The curve that measures a gradient

The three curves the rung below separates all answer versions of one question. A streamline says where fluid at this point is heading now; a pathline says where one particle went; a streakline says where everything from one place has got to. All three are about position, and the differences between them are differences about which fluid and which instant.

The fourth curve answers a different question, and the difference is not a refinement.

What a straight line becomes #

Release a set of particles that lie on a straight segment at one instant and watch. Each is carried by whatever velocity it finds, so after a time $t$ the particle that started at height $y$ has moved downstream by

$$\Delta x(y) = \int_0^{t} u\bigl(y(t')\bigr)\,dt',$$

which in a parallel shear flow, where a particle stays at its own height, is simply $u(y)\,t$.

So the line is the profile, scaled by the elapsed time. Not approximately, not to leading order: the displacement divided by $t$ is $u(y)$, and the check behind these figures verifies it at two hundred heights to fifteen decimal places, which is the precision of the arithmetic rather than of a model.

That is a genuinely different kind of measurement from the other three. A pathline requires following one particle and gives one number at a time. A timeline gives the whole profile in one exposure, because the particles that are far apart in the picture are the ones that were going at different speeds, and their separation is the difference in speed.

Why this is the curve a boundary layer wants #

A boundary layer is a velocity gradient anchored at a wall, and every number worth having about one is a property of that gradient rather than of the velocity itself. The wall shear is $\mu(\partial u/\partial y)$ at the surface. The displacement and momentum thicknesses are integrals of the deficit. The shape factor is a ratio of two of them, and it is what says whether separation is coming.

None of those is visible in a streak photograph. A smoke filament released into a boundary layer runs downstream and shows that the fluid runs downstream. A timeline released across the same layer draws the profile the numbers are computed from.

The second derivative deserves the emphasis. A timeline’s curvature is $\partial^2u/\partial y^2$ times the elapsed time, and that quantity is the one an inflexion-point criterion is about: a profile with an interior inflexion is unstable to a disturbance in a way one without is not. So a bubble line that visibly changes the sense of its curvature somewhere in the middle of a layer is showing an inflexion point directly, and no other visualisation technique shows one at all.

The two errors that are inside the photograph #

Every instrument applies an operator to the flow, and what a measurement reports is the flow convolved with something the experimenter chose. For a hydrogen-bubble wire the two operators are unusually easy to write down and unusually rarely printed beside the picture.

The bubbles rise. They are hydrogen in water and they are buoyant, so each one climbs at a terminal speed set by its size while it is being carried downstream. By the time the photograph is taken the line is no longer at the heights it was released at — it is at those heights plus the rise speed times the elapsed time — and the profile it reports is therefore the true profile shifted. Near a wall, where the profile is steep and the distance available is small, a shift of a fraction of a millimetre is a large fraction of the measurement.

And the release is not instantaneous. A pulse of finite duration releases bubbles over a window rather than at a moment, so what appears is the timeline convolved with that window: a band rather than a line, whose width is the local velocity times the pulse duration. It is therefore thickest where the flow is fastest, which is the free stream, and thinnest at the wall — the opposite of what would be convenient, since the free stream is where the answer is already known.

Both errors are computed rather than asserted, and the assertion behind the figure checks the thing worth checking: that the largest displacement error falls inside the sheared part of the profile rather than out in the uniform stream. A wire whose worst error were in the free stream would be a wire measuring something other than a boundary layer, and the check would have found it.

What the errors do not do #

It is worth being precise about the size of the complaint, because a technique that has been in use for sixty years is not being dismissed here.

The rise speed is a property of the bubbles and is roughly known, so the height shift is a correctable error — subtract it. The pulse width is set by the electronics and is known exactly, so the band’s thickness is a known uncertainty rather than an unknown one. Neither is a bias hidden in the method; both are terms that can be written down, and this essay’s contribution is to write them down rather than to object.

What they do is set a floor on how close to a wall the technique works. The rise displacement grows with elapsed time and the resolution needed grows as the wall is approached, so the two meet somewhere, and below that height the line is reporting fluid that came from further out. Where exactly depends on the bubble size and the shear, and the figure’s numbers are illustrative rather than a calibration of any real apparatus.

