Introduction
The excess charge Q : = N − Z Q := N-Z Q := N − Z of an atomic anion is the number of bound electrons N N N in excess of the nuclear charge Z Z Z . Experimentally, no stable doubly or more highly charged isolated atomic anions are known; see [12 , 13 ] . Explaining the observed bound Q ≤ 1 Q \le1 Q ≤ 1 from a microscopic model of ions remains a challenge to mathematical physics. A weaker formulation, the ionization conjecture, asks for a bound on Q Q Q independent of Z Z Z .
Simple density functionals already capture parts of this picture. Thomas–Fermi theory does not bind negative ions, whereas its Fermi–Amaldi modification permits at most one excess electron. For the Thomas–Fermi–Weizsäcker functional, Benguria and Lieb [1 ] proved charge-independent bounds for atoms and bounds proportional to the number of nuclei for molecules. Solovej established such a uniform bound in reduced Hartree–Fock theory and subsequently in Hartree–Fock theory with exchange [19 , 20 ] . These results give Q = O ( 1 ) Q=O(1) Q = O ( 1 ) as Z → ∞ Z\to\infty Z → ∞ , with a universal bound valid for all Z Z Z , but do not establish the sharp value 1. More recently, an OpenAI preprint [14 ] established a corresponding uniform bound for strictly bound states of the full nonrelativistic many-electron Coulomb Hamiltonian with two spin states. For K K K fixed nuclei with real charges Z k ≥ 1 Z_k\ge1 Z k ≥ 1 at arbitrary distinct positions, the result is N ≤ Z t o t + C K N\le Z_{\mathrm{tot}}+CK N ≤ Z tot + C K , with a universal constant C C C . In particular, it gives Q = O ( 1 ) Q=O(1) Q = O ( 1 ) for atoms as Z → ∞ Z\to\infty Z → ∞ , and hence the analogue of Solovej’s Hartree–Fock bound for the full N N N -particle model. Earlier quantitative bounds for that model grew with Z Z Z . The new result does not provide the sharp bound Q ≤ 1 Q\le1 Q ≤ 1 .
For heavy atoms, relativistic effects must also be taken into account. Although the highest velocities occur near the nucleus, their influence extends to the outer shells and to chemical and optical properties; the colour of gold is a familiar example [15 ] . Rigorous excess-charge results for relativistic microscopic models are comparatively scarce. Schulz [16 Section 2.2]
Date: Draft, 6 November 2026.
Support by the Deutsche Forschungsgemeinschaft (DFG), grant EXC-2111-390814868 and DFG TRR 352, grant 470903074, and FONDECYT (Chile) Project # 124–1863 is gratefully acknowledged. RB thanks the Mathematical Institute of LMU for hospitality, and HS thanks the Institute of Physics of the Pontificia Universidad Católica de Chile for hospitality. The authors acknowledge the use of OpenAI’s ChatGPT and Google’s Gemini for mathematical discussions, the development and checking of arguments, and assistance in preparing the manuscript. The authors take full responsibility for the final text and proofs. reviews this problem and proves, for the fermionic Chandrasekhar model,
Q ≤ 1 2 ( 2 − 1 ) Z + O ( Z 1 / 3 ) . Q \le\frac{1}{2}(\sqrt{2}-1)Z + O(Z^{1/3}). Q ≤ 2 1 ( 2 − 1 ) Z + O ( Z 1/3 ) .
Here the remainder is uniform when α Z \alpha Z α Z stays a fixed distance below the critical coupling 2 / π 2/\pi 2/ π ; more precisely, his Theorem 2.2.1 assumes α Z < 2 / π \alpha Z < 2/\pi α Z < 2/ π for a fixed a > 1 a>1 a > 1 .
Motivated by these questions, we consider a relativistic statistical model with a gradient correction. The relativistic Thomas–Fermi model of Vallarta and Rosen [21 ] suffers from collapse at a point nucleus, as discussed by Jensen [9 ] ; see also Gombás [8 ] , § 14, pp. 120 ff. Engel and Dreizler later obtained a relativistic analogue of the Thomas–Fermi–Weizsäcker functional through a formal derivation from the Hartree–Fock approximation to quantum electrodynamics [6 ] . They subsequently studied neutral and positively charged atomic solutions numerically [7 ] . We omit the exchange term and ask whether this model admits an explicit excess-charge bound uniform in the nuclear charges within a specified coupling domain.
The exterior estimates used below follow the Benguria–Lieb strategy, with particular use of Solovej’s refined Sommerfeld comparison after Lemma 10 of [18 ] : see Lemma 11, Proposition 12, and Lemma 13 there. His Proposition 14 illustrates how the improved comparison reduces the excess-charge constant. Uniform estimates for a broader family of nonrelativistic Thomas–Fermi–Weizsäcker functionals were obtained by Benguria and Siedentop [3 ] .
There are already rigorous excess-charge estimates for the Engel–Dreizler model. Chen and Siedentop proved, for the functional without exchange, a molecular bound N < 2.56 ∑ k Z k N<2.56\sum_k Z_k N < 2.56 ∑ k Z k [5 ] . For the atomic functional including exchange, Chen, Frank, and Siedentop obtained N ≤ 2 Z + O ( Z 2 / 3 ) N\le2Z+O(Z^{2/3}) N ≤ 2 Z + O ( Z 2/3 ) as Z → ∞ Z\to\infty Z → ∞ with κ = Z / ( c λ ) \kappa=Z/(c\sqrt{\lambda}) κ = Z / ( c λ ) and λ \lambda λ fixed, where c c c is the speed of light in the Hartree units used there [4 ] . These results bound the total electron number by a quantity of order the nuclear charge. The present work estimates the difference Q = N − ∑ k Z k Q=N-\sum_k Z_k Q = N − ∑ k Z k by an explicit constant times the number of nuclei.
In the molecular argument, the repulsion of the nuclei is essential to the energy estimate used to impose restrictions on individual charges. This use of nuclear repulsion has a precedent in the ultrarelativistic Thomas–Fermi–Weizsäcker stability analysis of Benguria, Loss, and Siedentop [2 ] . Their homogeneous model differs from the massive Engel–Dreizler functional considered here. The relevant electrostatic localization inequalities originate in the work of Lieb and Yau [11 ] ; we use the formulation of Lieb, Loss, and Siedentop [10 ] . Stability estimates for the Engel–Dreizler functional were also developed by Siedentop [17 ] .
Our main result, Theorem 2.1 , bounds the excess charge by ( 5913 / 1000 ) λ 3 / 2 K (5913/1000)\lambda^{3/2}K ( 5913/1000 ) λ 3/2 K in an explicitly stated coupling domain. At λ = 1 / 9 \lambda=1/9 λ = 1/9 the constant per nucleus is 219 / 1000 219/1000 219/1000 . The atomic result is a direct corollary of the molecular theorem. The molecular energy hypothesis is the nonpositivity of the actual total energy, including nuclear repulsion. A virial identity makes this hypothesis automatic at local minima of the Born–Oppenheimer energy, whenever the electronic infimum is attained.
There are two main steps in the proof. First, the exterior comparison estimates depend only on α λ \alpha\lambda α λ and on the function H ( x ) = ∑ k = 1 K ∣ x − R k ∣ − 1 H(x)=\sum_{k=1}^{K}|x-R_k|^{-1} H ( x ) = ∑ k = 1 K ∣ x − R k ∣ − 1 , which is harmonic away from the nuclear positions. They do not depend on individual nuclear charges. Second, a nonlinear comparison between the density and the screened potential is allowed to have a nonnegative Newton-potential correction. Its source is supported where the local Fermi momentum exceeds an explicit exponentially large cutoff. A quartic energy estimate makes the electron number on this set small enough to absorb the correction. The nuclear repulsion enters only in this second step, through a restriction on each nuclear charge. Unequal charges are handled by an exact averaging over subsets of nuclei.
Section 2 states the result and specifies the rescaling from Hartree units used throughout the paper. Section 3 supplies the variational framework and the exact covariance. The exterior estimates, coercivity bound, and nonlinear comparison are proved in Sections 4, 5, and 6, respectively. Section 7 completes the charge estimate, and Section 8 treats the nuclear energy condition and the geometric factor K K K . The appendices give the algebra behind the localization constant and a reproducible specification of the finite numerical checks. The large-coupling regime at fixed α \alpha α is outside the present theorem and is left to a separate study.
The functional and the main results
In Hartree units, ℏ = m e = e 2 / ( 4 π ε 0 ) = 1 \hbar=m_e=e^2/(4\pi\varepsilon_0)=1 ℏ = m e = e 2 / ( 4 π ε 0 ) = 1 and c = α − 1 c=\alpha^{-1} c = α − 1 . We rescale the spatial and density variables by
x = c x H , R k = c R k , H , p ( x ) = c − 1 p H ( x / c ) , ρ ( x ) = c − 3 ρ H ( x / c ) , x=cx_{\mathrm H},\qquad R_k=cR_{k,\mathrm H},\qquad p(x)=c^{-1}p_{\mathrm H}(x/c),\qquad\rho(x)=c^{-3}\rho_{\mathrm H}(x/c), x = c x H , R k = c R k , H , p ( x ) = c − 1 p H ( x / c ) , ρ ( x ) = c − 3 ρ H ( x / c ) ,
where the subscript H denotes Hartree variables. This change preserves the particle number. The Hartree energy and the functional E \mathcal{E} E defined in (3) are related by
E H [ p H ; R H , Z ] = c 2 E [ p ; R , Z ] . \mathcal{E}_{\mathrm H}[p_{\mathrm H};R_{\mathrm H},Z]=c^2\mathcal{E}[p;R,Z]. E H [ p H ; R H , Z ] = c 2 E [ p ; R , Z ] .
We omit the positive overall factor c 2 c^2 c 2 : it rescales the energy but preserves the corresponding minimizers, their particle numbers, and the sign of the total energy. Thus the formulas below use ℏ = m e = c = 1 \hbar=m_e=c=1 ℏ = m e = c = 1 , with Coulomb coupling α \alpha α . Let α , λ > 0 \alpha,\lambda>0 α , λ > 0 , and set
f ( p ) 2 = p 1 + p 2 + 2 p 2 1 + p 2 arsinh p , F ( p ) = ∫ 0 p f ( s ) d s , (1) f(p)^2=\frac{p}{\sqrt{1+p^2}}+\frac{2p^2}{1+p^2}\operatorname{arsinh}p,\qquad F(p)=\int_0^p f(s)\,\mathrm{d}s, \tag*{(1)} f ( p ) 2 = 1 + p 2 p + 1 + p 2 2 p 2 arsinh p , F ( p ) = ∫ 0 p f ( s ) d s , (1)
T ( p ) = 1 + p 2 − 1 , τ ( p ) = 8 ∫ 0 p s 2 T ( s ) d s . (2) T(p)=\sqrt{1+p^2}-1,\qquad\tau(p)=8\int_0^p s^2T(s)\,\mathrm{d}s. \tag*{(2)} T ( p ) = 1 + p 2 − 1 , τ ( p ) = 8 ∫ 0 p s 2 T ( s ) d s . (2)
The coefficient 2 in f 2 f^2 f 2 is the standard Engel–Dreizler coefficient. Write
P = F ( p ) , ρ = p 3 3 π 2 , N = ∫ R 3 ρ , Z t o t = ∑ k = 1 K Z k , Q = N − Z t o t . P=F(p),\qquad\rho=\frac{p^3}{3\pi^2},\qquad N=\int_{\mathbb{R}^3}\rho,\qquad Z_{\mathrm{tot}}=\sum_{k=1}^{K}Z_k,\qquad Q=N-Z_{\mathrm{tot}}. P = F ( p ) , ρ = 3 π 2 p 3 , N = ∫ R 3 ρ , Z tot = k = 1 ∑ K Z k , Q = N − Z tot .
The nuclei have pairwise distinct positions R 1 , … , R K R_1,\ldots,R_K R 1 , … , R K and charges Z k > 0 Z_k>0 Z k > 0 . Define
V ( x ) = ∑ k = 1 K Z k ∣ x − R k ∣ , U n n = α ∑ 1 ≤ k < l ≤ K Z k Z l ∣ R k − R l ∣ , V(x)=\sum_{k=1}^{K}\frac{Z_k}{|x-R_k|},\qquad U_{\mathrm{nn}}=\alpha\sum_{1\leq k<l\leq K}\frac{Z_kZ_l}{|R_k-R_l|}, V ( x ) = k = 1 ∑ K ∣ x − R k ∣ Z k , U nn = α 1 ≤ k < l ≤ K ∑ ∣ R k − R l ∣ Z k Z l ,
so that each pair is counted once. The total energy is
E [ p ; R , Z ] = 3 λ 8 π 2 ∫ ∣ ∇ F ( p ) ∣ 2 + 1 8 π 2 ∫ τ ( p ) − α ∫ V ρ + α D [ ρ ] + U n n , (3) \mathcal{E}[p;R,Z]=\frac{3\lambda}{8\pi^2}\int|\nabla F(p)|^2+\frac{1}{8\pi^2}\int\tau(p)-\alpha\int V\rho+\alpha D[\rho]+U_{\mathrm{nn}}, \tag*{(3)} E [ p ; R , Z ] = 8 π 2 3 λ ∫ ∣∇ F ( p ) ∣ 2 + 8 π 2 1 ∫ τ ( p ) − α ∫ V ρ + α D [ ρ ] + U nn , (3)
where
D [ ρ ] = 1 2 ∬ ρ ( x ) ρ ( y ) ∣ x − y ∣ d x d y . D[\rho]=\frac{1}{2}\iint\frac{\rho(x)\rho(y)}{|x-y|}\,\mathrm{d}x\,\mathrm{d}y. D [ ρ ] = 2 1 ∬ ∣ x − y ∣ ρ ( x ) ρ ( y ) d x d y .
Apart from the gradient term and the fixed nuclear repulsion, (3) is the relativistic Thomas–Fermi functional introduced by Vallarta and Rosen [21 ] . In contrast to the nonrelativistic Thomas–Fermi kinetic energy, the local term involving τ \tau τ does not stabilize the energy against collapse at a point nucleus: without a gradient correction, the electronic energy is unbounded below. Various ad hoc modifications were proposed to address this difficulty, including the addition of the nonrelativistic Weizsäcker term; see Gombás [8 § 14, pp. 120 ff.] for the early approaches. The advantage of the first term in (3) is that its form comes from Engel and Dreizler’s formal QED-based gradient expansion [6 ] , rather than an ad hoc choice. The coefficient λ \lambda λ is retained as a parameter; its standard value is 1 / 9 1/9 1/9 .
There is no exchange term. The finite-mass energy class is p ≥ 0 p\geq0 p ≥ 0 , p ∈ L 3 ∩ L 4 p\in L^3\cap L^4 p ∈ L 3 ∩ L 4 , F ( p ) ∈ D 1 , 2 F(p)\in D^{1,2} F ( p ) ∈ D 1 , 2 , and D [ p 3 ] < ∞ D[p^3]<\infty D [ p 3 ] < ∞ .
Here D 1 , 2 ( R 3 ) D^{1,2}(\mathbb{R}^3) D 1 , 2 ( R 3 ) is the completion of C c ∞ ( R 3 ) C^\infty_c(\mathbb{R}^3) C c ∞ ( R 3 ) in the gradient norm, and all unmarked integrals are over R 3 \mathbb{R}^3 R 3 . The first two terms of (3) are denoted by T W [ p ] T_{\mathrm W}[p] T W [ p ] and T T F [ p ] T_{\mathrm{TF}}[p] T TF [ p ] , respectively.
An electronic minimizer always means an attained infimum for fixed R R R and Z Z Z over this energy class, with either N ≤ M N\leq M N ≤ M for a prescribed finite M M M , N = M N=M N = M for a prescribed finite M M M , or no constraint on N N N . In the last case we explicitly require finite mass. We use N N N for the actual electron number of the minimizer, including when the constraint is relaxed. The argument applies to each such minimizer and does not require uniqueness. The underlying existence and ionization theory for the exchange-free functional is developed in [5 ] .
Theorem 2.1 (Excess charge). Let K ≥ 1 K \ge1 K ≥ 1 , let R 1 , … , R K ∈ R 3 R_1,\ldots,R_K \in\mathbb{R}^3 R 1 , … , R K ∈ R 3 be pairwise distinct, and let Z 1 , … , Z K > 0 Z_1,\ldots,Z_K > 0 Z 1 , … , Z K > 0 . Suppose
α λ ≤ 1 1233 , α 3 Z k 2 ≤ 1 137 ( 1 ≤ k ≤ K ) . (4) \alpha\lambda\le\frac{1}{1233}, \qquad\alpha^3 Z_k^2 \le\frac{1}{137} \qquad(1 \le k \le K). \tag*{(4)} α λ ≤ 1233 1 , α 3 Z k 2 ≤ 137 1 ( 1 ≤ k ≤ K ) . (4) If p p p is a finite-mass electronic minimizer and its actual total energy satisfies
E [ p ; R , Z ] ≤ 0 , (5) \mathcal{E}[p;\mathbf{R},\mathbf{Z}] \le0, \tag*{(5)} E [ p ; R , Z ] ≤ 0 , (5) then
N − Z t o t < 5913 1000 λ 3 / 2 K . (6) N-Z_{\mathrm{tot}} < \frac{5913}{1000}\lambda^{3/2}K. \tag*{(6)} N − Z tot < 1000 5913 λ 3/2 K . (6) The bound is independent of the nuclear positions and of the total nuclear charge. No lower bound on the nuclear separations is assumed.
Corollary 2.2 (The physical gradient coefficient). For λ = 1 / 9 \lambda=1/9 λ = 1/9 , 0 < α ≤ 1 / 137 0<\alpha\le1/137 0 < α ≤ 1/137 , and α Z k ≤ 1 \alpha Z_k\le1 α Z k ≤ 1 for every nucleus, an electronic minimizer satisfying (5) obeys
N − Z t o t < 219 1000 K . (7) N-Z_{\mathrm{tot}} < \frac{219}{1000}K. \tag*{(7)} N − Z tot < 1000 219 K . (7)
Proof. The stated assumptions imply (4) , since α 3 Z k 2 = α ( α Z k ) 2 ≤ 1 / 137 \alpha^3 Z_k^2=\alpha(\alpha Z_k)^2\le1/137 α 3 Z k 2 = α ( α Z k ) 2 ≤ 1/137 . Moreover ( 5913 / 1000 ) ( 1 / 9 ) 3 / 2 = 219 / 1000 (5913/1000)(1/9)^{3/2}=219/1000 ( 5913/1000 ) ( 1/9 ) 3/2 = 219/1000 . □ \square □
Corollary 2.3 (Atoms). Let Z , α , λ > 0 Z,\alpha,\lambda>0 Z , α , λ > 0 satisfy
α λ ≤ 1 1233 , α 3 Z 2 ≤ 1 137 . \alpha\lambda\le\frac{1}{1233}, \qquad\alpha^3 Z^2 \le\frac{1}{137}. α λ ≤ 1233 1 , α 3 Z 2 ≤ 137 1 . Every finite-mass atomic minimizer, with free mass, a relaxed mass constraint, or an attained exact mass constraint, satisfies
N − Z < 5913 1000 λ 3 / 2 . N-Z < \frac{5913}{1000}\lambda^{3/2}. N − Z < 1000 5913 λ 3/2 . In particular, for λ = 1 / 9 \lambda=1/9 λ = 1/9 , 0 < α ≤ 1 / 137 0<\alpha\le1/137 0 < α ≤ 1/137 , and α Z ≤ 1 \alpha Z\le1 α Z ≤ 1 , one has N − Z < 219 / 1000 N-Z<219/1000 N − Z < 219/1000 .
Proof. For one nucleus the nuclear repulsion is zero. In the free and relaxed problems, the zero density is admissible, so the minimizing energy is nonpositive. For an exact particle number M > 0 M>0 M > 0 , choose a smooth nonnegative compactly supported momentum q q q with ∫ q 3 / ( 3 π 2 ) = M \int q^3/(3\pi^2)=M ∫ q 3 / ( 3 π 2 ) = M , and set q L ( x ) = L − 1 q ( x / L ) q_L(x)=L^{-1}q(x/L) q L ( x ) = L − 1 q ( x / L ) , with the nucleus placed at the origin. The particle number is unchanged. As L → ∞ L\to\infty L → ∞ , both kinetic energies are O ( L − 2 ) O(L^{-2}) O ( L − 2 ) and both Coulomb contributions are O ( L − 1 ) O(L^{-1}) O ( L − 1 ) . Thus E [ q L ] → 0 \mathcal{E}[q_L]\to0 E [ q L ] → 0 , and an attained exact-mass minimum also has nonpositive energy. The case M = 0 M=0 M = 0 is immediate. The result now follows from Theorem 2.1 with K = 1 K=1 K = 1 in its invariant parameter domain. The stated physical assumptions imply that domain because α 3 Z 2 = α ( α Z ) 2 ≤ 1 / 137 \alpha^3Z^2=\alpha(\alpha Z)^2\le1/137 α 3 Z 2 = α ( α Z ) 2 ≤ 1/137 and α λ ≤ 1 / 1233 \alpha\lambda\le1/1233 α λ ≤ 1/1233 . □ \square □
The nuclear repulsion does not change the electronic Euler equation. It supplies the energy control needed in the molecular proof. In particular, Theorem 2.1 applies at a local minimum of the Born–Oppenheimer energy whenever the electronic infimum is attained; this follows from Corollary 8.2 below. The theorem does not assert existence of such a nuclear minimum. For an arbitrary externally fixed configuration, (5) remains an explicit hypothesis.
The numerical coefficient is an upper bound established by the present comparison argument; no optimality is claimed. Uniformity in the charges means uniformity within (4) . This domain places no restriction on Z t o t Z_{\mathrm{tot}} Z tot when the number of nuclei is allowed to vary.
Variational framework and covariance
The Euler equation and regularity
We record the prerequisites to distinguish qualitative regularity from a uniform maximum-density estimate. No uniform bound on ∥ p ∥ ∞ \lVert p\rVert_{\infty} ∥ p ∥ ∞ will be used.
Lemma 3.1. The finite-mass energy class is equivalent to P = F ( p ) ∈ H 1 ( R 3 ) P=F(p)\in H^{1}(\mathbb{R}^{3}) P = F ( p ) ∈ H 1 ( R 3 ) , P ≥ 0 P\geq0 P ≥ 0 . In this class p ∈ L 6 p\in L^{6} p ∈ L 6 and ρ ∈ L 1 ∩ L 2 \rho\in L^{1}\cap L^{2} ρ ∈ L 1 ∩ L 2 .
Proof. The elementary bound f ( p ) 2 ≤ 3 p f(p)^{2}\leq3p f ( p ) 2 ≤ 3 p implies F ( p ) 2 ≤ ( 4 / 3 ) p 3 F(p)^{2}\leq(4/3)p^{3} F ( p ) 2 ≤ ( 4/3 ) p 3 . Conversely, F ( p ) ≍ p 3 / 2 F(p)\asymp p^{3/2} F ( p ) ≍ p 3/2 for 0 ≤ p ≤ 1 0\leq p\leq1 0 ≤ p ≤ 1 , and F ( p ) ≥ c p F(p)\geq cp F ( p ) ≥ c p for p ≥ 1 p\geq1 p ≥ 1 . Thus
p 3 ≤ C ( P 2 + P 3 ) , p 4 ≤ C ( P 8 / 3 + P 4 ) , p 6 ≤ C ( P 4 + P 6 ) . p^{3}\leq C(P^{2}+P^{3}),\qquad p^{4}\leq C(P^{8/3}+P^{4}),\qquad p^{6}\leq C(P^{4}+P^{6}). p 3 ≤ C ( P 2 + P 3 ) , p 4 ≤ C ( P 8/3 + P 4 ) , p 6 ≤ C ( P 4 + P 6 ) .
