Weak Solutions and Distributional Formulation of PDEs

The distributional reformulation is the quiet engine under an enormous amount of modern computation and open theory. Every finite element solver — deal.II, FEniCS, the codes that certify aircraft, chips, and MRI magnets — discretizes the weak form (6), never the strong one, precisely because it asks for only first derivatives and so tolerates the simple continuous "tent" functions that make the method cheap and robust. Shock-capturing schemes in computational fluid dynamics (Godunov, ENO/WENO) are engineered to converge to the entropy weak solution and nothing else, which is why they resolve sonic booms and supernova blast waves without spurious oscillations. And the weak formulation is where the deepest open problem in classical physics lives: Jean Leray's 1934 weak solutions of the incompressible Navier–Stokes equations are known to exist globally in three dimensions, but whether they are unique and smooth — whether turbulence can spontaneously blow up — remains unknown, one of the Clay Institute's million-dollar Millennium Prize Problems. And in quantum field theory a field is an operator-valued distribution, smeared against test functions before it means anything, with renormalization the controlled extension of singular distributions to coincident points.

You never read a distribution's value at a point — you interrogate it, integrating it against a smooth bump $\varphi$ and recording the single number that comes back. A function is known not by its graph but by the answers it gives to all probes, and its derivative is defined by differentiating the probe instead, absorbing the minus sign integration by parts hands you. The payoff is startling: a step function, with no classical derivative at its jump, has a perfectly good distributional one — a unit-weight spike sitting exactly at the jump, the Dirac delta. The first figure is that statement in two panels.

Sobolev spaces are where energy lives. Demanding $\int|\nabla u|^2<\infty$ is the statement that the field has finite Dirichlet energy, and minimizing that energy is precisely solving Laplace's equation weakly; the completeness of $H^1$ guarantees the minimizer exists, turning the physicist's "extremize the action" into a theorem. The same integration-by-parts move that made a step differentiable makes a discontinuity legible to the equation: integrate a conservation law across a moving jump and the boundary term you collect is the Rankine–Hugoniot condition, fixing the shock's speed from the states on either side. The weak form is the conservation law; the differential form is only its smooth-region shadow. The second figure shows this for the inviscid Burgers equation, where fast characteristics overtake slow ones and a shock forms with a speed the weak form pins to a number.

A distribution is not a function but a rule that eats a smooth probe and returns a number, linearly and continuously. This single reinterpretation is the whole subject; everything below either defines the pairing or computes with it.

where: $T$ — a distribution, i.e. a continuous linear functional; $\varphi$ — a test function, smooth and compactly supported inside the open set $\Omega$; $\langle T,\varphi\rangle$ — the action (pairing) of $T$ on $\varphi$; $\mathcal{D}(\Omega)$ — the space of all such test functions. Compact support means $\varphi$ never touches $\partial\Omega$, so every boundary term from integration by parts dies. Every ordinary function you would ever call a field already is a distribution, obtained by pairing through an integral. This is the bridge that lets distributions extend, rather than replace, classical calculus.

where: $f$ — a locally integrable function, $f\in L^1_{\mathrm{loc}}(\Omega)$; $T_f$ — the distribution it induces. Because $\varphi$ has compact support, the integral runs only over a bounded set where $f$ is integrable: mere local integrability is enough to be a distribution. Now the master move. The derivative of a distribution is defined so that integration by parts holds by fiat, the minus sign carrying the derivative from $T$ over to the smooth $\varphi$.

where: $\partial_i T$ — the $i$-th weak derivative of $T$, which always exists; $\partial_i\varphi$ — the ordinary derivative of the smooth probe. Since $\varphi$ can be differentiated forever, so can $T$: every distribution is infinitely differentiable — a clean gain over classical calculus, where differentiation destroys smoothness. Apply that definition to the humble step and the Dirac delta drops out, the prototype of a singular distribution that is no function at all.

where: $H$ — the Heaviside step, $H=1$ for $x>0$ and $0$ otherwise; $\delta$ — the Dirac delta, the functional that reads a probe's value at the origin. No locally integrable $f$ satisfies $\int f\varphi=\varphi(0)$ for all $\varphi$, so $\delta$ lives only in the enlarged world $\mathcal{D}'$. To do analysis — existence, uniqueness, stability — we need a complete normed space graded by how many weak derivatives stay square-integrable. That is the Sobolev space.

where: $H^k=W^{k,2}$ — functions with weak derivatives through order $k$ in $L^2$; $\alpha$ — a multi-index of order $|\alpha|$; $D^\alpha u$ — the weak derivative of order $\alpha$. The norm is literally an energy, and the completeness of $H^1$ is what makes minimizers exist. Pairing $-\nabla^2u=f$ against a test function and integrating by parts once produces the symmetric weak form of Poisson's equation, requiring only one derivative of $u$ instead of two.

where: $u$ — the sought weak solution in $H_0^1(\Omega)$; $v$ — an arbitrary test function in $H_0^1(\Omega)$; $f$ — the source, $f\in L^2(\Omega)$; $H_0^1$ — the closure of $C_c^\infty(\Omega)$ in the $H^1$ norm, building in homogeneous Dirichlet data $u=0$ on $\partial\Omega$. Lowering the derivative count is why piecewise-linear finite elements work. Casting the weak form abstractly as "bilinear form equals linear functional" lets one theorem settle existence, uniqueness, and stability at a stroke.

where: $a(\cdot,\cdot)$ — a bounded, coercive bilinear form with $a(v,v)\ge c\|v\|_V^2$ and $|a(u,v)|\le M\|u\|_V\|v\|_V$; $\ell$ — a bounded linear functional; $V$ — a Hilbert space such as $H_0^1(\Omega)$; $c>0$ — the coercivity constant. This is the Lax–Milgram theorem: coercivity is the abstract shadow of "positive energy", guaranteeing the solution exists and depends continuously on the data.

