Introduction and statement of main results
Let L = g Z 3 ⊂ R 3 L=g\mathbb{Z}^{3}\subset\mathbb{R}^{3} L = g Z 3 ⊂ R 3 be a Bravais lattice of covolume ∣ L ∣ = ∣ det g ∣ |L|=|\det g| ∣ L ∣ = ∣ det g ∣ . Lots of physical and mathematical problems can be formulated as the minimization of the f f f -potential lattice energy as follows:
E f ( L ) : = ∑ v ∈ L ∖ { 0 } f ( ∣ v ∣ 2 ) , inf ∣ L ∣ = V E f ( L ) , V > 0. (1) E_f(L):=\sum_{v\in L\setminus\{0\}}f(|v|^2),\qquad\inf_{|L|=V}E_f(L),\qquad V>0. \tag*{(1)} E f ( L ) := v ∈ L ∖ { 0 } ∑ f ( ∣ v ∣ 2 ) , ∣ L ∣ = V inf E f ( L ) , V > 0. (1)
Here ∣ v ∣ |v| ∣ v ∣ denotes Euclidean length, V − 1 V^{-1} V − 1 is the number density, and the sum is assumed absolutely convergent. We identify lattices up to orthogonal equivalence, denoted by ≅ \cong ≅ .
Considering the Riesz potential f ( t ) = t − s f(t)=t^{-s} f ( t ) = t − s in (1) , we have the Epstein zeta function
E f ( L ) = E ( L , s ) : = ∑ v ∈ L ∖ { 0 } ∣ v ∣ − 2 s , s > 3 2 . (2) E_f(L)=E(L,s):=\sum_{v\in L\setminus\{0\}}|v|^{-2s},\qquad s>\frac{3}{2}. \tag*{(2)} E f ( L ) = E ( L , s ) := v ∈ L ∖ { 0 } ∑ ∣ v ∣ − 2 s , s > 2 3 . (2)
In our previous work [23 Theorem 1.2] , we proved the minima of 3D Epstein zeta functions conjectured by Sarnak and Strömbergsson [28 Section 5] . It is stated as follows:
Theorem A ([23 ] ). For every Bravais lattice L ⊂ R 3 L\subset\mathbb{R}^{3} L ⊂ R 3 of covolume one,
E ( L , s ) ≥ E ( F C C , s ) , s > 3 2 , (3) E(L,s)\geq E(\mathrm{FCC},s),\qquad s>\frac{3}{2}, \tag*{(3)} E ( L , s ) ≥ E ( FCC , s ) , s > 2 3 , (3) with equality if and only if L ≅ F C C L\cong\mathrm{FCC} L ≅ FCC . Here FCC denotes the face-centred cubic lattice of covolume one.
In this paper, we consider the widely used Lennard–Jones potential ([15 , 31 , 17 ] ) in dimension three:
f ( r 2 ) = 4 ε [ ( σ r ) 12 − ( σ r ) 6 ] , r > 0 , ε , σ > 0. (4) f(r^2)=4\varepsilon\left[\left(\frac{\sigma}{r}\right)^{12}-\left(\frac{\sigma}{r}\right)^6\right],\qquad r>0,\quad\varepsilon,\sigma>0. \tag*{(4)} f ( r 2 ) = 4 ε [ ( r σ ) 12 − ( r σ ) 6 ] , r > 0 , ε , σ > 0. (4)
The constants ε \varepsilon ε and σ \sigma σ specify the well depth and length scale; see . Measuring lengths in units 2 1 / 6 σ 2^{1/6}\sigma 2 1/6 σ and energies in units ε \varepsilon ε gives r − 12 − 2 r − 6 r^{-12}-2r^{-6} r − 12 − 2 r − 6 . Under the Lennard-Jones potential, the lattice energy in (1) becomes
E ( L ) = 1 2 E ( L , 6 ) − E ( L , 3 ) , m ( V ) = inf ∣ L ∣ = V E ( L ) . (5) \mathcal{E}(L)=\frac{1}{2}E(L,6)-E(L,3),\qquad m(V)=\inf_{|L|=V}\mathcal{E}(L). \tag*{(5)} E ( L ) = 2 1 E ( L , 6 ) − E ( L , 3 ) , m ( V ) = ∣ L ∣ = V inf E ( L ) . (5)
After normalization to covolume one, we have
2 V 4 m ( V ) = inf ∣ L ∣ = 1 { E ( L , 6 ) − 2 V 2 E ( L , 3 ) } , inf a > 0 E ( a L ) = − E ( L , 3 ) 2 2 E ( L , 6 ) . 2V^{4}m(V)=\inf_{|L|=1}\{E(L,6)-2V^{2}E(L,3)\},\qquad\inf_{a>0}\mathcal{E}(aL)=-\frac{E(L,3)^{2}}{2E(L,6)}. 2 V 4 m ( V ) = ∣ L ∣ = 1 inf { E ( L , 6 ) − 2 V 2 E ( L , 3 )} , a > 0 inf E ( a L ) = − 2 E ( L , 6 ) E ( L , 3 ) 2 .
The Lennard–Jones potential is a basic model for intermolecular forces, clusters and condensed matter, and has a long history; see a recent review by Schwerdtfeger-Wales [33 ] in 100 Years of the Lennard-Jones Potential . The crystal-energy problem dates back to Jones-Ingham’s 1925 study of minimum-energy cubic crystals [18 ] . There are tons of numerical simulations of Lennard–Jones energy; see, for example in [29 , 10 ] and the following citing references.
To state our main result, we introduce various important lattices and their notations. We use the unit-covolume FCC, body-centred cubic (BCC) and simple cubic lattices
D 3 = { n ∈ Z 3 : n 1 + n 2 + n 3 ∈ 2 Z } , F = F C C = 2 − 1 / 3 D 3 , D_{3}=\{n\in\mathbb{Z}^{3}:n_{1}+n_{2}+n_{3}\in2\mathbb{Z}\},\qquad F=\mathrm{FCC}=2^{-1/3}D_{3}, D 3 = { n ∈ Z 3 : n 1 + n 2 + n 3 ∈ 2 Z } , F = FCC = 2 − 1/3 D 3 ,
B C C = 2 1 / 3 [ Z 3 ∪ ( Z 3 + ( 1 2 , 1 2 , 1 2 ) ) ] , S C = Z 3 . \mathrm{BCC}=2^{1/3}\left[\mathbb{Z}^{3}\cup\left(\mathbb{Z}^{3}+\left(\frac{1}{2},\frac{1}{2},\frac{1}{2}\right)\right)\right],\qquad\mathrm{SC}=\mathbb{Z}^{3}. BCC = 2 1/3 [ Z 3 ∪ ( Z 3 + ( 2 1 , 2 1 , 2 1 ) ) ] , SC = Z 3 .
See Bétermin–Šamaj–Travěnec [7 ] . The triangular and simple-hexagonal families are denoted by
T A = 2 A 3 span Z { ( 1 , 0 ) , ( 1 2 , 3 2 ) } , A > 0 , T_{A}=\sqrt{\frac{2A}{\sqrt{3}}}\operatorname{span}_{\mathbb{Z}}\left\{(1,0),\left(\frac{1}{2},\frac{\sqrt{3}}{2}\right)\right\},\qquad A>0, T A = 3 2 A span Z { ( 1 , 0 ) , ( 2 1 , 2 3 ) } , A > 0 ,
H Y = ( 4 3 Y ) 1 / 6 span Z { ( 1 , 0 , 0 ) , ( 1 2 , 3 2 , 0 ) , ( 0 , 0 , Y ) } , Y > 0. H_{Y}=\left(\frac{4}{3Y}\right)^{1/6}\operatorname{span}_{\mathbb{Z}}\left\{(1,0,0),\left(\frac{1}{2},\frac{\sqrt{3}}{2},0\right),(0,0,\sqrt{Y})\right\},\qquad Y>0. H Y = ( 3 Y 4 ) 1/6 span Z { ( 1 , 0 , 0 ) , ( 2 1 , 2 3 , 0 ) , ( 0 , 0 , Y ) } , Y > 0.
Then ∣ T A ∣ = A |T_{A}|=A ∣ T A ∣ = A , ∣ L X ∣ = ∣ H Y ∣ = 1 |L_{X}|=|H_{Y}|=1 ∣ L X ∣ = ∣ H Y ∣ = 1 , and
V 1 / 3 H Y = T A ⊕ ( V / A ) Z , A = ( 3 V 2 2 Y ) 1 / 3 . V^{1/3}H_{Y}=T_{A}\oplus(V/A)\mathbb{Z},\qquad A=\left(\frac{\sqrt{3}V^{2}}{2Y}\right)^{1/3}. V 1/3 H Y = T A ⊕ ( V / A ) Z , A = ( 2 Y 3 V 2 ) 1/3 .
We shall also introduce a deformation family of the FCC lattice, namely, the unit-covolume body-centered tetragonal family:
L X = ( 4 X ) − 1 / 6 diag ( X , 1 , 1 ) D 3 , X > 0 ; L 1 = F , L 1 / 2 ≅ B C C . L_{X}=(4X)^{-1/6}\operatorname{diag}(\sqrt{X},1,1)D_{3},\qquad X>0;\qquad L_{1}=F,\quad L_{1/2}\cong\mathrm{BCC}. L X = ( 4 X ) − 1/6 diag ( X , 1 , 1 ) D 3 , X > 0 ; L 1 = F , L 1/2 ≅ BCC .
The three-dimensional Lennard-Jones lattice problem (5) attracts lots of attentions over the years. Bétermin–Šamaj–Travěnec [7 ] made a conjecture for general Lennard-Jones potential by numerical simulations.
Conjecture 1 (Bétermin–Šamaj–Travěnec [7 Conjecture 2] ). Assume that p > q > 3 p>q>3 p > q > 3 . Then there exist V p q > 0 V_{pq}>0 V pq > 0 such that
argmin ∣ L ∣ = V q E ( L , p / 2 ) − p E ( L , q / 2 ) 2 ( p − q ) = { { F } , 0 < V < V p q , { H Y ( V ) } , V > V p q . \underset{|L|=V}{\operatorname{argmin}}\frac{qE(L,p/2)-pE(L,q/2)}{2(p-q)}
=
\begin{cases}
\{F\}, & 0<V<V_{pq},\\
\{H_{Y}(V)\}, & V>V_{pq}.
\end{cases} ∣ L ∣ = V argmin 2 ( p − q ) q E ( L , p /2 ) − pE ( L , q /2 ) = { { F } , { H Y ( V )} , 0 < V < V pq , V > V pq .
For the classical (12, 6) potential, Bétermin–Šamaj–Travěnec [7 ] predicted a phase transition from the FCC to the simple-hexagonal lattice as the inverse density increases:
F C C ⟶ S H ( V ↑ ) . (6) \mathrm{FCC}\longrightarrow\mathrm{SH}\qquad(V\uparrow). \tag*{(6)} FCC ⟶ SH ( V ↑ ) . (6)
The three-dimensional Lennard-Jones lattice problem (5) is also listed as an open problem by by Blanc-Lewin [9 Section 2.5, (27), pp. 268–269] . It is also listed as an open problem as follows:
Question 1 (Bétermin–Petrache [6 Question 1.15] ). For the classical potential f ( r 2 ) = r − 12 − r − 6 f(r^{2})=r^{-12}-r^{-6} f ( r 2 ) = r − 12 − r − 6 , which unit-covolume Bravais lattice shapes minimize E f ( λ L ) E_{f}(\lambda L) E f ( λ L ) for each λ > 0 \lambda>0 λ > 0 ?
The change of scale between this question and (5) is
E f ( λ L ) = 1 2 E ( 2 − 1 / 6 λ L ) , ∣ L ∣ = 1 , V = λ 3 2 . E_{f}(\lambda L)=\frac{1}{2}\mathcal{E}(2^{-1/6}\lambda L),\qquad|L|=1,\qquad V=\frac{\lambda^{3}}{\sqrt{2}}. E f ( λ L ) = 2 1 E ( 2 − 1/6 λ L ) , ∣ L ∣ = 1 , V = 2 λ 3 .
We completely classify the minima of problem (5) , thereby providing an positive answer to Question 1, on open problem of Blanc–Lewin on 3D Lennard-Jones lattice energy ([9, Section 2.5, (27), pp. 268–269]), and identifying the intervening tetragonal phase beyond Conjecture 1.
Theorem 1.1 (Minima of 3D Lennard-Jones lattice energy). Let ∣ L ∣ = V |L| = V ∣ L ∣ = V . Up to orthogonal equivalence, there exist two thresholds 0 < V F T < V T H < ∞ 0 < V_{\mathrm{FT}} < V_{\mathrm{TH}} < \infty 0 < V FT < V TH < ∞ such that:
argmin ∣ L ∣ = V ( 1 2 E ( L , 6 ) − E ( L , 3 ) ) = { { F C C } , 0 < V < V F T , { F C C , L X ( V ) } , V = V F T , { L X ( V ) } , V F T < V < V T H , { L X ( V ) , H Y ( V ) } , V = V T H , { H Y ( V ) } , V > V T H . \underset{|L|=V}{\operatorname{argmin}}\left(\frac{1}{2}E(L,6)-E(L,3)\right)=
\begin{cases}
\{FCC\}, & 0<V<V_{\mathrm{FT}},\\
\{FCC,L_{X(V)}\}, & V=V_{\mathrm{FT}},\\
\{L_{X(V)}\}, & V_{\mathrm{FT}}<V<V_{\mathrm{TH}},\\
\{L_{X(V)},H_{Y(V)}\}, & V=V_{\mathrm{TH}},\\
\{H_{Y(V)}\}, & V>V_{\mathrm{TH}}.
\end{cases} ∣ L ∣ = V argmin ( 2 1 E ( L , 6 ) − E ( L , 3 ) ) = ⎩ ⎨ ⎧ { F C C } , { F C C , L X ( V ) } , { L X ( V ) } , { L X ( V ) , H Y ( V ) } , { H Y ( V ) } , 0 < V < V FT , V = V FT , V FT < V < V TH , V = V TH , V > V TH . Numerically,
V F T = 1.0648235668 ⋯ , V T H = 1.0759906200 ⋯ . V_{\mathrm{FT}}=1.0648235668\cdots,\qquad V_{\mathrm{TH}}=1.0759906200\cdots. V FT = 1.0648235668 ⋯ , V TH = 1.0759906200 ⋯ . Analytically, the thresholds V F T V_{\mathrm{FT}} V FT and V T H V_{\mathrm{TH}} V TH are the v v v -components of the unique solutions ( v , x ) (v,x) ( v , x ) and ( v , x , y ) (v,x,y) ( v , x , y ) of, respectively,
{ d d x [ E ( L x , 6 ) − 2 v 2 E ( L x , 3 ) ] = 0 , E ( L x , 6 ) − E ( F , 6 ) = 2 v 2 [ E ( L x , 3 ) − E ( F , 3 ) ] , (7) \begin{cases}
\dfrac{\mathrm{d}}{\mathrm{d}x}\left[E(L_x,6)-2v^2E(L_x,3)\right]=0,\\
E(L_x,6)-E(F,6)=2v^2\left[E(L_x,3)-E(F,3)\right], \tag*{(7)}
\end{cases} ⎩ ⎨ ⎧ d x d [ E ( L x , 6 ) − 2 v 2 E ( L x , 3 ) ] = 0 , E ( L x , 6 ) − E ( F , 6 ) = 2 v 2 [ E ( L x , 3 ) − E ( F , 3 ) ] , (7) { d d x [ E ( L x , 6 ) − 2 v 2 E ( L x , 3 ) ] = 0 , d d y [ E ( H y , 6 ) − 2 v 2 E ( H y , 3 ) ] = 0 , E ( H y , 6 ) − E ( L x , 6 ) = 2 v 2 [ E ( H y , 3 ) − E ( L x , 3 ) ] , (8) \begin{cases}
\dfrac{\mathrm{d}}{\mathrm{d}x}\left[E(L_x,6)-2v^2E(L_x,3)\right]=0,\\
\dfrac{\mathrm{d}}{\mathrm{d}y}\left[E(H_y,6)-2v^2E(H_y,3)\right]=0,\\
E(H_y,6)-E(L_x,6)=2v^2\left[E(H_y,3)-E(L_x,3)\right], \tag*{(8)}
\end{cases} ⎩ ⎨ ⎧ d x d [ E ( L x , 6 ) − 2 v 2 E ( L x , 3 ) ] = 0 , d y d [ E ( H y , 6 ) − 2 v 2 E ( H y , 3 ) ] = 0 , E ( H y , 6 ) − E ( L x , 6 ) = 2 v 2 [ E ( H y , 3 ) − E ( L x , 3 ) ] , (8) in the ranges
21 20 < v < 11 10 , 23 100 < x < 307 1000 , 103 100 < y < 103 50 . \frac{21}{20}<v<\frac{11}{10},\qquad\frac{23}{100}<x<\frac{307}{1000},\qquad\frac{103}{100}<y<\frac{103}{50}. 20 21 < v < 10 11 , 100 23 < x < 1000 307 , 100 103 < y < 50 103 . Alternatively, the first transition can be located as follows:
V F T 2 = 1 2 inf ∣ L ∣ = 1 L ≇ F E ( L , 6 ) − E ( F , 6 ) E ( L , 3 ) − E ( F , 3 ) . (9) V_{\mathrm{FT}}^2=\frac{1}{2}\inf_{\substack{|L|=1\\L\not\cong F}}\frac{E(L,6)-E(F,6)}{E(L,3)-E(F,3)}. \tag*{(9)} V FT 2 = 2 1 ∣ L ∣ = 1 L ≅ F inf E ( L , 3 ) − E ( F , 3 ) E ( L , 6 ) − E ( F , 6 ) . (9) Furthermore, the parameters X ( V ) X(V) X ( V ) and Y ( V ) Y(V) Y ( V ) can be located by solving equations.
(i) For 21 / 20 ≤ V ≤ 11 / 10 21/20\le V\le11/10 21/20 ≤ V ≤ 11/10 , the parameters X ( V ) X(V) X ( V ) and Y ( V ) Y(V) Y ( V ) are uniquely determined by
d E ( L X , 6 ) d log X = 2 V 2 d E ( L X , 3 ) d log X , (10) \frac{\mathrm{d}E(L_X,6)}{\mathrm{d}\log X}=2V^2\frac{\mathrm{d}E(L_X,3)}{\mathrm{d}\log X}, \tag*{(10)} d log X d E ( L X , 6 ) = 2 V 2 d log X d E ( L X , 3 ) , (10) d E ( H Y , 6 ) d log Y = 2 V 2 d E ( H Y , 3 ) d log Y , (11) \frac{\mathrm{d}E(H_Y,6)}{\mathrm{d}\log Y}=2V^2\frac{\mathrm{d}E(H_Y,3)}{\mathrm{d}\log Y}, \tag*{(11)} d log Y d E ( H Y , 6 ) = 2 V 2 d log Y d E ( H Y , 3 ) , (11) in 23 / 100 < X < 307 / 1000 23/100<X<307/1000 23/100 < X < 307/1000 and 103 / 100 < Y < 103 / 50 103/100<Y<103/50 103/100 < Y < 103/50 , respectively, with X ′ ( V ) < 0 < Y ′ ( V ) X'(V)<0<Y'(V) X ′ ( V ) < 0 < Y ′ ( V ) .
(ii) For 11 / 10 ≤ V ≤ 89 / 50 11/10\le V\le89/50 11/10 ≤ V ≤ 89/50 , Y ( V ) Y(V) Y ( V ) is the unique solution of (11) in 11 / 10 < Y < 217 / 50 11/10<Y<217/50 11/10 < Y < 217/50 ,
(iii) For V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 , the simple-hexagonal minimizer is
L V = T A ( V ) ⊕ V A ( V ) Z , Y ( V ) = 3 V 2 2 A ( V ) 3 , L_V=T_{A(V)}\oplus\frac{V}{A(V)}\mathbb{Z},\qquad Y(V)=\frac{\sqrt{3}V^2}{2A(V)^3}, L V = T A ( V ) ⊕ A ( V ) V Z , Y ( V ) = 2 A ( V ) 3 3 V 2 , where A ( V ) A(V) A ( V ) is the unique solution of
d d A E ( T A ⊕ ( V / A ) Z ) = 0 , 2 5 < A < 89 100 . \frac{\mathrm{d}}{\mathrm{d}A}\mathcal{E}\left(T_A\oplus(V/A)\mathbb{Z}\right)=0,\qquad\frac{2}{5}<A<\frac{89}{100}. d A d E ( T A ⊕ ( V / A ) Z ) = 0 , 5 2 < A < 100 89 .
As V → ∞ V \to\infty V → ∞ ,
A ( V ) = A ∗ + π ζ ( 4 ) S 6 A ∗ 4 3 S 3 2 V − 4 + O ( V − 8 ) , (12) A(V)=A_*+\frac{\pi\zeta(4)S_6A_*^4}{3S_3^2}V^{-4}+O(V^{-8}), \tag*{(12)} A ( V ) = A ∗ + 3 S 3 2 π ζ ( 4 ) S 6 A ∗ 4 V − 4 + O ( V − 8 ) , (12)
m ( V ) = e 2 − π ζ ( 4 ) A ∗ 3 V − 4 − π 2 ζ ( 4 ) 2 S 6 A ∗ 6 2 S 3 2 V − 8 + π ζ ( 10 ) A ∗ 9 5 V − 10 + O ( V − 12 ) . (13) m(V)=e_2-\pi\zeta(4)A_*^3V^{-4}-\frac{\pi^2\zeta(4)^2S_6A_*^6}{2S_3^2}V^{-8}+\frac{\pi\zeta(10)A_*^9}{5}V^{-10}+O(V^{-12}). \tag*{(13)} m ( V ) = e 2 − π ζ ( 4 ) A ∗ 3 V − 4 − 2 S 3 2 π 2 ζ ( 4 ) 2 S 6 A ∗ 6 V − 8 + 5 π ζ ( 10 ) A ∗ 9 V − 10 + O ( V − 12 ) . (13)
Here,
S s = ∑ ( i , j ) ∈ Z 2 ( i , j ) ≠ ( 0 , 0 ) ( i 2 + i j + j 2 ) − s ( s = 3 , 6 ) , A ∗ = 3 2 ( S 6 S 3 ) 1 / 3 , e 2 = − S 3 2 2 S 6 . S_s=\sum_{\substack{(i,j)\in\mathbb{Z}^2\\(i,j)\ne(0,0)}}(i^2+ij+j^2)^{-s}\quad(s=3,6),\quad A_*=\frac{\sqrt{3}}{2}\left(\frac{S_6}{S_3}\right)^{1/3},\quad e_2=-\frac{S_3^2}{2S_6}. S s = ( i , j ) ∈ Z 2 ( i , j ) = ( 0 , 0 ) ∑ ( i 2 + ij + j 2 ) − s ( s = 3 , 6 ) , A ∗ = 2 3 ( S 3 S 6 ) 1/3 , e 2 = − 2 S 6 S 3 2 .
(iv) m ( V ) m(V) m ( V ) in (5) increases on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) .
For the Lennard-Jones potential with free parameter b b b (r − 12 − b r − 6 r^{-12}-br^{-6} r − 12 − b r − 6 , b > 0 b>0 b > 0 ), the corresponding result follows from
E b ( L ) = b 2 4 E ( ( b / 2 ) 1 / 6 L ) , V F T ( b ) = 2 / b V F T , V T H ( b ) = 2 / b V T H . \mathcal{E}_b(L)=\frac{b^2}{4}\mathcal{E}\left((b/2)^{1/6}L\right),\quad V_{\rm FT}(b)=\sqrt{2/b}\,V_{\rm FT},\quad V_{\rm TH}(b)=\sqrt{2/b}\,V_{\rm TH}. E b ( L ) = 4 b 2 E ( ( b /2 ) 1/6 L ) , V FT ( b ) = 2/ b V FT , V TH ( b ) = 2/ b V TH .
By Theorem 1.1 , we provide a complete solution the seemingly simple three-dimensional Lennard-Jones lattice problem (5) .
In dimension two, Montgomery [24 ] proved triangular optimality for every Gaussian scale. For 2D Lennard–Jones energy, we refer to the Bétérmin and his collaborators [1 , 8 , 4 , 6 ] for pioneering works; our previous work [22 ] gives the complete classification. In three dimensions, some local minimal and stability results can be found in [2 , 6 , 5 ] ; see also [10 ] for cuboidal BCC instability.
In dimensions 8 and 24, Cohn-Kumar-Miller-Radchenko-Viazovska [12 ] proved universal optimality of E 8 E_8 E 8 and the Leech lattice. Petrache-Serfaty [26 ] transferred universal optimality to Coulomb and Riesz renormalized and jellium energies, obtaining the E 8 E_8 E 8 and Leech minima in dimensions 8 and 24.
There are many related crystallization results, and we refer to Blanc-Lewin [9 ] and Lewin [21 ] for reviews. Sandier-Serfaty [27 ] proved triangular optimality among lattices of fixed density for their established Coulomb renormalized energy; Chen-Oshita [11 ] proved triangular lattice in di-block copolymers. Petrache-Serfaty [25 ] identified the renormalized energy governing the microscopic next-order asymptotics of confined Riesz systems. Heitmann-Radin [16 ] proved triangular crystallization for sticky disks; Theil [30 ] proved a triangular crystallization, and E-Li [13 ] proved a hexagonal crystallization for a three-body potential. Flatley-Theil [14 ] proved FCC crystallization for a class of three-dimensional two- and three-body energies.
We now discuss some ideas in the proof of Theorem 1.1 . Theorem A provides the s = 3 s=3 s = 3 comparison used for the high-density phase. For the dilute phase, we use Bétérmin’s theorem [4 Theorem 1.1] and its fixed-area comparison [4 Proposition 1.2] : for every unit-area planar lattice Γ \Gamma Γ ,
E ( Γ , 6 ) − E ( T 1 , 6 ) ≥ 2 ( E ( Γ , 3 ) − E ( T 1 , 3 ) ) . (14) E(\Gamma,6)-E(T_1,6)\ge2\bigl(E(\Gamma,3)-E(T_1,3)\bigr). \tag*{(14)} E ( Γ , 6 ) − E ( T 1 , 6 ) ≥ 2 ( E ( Γ , 3 ) − E ( T 1 , 3 ) ) . (14)
The proof uses a planar reduction for large covolumes and a profile comparison on the remaining finite covolume intervals. For large V V V , write L = Γ + Z ( u + h ν ) L=\Gamma+\mathbb{Z}(u+h\nu) L = Γ + Z ( u + h ν ) , A = ∣ Γ ∣ A=|\Gamma| A = ∣Γ∣ , A h = V Ah=V A h = V . The planar comparison and Fourier inequalities give
E ( L ) ≥ E ( T A ⊕ ( V / A ) Z ) , ∂ A 2 E ( T A ⊕ ( V / A ) Z ) > 0 \mathcal{E}(L)\ge\mathcal{E}\left(T_A\oplus(V/A)\mathbb{Z}\right),\qquad\partial_A^2\mathcal{E}\left(T_A\oplus(V/A)\mathbb{Z}\right)>0 E ( L ) ≥ E ( T A ⊕ ( V / A ) Z ) , ∂ A 2 E ( T A ⊕ ( V / A ) Z ) > 0
on the localized area interval. On each remaining covolume interval [ V 0 , V 1 ] [V_0,V_1] [ V 0 , V 1 ] , let f ( V ) f(V) f ( V ) be the least candidate value of E ( L , 6 ) − 2 V 2 E ( L , 3 ) E(L,6)-2V^2E(L,3) E ( L , 6 ) − 2 V 2 E ( L , 3 ) . Then
g ( t ) = max V 0 ≤ V ≤ V 1 { f ( V ) + 2 V 2 t } , E ( L , 6 ) − g ( E ( L , 3 ) ) ≥ 0 ( ∣ L ∣ = 1 ) . g(t)=\max_{V_0\le V\le V_1}\{f(V)+2V^2t\},\qquad E(L,6)-g(E(L,3))\ge0\quad(|L|=1). g ( t ) = V 0 ≤ V ≤ V 1 max { f ( V ) + 2 V 2 t } , E ( L , 6 ) − g ( E ( L , 3 )) ≥ 0 ( ∣ L ∣ = 1 ) .
Every profile minimum minimizes E ( L , 6 ) − 2 V 2 E ( L , 3 ) E(L,6)-2V^2E(L,3) E ( L , 6 ) − 2 V 2 E ( L , 3 ) at each supporting V V V . Moment bounds give compactness; Fourier and stationary estimates determine the equality cases. Comparison (3) extends the result to V < 21 / 20 V<21/20 V < 21/20 .
We collect some preliminaries in Section 2. Section 3 deals with the dilute phase. Section 4 establishes the endpoint comparisons; Sections 5 and 6 prove the transition and intervening simple-hexagonal phases. The long but finite computations are collected in the Appendix.
Preliminaries and stationary branches
Packing, convexity and Ewald estimates give existence, fixed-systole comparisons and the stationary branches.
For 1 ≤ rank L ≤ 3 1 \le\operatorname{rank} L \le3 1 ≤ rank L ≤ 3 , put λ = λ 1 ( L ) : = min v ∈ L ∖ { 0 } ∣ v ∣ > 0 \lambda=\lambda_1(L):=\min_{v\in L\setminus\{0\}}|v|>0 λ = λ 1 ( L ) := min v ∈ L ∖ { 0 } ∣ v ∣ > 0 . Disjoint balls of radius λ / 2 \lambda/2 λ /2 give
# ( L ∩ B R ) ≤ ( 1 + 2 R / λ ) 3 . \#(L\cap B_R)\le(1+2R/\lambda)^3. # ( L ∩ B R ) ≤ ( 1 + 2 R / λ ) 3 .
Summation over the annuli 2 j λ ≤ ∣ v ∣ < 2 j + 1 λ 2^j\lambda\le|v|<2^{j+1}\lambda 2 j λ ≤ ∣ v ∣ < 2 j + 1 λ yields
E ( L , s ) ≤ 125 1 − 2 3 − 2 s λ − 2 s ( s > 3 / 2 ) . (15) E(L,s)\le\frac{125}{1-2^{3-2s}}\lambda^{-2s}\qquad(s>3/2). \tag*{(15)} E ( L , s ) ≤ 1 − 2 3 − 2 s 125 λ − 2 s ( s > 3/2 ) . (15)
For λ ≥ ρ > 0 \lambda\ge\rho>0 λ ≥ ρ > 0 and R ≥ ρ R\ge\rho R ≥ ρ , similarly,
∑ v ∈ L ∣ v ∣ > R ∣ v ∣ − 2 s ≤ 125 1 − 2 3 − 2 s ρ − 3 R 3 − 2 s . (16) \sum_{\substack{v\in L\\ |v|>R}}|v|^{-2s}\le\frac{125}{1-2^{3-2s}}\rho^{-3}R^{3-2s}. \tag*{(16)} v ∈ L ∣ v ∣ > R ∑ ∣ v ∣ − 2 s ≤ 1 − 2 3 − 2 s 125 ρ − 3 R 3 − 2 s . (16)
Integration gives
E ( L , 3 ) = 6 ∫ λ ∞ # { v ∈ L : 0 < ∣ v ∣ ≤ t } t − 7 d t ≤ 6 ∫ λ ∞ [ ( 1 + 2 t / λ ) 3 − 1 ] t − 7 d t = 206 5 λ − 6 . E(L,3)=6\int_\lambda^\infty\#\{v\in L:0<|v|\le t\}t^{-7}\,\mathrm{d}t\le6\int_\lambda^\infty[(1+2t/\lambda)^3-1]t^{-7}\,\mathrm{d}t=\frac{206}{5}\lambda^{-6}. E ( L , 3 ) = 6 ∫ λ ∞ # { v ∈ L : 0 < ∣ v ∣ ≤ t } t − 7 d t ≤ 6 ∫ λ ∞ [( 1 + 2 t / λ ) 3 − 1 ] t − 7 d t = 5 206 λ − 6 .
Hence
E ( L ) ≥ λ − 12 − 206 5 λ − 6 . (17) \mathcal{E}(L)\ge\lambda^{-12}-\frac{206}{5}\lambda^{-6}. \tag*{(17)} E ( L ) ≥ λ − 12 − 5 206 λ − 6 . (17)
Thus
inf L E ( L ) > − ∞ , sup n E ( L n ) < ∞ ⟹ inf n λ 1 ( L n ) > 0. \inf_L\mathcal{E}(L)>-\infty,\qquad\sup_n\mathcal{E}(L_n)<\infty\Longrightarrow\inf_n\lambda_1(L_n)>0. L inf E ( L ) > − ∞ , n sup E ( L n ) < ∞ ⟹ n inf λ 1 ( L n ) > 0.
Lemma 2.1 (A quantitative separation bound). *For a lattice L ⊂ R 3 L\subset\mathbb{R}^3 L ⊂ R 3 ,
E ( L ) ≤ − 3 ⟹ min v ∈ L ∖ { 0 } ∣ v ∣ > 7 10 . \mathcal{E}(L)\le-3\quad\Longrightarrow\quad\min_{v\in L\setminus\{0\}}|v|>\frac{7}{10}. E ( L ) ≤ − 3 ⟹ v ∈ L ∖ { 0 } min ∣ v ∣ > 10 7 . Proof. Suppose 0 < λ ≤ 7 / 10 0<\lambda\le7/10 0 < λ ≤ 7/10 and ∣ v 0 ∣ = λ |v_0|=\lambda ∣ v 0 ∣ = λ , v 0 ∈ L v_0\in L v 0 ∈ L . Disjoint radius-λ / 2 \lambda/2 λ /2 balls give
# { v ∈ L ∖ { 0 , ± v 0 } : ∣ v ∣ ≤ r } ≤ ( 1 + 2 r / λ ) 3 − 3 ( r ≥ 1 ) . \#\{v\in L\setminus\{0,\pm v_0\}:|v|\le r\}\le(1+2r/\lambda)^3-3\qquad(r\ge1). # { v ∈ L ∖ { 0 , ± v 0 } : ∣ v ∣ ≤ r } ≤ ( 1 + 2 r / λ ) 3 − 3 ( r ≥ 1 ) . For every r > 0 r>0 r > 0 ,
r − 12 − 2 r − 6 ≥ − 12 ∫ max { 1 , r } ∞ ( t − 7 − t − 13 ) d t . r^{-12}-2r^{-6}\ge-12\int_{\max\{1,r\}}^\infty(t^{-7}-t^{-13})\,\mathrm{d}t. r − 12 − 2 r − 6 ≥ − 12 ∫ m a x { 1 , r } ∞ ( t − 7 − t − 13 ) d t . Tonelli’s theorem gives
E ( L ) ≥ λ − 12 − 2 λ − 6 − 6 ∫ 1 ∞ ( t − 7 − t − 13 ) ( ( 1 + 2 t / λ ) 3 − 3 ) d t = λ − 12 − 2 λ − 6 + 1 − 216 55 λ − 54 5 λ 2 − 32 3 λ 3 . \begin{aligned}
\mathcal{E}(L)&\ge\lambda^{-12}-2\lambda^{-6}-6\int_1^\infty(t^{-7}-t^{-13})((1+2t/\lambda)^3-3)\,\mathrm{d}t\\
&=\lambda^{-12}-2\lambda^{-6}+1-\frac{216}{55\lambda}-\frac{54}{5\lambda^2}-\frac{32}{3\lambda^3}.
\end{aligned} E ( L ) ≥ λ − 12 − 2 λ − 6 − 6 ∫ 1 ∞ ( t − 7 − t − 13 ) (( 1 + 2 t / λ ) 3 − 3 ) d t = λ − 12 − 2 λ − 6 + 1 − 55 λ 216 − 5 λ 2 54 − 3 λ 3 32 . For x = 1 / λ x=1/\lambda x = 1/ λ , the x − 5 x^{-5} x − 5 -weighted derivative is
12 x 6 − 12 − 216 55 x 5 − 108 5 x 4 − 32 x 3 , 12x^6-12-\frac{216}{55x^5}-\frac{108}{5x^4}-\frac{32}{x^3}, 12 x 6 − 12 − 55 x 5 216 − 5 x 4 108 − x 3 32 , strictly increasing on ( 0 , ∞ ) (0,\infty) ( 0 , ∞ ) and positive at 10 / 7 10/7 10/7 . The energy lower bound at x = 10 / 7 x=10/7 x = 10/7 is > − 3 >-3 > − 3 ; hence E ( L ) > − 3 \mathcal{E}(L)>-3 E ( L ) > − 3 , contradicting E ( L ) ≤ − 3 \mathcal{E}(L)\le-3 E ( L ) ≤ − 3 . This completes the proof of Lemma 2.1. □ \square □
Mahler’s compactness criterion [34 ] states that, for a > 0 a > 0 a > 0 and 0 < V 1 ≤ V 2 < ∞ 0 < V_1 \le V_2 < \infty 0 < V 1 ≤ V 2 < ∞ , the set
{ g Z 3 : g ∈ G L 3 ( R ) , V 1 ≤ ∣ det g ∣ ≤ V 2 , min n ∈ Z 3 ∖ { 0 } ∣ g n ∣ ≥ a } \left\{g\mathbb{Z}^3 : g \in\mathrm{GL}_3(\mathbb{R}),\ V_1 \le\lvert\det g\rvert\le V_2,\ \min_{n\in\mathbb{Z}^3\setminus\{0\}}\lvert gn\rvert\ge a\right\} { g Z 3 : g ∈ GL 3 ( R ) , V 1 ≤ ∣ det g ∣ ≤ V 2 , n ∈ Z 3 ∖ { 0 } min ∣ g n ∣ ≥ a }
is compact in G L 3 ( R ) / G L 3 ( Z ) \mathrm{GL}_3(\mathbb{R})/\mathrm{GL}_3(\mathbb{Z}) GL 3 ( R ) / GL 3 ( Z ) . Equivalently, every sequence g j Z 3 g_j\mathbb{Z}^3 g j Z 3 in this set admits
j k ↑ ∞ , A k ∈ G L 3 ( Z ) , g j k A k ⟶ g ∈ G L 3 ( R ) . j_k \uparrow\infty,\qquad A_k \in\mathrm{GL}_3(\mathbb{Z}),\qquad g_{j_k}A_k \longrightarrow g \in\mathrm{GL}_3(\mathbb{R}). j k ↑ ∞ , A k ∈ GL 3 ( Z ) , g j k A k ⟶ g ∈ GL 3 ( R ) .
By (17) , (16) and this criterion, m ( V ) m(V) m ( V ) is attained. The chain and planar minima are
e 1 = − ζ ( 6 ) 2 ζ ( 12 ) , e 2 = E ( T A ∗ ) = − S 3 2 2 S 6 ; e_1=-\frac{\zeta(6)^2}{\zeta(12)},\qquad e_2=\mathcal{E}(T_{A_*})=-\frac{S_3^2}{2S_6}; e 1 = − ζ ( 12 ) ζ ( 6 ) 2 , e 2 = E ( T A ∗ ) = − 2 S 6 S 3 2 ;
the second identity and planar uniqueness follow from [4 ] . The following fixed-systole comparison extends the prescribed-bond viewpoint of [3 ] and excludes large planar areas.