Sobolev embedding gives these integrability statements. Interpolation and the Hardy–Littlewood–Sobolev inequality give finite direct energy; each nuclear attraction is finite by ρ ∈ L 1 ∩ L 2 \rho\in L^{1}\cap L^{2} ρ ∈ L 1 ∩ L 2 .
Lemma 3.2. Every nonzero finite-mass electronic minimizer as defined in § 2 satisfies
3 λ 4 Δ P = j ( p ) ( T ( p ) − t ) , j ( p ) = p 2 f ( p ) , t = α φ − η , φ = V − ρ ∗ ∣ ⋅ ∣ − 1 , (8) \frac{3\lambda}{4}\Delta P=j(p)(T(p)-t),\qquad j(p)=\frac{p^{2}}{f(p)},\qquad t=\alpha\varphi-\eta,\qquad\varphi=V-\rho*|\cdot|^{-1}, \tag*{(8)} 4 3 λ Δ P = j ( p ) ( T ( p ) − t ) , j ( p ) = f ( p ) p 2 , t = α φ − η , φ = V − ρ ∗ ∣ ⋅ ∣ − 1 , (8) with η ≥ 0 \eta\geq0 η ≥ 0 . For a relaxed cutoff M M M , η ( N − M ) = 0 \eta(N-M)=0 η ( N − M ) = 0 ; for a free minimizer η = 0 \eta=0 η = 0 . Also
Δ t = 4 π α ρ − 4 π α ∑ k Z k δ R k (9) \Delta t=4\pi\alpha\rho-4\pi\alpha\sum_{k}Z_{k}\delta_{R_{k}} \tag*{(9)} Δ t = 4 π α ρ − 4 π α k ∑ Z k δ R k (9) in distributions.
Proof. Extend the energy evenly to real P ∈ H 1 P\in H^{1} P ∈ H 1 , with n ( P ) = F − 1 ( ∣ P ∣ ) 3 / ( 3 π 2 ) n(P)=F^{-1}(|P|)^{3}/(3\pi^{2}) n ( P ) = F − 1 ( ∣ P ∣ ) 3 / ( 3 π 2 ) and k ( P ) = τ ( F − 1 ( ∣ P ∣ ) ) / ( 8 π 2 ) k(P)=\tau(F^{-1}(|P|))/(8\pi^{2}) k ( P ) = τ ( F − 1 ( ∣ P ∣ )) / ( 8 π 2 ) . Their derivatives at P > 0 P>0 P > 0 are n ′ = p 2 / ( π 2 f ) n'=p^{2}/(\pi^{2}f) n ′ = p 2 / ( π 2 f ) and k ′ = p 2 T / ( π 2 f ) k'=p^{2}T/(\pi^{2}f) k ′ = p 2 T / ( π 2 f ) . They extend continuously through zero, with growth at most C ( ∣ P ∣ + ∣ P ∣ 2 ) C(|P|+|P|^{2}) C ( ∣ P ∣ + ∣ P ∣ 2 ) and C ( ∣ P ∣ + ∣ P ∣ 3 ) C(|P|+|P|^{3}) C ( ∣ P ∣ + ∣ P ∣ 3 ) , respectively. The scalar bounds below give ∣ n ′ ′ ( P ) ∣ ≤ C ( 1 + ∣ P ∣ ) |n''(P)|\leq C(1+|P|) ∣ n ′′ ( P ) ∣ ≤ C ( 1 + ∣ P ∣ ) . Sobolev embedding, Hardy’s inequality at each of finitely many nuclei, and the Coulomb inequality make the mass and energy C 1 C^{1} C 1 on H 1 H^{1} H 1 . At a nonzero minimizer the mass derivative is nonzero. The multiplier rule for a relaxed constraint gives η ≥ 0 \eta\geq0 η ≥ 0 and (8) .
The exact and relaxed mass infima agree. To see this, multiply a relaxed competitor P P P by smooth compactly supported cutoffs taking values in [ 0 , 1 ] [0,1] [ 0 , 1 ] and tending to one, with convergence to P P P in H 1 H^{1} H 1 . The resulting particle numbers do not exceed the original one, and the continuity established above gives convergence of the masses and energies, including the nuclear and Hartree terms. For each such approximation, restore the missing mass with a smooth nonnegative momentum cloud
q L ( x ) = L − 1 q ( ( x − a L ) / L ) . q_{L}(x)=L^{-1}q((x-a_{L})/L). q L ( x ) = L − 1 q (( x − a L ) / L ) .
where q ∈ C c ∞ ( R 3 ) q\in C^{\infty}_{c}(\mathbb{R}^{3}) q ∈ C c ∞ ( R 3 ) is nonnegative and normalized to that missing mass. Choose ∣ a L ∣ / L → ∞ |a_{L}|/L\to\infty ∣ a L ∣/ L → ∞ so that the cloud and the compactly supported approximation have disjoint supports and their distance tends to infinity. The cloud’s kinetic energies are O ( L − 2 ) O(L^{-2}) O ( L − 2 ) , its direct energy is O ( L − 1 ) O(L^{-1}) O ( L − 1 ) , and its interactions with the nuclei and the original density tend to zero. First letting L → ∞ L\to\infty L → ∞ and then removing the cutoff proves equality of the two infima. Nuclear repulsion is fixed in this argument. An attained exact-mass minimizer is consequently a relaxed minimizer as well. Free variations give η = 0 \eta=0 η = 0 in the free case.
Lemma 3.3. Every nonzero finite-mass minimizer has P , p P,p P , p continuous and strictly positive on R 3 \mathbb{R}^{3} R 3 , smooth off the nuclei, with
P ∈ W l o c 2 , q ( 1 < q < 3 ) , ∥ p ∥ ∞ < ∞ , p ( x ) ⟶ 0 uniformly as ∣ x ∣ → ∞ . P\in W_{\mathrm{loc}}^{2,q}\ (1<q<3),\qquad\lVert p\rVert_{\infty}<\infty,\qquad p(x)\longrightarrow0\quad\text{uniformly as }|x|\to\infty. P ∈ W loc 2 , q ( 1 < q < 3 ) , ∥ p ∥ ∞ < ∞ , p ( x ) ⟶ 0 uniformly as ∣ x ∣ → ∞. The Hartree potential belongs to C 0 ∩ L ∞ C_{0}\cap L^{\infty} C 0 ∩ L ∞ .
Proof. The last assertion follows first from ρ ∈ L 1 ∩ L 2 \rho\in L^1 \cap L^2 ρ ∈ L 1 ∩ L 2 , by splitting the Coulomb kernel and approximating the density in both norms. The quotient p 2 / ( f ( p ) F ( p ) ) p^2/(f(p)F(p)) p 2 / ( f ( p ) F ( p )) extends to 3 / 2 3/2 3/2 at zero and is at most 4 + 2 p 4+2p 4 + 2 p . The Euler equation therefore gives
− Δ P ≤ C ( 1 + p ) V P . -\Delta P \le C(1+p)VP. − Δ P ≤ C ( 1 + p ) V P .
The coefficient lies locally in L 8 / 5 L^{8/5} L 8/5 , since p ∈ L 4 p \in L^4 p ∈ L 4 and each ∣ x − R k ∣ − 1 |x-R_k|^{-1} ∣ x − R k ∣ − 1 is locally in L 8 / 3 L^{8/3} L 8/3 , with 1 / 4 + 3 / 8 = 5 / 8 1/4+3/8=5/8 1/4 + 3/8 = 5/8 . We use the local L 2 L^2 L 2 -to-L ∞ L^\infty L ∞ estimate for nonnegative weak subsolutions of − Δ u ≤ a u -\Delta u \le au − Δ u ≤ a u : if a ∈ L q ( B 2 ( x ) ) a \in L^q(B_2(x)) a ∈ L q ( B 2 ( x )) with q > 3 / 2 q>3/2 q > 3/2 , then
sup B 1 ( x ) u ≤ C ∥ u ∥ L 2 ( B 2 ( x ) ) , \sup_{B_1(x)} u \le C\lVert u\rVert_{L^2(B_2(x))}, B 1 ( x ) sup u ≤ C ∥ u ∥ L 2 ( B 2 ( x )) ,
where C C C depends only on q q q and ∥ a ∥ L q ( B 2 ( x ) ) \lVert a\rVert_{L^q(B_2(x))} ∥ a ∥ L q ( B 2 ( x )) . It applies here with q = 8 / 5 q=8/5 q = 8/5 . On unit balls far from all nuclei the coefficient norms are uniformly bounded. Consequently,
sup B 1 ( x ) P ≤ C ∥ P ∥ L 2 ( B 2 ( x ) ) ⟶ 0 \sup_{B_1(x)} P \le C\lVert P\rVert_{L^2(B_2(x))} \longrightarrow0 B 1 ( x ) sup P ≤ C ∥ P ∥ L 2 ( B 2 ( x )) ⟶ 0
gives boundedness and decay. The right side of (8) now lies locally in L q L^q L q for q < 3 q<3 q < 3 , proving continuity and the asserted regularity. Writing Δ P = a ( x ) P \Delta P=a(x)P Δ P = a ( x ) P gives a ∈ L l o c q a \in L^q_{\mathrm{loc}} a ∈ L loc q for some q > 3 / 2 q>3/2 q > 3/2 ; Harnack’s principle and bootstrapping complete the proof. All constants in this qualitative argument may depend on the configuration.
Exact covariance and the parameter domain
The following simultaneous change of lengths, charges, and parameters preserves the form of the functional. Subscripts on E \mathcal{E} E display its parameters explicitly.
Proposition 3.4 (Exact spatial covariance). Let s > 0 s>0 s > 0 , put x = s y x=sy x = sy , and define
p ~ ( y ) = p ( s y ) , R ~ k = R k / s , Z ~ k = Z k / s 3 , (10) \widetilde{p}(y)=p(sy),\qquad\widetilde{R}_k=R_k/s,\qquad\widetilde{Z}_k=Z_k/s^3, \tag*{(10)} p ( y ) = p ( sy ) , R k = R k / s , Z k = Z k / s 3 , (10) λ ~ = λ / s 2 , α ~ = α s 2 . (11) \widetilde{\lambda}=\lambda/s^2,\qquad\widetilde{\alpha}=\alpha s^2. \tag*{(11)} λ = λ / s 2 , α = α s 2 . (11) Then
E λ , α , Z , R [ p ] = s 3 E λ ~ , α ~ , Z ~ , R ~ [ p ~ ] . (12) \mathcal{E}_{\lambda,\alpha,Z,R}[p]=s^3\mathcal{E}_{\widetilde{\lambda},\widetilde{\alpha},\widetilde{Z},\widetilde{R}}[\widetilde{p}]. \tag*{(12)} E λ , α , Z , R [ p ] = s 3 E λ , α , Z , R [ p ] . (12) Moreover,
N = s 3 N ~ , Q = s 3 Q ~ , η ~ = η , (13) N=s^3\widetilde{N},\qquad Q=s^3\widetilde{Q},\qquad\widetilde{\eta}=\eta, \tag*{(13)} N = s 3 N , Q = s 3 Q , η = η , (13) where the chemical potential is written as − η -\eta − η . In particular, the sign of the total energy is preserved, as is the sign of η \eta η .
Proof. The gradient, local kinetic, and Hartree integrals acquire the respective factors s s s , s 3 s^3 s 3 , and s 5 s^5 s 5 . Since ρ ~ ( y ) = ρ ( s y ) \widetilde{\rho}(y)=\rho(sy) ρ ( y ) = ρ ( sy ) and
∑ k Z k ∣ s y − R k ∣ = s 2 ∑ k Z ~ k ∣ y − R ~ k ∣ , \sum_k \frac{Z_k}{|sy-R_k|}=s^2\sum_k\frac{\widetilde{Z}_k}{|y-\widetilde{R}_k|}, k ∑ ∣ sy − R k ∣ Z k = s 2 k ∑ ∣ y − R k ∣ Z k ,
the electron–nuclear integral also acquires the factor s 5 s^5 s 5 when written with the transformed charges. The nuclear repulsion transforms consistently:
s 3 α ~ ∑ k ≠ l Z ~ k Z ~ l ∣ R ~ k − R ~ l ∣ = α ∑ k ≠ l Z k Z l ∣ R k − R l ∣ . s^3\widetilde{\alpha}\sum_{k\ne l}\frac{\widetilde{Z}_k\widetilde{Z}_l}{|\widetilde{R}_k-\widetilde{R}_l|}
=\alpha\sum_{k\ne l}\frac{Z_kZ_l}{|R_k-R_l|}. s 3 α k = l ∑ ∣ R k − R l ∣ Z k Z l = α k = l ∑ ∣ R k − R l ∣ Z k Z l .
These identities prove (12) . The particle-number formula follows by integration, and the charge formula follows because Z t o t = s 3 Z ~ t o t Z_{\mathrm{tot}}=s^3\widetilde{Z}_{\mathrm{tot}} Z tot = s 3 Z tot . Finally, energy and particle number have the same prefactor s 3 s^3 s 3 , so the Lagrange multiplier is unchanged. Equivalently, for the screened potentials,
ϕ ~ ( y ) = s − 2 ϕ ( s y ) , α ~ ϕ ~ ( y ) = α ϕ ( s y ) . \widetilde{\phi}(y)=s^{-2}\phi(sy),\qquad\widetilde{\alpha}\widetilde{\phi}(y)=\alpha\phi(sy). ϕ ( y ) = s − 2 ϕ ( sy ) , α ϕ ( y ) = α ϕ ( sy ) .
The transformation is a bijection between the corresponding admissible classes. It therefore transports electronic minimizers with exact, relaxed, or free particle number, with the particle-number constraint rescaled as in (13) . It also transports joint minimization over the electronic density and nuclear positions. Pairwise distinct nuclear positions remain pairwise distinct.
Useful invariant quantities are
γ = α λ , κ k = α Z k λ , χ k = α 3 Z k 2 = γ κ k 2 , (14) \gamma=\alpha\lambda,\qquad\kappa_k=\frac{\alpha Z_k}{\sqrt{\lambda}},\qquad\chi_k=\alpha^3Z_k^2=\gamma\kappa_k^2, \tag*{(14)} γ = α λ , κ k = λ α Z k , χ k = α 3 Z k 2 = γ κ k 2 , (14)
and
Z k λ 3 / 2 , R k λ , Q λ 3 / 2 . \frac{Z_k}{\lambda^{3/2}},\qquad\frac{R_k}{\sqrt{\lambda}},\qquad\frac{Q}{\lambda^{3/2}}. λ 3/2 Z k , λ R k , λ 3/2 Q .
Taking s = λ s=\sqrt{\lambda} s = λ sets λ ~ = 1 \widetilde{\lambda}=1 λ = 1 , but changes the coupling, charges, and nuclear geometry simultaneously. The factor λ 3 / 2 \lambda^{3/2} λ 3/2 in an excess-charge bound is consistent with this exact covariance.
Lemma 3.5 (Reduction of the parameter domain). To prove (6) under (4) , it suffices to prove it under
0 < α ≤ 1 137 , γ : = α λ ≤ γ 0 : = 1 1233 , α Z k ≤ 1 ( 1 ≤ k ≤ K ) . (15) 0<\alpha\leq\frac{1}{137},\qquad\gamma:=\alpha\lambda\leq\gamma_0:=\frac{1}{1233},\qquad\alpha Z_k\leq1\quad(1\leq k\leq K). \tag*{(15)} 0 < α ≤ 137 1 , γ := α λ ≤ γ 0 := 1233 1 , α Z k ≤ 1 ( 1 ≤ k ≤ K ) . (15) The electronic minimizing and total-energy hypotheses are preserved.
Proof. Take s = ( 137 α ) − 1 / 2 s=(137\alpha)^{-1/2} s = ( 137 α ) − 1/2 in Proposition 3.4 . Then
α ~ = 1 137 , λ ~ = 137 α λ ≤ 1 9 , α ~ Z k = 137 α 3 Z k 2 ≤ 1. \widetilde{\alpha}=\frac{1}{137},\qquad\widetilde{\lambda}=137\alpha\lambda\leq\frac{1}{9},\qquad\widetilde{\alpha}Z_k=\sqrt{137\alpha^3Z_k^2}\leq1. α = 137 1 , λ = 137 α λ ≤ 9 1 , α Z k = 137 α 3 Z k 2 ≤ 1.
The transformed parameters satisfy (15) . Multiplying Q ~ < ( 5913 / 1000 ) λ ~ 3 / 2 K \widetilde{Q}< (5913/1000)\widetilde{\lambda}^{3/2}K Q < ( 5913/1000 ) λ 3/2 K by s 3 s^3 s 3 gives (6) . The remaining assertions follow from the covariance. □
In fact, (4) is precisely the set of parameters that can be mapped into (15) by this covariance: the possible scaling factors satisfy α max k Z k ≤ s ≤ ( 137 α ) − 1 / 2 \alpha\max_k Z_k\leq s\leq(137\alpha)^{-1/2} α max k Z k ≤ s ≤ ( 137 α ) − 1/2 . In particular, there is no separate restriction on α \alpha α in Theorem 2.1 .
Universal exterior estimates
In this section γ = α λ ≤ γ 0 = 1 / 1233 \gamma=\alpha\lambda\leq\gamma_0=1/1233 γ = α λ ≤ γ 0 = 1/1233 . We use the functions in (1) –(2) , the Euler equation (8) , and the qualitative regularity of Lemma 3.3 . Away from the nuclei,
3 λ 4 Δ P = j ˙ ( p ) ( T ( p ) − t ) , Δ t = 4 α 3 π p 3 . (16) \frac{3\lambda}{4}\Delta P=\dot{j}(p)(T(p)-t),\qquad\Delta t=\frac{4\alpha}{3\pi}p^3. \tag*{(16)} 4 3 λ Δ P = j ˙ ( p ) ( T ( p ) − t ) , Δ t = 3 π 4 α p 3 . (16)
No restriction on the nuclear charges, no separation condition, and no global comparison between P P P and t t t is used in this section.
The molecular geometry is encoded by the unweighted function
H ( x ) = ∑ k = 1 K 1 ∣ x − R k ∣ . (17) H(x)=\sum_{k=1}^{K}\frac{1}{\lvert x-R_k\rvert}. \tag*{(17)} H ( x ) = k = 1 ∑ K ∣ x − R k ∣ 1 . (17)
It is harmonic on R 3 ∖ { R 1 , … , R K } \mathbb{R}^3\setminus\{R_1,\ldots,R_K\} R 3 ∖ { R 1 , … , R K } , where
Δ H = 0 , ∣ ∇ H ∣ ≤ ∑ k ∣ x − R k ∣ − 2 ≤ H 2 , τ H : = ∣ ∇ H ∣ 2 H 4 ∈ [ 0 , 1 ] . (18) \Delta H=0,\qquad\lvert\nabla H\rvert\leq\sum_k\lvert x-R_k\rvert^{-2}\leq H^2,\qquad\tau_H:=\frac{\lvert\nabla H\rvert^2}{H^4}\in[0,1]. \tag*{(18)} Δ H = 0 , ∣ ∇ H ∣ ≤ k ∑ ∣ x − R k ∣ − 2 ≤ H 2 , τ H := H 4 ∣ ∇ H ∣ 2 ∈ [ 0 , 1 ] . (18)
Scalar estimates and the form inequality
Lemma 4.1. For p > 0 p>0 p > 0 , f f f is increasing, and
f ( p ) 2 ≤ p + 2 p 3 , f ( p ) 2 p ≤ 17 8 , 0 < p f ′ ( p ) f ( p ) ≤ 7 10 . (19) f(p)^2\leq p+2p^3,\qquad\frac{f(p)^2}{p}\leq\frac{17}{8},\qquad0<\frac{pf'(p)}{f(p)}\leq\frac{7}{10}. \tag*{(19)} f ( p ) 2 ≤ p + 2 p 3 , p f ( p ) 2 ≤ 8 17 , 0 < f ( p ) p f ′ ( p ) ≤ 10 7 . (19) In particular, j ( p ) : = p 2 / f ( p ) j(p):=p^2/f(p) j ( p ) := p 2 / f ( p ) is strictly increasing, and 10 / 17 ≤ F ( p ) / ( p f ( p ) ) ≤ 1 10/17\leq F(p)/(pf(p))\leq1 10/17 ≤ F ( p ) / ( p f ( p )) ≤ 1 . Also
3 4 f ( p ) 2 p ( 2 − p f ′ ( p ) / f ( p ) ) ≤ 43 50 . (20) \frac{3}{4}\frac{f(p)^2}{p(2-pf'(p)/f(p))}\leq\frac{43}{50}. \tag*{(20)} 4 3 p ( 2 − p f ′ ( p ) / f ( p )) f ( p ) 2 ≤ 50 43 . (20)
Proof. Set v = p / 1 + p 2 v = p/\sqrt{1+p^2} v = p / 1 + p 2 , u = v 2 u = v^2 u = v 2 , and L = artanh v = arsinh p L = \operatorname{artanh} v = \operatorname{arsinh} p L = artanh v = arsinh p . If h = f 2 h = f^2 h = f 2 , then
h = v + 2 v 2 L , p h ′ = v ( 1 + v 2 ) + 4 v 2 ( 1 − v 2 ) L > 0 , h = v + 2v^2L,\qquad ph' = v(1+v^2) + 4v^2(1-v^2)L > 0, h = v + 2 v 2 L , p h ′ = v ( 1 + v 2 ) + 4 v 2 ( 1 − v 2 ) L > 0 ,
and
7 h − 5 p h ′ = v ( 2 − 5 u ) + 2 v 2 ( 10 u − 3 ) L . 7h - 5ph' = v(2-5u) + 2v^2(10u-3)L. 7 h − 5 p h ′ = v ( 2 − 5 u ) + 2 v 2 ( 10 u − 3 ) L .
For u ≥ 3 / 10 u \ge3/10 u ≥ 3/10 , use L ≥ v L \ge v L ≥ v to bound the last expression below by v ( 2 − 11 u + 20 u 2 ) > 0 v(2-11u+20u^2)>0 v ( 2 − 11 u + 20 u 2 ) > 0 . For u ≤ 3 / 10 u \le3/10 u ≤ 3/10 , use L ≤ v / ( 1 − u ) L \le v/(1-u) L ≤ v / ( 1 − u ) , taking account of the negative coefficient, to obtain v ( 2 − 13 u + 25 u 2 ) / ( 1 − u ) > 0 v(2-13u+25u^2)/(1-u)>0 v ( 2 − 13 u + 25 u 2 ) / ( 1 − u ) > 0 . Both quadratics have negative discriminants. Thus p f ′ / f = p h ′ / ( 2 h ) ≤ 7 / 10 pf'/f = ph'/(2h) \le7/10 p f ′ / f = p h ′ / ( 2 h ) ≤ 7/10 . Integrating the logarithmic derivative bound gives f ( s ) ≥ f ( p ) ( s / p ) 7 / 10 f(s) \ge f(p)(s/p)^{7/10} f ( s ) ≥ f ( p ) ( s / p ) 7/10 for s ≤ p s \le p s ≤ p , and hence F ( p ) ≥ ( 10 / 17 ) p f ( p ) F(p) \ge(10/17)pf(p) F ( p ) ≥ ( 10/17 ) p f ( p ) . Monotonicity gives F ( p ) ≤ p f ( p ) F(p) \le pf(p) F ( p ) ≤ p f ( p ) . The inequality arsinh p ≤ p \operatorname{arsinh} p \le p arsinh p ≤ p gives h ≤ p + 2 p 2 h \le p+2p^2 h ≤ p + 2 p 2 . Putting w = ( 1 + p 2 ) − 1 / 2 w=(1+p^2)^{-1/2} w = ( 1 + p 2 ) − 1/2 also gives h / p ≤ w + 2 − 2 w 2 ≤ 17 / 8 h/p \le w+2-2w^2 \le17/8 h / p ≤ w + 2 − 2 w 2 ≤ 17/8 .
