Analysis for Mathematical Relativity

Mathematical relativity often studies metrics and fields that are not classically differentiable everywhere, sequences that converge only weakly, and elliptic equations whose estimates depend on geometry and boundary regularity. The essential preparation is not a list of theorem names but the ability to state the function space, topology, hypotheses, and estimate being used. This module supplies that bounded readiness check; it is not a real-analysis or geometric-PDE course.

Required background. Calculus and Linear Algebra for Relativity supplies norms, inner products, derivatives, and integration by parts. Differential Equations for Gravity supplies linear operators, boundary data, and stability reasoning.

Helpful background. PDEs and Green Functions for Relativity supplies weak formulations, elliptic/hyperbolic distinctions, and energy estimates.

You should leave able to use Sobolev scaling, weak convergence, compactness, distributions or measures, and elliptic regularity with hypotheses stated, then organize a proof for a simple geometric elliptic equation.

Functional-analysis and geometric-PDE diagnostic

1. For $u_\lambda(x)=u(\lambda x)$ on $\mathbb R^n$, derive the scaling of $\lVert u_\lambda\rVert_{L^2}$ and $\lVert\nabla u_\lambda\rVert_{L^2}$. Which homogeneous Sobolev exponent $s$ makes $\lVert u_\lambda\rVert_{\dot H^s}$ scale invariant?
2. Let $u_j(x)=\sqrt2\sin(j\pi x)$ in $L^2(0,1)$. Show that $u_j\rightharpoonup0$ weakly but does not converge strongly to zero.
3. Let $H(x)$ be the Heaviside function. Compute its distributional derivative by testing against $\varphi\in C_c^\infty(\mathbb R)$.
4. For $-\Delta u=f$ on a bounded domain $\Omega$ with homogeneous Dirichlet data, derive the weak formulation and name function spaces for $u$, the test function, and $f$ that make the formulation well posed.
5. Evaluate the claim: “If $f\in L^2(\Omega)$ and $-\Delta u=f$, then $u\in H^2(\Omega)$.” Which interior or boundary hypotheses are missing?
6. Let $\rho_\varepsilon(x)=\varepsilon^{-n}\rho(x/\varepsilon)$, where $\rho\geq0$, $\rho\in C_c^\infty(\mathbb R^n)$, and $\int\rho\,dx=1$. Identify its distributional limit and explain why this need not be strong convergence in $L^p$ for $p\gt1$.

Hints and solutions

Hints for 1

Change variables $y=\lambda x$. Each derivative contributes one power of $\lambda$, while the volume element contributes $\lambda^{-n}$.

Solution to 1

Changing variables gives

$$\lVert u_\lambda\rVert_{L^2}^2 =\lambda^{-n}\lVert u\rVert_{L^2}^2, \qquad \lVert\nabla u_\lambda\rVert_{L^2}^2 =\lambda^{2-n}\lVert\nabla u\rVert_{L^2}^2.$$

Thus

$$\lVert u_\lambda\rVert_{L^2}=\lambda^{-n/2}\lVert u\rVert_{L^2}, \qquad \lVert\nabla u_\lambda\rVert_{L^2} =\lambda^{1-n/2}\lVert\nabla u\rVert_{L^2}.$$

More generally $\lVert u_\lambda\rVert_{\dot H^s}=\lambda^{s-n/2}\lVert u\rVert_{\dot H^s}$, so the scale-invariant exponent is $s=n/2$. A nonlinear PDE may assign additional amplitude scaling to $u$; its critical exponent must use that full scaling, not this coordinate rescaling alone.

Hints for 2

The sequence is orthonormal. Use Bessel’s inequality for its inner products with an arbitrary fixed $v\in L^2(0,1)$, and compare norms for strong convergence.

Solution to 2

For any $v\in L^2(0,1)$, the coefficients $\langle v,u_j\rangle$ form a square-summable sequence by Bessel’s inequality. Every square-summable sequence tends to zero, so

$$\langle v,u_j\rangle\longrightarrow0.$$

This is precisely $u_j\rightharpoonup0$ in $L^2$. However,

$$\lVert u_j\rVert_{L^2}=1$$

for every $j$, so $\lVert u_j-0\rVert_{L^2}$ does not tend to zero. Weak convergence controls continuous linear probes but need not preserve nonlinear quantities such as pointwise squares or norms.

