betaFeedback

WorksMathematics

Heat flow and the fundamental solution of the Baouendi-Grushin operator

Abstract

For the Baouendi-Grushin operator Lγ=Δx+∣x∣2γΔy\mathscr L_γ= Δ_x + |x|^{2γ}Δ_y in Rm+k\R^{m+k}, with γ>0γ>0, we prove that the time integral of the heat kernel with pole at the origin equals Cγ,Q Nγ2−QC_{γ,Q}\, N_γ^{2-Q}, where NγN_γ is the gauge and QQ the homogeneous dimension. When γ≠1γ\not=1 no closed form of the heat kernel is known. After a partial Fourier transform in yy, the time integral of the heat kernel becomes the Green function of an anharmonic oscillator, which solves a generalized modified Bessel equation, and the gauge appears when the inverse Fourier transform is computed as a Hankel transform. In particular, the heat flow provides a new proof that Cγ,Q Nγ2−QC_{γ,Q}\, N_γ^{2-Q} is a fundamental solution of −Lγ-\mathscr L_γ with pole at the origin.

Introduction

Consider the heat kernel p(x,t)=(4πt)−n2e−∣x∣24tp(x,t) = (4\pi t)^{-\frac{n}{2}}e^{-\frac{|x|^2}{4t}} in Rn\mathbb{R}^n, with n≥3n \ge3, and denote by σn−1=2πn/2Γ(n/2)\sigma_{n-1} = \frac{2\pi^{n/2}}{\Gamma(n/2)} the (n−1)(n-1)-dimensional measure of the unit sphere. A simple computation shows that for every x≠0x \ne0

∫0∞p(x,t) dt=1(n−2)σn−1∣x∣n−2,\int_0^\infty p(x,t)\,dt = \frac{1}{(n-2)\sigma_{n-1}|x|^{n-2}},

the fundamental solution of −Δ-\Delta with pole at the origin. That the time integral of the heat kernel is a fundamental solution is a general principle, but here the explicit knowledge of the heat kernel also reveals the explicit form of the Newtonian potential. The same happens in the Heisenberg group, where the time integral of the heat kernel found by Gaveau [19] and Hulanicki [22] gives Folland’s fundamental solution [12], and more generally in groups of Heisenberg type, where one obtains Kaplan’s fundamental solution [23], see [17]. This note is about a similar conclusion for a class of degenerate differential operators whose heat kernels are, with one exception, not known in closed form.

Given γ>0\gamma> 0 and m,k∈Nm,k \in\mathbb{N}, we consider in Rxm×Ryk\mathbb{R}^m_x \times\mathbb{R}^k_y the partial differential operator

Lγ=Δx+∣x∣2γΔy,(1)\mathcal{L}_\gamma= \Delta_x + |x|^{2\gamma}\Delta_y, \tag*{(1)}

introduced by Baouendi [1] and Grushin [21]. For its role in analysis and geometry, and for some relevant literature, we refer to [14, 10, 2]. The operator (1) is translation invariant in yy, and homogeneous of degree two with respect to the dilations δρ(x,y)=(ρx,ργ+1y)\delta_\rho(x,y)=(\rho x,\rho^{\gamma+1}y), for which d(δρ(x,y))=ρQdxdyd(\delta_\rho(x,y))=\rho^Q dxdy, with

Q=m+(γ+1)k.(2)Q=m+(\gamma+1)k. \tag*{(2)}

Let pγ((x,y),t)p_\gamma((x,y),t) denote the heat kernel for (1) with pole at (0,0)(0,0). When γ≠1\gamma\ne1 no closed form of pγp_\gamma is known, see [4 pp. 189-190]. The only exception is when γ=1\gamma=1. In [18 Theorem 3.4] the following heat kernel was found for the operator Δx+∣x∣24Δy\Delta_x+\frac{|x|^2}{4}\Delta_y, see also the references therein:

q((x,y),(x′,y′),t)=2k(4πt)m2+k∫Rke−it⟨λ,y′−y⟩(∣λ∣sinh⁡∣λ∣)m2e−∣λ∣4ttanh⁡∣λ∣((∣x∣2+∣x′∣2)−2⟨x,x′⟩sech⁡∣λ∣)dλ.(3)q((x,y),(x',y'),t)=\frac{2^k}{(4\pi t)^{\frac{m}{2}+k}}\int_{\mathbb{R}^k}e^{-\frac{i}{t}\langle\lambda,y'-y\rangle}\left(\frac{|\lambda|}{\sinh|\lambda|}\right)^{\frac{m}{2}}e^{-\frac{|\lambda|}{4t\tanh|\lambda|}\left((|x|^2+|x'|^2)-2\langle x,x'\rangle\operatorname{sech}|\lambda|\right)}d\lambda. \tag*{(3)}

and one has P1((x,y),(x′,y′),t)=2−kq((x,y2),(x′,y′2),t)P_1((x,y),(x',y'),t)=2^{-k}q((x,\frac{y}{2}),(x',\frac{y'}{2}),t), with P1P_1 as in (18) below. Notwithstanding the lack of a closed form of pγp_\gamma when γ≠1\gamma\ne1, our main result in this note is the following.

Theorem 1.1. Let m,k∈Nm,k\in\mathbb{N} and γ>0\gamma>0, and let ν=m−22(γ+1)\nu=\frac{m-2}{2(\gamma+1)}. For every (x,y)∈Rm+k(x,y)\in\mathbb{R}^{m+k} with x≠0x\ne0, one has

∫0∞pγ((x,y),t) dt=Cγ,QNγ(x,y)2−Q,(4)\int_0^\infty p_\gamma((x,y),t)\,dt=C_{\gamma,Q}N_\gamma(x,y)^{2-Q}, \tag*{(4)}

where QQ is as in (2),

Nγ(x,y)=(∣x∣2(γ+1)+(γ+1)2∣y∣2)12(γ+1),(5)N_\gamma(x,y)=\left(|x|^{2(\gamma+1)}+(\gamma+1)^2|y|^2\right)^{\frac{1}{2(\gamma+1)}}, \tag*{(5)}

and the constant in (4) is given by

Cγ,Q=(γ+1)k−1Γ(m2)Γ(Q−22(γ+1))4πm+k2Γ(ν+1).(6)C_{\gamma,Q}=\frac{(\gamma+1)^{k-1}\Gamma\left(\frac{m}{2}\right)\Gamma\left(\frac{Q-2}{2(\gamma+1)}\right)}{4\pi^{\frac{m+k}{2}}\Gamma(\nu+1)}. \tag*{(6)}

Concerning Theorem 1.1 we mention that the gauge (5) was first introduced in the paper [13]. In Proposition 2.1 of the same paper it was proved that

Γγ(x,y)=Cγ,Q∗Nγ(x,y)2−Q(7)\Gamma_\gamma(x,y)=C_{\gamma,Q}^*N_\gamma(x,y)^{2-Q} \tag*{(7)}

is a fundamental solution of −Lγ-\mathcal{L}_\gamma with pole at the origin, where

[Cγ,Q∗]−1=(Q−2)(Q+2γ)∫Rm+k∣x∣2γ[1+Nγ(x,y)2(γ+1)]1+Q+2γ2(γ+1) dxdy.(8)[C_{\gamma,Q}^*]^{-1}=(Q-2)(Q+2\gamma)\int_{\mathbb{R}^{m+k}}\frac{|x|^{2\gamma}}{\left[1+N_\gamma(x,y)^{2(\gamma+1)}\right]^{1+\frac{Q+2\gamma}{2(\gamma+1)}}}\,dxdy. \tag*{(8)}

The gauge (5) and the Baouendi-Grushin Newtonian potential (7) were instrumental in the discovery in [13] of a far-reaching monotonicity formula of Almgren type for the degenerate operators (1).

