How Science Works | How Pipe Flow Spreads a Small Drop of Dye

Consider steady, fully developed laminar flow of a Newtonian liquid in a straight circular pipe with stationary no-slip walls. Its axial velocity profile is parabolic: u(r) = 2U[1 − (r/a)²], where r is distance from the centre, a is pipe radius and U is cross-sectional mean speed.

At the centre, the fluid moves at 2U. At the wall, the ideal no-slip velocity is zero. At r = a/√2, it equals U. These are local speeds within the same steady flow, not separate pumps or a time-varying average.

A dye molecule near the centre can therefore move downstream faster than one near the wall. A narrow initial band becomes stretched. If every molecule remained on its original streamline, that difference in speed would keep adding to the difference in distance travelled.

The relevant width is downstream, along the pipe axis. The tracer can simultaneously be spreading across the pipe. Axial and transverse spreading answer different questions and operate on different length scales.

For an original diagram, draw a short coloured stripe across a pipe and a set of velocity arrows that are longest in the middle. In a second panel, stretch the stripe forward in the middle and leave its edges behind. Label this panel shear stretching before cross-stream exchange has erased the initial travel histories. It shows one ingredient, but not yet the complete long-time dispersion mechanism.

Contents · Next chapter

Molecular diffusion produces random motion relative to the mean liquid flow. Some dye that was travelling near the wall reaches faster regions; some that was near the centre reaches slower regions. Over time, a molecule samples a range of axial speeds.

This exchange is crucial. The molecule does not remain permanently assigned to a fast or slow lane. Its downstream displacement reflects a history of velocities whose memory is gradually lost as it moves across the cross-section.

In the appropriate long-time regime, many such histories produce a cross-sectionally averaged cloud that spreads approximately like a one-dimensional diffusion process while its centre is carried downstream. The effective axial coefficient can greatly exceed the molecular diffusion coefficient.

That does not mean the molecules themselves acquired a new microscopic random-walk speed. The enlarged coefficient summarises the combined effect of the organised velocity profile and transverse molecular motion.

G. I. Taylor’s 1953 work established the classical tube-flow mechanism. Aris later reformulated the problem using moments of the solute distribution, connecting its mean position and variance with a more general dispersion description. The original Taylor paper and Aris’s original analysis, reproduced in a collected volume are the historical primary sources.

A good explanation therefore keeps both arrows in the diagram: downstream motion with a nonuniform speed and random cross-stream exchange. Erasing either arrow changes the problem, sometimes enough to invalidate the familiar formula entirely.

Contents · Previous chapter · Next chapter

The mean position of the dye cloud tells us where its centre of mass is. Its variance tells us how widely the dye is distributed around that centre. Neither alone tells us every detail of the local concentration field.

For a sufficiently localised passive pulse in the long-time effective description, the mean advances at U. The axial variance grows at rate 2Deff, where Deff is the effective dispersion coefficient. Variance has units of length squared, so its rate has units of length squared per time.

A broad cloud is not necessarily well mixed at every smaller scale. Early in the process, a long stretched dye pattern can still contain strong concentration differences across the pipe. The coarse one-dimensional model intentionally averages over that transverse information.

Imagine two invented records. Cloud A moves 10 cm with little change in width. Cloud B moves the same 10 cm but becomes much broader. Their mean transport is similar; their dispersion differs. A third cloud could broaden without any imposed mean flow through ordinary molecular diffusion.

Likewise, a lower peak concentration need not mean that dye has chemically disappeared. With conserved tracer mass, broadening can lower the peak while leaving the area under the cross-sectionally averaged concentration-versus-distance curve unchanged, after accounting for pipe area.

This separation of centre, width and total amount is an evidence tool. If a model fits the peak but loses tracer mass, or predicts the centre correctly while missing the variance growth, it has not explained the whole observation.

Contents · Previous chapter · Next chapter

For the circular, fully developed, steady laminar flow described here, with a dilute passive solute of constant molecular diffusivity D, the long-time coefficient is Deff = D + U²a²/(48D). The first term is molecular axial diffusion; the second is the shear-enhanced contribution.

U is the cross-sectional mean speed, not the centreline speed. The length a is radius, not diameter. Replacing either with a value twice as large makes the enhancement four times too large. Making both mistakes multiplies it by sixteen.

The model assumes a straight uniform tube, no solute reaction or wall uptake, and a tracer that does not change the flow. It also requires sufficient time for transverse diffusion to establish the long-time regime. A useful order-of-magnitude cross-stream timescale is τD = a²/D; the asymptotic description needs times well beyond this scale and appropriate axial scale separation.

The exact approach to the regime depends on initial conditions and what feature is being measured. A single multiple of τD is not a universal guarantee of accuracy. In practice, researchers can examine whether the late variance-growth rate has settled.

