The wall that cancels its own waves

Everything this site has said about a nozzle so far is one-dimensional. Area in, Mach number out; a throat that stops listening; two Mach numbers for one area. All of it is right and none of it mentions a shape.

That silence is not a small gap. A duct with the correct area ratio and the wrong contour delivers a stream that is neither uniform nor parallel, at exactly the design Mach number.

Two invariants, and why they make the problem easy #

In a steady supersonic flow that is irrotational and isentropic, the equations are hyperbolic: information travels along Mach lines, at the Mach angle μ = arcsin(1/M) to the local flow direction, and nothing travels any other way. Along those lines two combinations are constant:

$$\theta + \nu(M) \quad\text{on a left-running line}, \qquad \theta - \nu(M)\quad\text{on a right-running one},$$

with ν the Prandtl–Meyer function that measures how far the flow has been turned by expansion.

This is what makes the whole calculation arithmetic. The flow at the crossing of two Mach lines is determined by the two invariants they carry: add them and halve to get θ, subtract and halve to get ν, and invert ν to get the Mach number. No differential equation is integrated anywhere — the integration was done once, analytically, when the invariants were found.

What has to be marched is the geometry: where the crossings are.

The design, and the one number it needs #

A minimum-length nozzle expands the flow at a sharp corner, which turns it through its maximum angle in one step, and then contracts the wall back to horizontal, absorbing each wave. Since the expansion must supply the whole turn ν(M_e) and the contraction must take it all back,

$$\theta_{\max} = \frac{\nu(M_e)}{2},$$

which is 18.37° for Mach 2.4 and is the only closed form in the construction.

Everything else is the net. Each characteristic leaving the corner crosses the reflection of every earlier one, and the wall point where each reflected wave lands is the point at which the wall must have turned to the wave’s own flow angle — because if it has not, the wave reflects and travels back into the flow, and the exit is not uniform.

The check that the two theories are one theory #

The exit height of the marched contour has to equal the area ratio the one-dimensional relation gives, because both are statements of mass conservation between the throat and a uniform exit. They share no arithmetic whatever: one marches a net of waves through the interior and reads a height off the last wall point, the other is an algebraic relation between area and Mach number.

That check earned its keep immediately, and the failure is worth recording because of its shape.

Two versions of this net had every Mach number and every Riemann invariant exactly right and every coordinate wrong. The first built each characteristic upward from the axis instead of downward from the corner; the second treated the corner as sonic for every characteristic, when in fact a characteristic leaving a Prandtl–Meyer fan has already been turned and carries its own Mach angle. The first collapsed the whole net onto the axis. The second produced a perfectly plausible nozzle whose exit area was nine per cent short, and which did not improve when the net was refined.

Nothing about either would have shown in a plot of the contour. The invariants held to machine precision throughout, the wall turned smoothly from θmax to zero, the exit Mach number was exactly 2.400 — and the only thing that caught them was a number computed a different way. This is the site’s habit doing precisely what it exists to do: an exactly-solved flow can be drawn in the wrong place, and a check on the values alone will never notice.

The convergence test is the second half of it. A discretisation error falls when the net is refined and a modelling error does not, so the slope of that figure is the evidence rather than any single value on it.

Three ducts, one area ratio #

The three failures are different and all three are real.

The cone leaves the flow diverging at the wall angle, so the exit momentum is not all axial. The thrust loss is $(1+\cos\alpha)/2$ — about 1.7 per cent for a 15-degree half-angle — and it is the reason bell nozzles exist. That correction is a standard piece of rocketry and it is a geometric consequence of exactly this: the area rule is silent about direction.

The duct that opens too fast cannot absorb its waves, so they reflect from the walls and cross the flow repeatedly. The exit has the right average Mach number and is full of oblique shocks; in a supersonic wind tunnel that means a test section whose Mach number varies across the working area, which is precisely what a tunnel cannot tolerate.

A duct that opens too slowly — not drawn — is simply longer and heavier than it needs to be, with more wall to grow a boundary layer on and more skin friction to pay for it. That is the trade the “minimum-length” in the name refers to.

What the net says about the flow inside #

The characteristic net is not only a construction device; it is a picture of where information can travel, and reading it is worth a section.