Since in Section 5 we show that Cγ,Q∗=Cγ,QC_{\gamma,Q}^*=C_{\gamma,Q}, we note explicitly that Theorem 1.1 says that the heat flow of Lγ\mathcal{L}_\gamma discovers the gauge (5). To provide the reader with an explanation of why this is possible, let Lγ\mathcal{L}_\gamma also denote the Friedrichs extension of (1) from C∞(Rm+k)C^\infty(\mathbb{R}^{m+k}), and let $p_\gamma be the heat kernel of ∂t−Lγ\partial_t-\mathcal{L}_\gamma with pole at the origin, defined in (19) below through the partial Fourier transform in yy and the heat kernels hλ(x,x′,t)h_\lambda(x,x',t) of the anharmonic oscillators

Hλ=−Δx+4π2∣λ∣2∣x∣2γ,λ∈Rk.H_\lambda=-\Delta_x+4\pi^2|\lambda|^2|x|^{2\gamma},\qquad\lambda\in\mathbb{R}^k.

When γ=1\gamma=1 the function hλh_\lambda is given by Mehler’s formula, see [18 Remark 3.3]. When γ≠1\gamma\ne1 no closed form of hλh_\lambda is known, and, besides the general facts collected in § 2, the only information on pγp_\gamma that we will use is the equation satisfied by its partial Fourier transform.

After such partial Fourier transform in yy, the function pγp_\gamma becomes the heat kernel hλ(x,0,t)h_\lambda(x,0,t) of the anharmonic oscillator HλH_\lambda, and in the variable s=∣x∣γ+1γ+1s=\frac{|x|^{\gamma+1}}{\gamma+1} the operator HλH_\lambda acting on functions which are radial in xx becomes, up to the factor ∣x∣2γ|x|^{2\gamma}, a modified Bessel operator in the fractal dimension m+2γγ+1\frac{m+2\gamma}{\gamma+1}. Such factor cannot be removed from the evolution equation, in which its reciprocal multiplies ∂t\partial_t, so that the evolution problem is not reduced to a Bessel heat equation. But it disappears from the stationary equation satisfied by ∫0∞hλ(x,0,t) dt\int_0^\infty h_\lambda(x,0,t)\,dt, the Green function of HλH_\lambda with pole at the origin, which is therefore s−νs^{-\nu} times a Macdonald function of order ν\nu in the variable 2π∣λ∣s2\pi|\lambda|s, see § 3. The inverse Fourier transform in λ\lambda is then a Hankel transform which can be computed in closed form, and it is at this stage that the gauge (5) appears, see § 4. The identity Cγ,Q∗=Cγ,QC^*_{\gamma,Q}=C_{\gamma,Q}, which also follows from [15 Lemma 3.4], is obtained in § 5 by a direct evaluation of the integral in (8). Finally, by Lemma 3.1 and Fubini’s theorem, ∫0∞pγ dt\int_0^\infty p_\gamma\,dt is a fundamental solution of −Lγ-\mathcal{L}_\gamma, and as we have mentioned above, the argument thus also gives an independent proof of [13 Prop. 2.1].

When γ=1\gamma=1, Theorem 1.1 is consistent with [17 Corollary 1.3] and [16 Theor. 1.3], see Remark 5.1. For the heat kernel of (1) when γ=1\gamma=1 we refer to [3, 7, 8, 18, 27]. When γ∈N\gamma\in\mathbb{N}, the operator (1) is a sum of squares of polynomial vector fields satisfying Hörmander’s condition, to which the lifting theorem in [26] applies, and Bauer, Furutani and Iwasaki found in [4 Cor. 7.4] a fundamental solution of Lγ\mathcal{L}_\gamma with pole at an arbitrary point, which for the pole at the origin reduces to (7). Their approach integrates in time the explicit heat kernel of an auxiliary operator obtained after a partial Fourier transform in yy and polar coordinates in xx, and uses γ∈N\gamma\in\mathbb{N} to sum a series of spherical harmonics; the heat kernel pγp_\gamma does not appear in [4]. See also [5] for other special values of γ\gamma.

Notation. Throughout the paper m,k∈Nm,k\in\mathbb{N} and γ>0\gamma>0 are fixed, QQ is as in (2), and we let

ν=m−22(γ+1),θ=k2+ν=Q−22(γ+1).(9)\nu=\frac{m-2}{2(\gamma+1)},\qquad\theta=\frac{k}{2}+\nu=\frac{Q-2}{2(\gamma+1)}. \tag*{(9)}

We note that ν>−12\nu>-\frac{1}{2}, that ν=0\nu=0 if and only if m=2m=2, and that θ>0\theta>0. We denote by ωm−1=2πm/2Γ(m/2)\omega_{m-1}=\frac{2\pi^{m/2}}{\Gamma(m/2)} the surface measure of the unit sphere Sm−1⊂Rm\mathbb{S}^{m-1}\subset\mathbb{R}^m, with the understanding that ω0=2\omega_0=2. We will use the function

s=s(x)=∣x∣γ+1γ+1,(10)s=s(x)=\frac{|x|^{\gamma+1}}{\gamma+1}, \tag*{(10)}

in terms of which the gauge (5) can be written as

s2+∣y∣2=Nγ(x,y)2(γ+1)(γ+1)2.(11)s^2+|y|^2=\frac{N_\gamma(x,y)^{2(\gamma+1)}}{(\gamma+1)^2}. \tag*{(11)}

The partial Fourier transform with respect to y∈Rky\in\mathbb{R}^k is

u^((x,λ),t)=∫Rke−2πi⟨λ,y⟩u((x,y),t) dy.\widehat{u}((x,\lambda),t)=\int_{\mathbb{R}^k}e^{-2\pi i\langle\lambda,y\rangle}u((x,y),t)\,dy.

Finally, we say that a function ff on Rk\mathbb{R}^k is spherically symmetric if f(λ)=f∗(∣λ∣)f(\lambda)=f^*(|\lambda|) for some f∗f^*; when k=1k=1 we simply mean that ff is even. We will use the following classical formula of Bochner, see [6 Theor. 40 on p. 69], which holds for k≥1k\geq1 and for every spherically symmetric f∈L1(Rk)f\in L^1(\mathbb{R}^k),

f^(ξ)=2π∣ξ∣k2−1∫0∞rk2f∗(r)Jk2−1(2π∣ξ∣r) dr,ξ≠0.(12)\widehat{f}(\xi)=\frac{2\pi}{|\xi|^{\frac{k}{2}-1}}\int_0^\infty r^{\frac{k}{2}}f^*(r)J_{\frac{k}{2}-1}(2\pi|\xi|r)\,dr,\qquad\xi\neq0. \tag*{(12)}

Since the inverse Fourier transform of a spherically symmetric function coincides with its Fourier transform, (12) also computes f~\widetilde{f}.

The heat kernel with pole at the origin

Since (1) is translation invariant in yy, the Cauchy problem

∂tu−Lγu=0,u((x,y),0)=φ(x,y),(13)\partial_tu-\mathcal{L}_\gamma u=0,\qquad u((x,y),0)=\varphi(x,y), \tag*{(13)}

is converted by the partial Fourier transform into the following one

∂tu^+Hλu^=0,u^((x,λ),0)=φ^(x,λ),(14)\partial_t\widehat{u}+H_\lambda\widehat{u}=0,\qquad\widehat{u}((x,\lambda),0)=\widehat{\varphi}(x,\lambda), \tag*{(14)}

where for λ∈Rk\lambda\in\mathbb{R}^k we have let

Hλ=−Δx+4π2∣λ∣2∣x∣2γ.(15)H_\lambda=-\Delta_x+4\pi^2|\lambda|^2|x|^{2\gamma}. \tag*{(15)}

We collect the facts about (15) that will be needed. Let λ≠0\lambda\neq0. Since the potential V(x):=4π2∣λ∣2∣x∣2γV(x):=4\pi^2|\lambda|^2|x|^{2\gamma} is continuous, nonnegative, and tends to ∞\infty as ∣x∣→∞|x|\to\infty, the operator HλH_\lambda is self-adjoint and nonnegative on L2(Rm)L^2(\mathbb{R}^m), with compact resolvent, see [25 Theor. XIII.67]. Its spectrum is thus discrete, and its first eigenvalue μ1(λ)\mu_1(\lambda) is strictly positive, since