The weak framework is not bolted on beside the classical one; it contains the classical picture as a special case and reduces to familiar objects in each regime.

1 · The weak derivatives of the absolute value (confidence builder)

The function $u(x)=|x|$ has a corner at the origin and no classical derivative there. Find its first and second weak derivatives on $\mathbb{R}$ and identify the singular object that appears.

Step 1. Test the first derivative. By definition $\langle u',\varphi\rangle=-\langle u,\varphi'\rangle=-\int_{-\infty}^\infty|x|\,\varphi'(x)\,\mathrm{d}x$. Split at the origin: $-\int_0^\infty x\varphi'\,\mathrm{d}x-\int_{-\infty}^0(-x)\varphi'\,\mathrm{d}x$, and integrate each piece by parts. The boundary terms vanish because $\varphi$ has compact support, leaving $\int_0^\infty\varphi\,\mathrm{d}x-\int_{-\infty}^0\varphi\,\mathrm{d}x=\int_{-\infty}^\infty\operatorname{sgn}(x)\,\varphi\,\mathrm{d}x$. Hence the first weak derivative is $\operatorname{sgn}(x)=2H(x)-1$, an honest $L^1_{\mathrm{loc}}$ function even though $u$ is not differentiable at $0$.

Step 2. Differentiate once more. Since $\operatorname{sgn}(x)=2H(x)-1$ and constants have zero derivative, $u''=2H'(x)=2\delta(x)$ by equation (4). So the second weak derivative is $2\delta(x)$ — a singular distribution, twice the unit spike, sitting at the corner.

Step 3. Sanity-check the coefficient. The corner of $|x|$ has slopes $-1$ left and $+1$ right, a jump in slope of $+2$; the second distributional derivative is a delta weighted by exactly that jump, so the factor $2$ is the kink strength. This is the one-dimensional cousin of $\nabla^2(1/r)=-4\pi\delta$: a non-smooth field whose top derivative is a delta whose weight is a geometric invariant of the singularity.

2 · Shock speed for the inviscid Burgers equation (fully worked)

Solve $u_t+\big(\tfrac12u^2\big)_x=0$ with Riemann data $u=2$ for $x<0$ and $u=0$ for $x>0$. Find the shock speed, and check that the shock is the physically admissible solution.

Step 1. Identify the flux and states. Here $f(u)=\tfrac12u^2$, so the characteristic speed is $f'(u)=u$: the left state $u_L=2$ sends characteristics right at speed $2$, the right state $u_R=0$ holds them still. The fast left characteristics overtake the stalled right ones immediately, so no continuous solution exists and a shock forms at $t=0^+$.

Step 2. Apply Rankine–Hugoniot. The jump condition from the weak form gives $s=\dfrac{f(u_R)-f(u_L)}{u_R-u_L}=\dfrac{0-\tfrac12(2)^2}{0-2}=\dfrac{-2}{-2}=\mathbf{1}$. For Burgers this is just the average of the states, $s=\tfrac12(u_L+u_R)=\tfrac12(2+0)=1$, so the shock travels to the right at speed $1$, exactly the red line of the second figure.

Step 3. Check the entropy condition. The Lax condition demands $f'(u_L)>s>f'(u_R)$, i.e. $u_L>s>u_R$. Substituting, $2>1>0$ — satisfied, so characteristics run into the shock from both sides and the discontinuity is the vanishing-viscosity limit, the unique physical solution. Had the data been reversed ($u_L=0$, $u_R=2$), Rankine–Hugoniot would still give $s=1$, but $0>1>2$ is false: that "shock" is unphysical, and the correct answer is a smooth rarefaction fan. s = 1, entropy-admissible; the weak form plus one inequality delivers the whole answer.

The weak formulation was assembled over half a century by physicists reaching for objects their calculus could not hold. Its founder gave the delta a rigorous home and, with it, a Fields Medal.

Texts. Stone & Goldbart, Mathematics for Physics (CUP 2009), Ch. 8–9 (distributions and Green's functions) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 9–10 · Hassani, Mathematical Physics, 2nd ed., Ch. 6–7 · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., Ch. 18–19 · L. C. Evans, Partial Differential Equations, 2nd ed. (AMS), Ch. 3, 5, 6 · R. S. Strichartz, A Guide to Distribution Theory and Fourier Transforms. Key standard results collected here for reference: $H'=\delta$; $\nabla^2(1/r)=-4\pi\delta$ with the $4\pi$ coming from the unit-sphere area in 3D; the Poincaré-based coercivity constant $c=1/(1+C_P^2)$ and energy estimate $\|u\|_{H^1}\le(1+C_P^2)\|f\|_{L^2}$; the Sobolev conjugate $1/p^*=1/p-1/n$ (giving $p^*=6$ for $p=2$, $n=3$); and the Burgers shock speed $s=\tfrac12(u_L+u_R)=1$ for $u_L=2$, $u_R=0$, with entropy check $2>1>0$.

Prerequisites. Classification of Second-Order Linear PDEs · Laplace's and Poisson's Equations · Boundary Conditions and Well-Posedness · Sturm–Liouville Theory and Eigenfunction Expansions. Leads to. Method of Characteristics · d'Alembert's Solution and Characteristics of the Wave Equation · The Helmholtz Equation.

Ask the tutor: Ask me why $\nabla^2(1/r)=-4\pi\delta$ pins the Coulomb Green's function, or to build the weak form of any PDE and check its Lax–Milgram hypotheses.