The Navier–Stokes Equations and Viscous Flow

The Navier–Stokes equations are the governing equations of real fluid motion — Euler's ideal-flow system completed by the internal friction every actual fluid exhibits. Conceptually they are nothing more than the continuum-mechanics program of this chapter run to its natural end: mass conservation from the kinematics article, momentum balance through the Cauchy stress tensor, and a constitutive law closing the system. The constitutive step is the dynamical cousin of Hooke's law in linear elasticity, with one decisive substitution — the viscous stress is linear in the rate of strain, not the strain. A fluid does not care how far it has been sheared, only how fast. The equations couple a vector velocity field to a scalar pressure through both the momentum balance and the incompressibility constraint, and the material derivative plants the quadratic term $(\mathbf{v}\cdot\nabla)\mathbf{v}$ in the middle of it all. That single term is responsible for nearly everything difficult and interesting in fluid mechanics: vortex stretching, the energy cascade, turbulence, and the fact that exact solutions exist only in geometries symmetric enough to kill it — pipe flow, sliding plates, a slowly settling sphere. One dimensionless group organizes the entire subject. The Reynolds number $Re=\rho UL/\mu$ weighs inertia against viscosity: honey at $Re\ll1$ creeps in linear, time-reversible Stokes flow, while air past a wing at $Re\sim10^7$ is nearly ideal everywhere except paper-thin boundary layers where the no-slip condition lives. Between the two regimes sits the transition to turbulence — and beneath the whole structure sits an open wound in mathematics: whether smooth three-dimensional solutions persist for all time is a Clay Millennium Problem, a million dollars still unclaimed for the equation that moves the weather.

Key takeaways

The Navier–Stokes equations are Newton's second law for a fluid continuum: material acceleration balanced by pressure gradient, viscous stress divergence, and body force.
The Newtonian constitutive law $\sigma_{ij}=-p\,\delta_{ij}+2\mu(e_{ij}-\tfrac13\delta_{ij}e_{kk})+\zeta\delta_{ij}e_{kk}$ is Hooke's law with strain replaced by strain rate; for incompressible flow it collapses the momentum equation to $\rho\,\mathrm{D}\mathbf{v}/\mathrm{D}t=-\nabla p+\mu\nabla^2\mathbf{v}+\mathbf{f}$ .
Viscosity is momentum diffusion: $\nu=\mu/\rho$ is a diffusivity with units $\mathrm{m^2/s}$ , and — counterintuitively — air ( $\nu\approx1.5\times10^{-5}$ ) out-diffuses water ( $\nu\approx1.0\times10^{-6}$ ).
The Reynolds number $Re=\rho UL/\mu$ sets the regime: linear reversible Stokes flow below $Re\sim1$ , boundary layers of thickness $\delta\sim L/\sqrt{Re}$ and turbulence at large $Re$ (pipes transition near $Re\approx2300$ ).
Exact solutions (Poiseuille's $Q\propto R^4$ pipe law, Stokes' drag $F=6\pi\mu RU$ ) exist only where symmetry kills the nonlinearity; global smoothness of 3D solutions is an unsolved Millennium Problem.
Key idea: the Navier–Stokes equations are Newton's second law for a fluid continuum — mass times material acceleration equals the divergence of a stress made of isotropic pressure plus a viscous part linear in the strain rate.

The core insight

The Navier–Stokes equations are Newton's second law for a fluid continuum: mass times the material acceleration of a fluid parcel equals the divergence of the stress tensor plus body forces, with the stress made of an isotropic pressure and a viscous part linear in the strain rate.

Where this lives today

The equations of this article consume a measurable fraction of the world's supercomputing. Weather and climate centers integrate filtered Navier–Stokes (with rotation and stratification) on planetary grids where direct simulation of all scales — down to Kolmogorov's millimeters at atmospheric $Re\sim10^9$ — would demand more floating-point operations than exist; the gap is bridged by turbulence closures whose calibration against the $k^{-5/3}$ cascade remains a live research industry. Aerospace runs the same equations as RANS and large-eddy solvers over full aircraft; Formula 1 teams are granted CFD budgets by regulation, wind-tunnel hours and core-hours being competitively equivalent currencies. The newest wave couples the solvers to machine learning — neural subgrid models and differentiable-CFD pipelines trained on resolved simulations — with the classical equations retained as the trusted arbiter. The mathematics is equally alive. The Clay Millennium Problem — do smooth 3D solutions exist for all time? — stands unclaimed, with the vortex-stretching term of equation (7) as the suspected culprit: 2D flow, where stretching vanishes, has been provably smooth since the 1960s. Recent progress runs through computer-assisted analysis: proofs of finite-time blow-up for the related Euler equations near boundaries, and self-similar singularity candidates found with physics-informed neural networks, have sharpened the question without settling it. A million dollars still waits on whether the equation that delivers your weather forecast can break down in finite time.