Theorem 2.2 (A convex fixed-systole comparison). Fix a > 0 a > 0 a > 0 . Suppose that f : [ a 2 , ∞ ) ⟶ ( 0 , ∞ ) f : [a^2,\infty) \longrightarrow(0,\infty) f : [ a 2 , ∞ ) ⟶ ( 0 , ∞ ) is continuous, strictly decreasing and strictly convex, and that for some s > 3 / 2 s > 3/2 s > 3/2 and C < ∞ C < \infty C < ∞ ,
f ( q ) ≤ C q − s ( q ≥ a 2 ) . f(q) \le Cq^{-s}\qquad(q \ge a^2). f ( q ) ≤ C q − s ( q ≥ a 2 ) . Let F 1 = 2 − 1 / 2 D 3 F_1=2^{-1/2}D_3 F 1 = 2 − 1/2 D 3 denote the FCC lattice of shortest length one. Every three-dimensional lattice with λ 1 ( L ) ≥ a \lambda_1(L) \ge a λ 1 ( L ) ≥ a satisfies
∑ v ∈ L ∖ { 0 } f ( ∣ v ∣ 2 ) ≤ ∑ v ∈ a F 1 ∖ { 0 } f ( ∣ v ∣ 2 ) , \sum_{v\in L\setminus\{0\}}f(\lvert v\rvert^2)\le\sum_{v\in aF_1\setminus\{0\}}f(\lvert v\rvert^2), v ∈ L ∖ { 0 } ∑ f (∣ v ∣ 2 ) ≤ v ∈ a F 1 ∖ { 0 } ∑ f (∣ v ∣ 2 ) , with equality if and only if L L L is an orthogonal image of a F 1 aF_1 a F 1 .
Proof. For a discrete additive subgroup L L L with separation at least a a a , disjoint balls of radius a / 2 a/2 a /2 give, for R > a / 2 R>a/2 R > a /2 ,
∑ v ∈ L ∣ v ∣ > R ∣ v ∣ − 2 s ≤ T s ( a , R ) : = 24 a 3 ( 2 s − 3 ) ( 1 + a 2 R ) 2 s ( R − a / 2 ) 3 − 2 s . (18) \sum_{\substack{v\in L\\ \lvert v\rvert>R}}\lvert v\rvert^{-2s}\le T_s(a,R):=\frac{24}{a^3(2s-3)}\left(1+\frac{a}{2R}\right)^{2s}(R-a/2)^{3-2s}. \tag*{(18)} v ∈ L ∣ v ∣ > R ∑ ∣ v ∣ − 2 s ≤ T s ( a , R ) := a 3 ( 2 s − 3 ) 24 ( 1 + 2 R a ) 2 s ( R − a /2 ) 3 − 2 s . (18)
Let M k M_k M k be the supremum over separated additive subgroups of rank ≤ k \le k ≤ k , 0 ≤ k ≤ 3 0 \le k \le3 0 ≤ k ≤ 3 , with M 0 = 0 M_0=0 M 0 = 0 . Point counts give local sequential compactness; separation, the group operations and vanishing ( k + 1 ) (k+1) ( k + 1 ) -fold exterior products are closed. The tail (18) gives continuity and attainment. Adjoining an orthogonal period of length a a a to an M k − 1 M_{k-1} M k − 1 maximizer preserves separation and adds positive terms, so
M k > M k − 1 , k = 1 , 2 , 3 , M_k>M_{k-1},\qquad k=1,2,3, M k > M k − 1 , k = 1 , 2 , 3 ,
and every maximizer has rank k k k .
Scaling by a / λ 1 ( L ) < 1 a/\lambda_1(L)<1 a / λ 1 ( L ) < 1 strictly increases the sum. Thus λ 1 ( L ) = a \lambda_1(L)=a λ 1 ( L ) = a , and its Gram matrix Q Q Q satisfies
n T Q n ≥ a 2 ( n ∈ Z 3 ∖ { 0 } ) , span { n n T : n T Q n = a 2 } = Sym 3 . n^TQn\ge a^2\qquad(n\in\mathbb{Z}^3\setminus\{0\}),\qquad\operatorname{span}\{nn^T:n^TQn=a^2\}=\operatorname{Sym}_3. n T Q n ≥ a 2 ( n ∈ Z 3 ∖ { 0 }) , span { n n T : n T Q n = a 2 } = Sym 3 .
Otherwise choose a nonzero symmetric D D D orthogonal to the span and
ε ∥ Q − 1 / 2 D Q − 1 / 2 ∥ o p < 1 2 , 0 < ε < min a 2 < n T Q n < 2 a 2 n T D n ≠ 0 n T Q n − a 2 ∣ n T D n ∣ , \varepsilon\lVert Q^{-1/2}DQ^{-1/2}\rVert_{\mathrm{op}}<\frac12,\qquad
0<\varepsilon<
\min_{\substack{a^2<n^TQn<2a^2\\ n^TDn\ne0}}
\frac{n^TQn-a^2}{\lvert n^TDn\rvert}, ε ∥ Q − 1/2 D Q − 1/2 ∥ op < 2 1 , 0 < ε < a 2 < n T Q n < 2 a 2 n T D n = 0 min ∣ n T D n ∣ n T Q n − a 2 ,
where the minimum of an empty set is + ∞ +\infty + ∞ . The second inequality handles n T Q n < 2 a 2 n^TQn<2a^2 n T Q n < 2 a 2 ; otherwise,
n T Q n ≥ 2 a 2 ⟹ n T ( Q ± ε D ) n ≥ ( 1 − ε ∥ Q − 1 / 2 D Q − 1 / 2 ∥ o p ) n T Q n > a 2 . n^TQn\ge2a^2\Longrightarrow n^T(Q\pm\varepsilon D)n\ge\left(1-\varepsilon\lVert Q^{-1/2}DQ^{-1/2}\rVert_{\mathrm{op}}\right)n^TQn>a^2. n T Q n ≥ 2 a 2 ⟹ n T ( Q ± ε D ) n ≥ ( 1 − ε ∥ Q − 1/2 D Q − 1/2 ∥ op ) n T Q n > a 2 .
Thus Q ± ε D ≻ 0 Q\pm\varepsilon D\succ0 Q ± ε D ≻ 0 preserve all constraints, with the active ones fixed.
Since D ≠ 0 D \ne0 D = 0 , some coordinate vector or pairwise sum satisfies n T D n ≠ 0 n^{T}Dn \ne0 n T D n = 0 . Strict convexity gives
∑ n ∈ Z 3 ∖ { 0 } f ( n T Q n ) < 1 2 ∑ n ∈ Z 3 ∖ { 0 } f ( n T ( Q + ϵ D ) n ) + 1 2 ∑ n ∈ Z 3 ∖ { 0 } f ( n T ( Q − ϵ D ) n ) . \sum_{n\in\mathbb{Z}^{3}\setminus\{0\}} f(n^{T}Qn)
<
\frac{1}{2}\sum_{n\in\mathbb{Z}^{3}\setminus\{0\}} f(n^{T}(Q+\epsilon D)n)
+
\frac{1}{2}\sum_{n\in\mathbb{Z}^{3}\setminus\{0\}} f(n^{T}(Q-\epsilon D)n). n ∈ Z 3 ∖ { 0 } ∑ f ( n T Q n ) < 2 1 n ∈ Z 3 ∖ { 0 } ∑ f ( n T ( Q + ϵD ) n ) + 2 1 n ∈ Z 3 ∖ { 0 } ∑ f ( n T ( Q − ϵD ) n ) . Convergence follows from the tail hypothesis. One strict term contradicts maximality, so at least six shortest-vector pairs occur.
For a superbase v 0 + v 1 + v 2 + v 3 = 0 v_{0}+v_{1}+v_{2}+v_{3}=0 v 0 + v 1 + v 2 + v 3 = 0 , every three being a basis, and { i , j , k , l } = { 0 , 1 , 2 , 3 } \{i,j,k,l\}=\{0,1,2,3\} { i , j , k , l } = { 0 , 1 , 2 , 3 } with v i ⋅ v j > 0 v_{i}\cdot v_{j}>0 v i ⋅ v j > 0 , Selling’s superbase move satisfies
( v ~ i , v ~ j , v ~ k , v ~ l ) = ( − v i , v j , v k + v i , v l + v i ) , ∑ m = 0 3 ∣ v ~ m ∣ 2 = ∑ m = 0 3 ∣ v m ∣ 2 − 2 v i ⋅ v j < ∑ m = 0 3 ∣ v m ∣ 2 . (\widetilde{v}_{i},\widetilde{v}_{j},\widetilde{v}_{k},\widetilde{v}_{l})=(-v_{i},v_{j},v_{k}+v_{i},v_{l}+v_{i}),\qquad
\sum_{m=0}^{3}|\widetilde{v}_{m}|^{2}
=
\sum_{m=0}^{3}|v_{m}|^{2}-2v_{i}\cdot v_{j}
<
\sum_{m=0}^{3}|v_{m}|^{2}. ( v i , v j , v k , v l ) = ( − v i , v j , v k + v i , v l + v i ) , m = 0 ∑ 3 ∣ v m ∣ 2 = m = 0 ∑ 3 ∣ v m ∣ 2 − 2 v i ⋅ v j < m = 0 ∑ 3 ∣ v m ∣ 2 . Finiteness under the squared-length bound gives termination at v i ⋅ v j ≤ 0 v_{i}\cdot v_{j}\le0 v i ⋅ v j ≤ 0 . Scale a a a to one and put p i j = − v i ⋅ v j ≥ 0 p_{ij}=-v_{i}\cdot v_{j}\ge0 p ij = − v i ⋅ v j ≥ 0 . The seven cut weights are
p 01 + p 02 + p 03 , p 01 + p 12 + p 13 , p_{01}+p_{02}+p_{03},\qquad p_{01}+p_{12}+p_{13}, p 01 + p 02 + p 03 , p 01 + p 12 + p 13 , p 02 + p 12 + p 23 , p 03 + p 13 + p 23 , p_{02}+p_{12}+p_{23},\qquad p_{03}+p_{13}+p_{23}, p 02 + p 12 + p 23 , p 03 + p 13 + p 23 , p 02 + p 03 + p 12 + p 13 , p 01 + p 03 + p 12 + p 23 , p 01 + p 02 + p 13 + p 23 . p_{02}+p_{03}+p_{12}+p_{13},\qquad p_{01}+p_{03}+p_{12}+p_{23},\qquad p_{01}+p_{02}+p_{13}+p_{23}. p 02 + p 03 + p 12 + p 13 , p 01 + p 03 + p 12 + p 23 , p 01 + p 02 + p 13 + p 23 . Every cut weight is at least one. Integer level sets give
∣ ∑ i = 0 3 n i v i ∣ 2 = ∑ 0 ≤ i < j ≤ 3 p i j ( n i − n j ) 2 ≥ max k = min 0 ≤ i ≤ 3 n i + 1 max 0 ≤ i ≤ 3 n i ∑ 0 ≤ i , j ≤ 3 n i ≥ k > n j p i j ≥ max 0 ≤ i ≤ 3 n i − min 0 ≤ i ≤ 3 n i . \left|\sum_{i=0}^{3}n_{i}v_{i}\right|^{2}
=
\sum_{0\le i<j\le3}p_{ij}(n_{i}-n_{j})^{2}
\ge
\max_{k=\min_{0\le i\le3}n_{i}+1}^{\max_{0\le i\le3}n_{i}}
\sum_{\substack{0\le i,j\le3\\ n_{i}\ge k>n_{j}}}p_{ij}
\ge
\max_{0\le i\le3}n_{i}-\min_{0\le i\le3}n_{i}. i = 0 ∑ 3 n i v i 2 = 0 ≤ i < j ≤ 3 ∑ p ij ( n i − n j ) 2 ≥ k = m i n 0 ≤ i ≤ 3 n i + 1 max m a x 0 ≤ i ≤ 3 n i 0 ≤ i , j ≤ 3 n i ≥ k > n j ∑ p ij ≥ 0 ≤ i ≤ 3 max n i − 0 ≤ i ≤ 3 min n i . Thus shortest vectors are binary cuts, with at least six weights equal to one. The first four and last three weights both sum to 2 ∑ 0 ≤ i < j ≤ 3 p i j 2\sum_{0\le i<j\le3}p_{ij} 2 ∑ 0 ≤ i < j ≤ 3 p ij ; hence not all seven equal one. If the missing equality were a singleton, its weight would be zero: the other three singleton weights and the last three weights both sum to three.
Relabeling the missing cut as the last one, the other six equations give
p 01 = p 02 = p 13 = p 23 = 1 2 , p 03 = p 12 = 0. p_{01}=p_{02}=p_{13}=p_{23}=\frac{1}{2},\qquad p_{03}=p_{12}=0. p 01 = p 02 = p 13 = p 23 = 2 1 , p 03 = p 12 = 0. In the basis ( v 0 , v 3 , v 1 ) (v_{0},v_{3},v_{1}) ( v 0 , v 3 , v 1 ) , the Gram matrix is
( 1 0 − 1 / 2 0 1 − 1 / 2 − 1 / 2 − 1 / 2 1 ) . \begin{pmatrix}
1 & 0 & -1/2\\
0 & 1 & -1/2\\
-1/2 & -1/2 & 1
\end{pmatrix}. 1 0 − 1/2 0 1 − 1/2 − 1/2 − 1/2 1 . the shortest-length-one FCC Gram matrix. Thus L ≅ a F 1 L\cong aF_{1} L ≅ a F 1 . This completes the proof of Theorem 2.2. □ \square □
Theorem 2.3 (Fixed-systole consequences). For λ 1 ( L ) ≥ a > 0 \lambda_{1}(L)\ge a>0 λ 1 ( L ) ≥ a > 0 ,
E ( L , s ) ≤ a − 2 s E ( F 1 , s ) ( s > 3 / 2 ) , E ( L ) ≥ E ( F 1 , 6 ) 2 a 12 − E ( F 1 , 3 ) a 6 ( a 6 ≥ 7 / 4 ) . E(L,s)\le a^{-2s}E(F_{1},s)\quad(s>3/2),\qquad
\mathcal{E}(L)\ge\frac{E(F_{1},6)}{2a^{12}}-\frac{E(F_{1},3)}{a^{6}}\quad(a^{6}\ge7/4). E ( L , s ) ≤ a − 2 s E ( F 1 , s ) ( s > 3/2 ) , E ( L ) ≥ 2 a 12 E ( F 1 , 6 ) − a 6 E ( F 1 , 3 ) ( a 6 ≥ 7/4 ) . In either inequality equality holds exactly for L ≅ a F 1 L\cong aF_{1} L ≅ a F 1 .
Proof. Apply Theorem 2.2 to q − s q^{-s} q − s and f ( q ) = 2 q − 3 − q − 6 f(q)=2q^{-3}-q^{-6} f ( q ) = 2 q − 3 − q − 6 , respectively. On q ≥ a 2 q\ge a^{2} q ≥ a 2 , a 6 ≥ 7 / 4 a^{6}\ge7/4 a 6 ≥ 7/4 ,
0 < f ( q ) ≤ 2 q − 3 , f ′ ( q ) = 6 q − 7 ( 1 − q 3 ) < 0 , f ′ ′ ( q ) = 6 q − 8 ( 4 q 3 − 7 ) ≥ 0. 0<f(q)\le2q^{-3},\qquad f'(q)=6q^{-7}(1-q^{3})<0,\qquad f''(q)=6q^{-8}(4q^{3}-7)\ge0. 0 < f ( q ) ≤ 2 q − 3 , f ′ ( q ) = 6 q − 7 ( 1 − q 3 ) < 0 , f ′′ ( q ) = 6 q − 8 ( 4 q 3 − 7 ) ≥ 0.
Thus f f f is strictly convex also at a 6 = 7 / 4 a^{6}=7/4 a 6 = 7/4 , and
E ( L ) = − 1 2 ∑ v ∈ L ∖ { 0 } f ( ∣ v ∣ 2 ) \mathcal{E}(L)=-\frac{1}{2}\sum_{v\in L\setminus\{0\}}f(|v|^{2}) E ( L ) = − 2 1 v ∈ L ∖ { 0 } ∑ f ( ∣ v ∣ 2 )
gives the second comparison and equality case. This completes the proof of Theorem 2.3. □ \square □
Lemma 2.4 (An explicit fixed-systole Riesz constant). For every three-dimensional lattice L L L ,
E ( L , 3 ) < 2901 200 λ 1 ( L ) − 6 , E ( L ) > λ 1 ( L ) − 12 − 2901 200 λ 1 ( L ) − 6 . E(L,3)<\frac{2901}{200}\lambda_1(L)^{-6},\qquad\mathcal{E}(L)>\lambda_1(L)^{-12}-\frac{2901}{200}\lambda_1(L)^{-6}. E ( L , 3 ) < 200 2901 λ 1 ( L ) − 6 , E ( L ) > λ 1 ( L ) − 12 − 200 2901 λ 1 ( L ) − 6 . Proof. Scale to shortest length one and apply Theorem 2.3 . With 2 − 1 / 6 F = 2 − 1 / 2 D 3 2^{-1/6}F=2^{-1/2}D_3 2 − 1/6 F = 2 − 1/2 D 3 , rational summation gives
∑ ( i , j , k ) ∈ { − 11 , … , 11 } 3 i + j + k ∈ 2 Z 0 < i 2 + j 2 + k 2 ≤ 128 8 ( i 2 + j 2 + k 2 ) 3 < 14443 1000 . \sum_{\substack{(i,j,k)\in\{-11,\ldots,11\}^3\\ i+j+k\in2\mathbb{Z}\\ 0<i^2+j^2+k^2\le128}}\frac{8}{(i^2+j^2+k^2)^3}<\frac{14443}{1000}. ( i , j , k ) ∈ { − 11 , … , 11 } 3 i + j + k ∈ 2 Z 0 < i 2 + j 2 + k 2 ≤ 128 ∑ ( i 2 + j 2 + k 2 ) 3 8 < 1000 14443 . There are 3054 retained points; the others have radius > 8 >8 > 8 . By (18) with ( a , s , R ) = ( 1 , 3 , 8 ) (a,s,R)=(1,3,8) ( a , s , R ) = ( 1 , 3 , 8 ) ,
E ( 2 − 1 / 6 F , 3 ) < 14443 1000 + T 3 ( 1 , 8 ) < 2901 200 . E(2^{-1/6}F,3)<\frac{14443}{1000}+T_3(1,8)<\frac{2901}{200}. E ( 2 − 1/6 F , 3 ) < 1000 14443 + T 3 ( 1 , 8 ) < 200 2901 . Thus E ( L , 6 ) / 2 ≥ λ 1 ( L ) − 12 E(L,6)/2\ge\lambda_1(L)^{-12} E ( L , 6 ) /2 ≥ λ 1 ( L ) − 12 . This completes the proof of Lemma 2.4. □
Lemma 2.5 (A bounded primitive planar section). Let λ 2 ( L ) \lambda_2(L) λ 2 ( L ) be the least radius containing two independent lattice vectors. Every rank-three lattice with E ( L ) ≤ e 2 \mathcal{E}(L)\le e_2 E ( L ) ≤ e 2 has a primitive rank-two section Γ \Gamma Γ satisfying
7 10 < λ 1 ( L ) < 8 5 , λ 2 ( L ) ≤ 25 8 , ∣ Γ ∣ < 5. \frac{7}{10}<\lambda_1(L)<\frac{8}{5},\qquad\lambda_2(L)\le\frac{25}{8},\qquad|\Gamma|<5. 10 7 < λ 1 ( L ) < 5 8 , λ 2 ( L ) ≤ 8 25 , ∣Γ∣ < 5. Writing L = Γ + Z ( u + h e 3 ) L=\Gamma+\mathbb{Z}(u+he_3) L = Γ + Z ( u + h e 3 ) , one has h > V / 5 h>V/5 h > V /5 . Proof.
e 2 < ∑ ( i , j ) ∈ Z 2 0 < i 2 + i j + j 2 ≤ 4 ( 1 2 ( i 2 + i j + j 2 ) 6 − 1 ( i 2 + i j + j 2 ) 3 ) < − 331 100 . e_2<\sum_{\substack{(i,j)\in\mathbb{Z}^2\\0<i^2+ij+j^2\le4}}\left(\frac{1}{2(i^2+ij+j^2)^6}-\frac{1}{(i^2+ij+j^2)^3}\right)<-\frac{331}{100}. e 2 < ( i , j ) ∈ Z 2 0 < i 2 + ij + j 2 ≤ 4 ∑ ( 2 ( i 2 + ij + j 2 ) 6 1 − ( i 2 + ij + j 2 ) 3 1 ) < − 100 331 . Lemma 2.1 gives λ 1 > 7 / 10 \lambda_1>7/10 λ 1 > 7/10 . If λ 1 ≥ 8 / 5 \lambda_1\ge8/5 λ 1 ≥ 8/5 , then
E ( L ) ≥ − 206 5 λ 1 − 6 > − 331 100 , \mathcal{E}(L)\ge-\frac{206}{5}\lambda_1^{-6}>-\frac{331}{100}, E ( L ) ≥ − 5 206 λ 1 − 6 > − 100 331 , a contradiction. If λ 2 > 25 / 8 \lambda_2>25/8 λ 2 > 25/8 , the radius-25 / 8 25/8 25/8 points lie on one primitive chain. Its partial energy is at least the full energy, which is ≥ e 1 \ge e_1 ≥ e 1 : omitted chain distances exceed one and their energies are negative. Disjoint balls of radius 7 / 20 7/20 7/20 give
∑ v ∈ L ∣ v ∣ > 25 / 8 ∣ v ∣ − 6 ≤ 8000 343 ( 139 125 ) 6 ( 111 40 ) − 3 . \sum_{\substack{v\in L\\ |v|>25/8}}|v|^{-6}\le\frac{8000}{343}\left(\frac{139}{125}\right)^6\left(\frac{111}{40}\right)^{-3}. v ∈ L ∣ v ∣ > 25/8 ∑ ∣ v ∣ − 6 ≤ 343 8000 ( 125 139 ) 6 ( 40 111 ) − 3 . Together with e 1 > − 21 / 20 e_1>-21/20 e 1 > − 21/20 , obtained from ζ ( 6 ) < 327 / 320 \zeta(6)<327/320 ζ ( 6 ) < 327/320 , this gives
E ( L ) ≥ e 1 − ∑ v ∈ L ∣ v ∣ > 25 / 8 ∣ v ∣ − 6 > − 331 100 > e 2 , \mathcal{E}(L)\ge e_1-\sum_{\substack{v\in L\\|v|>25/8}}|v|^{-6}>-\frac{331}{100}>e_2, E ( L ) ≥ e 1 − v ∈ L ∣ v ∣ > 25/8 ∑ ∣ v ∣ − 6 > − 100 331 > e 2 , again a contradiction.
Gauss reduction of Γ = L ∩ span R { v 1 , v 2 } \Gamma=L\cap\operatorname{span}_{\mathbb{R}}\{v_1,v_2\} Γ = L ∩ span R { v 1 , v 2 } for shortest independent v 1 , v 2 v_1,v_2 v 1 , v 2 gives basis lengths λ 1 \lambda_1 λ 1 and ≤ λ 2 \le\lambda_2 ≤ λ 2 . Hence
∣ Γ ∣ ≤ λ 1 λ 2 < 5 , h = V / ∣ Γ ∣ > V / 5. |\Gamma|\le\lambda_1\lambda_2<5,\qquad h=V/|\Gamma|>V/5. ∣Γ∣ ≤ λ 1 λ 2 < 5 , h = V /∣Γ∣ > V /5. The saturated section Γ \Gamma Γ is primitive and extends to a basis. This completes the proof of Lemma 2.5. □
Lemma 2.6 (The stationary branches). For 21 / 20 ≤ V ≤ 11 / 10 21/20 \le V \le11/10 21/20 ≤ V ≤ 11/10 , equations (10) and (11) have unique real analytic solutions X ( V ) , Y ( V ) X(V),Y(V) X ( V ) , Y ( V ) in the stated intervals, with X ′ ( V ) < 0 < Y ′ ( V ) X'(V)<0<Y'(V) X ′ ( V ) < 0 < Y ′ ( V ) .
Let
D T F ( V ) = 2 V 4 [ E ( V 1 / 3 L X ( V ) ) − E ( V 1 / 3 F ) ] . \mathcal{D}_{T_F}(V)=2V^4\left[\mathcal{E}(V^{1/3}L_X(V))-\mathcal{E}(V^{1/3}F)\right]. D T F ( V ) = 2 V 4 [ E ( V 1/3 L X ( V )) − E ( V 1/3 F ) ] . D T H ( V ) = 2 V 4 [ E ( V 1 / 3 L X ( V ) ) − E ( V 1 / 3 H Y ( V ) ) ] . \mathcal{D}_{T_H}(V)=2V^4\left[\mathcal{E}(V^{1/3}L_X(V))-\mathcal{E}(V^{1/3}H_Y(V))\right]. D T H ( V ) = 2 V 4 [ E ( V 1/3 L X ( V )) − E ( V 1/3 H Y ( V )) ] . Then
D T F ′ ( V ) < − 16 25 V , D T H ′ ( V ) > 1 5 V . \mathcal{D}'_{T_F}(V)<-\frac{16}{25}V,\qquad\mathcal{D}'_{T_H}(V)>\frac{1}{5}V. D T F ′ ( V ) < − 25 16 V , D T H ′ ( V ) > 5 1 V . Proof. For s = 3 , 6 s=3,6 s = 3 , 6 , termwise differentiation gives
d d x E ( L x , s ) = s ( 4 x ) s / 3 3 x ∑ n ∈ D 3 ∖ { 0 } n 2 2 + n 3 2 − 2 x n 1 2 ( x n 1 2 + n 2 2 + n 3 2 ) s + 1 . \frac{\mathrm{d}}{\mathrm{d}x}\mathcal{E}(L_x,s)=\frac{s(4x)^{s/3}}{3x}\sum_{n\in D_3\setminus\{0\}}\frac{n_2^2+n_3^2-2xn_1^2}{(xn_1^2+n_2^2+n_3^2)^{s+1}}. d x d E ( L x , s ) = 3 x s ( 4 x ) s /3 n ∈ D 3 ∖ { 0 } ∑ ( x n 1 2 + n 2 2 + n 3 2 ) s + 1 n 2 2 + n 3 2 − 2 x n 1 2 . d d y E ( H y , s ) = s ( 3 y / 4 ) s / 3 3 y ∑ ( i , j , k ) ∈ Z 3 ∖ { 0 } i 2 + i j + j 2 − 2 y k 2 ( i 2 + i j + j 2 + y k 2 ) s + 1 . \frac{\mathrm{d}}{\mathrm{d}y}\mathcal{E}(H_y,s)=\frac{s(3y/4)^{s/3}}{3y}\sum_{(i,j,k)\in\mathbb{Z}^3\setminus\{0\}}\frac{i^2+ij+j^2-2yk^2}{(i^2+ij+j^2+yk^2)^{s+1}}. d y d E ( H y , s ) = 3 y s ( 3 y /4 ) s /3 ( i , j , k ) ∈ Z 3 ∖ { 0 } ∑ ( i 2 + ij + j 2 + y k 2 ) s + 1 i 2 + ij + j 2 − 2 y k 2 . The series and their fixed-order derivatives converge uniformly on the stated compact positive intervals.
Numerical Lemma A.8 gives
∂ log X E ( L X , 3 ) < 0 , ∂ log X ( ∂ log X E ( L X , 6 ) ∂ log X E ( L X , 3 ) ) < 0. \partial_{\log X}\mathcal{E}(L_X,3)<0,\qquad
\partial_{\log X}\left(\frac{\partial_{\log X}\mathcal{E}(L_X,6)}{\partial_{\log X}\mathcal{E}(L_X,3)}\right)<0. ∂ l o g X E ( L X , 3 ) < 0 , ∂ l o g X ( ∂ l o g X E ( L X , 3 ) ∂ l o g X E ( L X , 6 ) ) < 0. ∂ log Y E ( H Y , 3 ) > 0 , ∂ log Y ( ∂ log Y E ( H Y , 6 ) ∂ log Y E ( H Y , 3 ) ) > 0. \partial_{\log Y}\mathcal{E}(H_Y,3)>0,\qquad
\partial_{\log Y}\left(\frac{\partial_{\log Y}\mathcal{E}(H_Y,6)}{\partial_{\log Y}\mathcal{E}(H_Y,3)}\right)>0. ∂ l o g Y E ( H Y , 3 ) > 0 , ∂ l o g Y ( ∂ l o g Y E ( H Y , 3 ) ∂ l o g Y E ( H Y , 6 ) ) > 0. The endpoint ratios bracket 2 V 2 2V^2 2 V 2 . At either root, with L ( t ) = L e t L(t)=L e^t L ( t ) = L e t or H e t H e^t H e t ,
∂ t 2 { E ( L ( t ) , 6 ) − 2 V 2 E ( L ( t ) , 3 ) } = ∂ t E ( L ( t ) , 3 ) ∂ t ( ∂ t E ( L ( t ) , 6 ) ∂ t E ( L ( t ) , 3 ) ) > 0. \partial_t^2\left\{\mathcal{E}(L(t),6)-2V^2\mathcal{E}(L(t),3)\right\}
=\partial_t\mathcal{E}(L(t),3)\,\partial_t\left(\frac{\partial_t\mathcal{E}(L(t),6)}{\partial_t\mathcal{E}(L(t),3)}\right)>0. ∂ t 2 { E ( L ( t ) , 6 ) − 2 V 2 E ( L ( t ) , 3 ) } = ∂ t E ( L ( t ) , 3 ) ∂ t ( ∂ t E ( L ( t ) , 3 ) ∂ t E ( L ( t ) , 6 ) ) > 0. Implicit differentiation gives analyticity and X ′ ( V ) < 0 < Y ′ ( V ) X'(V)<0<Y'(V) X ′ ( V ) < 0 < Y ′ ( V ) .
Stationarity and homogeneity give
1 4 V ( D T F ′ ( V ) D T H ′ ( V ) ) = ( E ( F , 3 ) − E ( L X ( V ) , 3 ) E ( H Y ( V ) , 3 ) − E ( L X ( V ) , 3 ) ) . \frac{1}{4V}\begin{pmatrix}\mathcal{D}'_{T_F}(V)\\ \mathcal{D}'_{T_H}(V)\end{pmatrix}
=
\begin{pmatrix}
\mathcal{E}(F,3)-\mathcal{E}(L_X(V),3)\\
\mathcal{E}(H_Y(V),3)-\mathcal{E}(L_X(V),3)
\end{pmatrix}. 4 V 1 ( D T F ′ ( V ) D T H ′ ( V ) ) = ( E ( F , 3 ) − E ( L X ( V ) , 3 ) E ( H Y ( V ) , 3 ) − E ( L X ( V ) , 3 ) ) . The certified moment bounds are
E ( F , 3 ) < 7.226961 < 7.39056 < E ( L X ( V ) , 3 ) < 7.88582 < 7.94245 < E ( H Y ( V ) , 3 ) . \mathcal{E}(F,3)<7.226961<7.39056<\mathcal{E}(L_X(V),3)<7.88582<7.94245<\mathcal{E}(H_Y(V),3). E ( F , 3 ) < 7.226961 < 7.39056 < E ( L X ( V ) , 3 ) < 7.88582 < 7.94245 < E ( H Y ( V ) , 3 ) . These gaps prove the derivative bounds. This completes the proof of Lemma 2.6. □
Proposition 2.7 (The complete comparison within two lattice families). For 21 / 20 ≤ V ≤ 11 / 10 21/20 \le V \le11/10 21/20 ≤ V ≤ 11/10 , the branch equations (10) –(11) have unique real analytic solutions, and the contact systems (7) and (8) have unique solutions in the stated rectangles:
23 100 < X ( V ) < 307 1000 , 103 100 < Y ( V ) < 103 50 , \frac{23}{100}<X(V)<\frac{307}{1000},\qquad
\frac{103}{100}<Y(V)<\frac{103}{50}, 100 23 < X ( V ) < 1000 307 , 100 103 < Y ( V ) < 50 103 , ( v , x ) = ( V F T , X ( V F T ) ) , ( v , x , y ) = ( V T H , X ( V T H ) , Y ( V T H ) ) , (v,x)=(V_{\mathrm{FT}},X(V_{\mathrm{FT}})),\qquad
(v,x,y)=(V_{\mathrm{TH}},X(V_{\mathrm{TH}}),Y(V_{\mathrm{TH}})), ( v , x ) = ( V FT , X ( V FT )) , ( v , x , y ) = ( V TH , X ( V TH ) , Y ( V TH )) , 21 20 < V F T < V T H < 11 10 . \frac{21}{20}<V_{\mathrm{FT}}<V_{\mathrm{TH}}<\frac{11}{10}. 20 21 < V FT < V TH < 10 11 .
Put m ~ ( V ) = min X , Y > 0 { E ( V 1 / 3 L X ) , E ( V 1 / 3 H Y ) } \widetilde m(V)=\min_{X,Y>0}\{\mathcal{E}(V^{1/3}L_X),\mathcal{E}(V^{1/3}H_Y)\} m ( V ) = min X , Y > 0 { E ( V 1/3 L X ) , E ( V 1/3 H Y )} . Up to orthogonal equivalence,
argmin Λ = L X , X > 0 or Λ = H Y , Y > 0 E ( V 1 / 3 Λ ) = { { F } , V < V F T , { F , L X ( V ) } , V = V F T , { L X ( V ) } , V F T < V < V T H , { L X ( V ) , H Y ( V ) } , V = V T H , { H Y ( V ) } , V > V T H , \underset{\substack{\Lambda=L_X,\ X>0\\ \text{or }\Lambda=H_Y,\ Y>0}}{\operatorname{argmin}}\ \mathcal{E}(V^{1/3}\Lambda)=
\begin{cases}
\{F\}, & V<V_{\mathrm{FT}},\\
\{F,L_X(V)\}, & V=V_{\mathrm{FT}},\\
\{L_X(V)\}, & V_{\mathrm{FT}}<V<V_{\mathrm{TH}},\\
\{L_X(V),H_Y(V)\}, & V=V_{\mathrm{TH}},\\
\{H_Y(V)\}, & V>V_{\mathrm{TH}},
\end{cases} Λ = L X , X > 0 or Λ = H Y , Y > 0 argmin E ( V 1/3 Λ ) = ⎩ ⎨ ⎧ { F } , { F , L X ( V )} , { L X ( V )} , { L X ( V ) , H Y ( V )} , { H Y ( V )} , V < V FT , V = V FT , V FT < V < V TH , V = V TH , V > V TH ,
and
m ~ + ′ ( V ) − m ~ − ′ ( V ) < 0 ( V ∈ { V F T , V T H } ) . \widetilde m_+'(V)-\widetilde m_-'(V)<0\qquad(V\in\{V_{\mathrm{FT}},V_{\mathrm{TH}}\}). m + ′ ( V ) − m − ′ ( V ) < 0 ( V ∈ { V FT , V TH }) .
Proof. Lemma 2.6 gives the two analytic branches. For the remaining parameters, use
E ( V 1 / 3 F ) < 31 20 ( 20 21 ) 4 − 36 5 ( 10 11 ) 2 < − 4 , \mathcal{E}(V^{1/3}F)<\frac{31}{20}\left(\frac{20}{21}\right)^4-\frac{36}{5}\left(\frac{10}{11}\right)^2<-4, E ( V 1/3 F ) < 20 31 ( 21 20 ) 4 − 5 36 ( 11 10 ) 2 < − 4 ,
E ( V 1 / 3 H 1 ) < 5 2 ( 20 21 ) 4 − 15 2 ( 10 11 ) 2 < − 4. \mathcal{E}(V^{1/3}H_1)<\frac{5}{2}\left(\frac{20}{21}\right)^4-\frac{15}{2}\left(\frac{10}{11}\right)^2<-4. E ( V 1/3 H 1 ) < 2 5 ( 21 20 ) 4 − 2 15 ( 11 10 ) 2 < − 4.
At a family minimum, Lemma 2.1 gives λ 1 > 7 / 10 \lambda_1>7/10 λ 1 > 7/10 , whereas
λ 1 ( V 1 / 3 L X ) 2 = V 2 / 3 ( 4 X ) − 1 / 3 min ( 4 X , X + 1 , 2 ) , \lambda_1(V^{1/3}L_X)^2=V^{2/3}(4X)^{-1/3}\min(4X,X+1,2), λ 1 ( V 1/3 L X ) 2 = V 2/3 ( 4 X ) − 1/3 min ( 4 X , X + 1 , 2 ) ,
λ 1 ( V 1 / 3 H Y ) 2 = V 2 / 3 ( 4 3 Y ) 1 / 3 min ( 1 , Y ) . \lambda_1(V^{1/3}H_Y)^2=V^{2/3}\left(\frac{4}{3Y}\right)^{1/3}\min(1,Y). λ 1 ( V 1/3 H Y ) 2 = V 2/3 ( 3 Y 4 ) 1/3 min ( 1 , Y ) .
Hence 77 / 1000 < X < 21 77/1000<X<21 77/1000 < X < 21 and 27 / 100 < Y < 14 27/100<Y<14 27/100 < Y < 14 . Lemma A.11 excludes every critical point outside the two branch windows and the two cubic windows 12 / 25 ≤ X ≤ 13 / 25 12/25\le X\le13/25 12/25 ≤ X ≤ 13/25 , 49 / 50 ≤ X ≤ 51 / 50 49/50\le X\le51/50 49/50 ≤ X ≤ 51/50 . Cubic symmetry fixes X = 1 / 2 , 1 X=1/2,1 X = 1/2 , 1 ; the strict second-derivative signs make BCC and FCC the unique stationary points in their windows, a maximum and a minimum, respectively. Thus argmin X > 0 E ( V 1 / 3 L X ) ⊂ { X ( V ) , 1 } \operatorname{argmin}_{X>0}\mathcal{E}(V^{1/3}L_X)\subset\{X(V),1\} argmin X > 0 E ( V 1/3 L X ) ⊂ { X ( V ) , 1 } and argmin Y > 0 E ( V 1 / 3 H Y ) = { Y ( V ) } \operatorname{argmin}_{Y>0}\mathcal{E}(V^{1/3}H_Y)=\{Y(V)\} argmin Y > 0 E ( V 1/3 H Y ) = { Y ( V )} .
In the stated rectangles,
\text{\hyperlink{eq:1.7}{(1.7)}}\Longleftrightarrow x=X(v),\quad D_{\mathrm{TF}}(v)=0,
\text{\hyperlink{eq:1.8}{(1.8)}}\Longleftrightarrow(x,y)=(X(v),Y(v)),\quad D_{\mathrm{TH}}(v)=0.
