Governing Physics & Scattering Mechanisms

Every RCS number in this project is a coherent complex-amplitude sum of three physical mechanisms, squared at the end, never added as independent powers. k=2\pi/\lambda is the free-space wavenumber, \hat s the unit look direction, \rho per-facet reflectivity (1 for bare PEC, <1 under a RAM/SIBC model).

1  What is a Facet, and Why Mesh the Aircraft at All?

What. Every aircraft surface here is discretized into flat triangular facets, the same faceted-mesh representation used in real RCS prediction codes (XPATCH, POFACETS, FEKO's PO solver). Each facet i carries a centroid \mathbf c_i, an outward unit normal \hat n_i, and an area A_i, exactly the same three numbers a structural-FEM mesh element carries, just used for a different physics.

Why not solve Maxwell's equations exactly on the real curved surface? Because an exact solution (the Method of Moments, MoM) requires solving a dense linear system with one unknown per roughly \lambda/10 of surface, for a full-size aircraft at X-band that is tens of millions of unknowns, computationally intractable on ordinary hardware. Faceting trades that exactness for tractability: replace the smooth surface with flat triangles small enough that, locally, the wave "sees" each one as if it were an infinite flat plate (the tangent-plane approximation underlying PO). The mesh-convergence studies on this site are exactly the check that this trade-off was made safely, i.e. that the facets ended up small enough relative to the wavelength that the approximation error is negligible.

How illumination is decided. PO sums a phased contribution from every illuminated facet (\hat n_i\cdot\hat s > 0, meaning the facet's outward normal has a positive component back toward the radar); shadowed facets (facing away) contribute nothing to first order. Finer meshing (higher subdivision) approximates curved/complex surfaces more accurately at the cost of O(n) to O(n^3) compute depending on which physics layer is active, PO's contribution is a simple vectorised sum, but double- and triple-bounce require checking pairs and triples of facets against each other, which is where the cost explodes.

2  Physical Optics (PO), Specular Reflection

What it models. The dominant "mirror-like" glint you get when a flat (or locally-flat) surface happens to be oriented so the incident ray reflects straight back at the radar. This is the single largest RCS contributor for almost any aircraft at almost any aspect, it is why a vertical tail fin lights up like a beacon at broadside and why a curved fuselage still returns a bright, narrow glint at the exact angle where its local surface normal points at the radar. The coherent PO scattered-field amplitude is a phased sum over all illuminated facets, each weighted by its area, its obliquity to the radar, and a propagation-phase term that encodes exactly how far that facet's return has to travel relative to every other facet's return (this phase term is what lets facets constructively or destructively interfere, the whole reason a coherent sum, not a power sum, is required):

How it scales, and why that matters for shaping. PO return scales with facet area and incidence angle as \sigma \propto A^2(\hat n\cdot\hat s)^2/\lambda^2, large flat panels facing the radar dominate (this is why a 3×2 m vertical stabiliser alone returns ~+57 dBsm broadside: doubling a panel's linear dimension quadruples its area, but because RCS goes as area squared, it actually increases the return by a factor of sixteen). This quadratic-in-area law is precisely why stealth designers prefer many small faceted panels, or curved/blended surfaces with no single large flat region, over one big smooth panel, splitting one large panel into several smaller, differently-angled ones can turn one huge specular spike into several much smaller ones scattered to different, less threatening angles.

Where it is used in practice. This is the workhorse layer of essentially every real high-frequency RCS prediction code, because it is cheap (a single vectorised sum, no per-pair or per-triple looping) and gets the dominant physics right for electrically-large smooth targets (aircraft, ships) at microwave frequencies.

Limitation. PO-only computation underestimates RCS: it misses edge and multi-bounce contributions present in real geometries, and it has no concept of a shadow boundary's diffracted field (it just switches a facet fully on or fully off at the illumination boundary, which is physically discontinuous). Every subsequent layer below corrects a specific, named blind spot of PO.

3  Physical Theory of Diffraction (PTD), Edge Diffraction

Why it is needed. PO's tangent-plane approximation is only valid deep inside a flat facet; right at an edge, a wingtip, a trailing edge, the rim of an engine inlet, the surface is discontinuous and PO's prediction is not just imprecise, it is formally wrong (it predicts a field discontinuity across the shadow boundary that cannot physically exist, since Maxwell's equations forbid an infinitely sharp field transition). Ufimtsev's Physical Theory of Diffraction fixes this by adding a correction "fringe wave" localized to the edge itself, so the combined PO+PTD field is continuous everywhere, matching what a full-wave solution shows.

How it is computed. Ufimtsev's fringe-wave diffraction coefficient depends on the wedge's interior angle via a parameter n (n=2 for a free/isolated edge such as a wingtip open to free space on both sides, n=1.5 for a right-angle convex ridge such as a wing-fuselage seam) and on the incidence angle \beta measured from the edge:

Diffraction lobes from a finite edge of length L follow a sinc envelope in aspect angle, the same sinc pattern that shows up in antenna-array theory and slit diffraction, because the underlying mathematics (a coherent sum along a line) is identical:

Where it matters for shaping. This is exactly why stealth aircraft leading and trailing edges are all swept to a small number of common angles ("edge alignment"), every edge's diffraction lobe then points in the same handful of directions, so there are only a few narrow "glint" aspects to defend against instead of edge energy smeared continuously across all azimuths.

