AN INTERACTIVE MATHEMATICAL JOURNEY

Visual Narrative of Navier-Stokes Paper

An educational visual guide to a proposed finite-time blow-up construction.

LIVE PARTICLE FIELDt → 1 / τ = 1 − t

Inward spiral.
Axial escape.

FOLLOW THE EQUATION

Pressure acceleration points toward lower pressure

001 TRACERSIllustrative flow · not a numerical solution
OUTER slowMIDDLE fasterINNER fastest
0 at rest72-second loopapproaching the limit 1

Drag to orbit · tilt & cutaway follow time · reset loop restores both

01/ 10
ONE BALANCE LAW · TEN CONNECTED STEPS

Follow the equation, then ask what each step makes possible.

Apply the chain rule along a moving point

The dot in the animation samples a field. The field tells it where to go; the chain rule tells us its acceleration. Navier–Stokes specifies which forces can produce that acceleration.

01THE QUESTION

Can smooth motion break down?

Start with a fluid at rest. Apply a smooth force. Can a tiny region become arbitrarily fast while total energy stays bounded?

THE CONNECTING THREAD − / +

Particle acceleration

Apply the chain rule along a moving point

X˙=u(X,t),X¨=tu+(u)u\dot X=u(X,t),\quad\ddot X=\partial_tu+(u\cdot\nabla)u

The dot in the animation samples a field. The field tells it where to go; the chain rule tells us its acceleration. Navier–Stokes specifies which forces can produce that acceleration.

Follow the dot: the chain rule becomes acceleration.

X labels a moving parcel, whereas x labels a fixed place. Differentiate u(X(t),t): the time derivative contributes ∂tu and the changing position contributes (u·∇)u. This is the exact bridge between particle trajectories and the velocity-field equation.

THE MATHEMATICSX˙(t)=u(X(t),t),X¨(t)=Dtu(X(t),t)\dot X(t)=u(X(t),t),\quad \ddot X(t)=D_tu(X(t),t)
EXPLORE IN 3Ddrag to orbit

Three tracers. Three visible time scales.

0%
A, B and C are tracers in three illustrative regions, not different matter. Prescribed paths show angular and axial speed differences; short trails reveal curvature. Heads recycle at each path endpoint. These paths are not a numerical solution.
A SIMPLE EXAMPLE

In steady circular motion the speed can be constant while velocity turns. For a circle of radius r at tangential speed v, acceleration has inward magnitude v²/r.

Go one level deeper +

Curvature is a property of a trajectory; the Laplacian is curvature of the velocity field with respect to spatial coordinates. They are different derivatives. A short particle trail visualizes the first, while the diffusion graph demonstrates the second.

Pressure enforces the volume constraint.

Take divergence of momentum balance. The time derivative and viscous term vanish because ∇·u=0. The nonlinear term remains and determines a Poisson equation for pressure, together with the domain and boundary or decay conditions.

THE MATHEMATICSΔp=i,j(iuj)(jui)f-\Delta p=\sum_{i,j}(\partial_i u_j)(\partial_j u_i)-\nabla\cdot f
EXPLORE THE IDEA

A small balance model

+A−Aremainder8.3 − 8.3 + 1 = 1
Change the two large opposing terms. Their sum remains 1. Values are illustrative.
A SIMPLE EXAMPLE

Pressure is not an extra independently chosen acceleration at each point: its spatial pattern must be compatible with the entire incompressible field.

Go one level deeper +

This identity shows the constraint connecting pressure to velocity. On a periodic domain, pressure is fixed only up to an additive function of time; its gradient is the physically relevant quantity.

Integrate the local law: the energy identity.

Dot the momentum equation with u and integrate. For smooth divergence-free fields with suitable decay, compact support or periodic boundaries, advection and pressure make no net contribution to total energy. Integration by parts turns viscosity into a nonnegative dissipation term.