The opposite signs in Numerical Lemma A.9 and the derivative bounds in Lemma 2.6 give unique contacts in disjoint brackets and the phase ordering. At either contact V V V , stationarity and equality of energies give the derivative jump
− 2 V − 3 { E ( Λ + , 3 ) − E ( Λ − , 3 ) } < 0 , -2V^{-3}\{\mathcal{E}(\Lambda_+,3)-\mathcal{E}(\Lambda_-,3)\}<0, − 2 V − 3 { E ( Λ + , 3 ) − E ( Λ − , 3 )} < 0 ,
where Λ − \Lambda_- Λ − and Λ + \Lambda_+ Λ + are the adjacent unit-covolume shapes. This completes the proof of Proposition 2.7. □
Lemma 2.8 (Ewald identity and rational kernel bounds). For a positive-definite 3 × 3 3\times3 3 × 3 Gram matrix Q Q Q with det Q = V 2 \det Q=V^2 det Q = V 2 , write
E ( Q , s ) = ∑ n ∈ Z 3 ∖ { 0 } ( n T Q n ) − s ( s > 3 / 2 ) , E ( Q ) = 1 2 E ( Q , 6 ) − E ( Q , 3 ) . E(Q,s)=\sum_{n\in\mathbb{Z}^3\setminus\{0\}}(n^{\mathsf T}Qn)^{-s}\qquad(s>3/2),\qquad\mathcal{E}(Q)=\frac{1}{2}E(Q,6)-E(Q,3). E ( Q , s ) = n ∈ Z 3 ∖ { 0 } ∑ ( n T Q n ) − s ( s > 3/2 ) , E ( Q ) = 2 1 E ( Q , 6 ) − E ( Q , 3 ) . Set
G a ( x ) = ∫ 1 ∞ e − x t t a − 1 d t . \mathcal{G}_a(x)=\int_1^\infty e^{-xt}t^{a-1}\,\mathrm{d}t. G a ( x ) = ∫ 1 ∞ e − x t t a − 1 d t .
Splitting Poisson summation at t = 1 t=1 t = 1 gives, absolutely convergently,
E ( Q ) = π 6 240 ( − 1 6 + 2 9 V ) − π 3 2 ( − 1 3 + 2 3 V ) + ∑ n ∈ Z 3 ∖ { 0 } { π 6 240 G 6 ( π n T Q n ) − π 3 2 G 3 ( π n T Q n ) } + 1 V ∑ n ∈ Z 3 ∖ { 0 } { π 6 240 G − 9 / 2 ( π n T Q − 1 n ) − π 3 2 G − 3 / 2 ( π n T Q − 1 n ) } . \mathcal{E}(Q)=\frac{\pi^{6}}{240}\left(-\frac{1}{6}+\frac{2}{9V}\right)-\frac{\pi^{3}}{2}\left(-\frac{1}{3}+\frac{2}{3V}\right)
+\sum_{n\in\mathbb{Z}^{3}\setminus\{0\}}\left\{\frac{\pi^{6}}{240}\mathcal{G}_{6}(\pi n^{\mathsf{T}}Qn)-\frac{\pi^{3}}{2}\mathcal{G}_{3}(\pi n^{\mathsf{T}}Qn)\right\}
+\frac{1}{V}\sum_{n\in\mathbb{Z}^{3}\setminus\{0\}}\left\{\frac{\pi^{6}}{240}\mathcal{G}_{-9/2}(\pi n^{\mathsf{T}}Q^{-1}n)-\frac{\pi^{3}}{2}\mathcal{G}_{-3/2}(\pi n^{\mathsf{T}}Q^{-1}n)\right\}. E ( Q ) = 240 π 6 ( − 6 1 + 9 V 2 ) − 2 π 3 ( − 3 1 + 3 V 2 ) + n ∈ Z 3 ∖ { 0 } ∑ { 240 π 6 G 6 ( π n T Q n ) − 2 π 3 G 3 ( π n T Q n ) } + V 1 n ∈ Z 3 ∖ { 0 } ∑ { 240 π 6 G − 9/2 ( π n T Q − 1 n ) − 2 π 3 G − 3/2 ( π n T Q − 1 n ) } .
The following rational enclosures hold for every kernel.
Proof. Use the alternating-series enclosures of 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 ) , with 37 and 11 terms, respectively. For x ≥ 0 x\geq0 x ≥ 0 and the least m ≥ 0 m\geq0 m ≥ 0 with x / 2 m ≤ 1 / 2 x/2^{m}\leq1/2 x / 2 m ≤ 1/2 ,
( ∑ j = 0 33 ( − x / 2 m ) j j ! ) 2 m ≤ e − x ≤ ( ∑ j = 0 32 ( − x / 2 m ) j j ! ) 2 m . \left(\sum_{j=0}^{33}\frac{(-x/2^{m})^{j}}{j!}\right)^{2^{m}}\leq e^{-x}\leq\left(\sum_{j=0}^{32}\frac{(-x/2^{m})^{j}}{j!}\right)^{2^{m}}. ( j = 0 ∑ 33 j ! ( − x / 2 m ) j ) 2 m ≤ e − x ≤ ( j = 0 ∑ 32 j ! ( − x / 2 m ) j ) 2 m .
For integers a ≥ 1 a\geq1 a ≥ 1 ,
G a ( x ) = e − x ∑ j = 0 a − 1 ( a − 1 ) ! j ! x a − j . \mathcal{G}_{a}(x)=e^{-x}\sum_{j=0}^{a-1}\frac{(a-1)!}{j!x^{a-j}}. G a ( x ) = e − x j = 0 ∑ a − 1 j ! x a − j ( a − 1 )! .
For x , k > 0 x,k>0 x , k > 0 and M ∈ Z ≥ 1 M\in\mathbb{Z}_{\geq1} M ∈ Z ≥ 1 , define the symmetric tridiagonal matrix
( J M , k ) j j = k + 2 j ( 0 ≤ j < M ) , ( J M , k ) j , j + 1 2 = ( j + 1 ) ( j + k ) ( 0 ≤ j < M − 1 ) . (J_{M,k})_{jj}=k+2j\quad(0\leq j<M),\qquad(J_{M,k})_{j,j+1}^{2}=(j+1)(j+k)\quad(0\leq j<M-1). ( J M , k ) j j = k + 2 j ( 0 ≤ j < M ) , ( J M , k ) j , j + 1 2 = ( j + 1 ) ( j + k ) ( 0 ≤ j < M − 1 ) .
For e 0 = ( 1 , 0 , … , 0 ) T ∈ R M e_{0}=(1,0,\ldots,0)^{\mathsf{T}}\in\mathbb{R}^{M} e 0 = ( 1 , 0 , … , 0 ) T ∈ R M , its rational resolvent is
L M , k ( x ) = e 0 T ( x I + J M , k ) − 1 e 0 = d 0 − 1 , L_{M,k}(x)=e_{0}^{\mathsf{T}}(xI+J_{M,k})^{-1}e_{0}=d_{0}^{-1}, L M , k ( x ) = e 0 T ( x I + J M , k ) − 1 e 0 = d 0 − 1 ,
d M − 1 = x + k + 2 ( M − 1 ) , d j = x + k + 2 j − ( j + 1 ) ( j + k ) d j + 1 ( j = M − 2 , … , 0 ) . d_{M-1}=x+k+2(M-1),\qquad d_{j}=x+k+2j-\frac{(j+1)(j+k)}{d_{j+1}}\quad(j=M-2,\ldots,0). d M − 1 = x + k + 2 ( M − 1 ) , d j = x + k + 2 j − d j + 1 ( j + 1 ) ( j + k ) ( j = M − 2 , … , 0 ) .
For every k > 0 k>0 k > 0 ,
e − x L M , k ( x ) < G 1 − k ( x ) < e − x x { 1 − k L M , k + 1 ( x ) } . e^{-x}L_{M,k}(x)<\mathcal{G}_{1-k}(x)<\frac{e^{-x}}{x}\{1-kL_{M,k+1}(x)\}. e − x L M , k ( x ) < G 1 − k ( x ) < x e − x { 1 − k L M , k + 1 ( x )} .
Indeed,
e x G 1 − k ( x ) = 1 Γ ( k ) ∫ 0 ∞ t k − 1 e − t x + t d t . e^{x}\mathcal{G}_{1-k}(x)=\frac{1}{\Gamma(k)}\int_{0}^{\infty}\frac{t^{k-1}e^{-t}}{x+t}\,dt. e x G 1 − k ( x ) = Γ ( k ) 1 ∫ 0 ∞ x + t t k − 1 e − t d t .
For the monic degree-M M M gamma-weight orthogonal polynomial P P P , Gauss quadrature is exact on ( 1 − P ( t ) 2 / P ( − x ) 2 ) / ( x + t ) (1-P(t)^{2}/P(-x)^{2})/(x+t) ( 1 − P ( t ) 2 / P ( − x ) 2 ) / ( x + t ) ; hence
e x G 1 − k ( x ) − L M , k ( x ) = 1 P ( − x ) 2 Γ ( k ) ∫ 0 ∞ P ( t ) 2 t 2 k − 1 e − t x + t d t > 0. e^{x}\mathcal{G}_{1-k}(x)-L_{M,k}(x)=\frac{1}{P(-x)^{2}\Gamma(k)}\int_{0}^{\infty}\frac{P(t)^{2}t^{2k-1}e^{-t}}{x+t}\,dt>0. e x G 1 − k ( x ) − L M , k ( x ) = P ( − x ) 2 Γ ( k ) 1 ∫ 0 ∞ x + t P ( t ) 2 t 2 k − 1 e − t d t > 0.
The upper bound follows from x e x G 1 − k ( x ) = 1 − k e x G − k ( x ) xe^{x}\mathcal{G}_{1-k}(x)=1-ke^{x}\mathcal{G}_{-k}(x) x e x G 1 − k ( x ) = 1 − k e x G − k ( x ) . For M = 48 , 32 , 24 , 16 M=48,32,24,16 M = 48 , 32 , 24 , 16 and x ∈ [ x − , x + ] x\in[x_{-},x_{+}] x ∈ [ x − , x + ] , the bounds for e x G 1 − k ( x ) e^{x}\mathcal{G}_{1-k}(x) e x G 1 − k ( x ) are L M , k ( x + ) L_{M,k}(x_{+}) L M , k ( x + ) and { 1 − k L M , k + 1 ( x − ) } / x − \{1-kL_{M,k+1}(x_{-})\}/x_{-} { 1 − k L M , k + 1 ( x − )} / x − , respectively. Use interval arithmetic for signed kernel combinations.
For N ∈ Z ≥ 0 N\in\mathbb{Z}_{\geq0} N ∈ Z ≥ 0 and b > 0 b>0 b > 0 , put
Θ N ( b ) = 6 e − π b ( N + 1 ) 2 ( 1 + b − 1 / 2 ) 2 1 − e − π b ( 2 N + 3 ) . \Theta_{N}(b)=\frac{6e^{-\pi b(N+1)^{2}}(1+b^{-1/2})^{2}}{1-e^{-\pi b(2N+3)}}. Θ N ( b ) = 1 − e − π b ( 2 N + 3 ) 6 e − π b ( N + 1 ) 2 ( 1 + b − 1/2 ) 2 .
The coordinate union bound and geometric summation give
∑ n ∈ Z 3 max 1 ≤ i ≤ 3 ∣ n i ∣ > N e − π b ∣ n ∣ 2 ≤ Θ N ( b ) , ∑ j ∈ Z e − π b j 2 ≤ 1 + b − 1 / 2 . \sum_{\substack{n\in\mathbb{Z}^{3}\\\max_{1\leq i\leq3}|n_{i}|>N}}e^{-\pi b|n|^{2}}\leq\Theta_{N}(b),\qquad\sum_{j\in\mathbb{Z}}e^{-\pi bj^{2}}\leq1+b^{-1/2}. n ∈ Z 3 m a x 1 ≤ i ≤ 3 ∣ n i ∣ > N ∑ e − π b ∣ n ∣ 2 ≤ Θ N ( b ) , j ∈ Z ∑ e − π b j 2 ≤ 1 + b − 1/2 .
For the cube N = 5 N=5 N = 5 , take a ≤ λ min ( Q ) a \leq\lambda_{\min}(Q) a ≤ λ m i n ( Q ) and a ∗ ≤ λ min ( Q − 1 ) a^* \leq\lambda_{\min}(Q^{-1}) a ∗ ≤ λ m i n ( Q − 1 ) . The omitted energy is bounded below and above, respectively, by
− π 3 2 ( 1 36 π a + 2 ( 36 π a ) 2 + 2 ( 36 π a ) 3 ) Θ 5 ( a ) − π 3 / 2 V ( 36 π a ∗ ) Θ 5 ( a ∗ ) -\frac{\pi^3}{2}\left(\frac{1}{36\pi a}+\frac{2}{(36\pi a)^2}+\frac{2}{(36\pi a)^3}\right)\Theta_5(a)-\frac{\pi^{3/2}}{V(36\pi a^*)}\Theta_5(a^*) − 2 π 3 ( 36 π a 1 + ( 36 π a ) 2 2 + ( 36 π a ) 3 2 ) Θ 5 ( a ) − V ( 36 π a ∗ ) π 3/2 Θ 5 ( a ∗ )
π 6 240 ( 1 36 π a + 5 ( 36 π a ) 2 + 20 ( 36 π a ) 3 + 60 ( 36 π a ) 4 + 120 ( 36 π a ) 5 + 120 ( 36 π a ) 6 ) Θ 5 ( a ) + π 6 / 240 V ( 36 π a ∗ ) Θ 5 ( a ∗ ) . \frac{\pi^6}{240}\left(\frac{1}{36\pi a}+\frac{5}{(36\pi a)^2}+\frac{20}{(36\pi a)^3}+\frac{60}{(36\pi a)^4}+\frac{120}{(36\pi a)^5}+\frac{120}{(36\pi a)^6}\right)\Theta_5(a)+\frac{\pi^6/240}{V(36\pi a^*)}\Theta_5(a^*). 240 π 6 ( 36 π a 1 + ( 36 π a ) 2 5 + ( 36 π a ) 3 20 + ( 36 π a ) 4 60 + ( 36 π a ) 5 120 + ( 36 π a ) 6 120 ) Θ 5 ( a ) + V ( 36 π a ∗ ) π 6 /240 Θ 5 ( a ∗ ) .
For cube N N N , replace 36, 13 by ( N + 1 ) 2 (N+1)^2 ( N + 1 ) 2 , 2 N + 3 2N+3 2 N + 3 . Lemma A.1 gives the common arithmetic. This completes the proof of Lemma 2.8. □
The complete coordinate derivatives and tails are given in Lemma 2.10 .
Lemma 2.9 (One polynomial construction for the stationary basins). For a reference lattice L 0 L_0 L 0 , use
S = ( 2 u x y x − u + v z y z − u − v ) , s = ( u , v , x , y , z ) T , (19) S=\begin{pmatrix}2u & x & y \\ x & -u+v & z \\ y & z & -u-v\end{pmatrix},\qquad\mathbf{s}=(u,v,x,y,z)^{\mathsf{T}}, \tag*{(19)} S = 2 u x y x − u + v z y z − u − v , s = ( u , v , x , y , z ) T , (19) ∥ S ∥ F 2 = 6 u 2 + 2 v 2 + 2 x 2 + 2 y 2 + 2 z 2 , D = det ( I + S ) . \lVert S\rVert_F^2=6u^2+2v^2+2x^2+2y^2+2z^2,\qquad D=\det(I+S). ∥ S ∥ F 2 = 6 u 2 + 2 v 2 + 2 x 2 + 2 y 2 + 2 z 2 , D = det ( I + S ) . and L ( S ) = D − 1 / 6 ( I + S ) 1 / 2 L 0 L(S)=D^{-1/6}(I+S)^{1/2}L_0 L ( S ) = D − 1/6 ( I + S ) 1/2 L 0 . For w ∈ L 0 ∖ { 0 } w\in L_0\setminus\{0\} w ∈ L 0 ∖ { 0 } set q w = ∣ w ∣ 2 q_w=\lvert w\rvert^2 q w = ∣ w ∣ 2 and ℓ w = w T S w / q w \ell_w=w^{\mathsf{T}}Sw/q_w ℓ w = w T S w / q w . Then
E ( L ( S ) ) = ∑ w ∈ L 0 ∖ { 0 } [ D 2 2 q w 6 ( 1 + ℓ w ) 6 − D q w 3 ( 1 + ℓ w ) 3 ] . \mathcal{E}(L(S))=\sum_{w\in L_0\setminus\{0\}}\left[\frac{D^2}{2q_w^6(1+\ell_w)^6}-\frac{D}{q_w^3(1+\ell_w)^3}\right]. E ( L ( S )) = w ∈ L 0 ∖ { 0 } ∑ [ 2 q w 6 ( 1 + ℓ w ) 6 D 2 − q w 3 ( 1 + ℓ w ) 3 D ] . Choose a finite symmetric F ⊂ L 0 ∖ { 0 } F\subset L_0\setminus\{0\} F ⊂ L 0 ∖ { 0 } and complete bounds
∑ w ∈ L 0 ∖ ( F ∪ { 0 } ) ∣ w ∣ − 2 s ≤ τ s , E ( L 0 , s ) ≤ E + ( L 0 , s ) ( s = 3 , 6 ) . \sum_{w\in L_0\setminus(F\cup\{0\})}|w|^{-2s}\leq\tau_s,\qquad E(L_0,s)\leq E^+(L_0,s)\qquad(s=3,6). w ∈ L 0 ∖ ( F ∪ { 0 }) ∑ ∣ w ∣ − 2 s ≤ τ s , E ( L 0 , s ) ≤ E + ( L 0 , s ) ( s = 3 , 6 ) . Put p s ( z ) = ∑ k = 0 6 ( − 1 ) k ( s + k − 1 k ) z k p_s(z)=\sum_{k=0}^{6}(-1)^k\binom{s+k-1}{k}z^k p s ( z ) = ∑ k = 0 6 ( − 1 ) k ( k s + k − 1 ) z k and
P ( S ) = ∑ w ∈ F [ D 2 2 q w 6 p 6 ( ℓ w ) − D q w 3 p 3 ( ℓ w ) ] . P(S)=\sum_{w\in F}\left[\frac{D^2}{2q_w^6}p_6(\ell_w)-\frac{D}{q_w^3}p_3(\ell_w)\right]. P ( S ) = w ∈ F ∑ [ 2 q w 6 D 2 p 6 ( ℓ w ) − q w 3 D p 3 ( ℓ w ) ] . Retain the determinant factors; deg P ≤ 12 \deg P\leq12 deg P ≤ 12 . For 0 ≤ R < 6 / 5 0\leq R<6/5 0 ≤ R < 6/5 and j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 , define
T s , j = ( 5 6 ) j E + ( L 0 , s ) { ( s + j − 1 ) ! ( s − 1 ) ! ( 1 − 5 R / 6 ) s + j − ∑ k = j 6 k ! ( k − j ) ! ( s + k − 1 k ) ( 5 R / 6 ) k − j } . T_{s,j}=\left(\frac{5}{6}\right)^j E^+(L_0,s)\left\{\frac{(s+j-1)!}{(s-1)!(1-5R/6)^{s+j}}-\sum_{k=j}^{6}\frac{k!}{(k-j)!}\binom{s+k-1}{k}(5R/6)^{k-j}\right\}. T s , j = ( 6 5 ) j E + ( L 0 , s ) ⎩ ⎨ ⎧ ( s − 1 )! ( 1 − 5 R /6 ) s + j ( s + j − 1 )! − k = j ∑ 6 ( k − j )! k ! ( k s + k − 1 ) ( 5 R /6 ) k − j ⎭ ⎬ ⎫ . On ∥ S ∥ F ≤ R \lVert S\rVert_F\leq R ∥ S ∥ F ≤ R , the complete Hessian remainder is at most
ε ( R ) = T 3 , 2 + 2 R 1 − 5 R / 6 T 3 , 1 + R 2 + 1 ( 1 − 5 R / 6 ) 2 T 3 , 0 + 1 2 ( T 6 , 2 + 4 R 1 − 5 R / 6 T 6 , 1 + 4 R 2 + 2 ( 1 − 5 R / 6 ) 2 T 6 , 0 ) + R 2 + 5 R + 28 / 3 ( 1 − 5 R / 6 ) 5 τ 3 + 4 R 2 + 20 R + 187 / 6 2 ( 1 − 5 R / 6 ) 8 τ 6 . (20) \begin{aligned}
\varepsilon(R)={}&T_{3,2}+\frac{2R}{1-5R/6}T_{3,1}+\frac{R^2+1}{(1-5R/6)^2}T_{3,0}\\
&+\frac{1}{2}\left(T_{6,2}+\frac{4R}{1-5R/6}T_{6,1}+\frac{4R^2+2}{(1-5R/6)^2}T_{6,0}\right)\\
&+\frac{R^2+5R+28/3}{(1-5R/6)^5}\tau_3+\frac{4R^2+20R+187/6}{2(1-5R/6)^8}\tau_6. \tag*{(20)}
\end{aligned} ε ( R ) = T 3 , 2 + 1 − 5 R /6 2 R T 3 , 1 + ( 1 − 5 R /6 ) 2 R 2 + 1 T 3 , 0 + 2 1 ( T 6 , 2 + 1 − 5 R /6 4 R T 6 , 1 + ( 1 − 5 R /6 ) 2 4 R 2 + 2 T 6 , 0 ) + ( 1 − 5 R /6 ) 5 R 2 + 5 R + 28/3 τ 3 + 2 ( 1 − 5 R /6 ) 8 4 R 2 + 20 R + 187/6 τ 6 . (20) Thus ∥ ∇ S 2 [ E ( L ( S ) ) − P ( S ) ] ∥ o p ≤ ε ( R ) \lVert\nabla_S^2[\mathcal{E}(L(S))-P(S)]\rVert_{\mathrm{op}}\leq\varepsilon(R) ∥ ∇ S 2 [ E ( L ( S )) − P ( S )] ∥ op ≤ ε ( R ) , where the operator norm is induced by ∥ ⋅ ∥ F \lVert\cdot\rVert_F ∥ ⋅ ∥ F on trace-free symmetric matrices. For E θ ( L ) = 1 2 E ( L , 6 ) − θ E ( L , 3 ) E_\theta(L)=\frac{1}{2}E(L,6)-\theta E(L,3) E θ ( L ) = 2 1 E ( L , 6 ) − θ E ( L , 3 ) , multiply the two s = 3 s=3 s = 3 contributions by an upper bound for ∣ θ ∣ \lvert\theta\rvert ∣ θ ∣ .
Proof. For trace-free symmetric H H H , ∥ H ∥ o p ≤ 2 / 3 ∥ H ∥ F ≤ 5 6 ∥ H ∥ F \lVert H\rVert_{\mathrm{op}} \le\sqrt{2/3}\lVert H\rVert_{F} \le\frac{5}{6}\lVert H\rVert_{F} ∥ H ∥ op ≤ 2/3 ∥ H ∥ F ≤ 6 5 ∥ H ∥ F . The positive binomial tails of orders j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 are bounded by T s , j T_{s,j} T s , j . For s = 3 , 6 s=3,6 s = 3 , 6 and ∥ H ∥ F = 1 \lVert H\rVert_{F}=1 ∥ H ∥ F = 1 ,
∂ t det ( I + S + t H ) s / 3 det ( I + S ) s / 3 ∣ t = 0 = s 3 tr ( ( I + S ) − 1 H ) , \left.\frac{\partial_t\det(I+S+tH)^{s/3}}{\det(I+S)^{s/3}}\right|_{t=0}
=\frac{s}{3}\operatorname{tr}\left((I+S)^{-1}H\right), det ( I + S ) s /3 ∂ t det ( I + S + t H ) s /3 t = 0 = 3 s tr ( ( I + S ) − 1 H ) ,
∂ t 2 det ( I + S + t H ) s / 3 det ( I + S ) s / 3 ∣ t = 0 = s 2 9 [ tr ( ( I + S ) − 1 H ) ] 2 − s 3 tr ( ( ( I + S ) − 1 H ) 2 ) . \left.\frac{\partial_t^2\det(I+S+tH)^{s/3}}{\det(I+S)^{s/3}}\right|_{t=0}
=\frac{s^2}{9}\left[\operatorname{tr}\left((I+S)^{-1}H\right)\right]^2
-\frac{s}{3}\operatorname{tr}\left(\left((I+S)^{-1}H\right)^2\right). det ( I + S ) s /3 ∂ t 2 det ( I + S + t H ) s /3 t = 0 = 9 s 2 [ tr ( ( I + S ) − 1 H ) ] 2 − 3 s tr ( ( ( I + S ) − 1 H ) 2 ) .
Since D ≤ 1 D\le1 D ≤ 1 , these identities give
∣ ∂ t j det ( I + S + t H ) s / 3 det ( I + S ) s / 3 ∣ t = 0 ∣ ≤ { s R 3 ( 1 − 5 R / 6 ) , j = 1 , s 2 R 2 / 9 + s / 3 ( 1 − 5 R / 6 ) 2 , j = 2. \left|\left.\partial_t^j\frac{\det(I+S+tH)^{s/3}}{\det(I+S)^{s/3}}\right|_{t=0}\right|
\le
\begin{cases}
\frac{sR}{3(1-5R/6)}, & j=1,\\
\frac{s^2R^2/9+s/3}{(1-5R/6)^2}, & j=2.
\end{cases} ∂ t j det ( I + S ) s /3 det ( I + S + t H ) s /3 t = 0 ≤ { 3 ( 1 − 5 R /6 ) s R , ( 1 − 5 R /6 ) 2 s 2 R 2 /9 + s /3 , j = 1 , j = 2.
The product rule gives (20) , including all omitted vectors and binomial terms. The finite matrix checks use
D = 1 − 3 u 2 − v 2 − x 2 − y 2 − z 2 + 2 u 3 − 2 u v 2 − 2 u z 2 D=1-3u^2-v^2-x^2-y^2-z^2+2u^3-2uv^2-2uz^2 D = 1 − 3 u 2 − v 2 − x 2 − y 2 − z 2 + 2 u 3 − 2 u v 2 − 2 u z 2
D = + u x 2 + v x 2 + u y 2 − v y 2 + 2 x y z , \phantom{D={}}+ux^2+vx^2+uy^2-vy^2+2xyz, D = + u x 2 + v x 2 + u y 2 − v y 2 + 2 x y z ,
max ∥ S ∥ F ≤ R ∣ S α ∣ = R ∣ α ∣ ( α 1 6 ∣ α ∣ ) α 1 / 2 ∏ i = 2 5 ( α i 2 ∣ α ∣ ) α i / 2 , ∣ α ∣ > 0. \max_{\lVert S\rVert_F\le R}|S^\alpha|
=R^{|\alpha|}\left(\frac{\alpha_1}{6|\alpha|}\right)^{\alpha_1/2}
\prod_{i=2}^{5}\left(\frac{\alpha_i}{2|\alpha|}\right)^{\alpha_i/2},
\qquad|\alpha|>0. ∥ S ∥ F ≤ R max ∣ S α ∣ = R ∣ α ∣ ( 6∣ α ∣ α 1 ) α 1 /2 i = 2 ∏ 5 ( 2∣ α ∣ α i ) α i /2 , ∣ α ∣ > 0.
Here α ∈ Z ≥ 0 5 \alpha\in\mathbb{Z}_{\ge0}^{5} α ∈ Z ≥ 0 5 and ∣ α ∣ = ∑ i = 1 5 α i |\alpha|=\sum_{i=1}^{5}\alpha_i ∣ α ∣ = ∑ i = 1 5 α i ; zero exponents contribute one. Apply them to ∇ S 2 P \nabla_S^2P ∇ S 2 P , retaining the nonnegative signs of even monomials. For the homogeneous decomposition P = ∑ d = 0 12 P d P=\sum_{d=0}^{12}P_d P = ∑ d = 0 12 P d , put
B i j ( S ) = ∑ d = 2 12 1 d − 1 ∂ 2 P d ∂ s i ∂ s j ( S ) , 1 ≤ i , j ≤ 5. B_{ij}(S)=\sum_{d=2}^{12}\frac{1}{d-1}\frac{\partial^2P_d}{\partial s_i\partial s_j}(S),
\qquad1\le i,j\le5. B ij ( S ) = d = 2 ∑ 12 d − 1 1 ∂ s i ∂ s j ∂ 2 P d ( S ) , 1 ≤ i , j ≤ 5.
Euler’s identity and integration give
s ⋅ ∇ S P ( S ) = s ⋅ ∇ S P ( 0 ) + s T B ( S ) s , \mathbf{s}\cdot\nabla_SP(S)=\mathbf{s}\cdot\nabla_SP(0)+\mathbf{s}^{\mathsf T}B(S)\mathbf{s}, s ⋅ ∇ S P ( S ) = s ⋅ ∇ S P ( 0 ) + s T B ( S ) s ,
∣ ∂ t [ E ( L ( t S ) ) − P ( t S ) ] ∣ t = 0 t = 1 ∣ = ∣ ∫ 0 1 ∂ t 2 [ E ( L ( t S ) ) − P ( t S ) ] d t ∣ ≤ ε ( R ) ∥ S ∥ F 2 . \left|\left.\partial_t\left[\mathcal{E}(L(tS))-P(tS)\right]\right|_{t=0}^{t=1}\right|
=\left|\int_0^1\partial_t^2\left[\mathcal{E}(L(tS))-P(tS)\right]\,dt\right|
\le\varepsilon(R)\lVert S\rVert_F^2. ∂ t [ E ( L ( tS )) − P ( tS ) ] ∣ t = 0 t = 1 = ∫ 0 1 ∂ t 2 [ E ( L ( tS )) − P ( tS ) ] d t ≤ ε ( R ) ∥ S ∥ F 2 .
The matrix and radial bounds give inner convexity and outer stationary-point exclusion. This completes the proof of Lemma 2.9. □ \square □
Lemma 2.10 (Coordinate Ewald formulas and uniform truncation bounds). Fix V > 0 V>0 V > 0 , N ∈ Z ≥ 0 N\in\mathbb{Z}_{\ge0} N ∈ Z ≥ 0 and a closed box B = ∏ i = 1 5 [ p i − , p i + ] ⊂ { a , δ > 0 } B=\prod_{i=1}^{5}[p_i^-,p_i^+]\subset\{a,\delta>0\} B = ∏ i = 1 5 [ p i − , p i + ] ⊂ { a , δ > 0 } , where p = ( a , δ , x , y , z ) p=(a,\delta,x,y,z) p = ( a , δ , x , y , z ) . Put
U ( p ) = ( a a x a y 0 δ δ z 0 0 h ) , h = V / ( a δ ) , U(p)=
\begin{pmatrix}
a & ax & ay\\
0 & \delta& \delta z\\
0 & 0 & h
\end{pmatrix},
\qquad h=V/(a\delta), U ( p ) = a 0 0 a x δ 0 a y δ z h , h = V / ( a δ ) , and let h − > 0 h_->0 h − > 0 be a lower bound for h h h on B B B . Set U i = ∂ p i U U_i=\partial_{p_i}U U i = ∂ p i U , U i j = ∂ p i ∂ p j U U_{ij}=\partial_{p_i}\partial_{p_j}U U ij = ∂ p i ∂ p j U . The identities (21) –(24) and the complete value, gradient and Hessian truncation bounds below hold uniformly on B B B , with no reduction hypothesis.
Proof. The squared lengths are
q p ∗ ( i , j , k ) = a 2 ( i + x j + y k ) 2 + δ 2 ( j + z k ) 2 + h 2 k 2 , (21) q_p^*(i,j,k)=a^2(i+xj+yk)^2+\delta^2(j+zk)^2+h^2k^2, \tag*{(21)} q p ∗ ( i , j , k ) = a 2 ( i + x j + y k ) 2 + δ 2 ( j + z k ) 2 + h 2 k 2 , (21)
q p ∗ ( i , j , k ) = a − 2 i 2 + δ − 2 ( j − x i ) 2 + h − 2 { k − z j + ( x z − y ) i } 2 . (22) q_p^*(i,j,k)=a^{-2}i^2+\delta^{-2}(j-xi)^2+h^{-2}\{k-zj+(xz-y)i\}^2. \tag*{(22)} q p ∗ ( i , j , k ) = a − 2 i 2 + δ − 2 ( j − x i ) 2 + h − 2 { k − z j + ( x z − y ) i } 2 . (22)
Put f ( p ) = 1 2 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) f(p)=\frac{1}{2}E(U(p)\mathbb{Z}^{3},6)-E(U(p)\mathbb{Z}^{3},3) f ( p ) = 2 1 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) and set
g ( q ) = π 6 240 G 6 ( π q ) − π 3 2 G 3 ( π q ) , g ~ ( q ) = π 6 240 G − 9 / 2 ( π q ) − π 3 2 G − 3 / 2 ( π q ) V . g(q)=\frac{\pi^{6}}{240}\mathcal{G}_{6}(\pi q)-\frac{\pi^{3}}{2}\mathcal{G}_{3}(\pi q),\qquad\widetilde{g}(q)=\frac{\frac{\pi^{6}}{240}\mathcal{G}_{-9/2}(\pi q)-\frac{\pi^{3}}{2}\mathcal{G}_{-3/2}(\pi q)}{V}. g ( q ) = 240 π 6 G 6 ( π q ) − 2 π 3 G 3 ( π q ) , g ( q ) = V 240 π 6 G − 9/2 ( π q ) − 2 π 3 G − 3/2 ( π q ) .
The exact identity of Lemma 2.8 is
f ( p ) = π 6 240 ( − 1 6 + 2 9 V ) − π 3 2 ( − 1 3 + 2 3 V ) + ∑ ( i , j , k ) ∈ Z 3 ∖ { 0 } { g ( q p ( i , j , k ) ) + g ~ ( q p ∗ ( i , j , k ) ) } . (23) f(p)=\frac{\pi^{6}}{240}\left(-\frac{1}{6}+\frac{2}{9V}\right)-\frac{\pi^{3}}{2}\left(-\frac{1}{3}+\frac{2}{3V}\right)+\sum_{(i,j,k)\in\mathbb{Z}^{3}\setminus\{0\}}\left\{g(q_p(i,j,k))+\widetilde{g}(q_p^{*}(i,j,k))\right\}. \tag*{(23)} f ( p ) = 240 π 6 ( − 6 1 + 9 V 2 ) − 2 π 3 ( − 3 1 + 3 V 2 ) + ( i , j , k ) ∈ Z 3 ∖ { 0 } ∑ { g ( q p ( i , j , k )) + g ( q p ∗ ( i , j , k )) } . (23)
For r = 0 , 1 , 2 r=0,1,2 r = 0 , 1 , 2 , with g ( r ) = d r g / d q r g^{(r)}=d^{r}g/dq^{r} g ( r ) = d r g / d q r and g ~ ( r ) = d r g ~ / d q r \widetilde{g}^{(r)}=d^{r}\widetilde{g}/dq^{r} g ( r ) = d r g / d q r ,
g ( r ) ( q ) = ( − π ) r { π 6 240 G 6 + r ( π q ) − π 3 2 G 3 + r ( π q ) } , g^{(r)}(q)=(-\pi)^{r}\left\{\frac{\pi^{6}}{240}\mathcal{G}_{6+r}(\pi q)-\frac{\pi^{3}}{2}\mathcal{G}_{3+r}(\pi q)\right\}, g ( r ) ( q ) = ( − π ) r { 240 π 6 G 6 + r ( π q ) − 2 π 3 G 3 + r ( π q ) } ,
g ~ ( r ) ( q ) = ( − π ) r V { π 6 240 G − 9 / 2 + r ( π q ) − π 3 2 G − 3 / 2 + r ( π q ) } . \widetilde{g}^{(r)}(q)=\frac{(-\pi)^{r}}{V}\left\{\frac{\pi^{6}}{240}\mathcal{G}_{-9/2+r}(\pi q)-\frac{\pi^{3}}{2}\mathcal{G}_{-3/2+r}(\pi q)\right\}. g ( r ) ( q ) = V ( − π ) r { 240 π 6 G − 9/2 + r ( π q ) − 2 π 3 G − 3/2 + r ( π q ) } .
For q = q p ( i 1 , i 2 , i 3 ) q=q_p(i_1,i_2,i_3) q = q p ( i 1 , i 2 , i 3 ) , q ∗ = q p ∗ ( i 1 , i 2 , i 3 ) q^{*}=q_p^{*}(i_1,i_2,i_3) q ∗ = q p ∗ ( i 1 , i 2 , i 3 ) and 1 ≤ i , j ≤ 5 1\leq i,j\leq5 1 ≤ i , j ≤ 5 ,
( f i , q i , q i ∗ ) = ∂ p i ( f , q , q ∗ ) , ( f i j , q i j , q i j ∗ ) = ∂ p i ∂ p j ( f , q , q ∗ ) . (f_i,q_i,q_i^{*})=\partial_{p_i}(f,q,q^{*}),\qquad(f_{ij},q_{ij},q_{ij}^{*})=\partial_{p_i}\partial_{p_j}(f,q,q^{*}). ( f i , q i , q i ∗ ) = ∂ p i ( f , q , q ∗ ) , ( f ij , q ij , q ij ∗ ) = ∂ p i ∂ p j ( f , q , q ∗ ) .
Lemma 2.8 encloses the kernels; differentiation gives
f i = ∑ ( i 1 , i 2 , i 3 ) ∈ Z 3 ∖ { ( 0 , 0 , 0 ) } { g ′ ( q ) q i + g ~ ′ ( q ∗ ) q i ∗ } , f_i=\sum_{(i_1,i_2,i_3)\in\mathbb{Z}^{3}\setminus\{(0,0,0)\}}\left\{g'(q)q_i+\widetilde{g}'(q^{*})q_i^{*}\right\}, f i = ( i 1 , i 2 , i 3 ) ∈ Z 3 ∖ {( 0 , 0 , 0 )} ∑ { g ′ ( q ) q i + g ′ ( q ∗ ) q i ∗ } ,
f i j = ∑ ( i 1 , i 2 , i 3 ) ∈ Z 3 ∖ { ( 0 , 0 , 0 ) } { g ′ ′ ( q ) q i q j + g ′ ( q ) q i j + g ~ ′ ′ ( q ∗ ) q i ∗ q j ∗ + g ~ ′ ( q ∗ ) q i j ∗ } . (24) f_{ij}=\sum_{(i_1,i_2,i_3)\in\mathbb{Z}^{3}\setminus\{(0,0,0)\}}\left\{g''(q)q_iq_j+g'(q)q_{ij}+\widetilde{g}''(q^{*})q_i^{*}q_j^{*}+\widetilde{g}'(q^{*})q_{ij}^{*}\right\}. \tag*{(24)} f ij = ( i 1 , i 2 , i 3 ) ∈ Z 3 ∖ {( 0 , 0 , 0 )} ∑ { g ′′ ( q ) q i q j + g ′ ( q ) q ij + g ′′ ( q ∗ ) q i ∗ q j ∗ + g ′ ( q ∗ ) q ij ∗ } . (24)
Complete tails. For p ∈ B p\in B p ∈ B , n ∈ Z 3 ∖ { 0 } n\in\mathbb{Z}^{3}\setminus\{0\} n ∈ Z 3 ∖ { 0 } , independently of reduction,
0 < α ≤ ( sup p ∈ B ∥ U ( p ) − 1 ∥ o p 2 ) − 1 , 0 < β ≤ ( sup p ∈ B ∥ U ( p ) ∥ o p 2 ) − 1 , 0<\alpha\leq\left(\sup_{p\in B}\|U(p)^{-1}\|_{\mathrm{op}}^{2}\right)^{-1},\qquad0<\beta\leq\left(\sup_{p\in B}\|U(p)\|_{\mathrm{op}}^{2}\right)^{-1}, 0 < α ≤ ( p ∈ B sup ∥ U ( p ) − 1 ∥ op 2 ) − 1 , 0 < β ≤ ( p ∈ B sup ∥ U ( p ) ∥ op 2 ) − 1 ,
q p ( n ) ≥ α ∣ n ∣ 2 , q p ∗ ( n ) ≥ β ∣ n ∣ 2 , ∥ A ∥ o p ≤ min { ∥ A ∥ F , ∥ A ∥ 1 ∥ A ∥ ∞ } . q_p(n)\geq\alpha\lvert n\rvert^{2},\qquad q_p^{*}(n)\geq\beta\lvert n\rvert^{2},\qquad\|A\|_{\mathrm{op}}\leq\min\{\|A\|_{F},\sqrt{\|A\|_{1}\|A\|_{\infty}}\}. q p ( n ) ≥ α ∣ n ∣ 2 , q p ∗ ( n ) ≥ β ∣ n ∣ 2 , ∥ A ∥ op ≤ min { ∥ A ∥ F , ∥ A ∥ 1 ∥ A ∥ ∞ } .
