Wave-front set of a distribution

Intuition Beginner

A distribution can fail to be smooth at a point, and the wave-front set records both where the failure happens and in which directions it is visible. The singular support tells you the bad points; the wave-front set tells you the bad points together with the bad directions at each of them.

Think of the Dirac delta at the origin on the line. It is singular at one point, the origin. Now ask: in which directions does that singularity propagate? Run a smooth bump centred at the origin against it, take a Fourier transform, and look at how fast the transform decays in each direction. The transform of a smoothly cut-off delta is just a smooth bump's transform, which does not decay rapidly in any direction. So every direction is a bad direction at the origin.

Why should you care about directions? Because most singularities of real distributions are directional. A shock front on a moving fluid is singular along the shock surface, but only in the direction normal to the surface; tangent to the surface, the distribution is perfectly smooth. The wave-front set is the right notion of singularity for partial differential equations because solutions inherit directional singularities from their data.

Visual Beginner

Picture a half-plane in $(x,\xi )$-space, with $x$ on the horizontal axis and $\xi$ on the vertical axis. The horizontal axis records position; the vertical axis records direction (a "frequency"). The wave-front set lives in this combined picture, not in position alone.

At the origin of position, the wave-front set of the delta function is the whole fan of directions in the upper half-plane. Every direction is bad. For a half-line indicator like the Heaviside step, the singularity is again at one position (the jump point), but the wave-front set is supported on positive directions only.

Worked example Beginner

Compute the wave-front set of ${\delta }_0$ on the real line. The Dirac delta is the distribution that pairs with a test function $\phi$ by giving back $\phi (0)$.

Step 1. Pick a smooth bump $\chi$ that equals $1$ near the origin and vanishes outside a small neighbourhood. The product $\chi {\delta }_0$ equals ${\delta }_0$, because the bump is $1$ at the origin.

Step 2. Take the Fourier transform of $\chi {\delta }_0={\delta }_0$. By the defining formula, the Fourier transform of ${\delta }_0$ is the constant function $1$. This holds on the whole real line, not just on one direction.

Step 3. Ask whether the transform decays rapidly. A function decays rapidly in a direction if it decays faster than every polynomial-reciprocal in that direction. The constant function $1$ does not decay at all. So the transform of the bump-cutoff ${\delta }_0$ fails to decay in every direction.

Step 4. By the definition of the wave-front set, a point $(0,\xi )$ with $\xi \ne 0$ is not in the complement of the wave-front set, because there is no bump and no conic open set around $\xi$ on which the transform decays rapidly. So $(0,\xi )$ is in the wave-front set for every non-zero $\xi$.

Step 5. Away from the origin, ${\delta }_0$ vanishes as a distribution: $\chi {\delta }_0=0$ for any bump $\chi$ supported away from $0$, and the transform of $0$ decays rapidly in every direction. So no point $(x_0,{\xi }_0)$ with $x_0\ne 0$ is in the wave-front set.

Conclusion: $\mathrm{WF}({\delta }_0)$ is the set of pairs $(0,\xi )$ with $\xi \ne 0$. In symbols, $\mathrm{WF}({\delta }_0)=\{0\}\times (\mathbb{R}\setminus \{0\})$. The singular point is the origin; the singular directions at that point are all non-zero directions.

What this tells us: the wave-front set lives in position-direction space, not in position alone. Even though the singular support is just one point, the wave-front set fans out into directions over that point. For more directional distributions, like the Heaviside step, the fan narrows to one ray; for solutions of wave equations, the fan picks out exactly the directions along which singularities travel.

Check your understanding Beginner

Formal definition Intermediate+

Let $X\subseteq {\mathbb{R}}^n$ be open and let $u\in {\mathcal{D}}^{\prime }(X)$ be a distribution on $X$ in the sense of L. Schwartz: a continuous linear functional on the test-function space $C_c^{\infty }(X)$ equipped with its usual inductive-limit topology. (The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ of rapidly decreasing smooth functions and its dual ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ of tempered distributions are the natural target on $X={\mathbb{R}}^n$; the wave-front set is defined the same way on ${\mathcal{D}}^{\prime }(X)$ and on ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ via localisation.) Recall the Fourier transform convention $$ \hat{f}(\xi) = \int_{\mathbb{R}^n} e^{-i x \cdot \xi} f(x) , dx, \qquad f \in \mathcal{S}(\mathbb{R}^n), $$ extended to tempered distributions by duality, and the notation that a function $g$ on ${\mathbb{R}}^n$ is rapidly decreasing on a set $V$ if for every $N\ge 0$ there is a constant $C_N$ with $|g(\xi )|\le C_N(1+|\xi |{)}^{-N}$ for all $\xi \in V$.

A subset $V\subseteq {\mathbb{R}}^n\setminus 0$ is conic if $\xi \in V$ implies $t\xi \in V$ for every $t>0$. The wave-front set lives in the punctured cotangent bundle $T^{\ast }X\setminus 0=X\times ({\mathbb{R}}^n\setminus 0)$, with the second factor regarded as conic.

Definition (wave-front set). Let $u\in {\mathcal{D}}^{\prime }(X)$. The wave-front set $\mathrm{WF}(u)\subseteq T^{\ast }X\setminus 0$ is the closed conic subset whose complement is characterised by: $(x_0,{\xi }_0)\in (T^{\ast }X\setminus 0)\setminus \mathrm{WF}(u)$ iff there exist $\phi \in C_c^{\infty }(X)$ with $\phi (x_0)\ne 0$ and an open conic neighbourhood $V\subset {\mathbb{R}}^n\setminus 0$ of ${\xi }_0$ such that $\widehat{\phi u}(\xi )$ is rapidly decreasing on $V$.