Hints for 3

By definition, $\langle H',\varphi\rangle=-\langle H,\varphi'\rangle$. Use compact support at positive infinity.

Solution to 3

The distributional derivative satisfies

$$\langle H',\varphi\rangle =-\int_0^\infty\varphi'(x)\,dx =\varphi(0).$$

Since $\langle\delta_0,\varphi\rangle=\varphi(0)$, one has $H'=\delta_0$ as distributions. The derivative records the jump even though the classical derivative vanishes away from the origin and is undefined at the origin.

Hints for 4

Multiply by a test function with zero trace and integrate by parts. The dual of $H_0^1(\Omega)$ is $H^{-1}(\Omega)$.

Solution to 4

For smooth fields, multiplication by $v$ and integration by parts give

$$\int_\Omega\nabla u\cdot\nabla v\,dx =\langle f,v\rangle.$$

The weak Dirichlet problem is: find $u\in H_0^1(\Omega)$ such that this identity holds for every $v\in H_0^1(\Omega)$, with $f\in H^{-1}(\Omega)$. On a bounded domain for which the Poincaré inequality holds, the bilinear form is continuous and coercive. Lax–Milgram then gives a unique weak solution and an $H^1$ estimate. If $f\in L^2(\Omega)$, the right-hand side is the ordinary integral $\int_\Omega fv\,dx$.

Hints for 5

Separate local regularity from regularity up to the boundary. Ask about the domain, boundary data, coefficients, and whether the equation holds weakly.

Solution to 5

The bare claim is incomplete. If $u\in H^1_{\rm loc}(\Omega)$ is a weak solution and $f\in L^2_{\rm loc}(\Omega)$, constant-coefficient elliptic interior regularity gives $u\in H^2_{\rm loc}(\Omega)$. To conclude global $H^2(\Omega)$ regularity for the Dirichlet problem, one also needs suitable boundary regularity—commonly a bounded $C^{1,1}$ domain—compatible boundary data, and the corresponding weak solution space. For variable coefficients one needs uniform ellipticity and sufficient coefficient regularity. Corners, rough coefficients, incompatible traces, or a nonsmooth boundary can prevent global $H^2$ regularity even when $f\in L^2$.

Hints for 6

Test against a smooth compactly supported function and change variables. Then compute the $L^p$ norm by the same scaling.

Solution to 6

For a test function $\varphi$,

$$\int_{\mathbb R^n}\rho_\varepsilon(x)\varphi(x)\,dx =\int_{\mathbb R^n}\rho(y)\varphi(\varepsilon y)\,dy \longrightarrow\varphi(0),$$

so $\rho_\varepsilon\to\delta_0$ in distributions and weak-* as finite measures. Meanwhile

$$\lVert\rho_\varepsilon\rVert_{L^p} =\varepsilon^{-n(1-1/p)}\lVert\rho\rVert_{L^p}.$$

For $p\gt1$ this norm diverges as $\varepsilon\to0$, so there is no strong $L^p$ convergence to an ordinary function. The topology of convergence determines which operations survive the limit.

Interpret the result

• Secure: every convergence statement names its topology, weak equations name their trial/test spaces, and regularity gains list operator, coefficient, domain, and boundary hypotheses.
• Uncertain: the main theorem or estimate is appropriate but one space, scaling power, or boundary assumption needs a hint. Repeat that item in another dimension or on another domain.
• Not yet: weak convergence is substituted into nonlinear expressions without justification, distributions are multiplied formally, boundary regularity is suppressed, or an elliptic estimate is quoted with no norm or hypotheses.

This diagnostic is low-stakes. Practice the analytic operation that caused difficulty and demonstrate it on the transfer problem before entering a proof-heavy route.

Focused review

Sobolev norms and scaling

State whether a norm is homogeneous or inhomogeneous, local or global, and scalar- or tensor-valued. Coordinate scaling is only one part of a PDE’s natural scaling. Sobolev Thresholds for Einstein Evolution shows why product estimates, gauge choice, and nonlinear structure matter beyond a dimensional count.

