AN-04 proof edition · CC0; exact source credit and separately licensed prerequisites

Resolved corner kernels and their exact inverse

These connected components retain AN03-U032, Totally characteristic operators on the half space, Sections 2–5, Section 6 through Example 6.7, and Sections 7–13. Original author: Claude Opus 5.5 (Anthropic), September 2026; editorial additions: Codex, September 2026. Both were dedicated to the public domain (CC0). Current prerequisite connections and proof clarifications: AN-04 course-writing task and OpenAI Codex, 5 October 2026, also CC0. The selected components retain every mathematical display, the full scalar and finite-matrix hypotheses, and all five original solved exercises.

The approved mathematical antecedent is Hörmander III, 2007 eBook, ISBN 978-3-540-49938-1, Section 18.3. Its use and ordinary citation are valid. Complete proofs are supplied in the components and the exact earlier programme proofs. The earlier linked components retain their individual licences.

The four components, in proof order, are Boundary tests, lacunary symbols and all normal jets, Resolved corner kernels and their exact inverse, Boundary adjoints, complete composition and distributional action, Boundary operator bounds, conormal action and the residual obstruction. Original section and equation numbers are retained across them. Sections 6.8 (polyhomogeneous corner characterization) and 14 (arbitrary positive-order Sobolev loss) are separate unadopted obligations; the theorems below do not substitute for those results or for the global compressed wave-front calculus.

The symbol class, Fourier-kernel construction, conventions and exact prerequisites are in Sections 1–5. This component proves all uniform estimates at the resolved corner, their converse and the finite negative-order Schur kernel bound.

6. Kernels near the corner

Coordinates at the corner

Kernels of totally characteristic operators live on

Q={(x,y)∈R2n:xn≥0, yn≥0},∂2Q={(x,y):xn=yn=0}, Q=\{(x,y)\in\mathbb R^{2n}:x_n\geq0,\ y_n\geq0\},\qquad \partial_2Q=\{(x,y):x_n=y_n=0\},

the quarter space and its distinguished boundary. Near ∂2Q\partial_2Q we use

t=xn+yn2,r=xn−ynt(t>0);xn=t(1+r2),yn=t(1−r2).(6.1) t=\frac{x_n+y_n}2,\qquad r=\frac{x_n-y_n}t\quad(t>0);\qquad x_n=t\Big(1+\frac r2\Big),\quad y_n=t\Big(1-\frac r2\Big). \tag{6.1}

Proposition 6.1 (Blow-up coordinates). Write Φ(t,r)=(t(1+r/2),t(1−r/2))\Phi(t,r)=(t(1+r/2),t(1-r/2)).

  1. Φ\Phi maps (0,∞)×R(0,\infty)\times\mathbb R diffeomorphically onto {xn+yn>0}\{x_n+y_n>0\}, with ∣det⁡Φ′∣=t|\det\Phi'|=t; hence dxn dyn=t dt drdx_n\,dy_n=t\,dt\,dr.
  2. Q∖{xn=yn=0}Q\setminus\{x_n=y_n=0\} corresponds to t>0t>0, ∣r∣≤2|r|\leq2. The face {xn=0<yn}\{x_n=0<y_n\} is r=−2r=-2, the face {yn=0<xn}\{y_n=0<x_n\} is r=2r=2, the diagonal xn=ynx_n=y_n is r=0r=0, and yn/xn=(2−r)/(2+r)y_n/x_n=(2-r)/(2+r).
  3. Φ\Phi extends smoothly to [0,∞)×R[0,\infty)\times\mathbb R and maps the whole line t=0t=0 to the corner: the corner is blown up into the front face t=0t=0, of which the segment ∣r∣≤2|r|\leq2 lies over QQ.
  4. Normal dilations (xn,yn)↦λ(xn,yn)(x_n,y_n)\mapsto\lambda(x_n,y_n) are (t,r)↦(λt,r)(t,r)\mapsto(\lambda t,r), and the radial field is xn∂xn+yn∂yn=t∂tx_n\partial_{x_n}+y_n\partial_{y_n}=t\partial_t.
  5. On QQ, ∣w∣/2≤t≤∣w∣|w|/2\leq t\leq|w| for w=(xn,yn)w=(x_n,y_n); rr is homogeneous of degree 0, so ∣∂wβr∣≤Cβ∣w∣−∣β∣|\partial_w^\beta r|\leq C_\beta|w|^{-|\beta|} on {xn+yn>0}∩{∣r∣≤3}\{x_n+y_n>0\}\cap\{|r|\leq3\}.
  6. The rescaled normal variable in (4.4) is a function of rr alone: (xn−yn)/xn=2r/(2+r)(x_n-y_n)/x_n=2r/(2+r).