Adding PTD raises the mean RCS by capturing edge diffraction from geometric discontinuities that PO alone is blind to, visible as the elevated diffuse floor in every azimuth sweep on the Scattering Mechanisms tab.

4  Double-Bounce, Corner Reflector

What it is and why it is dangerous. Two flat facets meeting at close to a right angle form a dihedral corner reflector, the same geometric principle used deliberately in navigation radar reflectors on small boats (a cheap, lightweight way to make a fibreglass boat visible to ship radar) and in truck-mounted calibration targets. On an aircraft, a wing-fuselage junction or a fin-fuselage junction is an accidental version of the same thing: a ray hits the first facet, reflects onto the second facet, and reflects a second time straight back along its incoming path, retroreflection, the same effect that makes a bicycle reflector or road sign so bright when lit head-on by car headlights. A facet pair (i,j) is an active double-bounce contributor when the retroreflection condition holds:

where \mathbf d_{ref,j} is the ray reflected off facet i then off facet j. Why this matters so much: a corner reflector's RCS does not fall off nearly as fast with mis-orientation as a single flat plate's does, it stays retroreflective over a much wider range of incidence angles, which is exactly why it introduces sharp, broad RCS spikes at quarter-on aspects (45°/135°) driven by dozens of simultaneously-active retroreflecting facet pairs, rather than the razor-thin spikes PO alone predicts from a single flat panel. PO+PTD alone completely miss this mechanism because it requires tracing a ray through two facets, not evaluating each facet independently.

5  Triple-Bounce, Trihedral Corner Reflector

What it is. Extending the ray trace one more reflection captures trihedral corners, three mutually near-perpendicular facets meeting at a point, exactly the geometry of a surveyor's or marine trihedral corner reflector (three metal or mesh panels folded like the inside corner of a cube), which is deliberately used as a radar calibration target precisely because it is such an efficient, wide-angle retroreflector. On an aircraft this three-way geometry shows up accidentally at a wing–fuselage–tail junction where three surfaces happen to meet at close to mutually perpendicular angles.

Why it returns even more energy than a dihedral. A trihedral reflects the incoming ray back along its exact incoming path over a much wider solid angle of incidence than a dihedral does (a dihedral only retroreflects within the plane perpendicular to its fold line; a trihedral retroreflects over a genuine 3-D cone of angles), and its peak RCS scales with the fourth power of its characteristic edge length L rather than a weaker power for a simple dihedral:

Textbook-consistent finding: trihedral (triple-bounce) returns exceed dihedral (double-bounce) returns at their respective peak angles, \sigma\propto L^4/\lambda^2 for a trihedral vs a weaker scaling for a simple dihedral. Confirmed numerically on the Multi-Bounce tab (not a bug) by explicitly counting how many contributing facet triples are active at the peak angle before accepting the result.

6  Bistatic Generalisation

What "bistatic" means and why it is a separate case at all. Every equation above assumes the transmitter and receiver are co-located (monostatic radar, the ordinary case, one antenna doing both jobs). A bistatic radar instead has a physically separate transmitter and receiver, sometimes hundreds of kilometres apart. This matters for stealth because shaping a target to defeat monostatic detection (redirecting energy away from the transmitter) does not automatically defeat a receiver sitting somewhere else entirely, energy redirected away from one direction has to go somewhere, and a bistatic receiver can be waiting exactly there.

Where this is used in the real world. "Passive bistatic radar" or "passive coherent location" systems deliberately exploit this: they use existing FM radio or digital TV broadcast towers as the illuminator and a separate, silent (hence undetectable) receiver to pick up the forward-scattered or bistatically-scattered energy , a technique specifically discussed as a potential counter-stealth approach in open literature, because stealth shaping is optimized primarily for the monostatic backscatter case.

How the model generalises. All mechanisms above generalise to separate transmitter/receiver directions via the incidence–scatter bisector, physically, a facet reflects specularly relative to the bisector of the incoming and outgoing ray directions, exactly like a mirror, so the obliquity and phase terms are evaluated against that bisector instead of a single direction:

At \hat s_{rx}=\hat s_{tx} (transmitter and receiver in the same place) this reduces exactly (bit-for-bit, verified in the self-check) to the original monostatic formula, a mandatory sanity check, since a bistatic formula that does not collapse correctly to its monostatic special case is provably wrong.

7  Polarimetric RCS

What it is and why a single RCS number is not the whole story. Radar antennas transmit a specific polarization (horizontal H or vertical V) and can also receive on either polarization. A real target can shift some of the reflected energy from the transmitted polarization into the orthogonal one, "depolarization" or "cross-pol return", and how much it does so is itself a distinguishing physical signature, not noise. A full 2×2 complex scattering matrix relates incident to scattered field in a fixed global (H,V) basis:

How depolarization physically arises. Each PEC facet reflects with the textbook TE/TM dichotomy (r_\perp=-1,\ r_\parallel=+1) in its own local plane-of-incidence basis; projecting that local reflection into the fixed global frame is exactly what produces cross-pol (HV/VH). A facet square-on to the radar has no defined plane of incidence and stays pure co-pol; a tilted facet or dihedral corner has a rotated local basis and leaks energy into the cross channel, the same mechanism that makes real corner reflectors depolarizing. Multi-bounce mechanisms use cascaded Jones matrices, e.g. double-bounce: S_{DB} \propto J_j J_i, triple-bounce: S_{TB}\propto J_k J_j J_i.