The cutoff $\phi$ localises $u$ to a small neighbourhood of $x_0$; the conic neighbourhood $V$ localises in direction at ${\xi }_0$. The complement of $\mathrm{WF}(u)$ collects the pairs $(x_0,{\xi }_0)$ for which $u$ is microlocally smooth — smooth at $x_0$ in direction ${\xi }_0$.

The definition is independent of the choice of cutoff: if $(x_0,{\xi }_0)$ admits one pair $(\phi ,V)$ with rapid decay, then every other cutoff supported in a small enough neighbourhood admits an open conic refinement of $V$ with rapid decay. The proof passes through Hörmander Vol. I §8.1 Prop. 8.1.3 by a careful estimate on the convolution of two compactly supported smooth functions and the resulting Fourier-decay transfer.

Notation and conventions

• ${\mathcal{D}}^{\prime }(X)$: Schwartz distributions on the open set $X\subseteq {\mathbb{R}}^n$. ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$: tempered distributions on ${\mathbb{R}}^n$.
• $T^{\ast }X\setminus 0$: the punctured cotangent bundle, here identified with $X\times ({\mathbb{R}}^n\setminus 0)$ for $X$ open in ${\mathbb{R}}^n$. The second factor is regarded up to positive scaling.
• $\mathrm{singsupp}(u)$: the singular support of $u$, the complement of the largest open subset of $X$ on which $u$ is given by a smooth function.
• Sign convention: $\hat{f}(\xi )=\int e^{-ix\cdot \xi }f(x)dx$. The opposite sign convention $e^{+ix\cdot \xi }$ used in some references flips the sign of every $\xi$ in the wave-front-set fibres; this matters when reading sources that adopt the physicist sign convention.

Projection to the singular support and conic property

Two structural facts:

Proposition. The projection ${\pi }_X(\mathrm{WF}(u))$ onto $X$ equals the singular support $\mathrm{singsupp}(u)$.

Proof. If $x_0\notin \mathrm{singsupp}(u)$, then $u$ is given by a smooth function on some open neighbourhood $U\ni x_0$. Pick $\phi \in C_c^{\infty }(U)$ with $\phi (x_0)\ne 0$. Then $\phi u\in C_c^{\infty }({\mathbb{R}}^n)$, so $\widehat{\phi u}$ is Schwartz, hence rapidly decreasing on every conic set. So $(x_0,\xi )\notin \mathrm{WF}(u)$ for every $\xi \ne 0$, and $x_0\notin {\pi }_X(\mathrm{WF}(u))$.

Conversely, if $x_0\notin {\pi }_X(\mathrm{WF}(u))$, then for every $\xi \in {\mathbb{R}}^n\setminus 0$ there is $({\phi }_{\xi },V_{\xi })$ giving rapid decay near $\xi$. Compactness of the unit sphere in $\xi$ furnishes finitely many $V_{\xi }$ covering the sphere, hence covering all of ${\mathbb{R}}^n\setminus 0$ by conicity. A common cutoff $\phi$ supported in the intersection of the (finitely many) cutoff supports works on all of them. So $\widehat{\phi u}$ is rapidly decreasing on ${\mathbb{R}}^n\setminus 0$, hence on ${\mathbb{R}}^n$, and $\phi u$ is Schwartz, hence smooth on a neighbourhood of $x_0$. $\square$

Proposition. *$\mathrm{WF}(u)$ is a closed conic subset of $T^{\ast }X\setminus 0$.*

Proof. Closedness: the complement is open by direct inspection of the definition (a rapid-decay estimate on an open cutoff and open conic neighbourhood persists under small perturbations of $(x_0,{\xi }_0)$). Conicity: if $(x_0,{\xi }_0)\notin \mathrm{WF}(u)$ with witness $(\phi ,V)$, then $(x_0,t{\xi }_0)\notin \mathrm{WF}(u)$ with the same witness, because $V$ is conic and contains $t{\xi }_0$ for every $t>0$. $\square$

Counterexamples to common slips

• The wave-front set is not a subset of $X\times {\mathbb{R}}^n$ — the zero direction is excluded. Conicity would make the zero direction redundant in any case, but the definition asks for ${\xi }_0\in {\mathbb{R}}^n\setminus 0$.
• Singular support and wave-front set differ in directional information, not in support location. The wave-front set of ${\delta }_0$ is the full fibre over $0$, not just one direction.
• A distribution can be smooth at $x_0$ in some directions while singular in others. The conormal of a smooth hypersurface is the prototype: a layer distribution on $\{x_n=0\}$ has wave-front set concentrated on the conormal direction $(x,e_n^{\ast })$, smooth in every tangential direction.