Proof. The Jacobian matrix of Φ\Phi has rows (1+r2,t2)(1+\tfrac r2,\tfrac t2) and (1−r2,−t2)(1-\tfrac r2,-\tfrac t2), with determinant −t-t; the inverse is (6.1). Parts 2–4 and 6 are direct substitutions. For part 5, xn+yn≥∣w∣x_n+y_n\geq|w| when both are nonnegative, and xn+yn≤2∣w∣x_n+y_n\leq\sqrt2|w|. The set {∣r∣≤3}={∣xn−yn∣≤32(xn+yn)}\{|r|\leq3\}=\{|x_n-y_n|\leq\tfrac32(x_n+y_n)\} is a closed cone that meets the line xn+yn=0x_n+y_n=0 only at the origin; its intersection with the unit circle is a compact subset of the open set where rr is smooth. The derivatives of order kk of rr are homogeneous of degree −k-k, so they are bounded by Ck∣w∣−kC_k|w|^{-k} on that cone. □\square

The point of these coordinates is part 6. The kernel formula (4.4) involves xn−1x_n^{-1} and a function of (xn−yn)/xn(x_n-y_n)/x_n, and neither is smooth at the corner. But tKtK becomes a smooth function of (t,r)(t,r), as Theorem 6.2 shows.

Residual kernels

Theorem 6.2 (Residual kernels).

(a) Let a∈Sla−∞a\in S^{-\infty}_{\mathrm{la}} and A(x,z)=(2π)−n∫eiz⋅ξa(x,ξ) dξA(x,z)=(2\pi)^{-n}\int e^{iz\cdot\xi}a(x,\xi)\,d\xi. Then A∈C∞(R‾+n×Rn)A\in C^\infty(\overline{\mathbb R}{}^n_+\times\mathbb R^n), A=0A=0 for zn≥1z_n\geq1, and for all α,β,N\alpha,\beta,N

∣∂xα∂zβA(x,z)∣≤CαβN(1+∣z∣)−N(1+xn)−N,(6.2) |\partial_x^\alpha\partial_z^\beta A(x,z)|\leq C_{\alpha\beta N}(1+|z|)^{-N}(1+x_n)^{-N}, \tag{6.2}

with CαβNC_{\alpha\beta N} bounded by seminorms of aa. Conversely every A∈C∞(R‾+n×Rn)A\in C^\infty(\overline{\mathbb R}{}^n_+\times\mathbb R^n) satisfying (6.2) and vanishing for zn>1z_n>1 comes in this way from exactly one a∈Sla−∞a\in S^{-\infty}_{\mathrm{la}}, namely a(x,ξ)=∫e−iz⋅ξA(x,z) dza(x,\xi)=\int e^{-iz\cdot\xi}A(x,z)\,dz.

(b) For a∈Sla−∞a\in S^{-\infty}_{\mathrm{la}} the kernel of TaT_a is the locally integrable function

K(x,y)=xn−1A(x, x′−y′, xn−ynxn)(xn>0),K(x,y)=0(xn<0).(6.3) K(x,y)=x_n^{-1}A\Big(x,\,x'-y',\,\frac{x_n-y_n}{x_n}\Big)\quad(x_n>0),\qquad K(x,y)=0\quad(x_n<0). \tag{6.3}

For xn>0x_n>0 and u∈Su\in\mathcal S, Tau(x)=∫K(x,y)u(y) dyT_au(x)=\int K(x,y)u(y)\,dy, and ∫∣K(x,y)∣ dy=∫∣A(x,z)∣ dz≤C(1+xn)−N\int|K(x,y)|\,dy=\int|A(x,z)|\,dz\leq C(1+x_n)^{-N}. Moreover supp⁡K⊂Q\operatorname{supp}K\subset Q and K∈C∞(R2n∖∂2Q)K\in C^\infty(\mathbb R^{2n}\setminus\partial_2Q).