Every point in the nozzle is influenced only by the region upstream between its two Mach lines. That wedge is the point’s domain of dependence, and it is the whole content of the flow being hyperbolic: a disturbance at the wall cannot affect anything upstream of the Mach line through it, which is why a supersonic nozzle can be designed marching downstream and a subsonic one cannot.

That asymmetry explains a fact this site established from the other end. A subsonic throat listens to the exit pressure and a supersonic one does not: in the supersonic section, no characteristic runs upstream, so nothing downstream of the throat can be felt at it. The net is that statement drawn.

It also explains why a design error stays where it is made. A kink in the wall at one station sends one wave into the flow, and that wave reaches the exit at a definite place — so a nozzle with a bad patch has a bad stripe in its exit flow rather than a uniformly degraded stream.

Why the length matters #

The “minimum-length” in the name is a real optimisation and it is worth saying what it trades.

A nozzle designed this way is the shortest one that produces a uniform parallel exit, because the expansion happens as fast as it possibly can — at a single sharp corner. Any nozzle that expands more gradually is longer, and length costs weight, wall area and boundary-layer growth.

What the sharp corner costs is a severe local expansion, which in practice separates the boundary layer or at least thickens it sharply. So real designs round the corner and accept the extra length, and the minimum-length construction becomes a bound rather than a blueprint: it says how short a nozzle could be if the boundary layer did not exist, and every real one is longer.

That is the useful shape of the result. An idealised design that cannot be built is worth having when it is a limit, in the same way that Betz’s limit for a wind turbine is worth having: nobody reaches it, and knowing where it is says how much of the gap belongs to the designer.

The same method outside a nozzle #

The construction is not about nozzles: it is about any steady supersonic flow whose boundary conditions are known, and it is worth naming the other places it does the work.

A supersonic aerofoil section. The flow over a diamond or biconvex section is a sequence of oblique shocks and expansion fans, and the characteristic net between them gives the surface pressure directly. This site computes that case in closed form for a thin section, and the net is what a thick one needs.

An overexpanded jet. A nozzle running below its design pressure ratio has a shock structure at its exit that repeats down the plume — the shock diamonds visible in an afterburning jet — and the periodic structure is a characteristic net reflecting alternately off the jet boundary and the axis.

A supersonic inlet. The oblique shock system on a ramp is designed by the same wave-by-wave bookkeeping, with the aim reversed: instead of cancelling waves at a wall the design places them so that they meet at a lip.

The unifying statement is the one the invariants make. Where a flow is supersonic and steady, the whole problem is bookkeeping along Mach lines, and the only question is what the boundaries do to the waves that arrive at them.

A contour is designed for one pressure ratio, and a rocket does not fly at one #

The construction produces a wall that cancels its waves exactly, and it does so for the pressure ratio it was designed at. A launch vehicle leaves that pressure ratio behind within seconds, and the consequences are severe enough to shape the hardware.

The thrust is $\dot m v_e + (p_e - p_a)A_e$, so a nozzle whose exit pressure matches ambient extracts everything available and any mismatch costs. The two mismatches fail differently.

Underexpanded, at altitude, the exit pressure exceeds ambient and the plume expands outside the nozzle through a fan. The gas is doing its expanding in the atmosphere instead of against the wall, so the thrust falls short of what a longer nozzle would have given. It is a loss, and it is benign.

Overexpanded, at sea level, the exit pressure is below ambient and the plume is compressed by oblique shocks at the lip. That is tolerable while it stays outside. Push it far enough — the usual rule of thumb puts the boundary near an exit-to-ambient pressure ratio of about 0.4 — and the shock system moves inside the nozzle, the boundary layer separates from the wall, and the separation line is not reliably symmetric. An asymmetric separation in a large nozzle produces a lateral force of tens of kilonewtons that flicks about unpredictably during start-up, and side loads of that kind are a structural design case rather than a performance one.

So a fixed contour is a compromise: expanded enough to be worth having at altitude, not so much that it separates on the pad. First stages carry modest area ratios and upper stages carry enormous ones, because an upper stage never sees an atmosphere to be overexpanded against.

The way out is to give up the wall. A plug or aerospike nozzle expands the gas along a central spike, with the outer boundary of the flow being the atmosphere itself — a free boundary rather than a designed one. The plume’s outer edge is then at ambient pressure by construction, at every altitude, and the expansion adjusts itself as the vehicle climbs. There is no contour to get wrong on that side because there is no contour.