Key theorem with proof Intermediate+

Theorem (Hörmander's product theorem; Hörmander Vol. I Theorem 8.2.10 ^{[Hörmander Vol. I §8.2]}). Let $u,v\in {\mathcal{D}}^{\prime }(X)$ and assume the diagonal-avoidance condition $$ \big{ (x, \xi) \in T^*X \setminus 0 : (x, \xi) \in \mathrm{WF}(u) \text{ and } (x, -\xi) \in \mathrm{WF}(v) \big} = \emptyset. $$ *Equivalently, the sumset $\mathrm{WF}(u)+\mathrm{WF}(v)$ defined fibrewise misses the zero section of $T^{\ast }X$. Then the pointwise product $u\cdot v$ extends uniquely from the smooth case to a well-defined distribution in ${\mathcal{D}}^{\prime }(X)$, with* $$ \mathrm{WF}(u \cdot v) \subseteq \mathrm{WF}(u) \cup \mathrm{WF}(v) \cup \big( \mathrm{WF}(u) + \mathrm{WF}(v) \big), $$ where the sum is fibrewise over $X$.

Proof. The argument has three steps. First, reduce the global product to a local product near each point of $X$ by partition of unity. Second, prove the local product is a distribution by a Fourier-convolution estimate. Third, bound the wave-front set of the product by a convolution estimate on the Fourier transforms.

Step 1: local reduction. Cover $X$ by open sets $U_{\alpha }$ on which both $u$ and $v$ have well-controlled localisations, and pick a smooth partition of unity $\{{\chi }_{\alpha }\}$ subordinate to the cover. Each ${\chi }_{\alpha }u$ and ${\chi }_{\alpha }v$ is compactly supported, and the product is determined by the local products $({\chi }_{\alpha }u)({\chi }_{\alpha }v)$ summed against test functions supported in $U_{\alpha }$. So it suffices to define $u\cdot v$ for $u,v\in {\mathcal{E}}^{\prime }(X)$ (compactly supported distributions) and check the wave-front-set bound there.

Step 2: the product as a Fourier convolution. For $u,v\in {\mathcal{E}}^{\prime }({\mathbb{R}}^n)$ both compactly supported, the candidate product $uv$, if it exists, must satisfy $$ \widehat{uv}(\xi) = (2\pi)^{-n} (\hat{u} * \hat{v})(\xi) = (2\pi)^{-n} \int \hat{u}(\xi - \eta) \hat{v}(\eta) , d\eta, $$ extending the convolution rule for Fourier transforms of smooth functions. Both $\hat{u}$ and $\hat{v}$ are smooth functions of polynomial growth (Paley-Wiener: the Fourier transform of a compactly supported distribution is an entire function of exponential type, polynomially bounded on real $\eta$). For the convolution to be a polynomially bounded smooth function on ${\mathbb{R}}^n$ — which is the requirement for it to be the Fourier transform of a distribution — we need the integrand $\hat{u}(\xi -\eta )\hat{v}(\eta )$ to be integrable on ${\mathbb{R}}_{\eta }^n$ uniformly in $\xi$.

The diagonal-avoidance condition is exactly what guarantees integrability. Decompose ${\mathbb{R}}_{\eta }^n=V\cup V^c$ where $V$ is an open conic neighbourhood of the bad directions of $v$ at the relevant base point and $V^c$ its conic complement. On $V^c$, $\hat{v}$ decays rapidly by definition of the wave-front set, so the integrand decays rapidly there. On $V$, the diagonal-avoidance condition forces $\xi -\eta$ to lie outside the bad directions of $u$ whenever $\eta$ lies in $V$ and $\xi$ is in the relevant conic neighbourhood; rapid decay of $\hat{u}(\xi -\eta )$ on this set makes the integrand integrable. The convolution converges absolutely and produces a smooth function of $\xi$ of polynomial growth, which is therefore the Fourier transform of a tempered distribution. This distribution is $uv$.

Step 3: wave-front-set bound. Take $(x_0,{\xi }_0)\notin \mathrm{WF}(u)\cup \mathrm{WF}(v)\cup (\mathrm{WF}(u)+\mathrm{WF}(v))$. Choose a cutoff $\phi$ supported in a small neighbourhood of $x_0$ and a conic neighbourhood $W\ni {\xi }_0$ such that:

• $\widehat{\phi u}(\zeta )$ is rapidly decreasing on $W$ (since $(x_0,{\xi }_0)\notin \mathrm{WF}(u)$);
• $\widehat{\phi v}(\eta )$ is rapidly decreasing on $W$;
• $\widehat{\phi u}(\xi -\eta )\widehat{\phi v}(\eta )$ is rapidly decreasing as $|\xi |\to \infty$ in $W$, uniformly in $\eta \in {\mathbb{R}}^n$, on account of the sumset condition.

The convolution estimate $\widehat{{\phi }^{2}uv}(\xi )=(2\pi {)}^{-n}(\widehat{\phi u}\ast \widehat{\phi v})(\xi )$ (with an extra power of $\phi$ absorbed into the cutoffs of $u$ and $v$) shows that the right-hand side is rapidly decreasing on a conic neighbourhood of ${\xi }_0$. So $(x_0,{\xi }_0)\notin \mathrm{WF}(uv)$. The bound on $\mathrm{WF}(uv)$ follows. $\square$

Bridge. Hörmander's product theorem builds toward every microlocal calculation that involves multiplying distributions on a manifold. The diagonal-avoidance condition is the microlocal generalisation of the classical fact that the product of two distributions with disjoint singular supports is always well-defined: the sumset condition replaces "disjoint singular supports" with "no cancelling conormal directions", which is the right notion for distributions with overlapping singular supports but compatible wave-front sets. This pattern appears again in 03.12.16 (Poincaré duality) and in 05.02.06 (geodesic flow on the cotangent bundle), where wave-front-set bookkeeping controls the pullback of singularities under maps and along Hamiltonian flow. The foundational insight is that the local-microlocal data $(\phi ,V)$ packages the singularity of a distribution as conic decay of its Fourier transform, and this packaging is functorial under multiplication, pullback by submersions, and propagation by pseudo-differential operators. Putting these together produces the entire microlocal calculus: wave-front sets are stable under $\Psi$DO action up to the characteristic variety, they propagate along Hamiltonian flow of the principal symbol, and they detect Hadamard states by their conormal structure on the diagonal of a Lorentzian manifold (Radzikowski 1996).