THE MATHEMATICSddt12u2dx+νu2dx=fudx\frac{d}{dt}\frac12\int|u|^2dx+\nu\int|\nabla u|^2dx=\int f\cdot u\,dx
EXPLORE THE IDEA

Peak speed versus total energy

200.50001ε=0.67 peak=1.5 energy=0.334
Orange: peak speed (left scale 0–20). Green: energy (right scale 0–0.5). Toy model: speed = 1/ε, volume = ε³, energy = ε/2. Not a Navier–Stokes solution.
A SIMPLE EXAMPLE

The local nonlinear term can redistribute energy into a smaller region even when its global integral cancels. A finite integral therefore does not itself impose a finite peak.

Go one level deeper +

The boundary assumptions matter: with other boundaries, energy flux terms must be retained. This identity connects the PDE to the L² estimate; it does not provide an L∞ bound in three dimensions.

One equation. Four competing effects.

The Navier–Stokes equation balances acceleration and force per unit mass. Velocity is both the quantity being transported and the thing doing the transporting. That self-interaction is the difficult part.

THE MATHEMATICStu+(u)uνΔu+p=f\partial_tu+(u\cdot\nabla)u-\nu\Delta u+\nabla p=f
EXPLORE THE IDEA

A small balance model

+A−Aremainder8.3 − 8.3 + 1 = 1
Change the two large opposing terms. Their sum remains 1. Values are illustrative.
A SIMPLE EXAMPLE

In a narrowing river a fixed sensor may see steady flow, while a drifting leaf accelerates into the faster region. Local change and material acceleration differ.

Go one level deeper +

Setting ν=0 gives the Euler equation (1.2). On the acceleration side, the viscous term is +νΔu and pressure acceleration is −∇p. The mini-visual is a teaching balance, not computed forces from the paper.

Narrower does not mean less volume.

Incompressibility preserves a travelling parcel's volume, but not its shape. It may narrow in two directions while stretching in a third. The initial condition says the whole flow starts at rest.

THE MATHEMATICSu=0,u(,0)=0\nabla\cdot u=0,\qquad u(\cdot,0)=0
EXPLORE IN 3Ddrag to orbit

Follow a deforming material parcel

0%
Exact toy flow u=(−x,−y,2z): X(s)=(e⁻ˢx₀,e⁻ˢy₀,e²ˢz₀). The volume is constant while the shape stretches. Each loop resets the parcel.
A SIMPLE EXAMPLE

u=(−x,−y,2z) has divergence −1−1+2=0. It compresses horizontally and stretches vertically. This is a toy field, not the paper's vortex.

Go one level deeper +

The shrinking high-speed core is a region through which fluid passes, not a single fixed material parcel. That distinction matters: an incompressible material parcel cannot simply lose volume.

A bounded whole. An unbounded peak.

The L² norm measures the entire spatial distribution; the L∞ norm measures its peak. An increasingly tiny region can contain increasingly high speeds without an increasing total kinetic energy.

THE MATHEMATICSsupt<1u(t)2<,lim supt1u(t)=\sup_{t<1}\lVert u(t)\rVert_2<\infty,\quad\limsup_{t\uparrow1}\lVert u(t)\rVert_\infty=\infty
EXPLORE THE IDEA

Peak speed versus total energy

200.50001ε=0.67 peak=1.5 energy=0.334
Orange: peak speed (left scale 0–20). Green: energy (right scale 0–0.5). Toy model: speed = 1/ε, volume = ε³, energy = ε/2. Not a Navier–Stokes solution.
A SIMPLE EXAMPLE

Speed 1/ε in a region of volume ε³ gives squared-integral ε. The peak tends to infinity while this integral tends to zero.

Go one level deeper +

The example illustrates concentration, not an exact divergence-free solution. The theorem's limsup does not imply monotone acceleration or blowup at every fixed location. A real fluid does not attain infinite speed.

The fine print is part of the mathematics.

The force must be smooth and compactly supported in space and positive time. The expression describes its input, output and regularity all at once.