The localization estimate (20) follows from the algebraic argument in Appendix § 9 .
Lemma 4.2. With A = ( 43 / 50 ) λ A=(43/50)\lambda A = ( 43/50 ) λ , the quadratic-form inequalities
− A Δ + T ( p ) − t ≥ 0 , − A Δ + 1 2 p 2 − t ≥ 0 (21) -A\Delta+ T(p)-t \ge0,\qquad-A\Delta+\tfrac{1}{2}p^2-t\ge0 \tag*{(21)} − A Δ + T ( p ) − t ≥ 0 , − A Δ + 2 1 p 2 − t ≥ 0 (21) hold. Their coefficient does not depend on the number or positions of the nuclei.
Proof. Let v ( P ) = f ( p ) / p 2 v(P)=f(p)/p^2 v ( P ) = f ( p ) / p 2 and D ( P ) = − v ′ ( P ) > 0 D(P)=-v'(P)>0 D ( P ) = − v ′ ( P ) > 0 . Testing (16) with a smooth function b b b supported away from all nuclei and completing a square gives
∫ ( T − t ) b 2 = 3 λ 4 ∫ v Δ P b 2 = 3 λ 4 ∫ D ∣ b ∇ P − v D ∇ b ∣ 2 − 3 λ 4 ∫ v 2 D ∣ ∇ b ∣ 2 . \begin{aligned}
\int(T-t)b^2 = \frac{3\lambda}{4}\int v\Delta_P b^2 \\
&= \frac{3\lambda}{4}\int D\left|b\nabla_P-\frac{v}{D}\nabla b\right|^2-\frac{3\lambda}{4}\int\frac{v^2}{D}|\nabla b|^2.
\end{aligned} ∫ ( T − t ) b 2 = 4 3 λ ∫ v Δ P b 2 = 4 3 λ ∫ D b ∇ P − D v ∇ b 2 − 4 3 λ ∫ D v 2 ∣∇ b ∣ 2 .
Here
v 2 D = f ( p ) 2 p ( 2 − p f ′ / f ) . \frac{v^2}{D}=\frac{f(p)^2}{p(2-pf'/f)}. D v 2 = p ( 2 − p f ′ / f ) f ( p ) 2 .
The first assertion follows from (20) ; the second uses T ( p ) ≤ p 2 / 2 T(p)\le p^2/2 T ( p ) ≤ p 2 /2 . The inequalities extend by density. Removing finitely many points by capacity cutoffs, with Hardy’s inequality at each point, does not change the coefficient. In the application below, all test balls are already disjoint from the nuclei.
The molecular Sommerfeld barrier
Lemma 4.3. Let k , ℓ > 0 k,\ell>0 k , ℓ > 0 and put
θ = 31 40 , M S = 692881 57600 , b ℓ ( r ) = M S 2 k 2 r 4 [ 1 − ( ℓ r ) θ ] − 4 , r > ℓ . (22) \theta=\frac{31}{40},\qquad M_S=\frac{692881}{57600},\qquad b_\ell(r)=\frac{M_S^2}{k^2r^4}\left[1-\left(\frac{\ell}{r}\right)^\theta\right]^{-4},\qquad r>\ell. \tag*{(22)} θ = 40 31 , M S = 57600 692881 , b ℓ ( r ) = k 2 r 4 M S 2 [ 1 − ( r ℓ ) θ ] − 4 , r > ℓ . (22) Then Δ b ℓ ≤ k b ℓ 3 / 2 \Delta b_\ell\le kb_\ell^{3/2} Δ b ℓ ≤ k b ℓ 3/2 in three dimensions.
Proof. For arbitrary M , θ > 0 M,\theta>0 M , θ > 0 , put s = ( ℓ / r ) θ s=(\ell/r)^\theta s = ( ℓ / r ) θ . Direct differentiation gives
Δ b ℓ − k b ℓ 3 / 2 = M 2 k 2 r − 6 ( 1 − s ) − 6 \Delta b_\ell-kb_\ell^{3/2}=\frac{M^2}{k^2}r^{-6}(1-s)^{-6} Δ b ℓ − k b ℓ 3/2 = k 2 M 2 r − 6 ( 1 − s ) − 6 × { 12 − M + 4 ( θ 2 + 7 θ − 6 ) s + ( 16 θ 2 − 28 θ + 12 ) s 2 } . {}\times\{12-M+4(\theta^2+7\theta-6)s+(16\theta^2-28\theta+12)s^2\}. × { 12 − M + 4 ( θ 2 + 7 θ − 6 ) s + ( 16 θ 2 − 28 θ + 12 ) s 2 } . For the values in (22) , the relevant quadratic satisfies
12 + 41 400 s − 9 100 s 2 = 692881 57600 − 9 100 ( s − 41 72 ) 2 . (23) 12+\frac{41}{400}s-\frac{9}{100}s^2=\frac{692881}{57600}-\frac{9}{100}\left(s-\frac{41}{72}\right)^2. \tag*{(23)} 12 + 400 41 s − 100 9 s 2 = 57600 692881 − 100 9 ( s − 72 41 ) 2 . (23) The asserted sign follows. The barrier diverges at r = ℓ r=\ell r = ℓ and tends to zero at infinity.
Proposition 4.4. For every 0 < β < 1 0<\beta<1 0 < β < 1 and every point away from the nuclei,
t ( x ) ≤ U β ( H ( x ) ) , U β ( h ) : = a β h 4 + b β h 2 , (24) t(x)\le U_\beta(H(x)),\qquad U_\beta(h):=a_\beta h^4+b_\beta h^2, \tag*{(24)} t ( x ) ≤ U β ( H ( x )) , U β ( h ) := a β h 4 + b β h 2 , (24) where
a β = 9 π 2 M S 2 128 α 2 ( 1 − β θ ) 4 , b β = 43 λ π 2 50 β 2 . (25) a_\beta=\frac{9\pi^2M_S^2}{128\alpha^2(1-\beta^\theta)^4},\qquad b_\beta=\frac{43\lambda\pi^2}{50\beta^2}. \tag*{(25)} a β = 128 α 2 ( 1 − β θ ) 4 9 π 2 M S 2 , b β = 50 β 2 43 λ π 2 . (25)
In particular, the coefficients are exactly those of the one-centre Sommerfeld comparison.
Proof. For a fixed smearing radius ℓ > 0 \ell>0 ℓ > 0 , use the normalized Dirichlet ball ground state
g ℓ ( y ) = sin ( π ∣ y ∣ / ℓ ) 2 π ℓ ∣ y ∣ 1 { ∣ y ∣ < ℓ } , ∥ g ℓ ∥ 2 = 1 , ∥ ∇ g ℓ ∥ 2 2 = π 2 ℓ 2 . g_{\ell}(y)=\frac{\sin(\pi|y|/\ell)}{\sqrt{2\pi\ell}|y|}\mathbf{1}_{\{|y|<\ell\}},\qquad\|g_{\ell}\|_{2}=1,\qquad\|\nabla g_{\ell}\|_{2}^{2}=\frac{\pi^{2}}{\ell^{2}}. g ℓ ( y ) = 2 π ℓ ∣ y ∣ sin ( π ∣ y ∣/ ℓ ) 1 { ∣ y ∣ < ℓ } , ∥ g ℓ ∥ 2 = 1 , ∥∇ g ℓ ∥ 2 2 = ℓ 2 π 2 .
Set ρ ~ = ρ ∗ g ℓ 2 \widetilde{\rho}=\rho*g_{\ell}^{2} ρ = ρ ∗ g ℓ 2 and u = t ∗ g ℓ 2 − A π 2 / ℓ 2 u=t*g_{\ell}^{2}-A\pi^{2}/\ell^{2} u = t ∗ g ℓ 2 − A π 2 / ℓ 2 . Testing (21) with translates of g ℓ g_{\ell} g ℓ and applying concavity of ρ 2 / 3 \rho^{2/3} ρ 2/3 gives
v ≤ 1 2 ( 3 π 2 ) 2 / 3 ρ ~ 2 / 3 . v\leq\frac{1}{2}(3\pi^{2})^{2/3}\widetilde{\rho}^{2/3}. v ≤ 2 1 ( 3 π 2 ) 2/3 ρ 2/3 .
On D ℓ : = { x : ∣ x − R k ∣ > ℓ for all k } D_{\ell}:=\{x:|x-R_{k}|>\ell\text{ for all }k\} D ℓ := { x : ∣ x − R k ∣ > ℓ for all k } ,
Δ v = 4 π α ρ ~ ≥ k ( v + ) 3 / 2 , k = 2 7 / 2 α 3 π . \Delta v=4\pi\alpha\widetilde{\rho}\geq k(v_{+})^{3/2},\qquad k=\frac{2^{7/2}\alpha}{3\pi}. Δ v = 4 π α ρ ≥ k ( v + ) 3/2 , k = 3 π 2 7/2 α .
Let r k = ∣ x − R k ∣ r_{k}=|x-R_{k}| r k = ∣ x − R k ∣ and B ℓ ( x ) = ∑ k b ℓ ( r k ) B_{\ell}(x)=\sum_{k}b_{\ell}(r_{k}) B ℓ ( x ) = ∑ k b ℓ ( r k ) . By Lemma 4.3,
Δ B ℓ ≤ k ∑ k b ℓ ( r k ) 3 / 2 ≤ k B ℓ 3 / 2 . \Delta B_{\ell}\leq k\sum_{k}b_{\ell}(r_{k})^{3/2}\leq kB_{\ell}^{3/2}. Δ B ℓ ≤ k k ∑ b ℓ ( r k ) 3/2 ≤ k B ℓ 3/2 .
On a positive set of v − B ℓ v-B_{\ell} v − B ℓ , both v v v and B ℓ B_{\ell} B ℓ are positive, and Δ ( v − B ℓ ) > 0 \Delta(v-B_{\ell})>0 Δ ( v − B ℓ ) > 0 . The sum of barriers diverges at every finite boundary point of D ℓ D_{\ell} D ℓ and tends to zero at infinity, whereas v → − η − A π 2 / ℓ 2 < 0 v\to-\eta-A\pi^{2}/\ell^{2}<0 v → − η − A π 2 / ℓ 2 < 0 . The maximum principle gives v ≤ B ℓ v\leq B_{\ell} v ≤ B ℓ .
Newton’s theorem for the radial unit-mass kernel gives ∣ ⋅ ∣ − 1 ∗ g ℓ 2 ≤ ∣ ⋅ ∣ − 1 |\cdot|^{-1}*g_{\ell}^{2}\leq|\cdot|^{-1} ∣ ⋅ ∣ − 1 ∗ g ℓ 2 ≤ ∣ ⋅ ∣ − 1 , with equality outside its support. Thus t ≤ t ∗ g ℓ 2 t\leq t*g_{\ell}^{2} t ≤ t ∗ g ℓ 2 on D ℓ D_{\ell} D ℓ , and consequently
t ( x ) ≤ A π 2 ℓ 2 + M S 2 k 2 ∑ k r k − 4 [ 1 − ( ℓ / r k ) θ ] − 4 . (26) t(x)\leq\frac{A\pi^{2}}{\ell^{2}}+\frac{M_{\mathrm{S}}^{2}}{k^{2}}\sum_{k}r_{k}^{-4}\left[1-\left(\ell/r_{k}\right)^{\theta}\right]^{-4}. \tag*{(26)} t ( x ) ≤ ℓ 2 A π 2 + k 2 M S 2 k ∑ r k − 4 [ 1 − ( ℓ / r k ) θ ] − 4 . (26)
Only the smearing kernel is radial; no symmetry of the electronic density has been used.
At a fixed evaluation point choose ℓ = β / H ( x ) \ell=\beta/H(x) ℓ = β / H ( x ) . This is smaller than every r k r_{k} r k . Put q k = ( r k H ( x ) ) − 1 q_{k}=(r_{k}H(x))^{-1} q k = ( r k H ( x ) ) − 1 , so q k ≥ 0 q_{k}\geq0 q k ≥ 0 and ∑ k q k = 1 \sum_{k}q_{k}=1 ∑ k q k = 1 . Then
∑ k r k − 4 [ 1 − ( ℓ / r k ) θ ] − 4 = H 4 ∑ k q k 4 [ 1 − ( β q k ) θ ] 4 ≤ H 4 ( 1 − β θ ) 4 , \sum_{k}r_{k}^{-4}\left[1-(\ell/r_{k})^{\theta}\right]^{-4}=H^{4}\sum_{k}\frac{q_{k}^{4}}{\left[1-(\beta q_{k})^{\theta}\right]^{4}}\leq\frac{H^{4}}{(1-\beta^{\theta})^{4}}, k ∑ r k − 4 [ 1 − ( ℓ / r k ) θ ] − 4 = H 4 k ∑ [ 1 − ( β q k ) θ ] 4 q k 4 ≤ ( 1 − β θ ) 4 H 4 ,
using q k ≤ 1 q_{k}\leq1 q k ≤ 1 and ∑ k q k 4 ≤ 1 \sum_{k}q_{k}^{4}\leq1 ∑ k q k 4 ≤ 1 . The localization term in (26) is A π 2 H 2 / β 2 A\pi^{2}H^{2}/\beta^{2} A π 2 H 2 / β 2 . This proves (24) –(25) . The localization error is incurred once, rather than once per nucleus. □
Density comparison on harmonic level sets
The next proposition follows the density-comparison strategy of Solovej [18 ] . Its differential calculation uses the exact relativistic functions F F F and T T T .
Proposition 4.5. If γ ≤ 1 / 1233 \gamma\leq1/1233 γ ≤ 1/1233 and 9 / 20 ≤ β ≤ 1 / 2 9/20\leq\beta\leq1/2 9/20 ≤ β ≤ 1/2 , then
T ( p ( x ) ) ≤ κ U β ( H ( x ) ) , κ = 103 100 . (27) T(p(x))\leq\kappa U_{\beta}(H(x)),\qquad\kappa=\frac{103}{100}. \tag*{(27)} T ( p ( x )) ≤ κ U β ( H ( x )) , κ = 100 103 . (27)
This estimate has no total-charge or separation hypothesis.
Proof. Abbreviate a = a β a=a_{\beta} a = a β , b = b β b=b_{\beta} b = b β , and U = a H 4 + b H 2 U=aH^{4}+bH^{2} U = a H 4 + b H 2 . Define
p κ = κ U ( κ U + 2 ) , P κ = F ( p κ ) , z = a H 2 b , p_{\kappa}=\sqrt{\kappa U(\kappa U+2)},\qquad P_{\kappa}=F(p_{\kappa}),\qquad z=\frac{aH^{2}}{b}, p κ = κ U ( κ U + 2 ) , P κ = F ( p κ ) , z = b a H 2 ,
and put
m ( z ) = 12 z + 2 1 + z , n ( z ) = 4 z + 2 1 + z . m(z)=\frac{12z+2}{1+z},\qquad n(z)=\frac{4z+2}{1+z}. m ( z ) = 1 + z 12 z + 2 , n ( z ) = 1 + z 4 z + 2 .
At p = p κ p=p_{\kappa} p = p κ , write s = 1 + p 2 s=\sqrt{1+p^{2}} s = 1 + p 2 , H = f ( p ) 2 / p H=f(p)^{2}/p H = f ( p ) 2 / p , and D = p f ′ ( p ) / f ( p ) D=pf'(p)/f(p) D = p f ′ ( p ) / f ( p ) . Direct differentiation, using Δ H = 0 \Delta H=0 Δ H = 0 , gives the exact identity
Δ P κ ȷ ( p κ ) U = τ H b R ( p , z ) . (28) \frac{\Delta P_{\kappa}}{\jmath(p_{\kappa})U}=\frac{\tau_{H}}{b}\mathcal{R}(p,z). \tag*{(28)} ( p κ ) U Δ P κ = b τ H R ( p , z ) . (28)
where
R ( p , z ) = H 1 + z { s s + 1 m ( z ) + s 2 D − 1 ( s + 1 ) 2 n ( z ) 2 } . (29) \mathcal{R}(p,z)=\frac{\mathcal{H}}{1+z}\left\{\frac{s}{s+1}m(z)+\frac{s^2D-1}{(s+1)^2}n(z)^2\right\}. \tag*{(29)} R ( p , z ) = 1 + z H { s + 1 s m ( z ) + ( s + 1 ) 2 s 2 D − 1 n ( z ) 2 } . (29)
For clarity, for any scalar function J J J , the chain rule gives Δ J ( H ) = J ′ ′ ( H ) ∣ ∇ H ∣ 2 \Delta J(H)=J^{\prime\prime}(H)|\nabla H|^2 Δ J ( H ) = J ′′ ( H ) ∣∇ H ∣ 2 . For its one-centre counterpart J ( 1 / r ) J(1/r) J ( 1/ r ) the radial Laplacian is r − 4 J ′ ′ ( 1 / r ) r^{-4}J^{\prime\prime}(1/r) r − 4 J ′′ ( 1/ r ) . Thus the full molecular Laplacian is the corresponding one-centre expression multiplied by τ H \tau_H τ H . Equivalently, the two coefficients obtained by differentiating F ( T − 1 ( κ U ) ) F(T^{-1}(\kappa U)) F ( T − 1 ( κ U )) , before division by U U U , are H s / ( s + 1 ) \mathcal{H}s/(s+1) H s / ( s + 1 ) multiplying Δ U / U \Delta U/U Δ U / U and H ( s 2 D − 1 ) / ( s + 1 ) 2 \mathcal{H}(s^2D-1)/(s+1)^2 H ( s 2 D − 1 ) / ( s + 1 ) 2 multiplying ∣ ∇ U ∣ 2 / U 2 |\nabla U|^2/U^2 ∣∇ U ∣ 2 / U 2 .
We next prove R < 4 / 3 \mathcal{R}<4/3 R < 4/3 along this barrier. Since θ < 4 / 5 \theta<4/5 θ < 4/5 and β ≥ 9 / 20 \beta\ge9/20 β ≥ 9/20 , β θ ≥ ( 9 / 20 ) 4 / 5 > 13 / 25 \beta^\theta\ge(9/20)^{4/5}>13/25 β θ ≥ ( 9/20 ) 4/5 > 13/25 ; indeed
( 9 / 20 ) 4 − ( 13 / 25 ) 5 = 7464617 2500000000 > 0. (9/20)^4-(13/25)^5=\frac{7464617}{2500000000}>0. ( 9/20 ) 4 − ( 13/25 ) 5 = 2500000000 7464617 > 0.
Using M S > 12 M_S>12 M S > 12 and 9 < π 2 < 10 9<\pi^2<10 9 < π 2 < 10 gives
a α 2 > 3515625 2048 > 1600 , b λ < 3440 81 < 43 , b 2 a < 1849 1600 γ 2 . (30) a\alpha^2>\frac{3515625}{2048}>1600,\qquad\frac{b}{\lambda}<\frac{3440}{81}<43,\qquad\frac{b^2}{a}<\frac{1849}{1600}\gamma^2. \tag*{(30)} a α 2 > 2048 3515625 > 1600 , λ b < 81 3440 < 43 , a b 2 < 1600 1849 γ 2 . (30)
If z ≥ 36 z\ge36 z ≥ 36 , (19) and m ≤ 12 m\le12 m ≤ 12 , n ≤ 4 n\le4 n ≤ 4 imply
R ≤ 17 8 12 + ( 7 / 10 ) 16 1 + z ≤ 493 370 < 4 3 . \mathcal{R}\le\frac{17}{8}\frac{12+(7/10)16}{1+z}\le\frac{493}{370}<\frac{4}{3}. R ≤ 8 17 1 + z 12 + ( 7/10 ) 16 ≤ 370 493 < 3 4 .
Here s / ( s + 1 ) ≤ 1 s/(s+1)\le1 s / ( s + 1 ) ≤ 1 and ( s 2 D − 1 ) / ( s + 1 ) 2 ≤ D (s^2D-1)/(s+1)^2\le D ( s 2 D − 1 ) / ( s + 1 ) 2 ≤ D .
If 0 ≤ z ≤ 36 0\le z\le36 0 ≤ z ≤ 36 , then
w : = T ( p κ ) = κ U = κ b 2 a z ( 1 + z ) < 7046539 6756840000 < 1 900 . w:=T(p_\kappa)=\kappa U=\kappa\frac{b^2}{a}z(1+z)<\frac{7046539}{6756840000}<\frac{1}{900}. w := T ( p κ ) = κ U = κ a b 2 z ( 1 + z ) < 6756840000 7046539 < 900 1 .
Consequently q = p κ 2 = w ( w + 2 ) < 1801 / 810000 < 1 / 400 q=p_\kappa^2=w(w+2)<1801/810000<1/400 q = p κ 2 = w ( w + 2 ) < 1801/810000 < 1/400 . On this range,
H ≤ 1 + 2 q , D ≤ 1 2 + 4 q , s s + 1 ≤ 1 2 + q 8 , s 2 D − 1 ( s + 1 ) 2 ≤ − 1 8 + 6 5 q . (31) \mathcal{H}\le1+2q,\qquad D\le\frac{1}{2}+4q,\qquad\frac{s}{s+1}\le\frac{1}{2}+\frac{q}{8},\qquad\frac{s^2D-1}{(s+1)^2}\le-\frac{1}{8}+\frac{6}{5}q. \tag*{(31)} H ≤ 1 + 2 q , D ≤ 2 1 + 4 q , s + 1 s ≤ 2 1 + 8 q , ( s + 1 ) 2 s 2 D − 1 ≤ − 8 1 + 5 6 q . (31)
For the derivative bound, use h = f 2 ≥ p / s h=f^2\ge p/s h = f 2 ≥ p / s and
h ′ ( p ) = 1 + 2 q ( 1 + q ) 3 / 2 + 4 p arsinh p ( 1 + q ) 2 ≤ 1 + 6 q , D ≤ 1 2 ( 1 + q / 2 ) ( 1 + 6 q ) ≤ 1 2 + 4 q . h'(p)=\frac{1+2q}{(1+q)^{3/2}}+\frac{4p\operatorname{arsinh}p}{(1+q)^2}\le1+6q,\qquad D\le\frac{1}{2}(1+q/2)(1+6q)\le\frac{1}{2}+4q. h ′ ( p ) = ( 1 + q ) 3/2 1 + 2 q + ( 1 + q ) 2 4 p arsinh p ≤ 1 + 6 q , D ≤ 2 1 ( 1 + q /2 ) ( 1 + 6 q ) ≤ 2 1 + 4 q .