Exercises Intermediate+

Advanced results Master

Theorem (intrinsic definition on a manifold; Hörmander Vol. I §8.2). *Let $M$ be a smooth manifold, $X\subseteq M$ open, and $u\in {\mathcal{D}}^{\prime }(X)$. The wave-front set $\mathrm{WF}(u)\subset T^{\ast }X\setminus 0$ is defined chart-by-chart: $(x_0,{\xi }_0)\in T^{\ast }X\setminus 0$ is in $\mathrm{WF}(u)$ iff, in any (equivalently every) chart $(U,\kappa )$ at $x_0$, the corresponding point in $T^{\ast }\kappa (U)\setminus 0$ lies in the Euclidean wave-front set of ${\kappa }_{\ast }u$. The chart-independence follows from Exercise 4: diffeomorphism invariance lifts to the cotangent action $(Df{)}^{\top }$, so the chart-by-chart definition glues to a closed conic subset of $T^{\ast }X\setminus 0$.*

This intrinsic version is what is used on a Lorentzian manifold in Gérard Ch. 2 ^[Gérard]. The wave-front set is a microlocal invariant of distributions on a manifold, recording the conormal directions along which the distribution is singular.

Theorem (pseudolocality and elliptic regularity for $\Psi$DOs; Hörmander Vol. III §18.1 ^{[Hörmander Vol. I §8 + Vol. III §18.1]}). Let $P\in {\Psi }^m(X)$ be a properly supported pseudo-differential operator of order $m$ on an open $X\subseteq {\mathbb{R}}^n$ (or on a manifold) with principal symbol $p_m\in S^m/S^{m-1}$. Then for every $u\in {\mathcal{D}}^{\prime }(X)$, $$ \mathrm{WF}(P u) \subseteq \mathrm{WF}(u) \quad \text{(pseudolocality)}. $$ If furthermore $P$ is elliptic at $(x_0,{\xi }_0)$, meaning $p_m(x_0,{\xi }_0)\ne 0$, then $$ (x_0, \xi_0) \in \mathrm{WF}(u) \iff (x_0, \xi_0) \in \mathrm{WF}(P u) \quad \text{(elliptic regularity)}. $$

Pseudolocality says $\Psi$DOs do not create new wave-front directions; elliptic regularity says away from the characteristic variety $\{p_m=0\}$ they do not destroy them either. The wave-front set is therefore the right notion of singularity for studying linear PDE: it is preserved by every elliptic operation and detects exactly the obstruction to local smoothness modulo elliptic correction.

Theorem (propagation of singularities along Hamiltonian flow; Hörmander 1971 Theorem 6.1.1 ^{[Hörmander 1971]}). Let $P\in {\Psi }^m(X)$ have real principal symbol $p_m$. For every $u\in {\mathcal{D}}^{\prime }(X)$, $$ \mathrm{WF}(u) \setminus \mathrm{WF}(P u) \subseteq p_m^{-1}(0) $$ *and $\mathrm{WF}(u)\setminus \mathrm{WF}(Pu)$ is invariant under the Hamiltonian flow ${\Phi }_t^{H_{p_m}}$ of $p_m$ on $T^{\ast }X\setminus 0$.*

This is the load-bearing technical theorem of microlocal analysis. The Hamiltonian flow of the principal symbol carries wave-front-set singularities along the bicharacteristic curves of the operator. For the wave operator ${\square }_g$ on a Lorentzian manifold, $p_m(x,\xi )=g^{\mu \nu }(x){\xi }_{\mu }{\xi }_{\nu }$, and the bicharacteristics are the null geodesics: singularities of a wave-equation solution propagate along null geodesics. For the Laplace operator on a Riemannian manifold, $p_m$ is positive definite away from the zero section, so the characteristic variety is empty and there are no propagation directions — this is the analytic reflection of ellipticity.

Theorem (Hadamard-state criterion; Radzikowski 1996 ^{[source pending]}). On a globally hyperbolic Lorentzian manifold $(M,g)$, a quasi-free state $\omega$ of the free scalar Klein-Gordon field is Hadamard iff its two-point function ${\omega }_2\in {\mathcal{D}}^{\prime }(M\times M)$ has wave-front set $$ \mathrm{WF}(\omega_2) = \big{ ((x_1, \xi_1), (x_2, -\xi_2)) \in T^(M \times M) \setminus 0 :\ (x_1, \xi_1) \sim_g (x_2, \xi_2),\ \xi_1 \in \overline{V^+{x_1}} \big},$$ *where ${\sim }_g$ denotes the null-geodesic equivalence relation (the points are connected by a null geodesic with $\xi$ the parallel transport of the cotangent) and $\overline{V^+{x_1}}$istheclosedfuturelightconein$T^*_{x_1} M$.

