betaFeedback

WorksMathematics

Isospectral Density Design and Distinct Global Shape Transitions in Weighted Yukawa Droplets

Abstract

We study Euclidean perimeter plus Yukawa repulsion with a fixed radial source density and a fixed ordinary volume. For each sufficiently small positive screening parameter, we construct two smooth, positive, strictly increasing densities whose constrained shape Hessians at the unit ball agree for every coupling. Their source masses, Yukawa moments, and ball self-energies also agree. Just above their common first global transition threshold, all minimizers form one rotation orbit of tetrahedral shapes for one density and one orbit of sectoral threefold shapes for the other. Both transitions are continuous, and the minimizing shapes are strictly convex. We obtain the two densities by fixing the boundary data that determine the Hessian and changing the higher derivatives that determine the reduced quartic energy. We compute a positive quartic coefficient, restore the three scalar observables by interior corrections, and add a finite exterior density barrier to control all finite-perimeter competitors.

The question and the main result

The shape Hessian at a sphere determines its unstable modes. It need not determine the shapes that minimize the nonlinear energy. We construct two radial source densities with the same ball Hessian, the same source mass, Yukawa moment, and ball self-energy, and the same first global transition threshold. The two models have different minimizing shapes just above that threshold.

For κ>0\kappa> 0, let

Gκ(x)=e−κ∣x∣4π∣x∣,Fβ,ρ(E)=P(E)+βIρ(E),Iρ(E)=12∬E×Eρ(∣x∣)Gκ(x−y)ρ(∣y∣) dx dy.(1)G_\kappa(x) = \frac{e^{-\kappa|x|}}{4\pi|x|}, \qquad F_{\beta,\rho}(E) = P(E) + \beta I_\rho(E), \qquad I_\rho(E) = \frac{1}{2}\iint_{E \times E} \rho(|x|)G_\kappa(x-y)\rho(|y|)\,\mathrm{d}x\,\mathrm{d}y. \tag*{(1)}

Here PP is Euclidean perimeter, β≥0\beta\ge0, and ρ\rho is fixed in space. The admissible sets have finite perimeter and ordinary volume ∣E∣=∣B∣|E| = |B|, where B=B1(0)B = B_1(0). Sets are identified up to null sets. The integral is allowed to be +∞+\infty; such sets are irrelevant for minimization when β>0\beta> 0. At β=0\beta= 0 the energy means perimeter alone. The density weights both sources in the interaction. The perimeter and volume are unweighted.

Write VℓV_\ell for the real spherical harmonics of degree ℓ\ell, with the L2(S2,dω)L^2(S^2,\mathrm{d}\omega) inner product. The constrained Hessian at BB is diagonal on ⨁ℓ≥1Vℓ\bigoplus_{\ell\ge1} V_\ell. We use “isospectral” to mean equality of the Hessian operators on each degree. Define

M0(ρ)=4π∫01r2ρ(r) dr,Mκ(ρ)=∫01rsinh⁡(κr)ρ(r) dr,(2)M_0(\rho)=4\pi\int_0^1 r^2\rho(r)\,\mathrm{d}r,\qquad M_\kappa(\rho)=\int_0^1 r\sinh(\kappa r)\rho(r)\,\mathrm{d}r, \tag*{(2)}

and let βg(ρ)\beta_g(\rho) be the supremum of the positive couplings at which BB is a global minimizer. The construction below proves that this set of couplings is exactly (0,βg](0,\beta_g].

The two unit critical harmonics used throughout are

eT=1054π xyz,eS=3532π(x3−3xy2) on S2.(3)e_T=\sqrt{\frac{105}{4\pi}}\,xyz,\qquad e_S=\sqrt{\frac{35}{32\pi}}\left(x^3-3xy^2\right)\text{ on }S^2. \tag*{(3)}

Their full orthogonal stabilizers are the tetrahedral group TdT_d and the sectoral threefold group D3hD_{3h}, respectively. The associated droplets are three-dimensional. The term “sectoral” describes the spherical harmonic.

Theorem 1.1 (Identical linear and scalar data, distinct first global shapes). There exists κ0>0\kappa_0>0 such that, for every fixed 0<κ<κ00<\kappa<\kappa_0, there are two radial densities ρT,ρS\rho_T,\rho_S, smooth as functions on R3\mathbb{R}^3, and numbers β∗,ϵ>0\beta_*,\epsilon>0 with the following properties.

  1. The densities are positive, satisfy ρi′(r)>0\rho_i'(r)>0 for r>0r>0, coincide outside a compact annulus, and have quadratic growth at infinity. Their constrained ball Hessians agree for every β≥0\beta\geq0, and

M0(ρT)=M0(ρS),Mκ(ρT)=Mκ(ρS),IρT(B)=IρS(B).M_0(\rho_T)=M_0(\rho_S),\qquad M_\kappa(\rho_T)=M_\kappa(\rho_S),\qquad I_{\rho_T}(B)=I_{\rho_S}(B).
  1. For each density a global minimizer exists for every β>0\beta>0. The centered ball is the unique global minimizer for 0<β≤β∗0<\beta\leq\beta_*. Its Hessian is positive definite for 0<β<β∗0<\beta<\beta_* and has kernel exactly V3V_3 at β∗\beta_*. Also,

βg(ρT)=βg(ρS)=β∗,lim⁡κ↓0β∗(κ)=175.\beta_g(\rho_T)=\beta_g(\rho_S)=\beta_*,\qquad\lim_{\kappa\downarrow0}\beta_*(\kappa)=175.
  1. For β∗<β<β∗+ϵ\beta_*<\beta<\beta_*+\epsilon, all global minimizers of the iith model form exactly one SO(3)\mathrm{SO}(3) orbit of nonspherical, strictly convex C2,γC^{2,\gamma} domains, for any fixed 0<γ<10<\gamma<1. Their full orthogonal stabilizers are TdT_d for i=Ti=T and D3hD_{3h} for i=Si=S.

  1. Put μ=β−β∗\mu=\beta-\beta_*. Up to rotation, the boundary is ri(ω)3=1+3wi(ω)r_i(\omega)^3=1+3w_i(\omega) with

wi=tiei+OC2,γ(ti2),ti2=αμ4ai+O(μ2),Fβ,ρi(Ei)−Fβ,ρi(B)=−α2μ216ai+O(μ3),(4)w_i=t_i e_i+O_{C^{2,\gamma}}(t_i^2),\qquad t_i^2=\frac{\alpha\mu}{4a_i}+O(\mu^2),\qquad F_{\beta,\rho_i}(E_i)-F_{\beta,\rho_i}(B)=-\frac{\alpha^2\mu^2}{16a_i}+O(\mu^3), \tag*{(4)}

where α\alpha, aTa_T, aSa_S are positive. The first global transition is continuous, with different leading angular shapes.

The densities depend on the fixed screening parameter and the unit reference volume. We do not require one pair to have the same Hessians at all screening parameters or radii. The theorem compares two functionals; it makes no claim that both shape types minimize the same functional. The boundary-layer width, exterior barrier strength, and transition interval may depend on κ\kappa; no uniform bounds on these choices as κ↓0\kappa\downarrow0 are claimed.

Relation to earlier work. Frank [1] constructed nonspherical volume-constrained equilibria of the Coulomb liquid drop model through bifurcation from a ball. De Regibus, Grossi, and Musso [2] study further bifurcations for nonlocal isoperimetric problems. For screened interactions, Bronsard, Merlet, and Pegon [3] establish connected nonball minimizers in Yukawa and truncated Coulomb models, with nonradial minimizers for the latter. Bronsard, DeMason, and Topaloglu [4] study existence and sharp ball stability for the unweighted Yukawa problem. These papers concern bifurcation or minimization for prescribed models. Here we choose the source density while fixing the ball Hessian and three scalar observables.

Weights and confinement have also been studied in related liquid drop problems. Alama, Bronsard, Topaloglu, and Zuniga [5] place the density in the perimeter, while Alama, Bronsard, Choksi, and Topaloglu [6] use a background potential. Their models differ from (1), whose ordinary perimeter and volume are fixed while a radial source density multiplies both factors of the interaction.

The O(3)O(3) bifurcation theory, the two orbit types, and the quartic invariants of harmonic cubics are classical [7, 8, 9, 10]. We realize the two orbit types using positive increasing Yukawa source densities with the same spectral and scalar data. We also prove that the first global thresholds agree and classify all minimizers just above them. We include the invariant calculations with the normalizations used in the proof.

Outline of the proof. The ball Hessian depends only on the boundary value, the first boundary derivative, and one interior Yukawa moment. We change higher boundary derivatives in a thin layer and use three interior corrections to restore the observables. We choose the second derivative so that the radial quartic coefficient is positive, then choose the third derivative to select either angular orbit. A common finite increase of the exterior density excludes distant competitors and leaves all near-ball energies unchanged.

Linear data and exact density realization

The complete ball Hessian

Let u=Gκ∗(ρ1B)u=G_\kappa\ast(\rho\mathbf{1}_B), q=ρ(1)q=\rho(1), p=ρ′(1)p=\rho'(1), and U=u(1)U=u(1). The radial Green formula gives

u(r)=e−κrκrMκ(ρ)(r≥1),U=e−κκMκ(ρ),u′(1)=−(κ+1)U.(5)u(r)=\frac{e^{-\kappa r}}{\kappa r}M_\kappa(\rho)\quad(r\ge1),\qquad U=\frac{e^{-\kappa}}{\kappa}M_\kappa(\rho),\qquad u'(1)=-(\kappa+1)U. \tag*{(5)}

Here u′(1)u'(1) differentiates the fixed ball potential in space.

Lemma 2.1 (Spectrum and its finite data). The ball is volume-constrained critical for every β\beta, and the Hessian eigenvalue on VℓV_\ell, with multiplicity 2ℓ+12\ell+1, is

λℓ(β)=ℓ(ℓ+1)−2+β{q2sℓ+Cρ},Cρ=U[p−(κ+1)q],ℓ≥1.(6)\lambda_\ell(\beta)=\ell(\ell+1)-2+\beta\{q^2s_\ell+C_\rho\},\qquad C_\rho=U[p-(\kappa+1)q],\qquad\ell\ge1. \tag*{(6)}

where sℓ=Iℓ+1/2(κ)Kℓ+1/2(κ)s_\ell=I_{\ell+1/2}(\kappa)K_{\ell+1/2}(\kappa). For fixed κ\kappa, sℓs_\ell decreases strictly to zero, and

12ℓ+1+κ2[(2ℓ+3)−1+(2ℓ−1)−1]≤sℓ≤12ℓ+1(ℓ≥1).(7)\frac{1}{2\ell+1+\kappa^2\left[(2\ell+3)^{-1}+(2\ell-1)^{-1}\right]}\le s_\ell\le\frac{1}{2\ell+1}\qquad(\ell\ge1). \tag*{(7)}

The data q,p,Mκq,p,M_\kappa determine the whole Hessian for every coupling.

