How long the fluid has been in there

Steady does not mean nothing is happening makes the point this essay starts from: photograph a steady flow twice and the pictures are identical, while every parcel in it is being thrown about. The material derivative is the reconciliation, and the convective term is where all of it lives.

This essay is about the simplest quantity that term produces, which is age. How long has the fluid at this point been inside the apparatus?

It is a well-posed question with a short answer. Age is carried by the fluid and increases at one second per second, so its material derivative is exactly 1 — the shortest transport equation in the subject. And it is a field like any other: steady in a steady flow, computable, and with boundary conditions at every inlet.

It is also invisible to every instrument that reads a field at a point.

The duct, and what does not happen in it #

The flow is a one-dimensional contraction: a duct whose area falls so that the speed rises linearly along it, from the inlet to four times the inlet speed at the exit. It is steady, incompressible and completely uninteresting as a flow. That is the point — everything below is true of the least surprising flow available.

Nothing in it depends on time. The rate of change of velocity at a fixed point is exactly zero — 0.000000000, and the check refuses anything above 10⁻⁹ — not approximately zero, at every station and every instant. A probe bolted to the wall records a flat line for ever, whatever it is measuring.

A parcel moving through it accelerates from three units at the inlet to twelve at the exit, in the units of the reference speed and length — a factor of four, which is the speed ratio, because the gradient is the same 3 everywhere along a linear speed rise. The two statements are both exactly true of the same flow.

The two halves of the derivative, station by station #

The material derivative has two terms and one of them is zero everywhere here, which makes the split unusually clean.

The local term — the rate of change at a fixed point — is zero at every station. The convective term — the parcel moving into a place where the flow is faster — carries all of it, and grows by a factor of four along the duct because it is the product of the speed and the speed’s gradient, and the speed has quadrupled while the gradient has not moved.

So the acceleration at the exit is four times the acceleration at the inlet — 12 against 3 — and both are entirely convective; the local term is 0 at both ends and at every station between them. An accelerometer riding the parcel reads a number that quadruples over 0.4621 of a reference time; a probe at the wall reads zero, to 0.000000000 rather than to a tolerance. Both instruments are working correctly.

The number the duct is actually about #

Now the quantity this essay exists for. How long does a parcel take to get through?

The trajectory has a closed form — the speed is proportional to position, so the position grows exponentially and the transit time is the logarithm of the area ratio divided by the rate. For a ratio of four it is ln(4)/3, or 0.4621 of the reference time, and the stepped trajectory matches the closed form to zero difference in double precision.

0.4621 is a property of the flow and it is in none of the flow’s fields. The velocity at each point does not contain it; nor does the pressure, nor the density, nor any derivative of any of them at any single station. It is an integral along a path — the reciprocal of the speed, integrated from inlet to exit — which is a functional of the whole streamline rather than a value anywhere on it.

Age is the first memory a flow has, in the sense that it is the simplest quantity whose value depends on history rather than on state. Everything else in this collection that a flow remembers — a stretch, a strain history, a dissolved scalar, a polymer’s stress — is a more elaborate version of the same construction: something integrated along a trajectory.

Making the duct sharper costs more than it saves #

The trade between the two quantities is worth a section, because it runs the way intuition does not expect.

Doubling the area ratio more than quadruples the acceleration at the exit — it goes as the ratio times the ratio less one, because it is speed times gradient and both rise — while the transit time falls only as the logarithm over that same factor. Five ratios, one length, one inlet speed:

A contraction eight times sharper than another — 32 against 4 — accelerates its fluid 82.7 times harder and keeps it inside for 24.2 per cent as long. The acceleration column spans a factor of 496 while the transit column spans a factor of 6.2.

That asymmetry is why a short contraction is a violent one and why the violence is not bought back in residence time. It is also the reason a wind-tunnel contraction is long: the requirement is a uniform exit flow, the boundary layers on its walls need a benign pressure history, and the way to get a large ratio without a large acceleration is to spread it over a distance.

Why age has its own equation, and what that buys #

Writing the age as a field with a transport equation rather than as a number attached to a parcel is what makes it computable on a grid, and it is worth seeing why it works.