8  RAM, Salisbury Screen & Dallenbach Layer

What RAM is and why shaping alone is not always enough. Radar-Absorbent Material is a coating or layer applied to a surface specifically to convert incident RF energy into heat (dielectric or resistive loss) instead of reflecting it. Shaping cannot help everywhere, some surfaces (a necessarily-flat weapons-bay door, an engine inlet lip) cannot be angled away from every possible threat direction without breaking some other requirement, and RAM is the tool that reduces the return from exactly those surfaces without changing their shape. Both classical designs modelled here are transmission-line reflection problems: a resistive/lossy layer over a PEC backplane (representing the aircraft's underlying metal structure), characterised by an input impedance Z_{in} and reflection coefficient

Salisbury screen, a single resistive sheet (R_s\approx Z_0) placed a quarter-wavelength in front of a PEC backplane, with an air gap in between. The short-circuited quarter-wave line transforms the PEC's zero impedance to infinite at the sheet, so all incident energy is dissipated resistively at that one frequency, deep but narrowband absorption. This is the oldest practical RAM design (1940s) and is still conceptually the basis for many radar-absorbing panels used at fixed test-range or calibration frequencies.

Dallenbach layer, a single lossy dielectric slab (complex \epsilon_r, meaning it has both a normal dielectric response and an imaginary loss component) of thickness d directly on the PEC backplane, no air gap required, simpler to bond directly onto an airframe skin:

Loss is distributed through the slab volume rather than concentrated at one sheet: broader but shallower absorption than Salisbury, i.e. it trades peak absorption depth for a wider useful frequency band, which is generally the more practical trade-off for an aircraft that must stay stealthy across an entire radar band, not just one calibration frequency.

9  Surface Impedance Boundary Condition (SIBC)

Every mechanism above assumes a Perfect Electric Conductor (\rho=1, total reflection). Real airframe skins are finite-conductivity metals/composites, the Leontovich SIBC replaces the PEC assumption with a frequency-dependent surface impedance.

\sigma here is electrical conductivity (S/m, not RCS), \mu permeability, \delta skin depth. As \sigma\to\infty, Z_s\to0 and \Gamma\to-1, recovering PEC exactly, verified in the self-check.

Why it's required. Modern airframes increasingly use CFRP composites (conductivity ~1000× lower than aluminum) instead of aluminum skins, a PEC-only model cannot distinguish them. SIBC plugs a real, measurable material property directly into every existing solver via the same interface used for Salisbury/ Dallenbach RAM, with zero physics-layer rewrite.

10  Geometry Optimization

Why automate the search at all. Everything up to this point treats the stealth aircraft's shape as fixed, a human designer picked the sweep angle, the tail cant, and the nose length by hand, informed by the design principles in §3–5 above. But those design principles only say the direction to move a parameter (sweep more, cant more); they do not say the specific optimal value, and RCS depends on all three parameters simultaneously through a landscape that is far too complex to search by hand. This section asks: if we let the computer search that parameter space automatically, can it beat the hand-tuned baseline, and by how much?

How the problem is set up. Stealth geometry is parametrized by wing leading-edge sweep angle \Lambda, V-tail cant angle \gamma, and nose length L_n, the three parameters the design-principles discussion above identified as the direct geometric levers on RCS. The objective is the linear-domain mean RCS over an azimuth sweep (correct averaging convention, dBsm values themselves must never be arithmetically averaged, since decibels are a logarithmic, not linear, scale; averaging them directly systematically under-weights the loud angles that actually matter most for detectability):

Why differential evolution, not gradient descent. Solved with \texttt{scipy.optimize.differential\_evolution}, a gradient-free, population-based global optimizer. This choice is deliberate, not arbitrary: a faceted PO/PTD RCS landscape is non-convex (many local minima as different corner reflectors and specular glints switch on and off with angle) and non-smooth (a facet flipping from illuminated to shadowed, or a double-bounce pair switching on, creates a genuine kink or jump in the objective function). Gradient-based methods (steepest descent, BFGS) implicitly assume the function is smooth enough to have a well-defined local slope everywhere and can get stuck in, or fail outright at, those discontinuities; differential evolution instead maintains a whole population of candidate parameter sets and combines/mutates them, needing no derivative information at all.

Where this mirrors real aircraft design practice. Modern aircraft design increasingly uses exactly this kind of multi-disciplinary automated optimization (MDO), letting an algorithm search a shape's free parameters against a stated objective (RCS, drag, structural weight) rather than relying purely on a human designer's iteration-by-hand, though a real MDO loop runs alongside aerodynamic and structural constraints this teaching-scale version deliberately leaves out to keep the search tractable and its result explainable.

11  Wideband Impulse Response

Why a single frequency is not enough. A single number cannot tell you where along the aircraft the energy is coming from, the nose, a wing leading edge, and the tail can all contribute to the same one-number total, indistinguishably. Real radars that need that spatial information (range profiling, target recognition) do not transmit one frequency; they step or sweep across a band and combine the results. This section reproduces exactly that processing chain.

How it works. Instead of a single-frequency RCS number, sweep the same coherent complex amplitude across a frequency band and inverse-Fourier-transform it into a time-domain range profile, the standard stepped-frequency radar processing chain used to resolve individual scattering centres along the target's line of sight. Physically: each frequency in the sweep is a different "comb tooth"; Fourier theory guarantees that combining many closely-spaced frequency-domain samples and inverse-transforming them synthesizes a short time-domain pulse, exactly as if a single very-short, very-wideband pulse had been transmitted and its echo recorded directly , without ever needing hardware capable of generating that short a pulse.