⟨Hλψ,ψ⟩=∫Rm∣∇ψ∣2+4π2∣λ∣2∫Rm∣x∣2γψ2=0\langle H_\lambda\psi,\psi\rangle=\int_{\mathbb{R}^m}|\nabla\psi|^2+4\pi^2|\lambda|^2\int_{\mathbb{R}^m}|x|^{2\gamma}\psi^2=0

forces ψ≡0\psi\equiv0. The change of variable x→∣λ∣−1γ+1xx\to|\lambda|^{-\frac{1}{\gamma+1}}x shows that HλH_\lambda is unitarily equivalent to ∣λ∣2γ+1Hλ0|\lambda|^{\frac{2}{\gamma+1}}H_{\lambda_0}, with ∣λ0∣=1|\lambda_0|=1, and therefore

μ1(λ)=μ1∣λ∣2γ+1,μ1=μ1(λ0)>0.(16)\mu_1(\lambda)=\mu_1|\lambda|^{\frac{2}{\gamma+1}},\qquad\mu_1=\mu_1(\lambda_0)>0. \tag*{(16)}

The semigroup e−tHλe^{-tH_\lambda} admits a kernel hλ(x,x′,t)h_\lambda(x,x',t), continuous on Rm×Rm×(0,∞)\mathbb{R}^m\times\mathbb{R}^m\times(0,\infty), symmetric in (x,x′)(x,x'), and such that

0≤hλ(x,x′,t)≤(4πt)−m2e−∣x−x′∣24t,(17)0\leq h_\lambda(x,x',t)\leq(4\pi t)^{-\frac{m}{2}}e^{-\frac{|x-x'|^2}{4t}}, \tag*{(17)}

see [9 Theor. 1.8.1 and Example 2.1.9]. For λ=0\lambda= 0 the kernel h0h_0 is the Gauss-Weierstrass kernel. The upper bound in (17) implies in particular that e−tHλe^{-tH_\lambda} is a contraction in L∞(Rm)L^\infty(\mathbb{R}^m). Since HλH_\lambda commutes with the rotations of Rm\mathbb{R}^m, one has for every R∈O(m)R \in O(m),

hλ(Rx,Rx′,t)=hλ(x,x′,t),h_\lambda(Rx,Rx',t)=h_\lambda(x,x',t),

and therefore the function x→hλ(x,0,t)x \to h_\lambda(x,0,t) is spherically symmetric.

Lemma 2.1. Let λ∈Rk∖{0}\lambda\in\mathbb{R}^k \setminus\{0\}. For every x,x′∈Rmx,x' \in\mathbb{R}^m and t>0t > 0 one has

hλ(x,x′,t)≤(2πt)−m2e−t2μ1∣λ∣2γ+1.h_\lambda(x,x',t) \leq(2\pi t)^{-\frac{m}{2}} e^{-\frac{t}{2}\mu_1|\lambda|^{\frac{2}{\gamma+1}}}.

Proof. By the semigroup property and the symmetry of the kernel, for 0<τ<t0 < \tau< t we have hλ(x,⋅,t)=e−(t−τ)Hλhλ(x,⋅,τ)h_\lambda(x,\cdot,t)=e^{-(t-\tau)H_\lambda}h_\lambda(x,\cdot,\tau). Since ∥e−sHλ∥L2→L2=e−sμ1(λ)\lVert e^{-sH_\lambda}\rVert_{L^2\to L^2}=e^{-s\mu_1(\lambda)}, we infer

hλ(x,x,2t)=∥hλ(x,⋅,t)∥L2(Rm)2≤e−2(t−τ)μ1(λ)∥hλ(x,⋅,τ)∥L2(Rm)2=e−2(t−τ)μ1(λ)hλ(x,x,2τ).h_\lambda(x,x,2t)=\lVert h_\lambda(x,\cdot,t)\rVert^2_{L^2(\mathbb{R}^m)} \leq e^{-2(t-\tau)\mu_1(\lambda)}\lVert h_\lambda(x,\cdot,\tau)\rVert^2_{L^2(\mathbb{R}^m)} =e^{-2(t-\tau)\mu_1(\lambda)}h_\lambda(x,x,2\tau).

Taking τ=t/2\tau=t/2, and using (17), we find hλ(x,x,2t)≤e−tμ1(λ)(4πt)−m2h_\lambda(x,x,2t)\leq e^{-t\mu_1(\lambda)}(4\pi t)^{-\frac{m}{2}}. On the other hand, by Cauchy-Schwarz

hλ(x,x′,2t)=∫Rmhλ(x,ξ,t)hλ(ξ,x′,t) dξ≤(hλ(x,x,2t)hλ(x′,x′,2t))12≤e−tμ1(λ)(4πt)−m2.h_\lambda(x,x',2t)=\int_{\mathbb{R}^m}h_\lambda(x,\xi,t)h_\lambda(\xi,x',t)\,d\xi \leq\bigl(h_\lambda(x,x,2t)h_\lambda(x',x',2t)\bigr)^{\frac{1}{2}} \leq e^{-t\mu_1(\lambda)}(4\pi t)^{-\frac{m}{2}}.

Replacing tt with t/2t/2, and keeping (16) in mind, we reach the desired conclusion.

Since C∞(Rm)C^\infty(\mathbb{R}^m) is a form core for each HλH_\lambda, the closure of the form of −Lγ-\mathcal{L}_\gamma on C∞(Rm+k)C^\infty(\mathbb{R}^{m+k}) is, after the partial Fourier transform, the direct integral in λ\lambda of the forms of HλH_\lambda, and therefore etLγφ(⋅,λ)=e−tHλφ^(⋅,λ)e^{t\mathcal{L}_\gamma}\varphi(\cdot,\lambda)=e^{-tH_\lambda}\widehat{\varphi}(\cdot,\lambda) for a.e. λ\lambda. Measurability in λ\lambda follows from

hλ(x,x′,t)=∣λ∣mγ+1hλ0(∣λ∣1γ+1x,∣λ∣1γ+1x′,∣λ∣2γ+1t).h_\lambda(x,x',t)=|\lambda|^{\frac{m}{\gamma+1}}h_{\lambda_0}\left(|\lambda|^{\frac{1}{\gamma+1}}x,|\lambda|^{\frac{1}{\gamma+1}}x',|\lambda|^{\frac{2}{\gamma+1}}t\right).

Hence the solution of (13) is given by

u((x,y),t)=∫Rke2πi⟨y,λ⟩∫Rmhλ(x,x′,t)φ^(x′,λ) dx′ dλ=∫Rm∫RkPγ((x,y),(x′,y′),t)φ(x′,y′) dy′ dx′,u((x,y),t)=\int_{\mathbb{R}^k}e^{2\pi i\langle y,\lambda\rangle}\int_{\mathbb{R}^m}h_\lambda(x,x',t)\widehat{\varphi}(x',\lambda)\,dx'\,d\lambda =\int_{\mathbb{R}^m}\int_{\mathbb{R}^k}P_\gamma((x,y),(x',y'),t)\varphi(x',y')\,dy'\,dx',

where

Pγ((x,y),(x′,y′),t)=∫Rke2πi⟨y−y′,λ⟩hλ(x,x′,t) dλ.(18)P_\gamma((x,y),(x',y'),t)=\int_{\mathbb{R}^k}e^{2\pi i\langle y-y',\lambda\rangle}h_\lambda(x,x',t)\,d\lambda. \tag*{(18)}

By Lemma 2.1 the integral in (18) converges absolutely. Accordingly, the heat kernel of ∂t−Lγ\partial_t-\mathcal{L}_\gamma with pole at the origin is the function

pγ((x,y),t)=Pγ((x,y),(0,0),t)=∫Rke2πi⟨y,λ⟩hλ(x,0,t) dλ.(19)p_\gamma((x,y),t)=P_\gamma((x,y),(0,0),t)=\int_{\mathbb{R}^k}e^{2\pi i\langle y,\lambda\rangle}h_\lambda(x,0,t)\,d\lambda. \tag*{(19)}

Integration in time at fixed frequency

For λ∈Rk∖{0}\lambda\in\mathbb{R}^{k} \setminus\{0\} and x∈Rm∖{0}x \in\mathbb{R}^{m} \setminus\{0\} we define

w(x,λ)=∫0∞hλ(x,0,t) dt.(20)w(x,\lambda)=\int_{0}^{\infty}h_{\lambda}(x,0,t)\,dt. \tag*{(20)}