If the material derivative of age is one, then in a steady flow the velocity dotted into the gradient of age is one. That is a linear first-order partial differential equation for a scalar, with the age set to zero at every inlet, and it can be solved by exactly the machinery any other scalar is solved by. Nothing in it is harder than a passive-scalar transport, and every code that carries a scalar can carry an age.

Two things follow immediately. The mean age in a region is the region’s volume divided by the flow rate through it — an exact statement from conservation alone, with no assumption about the flow inside. And the distribution of ages is completely free, because the constraint is one integral over a region and the distribution is a function.

That is the same structure as exact in the total, free in the profile, arriving in one of its most consequential settings: two vessels with identical volumes and identical flow rates have identical mean residence times and can differ by everything in what fraction of the fluid passes straight through.

Why no field measurement can supply it #

It is worth being precise about the impossibility claim, because it is stronger than “it is difficult”.

A field measurement returns a value and, at best, its derivatives at a point. The age at that point is an integral of the reciprocal speed along the streamline that reaches it — so recovering it from local data would require the streamline, and the streamline is a global object obtained by integrating the field. In a steady flow that integration is at least possible in principle, given the field everywhere; in an unsteady one the streamline is not the parcel’s path at all, which is streamlines are not the paths particles take, and the required object is the flow map.

So the age is computable from the whole field and is not readable at a point, and the distinction is the one that matters for instrumentation. This is exactly the position a scalar is a record of where its fluid was describes for any conserved scalar: the value is a look-up along a trajectory rather than a function of the local state.

The way age is actually measured reflects that. Nobody measures it; they measure a tracer. A pulse of dye at the inlet and a concentration record at the outlet gives the residence-time distribution directly, because the tracer’s arrival is the transit time by construction — and the measurement is a measurement of the map rather than of the field.

Where the distribution is what matters #

The mean is the quantity that is easy to compute and it is very often the wrong one.

A reactor. Conversion depends on how long each parcel spent, and a reactor with the right mean and a short-circuiting path through it converts far less than one with the same mean and a narrow spread — 76.9 per cent against 84.5 at the same mean residence time, on the packed-bed numbers of the outlet is the inlet a while ago — because the parcels that pass quickly are unreacted and the ones that linger are already finished.

A heat exchanger. The same argument with heat instead of reaction: a stream whose mean residence time is right and whose distribution is broad delivers a broad range of exit temperatures, and the mixed exit temperature is not the temperature of a parcel with the mean residence time.

A settling tank, a clean room, a ventilated space. Every one of them is specified by an air change rate, which is a mean age, and every one of them fails in the same way — a corner where the age is ten times the mean while the mean itself is exactly as designed. Six air changes an hour is a mean age of ten minutes and says nothing about whether some of the air has been there for a hundred.

A mixing vessel. The quality of mixing is a statement about the spread of ages, not about the mean of them — two vessels with a mean residence time of exactly 1 can have their middle eighty per cent of fluid leaving between 0.36 and 1.89 or between 0.91 and 1.09, a factor of 8.5 in the window at the same mean, and the spread is set by how the flow map folds fluid rather than by how much power the impeller delivers — which is the argument two strainings, and the order they came in makes about stretching, applied to time instead of length.

And a sampling line. A probe connected to an analyser by a metre of tube reports the composition of gas that entered the tube some seconds ago, and the smearing is the tube’s own residence-time distribution. It is the reason a fast analyser on a slow line is a slow measurement, and the reason the correction is a deconvolution rather than a time shift.

What the solver computed, and how it was checked #

The velocity field is written down and the trajectory is integrated from it, so the checks are aimed at the claims rather than at the discretisation.

That the local derivative really is zero. It is required to be the number zero rather than a small number, because the field has no time in it and anything else would be a bug.

That the exit speed is the area ratio. The check refuses a departure above 10⁻⁹ from 4 and reads 4.000000000, which is continuity holding rather than an accident of the parameters.

That the stepped trajectory matches the closed form. The exponential solution is available, and the two agree to zero difference — which is what allows the transit time to be quoted to the digits it is quoted to.