Proof. The first variations are DP[h]=∫HhDP[h]=\int Hh, DI[h]=∫ρuEhDI[h]=\int\rho uEh, and D∣E∣[h]=∫hD|E|[h]=\int h. At the sphere the multiplier is 2+βqU2+\beta qU. The second variation of the Lagrangian is

∫S2(∣∇h∣2−2h2)+βCρ∫S2h2+βq2∬S2×S2Gκ(ω−σ)h(ω)h(σ) dω dσ.\int_{\mathbb{S}^{2}} \left(|\nabla h|^{2}-2h^{2}\right)+\beta C_{\rho}\int_{\mathbb{S}^{2}} h^{2}+\beta q^{2}\iint_{\mathbb{S}^{2}\times\mathbb{S}^{2}}G_{\kappa}(\omega-\sigma)h(\omega)h(\sigma)\,\mathrm{d}\omega\,\mathrm{d}\sigma.

The local term is ∂r(ρu)(1)\partial_{r}(\rho u)(1); the remaining term is the single-layer operator. On VℓV_{\ell} its eigenvalue is the reciprocal of the jump between the interior and exterior logarithmic radial derivatives. The normalized solutions are proportional to r−1/2Iℓ+1/2(κr)r^{-1/2}I_{\ell+1/2}(\kappa r) and r−1/2Kℓ+1/2(κr)r^{-1/2}K_{\ell+1/2}(\kappa r), respectively. Their Wronskian gives sℓs_{\ell} as stated.

The same eigenvalue has the variational formula: sℓ−1s_{\ell}^{-1} is the minimum over f(1)=1f(1)=1 of

∫0∞{r2∣f′∣2+ℓ(ℓ+1)∣f∣2+κ2r2∣f∣2} dr.\int_{0}^{\infty}\left\{r^{2}|f'|^{2}+\ell(\ell+1)|f|^{2}+\kappa^{2}r^{2}|f|^{2}\right\}\,\mathrm{d}r.

Comparing degrees at the higher-degree minimizer proves strict monotonicity. Completing squares on (0,1)(0,1) and (1,∞)(1,\infty) gives the lower energy bound 2ℓ+12\ell+1. Testing with rℓr^{\ell} inside and r−ℓ−1r^{-\ell-1} outside gives the upper bound on the minimum and proves (7). These trial functions have finite massive energy for ℓ≥1\ell\geq1.

At a known nonzero coupling, two distinct labeled eigenvalues determine q2q^{2} and CρC_{\rho}: if Tℓ=(λℓ−ℓ(ℓ+1)+2)/βT_{\ell}=(\lambda_{\ell}-\ell(\ell+1)+2)/\beta, then q2=(Tℓ−Tj)/(sℓ−sj)q^{2}=(T_{\ell}-T_{j})/(s_{\ell}-s_{j}) and Cρ=Tℓ−q2sℓC_{\rho}=T_{\ell}-q^{2}s_{\ell}. We preserve the stronger data q,p,Mκq,p,M_{\kappa}. The translation modes V1V_{1} are retained, since the density is fixed in space.

Lemma 2.2 (A common degree-three crossing). Set

ρ0(r)=a0+b0r2,a0=7−12,b0=3−72.(8)\rho_{0}(r)=a_{0}+b_{0}r^{2},\qquad a_{0}=\frac{\sqrt{7}-1}{2},\qquad b_{0}=\frac{3-\sqrt{7}}{2}. \tag*{(8)}

For all sufficiently small positive κ\kappa, this density has a first positive Hessian crossing at β∗(κ)\beta_{*}(\kappa) with kernel exactly V3V_{3}. At the crossing the complement is uniformly coercive in H1H^{1}, and β∗(κ)→175\beta_{*}(\kappa)\to175.

Proof. At κ=0\kappa=0, q=1q=1, p=3−7p=3-\sqrt{7}, U=(2+7)/15U=(2+\sqrt{7})/15, and Cρ=U(p−1)=−1/5C_{\rho}=U(p-1)=-1/5. This gives

λℓ(β)=ℓ(ℓ+1)−2+β(12ℓ+1−15),λℓ(175)=(ℓ−3)(2ℓ2+9ℓ−46)2ℓ+1.\lambda_{\ell}(\beta)=\ell(\ell+1)-2+\beta\left(\frac{1}{2\ell+1}-\frac{1}{5}\right),\qquad\lambda_{\ell}(175)=\frac{(\ell-3)(2\ell^{2}+9\ell-46)}{2\ell+1}.

For integer ℓ≥1\ell\geq1 this is zero only at ℓ=3\ell=3 and positive otherwise. The degree-three slope is −2/35-2/35, while the degree-one slope is 2/15>02/15>0. At small κ>0\kappa>0 set

α=−(s3+Cρ0)>0,β∗=10/α.(9)\alpha=-(s_{3}+C_{\rho_{0}})>0,\qquad\beta_{*}=10/\alpha. \tag*{(9)}

Continuity preserves the positive gap for every finite set of noncritical degrees. For the remaining degrees, sℓ≥0s_{\ell}\geq0 and Cρ0C_{\rho_{0}} and β∗\beta_{*} are bounded, so λℓ(β∗)≥ℓ(ℓ+1)−C\lambda_{\ell}(\beta_{*})\geq\ell(\ell+1)-C uniformly. This proves the H1H^{1} gap for all noncritical degrees. Since every eigenvalue is affine in β\beta, positivity at β=0\beta=0 and β=β∗\beta=\beta_{*}, together with the positive degree-one slope, proves positivity for 0<β<β∗0<\beta<\beta_{*}. The degree-three eigenvalue is exactly −α(β−β∗)-\alpha(\beta-\beta_{*}).

Changing finite boundary derivatives with exact compensation

Lemma 2.3 (Finite-jet realization). Fix κ>0\kappa>0. Given arbitrary finite values for ρ′′(1)\rho''(1), ρ′′′(1)\rho'''(1), there is a smooth positive radial density ρ\rho realizing them and preserving

ρ(1),ρ′(1),M0,Mκ,I(B)\rho(1),\quad\rho'(1),\quad M_{0},\quad M_{\kappa},\quad I(B)

of the baseline (8). Its difference from the baseline has compact support away from the origin, and can be arbitrarily small in C1C^{1}. The density satisfies ρ′(r)>0\rho'(r)>0 for every r>0r>0.

Proof. With a cutoff χ=1\chi=1 near zero, first add

dϵ(r)=χ((r−1)/ϵ){a2(r−1)2+b6(r−1)3}.d_{\epsilon}(r)=\chi((r-1)/\epsilon)\left\{\frac{a}{2}(r-1)^2+\frac{b}{6}(r-1)^3\right\}.

The constants a,ba,b are the prescribed increments. This changes the second and third derivatives, leaves the value and first derivative fixed, and tends to zero in C1C^1 as ϵ→0\epsilon\to0 for fixed a,ba,b.

Choose an interval J⋐(0,1)J\Subset(0,1) disjoint from this layer. On Cc∞(J)C^\infty_c(J) consider the derivatives of the three observables:

L0(d)=4π∫r2d,LY(d)=∫rsinh⁡(κr)d,Lu(d)=4π∫r2u0d.L_0(d)=4\pi\int r^2d,\qquad L_Y(d)=\int r\sinh(\kappa r)d,\qquad L_u(d)=4\pi\int r^2u_0d.

where u0=Gκ∗(ρ01B)u_0=G_\kappa*(\rho_0\mathbf{1}_B). These functionals are independent. The first two weights are independent because sinh⁡(κr)/r\sinh(\kappa r)/r is strictly increasing. Dependence of the third would imply u0=A+Bsinh⁡(κr)/ru_0=A+B\sinh(\kappa r)/r on JJ; applying −Δ+κ2-\Delta+\kappa^2 would make ρ0\rho_0 constant there, a contradiction. Choose three test functions d1,d2,d3d_1,d_2,d_3 with invertible derivative matrix. The finite-dimensional implicit function theorem applied to

(c1,c2,c3)⟼(M0,Mκ,I(B))(ρ0+dϵ+∑cjdj)(c_1,c_2,c_3)\longmapsto(M_0,M_\kappa,I(B))\left(\rho_0+d_\epsilon+\sum c_jd_j\right)

gives small corrections restoring all three values exactly. We apply the theorem with the density in C0C^0 on the fixed ball. The first two functionals are linear, and I(B)I(B) is a continuous quadratic form. The higher derivatives of dϵd_\epsilon need not be small. The corrections are C1C^1 small and supported away from the boundary, so they leave the prescribed derivatives unchanged. Positivity and strict increase follow from the positive minima of ρ0\rho_0 and ρ0′\rho'_0 on the compact perturbation support. Outside that support the baseline is unchanged.

The total perturbation preserves MκM_\kappa. An individual interior correction may have nonzero moment to compensate for the boundary layer. If an interior perturbation dd supported in BaB_a has zero moment, then Gκ∗d=0G_\kappa*d=0 outside BaB_a by (5). For every E⊃BaE\supset B_a, the resulting change in I(E)I(E) is independent of EE, so relative shape energies are unchanged.

Exact reduction and quartic control

Reduction in a strong space, comparison in a weaker norm

Use the exact volume coordinate

rw3=1+3w,∫S2w=0,Ew={rω:0≤r<rw(ω)}.(10)r_w^3=1+3w,\qquad\int_{\mathbb{S}^2}w=0,\qquad E_w=\{r\omega:0\le r<r_w(\omega)\}. \tag*{(10)}

Then ∣Ew∣=∣B∣|E_w|=|B|, and rw−1=w+O(w2)r_w-1=w+O(w^2). Fix 0<γ<10<\gamma<1, put V=V3V=V_3, and let WW be its L2L^2 orthogonal complement among mean-zero functions. For μ=β−β∗\mu=\beta-\beta_* define E(w,μ)=Fβ∗+μ,ρ(Ew)−Fβ∗+μ,ρ(B)\mathcal{E}(w,\mu)=F_{\beta_*+\mu,\rho}(E_w)-F_{\beta_*+\mu,\rho}(B).