By (17) and Lemma 2.1 we have

0≤w(x,λ)≤∫01(4πt)−m2e−∣x∣24t dt+∫1∞(2πt)−m2e−t2μ1(λ)e2γ+1 dt,(21)0\leq w(x,\lambda)\leq\int_{0}^{1}(4\pi t)^{-\frac{m}{2}}e^{-\frac{|x|^{2}}{4t}}\,dt+\int_{1}^{\infty}(2\pi t)^{-\frac{m}{2}}e^{-\frac{t}{2}\mu_{1}(\lambda)}e^{\frac{2}{\gamma+1}}\,dt, \tag*{(21)}

which shows that the integral in (20): (a) converges for x≠0x\neq0, (b) that w(⋅,λ)∈Lloc1(Rm)w(\cdot,\lambda)\in L^{1}_{\mathrm{loc}}(\mathbb{R}^{m}) (the first integral in the right-hand side of (21) is bounded by C∣x∣2−mC|x|^{2-m} when m≥3m\geq3, by C(1+∣log⁡∣x∣∣)C(1+|\log|x||) when m=2m=2, and by a constant when m=1m=1), (c) and that, by dominated convergence (for each t>0t>0 one has hλ(x,0,t)→0h_{\lambda}(x,0,t)\to0 as ∣x∣→∞|x|\to\infty by (17), and for ∣x∣≥1|x|\geq1 the integrand is dominated by the minimum of the two bounds in (17) and Lemma 2.1, which is integrable on (0,∞)(0,\infty)),

w(x,λ)→0as ∣x∣→∞.(22)w(x,\lambda)\to0 \quad\text{as } |x|\to\infty. \tag*{(22)}

Moreover, w(⋅,λ)w(\cdot,\lambda) is spherically symmetric, since so is hλ(⋅,0,t)h_{\lambda}(\cdot,0,t).

Lemma 3.1. Let λ∈Rk∖{0}\lambda\in\mathbb{R}^{k}\setminus\{0\}. For every ψ∈C0∞(Rm)\psi\in C_{0}^{\infty}(\mathbb{R}^{m}) one has

∫Rmw(x,λ)Hλψ(x) dx=ψ(0).(23)\int_{\mathbb{R}^{m}}w(x,\lambda)H_{\lambda}\psi(x)\,dx=\psi(0). \tag*{(23)}

Proof. Since ψ\psi belongs to the domain of HλH_{\lambda}, we have ∂te−tHλψ=−e−tHλHλψ\partial_{t}e^{-tH_{\lambda}}\psi=-e^{-tH_{\lambda}}H_{\lambda}\psi. Using the symmetry of the kernel hλh_{\lambda}, for 0<ε<T<∞0<\varepsilon<T<\infty we thus find

∫εT∫Rmhλ(x,0,t)Hλψ(x) dx dt=∫εT(e−tHλHλψ)(0) dt=(e−εHλψ)(0)−(e−THλψ)(0).\int_{\varepsilon}^{T}\int_{\mathbb{R}^{m}}h_{\lambda}(x,0,t)H_{\lambda}\psi(x)\,dx\,dt=\int_{\varepsilon}^{T}(e^{-tH_{\lambda}}H_{\lambda}\psi)(0)\,dt=(e^{-\varepsilon H_{\lambda}}\psi)(0)-(e^{-T H_{\lambda}}\psi)(0).

By Lemma 2.1 we have

∣(e−THλψ)(0)∣≤(2πT)−m2e−T2μ1(λ)∥ψ∥L1(Rm)→0,as T→∞.|(e^{-T H_{\lambda}}\psi)(0)|\leq(2\pi T)^{-\frac{m}{2}}e^{-\frac{T}{2}\mu_{1}(\lambda)}\|\psi\|_{L^{1}(\mathbb{R}^{m})}\to0,\qquad\text{as }T\to\infty.

On the other hand, since e−sHλe^{-sH_{\lambda}} is a contraction in L∞(Rm)L^{\infty}(\mathbb{R}^{m}), we have

∣(e−εHλψ)(0)−ψ(0)∣=∣∫0ε(e−sHλHλψ)(0) ds∣≤ε∥Hλψ∥L∞(Rm)→0as ε→0+.|(e^{-\varepsilon H_{\lambda}}\psi)(0)-\psi(0)|=\left|\int_{0}^{\varepsilon}(e^{-sH_{\lambda}}H_{\lambda}\psi)(0)\,ds\right|\leq\varepsilon\|H_{\lambda}\psi\|_{L^{\infty}(\mathbb{R}^{m})}\to0\qquad\text{as }\varepsilon\to0^{+}.

Finally, since hλ≥0h_{\lambda}\geq0 and w(⋅,λ)∈Lloc1(Rm)w(\cdot,\lambda)\in L^{1}_{\mathrm{loc}}(\mathbb{R}^{m}), Tonelli’s theorem gives

∫0∞∫Rmhλ(x,0,t)∣Hλψ(x)∣ dx dt≤∥Hλψ∥L∞(Rm)∫supp⁡ψw(x,λ) dx<∞,\int_{0}^{\infty}\int_{\mathbb{R}^{m}}h_{\lambda}(x,0,t)|H_{\lambda}\psi(x)|\,dx\,dt\leq\|H_{\lambda}\psi\|_{L^{\infty}(\mathbb{R}^{m})}\int_{\operatorname{supp}\psi}w(x,\lambda)\,dx<\infty,

and we can let ε→0+\varepsilon\to0^{+} and T→∞T\to\infty in the above identity by dominated convergence. This proves (23).

We now solve (23). Lemma 3.1 says that w(⋅,λ)w(\cdot,\lambda) is a distributional solution of Hλw=δH_\lambda w=\delta in Rm\mathbb{R}^{m}, where δ\delta is the Dirac delta at the origin. In particular, since the potential in (15) is C∞C^\infty in Rm∖{0}\mathbb{R}^{m}\setminus\{0\}, we have Hλw=0H_\lambda w=0 there, and by hypoellipticity w(⋅,λ)∈C∞(Rm∖{0})w(\cdot,\lambda)\in C^\infty(\mathbb{R}^{m}\setminus\{0\}). By the spherical symmetry of w(⋅,λ)w(\cdot,\lambda), we can write w(x,λ)=W(∣x∣)w(x,\lambda)=W(|x|), where W∈C∞(0,∞)W\in C^\infty(0,\infty) solves the ordinary differential equation

W′′+m−1rW′−4π2∣λ∣2r2γW=0,r>0.(24)W''+\frac{m-1}{r}W'-4\pi^2|\lambda|^2r^{2\gamma}W=0,\qquad r>0. \tag*{(24)}

If we make the change of variables

W(r)=r1−m2Φ(ζ),ζ=2π∣λ∣γ+1rγ+1=2π∣λ∣s,(25)W(r)=r^{1-\frac{m}{2}}\Phi(\zeta),\qquad\zeta=\frac{2\pi|\lambda|}{\gamma+1}r^{\gamma+1}=2\pi|\lambda|s, \tag*{(25)}

a direct computation shows that

rm2+1(γ+1)2(W′′+m−1rW′−4π2∣λ∣2r2γW)=ζ2Φ′′(ζ)+ζΦ′(ζ)−(ζ2+ν2)Φ(ζ),\frac{r^{\frac{m}{2}+1}}{(\gamma+1)^2}\left(W''+\frac{m-1}{r}W'-4\pi^2|\lambda|^2r^{2\gamma}W\right)=\zeta^2\Phi''(\zeta)+\zeta\Phi'(\zeta)-(\zeta^2+\nu^2)\Phi(\zeta),

with ν\nu as in (9). Therefore, (24) is equivalent to the modified Bessel equation of order ν\nu for Φ\Phi, whose general solution is Φ(ζ)=AKν(ζ)+BIν(ζ)\Phi(\zeta)=AK_\nu(\zeta)+BI_\nu(\zeta), see [24 Sec. 5.7]. Here, IνI_\nu and KνK_\nu respectively denote the modified Bessel functions of the first and third kind, and we recall that K−ν=KνK_{-\nu}=K_\nu. Since as ζ→∞\zeta\to\infty one has Iν(ζ)∼(2πζ)−12eζI_\nu(\zeta)\sim(2\pi\zeta)^{-\frac{1}{2}}e^\zeta, whereas Kν(ζ)∼(π2ζ)12e−ζK_\nu(\zeta)\sim\left(\frac{\pi}{2\zeta}\right)^{\frac{1}{2}}e^{-\zeta}, (22) forces B=0B=0. We conclude that

w(x,λ)=A(λ)∣x∣1−m2Kν(2π∣λ∣s),(26)w(x,\lambda)=A(\lambda)|x|^{1-\frac{m}{2}}K_\nu(2\pi|\lambda|s), \tag*{(26)}

for some constant A(λ)A(\lambda) which we now determine using (23). Fix ψ∈C∞(Rm)\psi\in C^\infty(\mathbb{R}^{m}) with ψ(0)=1\psi(0)=1. Since Hλw=0H_\lambda w=0 in Rm∖{0}\mathbb{R}^{m}\setminus\{0\}, integrating by parts in {∣x∣>ε}\{|x|>\varepsilon\} we find