The genuinely uncorrectable one is a third that this model does not contain: bubbles are particles with inertia, and a particle follows a flow only up to a frequency its own inertia sets. For hydrogen bubbles in water that corner is high and the approximation is excellent, which is why the technique works at all; in a gas it is not, which is why nobody does this in air.

Reading a number off the line, and the one that is hard #

Since the whole claim is that a timeline is a profile, it is worth walking through what is actually read off one and where the reading gets difficult.

The free-stream speed is trivial. The straight, uniform part of the line, far from the wall, has been displaced by $U t$, so measuring that displacement and dividing by the interval gives $U$. It is the easiest measurement in the picture and it is the one already known from a pitot tube.

The layer thickness is easy and slightly arbitrary. Where the line stops being displaced by less than the free stream is the edge of the layer, and — as everywhere in this subject — that height has to be defined by a convention, because the approach is asymptotic. Ninety-nine per cent is the usual choice and it is a choice.

The wall shear is the hard one, and it is the one that matters. It is the slope of the line at the wall, which means differencing a picture in the region where the line is closest to a surface, shortest, most affected by the bubbles’ rise, and most likely to be obscured by the wire’s own wake. Every error in the technique is concentrated exactly where the quantity of most interest is measured.

The usual repair is not to measure it there. Fit a known profile shape to the part of the line that is legible, and take the wall slope from the fit — which works, and which means the wall shear reported from a bubble photograph is partly a consequence of the profile family assumed. That is a perfectly respectable measurement and it is not a direct one, and the distinction is worth keeping because the picture looks like a direct measurement of exactly the thing the number is not directly measured from.

And the momentum thickness is the surprising one, because it is easier than the wall shear rather than harder. It is an integral of the deficit across the whole layer, and an integral of a curve read off a photograph is a robust operation: the errors that ruin a derivative average out in a quadrature. So a timeline photograph gives the momentum thickness well and the wall shear badly, and the drag follows from the momentum thickness rather than from the wall shear — which means the technique is at its best exactly where the engineering question is.

The curve as a material line #

There is a second way to look at a timeline, and it connects this essay to a part of the collection that does not look related.

A timeline is a material line: a curve made of the same fluid particles for ever after, which is exactly the object whose length grows under strain with nothing pulling it. So everything that collection knows about material lines applies — the line’s length is a record of the deformation it has been through, its stretching rate is a distribution rather than a number, and in a flow with any chaos in it the line’s length grows exponentially and the picture becomes unreadable within a few timescales.

That last point is the practical limit of the technique and it is not a defect of the wire. A timeline is legible only while it is short-lived. In a laminar shear flow it deforms smoothly and stays readable for a long time; in a transitional or turbulent one it is folded past recognition within a fraction of a second, which is why hydrogen-bubble photographs of turbulent boundary layers show the low-speed streaks near the wall and nothing intelligible above them.

And that limitation is itself a measurement. How long a timeline stays legible is how long the flow takes to fold it, which is a timescale of the flow rather than of the apparatus. A picture that becomes a tangle in a tenth of a second has said something quantitative about the flow, and the number is usually thrown away.

Which part of the gradient it reports #

A velocity gradient is a tensor and a timeline is one line, so the technique cannot be reporting all of it. Being precise about which part is reported is the difference between a picture and a measurement.

A line released across the flow, perpendicular to the streamwise direction, is displaced by $u(y)$ and reports $\partial u/\partial y$. That is the one component a wall-bounded flow is mostly made of, which is why the wire is always strung that way, and it is why the technique looks more complete than it is.

Release the line along the flow instead and it reports $\partial u/\partial x$ — the streamwise stretching — and in a parallel flow that component is zero, so the line stays the length it was and the picture is empty. In a converging channel it is not zero, the line lengthens, and its lengthening is the acceleration.

So one wire measures one component, and the two orientations measure different ones. Getting the whole gradient means two wires or two runs, and even then two of the nine components are out of the plane and invisible to any arrangement of wires in it. This is the same limitation a strain rate splits into pieces has always had, seen from the instrument’s end: the symmetric and antisymmetric parts of the gradient are not separately visible in a single line’s distortion, because a rotation and a shear deform a line in ways that look alike over a short interval.