At earlier times, the injection’s radial position and shape can matter strongly. A concentrated centreline release initially samples different velocities from a release spread uniformly across the cross-section. The late effective description does not say those early histories were identical.

The coefficient 1/48 belongs to the stated circular-tube model. Other cross-sections, changing flow profiles, oscillatory forcing, surface interactions and non-Newtonian liquids require their own analysis rather than a cosmetic relabelling of the same equation.

Contents · Previous chapter · Next chapter

Choose an original hypothetical tube with radius a = 100 micrometres, mean speed U = 1.0 mm/s and molecular diffusivity D = 1.0 × 10⁻⁹ m²/s. Assume the flow is already fully developed and the tracer is passive.

The cross-stream diffusion scale is a²/D = (10⁻⁴)²/10⁻⁹ = 10 s. The radius-based Péclet number Ua/D is 100. This number compares axial transport over the radius scale with molecular diffusion; it does not by itself certify that a particular measurement is late enough.

The enhancement is U²a²/(48D) = 2.0833 × 10⁻⁷ m²/s. Adding D gives Deff = 2.0933 × 10⁻⁷ m²/s, about 209 times the molecular value.

Now suppose a late observation is made at a reference time of 1000 s, after the initial injection transient. We use the effective model from that reference time onward, not from the instant of injection. Suppose its measured axial standard deviation is 25.0 mm, including earlier dispersion and the initial injection width. This is consistent with substantial spreading already having occurred. What does the model predict after another 100 s?

Variance increases by 2DeffΔt = 4.1867 × 10⁻⁵ m². Adding the initial variance, (0.025)² m², and taking the square root gives a new standard deviation of 25.824 mm. During the same interval, the centroid advances UΔt = 0.10 m.

This requires a tube long enough that the cloud is not affected by an end. If density is hypothetically 1000 kg/m³ and viscosity 0.001 Pa s, the diameter-based Reynolds number is 0.2, consistent with a laminar model. The calculation remains a model example, not a validated apparatus design.

Contents · Previous chapter · Next chapter

The enhancement term is inversely proportional to D. At first this seems backwards: how could faster diffusion produce less spreading? The answer is that transverse diffusion controls how long a molecule retains an unusually fast or slow downstream history.

Faster cross-stream exchange shortens that memory. Molecules move between fast and slow regions more frequently, reducing the extra axial spreading caused by persistent velocity differences. Meanwhile, their direct molecular axial diffusion becomes stronger. The total coefficient includes both effects.

In the chapter 5 example, doubling D to 2.0 × 10⁻⁹ m²/s while keeping U and a fixed halves the enhancement to 1.0417 × 10⁻⁷ m²/s. The total becomes 1.0617 × 10⁻⁷ m²/s. It is smaller than before because the enhancement dominates in these chosen conditions. The transverse diffusion scale also halves, from 10 to 5 s.

Do not generalise this comparison to every value of D. If molecular diffusion becomes sufficiently dominant, increasing D increases the total coefficient. The statement concerns a competition between D and a constant divided by D.

Most importantly, do not set D = 0 in the long-time formula and announce infinite physical spreading. As D approaches zero, τD grows without bound. A fixed observation time then ceases to satisfy the condition used to derive the expression.

The formula’s divergence is a warning about exchanging limits: taking the late-time regime first is not the same as eliminating cross-stream diffusion first. The mechanism itself tells us why that mathematical shortcut is invalid.

Contents · Previous chapter · Next chapter

We can calculate a contrasting original idealisation. Release a very thin dye sheet uniformly across the circular area and set diffusion to zero. Assume no wall uptake, no flow fluctuations and the same parabolic velocity profile. Each molecule now stays on its original streamline.

For particles distributed uniformly over circular area, the quantity r²/a² is uniformly distributed between zero and one. Because u = 2U[1 − r²/a²], the sampled axial velocities are uniformly distributed from zero to 2U.

Their mean velocity is U and their velocity variance is U²/3. After time t, an initially negligible axial width therefore develops a spatial variance U²t²/3. The standard deviation grows proportionally to t, rather than to the square root of time characteristic of late diffusive spreading.

With U = 1 mm/s and t = 10 s, the standard deviation is Ut/√3, approximately 5.77 mm. This is a result for the zero-diffusion, area-uniform initial sheet, not a prediction from the long-time Taylor–Aris coefficient and not a universal early-time value for every injection.

The comparison explains why “shear stretches dye” is true but incomplete. Without transverse exchange, travel histories remain correlated indefinitely in this ideal model. With diffusion and enough time, those histories decorrelate and a constant late dispersion coefficient emerges.