∫{∣x∣>ε}wHλψ dx=∫{∣x∣>ε}(ψΔw−wΔψ) dx=∫{∣x∣=ε}(w∂rψ−ψ∂rw) dSε=εm−1W(ε)∫Sm−1∂rψ(εω) dω−εm−1W′(ε)∫Sm−1ψ(εω) dω.\begin{aligned} \int_{\{|x|>\varepsilon\}}wH_\lambda\psi\,dx=\int_{\{|x|>\varepsilon\}}(\psi\Delta w-w\Delta\psi)\,dx=\int_{\{|x|=\varepsilon\}}(w\partial_r\psi-\psi\partial_rw)\,dS_\varepsilon\\ &=\varepsilon^{m-1}W(\varepsilon)\int_{S^{m-1}}\partial_r\psi(\varepsilon\omega)\,d\omega-\varepsilon^{m-1}W'(\varepsilon)\int_{S^{m-1}}\psi(\varepsilon\omega)\,d\omega. \end{aligned}

When m=1m=1 the integrals over S0={±1}S^0=\{\pm1\} are to be interpreted as sums. Letting ε→0+\varepsilon\to0^+ in the latter identity, and using (23) and w∈Lloc1(Rm)w\in L^1_{\mathrm{loc}}(\mathbb{R}^{m}), we obtain

lim⁡ε→0+εm−1W(ε)∫Sm−1∂rψ(εω) dω−ωm−1lim⁡ε→0+εm−1W′(ε)=1,(27)\lim_{\varepsilon\to0^+}\varepsilon^{m-1}W(\varepsilon)\int_{S^{m-1}}\partial_r\psi(\varepsilon\omega)\,d\omega-\omega_{m-1}\lim_{\varepsilon\to0^+}\varepsilon^{m-1}W'(\varepsilon)=1, \tag*{(27)}

provided that the two limits exist. We now use the behavior of KνK_\nu near ζ=0\zeta=0. We treat separately three cases.

Case m≥3m\ge3. In this case ν>0\nu>0, and we have Kν(ζ)=12Γ(ν)(ζ/2)−ν(1+o(1))K_\nu(\zeta)=\frac{1}{2}\Gamma(\nu)(\zeta/2)^{-\nu}(1+o(1)) and Kν′(ζ)=−12Γ(ν)(ζ/2)−νζ−1(1+o(1))K_\nu'(\zeta)=-\frac{1}{2}\Gamma(\nu)(\zeta/2)^{-\nu}\zeta^{-1}(1+o(1)) as ζ→0+\zeta\to0^+, see [24 Sec. 5.7]. Since by (9) we have (γ+1)ν=m−22(\gamma+1)\nu=\frac{m-2}{2}, from (25) we find

W(r)=A(λ)2Γ(ν)(π∣λ∣γ+1)−νr2−m(1+o(1)),(28)W(r)=\frac{A(\lambda)}{2}\Gamma(\nu)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{-\nu}r^{2-m}(1+o(1)), \tag*{(28)}
W′(r)=(2−m)A(λ)2Γ(ν)(π∣λ∣γ+1)−νr1−m(1+o(1)),W'(r)=(2-m)\frac{A(\lambda)}{2}\Gamma(\nu)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{-\nu}r^{1-m}(1+o(1)),

as r→0+r\to0^{+}. The first limit in (27) vanishes, and we obtain

A(λ)=2(m−2)ωm−1Γ(ν)(π∣λ∣γ+1)ν.(29)A(\lambda)=\frac{2}{(m-2)\omega_{m-1}\Gamma(\nu)}\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{\nu}. \tag*{(29)}

Case m=2m=2. In this case ν=0\nu=0, and K0(ζ)=−log⁡ζ+O(1)K_{0}(\zeta)=-\log\zeta+O(1), K0′(ζ)=−K1(ζ)=−ζ−1(1+o(1))K_{0}'(\zeta)=-K_{1}(\zeta)=-\zeta^{-1}(1+o(1)) as ζ→0+\zeta\to0^{+}, see [24 Sec. 5.7]. Since ζ=2π∣λ∣γ+1rγ+1\zeta=\frac{2\pi|\lambda|}{\gamma+1}r^{\gamma+1}, from (25) we obtain W(r)=−(γ+1)A(λ)log⁡r+O(1)W(r)=-(\gamma+1)A(\lambda)\log r+O(1), and W′(r)=−(γ+1)A(λ)r−1(1+o(1))W'(r)=-(\gamma+1)A(\lambda)r^{-1}(1+o(1)) as r→0+r\to0^{+}. Again, the first limit in (27) vanishes, and we find

A(λ)=1(γ+1)ω1=12π(γ+1).(30)A(\lambda)=\frac{1}{(\gamma+1)\omega_{1}}=\frac{1}{2\pi(\gamma+1)}. \tag*{(30)}

Case m=1m=1. In this case −12<ν=−12(γ+1)<0-\frac{1}{2}<\nu=-\frac{1}{2(\gamma+1)}<0, and Kν=K∣ν∣K_{\nu}=K_{|\nu|}, with

K∣ν∣(ζ)=12Γ(∣ν∣)(ζ2)−∣ν∣(1+O(ζ2))+12Γ(−∣ν∣)(ζ2)∣ν∣(1+O(ζ2)),K_{|\nu|}(\zeta)=\frac{1}{2}\Gamma(|\nu|)\left(\frac{\zeta}{2}\right)^{-|\nu|}(1+O(\zeta^{2}))+\frac{1}{2}\Gamma(-|\nu|)\left(\frac{\zeta}{2}\right)^{|\nu|}(1+O(\zeta^{2})),

as ζ→0+\zeta\to0^{+}, which follows from the series expansion of I±∣ν∣I_{\pm|\nu|}, from K∣ν∣=π2I−∣ν∣−I∣ν∣sin⁡π∣ν∣K_{|\nu|}=\frac{\pi}{2}\frac{I_{-|\nu|}-I_{|\nu|}}{\sin\pi|\nu|}, and from the reflection formula for the gamma function, see [24 Sec. 5.7]. Since now (γ+1)∣ν∣=12(\gamma+1)|\nu|=\frac{1}{2}, from (25) we find

W(r)=A(λ)2Γ(∣ν∣)(π∣λ∣γ+1)−∣ν∣+A(λ)2Γ(−∣ν∣)(π∣λ∣γ+1)∣ν∣r+O(r2γ+2),W(r)=\frac{A(\lambda)}{2}\Gamma(|\nu|)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{-|\nu|}+\frac{A(\lambda)}{2}\Gamma(-|\nu|)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{|\nu|}r+O(r^{2\gamma+2}),

which shows that WW is continuous up to r=0r=0, and that W′(0+)=A(λ)2Γ(−∣ν∣)(π∣λ∣γ+1)∣ν∣W'(0+)=\frac{A(\lambda)}{2}\Gamma(-|\nu|)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{|\nu|}. Since ∂rψ(ε)+∂rψ(−ε)=ψ′(ε)−ψ′(−ε)→0\partial_{r}\psi(\varepsilon)+\partial_{r}\psi(-\varepsilon)=\psi'(\varepsilon)-\psi'(-\varepsilon)\to0, the first limit in (27) vanishes also in this case, and we obtain A(λ)Γ(−∣ν∣)(π∣λ∣γ+1)∣ν∣=−1A(\lambda)\Gamma(-|\nu|)\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{|\nu|}=-1. Recalling that ν=−∣ν∣\nu=-|\nu|, we have thus found