It is a striking inversion of this essay. The method above spends its whole effort shaping a wall so that the waves arriving at it are absorbed; the aerospike removes the wall and lets the waves terminate on a boundary that cannot reflect anything, since a free surface has no pressure to impose. What that buys is altitude compensation with no moving parts, and what it costs is a spike that has to be cooled along its whole length in the hottest part of the flow.

What the model does not contain #

No boundary layer. The contour computed here is the shape of the inviscid streamline, and a real nozzle’s wall must be displaced outward by the boundary-layer displacement thickness — a correction of a few per cent that grows along the nozzle and depends on the Reynolds number. Every practical design does the inviscid calculation first and then corrects it, which is the same division of labour the boundary-layer essays describe.

No throat. The construction begins at a sonic line that is assumed straight and vertical, with a sharp corner at its edge. A real throat is rounded, its sonic line is curved, and the flow just downstream of it is not the uniform sonic flow assumed here. This is the largest idealisation in the essay and it is why practical designs start the characteristic net from a computed transonic solution rather than from a line.

Two dimensions only. A round nozzle is axisymmetric, and the axisymmetric method of characteristics has an extra term — the flow’s divergence is no longer captured by the plane invariants — so the invariants are no longer exactly constant along a characteristic and have to be integrated. The structure of the method survives and the arithmetic gets harder.

Design point only. The whole construction assumes the nozzle runs at the pressure ratio it was designed for. Off design it is over- or under-expanded, with shocks or expansion fans at the exit; none of that is here, and it is what the area–Mach essay covers.

Isentropic and irrotational by assumption. If a shock forms anywhere in the nozzle, the flow behind it is rotational and the whole method fails — the invariants are invariants only for an irrotational flow.

How uniform is uniform #

The construction promises a uniform exit and it is fair to ask what that is worth at a finite number of waves, since a real design has to choose one.

Two quantities answer it. The exit angle is zero by construction — the last wall point is set to the flow angle of the last wave, which is zero — so the flow is parallel to whatever precision the arithmetic holds, and the solver reports 10⁻¹⁵ of a degree. The exit Mach number is exactly the design value on the last characteristic, again by construction.

What is not exact is everything between the waves: the flow there is interpolated by the net rather than resolved, so a nozzle built to an 8-wave design has a stream whose Mach number ripples between the waves at the per-cent level, and a 60-wave design at the per-mille level. The convergence figure is the measure of that, since the exit area error and the interior ripple share a cause.

The practical resolution is not chosen by this curve, though. It is chosen by the boundary-layer correction, which is a per-cent-level adjustment to the contour: refining the inviscid net below the size of a correction one is about to apply anyway buys nothing. A design is converged when its remaining error is smaller than the physics it is about to add, and for a supersonic tunnel that is usually a few dozen characteristics. The same reasoning decides how many waves this essay’s figures are drawn with: eighteen, which is enough for the net to be legible and far too few for the contour to be built from.

Who found it, and when #

The method of characteristics for supersonic flow is Prandtl and Busemann’s, from 1929, and it was for thirty years the only way to design a supersonic nozzle. The calculation was done graphically: draughtsmen drew Mach lines on large sheets with the angles read from tables, and a single nozzle design was days of work. The tables that made it possible — ν(M) and μ(M) — are exactly the two functions this essay’s solver evaluates.

The remarkable thing is how well the graphical method worked. The wind tunnels of the 1940s and 50s were designed this way and produced test sections uniform to a fraction of a per cent in Mach number, which is a standard a modern computation has to work to match. The method is exact; the discretisation is the only approximation, and a careful draughtsman’s hand was accurate enough for it.

Its modern descendant is unchanged in principle. Rocket nozzle contours are still designed by characteristics — with a transonic starting solution, an axisymmetric formulation and a boundary-layer correction — because the method computes the shape directly rather than iterating a guess through a solver.

Where the ladder goes next #

The wall in this essay was designed to cancel waves. There is a case where a wall does the opposite and is useful for it: a sharp leading edge that a flow cannot follow, which separates deliberately and rolls into a vortex that stays attached above the surface. That separation is normally a failure; on a slender delta wing it is where most of the lift comes from.