Use Θ N \Theta_N Θ N from Lemma 2.8; for s ∈ { 3 , 6 } s\in\{3,6\} s ∈ { 3 , 6 } , r ∈ { 0 , 1 , 2 } r\in\{0,1,2\} r ∈ { 0 , 1 , 2 } , set
A s , r ( b ) = ∑ j = 0 s + r − 1 ( s + r − 1 ) ! j ! { { π b ( N + 1 ) 2 } j − s Θ N ( b ) , j ≤ s , Θ N ( b / 2 ) , j = s + 1. A_{s,r}(b)=\sum_{j=0}^{s+r-1}\frac{(s+r-1)!}{j!}\begin{cases}\{\pi b(N+1)^{2}\}^{j-s}\Theta_N(b),&j\leq s,\\ \Theta_N(b/2),&j=s+1.\end{cases} A s , r ( b ) = j = 0 ∑ s + r − 1 j ! ( s + r − 1 )! { { π b ( N + 1 ) 2 } j − s Θ N ( b ) , Θ N ( b /2 ) , j ≤ s , j = s + 1.
Indeed,
x r G s + r ( x ) = e − x ∑ j = 0 s + r − 1 ( s + r − 1 ) ! j ! x j − s , x e − x ≤ e − x / 2 , x^{r}\mathcal{G}_{s+r}(x)=e^{-x}\sum_{j=0}^{s+r-1}\frac{(s+r-1)!}{j!}x^{j-s},\qquad xe^{-x}\leq e^{-x/2}, x r G s + r ( x ) = e − x j = 0 ∑ s + r − 1 j ! ( s + r − 1 )! x j − s , x e − x ≤ e − x /2 ,
and G ν ( x ) ≤ e − x / x \mathcal{G}_{\nu}(x)\leq e^{-x}/x G ν ( x ) ≤ e − x / x for ν ≤ 1 \nu\leq1 ν ≤ 1 . Thus, uniformly on B B B ,
∑ n ∈ Z 3 ∥ n ∥ ∞ > N q p ( n ) r ∣ g ( r ) ( q p ( n ) ) ∣ ≤ P r = π 3 2 A 3 , r ( α ) + π 6 240 A 6 , r ( α ) , \sum_{\substack{n\in\mathbb{Z}^{3}\\ \|n\|_{\infty}>N}}q_p(n)^{r}\lvert g^{(r)}(q_p(n))\rvert\leq P_r=\frac{\pi^{3}}{2}A_{3,r}(\alpha)+\frac{\pi^{6}}{240}A_{6,r}(\alpha), n ∈ Z 3 ∥ n ∥ ∞ > N ∑ q p ( n ) r ∣ g ( r ) ( q p ( n ))∣ ≤ P r = 2 π 3 A 3 , r ( α ) + 240 π 6 A 6 , r ( α ) ,
∑ n ∈ Z 3 ∥ n ∥ ∞ > N q p ∗ ( n ) r ∣ g ~ ( r ) ( q p ∗ ( n ) ) ∣ ≤ D r = π 3 2 + π 6 240 { Θ N ( β ) / { V π β ( N + 1 ) 2 } , r = 0 , Θ N ( β ) , r = 1 , Θ N ( β / 2 ) , r = 2. \sum_{\substack{n\in\mathbb{Z}^{3}\\ \|n\|_{\infty}>N}}q_p^{*}(n)^{r}\lvert\widetilde{g}^{(r)}(q_p^{*}(n))\rvert\leq D_r=\frac{\pi^{3}}{2}+\frac{\pi^{6}}{240}\begin{cases}\Theta_N(\beta)/\{V\pi\beta(N+1)^{2}\},&r=0,\\ \Theta_N(\beta),&r=1,\\ \Theta_N(\beta/2),&r=2.\end{cases} n ∈ Z 3 ∥ n ∥ ∞ > N ∑ q p ∗ ( n ) r ∣ g ( r ) ( q p ∗ ( n ))∣ ≤ D r = 2 π 3 + 240 π 6 ⎩ ⎨ ⎧ Θ N ( β ) / { V π β ( N + 1 ) 2 } , Θ N ( β ) , Θ N ( β /2 ) , r = 0 , r = 1 , r = 2.
The positive attraction/repulsion tails are
π 3 2 { A 3 , 0 ( α ) + Θ N ( β ) V π β ( N + 1 ) 2 } , π 6 240 { A 6 , 0 ( α ) + Θ N ( β ) V π β ( N + 1 ) 2 } . \frac{\pi^{3}}{2}\left\{A_{3,0}(\alpha)+\frac{\Theta_N(\beta)}{V\pi\beta(N+1)^{2}}\right\},\qquad\frac{\pi^{6}}{240}\left\{A_{6,0}(\alpha)+\frac{\Theta_N(\beta)}{V\pi\beta(N+1)^{2}}\right\}. 2 π 3 { A 3 , 0 ( α ) + V π β ( N + 1 ) 2 Θ N ( β ) } , 240 π 6 { A 6 , 0 ( α ) + V π β ( N + 1 ) 2 Θ N ( β ) } .
For box endpoints − , + -,+ − , + , use
( γ a , γ δ , γ y , γ z ) = ( 2 a − , 2 δ − , a + h − , δ + h − ) , (\gamma_a,\gamma_\delta,\gamma_y,\gamma_z)=\left(\frac{2}{a_-},\frac{2}{\delta_-},\frac{a_+}{h_-},\frac{\delta_+}{h_-}\right), ( γ a , γ δ , γ y , γ z ) = ( a − 2 , δ − 2 , h − a + , h − δ + ) ,
γ x = ( a + δ − ) 2 + ( a + max { ∣ z − ∣ , ∣ z + ∣ } h − ) 2 . \gamma_x=\sqrt{\left(\frac{a_+}{\delta_-}\right)^2+\left(\frac{a_+\max\{|z_-|,|z_+|\}}{h_-}\right)^2}. γ x = ( δ − a + ) 2 + ( h − a + max { ∣ z − ∣ , ∣ z + ∣ } ) 2 .
For p ∈ B p\in B p ∈ B , n ≠ 0 n\ne0 n = 0 and 1 ≤ i , j ≤ 5 1\le i,j\le5 1 ≤ i , j ≤ 5 , differentiation gives
∣ q i ∣ ≤ γ i q , ∣ q i ∗ ∣ ≤ γ i q ∗ , t i ≥ sup p ∈ B ∥ U i U − 1 ∥ o p , t i j ≥ sup p ∈ B ∥ U i j U − 1 ∥ o p , |q_i|\le\gamma_iq,\qquad|q_i^*|\le\gamma_iq^*,\qquad t_i\ge\sup_{p\in B}\|U_iU^{-1}\|_{\mathrm{op}},\qquad t_{ij}\ge\sup_{p\in B}\|U_{ij}U^{-1}\|_{\mathrm{op}}, ∣ q i ∣ ≤ γ i q , ∣ q i ∗ ∣ ≤ γ i q ∗ , t i ≥ p ∈ B sup ∥ U i U − 1 ∥ op , t ij ≥ p ∈ B sup ∥ U ij U − 1 ∥ op ,
∣ q i j ∣ ≤ 2 ( t i t j + t i j ) q , ∣ q i j ∗ ∣ ≤ 2 ( 3 t i t j + t i j ) q ∗ . |q_{ij}|\le2(t_it_j+t_{ij})q,\qquad|q_{ij}^*|\le2(3t_it_j+t_{ij})q^*. ∣ q ij ∣ ≤ 2 ( t i t j + t ij ) q , ∣ q ij ∗ ∣ ≤ 2 ( 3 t i t j + t ij ) q ∗ .
Use the preceding norm bounds and 2 ∣ u v ∣ ≤ u 2 + v 2 2|uv|\le u^2+v^2 2∣ uv ∣ ≤ u 2 + v 2 . Consequently,
∑ n ∈ Z 3 ∥ n ∥ ∞ > N ( ∣ g ′ ( q ) q i ∣ + ∣ g ~ ′ ( q ∗ ) q i ∗ ∣ ) ≤ γ i ( P 1 + D 1 ) , \sum_{\substack{n\in\mathbb{Z}^3\\ \|n\|_\infty>N}}\left(|g'(q)q_i|+|\widetilde{g}'(q^*)q_i^*|\right)\le\gamma_i(P_1+D_1), n ∈ Z 3 ∥ n ∥ ∞ > N ∑ ( ∣ g ′ ( q ) q i ∣ + ∣ g ′ ( q ∗ ) q i ∗ ∣ ) ≤ γ i ( P 1 + D 1 ) ,
∑ n ∈ Z 3 ∥ n ∥ ∞ > N ( ∣ g ′ ′ ( q ) q i q j ∣ + ∣ g ′ ( q ) q i j ∣ + ∣ g ~ ′ ′ ( q ∗ ) q i ∗ q j ∗ ∣ + ∣ g ~ ′ ( q ∗ ) q i j ∗ ∣ ) ≤ γ i γ j ( P 2 + D 2 ) + 2 ( t i t j + t i j ) P 1 + 2 ( 3 t i t j + t i j ) D 1 . (25) \begin{aligned}
\sum_{\substack{n\in\mathbb{Z}^3\\ \|n\|_\infty>N}}\bigl(&|g''(q)q_iq_j|+|g'(q)q_{ij}|\\
&+|\widetilde{g}''(q^*)q_i^*q_j^*|+|\widetilde{g}'(q^*)q_{ij}^*|\bigr)\\
&\le\gamma_i\gamma_j(P_2+D_2)+2(t_it_j+t_{ij})P_1+2(3t_it_j+t_{ij})D_1. \tag*{(25)}
\end{aligned} n ∈ Z 3 ∥ n ∥ ∞ > N ∑ ( ∣ g ′′ ( q ) q i q j ∣ + ∣ g ′ ( q ) q ij ∣ + ∣ g ′′ ( q ∗ ) q i ∗ q j ∗ ∣ + ∣ g ′ ( q ∗ ) q ij ∗ ∣ ) ≤ γ i γ j ( P 2 + D 2 ) + 2 ( t i t j + t ij ) P 1 + 2 ( 3 t i t j + t ij ) D 1 . (25)
This completes the proof of Lemma 2.10. □
The simple-hexagonal phase
For V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 , planar reduction identifies the simple-hexagonal minimizer, area convexity gives uniqueness, and Fourier expansion yields its dilute asymptotics. Put F V ( A ) = E ( T A ⊕ ( V / A ) Z ) F_V(A)=\mathcal{E}(T_A\oplus(V/A)\mathbb{Z}) F V ( A ) = E ( T A ⊕ ( V / A ) Z ) . Since A ∗ < 1 A_*<1 A ∗ < 1 ,
m ( V ) ≤ F V ( A ∗ ) < e 2 . m(V)\le F_V(A_*)<e_2. m ( V ) ≤ F V ( A ∗ ) < e 2 .
The reduction is
L ∈ argmin ∣ L ∣ = V E ( L ) ⟹ L ≅ T A ⊕ ( V / A ) Z , 2 5 < A < 89 100 , F V ′ ( A ) = 0 , L\in\operatorname*{argmin}_{|L|=V}\mathcal{E}(L)\Longrightarrow L\cong T_A\oplus(V/A)\mathbb{Z},\qquad\frac{2}{5}<A<\frac{89}{100},\qquad F_V'(A)=0, L ∈ ∣ L ∣ = V argmin E ( L ) ⟹ L ≅ T A ⊕ ( V / A ) Z , 5 2 < A < 100 89 , F V ′ ( A ) = 0 ,
with uniqueness from F V ′ ′ > 34 F_V''>34 F V ′′ > 34 .
For y > 0 y>0 y > 0 , set
Γ x , y = y − 1 / 2 span Z { ( 1 , 0 ) , ( x , y ) } . \Gamma_{x,y}=y^{-1/2}\operatorname{span}_{\mathbb{Z}}\{(1,0),(x,y)\}. Γ x , y = y − 1/2 span Z {( 1 , 0 ) , ( x , y )} .
Thus Γ 1 / 2 , 3 / 2 = T 1 \Gamma_{1/2,\sqrt{3}/2}=T_1 Γ 1/2 , 3 /2 = T 1 , and for s = 3 , 6 s=3,6 s = 3 , 6 ,
E ( Γ x , y , s ) = ∑ ( i , j ) ∈ Z 2 ( i , j ) ≠ ( 0 , 0 ) y s ( ( i + j x ) 2 + j 2 y 2 ) s . E(\Gamma_{x,y},s)=\sum_{\substack{(i,j)\in\mathbb{Z}^2\\(i,j)\ne(0,0)}}\frac{y^s}{\left((i+jx)^2+j^2y^2\right)^s}. E ( Γ x , y , s ) = ( i , j ) ∈ Z 2 ( i , j ) = ( 0 , 0 ) ∑ ( ( i + j x ) 2 + j 2 y 2 ) s y s .
Lemma 3.1 (Watson [32 ] ). For z > 0 z>0 z > 0 and ν ≥ 0 \nu\ge0 ν ≥ 0 ,
K ν ( z ) = ∫ 0 ∞ e − z cosh a cosh ( ν a ) d a = π 2 z e − z Γ ( ν + 1 / 2 ) ∫ 0 ∞ e − b b ν − 1 / 2 ( 1 + b 2 z ) ν − 1 / 2 d b . (26) K_\nu(z)=\int_0^\infty e^{-z\cosh a}\cosh(\nu a)\,\mathrm{d}a
=\sqrt{\frac{\pi}{2z}}\frac{e^{-z}}{\Gamma(\nu+1/2)}\int_0^\infty e^{-b}b^{\nu-1/2}\left(1+\frac{b}{2z}\right)^{\nu-1/2}\,\mathrm{d}b. \tag*{(26)} K ν ( z ) = ∫ 0 ∞ e − z c o s h a cosh ( ν a ) d a = 2 z π Γ ( ν + 1/2 ) e − z ∫ 0 ∞ e − b b ν − 1/2 ( 1 + 2 z b ) ν − 1/2 d b . (26) For j ∈ Z ≥ 0 j\in\mathbb{Z}_{\ge0} j ∈ Z ≥ 0 and 0 ≤ μ ≤ ν 0\le\mu\le\nu 0 ≤ μ ≤ ν ,
( − 1 ) j K ν ( j ) ( z ) = ∫ 0 ∞ e − z cosh a ( cosh a ) j cosh ( ν a ) d a > 0 , (-1)^jK_\nu^{(j)}(z)=\int_0^\infty e^{-z\cosh a}(\cosh a)^j\cosh(\nu a)\,\mathrm{d}a>0, ( − 1 ) j K ν ( j ) ( z ) = ∫ 0 ∞ e − z c o s h a ( cosh a ) j cosh ( ν a ) d a > 0 , K − ν ( z ) = K ν ( z ) , K μ ( z ) ≤ K ν ( z ) . K_{-\nu}(z)=K_\nu(z),\qquad K_\mu(z)\le K_\nu(z). K − ν ( z ) = K ν ( z ) , K μ ( z ) ≤ K ν ( z ) .
The differential and recurrence identities are
z 2 K ν ′ ′ + z K ν ′ − ( z 2 + ν 2 ) K ν = 0 , K 0 ′ = − K 1 , z^{2}K_{\nu}^{\prime\prime}+zK_{\nu}^{\prime}-(z^{2}+\nu^{2})K_{\nu}=0,\qquad K_{0}^{\prime}=-K_{1}, z 2 K ν ′′ + z K ν ′ − ( z 2 + ν 2 ) K ν = 0 , K 0 ′ = − K 1 ,
K ν + 1 = K ν − 1 + 2 ν z K ν , d d z { z ν K ν ( z ) } = − z ν K ν − 1 ( z ) . K_{\nu+1}=K_{\nu-1}+\frac{2\nu}{z}K_{\nu},\qquad\frac{\mathrm{d}}{\mathrm{d}z}\{z^{\nu}K_{\nu}(z)\}=-z^{\nu}K_{\nu-1}(z). K ν + 1 = K ν − 1 + z 2 ν K ν , d z d { z ν K ν ( z )} = − z ν K ν − 1 ( z ) .
For n ∈ Z ≥ 0 n\in\mathbb{Z}_{\geq0} n ∈ Z ≥ 0 ,
P n ( z ) = ∑ ℓ = 0 n ( n + ℓ ) ! ℓ ! ( n − ℓ ) ! ( 2 z ) ℓ , K n + 1 / 2 ( z ) = π 2 z e − z P n ( z ) . P_{n}(z)=\sum_{\ell=0}^{n}\frac{(n+\ell)!}{\ell!(n-\ell)!(2z)^{\ell}},\qquad K_{n+1/2}(z)=\sqrt{\frac{\pi}{2z}}e^{-z}P_{n}(z). P n ( z ) = ℓ = 0 ∑ n ℓ ! ( n − ℓ )! ( 2 z ) ℓ ( n + ℓ )! , K n + 1/2 ( z ) = 2 z π e − z P n ( z ) .
For z ≥ 1 z\geq1 z ≥ 1 , ν ≥ 0 \nu\geq0 ν ≥ 0 and j ∈ Z ≥ 0 j\in\mathbb{Z}_{\geq0} j ∈ Z ≥ 0 ,
z − 1 / 2 e − z − 1 ≤ K ν ( z ) , ∣ K ν ( j ) ( z ) ∣ ≤ 2 π z e − z + ( j + ν ) 2 / ( 2 z ) . (27) z^{-1/2}e^{-z-1}\leq K_{\nu}(z),\qquad\left|K_{\nu}^{(j)}(z)\right|\leq\sqrt{\frac{2\pi}{z}}e^{-z+(j+\nu)^{2}/(2z)}. \tag*{(27)} z − 1/2 e − z − 1 ≤ K ν ( z ) , K ν ( j ) ( z ) ≤ z 2 π e − z + ( j + ν ) 2 / ( 2 z ) . (27)
Lemma 3.2 (Watson [32 ] ). For z > 0 z>0 z > 0 , set
K 2 ( z ) = π 2 z e − z ( 1 + 15 8 z + 105 128 z 2 ) , \mathcal{K}_{2}(z)=\sqrt{\frac{\pi}{2z}}e^{-z}\left(1+\frac{15}{8z}+\frac{105}{128z^{2}}\right), K 2 ( z ) = 2 z π e − z ( 1 + 8 z 15 + 128 z 2 105 ) , K 5 ( z ) = π 2 z e − z ( 1 + 99 8 z + 9009 128 z 2 + 225225 1024 z 3 + 11486475 32768 z 4 + 43648605 262144 z 5 ) . \mathcal{K}_{5}(z)=\sqrt{\frac{\pi}{2z}}e^{-z}\left(1+\frac{99}{8z}+\frac{9009}{128z^{2}}+\frac{225225}{1024z^{3}}+\frac{11486475}{32768z^{4}}+\frac{43648605}{262144z^{5}}\right). K 5 ( z ) = 2 z π e − z ( 1 + 8 z 99 + 128 z 2 9009 + 1024 z 3 225225 + 32768 z 4 11486475 + 262144 z 5 43648605 ) . They satisfy K ν ( z ) ≤ K ν ( z ) K_{\nu}(z)\leq\mathcal{K}_{\nu}(z) K ν ( z ) ≤ K ν ( z ) for ν = 2 , 5 \nu=2,5 ν = 2 , 5 and z > 0 z>0 z > 0 .
Lemma 3.3 (An explicit planar Epstein gap). For 0 ≤ x ≤ 1 / 2 0\leq x\leq1/2 0 ≤ x ≤ 1/2 and y ≥ 3 / 2 y\geq\sqrt{3}/2 y ≥ 3 /2 ,
E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ≥ 9 5 π 3 e − 2 π y ( 1 + cos ( 2 π x ) ) + 3 3 ( y − 3 2 ) 2 E(\Gamma_{x,y},3)-E(T_{1},3)\geq\frac{9}{5}\pi^{3}e^{-2\pi y}(1+\cos(2\pi x))+3\sqrt{3}\left(y-\frac{\sqrt{3}}{2}\right)^{2} E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ≥ 5 9 π 3 e − 2 π y ( 1 + cos ( 2 π x )) + 3 3 ( y − 2 3 ) 2 ≥ 1 4 [ ( x − 1 2 ) 2 + ( y − 3 2 ) 2 ] . \geq\frac{1}{4}\left[\left(x-\frac{1}{2}\right)^{2}+\left(y-\frac{\sqrt{3}}{2}\right)^{2}\right]. ≥ 4 1 ( x − 2 1 ) 2 + ( y − 2 3 ) 2 . If 0 < δ ≤ 1 / 10 0<\delta\leq1/10 0 < δ ≤ 1/10 , then
( x − 1 2 ) 2 + ( y − 3 2 ) 2 ≥ δ 2 ⟹ E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ≥ δ 2 2 . \left(x-\frac{1}{2}\right)^{2}+\left(y-\frac{\sqrt{3}}{2}\right)^{2}\geq\delta^{2}\quad\Longrightarrow\quad E(\Gamma_{x,y},3)-E(T_{1},3)\geq\frac{\delta^{2}}{2}. ( x − 2 1 ) 2 + ( y − 2 3 ) 2 ≥ δ 2 ⟹ E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ≥ 2 δ 2 . Proof. Poisson summation and Lemma 3.1 with n = 2 n=2 n = 2 give
E ( Γ x , y , 3 ) = 2 ζ ( 6 ) y 3 + 3 π 4 ζ ( 5 ) y − 2 + 2 π 3 ∑ n = 1 ∞ n 2 ( ∑ d ∈ Z ≥ 1 d ∣ n d − 5 ) e − 2 π n y ( 1 + 3 2 π n y + 3 ( 2 π n y ) 2 ) cos ( 2 π n x ) . E(\Gamma_{x,y},3)=2\zeta(6)y^{3}+\frac{3\pi}{4}\zeta(5)y^{-2}
+2\pi^{3}\sum_{n=1}^{\infty}n^{2}\left(\sum_{\substack{d\in\mathbb{Z}_{\geq1}\\ d\mid n}}d^{-5}\right)e^{-2\pi ny}\left(1+\frac{3}{2\pi ny}+\frac{3}{(2\pi ny)^{2}}\right)\cos(2\pi nx). E ( Γ x , y , 3 ) = 2 ζ ( 6 ) y 3 + 4 3 π ζ ( 5 ) y − 2 + 2 π 3 n = 1 ∑ ∞ n 2 d ∈ Z ≥ 1 d ∣ n ∑ d − 5 e − 2 π n y ( 1 + 2 π n y 3 + ( 2 π n y ) 2 3 ) cos ( 2 π n x ) . The Fourier series and its derivatives converge absolutely and uniformly for y ≥ 3 / 2 y\geq\sqrt{3}/2 y ≥ 3 /2 . Here e − 2 π y < 1 / 220 < 1 / 200 e^{-2\pi y}<1/220<1/200 e − 2 π y < 1/220 < 1/200 , and
∑ d ∈ Z ≥ 1 d ∣ n d − 5 < ζ ( 5 ) < 26 25 , 26 25 ∑ n = 2 ∞ n 4 ( 1 / 200 ) n − 1 < 1 10 . \sum_{\substack{d\in\mathbb{Z}_{\geq1}\\ d\mid n}}d^{-5}<\zeta(5)<\frac{26}{25},\qquad\frac{26}{25}\sum_{n=2}^{\infty}n^{4}(1/200)^{n-1}<\frac{1}{10}. d ∈ Z ≥ 1 d ∣ n ∑ d − 5 < ζ ( 5 ) < 25 26 , 25 26 n = 2 ∑ ∞ n 4 ( 1/200 ) n − 1 < 10 1 . The inequalities ∣ sin ( n θ ) ∣ ≤ n sin θ |\sin(n\theta)|\leq n\sin\theta ∣ sin ( n θ ) ∣ ≤ n sin θ (0 ≤ θ ≤ π 0\leq\theta\leq\pi 0 ≤ θ ≤ π ) and ( 1 + 3 / t + 3 / t 2 ) ′ < 0 (1+3/t+3/t^{2})'<0 ( 1 + 3/ t + 3/ t 2 ) ′ < 0 (t > 0 t>0 t > 0 ) give
− ∂ x E ( Γ x , y , 3 ) ≥ 18 5 π 4 e − 2 π y sin ( 2 π x ) . -\partial_{x}E(\Gamma_{x,y},3)\geq\frac{18}{5}\pi^{4}e^{-2\pi y}\sin(2\pi x). − ∂ x E ( Γ x , y , 3 ) ≥ 5 18 π 4 e − 2 π y sin ( 2 π x ) . Integration from x x x to 1 / 2 1/2 1/2 yields
E ( Γ x , y , 3 ) − E ( Γ 1 / 2 , y , 3 ) ≥ 9 5 π 3 e − 2 π y ( 1 + cos ( 2 π x ) ) . E(\Gamma_{x,y},3)-E(\Gamma_{1/2,y},3)\geq\frac{9}{5}\pi^{3}e^{-2\pi y}(1+\cos(2\pi x)). E ( Γ x , y , 3 ) − E ( Γ 1/2 , y , 3 ) ≥ 5 9 π 3 e − 2 π y ( 1 + cos ( 2 π x )) .
Twice differentiating and using cos ( 2 π n x ) ≥ − 1 \cos(2\pi n x) \ge-1 cos ( 2 π n x ) ≥ − 1 gives
∂ y y E ( Γ x , y , 3 ) ≥ 12 ζ ( 6 ) y + 9 π 2 ζ ( 5 ) y − 4 − 8 π 5 ζ ( 5 ) ∑ n = 1 ∞ n 4 e − 2 π n y ( 1 + 3 2 π n y + 9 ( 2 π n y ) 2 + 18 ( 2 π n y ) 3 + 18 ( 2 π n y ) 4 ) . \partial_{yy}E(\Gamma_{x,y},3) \ge12\zeta(6)y+\frac{9\pi}{2}\zeta(5)y^{-4}
-8\pi^{5}\zeta(5)\sum_{n=1}^{\infty} n^{4}e^{-2\pi n y}\left(1+\frac{3}{2\pi n y}+\frac{9}{(2\pi n y)^{2}}+\frac{18}{(2\pi n y)^{3}}+\frac{18}{(2\pi n y)^{4}}\right). ∂ y y E ( Γ x , y , 3 ) ≥ 12 ζ ( 6 ) y + 2 9 π ζ ( 5 ) y − 4 − 8 π 5 ζ ( 5 ) n = 1 ∑ ∞ n 4 e − 2 π n y ( 1 + 2 π n y 3 + ( 2 π n y ) 2 9 + ( 2 π n y ) 3 18 + ( 2 π n y ) 4 18 ) .
Since
d d t { e − t ( t 4 + 3 t 3 + 9 t 2 + 18 t + 18 ) } = − e − t t 3 ( t − 1 ) ≤ 0 ( t ≥ 1 ) , \frac{\mathrm{d}}{\mathrm{d}t}\left\{e^{-t}(t^{4}+3t^{3}+9t^{2}+18t+18)\right\}=-e^{-t}t^{3}(t-1)\le0\qquad(t\ge1), d t d { e − t ( t 4 + 3 t 3 + 9 t 2 + 18 t + 18 ) } = − e − t t 3 ( t − 1 ) ≤ 0 ( t ≥ 1 ) ,
the negative Fourier part, divided by ( 9 π / 2 ) ζ ( 5 ) y − 4 (9\pi/2)\zeta(5)y^{-4} ( 9 π /2 ) ζ ( 5 ) y − 4 , has absolute value at most
1 9 ∑ n = 1 ∞ 2 20 − n [ ( 27 n 5 ) 4 + 3 ( 27 n 5 ) 3 + 9 ( 27 n 5 ) 2 + 18 ( 27 n 5 ) + 18 ] < 1. \frac{1}{9}\sum_{n=1}^{\infty}2^{20-n}\left[\left(\frac{27n}{5}\right)^{4}+3\left(\frac{27n}{5}\right)^{3}+9\left(\frac{27n}{5}\right)^{2}+18\left(\frac{27n}{5}\right)+18\right]<1. 9 1 n = 1 ∑ ∞ 2 20 − n [ ( 5 27 n ) 4 + 3 ( 5 27 n ) 3 + 9 ( 5 27 n ) 2 + 18 ( 5 27 n ) + 18 ] < 1.
Thus ∂ y y E ( Γ x , y , 3 ) ≥ 12 y ≥ 6 3 \partial_{yy}E(\Gamma_{x,y},3)\ge12y\ge6\sqrt{3} ∂ y y E ( Γ x , y , 3 ) ≥ 12 y ≥ 6 3 . Modular symmetry gives ∂ y E ( Γ 1 / 2 , y , 3 ) ∣ y = 3 / 2 = 0 \left.\partial_yE(\Gamma_{1/2,y},3)\right|_{y=\sqrt{3}/2}=0 ∂ y E ( Γ 1/2 , y , 3 ) y = 3 /2 = 0 ; integrating twice from 3 / 2 \sqrt{3}/2 3 /2 yields
E ( Γ 1 / 2 , y , 3 ) − E ( T 1 , 3 ) ≥ 3 3 ( y − 3 2 ) 2 . E(\Gamma_{1/2,y},3)-E(T_1,3)\ge3\sqrt{3}\left(y-\frac{\sqrt{3}}{2}\right)^2. E ( Γ 1/2 , y , 3 ) − E ( T 1 , 3 ) ≥ 3 3 ( y − 2 3 ) 2 .
For 3 / 2 ≤ y ≤ 1 \sqrt{3}/2\le y\le1 3 /2 ≤ y ≤ 1 , 1 + cos ( 2 π x ) ≥ 8 ( x − 1 / 2 ) 2 1+\cos(2\pi x)\ge8(x-1/2)^2 1 + cos ( 2 π x ) ≥ 8 ( x − 1/2 ) 2 gives
9 5 π 3 e − 2 π y ( 1 + cos ( 2 π x ) ) ≥ 1 2 ( x − 1 / 2 ) 2 . \frac{9}{5}\pi^{3}e^{-2\pi y}(1+\cos(2\pi x))\ge\frac{1}{2}(x-1/2)^2. 5 9 π 3 e − 2 π y ( 1 + cos ( 2 π x )) ≥ 2 1 ( x − 1/2 ) 2 .
Hence the lower bound is ≥ 1 2 [ ( x − 1 / 2 ) 2 + ( y − 3 / 2 ) 2 ] \ge\frac{1}{2}[(x-1/2)^2+(y-\sqrt{3}/2)^2] ≥ 2 1 [( x − 1/2 ) 2 + ( y − 3 /2 ) 2 ] . For y ≥ 1 y\ge1 y ≥ 1 , y − 3 / 2 > 1 / 8 y-\sqrt{3}/2>1/8 y − 3 /2 > 1/8 and ( x − 1 / 2 ) 2 ≤ 1 / 4 (x-1/2)^2\le1/4 ( x − 1/2 ) 2 ≤ 1/4 , so
3 3 ( y − 3 / 2 ) 2 > 5 ( y − 3 / 2 ) 2 ≥ 1 4 [ ( x − 1 / 2 ) 2 + ( y − 3 / 2 ) 2 ] . 3\sqrt{3}(y-\sqrt{3}/2)^2>5(y-\sqrt{3}/2)^2\ge\frac{1}{4}[(x-1/2)^2+(y-\sqrt{3}/2)^2]. 3 3 ( y − 3 /2 ) 2 > 5 ( y − 3 /2 ) 2 ≥ 4 1 [( x − 1/2 ) 2 + ( y − 3 /2 ) 2 ] .
This exceeds 5 / 64 > δ 2 / 2 5/64>\delta^2/2 5/64 > δ 2 /2 for δ ≤ 1 / 10 \delta\le1/10 δ ≤ 1/10 . This completes the proof of Lemma 3.3. □ \square □
Lemma 3.4 (Planar constants). With the notation of the planar triangular minimum,
S 3 < 13 2 , 9 3 26 < A ∗ 3 < 611 972 < ( 43 50 ) 3 , S 3 2 S 6 > 6 , E ( T 1 , 3 ) ≥ 9 3 4 . S_3<\frac{13}{2},\qquad\frac{9\sqrt{3}}{26}<A_*^3<\frac{611}{972}<\left(\frac{43}{50}\right)^3,\qquad\frac{S_3^2}{S_6}>6,\qquad E(T_1,3)\ge\frac{9\sqrt{3}}{4}. S 3 < 2 13 , 26 9 3 < A ∗ 3 < 972 611 < ( 50 43 ) 3 , S 6 S 3 2 > 6 , E ( T 1 , 3 ) ≥ 4 9 3 . Proof. For q = i 2 + i j + j 2 = 1 , 3 , 4 , 7 , 9 q=i^2+ij+j^2=1,3,4,7,9 q = i 2 + ij + j 2 = 1 , 3 , 4 , 7 , 9 , the multiplicities are 6 , 6 , 6 , 12 , 6 6,6,6,12,6 6 , 6 , 6 , 12 , 6 , respectively; disjoint radius-1 / 2 1/2 1/2 disks give
S 3 ≤ 6 + 6 27 + 6 64 + 12 343 + 6 729 + 2 ( 7 6 ) 6 ( 5 2 ) − 4 < 13 2 . S_3\le6+\frac{6}{27}+\frac{6}{64}+\frac{12}{343}+\frac{6}{729}+2\left(\frac{7}{6}\right)^6\left(\frac{5}{2}\right)^{-4}<\frac{13}{2}. S 3 ≤ 6 + 27 6 + 64 6 + 343 12 + 729 6 + 2 ( 6 7 ) 6 ( 2 5 ) − 4 < 2 13 . Since S 6 ≥ 6 S_6\ge6 S 6 ≥ 6 and the six radius-3 \sqrt{3} 3 vectors give S 3 − S 6 ≥ 52 / 243 S_3-S_6\ge52/243 S 3 − S 6 ≥ 52/243 ,
A ∗ 3 = 3 3 8 S 6 S 3 > 9 3 26 , S 6 S 3 < 3055 3159 . A_*^3=\frac{3\sqrt{3}}{8}\frac{S_6}{S_3}>\frac{9\sqrt{3}}{26},\qquad\frac{S_6}{S_3}<\frac{3055}{3159}. A ∗ 3 = 8 3 3 S 3 S 6 > 26 9 3 , S 3 S 6 < 3159 3055 . Hence A ∗ 3 < 611 / 972 < ( 43 / 50 ) 3 A_*^3<611/972<(43/50)^3 A ∗ 3 < 611/972 < ( 43/50 ) 3 , while S 3 2 / S 6 > S 3 > 6 S_3^2/S_6>S_3>6 S 3 2 / S 6 > S 3 > 6 and E ( T 1 , 3 ) = ( 3 3 / 8 ) S 3 ≥ 9 3 / 4 E(T_1,3)=(3\sqrt{3}/8)S_3\ge9\sqrt{3}/4 E ( T 1 , 3 ) = ( 3 3 /8 ) S 3 ≥ 9 3 /4 . This completes the proof of Lemma 3.4. □ \square □
Along triangular planes,
U ( A ) : = E ( T A ) = S 6 2 ( 3 2 A ) 6 − S 3 ( 3 2 A ) 3 = e 2 + S 3 2 2 S 6 [ ( A ∗ / A ) 3 − 1 ] 2 . (28) U(A):=\mathcal{E}(T_A)=\frac{S_6}{2}\left(\frac{\sqrt{3}}{2A}\right)^6-S_3\left(\frac{\sqrt{3}}{2A}\right)^3=e_2+\frac{S_3^2}{2S_6}\left[(A_*/A)^3-1\right]^2. \tag*{(28)} U ( A ) := E ( T A ) = 2 S 6 ( 2 A 3 ) 6 − S 3 ( 2 A 3 ) 3 = e 2 + 2 S 6 S 3 2 [ ( A ∗ / A ) 3 − 1 ] 2 . (28)
Since 6 < S 6 < S 3 < 13 / 2 6<S_6<S_3<13/2 6 < S 6 < S 3 < 13/2 , we have 6 < S 3 2 / S 6 < 11 6<S_3^2/S_6<11 6 < S 3 2 / S 6 < 11 .
Lemma 3.5 (Compression to planar area one). For a planar lattice Γ ⊂ R 2 \Gamma\subset\mathbb{R}^{2} Γ ⊂ R 2 ,
T 1 = 2 3 ( 1 1 / 2 0 3 / 2 ) Z 2 , ∣ T 1 ∣ = 1 , T_{1}=\sqrt{\frac{2}{\sqrt{3}}}\begin{pmatrix}1 & 1/2 \\ 0 & \sqrt{3}/2\end{pmatrix}\mathbb{Z}^{2}, \qquad|T_{1}|=1, T 1 = 3 2 ( 1 0 1/2 3 /2 ) Z 2 , ∣ T 1 ∣ = 1 , A = ∣ Γ ∣ ≥ 1 ⟹ E ( Γ ) ≥ E ( T 1 ) . A=|\Gamma|\ge1 \quad\Longrightarrow\quad\mathcal{E}(\Gamma)\ge\mathcal{E}(T_{1}). A = ∣Γ∣ ≥ 1 ⟹ E ( Γ ) ≥ E ( T 1 ) .
Proof. Choose a basis ( a , 0 ) , ( b , h ) (a,0),(b,h) ( a , 0 ) , ( b , h ) with a = min v ∈ Γ ∖ { 0 } ∣ v ∣ a=\min_{v\in\Gamma\setminus\{0\}}|v| a = min v ∈ Γ ∖ { 0 } ∣ v ∣ and a h = A ah=A ah = A . For a ≤ 1 a\le1 a ≤ 1 , decrease h h h to 1 / a 1/a 1/ a ; changed vectors retain vertical component ≥ 1 / a ≥ 1 \ge1/a\ge1 ≥ 1/ a ≥ 1 , and
d d r ( r − 12 − 2 r − 6 ) = 12 r − 13 ( r 6 − 1 ) ≥ 0 ( r ≥ 1 ) . \frac{\mathrm{d}}{\mathrm{d}r}(r^{-12}-2r^{-6})=12r^{-13}(r^{6}-1)\ge0 \qquad(r\ge1). d r d ( r − 12 − 2 r − 6 ) = 12 r − 13 ( r 6 − 1 ) ≥ 0 ( r ≥ 1 ) .
For a > 1 a>1 a > 1 , scale down until A = 1 A=1 A = 1 or a = 1 a=1 a = 1 , then compress vertically if needed. Changed lengths remain ≥ 1 \ge1 ≥ 1 . At area one, (14) gives E ( Γ ) ≥ E ( T 1 ) \mathcal{E}(\Gamma)\ge\mathcal{E}(T_{1}) E ( Γ ) ≥ E ( T 1 ) . This completes the proof of Lemma 3.5. □
Write the layered lattice as
L = Γ + Z ( u + h e 3 ) , A = ∣ Γ ∣ , h = V / A , u ∈ R 2 . L=\Gamma+\mathbb{Z}(u+he_{3}), \qquad A=|\Gamma|,\quad h=V/A,\quad u\in\mathbb{R}^{2}. L = Γ + Z ( u + h e 3 ) , A = ∣Γ∣ , h = V / A , u ∈ R 2 .