Lemma 3.1 (Exact reduction and energy comparison). For every fixed smooth radial density with the ball data of Lemma 2.2, there is a smooth, O(3)O(3)-equivariant map z:V×R→W∩C2,γz:V\times\mathbb{R}\to W\cap C^{2,\gamma} near (0,0)(0,0) such that

z(0,μ)=0,Dvz(0,μ)=0,PW∇E(v+z(v,μ),μ)=0.z(0,\mu)=0,\qquad D_vz(0,\mu)=0,\qquad P_W\nabla\mathcal{E}(v+z(v,\mu),\mu)=0.

The reduced energy Ψ(v,μ)=E(v+z(v,μ),μ)\Psi(v,\mu)=\mathcal{E}(v+z(v,\mu),\mu) is even in vv and has the expansion

Ψ(v,μ)=−α2μ∥v∥22+Qρ(v)+O(∥v∥26+∣μ∣∥v∥24).(11)\Psi(v,\mu)=-\frac{\alpha}{2}\mu\lVert v\rVert_2^2+Q_\rho(v)+O(\lVert v\rVert_2^6+|\mu|\lVert v\rVert_2^4). \tag*{(11)}

with the corresponding bounds after up to two vv derivatives. For every sufficiently small mean-zero C1C^1 graph, write w=v+z(v,μ)+yw=v+z(v,\mu)+y, y∈Wy\in W. There are c,C>0c,C>0 such that

Ψ(v,μ)+c∥y∥H12≤E(w,μ)≤Ψ(v,μ)+C∥y∥H12.(12)\Psi(v,\mu)+c\|y\|_{H^1}^2\leq\mathcal{E}(w,\mu)\leq\Psi(v,\mu)+C\|y\|_{H^1}^2. \tag*{(12)}

A positive definite QρQ_\rho then makes the critical ball a strict local minimizer among small C1C^1 equal-volume graphs.

Proof. The exact area and interaction first variations are

P(Ew)=∫rw21+rw−6∣∇w∣2,DIρ(w)[η]=∫ρ(rw)uw(rwω)η,uw=Gκ∗(ρ1Ew).(13)P(E_w)=\int r_w^2\sqrt{1+r_w^{-6}|\nabla w|^2},\qquad DI_\rho(w)[\eta]=\int\rho(r_w)u_w(r_w\omega)\eta,\quad u_w=G_\kappa*(\rho1_{E_w}). \tag*{(13)}

The projected Euler operator is smooth from mean-zero C2,γC^{2,\gamma} to mean-zero C0,γC^{0,\gamma}. For the nonlocal term, we account for the moving singularity as follows. Extend the boundary map to a smoothly ww-dependent, uniformly bi-Lipschitz map Φw\Phi_w of space, equal to the identity near zero and outside a fixed annulus. The potential trace becomes

∫BGκ(Φw(ω)−Φw(y))ρ(Φw(y))det⁡DΦw(y) dy.\int_B G_\kappa(\Phi_w(\omega)-\Phi_w(y))\rho(\Phi_w(y))\det D\Phi_w(y)\,dy.

Each parameter derivative of the kernel contains differences of parameter derivatives of Φw\Phi_w. Each such difference is O(∣ω−y∣)O(|\omega-y|), with a bound multilinear in the C2,γC^{2,\gamma} norms of the directions. So the differentiated integrands are bounded by C∣ω−y∣−1C|\omega-y|^{-1}; one tangential derivative is bounded by C∣ω−y∣−2C|\omega-y|^{-2}. Both are uniformly integrable in three dimensions. Taylor’s formula along parameter segments gives the same bounds for Fréchet remainders, with a factor tending to zero. Integration then controls these remainders in the Lipschitz norm of the trace, and so in C0,γC^{0,\gamma}. This proves smoothness to every finite order. Smoothness of the local area operator is immediate from (13).

The linearization on WW is −ΔS2−2-\Delta_{S^2}-2 plus a constant and a lower-order single-layer operator. It is Fredholm of index zero between these Hölder spaces, and the spectral gap makes it invertible by elliptic estimates. The implicit function theorem gives zz. The ball is critical for all μ\mu and the linearization preserves VV and WW, proving the two vanishing identities. Uniqueness gives equivariance. Central inversion acts as −1-1 on V3V_3, so Ψ\Psi is even. Its quadratic term is exactly −αμ∥v∥22/2-\alpha\mu\|v\|_2^2/2, which proves (11) by Taylor expansion.

To compare C1C^1 graphs, we prove that the Hessian is continuous as an H1H^1 bilinear form. Differentiating (13) gives, with Xw=rwωX_w=r_w\omega,

D2Iρ(w)[ξ,η]=∫ρ′(rw)uw(Xw)+ρ(rw)ω⋅∇uw(Xw)rw2ξη dω+∬ρ(rw(ω))ρ(rw(σ))Gκ(Xw(ω)−Xw(σ))ξ(ω)η(σ) dω dσ.(14)D^2I_\rho(w)[\xi,\eta]=\int\frac{\rho'(r_w)u_w(X_w)+\rho(r_w)\omega\cdot\nabla u_w(X_w)}{r_w^2}\xi\eta\,d\omega +\iint\rho(r_w(\omega))\rho(r_w(\sigma))G_\kappa(X_w(\omega)-X_w(\sigma))\xi(\omega)\eta(\sigma)\,d\omega\,d\sigma. \tag*{(14)}

The derivative in the first line is radial. The volume potential has a continuous gradient across the boundary. If δ=∥w∥C0\delta=\|w\|_{C^0}, then the source difference is bounded and supported on a set of volume O(δ)O(\delta); the kernels ∣x∣−1|x|^{-1} and ∣x∣−2|x|^{-2} give ∥uw−u0∥∞≤Cδ2/3\|u_w-u_0\|_\infty\leq C\delta^{2/3} and ∥∇uw−∇u0∥∞≤Cδ1/3\|\nabla u_w-\nabla u_0\|_\infty\leq C\delta^{1/3}. The first coefficient in (14) then converges uniformly. For the second line,

∣Xw(ω)−Xw(σ)∣≥min⁡rw ∣ω−σ∣,∣(Xw−Id⁡)(ω)−(Xw−Id⁡)(σ)∣≤C∥w∥C1∣ω−σ∣.|X_w(\omega)-X_w(\sigma)|\geq\min r_w\,|\omega-\sigma|,\qquad|(X_w-\operatorname{Id})(\omega)-(X_w-\operatorname{Id})(\sigma)|\leq C\|w\|_{C^1}|\omega-\sigma|.

The difference from the ball kernel is at most C∥w∥C1∣ω−σ∣−1C\|w\|_{C^1}|\omega-\sigma|^{-1}. Its L2L^2 operator norm tends to zero by Schur’s bound. The perimeter Hessian has coefficients continuous in w,∇ww,\nabla w. Combining these estimates,

∣(D2E(w,μ)−D2E(0,0))[ξ,η]∣≤o(1)∥ξ∥H1∥η∥H1(15)\left|(D^2\mathcal{E}(w,\mu)-D^2\mathcal{E}(0,0))[\xi,\eta]\right|\leq o(1)\|\xi\|_{H^1}\|\eta\|_{H^1} \tag*{(15)}

as ∥w∥C1+∣μ∣→0\lVert w\rVert_{C^1} + |\mu| \to0. The integral formulas and Taylor’s formula extend to C1C^1 graphs by smooth approximation and the displayed uniform bounds.

The spectral gap is coercive on WW in H1H^1, so (15) preserves that coercivity along the segment v+z(v,μ)+syv+z(v,\mu)+sy. The first derivative along yy vanishes at s=0s=0. Integrating the second derivative proves (12). If Qρ(v)≥a∥v∥24Q_\rho(v) \geq a\lVert v\rVert_2^4, then at μ=0\mu=0 this gives E(w,0)≥(a/2)∥v∥22+c∥y∥H12E(w,0) \geq(a/2)\lVert v\rVert_2^2+c\lVert y\rVert_{H^1}^2 after shrinking the neighborhood. Equality forces w=0w=0.

The graph topology and the coercivity norm differ, as in the shape stability results of [11]. Here the needed continuity estimate is (15).

One positive radial coefficient and one angular control

Set N(v)=∥v∥24\mathcal{N}(v)=\lVert v\rVert_2^4 and J(v)=∫S2v4\mathcal{J}(v)=\int_{S^2}v^4. The invariant calculation in § 4 shows that every quartic O(3)O(3) invariant on V3V_3 is uniquely AN+BJA\mathcal{N}+B\mathcal{J}.

Proposition 3.2 (The two quartic forms needed for the construction). For every sufficiently small fixed positive κ\kappa there exist two densities satisfying Lemma 2.3 whose reduced quartics are

QT=AκN+bκJ,QS=AκN−bκJ,Aκ>0,0<bκ<Aκ/jS,jS=175286π.(16)Q_T=A_\kappa\mathcal{N}+b_\kappa\mathcal{J},\qquad Q_S=A_\kappa\mathcal{N}-b_\kappa\mathcal{J},\qquad A_\kappa>0,\qquad0<b_\kappa<A_\kappa/j_S,\qquad j_S=\frac{175}{286\pi}. \tag*{(16)}

Both can be realized while fixing ρ′′(1)=ρ0′′(1)+1\rho''(1)=\rho_0''(1)+1.

Proof. First fix this second derivative and set ρ′′′(1)=0\rho'''(1)=0, using Lemma 2.3. Finite-order shape derivatives depend only on the boundary derivatives of the required orders and the ball Yukawa moment. This also follows from the signed-shell argument below. The coefficient AκA_\kappa of N\mathcal{N} does not depend on the particular interior correction. The direct transmission calculation in Appendix § 7 gives

lim⁡κ↓0Aκ=663451007−842683151672704π>0.(17)\lim_{\kappa\downarrow0}A_\kappa=\frac{66345100\sqrt{7}-84268315}{1672704\pi}>0. \tag*{(17)}

It is positive because 7>2\sqrt{7}>2. The coefficients involve only finitely many homogeneous radial solutions and finitely many boundary derivatives, and are continuous as κ↓0\kappa\downarrow0. So Aκ>0A_\kappa>0 for small positive κ\kappa.

