Boussinesq Theorem: Smooth Evolution, Terminal Regularity, and Derivative Blowup

Good to see you again. Last time, taking curl of the forced Boussinesq momentum equation gave us

$$D_t\omega ={\partial }_1\theta +\mathrm{curl}{f}_u.$$

Thus a horizontal temperature gradient produces vorticity, while external velocity forcing affects vorticity only through its rotational part. That equation makes it plausible that increasingly fine temperature structure could drive increasingly strong vorticity. But a blowup theorem makes a more delicate claim than “some quantity gets large.”

This lesson separates three statements that must be held apart when reading the Boussinesq construction: the solution is smooth at every preterminal time; the externally supplied forcing remains smooth even at the terminal time; nevertheless, selected norms of derivatives of the solution become unbounded as the terminal time is approached.

---

One theorem, three logically different assertions

Keep the forced system in view:

$${\partial }_t\theta +u\cdot \nabla \theta =f_{\theta },$$ $${\partial }_{t}u+u\cdot \nabla u+\nabla p=\theta e_2+f_u,$$ $$\mathrm{div}u=0.$$

A forced blowup result can be read schematically as saying that there exist:

• smooth initial data $(u_0,{\theta }_0)$,
• a finite time $T>0$,
• carefully constructed forcing terms $(f_{\theta },f_u)$, and
• a corresponding solution $(u,\theta ,p)$,

for which all of the following hold:

$$(u,\theta ,p)\in C^{\infty }([0,T)\times \Omega ),$$ $$(f_{\theta },f_u)\in C^{\infty }([0,T]\times \Omega ),$$

but some derivative-based solution norm becomes unbounded as $t\uparrow T$. In the present setting, the relevant quantities are naturally of the form

$$\Vert \nabla \theta (\cdot ,t){\Vert }_{L^{\infty }},\Vert \omega (\cdot ,t){\Vert }_{L^{\infty }},\Vert \nabla u(\cdot ,t){\Vert }_{L^{\infty }},$$

or higher-order analogues.

The exact theorem specifies its function spaces, domain, and particular blowup quantity. For conceptual reading, the decisive logical structure is:

Assertion	What it says	What it does not say
Smooth before $T$	At every fixed $t<T$, all required derivatives exist and are finite.	That there is one uniform bound valid for all $t<T$.
Forcing smooth through $T$	The source terms extend regularly to $t=T$, with no singularity inserted there.	That the force is zero, physically generic, or independent of the construction.
Blowup at $T$	A solution derivative norm becomes unbounded as $t\uparrow T$, preventing smooth continuation through $T$.	Necessarily that the fields $u$ or $\theta$ themselves diverge in amplitude.

The apparent tension between the first and third rows is resolved by a basic quantifier distinction:

$$\forall t<T,\Vert \nabla \theta (\cdot ,t){\Vert }_{L^{\infty }}<\infty$$

does not imply

$$\underset{0\le t<T}{\sup }\Vert \nabla \theta (\cdot ,t){\Vert }_{L^{\infty }}<\infty .$$

A familiar scalar analogy is

$$g(t)=\frac{1}{T-t}.$$

For every particular $t<T$, $g(t)$ is finite. Yet its values have no finite bound over the entire half-open interval $0\le t<T$. A PDE singularity theorem has this same logical shape, except that the divergent object is a norm of a solution derivative, not necessarily a scalar value.

---

Smooth preterminal evolution is genuine smoothness

Saying that the solution is smooth on $[0,T)$ is not a euphemism for “approximately regular.” At any selected time $t_{\ast }=T-\epsilon$, with $\epsilon >0$, it is an ordinary smooth solution. One may differentiate the equations, evaluate its vorticity, solve for pressure, and use the usual local well-posedness framework.

What fails is uniform regularity as the endpoint is approached. For example, one could have a sequence of times $t_n<T$ tending to $T$ for which

$$\Vert \nabla \theta (\cdot ,t_n){\Vert }_{L^{\infty }}\to \infty .$$

At each $t_n$, the gradient is still finite. The issue is that no single finite constant controls those gradients all the way to $T$.