As in (5) , E ( Γ ) = 1 2 E ( Γ , 6 ) − E ( Γ , 3 ) \mathcal{E}(\Gamma)=\frac{1}{2}E(\Gamma,6)-E(\Gamma,3) E ( Γ ) = 2 1 E ( Γ , 6 ) − E ( Γ , 3 ) . Its reciprocal lattice is
Γ ∗ = { w ∈ R 2 : w ⋅ v ∈ Z for every v ∈ Γ } . \Gamma^{*}=\{w\in\mathbb{R}^{2}:w\cdot v\in\mathbb{Z}\text{ for every }v\in\Gamma\}. Γ ∗ = { w ∈ R 2 : w ⋅ v ∈ Z for every v ∈ Γ } .
For d > 0 d>0 d > 0 define
f d ( x ) = ( ∣ x ∣ 2 + d 2 ) − 6 − 2 ( ∣ x ∣ 2 + d 2 ) − 3 , x ∈ R 2 . f_{d}(x)=(|x|^{2}+d^{2})^{-6}-2(|x|^{2}+d^{2})^{-3}, \qquad x\in\mathbb{R}^{2}. f d ( x ) = ( ∣ x ∣ 2 + d 2 ) − 6 − 2 ( ∣ x ∣ 2 + d 2 ) − 3 , x ∈ R 2 .
We use the Fourier transform with kernel e − 2 π i w ⋅ x e^{-2\pi i w\cdot x} e − 2 π i w ⋅ x . The Gaussian Laplace integral gives, for s > 1 s>1 s > 1 ,
( ∣ x ∣ 2 + d 2 ) − s ( w ) = π Γ ( s ) ∫ 0 ∞ t s − 2 e − d 2 t − π 2 ∣ w ∣ 2 / t d t . (29) (|x|^{2}+d^{2})^{-s}(w)=\frac{\pi}{\Gamma(s)}\int_{0}^{\infty}t^{s-2}e^{-d^{2}t-\pi^{2}|w|^{2}/t}\,\mathrm{d}t. \tag*{(29)} ( ∣ x ∣ 2 + d 2 ) − s ( w ) = Γ ( s ) π ∫ 0 ∞ t s − 2 e − d 2 t − π 2 ∣ w ∣ 2 / t d t . (29)
At w = 0 w=0 w = 0 the value is π d 2 − 2 s / ( s − 1 ) \pi d^{2-2s}/(s-1) π d 2 − 2 s / ( s − 1 ) ; for r = ∣ w ∣ > 0 r=|w|>0 r = ∣ w ∣ > 0 , Lemma 3.1 gives
( ∣ x ∣ 2 + d 2 ) − s ( w ) = 2 π s Γ ( s ) ( r / d ) s − 1 K s − 1 ( 2 π d r ) . (|x|^{2}+d^{2})^{-s}(w)=\frac{2\pi^{s}}{\Gamma(s)}(r/d)^{s-1}K_{s-1}(2\pi dr). ( ∣ x ∣ 2 + d 2 ) − s ( w ) = Γ ( s ) 2 π s ( r / d ) s − 1 K s − 1 ( 2 π d r ) .
The layers m m m and − m -m − m combine to give
E ( L ) = E ( Γ ) + 1 A ∑ m = 1 ∞ ∑ w ∈ Γ ∗ f ^ m h ( w ) cos ( 2 π m w ⋅ u ) . (30) \mathcal{E}(L)=\mathcal{E}(\Gamma)+\frac{1}{A}\sum_{m=1}^{\infty}\sum_{w\in\Gamma^{*}}\widehat{f}_{mh}(w)\cos(2\pi m w\cdot u). \tag*{(30)} E ( L ) = E ( Γ ) + A 1 m = 1 ∑ ∞ w ∈ Γ ∗ ∑ f mh ( w ) cos ( 2 π m w ⋅ u ) . (30)
The zero-frequency terms are
− π ζ ( 4 ) A h 4 + π ζ ( 10 ) 5 A h 10 = − π ζ ( 4 ) A 3 V 4 + π ζ ( 10 ) 5 A 9 V 10 . (31) -\frac{\pi\zeta(4)}{Ah^{4}}+\frac{\pi\zeta(10)}{5Ah^{10}}=-\pi\zeta(4)\frac{A^{3}}{V^{4}}+\frac{\pi\zeta(10)}{5}\frac{A^{9}}{V^{10}}. \tag*{(31)} − A h 4 π ζ ( 4 ) + 5 A h 10 π ζ ( 10 ) = − π ζ ( 4 ) V 4 A 3 + 5 π ζ ( 10 ) V 10 A 9 . (31)
For Γ \Gamma Γ in a compact neighborhood of T A ∗ T_{A_*} T A ∗ , there is ρ > 0 \rho>0 ρ > 0 such that
λ 1 ( Γ ) , λ 1 ( Γ ∗ ) ≥ ρ , # ( Γ ∩ B R ) + # ( Γ ∗ ∩ B R ) ≤ C ( 1 + R ) 2 . \lambda_{1}(\Gamma),\lambda_{1}(\Gamma^{*})\ge\rho,\qquad\#(\Gamma\cap B_{R})+\#(\Gamma^{*}\cap B_{R})\le C(1+R)^{2}. λ 1 ( Γ ) , λ 1 ( Γ ∗ ) ≥ ρ , # ( Γ ∩ B R ) + # ( Γ ∗ ∩ B R ) ≤ C ( 1 + R ) 2 .
Choose a smooth basis and lattice coordinates for u u u . For any fixed mixed partial derivative ∂ β \partial^{\beta} ∂ β in the area and shape coordinates, (29) and (27) give, for h ≥ 1 h\ge1 h ≥ 1 ,
∣ ∂ β { A − 1 f ^ m h ( w ) cos ( 2 π m w ⋅ u ) } ∣ ≤ C β ( 1 + m h + ∣ w ∣ ) N β e − 2 π m h ∣ w ∣ . \left|\partial^{\beta}\left\{A^{-1}\widehat{f}_{mh}(w)\cos(2\pi m w\cdot u)\right\}\right|\le C_{\beta}(1+mh+|w|)^{N_{\beta}}e^{-2\pi mh|w|}. ∂ β { A − 1 f mh ( w ) cos ( 2 π m w ⋅ u ) } ≤ C β ( 1 + mh + ∣ w ∣ ) N β e − 2 π mh ∣ w ∣ .
Quadratic counting gives a uniform O ( e − c h ) O(e^{-ch}) O ( e − c h ) sum over m ≥ 1 m\ge1 m ≥ 1 , w ≠ 0 w\ne0 w = 0 , with any fixed number of derivatives (after decreasing c > 0 c>0 c > 0 ).
Lemma 3.6 (Planar and cross-layer energy bounds). For a lattice L ⊂ R 3 L\subset\mathbb{R}^{3} L ⊂ R 3 , choose
v 1 ∈ argmin v ∈ L ∖ { 0 } ∣ v ∣ , v 2 ∈ argmin v ∈ L ∖ R v 1 ∣ v ∣ , v_{1}\in\operatorname*{argmin}_{v\in L\setminus\{0\}}|v|,\qquad v_{2}\in\operatorname*{argmin}_{v\in L\setminus\mathbb{R}v_{1}}|v|, v 1 ∈ v ∈ L ∖ { 0 } argmin ∣ v ∣ , v 2 ∈ v ∈ L ∖ R v 1 argmin ∣ v ∣ , Γ = L ∩ span R { v 1 , v 2 } , \Gamma=L\cap\operatorname{span}_{\mathbb{R}}\{v_{1},v_{2}\}, Γ = L ∩ span R { v 1 , v 2 } , a = ∣ v 1 ∣ , b = ∣ v 2 ∣ , A = ∣ Γ ∣ , h = V / A , V = ∣ L ∣ . a=|v_{1}|,\qquad b=|v_{2}|,\qquad A=|\Gamma|,\qquad h=V/A,\qquad V=|L|. a = ∣ v 1 ∣ , b = ∣ v 2 ∣ , A = ∣Γ∣ , h = V / A , V = ∣ L ∣.
If E ( L ) ≤ e 2 \mathcal{E}(L) \le e_2 E ( L ) ≤ e 2 , the short-plane bounds give 7 / 10 < a < 8 / 5 7/10 < a < 8/5 7/10 < a < 8/5 and A < 5 A < 5 A < 5 . Moreover
h 2 ≥ 3 4 b 2 − 1 4 a 2 > 75 256 A 2 − 32 25 . h^2 \ge\frac{3}{4}b^2-\frac{1}{4}a^2 > \frac{75}{256}A^2-\frac{32}{25}. h 2 ≥ 4 3 b 2 − 4 1 a 2 > 256 75 A 2 − 25 32 .
The planar energy satisfies
E ( Γ ) ≥ a − 12 − 11 5 a − 6 − 11 5 a 6 A 6 − 363 140 a 4 A 5 , E ( Γ ) > − 21 20 − 7 A 3 . \mathcal{E}(\Gamma) \ge a^{-12}-\frac{11}{5}a^{-6}-\frac{11}{5}\frac{a^6}{A^6}-\frac{363}{140}\frac{a^4}{A^5}, \qquad\mathcal{E}(\Gamma)>-\frac{21}{20}-\frac{7}{A^3}. E ( Γ ) ≥ a − 12 − 5 11 a − 6 − 5 11 A 6 a 6 − 140 363 A 5 a 4 , E ( Γ ) > − 20 21 − A 3 7 .
The full cross-layer energy, with any layer shift, satisfies
E ( L ) − E ( Γ ) ≥ − 11 5 h 6 − 363 140 ( 1 / a + a / A ) h − 5 − 121 35 A h − 4 ≥ − J ( A , h ) , \mathcal{E}(L)-\mathcal{E}(\Gamma) \ge-\frac{11}{5h^6}-\frac{363}{140}(1/a+a/A)h^{-5}-\frac{121}{35A}h^{-4} \ge-J(A,h), E ( L ) − E ( Γ ) ≥ − 5 h 6 11 − 140 363 ( 1/ a + a / A ) h − 5 − 35 A 121 h − 4 ≥ − J ( A , h ) ,
where
J ( A , h ) = 11 5 h 6 + 363 140 ( 10 7 + 8 5 A ) h − 5 + 121 35 A h − 4 . J(A,h)=\frac{11}{5h^6}+\frac{363}{140}\left(\frac{10}{7}+\frac{8}{5A}\right)h^{-5}+\frac{121}{35A}h^{-4}. J ( A , h ) = 5 h 6 11 + 140 363 ( 7 10 + 5 A 8 ) h − 5 + 35 A 121 h − 4 .
Proof. Horizontal reduction gives
[ Γ : Z v 1 + Z v 2 ] > 1 ⟹ ∃ v ∈ Γ ∖ R v 1 : ∣ v ∣ 2 ≤ a 2 + b 2 4 < b 2 , [\Gamma:\mathbb{Z}v_1+\mathbb{Z}v_2]>1 \Longrightarrow\exists v\in\Gamma\setminus\mathbb{R}v_1: |v|^2\le\frac{a^2+b^2}{4}<b^2, [ Γ : Z v 1 + Z v 2 ] > 1 ⟹ ∃ v ∈ Γ ∖ R v 1 : ∣ v ∣ 2 ≤ 4 a 2 + b 2 < b 2 ,
contradicting the choice of v 2 v_2 v 2 . Choose its sign so that 0 ≤ t = v 1 ⋅ v 2 ≤ a 2 / 2 0\le t=v_1\cdot v_2\le a^2/2 0 ≤ t = v 1 ⋅ v 2 ≤ a 2 /2 . The two nonobtuse triangles in the reduced cell have common circumradius
R 2 = a 2 b 2 ( a 2 + b 2 − 2 t ) 4 ( a 2 b 2 − t 2 ) ≤ a 2 + b 2 4 . R^2=\frac{a^2b^2(a^2+b^2-2t)}{4(a^2b^2-t^2)}\le\frac{a^2+b^2}{4}. R 2 = 4 ( a 2 b 2 − t 2 ) a 2 b 2 ( a 2 + b 2 − 2 t ) ≤ 4 a 2 + b 2 .
Their radius-R R R vertex disks cover the cell; reducing a height-h h h vector modulo Γ \Gamma Γ gives b 2 ≤ h 2 + R 2 b^2\le h^2+R^2 b 2 ≤ h 2 + R 2 .
Thus
h 2 ≥ b 2 − R 2 ≥ 3 4 b 2 − 1 4 a 2 ≥ 3 A 2 4 a 2 − a 2 4 > 75 256 A 2 − 32 25 ( b ≥ A / a , a < 8 / 5 ) . h^2\ge b^2-R^2\ge\frac{3}{4}b^2-\frac{1}{4}a^2\ge\frac{3A^2}{4a^2}-\frac{a^2}{4}>\frac{75}{256}A^2-\frac{32}{25}\qquad(b\ge A/a,\ a<8/5). h 2 ≥ b 2 − R 2 ≥ 4 3 b 2 − 4 1 a 2 ≥ 4 a 2 3 A 2 − 4 a 2 > 256 75 A 2 − 25 32 ( b ≥ A / a , a < 8/5 ) .
For any c , z > 0 c,z>0 c , z > 0 and any shift t t t , positivity of the Fourier coefficients and monotonicity on the positive half-line give
∑ k ∈ Z ( ( k c + t ) 2 + z 2 ) − 3 ≤ z − 6 + 3 π 8 c z − 5 , \sum_{k\in\mathbb{Z}}\left((kc+t)^2+z^2\right)^{-3}\le z^{-6}+\frac{3\pi}{8c}z^{-5}, k ∈ Z ∑ ( ( k c + t ) 2 + z 2 ) − 3 ≤ z − 6 + 8 c 3 π z − 5 ,
∑ k ∈ Z ( ( k c + t ) 2 + z 2 ) − 5 / 2 ≤ z − 5 + 4 3 c z − 4 . \sum_{k\in\mathbb{Z}}\left((kc+t)^2+z^2\right)^{-5/2}\le z^{-5}+\frac{4}{3c}z^{-4}. k ∈ Z ∑ ( ( k c + t ) 2 + z 2 ) − 5/2 ≤ z − 5 + 3 c 4 z − 4 .
Applying these along ( a , 0 ) (a,0) ( a , 0 ) and the spacing A / a A/a A / a gives
∑ p ∈ Γ ( ∣ p + t ∣ 2 + z 2 ) − 3 ≤ z − 6 + 3 π 8 ( 1 / a + a / A ) z − 5 + π 2 A z − 4 . \sum_{p\in\Gamma}\left(|p+t|^2+z^2\right)^{-3}\le z^{-6}+\frac{3\pi}{8}(1/a+a/A)z^{-5}+\frac{\pi}{2A}z^{-4}. p ∈ Γ ∑ ( ∣ p + t ∣ 2 + z 2 ) − 3 ≤ z − 6 + 8 3 π ( 1/ a + a / A ) z − 5 + 2 A π z − 4 .
Sum the nonzero layers and omit repulsion; π < 22 / 7 \pi<22/7 π < 22/7 and ζ ( 4 ) , ζ ( 5 ) , ζ ( 6 ) < 11 / 10 \zeta(4),\zeta(5),\zeta(6)<11/10 ζ ( 4 ) , ζ ( 5 ) , ζ ( 6 ) < 11/10 give J J J . Separating the plane itself into horizontal chains gives
E ( Γ ) ≥ E ( a Z ) − 2 ζ ( 6 ) ( a / A ) 6 − 3 π 4 ζ ( 5 ) a 4 / A 5 . \mathcal{E}(\Gamma)\ge\mathcal{E}(a\mathbb{Z})-2\zeta(6)(a/A)^6-\frac{3\pi}{4}\zeta(5)a^4/A^5. E ( Γ ) ≥ E ( a Z ) − 2 ζ ( 6 ) ( a / A ) 6 − 4 3 π ζ ( 5 ) a 4 / A 5 .
Use E ( a Z ) ≥ a − 12 − ( 11 / 5 ) a − 6 \mathcal{E}(a\mathbb{Z})\ge a^{-12}-(11/5)a^{-6} E ( a Z ) ≥ a − 12 − ( 11/5 ) a − 6 , or E ( a Z ) ≥ e 1 > − 21 / 20 \mathcal{E}(a\mathbb{Z})\ge e_1>-21/20 E ( a Z ) ≥ e 1 > − 21/20 with a 2 ≤ A / ( 3 / 2 ) a^2\le A/(\sqrt{3}/2) a 2 ≤ A / ( 3 /2 ) , respectively;
2 ζ ( 6 ) ( 3 / 2 ) 3 + π ζ ( 5 ) < 7. \frac{2\zeta(6)}{(\sqrt{3}/2)^3}+\pi\zeta(5)<7. ( 3 /2 ) 3 2 ζ ( 6 ) + π ζ ( 5 ) < 7.
This completes the proof of Lemma 3.6 . □
Lemma 3.7 (A fixed-systole planar maximum). Let Γ ⊂ R 2 \Gamma\subset\mathbb{R}^{2} Γ ⊂ R 2 be a planar lattice with
min v ∈ Γ \ { 0 } ∣ v ∣ = ρ > 0 , \min_{v\in\Gamma\backslash\{0\}} |v|=\rho>0, v ∈ Γ\ { 0 } min ∣ v ∣ = ρ > 0 , and let f : ( 0 , ∞ ) → ( 0 , ∞ ) f:(0,\infty)\to(0,\infty) f : ( 0 , ∞ ) → ( 0 , ∞ ) be decreasing and convex. If the following sums converge, then
∑ v ∈ Γ \ { 0 } f ( ∣ v ∣ 2 ) ≤ ∑ ( m , n ) ∈ Z 2 \ { ( 0 , 0 ) } f ( ρ 2 ( m 2 + m n + n 2 ) ) . \sum_{v\in\Gamma\backslash\{0\}}f(|v|^{2})\leq\sum_{(m,n)\in\mathbb{Z}^{2}\backslash\{(0,0)\}}f\left(\rho^{2}(m^{2}+mn+n^{2})\right). v ∈ Γ\ { 0 } ∑ f ( ∣ v ∣ 2 ) ≤ ( m , n ) ∈ Z 2 \ {( 0 , 0 )} ∑ f ( ρ 2 ( m 2 + mn + n 2 ) ) . Proof. Gauss reduction gives the Gram matrix
ρ 2 ( 1 b b c ) , ∣ b ∣ ≤ 1 2 , c ≥ 1. \rho^{2}\begin{pmatrix}1&b\\b&c\end{pmatrix},\qquad|b|\leq\frac{1}{2},\quad c\geq1. ρ 2 ( 1 b b c ) , ∣ b ∣ ≤ 2 1 , c ≥ 1. Since f f f decreases,
∑ ( m , n ) ∈ Z 2 \ { ( 0 , 0 ) } f ( ρ 2 [ m 2 + 2 b m n + c n 2 ] ) ≤ ∑ ( m , n ) ∈ Z 2 \ { ( 0 , 0 ) } f ( ρ 2 [ m 2 + 2 b m n + n 2 ] ) . \sum_{(m,n)\in\mathbb{Z}^{2}\backslash\{(0,0)\}}f\left(\rho^{2}[m^{2}+2bmn+cn^{2}]\right)\leq\sum_{(m,n)\in\mathbb{Z}^{2}\backslash\{(0,0)\}}f\left(\rho^{2}[m^{2}+2bmn+n^{2}]\right). ( m , n ) ∈ Z 2 \ {( 0 , 0 )} ∑ f ( ρ 2 [ m 2 + 2 bmn + c n 2 ] ) ≤ ( m , n ) ∈ Z 2 \ {( 0 , 0 )} ∑ f ( ρ 2 [ m 2 + 2 bmn + n 2 ] ) . Evenness and convexity in b b b give the maximum at b = ± 1 / 2 b=\pm1/2 b = ± 1/2 , the triangular Gram matrix. This completes the proof of Lemma 3.7 . □
Lemma 3.8 (A triangular majorant for arbitrary shifted layers). Let Γ ⊂ R 2 \Gamma\subset\mathbb{R}^{2} Γ ⊂ R 2 be a planar lattice, with A = ∣ Γ ∣ A=|\Gamma| A = ∣Γ∣ and a = min v ∈ Γ \ { 0 } ∣ v ∣ a=\min_{v\in\Gamma\backslash\{0\}}|v| a = min v ∈ Γ\ { 0 } ∣ v ∣ . Put ρ = a / A \rho=a/A ρ = a / A and q i , j = i 2 + i j + j 2 q_{i,j}=i^{2}+ij+j^{2} q i , j = i 2 + ij + j 2 . For any shift u u u and any d > 0 d>0 d > 0 ,
∑ p ∈ Γ ( ∣ p + u ∣ 2 + d 2 ) − 3 ≤ π 2 A d 4 + π 3 ρ 2 A d 2 ∑ ( i , j ) ∈ Z 2 \ { ( 0 , 0 ) } q i , j K 2 ( 2 π d ρ q i , j ) . \sum_{p\in\Gamma}\left(|p+u|^{2}+d^{2}\right)^{-3}\leq\frac{\pi}{2Ad^{4}}+\frac{\pi^{3}\rho^{2}}{Ad^{2}}\sum_{(i,j)\in\mathbb{Z}^{2}\backslash\{(0,0)\}}q_{i,j}K_{2}\left(2\pi d\rho\sqrt{q_{i,j}}\right). p ∈ Γ ∑ ( ∣ p + u ∣ 2 + d 2 ) − 3 ≤ 2 A d 4 π + A d 2 π 3 ρ 2 ( i , j ) ∈ Z 2 \ {( 0 , 0 )} ∑ q i , j K 2 ( 2 π d ρ q i , j ) . The cross-layer energy is therefore ≥ − J △ ( a , A , h ) \geq-J_{\triangle}(a,A,h) ≥ − J △ ( a , A , h ) , where
J △ ( a , A , h ) = π ζ ( 4 ) A h 4 + 2 π 3 ρ 2 A h 2 ∑ m = 1 ∞ 1 m 2 ∑ ( i , j ) ∈ Z 2 \ { ( 0 , 0 ) } q i , j K 2 ( 2 π m h ρ q i , j ) . J_{\triangle}(a,A,h)=\frac{\pi\zeta(4)}{Ah^{4}}+\frac{2\pi^{3}\rho^{2}}{Ah^{2}}\sum_{m=1}^{\infty}\frac{1}{m^{2}}\sum_{(i,j)\in\mathbb{Z}^{2}\backslash\{(0,0)\}}q_{i,j}K_{2}\left(2\pi mh\rho\sqrt{q_{i,j}}\right). J △ ( a , A , h ) = A h 4 π ζ ( 4 ) + A h 2 2 π 3 ρ 2 m = 1 ∑ ∞ m 2 1 ( i , j ) ∈ Z 2 \ {( 0 , 0 )} ∑ q i , j K 2 ( 2 π mh ρ q i , j ) . Proof. Positive Fourier coefficients give the maximum at u = 0 u=0 u = 0 , and Γ ∗ ≅ A − 1 Γ \Gamma^{*}\cong A^{-1}\Gamma Γ ∗ ≅ A − 1 Γ gives λ 1 ( Γ ∗ ) = ρ \lambda_{1}(\Gamma^{*})=\rho λ 1 ( Γ ∗ ) = ρ . The Fourier kernel
F 3 ( q , d ) = π 2 ∫ 0 ∞ t e − d 2 t − π 2 q / t d t F_{3}(q,d)=\frac{\pi}{2}\int_{0}^{\infty}t e^{-d^{2}t-\pi^{2}q/t}\,dt F 3 ( q , d ) = 2 π ∫ 0 ∞ t e − d 2 t − π 2 q / t d t satisfies ∂ q F 3 < 0 < ∂ q q F 3 \partial_{q}F_{3}<0<\partial_{qq}F_{3} ∂ q F 3 < 0 < ∂ q q F 3 for q > 0 q>0 q > 0 , and F 3 ( r 2 , d ) = π 3 r 2 d − 2 K 2 ( 2 π d r ) F_{3}(r^{2},d)=\pi^{3}r^{2}d^{-2}K_{2}(2\pi dr) F 3 ( r 2 , d ) = π 3 r 2 d − 2 K 2 ( 2 π d r ) . Lemma 3.7 , with zero mode π / ( 2 d 4 ) \pi/(2d^{4}) π / ( 2 d 4 ) , gives the first bound. Summing over d = m h d=mh d = mh and dropping repulsion gives the second. This completes the proof of Lemma 3.8 . □
Lemma 3.10 (An improved elementary planar lower bound). For a planar lattice Γ ⊂ R 2 \Gamma\subset\mathbb{R}^{2} Γ ⊂ R 2 , set A = ∣ Γ ∣ A=|\Gamma| A = ∣Γ∣ and a = min v ∈ Γ \ { 0 } ∣ v ∣ a=\min_{v\in\Gamma\backslash\{0\}}|v| a = min v ∈ Γ\ { 0 } ∣ v ∣ . Then
E ( Γ ) ≥ a − 12 − 11 5 a − 6 − 8349 2800 a 4 A 5 . \mathcal{E}(\Gamma)\geq a^{-12}-\frac{11}{5}a^{-6}-\frac{8349}{2800}\frac{a^{4}}{A^{5}}. E ( Γ ) ≥ a − 12 − 5 11 a − 6 − 2800 8349 A 5 a 4 .
If A ≥ 1 A \ge1 A ≥ 1 , two further lower bounds are
E ( Γ ) > − 947 320 , E ( Γ ) ≥ a − 12 − 13 2 a − 6 . \mathcal{E}(\Gamma)>-\frac{947}{320},\qquad\mathcal{E}(\Gamma)\ge a^{-12}-\frac{13}{2}a^{-6}. E ( Γ ) > − 320 947 , E ( Γ ) ≥ a − 12 − 2 13 a − 6 .
Proof. Choose a Gauss-reduced basis ( a , 0 ) , ( b , c ) (a,0),(b,c) ( a , 0 ) , ( b , c ) , with c = A / a c=A/a c = A / a and c / a ≥ 3 / 2 c/a\ge\sqrt{3}/2 c / a ≥ 3 /2 . One-dimensional Poisson summation gives, for s = 2 π z / a s=2\pi z/a s = 2 π z / a ,
∑ k ∈ Z ( ( k a + t ) 2 + z 2 ) − 3 ≤ 3 π 8 a z 5 [ 1 + 2 ∑ n = 1 ∞ e − n s ( 1 + n s + n 2 s 2 3 ) ] . \sum_{k\in\mathbb{Z}}\left((ka+t)^2+z^2\right)^{-3}\le\frac{3\pi}{8az^5}\left[1+2\sum_{n=1}^{\infty}e^{-ns}\left(1+ns+\frac{n^2s^2}{3}\right)\right]. k ∈ Z ∑ ( ( k a + t ) 2 + z 2 ) − 3 ≤ 8 a z 5 3 π [ 1 + 2 n = 1 ∑ ∞ e − n s ( 1 + n s + 3 n 2 s 2 ) ] .
Each summand in the correction decreases with s > 0 s>0 s > 0 . If z ≥ c z\ge c z ≥ c , then s ≥ π 3 > 27 / 5 s\ge\pi\sqrt{3}>27/5 s ≥ π 3 > 27/5 and e 27 / 5 > 220 e^{27/5}>220 e 27/5 > 220 . Thus the correction is at most
2 ∑ n = 1 ∞ 220 − n ( 1 + 27 n 5 + 243 n 2 25 ) < 3 20 . 2\sum_{n=1}^{\infty}220^{-n}\left(1+\frac{27n}{5}+\frac{243n^2}{25}\right)<\frac{3}{20}. 2 n = 1 ∑ ∞ 22 0 − n ( 1 + 5 27 n + 25 243 n 2 ) < 20 3 .
Summing over the nonzero planar rows gives
E ( Γ , 3 ) ≤ 2 ζ ( 6 ) a − 6 + 69 π ζ ( 5 ) 80 a 4 A 5 < 11 5 a − 6 + 8349 2800 a 4 A 5 . E(\Gamma,3)\le2\zeta(6)a^{-6}+\frac{69\pi\zeta(5)}{80}\frac{a^4}{A^5}<\frac{11}{5}a^{-6}+\frac{8349}{2800}\frac{a^4}{A^5}. E ( Γ , 3 ) ≤ 2 ζ ( 6 ) a − 6 + 80 69 π ζ ( 5 ) A 5 a 4 < 5 11 a − 6 + 2800 8349 A 5 a 4 .
Since 1 2 E ( Γ , 6 ) ≥ a − 12 \frac{1}{2}E(\Gamma,6)\ge a^{-12} 2 1 E ( Γ , 6 ) ≥ a − 12 , the first bound follows. For A ≥ 1 A\ge1 A ≥ 1 , Lemmas 3.5 and 3.4 give E ( Γ ) ≥ U ( 1 ) > − 947 / 320 \mathcal{E}(\Gamma)\ge U(1)>-947/320 E ( Γ ) ≥ U ( 1 ) > − 947/320 ; the fixed-systole bound E ( Γ , 3 ) ≤ S 3 a − 6 < ( 13 / 2 ) a − 6 E(\Gamma,3)\le S_3a^{-6}<(13/2)a^{-6} E ( Γ , 3 ) ≤ S 3 a − 6 < ( 13/2 ) a − 6 gives the last inequality. This completes the proof of Lemma 3.10. □
Lemma 3.11 (A rational planar benchmark). The planar minimum is
e 2 = min A > 0 E ( T A ) = E ( T A ∗ ) = − S 3 2 2 S 6 < − 337 100 . e_2=\min_{A>0}\mathcal{E}(T_A)=\mathcal{E}(T_{A_*})=-\frac{S_3^2}{2S_6}<-\frac{337}{100}. e 2 = A > 0 min E ( T A ) = E ( T A ∗ ) = − 2 S 6 S 3 2 < − 100 337 .
Proof. At nearest-neighbor distance 99 / 100 99/100 99/100 , every vector omitted from the following cutoff has length > 1 >1 > 1 and contributes negatively:
e 2 < ∑ ( i , j ) ∈ Z 2 0 < q = i 2 + i j + j 2 ≤ 16 [ ( 100 / 99 ) 12 2 q 6 − ( 100 / 99 ) 6 q 3 ] < − 337 100 . e_2<\sum_{\substack{(i,j)\in\mathbb{Z}^2\\0<q=i^2+ij+j^2\le16}}\left[\frac{(100/99)^{12}}{2q^6}-\frac{(100/99)^6}{q^3}\right]<-\frac{337}{100}. e 2 < ( i , j ) ∈ Z 2 0 < q = i 2 + ij + j 2 ≤ 16 ∑ [ 2 q 6 ( 100/99 ) 12 − q 3 ( 100/99 ) 6 ] < − 100 337 .
This completes the proof of Lemma 3.11. □
Lemma 3.12 (A common triangular reference bound). For V > 0 V>0 V > 0 ,
E ( T 17 / 20 ⊕ V 17 / 20 ) ≤ U ( 17 / 20 ) − π ζ ( 4 ) ( 17 / 20 ) 3 V 4 + ζ ( 10 ) 5 ( 49 / 100 ) 2 ( 17 / 20 V ) 10 ( 1 + ( 49 / 100 ) ( 17 / 20 ) V ) 12 . \mathcal{E}\left(T_{17/20}\oplus\frac{V}{17/20}\right)\le U(17/20)-\frac{\pi\zeta(4)(17/20)^3}{V^4}+\frac{\zeta(10)}{5(49/100)^2}\left(\frac{17/20}{V}\right)^{10}\left(1+\frac{(49/100)(17/20)}{V}\right)^{12}. E ( T 17/20 ⊕ 17/20 V ) ≤ U ( 17/20 ) − V 4 π ζ ( 4 ) ( 17/20 ) 3 + 5 ( 49/100 ) 2 ζ ( 10 ) ( V 17/20 ) 10 ( 1 + V ( 49/100 ) ( 17/20 ) ) 12 . Proof. Since λ 1 ( T 17 / 20 ) > 98 / 100 = 2 ( 49 / 100 ) \lambda_1(T_{17/20})>98/100=2(49/100) λ 1 ( T 17/20 ) > 98/100 = 2 ( 49/100 ) , integration over disjoint radius-( 49 / 100 ) (49/100) ( 49/100 ) disks gives
∑ p ∈ T 17 / 20 ( ∣ p ∣ 2 + d 2 ) − 6 ≤ d − 10 5 ( 49 / 100 ) 2 ( 1 + 49 / 100 d ) 12 . \sum_{p\in T_{17/20}}(|p|^2+d^2)^{-6}\le\frac{d^{-10}}{5(49/100)^2}\left(1+\frac{49/100}{d}\right)^{12}. p ∈ T 17/20 ∑ ( ∣ p ∣ 2 + d 2 ) − 6 ≤ 5 ( 49/100 ) 2 d − 10 ( 1 + d 49/100 ) 12 . Sum over d = m V / ( 17 / 20 ) d=mV/(17/20) d = mV / ( 17/20 ) , m ≥ 1 m\ge1 m ≥ 1 , and omit the nonpositive attractive Fourier terms, retaining − π ζ ( 4 ) ( 17 / 20 ) 3 / V 4 -\pi\zeta(4)(17/20)^3/V^4 − π ζ ( 4 ) ( 17/20 ) 3 / V 4 . This completes the proof of Lemma 3.12. □
Lemma 3.13 (The uniform triangular-dual lower envelope). Suppose
7 10 ≤ a ≤ 8 5 , 1 ≤ A ≤ 7 2 , V ≥ 2 , A ≥ 433 500 a 2 , ( V A ) 2 ≥ 3 A 2 4 a 2 − a 2 4 . \frac{7}{10}\le a\le\frac{8}{5},\qquad1\le A\le\frac{7}{2},\qquad V\ge2,\qquad A\ge\frac{433}{500}a^2,\qquad\left(\frac{V}{A}\right)^2\ge\frac{3A^2}{4a^2}-\frac{a^2}{4}. 10 7 ≤ a ≤ 5 8 , 1 ≤ A ≤ 2 7 , V ≥ 2 , A ≥ 500 433 a 2 , ( A V ) 2 ≥ 4 a 2 3 A 2 − 4 a 2 .
Put h = V / A h=V/A h = V / A and define
P ( a , A ) = max { a − 12 − 11 5 a − 6 − 8349 a 4 2800 A 5 , a − 12 − 13 2 a − 6 − 947 320 } . P(a,A)=\max\left\{a^{-12}-\frac{11}{5}a^{-6}-\frac{8349a^{4}}{2800A^{5}},\ a^{-12}-\frac{13}{2}a^{-6}-\frac{947}{320}\right\}. P ( a , A ) = max { a − 12 − 5 11 a − 6 − 2800 A 5 8349 a 4 , a − 12 − 2 13 a − 6 − 320 947 } .
With J △ J_{\triangle} J △ as in Lemma 3.8 ,
max { P ( a , A ) − J △ ( a , A , h ) , a − 12 − 2901 200 a − 6 } > − 10 3 > − 337 100 . \max\left\{P(a,A)-J_{\triangle}(a,A,h),\ a^{-12}-\frac{2901}{200}a^{-6}\right\}>-\frac{10}{3}>-\frac{337}{100}. max { P ( a , A ) − J △ ( a , A , h ) , a − 12 − 200 2901 a − 6 } > − 3 10 > − 100 337 .
Proof. Lemma 3.1 gives
π 3 r 2 d 2 K 2 ( 2 π d r ) < U ( r , d ) : = π 3 2 r 3 / 2 d 5 / 2 e − 2 π d r ( 1 + 3 2 π d r + 3 ( 2 π d r ) 2 ) . \frac{\pi^{3}r^{2}}{d^{2}}K_{2}(2\pi dr)<U(r,d):=\frac{\pi^{3}}{2}\frac{r^{3/2}}{d^{5/2}}e^{-2\pi dr}\left(1+\frac{3}{2\pi dr}+\frac{3}{(2\pi dr)^{2}}\right). d 2 π 3 r 2 K 2 ( 2 π d r ) < U ( r , d ) := 2 π 3 d 5/2 r 3/2 e − 2 π d r ( 1 + 2 π d r 3 + ( 2 π d r ) 2 3 ) .
The expression decreases in r , d r,d r , d : for z = 2 π d r z=2\pi dr z = 2 π d r ,
d d z { e − z ( z 3 / 2 + 3 z 1 / 2 + 3 z − 1 / 2 ) } = − e − z ( z 3 / 2 + 3 2 z 1 / 2 + 3 2 z − 1 / 2 + 3 2 z − 3 / 2 ) < 0. \frac{d}{dz}\left\{e^{-z}\left(z^{3/2}+3z^{1/2}+3z^{-1/2}\right)\right\}
=-e^{-z}\left(z^{3/2}+\frac{3}{2}z^{1/2}+\frac{3}{2}z^{-1/2}+\frac{3}{2}z^{-3/2}\right)<0. d z d { e − z ( z 3/2 + 3 z 1/2 + 3 z − 1/2 ) } = − e − z ( z 3/2 + 2 3 z 1/2 + 2 3 z − 1/2 + 2 3 z − 3/2 ) < 0.
Monotonicity in d d d follows termwise.
Put ρ = a / A \rho=a/A ρ = a / A . The hypotheses give
( h ρ ) 2 ≥ 3 4 − a 4 4 A 2 > ( 16 25 ) 2 . (h\rho)^{2}\geq\frac{3}{4}-\frac{a^{4}}{4A^{2}}>\left(\frac{16}{25}\right)^{2}. ( h ρ ) 2 ≥ 4 3 − 4 A 2 a 4 > ( 25 16 ) 2 .
Thus 2 π m h ρ q > 4 m q 2\pi mh\rho\sqrt{q}>4m\sqrt{q} 2 π mh ρ q > 4 m q , using π > 25 / 8 \pi>25/8 π > 25/8 . The factor 1 + 3 / z + 3 / z 2 1+3/z+3/z^{2} 1 + 3/ z + 3/ z 2 is less than 2 when z > 4 z>4 z > 4 . Furthermore
π 3 ρ 3 / 2 A h 5 / 2 = π 3 a 3 / 2 V 5 / 2 < 12. \frac{\pi^{3}\rho^{3/2}}{Ah^{5/2}}=\frac{\pi^{3}a^{3/2}}{V^{5/2}}<12. A h 5/2 π 3 ρ 3/2 = V 5/2 π 3 a 3/2 < 12.
Since min ( i , j ) ∈ Z 2 ∖ { ( 0 , 0 ) } ( i 2 + i j + j 2 ) = 1 \min_{(i,j)\in\mathbb{Z}^{2}\setminus\{(0,0)\}}(i^{2}+ij+j^{2})=1 min ( i , j ) ∈ Z 2 ∖ {( 0 , 0 )} ( i 2 + ij + j 2 ) = 1 , disjoint radius-1 / 2 1/2 1/2 disks give
# { ( i , j ) ∈ Z 2 : n ≤ i 2 + i j + j 2 < n + 1 } < 25 n 2 ( n ∈ Z ≥ 1 ) . \#\left\{(i,j)\in\mathbb{Z}^{2}:n\leq\sqrt{i^{2}+ij+j^{2}}<n+1\right\}<25n^{2}\qquad(n\in\mathbb{Z}_{\geq1}). # { ( i , j ) ∈ Z 2 : n ≤ i 2 + ij + j 2 < n + 1 } < 25 n 2 ( n ∈ Z ≥ 1 ) .