The physical picture

Picture a small cube of fluid drifting with the flow. Its momentum changes for three reasons: pressure pushes harder on one face than the opposite one ( $-\nabla p$ ), neighboring layers moving at different speeds drag on its sides through friction ( $\mu\nabla^2\mathbf{v}$ ), and gravity acts on its mass. Newton's law for the cube is the Navier–Stokes equation — the only new physics relative to Euler is the drag between layers. That drag is momentum diffusion. Just as heat diffuses down a temperature gradient, transverse momentum diffuses down a velocity gradient: molecules wandering from the fast layer into the slow one deliver excess momentum, and vice versa. The Laplacian is precisely the diffusion operator, and $\nu=\mu/\rho$ is the diffusivity of momentum — which is why air, thin as it is, spreads momentum fifteen times faster than water.

Figure

Viscosity as momentum diffusion between sliding layers — the fast layer drags the slow one through the shear stress mu du/dy, momentum leaking down the velocity gradient exactly like heat down a temperature gradient

The convective term $(\mathbf{v}\cdot\nabla)\mathbf{v}$ carries over from the ideal-flow articles: parcels accelerate through steady patterns, the equations are quadratic in $\mathbf{v}$ , and small disturbances can amplify, fold, and cascade to ever finer scales. What is new is the referee. Viscosity damps the smallest scales, inertia feeds them, and the Reynolds number $Re=UL/\nu$ scores the contest: honey poured slowly ( $Re\sim10^{-1}$ ) is all viscosity — creeping, reversible, and eerily time-symmetric; air past an aircraft ( $Re\sim10^7$ ) is all inertia except in a millimeters-thin boundary layer where the no-slip condition must be honored. Every named regime of fluid mechanics is an entry in the table of $Re$ .

The core equations

Mass conservation carries over unchanged from the kinematics article — no constitutive physics ever touches it:

\frac{\partial\rho}{\partial t}+\nabla\cdot(\rho\mathbf{v})=0 \qquad (1)

where: $\rho$ — density; $\rho\mathbf{v}$ — mass flux. For incompressible flow this is the constraint $\nabla\cdot\mathbf{v}=0$ , which the pressure field enforces instantaneously — pressure in incompressible flow is a Lagrange multiplier, not a thermodynamic variable. Momentum balance is Cauchy's equation of motion, the continuum $\mathbf{F}=m\mathbf{a}$ built in the Stress Tensor and Cauchy's Relation article, now with the material acceleration written out:

\rho\left(\frac{\partial\mathbf{v}}{\partial t}+(\mathbf{v}\cdot\nabla)\mathbf{v}\right)=\nabla\cdot\boldsymbol{\sigma}+\mathbf{f} \qquad (2)

where: $\boldsymbol{\sigma}$ — the Cauchy stress tensor; $\mathbf{f}$ — body force per unit volume. This holds for any continuum whatsoever — solid, fluid, or silly putty. The physics of "fluid" enters only through the choice of $\boldsymbol{\sigma}$ . That choice is the Newtonian constitutive law. Demand that the viscous stress be linear in the velocity gradient, vanish for rigid rotation (only the strain-rate tensor $e_{ij}=\tfrac12(\partial_iv_j+\partial_jv_i)$ from the vorticity article's decomposition can enter), and respect isotropy. Two coefficients survive:

\sigma_{ij}=-p\,\delta_{ij}+2\mu\left(e_{ij}-\tfrac13\delta_{ij}\,e_{kk}\right)+\zeta\,\delta_{ij}\,e_{kk} \qquad (3)

where: $p$ — pressure; $\mu$ — dynamic (shear) viscosity, resisting shape change; $\zeta$ — bulk viscosity, resisting compression. This is Hooke's law for solids with strain replaced by strain rate — the deep structural parallel of this chapter. A fluid holds no memory of how far it has deformed, only of how fast it is deforming; that one substitution turns elastostatics into fluid dynamics. Insert (3) into (2) and impose incompressibility (Derivation A) and the system takes its famous form:

\rho\left(\frac{\partial\mathbf{v}}{\partial t}+(\mathbf{v}\cdot\nabla)\mathbf{v}\right)=-\nabla p+\mu\nabla^2\mathbf{v}+\mathbf{f},\qquad \nabla\cdot\mathbf{v}=0 \qquad (4)

where: $\mu\nabla^2\mathbf{v}$ — viscous diffusion of momentum; four scalar equations for the four unknowns $(\mathbf{v},p)$ . Strike the viscous term and Euler's equation from two articles ago reappears; the whole ideal-flow edifice is the $\mu\to0$ face of this system — with one treacherous subtlety about walls that the warning box below spells out. Dividing by $\rho$ exposes the diffusivity and the anchor numbers worth memorizing:

\frac{\partial\mathbf{v}}{\partial t}+(\mathbf{v}\cdot\nabla)\mathbf{v}=-\frac{1}{\rho}\nabla p+\nu\nabla^2\mathbf{v},\qquad \nu=\frac{\mu}{\rho} \qquad (5)

where: $\nu$ — kinematic viscosity, units $\mathrm{m^2/s}$ . Water at 20°C: $\mu=1.0\times10^{-3}\ \mathrm{Pa\,s}$ , $\nu=1.0\times10^{-6}\ \mathrm{m^2/s}$ . Air: $\mu=1.8\times10^{-5}\ \mathrm{Pa\,s}$ but $\nu=1.5\times10^{-5}\ \mathrm{m^2/s}$ — fifteen times water's. In the momentum-diffusion race the relevant stake is $\nu$ , not $\mu$ . Non-dimensionalize equation (5) with a characteristic speed $U$ and length $L$ and a single group survives, the ratio of the inertial term $\rho U^2/L$ to the viscous term $\mu U/L^2$ :

Re=\frac{\rho U L}{\mu}=\frac{UL}{\nu} \qquad (6)

where: $Re$ — the Reynolds number. Two flows with the same geometry and the same $Re$ are dynamically identical — the similarity principle that makes wind tunnels and scale models legitimate. Bacteria swim at $Re\sim10^{-5}$ , blood flows at $Re\sim10^3$ , cars drive at $Re\sim10^6$ , and the atmosphere churns at $Re\sim10^{9}$ and beyond.

Figure

The two canonical faces of viscous flow — the Poiseuille parabola pinned to zero at both no-slip walls, and the high Reynolds number boundary layer squeezing all of viscosity's work into a thin sheet of thickness L over root Re

Taking the curl of equation (5) eliminates pressure entirely and upgrades the ideal vorticity equation of the previous article with a diffusion term:

\frac{\mathrm{D}\boldsymbol{\omega}}{\mathrm{D}t}=(\boldsymbol{\omega}\cdot\nabla)\mathbf{v}+\nu\nabla^2\boldsymbol{\omega} \qquad (7)

where: $(\boldsymbol{\omega}\cdot\nabla)\mathbf{v}$ — vortex stretching, the 3D amplifier; $\nu\nabla^2\boldsymbol{\omega}$ — viscous diffusion, the term that finally lets vorticity be created (at no-slip walls) and destroyed, breaking Kelvin's theorem exactly where real flows demand it. The contest between these two terms — stretching feeding vorticity to small scales, diffusion burning it there — is turbulence in one line. The energy ledger closes the story. Dotting equation (4) with $\mathbf{v}$ and integrating gives

\frac{\mathrm{d}}{\mathrm{d}t}\int_V\tfrac12\rho v^2\,\mathrm{d}V=-\int_V 2\mu\,e_{ij}e_{ij}\,\mathrm{d}V+(\text{work by surface and body forces}) \qquad (8)

where: $2\mu\,e_{ij}e_{ij}$ — the viscous dissipation rate per unit volume, a sum of squares and hence never negative. Viscosity is a one-way street: kinetic energy converts irreversibly to heat, the arrow of time that ideal flow lacked, and the reason equation (4) — unlike Euler — cannot be run backwards.

Derivation AFrom Cauchy momentum to the Navier-Stokes equation

We assemble the master equation by inserting the Newtonian constitutive law into the Cauchy momentum balance and imposing incompressibility. The strategy is pure index bookkeeping — differentiate the stress, watch which terms the constraint $\partial_kv_k=0$ kills, and let the survivors organize themselves into a Laplacian. Two structural facts fall out along the way: the bulk viscosity never enters incompressible flow, and the viscous force is proportional to the second derivative of velocity, making the equation diffusive.

Step 1. Start from momentum balance with the stress divergence written in components:

\rho\,\frac{\mathrm{D}v_i}{\mathrm{D}t}=\partial_j\sigma_{ij}+f_i

Step 2. Differentiate the constitutive law (3), treating $\mu$ and $\zeta$ as constants:

\partial_j\sigma_{ij}=-\partial_i p+\mu\,\partial_j\!\left(\partial_i v_j+\partial_j v_i-\tfrac23\delta_{ij}\partial_k v_k\right)+\zeta\,\partial_i(\partial_k v_k)

Step 3. Impose incompressibility, $\partial_kv_k=0$ . Every term carrying the divergence dies — including the entire bulk-viscosity contribution:

\partial_j\sigma_{ij}=-\partial_i p+\mu\left(\partial_i\partial_j v_j+\partial_j\partial_j v_i\right)=-\partial_i p+\mu\,\nabla^2 v_i

Why $\zeta$ vanishes: bulk viscosity resists volume change, and incompressible parcels change no volume; this is why sound absorption — a compressive phenomenon — is where $\zeta$ finally earns its keep. *Why a Laplacian: the term $\partial_i\partial_jv_j$ is the gradient of the divergence, zero by the constraint; the remaining $\partial_j\partial_jv_i$ is the diffusion operator acting on each velocity component.*

Step 4. Assemble, restoring the material derivative:

\rho\left(\partial_t\mathbf{v}+(\mathbf{v}\cdot\nabla)\mathbf{v}\right)=-\nabla p+\mu\nabla^2\mathbf{v}+\mathbf{f}

together with $\nabla\cdot\mathbf{v}=0$ : four equations, four unknowns, and no equation of state — pressure is whatever field keeps the velocity divergence-free. Two quick checks: setting $\mu=0$ recovers Euler's equation exactly, and dimensionally $[\mu\nabla^2 v]=\mathrm{Pa\,s\cdot(m/s)/m^2}=\mathrm{N/m^3}$ , matching $[\nabla p]$ as required.

Go deeper · Derivation B · 1 minPlane Poiseuille flow between parallel plates

Derivation BPlane Poiseuille flow between parallel plates

We solve the Navier–Stokes equations exactly for steady, pressure-driven flow between two stationary plates at $y=\pm h$ — the archetype of every exact viscous solution. The strategy: adopt the symmetry ansatz $\mathbf{v}=(u(y),0,0)$ , verify it kills both the incompressibility worry and the nonlinear term identically, and integrate the surviving linear ODE against the no-slip boundary condition. Watch how the nonlinearity dies by geometry, not by approximation — that is the only reason an exact solution exists at all.

Step 1. Check the ansatz against continuity: $\nabla\cdot\mathbf{v}=\partial_x u(y)=0$ identically. Consistent.

Step 2. Check the nonlinear term:

(\mathbf{v}\cdot\nabla)\mathbf{v}=u\,\partial_x\!\left(u(y)\right)\hat{\mathbf{x}}=0

Why it vanishes: the only nonzero velocity component varies only transversely to itself — parcels move along lines of constant velocity, so no parcel ever accelerates. The full nonlinear problem has become linear by symmetry alone.

Step 3. The $x$ -momentum equation, steady and body-force-free, with constant pressure gradient $\mathrm{d}p/\mathrm{d}x=-G$ ( $G>0$ drives flow toward $+x$ ):

0=G+\mu\,\frac{\mathrm{d}^2u}{\mathrm{d}y^2}

The $y$ -momentum equation gives $\partial_yp=0$ , so $p$ varies only along the channel — the transverse problem and the streamwise drive decouple.

Step 4. Integrate twice and apply no-slip $u(\pm h)=0$ :

u(y)=-\frac{G}{2\mu}y^2+Ay+B \quad\xrightarrow{\ u(\pm h)=0\ }\quad u(y)=\frac{G}{2\mu}\left(h^2-y^2\right)

Why no-slip: molecularly, fluid adheres to the wall; the boundary condition $\mathbf{v}=0$ on stationary solids is an experimental fact honored by every viscous solution, and symmetry then forces $A=0$ .

Step 5. Integrate the parabola for the flux per unit width:

Q=\int_{-h}^{h}u\,\mathrm{d}y=\frac{G}{2\mu}\left[2h^3-\frac{2h^3}{3}\right]=\frac{2Gh^3}{3\mu}

The cubic growth with channel half-width is the viscous signature; repeating the calculation in a circular pipe of radius $R$ gives the Hagen–Poiseuille law $Q=\pi R^4G/8\mu$ — a fourth power, the fact that dominates physiology and plumbing alike. Two internal checks on the solution: the maximum speed $u(0)=Gh^2/2\mu$ is $\tfrac32$ the mean speed $Q/2h$ — the standard parabolic-profile ratio — and the wall shear $\mu|u'(\pm h)|=Gh$ times the wetted perimeter exactly balances the pressure force $G\times$ cross-sectional area per unit length.

Go deeper · Derivation C · 2 minStokes drag on a slowly moving sphere

Derivation CStokes drag on a slowly moving sphere

We extract the drag law of creeping flow — the $Re\ll1$ regime where inertia is negligible and the equations linearize. The strategy: drop the inertial terms to get the Stokes equations, quote the classical axisymmetric stream-function solution satisfying no-slip on the sphere and uniform flow far away, and assemble the force by integrating the traction over the surface. The reward is the linear-in-velocity drag that underlies sedimentation, aerosol physics, and Millikan's electron-charge measurement.