THE MATHEMATICSfCc(R3×(0,);R3)f\in C_c^\infty(\mathbb R^3\times(0,\infty);\mathbb R^3)
EXPLORE THE IDEA

A smooth window

01support |x| < 0.48
The bump exp(1−1/(1−x²)) for |x|<1 is smoothly zero outside. Adjust its support width.
A SIMPLE EXAMPLE

Read the semicolon as 'taking values in', the c as 'compact support', and ∈ as 'belongs to'. The three real output components make f a vector.

Go one level deeper +

The historical section distinguishes smooth classical solutions from weak solutions, and introduces mixed space-time norms. This website explains the paper's construction; it does not independently certify its proof or adjudicate prize acceptance.

02THE PHYSICAL PICTURE

A vortex with nowhere to hide.

Radial inflow, rotation and axial outflow combine. The core becomes more slender as characteristic speeds increase.

THE CONNECTING THREAD − / +

One vector equation

Resolve it into radial, angular and axial directions

u=1rr(rur)+zuz=0\nabla\cdot u=\frac1r\partial_r(ru_r)+\partial_z u_z=0

For an axisymmetric flow, radial convergence must be balanced by axial extension. The spirals show components of one velocity field, not three different fluids. Changing coordinates changes the description, not the law.

Three directions, one flow.

Red/orange tracers follow fast inner spirals, blue the middle, green the slower outer regions. Fluid exits axially above and below a dividing layer. The actual construction has a slight axial asymmetry.

THE MATHEMATICSx=(rcosθ,rsinθ,z),ur, uθ, uzx=(r\cos\theta,r\sin\theta,z),\quad u_r,\ u_\theta,\ u_z
EXPLORE IN 3Ddrag to orbit

Three tracers. Three visible time scales.

0%
A, B and C are tracers in three illustrative regions, not different matter. Prescribed paths show angular and axial speed differences; short trails reveal curvature. Heads recycle at each path endpoint. These paths are not a numerical solution.
A SIMPLE EXAMPLE

With no torque, halving a parcel's radius doubles its tangential speed. That analogy is incomplete here: viscosity also transports angular momentum outward.

Go one level deeper +

The animation retains the approved nine trajectory families and 972-particle balance. It is not a PDE solver. In a smooth axisymmetric field, tangential speed vanishes on the exact axis; fastest swirl is near it, not literally at r=0.

Both lengths shrink. One shrinks faster.

Self-similarity keeps the profile fixed in rescaled coordinates. The radial scale shrinks faster than the axial scale. 'Elongation' here means increasing slenderness, not an absolute height growing without limit.

THE MATHEMATICSrτ1/2,zτ1/2h,τ=1t\ell_r\asymp\tau^{1/2},\quad\ell_z\asymp\tau^{1/2-h},\quad\tau=1-t
EXPLORE THE IDEA

The exact scaling exponents

30−306τ = 10⁻2
Log₁₀ scale factors versus −log₁₀τ, with h=0.005. Orange: speed; blue: height; green: radius.
A SIMPLE EXAMPLE

The core energy scale is U²V: τ^(−1−2h) × τ^(3/2−h) = τ^(1/2−3h), tending to zero despite rising speed.

Go one level deeper +

This is core energy, not a statement that the entire flow's energy vanishes. The particle display exaggerates shape changes for visibility; the mini-graph uses the stated powers with h=0.005.

Viscosity is still in the contest.

One Reynolds number grows while another stays bounded. Radial diffusion remains in the leading balance. The construction does not simply discard viscosity.

THE MATHEMATICSReθ=uθrντh,Rer=urrν=O(1)Re_\theta=\frac{|u_\theta|\ell_r}{\nu}\asymp\tau^{-h},\quad Re_r=\frac{|u_r|\ell_r}{\nu}=O(1)
EXPLORE THE IDEA

Why fine ripples fade faster

0t = 1.05
The scalar heat mode exp(−0.15 k²t) sin(kx). k=1 in green; k=3 in blue.
A SIMPLE EXAMPLE

Doubling the wavenumber k of a sine ripple multiplies its viscous damping rate by four.

Go one level deeper +

Background shear can amplify pulses initially, then shorten their wavelengths until damping overtakes growth. The mini-graph uses a scalar heat-equation mode, not the full cylindrical vector exterior.