(c) The function F(x′,y′,t,r)=t K(x′,t(1+r2),y′,t(1−r2))F(x',y',t,r)=t\,K(x',t(1+\tfrac r2),y',t(1-\tfrac r2)), t>0t>0, extends to a C∞C^\infty function on {t≥0}×Rx′n−1×Ry′n−1×Rr\{t\geq0\}\times\mathbb R^{n-1}_{x'}\times\mathbb R^{n-1}_{y'}\times\mathbb R_r, which vanishes for ∣r∣≥2|r|\geq2. For all α,β,τ,ρ,ν\alpha,\beta,\tau,\rho,\nu and r>−2r>-2,

∣Dx′αDy′βDtτDrρF∣≤C(1+∣x′−y′∣+t)−ν(2+r)ν,(6.4) |D^\alpha_{x'}D^\beta_{y'}D^\tau_tD^\rho_rF|\leq C(1+|x'-y'|+t)^{-\nu}(2+r)^{\nu}, \tag{6.4}

and in particular

∣Dx′αDy′βDtτDrρF∣≤C′(1+∣x′−y′∣+t)−ν.(6.5) |D^\alpha_{x'}D^\beta_{y'}D^\tau_tD^\rho_rF|\leq C'(1+|x'-y'|+t)^{-\nu}. \tag{6.5}

On the front face,

F(x′,y′,0,r)=22+r A(x′,0, x′−y′, 2r2+r)(r>−2).(6.6) F(x',y',0,r)=\frac{2}{2+r}\,A\Big(x',0,\,x'-y',\,\frac{2r}{2+r}\Big)\quad(r>-2). \tag{6.6}

(d) Conversely, let K∈Lloc1(R2n)K\in L^1_{\mathrm{loc}}(\mathbb R^{2n}) with supp⁡K⊂Q\operatorname{supp}K\subset Q, and suppose that the function FF of (c) agrees almost everywhere on t>0t>0 with a function in C∞({t≥0})C^\infty(\{t\geq0\}) that vanishes for ∣r∣≥2|r|\geq2 and satisfies (6.5). Then KK is, almost everywhere, the kernel of TaT_a for exactly one a∈Sla−∞a\in S^{-\infty}_{\mathrm{la}}.

Proof. (a) For a∈S+−∞a\in S^{-\infty}_+, integration by parts gives zγ∂zβ∂xαA=(2π)−n∫eiz⋅ξ(−Dξ)γ[(iξ)β∂xαa] dξz^\gamma\partial^\beta_z\partial^\alpha_xA=(2\pi)^{-n}\int e^{iz\cdot\xi}(-D_\xi)^\gamma[(i\xi)^\beta\partial^\alpha_xa]\,d\xi, with an integrand bounded by C(1+∣ξ∣)−n−1(1+xn)−NC(1+|\xi|)^{-n-1}(1+x_n)^{-N}. This proves smoothness and (6.2). For fixed (x,ξ′)(x,\xi'), a(x,ξ′,⋅)∈S(R)a(x,\xi',\cdot)\in\mathcal S(\mathbb R), so Fna\mathcal F_na is a continuous function; by (4.5) it vanishes for t<−1t<-1. Since A(x,z)=(2π)−n∫eiz′⋅ξ′(Fna)(x,ξ′,−zn) dξ′A(x,z)=(2\pi)^{-n}\int e^{iz'\cdot\xi'}(\mathcal F_na)(x,\xi',-z_n)\,d\xi', A=0A=0 for zn>1z_n>1, and by continuity for zn≥1z_n\geq1. Conversely, if AA satisfies (6.2), then ξγ∂ξβ∂xαa=∫e−iz⋅ξDzγ[(−iz)β∂xαA] dz\xi^\gamma\partial_\xi^\beta\partial_x^\alpha a=\int e^{-iz\cdot\xi}D_z^\gamma[(-iz)^\beta\partial_x^\alpha A]\,dz is bounded by C(1+xn)−NC(1+x_n)^{-N}, so a∈S+−∞a\in S^{-\infty}_+; Fourier inversion recovers AA from aa; and Fna(x,ξ′,t)=2π∫e−iz′⋅ξ′A(x,z′,−t) dz′\mathcal F_na(x,\xi',t)=2\pi\int e^{-iz'\cdot\xi'}A(x,z',-t)\,dz' vanishes for t<−1t<-1. Uniqueness is Fourier inversion.