The Radzikowski criterion is the wave-front-set formulation of the older Hadamard condition (Kay-Wald 1991 ^{[source pending]}) and is the central definition of Gérard 2019 ^[Gérard]. Every existence theorem for Hadamard states on a globally hyperbolic spacetime — Fulling-Narcowich-Wald 1981, Junker-Olbermann, Gérard-Wrochna 2014 — verifies the wave-front-set condition above.

Theorem (analytic wave-front set; Sato 1969 / Hörmander Vol. I §8.4 ^{[Hörmander Vol. I]}). There is a refinement ${\mathrm{WF}}_A(u)\supseteq \mathrm{WF}(u)$ defined by replacing rapid decay of $\widehat{\phi u}$ with exponential decay on a conic neighbourhood, using a family of analytic cutoffs ${\phi }_N$ with derivative-bound $|{\partial }^{\alpha }{\phi }_N|\le C{N}^{|\alpha |}$ instead of compactly supported cutoffs. The analytic wave-front set records the directions in which $u$ fails to be real-analytic, refining the $C^{\infty }$ wave-front set. Sato's singular spectrum $\mathrm{SS}(u)$ in the hyperfunction setting agrees with ${\mathrm{WF}}_A(u)$ for distributions.

The analytic wave-front set captures sharper information than the $C^{\infty }$ wave-front set but is harder to compute. It is the right object for analytic-PDE problems and for the Sato-Kashiwara-Kawai theory of microfunctions and analytic pseudo-differential equations ^{[Sato-Kawai-Kashiwara]}.

Theorem (wave-front-set pullback; Hörmander Vol. I §8.2 ^{[Hörmander Vol. I]}). Let $f:Y\to X$ be a smooth map between open subsets of Euclidean space. For $u\in {\mathcal{D}}^{\prime }(X)$ with $\mathrm{WF}(u)\cap N^{\ast }f=\varnothing$, where $N^{\ast }f=\{(f(y),\eta ):(Df(y){)}^{\top }\eta =0\}$ is the conormal of $f$, the pullback $f^{\ast }u\in {\mathcal{D}}^{\prime }(Y)$ is defined and* $$ \mathrm{WF}(f^ u) \subseteq f^* \mathrm{WF}(u) = \big{ (y, (Df(y))^\top \xi) : (f(y), \xi) \in \mathrm{WF}(u) \big}. $$

The conormal condition $\mathrm{WF}(u)\cap N^{\ast }f=\varnothing$ rules out cancellation between the singularities of $u$ and the conormal directions of $f$, exactly as the diagonal-avoidance condition rules out cancellation in the product theorem. The two together — pullback and product — give a calculus of distributions under smooth-map operations, controlled by wave-front-set bookkeeping.

Synthesis. The wave-front set is the foundational microlocal invariant of a distribution: it records both the position and the conormal direction of every singularity. The central insight is that conic decay of the localised Fourier transform $\widehat{\phi u}$ on a direction ${\xi }_0$ exactly captures microlocal smoothness at $(x_0,{\xi }_0)$, and the resulting closed conic subset of $T^{\ast }X\setminus 0$ is functorial under all the operations of microlocal analysis. This is exactly the same local-to-global principle that appears again in 03.12.16 (Poincaré duality), where local data on a manifold is assembled into a global invariant by a partition-of-unity argument identical in structure to the partition-of-unity argument that pieces together the wave-front set of a distribution on a manifold. Putting these together, one wave-front-set framework controls every microlocal operation — multiplication by Hörmander's product theorem, pullback by submersions and other admissible maps, propagation under $\Psi$DOs along Hamiltonian flow, and the conormal-bundle characterisation of layer distributions on hypersurfaces — and produces every classical regularity theorem of linear PDE as a consequence. The bridge is the recognition that a singularity is not just a point but a direction in cotangent space, and the right calculus is therefore microlocal, not local.

The wave-front set identifies several singularity notions that look distinct at first inspection. The singular support records position; the conormal bundle records direction at a smooth submanifold; the spectrum condition of a quantum-mechanical two-point function records the energy-momentum support of a positive-energy distribution on Minkowski. The wave-front set unifies all three: position is the projection to $X$, direction is the cotangent fibre, and the Hadamard criterion of Radzikowski 1996 reads the wave-front set of the two-point function on a curved Lorentzian background as the exact analogue of the Minkowski spectrum condition. Sato's analytic singular spectrum and Hörmander's $C^{\infty }$ wave-front set are two sharpness levels of the same idea, differing only in whether real-analytic or $C^{\infty }$ regularity is the target. Atiyah-Singer-style index theorems for elliptic $\Psi$DOs make sense exactly because the wave-front set transforms predictably under $\Psi$DO action, with the characteristic variety $\{p_m=0\}$ playing the role of the locus where ellipticity can fail. The wave-front set generalises to vector-valued and operator-valued distributions in QFT, to wave-front-set-with-coefficients in microlocal sheaf theory (Kashiwara-Schapira), and to the conic Lagrangian submanifolds of Hörmander's Fourier integral operator theory; every one of these specialisations is controlled by the same conic-decay condition with which we began.