03THE PROOF STRATEGY

Make the imbalance disappear.

Inventing a fast vortex is easy. Keeping its residual force smooth is the challenge.

THE CONNECTING THREAD − / +

A candidate vortex

Substitute it into the original equation

R(U,P):=tU+(U)UνΔU+PR(U,P):=\partial_tU+(U\cdot\nabla)U-\nu\Delta U+\nabla P

R is an audit of the candidate: the force this proposed motion would need. An unbounded R is not an admissible smooth force. This is why a plausible vortex is only the start of the construction.

The force is what remains.

Insert candidate velocity and pressure into the equation. What remains is the momentum residual. It may equal a smooth nonzero force; the construction removes its singular part, including in every derivative.

THE MATHEMATICSR(u,p)=tu+(u)uΔu+pR(u,p)=\partial_tu+(u\cdot\nabla)u-\Delta u+\nabla p
EXPLORE THE IDEA

An improving decay bound

015cycle 5 bound = 3.98e-2
A toy bound q^(0.2+j/10), q=0.01. Each cycle changes the exponent, not a fixed percentage.
A SIMPLE EXAMPLE

The toy identity 1/τ−1/τ+1=1 shows how individually large terms can leave a bounded result. A PDE demands far more than this arithmetic cancellation.

Go one level deeper +

The proof builds a background, identifies annular stress, realizes it with waves, corrects smaller errors, sums, and localizes. These are logical dependencies, not physical events in time.

04THE LEADING FLOW

Change the coordinates. Keep the shape.

Rescaled variables reveal a fixed profile underneath concentrating motion.

THE CONNECTING THREAD − / +

A shrinking physical core

Factor out lengths and velocity scales

U(x,t)=A(t)V(y,t),yi=xi/i(t)U(x,t)=A(t)V(y,t),\quad y_i=x_i/\ell_i(t)

At fixed x, ∂tU=A′V+A∂tV−AΣ(ℓ′i/ℓi)yi∂yiV. These extra terms record the moving coordinate grid. The profile equations are the same momentum balance viewed through a changing magnifying glass; the paper uses component-dependent scales.

Coordinates that travel with the scales.

Profile coordinates resolve unequal radial and axial concentration. The chain rule converts profile derivatives back into physical space-time derivatives.

THE MATHEMATICSz=qDη,τ=q(1η2),X=r22qz=q^D\eta,\quad\tau=q(1-\eta^2),\quad X=\frac{r^2}{2q}
EXPLORE THE IDEA

Physical view / rescaled view

01physical width = 0.57
A narrowing Gaussian compared with its fixed normalized shape. An illustrative profile, not the paper's E.
A SIMPLE EXAMPLE

At z=0, η=0 and q=τ. Fixed X=1 means physical radius r=√(2τ): fixed in one view, shrinking in the other.

Go one level deeper +

Equation (4.3) writes uθ⁽⁰⁾=q⁻ᴬE, uz⁽⁰⁾=q⁻ᴬU, rur⁽⁰⁾=V₀, p⁽⁰⁾=q⁻²ᴬΠ. Equation (4.7) fixes V₀ through incompressibility and gives ΠX=E²/(2X). The source determines the allowed profiles.

Two wave families, one target.

Two independent wave families carry different momentum-flux ratios. The desired stress must lie inside the positive cone that their directions generate.

THE MATHEMATICST=c1v1+c2v2,c1,c2>0T=c_1v_1+c_2v_2,\qquad c_1,c_2>0
EXPLORE THE IDEA

Positive combinations

v₁v₂Tc₁=0.73 c₂=1.27
A target T=c₁v₁+c₂v₂ remains between its two generators while c₁,c₂ are positive.
A SIMPLE EXAMPLE

Change the coefficient below: the resulting vector stays between the generating directions when both coefficients are positive.

Go one level deeper +

The paper's exact cone inequalities involve profile and shear parameters. This vector diagram illustrates positive representation, not the full numerical admissibility region in (4.22).

05HIGHER-ORDER CONSTRUCTION

Correct without breaking the balance.

