This page is organized around one customer-facing question: once timestep becomes coarse enough to matter in practice, does TDSE On remain closer to gold than TDSE Off? We begin with one flagship public case, then extend the evidence through cross-case support, multiport progression, supporting event families, and deliberately caveated scale results.
Flagship Benchmark
One public benchmark, pushed until the separation becomes undeniable.
The flagship benchmark is built to answer one customer-facing question with as little storytelling overhead as possible: once timestep or event bandwidth becomes coarse enough to matter in real runtime, does TDSE On stay closer to gold than TDSE Off? Everything in this chapter stays anchored to one public IEEE118 setup, then stresses it in two controlled directions.
Flagship Setup
IEEE118, Gaussian burst, one-port at Bus 12.
The public test system is IEEE118. The exposed boundary is a single port at Bus 12. The disturbance is a Gaussian burst injected through that same boundary while runtime semantics, boundary definition, and solver context stay fixed.
This is the controlled lab bench for the whole page. We are not switching systems, changing boundaries, or moving between unrelated disturbances. We start with one clean public setup, then vary one axis at a time: first burst frequency, then timestep.
Frequency Sweep
One fixed benchmark, six increasingly faster disturbances.
This section keeps the IEEE118 Bus 12 one-port benchmark fixed and increases only the burst frequency. The purpose is simple: show, in the same case, when TDSE On and TDSE Off stop looking similar and begin to separate in a way a customer would actually care about.
Each tab is one representative frequency from the same study. Waveform appears on the left, spectrum on the right, and the corresponding waveform and spectrum error are shown underneath.
Waveform at 1000 Hz
Spectrum at 1000 Hz
GoldTDSE OnTDSE Off
Waveform error
TDSE On0.0268
TDSE Off0.2939
Spectrum error
TDSE On0.0082
TDSE Off0.0599
The tabs above show the raw evidence at six representative frequencies. The two summary curves below compress the same sweep into one waveform-error trend and one spectrum-error trend. Taken together, they show the same pattern: once the disturbance stops being gentle, TDSE Off degrades much faster than TDSE On.
Waveform error vs burst frequency
Spectrum error vs burst frequency
TDSE OnTDSE Off
Timestep Sensitivity
The same case crosses over once timestep leaves the fine-step regime.
Frequency sweep answers what happens when the disturbance gets faster. The next customer question is what happens when the timestep itself gets coarser. This study keeps the same IEEE118 Gaussian benchmark fixed at a 2000 Hz burst and repeats it across 22 timestep settings, each with 50 randomized event centers, so the result is a statistical trend rather than a single lucky waveform.
The pattern is now clear. In the fine-step regime, TDSE Off can still sit closer to gold. Around 20 to 22 microseconds, the two paths cross. After that, TDSE Off accumulates burst-core error much faster, while TDSE On remains comparatively controlled across the coarse-step range.
Error inside the burst core vs timestep
TDSE OnTDSE Off
Multiport Evidence
The clean next rung is a two-port operator on the same full public system.
Before asking TDSE to carry a partitioned host-versus-equivalent split, the cleaner question is whether the flagship one-port story survives when the same full IEEE118 system is promoted to a two-port operator. This chapter therefore starts with the stronger control experiment: the whole system still goes into TDSE, one port is driven, and both boundary voltages are checked. The harder partition benchmark remains important, but it is a later rung, not the first multiport claim.
Full-Deck Two-Port Benchmark
Keep the whole IEEE118 system inside TDSE, then promote the interface from one port to two.
This is the cleanest way to test the move from one port to two. Nothing is partitioned out yet. The whole IEEE118 network is still compressed into TDSE, the exposed interface is promoted to ports 12 and 16, and the Gaussian burst is injected through port 12 while both port voltages are observed.
That matters because it isolates the operator question. If the two-port operator is already clean in the full-deck setting, then later difficulties in a partitioned benchmark should be read as host-coupling and boundary-selection issues rather than as a failure of TDSE to move beyond one port.
System scope
The entire IEEE118 network remains inside the TDSE operator, exactly as in the one-port flagship line.
Observed ports
The interface is promoted to buses 12 and 16, so both boundary voltages can be checked under the same burst.
Drive model
Two Gaussian bursts are injected with partial overlap: port 12 leads with a 1600 Hz burst, and port 16 follows 0.5 ms later with a 2500 Hz burst.
Waveform at Bus 12
Spectrum at Bus 12
GoldTDSE OnTDSE Off
Waveform error
TDSE On0.0202
TDSE Off1.3112
Spectrum error
TDSE On0.0014
TDSE Off0.4806
Waveform at Bus 16
Spectrum at Bus 16
GoldTDSE OnTDSE Off
Waveform error
TDSE On0.0222
TDSE Off0.7258
Spectrum error
TDSE On0.0097
TDSE Off0.3194
What is already true
This result still removes partition complexity from the question. The whole public system lives inside TDSE, so the main step up from the flagship line is the move from one-port drive to a genuinely coupled two-port excitation.
What the figure shows
At both Bus 12 and Bus 16, TDSE On stays close to Gold even after the second burst starts to overlap the first, while same-step TDSE Off departs much more strongly. The overlap makes the interaction visibly more two-port, but the operator still stays controlled.
Why it matters
This is a stronger first multiport claim for the website than a single driven port. It shows that TDSE can stay close to Gold even when both ports are actively excited with overlapping bursts, so later partitioned difficulty can be read as a harder host-coupling problem rather than a failure of two-port TDSE itself.