This distinction matters particularly for the high-frequency mechanism that this module will develop. A mode can retain a modest field amplitude while its wavelength shrinks. If a temperature contribution has schematic form

$${\theta }_k(x,t)=a_k(t)\cos (k\varphi (x,t)),$$

then its size is controlled roughly by $|a_k(t)|$, whereas a spatial derivative has scale roughly

$$|\nabla {\theta }_k|\sim k|a_k(t)||\nabla \varphi |.$$

Thus, amplitude and derivative size are separate currencies. A sequence of increasingly fine oscillations can leave $\theta$ bounded while making $\nabla \theta$ large.

That derivative growth is dynamically relevant, not merely a choice of norm: from the vorticity equation,

$$D_t\omega ={\partial }_1\theta +\mathrm{curl}{f}_u,$$

a large horizontal derivative of temperature is exactly a large vorticity source.

---

Smooth forcing at the terminal time is the striking part

The phrase “with smooth forcing” deserves precision. It does not merely say that the force is smooth at every time before $T$. It says that the forcing has a regular extension to the closed time interval, schematically

$$(f_{\theta },f_u)\in C^{\infty }([0,T]\times \Omega ).$$

On a compact or periodic spatial domain, this means that every fixed mixed derivative has a finite bound up to time $T$:

$$\underset{0\le t\le T}{\sup }{\Vert {\partial }_t^j{\partial }_x^{\beta }{f}_{\theta }(\cdot ,t)\Vert }_{L^{\infty }}<\infty ,$$ $$\underset{0\le t\le T}{\sup }{\Vert {\partial }_t^j{\partial }_x^{\beta }{f}_u(\cdot ,t)\Vert }_{L^{\infty }}<\infty$$

for every fixed nonnegative integer $j$ and multi-index $\beta$. On a noncompact domain, one states the corresponding conclusion in the theorem’s specified global or local smooth topology.

So the singularity is not caused by placing a term such as

$$\frac{1}{T-t}$$

directly into the forcing. Nor is it simply an initial-data singularity carried along by the flow: the initial data are regular as well. Rather, the smooth forcing sustains a nonlinear evolution that transfers activity to finer and finer scales, until the solution itself loses derivative regularity at the finite terminal time.

There is a methodological subtlety here. The force is generally engineered, along with the solution, to make the construction work. “Smooth forcing” does not mean “an arbitrary physical forcing will create a singularity.” It means that there exists a force with no singular behavior of its own for which the equation nevertheless produces one. This is already a mathematically strong conclusion.

Terry Tao’s overview gives the right high-level lens: successive high-frequency corrections make the solution increasingly singular, while their contribution to the forcing is controlled far more strongly.

---

What “blowup” says, and what it leaves open

In this context, finite-time blowup means that the solution cannot be continued through $T$ in the asserted regularity class. A typical quantitative form is

$$\underset{t\uparrow T}{\mathrm{limsup}}\mathcal{N}(u,\theta )(t)=\infty ,$$

where $\mathcal{N}$ contains one or more derivative norms.

For Boussinesq, $\Vert \nabla \theta {\Vert }_{L^{\infty }}$ is especially natural because it directly drives vorticity. The velocity is recovered nonlocally from vorticity through the incompressibility constraint and Biot–Savart law, so growth of $\omega$ also threatens control of $\nabla u$.

The blowup statement should not automatically be inflated into stronger claims:

1. It need not be amplitude blowup.
   It may remain possible that $\Vert u(t){\Vert }_{L^{\infty }}$ and $\Vert \theta (t){\Vert }_{L^{\infty }}$ stay bounded while gradients become unbounded. Fine-scale oscillation permits exactly this separation.

2. It need not mean every derivative diverges at the same rate.
   A theorem identifies the regularity threshold it proves is lost. Higher derivatives may of course become problematic too, but the proof’s stated norm is what one should quote.

3. It need not establish an unforced singularity.
   A forced construction answers a different question from the classical open problem for unforced 3D Euler. Its force is smooth, which makes the result highly nontrivial, but it remains a force.

4. It need not give a classical solution at $t=T$.
   There may be weak, distributional, or otherwise generalized objects beyond $T$, but that is not what a smooth blowup theorem asserts. The immediate conclusion is the failure of continuation in the indicated smooth class.

A useful verbal test is:

Every frame before the deadline is smooth; no uniform derivative bound survives up to the deadline; and the external input itself remains regular at the deadline.

That is the theorem’s central surprise.