Full proof set Master

Theorem (Hörmander's product theorem), proof. Given in the Intermediate-tier section: local reduction by partition of unity, the convolution representation $\widehat{uv}=(2\pi {)}^{-n}\hat{u}\ast \hat{v}$ with absolute integrability guaranteed by the diagonal-avoidance condition, and the conic decay estimate on the convolution localised by cutoffs. The full argument is in Hörmander Vol. I §8.2 Theorem 8.2.10 ^{[Hörmander Vol. I §8.2]}. $\square$

Proposition (consistency of the definition under cutoff change). If $(x_0,{\xi }_0)\notin \mathrm{WF}(u)$ with witness $({\phi }_1,V_1)$, then for every ${\phi }_2\in C_c^{\infty }(X)$ with ${\phi }_2(x_0)\ne 0$ and supported in a sufficiently small neighbourhood of $x_0$, there exists a conic neighbourhood $V_2\subseteq V_1$ of ${\xi }_0$ on which $\widehat{{\phi }_{2}u}$ is rapidly decreasing.

Proof. Write ${\phi }_2={\phi }_1\cdot ({\phi }_2/{\phi }_1)$ on a small neighbourhood of $x_0$ where ${\phi }_1$ does not vanish; the quotient is smooth there. Extend the quotient to a smooth function $\chi \in C_c^{\infty }$ supported on the neighbourhood where ${\phi }_1\ne 0$. Then ${\phi }_{2}u=\chi \cdot ({\phi }_{1}u)$ in ${\mathcal{D}}^{\prime }(X)$ on the support of ${\phi }_2$. Take Fourier transforms: $$ \widehat{\varphi_2 u}(\xi) = (2\pi)^{-n} (\hat\chi * \widehat{\varphi_1 u})(\xi) = (2\pi)^{-n} \int \hat\chi(\xi - \eta) \widehat{\varphi_1 u}(\eta) , d\eta. $$ Split ${\mathbb{R}}_{\eta }^n=V_1\cup V_1^c$. On $V_1$, $\widehat{{\phi }_{1}u}$ decays rapidly, so the integrand decays rapidly in $\eta$ uniformly in $\xi$. On $V_1^c$, $\widehat{{\phi }_{1}u}$ has polynomial growth (it is the Fourier transform of a compactly supported distribution, hence at most polynomially growing). The Schwartz factor $\hat{\chi }(\xi -\eta )$ provides rapid decay in $|\xi |$ for $\xi$ in any conic neighbourhood $V_2\subset V_1$ of ${\xi }_0$ at positive distance from $V_1^c$, since then $|\xi -\eta |\ge c|\xi |$ for $\eta \in V_1^c$ and $\xi \in V_2$ with $|\xi |$ large. Combining the two estimates gives rapid decay of $\widehat{{\phi }_{2}u}$ on $V_2$. $\square$

Proposition (wave-front set of ${\delta }_0$ on ${\mathbb{R}}^n$). $\mathrm{WF}({\delta }_0)=\{0\}\times ({\mathbb{R}}^n\setminus 0)$.

Proof. Forward inclusion: ${\hat{\delta }}_0=1$, the constant function. For any cutoff $\phi$ with $\phi (0)\ne 0$, $\widehat{\phi {\delta }_0}(\xi )=\phi (0)$, a non-zero constant, which does not decay on any conic set in ${\mathbb{R}}^n\setminus 0$. So every $(0,\xi )$ with $\xi \ne 0$ is in $\mathrm{WF}({\delta }_0)$.

Backward inclusion: for $x_0\ne 0$, pick $\phi$ supported in a neighbourhood of $x_0$ that misses the origin. Then $\phi {\delta }_0=0$, $\widehat{\phi {\delta }_0}=0$, which decays rapidly. So no $(x_0,\xi )$ with $x_0\ne 0$ is in $\mathrm{WF}({\delta }_0)$. $\square$

Proposition (wave-front set of the Heaviside step on $\mathbb{R}$). $\mathrm{WF}(H)=\{0\}\times (\mathbb{R}\setminus 0)$.

Proof. For $x_0\ne 0$, $H$ is smooth on a neighbourhood of $x_0$, so $(x_0,{\xi }_0)\notin \mathrm{WF}(H)$. For $x_0=0$: pick $\phi \in C_c^{\infty }$ with $\phi (0)\ne 0$. Then $\phi H=\phi \cdot 1_{(0,\infty )}$, a compactly supported function that jumps at $0$. Its Fourier transform is computed by partial integration: $$ \widehat{\varphi H}(\xi) = \int_0^\infty \varphi(x) e^{-i x \xi} , dx = \frac{\varphi(0)}{i\xi} + \int_0^\infty \frac{\varphi'(x)}{i\xi} e^{-i x \xi} , dx $$ for $\xi \ne 0$. The first term decays like $|\xi {|}^{-1}$ on both rays $\xi >0$ and $\xi <0$; the second is a Fourier transform of a smooth function with a jump in its derivative at $0$, which decays like $|\xi {|}^{-1}$ on both rays as well (and not rapidly). So $\widehat{\phi H}$ decays as $|\xi {|}^{-1}$ on every conic set, but not rapidly. So $(0,\xi )\in \mathrm{WF}(H)$ for every $\xi \ne 0$, in agreement with the formula. $\square$

Proposition (wave-front set of $1/(x+i0)$ on $\mathbb{R}$). $\mathrm{WF}(1/(x+i0))=\{0\}\times (0,\infty )$.