Corrections must improve accuracy while preserving incompressibility, support and the exterior.

THE CONNECTING THREAD − / +

A leading profile with a defect

Linearize the residual, retaining the quadratic remainder

R(U+w,P+q)=R(U,P)+LUw+q+(w)wR(U+w,P+q)=R(U,P)+L_Uw+\nabla q+(w\cdot\nabla)w

Here LUw=∂tw+(U·∇)w+(w·∇)U−νΔw. A correction changes the old error and creates new interactions. The construction must estimate both, at every derivative order it needs.

Writing an infinite series is not enough.

An asymptotic expansion arranges increasingly small orders. The dots are not a convergence proof: coefficients, derivatives and support must all be controlled.

THE MATHEMATICSu(0)+εu(1)+ε2u(2)+u^{(0)}+\varepsilon u^{(1)}+\varepsilon^2u^{(2)}+\cdots
EXPLORE THE IDEA

How the partial sums approach

1104 terms: 1.62400 limit: 1.66667
Geometric series Σ εⁿ, ε=0.4. Unlike an arbitrary asymptotic series, this example converges.
A SIMPLE EXAMPLE

The geometric series converges for |ε|<1. The formal series Σn!εⁿ need not converge for any nonzero ε.

Go one level deeper +

Sections 5.1–5.5 solve coefficients near the axis, repair radial integrals, prove finite-order estimates and sum the fields with shrinking cutoffs. The paper applies cutoffs to potentials before differentiating.

Build incompressibility in.

A vector potential constructs velocity through its curl. For smooth fields, the result automatically has zero divergence.

THE MATHEMATICS(×A)=0\nabla\cdot(\nabla\times A)=0
EXPLORE IN 3Ddrag to orbit

Follow a deforming material parcel

0%
Exact toy flow u=(−x,−y,2z): X(s)=(e⁻ˢx₀,e⁻ˢy₀,e²ˢz₀). The volume is constant while the shape stretches. Each loop resets the parcel.
A SIMPLE EXAMPLE

A=(0,0,xy) gives curl A=(x,−y,0), whose divergence is zero.

Go one level deeper +

The paper's complete representation also retains an axisymmetric azimuthal component. This identity teaches one construction tool rather than the full decomposition.

06AUXILIARY GEOMETRY

Give the waves their own ledger.

Periodic bookkeeping coordinates organize pulses and control unwanted interactions.

THE CONNECTING THREAD − / +

Several oscillatory corrections

Use auxiliary periodic phases to organize them

w=0⇏ww=0\langle w\rangle=0\quad\not\Rightarrow\quad\langle w\otimes w\rangle=0

A periodic phase is a bookkeeping coordinate. It is distinct from making physical space periodic. Even a wave with zero mean can leave a nonzero quadratic mean, and that is precisely the useful part.

An extra space, not extra physics.

The auxiliary torus records fast phases. Separated supports in this space help organize packets; physical fields are recovered through a phase map.

THE MATHEMATICSYT2,YY+n, nZ2Y\in\mathbb T^2,\quad Y\sim Y+n,\ n\in\mathbb Z^2
EXPLORE IN 3Ddrag to orbit

A periodic cube, seen in three dimensions

0%
This cube represents T³ by identifying opposite faces. It is not a doughnut surface: a doughnut depicts T², which has two periodic coordinates. Tracers wrap without colliding with a wall.
A SIMPLE EXAMPLE

The coordinate 1.2 wraps to 0.2 in a unit periodic cell. This identifies opposite boundaries; it is not a new physical dimension.

Go one level deeper +

The section constructs dyadic charts, slow cutoffs, disjoint auxiliary rectangles and compatible common tori for overlapping bands, then establishes coefficient bounds after physical phase evaluation.

07OSCILLATORY MOMENTUM TRANSFER

Zero average. Nonzero effect.

A fluctuation can average to zero while its nonlinear momentum transport remains.