The next bound uses s / ( s + 1 ) = 1 / 2 + q / [ 2 ( s + 1 ) 2 ] s/(s+1)=1/2+q/[2(s+1)^2] s / ( s + 1 ) = 1/2 + q / [ 2 ( s + 1 ) 2 ] . For the last one, writing B = ( s 2 D − 1 ) / ( s + 1 ) 2 B=(s^2D-1)/(s+1)^2 B = ( s 2 D − 1 ) / ( s + 1 ) 2 ,
B + 1 8 = 8 ( 1 + q ) D − 8 + ( s + 1 ) 2 8 ( s + 1 ) 2 ≤ 38 q + 32 q 2 32 = 19 16 q + q 2 ≤ 6 5 q . B+\frac{1}{8}=\frac{8(1+q)D-8+(s+1)^2}{8(s+1)^2}\le\frac{38q+32q^2}{32}=\frac{19}{16}q+q^2\le\frac{6}{5}q. B + 8 1 = 8 ( s + 1 ) 2 8 ( 1 + q ) D − 8 + ( s + 1 ) 2 ≤ 32 38 q + 32 q 2 = 16 19 q + q 2 ≤ 5 6 q .
Substitution into (29) gives
R ≤ ( 1 + 2 q ) { g 0 ( z ) + 207 10 q } , g 0 ( z ) = 4 z 2 + 5 z + 1 / 2 ( 1 + z ) 3 . \mathcal{R}\le(1+2q)\left\{g_0(z)+\frac{207}{10}q\right\},\qquad g_0(z)=\frac{4z^2+5z+1/2}{(1+z)^3}. R ≤ ( 1 + 2 q ) { g 0 ( z ) + 10 207 q } , g 0 ( z ) = ( 1 + z ) 3 4 z 2 + 5 z + 1/2 .
The maximum of g 0 g_0 g 0 occurs at z = ( 15 − 1 ) / 4 z=(\sqrt{15}-1)/4 z = ( 15 − 1 ) /4 and is ( 40 15 − 144 ) / 9 < 5 / 4 (40\sqrt{15}-144)/9<5/4 ( 40 15 − 144 ) /9 < 5/4 . Therefore
R ≤ 201 200 ( 5 4 + 207 4000 ) = 1046607 800000 < 4 3 . \mathcal{R}\le\frac{201}{200}\left(\frac{5}{4}+\frac{207}{4000}\right)=\frac{1046607}{800000}<\frac{4}{3}. R ≤ 200 201 ( 4 5 + 4000 207 ) = 800000 1046607 < 3 4 .
This completes the scalar estimate.
At this point it is essential to retain the sign of R \mathcal{R} R correctly. There is no need to assert that R \mathcal{R} R is nonnegative: by (18) ,
τ H R ≤ max { R , 0 } ≤ 4 3 . (32) \tau_H\mathcal{R}\le\max\{\mathcal{R},0\}\le\frac{4}{3}. \tag*{(32)} τ H R ≤ max { R , 0 } ≤ 3 4 . (32)
Thus (28) remains an upper bound even where the second derivative of the scalar composite barrier is negative. Since β ≤ 1 / 2 \beta\le1/2 β ≤ 1/2 ,
λ b ≤ 25 86 π 2 < 3 100 = κ − 1. \frac{\lambda}{b}\le\frac{25}{86\pi^2}<\frac{3}{100}=\kappa-1. b λ ≤ 86 π 2 25 < 100 3 = κ − 1.
It follows that
3 λ 4 Δ P κ ≤ ( κ − 1 ) U j ( p κ ) . \frac{3\lambda}{4}\Delta P_{\kappa}\leq(\kappa-1)Uj(p_{\kappa}). 4 3 λ Δ P κ ≤ ( κ − 1 ) U j ( p κ ) .
On a positive set of P − P κ P-P_{\kappa} P − P κ , one has p > p κ p>p_{\kappa} p > p κ , T ( p ) > κ U T(p)>\kappa U T ( p ) > κ U , and t ≤ U t\leq U t ≤ U . The monotonicity of j j j and (16) yield
3 λ 4 Δ P = j ( p ) ( T ( p ) − t ) > j ( p κ ) ( κ − 1 ) U ≥ 3 λ 4 Δ P κ . \frac{3\lambda}{4}\Delta P=j(p)(T(p)-t)>j(p_{\kappa})(\kappa-1)U\geq\frac{3\lambda}{4}\Delta P_{\kappa}. 4 3 λ Δ P = j ( p ) ( T ( p ) − t ) > j ( p κ ) ( κ − 1 ) U ≥ 4 3 λ Δ P κ .
Near every nucleus P P P is bounded and P κ → ∞ P_{\kappa}\to\infty P κ → ∞ . At infinity both tend to zero. A positive maximum of P − P κ P-P_{\kappa} P − P κ is therefore impossible, proving (27) . □
The exterior subharmonic function
Define
I ( p ) = ∫ 0 p s f ( s ) 2 d s , g ( p ) 2 = 2 I ( p ) , B η = λ α π g ( p ) 2 + ( φ − η / α ) 2 − η α . (33) I(p)=\int_{0}^{p}sf(s)^{2}\,ds,\qquad g(p)^{2}=2I(p),\qquad B_{\eta}=\sqrt{\frac{\lambda}{\alpha\pi}}g(p)^{2}+(\varphi-\eta/\alpha)^{2}-\frac{\eta}{\alpha}. \tag*{(33)} I ( p ) = ∫ 0 p s f ( s ) 2 d s , g ( p ) 2 = 2 I ( p ) , B η = α π λ g ( p ) 2 + ( φ − η / α ) 2 − α η . (33)
Lemma 4.6. The function B η B_{\eta} B η is subharmonic wherever 0 < p ≤ 1 / 2 0<p\leq1/2 0 < p ≤ 1/2 away from the nuclei. Furthermore B η ≥ − φ B_{\eta}\geq-\varphi B η ≥ − φ , and B η → 0 B_{\eta}\to0 B η → 0 uniformly at infinity.
Proof. Put B = B η + η / α B=B_{\eta}+\eta/\alpha B = B η + η / α . The nonnegative Hessian term in the Laplacian of a Euclidean norm, combined with (16) , gives
B Δ B η ≥ λ α π g g ′ ( g ′ ′ g ′ − f ′ f ) ∣ ∇ p ∣ 2 + 4 p 3 T ( p ) 3 α π . (34) B\Delta B_{\eta}\geq\frac{\lambda}{\alpha\pi}gg'\left(\frac{g''}{g'}-\frac{f'}{f}\right)|\nabla p|^{2}+\frac{4p^{3}T(p)}{3\alpha\pi}. \tag*{(34)} B Δ B η ≥ α π λ g g ′ ( g ′ g ′′ − f f ′ ) ∣∇ p ∣ 2 + 3 α π 4 p 3 T ( p ) . (34)
The potential terms cancel exactly because g g ′ = p f 2 gg'=pf^{2} g g ′ = p f 2 . To check the remaining gradient sign, set e ( p ) = p h ′ ( p ) / h ( p ) e(p)=ph'(p)/h(p) e ( p ) = p h ′ ( p ) / h ( p ) , where h = f 2 h=f^{2} h = f 2 , and use v = p / 1 + p 2 v=p/\sqrt{1+p^{2}} v = p / 1 + p 2 and w h ( v ) = v artanh v w_{h}(v)=v\operatorname{artanh}v w h ( v ) = v artanh v . Differentiation gives
e = 1 + v 2 + 4 ( 1 − v 2 ) w h 1 + 2 w h , d e d v = 2 { v ( 1 − 2 w h − 8 w h 2 ) + ( 1 − 3 v 2 ) w h ′ } ( 1 + 2 w h ) 2 . e=\frac{1+v^{2}+4(1-v^{2})w_{h}}{1+2w_{h}},\qquad
\frac{de}{dv}=\frac{2\{v(1-2w_{h}-8w_{h}^{2})+(1-3v^{2})w_{h}'\}}{(1+2w_{h})^{2}}. e = 1 + 2 w h 1 + v 2 + 4 ( 1 − v 2 ) w h , d v d e = ( 1 + 2 w h ) 2 2 { v ( 1 − 2 w h − 8 w h 2 ) + ( 1 − 3 v 2 ) w h ′ } .
For p ≤ 1 / 2 p\leq1/2 p ≤ 1/2 , one has v 2 ≤ 1 / 5 v^{2}\leq1/5 v 2 ≤ 1/5 , 0 ≤ w h ≤ v 2 / ( 1 − v 2 ) = p 2 ≤ 1 / 4 0\leq w_{h}\leq v^{2}/(1-v^{2})=p^{2}\leq1/4 0 ≤ w h ≤ v 2 / ( 1 − v 2 ) = p 2 ≤ 1/4 , and w h ′ > 0 w_{h}'>0 w h ′ > 0 . Thus e e e is increasing. As ( p 2 h ) ′ = p h ( 2 + e ) (p^{2}h)'=ph(2+e) ( p 2 h ) ′ = p h ( 2 + e ) , integration gives
I ( p ) ≥ p 2 h ( p ) 2 + e ( p ) . I(p)\geq\frac{p^{2}h(p)}{2+e(p)}. I ( p ) ≥ 2 + e ( p ) p 2 h ( p ) .
For K ( p ) = p 2 h ( p ) / I ( p ) \mathcal{K}(p)=p^{2}h(p)/I(p) K ( p ) = p 2 h ( p ) / I ( p ) this implies p K ′ / K = 2 + e − K ≥ 0 p\mathcal{K}'/\mathcal{K}=2+e-\mathcal{K}\geq0 p K ′ / K = 2 + e − K ≥ 0 . Since g ′ ′ / g ′ − f ′ / f = 1 2 ( log K ) ′ g''/g'-f'/f=\frac{1}{2}(\log\mathcal{K})' g ′′ / g ′ − f ′ / f = 2 1 ( log K ) ′ , the first term on the right of (34) is nonnegative.
Finally, B η ≥ ∣ φ − η / α ∣ − η / α ≥ − φ B_{\eta}\geq|\varphi-\eta/\alpha|-\eta/\alpha\geq-\varphi B η ≥ ∣ φ − η / α ∣ − η / α ≥ − φ , and the stated limit follows from p , φ → 0 p,\varphi\to0 p , φ → 0 and η ≥ 0 \eta\geq0 η ≥ 0 . In particular B η B_{\eta} B η need not be nonnegative. □
Quartic coercivity with nuclear repulsion
We now use the total energy (3) , including U n n U_{\mathrm{nn}} U nn . We prove the following bound, retaining a positive multiple of ∫ p 4 \int p^{4} ∫ p 4 with constants independent of the number and positions of nuclei. This argument modifies the localization strategy of [17 ] ; the negative Hardy correction is explicitly estimated.
Proposition 5.1 (Quartic coercivity). For every λ > 0 \lambda>0 λ > 0 , 0 < α ≤ 1 / 137 0<\alpha\leq1/137 0 < α ≤ 1/137 , and 0 < α Z k ≤ 1 0<\alpha Z_{k}\leq1 0 < α Z k ≤ 1 , every admissible density satisfies
E [ p ; R , Z ] ≥ 1 16 π 2 ∫ p 4 − N − α C λ Z t o t , C λ = 8 π sinh ( 16 3 λ ) . (35) \mathcal{E}[p;\mathbf{R},\mathbf{Z}]\geq\frac{1}{16\pi^{2}}\int p^{4}-N-\alpha C_{\lambda}Z_{\mathrm{tot}},\qquad C_{\lambda}=\frac{8}{\pi}\sinh\left(\frac{16}{3\lambda}\right). \tag*{(35)} E [ p ; R , Z ] ≥ 16 π 2 1 ∫ p 4 − N − α C λ Z tot , C λ = π 8 sinh ( 3 λ 16 ) . (35)
Consequently, for E + = max { E [ p ; R , Z ] , 0 } E_{+}=\max\{\mathcal{E}[p;\mathbf{R},\mathbf{Z}],0\} E + = max { E [ p ; R , Z ] , 0 } and M > 0 M>0 M > 0 ,
N { p ≥ M } : = 1 3 π 2 ∫ { p ≥ M } p 3 ≤ 16 3 M ( N + α C λ Z t o t + E + ) . (36) N_{\{p\geq M\}}:=\frac{1}{3\pi^{2}}\int_{\{p\geq M\}}p^{3}\leq\frac{16}{3M}\left(N+\alpha C_{\lambda}Z_{\mathrm{tot}}+E_{+}\right). \tag*{(36)} N { p ≥ M } := 3 π 2 1 ∫ { p ≥ M } p 3 ≤ 3 M 16 ( N + α C λ Z tot + E + ) . (36)
We give the proof in four steps. None of these steps requires electronic minimality or the sign condition (5) .
The electrostatic inequality, with its normalization
Suppose first that all nuclear charges equal Z > 0 Z>0 Z > 0 and K ≥ 2 K\ge2 K ≥ 2 . Write Γ k \Gamma_k Γ k for the Voronoi cell of R k R_k R k and
r k ( x ) = ∣ x − R k ∣ , D k = 1 2 min l ≠ k ∣ R k − R l ∣ , B k = B ( R k , D k ) . r_k(x)=|x-R_k|,\qquad D_k=\frac{1}{2}\min_{l\ne k}|R_k-R_l|,\qquad B_k=B(R_k,D_k). r k ( x ) = ∣ x − R k ∣ , D k = 2 1 l = k min ∣ R k − R l ∣ , B k = B ( R k , D k ) .
The balls B k B_k B k are disjoint and are contained in their respective cells. The electrostatic inequality of Lieb–Yau, in the form also used by Lieb–Loss–Siedentop [10 Eq. (3.11)] , gives
D [ ρ ] − Z ∑ k ∫ ρ ( x ) r k ( x ) d x + Z 2 ∑ k < l 1 ∣ R k − R l ∣ ≥ − Z ∑ k ∫ Γ k ρ ( x ) r k ( x ) d x + Z 2 8 ∑ k 1 D k . (37) D[\rho]-Z\sum_k\int\frac{\rho(x)}{r_k(x)}\,dx+Z^2\sum_{k<l}\frac{1}{|R_k-R_l|}\ge-Z\sum_k\int_{\Gamma_k}\frac{\rho(x)}{r_k(x)}\,dx+\frac{Z^2}{8}\sum_k\frac{1}{D_k}. \tag*{(37)} D [ ρ ] − Z k ∑ ∫ r k ( x ) ρ ( x ) d x + Z 2 k < l ∑ ∣ R k − R l ∣ 1 ≥ − Z k ∑ ∫ Γ k r k ( x ) ρ ( x ) d x + 8 Z 2 k ∑ D k 1 . (37)
For clarity, here is a derivation of precisely this normalization. Set
V = ∑ k Z r k , W = max k Z r k , Φ = V − W , U = Z 2 ∑ k < l ∣ R k − R l ∣ − 1 . V=\sum_k\frac{Z}{r_k},\qquad W=\max_k\frac{Z}{r_k},\qquad\Phi=V-W,\qquad U=Z^2\sum_{k<l}|R_k-R_l|^{-1}. V = k ∑ r k Z , W = k max r k Z , Φ = V − W , U = Z 2 k < l ∑ ∣ R k − R l ∣ − 1 .
The nuclear singularities of Φ \Phi Φ cancel. The distribution σ = − Δ Φ / ( 4 π ) \sigma=-\Delta\Phi/(4\pi) σ = − ΔΦ/ ( 4 π ) is a nonnegative measure supported on the Voronoi faces, and Φ = σ ∗ ∣ ⋅ ∣ − 1 \Phi=\sigma*|\cdot|^{-1} Φ = σ ∗ ∣ ⋅ ∣ − 1 . On a face shared by Γ k \Gamma_k Γ k and Γ l \Gamma_l Γ l , with d k l = ∣ R k − R l ∣ d_{kl}=|R_k-R_l| d k l = ∣ R k − R l ∣ and r k = r l = r r_k=r_l=r r k = r l = r ,
d σ = Z d k l 4 π r 3 d S . d\sigma=\frac{Zd_{kl}}{4\pi r^3}\,dS. d σ = 4 π r 3 Z d k l d S .
The measure σ \sigma σ has finite mass: its density on each of the finitely many faces is integrable. Moreover, Φ \Phi Φ is bounded, and Φ ( x ) = ( K − 1 ) Z / ∣ x ∣ + O ( ∣ x ∣ − 2 ) \Phi(x)=(K-1)Z/|x|+O(|x|^{-2}) Φ ( x ) = ( K − 1 ) Z /∣ x ∣ + O ( ∣ x ∣ − 2 ) at infinity. The flux of − ∇ Φ -\nabla\Phi − ∇Φ through large spheres therefore gives σ ( R 3 ) = ( K − 1 ) Z \sigma(\mathbb{R}^3)=(K-1)Z σ ( R 3 ) = ( K − 1 ) Z . Consequently,
D [ σ ] = 1 2 ∫ Φ d σ ≤ 1 2 ∥ Φ ∥ ∞ ( K − 1 ) Z < ∞ . D[\sigma]=\frac{1}{2}\int\Phi\,d\sigma\le\frac{1}{2}\|\Phi\|_\infty(K-1)Z<\infty. D [ σ ] = 2 1 ∫ Φ d σ ≤ 2 1 ∥Φ ∥ ∞ ( K − 1 ) Z < ∞.
Thus the Coulomb square is well defined, and its nonnegativity gives
D [ ρ ] − ∫ Φ ρ ≥ − D [ σ ] . D[\rho]-\int\Phi\rho\ge-D[\sigma]. D [ ρ ] − ∫ Φ ρ ≥ − D [ σ ] .
The two distributional identities
− Δ Φ = 4 π σ , − Δ W = 4 π ∑ k Z δ R k − 4 π σ -\Delta\Phi=4\pi\sigma,\qquad-\Delta W=4\pi\sum_k Z\delta_{R_k}-4\pi\sigma − ΔΦ = 4 π σ , − Δ W = 4 π k ∑ Z δ R k − 4 π σ
give, by integration by parts,
4 π ∫ W d σ = ∫ ∇ W ⋅ ∇ Φ = 8 π ( U − D [ σ ] ) . 4\pi\int W\,d\sigma=\int\nabla W\cdot\nabla\Phi=8\pi(U-D[\sigma]). 4 π ∫ W d σ = ∫ ∇ W ⋅ ∇Φ = 8 π ( U − D [ σ ]) .
These identities can first be justified with small balls about the nuclei removed and a large exterior cutoff, then by passage to the limit. It follows that
U − D [ σ ] = 1 2 ∫ W d σ = Z 2 8 π ∑ k , l ∫ faces k , l d k l r k 4 d S = Z 2 8 π ∑ k ∫ ∂ Γ k n k ⋅ ( x − R k ) r k 4 d S . \begin{aligned}
U-D[\sigma]=\frac{1}{2}\int W\,d\sigma=\frac{Z^2}{8\pi}\sum_{k,l}\int_{\text{faces }k,l}\frac{d_{kl}}{r_k^4}\,dS \\
&=\frac{Z^2}{8\pi}\sum_k\int_{\partial\Gamma_k}\frac{n_k\cdot(x-R_k)}{r_k^4}\,dS.
\end{aligned} U − D [ σ ] = 2 1 ∫ W d σ = 8 π Z 2 k , l ∑ ∫ faces k , l r k 4 d k l d S = 8 π Z 2 k ∑ ∫ ∂ Γ k r k 4 n k ⋅ ( x − R k ) d S .
In the last equality, the two contributions of a shared face sum to d k l / r k 4 d_{kl}/r_k^4 d k l / r k 4 . The divergence theorem on Γ k ∖ B k \Gamma_k\setminus B_k Γ k ∖ B k yields
∫ ∂ Γ k n k ⋅ ( x − R k ) r k 4 d S = 4 π D k − ∫ Γ k ∖ B k d x r k 4 . \int_{\partial\Gamma_k}\frac{n_k\cdot(x-R_k)}{r_k^4}\,dS=\frac{4\pi}{D_k}-\int_{\Gamma_k\setminus B_k}\frac{dx}{r_k^4}. ∫ ∂ Γ k r k 4 n k ⋅ ( x − R k ) d S = D k 4 π − ∫ Γ k ∖ B k r k 4 d x .
The cell lies in the half-space bounded by the bisector to a nearest neighbour. At radius r ≥ D k r\ge D_k r ≥ D k , the solid angle of that half-space is 2 π ( 1 + D k / r ) 2\pi(1+D_k/r) 2 π ( 1 + D k / r ) . Therefore
∫ Γ k ∖ B k d x r k 4 ≤ ∫ D k ∞ 2 π ( r − 2 + D k r − 3 ) d r = 3 π D k . (38) \int_{\Gamma_k\setminus B_k}\frac{dx}{r_k^4}\le\int_{D_k}^{\infty}2\pi(r^{-2}+D_kr^{-3})\,dr=\frac{3\pi}{D_k}. \tag*{(38)} ∫ Γ k ∖ B k r k 4 d x ≤ ∫ D k ∞ 2 π ( r − 2 + D k r − 3 ) d r = D k 3 π . (38)
Thus U − D [ σ ] ≥ ( Z 2 / 8 ) ∑ k D k − 1 U-D[\sigma]\ge(Z^2/8)\sum_kD_k^{-1} U − D [ σ ] ≥ ( Z 2 /8 ) ∑ k D k − 1 , proving (37) . No restriction on Z Z Z , K K K , or the separations is used in this electrostatic lemma.
A signed local Hardy inequality
For 0 < r k < D k 0<r_k<D_k 0 < r k < D k define
A k ( x ) = ( 1 2 r k − r k 2 D k 2 ) x − R k r k , A_k(x)=\left(\frac{1}{2r_k}-\frac{r_k}{2D_k^2}\right)\frac{x-R_k}{r_k}, A k ( x ) = ( 2 r k 1 − 2 D k 2 r k ) r k x − R k ,
and extend A k A_k A k by zero outside B k B_k B k . Its normal component vanishes at ∂ B k \partial B_k ∂ B k . Expanding the nonnegative square ∫ ∣ ∇ v s + ( ∑ k A k ) v s ∣ 2 \int|\nabla v_s+(\sum_k A_k)v_s|^2 ∫ ∣∇ v s + ( ∑ k A k ) v s ∣ 2 gives
∫ ∣ ∇ v s ∣ 2 ≥ ∑ k ∫ B k [ 1 4 r k 2 − 1 D k 2 Y ( r k D k ) ] v s 2 , (39) \int|\nabla v_s|^2\geq\sum_k\int_{B_k}\left[\frac{1}{4r_k^2}-\frac{1}{D_k^2}Y\left(\frac{r_k}{D_k}\right)\right]v_s^2, \tag*{(39)} ∫ ∣∇ v s ∣ 2 ≥ k ∑ ∫ B k [ 4 r k 2 1 − D k 2 1 Y ( D k r k ) ] v s 2 , (39)
where Y ( t ) = 1 + t 2 / 4 Y(t)=1+t^2/4 Y ( t ) = 1 + t 2 /4 .
Indeed,
div A k − ∣ A k ∣ 2 = 1 4 r k 2 − 1 D k 2 − r k 2 4 D k 4 . \operatorname{div}A_k-|A_k|^2=\frac{1}{4r_k^2}-\frac{1}{D_k^2}-\frac{r_k^2}{4D_k^4}. div A k − ∣ A k ∣ 2 = 4 r k 2 1 − D k 2 1 − 4 D k 4 r k 2 .