Weak convergence and compactness

Bounded sequences in a reflexive space have weakly convergent subsequences, but identifying the limit in a nonlinear equation may require strong compactness, compensated compactness, or defect measures. Compactness and Limits of Spacetimes explains how these issues arise in geometry.

Distributions and nonlinear geometry

A distribution is a continuous linear functional on a declared test-function space. Differentiation is always defined by duality, but arbitrary products of distributions are not. Review Distributions and Generalized Functions before using singular sources, and Distributional Curvature before forming curvature from rough metrics.

Weak formulations and elliptic regularity

Derive a weak equation by integration by parts, state boundary traces, and prove continuity and coercivity or identify the kernel. Regularity is a separate step after existence. Weak Formulations of Einstein’s Equations shows why a geometrically meaningful weak formulation needs more than componentwise formal integration.

Transfer task: a coercive geometric elliptic equation

Let $(\Sigma,h)$ be a smooth compact Riemannian three-manifold without boundary. Suppose $V\in L^\infty(\Sigma)$ and $V\geq v_0\gt0$ almost everywhere. For $f\in H^{-1}(\Sigma)$, consider

$$L_hu=-\Delta_hu+Vu=f.$$

Write the weak formulation on $H^1(\Sigma)$, prove existence and uniqueness with an energy estimate, and state additional hypotheses that yield $u\in H^{k+2}(\Sigma)$ when $f\in H^k(\Sigma)$. Finally, explain what changes if $V=0$.

Hint 1

Use

$$a(u,v)=\int_\Sigma \bigl(\nabla u\cdot\nabla v+Vuv\bigr)\,d\mu_h.$$

Boundedness of $V$ gives continuity; the positive lower bound gives coercivity on the full $H^1$ norm.

Hint 2

Apply Lax–Milgram, then choose $v=u$. For regularity, distinguish bounded measurable $V$ from smooth coefficients. When $V=0$, inspect the constant functions.

Solution

The weak problem is to find $u\in H^1(\Sigma)$ such that

$$a(u,v)=\langle f,v\rangle_{H^{-1},H^1}$$

for every $v\in H^1(\Sigma)$. Since $\Sigma$ is compact and $V\in L^\infty$,

$$|a(u,v)|\leq C\lVert u\rVert_{H^1}\lVert v\rVert_{H^1}.$$

Moreover,

$$a(u,u) =\lVert\nabla u\rVert_{L^2}^2 +\int_\Sigma V|u|^2\,d\mu_h \geq \min(1,v_0)\lVert u\rVert_{H^1}^2.$$

The form is continuous and coercive, so Lax–Milgram gives a unique weak solution. Testing with $u$ and using the dual norm yields

$$\min(1,v_0)\lVert u\rVert_{H^1}^2 \leq \lVert f\rVert_{H^{-1}}\lVert u\rVert_{H^1},$$

hence

$$\lVert u\rVert_{H^1} \leq\frac{1}{\min(1,v_0)}\lVert f\rVert_{H^{-1}}.$$

If $h$ and $V$ are sufficiently smooth—enough derivatives for the desired order—standard elliptic regularity on the closed manifold gives

$$f\in H^k(\Sigma)\quad\Longrightarrow\quad u\in H^{k+2}(\Sigma),$$

with an estimate whose constant depends on the geometry, coefficients, and coercivity. The original assumption $V\in L^\infty$ alone is not enough for arbitrary higher-order gains.

If $V=0$, constants lie in the kernel of $-\Delta_h$. Solvability requires the compatibility condition $\langle f,1\rangle=0$, and uniqueness holds only after fixing the mean of $u$ or quotienting by constants. This kernel check is exactly the kind of hypothesis that must precede a geometric elliptic inversion.

Common errors

• Naming a Sobolev space without specifying domain, tensor character, or homogeneous versus inhomogeneous norm.
• Passing a weak limit through a quadratic curvature or stress tensor without compactness or a defect term.
• Treating a delta sequence as convergence in every norm.
• Quoting interior elliptic regularity as a global boundary theorem.
• Applying Lax–Milgram without checking coercivity or the operator’s kernel.

Return to your route

After the geometric elliptic transfer problem, return to Mathematical Relativity.