To change the third derivative by η\eta, use another density with the same second and lower derivatives and the same moment. Write its difference as dd, let h=rw−1h=r_w-1, σh=1Ew−1B\sigma_h=\mathbf{1}_{E_w}-\mathbf{1}_B, and put ud=Gκ∗(d1B)u_d=G_\kappa*(d\mathbf{1}_B). Since Mκ(d)=0M_\kappa(d)=0, ud(1)=ud′(1)=0u_d(1)=u_d'(1)=0. The radial equation (rud)′′−κ2(rud)=−rd(ru_d)''-\kappa^2(ru_d)=-rd implies ud(1+s)=O(∣s∣5)u_d(1+s)=O(|s|^5) inside, while it vanishes outside. The exact difference of the relative interaction energies is

∫σh{duB+(ρ+d)ud}+⟨dσh,Gκ∗(ρσh)⟩+12⟨dσh,Gκ∗(dσh)⟩.(18)\int\sigma_h\{d u_B+(\rho+d)u_d\}+\langle d\sigma_h,G_\kappa*(\rho\sigma_h)\rangle+\frac{1}{2}\langle d\sigma_h,G_\kappa*(d\sigma_h)\rangle. \tag*{(18)}

Here d(1+s)=ηs3/6+O(s4)d(1+s)=\eta s^3/6+O(s^4). The first contribution is Uη∫h4/24+O(∥h∥∞5)U_\eta\int h^4/24+O(\lVert h\rVert_\infty^5). A spherical shell of thickness δ\delta has Coulomb potential O(δ)O(\delta) uniformly; since Gκ≤G0G_\kappa\leq G_0, the cross terms in (18) are O(∥h∥∞5)O(\lVert h\rVert_\infty^5). So the first changed energy term is β∗UηJ(v)/24\beta_*U_\eta\mathcal{J}(v)/24 at degree four. More generally the same argument shows that if the boundary derivatives through order three and the moment agree, all derivatives of relative energy through order four agree.

The quadratic complement correction is determined by the cubic energy, which has not changed. The two reduction graphs first differ at order three, and the energy cost of that difference is order six. The change in the reduced quartic is

Qρ+d−Qρ=β∗U24ηJ.(19)Q_{\rho+d}-Q_\rho=\frac{\beta_*U}{24}\eta\mathcal{J}. \tag*{(19)}

Since β∗U>0\beta_* U > 0, this sets its J\mathcal{J} coefficient to either bκb_\kappa or −bκ-b_\kappa, for example with bκ=Aκ/(2jS)b_\kappa= A_\kappa/(2j_S). Both prescribed third derivatives are finite for the fixed κ\kappa and are realized by Lemma 2.3. We take the limit only in the local coefficient formula. The interior corrections need not depend continuously on κ\kappa at zero.

Angular selection and the unique near-ball minimizing orbit

We compute the angular bounds and Hessians for harmonic cubics in our normalization. The invariant theory and orbit types are described in [7, 8, 9, 10].

Two invariants and their extremal orbits

Write v=Tijkxixjxkv = T_{ijk}x_i x_j x_k with TT symmetric and trace-free, and set

Aij=TikℓTjkℓ,I=(tr⁡A)2,K=tr⁡(A2),Qℓ(v)=∥Πℓ(v2)∥22.A_{ij} = T_{ik\ell}T_{jk\ell}, \qquad I = (\operatorname{tr} A)^2, \qquad K = \operatorname{tr}(A^2), \qquad Q_\ell(v) = \lVert\Pi_\ell(v^2)\rVert_2^2.

Repeated indices are summed.

Lemma 4.1 (Quartic invariants, equality orbits, and angular Hessians). The quartic O(3)O(3) invariants on V3V_3 have basis (N,J)(\mathcal{N},\mathcal{J}), and

jT:=315572π≤J(v)≤175286π=:jS(∥v∥2=1).(20)j_T := \frac{315}{572\pi} \leq\mathcal{J}(v) \leq\frac{175}{286\pi} =: j_S \qquad(\lVert v\rVert_2 = 1). \tag*{(20)}

The equality sets are respectively SO(3)eTSO(3)e_T and SO(3)eSSO(3)e_S. On the three-dimensional normal space to either orbit inside the unit sphere of V3V_3, the Hessian of J\mathcal{J} has eigenvalues

eT:56143π  (triple),eS:−63143π  (simple),−42143π  (double).(21)e_T : \frac{56}{143\pi}\;(\text{triple}), \qquad e_S : -\frac{63}{143\pi}\;(\text{simple}), \qquad-\frac{42}{143\pi}\;(\text{double}). \tag*{(21)}

Proof. The complexified V3V_3 has rotation weights −3,−2,…,3-3,-2,\ldots,3. In its fourth symmetric power the multiplicities at weights zero and one are respectively 18 and 16: count nondecreasing quadruples in {0,…,6}\{0,\ldots,6\} of sums 12 and 13. Their difference, two, is the multiplicity of the trivial representation. Quartics are even under central inversion, so the SO(3)SO(3) and O(3)O(3) invariant spaces coincide. The invariants I,KI,K are independent, having ratios K/I=1/3K/I = 1/3 for xyzxyz and 1/21/2 for z3−3zy2z^3 - 3zy^2.

The moment formula

∫S2x2ay2bz2c dω=4π(2a−1)!!(2b−1)!!(2c−1)!!(2a+2b+2c+1)!!(22)\int_{S^2} x^{2a}y^{2b}z^{2c}\,\mathrm{d}\omega= 4\pi\frac{(2a-1)!!(2b-1)!!(2c-1)!!}{(2a+2b+2c+1)!!} \tag*{(22)}

and vanishing of odd-coordinate moments give

∥v∥22=8π35tr⁡A,J(v)=16π5005(7I+6K),Q2=32π735(K−I/3),J=jTN+63143Q2.(23)\lVert v\rVert_2^2 = \frac{8\pi}{35}\operatorname{tr} A,\qquad\mathcal{J}(v) = \frac{16\pi}{5005}(7I+6K),\qquad Q_2 = \frac{32\pi}{735}(K-I/3),\qquad\mathcal{J} = j_T\mathcal{N}+\frac{63}{143}Q_2. \tag*{(23)}

For example Π2(v2)=27(A−13tr⁡AId⁡)ijxixj\Pi_2(v^2) = \frac{2}{7}(A-\frac{1}{3}\operatorname{tr} A\operatorname{Id})_{ij}x_i x_j; this follows by applying the Laplacian twice to the homogeneous polynomial v2v^2. These identities prove that (N,J)(\mathcal{N},\mathcal{J}) is a basis.

The matrix AA is nonnegative, so K≥I/3K \geq I/3. For the upper bound choose a positive maximum of vv as north pole and diagonalize its tangential quadratic part. Then T133=T233=T123=0T_{133}=T_{233}=T_{123}=0. Put a=T333>0a=T_{333}>0, b=T113b=T_{113}, p=T111p=T_{111}, q=T112q=T_{112}, s=p2+q2s=p^2+q^2. The trace conditions give

A=(2(s+b2)0p(a+2b)02(s+(a+b)2)q(a+2b)p(a+2b)q(a+2b)a2+b2+(a+b)2),I−2K=12{b2(a+b)2+a2(p2+q2)}.A=\begin{pmatrix} 2(s+b^2) & 0 & p(a+2b)\\ 0 & 2(s+(a+b)^2) & q(a+2b)\\ p(a+2b) & q(a+2b) & a^2+b^2+(a+b)^2 \end{pmatrix}, \qquad I-2K=12\{b^2(a+b)^2+a^2(p^2+q^2)\}.

So K≤I/2K \le I/2, with equality exactly when p=q=0p=q=0 and b=0b=0 or b=−ab=-a; both give rotations and multiples of z3−3zy2z^3-3zy^2. If AA is scalar, then b=−a/2b=-a/2 and p2+q2=a2/2p^2+q^2=a^2/2. A rotation about the pole makes q=0q=0, p=a/2p=a/\sqrt{2}, giving

a[z3−32(x2+y2)z+12(x3−3xy2)]=33aXYZ,a\left[z^3-\frac{3}{2}(x^2+y^2)z+\frac{1}{\sqrt{2}}(x^3-3xy^2)\right]=3\sqrt{3}aXYZ,

where X=2/3x+z/3X=\sqrt{2/3}x+z/\sqrt{3}, Y=−x/6+y/2+z/3Y=-x/\sqrt{6}+y/\sqrt{2}+z/\sqrt{3}, and Z=−x/6−y/2+z/3Z=-x/\sqrt{6}-y/\sqrt{2}+z/\sqrt{3} are orthogonal coordinates. This proves the lower equality classification. Formula (23) now gives (20). The negatives of both representatives are proper rotations of them: combine a coordinate interchange with central inversion for eTe_T, and use a π/3\pi/3 rotation about zz for eSe_S.

At a unit critical harmonic ee the constrained angular Hessian is

He[h,h]=12∫e2h2−4J(e)∥h∥2,⟨e,h⟩=0.(24)H_e[h,h]=12\int e^2h^2-4\mathcal{J}(e)\lVert h\rVert^2,\qquad\langle e,h\rangle=0. \tag*{(24)}

At eTe_T, a transverse orthonormal basis consists of 7x(5x2−3∣x∣2)/(4π)\sqrt{7}x(5x^2-3|x|^2)/(4\sqrt{\pi}) and its cyclic partners. At eSe_S, use

74π(5z3−3z∣x∣2),2132πx(5z2−∣x∣2),2132πy(5z2−∣x∣2).\frac{\sqrt{7}}{4\sqrt{\pi}}(5z^3-3z|x|^2),\qquad\sqrt{\frac{21}{32\pi}}x(5z^2-|x|^2),\qquad\sqrt{\frac{21}{32\pi}}y(5z^2-|x|^2).

The corresponding diagonal moments ∫e2h2\int e^2h^2 are

eT:3711716π in each direction;eS:2871716π, 739π, 739π.e_T:\quad\frac{371}{1716\pi}\text{ in each direction};\qquad e_S:\quad\frac{287}{1716\pi},\ \frac{7}{39\pi},\ \frac{7}{39\pi}.

All mixed moments vanish by parity. Equation (24) gives (21). The three rotational directions are exact zero modes by invariance. Their polynomial spans are x(z2−y2),y(x2−z2),z(y2−x2)x(z^2-y^2),y(x^2-z^2),z(y^2-x^2) at eTe_T, and 3x2y−y3,xyz,z(x2−y2)3x^2y-y^3,xyz,z(x^2-y^2) at eSe_S. These are orthogonal to the transverse vectors displayed above. Since the unit sphere has dimension six and each orbit dimension three, no further directions remain.

