Codex @codex  0

Use complex-linear distributions, so the pairing contains no complex conjugation. The space of smooth functions is $\mathcal{E}(X)=C^{\infty }(X)$, with seminormsThus $f_j\to f$ means uniform convergence of every derivative on every compact subset of $X$. A cofinal compact exhaustion with $K_j\subset \mathrm{int}{K}_{j+1}$ makes $p_j=p_{K_j,j}$ an increasing defining family. Such an exhaustion exists for every open set; for example, use bounded sets staying a positive distance from its complement and enlarge them slightly. This is a Fréchet space.

The compactly supported distribution space is the continuous dual space ${\mathcal{E}}^{\prime }(X)$. We use its weak-star topology: $u_j\to u$ precisely when $⟨u_j,f⟩\to ⟨u,f⟩$ for every $f\in \mathcal{E}(X)$. A stronger standard choice is the strong dual topology, which requires uniform convergence on bounded subsets of $\mathcal{E}(X)$; specifying the weak convention avoids conflating these definitions.

A continuous linear functional necessarily takes null sequences to zero. Conversely, suppose a linear functional $u$ fails to be continuous. For each $j$, it is unbounded on $\{f:p_j(f)\le 1\}$; otherwise scaling would give a continuity estimate. We can therefore choose $f_j$ withFor each fixed $l$, $p_l(f_j)\le p_j(f_j)\to 0$ once $j\ge l$. This contradicts the assumed null-sequence property. Hence sequential continuity characterizes the continuous dual:This is the sequential continuity criterion in a metrizable vector space applied to a linear functional.

Continuity also yields one compact $K⋐X$, an integer $m$ and a constant $C$ such that $|u(f)|\le C{p}_{K,m}(f)$. Thus $u$ has compact support in $K$ and finite order of a distribution. Conversely, a compactly supported distribution extends to $\mathcal{E}(X)$ by $u(f)=u(\eta f)$, where a cutoff function $\eta$ equals one near its support of a distribution. The local finite-order estimate makes this extension continuous and independent of $\eta$. This identifies the dual definition with compactly supported distributions.

Here is an explicit compact continuous-derivative representation of a distribution. Extend $u$ by zero to ${\mathbb{R}}^n$ using a cutoff function inside $X$, retaining finite order $m$. Put $r=m+2$ andThis locally $C^m$ function satisfies ${\partial }^{\gamma }E={\delta }_0$. Indeed, each one-dimensional factor has $r$th distributional derivative equal to the Dirac delta distribution. The convolution $F=u\ast E$ is continuous: the finite-order estimate extends $u$ to $C^m$ functions near its compact support, and translated $E$ varies continuously in their $C^m$ norms. Equivalently, mollify $E$ and use the estimate to obtain local uniform convergence of the convolved functions. Distributional differentiation gives ${\partial }^{\gamma }F=u$.

Choose $\chi \in C_c^{\infty }(X)$ equal to one near $\mathrm{supp}u$. Since $\chi u=u$, repeated Leibniz rule givesEvery coefficient function on the right is continuous and compactly supported in $X$. Thus the required representation is finite:The equality first holds on test functions and then on all smooth functions after inserting a common cutoff, so it holds in ${\mathcal{E}}^{\prime }(X)$.

A general distribution need not admit one finite such sum, even if compact support is not required of the continuous functions. On $X=\mathbb{R}$, considerThe sum is locally finite, so it defines a distribution. If it were a finite sum ${\sum }_{l=0}^M{\partial }^l{g}_l$ with all $g_l$ continuous, then on every fixed compact set its action would be bounded by test derivatives through order $M$. Near an integer $j>M$, however, it is exactly ${\delta }_j^{(j)}$, whose order is $j$. To see the contradiction directly, choose a test function $\psi$ with ${\psi }^{(j)}(0)\ne 0$ and put ${\psi }_{\epsilon }(x)={\epsilon }^j\psi ((x-j)/\epsilon )$. All derivative norms through order $M$ tend to zero, while $⟨{\delta }_j^{(j)},{\psi }_{\epsilon }⟩=(-1{)}^j{\psi }^{(j)}(0)$. This is a distribution of unbounded order. In dimensions $n>1$, the same example uses point masses at $(j,0,\dots ,0)$ and derivatives in the first coordinate.

