	Y21 += y21
	Y22 += y22
	end
	return (Y11, Y12, Y21, Y22)
	end

The new collision cases violate that invariant by placing a branch oriented new_arc = (A,B) into a group keyed reverse_new_arc = (B,A):

PowerNetworkMatrices.jl/src/Ybus.jl

Lines 1631 to 1645 in 59faae9

	existing = pop!(nr.direct_branch_map, reverse_new_arc)
	@debug "Bus merge created anti-parallel collision: remapped arc $new_arc conflicts with existing $reverse_new_arc; promoting $(get_name(existing)) and $(get_name(val)) to a parallel group under $reverse_new_arc."
	if haskey(nr.parallel_branch_map, reverse_new_arc)
	_push_parallel_branch!(nr.parallel_branch_map, reverse_new_arc, existing)
	_push_parallel_branch!(nr.parallel_branch_map, reverse_new_arc, val)
	else
	nr.parallel_branch_map[reverse_new_arc] =
	_make_parallel_branch_pair(existing, val, reverse_new_arc)
	end
	elseif haskey(nr.parallel_branch_map, reverse_new_arc)
	@debug "Bus merge created anti-parallel collision: remapped arc $new_arc conflicts with existing parallel group at $reverse_new_arc; adding $(get_name(val)) to that group."
	_push_parallel_branch!(nr.parallel_branch_map, reverse_new_arc, val)
	else
	nr.direct_branch_map[new_arc] = val
	end

(same pattern in the parallel→reverse cases at L1676-L1685). For symmetric lines Y11=Y22 and Y12=Y21, so the positional sum is unaffected. For asymmetric branches (transformers, phase shifters), Y11≠Y22 and Y12≠Y21, so the reverse-oriented member's tap/shunt admittance lands on the wrong port and get_equivalent_physical_branch_parameters returns an incorrect equivalent Ybus. Consider swapping the reverse member's (Y11,Y12,Y21,Y22) to (Y22,Y21,Y12,Y11) before grouping, or normalizing orientation, so the L478 invariant continues to hold.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}