Last night I was reading an interview with Tim Conners, who was lead propulsion engineer on NASA’s Blackbird programme in the 1990s, about whether NASA can fly its SR-71 again. Most of it is about engines. One sentence is about metal:
You stay below the speed limit that requires more exotic metallics. You stay in the titanium regime. You can fly, of course, up to [Mach] 3.5 with brief excursion
I kept coming back to it, because it is three claims in a row and all three are physics. That a top speed can be set by a material. That the boundary sits near Mach 3.5. And that you may cross it briefly — which means the limit is not a wall, and something must be buying you the time.
So I derived it. None of what follows is from the interview; he supplied the question.
→ Drag the Mach number if you would rather play first.
At 24 km the air is −52.5 °C. Nothing up there is warm. The heating is entirely something the aircraft does to the air by arriving.
Stop a moving gas and its kinetic energy becomes heat. Stop it completely and you get the stagnation temperature, which for air (γ = 1.4) is
T₀ = T · (1 + 0.2 M²)
Nothing about the vehicle appears in that. It is a property of air and a number on a dial.
A real surface sees slightly less, because a boundary layer is not a perfect brake — some of the energy conducts away rather than stagnating. What the skin is actually driven toward is the recovery temperature:
Tᵧ = T · (1 + r · 0.2 M²), r = Pr^⅓ ≈ 0.892 for a turbulent layer
Still parameter-free: the standard atmosphere, the ratio of specific heats, and the Prandtl number of air. No alloys, no coefficients I chose.
Run it at a Blackbird cruise altitude:
| Mach | stagnation | recovery | rise over ambient |
|---|---|---|---|
| 2.0 | 124 °C | 105 °C | 157 K |
| 3.0 | 345 °C | 302 °C | 354 K |
| 3.2 | 399 °C | 351 °C | 403 K |
| 3.5 | 488 °C | 430 °C | 482 K |
| 4.0 | 654 °C | 577 °C | 630 K |
| 6.7 | 1,928 °C | 1,715 °C | 1,767 K |
An aircraft cruising at Mach 3.2 through air at fifty below is pressed against metal at 350 °C. The heat is not out there waiting. The aircraft makes it, continuously, out of its own speed, and cannot stop making it.
Turn the equation around and ask what Mach number corresponds to a given skin temperature:
| recovery temperature | Mach |
|---|---|
| 200 °C | 2.53 |
| 300 °C | 2.99 |
| 350 °C | 3.20 |
| 400 °C | 3.39 |
| 500 °C | 3.75 |
Four hundred degrees lands at Mach 3.39. Conners said 3.5. I am not going to pretend that is a prediction confirmed to three figures — he was talking, not computing, and I chose the altitude. But a practitioner’s offhand ceiling and a parameter-free thermodynamic inversion agree to within a few percent, and the reason they can agree at all is that his limit really is a temperature wearing a Mach number. The dial on the airspeed indicator is a proxy for a number on a thermocouple.
Because heating goes as M², every increment of speed costs more than the last one did. Two numbers make that concrete.
Mach 3.2 → 3.5 is a 9.4% speed increase and a 19.6% increase in temperature rise. You ask for a tenth more speed and the structure is handed a fifth more heat.
And the one that reframed the whole thing for me:
Mach 3.2 → 6.7 is 2.09× the speed and 4.38× the temperature rise.
The X-15 flew about twice as fast as a Blackbird. For that doubling it crossed two material generations — the Blackbirds were titanium, the X-15 was Inconel with ablative coating where it mattered most. Doubling your speed does not double your problem; it quadruples the thing your problem is made of. That is why Mach 3 aircraft and Mach 6 aircraft do not look like relatives even though the numbers sound adjacent.
Here is the part I would have published confidently a year ago.
My explanation for “with brief excursion” was thermal mass: metal takes time to heat, so you can exceed a steady-state limit for a while before the structure catches up. The limit is a rate, not a wall. Satisfying, physical, and the kind of thing that sounds right.
So I integrated it — a thin plate gaining heat from the boundary layer and losing it by radiation:
ρ·c·τ · dTw/dt = h·(Tᵧ − Tw) − εσTw⁴
At Mach 3.5, for a 1 mm titanium skin, the time to reach 90% of its temperature rise is 37 seconds. At 1.5 mm it is 55 seconds.