THE CONNECTING THREAD − / +

Zero-mean waves

Average the nonlinear momentum flux

(w)w=ww\langle(w\cdot\nabla)w\rangle=\nabla\cdot\langle w\otimes w\rangle

This identity assumes divergence-free w and an average that commutes with spatial derivatives. The tensor records correlated velocity components. Its divergence contributes to mean momentum balance, allowing waves to address the background defect.

The product changes everything.

Outward motion carrying a positive tangential surplus has the same product sign as inward motion carrying a deficit. First-order oscillations cancel; their quadratic products need not.

THE MATHEMATICSsinθ=0,sin2θ=12\langle\sin\theta\rangle=0,\qquad\langle\sin^2\theta\rangle=\frac12
EXPLORE THE IDEA

Zero-mean waves can transport momentum

0mean product = 0.227
Green: sin θ. Blue: sin(θ+φ). Orange: their product. Dashed: its mean, cos φ / 2.
A SIMPLE EXAMPLE

The sine wave crosses zero. Its square stays nonnegative, with mean one half. Adjust the phase to see how correlation changes the average product.

Go one level deeper +

Section 7 constructs wave phases and an orthogonal frame, solves pulse evolution, computes covariance and its linearization, and takes exact curls while retaining tails. Random tracer jitter is a design choice, not these carefully constructed pulses.

Amplification, then damping.

Shear initially amplifies a pulse. As its wavevector changes, its radial wavelength shortens and viscous damping grows. Eventually damping wins.

THE MATHEMATICSeνk2tsin(kx)e^{-\nu k^2t}\sin(kx)
EXPLORE THE IDEA

Why fine ripples fade faster

0t = 1.05
The scalar heat mode exp(−0.15 k²t) sin(kx). k=1 in green; k=3 in blue.
A SIMPLE EXAMPLE

A mode with twice the frequency loses amplitude four times faster in this scalar diffusion model.

Go one level deeper +

The scalar curve below demonstrates damping only. The full pulse equation also contains background-driven amplification, changing orientation, pressure and support controls.

08MEANS AND COMPATIBILITY

Repair what the waves leave behind.

Angular means and integral constraints require separate corrections.

THE CONNECTING THREAD − / +

A desired local correction

Integrate its differential equation across the support

r(reϕ)=reg    regdr=0\partial_r(r^e\phi)=r^eg\;\Rightarrow\;\int r^eg\,dr=0

If rᵉφ vanishes at both boundaries, the source must have zero weighted integral. Moments are therefore solvability conditions, not decorative extra equations. A local differential repair has a global obligation.

A local repair has integral obligations.

Multiply the first equation by rᵉ: the left becomes a radial derivative. If the primitive is compactly supported, integrating across the support forces the second condition.

THE MATHEMATICS(r+e/r)ϕ=g,reg(r)dr=0(\partial_r+e/r)\phi=g,\qquad\int r^e g(r)\,dr=0
EXPLORE THE IDEA

Zero area is a compatibility condition

+10.76signed area = 0.24
Two signed triangular bumps. Adjust the negative contribution until the signed areas cancel.
A SIMPLE EXAMPLE

A derivative of a compactly supported function integrates to zero. An everywhere-positive source cannot be such a derivative.

Go one level deeper +

The section reconstructs pressure and mean velocity, inverts zero-mean auxiliary-time derivatives, and uses five correction directions: three support-related defects are repaired while two moment constraints are preserved.

09ITERATION AND FLATNESS

Smaller errors, at every scale.

Every correction creates new errors. A complete cycle must still deliver a net gain.

THE CONNECTING THREAD − / +

A smaller residual

Repeat with derivative estimates and controlled summation

αRCα,NqN(every fixed α,N)|\partial^\alpha R|\le C_{\alpha,N}q^N\quad(\text{every fixed }\alpha,N)

Small amplitude alone does not guarantee smoothness: rapid oscillations can make derivatives large. Flatness means every fixed derivative decays faster than every fixed power, with constants allowed to depend on the orders.

A gain in the exponent—not ten percent.

Each full cycle improves the residual decay order. Raising an exponent of a small positive parameter makes its power smaller.