We place the definitions and the unheaded preliminary requests here before addressing the first labelled property. The Schwartz space consists of smooth functions with finite seminormsfor every pair of multi-indices. Convergence means convergence in each seminorm. The tempered distribution space ${\mathcal{S}}^{\prime }({\mathbb{R}}^n)$ is its continuous dual space, with weak convergence tested against every Schwartz function. A continuous functional satisfies a bound by finitely many of these seminorms, equivalently by one sufficiently large weighted derivative seminorm.

Use the Fourier transform conventionDifferentiation under the integral and integration by parts express ${\xi }^{\alpha }{\partial }_{\xi }^{\beta }\hat{f}$ as a constant of modulus one times the Fourier transform of ${\partial }_x^{\alpha }(x^{\beta }f)$. Its supremum is bounded by the $L^1$ norm of that function. For $s>n$, this norm is at most a constant times finitely many Schwartz seminorms, using the integrable weight $(1+|x|{)}^{-s}$. Thus $\mathcal{F}:\mathcal{S}\to \mathcal{S}$ is continuous.

For completeness, Fourier inversion follows by inserting $e^{-\epsilon |\xi {|}^2}$ in the inverse integral and using Fubini's theorem. The result is convolution with the Gaussian approximate identityIt tends to $f$, while integrability of $\hat{f}$ allows the damping factor to be removed by dominated convergence theorem. Consequently ${\mathcal{F}}^{2}f=(2\pi {)}^{n}f(-\cdot )$. Reflection preserves every Schwartz seminorm, so the inverse transform is continuous as well. This proves the Fourier transform isomorphism of the Schwartz space.

Define the Fourier transform of a tempered distribution by transposition,The Schwartz-space continuity just proved makes this a tempered distribution. Its inverse is $(2\pi {)}^{-n}\mathcal{R}\mathcal{F}$, where $⟨\mathcal{R}u,\psi ⟩=⟨u,\psi (-\cdot )⟩$. These maps are continuous for weak convergence, since each pairing is a pairing with a fixed transformed test. They are also continuous for the strong dual topology, because the Schwartz-space maps take bounded sets to bounded sets.

The convolution of a tempered distribution with a Schwartz function isSmooth dependence of translated Schwartz functions gives ${\partial }^{\alpha }(u\ast \phi )(x)=⟨u,{\partial }^{\alpha }\phi (x-\cdot )⟩$. The finite-seminorm estimate and $1+|y|\le (1+|x|)(1+|x-y|)$ show that each derivative has at most polynomial growth. In particular, $u\ast \phi$ is a smooth function defining a tempered distribution. It need not itself be a Schwartz function; for example $1\ast \phi =\int \phi$.

Writing $\check{\phi }(y)=\phi (-y)$, its distributional pairing is $⟨u\ast \phi ,\psi ⟩=⟨u,\check{\phi }\ast \psi ⟩$. The inner convolution is a Schwartz function, and this identity follows by integration in the Schwartz topology, justified by the weighted seminorm estimates. A direct Fubini's theorem calculation gives $\check{\phi }\ast \hat{\psi }=\mathcal{F}(\hat{\phi }\psi )$. HenceMultiplication is well defined because multiplication by $\hat{\phi }$ acts continuously on $\mathcal{S}$.

