Log-Scaled Hilbert Curve → SWMM5 Network Generator

Follows the traversal order of a Hilbert curve with step length = scale × [ln(k+2) − ln(k+1)] and exports the result as a real SWMM5 conduit chain — built to stress-test the dynamic wave solver across an extreme conduit-length ratio.

UnitsiSets labels and the SWMM5 FLOW_UNITS option: US (ft, cfs → CFS) or SI (m, cms → CMS). Switching resets scale, floor, invert, diameter, inflow, and celerity to unit-appropriate defaults. Settings persist between visits where the browser allows it.

Curve order (n)iHow many recursion levels the Hilbert curve has. Cell count = 4^n, so each +1 quadruples the network size. 4

4^n cells → 4^n−1 junctions + 1 outfall, 4^n−1 conduits. Order 6 = 4,095 conduits.

Length scaleiReal-world multiplier on the raw log step. Bigger scale pushes more conduits into the "fine" (blue) band; smaller scale pushes more into "critical" (red). 150 ft

Multiplies the raw log-step ln(k+2)−ln(k+1).

Minimum conduit lengthiThe raw log step shrinks toward zero as k grows. This floor keeps every conduit a real, nonzero length — SWMM5 won't accept a literal 0. 0.5 ft

Floor to avoid a literal zero-length conduit as k grows.

DirectioniThe curve is one continuous path, J0 → outfall. This just picks which end of that path gets the long conduits and which gets the tiny ones.

Flips which end of the single flow path carries the long vs. tiny conduits.

Global slopeiConstant drop per unit length applied along the whole chain, so every junction's invert steps down toward the outfall — no adverse slopes. 0.5%

Starting invertiInvert elevation at J0, the head junction. Every downstream invert is derived from this minus slope × cumulative length. 100 ft

Pipe materialiPresets that fill the Manning's n field below with a typical value for that material. You can still edit n by hand afterward.

Manning's niRoughness written to every conduit in [CONDUITS] — one uniform value along the whole chain. Typical closed-pipe range 0.011–0.024. Editing this switches the material to Custom. 0.013

Pipe diameteriUniform CIRCULAR diameter applied to every conduit in [XSECTIONS] — this generator doesn't taper diameter along the chain. 1.0 ft

Peak inflow pulseiPeak of the triangular hydrograph injected at head junction J0: zero at 0:00, peak at 0:15, back to zero at 0:30 — written to [TIMESERIES] and [INFLOWS]. 2.0 cfs

Triangular pulse at the head junction (peaks at 15 min) so the solver has something to route.

Assumed wave celerityiRough wave speed for the on-screen Δt = L/c color bands only. SWMM5 computes its own celerity from depth and geometry at runtime — this value never enters the .inp. 8 ft/s

Used only for the on-screen Courant estimate below — not fed to SWMM5.

LENGTHENING_STEPiExported to [OPTIONS]. 0 = no artificial lengthening (baseline). Nonzero values lengthen short conduits for stability at the cost of phantom storage. Export once at 0 and once at 1 s for an A/B pair — the filename carries the value.

INERTIAL_DAMPINGiExported explicitly so runs are reproducible across engines. NONE keeps the full inertial terms (hardest test), PARTIAL dampens them as the pipe nears full, FULL drops them entirely.

SURCHARGE_METHODiExported explicitly and stamped into the filename when non-default, so the surcharge method is a controlled variable, not a hidden one. Keeping SLOT fixed controls comparisons; results can differ under EXTRAN.

REPORT_STEPiOutput interval only; the routing step is separate. Shorter intervals give finer hydrographs but larger .out files. No report interval resolves subsecond routing oscillations; use the routing-step summary and instability indexes for that.

Batch experiment exportiExports one .inp per combination of the checked values (other settings from the controls above), plus a JSON manifest with an experiment set ID, parameters, and sha256 checksums. Filenames carry the varied options.

LENGTHENING_STEP (s)

0 0.5 1 5

INERTIAL_DAMPING

NONE PARTIAL FULL

SURCHARGE_METHOD

SLOT EXTRAN

fine (ΔtC ≥ 5 s) marginal (0.5–5 s) critical (ΔtC < 0.5 s min step)

Celerity-only screening index, not the SWMM routing timestep. ΔtC = 0.75 × L / c with your assumed c. Hover any segment for its conduit ID, length, and ΔtC; the .rpt is the truth.

Benchmark — SWMM .rpt reader iRun the exported .inp files in SWMM 5.2+, then drop the .rpt reports here. Parses continuity error, timestep stats, non-convergence, instability indexes, outfall peaks, and warnings, side by side. Green = best, red = worst.

Drop .rpt reports, manifests (.json), or .inp files — whole folders work — or

Adaptive loop — in-browser SWMM iCloses propose → run → score → propose without leaving the page: a genuine EPA SWMM engine (WebAssembly, build 5.2.2) runs each probe and a bisection strategy narrows the bracket. Assumes the metric responds monotonically to the knob; the two endpoint probes test that assumption before any bisecting. Every probe records its experiment ID, input and report sha256, and engine build.

Engine: not loaded

Objective

Bracket [lo, hi] and max probes

Endpoints run first; bisection only proceeds if lo fails and hi passes.

The WASM engine is EPA SWMM 5.2.2; the verification corpus pins 5.2.4. Stable configurations reproduce exactly across builds, chaotic ones can differ (order-4 len0 measured 4.81% continuity on 5.2.2 vs 3.30% on 5.2.4) — the engine build is recorded per probe and shown in the log for exactly this reason.

Why this stresses SWMM5 (and what it actually is)

Precision about the geometry: the app takes the traversal order of a Hilbert curve, then assigns each step a different physical length. Once the lengths vary, the drawn path is no longer a true Hilbert curve — in the length-proportional layout it can even cross itself (verified: 2 crossings at order 4 defaults, 58 at order 5). The SWMM topology doesn't care: crossing lines in [COORDINATES] connect nothing unless they share a node, and this network remains a clean serial chain, J0 through the outfall. The "True Hilbert grid" mode draws the classic non-crossing curve on the uniform grid instead, exploiting the fact that a SWMM conduit's Length attribute is independent of its map distance — every real model with schematic coordinates works the same way.

The network is one continuous serial flow path, not a branching collection system. What it gives you, in a single connected model, is a conduit-length ratio spanning two to three orders of magnitude. That geometry exercises SWMM5's VARIABLE_STEP Courant adaptation, LENGTHENING_STEP (settable in the controls; export at 0 and again at 1 s for an A/B pair), INERTIAL_DAMPING, and the very-short-conduit handling in the dynamic wave solver — things a normal hand-built test network rarely covers in one file. The color bands use ΔtC = 0.75 × L / c with your assumed celerity, mirroring the VARIABLE_STEP safety factor, but they are screening estimates: the .rpt file is the truth.