That last point has a clean consequence. A timeline in a solid-body rotation and a timeline in a pure shear at the same rate both start out bending the same way, and they separate only after the rotation has carried the line round far enough to distinguish them — which is precisely the distinction spin is not going round is about. A single short-exposure timeline cannot tell them apart, and that is a statement about the instrument rather than about the flow.

The one flow in which a timeline says nothing #

A measurement is best understood by the case where it returns nothing, and this one has a clean one.

In a flow with no shear a timeline does not deform. Every particle moves by the same amount, the released line is translated bodily, and after any interval it is a straight segment parallel to where it started. That is not a failure of the technique; it is the technique correctly reporting that the velocity gradient is zero.

The uniform oscillating stream of the rung below is exactly this case, and it is worth noticing what happens there. That flow produces streaklines that fold, cross themselves and close into ellipses — the richest set of pictures in this anchor — and it produces a timeline that is a straight segment being carried about unchanged. The two curves are sensitive to completely different things: the streakline to the history of the velocity at a point, and the timeline to its variation across space, and a flow can have as much of one as it likes with none of the other.

That is the sharpest statement of why there are four curves rather than three plus a refinement. Two of them are instruments pointed at time and two at space, and a flow that is trivial to one may be the interesting case for the other.

What the picture cannot show #

The profiles are prescribed. Nothing here solves a boundary layer. The layer-shaped profile is a sine curve within about a per cent of the tabulated Blasius solution, chosen because the argument is about the shape a timeline draws rather than about the profile’s own accuracy, and any claim about a real layer’s numbers should be taken from the essays that solve one instead.

The flow is parallel and a real one is not. Every particle is assumed to stay at its own height, which is what makes the displacement exactly $u(y)t$. In a growing boundary layer there is a small wall-normal velocity, particles drift outward, and the reading acquires a bias this model does not have.

Nothing is unsteady. The whole essay treats profiles that do not change while the line is travelling, which is the case a timeline is easiest to read in and is not the case the technique is most often used in.

And a bubble is not a fluid particle. The model treats each as a marker carried exactly, with the rise added as a separate term. A real bubble also disturbs the flow it is in, sits at a slightly different place than the fluid it marks, and grows on the wire before it leaves. None of that is here.

Who used it, and when #

The hydrogen-bubble technique was developed in the late 1950s and came into general use through the 1960s, most famously in the Stanford work on the structure of the turbulent boundary layer — the experiments that found the low-speed streaks, and that established the near-wall region as having structure rather than being merely random. Those photographs are timelines, and the streaks in them are places where a released line has been dragged backwards relative to its neighbours.

The idea of releasing a line rather than a point is much older than the apparatus. It appears wherever somebody has had a row of markers and a shear: a line of floats across a river, a row of dye spots, the smoke from a wire. What the electrolysis gave was control — a line released at a chosen instant, in a chosen place, repeatedly, with a pulse whose duration is a dial.

And the technique is the reason a particular picture is in every textbook. The image of a boundary layer as a set of bent lines, thickening downstream, is a hydrogen-bubble photograph, and its persuasive power comes from the one property this essay is about: the reader is looking straight at the profile rather than at something the profile caused.

Where the ladder goes next #

This anchor now has all four curves: the three that report position and the one that reports a gradient. What it does not have is the comparison a reader would want next.

The rung above is the instrument in general, and it is the question this rung has kept touching. Each of these techniques applies a different operator to the same flow — an exposure, a release window, a probe volume, a rise speed — and each is legible over a different band of timescales. Setting the four side by side with their transfer functions written down, and computing which flow features each one can and cannot resolve, is a rung with a real result in it: the bands overlap much less than the interchangeable use of the pictures suggests.

The one beside it is the inverse problem. Given a timeline photograph and nothing else, what can be recovered — and in particular, can the two errors above be fitted rather than corrected, by using the band’s own thickness to measure the pulse and the line’s own drift to measure the rise? That would make the instrument self-calibrating, and it is the kind of question that turns a visualisation into a measurement.