The inequality holds for v s ∈ H 1 ( R 3 ) v_s\in H^1(\mathbb{R}^3) v s ∈ H 1 ( R 3 ) by approximation. The negative part in (39) is essential and is retained.
Truncation of the density and explicit coercivity
Put ζ = α Z \zeta=\alpha Z ζ = α Z and choose
s = sinh ( 16 ζ 2 3 λ ) , v s = ( p − s ) + , a W = 2 ζ 2 π 2 , C = 8 s ζ 3 π , A l o c = 27308 945 π . (40) s=\sinh\left(\frac{16\zeta^2}{3\lambda}\right),\qquad v_s=(p-s)_+,\qquad a_W=\frac{2\zeta^2}{\pi^2},\qquad C=\frac{8s\zeta^3}{\pi},\qquad A_{\mathrm{loc}}=\frac{27308}{945\pi}. \tag*{(40)} s = sinh ( 3 λ 16 ζ 2 ) , v s = ( p − s ) + , a W = π 2 2 ζ 2 , C = π 8 s ζ 3 , A loc = 945 π 27308 . (40)
For all p ≥ 0 p\geq0 p ≥ 0 ,
f ( p ) 2 = p 1 + p 2 + 2 p 2 1 + p 2 arsinh p ≥ arsinh p . f(p)^2=\frac{p}{\sqrt{1+p^2}}+\frac{2p^2}{1+p^2}\operatorname{arsinh}p\geq\operatorname{arsinh}p. f ( p ) 2 = 1 + p 2 p + 1 + p 2 2 p 2 arsinh p ≥ arsinh p .
For example, p 1 + p 2 ≥ p ≥ arsinh p p\sqrt{1+p^2}\geq p\geq\operatorname{arsinh}p p 1 + p 2 ≥ p ≥ arsinh p bounds the first term from below by arsinh p / ( 1 + p 2 ) \operatorname{arsinh}p/(1+p^2) arsinh p / ( 1 + p 2 ) . Consequently
T W ≥ 3 λ 8 π 2 arsinh ( s ) ∫ ∣ ∇ v s ∣ 2 = a W ∫ ∣ ∇ v s ∣ 2 . T_W\geq\frac{3\lambda}{8\pi^2}\operatorname{arsinh}(s)\int|\nabla v_s|^2=a_W\int|\nabla v_s|^2. T W ≥ 8 π 2 3 λ arsinh ( s ) ∫ ∣∇ v s ∣ 2 = a W ∫ ∣∇ v s ∣ 2 .
For s > 0 s>0 s > 0 the truncation belongs to H 1 H^1 H 1 : its gradient is controlled by this estimate, and v s 2 ≤ s − 1 p 3 v_s^2\leq s^{-1}p^3 v s 2 ≤ s − 1 p 3 on { p > s } \{p>s\} { p > s } gives its L 2 L^2 L 2 bound. The elementary kinetic bound τ ( p ) ≥ 2 p 4 − 8 3 p 3 \tau(p)\geq2p^4-\frac{8}{3}p^3 τ ( p ) ≥ 2 p 4 − 3 8 p 3 gives
T T F ≥ 1 4 π 2 ∫ p 4 − N . T_{\mathrm{TF}}\geq\frac{1}{4\pi^2}\int p^4-N. T TF ≥ 4 π 2 1 ∫ p 4 − N .
Retain 1 16 π 2 ∫ p 4 \frac{1}{16\pi^2}\int p^4 16 π 2 1 ∫ p 4 . In each ball, allocate the remaining quartic term as p 4 / ( 8 π 2 ) + p 4 / ( 16 π 2 ) p^4/(8\pi^2)+p^4/(16\pi^2) p 4 / ( 8 π 2 ) + p 4 / ( 16 π 2 ) ; outside the balls it is 3 p 4 / ( 16 π 2 ) 3p^4/(16\pi^2) 3 p 4 / ( 16 π 2 ) . Combining (37) and (39) leaves three estimates.
The singular term in a ball. On p ≥ 3 s p\geq3s p ≥ 3 s one has v s ≥ 2 p / 3 v_s\geq2p/3 v s ≥ 2 p /3 , and hence
a W v s 2 4 r k 2 + p 4 8 π 2 − ζ p 3 3 π 2 r k ≥ 1 8 π 2 ( p 2 − 4 ζ p 3 r k ) 2 ≥ 0. \frac{a_Wv_s^2}{4r_k^2}+\frac{p^4}{8\pi^2}-\frac{\zeta p^3}{3\pi^2r_k}\geq\frac{1}{8\pi^2}\left(p^2-\frac{4\zeta p}{3r_k}\right)^2\geq0. 4 r k 2 a W v s 2 + 8 π 2 p 4 − 3 π 2 r k ζ p 3 ≥ 8 π 2 1 ( p 2 − 3 r k 4 ζ p ) 2 ≥ 0.
On p < 3 s p<3s p < 3 s , discard the nonnegative Hardy term. Extend the pointwise negative lower bound to all of R 3 \mathbb{R}^3 R 3 . The minimizing value is p min ( r ) = min { 3 s , 2 ζ / r } p_{\min}(r)=\min\{3s,2\zeta/r\} p m i n ( r ) = min { 3 s , 2 ζ / r } , with transition radius r ∗ = 2 ζ / ( 3 s ) r_*=2\zeta/(3s) r ∗ = 2 ζ / ( 3 s ) . Direct integration gives
∫ B k ∩ { p < 3 s } ( p 4 8 π 2 − ζ p 3 3 π 2 r k ) d x ≥ − 8 s ζ 3 π = − C . (41) \int_{B_k\cap\{p<3s\}}\left(\frac{p^4}{8\pi^2}-\frac{\zeta p^3}{3\pi^2r_k}\right)\,dx\geq-\frac{8s\zeta^3}{\pi}=-C. \tag*{(41)} ∫ B k ∩ { p < 3 s } ( 8 π 2 p 4 − 3 π 2 r k ζ p 3 ) d x ≥ − π 8 s ζ 3 = − C . (41)
The radial integrals inside and outside r ∗ r_* r ∗ each equal − 4 s ζ 3 / π -4s\zeta^3/\pi − 4 s ζ 3 / π .
The negative Hardy correction. Use v s 2 ≤ p 2 v_s^2\leq p^2 v s 2 ≤ p 2 and pointwise minimization to obtain
p 4 16 π 2 − a W D k 2 Y ( r k / D k ) v s 2 ≥ − 16 ζ 4 π 2 D k 4 Y ( r k / D k ) 2 . \frac{p^4}{16\pi^2}-\frac{a_W}{D_k^2}Y(r_k/D_k)v_s^2\geq-\frac{16\zeta^4}{\pi^2D_k^4}Y(r_k/D_k)^2. 16 π 2 p 4 − D k 2 a W Y ( r k / D k ) v s 2 ≥ − π 2 D k 4 16 ζ 4 Y ( r k / D k ) 2 .
Since
∫ 0 1 r 2 ( 1 + r 2 / 4 ) 2 d r = 743 1680 , \int_0^1r^2(1+r^2/4)^2\,dr=\frac{743}{1680}, ∫ 0 1 r 2 ( 1 + r 2 /4 ) 2 d r = 1680 743 ,
the integrated correction is at least
64 ζ 4 π D k 743 1680 = − 2972 ζ 4 105 π D k . (42) \frac{64\zeta^{4}}{\pi D_k}\frac{743}{1680}=-\frac{2972\zeta^{4}}{105\pi D_k}. \tag*{(42)} π D k 64 ζ 4 1680 743 = − 105 π D k 2972 ζ 4 . (42)
The exterior of a ball in its Voronoi cell. Pointwise minimization gives
3 p 4 16 π 2 − ζ p 3 3 π 2 r k ≥ − 16 ζ 4 81 π 2 r k 4 . \frac{3p^{4}}{16\pi^{2}}-\frac{\zeta p^{3}}{3\pi^{2}r_k}\geq-\frac{16\zeta^{4}}{81\pi^{2}r_k^{4}}. 16 π 2 3 p 4 − 3 π 2 r k ζ p 3 ≥ − 81 π 2 r k 4 16 ζ 4 .
By (38) , its integral is at least
− 16 ζ 4 27 π D k . (43) -\frac{16\zeta^{4}}{27\pi D_k}. \tag*{(43)} − 27 π D k 16 ζ 4 . (43)
Combining (41) –(43) proves
E [ p ; R , ( Z , … , Z ) ] ≥ 1 16 π 2 ∫ p 4 − N − C K + ( α Z 2 8 − A l o c ζ 4 ) ∑ k 1 D k . (44) \mathcal{E}[p;\mathbf{R},(Z,\ldots,Z)]\geq\frac{1}{16\pi^{2}}\int p^{4}-N-CK+\left(\frac{\alpha Z^{2}}{8}-A_{\mathrm{loc}}\zeta^{4}\right)\sum_k\frac{1}{D_k}. \tag*{(44)} E [ p ; R , ( Z , … , Z )] ≥ 16 π 2 1 ∫ p 4 − N − C K + ( 8 α Z 2 − A loc ζ 4 ) k ∑ D k 1 . (44)
For K = 1 K=1 K = 1 the same estimate holds with the distance term omitted, using the ordinary global Hardy inequality and the same high/low truncation. For K = 0 K=0 K = 0 , E ≥ ( 16 π 2 ) − 1 ∫ p 4 − N \mathcal{E}\geq(16\pi^{2})^{-1}\int p^{4}-N E ≥ ( 16 π 2 ) − 1 ∫ p 4 − N is immediate.
Individual and unequal nuclear charges
Suppose 0 < α ≤ 1 / 137 0<\alpha\leq1/137 0 < α ≤ 1/137 and 0 < α Z ≤ 1 0<\alpha Z\leq1 0 < α Z ≤ 1 . The distance coefficient in (44) is nonnegative, because A l o c < 10 A_{\mathrm{loc}}<10 A loc < 10 and
α Z 2 8 − A l o c ζ 4 = ζ 2 ( 1 8 α − A l o c ζ 2 ) ≥ ζ 2 ( 137 8 − A l o c ) > 0. \frac{\alpha Z^{2}}{8}-A_{\mathrm{loc}}\zeta^{4}=\zeta^{2}\left(\frac{1}{8\alpha}-A_{\mathrm{loc}}\zeta^{2}\right)\geq\zeta^{2}\left(\frac{137}{8}-A_{\mathrm{loc}}\right)>0. 8 α Z 2 − A loc ζ 4 = ζ 2 ( 8 α 1 − A loc ζ 2 ) ≥ ζ 2 ( 8 137 − A loc ) > 0.
The constant in (40) satisfies C ≤ C λ = ( 8 / π ) sinh ( 16 / ( 3 λ ) ) C\leq C_{\lambda}=(8/\pi)\sinh(16/(3\lambda)) C ≤ C λ = ( 8/ π ) sinh ( 16/ ( 3 λ )) . After discarding the distance term, the equal-charge bound is therefore
E ≥ 1 16 π 2 ∫ p 4 − N − C λ K . \mathcal{E}\geq\frac{1}{16\pi^{2}}\int p^{4}-N-C_{\lambda}K. E ≥ 16 π 2 1 ∫ p 4 − N − C λ K .
For unequal charges, put Z ∗ = 1 / α Z_{*}=1/\alpha Z ∗ = 1/ α and take independent random variables ξ k \xi_k ξ k equal to Z ∗ Z_{*} Z ∗ with probability α Z k \alpha Z_k α Z k , and zero otherwise. For each fixed density and geometry the energy is separately affine in the nuclear charges. Independence gives
E E [ p ; R , ξ ] = E [ p ; R , Z ] . \mathbb{E}\,\mathcal{E}[p;\mathbf{R},\boldsymbol{\xi}]=\mathcal{E}[p;\mathbf{R},\mathbf{Z}]. E E [ p ; R , ξ ] = E [ p ; R , Z ] .
For each realization, let K ξ = # { k : ξ k = Z ∗ } K_{\xi}=\#\{k:\xi_k=Z_{*}\} K ξ = # { k : ξ k = Z ∗ } . The active nuclei all have charge Z ∗ Z_{*} Z ∗ . Applying the preceding equal-charge bound with Z = Z ∗ Z=Z_{*} Z = Z ∗ and K = K ξ K=K_{\xi} K = K ξ gives
E [ p ; R , ξ ] ≥ 1 16 π 2 ∫ p 4 − N − C λ K ξ . \mathcal{E}[p;\mathbf{R},\boldsymbol{\xi}]\geq\frac{1}{16\pi^{2}}\int p^{4}-N-C_{\lambda}K_{\xi}. E [ p ; R , ξ ] ≥ 16 π 2 1 ∫ p 4 − N − C λ K ξ .
The cases of one or no active nuclei were covered above. Taking expectations in this inequality and using
E K ξ = ∑ k = 1 K P ( ξ k = Z ∗ ) = α ∑ k = 1 K Z k = α Z t o t \mathbb{E}K_{\xi}=\sum_{k=1}^{K}\mathbb{P}(\xi_k=Z_{*})=\alpha\sum_{k=1}^{K}Z_k=\alpha Z_{\mathrm{tot}} E K ξ = k = 1 ∑ K P ( ξ k = Z ∗ ) = α k = 1 ∑ K Z k = α Z tot
proves (35) . Finally, p 3 ≤ p 4 / M p^{3}\leq p^{4}/M p 3 ≤ p 4 / M on { p ≥ M } \{p\geq M\} { p ≥ M } gives (36) and completes the proof of Proposition 5.1.
A nonlinear comparison with an integrable source
We continue to assume only 0 < γ = α λ ≤ γ 0 = 1 / 1233 0<\gamma=\alpha\lambda\leq\gamma_{0}=1/1233 0 < γ = α λ ≤ γ 0 = 1/1233 until the coercivity estimate is invoked in § 7 . The comparison function below agrees with a multiple of P 4 / 3 P^{4/3} P 4/3 at small density and with its tangent thereafter. Convexity preserves the sign of the gradient term in the Euler equation. The scalar comparison holds up to a large explicit momentum; any failure above it is represented by a Newton potential.
The scalar comparison and its high-density cutoff
Lemma 6.1 (A convex comparison with a tangent continuation). Let
0 < γ = α λ ≤ γ 0 : = 1 1233 , L = γ 0 γ ≥ 1 , p 0 = 4 γ , P 0 = F ( p 0 ) , 0 < \gamma= \alpha\lambda\le\gamma_0 := \frac{1}{1233}, \qquad L = \frac{\gamma_0}{\gamma} \ge1, \qquad p_0 = 4\gamma, \qquad P_0 = F(p_0), 0 < γ = α λ ≤ γ 0 := 1233 1 , L = γ γ 0 ≥ 1 , p 0 = 4 γ , P 0 = F ( p 0 ) , and put
β 0 = 3 4 , c = β 0 2 ( 3 2 ) 4 / 3 , m = 4 c 3 P 0 1 / 3 , b = c 3 P 0 4 / 3 . \beta_0 = \frac{3}{4}, \qquad c = \frac{\beta_0}{2}\left(\frac{3}{2}\right)^{4/3}, \qquad m = \frac{4c}{3}P_0^{1/3}, \qquad b = \frac{c}{3}P_0^{4/3}. β 0 = 4 3 , c = 2 β 0 ( 2 3 ) 4/3 , m = 3 4 c P 0 1/3 , b = 3 c P 0 4/3 . Define on [ 0 , ∞ ) [0,\infty) [ 0 , ∞ )
G ( P ) = { c P 4 / 3 , 0 ≤ P ≤ P 0 , m P − b , P ≥ P 0 . (45) G(P) =
\begin{cases}
cP^{4/3}, & 0 \le P \le P_0,\\
mP-b, & P \ge P_0.
\end{cases} \tag*{(45)} G ( P ) = { c P 4/3 , m P − b , 0 ≤ P ≤ P 0 , P ≥ P 0 . (45) Then G G G is nonnegative, convex, and continuously differentiable, with G ( 0 ) = G ′ ( 0 ) = 0 G(0)=G'(0)=0 G ( 0 ) = G ′ ( 0 ) = 0 , 0 < G ′ ( P ) ≤ m 0<G'(P)\le m 0 < G ′ ( P ) ≤ m for P > 0 P>0 P > 0 , and G ( P ) ≤ m P G(P)\le mP G ( P ) ≤ m P . The function
D G ( p ) = G ( F ( p ) ) + γ p f ( p ) π G ′ ( F ( p ) ) − T ( p ) ( p > 0 ) (46) D_G(p)=G(F(p))+\frac{\gamma p f(p)}{\pi G'(F(p))}-T(p) \qquad(p>0) \tag*{(46)} D G ( p ) = G ( F ( p )) + π G ′ ( F ( p )) γ p f ( p ) − T ( p ) ( p > 0 ) (46) extends continuously by D G ( 0 ) = 0 D_G(0)=0 D G ( 0 ) = 0 . With
d = 361 1000 γ 2 , p c u t = e 180 L , d=\frac{361}{1000}\gamma^2,\qquad p_{\mathrm{cut}}=e^{180L}, d = 1000 361 γ 2 , p cut = e 180 L , one has
D G ( p ) ≤ d ( 0 ≤ p ≤ p c u t ) , D G ( p ) < 0 ( p 0 ≤ p ≤ p c u t ) . (47) D_G(p)\le d \quad(0\le p\le p_{\mathrm{cut}}),\qquad D_G(p)<0 \quad(p_0\le p\le p_{\mathrm{cut}}). \tag*{(47)} D G ( p ) ≤ d ( 0 ≤ p ≤ p cut ) , D G ( p ) < 0 ( p 0 ≤ p ≤ p cut ) . (47) Moreover,
3 2 γ ≤ m ≤ 15001 10000 γ , b ≥ 2 γ 2 . (48) \frac{3}{2}\sqrt{\gamma}\le m\le\frac{15001}{10000}\sqrt{\gamma},\qquad b\ge2\gamma^2. \tag*{(48)} 2 3 γ ≤ m ≤ 10000 15001 γ , b ≥ 2 γ 2 . (48)
Proof. The two expressions in (45) have the same value and derivative at P 0 P_0 P 0 . Their derivative increases from zero to m m m and then remains constant. This proves the asserted regularity and convexity. The inequality G ( P ) ≤ m P G(P)\le mP G ( P ) ≤ m P follows directly on each branch.
We first record the elementary small-p p p bounds
p ≤ f ( p ) ≤ p ( 1 + p 2 ) , 2 3 p 3 / 2 ≤ F ( p ) ≤ 2 3 p 3 / 2 ( 1 + 3 7 p 2 ) ( 0 ≤ p ≤ 1 / 100 ) . (49) \sqrt{p}\le f(p)\le\sqrt{p}(1+p^2),\qquad\frac{2}{3}p^{3/2}\le F(p)\le\frac{2}{3}p^{3/2}\left(1+\frac{3}{7}p^2\right)\qquad(0\le p\le1/100). \tag*{(49)} p ≤ f ( p ) ≤ p ( 1 + p 2 ) , 3 2 p 3/2 ≤ F ( p ) ≤ 3 2 p 3/2 ( 1 + 7 3 p 2 ) ( 0 ≤ p ≤ 1/100 ) . (49)
The upper bounds follow from f ( p ) 2 ≤ p + 2 p 3 f(p)^2\le p+2p^3 f ( p ) 2 ≤ p + 2 p 3 . To prove the lower bound, use arsinh p ≥ p 1 + p 2 \operatorname{arsinh}p\ge\frac{p}{\sqrt{1+p^2}} arsinh p ≥ 1 + p 2 p , which gives
f ( p ) 2 ≥ p 1 + 3 p 2 ( 1 + p 2 ) 3 / 2 ≥ p ( 0 ≤ p ≤ 1 ) . f(p)^2\ge p\frac{1+3p^2}{(1+p^2)^{3/2}}\ge p\qquad(0\le p\le1). f ( p ) 2 ≥ p ( 1 + p 2 ) 3/2 1 + 3 p 2 ≥ p ( 0 ≤ p ≤ 1 ) .
Indeed, for x = p 2 ∈ [ 0 , 1 ] x=p^2\in[0,1] x = p 2 ∈ [ 0 , 1 ] , ( 1 + 3 x ) 2 − ( 1 + x ) 3 = x ( 3 + 6 x − x 2 ) ≥ 0 (1+3x)^2-(1+x)^3=x(3+6x-x^2)\ge0 ( 1 + 3 x ) 2 − ( 1 + x ) 3 = x ( 3 + 6 x − x 2 ) ≥ 0 . Integration proves the lower bound on F F F . We shall also use
T ( p ) ≥ 1 2 p 2 − 1 8 p 4 . (50) T(p)\ge\frac{1}{2}p^2-\frac{1}{8}p^4. \tag*{(50)} T ( p ) ≥ 2 1 p 2 − 8 1 p 4 . (50)
This follows from T ( p ) = p 2 1 + p 2 + 1 ≥ p 2 2 + p 2 / 2 T(p)=\frac{p^2}{\sqrt{1+p^2}+1}\ge\frac{p^2}{2+p^2/2} T ( p ) = 1 + p 2 + 1 p 2 ≥ 2 + p 2 /2 p 2 and ( 1 + v ) − 1 ≥ 1 − v (1+v)^{-1}\ge1-v ( 1 + v ) − 1 ≥ 1 − v .
Since p 0 ≤ 4 / 1233 < 1 / 100 p_0\le4/1233<1/100 p 0 ≤ 4/1233 < 1/100 , (49) applies at p 0 p_0 p 0 . Substitution into m m m and b b b gives
3 2 ≤ m γ ≤ 3 2 ( 1 + 3 7 p 0 2 ) 1 / 3 , b ≥ 1 8 p 0 2 = 2 γ 2 . \frac{3}{2}\le\frac{m}{\sqrt{\gamma}}\le\frac{3}{2}\left(1+\frac{3}{7}p_0^2\right)^{1/3},\qquad b\ge\frac{1}{8}p_0^2=2\gamma^2. 2 3 ≤ γ m ≤ 2 3 ( 1 + 7 3 p 0 2 ) 1/3 , b ≥ 8 1 p 0 2 = 2 γ 2 .
Set
Δ 0 = 16 1233 2 = 16 1520289 , M 0 = 15001 10000 , π − = 333 106 < π . \Delta_0=\frac{16}{1233^2}=\frac{16}{1520289},\qquad M_0=\frac{15001}{10000},\qquad\pi_-=\frac{333}{106}<\pi. Δ 0 = 123 3 2 16 = 1520289 16 , M 0 = 10000 15001 , π − = 106 333 < π .
The upper slope bound follows from the exact positive gap
M 0 3 − ( 3 2 ) 3 ( 1 + 3 7 Δ 0 ) = 86689437366383 131383000000000000 > 0. M_0^3-\left(\frac{3}{2}\right)^3\left(1+\frac{3}{7}\Delta_0\right)=\frac{86689437366383}{131383000000000000}>0. M 0 3 − ( 2 3 ) 3 ( 1 + 7 3 Δ 0 ) = 131383000000000000 86689437366383 > 0.