From quartic preference to exact minimizing shapes

Proposition 4.2 (Unique near-ball minimum modulo rotations). Suppose the reduced quartic is one of (16), and let e=eTe=e_T or e=eSe=e_S for the chosen quartic. Put a=Q(e)>0a=Q(e)>0. At μ=0\mu=0 the ball is a strict C1C^1 local minimizer. For every sufficiently small μ>0\mu>0, the shapes attaining the least energy in a fixed sufficiently small C1C^1 graph neighborhood of the ball form exactly one rotation orbit, with representative

w=te+z(te,μ),t2=αμ4a+O(μ2).w=te+z(t e,\mu),\qquad t^2=\frac{\alpha\mu}{4a}+O(\mu^2).

Its energy is −α2μ2/(16a)+O(μ3)-\alpha^2\mu^2/(16a)+O(\mu^3) relative to the ball. Its full stabilizer is that of ee, and it is strictly convex for small μ\mu.

Proof. Lemma 4.1 shows that Q(v)≥a∥v∥2Q(v)\ge a\lVert v\rVert^2 with equality precisely along the chosen angular orbit. Positivity at μ=0\mu=0 follows from Lemma 3.1. The fixed-point space of the stabilizer of ee in V3V_3 is Re\mathbb{R}e. For TdT_d, the even coordinate sign changes and permutations leave only xyzxyz; for D3hD_{3h}, reflection in zz and the planar rotation and reflection leave only the cosine mode of frequency three. By equivariance, a stationary point on that line is a stationary point of the full reduced energy. If ψ(t,μ)=Ψ(te,μ)\psi(t,\mu)=\Psi(te,\mu), its equation is

∂tψ(t,μ)t=−αμ+4at2+O(t4+∣μ∣t2)=0.\frac{\partial_t\psi(t,\mu)}{t}=-\alpha\mu+4at^2+O(t^4+|\mu|t^2)=0.

The implicit function theorem in μ\mu gives μ(t)=4at2/α+O(t4)\mu(t) = 4at^2/\alpha+ O(t^4). This function is even and strictly increasing for small t>0t > 0, so every sufficiently small μ>0\mu> 0 has exactly one small positive amplitude, with the stated expansion.

To prove angular uniqueness near the selected orbit, take a normal slice to the orbit in the finite-dimensional unit sphere of V3V_3. Divide the angular equation at v=tωv = t\omega by t4t^4. It extends smoothly to t=μ=0t = \mu= 0, with invertible angular derivative given by ±bκHe\pm b_\kappa H_e. Symmetry makes the slice origin a solution for all small t,μt,\mu. The implicit function theorem makes it the unique one there. The radial equation above then gives a single critical orbit in the cone. The slice argument takes place in V3V_3, so no differentiability of the rotation action on the Hölder graph space is needed.

We now show that every near-ball minimizer lies in that cone. Choose δ>0\delta> 0 and then μ0>0\mu_0 > 0 so that the remainder in (11) is at most (a/2)∥v∥24(a/2)\lVert v\rVert_2^4 for ∥v∥2≤δ\lVert v\rVert_2 \le\delta, 0<μ<μ00 < \mu< \mu_0. Then

Ψ(v,μ)≥−α2μ∥v∥22+a2∥v∥24.\Psi(v,\mu) \ge-\frac{\alpha}{2}\mu\lVert v\rVert_2^2 + \frac{a}{2}\lVert v\rVert_2^4.

Choose K2>α/aK^2 > \alpha/a and decrease μ0\mu_0 so that Kμ0<δK\sqrt{\mu_0} < \delta. The energy is positive when Kμ≤∥v∥≤δK\sqrt{\mu} \le\lVert v\rVert\le\delta, while the constructed branch has negative energy. The minimum on the closed δ\delta ball is then interior and has ∥v∥2<Kμ\lVert v\rVert_2 < K\sqrt{\mu}. Under v=μ yv = \sqrt{\mu}\,y, the energies divided by μ2\mu^2 converge in C2C^2 on that fixed yy ball to

Φ(y)=−α2∥y∥22+Q(y).\Phi(y) = -\frac{\alpha}{2}\lVert y\rVert_2^2 + Q(y).

Its global minimizing set is exactly α/(4a) SO(3)e\sqrt{\alpha/(4a)}\,\mathrm{SO}(3)e. Any sequence of minimizers of the scaled energies has a convergent subsequence; comparison with each fixed yy makes its limit a minimum of Φ\Phi. It follows, uniformly over all minimizers, that they enter the previously chosen cone and stay away from zero in scaled coordinates. Angular and radial uniqueness identify them with the constructed orbit. This step uses the angular bounds on the whole unit sphere.

Choose the fixed C1C^1 graph neighborhood so that its center projections lie in the δ\delta ball and (12) holds. That estimate compares every graph in the neighborhood with its exact reduction graph. Each graph has energy at least the reduced minimum, with equality only if y=0y = 0 and its V3V_3 coordinate lies on that orbit. The constructed graphs lie in the chosen neighborhood for small μ\mu, so they attain the minimum. Substitution in (11) gives the stated energy expansion. Equivariance gives every symmetry of ee as a symmetry of the graph; projecting the graph onto V3V_3 shows that any symmetry of the graph also fixes ee. Finally z(te,μ)=OC2,γ(t2)z(te,\mu) = O_{C^{2,\gamma}}(t^2) and t→0t \to0, so the boundaries are nonspherical strictly convex perturbations of the unit sphere. □

From local selection to the first global transition

Fix κ>0\kappa> 0 and write V=∣B∣V = |B|. In this section we consider smooth positive densities satisfying

cρ(1+∣x∣2)≤ρ(x)≤Cρ(1+∣x∣2),x∈R3,(25)c_\rho(1+|x|^2) \le\rho(x) \le C_\rho(1+|x|^2), \qquad x \in\mathbb{R}^3, \tag*{(25)}

with positive constants depending on the density. All densities constructed above satisfy these bounds. So do their modifications by any fixed bounded nonnegative function. Sets are identified up to null sets. To pass from graph minimality to L1L^1 minimality, we use penalization as in [12, 13]. We give the argument for a Hessian with a kernel. For BV compactness and perimeter slicing, we use [15].

Lemma 5.1 (Existence and parameter compactness). For every β>0\beta> 0, Fβ,ρF_{\beta,\rho} has a global minimizer among finite-perimeter sets of volume VV. If βj→β0>0\beta_j \to\beta_0 > 0 and EjE_j are corresponding global minimizers, a subsequence converges in L1(R3)L^1(\mathbb{R}^3) to a global minimizer at β0\beta_0.

Proof. For z∈Z3z \in\mathbb{Z}^{3}, let Qz=z+[−1/2,1/2)3Q_z=z+[-1/2,1/2)^{3}, mz=inf⁡Qzρm_z=\inf_{Q_z}\rho, and g=e−κ3/(4π3)g=e^{-\kappa\sqrt{3}}/(4\pi\sqrt{3}). Positivity of the kernel gives

2Iρ(E)≥g∑zmz2∣E∩Qz∣2,∑z∈Z3∣E∩Qz∣≤(2Iρ(E)g)1/2(∑z∈Z3mz−2)1/2.2I_{\rho}(E)\ge g\sum_{z}m_z^{2}|E\cap Q_z|^{2},\qquad\sum_{z\in\mathbb{Z}^{3}}|E\cap Q_z|\le\left(\frac{2I_{\rho}(E)}{g}\right)^{1/2}\left(\sum_{z\in\mathbb{Z}^{3}}m_z^{-2}\right)^{1/2}.

The lower bound in (25) implies ∑zmz−2<∞\sum_zm_z^{-2}<\infty. Bounded self-energy then gives uniform tightness of the ordinary volume. Comparison with BB bounds the perimeter and self-energy of a minimizing sequence. Local BV compactness and tightness give global L1L^{1} convergence preserving volume. Perimeter is lower semicontinuous, and Fatou’s lemma treats the positive double integral. This proves existence.

For varying parameters, comparison with BB and P(Ej)≥P(B)P(E_j)\ge P(B) give Iρ(Ej)≤Iρ(B)I_{\rho}(E_j)\le I_{\rho}(B) and a uniform perimeter bound. The same compactness argument applies. Since Iρ(Ej)I_{\rho}(E_j) is bounded, replacing βj\beta_j by β0\beta_0 changes the energy by o(1)o(1). Lower semicontinuity and comparison with each fixed finite-energy competitor then prove that the limit is a minimizer. Infinite-energy competitors impose no additional condition at a positive coupling.

We need the following local estimate for the unbounded density. For a compact set KK and any ∣E∣≤V+1|E|\le V+1,

sup⁡x∈KGκ∗(ρ1E)(x)≤CK,V.(26)\sup_{x\in K}G_{\kappa}*(\rho\mathbf{1}_{E})(x)\le C_{K,V}. \tag*{(26)}

The contribution from ∣x−y∣<1|x-y|<1 is bounded by local boundedness of ρ\rho and integrability of GκG_{\kappa}. On ∣x−y∣≥1|x-y|\ge1, the function ρ(y)Gκ(x−y)\rho(y)G_{\kappa}(x-y) is uniformly bounded for x∈Kx\in K, by the upper quadratic bound and exponential decay. Its integral over EE is bounded by that supremum times ∣E∣|E|. So if Iρ(E)<∞I_{\rho}(E)<\infty and F△E⊂KF\triangle E\subset K, then

∣Iρ(F)−Iρ(E)∣≤CK,V∣F△E∣.(27)|I_{\rho}(F)-I_{\rho}(E)|\le C_{K,V}|F\triangle E|. \tag*{(27)}

To see this, set a=ρ(1F−1E)a=\rho(\mathbf{1}_{F}-\mathbf{1}_{E}) and expand the difference as ∫a(Gκ∗(ρ1E))+12∫a(Gκ∗a)\int a(G_{\kappa}*(\rho\mathbf{1}_{E}))+\frac{1}{2}\int a(G_{\kappa}*a). The first term is controlled by (26); the second by 12(sup⁡Kρ)2∥Gκ∥1∣F△E∣\frac{1}{2}(\sup_{K}\rho)^{2}\|G_{\kappa}\|_{1}|F\triangle E|. So bounded local modifications preserve finiteness of the energy.

Lemma 5.2 (Uniform regularity near the ball). Let βj\beta_j lie in a compact subinterval of (0,∞)(0,\infty), and let HjH_j be real-valued penalties with a common Lipschitz constant for the distance ∣E△F∣|E\triangle F|. Suppose AjA_j minimize Fβj,ρ+HjF_{\beta_j,\rho}+H_j at volume VV and Aj→BA_j\to B in L1L^{1}. For all sufficiently large jj, their essential supports are uniformly bounded, and there exist Λ,r0>0\Lambda,r_0>0 such that

P(Aj;Br(x))≤P(F;Br(x))+Λ∣F△Aj∣(28)P(A_j;B_r(x))\le P(F;B_r(x))+\Lambda|F\triangle A_j| \tag*{(28)}

whenever r≤r0r\le r_0 and F△Aj⋐Br(x)F\triangle A_j\Subset B_r(x). Their boundaries are eventually spherical graphs converging to ∂B\partial B in C1,αC^{1,\alpha} for every 0<α<1/20<\alpha<1/2.