Here q = i 2 + i j + j 2 q=i^{2}+ij+j^{2} q = i 2 + ij + j 2 . Also q 3 / 4 < 3 n 3 / 2 q^{3/4}<3n^{3/2} q 3/4 < 3 n 3/2 . The complete tail m ≥ 6 m\geq6 m ≥ 6 or q > 36 q>36 q > 36 is < 1 / 1000 <1/1000 < 1/1000 : ∑ n = 1 ∞ n 4 501 − n < 2 \sum_{n=1}^{\infty}n^{4}501^{-n}<2 ∑ n = 1 ∞ n 4 50 1 − n < 2 , and the two regions have bounds 4 e − 24 4e^{-24} 4 e − 24 and 4 ⋅ 64 e − 24 4\cdot64e^{-24} 4 ⋅ 64 e − 24 , respectively; their overlap may be counted twice. Use the actual a , A , h a,A,h a , A , h before rounding cell endpoints.
Use the rational grid
a i = 1792 + 9 i 2560 , A j = 1 + 5 j 512 , i , j = 0 , 1 , … , 256. a_i=\frac{1792+9i}{2560},\qquad A_j=1+\frac{5j}{512},\qquad i,j=0,1,\ldots,256. a i = 2560 1792 + 9 i , A j = 1 + 512 5 j , i , j = 0 , 1 , … , 256.
Discard only A j + 1 < ( 433 / 500 ) a i 2 A_{j+1}<(433/500)a_i^{2} A j + 1 < ( 433/500 ) a i 2 : 8562 cells are discarded and 56974 retained. On each retained cell put
H i j = 1 10 6 ⌊ 10 6 max { 4 A j + 1 2 , 3 A j 2 4 a i + 1 2 − a i + 1 2 4 } ⌋ , ρ i j = a i A j + 1 . H_{ij}=\frac{1}{10^{6}}\left\lfloor10^{6}\sqrt{\max\left\{\frac{4}{A_{j+1}^{2}},\frac{3A_j^{2}}{4a_{i+1}^{2}}-\frac{a_{i+1}^{2}}{4}\right\}}\right\rfloor,\qquad\rho_{ij}=\frac{a_i}{A_{j+1}}. H ij = 1 0 6 1 1 0 6 max { A j + 1 2 4 , 4 a i + 1 2 3 A j 2 − 4 a i + 1 2 } , ρ ij = A j + 1 a i .
Then h ≥ H i j h\geq H_{ij} h ≥ H ij and a / A ≥ ρ i j a/A\geq\rho_{ij} a / A ≥ ρ ij on the admissible part of that cell. Define
J ‾ i j = π 5 90 A j H i j 4 + 1 1000 + 2 A j 5 ∑ m = 1 5 [ 6 ∑ q ∈ { 1 , 3 , 4 , 9 , 12 , 16 , 25 , 27 , 36 } U ( ρ i j q , m H i j ) + 12 ∑ q ∈ { 7 , 13 , 19 , 21 , 28 , 31 } U ( ρ i j q , m H i j ) ] . \overline{J}_{ij}=\frac{\pi^{5}}{90A_jH_{ij}^{4}}+\frac{1}{1000}
+\frac{2}{A_j^{5}}\sum_{m=1}^{5}\left[6\sum_{q\in\{1,3,4,9,12,16,25,27,36\}}U(\rho_{ij}\sqrt{q},mH_{ij})
+12\sum_{q\in\{7,13,19,21,28,31\}}U(\rho_{ij}\sqrt{q},mH_{ij})\right]. J ij = 90 A j H ij 4 π 5 + 1000 1 + A j 5 2 m = 1 ∑ 5 6 q ∈ { 1 , 3 , 4 , 9 , 12 , 16 , 25 , 27 , 36 } ∑ U ( ρ ij q , m H ij ) + 12 q ∈ { 7 , 13 , 19 , 21 , 28 , 31 } ∑ U ( ρ ij q , m H ij ) .
The multiplicities enumerate 0 < i 2 + i j + j 2 ≤ 36 0<i^{2}+ij+j^{2}\leq36 0 < i 2 + ij + j 2 ≤ 36 ; monotonicity and the complete tail give J △ ( a , A , h ) < J ‾ i j J_{\triangle}(a,A,h)<\overline{J}_{ij} J △ ( a , A , h ) < J ij . On each retained cell use the lower bound
L i j = max { max { a i + 1 − 12 − 11 5 a i 6 − 8349 a i + 1 4 2800 A j 5 , a i + 1 − 12 − 13 2 a i 6 , − 947 320 } − J ‾ i j , a i + 1 − 12 − 2901 200 a i 6 } . L_{ij}=\max\left\{\max\left\{a_{i+1}^{-12}-\frac{11}{5a_i^6}-\frac{8349a_{i+1}^4}{2800A_j^5},a_{i+1}^{-12}-\frac{13}{2a_i^6},-\frac{947}{320}\right\}-\overline{J}_{ij},a_{i+1}^{-12}-\frac{2901}{200a_i^6}\right\}. L ij = max { max { a i + 1 − 12 − 5 a i 6 11 − 2800 A j 5 8349 a i + 1 4 , a i + 1 − 12 − 2 a i 6 13 , − 320 947 } − J ij , a i + 1 − 12 − 200 a i 6 2901 } .
All 56974 comparisons give L i j > − 10 / 3 L_{ij}>-10/3 L ij > − 10/3 ; more precisely,
− 3.319871 < min 0 ≤ i , j < 256 A j + 1 ≥ ( 433 / 500 ) a i 2 L i j < − 3.319870. -3.319871<\min_{\substack{0\le i,j<256\\ A_{j+1}\ge(433/500)a_i^2}}L_{ij}<-3.319870. − 3.319871 < 0 ≤ i , j < 256 A j + 1 ≥ ( 433/500 ) a i 2 min L ij < − 3.319870.
The minimum occurs at ( i , j ) = ( 118 , 18 ) (i,j)=(118,18) ( i , j ) = ( 118 , 18 ) , with
a i = 1427 1280 , a i + 1 = 2863 2560 , A j = 301 256 , A j + 1 = 607 512 , H i j = 1686985 10 6 . a_i=\frac{1427}{1280},\qquad a_{i+1}=\frac{2863}{2560},\qquad A_j=\frac{301}{256},\qquad A_{j+1}=\frac{607}{512},\qquad H_{ij}=\frac{1686985}{10^6}. a i = 1280 1427 , a i + 1 = 2560 2863 , A j = 256 301 , A j + 1 = 512 607 , H ij = 1 0 6 1686985 .
Evaluate the finite sums and complete 1 / 1000 1/1000 1/1000 tail by interval addition and multiplication, using Lemma A.1 : the rational π \pi π enclosure, exponential bounds of degrees 21, 20, scaling and checked roots. Compute the floors in H i j H_{ij} H ij by exact integer square roots. This completes the proof of Lemma 3.13. □
Lemma 3.14 (A uniform large-area envelope). With v 1 , v 2 , Γ v_1,v_2,\Gamma v 1 , v 2 , Γ as in Lemma 3.6 ,
∣ L ∣ = V ≥ 2 , E ( L ) = m ( V ) ⟹ A = ∣ Γ ∣ < 1. |L|=V\ge2,\qquad\mathcal{E}(L)=m(V)\quad\Longrightarrow\quad A=|\Gamma|<1. ∣ L ∣ = V ≥ 2 , E ( L ) = m ( V ) ⟹ A = ∣Γ∣ < 1.
Proof. Since V / A ∗ > 1 V/A_*>1 V / A ∗ > 1 , the competitor T A ∗ ⊕ ( V / A ∗ ) Z T_{A_*}\oplus(V/A_*)\mathbb{Z} T A ∗ ⊕ ( V / A ∗ ) Z has negative cross-layer energy. Lemmas 3.11, 2.1 and 2.5 give m ( V ) < e 2 < − 337 / 100 < − 3 m(V)<e_2<-337/100<-3 m ( V ) < e 2 < − 337/100 < − 3 , 7 / 10 < a < 8 / 5 7/10<a<8/5 7/10 < a < 8/5 and A < 5 A<5 A < 5 . Theorem 2.3 and Lemma 2.4 also give
E ( L ) ≥ a − 12 − 2901 200 a − 6 . \mathcal{E}(L)\ge a^{-12}-\frac{2901}{200}a^{-6}. E ( L ) ≥ a − 12 − 200 2901 a − 6 .
For A ≥ 7 / 2 A\ge7/2 A ≥ 7/2 , Lemma 3.6 gives h > 3 A / 7 h>3A/7 h > 3 A /7 . Every term of 7 A − 3 + J ( A , 3 A / 7 ) 7A^{-3}+J(A,3A/7) 7 A − 3 + J ( A , 3 A /7 ) decreases with A A A , and
7 ( 7 / 2 ) 3 + J ( 7 / 2 , 3 / 2 ) < 113 50 . \frac{7}{(7/2)^3}+J(7/2,3/2)<\frac{113}{50}. ( 7/2 ) 3 7 + J ( 7/2 , 3/2 ) < 50 113 .
The planar and cross-layer bounds give E ( L ) > − 331 / 100 > e 2 \mathcal{E}(L)>-331/100>e_2 E ( L ) > − 331/100 > e 2 , a contradiction.
For 1 ≤ A ≤ 7 / 2 1\le A\le7/2 1 ≤ A ≤ 7/2 , the planar, triangular-dual and FCC bounds give the maximum in Lemma 3.13; its hypotheses follow from
A ≥ 3 2 a 2 > 433 500 a 2 , h 2 ≥ 3 A 2 4 a 2 − a 2 4 . A\ge\frac{\sqrt{3}}{2}a^2>\frac{433}{500}a^2,\qquad h^2\ge\frac{3A^2}{4a^2}-\frac{a^2}{4}. A ≥ 2 3 a 2 > 500 433 a 2 , h 2 ≥ 4 a 2 3 A 2 − 4 a 2 .
Lemma 3.13 now gives
E ( L ) > − 10 3 > − 337 100 > e 2 , \mathcal{E}(L)>-\frac{10}{3}>-\frac{337}{100}>e_2, E ( L ) > − 3 10 > − 100 337 > e 2 ,
again a contradiction. Hence A < 1 A<1 A < 1 . This completes the proof of Lemma 3.14 . □
Lemma 3.15 (Exact layer alignment). Let Γ ⊂ R 2 \Gamma\subset\mathbb{R}^2 Γ ⊂ R 2 be a planar lattice with
A = ∣ Γ ∣ ≤ 9 10 , min v ∈ Γ ∖ { 0 } ∣ v ∣ ≥ 7 10 , h ≥ 2. A=|\Gamma|\le\frac{9}{10},\qquad\min_{v\in\Gamma\setminus\{0\}}|v|\ge\frac{7}{10},\qquad h\ge2. A = ∣Γ∣ ≤ 10 9 , v ∈ Γ ∖ { 0 } min ∣ v ∣ ≥ 10 7 , h ≥ 2. Then
arg min u ∈ R 2 / Γ E ( Γ + Z ( u + h e 3 ) ) = { 0 } . \underset{u\in\mathbb{R}^2/\Gamma}{\arg\min}\,\mathcal{E}\bigl(\Gamma+\mathbb{Z}(u+he_3)\bigr)=\{0\}. u ∈ R 2 /Γ arg min E ( Γ + Z ( u + h e 3 ) ) = { 0 } .
Proof. For a reduced reciprocal basis w 1 , w 2 w_1,w_2 w 1 , w 2 , put ρ = 3 / 4 \rho=3/4 ρ = 3/4 and α = 4 / 3 \alpha=4/3 α = 4/3 . Then
∣ w j ∣ ≤ 2 3 λ 1 ( Γ ) ≤ 20 7 3 < 5 3 ( j = 1 , 2 ) , λ 1 ( Γ ∗ ) ≥ 7 9 > ρ , |w_j| \leq\frac{2}{\sqrt{3}\lambda_1(\Gamma)} \leq\frac{20}{7\sqrt{3}} < \frac{5}{3}\quad(j=1,2),\qquad\lambda_1(\Gamma^*) \geq\frac{7}{9}>\rho, ∣ w j ∣ ≤ 3 λ 1 ( Γ ) 2 ≤ 7 3 20 < 3 5 ( j = 1 , 2 ) , λ 1 ( Γ ∗ ) ≥ 9 7 > ρ ,
det ( w i ⋅ w j ) i , j = 1 2 = A − 2 ≥ 100 81 , ∣ w 1 ∣ 2 + ∣ w 2 ∣ 2 ≤ 50 9 , \det(w_i\cdot w_j)_{i,j=1}^2=A^{-2}\geq\frac{100}{81},\qquad|w_1|^2+|w_2|^2\leq\frac{50}{9}, det ( w i ⋅ w j ) i , j = 1 2 = A − 2 ≥ 81 100 , ∣ w 1 ∣ 2 + ∣ w 2 ∣ 2 ≤ 9 50 ,
n 2 + m 2 ≤ 9 2 ∣ n w 1 + m w 2 ∣ 2 ( n , m ∈ Z ) . n^2+m^2\leq\frac{9}{2}|nw_1+mw_2|^2\qquad(n,m\in\mathbb{Z}). n 2 + m 2 ≤ 2 9 ∣ n w 1 + m w 2 ∣ 2 ( n , m ∈ Z ) .
For r = ∣ n w 1 + m w 2 ∣ r=|nw_1+mw_2| r = ∣ n w 1 + m w 2 ∣ , d = k h d=kh d = k h , ( n , m ) ≠ ( 0 , 0 ) (n,m)\ne(0,0) ( n , m ) = ( 0 , 0 ) and k ≥ 1 k\geq1 k ≥ 1 ,
C ( r , d ) = π 6 r 5 60 d 5 K 5 ( 2 π r d ) − 2 π 3 r 2 d 2 K 2 ( 2 π r d ) . C(r,d)=\frac{\pi^6r^5}{60d^5}K_5(2\pi rd)-\frac{2\pi^3r^2}{d^2}K_2(2\pi rd). C ( r , d ) = 60 d 5 π 6 r 5 K 5 ( 2 π r d ) − d 2 2 π 3 r 2 K 2 ( 2 π r d ) .
Lemma 3.1 gives, for z = 2 π r d ≥ 3 π > 47 / 5 z=2\pi rd\geq3\pi>47/5 z = 2 π r d ≥ 3 π > 47/5 ,
K 5 ( z ) ≤ π 2 z e − z P 5 ( z ) , K 2 ( z ) ≥ π 2 z e − z , P 5 ( z ) < 9 2 . K_5(z)\leq\sqrt{\frac{\pi}{2z}}e^{-z}P_5(z),\qquad K_2(z)\geq\sqrt{\frac{\pi}{2z}}e^{-z},\qquad P_5(z)<\frac{9}{2}. K 5 ( z ) ≤ 2 z π e − z P 5 ( z ) , K 2 ( z ) ≥ 2 z π e − z , P 5 ( z ) < 2 9 .
For r = ∣ w j ∣ r=|w_j| r = ∣ w j ∣ and d ≥ h ≥ 2 d\geq h\geq2 d ≥ h ≥ 2 ,
π 3 120 ( r / d ) 3 P 5 ( 2 π r d ) < 3 4 , − C ( ∣ w j ∣ , h ) ≥ b ( h ) : = π 3 ρ 3 / 2 4 h 5 / 2 e − ( 10 π / 3 ) h . \frac{\pi^3}{120}(r/d)^3P_5(2\pi rd)<\frac{3}{4},\qquad-C(|w_j|,h)\geq b(h):=\frac{\pi^3\rho^{3/2}}{4h^{5/2}}e^{-(10\pi/3)h}. 120 π 3 ( r / d ) 3 P 5 ( 2 π r d ) < 4 3 , − C ( ∣ w j ∣ , h ) ≥ b ( h ) := 4 h 5/2 π 3 ρ 3/2 e − ( 10 π /3 ) h .
Moreover ∂ r { r 3 P 5 ( 2 π d r ) } > 0 \partial_r\{r^3P_5(2\pi dr)\}>0 ∂ r { r 3 P 5 ( 2 π d r )} > 0 , since
3 + 30 / z + 105 / z 2 − 945 / z 4 − 1890 / z 5 > 0 ( z ≥ 47 / 5 ) . 3+30/z+105/z^2-945/z^4-1890/z^5>0\qquad(z\geq47/5). 3 + 30/ z + 105/ z 2 − 945/ z 4 − 1890/ z 5 > 0 ( z ≥ 47/5 ) .
Thus
r ≤ α d ⟹ π 3 120 ( r / d ) 3 P 5 ( 2 π r d ) ≤ π 3 α 3 120 P 5 ( 2 π α d 2 ) < 1 , r\leq\alpha d\Longrightarrow\frac{\pi^3}{120}(r/d)^3P_5(2\pi rd)\leq\frac{\pi^3\alpha^3}{120}P_5(2\pi\alpha d^2)<1, r ≤ α d ⟹ 120 π 3 ( r / d ) 3 P 5 ( 2 π r d ) ≤ 120 π 3 α 3 P 5 ( 2 π α d 2 ) < 1 ,
C ( r , d ) > 0 ⟹ r > α d , C ( r , d ) + ≤ 5 π 6 384 r 9 / 2 d − 11 / 2 e − 2 π r d . C(r,d)>0\Longrightarrow r>\alpha d,\qquad C(r,d)_+\leq\frac{5\pi^6}{384}r^{9/2}d^{-11/2}e^{-2\pi rd}. C ( r , d ) > 0 ⟹ r > α d , C ( r , d ) + ≤ 384 5 π 6 r 9/2 d − 11/2 e − 2 π r d .
Set δ j = 1 − cos ( 2 π w j ⋅ u ) \delta_j=1-\cos(2\pi w_j\cdot u) δ j = 1 − cos ( 2 π w j ⋅ u ) , j = 1 , 2 j=1,2 j = 1 , 2 . Then
1 − cos ( 2 π k ( n w 1 + m w 2 ) ⋅ u ) ≤ 9 2 k 2 r 2 ( δ 1 + δ 2 ) . 1-\cos(2\pi k(nw_1+mw_2)\cdot u)\leq\frac{9}{2}k^2r^2(\delta_1+\delta_2). 1 − cos ( 2 π k ( n w 1 + m w 2 ) ⋅ u ) ≤ 2 9 k 2 r 2 ( δ 1 + δ 2 ) .
Equation (30) yields
E ( Γ + Z ( u + h e 3 ) ) − E ( Γ ⊕ h Z ) ≥ 2 b ( h ) − S ( h ) A ( δ 1 + δ 2 ) . \mathcal{E}(\Gamma+\mathbb{Z}(u+he_3))-\mathcal{E}(\Gamma\oplus h\mathbb{Z})\geq\frac{2b(h)-S(h)}{A}(\delta_1+\delta_2). E ( Γ + Z ( u + h e 3 )) − E ( Γ ⊕ h Z ) ≥ A 2 b ( h ) − S ( h ) ( δ 1 + δ 2 ) .
where
S ( h ) : = 9 2 ∑ k = 1 ∞ k 2 ∑ w ∈ Γ ∗ ∖ { 0 } ∣ w ∣ 2 C ( ∣ w ∣ , k h ) + ≤ 15 π 6 256 ∑ k = 1 ∞ k 2 ( k h ) − 11 / 2 ∑ w ∈ Γ ∗ ∣ w ∣ > α k h ∣ w ∣ 13 / 2 e − 2 π k h ∣ w ∣ . \begin{aligned}
S(h):=\frac{9}{2}\sum_{k=1}^{\infty}k^2\sum_{w\in\Gamma^*\setminus\{0\}}|w|^2C(|w|,kh)_+ \\
&\leq\frac{15\pi^6}{256}\sum_{k=1}^{\infty}k^2(kh)^{-11/2}\sum_{\substack{w\in\Gamma^*\\ |w|>\alpha kh}}|w|^{13/2}e^{-2\pi kh|w|}.
\end{aligned} S ( h ) := 2 9 k = 1 ∑ ∞ k 2 w ∈ Γ ∗ ∖ { 0 } ∑ ∣ w ∣ 2 C ( ∣ w ∣ , k h ) + ≤ 256 15 π 6 k = 1 ∑ ∞ k 2 ( k h ) − 11/2 w ∈ Γ ∗ ∣ w ∣ > α k h ∑ ∣ w ∣ 13/2 e − 2 π k h ∣ w ∣ .
For d ≥ 2 d\geq2 d ≥ 2 , j ∈ Z ≥ 0 j\in\mathbb{Z}_{\geq0} j ∈ Z ≥ 0 ,
# { w ∈ Γ ∗ : α d + j ρ ≤ ∣ w ∣ < α d + ( j + 1 ) ρ } ≤ ( 2 α d / ρ + 2 j + 3 ) 2 ≤ 26 d 2 ( j + 1 ) 2 . \#\{w\in\Gamma^*:\alpha d+j\rho\leq|w|<\alpha d+(j+1)\rho\}\leq(2\alpha d/\rho+2j+3)^2\leq26d^2(j+1)^2. # { w ∈ Γ ∗ : α d + j ρ ≤ ∣ w ∣ < α d + ( j + 1 ) ρ } ≤ ( 2 α d / ρ + 2 j + 3 ) 2 ≤ 26 d 2 ( j + 1 ) 2 .
Since e − 2 π ρ d < 1 / 10000 e^{-2\pi\rho d}<1/10000 e − 2 π ρ d < 1/10000 and the radial summand decreases, geometric comparison gives
∑ w ∈ Γ ∗ ∣ w ∣ > α d ∣ w ∣ 13 / 2 e − 2 π d ∣ w ∣ < 52 α 13 / 2 d 17 / 2 e − 2 π α d 2 , \sum_{\substack{w\in\Gamma^*\\ |w|>\alpha d}}|w|^{13/2}e^{-2\pi d|w|}<52\alpha^{13/2}d^{17/2}e^{-2\pi\alpha d^2}, w ∈ Γ ∗ ∣ w ∣ > α d ∑ ∣ w ∣ 13/2 e − 2 π d ∣ w ∣ < 52 α 13/2 d 17/2 e − 2 π α d 2 ,
∑ k = 1 ∞ k 5 e − 2 π α k 2 h 2 < 2 e − 2 π α h 2 , \sum_{k=1}^{\infty}k^5e^{-2\pi\alpha k^2h^2}<2e^{-2\pi\alpha h^2}, k = 1 ∑ ∞ k 5 e − 2 π α k 2 h 2 < 2 e − 2 π α h 2 ,
S ( h ) 2 b ( h ) < 195 π 3 α 13 / 2 h 11 / 2 16 ρ 3 / 2 e − 2 π h ( α h − 5 / 3 ) < 1 ( h ≥ 2 ) . \frac{S(h)}{2b(h)}<\frac{195\pi^3\alpha^{13/2}h^{11/2}}{16\rho^{3/2}}e^{-2\pi h(\alpha h-5/3)}<1\qquad(h\geq2). 2 b ( h ) S ( h ) < 16 ρ 3/2 195 π 3 α 13/2 h 11/2 e − 2 π h ( α h − 5/3 ) < 1 ( h ≥ 2 ) .
The last majorant decreases for h ≥ 2 h \ge2 h ≥ 2 . Since w 1 , w 2 w_1,w_2 w 1 , w 2 form a reciprocal basis, δ 1 + δ 2 = 0 ⟺ u ∈ Γ \delta_1+\delta_2=0 \Longleftrightarrow u\in\Gamma δ 1 + δ 2 = 0 ⟺ u ∈ Γ , giving the stated equality case. This completes the proof of Lemma 3.15 . □ \square □
Lemma 3.16 (Explicit nonzero Fourier bounds). For a planar lattice Γ ⊂ R 2 \Gamma\subset\mathbb{R}^2 Γ ⊂ R 2 , set A = ∣ Γ ∣ A=|\Gamma| A = ∣Γ∣ and
R ( Γ , h ) = 1 A ∑ k = 1 ∞ ∑ w ∈ Γ ∗ ∖ { 0 } [ π 6 ∣ w ∣ 5 60 ( k h ) 5 K 5 ( 2 π k h ∣ w ∣ ) − 2 π 3 ∣ w ∣ 2 ( k h ) 2 K 2 ( 2 π k h ∣ w ∣ ) ] . R(\Gamma,h)=\frac{1}{A}\sum_{k=1}^{\infty}\sum_{w\in\Gamma^*\setminus\{0\}}\left[\frac{\pi^6|w|^5}{60(kh)^5}K_5(2\pi kh|w|)-\frac{2\pi^3|w|^2}{(kh)^2}K_2(2\pi kh|w|)\right]. R ( Γ , h ) = A 1 k = 1 ∑ ∞ w ∈ Γ ∗ ∖ { 0 } ∑ [ 60 ( k h ) 5 π 6 ∣ w ∣ 5 K 5 ( 2 π k h ∣ w ∣ ) − ( k h ) 2 2 π 3 ∣ w ∣ 2 K 2 ( 2 π k h ∣ w ∣ ) ] .
For this nonzero aligned-layer Fourier term and h ≥ 2 h\ge2 h ≥ 2 ,
If
Γ = A Γ x , y , 2 5 ≤ A ≤ 9 10 , 0 ≤ x ≤ 1 2 , 3 2 ≤ y ≤ 1 , \Gamma=\sqrt{A}\Gamma_{x,y},\qquad\frac{2}{5}\le A\le\frac{9}{10},\qquad0\le x\le\frac{1}{2},\qquad\frac{\sqrt{3}}{2}\le y\le1, Γ = A Γ x , y , 5 2 ≤ A ≤ 10 9 , 0 ≤ x ≤ 2 1 , 2 3 ≤ y ≤ 1 ,
then, with A , h A,h A , h fixed,
∥ ∇ x , y 2 R ( Γ , h ) ∥ o p < 3 8 . \left\|\nabla^2_{x,y}R(\Gamma,h)\right\|_{\mathrm{op}}<\frac{3}{8}. ∇ x , y 2 R ( Γ , h ) op < 8 3 .
If A ≤ 9 / 10 A\le9/10 A ≤ 9/10 and min v ∈ Γ ∖ { 0 } ∣ v ∣ ≥ 7 / 10 \min_{v\in\Gamma\setminus\{0\}}|v|\ge7/10 min v ∈ Γ ∖ { 0 } ∣ v ∣ ≥ 7/10 , then
∣ R ( Γ , h ) ∣ < 1 80 . |R(\Gamma,h)|<\frac{1}{80}. ∣ R ( Γ , h ) ∣ < 80 1 .
Proof. For r = ∣ w ∣ r=|w| r = ∣ w ∣ , d = k h d=kh d = k h , q = r 2 q=r^2 q = r 2 , write
C ( q , d ) = π 6 r 5 60 d 5 K 5 ( 2 π r d ) − 2 π 3 r 2 d 2 K 2 ( 2 π r d ) . C(q,d)=\frac{\pi^6r^5}{60d^5}K_5(2\pi rd)-\frac{2\pi^3r^2}{d^2}K_2(2\pi rd). C ( q , d ) = 60 d 5 π 6 r 5 K 5 ( 2 π r d ) − d 2 2 π 3 r 2 K 2 ( 2 π r d ) .
The derivative identity in Lemma 3.1 gives
C q = − π 7 r 4 60 d 4 K 4 ( 2 π r d ) + 2 π 4 r d K 1 ( 2 π r d ) , C q q = π 8 r 3 60 d 3 K 3 ( 2 π r d ) − 2 π 5 K 0 ( 2 π r d ) . C_q=-\frac{\pi^7r^4}{60d^4}K_4(2\pi rd)+\frac{2\pi^4r}{d}K_1(2\pi rd),\qquad C_{qq}=\frac{\pi^8r^3}{60d^3}K_3(2\pi rd)-2\pi^5K_0(2\pi rd). C q = − 60 d 4 π 7 r 4 K 4 ( 2 π r d ) + d 2 π 4 r K 1 ( 2 π r d ) , C q q = 60 d 3 π 8 r 3 K 3 ( 2 π r d ) − 2 π 5 K 0 ( 2 π r d ) .
In (i), the reciprocal squared lengths satisfy
q = 1 A ( y n 2 + ( m + n x ) 2 y ) , ∣ ∇ x , y q ∣ = q / y , ∥ ∇ x , y 2 q ∥ o p = 2 q / y 2 . q=\frac{1}{A}\left(yn^2+\frac{(m+nx)^2}{y}\right),\qquad|\nabla_{x,y}q|=q/y,\qquad\left\|\nabla^2_{x,y}q\right\|_{\mathrm{op}}=2q/y^2. q = A 1 ( y n 2 + y ( m + n x ) 2 ) , ∣ ∇ x , y q ∣ = q / y , ∇ x , y 2 q op = 2 q / y 2 .
Thus r ≥ ρ : = 19 / 20 r\ge\rho:=19/20 r ≥ ρ := 19/20 ; n = 0 n=0 n = 0 gives q ≥ 1 / ( A y ) ≥ 10 / 9 q\ge1/(Ay)\ge10/9 q ≥ 1/ ( A y ) ≥ 10/9 , and n ≠ 0 n\ne0 n = 0 gives q ≥ y / A ≥ 5 3 / 9 > 361 / 400 q\ge y/A\ge5\sqrt{3}/9>361/400 q ≥ y / A ≥ 5 3 /9 > 361/400 . In particular z = 2 π r d > 19 π / 5 > 119 / 10 z=2\pi rd>19\pi/5>119/10 z = 2 π r d > 19 π /5 > 119/10 . Lemma 3.1 gives
K ν ( 2 π r d ) ≤ 9 8 r d e − 2 π r d ( ν ∈ { 0 , 1 , 3 , 4 } ) . K_\nu(2\pi rd)\le\frac{9}{8\sqrt{rd}}e^{-2\pi rd}\qquad(\nu\in\{0,1,3,4\}). K ν ( 2 π r d ) ≤ 8 r d 9 e − 2 π r d ( ν ∈ { 0 , 1 , 3 , 4 }) .
Therefore
∥ ∇ x , y 2 C ( q , d ) ∥ o p ≤ 9 e − 2 π r d 8 y 2 [ 8 r 13 / 2 60 d 7 / 2 + 2 π 5 r 7 / 2 d 1 / 2 + π 7 r 11 / 2 30 d 9 / 2 + 4 π 4 r 5 / 2 d 3 / 2 ] . \left\|\nabla^2_{x,y}C(q,d)\right\|_{\mathrm{op}}\le\frac{9e^{-2\pi rd}}{8y^2}\left[\frac{8r^{13/2}}{60d^{7/2}}+\frac{2\pi^5r^{7/2}}{d^{1/2}}+\frac{\pi^7r^{11/2}}{30d^{9/2}}+\frac{4\pi^4r^{5/2}}{d^{3/2}}\right]. ∇ x , y 2 C ( q , d ) op ≤ 8 y 2 9 e − 2 π r d [ 60 d 7/2 8 r 13/2 + d 1/2 2 π 5 r 7/2 + 30 d 9/2 π 7 r 11/2 + d 3/2 4 π 4 r 5/2 ] .
Disjoint reciprocal disks of radius ρ / 2 \rho/2 ρ /2 show that the annulus j ρ ≤ r < ( j + 1 ) ρ j\rho\le r<(j+1)\rho j ρ ≤ r < ( j + 1 ) ρ contains at most 25 j 2 25j^2 25 j 2 points. The radial summands decrease in this range. Put q 0 = e − 2 π ρ h q_0=e^{-2\pi\rho h} q 0 = e − 2 π ρ h . For every positive exponent a a a and each radial power p ≤ 13 / 2 p\le13/2 p ≤ 13/2 used here,
∑ k = 1 ∞ ( k h ) − a ∑ w ∈ Γ ∗ ∖ { 0 } ∣ w ∣ p e − 2 π k h ∣ w ∣ ≤ 25 ρ p h − a q 0 ( 1 − 512 q 0 ) ( 1 − q 0 ) < 26 ρ p h − a q 0 ( q 0 ≤ 1 / 15000 ) . \sum_{k=1}^{\infty}(kh)^{-a}\sum_{w\in\Gamma^*\setminus\{0\}}|w|^pe^{-2\pi kh|w|}
\le\frac{25\rho^ph^{-a}q_0}{(1-512q_0)(1-q_0)}
<26\rho^ph^{-a}q_0\qquad(q_0\le1/15000). k = 1 ∑ ∞ ( k h ) − a w ∈ Γ ∗ ∖ { 0 } ∑ ∣ w ∣ p e − 2 π k h ∣ w ∣ ≤ ( 1 − 512 q 0 ) ( 1 − q 0 ) 25 ρ p h − a q 0 < 26 ρ p h − a q 0 ( q 0 ≤ 1/15000 ) .
The annulus majorant ∑ j = 1 ∞ j 9 q 0 k j \sum_{j=1}^{\infty}j^9q_0^{kj} ∑ j = 1 ∞ j 9 q 0 k j has successive-term ratio ≤ 512 q 0 k \le512q_0^k ≤ 512 q 0 k ; the layer series is geometric, with q 0 < 1 / 140000 q_0<1/140000 q 0 < 1/140000 . Since ( A y 2 ) − 1 ≤ 10 / 3 (Ay^2)^{-1}\le10/3 ( A y 2 ) − 1 ≤ 10/3 , and the coefficient bracket including 9 / 8 9/8 9/8 is less than 580 580 580 ,
∥ ∇ x , y 2 R ∥ o p < 3 8 . \left\|\nabla^2_{x,y}R\right\|_{\mathrm{op}}<\frac{3}{8}. ∇ x , y 2 R op < 8 3 .
For (ii), use the reciprocal separation ρ A = 7 / ( 10 A ) \rho_A=7/(10A) ρ A = 7/ ( 10 A ) , since Γ ∗ ≅ A − 1 Γ \Gamma^*\cong A^{-1}\Gamma Γ ∗ ≅ A − 1 Γ . Now 2 π r d ≥ 2 π ρ A h ≥ 28 π / 9 > 9.7 2\pi rd\ge2\pi\rho_Ah\ge28\pi/9>9.7 2 π r d ≥ 2 π ρ A h ≥ 28 π /9 > 9.7 . Lemma 3.1 gives
∣ C ( q , d ) ∣ ≤ e − 2 π r d ( 3 π 6 80 r 9 / 2 d − 11 / 2 + 3 π 3 2 r 3 / 2 d − 5 / 2 ) . |C(q,d)|\le e^{-2\pi rd}\left(\frac{3\pi^6}{80}r^{9/2}d^{-11/2}+\frac{3\pi^3}{2}r^{3/2}d^{-5/2}\right). ∣ C ( q , d ) ∣ ≤ e − 2 π r d ( 80 3 π 6 r 9/2 d − 11/2 + 2 3 π 3 r 3/2 d − 5/2 ) .
Here P 5 ( 2 π r d ) < 9 / 2 P_5(2\pi r d)<9/2 P 5 ( 2 π r d ) < 9/2 and P 2 ( 2 π r d ) < 3 / 2 P_2(2\pi r d)<3/2 P 2 ( 2 π r d ) < 3/2 . The same disk and layer comparisons give
∣ R ( Γ , h ) ∣ ≤ 26 A e − 2 π ρ A h ( 3 π 6 80 ρ 9 A 9 / 2 h − 11 / 2 + 3 π 3 2 ρ 3 A 3 / 2 h − 5 / 2 ) . |R(\Gamma,h)|\leq\frac{26}{A}e^{-2\pi\rho Ah}\left(\frac{3\pi^6}{80\rho^9}A^{9/2}h^{-11/2}+\frac{3\pi^3}{2\rho^3}A^{3/2}h^{-5/2}\right). ∣ R ( Γ , h ) ∣ ≤ A 26 e − 2 π ρ A h ( 80 ρ 9 3 π 6 A 9/2 h − 11/2 + 2 ρ 3 3 π 3 A 3/2 h − 5/2 ) .
Each summand decreases in h ≥ 2 h\geq2 h ≥ 2 and increases in A ≤ 9 / 10 A\leq9/10 A ≤ 9/10 , since the logarithmic derivative of A − p − 1 e − ( 7 π / 5 ) h / A A^{-p-1}e^{-(7\pi/5)h/A} A − p − 1 e − ( 7 π /5 ) h / A has sign ( 7 π / 5 ) h / A − p − 1 > 0 (7\pi/5)h/A-p-1>0 ( 7 π /5 ) h / A − p − 1 > 0 . At ( A , h ) = ( 9 / 10 , 2 ) (A,h)=(9/10,2) ( A , h ) = ( 9/10 , 2 ) , the exponential factor is less than 1 / 16000 1/16000 1/16000 and the coefficient bracket is less than 25 / 4 25/4 25/4 . Hence ∣ R ∣ < 1 / 80 |R|<1/80 ∣ R ∣ < 1/80 .