Geometry & Meshing

Six triangulated-mesh PEC geometries, all built from first-principles facet primitives (\texttt{geometries.py}): quads subdivided into paired triangles, triangular nose/tail caps, welded only within each primitive so wing–fuselage junction edges stay correctly classified as boundary edges.

Why six shapes and not just "conventional vs stealth"? Each of the other four isolates one specific electromagnetic idea so its effect can be seen in isolation before looking at the two full aircraft: the sphere is the one shape with an exact closed-form answer (used purely to validate the solver, see §4 below); the flat plate is the simplest possible specular reflector, used as a textbook-formula sanity check; the flying wing demonstrates a shape with zero vertical surfaces at all (the most radical stealth shaping choice, used on the real B-2 Spirit); and the chined-nose shape demonstrates blending the fuselage directly into the wing leading edge (used on the real F-22/F-35 family) as an alternative to a separate, sharply-defined nose cone.

1  Target Geometries

Sphere, iso

Flat plate, iso

Conventional, iso

Stealth, iso

Flying wing, iso

Chined nose, iso

2  Multi-View Detail (Conventional vs Stealth)

Conventional, front

Conventional, side

Conventional, top

Conventional, wireframe (facets)

Stealth, front

Stealth, side

Stealth, top

Stealth, wireframe (facets)

Flying wing & chined nose

Flying wing, wireframe

Flying wing, top

Chined nose, wireframe

Chined nose, top

3  Edge Classification

Why classify edges at all. Recall from the Governing Physics tab that PTD's diffraction strength depends on the wedge parameter n, and n in turn depends on whether an edge is a "boundary" edge (open to free space on one side, like a wingtip or trailing edge, typically n=2) or an "internal" edge (a ridge where two facets of the same solid meet at some interior angle, like a wing-fuselage seam, typically n\approx1.5 for a right-angle ridge). Internal right-angle ridges diffract more strongly than free boundary edges, so simply counting how many of each type a geometry has is a fast, purely-geometric predictor of how "loud" its PTD layer is, before a single RCS calculation is run.

The conventional aircraft's 41 internal edges (many at n=1.5, right-angle convex ridges) drive elevated diffraction; the stealth aircraft's uniform n=2.00 reflects zero internal right-angle ridge structure , every single edge on the stealth geometry is a free boundary edge, a direct, quantitative fingerprint of its blended, seamless shaping philosophy that isn't obvious from the pictures alone.

Conventional, edge classification

Stealth, edge classification

4  Mie Validation (Why Trust the Solver)

Sphere, Mie series exact RCS, X-band

Sphere mesh (2304 facets) used for PO cross-check

5  Mesh Convergence, Baseline (PO+PTD+DoubleBounce)

Why bother checking this. The report's headline numbers use deliberately coarse baseline meshes (18–34 facets) chosen to keep the O(n²) double-bounce and O(n³) triple-bounce solvers fast enough for interactive sweeps across hundreds of angles. That raises an obvious, fair question a reviewer should ask: is 20–34 facets actually enough, or are the headline numbers an artifact of too coarse a mesh? A mesh-convergence study is the standard numerical-methods answer to exactly that question, refine the mesh repeatedly and check whether the answer stops changing. If it does not stop changing, the original result cannot be trusted regardless of how physically reasonable it looks.

Does RCS stabilise as facet count grows and facet size shrinks well below \lambda (X-band, 3 cm)? Three solver tiers by computational complexity class: PO alone (vectorised, O(n), scales to ~10⁸ facets), PO+PTD (edge extraction, Python loop, ~10⁶), PO+PTD+DoubleBounce (O(n^2) nested loop, ~10⁴). Click legend entries to toggle series.

Why the curve suddenly jumps at two points instead of smoothly flattening. Look closely and every convergence chart on this site has one or two sudden steps rather than a single smooth settle, those steps sit exactly where the plot crosses from one solver tier to the next (DoubleBounce→PTD around 8,704→34,816 facets, PTD→PO around 557,056→3,060,000 facets). This is not numerical error and not the mesh failing to converge. Double-bounce and PTD are too expensive to run at millions of facets (O(n^2) and a per-edge Python loop respectively), so past each tier boundary that mechanism is silently dropped from the sum purely to stay computationally tractable, the PO+PTD+DoubleBounce tier includes all three mechanisms, the PTD tier includes only PO+PTD, and the PO tier includes PO alone. Each step is therefore a real, honest accounting change (one whole physical mechanism switching off), not a defect in the plot; within each tier on its own, the curve does settle smoothly, which is the actual convergence claim being tested.

Scattering Mechanism Breakdown

Isolating each physics layer (PO-only, PTD edge, double-bounce) on the same axes reveals which geometric feature drives which mechanism, the diagnostic result the problem statement explicitly demanded (Challenge 3, Overview §2). This is the tab that turns "the total RCS is X dBsm" into "the total RCS is X dBsm, and here specifically is what caused it and where on the airframe."

The method is simple: compute the same azimuth sweep three separate times, once with only the PO layer switched on, once with only PTD, once with only double-bounce, and plot each against the full-physics total as a faint reference line. Wherever a single-mechanism curve tracks close to the full-physics line, that mechanism is the dominant contributor at that angle; wherever it sits far below, something else (usually one of the other two layers) is doing most of the work there instead.