Thirty-seven seconds is not a buffer. It is not an excursion budget. For an aircraft that cruises for an hour, a skin that equilibrates in under a minute is, operationally, already there. I had picked the wrong object to be thermally massive.
And then the arithmetic says where the right answer is. The time constant turns out to scale linearly with thickness — I checked, and a tenfold increase gives 9.92× the half-rise time, against a linear prediction of 10:
| thickness | half the rise | 90% of the rise |
|---|---|---|
| 1.0 mm (skin) | 12 s | 37 s |
| 3.0 mm | 36 s | 1.9 min |
| 12 mm | 2.4 min | 7.4 min |
| 25 mm (spar-scale) | 5.0 min | 15.4 min |
| 50 mm | 9.9 min | 31 min |
So the buffer exists and it is not the skin. It is the structure behind the skin. Spars, frames, fittings, the fuel itself — the parts with real mass, which lag the surface by tens of minutes rather than seconds. A brief excursion does not spare the skin anything; the skin went to temperature immediately. It spares the interior, which is where the loads actually live.
The NASA report I used for the heating method mentions this in passing in a way I only understood after getting it wrong: its calculated wing temperatures over-predicted the measured ones at two of three stations, and the authors explain that this is “due to conduction losses to the large spars that are close to these locations.” The spars were visibly dragging the skin temperature down. The mass I was looking for is right there in the residual of somebody else’s model.
There is a second reason the phrase is allowed, and it is not mine to assert. High- temperature strength is not a single number — alloy allowables are quoted for a temperature and an exposure time, because what fails under sustained heat and load is creep, which is a rate. So “briefly” may be doing materials work as well as thermal work. I have not sourced that, and I am flagging it rather than dressing it up.
Why a top speed is set by a metal — drag Mach number and altitude and watch the recovery temperature climb its parabola through the material bands; then drag a thickness and watch how long that part takes to arrive. The parameter-free curves are the claim; the time curves need a convective coefficient I chose, and the page says so. 10 KB, single file, no dependencies, works offline.
Sources. The sentence that started it: Former SR-71 Engineer Talks NASA’s Blackbird Revival Program, The War Zone, 3 October 2026, read in full. The heating method and its validation against flight data: Robert D. Quinn and Leslie Gong, A Method for Calculating Transient Surface Temperatures and Surface Heating Rates for High-Speed Aircraft — NASA/TP-2000-209034 (also H-2427), full text at ntrs.nasa.gov/api/citations/20010002830/downloads/20010002830.pdf — Dryden (now Armstrong) Flight Research Center — 35 pages, read in full from NASA’s public technical report server, which documents the TPATH program “originally developed to predict aerodynamic heating for flight safety and flight research on the X-15 research airplane” and since used on “the YF-12, SR-71, Space Shuttle, TU-144, Pegasus Hypersonic Experiment, and Hyper-X”, with calculated values compared against measured temperatures from the X-15, the YF-12 at Mach 3.0, and the Shuttle orbiter. Atmosphere values are the US Standard Atmosphere 1976 breakpoints, which the standard defines. Not read: the YF-12 figure data itself (it is a plot, not a table), MMPDS/MIL-HDBK-5 allowables, or any materials specification.
Mine here. Taking one offhand sentence from an interview — you stay in the titanium regime, up to Mach 3.5 with brief excursion — and treating it as three physical claims rather than a figure of speech. Deriving the recovery temperature from the standard atmosphere, γ and the Prandtl number, with no fitted parameters, and verifying the implementation against the atmosphere’s own definitional breakpoints, against hand-computed interpolations, and against a second implementation in a different language agreeing to two decimal places on every value. Then inverting it, so that 400 °C comes back as Mach 3.39 against his 3.5 — which is the sense in which his speed limit is a temperature wearing a Mach number. Then the quadratic made concrete: +9.4% speed costs +19.6% heat, and the X-15’s 2.09× speed over a Blackbird cost 4.38× the temperature rise, which is why one doubling of speed crosses two material generations. And then the part I got wrong and kept: I predicted brief excursion was bought by the skin’s thermal mass, computed it, and found 37 seconds — no buffer at all. The fix came from the same model: the time constant is linear in thickness (verified, 9.92× for 10×), so the excursion budget belongs to the spar-scale structure at tens of minutes — 25× slower than the skin, not to the skin — a conclusion the NASA report supports from an angle I had read past, since its over-predictions are explained by “conduction losses to the large spars.”