(b) For xn>0x_n>0 fixed, a♭(x,⋅)∈S(Rn)a^\flat(x,\cdot)\in\mathcal S(\mathbb R^n), because aa decreases rapidly in (ξ′,xnξn)(\xi',x_n\xi_n). So K(x,y)=(2π)−n∫ei(x−y)⋅ξa♭(x,ξ)dξK(x,y)=(2\pi)^{-n}\int e^{i(x-y)\cdot\xi}a^\flat(x,\xi)d\xi converges absolutely, and the substitution ηn=xnξn\eta_n=x_n\xi_n gives (6.3). Fubini gives Tau(x)=∫K(x,y)u(y)dyT_au(x)=\int K(x,y)u(y)dy. The change of variables z=(x′−y′,(xn−yn)/xn)z=(x'-y',(x_n-y_n)/x_n), dy=xn dzdy=x_n\,dz, gives ∫∣K(x,y)∣dy=∫∣A(x,z)∣dz\int|K(x,y)|dy=\int|A(x,z)|dz. Hence (Tau,v)=∬K(x,y)u(y)v(x)‾ dy dx(T_au,v)=\iint K(x,y)u(y)\overline{v(x)}\,dy\,dx for u,v∈Su,v\in\mathcal S, with absolute convergence; so KK, which is locally integrable, is the Schwartz kernel. If xn<0x_n<0, K=0K=0. If xn>0>ynx_n>0>y_n, then (xn−yn)/xn>1(x_n-y_n)/x_n>1 and A=0A=0. So KK vanishes outside QQ, up to the null set xn=0x_n=0.

Smoothness off ∂2Q\partial_2Q. Near a point with xn>0x_n>0, (6.3) is smooth. Near a point with xn<0x_n<0, or with xn=0x_n=0 and yn<0y_n<0, K=0K=0. Let xn0=0<yn0x^0_n=0<y^0_n. For xn>0x_n>0 small and yny_n near yn0y^0_n, the normal argument zn=(xn−yn)/xnz_n=(x_n-y_n)/x_n satisfies ∣zn∣≥yn0/(2xn)|z_n|\geq y^0_n/(2x_n). By the chain rule, a derivative of order ∣γ∣|\gamma| of KK is a finite sum of terms xn−kynl(∂A)(x,z)x_n^{-k}y_n^l(\partial A)(x,z) with k≤1+2∣γ∣k\leq1+2|\gamma|, l≤∣γ∣l\leq|\gamma|, and ∣∂A∣≤CM(1+∣zn∣)−M≤CM(2xn/yn0)M|\partial A|\leq C_M(1+|z_n|)^{-M}\leq C_M(2x_n/y^0_n)^M. So KK and all its derivatives tend to 0 as xn→0+x_n\to0+, uniformly near the point. Since K=0K=0 for xn≤0x_n\leq0, KK is smooth there.

(c) For t>0t>0 and r>−2r>-2 we have xn=t(1+r/2)>0x_n=t(1+r/2)>0, (xn−yn)/xn=2r/(2+r)(x_n-y_n)/x_n=2r/(2+r) and t/xn=2/(2+r)t/x_n=2/(2+r). So by (6.3)

F=22+r A(x′, t(1+r2), x′−y′, 2r2+r)=:G.(6.7) F=\frac2{2+r}\,A\Big(x',\,t\big(1+\tfrac r2\big),\,x'-y',\,\frac{2r}{2+r}\Big)=:G . \tag{6.7}

The right side is smooth on {t≥0, r>−2}\{t\geq0,\ r>-2\}, because AA is smooth up to xn=0x_n=0. It vanishes for r≥2r\geq2, where 2r/(2+r)≥12r/(2+r)\geq1. For t>0t>0, r<−2r<-2 we have xn<0x_n<0 and F=0F=0.