Step 1. For $Re\ll1$ discard inertia entirely. The momentum equation becomes the linear Stokes system:

\nabla p=\mu\nabla^2\mathbf{v},\qquad \nabla\cdot\mathbf{v}=0

Why this is legitimate: the neglected term $\rho(\mathbf{v}\cdot\nabla)\mathbf{v}\sim\rho U^2/L$ stands to the kept term $\mu U/L^2$ exactly in the ratio $Re$ — small by hypothesis. Taking the divergence shows $\nabla^2p=0$ ; taking the curl shows vorticity is harmonic too.

Step 2. For a sphere of radius $R$ held in a stream $U\hat{\mathbf{z}}$ , the axisymmetric solution regular at infinity and obeying no-slip at $r=R$ has the Stokes stream function

\psi=U\sin^2\!\theta\left(\frac{r^2}{2}-\frac{3Rr}{4}+\frac{R^3}{4r}\right)

Why these three powers: $r^2$ is the uniform stream, $1/r$ a potential dipole, and $r$ the "Stokeslet" — the signature slow-flow term that decays so lazily that the sphere's influence reaches tens of radii, the root of the famous long-range interactions of suspensions.

Step 3. From $\psi$ compute the surface pressure and viscous traction:

p=p_\infty-\frac{3\mu U R}{2r^2}\cos\theta

with the corresponding shear stress on $r=R$ . Both distributions point systematically downstream.

Step 4. Integrate the total traction $\boldsymbol{\sigma}\cdot\hat{\mathbf{n}}$ over the sphere. Pressure contributes $2\pi\mu RU$ (one third) and viscous shear $4\pi\mu RU$ (two thirds):

F_{\text{drag}}=6\pi\mu R U

Why linear in $U$ : the equations are linear, so doubling the stream doubles every field and hence the force — in stark contrast to the $\rho U^2$ drag of high- $Re$ flow. Linearity also makes creeping flow time-reversible, with consequences the FAQ explores.

Step 5. Balance against gravity for a falling sphere of density $\rho_s$ in fluid of density $\rho_f$ : $6\pi\mu Rv_t=\tfrac43\pi R^3(\rho_s-\rho_f)g$ gives the terminal velocity

v_t=\frac{2(\rho_s-\rho_f)gR^2}{9\mu}

The $R^2$ scaling is the workhorse of viscometry and sedimentation analysis; Millikan's oil drops ( $R\sim1\ \mu\mathrm{m}$ ) fall at the $\sim10^{-4}\ \mathrm{m/s}$ this formula predicts, slow enough to watch — which is precisely why the experiment worked.

Limiting cases

Case	Result	Meaning
Creeping flow, $Re\ll1$	$\nabla p=\mu\nabla^2\mathbf{v}$ , $\nabla\cdot\mathbf{v}=0$	Linear, time-reversible Stokes flow: sedimentation, microswimmers, lubrication; sphere drag $6\pi\mu RU$ .
Ideal limit, $Re\to\infty$	Euler equations + boundary layers $\delta\sim L/\sqrt{Re}$	Singular limit: viscosity survives at walls however large $Re$ ; ignoring it gives d'Alembert's zero drag.
Incompressible limit, $M\ll1$	$\nabla\cdot\mathbf{v}=0$ , $\rho$ const	Below Mach $\approx0.3$ pressure enforces the constraint instantaneously; no equation of state needed.
Two dimensions	stretching term $(\boldsymbol{\omega}\cdot\nabla)\mathbf{v}=0$	Vorticity merely advects and diffuses; global smoothness is a theorem in 2D — the Millennium Problem is strictly 3D.
Pipe transition	$Re\approx2300$	Laminar Poiseuille flow destabilizes; friction rises sharply and empirical correlations take over.
Fully turbulent, $Re\gg10^4$	$E(k)\sim\varepsilon^{2/3}k^{-5/3}$	Kolmogorov's 1941 cascade: energy injected at large scales flows through an inertial range to viscous dissipation at the smallest.

Common trapThe inviscid limit is singular — a large Reynolds number never licenses deleting viscosity at a wall

Setting $\mu=0$ (Euler) and taking $Re\to\infty$ (Navier–Stokes at high Reynolds number) are different theories at every wall. The viscous term multiplies the highest derivative in the equation, so removing it changes the boundary conditions the system can honor: Euler flow slips along surfaces, real flow cannot. However large $Re$ grows, a boundary layer of thickness $\delta\sim L/\sqrt{Re}$ survives to enforce no-slip, and inside it vorticity is manufactured, skin friction is exerted, and separation is decided — the phenomena behind stall, form drag, and the resolution of d'Alembert's paradox. Dropping the viscous term because " $Re$ is huge" throws away exactly the physics that distinguishes a wing that flies from one that has stalled.