The bounds justify differentiation and rearrangement. This completes the proof of Lemma 3.16. □
Lemma 3.17 (The global fixed-area comparison). Let Γ ⊂ R 2 \Gamma\subset\mathbb{R}^2 Γ ⊂ R 2 be a planar lattice with
A = ∣ Γ ∣ ≤ 9 10 , min v ∈ Γ ∖ { 0 } ∣ v ∣ ≥ 7 10 , h ≥ 2. A=|\Gamma|\leq\frac{9}{10},\qquad\min_{v\in\Gamma\setminus\{0\}}|v|\geq\frac{7}{10},\qquad h\geq2. A = ∣Γ∣ ≤ 10 9 , v ∈ Γ ∖ { 0 } min ∣ v ∣ ≥ 10 7 , h ≥ 2. Then
E ( Γ ⊕ h Z ) ≥ E ( T A ⊕ h Z ) , \mathcal{E}(\Gamma\oplus h\mathbb{Z})\geq\mathcal{E}(T_A\oplus h\mathbb{Z}), E ( Γ ⊕ h Z ) ≥ E ( T A ⊕ h Z ) , E ( Γ ⊕ h Z ) = E ( T A ⊕ h Z ) ⟺ Γ ≅ T A . \mathcal{E}(\Gamma\oplus h\mathbb{Z})=\mathcal{E}(T_A\oplus h\mathbb{Z})\quad\Longleftrightarrow\quad\Gamma\cong T_A. E ( Γ ⊕ h Z ) = E ( T A ⊕ h Z ) ⟺ Γ ≅ T A . Proof. Write
Γ = A Γ x , y , 0 ≤ x ≤ 1 2 , x 2 + y 2 ≥ 1. \Gamma=\sqrt{A}\Gamma_{x,y},\qquad0\leq x\leq\frac12,\qquad x^2+y^2\geq1. Γ = A Γ x , y , 0 ≤ x ≤ 2 1 , x 2 + y 2 ≥ 1. Then
A ≥ 3 2 ( 7 10 ) 2 > 2 5 , y = A λ 1 ( Γ ) 2 ≤ 90 49 < 2. A\geq\frac{\sqrt{3}}{2}\left(\frac{7}{10}\right)^2>\frac25,\qquad y=\frac{A}{\lambda_1(\Gamma)^2}\leq\frac{90}{49}<2. A ≥ 2 3 ( 10 7 ) 2 > 5 2 , y = λ 1 ( Γ ) 2 A ≤ 49 90 < 2. By homogeneity and the planar input (14) ,
E ( Γ ) − E ( T A ) ≥ ( A − 6 − A − 3 ) [ E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ] . \mathcal{E}(\Gamma)-\mathcal{E}(T_A)\geq\left(A^{-6}-A^{-3}\right)\left[\mathcal{E}(\Gamma_{x,y},3)-\mathcal{E}(T_1,3)\right]. E ( Γ ) − E ( T A ) ≥ ( A − 6 − A − 3 ) [ E ( Γ x , y , 3 ) − E ( T 1 , 3 ) ] . For y ≤ 1 y\leq1 y ≤ 1 , Lemma 3.3 and A − 6 − A − 3 > 1 / 2 A^{-6}-A^{-3}>1/2 A − 6 − A − 3 > 1/2 give 1 4 [ ( x − 1 / 2 ) 2 + ( y − 3 / 2 ) 2 ] \frac14[(x-1/2)^2+(y-\sqrt{3}/2)^2] 4 1 [( x − 1/2 ) 2 + ( y − 3 /2 ) 2 ] . Triangular symmetry gives ∇ x , y R ( A Γ x , y , h ) = 0 \nabla_{x,y}R(\sqrt{A}\Gamma_{x,y},h)=0 ∇ x , y R ( A Γ x , y , h ) = 0 at ( 1 / 2 , 3 / 2 ) (1/2,\sqrt{3}/2) ( 1/2 , 3 /2 ) . The segment to ( x , y ) (x,y) ( x , y ) lies in the rectangle of Lemma 3.16(i); Taylor’s formula bounds the Fourier loss by 3 / 16 3/16 3/16 times this squared distance. For y ≥ 1 y\geq1 y ≥ 1 , instead,
E ( Γ ) − E ( T A ) ≥ 3 3 2 ( y − 3 / 2 ) 2 > 5 128 , A = y λ 1 ( Γ ) 2 ≥ 49 100 . \mathcal{E}(\Gamma)-\mathcal{E}(T_A)\geq\frac{3\sqrt{3}}{2}(y-\sqrt{3}/2)^2>\frac{5}{128},\qquad A=y\lambda_1(\Gamma)^2\geq\frac{49}{100}. E ( Γ ) − E ( T A ) ≥ 2 3 3 ( y − 3 /2 ) 2 > 128 5 , A = y λ 1 ( Γ ) 2 ≥ 100 49 . Hence λ 1 ( T A ) 2 = 2 A / 3 ≥ 49 / 100 \lambda_1(T_A)^2=2A/\sqrt{3}\geq49/100 λ 1 ( T A ) 2 = 2 A / 3 ≥ 49/100 , and Lemma 3.16(ii) bounds the Fourier loss by 1 / 40 1/40 1/40 . The zero mode is independent of shape, so
E ( Γ ⊕ h Z ) − E ( T A ⊕ h Z ) ≥ { 1 16 [ ( x − 1 / 2 ) 2 + ( y − 3 / 2 ) 2 ] , y ≤ 1 , 9 640 , y > 1. \mathcal{E}(\Gamma\oplus h\mathbb{Z})-\mathcal{E}(T_A\oplus h\mathbb{Z})\geq
\begin{cases}
\frac{1}{16}[(x-1/2)^2+(y-\sqrt{3}/2)^2],&y\leq1,\\
\frac{9}{640},&y>1.
\end{cases} E ( Γ ⊕ h Z ) − E ( T A ⊕ h Z ) ≥ { 16 1 [( x − 1/2 ) 2 + ( y − 3 /2 ) 2 ] , 640 9 , y ≤ 1 , y > 1. Equality therefore requires Γ ≅ T A \Gamma\cong T_A Γ ≅ T A . This completes the proof of Lemma 3.17. □
Lemma 3.18 (Uniform area convexity). For V ≥ 89 / 50 V\geq89/50 V ≥ 89/50 ,
F V ( A ) : = E ( T A ⊕ V A Z ) , F V ′ ′ ( A ) > 34 ( 2 / 5 ≤ A ≤ 89 / 100 ) . F_V(A):=\mathcal{E}\left(T_A\oplus\frac{V}{A}\mathbb{Z}\right),\qquad F_V^{\prime\prime}(A)>34\qquad(2/5\leq A\leq89/100). F V ( A ) := E ( T A ⊕ A V Z ) , F V ′′ ( A ) > 34 ( 2/5 ≤ A ≤ 89/100 ) . Proof.
min v ∈ T 1 ∖ { 0 } ∣ v ∣ 2 = 2 3 , E ( T 1 , 6 ) ≥ 81 32 , \min_{v\in T_1\setminus\{0\}}|v|^2=\frac{2}{\sqrt{3}},\qquad E(T_1,6)\geq\frac{81}{32}, v ∈ T 1 ∖ { 0 } min ∣ v ∣ 2 = 3 2 , E ( T 1 , 6 ) ≥ 32 81 , { ( i , j ) ∈ Z 2 : i 2 + i j + j 2 = 1 } = { ± ( 1 , 0 ) , ± ( 0 , 1 ) , ± ( 1 , − 1 ) } . \{(i,j)\in\mathbb{Z}^2:i^2+ij+j^2=1\}=\{\pm(1,0),\pm(0,1),\pm(1,-1)\}. {( i , j ) ∈ Z 2 : i 2 + ij + j 2 = 1 } = { ± ( 1 , 0 ) , ± ( 0 , 1 ) , ± ( 1 , − 1 )} . Lemma 3.4 gives E ( T 1 , 3 ) < 17 / 4 E(T_1,3)<17/4 E ( T 1 , 3 ) < 17/4 , hence
U ′ ′ ( A ) = 21 E ( T 1 , 6 ) A − 8 − 12 E ( T 1 , 3 ) A − 5 > A − 8 ( 1701 / 32 − 51 A 3 ) > 37. U^{\prime\prime}(A)=21E(T_1,6)A^{-8}-12E(T_1,3)A^{-5}>A^{-8}(1701/32-51A^3)>37. U ′′ ( A ) = 21 E ( T 1 , 6 ) A − 8 − 12 E ( T 1 , 3 ) A − 5 > A − 8 ( 1701/32 − 51 A 3 ) > 37.
The interlayer terms are
π ζ ( 10 ) A 9 5 V 10 − π ζ ( 4 ) A 3 V 4 + ∑ k = 1 ∞ ∑ w ∈ Γ 1 ∗ ∖ { 0 } [ π 6 ∣ w ∣ 5 A 3 / 2 60 ( k V ) 5 K 5 ( 2 π ∣ w ∣ k V A − 3 / 2 ) − 2 π 3 ∣ w ∣ 2 ( k V ) 2 K 2 ( 2 π ∣ w ∣ k V A − 3 / 2 ) ] . \frac{\pi\zeta(10)A^9}{5V^{10}}-\frac{\pi\zeta(4)A^3}{V^4}+\sum_{k=1}^{\infty}\sum_{w\in\Gamma_1^*\setminus\{0\}}\left[\frac{\pi^6|w|^5A^{3/2}}{60(kV)^5}K_5(2\pi|w|kVA^{-3/2})-\frac{2\pi^3|w|^2}{(kV)^2}K_2(2\pi|w|kVA^{-3/2})\right]. 5 V 10 π ζ ( 10 ) A 9 − V 4 π ζ ( 4 ) A 3 + k = 1 ∑ ∞ w ∈ Γ 1 ∗ ∖ { 0 } ∑ [ 60 ( k V ) 5 π 6 ∣ w ∣ 5 A 3/2 K 5 ( 2 π ∣ w ∣ k V A − 3/2 ) − ( k V ) 2 2 π 3 ∣ w ∣ 2 K 2 ( 2 π ∣ w ∣ k V A − 3/2 ) ] .
The repulsive zero mode is strictly convex; by Lemma 3.1, for nonzero modes, z = κ A − 3 / 2 z=\kappa A^{-3/2} z = κ A − 3/2 gives
d 2 d A 2 { A 3 / 2 K 5 ( z ) } = A − 1 / 2 [ 3 4 ( K 5 − z K 5 ′ ) + 9 4 z 2 K 5 ′ ′ ] > 0. \frac{\mathrm{d}^2}{\mathrm{d}A^2}\{A^{3/2}K_5(z)\}=A^{-1/2}\left[\frac{3}{4}(K_5-zK_5')+\frac{9}{4}z^2K_5''\right]>0. d A 2 d 2 { A 3/2 K 5 ( z )} = A − 1/2 [ 4 3 ( K 5 − z K 5 ′ ) + 4 9 z 2 K 5 ′′ ] > 0.
The attractive zero-mode loss is less than
6 π ζ ( 4 ) A V 4 < 19 10 . \frac{6\pi\zeta(4)A}{V^4}<\frac{19}{10}. V 4 6 π ζ ( 4 ) A < 10 19 .
The differential equation and K 2 ′ < 0 K_2'<0 K 2 ′ < 0 in Lemma 3.1 give, for c > 0 c>0 c > 0 independent of A A A ,
d 2 d A 2 [ − c K 2 ( z ) ] = − c 4 A 2 [ 9 ( z 2 + 4 ) K 2 + 6 z K 2 ′ ] ≥ − 9 c 4 A 2 ( z 2 + 4 ) K 2 . \frac{\mathrm{d}^2}{\mathrm{d}A^2}[-cK_2(z)]=-\frac{c}{4A^2}[9(z^2+4)K_2+6zK_2']\geq-\frac{9c}{4A^2}(z^2+4)K_2. d A 2 d 2 [ − c K 2 ( z )] = − 4 A 2 c [ 9 ( z 2 + 4 ) K 2 + 6 z K 2 ′ ] ≥ − 4 A 2 9 c ( z 2 + 4 ) K 2 .
Put h = V / A h=V/A h = V / A , d = k h d=kh d = k h and r = ∣ w ∣ / A r=|w|/\sqrt{A} r = ∣ w ∣/ A . Then c = 2 π 3 r 2 / ( A d 2 ) c=2\pi^3r^2/(Ad^2) c = 2 π 3 r 2 / ( A d 2 ) and
z = 2 π r d ≥ 2 π V A − 3 / 2 > 333 25 , K 2 ( z ) ≤ K 5 / 2 ( z ) < 3 e − 2 π r d 4 r d . z=2\pi rd\geq2\pi VA^{-3/2}>\frac{333}{25},\qquad K_2(z)\leq K_{5/2}(z)<\frac{3e^{-2\pi rd}}{4\sqrt{rd}}. z = 2 π r d ≥ 2 π V A − 3/2 > 25 333 , K 2 ( z ) ≤ K 5/2 ( z ) < 4 r d 3 e − 2 π r d .
The total nonzero-mode loss is bounded above by
27 2 A 3 ∑ k = 1 ∞ ∑ w ∈ Γ 1 ∗ ∖ { 0 } e − 2 π r d ( π 5 r 7 / 2 d − 1 / 2 + π 3 r 3 / 2 d − 5 / 2 ) < 351 ( π 5 V − 1 / 2 A − 17 / 4 + π 3 V − 5 / 2 A − 5 / 4 ) e − 2 π V A − 3 / 2 . \begin{aligned}
\frac{27}{2A^3}\sum_{k=1}^{\infty}\sum_{w\in\Gamma_1^*\setminus\{0\}}e^{-2\pi rd}\left(\pi^5r^{7/2}d^{-1/2}+\pi^3r^{3/2}d^{-5/2}\right) \\
&<351\left(\pi^5V^{-1/2}A^{-17/4}+\pi^3V^{-5/2}A^{-5/4}\right)e^{-2\pi VA^{-3/2}}.
\end{aligned} 2 A 3 27 k = 1 ∑ ∞ w ∈ Γ 1 ∗ ∖ { 0 } ∑ e − 2 π r d ( π 5 r 7/2 d − 1/2 + π 3 r 3/2 d − 5/2 ) < 351 ( π 5 V − 1/2 A − 17/4 + π 3 V − 5/2 A − 5/4 ) e − 2 π V A − 3/2 .
Use Lemma 3.16, since e − 2 π V A − 3 / 2 < 1 / 600000 < 1 / 15000 e^{-2\pi VA^{-3/2}}<1/600000<1/15000 e − 2 π V A − 3/2 < 1/600000 < 1/15000 . The last majorant decreases in V V V and increases in A A A . At ( V , A ) = ( 89 / 50 , 89 / 100 ) (V,A)=(89/50,89/100) ( V , A ) = ( 89/50 , 89/100 ) this loss is less than 1 / 4 1/4 1/4 , hence F V ′ ′ > 34 F_V''>34 F V ′′ > 34 . The same exponential bounds justify termwise differentiation twice. This completes the proof of Lemma 3.18. □
Lemma 3.19 (Three finite area comparisons and common scalar bounds). For the primitive section Γ \Gamma Γ of Lemma 3.6 , set a = min v ∈ Γ ∖ { 0 } ∣ v ∣ a=\min_{v\in\Gamma\setminus\{0\}}|v| a = min v ∈ Γ ∖ { 0 } ∣ v ∣ , A = ∣ Γ ∣ A=|\Gamma| A = ∣Γ∣ and h = V / A h=V/A h = V / A . Suppose
7 10 ≤ a ≤ 8 5 , 1 ≤ A ≤ 7 2 , A ≥ 433 500 a 2 , h 2 ≥ 3 A 2 4 a 2 − a 2 4 . \frac{7}{10}\leq a\leq\frac{8}{5},\qquad1\leq A\leq\frac{7}{2},\qquad A\geq\frac{433}{500}a^2,\qquad h^2\geq\frac{3A^2}{4a^2}-\frac{a^2}{4}. 10 7 ≤ a ≤ 5 8 , 1 ≤ A ≤ 2 7 , A ≥ 500 433 a 2 , h 2 ≥ 4 a 2 3 A 2 − 4 a 2 . For each of the three parameter triples
( u , v , τ ) ∈ { ( 89 50 , 9 5 , − 357 100 ) , ( 9 5 , 37 20 , − 71 20 ) , ( 37 20 , 2 , − 7 2 ) } , (u,v,\tau)\in\left\{\left(\frac{89}{50},\frac{9}{5},-\frac{357}{100}\right),\left(\frac{9}{5},\frac{37}{20},-\frac{71}{20}\right),\left(\frac{37}{20},2,-\frac{7}{2}\right)\right\}, ( u , v , τ ) ∈ { ( 50 89 , 5 9 , − 100 357 ) , ( 5 9 , 20 37 , − 20 71 ) , ( 20 37 , 2 , − 2 7 ) } , if u ≤ V ≤ v u\leq V\leq v u ≤ V ≤ v , then E ( L ) > τ \mathcal{E}(L)>\tau E ( L ) > τ , whereas
E ( T 17 / 20 ⊕ ( v / ( 17 / 20 ) ) Z ) < τ . \mathcal{E}\left(T_{17/20}\oplus\left(v/(17/20)\right)\mathbb{Z}\right)<\tau. E ( T 17/20 ⊕ ( v / ( 17/20 ) ) Z ) < τ . The scalar and Fourier bounds used in Proposition 3.20 hold for all V ≥ 89 / 50 V\geq89/50 V ≥ 89/50 .
Proof. For z = 2 π r d z=2\pi rd z = 2 π r d , use
U 2 ( r , d ) = π 3 r 2 d 2 K 2 ( z ) . U_2(r,d)=\frac{\pi^3r^2}{d^2}K_2(z). U 2 ( r , d ) = d 2 π 3 r 2 K 2 ( z ) .
Lemma 3.2 gives π 3 r 2 d − 2 K 2 ( 2 π r d ) ≤ U 2 ( r , d ) \pi^3r^2d^{-2}K_2(2\pi rd)\leq U_2(r,d) π 3 r 2 d − 2 K 2 ( 2 π r d ) ≤ U 2 ( r , d ) . Monotonicity in r , d r,d r , d reduces, for the radial derivative, to
z 3 + 3 8 z 2 − 15 128 z + 105 256 > 0 ; z^3+\frac{3}{8}z^2-\frac{15}{128}z+\frac{105}{256}>0; z 3 + 8 3 z 2 − 128 15 z + 256 105 > 0 ;
the quadratic formed by the last three terms has negative discriminant. For ρ = a / A \rho=a/A ρ = a / A , Lemma 3.8 gives E ( L ) − E ( Γ ) ≥ − J △ ( a , A , h ) \mathcal{E}(L)-\mathcal{E}(\Gamma)\ge-J_{\triangle}(a,A,h) E ( L ) − E ( Γ ) ≥ − J △ ( a , A , h ) .
For a reduced basis ( a , 0 ) , ( t , A / a ) (a,0),(t,A/a) ( a , 0 ) , ( t , A / a ) , 0 ≤ t ≤ a / 2 0\le t\le a/2 0 ≤ t ≤ a /2 , the independent pair has squared length ≤ A 2 / a 2 + a 2 / 4 \le A^2/a^2+a^2/4 ≤ A 2 / a 2 + a 2 /4 . One-dimensional Poisson summation gives
E ( Γ , 3 ) ≤ 2 ζ ( 6 ) a − 6 + 3 π ζ ( 5 ) 4 B ( A / a 2 ) a 4 A 5 . E(\Gamma,3)\le2\zeta(6)a^{-6}+\frac{3\pi\zeta(5)}{4}B(A/a^2)\frac{a^4}{A^5}. E ( Γ , 3 ) ≤ 2 ζ ( 6 ) a − 6 + 4 3 π ζ ( 5 ) B ( A / a 2 ) A 5 a 4 .
where
B ( y ) = 1 + 2 ∑ n = 1 ∞ e − 2 π n y ( 1 + 2 π n y + ( 2 π n y ) 2 3 ) . B(y)=1+2\sum_{n=1}^{\infty}e^{-2\pi ny}\left(1+2\pi ny+\frac{(2\pi ny)^2}{3}\right). B ( y ) = 1 + 2 n = 1 ∑ ∞ e − 2 π n y ( 1 + 2 π n y + 3 ( 2 π n y ) 2 ) .
Geometric summation, 2 ζ ( 6 ) < 407 / 200 2\zeta(6)<407/200 2 ζ ( 6 ) < 407/200 and ζ ( 5 ) < 1037 / 1000 \zeta(5)<1037/1000 ζ ( 5 ) < 1037/1000 give
B ( y ) < 23 20 ( y ≥ 3 / 2 ) , B(y)<\frac{23}{20}\qquad(y\ge\sqrt{3}/2), B ( y ) < 20 23 ( y ≥ 3 /2 ) ,
3 π ζ ( 5 ) 4 B ( y ) < { 53 / 20 , y ≥ 1 , 2813 / 1000 , 3 / 2 ≤ y < 1. \frac{3\pi\zeta(5)}{4}B(y)<\begin{cases}
53/20,& y\ge1,\\
2813/1000,& \sqrt{3}/2\le y<1.
\end{cases} 4 3 π ζ ( 5 ) B ( y ) < { 53/20 , 2813/1000 , y ≥ 1 , 3 /2 ≤ y < 1.
Retaining the independent pair’s repulsion gives
E ( Γ ) ≥ max { a − 12 − 407 200 a 6 − { 53 / 20 , A ≥ a 2 , 2813 / 1000 , A < a 2 + ( A 2 a 2 + a 2 4 ) − 6 , a − 12 − 13 2 a 6 , − 947 320 } . \mathcal{E}(\Gamma)\ge\max\left\{a^{-12}-\frac{407}{200a^6}-\begin{cases}
53/20,& A\ge a^2,\\
2813/1000,& A<a^2
\end{cases}
+\left(\frac{A^2}{a^2}+\frac{a^2}{4}\right)^{-6},\ a^{-12}-\frac{13}{2a^6},\ -\frac{947}{320}\right\}. E ( Γ ) ≥ max { a − 12 − 200 a 6 407 − { 53/20 , 2813/1000 , A ≥ a 2 , A < a 2 + ( a 2 A 2 + 4 a 2 ) − 6 , a − 12 − 2 a 6 13 , − 320 947 } .
The covering bound R 2 ≤ 5 a 2 / 16 + A 2 / ( 4 a 2 ) R^2\le5a^2/16+A^2/(4a^2) R 2 ≤ 5 a 2 /16 + A 2 / ( 4 a 2 ) retains from the nearest point in the first shifted layer the repulsion
{ ( v / A ) 2 + 5 a 2 16 + A 2 4 a 2 } − 6 , \left\{\left(v/A\right)^2+\frac{5a^2}{16}+\frac{A^2}{4a^2}\right\}^{-6}, { ( v / A ) 2 + 16 5 a 2 + 4 a 2 A 2 } − 6 ,
since V ≤ v V\le v V ≤ v ; the opposite layers give coefficient one. For a ≥ 9 / 8 a\ge9/8 a ≥ 9/8 , the fixed-systole comparison also gives E ( L ) ≥ 6 a − 12 − ( 2901 / 200 ) a − 6 \mathcal{E}(L)\ge6a^{-12}-(2901/200)a^{-6} E ( L ) ≥ 6 a − 12 − ( 2901/200 ) a − 6 .
All three cases use the same 256 by 256 rational grid
a i = 1792 + 9 i 2560 , A j = 1 + 5 j 512 ( 0 ≤ i , j ≤ 256 ) . a_i=\frac{1792+9i}{2560},\qquad A_j=1+\frac{5j}{512}\qquad(0\le i,j\le256). a i = 2560 1792 + 9 i , A j = 1 + 512 5 j ( 0 ≤ i , j ≤ 256 ) .
Discard only cells with A j + 1 < ( 433 / 500 ) a i 2 A_{j+1}<(433/500)a_i^2 A j + 1 < ( 433/500 ) a i 2 . For each retained cell set
H i j = 10 − 6 ⌊ 10 6 max { v 2 A j + 1 2 , 3 A j 2 4 a i + 1 2 − a i + 1 2 4 } ⌋ , ρ i j = a i / A j + 1 . H_{ij}=10^{-6}\left\lfloor10^6\sqrt{\max\left\{\frac{v^2}{A_{j+1}^2},\frac{3A_j^2}{4a_{i+1}^2}-\frac{a_{i+1}^2}{4}\right\}}\right\rfloor,\qquad\rho_{ij}=a_i/A_{j+1}. H ij = 1 0 − 6 1 0 6 max { A j + 1 2 v 2 , 4 a i + 1 2 3 A j 2 − 4 a i + 1 2 } , ρ ij = a i / A j + 1 .
Define
J ‾ i j = π 5 90 A j H i j 4 + 1 1000 + 2 A j ∑ m = 1 5 [ 6 ∑ q ∈ { 1 , 3 , 4 , 9 , 12 , 16 , 25 , 27 , 36 } u 2 ( ρ i j q , m H i j ) + 12 ∑ q ∈ { 7 , 13 , 19 , 21 , 28 , 31 } u 2 ( ρ i j q , m H i j ) ] . \begin{aligned}
\overline{J}_{ij}=\frac{\pi^5}{90A_jH_{ij}^4}+\frac{1}{1000} \\
&+\frac{2}{A_j}\sum_{m=1}^{5}\left[6\sum_{q\in\{1,3,4,9,12,16,25,27,36\}}u_2(\rho_{ij}\sqrt{q},mH_{ij})+12\sum_{q\in\{7,13,19,21,28,31\}}u_2(\rho_{ij}\sqrt{q},mH_{ij})\right].
\end{aligned} J ij = 90 A j H ij 4 π 5 + 1000 1 + A j 2 m = 1 ∑ 5 6 q ∈ { 1 , 3 , 4 , 9 , 12 , 16 , 25 , 27 , 36 } ∑ u 2 ( ρ ij q , m H ij ) + 12 q ∈ { 7 , 13 , 19 , 21 , 28 , 31 } ∑ u 2 ( ρ ij q , m H ij ) .
These multiplicities enumerate exactly 0 < k 2 + k l + l 2 ≤ 36 0<k^2+kl+l^2\le36 0 < k 2 + k l + l 2 ≤ 36 . The reduced-plane bounds give h ρ > 16 / 25 h\rho>16/25 h ρ > 16/25 and π 3 a 3 / 2 V − 5 / 2 < 15 \pi^3a^{3/2}V^{-5/2}<15 π 3 a 3/2 V − 5/2 < 15 . The complete tail m ≥ 6 m\ge6 m ≥ 6 or q > 36 q>36 q > 36 is < 1 / 1000 <1/1000 < 1/1000 , hence J △ < J ‾ i j J_{\triangle}<\overline{J}_{ij} J △ < J ij throughout the cell.
For the k k k th parameter triple, put
d i j = { 53 / 20 , A j ≥ a i + 1 2 , 2813 / 1000 , A j < a i + 1 2 , d_{ij}=\begin{cases}
53/20,& A_j\ge a_{i+1}^2,\\
2813/1000,& A_j<a_{i+1}^2,
\end{cases} d ij = { 53/20 , 2813/1000 , A j ≥ a i + 1 2 , A j < a i + 1 2 ,
and define
L i j ( k ) = max { [ ( v / A j ) 2 + 5 a i + 1 2 16 + A j + 1 2 4 a i 2 ] − 6 − J ‾ i j + max { a i + 1 − 12 − 407 200 a i 6 − d i j a i + 1 4 A j 5 + ( A j + 1 2 a i 2 + a i + 1 2 4 ) − 6 , a i + 1 − 12 − 13 2 a i 6 , − 947 320 } , a i + 1 − 12 − 2901 200 a i 6 } . L_{ij}^{(k)}=\max\left\{\left[\left(v/A_j\right)^2+\frac{5a_{i+1}^2}{16}+\frac{A_{j+1}^2}{4a_i^2}\right]^{-6}-\overline{J}_{ij}
+\max\left\{
\begin{array}{c}
a_{i+1}^{-12}-\frac{407}{200a_i^6}-d_{ij}\frac{a_{i+1}^4}{A_j^5}+\left(\frac{A_{j+1}^2}{a_i^2}+\frac{a_{i+1}^2}{4}\right)^{-6},\\
a_{i+1}^{-12}-\frac{13}{2a_i^6},\quad-\frac{947}{320}
\end{array}
\right\},
a_{i+1}^{-12}-\frac{2901}{200a_i^6}
\right\}. L ij ( k ) = max ⎩ ⎨ ⎧ [ ( v / A j ) 2 + 16 5 a i + 1 2 + 4 a i 2 A j + 1 2 ] − 6 − J ij + max ⎩ ⎨ ⎧ a i + 1 − 12 − 200 a i 6 407 − d ij A j 5 a i + 1 4 + ( a i 2 A j + 1 2 + 4 a i + 1 2 ) − 6 , a i + 1 − 12 − 2 a i 6 13 , − 320 947 ⎭ ⎬ ⎫ , a i + 1 − 12 − 200 a i 6 2901 ⎭ ⎬ ⎫ .
For a i ≥ 9 / 8 a_i \ge9/8 a i ≥ 9/8 , include the additional candidate 6 a i − 12 − ( 2901 / 200 ) a i − 6 6a_i^{-12}-(2901/200)a_i^{-6} 6 a i − 12 − ( 2901/200 ) a i − 6 in this maximum. Each case has 56974 retained and 8562 discarded cells, with
− 3.568443985 < min L i j ( 1 ) < − 3.568443984 , -3.568443985 < \min L_{ij}^{(1)} < -3.568443984, − 3.568443985 < min L ij ( 1 ) < − 3.568443984 ,
− 3.515188971 < min L i j ( 2 ) < − 3.515188970 , -3.515188971 < \min L_{ij}^{(2)} < -3.515188970, − 3.515188971 < min L ij ( 2 ) < − 3.515188970 ,
− 3.404198849 < min L i j ( 3 ) < − 3.404198848. -3.404198849 < \min L_{ij}^{(3)} < -3.404198848. − 3.404198849 < min L ij ( 3 ) < − 3.404198848.
The minimizing index pairs for the three triples above are (151, 73), (154, 75) and (118, 14), respectively.
For the scalar checks, let s 3 , s 6 s_3,s_6 s 3 , s 6 sum q − 3 , q − 6 q^{-3},q^{-6} q − 3 , q − 6 over − 16 ≤ k , l ≤ 16 -16 \le k,l \le16 − 16 ≤ k , l ≤ 16 , 0 < q = k 2 + k l + l 2 ≤ 64 0<q=k^2+kl+l^2\le64 0 < q = k 2 + k l + l 2 ≤ 64 . The disjoint-disk estimate gives
s 3 ≤ S 3 ≤ s 3 + 2 ( 17 / 16 ) 6 ( 15 / 2 ) − 4 < 63771 10000 , s_3\le S_3\le s_3+2(17/16)^6(15/2)^{-4}<\frac{63771}{10000}, s 3 ≤ S 3 ≤ s 3 + 2 ( 17/16 ) 6 ( 15/2 ) − 4 < 10000 63771 ,
s 6 ≤ S 6 ≤ s 6 + 4 5 ( 17 / 16 ) 12 ( 15 / 2 ) − 10 . s_6\le S_6\le s_6+\frac{4}{5}(17/16)^{12}(15/2)^{-10}. s 6 ≤ S 6 ≤ s 6 + 5 4 ( 17/16 ) 12 ( 15/2 ) − 10 .
Lemma 3.12 bounds the endpoint energy by
[ 27 ( s 6 + 4 5 ( 17 / 16 ) 12 ( 15 / 2 ) − 10 ) 128 ( 17 / 20 ) 6 − 3 3 s 3 8 ( 17 / 20 ) 3 ] π 5 ( 17 / 20 ) 3 90 v 4 + 11 / 10 5 ( 49 / 100 ) 2 ( 17 / 20 v ) 10 ( 1 + ( 49 / 100 ) ( 17 / 20 ) v ) 12 < { − 3.5744333 , v = 9 / 5 , − 3.5555334 , v = 37 / 20 , − 3.5101664 , v = 2. \left[\frac{27\left(s_6+\frac{4}{5}(17/16)^{12}(15/2)^{-10}\right)}{128(17/20)^6}-\frac{3\sqrt{3}s_3}{8(17/20)^3}\right]\frac{\pi^5(17/20)^3}{90v^4}
+\frac{11/10}{5(49/100)^2}\left(\frac{17/20}{v}\right)^{10}\left(1+\frac{(49/100)(17/20)}{v}\right)^{12}
<
\begin{cases}
-3.5744333, & v=9/5,\\
-3.5555334, & v=37/20,\\
-3.5101664, & v=2.
\end{cases} [ 128 ( 17/20 ) 6 27 ( s 6 + 5 4 ( 17/16 ) 12 ( 15/2 ) − 10 ) − 8 ( 17/20 ) 3 3 3 s 3 ] 90 v 4 π 5 ( 17/20 ) 3 + 5 ( 49/100 ) 2 11/10 ( v 17/20 ) 10 ( 1 + v ( 49/100 ) ( 17/20 ) ) 12 < ⎩ ⎨ ⎧ − 3.5744333 , − 3.5555334 , − 3.5101664 , v = 9/5 , v = 37/20 , v = 2.
The common nonzero attractive Fourier upper bound is the finite sum
2 89 / 100 ∑ m = 1 8 ∑ ( k , l ) ∈ Z 2 − 10 ≤ k , l ≤ 10 ( k , l ) ≠ ( 0 , 0 ) u 2 ( 7 10 k 2 + k l + l 2 , m ( 89 / 50 ) ) + 10 − 12 < 73 5000 . \frac{2}{89/100}\sum_{m=1}^{8}\sum_{\substack{(k,l)\in\mathbb{Z}^2\\-10\le k,l\le10\\(k,l)\ne(0,0)}}u_2\left(\frac{7}{10}\sqrt{k^2+kl+l^2},m(89/50)\right)+10^{-12}<\frac{73}{5000}. 89/100 2 m = 1 ∑ 8 ( k , l ) ∈ Z 2 − 10 ≤ k , l ≤ 10 ( k , l ) = ( 0 , 0 ) ∑ u 2 ( 10 7 k 2 + k l + l 2 , m ( 89/50 ) ) + 1 0 − 12 < 5000 73 .
The tail allowance follows from the block bound 2000 ( K 6 + K 3 ) e − 5 m K 2000(K^6+K^3)e^{-5mK} 2000 ( K 6 + K 3 ) e − 5 m K , K = max ( ∣ k ∣ , ∣ l ∣ ) K=\max(|k|,|l|) K = max ( ∣ k ∣ , ∣ l ∣ ) . Its two omitted regions are at most
100 99 50 49 2000 ( 11 6 + 11 3 ) 100 − 11 + 200 99 4000 100 − 9 < 10 − 12 . \frac{100}{99}\frac{50}{49}2000(11^6+11^3)100^{-11}+\frac{200}{99}4000\,100^{-9}<10^{-12}. 99 100 49 50 2000 ( 1 1 6 + 1 1 3 ) 10 0 − 11 + 99 200 4000 10 0 − 9 < 1 0 − 12 .
For the remaining comparisons, use
3 3 8 s 3 ≤ E ( T 1 , 3 ) ≤ 3 3 8 63771 10000 , \frac{3\sqrt{3}}{8}s_3\le E(T_1,3)\le\frac{3\sqrt{3}}{8}\frac{63771}{10000}, 8 3 3 s 3 ≤ E ( T 1 , 3 ) ≤ 8 3 3 10000 63771 ,
27 64 s 6 ≤ E ( T 1 , 6 ) ≤ 27 64 ( s 6 + 4 5 ( 17 / 16 ) 12 ( 15 / 2 ) − 10 ) . \frac{27}{64}s_6\le E(T_1,6)\le\frac{27}{64}\left(s_6+\frac{4}{5}(17/16)^{12}(15/2)^{-10}\right). 64 27 s 6 ≤ E ( T 1 , 6 ) ≤ 64 27 ( s 6 + 5 4 ( 17/16 ) 12 ( 15/2 ) − 10 ) .
Then
21 ( 27 64 s 6 ) − 12 ( 3 3 8 63771 10000 ) > 0 , 21\left(\frac{27}{64}s_6\right)-12\left(\frac{3\sqrt{3}}{8}\frac{63771}{10000}\right)>0, 21 ( 64 27 s 6 ) − 12 ( 8 3 3 10000 63771 ) > 0 ,
27 64 ( s 6 + 4 5 ( 17 / 16 ) 12 ( 15 / 2 ) − 10 ) 2 [ ( 89 / 100 ) − 6 − ( 17 / 20 ) − 6 ] − ( 3 3 8 s 3 ) [ ( 89 / 100 ) − 3 − ( 17 / 20 ) − 3 ] > 291 5000 , \frac{\frac{27}{64}\left(s_6+\frac{4}{5}(17/16)^{12}(15/2)^{-10}\right)}{2}\left[(89/100)^{-6}-(17/20)^{-6}\right]
-\left(\frac{3\sqrt{3}}{8}s_3\right)\left[(89/100)^{-3}-(17/20)^{-3}\right]>\frac{291}{5000}, 2 64 27 ( s 6 + 5 4 ( 17/16 ) 12 ( 15/2 ) − 10 ) [ ( 89/100 ) − 6 − ( 17/20 ) − 6 ] − ( 8 3 3 s 3 ) [ ( 89/100 ) − 3 − ( 17/20 ) − 3 ] > 5000 291 ,
3 ( 89 / 100 ) − 7 [ ( 3 3 8 s 3 ) ( 89 / 100 ) 3 − 27 64 ( s 6 + 4 5 ( 17 / 16 ) 12 ( 15 / 2 ) − 10 ) ] − 3 ( 121 / 35 ) ( 89 / 50 ) 4 > 157 100 . 3(89/100)^{-7}\left[\left(\frac{3\sqrt{3}}{8}s_3\right)(89/100)^3-\frac{27}{64}\left(s_6+\frac{4}{5}(17/16)^{12}(15/2)^{-10}\right)\right]
-\frac{3(121/35)}{(89/50)^4}>\frac{157}{100}. 3 ( 89/100 ) − 7 [ ( 8 3 3 s 3 ) ( 89/100 ) 3 − 64 27 ( s 6 + 5 4 ( 17/16 ) 12 ( 15/2 ) − 10 ) ] − ( 89/50 ) 4 3 ( 121/35 ) > 100 157 .
The first inequality gives U ′ ′ > 0 U^{\prime\prime}>0 U ′′ > 0 up to area one; the third gives ∂ A { U ( A ) − π ζ ( 4 ) A 3 / V 4 } > 157 / 100 \partial_A\{U(A)-\pi\zeta(4)A^3/V^4\}>157/100 ∂ A { U ( A ) − π ζ ( 4 ) A 3 / V 4 } > 157/100 for 89 / 100 ≤ A ≤ 1 89/100\le A\le1 89/100 ≤ A ≤ 1 and V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 . For V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 , the zero-mode and reference-energy losses satisfy
π ζ ( 4 ) V 4 [ ( 89 / 100 ) 3 − ( 17 / 20 ) 3 ] < 157 5000 , \frac{\pi\zeta(4)}{V^4}\left[(89/100)^3-(17/20)^3\right]<\frac{157}{5000}, V 4 π ζ ( 4 ) [ ( 89/100 ) 3 − ( 17/20 ) 3 ] < 5000 157 ,
E ( T 17 / 20 ⊕ V 17 / 20 Z ) − U ( 17 / 20 ) + π ζ ( 4 ) ( 17 / 20 ) 3 V 4 < 71 10000 . \mathcal{E}\left(T_{17/20}\oplus\frac{V}{17/20}\mathbb{Z}\right)-U(17/20)+\frac{\pi\zeta(4)(17/20)^3}{V^4}<\frac{71}{10000}. E ( T 17/20 ⊕ 17/20 V Z ) − U ( 17/20 ) + V 4 π ζ ( 4 ) ( 17/20 ) 3 < 10000 71 .
The second bound uses Lemma 3.12; Lemma 3.18 supplies the area Hessian bounds for all modes.
Enclose every finite sum, tail and scalar bound as in Lemma 3.13, using the rational π \pi π enclosure and binary-scaled bounds T 21 ≤ e − z ≤ T 20 T_{21}\le e^{-z}\le T_{20} T 21 ≤ e − z ≤ T 20 of Lemma A.1 . Check radical endpoints by integer squaring and height floors by exact integer square roots. This completes the proof of Lemma 3.19. □
Proposition 3.20 (The effective simple-hexagonal phase). For V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 ,
∃ ! A ( V ) ∈ ( 2 / 5 , 89 / 100 ) : F V ′ ( A ( V ) ) = 0 , V A ( V ) > 2 , \exists! A(V)\in(2/5,89/100):\quad F_V^{\prime}(A(V))=0,\qquad\frac{V}{A(V)}>2, ∃ ! A ( V ) ∈ ( 2/5 , 89/100 ) : F V ′ ( A ( V )) = 0 , A ( V ) V > 2 , and
argmin ∣ L ∣ = V E ( L ) = { Q ( T A ( V ) ⊕ V A ( V ) Z ) : Q ∈ O ( 3 ) } . \underset{|L|=V}{\operatorname{argmin}}\mathcal{E}(L)=\left\{Q\left(T_{A(V)}\oplus\frac{V}{A(V)}\mathbb{Z}\right):Q\in O(3)\right\}. ∣ L ∣ = V argmin E ( L ) = { Q ( T A ( V ) ⊕ A ( V ) V Z ) : Q ∈ O ( 3 ) } .