1  PO-Only (Specular)

Conventional, PO only

Stealth, PO only (deep null at tail-on, 180°)

2  PTD Edge Diffraction

Conventional, PTD edge only

Stealth, PTD edge only (peaks at swept LE alignment)

3  Double-Bounce (Corner Reflector)

Conventional, double-bounce (4 sharp lobes at 0/90/180/270)

Stealth, double-bounce (broader, lower, canted-tail geometry)

Double-bounce hotspot count. Conventional aircraft: 48 active retroreflecting facet pairs clustered at the fin–fuselage 90° dihedral. Stealth aircraft: ~22 pairs at roughly half peak amplitude with no dominant cluster, the 35°-canted V-tail eliminates the dominant corner reflector rather than merely weakening it.

4  RAM Physics Diagrams

Salisbury screen, quarter-wave resistive sheet over PEC

Dallenbach layer, lossy dielectric slab over PEC

5  Design Principles Derived From the Breakdown

• Sweep leading edges. Shifts the diffraction peak by angle \Lambda off nose-on, moving the return outside the frontal threat corridor for \Lambda>60^\circ. Evidence: stealth aircraft's ~65° swept delta wings with aligned trailing edges.
• Cant vertical surfaces. A single design change simultaneously reduces specular return and eliminates the 90° fin–fuselage double-bounce corner. Evidence: stealth V-tails canted 35° outward.
• Combine shaping with RAM. Shaping redirects scattered energy away from the threat sector; RAM reduces the amplitude of remaining returns, Salisbury screens give deep narrowband absorption, Dallenbach layers give broader shallower absorption.

Salisbury vs Dallenbach absorption curves

Stealth aircraft + Salisbury screen @10 GHz

Bistatic & Polarimetric RCS

Extending every solver (PO, PTD, double-bounce, triple-bounce) to independent transmit/receive directions and to the full 2×2 polarimetric scattering matrix, both generalisations verified bit-for-bit to reduce to their monostatic/scalar forms when tx=rx. Formulation and real-world motivation for both extensions is in Governing Physics §6–7; this tab is the results.

1  Bistatic Sweep

Bistatic RCS vs receiver azimuth, tx fixed nose-on. 0°=monostatic backscatter, 180°=forward scatter (Babinet regime, not shape-reducible)

RCS vs bistatic angle β, tx fixed broadside

2  Polarimetric Signature

HH/VV co-pol vs HV/VH cross-pol, full azimuth sweep

Polarimetric signature by aspect, corner-reflector depolarisation

Flat plate at normal incidence: HH=VV, HV<−60 dB below co-pol (numerically zero, no plane of incidence, no depolarisation). Corner reflector at 45°: full-physics HV exceeds PO-only HV by >3 dB, the double-bounce cascaded-Jones term is what produces real cross-pol.

3  Mesh Convergence, Bistatic RCS

Same tiered solver strategy; bistatic pair az_tx=0°/el_tx=0°, az_rx=30°/el_rx=10°.

4  Mesh Convergence, Polarimetric (HH / HV)

Multi-Bounce: Double-Bounce vs Triple-Bounce

Extending the double-bounce ray trace one more reflection captures trihedral corners (wing–fuselage–tail three-way junctions), a real conventional-aircraft mechanism absent from the two-bounce model. Why trihedrals matter and how the peak-RCS formula is derived is covered in Governing Physics §5; this tab is the numerical evidence, including the honest admission that the triple-bounce mesh-convergence study below did not fully converge given practical compute limits.

1  Corner-Reflector Comparison (Full Azimuth)

All six geometries, overlaid polar RCS

2  Mesh Convergence, Triple-Bounce (Scalar)

Peak trihedral angles located by coarse sphere sweep: Conventional az=85°/el=60°, Stealth az=340°/el=60°. Dashed = double-bounce reference at the same angle for scale.

3  Mesh Convergence, Triple-Bounce Polarimetric

4  Physical Interpretation

Peak trihedral RCS exceeds peak dihedral RCS for both geometries at their respective optimum angles, textbook-consistent (\sigma_{trihedral}\propto L^4/\lambda^2, stronger scaling than a simple dihedral). Verified this is genuine physics (via contributing-triple counts), not a bug, before reporting.

RAM & Surface Impedance Boundary Condition

PEC vs finite-conductivity real aerospace skins, X-band and swept frequency, using the Leontovich SIBC (see Governing Physics §9) plugged into every existing solver via the same interface as Salisbury/ Dallenbach RAM.

Every plot before this tab assumed a Perfect Electric Conductor, a mathematical idealisation with infinite conductivity that reflects 100% of incident energy. Real airframe skins are not PEC: aluminum, titanium, and especially modern carbon-fibre composite (CFRP) skins are finite conductors. This tab asks the natural next question a materials-aware reviewer asks: does that idealisation actually matter, or is PEC a safe enough approximation for RCS work? The answer, worked out quantitatively below rather than assumed, turns out to be aspect- and frequency-dependent, PEC is an excellent approximation at X-band for every material tested, but the approximation visibly degrades as frequency rises.

1  Real Material Conductivities Used

Approximate, room-temperature, commonly-cited engineering reference values.