Worked examples

Example 1 — an IV drip through a fine needle (confidence builder). Saline ( $\rho=1000\ \mathrm{kg/m^3}$ , $\mu=1.0\times10^{-3}\ \mathrm{Pa\,s}$ ) is pushed through a needle of radius $R=0.40\ \mathrm{mm}$ and length $L=40\ \mathrm{mm}$ by a pressure difference $\Delta p=4.0\ \mathrm{kPa}$ . Find the flow rate and check the flow is laminar.

Step 1. Hagen–Poiseuille (the pipe version of Derivation B): $Q=\dfrac{\pi R^4\,\Delta p}{8\mu L}$ . Compute $R^4=(4.0\times10^{-4})^4=2.56\times10^{-14}\ \mathrm{m^4}$ .

Step 2. $Q=\dfrac{\pi\times2.56\times10^{-14}\times4.0\times10^{3}}{8\times1.0\times10^{-3}\times0.040}=\dfrac{3.22\times10^{-10}}{3.2\times10^{-4}}=1.0\times10^{-6}\ \mathrm{m^3/s}$ — that is $1.0\ \mathrm{mL/s}$ .

Step 3. Mean speed: $\bar v=Q/\pi R^2=1.0\times10^{-6}/(5.03\times10^{-7})=2.0\ \mathrm{m/s}$ .

Step 4. Reynolds check: $Re=\rho\bar v(2R)/\mu=1000\times2.0\times8.0\times10^{-4}/10^{-3}=1600<2300$ — laminar, so Poiseuille was legitimate.

Step 5. $Q=1.0\ \mathrm{mL/s}$ at $Re=1600$ . The $R^4$ law is the clinical headline: switching to a needle of half the radius at the same pressure would cut delivery sixteen-fold — and the same law is why a 25% narrowing of a coronary artery raises the pressure cost of the same blood flow by a factor of $(1/0.75)^4\approx3.2$ .

Example 2 — why fog does not fall (exam level). A fog droplet of radius $r=10\ \mu\mathrm{m}$ (water, $\rho_s=1000\ \mathrm{kg/m^3}$ ) settles in still air ( $\rho_f=1.2\ \mathrm{kg/m^3}$ , $\mu=1.8\times10^{-5}\ \mathrm{Pa\,s}$ ). Find its terminal speed, verify the Stokes regime, and estimate the settling time from a 100 m cloud base.

Step 1. Terminal velocity from Derivation C (buoyancy negligible since $\rho_f\ll\rho_s$ ):

v_t=\frac{2\rho_s g r^2}{9\mu}=\frac{2\times1000\times9.81\times(1.0\times10^{-5})^2}{9\times1.8\times10^{-5}}

Step 2. Numerator: $2\times1000\times9.81\times10^{-10}=1.96\times10^{-6}$ ; denominator: $1.62\times10^{-4}$ . So $v_t=1.96\times10^{-6}/1.62\times10^{-4}=1.2\times10^{-2}\ \mathrm{m/s}$ — about 1.2 cm/s.

Step 3. Validity check: $Re=\rho_f v_t(2r)/\mu=1.2\times0.012\times2\times10^{-5}/1.8\times10^{-5}=1.6\times10^{-2}\ll1$ . Deep in the Stokes regime — the linear drag law was legitimate.

Step 4. Settling time from 100 m: $t=100/0.012=8.3\times10^{3}\ \mathrm{s}\approx2.3$ hours — and the gentlest updraft (a few cm/s) suspends the droplet indefinitely.

Step 5. $v_t\approx1.2\ \mathrm{cm/s}$ , $Re\approx0.016$ , settling time hours. The $r^2$ scaling is the meteorology: double the droplet radius and it falls four times faster, which is why fog persists while drizzle ( $r\sim100\ \mu\mathrm{m}$ , borderline beyond Stokes) reaches the ground.

Origins · 1 minClaude-Louis Navier and how this was discovered

Claude-Louis Navier

1785–1836 · École des Ponts et Chaussées, Paris

Navier was a bridge engineer — France's foremost, author of the standard texts on structural mechanics and builder of suspension bridges — who in 1822 added molecular attraction and repulsion to Euler's fluid equations and produced, from a microscopic model now known to be wrong, exactly the right macroscopic equations. His viscosity coefficient emerged from hypothetical intermolecular forces; the equations survived because their form is fixed by symmetry and conservation, not by the microscopic story. Stokes rederived them in 1845 from the pure continuum argument used in this article — stress linear in strain rate — with no molecular hypothesis at all, which is why both names attach to the equations. Navier's own masterpiece, the Pont des Invalides, was cancelled mid-construction after a sewer collapse; his equations, indifferent to the irony, now simulate the Seine flowing past its replacement.