Now let −2<r<2-2<r<2 and write zn=2r/(2+r)z_n=2r/(2+r), z′=x′−y′z'=x'-y', xn=t(2+r)/2x_n=t(2+r)/2. Then

12+r≤1+∣zn∣2,t=2xn2+r≤xn(1+∣zn∣).(6.8) \frac1{2+r}\leq\frac{1+|z_n|}2,\qquad t=\frac{2x_n}{2+r}\leq x_n(1+|z_n|). \tag{6.8}

The first inequality holds because 1+∣zn∣≥1≥2/(2+r)1+|z_n|\geq1\geq2/(2+r) when r≥0r\geq0, and 1+∣zn∣=(2−r)/(2+r)≥2/(2+r)1+|z_n|=(2-r)/(2+r)\geq2/(2+r) when r<0r<0. By the chain rule, Dx′αDy′βDtτDrρGD^\alpha_{x'}D^\beta_{y'}D^\tau_tD^\rho_rG is a finite sum of terms c tk(2+r)−l(∂xμ∂zκA)(x′,xn,z′,zn)c\,t^k(2+r)^{-l}(\partial_x^\mu\partial_z^\kappa A)(x',x_n,z',z_n) with 0≤k≤ρ0\leq k\leq\rho and l≤1+2ρl\leq1+2\rho: an rr-derivative may hit (2+r)−l(2+r)^{-l}, the argument t(1+r/2)t(1+r/2) (factor t/2t/2) or the argument 2r/(2+r)2r/(2+r) (factor 4/(2+r)24/(2+r)^2); a tt-derivative brings the factor (2+r)/2(2+r)/2. By (6.8) and (6.2), each term is at most

C xnk(1+∣zn∣)k+l+(1+∣z∣)−M(1+xn)−M. C\,x_n^k(1+|z_n|)^{k+l_+}(1+|z|)^{-M}(1+x_n)^{-M}.

On the other hand 1+∣x′−y′∣+t≤(1+∣z′∣)(1+xn)(1+∣zn∣)≤(1+∣z∣)2(1+xn)1+|x'-y'|+t\leq(1+|z'|)(1+x_n)(1+|z_n|)\leq(1+|z|)^2(1+x_n) and (2+r)−ν≤(1+∣z∣)ν(2+r)^{-\nu}\leq(1+|z|)^\nu. Choosing MM large gives (6.4) for −2<r<2-2<r<2; for r≥2r\geq2 the left side vanishes. In particular every derivative of GG tends to 0 as r↓−2r\downarrow-2, locally uniformly (also at t=0t=0); so GG, extended by 0 to r≤−2r\leq-2, is smooth, and it equals FF. Since 2+r≤42+r\leq4 on the support, (6.4) implies (6.5). Formula (6.6) is (6.7) at t=0t=0.

(d) Taylor's formula at r=−2r=-2, where FF vanishes to infinite order, turns (6.5) into (6.4): ∣∂rρF(r)∣≤sup⁡[−2,r]∣∂rρ+νF∣ (r+2)ν/ν!|\partial_r^\rho F(r)|\leq\sup_{[-2,r]}|\partial_r^{\rho+\nu}F|\,(r+2)^\nu/\nu!. Define, for x∈R‾+nx\in\overline{\mathbb R}{}^n_+,

A(x,z)=22−zn F(x′, x′−z′, xn(2−zn)2, 2zn2−zn)(zn<2),A(x,z)=0(zn>1).(6.9) A(x,z)=\frac2{2-z_n}\,F\Big(x',\,x'-z',\,\frac{x_n(2-z_n)}2,\,\frac{2z_n}{2-z_n}\Big)\quad(z_n<2),\qquad A(x,z)=0\quad(z_n>1). \tag{6.9}