The power branch: 0 ≤ p ≤ p 0 0 \le p \le p_0 0 ≤ p ≤ p 0 . For 0 ≤ x ≤ 3 0 \le x \le3 0 ≤ x ≤ 3 one has ( 1 + 3 x / 7 ) 4 / 3 ≤ 1 + x (1+3x/7)^{4/3} \le1+x ( 1 + 3 x /7 ) 4/3 ≤ 1 + x ; this follows, for example, by differentiating log ( 1 + x ) − ( 4 / 3 ) log ( 1 + 3 x / 7 ) \log(1+x) - (4/3)\log(1+3x/7) log ( 1 + x ) − ( 4/3 ) log ( 1 + 3 x /7 ) . Using (49) therefore gives
3 8 p 2 ≤ G ( F ( p ) ) ≤ 3 8 ( 1 + p 0 2 ) p 2 , G ′ ( F ( p ) ) ≥ 3 4 p . \frac{3}{8}p^2 \le G(F(p)) \le\frac{3}{8}(1+p_0^2)p^2,\qquad G'(F(p)) \ge\frac{3}{4}\sqrt{p}. 8 3 p 2 ≤ G ( F ( p )) ≤ 8 3 ( 1 + p 0 2 ) p 2 , G ′ ( F ( p )) ≥ 4 3 p .
Together with (50) ,
D G ( p ) ≤ − ( 1 8 − 1 2 p 0 2 ) p 2 + γ ( 1 + p 0 2 ) π β 0 p ≤ − A 0 p 2 + B 0 γ p , D_G(p) \le-\left(\frac{1}{8}-\frac{1}{2}p_0^2\right)p^2+\frac{\gamma(1+p_0^2)}{\pi\beta_0}p \le-A_0p^2+B_0\gamma p, D G ( p ) ≤ − ( 8 1 − 2 1 p 0 2 ) p 2 + π β 0 γ ( 1 + p 0 2 ) p ≤ − A 0 p 2 + B 0 γ p ,
where
A 0 = 1 8 − 1 2 Δ 0 > 0 , B 0 = 1 + Δ 0 π − β 0 . A_0=\frac{1}{8}-\frac{1}{2}\Delta_0>0,\qquad B_0=\frac{1+\Delta_0}{\pi-\beta_0}. A 0 = 8 1 − 2 1 Δ 0 > 0 , B 0 = π − β 0 1 + Δ 0 .
The maximum of the last quadratic on p ≥ 0 p\ge0 p ≥ 0 is B 0 2 γ 2 / ( 4 A 0 ) B_0^2\gamma^2/(4A_0) B 0 2 γ 2 / ( 4 A 0 ) . The exact bound is
B 0 2 4 A 0 = 33241694034468992 92262451740651801 < 361 1000 , \frac{B_0^2}{4A_0}=\frac{33241694034468992}{92262451740651801}<\frac{361}{1000}, 4 A 0 B 0 2 = 92262451740651801 33241694034468992 < 1000 361 ,
with gap
361 1000 − B 0 2 4 A 0 = 65051043906308161 92262451740651801000 > 0. \frac{361}{1000}-\frac{B_0^2}{4A_0}=\frac{65051043906308161}{92262451740651801000}>0. 1000 361 − 4 A 0 B 0 2 = 92262451740651801000 65051043906308161 > 0.
The same estimates give the stated continuous extension at p = 0 p=0 p = 0 : G ( F ( p ) ) = O ( p 2 ) G(F(p))=O(p^2) G ( F ( p )) = O ( p 2 ) , G ′ ( F ( p ) ) ∼ β 0 p G'(F(p))\sim\beta_0\sqrt{p} G ′ ( F ( p )) ∼ β 0 p , and the quotient term in (46) is O ( γ p ) O(\gamma p) O ( γ p ) .
The initial tangent branch: p 0 ≤ p ≤ 1 / 100 p_0\le p\le1/100 p 0 ≤ p ≤ 1/100 . Here
D G ( p ) = m F ( p ) − b + γ p f ( p ) π m − T ( p ) . D_G(p)=mF(p)-b+\frac{\gamma p f(p)}{\pi m}-T(p). D G ( p ) = m F ( p ) − b + π m γ p f ( p ) − T ( p ) .
The preceding estimates imply
D G ( p ) ≤ C 0 γ p 3 / 2 − 39999 80000 p 2 − 2 γ 2 , D_G(p)\le C_0\sqrt{\gamma}\,p^{3/2}-\frac{39999}{80000}p^2-2\gamma^2, D G ( p ) ≤ C 0 γ p 3/2 − 80000 39999 p 2 − 2 γ 2 ,
where
C 0 = M 0 ( 2 3 + 2 70000 ) + 1 + 1 / 10000 π − ( 3 / 2 ) < 49 40 . C_0=M_0\left(\frac{2}{3}+\frac{2}{70000}\right)+\frac{1+1/10000}{\pi-(3/2)}<\frac{49}{40}. C 0 = M 0 ( 3 2 + 70000 2 ) + π − ( 3/2 ) 1 + 1/10000 < 40 49 .
For reproducibility, 49 / 40 − C 0 = 632219143 / 49950000000 > 0 49/40-C_0=632219143/49950000000>0 49/40 − C 0 = 632219143/49950000000 > 0 . Writing u = p / γ ≥ 4 u=p/\gamma\ge4 u = p / γ ≥ 4 , we obtain
D G ( p ) γ 2 ≤ h ( u ) : = 49 40 u 3 / 2 − 39999 80000 u 2 − 2. \frac{D_G(p)}{\gamma^2}\le h(u):=\frac{49}{40}u^{3/2}-\frac{39999}{80000}u^2-2. γ 2 D G ( p ) ≤ h ( u ) := 40 49 u 3/2 − 80000 39999 u 2 − 2.
One has h ( 4 ) = − 999 / 5000 < 0 h(4)=-999/5000<0 h ( 4 ) = − 999/5000 < 0 and
h ′ ( u ) = u ( 147 80 − 39999 40000 u ) < 0 ( u ≥ 4 ) . h'(u)=\sqrt{u}\left(\frac{147}{80}-\frac{39999}{40000}\sqrt{u}\right)<0\qquad(u\ge4). h ′ ( u ) = u ( 80 147 − 40000 39999 u ) < 0 ( u ≥ 4 ) .
This proves strict negativity throughout this interval.
The remaining tangent branch. The global scalar inequality F ( p ) ≤ p f ( p ) F(p)\le pf(p) F ( p ) ≤ p f ( p ) gives
D G ( p ) ≤ γ C m p f ( p ) − T ( p ) , C m : = m γ + γ π m . (51) D_G(p)\le\sqrt{\gamma}\,C_m pf(p)-T(p),\qquad C_m:=\frac{m}{\sqrt{\gamma}}+\frac{\sqrt{\gamma}}{\pi m}. \tag*{(51)} D G ( p ) ≤ γ C m p f ( p ) − T ( p ) , C m := γ m + π m γ . (51)
By (48) ,
C m ≤ M 0 + 1 π − ( 3 / 2 ) < 12 7 , 12 7 − M 0 − 1 π − ( 3 / 2 ) = 138007 69930000 > 0. C_m\le M_0+\frac{1}{\pi-(3/2)}<\frac{12}{7},\qquad\frac{12}{7}-M_0-\frac{1}{\pi-(3/2)}=\frac{138007}{69930000}>0. C m ≤ M 0 + π − ( 3/2 ) 1 < 7 12 , 7 12 − M 0 − π − ( 3/2 ) 1 = 69930000 138007 > 0.
For 1 / 100 ≤ p ≤ 1 / 10 1/100\le p\le1/10 1/100 ≤ p ≤ 1/10 , use T ( p ) ≥ 399 p 2 / 800 T(p)\ge399p^2/800 T ( p ) ≥ 399 p 2 /800 and f ( p ) ≤ ( 101 / 100 ) p f(p)\le(101/100)\sqrt{p} f ( p ) ≤ ( 101/100 ) p to obtain
T ( p ) p f ( p ) ≥ 399 8080 > 12 7 γ 0 . \frac{T(p)}{pf(p)}\ge\frac{399}{8080}>\frac{12}{7}\sqrt{\gamma_0}. p f ( p ) T ( p ) ≥ 8080 399 > 7 12 γ 0 .
The last comparison follows by squaring, with exact gap
( 399 8080 ) 2 − ( 12 7 ) 2 1 1233 = 24133913 438267603200 > 0. \left(\frac{399}{8080}\right)^2-\left(\frac{12}{7}\right)^2\frac{1}{1233}=\frac{24133913}{438267603200}>0. ( 8080 399 ) 2 − ( 7 12 ) 2 1233 1 = 438267603200 24133913 > 0.
On [ 1 / 10 , 1 ] [1/10,1] [ 1/10 , 1 ] and [ 1 , 5 ] [1,5] [ 1 , 5 ] the required lower bounds are, respectively,
T ( p ) p f ( p ) ≥ 2 25 , T ( p ) p f ( p ) ≥ 2 21 . \frac{T(p)}{pf(p)} \ge\frac{2}{25}, \qquad\frac{T(p)}{pf(p)} \ge\frac{2}{21}. p f ( p ) T ( p ) ≥ 25 2 , p f ( p ) T ( p ) ≥ 21 2 .
For clarity these follow from f ( p ) 2 / p ≤ 17 / 8 f(p)^2/p \le17/8 f ( p ) 2 / p ≤ 17/8 , using 17 / 8 < 3 / 2 \sqrt{17/8}<3/2 17/8 < 3/2 , 1 / 10 > 3 / 10 \sqrt{1/10}>3/10 1/10 > 3/10 , 2 + 1 < 5 / 2 \sqrt{2}+1<5/2 2 + 1 < 5/2 , and 26 + 1 < 7 \sqrt{26}+1<7 26 + 1 < 7 in
T ( p ) p f ( p ) ≥ p 17 / 8 ( 1 + p 2 + 1 ) . \frac{T(p)}{pf(p)} \ge\frac{\sqrt{p}}{\sqrt{17/8}(\sqrt{1+p^2}+1)}. p f ( p ) T ( p ) ≥ 17/8 ( 1 + p 2 + 1 ) p .
Both constants exceed ( 12 / 7 ) γ 0 (12/7)\sqrt{\gamma_0} ( 12/7 ) γ 0 . Thus (51) is strictly negative up to p = 5 p=5 p = 5 .
For 5 ≤ p ≤ 100 5\le p\le100 5 ≤ p ≤ 100 , the estimates
f ( p ) 2 ≤ 1 + 2 arsinh p < 13 , T ( p ) p > 1 − 1 p ≥ 4 5 f(p)^2\le1+2\operatorname{arsinh}p<13,\qquad\frac{T(p)}{p}>1-\frac{1}{p}\ge\frac{4}{5} f ( p ) 2 ≤ 1 + 2 arsinh p < 13 , p T ( p ) > 1 − p 1 ≥ 5 4
give strict negativity. The sufficient squared comparison is
( 4 5 ) 2 − ( 12 7 ) 2 13 1233 = 102208 167825 > 0. \left(\frac{4}{5}\right)^2-\left(\frac{12}{7}\right)^2\frac{13}{1233}
=\frac{102208}{167825}>0. ( 5 4 ) 2 − ( 7 12 ) 2 1233 13 = 167825 102208 > 0.
Here arsinh 100 < log 300 < 6 \operatorname{arsinh}100<\log300<6 arsinh 100 < log 300 < 6 , and log 300 < 6 \log300<6 log 300 < 6 follows from e > 8 / 3 e>8/3 e > 8/3 and ( 8 / 3 ) 6 > 300 (8/3)^6>300 ( 8/3 ) 6 > 300 .
Finally, for 100 ≤ p ≤ e 180 L 100\le p\le e^{180L} 100 ≤ p ≤ e 180 L ,
f ( p ) 2 ≤ 1 + 2 arsinh p < 1 + 2 log ( 3 p ) < 360 L + 4 ≤ 364 L . f(p)^2\le1+2\operatorname{arsinh}p<1+2\log(3p)<360L+4\le364L. f ( p ) 2 ≤ 1 + 2 arsinh p < 1 + 2 log ( 3 p ) < 360 L + 4 ≤ 364 L .
Since γ L = γ 0 \gamma L=\gamma_0 γ L = γ 0 , one obtains
C m γ f ( p ) < 12 7 364 1233 < 99 100 < T ( p ) p . C_m\sqrt{\gamma}f(p)<\frac{12}{7}\sqrt{\frac{364}{1233}}<\frac{99}{100}<\frac{T(p)}{p}. C m γ f ( p ) < 7 12 1233 364 < 100 99 < p T ( p ) .
The exact middle squared gap is
( 99 100 ) 2 − ( 12 7 ) 2 364 1233 = 1079159 9590000 > 0. \left(\frac{99}{100}\right)^2-\left(\frac{12}{7}\right)^2\frac{364}{1233}
=\frac{1079159}{9590000}>0. ( 100 99 ) 2 − ( 7 12 ) 2 1233 364 = 9590000 1079159 > 0.
This proves (47) on the full asserted interval. □
The global elliptic comparison and its Newton source
Lemma 6.2 (Nonlinear comparison with a Newton source). Put
q = ( d − η ) + , c G ( p ) = 4 G ′ ( F ( p ) ) 3 λ p 2 f ( p ) , c G ( 0 ) = 0 , (52) q=(d-\eta)_+,\qquad c_G(p)=\frac{4G'(F(p))}{3\lambda}\frac{p^2}{f(p)},\qquad c_G(0)=0, \tag*{(52)} q = ( d − η ) + , c G ( p ) = 3 λ 4 G ′ ( F ( p )) f ( p ) p 2 , c G ( 0 ) = 0 , (52) S ( x ) = c G ( p ( x ) ) ( D G ( p ( x ) ) − d ) + , Ψ ( x ) = 1 4 π ∫ S ( y ) ∣ x − y ∣ d y . (53) S(x)=c_G(p(x))(D_G(p(x))-d)_+,\qquad\Psi(x)=\frac{1}{4\pi}\int\frac{S(y)}{|x-y|}\,dy. \tag*{(53)} S ( x ) = c G ( p ( x )) ( D G ( p ( x )) − d ) + , Ψ ( x ) = 4 π 1 ∫ ∣ x − y ∣ S ( y ) d y . (53) Then S S S is nonnegative, bounded, and compactly supported, with S = 0 S=0 S = 0 on { p ≤ p c u t } \{p\le p_{\mathrm{cut}}\} { p ≤ p cut } , and
G ( P ) ≤ α φ + q + Ψ . (54) G(P)\le\alpha\varphi+q+\Psi. \tag*{(54)} G ( P ) ≤ α φ + q + Ψ. (54) No upper restriction on η ≥ 0 \eta\ge0 η ≥ 0 is required.
Proof. Near zero, c G ( p ) = O ( p 2 ) c_G(p)=O(p^2) c G ( p ) = O ( p 2 ) , so its stated extension is continuous. The qualitative boundedness and decay of p p p , together with (47) , give all claims about S S S . Its Newton potential satisfies Ψ ≥ 0 \Psi\ge0 Ψ ≥ 0 , − Δ Ψ = S -\Delta\Psi=S − ΔΨ = S , and Ψ ∈ C 0 ( R 3 ) \Psi\in C_0(\mathbb{R}^3) Ψ ∈ C 0 ( R 3 ) .
The convex chain rule gives
Δ G ( P ) = G ′ ( P ) Δ P + G ′ ′ ( P ) ∣ ∇ P ∣ 2 ≥ G ′ ( P ) Δ P \Delta G(P)=G'(P)\Delta P+G''(P)|\nabla P|^2\ge G'(P)\Delta P Δ G ( P ) = G ′ ( P ) Δ P + G ′′ ( P ) ∣∇ P ∣ 2 ≥ G ′ ( P ) Δ P
in distributions. To justify it directly, on each compact set P P P has a positive lower bound, and G ′ G' G ′ is Lipschitz on the corresponding range of P P P . One can use P ∈ W l o c 2 , 2 P\in W_{\mathrm{loc}}^{2,2} P ∈ W loc 2 , 2 and the ordinary Sobolev chain rule, or approximate G G G by smooth convex functions. There is no surface measure at P = P 0 P=P_0 P = P 0 , because the two derivatives in (45) agree there. The value assigned to G ′ ′ G'' G ′′ at that level is immaterial. The possible singularity of G ′ ′ G'' G ′′ at P = 0 P=0 P = 0 does not affect a compact-set argument for a positive minimizer.
Set w = G ( P ) − α φ − q w=G(P)-\alpha\varphi-q w = G ( P ) − α φ − q . The Euler and Poisson equations yield
( − Δ + c G ( p ) ) w = c G ( p ) ( D G ( p ) − η − q ) − G ′ ′ ( P ) ∣ ∇ P ∣ 2 − 4 π α ∑ k Z k δ R k (55) \left(-\Delta+c_G(p)\right)w=c_G(p)\left(D_G(p)-\eta-q\right)-G^{\prime\prime}(P)|\nabla P|^2-4\pi\alpha\sum_k Z_k\delta_{R_k} \tag*{(55)} ( − Δ + c G ( p ) ) w = c G ( p ) ( D G ( p ) − η − q ) − G ′′ ( P ) ∣∇ P ∣ 2 − 4 π α k ∑ Z k δ R k (55)
≤ S . (56) \le S. \tag*{(56)} ≤ S . (56)
Here the exact cancellation is
c G ( p ) γ p f ( p ) π G ′ ( F ( p ) ) = 4 α 3 π p 3 = 4 π α ρ , c_G(p)\frac{\gamma p f(p)}{\pi G^{\prime}(F(p))}=\frac{4\alpha}{3\pi}p^3=4\pi\alpha\rho, c G ( p ) π G ′ ( F ( p )) γ p f ( p ) = 3 π 4 α p 3 = 4 π α ρ ,
and η + q = max { η , d } ≥ d \eta+q=\max\{\eta,d\}\ge d η + q = max { η , d } ≥ d . The term involving G ′ ′ G^{\prime\prime} G ′′ has the favorable sign. No derivative of c G c_G c G is taken: it is only a zeroth-order coefficient. Since
( − Δ + c G ( p ) ) Ψ = S + c G ( p ) Ψ ≥ S , \left(-\Delta+c_G(p)\right)\Psi=S+c_G(p)\Psi\ge S, ( − Δ + c G ( p ) ) Ψ = S + c G ( p ) Ψ ≥ S ,
we have ( − Δ + c G ( p ) ) ( w − Ψ ) ≤ 0 \left(-\Delta+c_G(p)\right)(w-\Psi)\le0 ( − Δ + c G ( p ) ) ( w − Ψ ) ≤ 0 . At a positively charged nucleus, w − Ψ → − ∞ w-\Psi\to-\infty w − Ψ → − ∞ because G ( P ) G(P) G ( P ) and the electronic Coulomb potential are bounded. At infinity, w − Ψ → − q ≤ 0 w-\Psi\to-q\le0 w − Ψ → − q ≤ 0 . The maximum principle on punctured bounded domains, followed by exhaustion, gives (54) . The coefficient c G ≥ 0 c_G\ge0 c G ≥ 0 is bounded; compact support of this coefficient is neither asserted nor required. □
Lemma 6.3 (Mass of the nonlinear defect). With N h i g h = ∫ { p > p c u t } ρ N_{\mathrm{high}}=\int_{\{p>p_{\mathrm{cut}}\}}\rho N high = ∫ { p > p cut } ρ , one has
1 4 π α ∫ S ≤ 9 N h i g h . (57) \frac{1}{4\pi\alpha}\int S\le9N_{\mathrm{high}}. \tag*{(57)} 4 π α 1 ∫ S ≤ 9 N high . (57) Proof. On the support of S S S we have p > e 180 L > 4 γ p>e^{180L}>4\gamma p > e 180 L > 4 γ , and therefore G ′ ( F ( p ) ) = m G^{\prime}(F(p))=m G ′ ( F ( p )) = m . As T , d ≥ 0 T,d\ge0 T , d ≥ 0 and G ( F ( p ) ) ≤ m F ( p ) ≤ m p f ( p ) G(F(p))\le mF(p)\le mpf(p) G ( F ( p )) ≤ m F ( p ) ≤ m p f ( p ) ,
S ≤ 4 m p 2 3 λ f ( p ) ( m + γ π m ) p f ( p ) 1 { p > p c u t } S\le\frac{4mp^2}{3\lambda f(p)}\left(m+\frac{\gamma}{\pi m}\right)pf(p)\mathbf{1}_{\{p>p_{\mathrm{cut}}\}} S ≤ 3 λ f ( p ) 4 m p 2 ( m + π m γ ) p f ( p ) 1 { p > p cut } = 4 π α ( 1 + π m 2 γ ) ρ 1 { p > p c u t } =4\pi\alpha\left(1+\frac{\pi m^2}{\gamma}\right)\rho\mathbf{1}_{\{p>p_{\mathrm{cut}}\}} = 4 π α ( 1 + γ π m 2 ) ρ 1 { p > p cut } ≤ 36 π α ρ 1 { p > p c u t } . \le36\pi\alpha\rho\mathbf{1}_{\{p>p_{\mathrm{cut}}\}}. ≤ 36 π α ρ 1 { p > p cut } . The last inequality follows from 1 + ( 22 / 7 ) ( 15001 / 10000 ) 2 < 9 1+(22/7)(15001/10000)^2<9 1 + ( 22/7 ) ( 15001/10000 ) 2 < 9 . Integration proves the claim. □
The excess-charge estimate
Boundary comparison, with the chemical potential retained
7.1. Boundary comparison, with the chemical potential retained. We use G , d , q , S , Ψ G,d,q,S,\Psi G , d , q , S , Ψ , and p c u t p_{\mathrm{cut}} p cut from § 6 . Set
x ∗ = 66 5 , β = 47 100 , r = x ∗ α λ , b ∗ = 2239 5000 , Ω r = { x : H ( x ) < 1 / r } . (58) x_*=\frac{66}{5},\qquad\beta=\frac{47}{100},\qquad r=\frac{x_*}{\alpha\sqrt{\lambda}},\qquad b_*=\frac{2239}{5000},\qquad\Omega_r=\{x:H(x)<1/r\}. \tag*{(58)} x ∗ = 5 66 , β = 100 47 , r = α λ x ∗ , b ∗ = 5000 2239 , Ω r = { x : H ( x ) < 1/ r } . (58)
Proposition 7.1 (Exterior boundary estimate). For 0 < γ ≤ γ 0 0<\gamma\le\gamma_0 0 < γ ≤ γ 0 , throughout Ω ‾ r \overline{\Omega}_r Ω r one has
α B η ≤ b ∗ γ 2 + Ψ , α B η = ( γ / π ) g ( p ) 2 + t 2 − η . (59) \alpha B\eta\le b_*\gamma^2+\Psi,\qquad\alpha B\eta=\sqrt{(\gamma/\pi)g(p)^2+t^2}-\eta. \tag*{(59)} α B η ≤ b ∗ γ 2 + Ψ , α B η = ( γ / π ) g ( p ) 2 + t 2 − η . (59)
The estimate holds for every η ≥ 0 \eta\ge0 η ≥ 0 .
Proof. Write d = δ γ 2 d=\delta\gamma^2 d = δ γ 2 , where δ = 361 / 1000 \delta=361/1000 δ = 361/1000 . Certified exterior density and potential bounds. Use Propositions 4.4 and 4.5 with κ = 103 / 100 \kappa=103/100 κ = 103/100 and M S = 692881 / 57600 M_S=692881/57600 M S = 692881/57600 . Put
V ‾ = 76607 250000 , z ∗ = 79451 100000 , c 3 = 2 3 ( 1 + 1 10 6 ) . \overline{V}=\frac{76607}{250000},\qquad z_*=\frac{79451}{100000},\qquad c_3=\frac{2}{3}\left(1+\frac{1}{10^6}\right). V = 250000 76607 , z ∗ = 100000 79451 , c 3 = 3 2 ( 1 + 1 0 6 1 ) .