A(λ)=−1Γ(ν)(π∣λ∣γ+1)ν.(31)A(\lambda)=-\frac{1}{\Gamma(\nu)}\left(\frac{\pi|\lambda|}{\gamma+1}\right)^{\nu}. \tag*{(31)}

Since ω0=2\omega_{0}=2, we see that (31) is (29) with m=1m=1. Furthermore, using Γ(ν+1)=νΓ(ν)\Gamma(\nu+1)=\nu\Gamma(\nu), and m2−1ν=1γ+1\frac{\frac{m}{2}-1}{\nu}=\frac{1}{\gamma+1}, we can rewrite (29) in the following form, which for m=2m=2 returns (30),

A(λ)=Γ(m2)2(γ+1)πm2Γ(ν+1)(πγ+1)ν∣λ∣ν.(32)A(\lambda)=\frac{\Gamma\left(\frac{m}{2}\right)}{2(\gamma+1)\pi^{\frac{m}{2}}\Gamma(\nu+1)}\left(\frac{\pi}{\gamma+1}\right)^{\nu}|\lambda|^{\nu}. \tag*{(32)}

Substituting (32) in (26), we conclude that for every m∈Nm \in\mathbb{N} one has

w(x,λ)=Γ(m2)2(γ+1)πm2Γ(ν+1)(πγ+1)ν∣λ∣ν∣x∣1−m2Kν(2π∣λ∣∣x∣γ+1γ+1).(33)w(x,\lambda)=\frac{\Gamma\left(\frac{m}{2}\right)}{2(\gamma+1)\pi^{\frac{m}{2}}\Gamma(\nu+1)}\left(\frac{\pi}{\gamma+1}\right)^\nu|\lambda|^\nu|x|^{1-\frac{m}{2}}K_\nu\left(\frac{2\pi|\lambda||x|^{\gamma+1}}{\gamma+1}\right). \tag*{(33)}

Integration in the frequency variable

We now fix x∈Rm∖{0}x \in\mathbb{R}^m \setminus\{0\} and compute the inverse Fourier transform of (33) with respect to λ\lambda. We first observe that w(x,⋅)∈L1(Rk)w(x,\cdot) \in L^1(\mathbb{R}^k). The exponential decay of KνK_\nu takes care of the integrability at infinity. Near λ=0\lambda=0, from (33) and the expansions of KνK_\nu recalled above, one has that w(x,λ)w(x,\lambda) is bounded when m≥3m \ge3, that ∣w(x,λ)∣≤C(1+∣log⁡∣λ∣∣)|w(x,\lambda)| \le C(1+|\log|\lambda||) when m=2m=2, and that ∣w(x,λ)∣≤C∣λ∣2ν=C∣λ∣γ+1k|w(x,\lambda)| \le C|\lambda|^{2\nu}=C|\lambda|^{\frac{\gamma+1}{k}} when m=1m=1. In all cases, w(x,⋅)w(x,\cdot) is integrable near λ=0\lambda=0 since k≥1>1γ+1k \ge1 > \frac{1}{\gamma+1}.

Since hλ≥0h_\lambda\ge0, we have from (19), (20) and Tonelli’s theorem

∫0∞∫Rk∣e2πi⟨y,λ⟩hλ(x,0,t)∣ dλ dt=∫Rkw(x,λ) dλ<∞.\int_0^\infty\int_{\mathbb{R}^k}\left|e^{2\pi i\langle y,\lambda\rangle}h_\lambda(x,0,t)\right|\,d\lambda\,dt=\int_{\mathbb{R}^k}w(x,\lambda)\,d\lambda<\infty.

By Fubini’s theorem we thus obtain

Eγ(x,y)=∫0∞pγ((x,y),t) dt=∫Rke2πi⟨y,λ⟩w(x,λ) dλ.(34)\mathcal{E}_\gamma(x,y)=\int_0^\infty p_\gamma((x,y),t)\,dt=\int_{\mathbb{R}^k}e^{2\pi i\langle y,\lambda\rangle}w(x,\lambda)\,d\lambda. \tag*{(34)}

In particular, Eγ(x,⋅)\mathcal{E}_\gamma(x,\cdot) is continuous in Rk\mathbb{R}^k. Since Γγ\Gamma_\gamma is also continuous in yy when x≠0x\ne0, it suffices to consider y≠0y\ne0. Since w(x,⋅)w(x,\cdot) is spherically symmetric in Rk\mathbb{R}^k, the inverse Fourier transform in (34) is computed by (12). Writing w(x,λ)=A0∣λ∣ν−m2Kν(2π∣λ∣s)w(x,\lambda)=A_0|\lambda|^{\nu-\frac{m}{2}}K_\nu(2\pi|\lambda|s), with ss as in (10) and

A0=Γ(m2)2(γ+1)πm2Γ(ν+1)(πγ+1)ν,(35)A_0=\frac{\Gamma\left(\frac{m}{2}\right)}{2(\gamma+1)\pi^{\frac{m}{2}}\Gamma(\nu+1)}\left(\frac{\pi}{\gamma+1}\right)^\nu, \tag*{(35)}

we obtain from (34)

Eγ(x,y)=2πA0∣x∣1−m2∣y∣k2−1∫0∞ρk2+νKν(2πsρ)Jk2−1(2π∣y∣ρ) dρ.(36)\mathcal{E}_\gamma(x,y)=\frac{2\pi A_0|x|^{1-\frac{m}{2}}}{|y|^{\frac{k}{2}-1}}\int_0^\infty\rho^{\frac{k}{2}+\nu}K_\nu(2\pi s\rho)J_{\frac{k}{2}-1}(2\pi|y|\rho)\,d\rho. \tag*{(36)}

At this point we recall the following classical formula, see [28 Sec. 13.45], or also [20 6.576 (3)],

∫0∞τ−ηKμ(aτ)Jν′(bτ) dτ=bν′Γ(ν′−η+μ+12)Γ(ν′−η−μ+12)2η+1aν′−η+1Γ(ν′+1) 2F1(ν′−η+μ+12,ν′−η−μ+12;ν′+1;−b2a2).(37)\int_0^\infty\tau^{-\eta}K_\mu(a\tau)J_{\nu'}(b\tau)\,d\tau= \frac{b^{\nu'}\Gamma\left(\frac{\nu'-\eta+\mu+1}{2}\right)\Gamma\left(\frac{\nu'-\eta-\mu+1}{2}\right)} {2^{\eta+1}a^{\nu'-\eta+1}\Gamma(\nu'+1)} \,{}_2F_1\left(\frac{\nu'-\eta+\mu+1}{2},\frac{\nu'-\eta-\mu+1}{2};\nu'+1;-\frac{b^2}{a^2}\right). \tag*{(37)}

valid for a>0a>0, b≥0b\ge0, and ν′−η+1>∣μ∣\nu'-\eta+1>|\mu|. We apply (37) with the choice

μ=ν,ν′=k2−1,η=−k2−ν,a=2πs,b=2π∣y∣.\mu=\nu,\qquad\nu'=\frac{k}{2}-1,\qquad\eta=-\frac{k}{2}-\nu,\qquad a=2\pi s,\qquad b=2\pi|y|.

The condition ν′−η+1=k+ν>∣ν∣\nu' - \eta+ 1 = k + \nu> |\nu| is satisfied, since k≥1>1γ+1=2∣ν∣k \ge1 > \frac{1}{\gamma+1} = 2|\nu| when m=1m = 1, and ν≥0\nu\ge0 when m≥2m \ge2. Keeping (9) in mind, we find

ν′−η+μ+12=k2+ν=θ,ν′−η−μ+12=k2=ν′+1.\frac{\nu' - \eta+ \mu+ 1}{2} = \frac{k}{2} + \nu= \theta,\qquad\frac{\nu' - \eta- \mu+ 1}{2} = \frac{k}{2} = \nu' + 1.

The hypergeometric function in (37) is therefore in the special form F(a,b;b;x)=(1−x)−aF(a,b;b;x)=(1-x)^{-a}, see [11 (4) on p. 101], and we have

F(θ,k2;k2;−∣y∣2s2)=(1+∣y∣2s2)−θ=s2θ(s2+∣y∣2)−θ.F\left(\theta,\frac{k}{2};\frac{k}{2};-\frac{|y|^2}{s^2}\right)=\left(1+\frac{|y|^2}{s^2}\right)^{-\theta}=s^{2\theta}(s^2+|y|^2)^{-\theta}.