On 1<zn<21<z_n<2 both definitions give 0, because then 2zn/(2−zn)>22z_n/(2-z_n)>2; so AA is smooth. For zn≤1z_n\leq1 put r=2zn/(2−zn)∈(−2,2]r=2z_n/(2-z_n)\in(-2,2] and t=xn(2−zn)/2t=x_n(2-z_n)/2. Then r+2=4/(2−zn)≤8/(1+∣zn∣)r+2=4/(2-z_n)\leq8/(1+|z_n|) and xn/2≤t≤xn(1+∣zn∣)x_n/2\leq t\leq x_n(1+|z_n|). Every derivative of AA is a finite sum of terms (polynomial in xnx_n) ×\times (smooth function of znz_n growing at most polynomially on zn≤1z_n\leq1) ×\times (a derivative of FF at the displayed point). By (6.4) such a term is at most C(1+xn)k(1+∣zn∣)k(1+∣z′∣+xn/2)−ν(1+∣zn∣)−νC(1+x_n)^{k}(1+|z_n|)^{k}(1+|z'|+x_n/2)^{-\nu}(1+|z_n|)^{-\nu}, and choosing ν\nu large gives (6.2). By (a), AA comes from a unique a∈Sla−∞a\in S^{-\infty}_{\mathrm{la}}. Finally, for xn>0<ynx_n>0<y_n, inserting z=(x′−y′,(xn−yn)/xn)z=(x'-y',(x_n-y_n)/x_n) into (6.9) gives 2/(2−zn)=xn/t2/(2-z_n)=x_n/t, xn(2−zn)/2=tx_n(2-z_n)/2=t and 2zn/(2−zn)=r2z_n/(2-z_n)=r, so the kernel (6.3) of TaT_a equals F/t=KF/t=K almost everywhere; for yn<0y_n<0 both vanish. □\square

Remark 6.3 (Singular kernels of order −∞-\infty). For an ordinary pseudodifferential operator of order −∞-\infty the kernel is smooth. Here, by (c), K=F/tK=F/t near the corner, and the leading part F(x′,y′,0,r)/tF(x',y',0,r)/t is homogeneous of degree −1-1 in (xn,yn)(x_n,y_n). It is not smooth, and not even bounded, unless FF vanishes on the front face. This singularity is what makes residual operators fail to improve regularity in Section 12.

A kernel bound at finite negative order

Proposition 6.4 (A kernel bound). Let a∈Sla−n−2a\in S^{-n-2}_{\mathrm{la}}. Then the kernel of TaT_a is a function, and

∣Ka(x,y)∣≤C (1+∣x′−y′∣)−n xnyn(xn+yn)3(xn,yn>0),Ka=0 elsewhere,(6.10) |K_a(x,y)|\leq C\,(1+|x'-y'|)^{-n}\,\frac{x_ny_n}{(x_n+y_n)^3}\quad(x_n,y_n>0),\qquad K_a=0\ \text{elsewhere}, \tag{6.10}

with CC bounded by a seminorm of aa. Consequently sup⁡x∫∣Ka(x,y)∣dy\sup_x\int|K_a(x,y)|dy and sup⁡y∫∣Ka(x,y)∣dx\sup_y\int|K_a(x,y)|dx are at most 12C∫Rn−1(1+∣z′∣)−ndz′\tfrac12C\int_{\mathbb R^{n-1}}(1+|z'|)^{-n}dz'.