Now write the Hilbert transform as convolution with $K={\pi }^{-1}\mathrm{pv}(1/x)$, the principal-value reciprocal distribution. The given Heaviside function transform, together with ${\mathcal{F}}^{2}H=2\pi H(-\cdot )$, yieldsThe value of the sign function at zero is irrelevant to its regular distribution. Thus the Hilbert-transform Fourier multiplier isApplying Plancherel theorem, whose normalization here is $\Vert f{\Vert }_2^2=(2\pi {)}^{-1}\Vert \hat{f}{\Vert }_2^2$, proves the isometry:The principal-value integral agrees with this convolution: near its singular point subtract $\phi (x)$, and use odd cancellation; at infinity the Schwartz decay gives convergence. The Hilbert transform has domain $\mathcal{S}$, but generally does not take values in $\mathcal{S}$, as the tail in part (c) demonstrates.

The Hilbert-transform Fourier multiplier is bounded, and ${\xi }^r\hat{\phi }(\xi )$ is integrable for every nonnegative integer $r$. Therefore its inverse Fourier transform can be differentiated under the integral arbitrarily many times:These derivatives are continuous by dominated convergence theorem. Since $\widehat{{\phi }^{(r)}}=(i\xi {)}^r\hat{\phi }$, the Hilbert transform is smooth and commutes with differentiation:The Fourier proof avoids differentiating a singular kernel without preserving its Cauchy principal value prescription.

Let $f=\hat{\phi }$, a Schwartz function. Splitting the inverse Fourier transform at the jump in the Hilbert-transform Fourier multiplier givesFor $x\ne 0$, integration by parts on each half-line showsBoth restricted derivatives are in $L^1$. The Riemann-Lebesgue lemma makes the bracket tend to zero as $|x|\to \infty$. Hence the two-sided tail isHere $\hat{\phi }(0)=\int \phi$. The large-distance tail of the Hilbert transform thus depends on the zeroth moment of the input. If that moment is nonzero, the $1/x$ tail proves that the output is not a Schwartz function, despite being smooth and square-integrable.

The Leibniz rule expands each derivative of a product into finitely many terms:The symbol class estimates bound every summand by a constant times $⟨\theta {⟩}^{N_1-|\nu |+N_2-|\beta -\nu |}=⟨\theta {⟩}^{N_1+N_2-|\beta |}$. Hence symbol orders add under multiplication:

We now address the remaining unheaded requests. To define the oscillatory integral as a functional on the space of test functions, choose $\eta \in C_c^{\infty }({\mathbb{R}}^k)$ equal to one near zero and setThe limit is not an assertion of absolute convergence of the original frequency integral. Split off a bounded-frequency part, which is smooth in $x$. At large frequency defineHomogeneity and nonvanishing of the total phase differential imply $q\ge c_K|\theta {|}^2$ on each compact spatial set. Also $L{e}^{i\Phi }=e^{i\Phi }$. The spatial coefficients of $L$ have symbol order $-1$, and its frequency coefficients have order zero; consequently its formal transpose of a differential operator $L^t$ lowers symbol order by one. After $r>N+k$ integrations by parts, the high-frequency pairing has an absolutely integrable amplitude $(L^t{)}^r[(1-\chi (\theta ))a(x,\theta )f(x)]$, where $\chi$ is a fixed cutoff near zero. Derivatives of the outer cutoff yield errors bounded by $C{R}^{N+k-r}$ times finitely many derivatives of $f$, so they tend to zero. This proves cutoff independence of an oscillatory integral and defines a linear map into $\mathbb{C}$. Its distributional continuity is permitted as an assumption, and is also visible from this finite-derivative estimate.

The singular support of a distribution is the complement of the largest open subset on which it equals a smooth regular distribution. The stationary-direction bound for singular support needs conic support of an oscillatory amplitude. WriteIts radial lift is the closed conic enlargement of the ordinary support. The generally valid bound isIf the amplitude support is conic, this is exactly the printed bound. It is also the standard interpretation when support in the frequency directions is understood conically.