Proof. The distribution $1/(x+i0)$ is the limit in ${\mathcal{D}}^{\prime }(\mathbb{R})$ of $1/(x+i\epsilon )$ as $\epsilon \to 0^{+}$. Its Fourier transform is computed by contour integration: $$ \widehat{1/(x + i 0)}(\xi) = \lim_{\varepsilon \to 0^+} \int \frac{e^{-i x \xi}}{x + i \varepsilon} , dx = -2\pi i \cdot H(-\xi), $$ where $H$ is the Heaviside step. The integral is computed by closing the contour in the lower half-plane for $\xi >0$ (no pole, integral $=0$) and in the upper half-plane for $\xi <0$ (one pole at $x=-i\epsilon$, residue $-2\pi i$). For $x_0\ne 0$, the distribution is smooth, so $(x_0,{\xi }_0)\notin \mathrm{WF}$. For $x_0=0$: with a cutoff $\phi$ at the origin, $\widehat{\phi /(x+i0)}$ equals (modulo a Schwartz error from the cutoff) $-2\pi iH(-\xi )$ multiplied by a Schwartz function. This is identically zero for $\xi >0$ (modulo rapidly decreasing terms), hence rapidly decreasing on the positive ray; and equal to a Schwartz function times $-2\pi i$ on the negative ray, which does not go to zero. So $(0,\xi )$ is in $\mathrm{WF}$ iff $\xi >0$. $\square$

Proposition (wave-front set of a hypersurface layer). Let $\Sigma =\{x_n=0\}\subset {\mathbb{R}}^n$ with surface measure ${\delta }_{\Sigma }$. Then $\mathrm{WF}(\delta_\Sigma) = N^\Sigma \setminus 0$, the punctured conormal bundle.*

Proof. Write ${\delta }_{\Sigma }={\delta }_0(x_n)\otimes 1_{x^{\prime }}$ in coordinates $x=(x^{\prime },x_n)$. The Fourier transform factorises: ${\hat{\delta }}_{\Sigma }({\xi }^{\prime },{\xi }_n)=(2\pi {)}^{n-1}{\delta }_0({\xi }^{\prime })\cdot 1_{{\xi }_n}$. For a cutoff $\phi (x^{\prime },x_n)$ at $(x_0^{\prime },0)$, the product $\phi {\delta }_{\Sigma }$ has Fourier transform that is a Schwartz function in ${\xi }^{\prime }$ (smooth-in-$x^{\prime }$ factor smoothed by the cutoff) times $1_{{\xi }_n}$ (the constant Fourier transform of ${\delta }_0$ in $x_n$). For ${\xi }^{\prime }\ne 0$, the ${\xi }^{\prime }$-decay is rapid (Schwartz). For ${\xi }^{\prime }=0$ and ${\xi }_n\ne 0$, the ${\xi }_n$-decay is non-existent (constant $1$). So $(x^{\prime },0;0,{\xi }_n)$ is in $\mathrm{WF}({\delta }_{\Sigma })$ for every ${\xi }_n\ne 0$, and these are the only wave-front-set points. This is the conormal bundle of $\Sigma$. $\square$

Connections Master

• Hilbert space 02.11.08. The Fourier transform extends from $\mathcal{S}({\mathbb{R}}^n)$ to a unitary involution on $L^2({\mathbb{R}}^n)$ by Plancherel's theorem, and the rapid-decay characterisation of the wave-front set generalises naturally to the $L^2$-Sobolev wave-front set ${\mathrm{WF}}_{H^s}(u)$, replacing rapid decay with $H^s$-Sobolev decay. The microlocal Sobolev refinement is what makes the wave-front set quantitatively useful in elliptic regularity theory, where one wants to track gain of derivatives along non-characteristic directions.

• Smooth manifold 03.02.01. The wave-front set lives in the punctured cotangent bundle $T^{\ast }X\setminus 0$ of a smooth manifold and is invariant under cotangent lifts of diffeomorphisms. The intrinsic chart-by-chart definition relies on the smooth-manifold structure for transition functions to glue chart-local wave-front sets into a globally defined closed conic subset; the chart-independence is exactly Exercise 4 above.

• Poincaré duality 03.12.16. Both the wave-front set and Poincaré duality use partition-of-unity arguments to assemble local microlocal or local cohomological data on a manifold into a global invariant. The proof structure is parallel: reduce the global claim to a local Euclidean computation, propagate by Mayer-Vietoris or its microlocal analogue, and use a five-lemma or compactness step to conclude. The wave-front set on a closed oriented manifold becomes the right microlocal partner to the cohomological Poincaré-duality framework.

• Banach space 02.11.04. The Schwartz space $\mathcal{S}({\mathbb{R}}^n)$ of rapidly decreasing smooth functions is a Fréchet space (a complete metrisable locally convex space), not a Banach space in any natural norm — the family of seminorms $\Vert x^{\alpha }{\partial }^{\beta }u{\Vert }_{\infty }$ for all multi-indices $\alpha ,\beta$ does not reduce to a single norm. The dual ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ of tempered distributions is the natural ambient space for the wave-front-set construction on ${\mathbb{R}}^n$, and is the universe of every Fourier-analytic argument in this unit.