Proof. Comparison with BB bounds P(Aj)P(A_j) and Iρ(Aj)I_{\rho}(A_j) uniformly, since ∣Hj(Aj)−Hj(B)∣≤2VLip⁡(Hj)|H_j(A_j)-H_j(B)|\le2V\operatorname{Lip}(H_j). Choose bounded neighborhoods U1,U2U_1,U_2 with disjoint closures near two separated sphere patches, and smooth fields XiX_i compactly supported in UiU_i, with bi=div⁡Xi>0b_i=\operatorname{div}X_i>0. Then ∫Ajdiv⁡Xi≥bi/2\int_{A_j}\operatorname{div}X_i\ge b_i/2 for large jj. For the flows Φi,t\Phi_{i,t}, bounded second derivatives of the Jacobian give a common interval on which the derivative of ∣Φi,t(Aj)∣|\Phi_{i,t}(A_j)| stays positive. Any volume correction ∣s∣<s0|s|<s_0 can be made with ∣t∣≤C∣s∣|t|\le C|s|. The tangential-Jacobian and local BV flow estimates give

∣P(Φi,t(Aj))−P(Aj)+∣Φi,t(Aj)△Aj∣∣≤C∣t∣P(Aj;Ui).\left|P(\Phi_{i,t}(A_j))-P(A_j)+|\Phi_{i,t}(A_j)\triangle A_j|\right|\le C|t|P(A_j;U_i).

Equations (26)–(27) and the penalty bound show that the full correction costs at most C∣s∣C|s|. If a preliminary modification leaves UiU_i unchanged, the volume-change function and local perimeter in this argument are unchanged. The same cost estimate applies to the truncations and bounded local modifications below; their intermediate volumes are at most V+1V+1.

For tt outside the two correction neighborhoods, put mj(t)=∣Aj∖Bt∣m_j(t)=|A_j\setminus B_t| and aj(t)=−mj′(t)=H2(Aj∩∂Bt)a_j(t)=-m_j'(t)=\mathcal{H}^2(A_j\cap\partial B_t) for almost every tt. The perimeter slicing identity reads

P(Aj∩Bt)−P(Aj)=2aj(t)−P(Aj∖Bt).P(A_j\cap B_t)-P(A_j)=2a_j(t)-P(A_j\setminus B_t).

Removing the tail decreases the positive interaction energy. Whenever mj(t)<s0m_j(t)<s_0, restore its volume using either flow. Minimality, the correction estimate, and the isoperimetric inequality yield

c3mj(t)2/3≤P(Aj∖Bt)≤2aj(t)+Cmj(t).c_3m_j(t)^{2/3}\le P(A_j\setminus B_t)\le2a_j(t)+Cm_j(t).

Choose a fixed TT containing BB, U‾1\overline{U}_1, U‾2\overline{U}_2. Since mj(T)→0m_j(T)\to0, the correction is allowed and Cmj≤(c3/2)mj2/3Cm_j\le(c_3/2)m_j^{2/3} for t≥Tt\ge T and large jj. Where mj>0m_j>0, it follows that −(mj1/3)′≥c3/12-(m_j^{1/3})'\ge c_3/12. So all these tails vanish before a common finite radius LL.

Choose r0r_0 smaller than one quarter of the distance between U‾1,U‾2\overline{U}_1,\overline{U}_2, with ∣Br0∣<min⁡(s0,1)|B_{r_0}|<\min(s_0,1). If Br(x)B_r(x) is disjoint from BLB_L, (28) is automatic. Otherwise all changes occur in a fixed enlarged ball. One UiU_i is disjoint from Br(x)B_r(x); use its flow to correct the volume error of FF, whose size is at most ∣F△Aj∣|F\mathbin{\triangle}A_j|. The local perimeter on UiU_i is still that of AjA_j, even when P(F)P(F) is large. The correction costs and (27) then give P(Aj)≤P(F)+C∣F△Aj∣P(A_j)\le P(F)+C|F\mathbin{\triangle}A_j| by minimality. The exterior perimeter terms cancel, proving (28).

Since ∣F△Aj∣≤∣Br∣|F\mathbin{\triangle}A_j|\le|B_r|, this is area almost minimality with a common error Cr3Cr^3. We apply the spherical convergence theorem in [14 Theorem 2.2]: uniform perimeter bounds and global L1L^1 convergence to BB imply eventual C1,1/2C^{1,1/2} spherical graphs and convergence of their graph functions in C1,αC^{1,\alpha}, α<1/2\alpha<1/2. □

Lemma 5.3 (Strict graph minimality implies strict L1L^1 minimality). Fix β>0\beta>0. If BB is a strict minimizer among sufficiently small C1C^1 normal graphs of volume VV, it is a strict L1L^1 local minimizer among all finite-perimeter sets of that volume. The constrained Hessian may have a kernel.

Proof. Choose a smooth compactly supported field XX with ∣X∣≤1|X|\le1 and X=νBX=\nu_B on ∂B\partial B. The divergence formula gives P(E)−P(B)≥−C∣E△B∣P(E)-P(B)\ge-C|E\mathbin{\triangle}B|. Positivity also gives Iρ(E)≥Iρ(E∩B)I_\rho(E)\ge I_\rho(E\cap B), while Iρ(B)−Iρ(E∩B)≤C∣B∖E∣I_\rho(B)-I_\rho(E\cap B)\le C|B\setminus E|. Writing d(E)=∣E△B∣d(E)=|E\mathbin{\triangle}B|, we get

Fβ,ρ(E)≥Fβ,ρ(B)−Cd(E).F_{\beta,\rho}(E)\ge F_{\beta,\rho}(B)-Cd(E).

If the assertion fails, choose 0<δj=d(Ej)→00<\delta_j=d(E_j)\to0 with Fβ,ρ(Ej)≤Fβ,ρ(B)F_{\beta,\rho}(E_j)\le F_{\beta,\rho}(B). For a fixed M>CM>C, minimize Jj(E)=Fβ,ρ(E)+M∣d(E)−δj∣J_j(E)=F_{\beta,\rho}(E)+M|d(E)-\delta_j| at volume VV. The direct method above applies to this nonnegative L1L^1-continuous penalty. Its minimizers satisfy

Jj(Aj)≤Fβ,ρ(B)<Jj(B),d(Aj)≤MδjM−C,Fβ,ρ(Aj)≤Fβ,ρ(B).J_j(A_j)\le F_{\beta,\rho}(B)<J_j(B),\qquad d(A_j)\le\frac{M\delta_j}{M-C},\qquad F_{\beta,\rho}(A_j)\le F_{\beta,\rho}(B).

The penalties have common Lipschitz constant MM. Lemma 5.2 makes the nonball sets AjA_j eventually small C1C^1 graphs of the prescribed volume, a contradiction. □

Proposition 5.4 (A common finite exterior density barrier). Consider finitely many radial densities satisfying (25). Suppose that, at a common β∗>0\beta_*>0, the ball is a strict C1C^1 graph minimizer in each model, while every slightly larger coupling admits arbitrarily small equal-volume graphs of lower

energy than the ball. There exist c>1c>1, a bounded smooth nondecreasing radial function hch_c, zero on BcB_c and positive outside, and a common finite s≥0s\geq0, such that the modified densities ρ^i=ρi+shc\widehat{\rho}_i=\rho_i+sh_c have βg(ρ^i)=β∗\beta_g(\widehat{\rho}_i)=\beta_*. The centered ball is their unique global minimizer for 0<β≤β∗0<\beta\leq\beta_*. All global minimizers converge to BB in L1L^1 as β↓β∗\beta\downarrow\beta_*, uniformly over the choice of minimizer. The barrier preserves the exact energy near the ball, all ball data, and strict radial monotonicity when present.

Proof. Lemma 5.3 provides strict L1L^1 radii; let δ\delta be their minimum. Choose c>1c>1 with 2(∣Bc∣−V)<δ2(|B_c|-V)<\delta. Every E⊂BcE\subset B_c of volume VV then satisfies ∣E△B∣<δ|E\mathbin{\triangle}B|<\delta, so BB is the unique minimizer in this hard-wall class. Fix hch_c, for example hc(r)=e−(r−c)−2h_c(r)=e^{-(r-c)^{-2}} for r>cr>c and zero otherwise.

Work first with one density. If arbitrarily large finite strengths fail, there are sj→∞s_j\to\infty and nonball global minimizers EjE_j for ρj=ρ+sjhc\rho_j=\rho+s_jh_c at β∗\beta_*. Existence holds for each fixed sjs_j. Comparison with BB and the isoperimetric inequality give a uniform perimeter bound and

Iρ(Ej)≤Iρj(Ej)≤Iρ(B),sj2Ihc(Ej)≤Iρ(B).I_\rho(E_j)\leq I_{\rho_j}(E_j)\leq I_\rho(B),\qquad s_j^2I_{h_c}(E_j)\leq I_\rho(B).

The tightness estimate for the original density gives global L1L^1 convergence along a subsequence. Fatou’s lemma forces its limit E∞E_\infty to satisfy Ihc(E∞)=0I_{h_c}(E_\infty)=0, so E∞⊂BcE_\infty\subset B_c by strict positivity of the kernel. Lower semicontinuity gives Fβ∗,ρ(E∞)≤Fβ∗,ρ(B)F_{\beta_*,\rho}(E_\infty)\leq F_{\beta_*,\rho}(B), so E∞=BE_\infty=B. Strict L1L^1 minimality for the original density now contradicts

Fβ∗,ρj(Ej)≥Fβ∗,ρ(Ej)>Fβ∗,ρ(B)=Fβ∗,ρj(B).F_{\beta_*,\rho_j}(E_j)\geq F_{\beta_*,\rho}(E_j)>F_{\beta_*,\rho}(B)=F_{\beta_*,\rho_j}(B).

This proves that every sufficiently large finite strength works. Take the largest required threshold for the finitely many models and fix one common ss. The contradiction uses strict L1L^1 minimality for the fixed original density, so it needs no regularity bound uniform in ss.