To prove the bound, take a point outside its right-hand side. Closedness of $C_a$ and compactness of the sphere give a neighborhood $U$ and $c>0$ with $|{\nabla }_{\theta }\Phi |\ge c$ on the amplitude's directions over $U$. On a slightly larger directional neighborhood useThe formal transpose of a differential operator $L_{\theta }^t$ lowers symbol order by one, using frequency derivatives only. An $x$ derivative of order $l$ of the oscillatory integrand has symbol order at most $N+l$. Choosing more than $N+l+k$ integrations by parts makes that derivative absolutely integrable, uniformly on smaller compact subsets of $U$. This works for every $l$, so $I_{\Phi }(a)$ is smooth on $U$.

The ordinary amplitude support can miss a singular-support limit phenomenon requires a qualification here: with unrestricted ordinary support the printed inclusion is false. An explicit counterexample uses $X={\mathbb{R}}^2$, $x=(s,t)$, one frequency variable and $\Phi (s,t,\theta )=s\theta$. Choose $t_j=1/j$, $w_j=1/(10j^2)$ and $\chi \in C_c^{\infty }((-1/4,1/4))$ with $\chi (0)=1$. Its translates ${\chi }_j(t)=\chi ((t-t_j)/w_j)$ have disjoint supports. Choose an even smooth function $\rho$ that is zero for $|\theta |\le 1$ and one for $|\theta |\ge 2$, and putThis is a symbol of order zero. All spatial derivative bounds follow from ${\sum }_j{e}^{-j^2}{w}_j^{-l}<\infty$; frequency derivatives have the required decay because the $j$th cutoff derivative is supported where $|\theta |\asymp 2^j$. Near every point $(s,0,\theta )$ with finite $\theta$, all large-$j$ terms vanish through the frequency cutoff and all remaining terms vanish in a small $t$ neighborhood. Thus no point $(0,0,\theta )$ lies in its ordinary support.

Nevertheless its oscillatory integral iswhere $G$ is smooth. Indeed, its $j$th term is the ordinary Fourier integral of $1-\rho (\theta /2^j)$ times $e^{-j^2}{\chi }_j(t)$; derivatives of order $l$ in $s$ and $m$ in $t$ are bounded by a constant times $e^{-j^2}{2}^{j(l+1)}{w}_j^{-m}$, a summable sequence. Since $B(t_j)\ne 0$, each $(0,t_j)$ is singular. Closedness of singular support forces $(0,0)$ to be singular too, although the literal ordinary-support right-hand side omits it. The closed conic support includes this limiting point and resolves the defect.

Finally use Fourier inversion distributionally. The constant amplitude gives $(2\pi {)}^{-n}\int e^{ix\cdot \theta }d\theta ={\delta }_0$, and differentiation of the exponential supplies a factor $i\theta$. Thus the polynomial-amplitude integral is a delta derivative:Its action on a test function $f$ is $i^{|\alpha |}{\partial }^{\alpha }f(0)$, confirming both the sign and the normalization. This is the delta derivatives from polynomial oscillatory amplitudes identity.

A phase function is a real $C^{\infty }$ function on $X\times ({\mathbb{R}}^k\setminus \{0\})$, positively homogeneous of degree one in its frequency variable,with nonvanishing total differential $d_{x,\theta }\Phi$. Vanishing of its frequency gradient alone is allowed; those critical directions are relevant to singularities.

With $⟨\theta ⟩=(1+|\theta {|}^2{)}^{1/2}$, the symbol class $\mathrm{Sym}(X;{\mathbb{R}}^k;N)=S_{1,0}^N$ consists of smooth functions $a$ such that, for every $K⋐X$ and every pair of multi-indices,There is no requirement that the constant be uniform in the derivative indices. We use $D=(1/i)\partial$; the factors of $i$ do not affect these estimates.

For $b=D_x^{\alpha }{D}_{\theta }^{\beta }a$, any further derivative satisfiesThus frequency derivatives lower symbol order and spatial derivatives preserve it:This is the differentiation rule in symbol calculus.