We thus obtain from (37)

∫0∞ρk2+νKν(2πsρ)Jk2−1(2π∣y∣ρ) dρ=2k2+ν−1(2π∣y∣)k2−1(2πs)−k−νΓ(θ)s2θ(s2+∣y∣2)−θ.(38)\int_0^\infty\rho^{\frac{k}{2}+\nu}K_\nu(2\pi s\rho)J_{\frac{k}{2}-1}(2\pi|y|\rho)\,d\rho =2^{\frac{k}{2}+\nu-1}(2\pi|y|)^{\frac{k}{2}-1}(2\pi s)^{-k-\nu}\Gamma(\theta)s^{2\theta}(s^2+|y|^2)^{-\theta}. \tag*{(38)}

Substituting (38) in (36), and observing that 2θ−k−ν=ν2\theta-k-\nu=\nu, we see that the powers of ∣y∣|y| cancel, and we are left with

Eγ(x,y)=A0Γ(θ)2πk2+ν∣x∣1−m2sν(s2+∣y∣2)−θ.(39)\mathcal{E}_\gamma(x,y)=\frac{A_0\Gamma(\theta)}{2\pi^{\frac{k}{2}+\nu}}|x|^{1-\frac{m}{2}}s^\nu(s^2+|y|^2)^{-\theta}. \tag*{(39)}

Now, by (10) and (γ+1)ν=m−22(\gamma+1)\nu=\frac{m-2}{2}, we have ∣x∣1−m2sν=(γ+1)−ν|x|^{1-\frac{m}{2}}s^\nu=(\gamma+1)^{-\nu}, and by (11) and 2(γ+1)θ=Q−22(\gamma+1)\theta=Q-2, we have (s2+∣y∣2)−θ=(γ+1)2θNγ(x,y)2−Q(s^2+|y|^2)^{-\theta}=(\gamma+1)^{2\theta}N_\gamma(x,y)^{2-Q}. Substituting these identities and (35) in (39), and using 2θ−2ν=k2\theta-2\nu=k, we finally obtain

Eγ(x,y)=(γ+1)k−1Γ(m2)Γ(θ)4πm+k2Γ(ν+1)Nγ(x,y)2−Q.(40)\mathcal{E}_\gamma(x,y)=\frac{(\gamma+1)^{k-1}\Gamma\left(\frac{m}{2}\right)\Gamma(\theta)}{4\pi^{\frac{m+k}{2}}\Gamma(\nu+1)}N_\gamma(x,y)^{2-Q}. \tag*{(40)}

Since θ=Q−22(γ+1)\theta=\frac{Q-2}{2(\gamma+1)}, the constant in the right-hand side of (40) is Cγ,QC_{\gamma,Q} in (6), and this proves (4). In the next section we show that Cγ,Q=Cγ,Q∗C_{\gamma,Q}=C^*_{\gamma,Q}.

Computing the constant

In this section we show that Cγ,Q∗=Cγ,QC^*_{\gamma,Q}=C_{\gamma,Q}, where Cγ,Q∗C^*_{\gamma,Q} is defined by (8). This also follows from [15 Lemma 3.4] with p=2p=2 and Euclidean norms, after the change of variable y→y/2y\to y/2; for completeness we give the direct evaluation of the integral in (8). By (9) we have

1+Q+2γ2(γ+1)=θ+2,θ+2−k2=ν+2.1+\frac{Q+2\gamma}{2(\gamma+1)}=\theta+2,\qquad\theta+2-\frac{k}{2}=\nu+2.

Using the elementary formula ∫Rk(a+∣ξ∣2)−e dξ=πk2Γ(e−k2)Γ(e)ak2−e\int_{\mathbb{R}^k}(a+|\xi|^2)^{-e}\,d\xi=\pi^{\frac{k}{2}}\frac{\Gamma(e-\frac{k}{2})}{\Gamma(e)}a^{\frac{k}{2}-e}, valid for a>0a>0 and e>k2e>\frac{k}{2}, with a=1+∣x∣2(γ+1)a=1+|x|^{2(\gamma+1)} and e=θ+2e=\theta+2, we first integrate in yy in (8), obtaining

∫Rkdy[1+∣x∣2(γ+1)+(γ+1)2∣y∣2]θ+2=πk2(γ+1)kΓ(ν+2)Γ(θ+2)(1+∣x∣2(γ+1))−ν−2.\int_{\mathbb{R}^k}\frac{dy}{\left[1+|x|^{2(\gamma+1)}+(\gamma+1)^2|y|^2\right]^{\theta+2}} =\frac{\pi^{\frac{k}{2}}}{(\gamma+1)^k}\frac{\Gamma(\nu+2)}{\Gamma(\theta+2)} \left(1+|x|^{2(\gamma+1)}\right)^{-\nu-2}.

Next, we integrate in xx. The change of variable u=r2(γ+1)u=r^{2(\gamma+1)}, and the identity m+2γ2(γ+1)−1=ν\frac{m+2\gamma}{2(\gamma+1)}-1=\nu, give

∫Rm∣x∣2γ dx(1+∣x∣2(γ+1))ν+2=ωm−1∫0∞rm−1+2γ dr(1+r2(γ+1))ν+2=ωm−12(γ+1)∫0∞uν du(1+u)ν+2=ωm−12(γ+1)(ν+1).\begin{aligned} \int_{\mathbb{R}^{m}}\frac{|x|^{2\gamma}\,dx}{\left(1+|x|^{2(\gamma+1)}\right)^{\nu+2}}=\omega_{m-1}\int_{0}^{\infty}\frac{r^{m-1+2\gamma}\,dr}{\left(1+r^{2(\gamma+1)}\right)^{\nu+2}} \\ &=\frac{\omega_{m-1}}{2(\gamma+1)}\int_{0}^{\infty}\frac{u^{\nu}\,du}{(1+u)^{\nu+2}}=\frac{\omega_{m-1}}{2(\gamma+1)(\nu+1)}. \end{aligned}

where in the last equality we have used ∫0∞uν(1+u)−ν−2du=B(ν+1,1)=1ν+1\int_{0}^{\infty}u^{\nu}(1+u)^{-\nu-2}du=B(\nu+1,1)=\frac{1}{\nu+1}. Combining the last two formulas, we find

[Cγ,Q⋆]−1=(Q−2)(Q+2γ)πk2ωm−12(γ+1)k+1Γ(ν+1)Γ(θ+2).[C_{\gamma,Q}^{\star}]^{-1}=(Q-2)(Q+2\gamma)\frac{\pi^{\frac{k}{2}}\omega_{m-1}}{2(\gamma+1)^{k+1}}\frac{\Gamma(\nu+1)}{\Gamma(\theta+2)}.

Since (Q−2)(Q+2γ)=4(γ+1)2θ(θ+1)(Q-2)(Q+2\gamma)=4(\gamma+1)^2\theta(\theta+1) and Γ(θ+2)=θ(θ+1)Γ(θ)\Gamma(\theta+2)=\theta(\theta+1)\Gamma(\theta), we conclude

[Cγ,Q⋆]−1=2πk2ωm−1Γ(ν+1)(γ+1)k−1Γ(θ)=4πm+k2Γ(ν+1)(γ+1)k−1Γ(m2)Γ(θ).[C_{\gamma,Q}^{\star}]^{-1}=\frac{2\pi^{\frac{k}{2}}\omega_{m-1}\Gamma(\nu+1)}{(\gamma+1)^{k-1}\Gamma(\theta)}=\frac{4\pi^{\frac{m+k}{2}}\Gamma(\nu+1)}{(\gamma+1)^{k-1}\Gamma\left(\frac{m}{2}\right)\Gamma(\theta)}.

which is the reciprocal of Cγ,QC_{\gamma,Q} in (6). This proves Cγ,Q⋆=Cγ,QC_{\gamma,Q}^{\star}=C_{\gamma,Q}. Finally, by Lemma 3.1 applied to ψ=φ^(⋅,λ)\psi=\hat{\varphi}(\cdot,\lambda) and Fubini’s theorem, Eγ\mathcal{E}_{\gamma} is a fundamental solution of −Lγ-\mathcal{L}_{\gamma} with pole at the origin, and together with (40) this gives an independent proof of [13 Prop. 2.1].