History

1755

— Euler publishes the equations of ideal flow; friction awaits a constitutive idea

1822

— Navier derives the viscous equations from a molecular-force model, introducing the viscosity coefficient

1845

— Stokes rederives them from the continuum hypothesis alone: stress linear in strain rate

1851

— Stokes solves creeping flow past a sphere: the 6-pi-mu-R-U drag law

1883

— Reynolds' dye experiments locate the laminar-turbulent transition and name the controlling number

1904

— Prandtl's boundary layer reconciles ideal flow with no-slip and resolves d'Alembert's paradox

1941

— Kolmogorov's cascade theory gives turbulence its minus-five-thirds spectrum

2000

— Global regularity of 3D solutions is named a Clay Millennium Problem, still open

Going further

References. Landau & Lifshitz, Fluid Mechanics (Vol. 6), Ch. I–II, §§15, 20 · Batchelor, An Introduction to Fluid Dynamics, Ch. 3–4 · Acheson, Elementary Fluid Dynamics, Ch. 2, 6 · Goldstein, Classical Mechanics, 3rd ed., Ch. 13 · Taylor, Classical Mechanics, Ch. 16 Key numerical landmarks: the Poiseuille channel flux $Q=2Gh^3/3\mu$ and pipe law $Q=\pi R^4G/8\mu$ ; Stokes drag $6\pi\mu RU$ splits as $2\pi\mu RU$ from pressure and $4\pi\mu RU$ from viscous shear; Example 1 gives $Q=1.0\times10^{-6}\ \mathrm{m^3/s}$ and $Re=1600$ ; Example 2 gives $v_t=1.2\times10^{-2}\ \mathrm{m/s}$ , $Re=1.6\times10^{-2}$ , and settling time $t\approx8.3\times10^3\ \mathrm{s}$ . Fluid properties at 20°C and the $Re\approx2300$ pipe-transition threshold are standard handbook values. The full story — boundary-layer theory, transition and instability, turbulence statistics and modeling, non-Newtonian fluids — lives in the Fluid Mechanics track (migrates later as its own subject); this chapter closes the mechanics thread at the equations themselves.

Builds on: Stress Tensor and Cauchy's Relation · Strain Tensor and Deformation of Continua · Linear Elasticity and Hooke's Law for Solids · Euler's Equations and Bernoulli's Theorem · Vorticity, Circulation, and Potential Flow Leads to: Elastic Waves in Solids · The One-Dimensional Wave Equation · Hamiltonian Density and Canonical Field Momentum

Where does the no-slip condition actually come from? It seems like an extra assumption smuggled in.

It is an empirical boundary condition, not derivable from the bulk equations — and it was controversial for a century, with Navier himself proposing partial slip. The molecular picture: fluid molecules adjacent to a solid exchange momentum with it so efficiently (adsorption, surface roughness at molecular scale) that the tangential velocity equilibrates to the wall's within a few molecular diameters. For continuum purposes that is $\mathbf{v}=\mathbf{v}_{\text{wall}}$ exactly. It fails measurably only where the continuum itself thins: rarefied gases (spacecraft reentry), nanochannels, and superhydrophobic surfaces show slip lengths of nanometers to microns. Everywhere else, no-slip is among the best-verified statements in fluid mechanics — and it is the sole reason boundary layers, skin friction, and vorticity generation at walls exist.

Honey is obviously more viscous than air. In what sense does air diffuse momentum faster?

Two different questions hide in "viscous." The shear stress produced by a given velocity gradient is set by the dynamic viscosity $\mu$ — honey's is a hundred thousand times air's, which is why stirring honey is hard work. But how fast a velocity disturbance spreads through the quiescent fluid is set by the kinematic viscosity $\nu=\mu/\rho$ : the momentum injected must be shared among the local mass, and air has very little mass to convince. Numbers: $\nu_{\text{air}}=1.5\times10^{-5}\ \mathrm{m^2/s}$ versus $\nu_{\text{water}}=1.0\times10^{-6}\ \mathrm{m^2/s}$ . A stopped fan's swirl dies in seconds across a room-scale eddy; the same-size eddy in a swimming pool coasts far longer. Whenever the question is "which term wins in the equation," it is $\nu$ — hence $Re=UL/\nu$ — that referees.

Why is Stokes drag linear in velocity when everyday drag — cars, cyclists — is quadratic?