This is the right place to stop the public multiport claim. The flagship one-port result already survives promotion to a genuinely coupled two-port interface on the same full IEEE118 system, and that is enough for this chapter to close on a clean customer-facing proof.
Event Family Support
The flagship claim already survives a fault-driven event family.
Gaussian remains the flagship family because it gives the cleanest controlled comparison. But the evidence line cannot stop there. The first supporting family to keep is a short-fault event on the same IEEE118 Bus 12 one-port setup, because it shows that the coarse-step advantage is still visible when the disturbance looks more like an operational fault than a synthetic burst.
Fault Support
The first supporting family is a short-fault event on the same IEEE118 one-port setup.
This study keeps the public IEEE118 Bus 12 one-port benchmark fixed and changes only the event family. The disturbance is a short fault instead of a Gaussian burst, so the page starts to move from a synthetic excitation toward a more operational event.
The audited supporting windows are the first-cycle view with switching edges excluded and the broader post-fault recovery window. Those are the two windows that remain clean enough to support the public claim, so they are the only ones shown here.
Representative Fault Case
A median-like 50 us fault case, shown in the same waveform-and-spectrum language as the flagship figures.
Before compressing the family into timestep trends, it helps to see one concrete short-fault realization. The example below is a representative 50 microsecond case drawn near the middle of the audited distribution rather than the cleanest-looking trial, so the picture stays honest.
Waveform at Bus 12
Spectrum
GoldTDSE OnTDSE Off
Waveform error
TDSE On0.1657
TDSE Off0.5391
Spectrum error
TDSE On0.0115
TDSE Off0.3360
Fault Current at Bus 12
Spectrum
GoldTDSE OnTDSE Off
Waveform error (2-point moving average)
TDSE On0.0579
TDSE Off0.0821
Spectrum error
TDSE On0.0248
TDSE Off0.0587
What the figure keeps
This section now stays at the representative-case level on purpose. The single fault example already shows the public claim we need: under a short fault on the same IEEE118 one-port setup, TDSE On still tracks Gold much more closely than same-step TDSE Off.
What it does not claim
Fault events are less uniform than the Gaussian flagship, especially once switching edges and very coarse steps start to dominate. That is exactly why this chapter stays as supporting evidence rather than turning into a second flagship benchmark.
Why it matters
The page no longer depends on one carefully chosen burst family. The same coarse-step separation is still visible under a fault-driven disturbance that reads more like engineering reality, which is enough to show that the flagship conclusion is not tied to Gaussian bursts alone.
Cross-Case Evidence
The flagship timestep pattern is visible beyond IEEE118, but not with identical maturity.
Cross-case evidence should stay narrow. The flagship chapter already did the heavy lifting on waveform, spectrum, and event design. What matters here is whether the same burst-core timestep story survives on other public systems, so each case is reduced to the same one summary view: error inside the burst core versus timestep.
Cross-Case Core Window
Each additional system is reduced to the same burst-core timestep view.
IEEE118 remains the flagship because it is easiest to audit, but it cannot be the only public-native case on the page. For cross-case support, the cleanest shared question is whether TDSE On stays below TDSE Off inside the burst core as timestep grows. That keeps the comparison narrow and avoids turning this section back into a gallery of unrelated figures.
ACTIVSg200: Error inside the burst core vs timestep
NREL240: Error inside the burst core vs timestep
TDSE OnTDSE Off
ACTIVSg200
ACTIVSg200 now uses the dense Bus 113 sweep rather than the old three-point screen. It stays in the same qualitative family as IEEE118, but the crossover comes later: TDSE Off remains better through 30 microseconds, the burst-core separation flips around 35 microseconds, and from there TDSE On stays below TDSE Off across the coarse-step range.
NREL240
NREL240 now has a full dense sweep rather than the old three-point screen. It behaves like IEEE118: TDSE Off is still better in the finest-step regime, the burst-core crossover lands around 20 to 22 microseconds, and from 22 microseconds onward TDSE On stays cleanly below TDSE Off as timestep grows.
Why it matters
The flagship timestep story is no longer confined to one public system. What changes across cases is not the existence of the advantage, but how early and how cleanly that advantage appears.
Scale
Scale is visible, but still deliberately caveated.
ACTIVSg2000 belongs on the site because it shows that the scale trajectory is real. It should not be flattened into the same claim tier as the flagship benchmark, because that qualification line is still in progress.
Representative scale-support result: port 5293
This figure shows that scale evidence is already visible on a 2000-bus public-native case.
Why it belongs here
It shows that scale is not a future promise. The evidence line already reaches a substantially larger case class.
Why it stays caveated
The broader scale line is still under qualification, so this figure supports the trajectory without pretending to close the entire claim.
Second representative scale-support result: port 5479
A second representative port helps keep the scale story from looking like one isolated result.
What it adds
It shows that the scale-support pattern has already started to replicate within the ACTIVSg2000 regime.
How to read it
This is support for scale visibility, not yet a headline customer-safe scale claim.
Why Trust This
The page should read like a governed program, not a pile of plots.
These figures are not being promoted ad hoc. They come from a shared superiority framework with explicit claim tiers, a common output schema, and visible caveat discipline.
Shared framework
The figures come from a shared registry, a shared driver, and a shared milestone pack rather than disconnected one-off experiments.
Claim boundaries
Flagship, supporting, technical-support, and scale-in-progress evidence are kept separate on purpose, so the website does not blur unlike evidence tiers into one another.
Customer-safe pack
The page is built from the current customer-safe figure pack, which keeps the public story strong without outrunning the qualification work behind it.