The rational choices π − = 333 / 106 \pi_-=333/106 π − = 333/106 , π + = 355 / 113 \pi_+=355/113 π + = 355/113 , and σ ‾ = 55703 / 100000 \overline{\sigma}=55703/100000 σ = 55703/100000 satisfy β 31 < σ ‾ 40 \beta^{31}<\overline{\sigma}^{40} β 31 < σ 40 and
π + 2 [ 9 M S 2 128 x ∗ 4 ( 1 − σ ‾ ) 4 + 43 50 β 2 x ∗ 2 ] < V ‾ , \pi_+^2\left[\frac{9M_S^2}{128x_*^4(1-\overline{\sigma})^4}+\frac{43}{50\beta^2x_*^2}\right]<\overline{V}, π + 2 [ 128 x ∗ 4 ( 1 − σ ) 4 9 M S 2 + 50 β 2 x ∗ 2 43 ] < V ,
κ V ‾ ( 2 + κ γ 0 2 V ‾ ) < z ∗ 2 . \kappa\overline{V}\left(2+\kappa\gamma_0^2\overline{V}\right)<z_*^2. κ V ( 2 + κ γ 0 2 V ) < z ∗ 2 .
For x ∈ Ω r x \in\Omega_r x ∈ Ω r , one has H ( x ) ≤ 1 / r H(x) \le1/r H ( x ) ≤ 1/ r and hence
t ( x ) ≤ U β ( H ( x ) ) ≤ U β ( 1 / r ) < γ 2 V ‾ , T ( p ( x ) ) ≤ κ U β ( H ( x ) ) < κ γ 2 V ‾ . t(x) \le U_\beta(H(x)) \le U_\beta(1/r) < \gamma^2\overline{V}, \qquad T(p(x)) \le\kappa U_\beta(H(x)) < \kappa\gamma^2\overline{V}. t ( x ) ≤ U β ( H ( x )) ≤ U β ( 1/ r ) < γ 2 V , T ( p ( x )) ≤ κ U β ( H ( x )) < κ γ 2 V .
Using p 2 = T ( p ) ( T ( p ) + 2 ) p^2=T(p)(T(p)+2) p 2 = T ( p ) ( T ( p ) + 2 ) and γ ≤ γ 0 \gamma\le\gamma_0 γ ≤ γ 0 , we obtain
p ( x ) 2 ≤ γ 2 κ V ‾ ( 2 + κ γ 0 2 V ‾ ) < γ 2 z ∗ 2 . p(x)^2 \le\gamma^2\kappa\overline{V}(2+\kappa\gamma_0^2\overline{V}) < \gamma^2 z_*^2. p ( x ) 2 ≤ γ 2 κ V ( 2 + κ γ 0 2 V ) < γ 2 z ∗ 2 .
Thus, throughout the closed exterior,
t ≤ γ 2 V ‾ , p < γ z ∗ < γ < 4 γ . (60) t \le\gamma^2\overline{V}, \qquad p < \gamma z_* < \gamma< 4\gamma. \tag*{(60)} t ≤ γ 2 V , p < γ z ∗ < γ < 4 γ . (60)
In particular, G G G is evaluated on its small-density branch there. The usual estimate f ( p ) 2 ≤ p + 2 p 3 f(p)^2 \le p+2p^3 f ( p ) 2 ≤ p + 2 p 3 gives
g ( p ) 2 ≤ 2 3 p 3 ( 1 + 6 5 p 2 ) ≤ c 3 p 3 , g(p)^2 \le\frac{2}{3}p^3\left(1+\frac{6}{5}p^2\right) \le c_3p^3, g ( p ) 2 ≤ 3 2 p 3 ( 1 + 5 6 p 2 ) ≤ c 3 p 3 ,
because ( 6 / 5 ) γ 0 2 < 10 − 6 (6/5)\gamma_0^2<10^{-6} ( 6/5 ) γ 0 2 < 1 0 − 6 . The lower bound in (49) also gives G ( P ) ≥ 3 p 2 / 8 G(P)\ge3p^2/8 G ( P ) ≥ 3 p 2 /8 on this exterior.
The positive-t t t part. For t ≥ 0 t\ge0 t ≥ 0 , the bounds above give
( γ / π ) g ( p ) 2 + t 2 < γ 4 [ ( c 3 / π − ) z ∗ 3 + V ‾ 2 ] < b ∗ 2 γ 4 . (\gamma/\pi)g(p)^2+t^2<\gamma^4\left[\left(c_3/\pi_-\right)z_*^3+\overline{V}^2\right]<b_*^2\gamma^4. ( γ / π ) g ( p ) 2 + t 2 < γ 4 [ ( c 3 / π − ) z ∗ 3 + V 2 ] < b ∗ 2 γ 4 .
The last inequality is a strict rational inequality; the exact expressions for this and the following numerical margins are recorded in Appendix B. Hence α B η < b ∗ γ 2 − η \alpha B\eta<b_*\gamma^2-\eta α B η < b ∗ γ 2 − η , which implies (59) .
The negative-t t t part, uniformly in the multiplier. At a fixed point let
a = η + q + Ψ = max { η , d } + Ψ ≥ d . a=\eta+q+\Psi=\max\{\eta,d\}+\Psi\ge d. a = η + q + Ψ = max { η , d } + Ψ ≥ d .
The comparison (54) and the lower bound for G G G yield
0 ≤ 3 8 p 2 ≤ t + a . 0\le\frac{3}{8}p^2\le t+a. 0 ≤ 8 3 p 2 ≤ t + a .
If t < 0 t<0 t < 0 , this implies − a ≤ t < 0 -a\le t<0 − a ≤ t < 0 and
( γ / π ) g ( p ) 2 + t 2 ≤ γ c 3 π [ 8 3 ( t + a ) ] 3 / 2 + t 2 . (\gamma/\pi)g(p)^2+t^2\le\frac{\gamma c_3}{\pi}\left[\frac{8}{3}(t+a)\right]^{3/2}+t^2. ( γ / π ) g ( p ) 2 + t 2 ≤ π γ c 3 [ 3 8 ( t + a ) ] 3/2 + t 2 .
The right-hand side is a convex function of t t t on [ − a , 0 ] [-a,0] [ − a , 0 ] . Its maximum is attained at one of the endpoints. Therefore
( γ / π ) g ( p ) 2 + t 2 ≤ M ( a ) , M ( a ) = max { a , C γ a 3 / 4 } , \sqrt{(\gamma/\pi)g(p)^2+t^2}\le M(a), \qquad M(a)=\max\{a,C\sqrt{\gamma}a^{3/4}\}, ( γ / π ) g ( p ) 2 + t 2 ≤ M ( a ) , M ( a ) = max { a , C γ a 3/4 } ,
where C = ( c 3 / π − ) 1 / 2 ( 8 / 3 ) 3 / 4 C=(c_3/\pi_-)^{1/2}(8/3)^{3/4} C = ( c 3 / π − ) 1/2 ( 8/3 ) 3/4 .
The following two rational checks are sufficient:
b ∗ > δ , b ∗ 4 > ( c 3 / π − ) 2 ( 8 / 3 ) 3 δ 3 , b_*>\delta, \qquad b_*^4>(c_3/\pi_-)^2(8/3)^3\delta^3, b ∗ > δ , b ∗ 4 > ( c 3 / π − ) 2 ( 8/3 ) 3 δ 3 ,
1 − 3 b ∗ 4 δ = 503 7220 > 0. 1-\frac{3b_*}{4\delta}=\frac{503}{7220}>0. 1 − 4 δ 3 b ∗ = 7220 503 > 0.
The first line gives M ( d ) < b ∗ γ 2 M(d)<b_*\gamma^2 M ( d ) < b ∗ γ 2 . For a ≥ d a\ge d a ≥ d , the derivative of the second branch is bounded by
3 4 C γ d − 1 / 4 < 3 b ∗ 4 δ < 1. \frac{3}{4}C\sqrt{\gamma}\,d^{-1/4}<\frac{3b_*}{4\delta}<1. 4 3 C γ d − 1/4 < 4 δ 3 b ∗ < 1.
The first branch has derivative one. Both branches are therefore increasing and 1-Lipschitz on [ d , ∞ ) [d,\infty) [ d , ∞ ) , and so is their maximum. It follows pointwise, without differentiating Ψ \Psi Ψ , that
α B η ≤ M ( a ) − η ≤ M ( d ) + a − d − η < b ∗ γ 2 + Ψ − min { η , d } ≤ b ∗ γ 2 + Ψ . \begin{aligned}
\alpha B\eta\le M(a)-\eta\le M(d)+a-d-\eta\\
&<b_*\gamma^2+\Psi-\min\{\eta,d\}\le b_*\gamma^2+\Psi.
\end{aligned} α B η ≤ M ( a ) − η ≤ M ( d ) + a − d − η < b ∗ γ 2 + Ψ − min { η , d } ≤ b ∗ γ 2 + Ψ.
This proves (59) for every η ≥ 0 \eta\ge0 η ≥ 0 . □
Harmonic comparison and extraction of the charge
Proposition 7.2 (Charge estimate before absorption). For 0 < γ ≤ γ 0 0 < \gamma\le\gamma_0 0 < γ ≤ γ 0 one has
Q ≤ 147774 25000 λ 3 / 2 K + 1 4 π α ∫ S ≤ 147774 25000 λ 3 / 2 K + 9 N h i g h , (61) Q \le\frac{147774}{25000}\lambda^{3/2}K+\frac{1}{4\pi\alpha}\int S \le\frac{147774}{25000}\lambda^{3/2}K+9N_{\mathrm{high}}, \tag*{(61)} Q ≤ 25000 147774 λ 3/2 K + 4 π α 1 ∫ S ≤ 25000 147774 λ 3/2 K + 9 N high , (61) where N h i g h = ∫ { p > p c u t } ρ N_{\mathrm{high}}=\int_{\{p>p_{\mathrm{cut}}\}}\rho N high = ∫ { p > p cut } ρ .
Proof. The density cap (60) gives p < γ < 1 / 2 p < \gamma< 1/2 p < γ < 1/2 on Ω ‾ r \overline{\Omega}_r Ω r . Lemma 4.6 shows that B η ≥ − φ ‾ B_\eta\ge-\overline{\varphi} B η ≥ − φ , B η → 0 B_\eta\to0 B η → 0 at infinity, and Δ B η ≥ 0 \Delta B_\eta\ge0 Δ B η ≥ 0 in this exterior. Since − Δ Ψ = S ≥ 0 -\Delta\Psi=S\ge0 − ΔΨ = S ≥ 0 , the function B η − Ψ / α B_\eta-\Psi/\alpha B η − Ψ/ α is subharmonic there. By Proposition 7.1, its boundary values are at most b ∗ γ 2 / α b_*\gamma^2/\alpha b ∗ γ 2 / α . The harmonic comparison therefore gives
− φ ≤ B η ≤ b ∗ γ 2 α r H + Ψ α on Ω r . (62) -\varphi\le B_\eta\le\frac{b_*\gamma^2}{\alpha}rH+\frac{\Psi}{\alpha}\qquad\text{on }\Omega_r. \tag*{(62)} − φ ≤ B η ≤ α b ∗ γ 2 r H + α Ψ on Ω r . (62)
For a direct exhaustion argument, first take a regular level h < 1 / r h<1/r h < 1/ r of H H H . The maximum principle on { H < h } \{H<h\} { H < h } , intersected with a large ball, compares B η − Ψ / α B_\eta-\Psi/\alpha B η − Ψ/ α with ( b ∗ γ 2 / α ) H / h (b_*\gamma^2/\alpha)H/h ( b ∗ γ 2 / α ) H / h . All functions vanish at infinity, so the outer boundary error tends to zero. Let regular values h ↑ 1 / r h\uparrow1/r h ↑ 1/ r to obtain (62) . This also handles a nonregular boundary of Ω r \Omega_r Ω r .
Let ⟨ u ⟩ R = ( 4 π ) − 1 ∫ S 2 u ( R ω ) d ω \langle u\rangle_R=(4\pi)^{-1}\int_{\mathbb{S}^2}u(R\omega)\,d\omega ⟨ u ⟩ R = ( 4 π ) − 1 ∫ S 2 u ( R ω ) d ω . For R R R enclosing every nucleus, Newton’s theorem gives
R ⟨ H ⟩ R = K , (63) R\langle H\rangle_R=K, \tag*{(63)} R ⟨ H ⟩ R = K , (63)
− R ⟨ φ ⟩ R = ∫ min { 1 , R / ∣ y ∣ } ρ ( y ) d y − Z t o t ⟶ Q , (64) -R\langle\varphi\rangle_R=\int\min\{1,R/|y|\}\rho(y)\,dy-Z_{\mathrm{tot}}\longrightarrow Q, \tag*{(64)} − R ⟨ φ ⟩ R = ∫ min { 1 , R /∣ y ∣ } ρ ( y ) d y − Z tot ⟶ Q , (64)
R ⟨ Ψ ⟩ R = 1 4 π ∫ min { 1 , R / ∣ y ∣ } S ( y ) d y ⟶ 1 4 π ∫ S . (65) R\langle\Psi\rangle_R=\frac{1}{4\pi}\int\min\{1,R/|y|\}S(y)\,dy\longrightarrow\frac{1}{4\pi}\int S. \tag*{(65)} R ⟨ Ψ ⟩ R = 4 π 1 ∫ min { 1 , R /∣ y ∣ } S ( y ) d y ⟶ 4 π 1 ∫ S . (65)
The limits use only the integrability of ρ \rho ρ and S S S . For sufficiently large R R R , the whole sphere lies in Ω r \Omega_r Ω r . Average (62) , multiply by R R R , and let R → ∞ R\to\infty R → ∞ . Since b ∗ γ 2 r / α = x ∗ b ∗ λ 3 / 2 b_*\gamma^2r/\alpha=x_*b_*\lambda^{3/2} b ∗ γ 2 r / α = x ∗ b ∗ λ 3/2 and x ∗ b ∗ = 147774 / 25000 x_*b_*=147774/25000 x ∗ b ∗ = 147774/25000 , the first inequality in (61) follows. The second is Lemma 6.3. □
Absorption and proof of Theorem 2.1
Proof of Theorem 2.1. If p = 0 p=0 p = 0 , then Q = − Z t o t < 0 Q=-Z_{\mathrm{tot}}<0 Q = − Z tot < 0 and the assertion is immediate. For a nonzero minimizer, Lemma 3.5 reduces the proof to (15) . We first establish the result in this domain. Since E ≤ 0 \mathcal{E}\le0 E ≤ 0 , Proposition 5.1 gives
N h i g h ≤ 16 3 p c u t ( N + α C λ Z t o t ) , C λ = 8 π sinh ( 16 3 λ ) . N_{\mathrm{high}}\le\frac{16}{3p_{\mathrm{cut}}}(N+\alpha C_\lambda Z_{\mathrm{tot}}),\qquad C_\lambda=\frac{8}{\pi}\sinh\left(\frac{16}{3\lambda}\right). N high ≤ 3 p cut 16 ( N + α C λ Z tot ) , C λ = π 8 sinh ( 3 λ 16 ) .
With ε = 48 e − 180 L \varepsilon=48e^{-180L} ε = 48 e − 180 L , N = Z t o t + Q N=Z_{\mathrm{tot}}+Q N = Z tot + Q , and Z t o t ≤ K / α Z_{\mathrm{tot}}\le K/\alpha Z tot ≤ K / α , (61) therefore yields
( 1 − ε ) Q ≤ [ 147774 25000 λ 3 / 2 + ε ( 1 α + C λ ) ] K . (66) (1-\varepsilon)Q\le\left[\frac{147774}{25000}\lambda^{3/2}+\varepsilon\left(\frac{1}{\alpha}+C_\lambda\right)\right]K. \tag*{(66)} ( 1 − ε ) Q ≤ [ 25000 147774 λ 3/2 + ε ( α 1 + C λ ) ] K . (66)
In particular, ε < 1 \varepsilon<1 ε < 1 . Put B o u t = 5913 / 1000 B_{\mathrm{out}}=5913/1000 B out = 5913/1000 . The parameter hypotheses imply
λ − 1 ≤ 9 L , C λ < 2 e 48 L , 1 α λ 3 / 2 ≤ 3699 L 3 / 2 , C λ λ 3 / 2 < 54 L 3 / 2 e 48 L . \lambda^{-1}\le9L,\qquad C_\lambda<2e^{48L},\qquad\frac{1}{\alpha\lambda^{3/2}}\le3699L^{3/2},\qquad\frac{C_\lambda}{\lambda^{3/2}}<54L^{3/2}e^{48L}. λ − 1 ≤ 9 L , C λ < 2 e 48 L , α λ 3/2 1 ≤ 3699 L 3/2 , λ 3/2 C λ < 54 L 3/2 e 48 L .
Since L 3 / 2 e − a L L^{3/2}e^{-aL} L 3/2 e − a L is decreasing for L ≥ 1 L\ge1 L ≥ 1 when a ≥ 3 / 2 a\ge3/2 a ≥ 3/2 ,
ε [ B o u t + 1 α λ 3 / 2 + C λ λ 3 / 2 ] < 48 ( B o u t + 3699 ) e − 180 + 2592 e − 132 < 2 − 162 + 2 − 120 < 2 − 119 < 2 − 9 < 51 25000 = B o u t − 147774 25000 . \begin{aligned}
\varepsilon\left[B_{\mathrm{out}}+\frac{1}{\alpha\lambda^{3/2}}+\frac{C_\lambda}{\lambda^{3/2}}\right]
<48(B_{\mathrm{out}}+3699)e^{-180}+2592e^{-132} \\
&<2^{-162}+2^{-120}<2^{-119}<2^{-9} \\
&<\frac{51}{25000}=B_{\mathrm{out}}-\frac{147774}{25000}.
\end{aligned} ε [ B out + α λ 3/2 1 + λ 3/2 C λ ] < 48 ( B out + 3699 ) e − 180 + 2592 e − 132 < 2 − 162 + 2 − 120 < 2 − 119 < 2 − 9 < 25000 51 = B out − 25000 147774 .
The powers of two use e > 2 e>2 e > 2 , 48 ( B o u t + 3699 ) < 2 18 48(B_{\mathrm{out}}+3699)<2^{18} 48 ( B out + 3699 ) < 2 18 , and 2592 < 2 12 2592<2^{12} 2592 < 2 12 . Combining this strict margin with (66) proves
Q < 5913 1000 λ 3 / 2 K . (67) Q<\frac{5913}{1000}\lambda^{3/2}K. \tag*{(67)} Q < 1000 5913 λ 3/2 K . (67)
At λ = 1 / 9 \lambda= 1/9 λ = 1/9 , this is exactly
Q < 219 1000 K . (68) Q < \frac{219}{1000}K. \tag*{(68)} Q < 1000 219 K . (68)
For the original invariant parameter domain, apply the change of variables in Lemma 3.5 . Electronic minimality, finiteness of the mass, and the energy sign are preserved; multiplication of the transformed bound by s 3 s^3 s 3 proves (6) .
Remark 7.3 (A positive-energy correction). The same proof under (15) gives, without the sign restriction on the total energy,
Q < 5913 1000 λ 3 / 2 K + ε 1 − ε E + , ε = 48 exp ( − 180 1233 α λ ) . Q < \frac{5913}{1000}\lambda^{3/2}K+\frac{\varepsilon}{1-\varepsilon}E_+,\qquad\varepsilon=48\exp\left(-\frac{180}{1233\alpha\lambda}\right). Q < 1000 5913 λ 3/2 K + 1 − ε ε E + , ε = 48 exp ( − 1233 α λ 180 ) .
Indeed, (36) adds ε E + \varepsilon E_+ ε E + to (66) . By covariance, the same formula holds under (4) : E + E_+ E + and Q Q Q acquire the same scaling factor, while ε \varepsilon ε is invariant. This correction can be used for electronically minimizing configurations whose positive total energies tend to zero. No claim about existence or convergence of such a sequence is needed for the estimate.
Nuclear configurations and the factor K K K
A mass-preserving dilation
The following variation keeps the nuclear charges and the parameters α , λ \alpha,\lambda α , λ fixed. It is therefore distinct from a covariance identity that changes the parameters of the functional. Let p p p be an admissible finite-mass density variable and set, for s > 0 s>0 s > 0 ,
p s ( x ) = s p ( s x ) , R k , s = s − 1 R k . (69) p_s(x)=s\,p(sx),\qquad R_{k,s}=s^{-1}R_k. \tag*{(69)} p s ( x ) = s p ( s x ) , R k , s = s − 1 R k . (69)
Then ρ s ( x ) = s 3 ρ ( s x ) \rho_s(x)=s^3\rho(sx) ρ s ( x ) = s 3 ρ ( s x ) , and consequently ∫ ρ s = ∫ ρ \int\rho_s=\int\rho ∫ ρ s = ∫ ρ . In particular, the Lagrange multiplier for the electron number contributes nothing to this variation.
Write P = F ( p ) P=F(p) P = F ( p ) and
ϑ ( p ) = p f ′ ( p ) f ( p ) , c λ = 3 λ 8 π 2 , c 0 = 1 8 π 2 , \vartheta(p)=\frac{pf'(p)}{f(p)},\qquad c_\lambda=\frac{3\lambda}{8\pi^2},\qquad c_0=\frac{1}{8\pi^2}, ϑ ( p ) = f ( p ) p f ′ ( p ) , c λ = 8 π 2 3 λ , c 0 = 8 π 2 1 ,
where ϑ ( 0 ) \vartheta(0) ϑ ( 0 ) is defined by its continuous limit. Denote the full Coulomb part of the energy by
C = − α ∫ V ρ + α D [ ρ ] + α ∑ k < l Z k Z l ∣ R k − R l ∣ . \mathcal{C}=-\alpha\int V\rho+\alpha D[\rho]+\alpha\sum_{k<l}\frac{Z_kZ_l}{|R_k-R_l|}. C = − α ∫ V ρ + α D [ ρ ] + α k < l ∑ ∣ R k − R l ∣ Z k Z l .
Every term in C \mathcal{C} C , including nuclear repulsion, is multiplied by s s s under (69) . The exact energy along this path is
E [ p s ; R s , Z ] = c λ s ∫ f ( s p ) 2 f ( p ) 2 ∣ ∇ P ∣ 2 + c 0 s − 3 ∫ τ ( s p ) + s C . (70) \mathcal{E}[p_s;R_s,Z]=c_\lambda s\int\frac{f(sp)^2}{f(p)^2}|\nabla P|^2+c_0s^{-3}\int\tau(sp)+s\mathcal{C}. \tag*{(70)} E [ p s ; R s , Z ] = c λ s ∫ f ( p ) 2 f ( s p ) 2 ∣∇ P ∣ 2 + c 0 s − 3 ∫ τ ( s p ) + s C . (70)
At p = 0 p=0 p = 0 the ratio in the first integral is defined by continuity and equals s s s . This form of the identity requires no unweighted H 1 H^1 H 1 regularity of p p p .