For a finite-energy competitor and 0<β<β∗0<\beta<\beta_*,

ΔFβ=ββ∗ΔFβ∗+(1−ββ∗)ΔP\Delta F_\beta=\frac{\beta}{\beta_*}\Delta F_{\beta_*}+\left(1-\frac{\beta}{\beta_*}\right)\Delta P

gives strict uniqueness, since ΔP≥0\Delta P\geq0. Infinite-energy competitors are irrelevant for positive couplings. Graphs sufficiently close to BB lie in BcB_c, so their exact energies are unchanged. The lower-energy competitors just above β∗\beta_* persist. For each such competitor at β1>β∗\beta_1>\beta_*, ΔP+β1ΔI<0\Delta P+\beta_1\Delta I<0 and ΔP≥0\Delta P\geq0 imply ΔI<0\Delta I<0; the same competitor defeats the ball at every β≥β1\beta\geq\beta_1. So the ball cannot become minimizing again, and βg=β∗\beta_g=\beta_*. Parameter compactness and uniqueness at β∗\beta_* give the stated uniform convergence. Finally, hc=0h_c=0 on BcB_c preserves local energies, and hc′≥0h_c'\geq0 preserves strict radial monotonicity. The common addition also preserves agreement of the densities at all sufficiently large radii. Its fixed finite size preserves (25). □

Proposition 5.5 (Every global minimizer lies on the exact reduction graph). Fix one model after the finite barrier construction, with the reduction and C1C^1 energy estimate of Lemma 3.1. For all sufficiently small μ>0\mu>0, every global minimizer has the form Ev+z(v,μ)E_{v+z(v,\mu)}, where vv minimizes Ψ(⋅,μ)\Psi(\cdot,\mu) over a fixed small center neighborhood.

Proof. For any sequence μj↓0\mu_j\downarrow0, global minimizers converge to BB in L1L^1 by Proposition 5.4. Lemma 5.2, with zero penalty, gives small C1C^1 graphs. In exact volume coordinates write wj=vj+z(vj,μj)+yjw_j=v_j+z(v_j,\mu_j)+y_j, yj∈Wy_j\in W. Compare the minimizer with the reduced shape at the same vjv_j. Equation (12) gives

Ψ(vj,μj)+c∥yj∥H12≤F(Ej)−F(B)≤Ψ(vj,μj).\Psi(v_j,\mu_j)+c\lVert y_j\rVert_{H^1}^2\leq F(E_j)-F(B)\leq\Psi(v_j,\mu_j).

The estimate forces yj=0y_j=0. Comparison with every other small reduced shape proves minimality in the center neighborhood. A sequence of counterexamples would contradict this argument, so the conclusion holds for every minimizer throughout a sufficiently small parameter interval. □

If the reduced minimizers form one rotation orbit, existence and rotation invariance imply that the global minimizers form the same orbit. All constants in this section concern fixed densities. We first fix the boundary derivatives, then the finite exterior barrier, and then shrink the parameter interval.

Proof of the main theorem

Proof of Theorem 1.1. Choose κ0\kappa_0 small enough for Lemma 2.2 and Proposition 3.2. Fix 0<κ<κ00<\kappa<\kappa_0 for the rest of the construction. Use Lemma 2.3 to realize (16), with the same second derivative ρ′′(1)=4−7\rho^{\prime\prime}(1)=4-\sqrt{7} and the two prescribed third derivatives. Call these preliminary densities ρT0,ρS0\rho_T^0,\rho_S^0. They preserve the common boundary value, first derivative, mass, Yukawa moment, and ball self-energy. Lemma 2.1 gives identical constrained Hessians for every coupling, including all degree-one and all high-degree modes. Both densities satisfy two-sided quadratic growth bounds and are strictly increasing.

Their quartics have positive angular minima

aT=Aκ+bκjT>0,aS=Aκ−bκjS>0.a_T=A_\kappa+b_\kappa j_T>0,\qquad a_S=A_\kappa-b_\kappa j_S>0.

Lemma 3.1 gives strict C1C^1 graph minimality of the critical ball, and Lemma 5.3 upgrades it to strict L1L^1 local minimality. Choose the common exterior radius and a common finite barrier strength as in Proposition 5.4, and set

ρT=ρT0+shc,ρS=ρS0+shc.\rho_T=\rho_T^0+sh_c,\qquad\rho_S=\rho_S^0+sh_c.

The increment is smooth, nonnegative, nondecreasing, bounded, and zero throughout BcB_c, where c>1c>1. It preserves strict increase, quadratic growth, all the ball data, and the entire exact energy of every shape contained in BcB_c. The two densities coincide away from the compact annuli supporting their preliminary perturbations.

Proposition 5.4 makes BB uniquely globally minimizing for 0<β≤β∗0<\beta\leq\beta_*. Proposition 4.2 gives near-ball competitors of lower energy for every sufficiently small μ>0\mu>0; these lie in BcB_c and remain unchanged by the barrier. Proposition 5.4 then gives βg=β∗\beta_g=\beta_*. The common linear threshold and its limit follow from Lemma 2.2.

Fix these two final densities. By compactness and uniqueness at β∗\beta_*, all their global minimizers converge to BB in L1L^1 as β↓β∗\beta\downarrow\beta_*. Proposition 5.5 places every such minimizer on the exact reduction graph. Its center coordinate minimizes the reduced energy in a fixed small neighborhood. Proposition 4.2 identifies that minimizing set as the single prescribed rotation orbit. Global existence gives a point of each orbit, and rotation invariance makes the whole orbit globally minimizing. Taking the smaller of the two parameter neighborhoods gives the common ε\varepsilon. The stabilizers, convexity, and expansions (4) follow from the same proposition. □

A finite transmission calculation for the positive coefficient

We prove (17) by expanding the original shape energy to fourth order and eliminating the noncritical variables. The reduction in Lemma 3.1 uses the full shape equation.

All quantities in this calculation are Newtonian limits. Set

q=1,p=3−7,r2=p+1,r3=0,U=2+715,β∗=175,Lℓ=ℓ(ℓ+1).q=1,\qquad p=3-\sqrt{7},\qquad r_2=p+1,\qquad r_3=0,\qquad U=\frac{2+\sqrt{7}}{15},\qquad\beta_*=175,\qquad L_\ell=\ell(\ell+1).

Along w=tvw=tv, v∈V3v\in V_3, the radial height is

h=tv−t2v2+53t3v3+O(t4).(29)h=tv-t^2v^2+\frac{5}{3}t^3v^3+O(t^4). \tag*{(29)}

Add homogeneous corrections of orders tjt^j to the interior and exterior radial ball potentials, with traces Aj,BjA_j,B_j at r=1r=1. The radial factors on degree ℓ\ell are rℓr^\ell and r−ℓ−1r^{-\ell-1}. Let [f]=f−(1)−f+(1)[f]=f^-(1)-f^+(1) and let D−D_- and D+D_+ have multipliers ℓ\ell and −ℓ−1-\ell-1. The Poisson equation gives

[u0′′]=−1,[u0′′′]=2−p,[u0′′′′]=−8+2p−r2.[u_0'']=-1,\qquad[u_0''']=2-p,\qquad[u_0'''']=-8+2p-r_2.

Expanding continuity of the potential and its radial derivative at 1+h1+h yields

A1=B1=v/7,A_1=B_1=v/7,
A2−B2=−v2/2,D−A2−D+B2=pv2/2,A_2-B_2=-v^2/2,\qquad D_-A_2-D_+B_2=pv^2/2,
A3−B3=(2−p)v3/3,D−A3−D+B3=(−6−p3+r26)v3+v2∑ℓ=0,2,4,6LℓΠℓ(v2).(30)A_3-B_3=(2-p)v^3/3,\qquad D_-A_3-D_+B_3=\left(-6-\frac{p}{3}+\frac{r_2}{6}\right)v^3+\frac{v}{2}\sum_{\ell=0,2,4,6}L_\ell\Pi_\ell(v^2). \tag*{(30)}

At second order the base solution contributes −v2/2-v^2/2 to the value jump, and the first-order correction contributes v2v^2. This gives A2−B2=−v2/2A_2-B_2=-v^2/2. In the third-order derivative equation one uses [u1′′]=18v[u_1'']=18v and [u2′′]=−pv2−12∑LℓΠℓ(v2)[u_2'']=-pv^2-\frac{1}{2}\sum L_\ell\Pi_\ell(v^2), which follow from the radial homogeneous equation. These identities fix the signs in (A.2).

For the system A−B=JA-B=J, ℓA+(ℓ+1)B=K\ell A+(\ell+1)B=K, the exterior trace is B=(K−ℓJ)/(2ℓ+1)B=(K-\ell J)/(2\ell+1). This gives

B2=∑ℓ=0,2,4,6p+ℓ2(2ℓ+1)Πℓ(v2),B_2=\sum_{\ell=0,2,4,6}\frac{p+\ell}{2(2\ell+1)}\Pi_\ell(v^2),
⟨v,B3⟩=−8+2p/3+r2/67J(v)+114∑ℓ=0,2,4,6LℓQℓ(v).(31)\langle v,B_3\rangle=\frac{-8+2p/3+r_2/6}{7}J(v)+\frac{1}{14}\sum_{\ell=0,2,4,6}L_\ell Q_\ell(v). \tag*{(31)}

If ϕj\phi_j is the coefficient of the exterior potential evaluated at the moving boundary, then

ϕ1=(−U+1/7)v,\phi_1=(-U+1/7)v,
ϕ2=(2U−4/7)v2+B2,\phi_2=(2U-4/7)v^2+B_2,
ϕ3=(−14U/3+2)v3−v∑ℓ=0,2,4,6(ℓ+1)ΠℓB2+B3.\phi_3=(-14U/3+2)v^3-v\sum_{\ell=0,2,4,6}(\ell+1)\Pi_\ell B_2+B_3.

The density at that boundary is

ρ(1+h)=1+ptv+(−p+r2/2)t2v2+(5p/3−r2+r3/6)t3v3+O(t4).\rho(1+h)=1+ptv+(-p+r_2/2)t^2v^2+(5p/3-r_2+r_3/6)t^3v^3+O(t^4).

Multiplication gives the nonlocal Euler gradient in exact volume coordinates, by (13).