Proof. For xn>0x_n>0, ∣a♭(x,ξ)∣≤C(1+∣(ξ′,xnξn)∣)−n−2|a^\flat(x,\xi)|\leq C(1+|(\xi',x_n\xi_n)|)^{-n-2} is integrable in ξ\xi, so Ka(x,y)=(2π)−n∫ei(x−y)⋅ξa♭dξK_a(x,y)=(2\pi)^{-n}\int e^{i(x-y)\cdot\xi}a^\flat d\xi converges absolutely, Tau(x)=∫Ka(x,y)u(y)dyT_au(x)=\int K_a(x,y)u(y)dy, and (6.3) holds with the bounded continuous function A(x,z)=(2π)−n∫eiz⋅ξa(x,ξ)dξA(x,z)=(2\pi)^{-n}\int e^{iz\cdot\xi}a(x,\xi)d\xi. As in Theorem 6.2(a), Fna\mathcal F_na is now a continuous function, so A=0A=0 for zn≥1z_n\geq1. Integration by parts gives ∣zγA∣≤C|z^\gamma A|\leq C for ∣γ∣≤n+2|\gamma|\leq n+2, because ∣∂ξγa∣≤C(1+∣ξ∣)−n−2−∣γ∣|\partial_\xi^\gamma a|\leq C(1+|\xi|)^{-n-2-|\gamma|} is integrable; so (1+∣z′∣)n(1+∣zn∣)2∣A∣≤C(1+|z'|)^n(1+|z_n|)^2|A|\leq C. Likewise ∂znA=(2π)−n∫eiz⋅ξiξna dξ\partial_{z_n}A=(2\pi)^{-n}\int e^{iz\cdot\xi}i\xi_na\,d\xi and ∣z′γ∂znA∣≤C|z'^\gamma\partial_{z_n}A|\leq C for ∣γ∣≤n|\gamma|\leq n. Since A(x,z′,1)=0A(x,z',1)=0, the mean value theorem gives ∣A(x,z)∣≤C(1+∣z′∣)−n∣1−zn∣|A(x,z)|\leq C(1+|z'|)^{-n}|1-z_n| for zn≤1z_n\leq1.

If 0<xn≤yn0<x_n\leq y_n, then zn=(xn−yn)/xn≤0z_n=(x_n-y_n)/x_n\leq0, 1+∣zn∣=yn/xn1+|z_n|=y_n/x_n, and ∣Ka∣≤xn−1C(1+∣z′∣)−n(xn/yn)2=C(1+∣z′∣)−nxn/yn2≤8C(1+∣z′∣)−nxnyn/(xn+yn)3|K_a|\leq x_n^{-1}C(1+|z'|)^{-n}(x_n/y_n)^2=C(1+|z'|)^{-n}x_n/y_n^2\leq8C(1+|z'|)^{-n}x_ny_n/(x_n+y_n)^3, because xn+yn≤2ynx_n+y_n\leq2y_n. If 0<yn<xn0<y_n<x_n, then ∣1−zn∣=yn/xn|1-z_n|=y_n/x_n and ∣Ka∣≤C(1+∣z′∣)−nyn/xn2≤8C(1+∣z′∣)−nxnyn/(xn+yn)3|K_a|\leq C(1+|z'|)^{-n}y_n/x_n^2\leq8C(1+|z'|)^{-n}x_ny_n/(x_n+y_n)^3. For the marginals, ∫Rn−1(1+∣z′∣)−ndz′<∞\int_{\mathbb R^{n-1}}(1+|z'|)^{-n}dz'<\infty and ∫0∞xnyn(xn+yn)−3dyn=∫0∞s(1+s)−3ds=12\int_0^\infty x_ny_n(x_n+y_n)^{-3}dy_n=\int_0^\infty s(1+s)^{-3}ds=\tfrac12; the bound is symmetric in xn,ynx_n,y_n. □\square

The two cases correspond to the two sides of the diagonal: for yn≥xny_n\geq x_n the decay of AA in znz_n is used, for yn<xny_n<x_n the vanishing of AA at zn=1z_n=1, that is, lacunarity.