Historical & philosophical context Master

The wave-front set in its modern form was introduced by Lars Hörmander in "Fourier integral operators I," Acta Math. 127 (1971) 79–183 ^{[Hörmander 1971]}, with the equivalent companion paper Duistermaat-Hörmander, "Fourier integral operators II," Acta Math. 128 (1972) 183–269 ^{[Duistermaat-Hörmander 1972]} establishing propagation of singularities for operators with real principal type and the calculus of Fourier integral operators built on the wave-front set. The construction packaged earlier ideas of Sato, Kashiwara, and Kawai on the singular spectrum of a hyperfunction (Sato 1969 RIMS preprint, refined and published as Sato-Kawai-Kashiwara, Microfunctions and Pseudo-Differential Equations, Lecture Notes in Mathematics 287, Springer 1973 ^{[Sato-Kawai-Kashiwara]}) into a distributional framework, replacing the analytic-functional sheaf-theoretic apparatus of microfunctions with the more elementary Fourier-decay characterisation suited to $C^{\infty }$ distributions. The two notions agree as analytic and $C^{\infty }$ refinements of the same microlocal phenomenon, and Hörmander's Vol. I §8.4 develops both side by side.

The applications of the wave-front set to partial differential equations and to quantum field theory have followed two parallel tracks. On the PDE side, Hörmander's four-volume monograph The Analysis of Linear Partial Differential Operators (Springer 1983–85) systematised the microlocal calculus that grew out of the 1971 paper, with Vols. I and III the technical anchors for distribution theory and pseudo-differential operators respectively. On the QFT side, M. J. Radzikowski's Comm. Math. Phys. 179 (1996) 529 paper "Micro-local approach to the Hadamard condition in quantum field theory on curved space-time" ^{[Radzikowski 1996]} identified the Hadamard condition on a quantum state's two-point function with a wave-front-set condition equivalent to the older Kay-Wald 1991 asymptotic-expansion formulation (Kay-Wald, Phys. Rep. 207 (1991) 49 ^{[Kay-Wald 1991]}). Brunetti-Fredenhagen-Köhler 1996 extended the wave-front-set criterion to higher $n$-point functions in the microlocal spectrum condition, and Brunetti-Fredenhagen-Verch 2003 absorbed the framework into the locally covariant functor formulation of algebraic QFT. Christian Gérard's textbook treatment in Microlocal Analysis of Quantum Fields on Curved Spacetimes (EMS Lectures 2019) ^[Gérard] consolidates this entire programme into a single volume, with the wave-front set developed from scratch in Chs. 1–2 as the entry point.

Bibliography Master

@article{Hormander1971FIO,
  author  = {H{\"o}rmander, Lars},
  title   = {Fourier integral operators. {I}},
  journal = {Acta Math.},
  volume  = {127},
  year    = {1971},
  pages   = {79--183}
}

@article{DuistermaatHormander1972,
  author  = {Duistermaat, J. J. and H{\"o}rmander, Lars},
  title   = {Fourier integral operators. {II}},
  journal = {Acta Math.},
  volume  = {128},
  year    = {1972},
  pages   = {183--269}
}

@book{HormanderVolI,
  author    = {H{\"o}rmander, Lars},
  title     = {The Analysis of Linear Partial Differential Operators, Vol. {I}},
  publisher = {Springer-Verlag},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {256},
  year      = {1983}
}

@book{HormanderVolIII,
  author    = {H{\"o}rmander, Lars},
  title     = {The Analysis of Linear Partial Differential Operators, Vol. {III}},
  publisher = {Springer-Verlag},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {274},
  year      = {1985}
}

@book{SatoKawaiKashiwara1973,
  author    = {Sato, Mikio and Kawai, Takahiro and Kashiwara, Masaki},
  title     = {Microfunctions and Pseudo-Differential Equations},
  publisher = {Springer-Verlag},
  series    = {Lecture Notes in Mathematics},
  volume    = {287},
  year      = {1973},
  pages     = {265--529}
}

@book{FriedlanderJoshiDistributions,
  author    = {Friedlander, F. G. and Joshi, M.},
  title     = {Introduction to the Theory of Distributions},
  publisher = {Cambridge University Press},
  edition   = {2},
  year      = {1998}
}

@book{Duistermaat1996FIO,
  author    = {Duistermaat, J. J.},
  title     = {Fourier Integral Operators},
  publisher = {Birkh{\"a}user},
  year      = {1996}
}

@book{Gerard2019Microlocal,
  author    = {G{\'e}rard, Christian},
  title     = {Microlocal Analysis of Quantum Fields on Curved Spacetimes},
  publisher = {European Mathematical Society},
  series    = {ESI Lectures in Mathematics and Physics},
  year      = {2019}
}

@article{Radzikowski1996,
  author  = {Radzikowski, Marek J.},
  title   = {Micro-local approach to the {H}adamard condition in quantum field theory on curved space-time},
  journal = {Comm. Math. Phys.},
  volume  = {179},
  year    = {1996},
  pages   = {529--553}
}

@article{KayWald1991,
  author  = {Kay, Bernard S. and Wald, Robert M.},
  title   = {Theorems on the uniqueness and thermal properties of stationary, nonsingular, quasifree states on spacetimes with a bifurcate {K}illing horizon},
  journal = {Phys. Rep.},
  volume  = {207},
  year    = {1991},
  pages   = {49--136}
}

@article{Sato1969Hyperfunctions,
  author  = {Sato, Mikio},
  title   = {Hyperfunctions and partial differential equations},
  journal = {Proc. Intern. Conf. on Functional Analysis and Related Topics (Tokyo 1969)},
  year    = {1969},
  pages   = {91--94}
}