2  PEC vs Real Materials, Azimuth Sweep, X-band

RCS vs azimuth, PEC vs 4 real materials, both aircraft

Mean RCS drop vs bare PEC, azimuth-averaged

Honest finding. At X-band, even CFRP (the worst real conductor tested) is near-indistinguishable from PEC (sub-0.06 dB), skin depth is orders of magnitude smaller than facet size for every material here. This is a textbook-consistent result, not an implementation shortfall.

3  Frequency Dependence

Reflection loss vs frequency, L–Ka band, per material

Reflection loss grows with frequency (not shrinks): CFRP loss goes −0.017 dB (1 GHz) → −0.053 dB (10 GHz) → −0.100 dB (36 GHz). Since Z_s\propto\sqrt f while Z_0 is fixed, the impedance mismatch grows with frequency, the opposite of a naive "skin effect gets stronger so loss should look bigger at low f" intuition, but correct once the actual Z_s(f) scaling is followed through.

4  Mesh Convergence, PEC vs CFRP

Broadside aspect (90°), full tiered solver sweep. Solid = PEC, dashed = CFRP composite.

Stealth Geometry Optimization

Gradient-free optimization of wing sweep, V-tail cant, and nose length to minimize azimuth-averaged mean RCS, formulation, and why differential evolution rather than gradient descent, in Governing Physics §10.

This tab answers a question the design-principles discussion (Scattering Mechanisms §5) could only answer qualitatively: given that sweeping the wing and canting the tail both help, what is the actual best combination of angles, and how much better is it than the hand-picked baseline? Letting an optimizer search the full 3-parameter space removes human guesswork from that specific question.

1  Optimizer Convergence

Differential-evolution best-so-far mean RCS vs generation

2  Result

Baseline vs DE-optimized azimuth RCS sweep, X-band

Baseline geometry kept bit-exact reproducible (parametrization defaults to the original hardcoded coordinates) so this optimization result is a genuine A/B comparison, not a re-derivation of the report's baseline numbers.

3  Why These Specific Values

The three parameters do not behave the same way, and a single 1D slice through the objective at the optimum, plus a 2D slice over sweep and cant, shows why each one landed where it did.

Mean RCS vs each parameter, the other two held at their optimal value, 120-point sweep per axis

Mean RCS landscape over sweep x cant, nose length fixed at its optimum

Why the cant notch is so narrow. The double-bounce retroreflection condition (Governing Physics §4, \mathbf d_{ref,j}\cdot\hat s>0.2) is a geometric threshold on a specific facet pair. A fraction of a degree of V-tail cant is enough to flip that facet pair between satisfying and failing the condition, so the coherent double-bounce contribution switches on or off almost like a step function rather than fading smoothly. That is also the direct, practical reason a gradient-free optimizer was required (Governing Physics §10): a gradient-based method sitting one grid step away from a notch this narrow has no local slope pointing toward it. It would need to already be inside the notch to find it, which a global population-based search like differential evolution can stumble into but a local descent method generally cannot.

None of this changes the headline result, 48.44 dBsm against a 51.58 dBsm baseline is real and reproducible at these exact parameter values. What it does show is that "the optimizer chose 59.79/19.95/0.50" is not three independently meaningful design decisions: it is one sharp, load-bearing cant angle, one parameter pinned against its search bound, and one direction that barely matters at all.

4  Mesh Convergence at the Optimum Parameter Set

Wideband Impulse Response

Formulation, and why a single frequency cannot show where on the airframe a return originates, in Governing Physics §11. 2–18 GHz sweep (S through Ku band, 16 GHz bandwidth), coherent PO+PTD+DoubleBounce amplitude at every frequency point.

1  Frequency-Domain Response

Nose-on wideband RCS vs frequency, 2–18 GHz, interference ripple from multiple coherent scattering centres

2  Range Profile (Impulse Response)

Unlike a single-frequency RCS number, the range profile shows where along the target the energy returns from, nose tip, wing leading edge, and tail each appear as a distinct peak instead of one aggregate scalar.

3  Mesh Convergence, Wideband Energy

Broadside aspect, mean |H(f)|^2 across the band (32-point reduced sweep for the O(n^2) DoubleBounce tier).

Results & Comparison

1  Mean RCS Ranking (X-band, ascending)

Mean RCS ranking, all six geometries

RCS reduction, Conventional → Stealth, by aspect

2  RCS Reduction: Conventional → Stealth

Nose-on: a frontal radar now sees only 6.5% of the energy a conventional aircraft returns. A radar detecting the conventional aircraft at 200 km first detects the stealth aircraft at only 100 km, a 50% reduction in detection range from an 11.9 dB nose-on RCS reduction ( R\propto\sigma^{1/4}).

Broadside is the one aspect where shaping does not help (−0.8 dB, i.e. slightly worse): the V-tail cant partially re-radiates energy back toward the radar at exactly this angle, an honest trade-off, not hidden in the summary numbers.

3  Frequency Sweep (Nose-on Reduction)

Conventional, RCS vs frequency, nose-on

Stealth, RCS vs frequency, nose-on (5.7 dB quieter at L-band)

Conventional, RCS vs frequency, broadside

Stealth, RCS vs frequency, broadside

4  RCS Heatmaps (Azimuth × Elevation)

Conventional, full angular heatmap

Stealth, full angular heatmap

5  Per-Geometry Full-Physics vs PO-Only

Conventional, full physics (PO+PTD+DB)

Conventional, PO only

Stealth, full physics (PO+PTD+DB)

Stealth, PO only

Flying wing, full physics

Chined nose, full physics

6  Conclusion

Bringing every mechanism, extension, and convergence study on this site together, the results support one consistent physical narrative: the stealth aircraft's radar quietness is a shaping story first, a materials story a distant second, and that story now holds up under mesh refinement, not just at the coarse report-baseline mesh it was originally measured on.