Examples of residual kernels

Example 6.5 (Without lacunarity the operator sees below the boundary). Let 0≤h∈C0∞(Rn)0\leq h\in C_0^\infty(\mathbb R^n) be supported near (0,32)(0,\tfrac32), with h(0,32)>0h(0,\tfrac32)>0, and a=e−xnh^(ξ)∈S+−∞a=e^{-x_n}\widehat h(\xi)\in S^{-\infty}_+. Then A=e−xnhA=e^{-x_n}h does not vanish on zn>1z_n>1, so aa is not lacunary. For 0≤u∈C0∞(R−n)0\leq u\in C_0^\infty(\mathbb R^n_-) supported near (0,−12)(0,-\tfrac12), with u(0,−12)>0u(0,-\tfrac12)>0, formula (6.3), whose derivation for xn>0x_n>0 in Theorem 6.2(b) does not use lacunarity, gives Tau(0,1)=e−1∫h(−y′,1−yn)u(y) dy>0T_au(0,1)=e^{-1}\int h(-y',1-y_n)u(y)\,dy>0, since 1−yn1-y_n is near 32\tfrac32 there. So Tau≠0T_au\neq0 in R+n\mathbb R^n_+ although u=0u=0 in R+n\mathbb R^n_+: lacunarity cannot be dropped from Theorem 5.1(a), in accordance with Proposition 4.3.

Example 6.6 (A residual kernel in one dimension). Let n=1n=1, 0≠h∈C0∞((−12,12))0\neq h\in C_0^\infty((-\tfrac12,\tfrac12)) and a(x,ξ)=e−xh^(ξ)a(x,\xi)=e^{-x}\widehat h(\xi). Then K(x,y)=e−xx−1h((x−y)/x)K(x,y)=e^{-x}x^{-1}h((x-y)/x) for x>0x>0, and

Tau(x)=e−x∫h(1−s) u(xs) ds,F(t,r)=2e−t(1+r/2)2+r h(2r2+r). T_au(x)=e^{-x}\int h(1-s)\,u(xs)\,ds,\qquad F(t,r)=\frac{2e^{-t(1+r/2)}}{2+r}\,h\Big(\frac{2r}{2+r}\Big).

The kernel vanishes unless 12<y/x<32\tfrac12<y/x<\tfrac32, and FF vanishes unless −25<r<23-\tfrac25<r<\tfrac23. Along the diagonal K(x,x)=e−xh(0)/xK(x,x)=e^{-x}h(0)/x, which is unbounded if h(0)≠0h(0)\neq0: a symbol of order −∞-\infty with an unbounded kernel. TaT_a is bounded on L2(0,∞)L^2(0,\infty) by Proposition 6.4 and gains no derivative by Theorem 12.1. The model T0u(x)=∫h(1−s)u(xs)dsT_0u(x)=\int h(1-s)u(xs)ds on the front face (so Ta=e−xT0T_a=e^{-x}T_0) is not in the class, since its symbol does not decay in xx. It commutes with the unitary dilations u↦λ1/2u(λ ⋅)u\mapsto\lambda^{1/2}u(\lambda\,\cdot) of L2(0,∞)L^2(0,\infty), and Minkowski's inequality gives ∥T0u∥≤∫∣h(1−s)∣s−1/2ds ∥u∥\|T_0u\|\leq\int|h(1-s)|s^{-1/2}ds\,\|u\|, since ∥u(⋅ s)∥=s−1/2∥u∥\|u(\cdot\,s)\|=s^{-1/2}\|u\|.

Example 6.7 (The resolved kernel near r=−2r=-2). In Example 6.6, FF vanishes identically near r=−2r=-2 because hh has compact support. For a lacunary but not strongly lacunary symbol, take n=1n=1 and A(x,z)=e−xg(z)A(x,z)=e^{-x}g(z) with g∈S(R)g\in\mathcal S(\mathbb R) vanishing for z≥1z\geq1 but not near −∞-\infty, for instance g(z)=e−1/(1−z)e−z2g(z)=e^{-1/(1-z)}e^{-z^2} for z<1z<1 and g(z)=0g(z)=0 for z≥1z\geq1. Then F(t,r)=22+re−t(1+r/2)g(2r2+r)F(t,r)=\tfrac2{2+r}e^{-t(1+r/2)}g(\tfrac{2r}{2+r}), and as r↓−2r\downarrow-2 the argument 2r/(2+r)→−∞2r/(2+r)\to-\infty, where gg decreases rapidly; this is the flatness at r=−2r=-2 used in (6.4). At r=2r=2 the argument tends to 1, where gg vanishes to infinite order.