---

A comparison with an unforced Boussinesq blowup result

The distinction between preterminal regularity and terminal failure also appears in earlier Boussinesq singularity results, even where the theorem is not a “smooth forcing” theorem. The Chen–Hou paper provides a useful contrast: its Boussinesq theorem concerns singularity formation from regular initial data in a boundary setting, and its final argument invokes a BKM-type continuation criterion involving the time integral of $\Vert \nabla \theta {\Vert }_{L^{\infty }}$.

The BKM-type perspective is worth retaining. A continuation criterion has the general form:

$${\int }_0^T\Vert \nabla \theta (\cdot ,t){\Vert }_{L^{\infty }}dt<\infty ⟹\text{smooth continuation past }T.$$

Therefore, proving that the controlling quantity cannot remain integrable up to $T$ rules out smooth continuation. In the cited unforced paper, the authors then obtain blowup of physical norms through the rescaling analysis. For the smooth-forcing construction, the same broad distinction applies: the force stays regular, but the solution reaches a derivative-level obstruction to continuation.

---

Reading theorem statements without conflating their clauses

When you meet the theorem statement underlying Tao’s post, parse it in the following order.

1. Identify the equation and the force

Ask whether the result concerns:

$$f_{\theta }=0,f_u=0,$$

or whether it constructs nonzero $(f_{\theta },f_u)$. In the latter case, immediately look for the topology in which the forcing is said to be smooth. “Smooth” is a claim about all finite derivative orders, not merely continuity or bounded amplitude.

2. Locate the interval of existence

Check whether the solution is stated on:

$$[0,T),$$

rather than on $[0,T]$. The excluded endpoint is meaningful. It signals that the theorem gives classical evolution arbitrarily close to $T$, but not a smooth extension at $T$.

3. Find the norm that fails

Do not replace “a $C^1$ norm blows up” with the vaguer claim “the velocity blows up.” Determine whether the theorem proves divergence of:

$$\Vert \omega {\Vert }_{L^{\infty }},\Vert \nabla \theta {\Vert }_{L^{\infty }},\Vert \nabla u{\Vert }_{L^{\infty }},$$

or a higher Sobolev or Hölder norm. These express materially different singularity scenarios.

4. Separate existence from genericity

An existence theorem says there is at least one designed configuration with the stated behavior. It is not yet a claim that a random perturbation, a broad open set of data, or physically typical forcing produces blowup. The stability or genericity question is separate.

---

Why the construction can thread this needle

At a high level, the forcing and the solution obey different summability requirements.

Imagine adding layers indexed by $n$, each with a very large frequency $k_n$. A solution component may have small amplitude $a_n$, but its gradient has scale $k_n{a}_n$. If the frequency growth wins over amplitude decay, the cumulative derivative field can become very large near $T$.

At the same time, the construction is arranged so that the residual error each layer creates in the PDE is much smaller than its visible effect on the targeted solution derivative. After appropriate corrections, the forcing contributions can converge in every fixed $C^m$ norm.

The key point is not a contradiction of differentiation. Indeed, differentiating a high-frequency term is expensive. The construction must pay for every derivative of the force. Its achievement is to make the error amplitude decay rapidly enough, relative to frequency, that the force is smooth even though the intended solution develops increasingly violent small scales.

We will later examine the estimates that make this possible. For now, avoid the misleading mental model that the force “blows up more slowly than the solution.” The proper statement is stronger:

$$\text{the forcing remains smooth at }T,$$

whereas

$$\text{the solution loses the specified derivative regularity at }T.$$

---

Takeaways and next lesson

A finite-time Boussinesq blowup theorem with smooth forcing contains three compatible facts:

• For every fixed $t<T$, the solution is a genuine smooth solution.
• The constructed forcing remains smooth through $t=T$; it does not carry a hidden singularity at the terminal time.
• Nevertheless, a derivative-level norm of the solution becomes unbounded as $t\uparrow T$, so the solution cannot be continued smoothly past $T$.

The crucial logical distinction is between pointwise-in-time finiteness before $T$ and a uniform bound all the way up to $T$. The mechanism we will study exploits the fact that bounded-amplitude fields can acquire enormous derivatives when their frequency rises.

Next, we turn that observation into a calculation: for a single temperature–vorticity plane wave, we will derive the Fourier-mode Biot–Savart velocity and track exactly how velocity and temperature-gradient size scale with frequency.