THE MATHEMATICSσj+1=σj+110\sigma_{j+1}=\sigma_j+\frac1{10}
EXPLORE THE IDEA

An improving decay bound

015cycle 5 bound = 3.98e-2
A toy bound q^(0.2+j/10), q=0.01. Each cycle changes the exponent, not a fixed percentage.
A SIMPLE EXAMPLE

For q=0.1, q²=0.01 and q³=0.001. Coefficient growth still needs control; powers alone are not a proof.

Go one level deeper +

The cycle handles angular harmonics, covariance changes, zero-auxiliary-mean fields and moment defects. A common domain is needed before the limiting summation.

Flatter than any power.

Every fixed derivative of the residual decays faster than any specified power. This flatness supports smooth extension of forcing through the singular time.

THE MATHEMATICSxαtbRCα,b,NqNfor every N|\partial_x^\alpha\partial_t^bR|\le C_{\alpha,b,N}q^N\quad\text{for every }N
EXPLORE THE IDEA

Flat is not identically zero

01q=0.38 flat function=1.08e-3
Orange: q². Green: exp(−1/q²). Both vanish at zero; only the green function is flat there.
A SIMPLE EXAMPLE

exp(−1/q²), extended by zero at q=0, has every derivative zero there, yet is positive for q>0.

Go one level deeper +

Constants may depend on derivative order and N. One bound shared by all orders is not claimed. The exact source also specifies domains and profile ranges.

10LOCALIZATION AND THE RESULT

Keep the core. Close the world.

The local construction becomes a compactly forced whole-space flow starting from rest.

THE CONNECTING THREAD − / +

A local field with controlled error

Localize while preserving the constraint

curl(cA)=0\nabla\cdot\operatorname{curl}(cA)=0

Multiplying velocity directly by c would add ∇c·u to its divergence. Cutting off a vector potential instead preserves the constraint identically. The extra curl term is accounted for in the final forcing.

Cut off the potential first.

A spatial window retains the core and switches off the exterior smoothly. A temporal window creates zero initial velocity. Taking curl after localization preserves incompressibility.

THE MATHEMATICSu=curl(cA)+cBeθ,p=cplocu=\operatorname{curl}(cA)+cB e_\theta,\qquad p=c\,p^{\rm loc}
EXPLORE THE IDEA

A smooth window

01support |x| < 0.48
The bump exp(1−1/(1−x²)) for |x|<1 is smoothly zero outside. Adjust its support width.
A SIMPLE EXAMPLE

curl(cA)=c curl A+∇c×A. The extra term is the price of localization; it cannot be discarded.

Go one level deeper +

The authors establish terminal force-derivative limits, construct a compact time extension, prove bounded energy and compare against hypothetical global smooth finite-energy solutions. Appropriate scaling handles any positive viscosity.

The drawing is not the proof.

The paper presents a forced finite-time blowup construction. This website teaches its logical structure and vocabulary. A convincing particle vortex cannot establish mathematical validity.

THE MATHEMATICSuν(x,t)=νu(x/ν,t)u_\nu(x,t)=\sqrt{\nu}\,u(x/\sqrt{\nu},t)
EXPLORE IN 3Ddrag to orbit

A periodic cube, seen in three dimensions

0%
This cube represents T³ by identifying opposite faces. It is not a doughnut surface: a doughnut depicts T², which has two periodic coordinates. Tracers wrap without colliding with a wall.
A SIMPLE EXAMPLE

A localized construction also yields a periodic version after selecting a suitable periodic box and scaling.

Go one level deeper +

This is a section-wide guided overview, not a line-by-line verification of every lemma. Use the source links for exact hypotheses, notation and proofs. The appendix chapters explain how supporting constraints are closed.

RETURN TO THE SAME EQUATION

Every correction answers the same question.

Does the proposed velocity leave a smooth force in the momentum balance? Geometry, scaling, waves, moments and flatness each remove an obstacle to that requirement. The animation is an illustration; the source contains the proof and its hypotheses.

Return to the complete paper (opens in a new tab)
THE QUESTION