They are opposite ends of the Reynolds axis, with different physics carrying the momentum. At $Re\ll1$ the drag is transmitted by viscous stresses in a linear equation: double $U$ , double every field, double the force — $F=6\pi\mu RU$ . At $Re\gg1$ the body must continuously accelerate the mass of fluid it plows through; momentum flux scales as $\rho U^2$ times frontal area, giving $F=\tfrac12C_D\rho AU^2$ with $C_D$ roughly constant. The crossover for a given object sits near $Re\sim1$ – $10^2$ . A dust grain, a bacterium, and a fog droplet live in the linear world (Example 2); a raindrop, a car, and a cyclist live in the quadratic one. Using the wrong law is among the commonest errors in estimation problems — always compute $Re$ first.

What is the scallop theorem, and why does it rule how microorganisms swim?

The Stokes equations contain no time derivative — inertia is gone — so creeping flow is quasi-static and time-reversible: run any sequence of boundary motions backwards and the fluid retraces itself. Purcell's scallop theorem follows: a swimmer whose stroke is reciprocal (looks identical played forward and backward, like a scallop opening slowly and closing fast) achieves zero net displacement at $Re=0$ — the fast half undoes the slow half exactly, because the equations do not know about "fast." Real microswimmers therefore break time-symmetry geometrically: E. coli rotates a rigid helical flagellum (a chiral, non-reciprocal motion), sperm pass bending waves down their tails, and engineered microrobots use magnetic corkscrews. The theorem is also why micro-mixing is hard — reversibility resists stirring — which microfluidic chips defeat with geometry instead of turbulence.

Is blood a Newtonian fluid? Physiology examples keep using Poiseuille anyway.

Blood is honestly non-Newtonian: it is a suspension of deformable cells (~45% by volume), so its effective viscosity falls with shear rate (red cells align and deform), and in vessels below ~300 μm the cells migrate toward the axis, leaving a plasma-rich lubricating layer that lowers resistance (the Fåhræus–Lindqvist effect). Poiseuille with a constant $\mu\approx3$ – $4\times10^{-3}\ \mathrm{Pa\,s}$ is nonetheless a good working model in large vessels at physiological shear rates, where the shear-thinning has saturated. The clinically robust content is the scaling: resistance $\propto R^{-4}$ survives all the rheological corrections, which is why modest arterial narrowing has outsized hemodynamic cost and why vasodilation is such a powerful control knob.

The Millennium Problem asks about smoothness. Why would a smooth equation produce a singularity at all?

The suspect is vortex stretching. In the vorticity equation $\mathrm{D}\boldsymbol{\omega}/\mathrm{D}t=(\boldsymbol{\omega}\cdot\nabla)\mathbf{v}+\nu\nabla^2\boldsymbol{\omega}$ , the stretching term feeds vorticity growth at a rate proportional to vorticity itself times the velocity gradient — which is itself built from vorticity. That self-amplifying loop could, in principle, drive $|\boldsymbol{\omega}|\to\infty$ in finite time, outrunning viscous smoothing; nobody has proven it does or does not. In 2D the term vanishes identically and global smoothness is a theorem. The physical stakes are real but bounded: a mathematical singularity would signal the model breaking (continuum hypothesis, incompressibility) rather than infinite physical velocities — molecular scales intervene first. Recent computer-assisted work has produced rigorous blow-up results for the inviscid Euler equations in special geometries and numerical singularity candidates for related systems, sharpening where the danger lies.

How can pressure be computed if there is no equation of state in incompressible flow?

Pressure changes job. In compressible flow $p$ is thermodynamic, tied to $\rho$ and $T$ by a state equation. In incompressible flow it is the Lagrange multiplier of the constraint $\nabla\cdot\mathbf{v}=0$ : taking the divergence of the momentum equation gives a Poisson equation, $\nabla^2p=-\rho\,\partial_i v_j\,\partial_j v_i$ , whose solution at each instant is exactly the pressure field needed to keep the velocity divergence-free. Two consequences: pressure responds instantaneously and globally (the infinite-sound-speed idealization), and numerical solvers must solve that global Poisson problem every time step — the pressure-projection step that dominates CFD budgets, as the kinematics article noted.

Everything here assumed constant viscosity. When does that fail, and what breaks first?

Two separate assumptions hide there: that $\mu$ is constant in space, and that stress is linear in strain rate at all. The first fails benignly whenever temperature varies — $\mu$ of liquids drops steeply with heating (engine oil is designed around this), gases the opposite — and the remedy is keeping $\mu$ inside the divergence, $\partial_j(2\mu e_{ij})$ . The second fails structurally for non-Newtonian fluids: polymer solutions and melts carry elastic memory (stress depends on deformation history), suspensions shear-thin or shear-thicken (cornstarch), and yield-stress fluids (toothpaste, drilling mud) refuse to flow below a threshold. Each needs its own constitutive law — Oldroyd-B, power-law, Bingham — slotted into the same Cauchy momentum balance. The balance equation is universal; the constitutive law is where the material's personality lives, exactly as in the elasticity articles.