Core validated findings

• Solver correctness. The Physical Optics implementation matches the exact Mie series solution for a conducting sphere to within 0.2 dBsm, the one scattering problem with a closed-form answer, and therefore the one no-excuses check the whole framework had to pass before any comparative claim about conventional vs.\ stealth could be trusted.
• Nose-on reduction. The stealth aircraft achieves an 11.9 dB RCS reduction at nose-on over the conventional aircraft, via the fourth-root range–RCS relation, equivalent to a radar detection range dropping from 200 km to 100 km. Tail-on shows the single largest reduction, 26.2 dB; broadside is the one aspect where shaping costs a small, honestly-reported penalty (−0.8 dB) rather than a gain, because the canted V-tail's redirected energy has to land somewhere.
• All three coherent mechanisms are load-bearing. Double-bounce alone accounts for sharp 45°-aspect spikes that PO+PTD completely miss, and triple-bounce peak returns were confirmed, at every facet count tested, to exceed double-bounce peak returns at each geometry's own optimum trihedral angle , consistent with the stronger L^4/\lambda^2 trihedral scaling law, and confirmed as genuine physics via an explicit count of contributing facet triples rather than accepted on faith.
• Geometry over material. Edge-classification diagnostics quantitatively confirm the stealth aircraft has zero internal right-angle ridge structure (wedge parameter n=2.00 uniformly, 44 boundary edges out of 52) against the conventional aircraft's 41 internal edges. The SIBC study then shows that at X-band, even the worst real conductor tested (CFRP composite) is near-indistinguishable from bare PEC (sub-0.06 dB), so the 11.9 dB nose-on reduction is a shaping effect almost in its entirety, not a coating effect. This matches how real low-observable programmes actually prioritise their effort.
• Automated shaping beats hand shaping, a little. A differential-evolution optimizer searching only three geometric parameters (wing sweep, V-tail cant, nose length) found a configuration with 3.13 dB lower mean RCS than the original hand-designed baseline in 2492 evaluations and under two minutes, evidence that even a small, cheap automated search is a useful complement to design intuition, not a replacement for the physical reasoning that produced the baseline shape in the first place.
• A single RCS number hides where the energy comes from. Wideband impulse-response processing resolves nose, wing-leading-edge, and tail as separate range-profile peaks instead of one aggregate scalar shows the conventional aircraft as a spatially coherent, radar-identifiable image while the stealth aircraft's image is diffuse and low-amplitude at the same aspect, the stealth shape denies a clean picture, not just a lower number.
• Convergence discipline held throughout. Every result above carries its own mesh-convergence study, run to the point where either the answer visibly stabilises or the study honestly says it does not (triple-bounce, capped by its O(n^3) cost). No claim on this site rests on an unrefined, unchecked mesh.

What this does and does not claim

This is a first-principles, high-frequency asymptotic (PO/PTD/ray-optics) study, not a full-wave solution and not a validated prediction for any real fielded aircraft, the six geometries are teaching-scale representative shapes, not CAD-accurate airframes, and the SIBC material set uses single reference conductivities rather than dispersive, temperature-dependent material data. Within that scope, every quantitative claim above is backed by a specific equation in the Governing Physics tab, a specific convergence plot, and, where applicable, a specific self-check assertion in \texttt{solvers.py} that runs automatically and fails loudly if violated.

Literature Survey & References

Sources consulted span the four eras of this project's scope: classical exact-solution electromagnetics (1900s–1960s), the high-frequency asymptotic methods that make faceted RCS prediction tractable (1950s–1980s), the applied RCS/RAM engineering handbooks that connect theory to real airframes.

References

1. E. F. Knott, J. F. Shaeffer, M. T. Tuley, Radar Cross Section, 2nd ed. SciTech Publishing, Raleigh, NC, 2004.
2. P. Ya. Ufimtsev, Fundamentals of the Physical Theory of Diffraction. John Wiley & Sons, Hoboken, NJ, 2007.
3. M. I. Skolnik, Introduction to Radar Systems, 3rd ed. McGraw-Hill, New York, 2001.
4. C. A. Balanis, Advanced Engineering Electromagnetics, 2nd ed. John Wiley & Sons, Hoboken, NJ, 2012.
5. R. F. Harrington, Field Computation by Moment Methods. Macmillan, New York, 1968.
6. J. B. Keller, "Geometrical Theory of Diffraction," Journal of the Optical Society of America, vol. 52, no. 2, pp. 116–130, 1962.
7. R. C. Kouyoumjian and P. H. Pathak, "A Uniform Geometrical Theory of Diffraction for an Edge in a Perfectly Conducting Surface," Proceedings of the IEEE, vol. 62, no. 11, pp. 1448–1461, 1974.
8. G. T. Ruck, D. E. Barrick, W. D. Stuart, C. K. Krichbaum, Radar Cross Section Handbook, vols. 1–2. Plenum Press, New York, 1970.
9. D. C. Jenn, Radar and Laser Cross Section Engineering, 2nd ed. AIAA Education Series, Reston, VA, 2005.
10. D. E. Weile and E. Michielssen, "Genetic Algorithm Optimization Applied to Electromagnetics: A Review," IEEE Transactions on Antennas and Propagation, vol. 45, no. 3, pp. 343–353, 1997.
11. D. C. Munson Jr., J. D. O'Brien, W. K. Jenkins, "A Tomographic Formulation of Spotlight-Mode Synthetic Aperture Radar," Proceedings of the IEEE, vol. 71, no. 8, pp. 917–925, 1983.
12. V. C. Chen and M. Martorella, Inverse Synthetic Aperture Radar Imaging: Principles, Algorithms and Applications. SciTech Publishing, Edison, NJ, 2014.
13. N. C. Currie (ed.), Techniques of Radar Reflectivity Measurement. Artech House, Norwood, MA, 1989.