And that the window is the transit. The first version of this calculation ran the parcel for a round unit of time rather than to the exit, and reported a speed thirty-seven times the inlet value in a duct of area ratio four. That is the model extrapolated far past its own geometry, and the window is now the transit time computed from the geometry rather than a number chosen for tidiness.

Age and the entrance length are the same question #

There is a neighbour to this essay in the viscous field and the connection is close enough to state.

How far before a duct forgets what was fed into it computes the entrance length: the distance a duct needs before its profile stops depending on what was pushed in at the inlet. That is a memory question with a length as its answer, and the length is the transit distance over which viscous diffusion has time to cross the duct.

Written as a ratio of two times it becomes obvious: the profile forgets when the residence time exceeds the diffusion time across the duct. The entrance length is that statement multiplied by the speed, and the Reynolds number appears in it for no other reason.

The same construction runs through this collection wherever a regime is decided. A Deborah number is a material’s memory divided by the flow’s; a Womersley number is a diffusion time divided by an oscillation period; a Damköhler number is a residence time divided by a reaction time. In every one of them the residence time is the flow’s contribution, and the quantity it is compared against belongs to whatever else is happening.

The one number the mean age is good for #

Having said the mean is often the wrong quantity, it is worth stating precisely what it is right for, because it is an exact result and it is free.

The mean age of the fluid leaving a vessel equals the vessel’s volume divided by the volumetric flow rate through it. That follows from conservation alone: the volume is the integral of what is inside, the flow rate is what crosses the boundary, and nothing about the arrangement inside enters. It holds for a stirred tank, a plug-flow tube, a vessel with a dead zone and a vessel with a short circuit, identically.

So a mean residence time measured by tracer and disagreeing with volume over flow rate is evidence of an error in one of the three numbers, not of an unusual flow. That is the standard use of the result in practice — it is a closure check on an experiment rather than a prediction — and it is a good example of a conservation law being most useful as an audit rather than as an answer.

What the picture cannot show #

The duct drawn in the first figure has walls that converge, and the parcel is drawn on the axis. A real contraction has a boundary layer on those walls in which the fluid is slower and its residence time is therefore longer — in principle unboundedly so, since the fluid at the wall does not move at all.

That is not a small correction to the residence-time distribution; it is the reason the distribution has a long tail in every real apparatus. Nothing here contains it, because the model is one-dimensional and has no wall in it.

Who found it, and when #

Residence-time distributions as a discipline are Danckwerts’, from 1953, in chemical engineering rather than fluid mechanics — which is where the practical questions were. The age equation in the form used here, as a scalar with unit source, came later and from the ventilation and oceanography literature, where “the age of the water” is a quantity people genuinely want and can measure with tracers.

The fluid-mechanical statement is older and is Euler’s: the distinction between what happens at a point and what happens to a parcel is the whole content of the material derivative, and everything above is a consequence of the convective term being the only one that survives in a steady flow.

Limits recorded rather than smoothed over #

One dimension, one streamline. The whole calculation follows a single parcel down the axis of a duct with no cross-stream variation. That is enough for the arithmetic about acceleration and transit and it is not a residence-time distribution, which needs the whole cross-section and the wall.

Incompressible. The speed rises as the area falls, exactly. In a compressible duct the density changes too and the relation between area and speed is the one the duct that works backwards is about, where the sign of the effect reverses above the speed of sound.

The age field is described and not solved. The transport equation for age is written down here and its two consequences are used, and no figure solves it on a grid: the duct is one-dimensional, so its age field is the transit integral and nothing more. A two-dimensional age field, with its corner where the fluid is ten times older than the mean, would need the wall this model does not have — and the wall is where the thin layer.

No pressure, no viscosity, no forces. The duct here is a kinematic object: a prescribed speed against position. What pressure gradient would be needed to produce that acceleration, and what the walls would have to do about it, is a separate calculation and is not made.

And the transit time is exact for this speed law. A different contraction shape gives a different integral, and the logarithm is a property of the linear speed rise rather than of contractions in general. The trade — acceleration as the square, time as the logarithm — is likewise this law’s, though the direction of it is not.