Quadratic forcing. The cubic perimeter energy is P3(w)=∫(4w3/3−2w∣∇w∣2)P_3(w)=\int(4w^3/3-2w|\nabla w|^2), whose gradient at v∈V3v\in V_3 is Δ(v2)−20v2\Delta(v^2)-20v^2. So the quadratic coefficient of the full projected gradient is

g2(v)=∑ℓ=2,4,6gℓΠℓ(v2),gℓ=−(Lℓ+20)+175[U(2−2p+r22)+p−47+p+ℓ2(2ℓ+1)].(32)g_2(v)=\sum_{\ell=2,4,6}g_\ell\Pi_\ell(v^2),\qquad g_\ell=-(L_\ell+20)+175\left[U\left(2-2p+\frac{r_2}{2}\right)+\frac{p-4}{7}+\frac{p+\ell}{2(2\ell+1)}\right]. \tag*{(32)}

The constant channel is removed by the exact volume constraint.

Direct quartic coefficient. Pairing the cubic Euler-gradient coefficient with vv and dividing by four gives the direct quartic interaction energy. The part multiplying QℓQ_\ell, modulo a multiple of J\mathcal{J}, is

1754[p2−p−Lℓ2(2ℓ+1)+Lℓ14]Qℓ.\frac{175}{4}\left[\frac{p^2-p-L_\ell}{2(2\ell+1)}+\frac{L_\ell}{14}\right]Q_\ell.

For the perimeter, expansion of the exact area integrand gives

P4(v)=∫(−73v4+7v2∣∇v∣2−18∣∇v∣4)=773J−18∑ℓ=0,2,4,6(12−Lℓ2)2Qℓ.P_4(v)=\int\left(-\frac{7}{3}v^4+7v^2|\nabla v|^2-\frac{1}{8}|\nabla v|^4\right)=\frac{77}{3}\mathcal{J}-\frac{1}{8}\sum_{\ell=0,2,4,6}\left(12-\frac{L_\ell}{2}\right)^2Q_\ell.

Here ∣∇v∣2=12Δ(v2)+12v2|\nabla v|^2=\frac{1}{2}\Delta(v^2)+12v^2 and ∫v2∣∇v∣2=4∫v4\int v^2|\nabla v|^2=4\int v^4. We obtain

Qdir≡∑ℓ=0,2,4,6dℓQℓ(modRJ),dℓ=−18(12−Lℓ2)2+1754[p2−p−Lℓ2(2ℓ+1)+Lℓ14].(33)Q_{\mathrm{dir}}\equiv\sum_{\ell=0,2,4,6}d_\ell Q_\ell\pmod{\mathbb{R}\mathcal{J}},\qquad d_\ell=-\frac{1}{8}\left(12-\frac{L_\ell}{2}\right)^2+\frac{175}{4}\left[\frac{p^2-p-L_\ell}{2(2\ell+1)}+\frac{L_\ell}{14}\right]. \tag*{(33)}

Exact complement elimination. The quadratic correction is z2=−LW−1g2z_2=-L_W^{-1}g_2. In degree four it contributes ⟨g2,z2⟩+⟨LWz2,z2⟩/2=−⟨g2,LW−1g2⟩/2\langle g_2,z_2\rangle+\langle L_Wz_2,z_2\rangle/2=-\langle g_2,L_W^{-1}g_2\rangle/2. This gives

Q=Qdir−12∑ℓ=2,4,6gℓ2λℓQℓ.(34)Q=Q_{\mathrm{dir}}-\frac{1}{2}\sum_{\ell=2,4,6}\frac{g_\ell^2}{\lambda_\ell}Q_\ell. \tag*{(34)}

There is a constant channel in (A.5), but none in this elimination sum.

Define L(Q)=10Q(eT)−9Q(eS)\mathcal{L}(Q)=10Q(e_T)-9Q(e_S). Since jS=10jT/9j_S=10j_T/9, L(N)=1\mathcal{L}(\mathcal{N})=1 and L(J)=0\mathcal{L}(\mathcal{J})=0, so L(Q)\mathcal{L}(Q) is exactly the coefficient AA of N\mathcal{N}. The required data are

ℓ\ellλℓ\lambda_\ellgℓg_\ellπL(Qℓ)\pi\mathcal{L}(Q_\ell)
0excludedexcluded1/41/4
24(674−1857)/6(674-185\sqrt{7})/6−5/4-5/4
422/922/9(1420−4157)/18(1420-415\sqrt{7})/1869/4469/44
6240/13240/13(3854−15657)/78(3854-1565\sqrt{7})/78−25/44-25/44

Table A.7.

The last column follows from these projection values:

ℓ\ell0246
πQℓ(eT)\pi Q_\ell(e_T)1/41/4021/12121/121200/1573200/1573
πQℓ(eS)\pi Q_\ell(e_S)1/41/45/365/369/4849/48411575/5662811575/56628

To compute them from (22), write the square of the homogeneous harmonic cubic PP as P2=H6+r2H4+r4H2+r6H0P^2=H_6+r^2H_4+r^4H_2+r^6H_0. Then H0=2tr⁡A/35H_0=2\operatorname{tr}A/35, H2=2(A−tr⁡AId⁡/3)ijxixj/7H_2=2(A-\operatorname{tr}A\operatorname{Id}/3)_{ij}x_ix_j/7, H4=(Δ(P2)−36r2H2−42r4H0)/22H_4=(\Delta(P^2)-36r^2H_2-42r^4H_0)/22, and the remainder is H6H_6. Restriction to the sphere gives each projection.

Finally, (A.5)–(A.7) give

πA=∑ℓ=0,2,4,6dℓπL(Qℓ)−12∑ℓ=2,4,6gℓ2λℓπL(Qℓ)=663451007−842683151672704>0.(35)\pi A=\sum_{\ell=0,2,4,6}d_\ell\pi\mathcal{L}(Q_\ell)-\frac{1}{2}\sum_{\ell=2,4,6}\frac{g_\ell^2}{\lambda_\ell}\pi\mathcal{L}(Q_\ell)=\frac{66345100\sqrt{7}-84268315}{1672704}>0. \tag*{(35)}

For positive κ\kappa, replace the radial powers by the normalized modified Bessel solutions and the Poisson equation by the Yukawa equation. Their boundary derivatives at the finitely many degrees involved converge to the data above. The denominators λ2,λ4,λ6\lambda_2,\lambda_4,\lambda_6 stay positive. This proves the finite coefficient continuity used in Proposition 3.2.

Independent algebraic check. The ancillary program check_transmission.py reconstructs the moving-boundary matching equations, the area expansion, and the Schur correction from polynomial operations over Q(7)\mathbb{Q}(\sqrt{7}). It does not use fitted numerical values. The program check_angular.py checks the harmonic projections and angular Hessians using exact spherical moments. These programs check the finite algebraic calculations. The reduction and global minimization arguments are proved in the text.

Use of generative AI. ChatGPT (OpenAI) was used for language refinement and consistency checks. The authors reviewed and verified all content and take full responsibility for the final manuscript.

References

  1. [1]R. L. Frank. Non-spherical equilibrium shapes in the liquid drop model. Journal of Mathematical Physics 60 (2019), 071506. doi:10.1063/1.5095603. arXiv:1903.04344.DOI
  2. [2]F. De Regibus, M. Grossi, and M. Musso. Bifurcations in Isoperimetric Problems with Nonlocal Interactions. Preprint, 2026. arXiv:2604.19170v1.arxiv.org/abs/2604.19170
  3. [3]L. Bronsard, B. Merlet, and M. Pegon. Symmetry breaking in the liquid drop model with screened interactions. Preprint, 2025; revised 6 September 2026. arXiv:2510.11893v2.arxiv.org/abs/2510.11893
  4. [4]L. Bronsard, K. DeMason, and I. Topaloglu. The Liquid Drop Model with a Yukawa Potential: Existence of Minimizers and Sharp Stability of the Ball. Preprint, 2026. arXiv:2608.19340v1.arxiv.org/abs/2608.19340
  5. [5]S. Alama, L. Bronsard, I. Topaloglu, and A. Zuniga. A nonlocal isoperimetric problem with density perimeter. Calculus of Variations and Partial Differential Equations 60 (2021), article 1. doi:10.1007/s00526-020-01865-8.DOI
  6. [6]S. Alama, L. Bronsard, R. Choksi, and I. Topaloglu. Droplet breakup in the liquid drop model with background potential. Communications in Contemporary Mathematics 21(3) (2019), 1850022. doi:10.1142/S0219199718500220. arXiv:1708.04292v2.DOI
  7. [7]E. Ihrig and M. Golubitsky. Pattern selection with O(3) symmetry. Physica D 13 (1984), 1–33. doi:10.1016/0167-2789(84)90268-9.DOI
  8. [8]P. Chossat, R. Lauterbach, and I. Melbourne. Steady-state bifurcation with O(3)-symmetry. Archive for Rational Mechanics and Analysis 113 (1991), 313–376. doi:10.1007/BF00374697.
  9. [9]Y. Chen, S. Hu, L. Qi, and W. Zou. Irreducible function bases of isotropic invariants of a third order three-dimensional symmetric and traceless tensor. Frontiers of Mathematics in China 14(1) (2019), 1–16. doi:10.1007/s11464-019-0748-x. arXiv:1712.02087.DOI
  10. [10]J.-P. Børnsen and A. E. M. van de Ven. Tangent Developable Orbit Space of an Octupole. Preprint, 2018. arXiv:1807.04817.
  11. [11]M. Dambrine and J. Lamboley. Stability in shape optimization with second variation. Journal of Differential Equations 267(5) (2019), 3009–3045. doi:10.1016/j.jde.2019.03.033. We use the corrected author version dated 20 September 2021: corrected full text.DOI
  12. [12]E. Acerbi, N. Fusco, and M. Morini. Minimality via second variation for a nonlocal isoperimetric problem. Communications in Mathematical Physics 322 (2013), 515–557. doi:10.1007/s00220-013-1733-y.DOI
  13. [13]M. Bonacini and R. Cristoferi. Local and global minimality results for a nonlocal isoperimetric problem on ℝ^N. SIAM Journal on Mathematical Analysis 46(4) (2014), 2310–2349. doi:10.1137/130929898.DOI
  14. [14]N. Fusco and V. Julin. A strong form of the quantitative isoperimetric inequality. Calculus of Variations and Partial Differential Equations 50 (2014), 925–937. doi:10.1007/s00526-013-0661-1. author manuscript. Theorem 2.2 records the spherical improved-convergence theorem of White used here.DOI
  15. [15]F. Maggi. Sets of Finite Perimeter and Geometric Variational Problems: An Introduction to Geometric Measure Theory. Cambridge Studies in Advanced Mathematics 135. Cambridge University Press, 2012. doi:10.1017/CBO9781139108133.DOI

Paper details

vlnc:W_faqrbr6f
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

Not recorded for this paper

View statement
Looking for related papers…
Contents