Remark 5.1. When γ=1\gamma=1, Mehler’s formula for hλ(x,0,t)h_{\lambda}(x,0,t), see [18 Remark 3.3], and the change of variable λ↦λ4πt\lambda\mapsto\frac{\lambda}{4\pi t} in (19) give 2kp1((x,2y),t)=Gm2+k((x,y),t)2^{k}p_{1}((x,2y),t)=G_{\frac{m}{2}+k}((x,y),t), where Gα,βG_{\alpha,\beta} is the generalized Mehler kernel in [16 Prop. 1.2]. With m=2αm=2\alpha one has N1(x,2y)=(∣x∣4+16∣y∣2)1/4N_{1}(x,2y)=(|x|^{4}+16|y|^{2})^{1/4}, Q=2α+2kQ=2\alpha+2k, ν=α−12\nu=\frac{\alpha-1}{2} and θ=α+k−12\theta=\frac{\alpha+k-1}{2}, and by Legendre’s duplication formula 2k2^{k} times the constant in (6) equals the constant in [16 Theor. 1.3], and in [17 Corollary 1.3] when mm is the dimension of the horizontal layer of a group of Heisenberg type.

Declarations

Competing interests. The author has no competing interests to declare that are relevant to the content of this article.

Use of AI: The author acknowledges the assistance of AI in organizing and revising the manuscript, and in reviewing the proofs and the language. The research ideas and the mathematical arguments are exclusively the author’s, and he takes full responsibility for the content and the final text.

Declarations

Competing interests. The author has no competing interests to declare that are relevant to the content of this article.

Use of AI: The author acknowledges the assistance of AI in organizing and revising the manuscript, and in reviewing the proofs and the language. The research ideas and the mathematical arguments are exclusively the author’s, and he takes full responsibility for the content and the final text.

School of Mathematical and Statistical Sciences, Arizona State University

Email address, Nicola Garofalo: [email protected]

References

  1. [1]M. S. Baouendi, Sur une classe d’opérateurs elliptiques dégénérés. Bull. Soc. Math. France 95 (1967), 45-87.DOI
  2. [2]A. Banerjee & N. Garofalo, An improved Rellich type estimate for the eigenfunctions of the Baouendi-Grushin operator, to appear in J. Anal. Math.
  3. [3]W. Bauer, K. Furutani & C. Iwasaki, Spectral analysis and geometry of sub-Laplacian and related Grushin-type operators. Partial differential equations and spectral theory, 183-290, Oper. Theory Adv. Appl., 211, Birkhäuser/Springer Basel AG, Basel, 2011.DOI
  4. [4]W. Bauer, K. Furutani & C. Iwasaki, Fundamental solution of a higher step Grushin type operator. Adv. Math. 271 (2015), 188-234. 2, 3DOI
  5. [5]R. Beals, B. Gaveau & P. Greiner, Green’s functions for some highly degenerate elliptic operators. J. Funct. Anal. 165 (1999), no. 2, 407-429. 3DOI
  6. [6]S. Bochner & K. Chandrasekharan, Fourier Transforms, Annals of Mathematics Studies, no. 19, Princeton University Press, Princeton, N. J.; Oxford University Press, London, 1949. ix+219 pp. 4
  7. [7]O. Calin, D.-C. Chang, K. Furutani & C. Iwasaki, Heat kernels for elliptic and sub-elliptic operators. Methods and techniques. Applied and Numerical Harmonic Analysis. Birkhäuser/Springer, New York, 2011. xviii+433 pp. 3DOI
  8. [8]C.-H. Chang, D.-C. Chang, B. Gaveau, P. Greiner & H.-P. Lee, Geometric analysis on a step 2 Grusin operator. Bull. Inst. Math. Acad. Sin. (N.S.) 4 (2009), no. 2, 119-188. 3
  9. [9]E. B. Davies, Heat kernels and spectral theory. Cambridge Tracts in Mathematics, 92. Cambridge University Press, Cambridge, 1989. x+197 pp. 5
  10. [10]F. Dragoni, N. Garofalo, G. Giovannardi & P. Salani, A fundamental solution for a subelliptic operator in Finsler geometry. Indiana Univ. Math. J. 75 (2026), no. 1, 201-226. 1
  11. [11]A. Erdélyi, W. Magnus, F. Oberhettinger & F. G. Tricomi, Higher transcendental functions. Vol. I. Based on notes left by Harry Bateman. With a preface by Mina Rees. With a foreword by E. C. Watson. Reprint of the 1953 original. Robert E. Krieger Publishing Co., Inc., Melbourne, Fla., 1981. xiii+302 pp. 10
  12. [12]G. B. Folland, A fundamental solution for a subelliptic operator. Bull. Amer. Math. Soc. 79 (1973), 373-376. 1DOI
  13. [13]N. Garofalo, Unique continuation for a class of elliptic operators which degenerate on a manifold of arbitrary codimension. J. Differential Equations 104 (1993), no. 1, 117-146. 2, 3, 11DOI
  14. [14]N. Garofalo, Fractional thoughts. Contemp. Math., 723 American Mathematical Society, [Providence], RI, 2019, 1-135. 1
  15. [15]N. Garofalo, On an evolution equation in sub-Finsler geometry. Anal. Geom. Metr. Spaces 12 (2024), Paper No. 20240012. 3, 10arxiv.org/abs/2402.16745
  16. [16]N. Garofalo, Fractal Mehler kernels and nonlinear geometric flows. Preprint, arXiv:2512.2830. 3, 11
  17. [17]N. Garofalo & G. Tralli, Feeling the heat in a group of Heisenberg type. Adv. Math. 381 (2021), Paper No. 107635, 42 pp. 1, 3, 11DOI
  18. [18]N. Garofalo & G. Tralli, Heat kernels for a class of hybrid evolution equations. Potential Anal. 59 (2023), no. 2, 823-856. 2, 3, 11arxiv.org/abs/2007.10862
  19. [19]B. Gaveau, Principe de moindre action, propagation de la chaleur et estimées sous elliptiques sur certains groupes nilpotents. Acta Math. 139 (1977), no. 1-2, 95-153. 1DOI
  20. [20]I. S. Gradshteyn & I. M. Ryzhik, Table of integrals, series, and products. Seventh edition. Elsevier/Academic Press, Amsterdam, 2007. xiii+1117 pp. 9
  21. [21]V. V. Grushin, A certain class of hypoelliptic operators. (Russian) Mat. Sb. (N.S.) 83 (125) (1970), 456-473. 1
  22. [22]A. Hulanicki, The distribution of energy in the Brownian motion in the Gaussian field and analytic-hypoellipticity of certain subelliptic operators on the Heisenberg group. Studia Math. 56 (1976), no. 2, 165-173. 1
  23. [23]A. Kaplan, Fundamental solutions for a class of hypoelliptic PDE generated by composition of quadratic forms. Trans. Amer. Math. Soc. 258 (1980), no. 1, 147-153. 1DOI
  24. [24]N. N. Lebedev, Special functions and their applications. Revised edition, translated from the Russian and edited by R. A. Silverman. Unabridged and corrected republication. Dover Publications, Inc., New York, 1972. xii+308 pp. 7, 8
  25. [25]M. Reed & B. Simon, Methods of modern mathematical physics. IV. Analysis of operators. Academic Press, New York-London, 1978. xv+396 pp. 4
  26. [26]L. P. Rothschild & E. M. Stein, Hypoelliptic differential operators and nilpotent groups. Acta Math. 137 (1976), no. 3-4, 247-320. 3DOI
  27. [27]K. Stempak, A transform for the Grushin operator with applications. J. Differential Equations 481 (2026), Paper No. 114630, 23 pp. 3
  28. [28]G. N. Watson, A treatise on the theory of Bessel functions. Reprint of the second (1944) edition. Cambridge Mathematical Library. Cambridge University Press, Cambridge, 1995. viii+804 pp. 9

Paper details

vlnc:W_auh5tc8x
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI

View statement
Looking for related papers…
Contents