Here are sufficient bounds to justify differentiating (70) . The previously established scalar estimate 0 < ϑ ( p ) ≤ 7 / 10 0<\vartheta(p)\le7/10 0 < ϑ ( p ) ≤ 7/10 gives, uniformly for s s s in a compact subinterval of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) , a bound on f ( s p ) 2 / f ( p ) 2 f(sp)^2/f(p)^2 f ( s p ) 2 / f ( p ) 2 and on the derivative of s f ( s p ) 2 / f ( p ) 2 sf(sp)^2/f(p)^2 s f ( s p ) 2 / f ( p ) 2 . Thus the gradient integrands are dominated by a constant times ∣ ∇ P ∣ 2 |\nabla P|^2 ∣∇ P ∣ 2 . The Sobolev chain rule applies because u ↦ F ( s F − 1 ( u ) ) u\mapsto F(sF^{-1}(u)) u ↦ F ( s F − 1 ( u )) has bounded derivative for each such range of s s s .
For the local kinetic term one has
4 τ ( p ) ≤ p τ ′ ( p ) ≤ 5 τ ( p ) . (71) 4\tau(p)\le p\tau'(p)\le5\tau(p). \tag*{(71)} 4 τ ( p ) ≤ p τ ′ ( p ) ≤ 5 τ ( p ) . (71)
Indeed, for p > 0 p>0 p > 0 ,
p T ′ ( p ) T ( p ) = 1 + 1 1 + p 2 ∈ [ 1 , 2 ] . \frac{pT'(p)}{T(p)}=1+\frac{1}{\sqrt{1+p^2}}\in[1,2]. T ( p ) p T ′ ( p ) = 1 + 1 + p 2 1 ∈ [ 1 , 2 ] .
It follows, for 0 ≤ u ≤ 1 0 \le u \le1 0 ≤ u ≤ 1 , that u 2 T ( p ) ≤ T ( u p ) ≤ u T ( p ) u^2T(p) \le T(up) \le uT(p) u 2 T ( p ) ≤ T ( u p ) ≤ u T ( p ) . Substitution in
τ ( p ) = 8 p 3 ∫ 0 1 u 2 T ( u p ) d u \tau(p)=8p^3\int_0^1u^2T(up)\,du τ ( p ) = 8 p 3 ∫ 0 1 u 2 T ( u p ) d u
gives p τ ′ ( p ) / 5 ≤ τ ( p ) ≤ p τ ′ ( p ) / 4 p\tau'(p)/5 \le\tau(p) \le p\tau'(p)/4 p τ ′ ( p ) /5 ≤ τ ( p ) ≤ p τ ′ ( p ) /4 , proving (71) . This estimate also bounds τ ( s p ) \tau(sp) τ ( s p ) and the derivative of s − 3 τ ( s p ) s^{-3}\tau(sp) s − 3 τ ( s p ) by a constant times τ ( p ) \tau(p) τ ( p ) when s s s stays in a compact subinterval of ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) . Dominated convergence therefore justifies the differentiation.
Proposition 8.1 (Virial identity and the energy sign). If an admissible nonzero pair ( p , R ) (p,\mathbf{R}) ( p , R ) is stationary along the dilation (69) , then
E [ p ; R , Z ] = − 3 λ 4 π 2 ∫ ϑ ( p ) ∣ ∇ P ∣ 2 − 1 8 π 2 ∫ ( p τ ′ ( p ) − 4 τ ( p ) ) < 0. (72) \mathcal{E}[p;\mathbf{R},\mathbf{Z}]=-\frac{3\lambda}{4\pi^2}\int\vartheta(p)|\nabla P|^2-\frac{1}{8\pi^2}\int\left(p\tau'(p)-4\tau(p)\right)<0. \tag*{(72)} E [ p ; R , Z ] = − 4 π 2 3 λ ∫ ϑ ( p ) ∣∇ P ∣ 2 − 8 π 2 1 ∫ ( p τ ′ ( p ) − 4 τ ( p ) ) < 0. (72)
Proof. The preceding bounds justify differentiation at s = 1 s=1 s = 1 and give
0 = c λ ∫ ( 1 + 2 ϑ ( p ) ) ∣ ∇ P ∣ 2 + c 0 ∫ ( p τ ′ ( p ) − 3 τ ( p ) ) + C . (73) 0=c_\lambda\int(1+2\vartheta(p))|\nabla P|^2+c_0\int\left(p\tau'(p)-3\tau(p)\right)+C. \tag*{(73)} 0 = c λ ∫ ( 1 + 2 ϑ ( p )) ∣∇ P ∣ 2 + c 0 ∫ ( p τ ′ ( p ) − 3 τ ( p ) ) + C . (73)
Subtracting this identity from the energy proves the equality in (72) . Both integrands are nonnegative. In fact, for h ( p ) = p τ ′ ( p ) − 4 τ ( p ) h(p)=p\tau'(p)-4\tau(p) h ( p ) = p τ ′ ( p ) − 4 τ ( p ) one has h ( 0 ) = 0 h(0)=0 h ( 0 ) = 0 and
h ′ ( p ) = 8 p 2 ( 1 − 1 1 + p 2 ) > 0 ( p > 0 ) . h'(p)=8p^2\left(1-\frac{1}{\sqrt{1+p^2}}\right)>0\qquad(p>0). h ′ ( p ) = 8 p 2 ( 1 − 1 + p 2 1 ) > 0 ( p > 0 ) .
Thus the local integral is strictly positive for a nonzero density. □
Local minima of the Born–Oppenheimer energy
Fix one of the mass constraints admitted in § 2 and define the electronic ground-state energy, including nuclear repulsion, by
E B O ( R ) = inf p E [ p ; R , Z ] . E_{\mathrm{BO}}(\mathbf{R})=\inf_p\mathcal{E}[p;\mathbf{R},\mathbf{Z}]. E BO ( R ) = p inf E [ p ; R , Z ] .
Corollary 8.2 (Locally minimizing nuclear configurations). Suppose that R \mathbf{R} R is a local minimum of E B O E_{\mathrm{BO}} E BO , that the electronic infimum at R \mathbf{R} R is attained by a nonzero finite-mass density p p p , and that (4) holds. Then
E [ p ; R , Z ] < 0 , N − Z t o t < 5913 1000 λ 3 / 2 K . \mathcal{E}[p;\mathbf{R},\mathbf{Z}]<0,\qquad N-Z_{\mathrm{tot}}<\frac{5913}{1000}\lambda^{3/2}K. E [ p ; R , Z ] < 0 , N − Z tot < 1000 5913 λ 3/2 K . In particular, the conclusion holds for any attained joint ground state with nonzero electronic density.
Proof. The dilation preserves the electron number, and hence the chosen mass constraint. For s s s near one,
E [ p s ; R / s , Z ] ≥ E B O ( R / s ) ≥ E B O ( R ) = E [ p ; R , Z ] . \mathcal{E}[p_s;\mathbf{R}/s,\mathbf{Z}]\ge E_{\mathrm{BO}}(\mathbf{R}/s)\ge E_{\mathrm{BO}}(\mathbf{R})=\mathcal{E}[p;\mathbf{R},\mathbf{Z}]. E [ p s ; R / s , Z ] ≥ E BO ( R / s ) ≥ E BO ( R ) = E [ p ; R , Z ] .
Its derivative at one vanishes. Proposition 8.1 and Theorem 2.1 give the assertions. □
This argument requires neither differentiability of E B O E_{\mathrm{BO}} E BO nor uniqueness of its electronic minimizer. The electronic minimizing hypothesis remains essential to the charge theorem. The corollary is conditional on attainment and does not require an existence theorem for nuclear minimizing configurations.
If a dissociation limit consists of attained minimizing fragments, the theorem can be applied separately to each fragment satisfying its energy hypothesis. Summing gives the same coefficient times the sum of their numbers of nuclei. A fragment cannot be omitted merely because its total energy is positive: the present theorem then supplies only the correction in Remark 7.3. This observation makes no assertion about the existence of a particular fragmentation.
The harmonic geometry of the factor K K K
The harmonic comparison uses the function H H H , harmonic away from the nuclei, and the associated level sets
H ( x ) = ∑ k = 1 K 1 ∣ x − R k ∣ , E r = { x : r H ( x ) ≥ 1 } , Ω r = R 3 ∖ E r . H(x)=\sum_{k=1}^{K}\frac{1}{\lvert x-R_k\rvert},\qquad E_r=\{x:rH(x)\ge1\},\qquad\Omega_r=\mathbb{R}^3\setminus E_r. H ( x ) = k = 1 ∑ K ∣ x − R k ∣ 1 , E r = { x : r H ( x ) ≥ 1 } , Ω r = R 3 ∖ E r .
In the Newtonian capacity convention cap ( B ( 0 , a ) ) = a \operatorname{cap}(B(0,a))=a cap ( B ( 0 , a )) = a , the function min { 1 , r H } \min\{1,rH\} min { 1 , r H } is the capacitary potential of E r E_r E r : it belongs to D 1 , 2 ( R 3 ) D^{1,2}(\mathbb{R}^3) D 1 , 2 ( R 3 ) , is superharmonic, equals one on E r E_r E r , is harmonic on its complement, and tends to zero at infinity. Since
r H ( x ) = K r ∣ x ∣ + O ( ∣ x ∣ − 2 ) ( ∣ x ∣ → ∞ ) , rH(x)=\frac{Kr}{\lvert x\rvert}+O(\lvert x\rvert^{-2})\qquad(\lvert x\rvert\to\infty), r H ( x ) = ∣ x ∣ K r + O (∣ x ∣ − 2 ) (∣ x ∣ → ∞ ) ,
one has exactly
cap ( E r ) = K r . (74) \operatorname{cap}(E_r)=Kr. \tag*{(74)} cap ( E r ) = K r . (74)
This remains true when components of E r E_r E r merge. A homothety rescales capacity and r r r by the same factor and hence does not improve K K K in this comparison. A smaller geometric coefficient would require a different comparison set together with adequate estimates of the density and potential on its boundary; it does not follow from scaling the present equipotential construction.
The localization coefficient
We prove the scalar estimate (20) used in Lemma 4.2. The argument is algebraic and holds for every p > 0 p>0 p > 0 .
Put u l o c = arsinh p u_{\mathrm{loc}}=\operatorname{arsinh}p u loc = arsinh p , c = cosh u l o c c=\cosh u_{\mathrm{loc}} c = cosh u loc , s = sinh u l o c s=\sinh u_{\mathrm{loc}} s = sinh u loc , and
U 0 = ( c + 2 u l o c s ) 2 , D 0 = c ( 3 c 2 − s 2 ) + 4 u l o c s ( c 2 + s 2 ) > 0. U_0=(c+2u_{\mathrm{loc}}s)^2,\qquad D_0=c(3c^2-s^2)+4u_{\mathrm{loc}}s(c^2+s^2)>0. U 0 = ( c + 2 u loc s ) 2 , D 0 = c ( 3 c 2 − s 2 ) + 4 u loc s ( c 2 + s 2 ) > 0.
Direct differentiation gives
3 4 f ( p ) 2 p ( 2 − p f ′ / f ) = 3 2 U 0 D 0 . \frac{3}{4}\frac{f(p)^2}{p(2-pf'/f)}=\frac{3}{2}\frac{U_0}{D_0}. 4 3 p ( 2 − p f ′ / f ) f ( p ) 2 = 2 3 D 0 U 0 .
It therefore suffices to prove J ( u l o c ) : = 43 D 0 − 75 U 0 > 0 J(u_{\mathrm{loc}}):=43D_0-75U_0>0 J ( u loc ) := 43 D 0 − 75 U 0 > 0 . Hyperbolic identities yield
J ( u l o c ) = 43 2 cosh ( 3 u l o c ) + 215 2 cosh u l o c − 75 2 − 75 2 cosh ( 2 u l o c ) + 86 u l o c ( sinh ( 3 u l o c ) − sinh u l o c ) − 150 u l o c sinh ( 2 u l o c ) − 150 u l o c 2 cosh ( 2 u l o c ) + 150 u l o c . \begin{aligned}
J(u_{\mathrm{loc}})={}&\frac{43}{2}\cosh(3u_{\mathrm{loc}})+\frac{215}{2}\cosh u_{\mathrm{loc}}-\frac{75}{2}-\frac{75}{2}\cosh(2u_{\mathrm{loc}})\\
&+86u_{\mathrm{loc}}\bigl(\sinh(3u_{\mathrm{loc}})-\sinh u_{\mathrm{loc}}\bigr)-150u_{\mathrm{loc}}\sinh(2u_{\mathrm{loc}})\\
&-150u_{\mathrm{loc}}^2\cosh(2u_{\mathrm{loc}})+150u_{\mathrm{loc}}.
\end{aligned} J ( u loc ) = 2 43 cosh ( 3 u loc ) + 2 215 cosh u loc − 2 75 − 2 75 cosh ( 2 u loc ) + 86 u loc ( sinh ( 3 u loc ) − sinh u loc ) − 150 u loc sinh ( 2 u loc ) − 150 u loc 2 cosh ( 2 u loc ) + 150 u loc .
For n ≥ 2 n\ge2 n ≥ 2 , its coefficients a n = ( 2 n ) ! [ u l o c 2 n ] J ( u l o c ) a_n=(2n)![u_{\mathrm{loc}}^{2n}]J(u_{\mathrm{loc}}) a n = ( 2 n )! [ u loc 2 n ] J ( u loc ) are
a n = A n − B n − C n , A n = 129 + 344 n 6 9 n , B n = 300 n 2 + 150 n + 75 2 4 n , C n = 172 n − 215 2 . a_n=A_n-B_n-C_n,\qquad A_n=\frac{129+344n}{6}9^n,\qquad B_n=\frac{300n^2+150n+75}{2}4^n,\qquad C_n=172n-\frac{215}{2}. a n = A n − B n − C n , A n = 6 129 + 344 n 9 n , B n = 2 300 n 2 + 150 n + 75 4 n , C n = 172 n − 2 215 .
One has a 3 = 37453 > 0 a_3=37453>0 a 3 = 37453 > 0 ; for n ≥ 3 n\ge3 n ≥ 3 , A n + 1 > 9 A n A_{n+1}>9A_n A n + 1 > 9 A n , B n + 1 ≤ 9 B n B_{n+1}\le9B_n B n + 1 ≤ 9 B n , and C n + 1 ≤ 9 C n C_{n+1}\le9C_n C n + 1 ≤ 9 C n . The second inequality reduces to 20 n 2 − 22 n − 19 ≥ 0 20n^2-22n-19\ge0 20 n 2 − 22 n − 19 ≥ 0 . Consequently all a n a_n a n with n ≥ 3 n\ge3 n ≥ 3 are positive, and J ( u l o c ) ≥ P 5 ( u l o c ) J(u_{\mathrm{loc}})\ge P_5(u_{\mathrm{loc}}) J ( u loc ) ≥ P 5 ( u loc ) , where
P 5 ( y ) = 54 − 105 2 y − 1807 24 y 2 + 37453 720 y 3 + 314779 13440 y 4 + 13933781 3628800 y 5 . P_5(y)=54-\frac{105}{2}y-\frac{1807}{24}y^2+\frac{37453}{720}y^3+\frac{314779}{13440}y^4+\frac{13933781}{3628800}y^5. P 5 ( y ) = 54 − 2 105 y − 24 1807 y 2 + 720 37453 y 3 + 13440 314779 y 4 + 3628800 13933781 y 5 .
For y = 1 + v ≥ 1 y=1+v\ge1 y = 1 + v ≥ 1 ,
3628800 P 5 ( 1 + v ) = 19912031 + 238970785 v + 942350750 v 2 + 668062250 v 3 + 154659235 v 4 + 13933781 v 5 > 0. 3628800P_5(1+v)=19912031+238970785v+942350750v^2+668062250v^3+154659235v^4+13933781v^5>0. 3628800 P 5 ( 1 + v ) = 19912031 + 238970785 v + 942350750 v 2 + 668062250 v 3 + 154659235 v 4 + 13933781 v 5 > 0.
For 0 ≤ y ≤ 1 0\le y\le1 0 ≤ y ≤ 1 , expand P 5 P_5 P 5 in the degree-five Bernstein basis on each of the following intervals. The minimum Bernstein coefficients are
Table ?.
and are strictly positive. These entries are obtained by substituting y = a + ( b − a ) v y = a + (b-a)v y = a + ( b − a ) v and using b j = ∑ k = 0 j q k ( j k ) / ( 5 k ) b_j = \sum_{k=0}^{j} q_k \binom{j}{k}/\binom{5}{k} b j = ∑ k = 0 j q k ( k j ) / ( k 5 ) when P 5 ( a + ( b − a ) v ) = ∑ k = 0 5 q k v k P_5(a + (b-a)v) = \sum_{k=0}^{5} q_kv^k P 5 ( a + ( b − a ) v ) = ∑ k = 0 5 q k v k . This proves P 5 > 0 P_5 > 0 P 5 > 0 and hence (20) .
Finite arithmetic checks
The formulae of this appendix have been checked by exact rational arithmetic in standard-library Python.
The bounds on π \pi π used throughout are
π − : = 333 106 < π < 355 113 = : π + . \pi_- := \frac{333}{106} < \pi< \frac{355}{113} =: \pi_+. π − := 106 333 < π < 113 355 =: π + .
They may be certified using Machin’s identity π = 16 arctan ( 1 / 5 ) − 4 arctan ( 1 / 239 ) \pi= 16\arctan(1/5) - 4\arctan(1/239) π = 16 arctan ( 1/5 ) − 4 arctan ( 1/239 ) and the alternating series
A n ( x ) = ∑ ν = 0 n − 1 ( − 1 ) ν x 2 ν + 1 2 ν + 1 . A_n(x) = \sum_{\nu=0}^{n-1} \frac{(-1)^\nu x^{2\nu+1}}{2\nu+1}. A n ( x ) = ν = 0 ∑ n − 1 2 ν + 1 ( − 1 ) ν x 2 ν + 1 .
For 0 < x < 1 0 < x < 1 0 < x < 1 , A 2 n ( x ) < arctan x < A 2 n + 1 ( x ) A_{2n}(x) < \arctan x < A_{2n+1}(x) A 2 n ( x ) < arctan x < A 2 n + 1 ( x ) ; substitution with n = 5 n = 5 n = 5 in each term already gives rational bounds inside ( π − , π + ) (\pi_-,\pi_+) ( π − , π + ) .
For the boundary estimate let
δ = 361 1000 , b ∗ = 2239 5000 , V ‾ = 76607 250000 , z ∗ = 79451 100000 , c 3 = 2 3 ( 1 + 10 − 6 ) , σ ‾ = 55703 100000 . \delta= \frac{361}{1000}, \quad b_* = \frac{2239}{5000}, \quad\overline{V} = \frac{76607}{250000}, \quad z_* = \frac{79451}{100000}, \quad c_3 = \frac{2}{3}(1+10^{-6}), \quad\overline{\sigma} = \frac{55703}{100000}. δ = 1000 361 , b ∗ = 5000 2239 , V = 250000 76607 , z ∗ = 100000 79451 , c 3 = 3 2 ( 1 + 1 0 − 6 ) , σ = 100000 55703 .
With x ∗ = 66 / 5 x_* = 66/5 x ∗ = 66/5 , β = 47 / 100 \beta= 47/100 β = 47/100 , κ = 103 / 100 \kappa= 103/100 κ = 103/100 , and M S = 692881 / 57600 M_S = 692881/57600 M S = 692881/57600 , define the following rational margins:
Δ V = V ‾ − π + 2 [ 9 M S 2 128 x ∗ 4 ( 1 − σ ‾ ) 4 + 43 50 β 2 x ∗ 2 ] . \Delta_V = \overline{V} - \pi_+^2\left[\frac{9M_S^2}{128x_*^4(1-\overline{\sigma})^4} + \frac{43}{50\beta^2x_*^2}\right]. Δ V = V − π + 2 [ 128 x ∗ 4 ( 1 − σ ) 4 9 M S 2 + 50 β 2 x ∗ 2 43 ] .
Δ p = z ∗ 2 − κ V ‾ ( 2 + κ γ 0 2 V ‾ ) , \Delta_p = z_*^2 - \kappa\overline{V}(2+\kappa\gamma_0^2\overline{V}), Δ p = z ∗ 2 − κ V ( 2 + κ γ 0 2 V ) ,
Δ + = b ∗ 2 − [ ( c 3 / π − ) z ∗ 3 + V ‾ 2 ] , \Delta_+ = b_*^2 - \left[(c_3/\pi_-)z_*^3+\overline{V}^2\right], Δ + = b ∗ 2 − [ ( c 3 / π − ) z ∗ 3 + V 2 ] ,
Δ − = b ∗ 4 − ( c 3 / π − ) 2 ( 8 / 3 ) 3 δ 3 . \Delta_- = b_*^4 - (c_3/\pi_-)^2(8/3)^3\delta^3. Δ − = b ∗ 4 − ( c 3 / π − ) 2 ( 8/3 ) 3 δ 3 .
The exact positivity checks can equivalently be bounded below by
Δ V > 9 ⋅ 10 − 7 , Δ p > 4 ⋅ 10 − 6 , Δ + > 19 ⋅ 10 − 5 , Δ − > 3 ⋅ 10 − 5 . \Delta_V > 9\cdot10^{-7}, \quad\Delta_p > 4\cdot10^{-6}, \quad\Delta_+ > 19\cdot10^{-5}, \quad\Delta_- > 3\cdot10^{-5}. Δ V > 9 ⋅ 1 0 − 7 , Δ p > 4 ⋅ 1 0 − 6 , Δ + > 19 ⋅ 1 0 − 5 , Δ − > 3 ⋅ 1 0 − 5 .
The remaining boundary checks are β 31 < σ ‾ 40 \beta^{31}<\overline{\sigma}^{40} β 31 < σ 40 , ( 6 / 5 ) γ 0 2 < 10 − 6 (6/5)\gamma_0^2<10^{-6} ( 6/5 ) γ 0 2 < 1 0 − 6 , b ∗ > δ b_*>\delta b ∗ > δ , and 1 − 3 b ∗ / ( 4 δ ) = 503 / 7220 > 0 1-3b_*/(4\delta)=503/7220>0 1 − 3 b ∗ / ( 4 δ ) = 503/7220 > 0 . The localization Bernstein coefficients are displayed in Appendix § 9 ; the scalar cutoff proof records its interval estimates directly. Finally, the absorption uses the exact gap
5913 1000 − 147774 25000 = 51 25000 > 2 − 9 , 1 27 5913 1000 = 219 1000 . \frac{5913}{1000}-\frac{147774}{25000}=\frac{51}{25000}>2^{-9}, \qquad\frac{1}{27}\frac{5913}{1000}=\frac{219}{1000}. 1000 5913 − 25000 147774 = 25000 51 > 2 − 9 , 27 1 1000 5913 = 1000 219 .
These formulae specify the finite arithmetic independently of the software.
References
Instituto de Física, Pontificia Universidad Católica de Chile, Av. Vicuña Mackenna 4860, Macul, 7820436, Santiago, Chile
Email address: rbenguri@uc.cl
Mathematisches Institut, Ludwig-Maximilians-Universität München, Theresienstrasse 39, 80333 München, and Munich Center for Quantum Science and Technology (MCQST), Schellingstr. 4, 80799 München, Germany
Email address: h.s@lmu.de