A recent primary theoretical study explicitly addressed streamwise dispersion over the evolving process using a Lagrangian framework with time-dependent transport descriptions. Its focus reinforces the importance of separating transient behaviour from a late asymptotic coefficient. Guan and Chen, Journal of Fluid Mechanics, 2024.

Contents · Previous chapter · Next chapter

The enhancement U²a²/(48D) grows as a² if mean speed U and D are held fixed. Doubling the radius would then quadruple the enhancement and multiply the transverse diffusion timescale by four.

But suppose instead that the volume flow rate Q is fixed. Since U = Q/(πa²), doubling radius reduces mean speed to one-quarter. The enhancement changes by (1/4)² × 2² = 1/4. The same radius change now decreases the enhancement.

For another ideal comparison, hold the pressure gradient and viscosity fixed while the fully developed Newtonian laminar model remains valid. Poiseuille flow gives U proportional to a². The enhancement then scales as a⁶, so doubling radius produces a factor of 64. This is a different experiment, and the regime and available length must still be checked.

These three answers do not contradict one another. They hold different quantities fixed. The algebra is useful because it exposes the hidden experimental choice behind a sentence such as “wider pipes disperse more.”

Radius changes also alter the time needed to reach the asymptotic regime. Comparing two tubes at the same elapsed time may compare different stages of their transient development. A width difference could therefore reflect both a different late coefficient and a different distance from that late regime.

When explaining an observed change, identify the control variable first: imposed pressure difference, pressure gradient, volume flow or mean speed. Then compare like with like, including diffusivity, cross-section, wall behaviour and the period over which variance was measured.

Contents · Previous chapter · Next chapter

A useful paper-based investigation begins with cross-sectionally averaged concentration profiles at several late times. Estimate total tracer amount, centroid and variance from each profile, retaining units and measurement uncertainty. Plot the variance against time.

In the effective regime, a straight-line slope supports a constant Deff, with slope 2Deff. For an original synthetic dataset, let variance be 625, 645.93 and 666.87 mm² at reference offsets 0, 50 and 100 s. The approximate slope is 0.4187 mm²/s, implying Deff ≈ 0.2093 mm²/s, or 2.093 × 10⁻⁷ m²/s.

A width-time fit alone does not prove the circular-pipe mechanism. Injection duration, detector averaging, wall adsorption, changing flow and other processes can broaden a signal. Compare several flow speeds: the model predicts that Deff − D varies as U² when radius, diffusivity and the other conditions stay fixed.

Also distinguish a spatial profile at one time from a time trace at a fixed downstream detector. Their widths are not interchangeable without a transport model. A slowly responding detector can broaden an arrival curve even if the fluid’s actual dispersion is unchanged.

Current research continues to test which assumptions can be relaxed. In June 2026, Boyko and Stone derived a dispersion description for shear-rate-dependent flows that couples viscosity and transverse diffusivity through a generalised Stokes–Einstein assumption. This is a model-dependent extension, not evidence that the Newtonian 1/48 formula applies unchanged to arbitrary complex fluids. Journal of Fluid Mechanics.

The best evidence checks the mechanism’s predicted changes under controlled conditions, rather than celebrating a single broad, colourful dye cloud.

Contents · Previous chapter · Next chapter

First, compare the late width calculation with molecular diffusion alone. Starting from the same 25.0 mm reference standard deviation and adding 100 s of diffusion with D = 10⁻⁹ m²/s gives √[6.25 × 10⁻⁴ + 2 × 10⁻⁷] m, approximately 25.004 mm. The effective-flow example gave 25.824 mm. Both retain the same initial width; neither assumes an infinitely narrow pulse at the reference time.

Second, ask what happens if the mean speed doubles while radius and diffusivity remain fixed. The shear enhancement quadruples, but the molecular term stays unchanged. The new Deff is approximately 8.343 × 10⁻⁷ m²/s, not exactly four times the previous total.

Third, audit a claim that a tube with diameter 200 micrometres has a = 200 micrometres. The formula needs its radius, 100 micrometres. Fixing this one definition changes the enhancement by a factor of four. Units alone would not catch the error because both radius and diameter are lengths.

Finally, label every result as one of three kinds: a measured profile, an inferred effective coefficient or a prediction conditional on the pipe model. An attractive fit cannot by itself establish the absence of adsorption, reaction or transient effects.

This is a safe investigation to do with supplied data, diagrams or a simulation. No pressure source, chemical tracer or environmental release is needed. Real laboratory work requires suitable materials, training and controls.

The central mechanism is now precise: shear creates different downstream speeds; transverse diffusion changes which speeds each molecule samples; long-time averaging produces enhanced axial dispersion. Return to Fluid Dynamics or the How Science Works master guide with a stronger habit: carry the geometry, the held-constant variables and the relevant timescale alongside every formula.

Contents · Previous chapter · Continue to Fluid Dynamics