Online & Supplementary Resources

General-audience and semi-technical resources used for building physical intuition and cross-checking qualitative claims (stealth shaping rationale, real-world radar band usage, demonstrations) alongside the primary academic literature above.

• Real Engineering (YouTube), video essays on stealth aircraft shaping principles and the aerodynamics/RCS design trade-off.
• Mustard (YouTube), historical deep-dives on the F-117 Nighthawk, B-2 Spirit, and SR-71 programs, useful for real airframe shaping context behind the design principles discussed on this site.
• The Efficient Engineer (YouTube), general engineering-fundamentals explainer style used for cross-checking how to present derivations clearly.
• MIT OpenCourseWare, 6.632 Electromagnetic Wave Theory lecture series, background review of Maxwell's equations and boundary-value problems underlying the SIBC/RAM derivations.
• IEEE Spectrum (blog), technology-journalism coverage of stealth shaping and radar-absorbent material developments, used for real-world material/application context.
• The Aviationist and Air & Space Forces Magazine (blogs/trade press), military aviation technical reporting used for real-world RCS reference figures and program history.
• NASA Technical Reports Server (NTRS), public archive of aerospace technical reports, consulted for supporting background on radar cross-section measurement methodology.

Computational Framework

Entire pipeline (PO, PTD, double/triple-bounce, RAM/SIBC, bistatic, polarimetric, optimizer, wideband) implemented from first principles in Python (NumPy/SciPy/Matplotlib), self-checked against analytical closed-form limits at every stage (Mie series, PEC limits, monostatic reduction of bistatic formulas, TE/TM reflection dichotomy) rather than validated by inspection alone.

Appendix

A1  Abbreviations & Shortforms

A2  Nomenclature

Every symbol used in the equations across this site, collected in one place.

Roman symbols

Greek symbols

A3  Project Directory Structure

oll/
├── main.py                          # PyQt6 GUI app — 11-tab tool wrapping every solver below
├── geometries.py                    # facet primitives (quads/tris/welding); the 6 target geometries
├── solvers.py                       # core physics engine — POAmplitudeSolver → PTDSolver → DoubleBounce → TripleBounce,
│                                    # SIBCMaterial, SalisburyScreen, DallenbachLayer, Mie series, self-check
├── plot_style.py                    # shared dark theme for every matplotlib figure on this site
│
├── convergence_study.py             # baseline PO/PTD/DoubleBounce mesh convergence
├── advanced_study.py                # bistatic + polarimetric demonstration plots
├── advanced_convergence_study.py    # bistatic + polarimetric mesh convergence
├── triple_bounce_convergence.py     # triple-bounce mesh convergence (scalar + polarimetric)
│
├── sibc_comparison.py               # SIBC vs PEC — azimuth sweep, frequency sweep, reduction bars
├── sibc_convergence.py              # SIBC vs PEC mesh convergence
│
├── stealth_optimizer.py             # differential-evolution geometry optimizer + its convergence study
│
├── presentation.html                # this website (single self-contained file)
│
└── images/rcs_prediction/                     # generated output — 85 PNGs, 14 numbered subfolders
    ├── 01_geometry/              (30 png)  # iso/front/side/top/wireframe views, all 6 geometries
    ├── 02_rcs_polar/              (10 png)  # polar RCS patterns — PO-only and full-physics, per geometry
    ├── 03_rcs_heatmap/            (2 png)   # azimuth×elevation RCS heatmaps, conventional + stealth
    ├── 04_frequency/              (4 png)   # RCS vs frequency, L→Ka band, nose-on + broadside
    ├── 05_ram/                    (5 png)   # Salisbury/Dallenbach absorption curves + applied examples
    ├── 06_breakdown/              (6 png)   # PO-only / PTD-only / double-bounce-only isolated plots
    ├── 07_comparison/             (3 png)   # overlay of all 6 geometries, ranking bars, reduction bars
    ├── 08_physics_diagrams/       (4 png)   # edge classification + Salisbury/Dallenbach schematic diagrams
    ├── 09_bistatic_polarimetric/  (4 png)   # bistatic sweeps, polarimetric azimuth + signature bars
    ├── 10_convergence/            (5 png)   # baseline/bistatic/polarimetric/triple-bounce mesh convergence
    ├── 11_sibc/                   (4 png)   # SIBC vs PEC — azimuth, frequency sweep, reduction, convergence
    ├── 12_optimization/           (3 png)   # DE optimizer convergence, baseline-vs-optimized, mesh convergence
    ├── 13_wideband/               (3 png)   # frequency response, range profile, mesh convergence