Proof. Since V / A ∗ > 1 V/A_*>1 V / A ∗ > 1 , m ( V ) ≤ F V ( A ∗ ) < e 2 m(V)\le F_V(A_*)<e_2 m ( V ) ≤ F V ( A ∗ ) < e 2 , so every minimizer has 7 / 10 < a < 8 / 5 7/10<a<8/5 7/10 < a < 8/5 and A < 5 A<5 A < 5 . The A ≥ 7 / 2 A\ge7/2 A ≥ 7/2 exclusion in Lemma 3.14 holds for all V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 .
For 1 ≤ A ≤ 7 / 2 1\le A\le7/2 1 ≤ A ≤ 7/2 and each case u ≤ V ≤ v u\le V\le v u ≤ V ≤ v of Lemma 3.19,
E ( L ) > τ > E ( T 17 / 20 ⊕ ( v / ( 17 / 20 ) ) Z ) ≥ E ( T 17 / 20 ⊕ ( V / ( 17 / 20 ) ) Z ) . \mathcal{E}(L)>\tau>\mathcal{E}\left(T_{17/20}\oplus\left(v/(17/20)\right)\mathbb{Z}\right)\ge\mathcal{E}\left(T_{17/20}\oplus\left(V/(17/20)\right)\mathbb{Z}\right). E ( L ) > τ > E ( T 17/20 ⊕ ( v / ( 17/20 ) ) Z ) ≥ E ( T 17/20 ⊕ ( V / ( 17/20 ) ) Z ) .
All affected distances are ≥ V / ( 17 / 20 ) > 1 \ge V/(17/20)>1 ≥ V / ( 17/20 ) > 1 , where r − 12 − 2 r − 6 r^{-12}-2r^{-6} r − 12 − 2 r − 6 increases with r r r . For V ≥ 2 V\ge2 V ≥ 2 , Lemma 3.13 gives E ( L ) > − 10 / 3 > e 2 \mathcal{E}(L)>-10/3>e_2 E ( L ) > − 10/3 > e 2 . Thus every minimizer has A < 1 A<1 A < 1 .
For 89 / 100 ≤ A ≤ 1 89/100\le A\le1 89/100 ≤ A ≤ 1 and V ≥ 89 / 50 V\ge89/50 V ≥ 89/50 , Lemma 3.19 gives
∂ A { U ( A ) − π ζ ( 4 ) A 3 / V 4 } > 157 / 100 > 0. \partial_A\{U(A)-\pi\zeta(4)A^3/V^4\}>157/100>0. ∂ A { U ( A ) − π ζ ( 4 ) A 3 / V 4 } > 157/100 > 0.
Compare the scalar part at A = 89 / 100 A=89/100 A = 89/100 with the reference at 17 / 20 17/20 17/20 . The planar gain is > 291 / 5000 >291/5000 > 291/5000 ; the zero-mode, attractive Fourier and reference-repulsion losses are respectively < 157 / 5000 <157/5000 < 157/5000 , 73 / 5000 73/5000 73/5000 , 71 / 10000 71/10000 71/10000 . Hence
E ( L ) − E ( T 17 / 20 ⊕ ( V / ( 17 / 20 ) ) Z ) > 51 10000 > 0. \mathcal{E}(L)-\mathcal{E}\left(T_{17/20}\oplus(V/(17/20))\mathbb{Z}\right)>\frac{51}{10000}>0. E ( L ) − E ( T 17/20 ⊕ ( V / ( 17/20 )) Z ) > 10000 51 > 0.
Consequently 2 / 5 < ( 3 / 2 ) ( 7 / 10 ) 2 ≤ A < 89 / 100 2/5<(\sqrt{3}/2)(7/10)^2\le A<89/100 2/5 < ( 3 /2 ) ( 7/10 ) 2 ≤ A < 89/100 and h = V / A > 2 h=V/A>2 h = V / A > 2 .
Lemmas 3.15 and 3.17 now give
L ≅ T A ⊕ ( V / A ) Z , F V ′ ( A ) = 0. L\cong T_A\oplus(V/A)\mathbb{Z},\qquad F_V^{\prime}(A)=0. L ≅ T A ⊕ ( V / A ) Z , F V ′ ( A ) = 0.
Attainment gives a stationary area in ( 2 / 5 , 89 / 100 ) (2/5,89/100) ( 2/5 , 89/100 ) ; F V ′ ′ > 34 F_V'' > 34 F V ′′ > 34 by Lemma Lemma 3.18 gives uniqueness. This completes the proof of Proposition Proposition 3.20 . □
For the dilute asymptotics, Lemma Lemma 3.4 gives 2 / 5 < A ∗ < 43 / 50 2/5 < A_* < 43/50 2/5 < A ∗ < 43/50 . By (28) ,
U ′ ( A ∗ ) = 0 , U ′ ′ ( A ∗ ) = 9 S 3 2 S 6 A ∗ 2 > 0. (32) U'(A_*) = 0,\qquad U''(A_*) = \frac{9S_3^2}{S_6A_*^2} > 0. \tag*{(32)} U ′ ( A ∗ ) = 0 , U ′′ ( A ∗ ) = S 6 A ∗ 2 9 S 3 2 > 0. (32)
Equations (30) –(31) give, uniformly with any fixed number of A A A -derivatives near A ∗ A_* A ∗ ,
F V ( A ) = U ( A ) − π ζ ( 4 ) A 3 V − 4 + π ζ ( 10 ) 5 A 9 V − 10 + O ( e − c V ) . (33) F_V(A) = U(A) - \pi\zeta(4)A^3V^{-4} + \frac{\pi\zeta(10)}{5}A^9V^{-10} + O(e^{-cV}). \tag*{(33)} F V ( A ) = U ( A ) − π ζ ( 4 ) A 3 V − 4 + 5 π ζ ( 10 ) A 9 V − 10 + O ( e − c V ) . (33)
Since A ∗ , A ( V ) ∈ ( 2 / 5 , 89 / 100 ) A_*, A(V) \in(2/5,89/100) A ∗ , A ( V ) ∈ ( 2/5 , 89/100 ) , F V ′ ( A ( V ) ) = 0 F_V'(A(V)) = 0 F V ′ ( A ( V )) = 0 and F V ′ ′ > 34 F_V'' > 34 F V ′′ > 34 , the mean value theorem yields
δ : = A ( V ) − A ∗ , ∣ δ ∣ ≤ ∣ F V ′ ( A ∗ ) ∣ 34 = O ( V − 4 ) . \delta:= A(V)-A_*,\qquad|\delta| \le\frac{|F_V'(A_*)|}{34} = O(V^{-4}). δ := A ( V ) − A ∗ , ∣ δ ∣ ≤ 34 ∣ F V ′ ( A ∗ ) ∣ = O ( V − 4 ) .
Consequently,
0 = 9 S 3 2 S 6 A ∗ 2 δ − 3 π ζ ( 4 ) A ∗ 2 V − 4 + O ( δ 2 + ∣ δ ∣ V − 4 + V − 10 ) , 0 = \frac{9S_3^2}{S_6A_*^2}\delta- 3\pi\zeta(4)A_*^2V^{-4} + O(\delta^2 + |\delta|V^{-4} + V^{-10}), 0 = S 6 A ∗ 2 9 S 3 2 δ − 3 π ζ ( 4 ) A ∗ 2 V − 4 + O ( δ 2 + ∣ δ ∣ V − 4 + V − 10 ) ,
δ = π ζ ( 4 ) S 6 A ∗ 4 3 S 3 2 V − 4 + O ( V − 8 ) , \delta= \frac{\pi\zeta(4)S_6A_*^4}{3S_3^2}V^{-4} + O(V^{-8}), δ = 3 S 3 2 π ζ ( 4 ) S 6 A ∗ 4 V − 4 + O ( V − 8 ) ,
m ( V ) = e 2 − π ζ ( 4 ) A ∗ 3 V − 4 + 9 S 3 2 2 S 6 A ∗ 2 δ 2 − 3 π ζ ( 4 ) A ∗ 2 δ V − 4 + π ζ ( 10 ) 5 A ∗ 9 V − 10 + O ( V − 12 ) . m(V) = e_2 - \pi\zeta(4)A_*^3V^{-4} + \frac{9S_3^2}{2S_6A_*^2}\delta^2 - 3\pi\zeta(4)A_*^2\delta V^{-4}
+ \frac{\pi\zeta(10)}{5}A_*^9V^{-10} + O(V^{-12}). m ( V ) = e 2 − π ζ ( 4 ) A ∗ 3 V − 4 + 2 S 6 A ∗ 2 9 S 3 2 δ 2 − 3 π ζ ( 4 ) A ∗ 2 δ V − 4 + 5 π ζ ( 10 ) A ∗ 9 V − 10 + O ( V − 12 ) .
Substitute δ \delta δ to obtain (1.12) and (1.13).
Stationarity and endpoint comparisons
Stationarity gives m ( V 1 ) < m ( V 2 ) m(V_1) < m(V_2) m ( V 1 ) < m ( V 2 ) for 1 ≤ V 1 < V 2 1 \le V_1 < V_2 1 ≤ V 1 < V 2 . Korkine–Zolotarev reduction [19 , 20 ] and stationary-basin estimates give the endpoint minima at V = 21 / 20 , 11 / 10 V = 21/20, 11/10 V = 21/20 , 11/10 .
Proposition 4.1 (Stationarity and stability at fixed covolume). For a lattice L ⊂ R 3 L \subset\mathbb{R}^3 L ⊂ R 3 , with E ( L ) = 1 2 E ( L , 6 ) − E ( L , 3 ) \mathcal{E}(L) = \frac{1}{2}\mathcal{E}(L,6) - \mathcal{E}(L,3) E ( L ) = 2 1 E ( L , 6 ) − E ( L , 3 ) , fixed-covolume stationarity is characterized by
d d t E ( e t H / 2 L ) ∣ t = 0 = 0 for every H = H T , tr H = 0 \left.\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{E}(e^{tH/2}L)\right|_{t=0} = 0 \qquad\text{for every } H = H^{\mathsf T},\ \operatorname{tr} H = 0 d t d E ( e t H /2 L ) t = 0 = 0 for every H = H T , tr H = 0 ⟺ ∑ v ∈ L ∖ { 0 } ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v v T = E ( L , 3 ) − E ( L , 6 ) 3 I . (34) \Longleftrightarrow\sum_{v\in L\setminus\{0\}}\left(|v|^{-8} - |v|^{-14}\right)vv^{\mathsf T} = \frac{\mathcal{E}(L,3)-\mathcal{E}(L,6)}{3}I. \tag*{(34)} ⟺ v ∈ L ∖ { 0 } ∑ ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v v T = 3 E ( L , 3 ) − E ( L , 6 ) I . (34) Under (34) , for H = H T H = H^{\mathsf T} H = H T , tr H = 0 \operatorname{tr} H = 0 tr H = 0 ,
d 2 d t 2 E ( e t H / 2 L ) ∣ t = 0 = ( E ( L , 3 ) − E ( L , 6 ) ) tr ( H 2 ) + 3 ∑ v ∈ L ∖ { 0 } ( 7 ∣ v ∣ − 16 − 4 ∣ v ∣ − 10 ) ( v T H v ) 2 . (35) \left.\frac{\mathrm{d}^2}{\mathrm{d}t^2}\mathcal{E}(e^{tH/2}L)\right|_{t=0} = \bigl(\mathcal{E}(L,3)-\mathcal{E}(L,6)\bigr)\operatorname{tr}(H^2)
+ 3\sum_{v\in L\setminus\{0\}}\bigl(7|v|^{-16} - 4|v|^{-10}\bigr)(v^{\mathsf T}Hv)^2. \tag*{(35)} d t 2 d 2 E ( e t H /2 L ) t = 0 = ( E ( L , 3 ) − E ( L , 6 ) ) tr ( H 2 ) + 3 v ∈ L ∖ { 0 } ∑ ( 7∣ v ∣ − 16 − 4∣ v ∣ − 10 ) ( v T H v ) 2 . (35) Consequently
L ∈ argmin ∣ L ~ ∣ = ∣ L ∣ E ( L ~ ) L \in\underset{|\widetilde L|=|L|}{\operatorname{argmin}}\mathcal{E}(\widetilde L) L ∈ ∣ L ∣ = ∣ L ∣ argmin E ( L ) ⟹ E ( L , 3 ) ≤ 3 E ( L , 6 ) , min v ∈ L ∖ { 0 } ∣ v ∣ < 3 1 / 6 . (36) \Longrightarrow\mathcal{E}(L,3) \le3\mathcal{E}(L,6),\qquad\min_{v\in L\setminus\{0\}} |v| < 3^{1/6}. \tag*{(36)} ⟹ E ( L , 3 ) ≤ 3 E ( L , 6 ) , v ∈ L ∖ { 0 } min ∣ v ∣ < 3 1/6 . (36)
Proof. For symmetric trace-free H H H ,
∂ ε E ( e ε H / 2 L ) ∣ ε = 0 = 3 ∑ v ∈ L ∖ { 0 } ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v T H v . \left.\partial_{\varepsilon}\mathcal{E}(e^{\varepsilon H/2}L)\right|_{\varepsilon=0}=3\sum_{v\in L\setminus\{0\}}\left(|v|^{-8}-|v|^{-14}\right)v^{T}Hv. ∂ ε E ( e ε H /2 L ) ε = 0 = 3 v ∈ L ∖ { 0 } ∑ ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v T H v .
∂ ε 2 E ( e ε H / 2 L ) ∣ ε = 0 = 3 ∑ v ∈ L ∖ { 0 } [ ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v T H 2 v + ( 7 ∣ v ∣ − 16 − 4 ∣ v ∣ − 10 ) ( v T H v ) 2 ] . \left.\partial_{\varepsilon}^{2}\mathcal{E}(e^{\varepsilon H/2}L)\right|_{\varepsilon=0}=3\sum_{v\in L\setminus\{0\}}\left[\left(|v|^{-8}-|v|^{-14}\right)v^{T}H^{2}v+\left(7|v|^{-16}-4|v|^{-10}\right)(v^{T}Hv)^{2}\right]. ∂ ε 2 E ( e ε H /2 L ) ε = 0 = 3 v ∈ L ∖ { 0 } ∑ [ ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) v T H 2 v + ( 7∣ v ∣ − 16 − 4∣ v ∣ − 10 ) ( v T H v ) 2 ] .
Thus stationarity is equivalent to (34) , and (35) follows. For a Frobenius-orthonormal basis H 1 , … , H 5 H_{1},\ldots,H_{5} H 1 , … , H 5 of { H ∈ R 3 × 3 : H = H T , tr H = 0 } \{H\in\mathbb{R}^{3\times3}:H=H^{T},\operatorname{tr}H=0\} { H ∈ R 3 × 3 : H = H T , tr H = 0 } ,
∑ j = 1 5 ( u T H j u ) 2 = ∥ u u T − I / 3 ∥ F 2 = 2 / 3 ( ∣ u ∣ = 1 ) . \sum_{j=1}^{5}(u^{T}H_{j}u)^{2}=\left\|uu^{T}-I/3\right\|_{F}^{2}=2/3\qquad(|u|=1). j = 1 ∑ 5 ( u T H j u ) 2 = u u T − I /3 F 2 = 2/3 ( ∣ u ∣ = 1 ) .
Consequently
∑ j = 1 5 ∂ ε 2 E ( e ε H j / 2 L ) ∣ ε = 0 = 9 E ( L , 6 ) − 3 E ( L , 3 ) . (37) \sum_{j=1}^{5}\left.\partial_{\varepsilon}^{2}\mathcal{E}(e^{\varepsilon H_{j}/2}L)\right|_{\varepsilon=0}=9E(L,6)-3E(L,3). \tag*{(37)} j = 1 ∑ 5 ∂ ε 2 E ( e ε H j /2 L ) ε = 0 = 9 E ( L , 6 ) − 3 E ( L , 3 ) . (37)
At a local minimum, ∑ j = 1 5 ∂ ε 2 E ( e ε H j / 2 L ) ∣ ε = 0 ≥ 0 \sum_{j=1}^{5}\left.\partial_{\varepsilon}^{2}\mathcal{E}(e^{\varepsilon H_{j}/2}L)\right|_{\varepsilon=0}\geq0 ∑ j = 1 5 ∂ ε 2 E ( e ε H j /2 L ) ε = 0 ≥ 0 and
E ( L , 3 ) = ∑ v ∈ L ∖ { 0 } ∣ v ∣ 6 ∣ v ∣ − 12 > λ 1 ( L ) 6 E ( L , 6 ) , E(L,3)=\sum_{v\in L\setminus\{0\}}|v|^{6}|v|^{-12}>\lambda_{1}(L)^{6}E(L,6), E ( L , 3 ) = v ∈ L ∖ { 0 } ∑ ∣ v ∣ 6 ∣ v ∣ − 12 > λ 1 ( L ) 6 E ( L , 6 ) ,
hence (36) . At every fixed order, C ( ∣ v ∣ − 6 + ∣ v ∣ − 12 ) C(|v|^{-6}+|v|^{-12}) C ( ∣ v ∣ − 6 + ∣ v ∣ − 12 ) bounds the local derivatives and justifies differentiation. This completes the proof of Proposition 4.1. □
Proposition 4.2 (A positive dilation derivative). If ∣ L ∣ ≥ 1 |L|\geq1 ∣ L ∣ ≥ 1 and (34) holds, then
E ( L , 3 ) > E ( L , 6 ) , d d t E ( e t L ) ∣ t = 0 = 6 ( E ( L , 3 ) − E ( L , 6 ) ) > 0. (38) E(L,3)>E(L,6),\qquad\left.\frac{\mathrm{d}}{\mathrm{d}t}\mathcal{E}(e^{t}L)\right|_{t=0}=6(E(L,3)-E(L,6))>0. \tag*{(38)} E ( L , 3 ) > E ( L , 6 ) , d t d E ( e t L ) t = 0 = 6 ( E ( L , 3 ) − E ( L , 6 )) > 0. (38) For L ∈ argmin loc , ∣ L ~ ∣ = ∣ L ∣ E ( L ~ ) L\in\operatorname*{argmin}_{\operatorname{loc},|\,\widetilde{L}\,|=|L|}\mathcal{E}(\widetilde{L}) L ∈ argmin loc , ∣ L ∣ = ∣ L ∣ E ( L ) and ∣ L ∣ ≥ 1 |L|\geq1 ∣ L ∣ ≥ 1 , one has E ( L ) < 0 \mathcal{E}(L)<0 E ( L ) < 0 and
− 6 E ( L ) 5 ≤ E ( L , 3 ) < − 2 E ( L ) , − 2 E ( L ) 5 ≤ E ( L , 6 ) < − 2 E ( L ) . -\frac{6\mathcal{E}(L)}{5}\leq E(L,3)<-2\mathcal{E}(L),\qquad-\frac{2\mathcal{E}(L)}{5}\leq E(L,6)<-2\mathcal{E}(L). − 5 6 E ( L ) ≤ E ( L , 3 ) < − 2 E ( L ) , − 5 2 E ( L ) ≤ E ( L , 6 ) < − 2 E ( L ) . Proof. Put W = span R { v ∈ L : ∣ v ∣ ≤ 1 } W=\operatorname{span}_{\mathbb{R}}\{v\in L:|v|\leq1\} W = span R { v ∈ L : ∣ v ∣ ≤ 1 } . For W ≠ R 3 W\ne\mathbb{R}^{3} W = R 3 , take n ∈ W ⊥ n\in W^{\perp} n ∈ W ⊥ , ∣ n ∣ = 1 |n|=1 ∣ n ∣ = 1 . Then (34) gives
E ( L , 3 ) − E ( L , 6 ) 3 = ∑ v ∈ L ∣ v ∣ > 1 ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) ( v ⋅ n ) 2 > 0. \frac{E(L,3)-E(L,6)}{3}=\sum_{\substack{v\in L\\ |v|>1}}\left(|v|^{-8}-|v|^{-14}\right)(v\cdot n)^{2}>0. 3 E ( L , 3 ) − E ( L , 6 ) = v ∈ L ∣ v ∣ > 1 ∑ ( ∣ v ∣ − 8 − ∣ v ∣ − 14 ) ( v ⋅ n ) 2 > 0. The sum is positive since span R L = R 3 \operatorname{span}_{\mathbb{R}}L=\mathbb{R}^{3} span R L = R 3 . If W = R 3 W=\mathbb{R}^{3} W = R 3 , choose v 1 , v 2 , v 3 ∈ L v_{1},v_{2},v_{3}\in L v 1 , v 2 , v 3 ∈ L with det ( v 1 , v 2 , v 3 ) ≠ 0 \det(v_{1},v_{2},v_{3})\ne0 det ( v 1 , v 2 , v 3 ) = 0 , max 1 ≤ i ≤ 3 ∣ v i ∣ ≤ 1 \max_{1\leq i\leq3}|v_{i}|\leq1 max 1 ≤ i ≤ 3 ∣ v i ∣ ≤ 1 , and k = [ L : ∑ i = 1 3 Z v i ] k=[L:\sum_{i=1}^{3}\mathbb{Z}v_{i}] k = [ L : ∑ i = 1 3 Z v i ] . Hadamard gives
1 ≤ ∣ L ∣ ≤ k ∣ L ∣ = ∣ det ( v 1 , v 2 , v 3 ) ∣ ≤ ∣ v 1 ∣ ∣ v 2 ∣ ∣ v 3 ∣ ≤ 1. 1\leq|L|\leq k|L|=|\det(v_{1},v_{2},v_{3})|\leq|v_{1}||v_{2}||v_{3}|\leq1. 1 ≤ ∣ L ∣ ≤ k ∣ L ∣ = ∣ det ( v 1 , v 2 , v 3 ) ∣ ≤ ∣ v 1 ∣∣ v 2 ∣∣ v 3 ∣ ≤ 1. Thus ∣ L ∣ = k = 1 |L|=k=1 ∣ L ∣ = k = 1 , L ≅ Z 3 L\cong\mathbb{Z}^{3} L ≅ Z 3 and E ( L , 3 ) > E ( L , 6 ) E(L,3)>E(L,6) E ( L , 3 ) > E ( L , 6 ) termwise. Homogeneity gives (38) ; at a local minimum,
E ( L , 6 ) < E ( L , 3 ) ≤ 3 E ( L , 6 ) , E ( L ) = 1 2 E ( L , 6 ) − E ( L , 3 ) , E(L,6)<E(L,3)\leq3E(L,6),\qquad\mathcal{E}(L)=\frac{1}{2}E(L,6)-E(L,3), E ( L , 6 ) < E ( L , 3 ) ≤ 3 E ( L , 6 ) , E ( L ) = 2 1 E ( L , 6 ) − E ( L , 3 ) , which give E ( L ) < 0 \mathcal{E}(L)<0 E ( L ) < 0 and the stated moment bounds. This completes the proof of Proposition 4.2. □
Proposition 4.3 (Monotonicity of the minimum energy). The minimum m ( V ) m(V) m ( V ) is strictly increasing on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) .
Proof. For 1 ≤ V 1 < V 2 1\leq V_{1}<V_{2} 1 ≤ V 1 < V 2 , (17) , (16) and Mahler’s criterion [34 ] give compact sublevel sets
{ L : V 1 ≤ ∣ L ∣ ≤ V 2 , 1 2 E ( L , 6 ) − E ( L , 3 ) ≤ C } ( C ∈ R ) . \left\{L:V_{1}\leq|L|\leq V_{2},\quad\frac{1}{2}E(L,6)-E(L,3)\leq C\right\}\qquad(C\in\mathbb{R}). { L : V 1 ≤ ∣ L ∣ ≤ V 2 , 2 1 E ( L , 6 ) − E ( L , 3 ) ≤ C } ( C ∈ R ) .
Hence a minimizer L ∗ L_\ast L ∗ on V 1 ≤ ∣ L ∣ ≤ V 2 V_1 \le|L| \le V_2 V 1 ≤ ∣ L ∣ ≤ V 2 exists and satisfies (34) . If ∣ L ∗ ∣ > V 1 |L_\ast| > V_1 ∣ L ∗ ∣ > V 1 , Proposition 4.2 gives, for small t > 0 t > 0 t > 0 ,
V 1 ≤ ∣ e − t L ∗ ∣ < ∣ L ∗ ∣ , E ( e − t L ∗ ) < E ( L ∗ ) , V_1 \le|e^{-tL_\ast}| < |L_\ast|,\qquad\mathcal{E}(e^{-tL_\ast}) < \mathcal{E}(L_\ast), V 1 ≤ ∣ e − t L ∗ ∣ < ∣ L ∗ ∣ , E ( e − t L ∗ ) < E ( L ∗ ) ,
a contradiction. Hence
argmin V 1 ≤ ∣ L ∣ ≤ V 2 E ( L ) ⊂ { L : ∣ L ∣ = V 1 } , m ( V 2 ) > m ( V 1 ) . \underset{V_1\le|L|\le V_2}{\operatorname{argmin}}\ \mathcal{E}(L) \subset\{L: |L|=V_1\},\qquad m(V_2)>m(V_1). V 1 ≤ ∣ L ∣ ≤ V 2 argmin E ( L ) ⊂ { L : ∣ L ∣ = V 1 } , m ( V 2 ) > m ( V 1 ) .
This completes the proof of Proposition 4.3 . □
Lemma 4.4 (Uniform separation on a covolume strip). Suppose
21 20 ≤ V = ∣ L ∣ ≤ 11 10 , E ( L ) ≤ − 124 25 . \frac{21}{20}\le V=|L|\le\frac{11}{10},\qquad\mathcal{E}(L)\le-\frac{124}{25}. 20 21 ≤ V = ∣ L ∣ ≤ 10 11 , E ( L ) ≤ − 25 124 . Then
λ 1 ( L ) > 4 5 , λ 1 ( L ∗ ) > 10 13 . (39) \lambda_1(L)>\frac{4}{5},\qquad\lambda_1(L^\ast)>\frac{10}{13}. \tag*{(39)} λ 1 ( L ) > 5 4 , λ 1 ( L ∗ ) > 13 10 . (39) Equivalently, every primitive planar section has layer spacing less than 13 / 10 13/10 13/10 .
Proof. The general separation bound gives λ 1 ( L ) > 7 / 10 \lambda_1(L)>7/10 λ 1 ( L ) > 7/10 .
For a primitive planar section Γ \Gamma Γ of area A = V / h A=V/h A = V / h , write L = Γ + Z ( u + h ν ) L=\Gamma+\mathbb{Z}(u+h\nu) L = Γ + Z ( u + h ν ) . If λ 1 ( L ) > a 0 \lambda_1(L)>a_0 λ 1 ( L ) > a 0 , then λ 1 ( Γ ∗ ) > a 0 / A \lambda_1(\Gamma^\ast)>a_0/A λ 1 ( Γ ∗ ) > a 0 / A . A Gauss-reduced reciprocal basis consequently gives
∣ w i , j ∣ ≥ a 0 h V i 2 + j 2 − ∣ i j ∣ . |w_{i,j}|\ge\frac{a_0h}{V}\sqrt{i^2+j^2-|ij|}. ∣ w i , j ∣ ≥ V a 0 h i 2 + j 2 − ∣ ij ∣ .
The bounds S 3 < 63771 / 10000 S_3<63771/10000 S 3 < 63771/10000 and S 6 ≥ 6 S_6\ge6 S 6 ≥ 6 give
E ( Γ ) ≥ e 2 = − S 3 2 2 S 6 > − ( 63771 / 10000 ) 2 12 > − 17 5 . \mathcal{E}(\Gamma)\ge e_2=-\frac{S_3^2}{2S_6}>-\frac{(63771/10000)^2}{12}>-\frac{17}{5}. E ( Γ ) ≥ e 2 = − 2 S 6 S 3 2 > − 12 ( 63771/10000 ) 2 > − 5 17 .
Using K 2 , K 5 K_2,K_5 K 2 , K 5 from Lemma 3.2 , set
r i , j = a 0 h V i 2 + j 2 − ∣ i j ∣ , r_{i,j}=\frac{a_0h}{V}\sqrt{i^2+j^2-|ij|}, r i , j = V a 0 h i 2 + j 2 − ∣ ij ∣ ,
and define
R ( V , h , a 0 ) = h V ∑ m = 1 ∞ ∑ ( i , j ) ∈ Z 2 ∖ { ( 0 , 0 ) } { π 6 60 ( r i , j m h ) 5 K 5 ( 2 π m h r i , j ) + 2 π 3 ( r i , j m h ) 2 K 2 ( 2 π m h r i , j ) } . R(V,h,a_0)=\frac{h}{V}\sum_{m=1}^{\infty}\sum_{(i,j)\in\mathbb{Z}^2\setminus\{(0,0)\}}\left\{\frac{\pi^6}{60}\left(\frac{r_{i,j}}{mh}\right)^5K_5(2\pi mhr_{i,j})+2\pi^3\left(\frac{r_{i,j}}{mh}\right)^2K_2(2\pi mhr_{i,j})\right\}. R ( V , h , a 0 ) = V h m = 1 ∑ ∞ ( i , j ) ∈ Z 2 ∖ {( 0 , 0 )} ∑ { 60 π 6 ( mh r i , j ) 5 K 5 ( 2 π mh r i , j ) + 2 π 3 ( mh r i , j ) 2 K 2 ( 2 π mh r i , j ) } .
Substitute the reciprocal-length floors by monotonicity of r ν K ν ( 2 π d r ) r^\nu K_\nu(2\pi dr) r ν K ν ( 2 π d r ) . Poisson summation then gives
E ( L ) > − 17 5 − π 5 90 V h 3 + π 11 467775 V h 9 − R ( V , h , a 0 ) . (40) \mathcal{E}(L)>-\frac{17}{5}-\frac{\pi^5}{90Vh^3}+\frac{\pi^{11}}{467775Vh^9}-R(V,h,a_0). \tag*{(40)} E ( L ) > − 5 17 − 90 V h 3 π 5 + 467775 V h 9 π 11 − R ( V , h , a 0 ) . (40)
The attractive nonzero-mode coefficient is 2 π 3 2\pi^3 2 π 3 . Use
( H , a 0 ) = ( 27 20 , 7 10 ) , ( H , a 0 ) = ( 13 10 , 4 5 ) . (H,a_0)=\left(\frac{27}{20},\frac{7}{10}\right),\qquad(H,a_0)=\left(\frac{13}{10},\frac{4}{5}\right). ( H , a 0 ) = ( 20 27 , 10 7 ) , ( H , a 0 ) = ( 10 13 , 5 4 ) .
Since H 6 > 3 ζ ( 10 ) / ( 5 ζ ( 4 ) ) H^6>3\zeta(10)/(5\zeta(4)) H 6 > 3 ζ ( 10 ) / ( 5 ζ ( 4 )) , the zero-mode bracket is bounded below at ( 21 / 20 , H ) (21/20,H) ( 21/20 , H ) for h ≥ H h\ge H h ≥ H . The summands of R R R are
V j − ν − 1 / 2 h − 2 j e − c h 2 / V , c > 0 , 0 ≤ j , ν ≤ 5 , V^{j-\nu-1/2}h^{-2j}e^{-ch^2/V},\qquad c>0,\quad0\le j,\nu\le5, V j − ν − 1/2 h − 2 j e − c h 2 / V , c > 0 , 0 ≤ j , ν ≤ 5 ,
with
∂ V log ( ⋅ ) = c h 2 / V + j − ν − 1 / 2 V > 0 , c h 2 / V > 11 / 2 , \partial_V\log(\cdot)=\frac{ch^2/V+j-\nu-1/2}{V}>0,\qquad ch^2/V>11/2, ∂ V log ( ⋅ ) = V c h 2 / V + j − ν − 1/2 > 0 , c h 2 / V > 11/2 ,
∂ h log ( ⋅ ) = − 2 ( j + c h 2 / V ) h < 0. \partial_h\log(\cdot)=-\frac{2(j+ch^2/V)}{h}<0. ∂ h log ( ⋅ ) = − h 2 ( j + c h 2 / V ) < 0.
Here c h 2 / V ≥ 2 π a 0 H 2 / ( 11 / 10 ) ch^2/V\ge2\pi a_0H^2/(11/10) c h 2 / V ≥ 2 π a 0 H 2 / ( 11/10 ) . Hence R ( V , h , a 0 ) ≤ R ( 11 / 10 , H , a 0 ) R(V,h,a_0)\le R(11/10,H,a_0) R ( V , h , a 0 ) ≤ R ( 11/10 , H , a 0 ) . The endpoint estimates are
− π 5 90 ( 21 / 20 ) H 3 + π 11 467775 ( 21 / 20 ) H 9 ≥ { − 319 / 250 , ( H , a 0 ) = ( 27 / 20 , 7 / 10 ) , − 709 / 500 , ( H , a 0 ) = ( 13 / 10 , 4 / 5 ) , -\frac{\pi^{5}}{90(21/20)H^{3}}+\frac{\pi^{11}}{467775(21/20)H^{9}}\ge
\begin{cases}
-319/250, & (H,a_{0})=(27/20,7/10),\\
-709/500, & (H,a_{0})=(13/10,4/5),
\end{cases} − 90 ( 21/20 ) H 3 π 5 + 467775 ( 21/20 ) H 9 π 11 ≥ { − 319/250 , − 709/500 , ( H , a 0 ) = ( 27/20 , 7/10 ) , ( H , a 0 ) = ( 13/10 , 4/5 ) ,
R ( 11 / 10 , H , a 0 ) < { 16 / 125 , ( H , a 0 ) = ( 27 / 20 , 7 / 10 ) , 107 / 1000 , ( H , a 0 ) = ( 13 / 10 , 4 / 5 ) . R(11/10,H,a_{0})<
\begin{cases}
16/125, & (H,a_{0})=(27/20,7/10),\\
107/1000, & (H,a_{0})=(13/10,4/5).
\end{cases} R ( 11/10 , H , a 0 ) < { 16/125 , 107/1000 , ( H , a 0 ) = ( 27/20 , 7/10 ) , ( H , a 0 ) = ( 13/10 , 4/5 ) .
Retain m ∈ { 1 , … , 8 } m\in\{1,\ldots,8\} m ∈ { 1 , … , 8 } and ( i , j ) ∈ [ − 10 , 10 ] 2 ∩ Z 2 ∖ { 0 } (i,j)\in[-10,10]^{2}\cap\mathbb{Z}^{2}\setminus\{0\} ( i , j ) ∈ [ − 10 , 10 ] 2 ∩ Z 2 ∖ { 0 } in R R R . For k = max { ∣ i ∣ , ∣ j ∣ } k=\max\{|i|,|j|\} k = max { ∣ i ∣ , ∣ j ∣ } , each omitted block is bounded by
2000 ( k 6 + k 3 ) e − 5 m k . 2000(k^{6}+k^{3})e^{-5mk}. 2000 ( k 6 + k 3 ) e − 5 mk .
Indeed 3 k 2 / 4 ≤ i 2 + j 2 − ∣ i j ∣ ≤ 2 k 2 3k^{2}/4\le i^{2}+j^{2}-|ij|\le2k^{2} 3 k 2 /4 ≤ i 2 + j 2 − ∣ ij ∣ ≤ 2 k 2 , the Bessel argument exceeds 5 m k 5mk 5 mk , and the two Bessel polynomials are at most 8 8 8 and 3 / 2 3/2 3/2 . Their square-root prefactors are less than one; the coefficients before ( i 2 + j 2 − ∣ i j ∣ ) 5 / 2 (i^{2}+j^{2}-|ij|)^{5/2} ( i 2 + j 2 − ∣ ij ∣ ) 5/2 and i 2 + j 2 − ∣ i j ∣ i^{2}+j^{2}-|ij| i 2 + j 2 − ∣ ij ∣ are less than 40 40 40 and 100 100 100 , respectively. Each block has 8 k 8k 8 k pairs; e − 5 < 1 / 100 e^{-5}<1/100 e − 5 < 1/100 bounds the omitted part by
100 99 50 49 2000 ( 11 6 + 11 3 ) 100 − 11 + 200 99 4000 100 − 9 < 10 − 12 . \frac{100}{99}\frac{50}{49}2000(11^{6}+11^{3})100^{-11}+\frac{200}{99}4000\,100^{-9}<10^{-12}. 99 100 49 50 2000 ( 1 1 6 + 1 1 3 ) 10 0 − 11 + 99 200 4000 10 0 − 9 < 1 0 − 12 .
Both endpoint tests use the rational Machin, square-root and exponential enclosures of Lemma 2.8.
At the first endpoint, (40) gives, if h ≥ 27 / 20 h\ge27/20 h ≥ 27/20 ,
E ( L ) > − 124 25 . \mathcal{E}(L)>-\frac{124}{25}. E ( L ) > − 25 124 .
Hence h < 27 / 20 h<27/20 h < 27/20 for every primitive layer. Let a = λ 1 ( L ) a=\lambda_{1}(L) a = λ 1 ( L ) and let δ \delta δ be the first projected Gauss-reduced length. The lifted pair gives a primitive plane with A = a δ A=a\delta A = a δ , h = V / A h=V/A h = V / A , and
h 2 ≥ 3 4 δ 2 , A 2 ≤ 2 a V 3 . h^{2}\ge\frac{3}{4}\delta^{2},\qquad A^{2}\le\frac{2aV}{\sqrt{3}}. h 2 ≥ 4 3 δ 2 , A 2 ≤ 3 2 aV .
For a ≤ 4 / 5 a\le4/5 a ≤ 4/5 , the layer bound and 3 > 173 / 100 \sqrt{3}>173/100 3 > 173/100 give
7 10 < a ≤ 4 5 , 7 9 < A < 101 100 . \frac{7}{10}<a\le\frac{4}{5},\qquad\frac{7}{9}<A<\frac{101}{100}. 10 7 < a ≤ 5 4 , 9 7 < A < 100 101 .
The nonnegative Fourier transform of ( x 2 + z 2 ) − 3 (x^{2}+z^{2})^{-3} ( x 2 + z 2 ) − 3 makes its periodization maximal at zero shift. For even nonnegative decreasing g g g , ∑ k ∈ Z g ( a k ) ≤ g ( 0 ) + a − 1 ∫ R g \sum_{k\in\mathbb{Z}}g(ak)\le g(0)+a^{-1}\int_{\mathbb{R}}g ∑ k ∈ Z g ( ak ) ≤ g ( 0 ) + a − 1 ∫ R g ; apply this along ( a , 0 ) , ( s , A / a ) (a,0),(s,A/a) ( a , 0 ) , ( s , A / a ) to obtain
E ( Γ ) ≥ ζ ( 12 ) a − 12 − 2 ζ ( 6 ) a − 6 − 2 ζ ( 6 ) ( a / A ) 6 − 3 π ζ ( 5 ) 4 a 4 A 5 , \mathcal{E}(\Gamma)\ge\zeta(12)a^{-12}-2\zeta(6)a^{-6}-2\zeta(6)(a/A)^{6}-\frac{3\pi\zeta(5)}{4}\frac{a^{4}}{A^{5}}, E ( Γ ) ≥ ζ ( 12 ) a − 12 − 2 ζ ( 6 ) a − 6 − 2 ζ ( 6 ) ( a / A ) 6 − 4 3 π ζ ( 5 ) A 5 a 4 ,
E ( L ) − E ( Γ ) ≥ − 2 ζ ( 6 ) h − 6 − 3 π ζ ( 5 ) 4 ( 1 / a + a / A ) h − 5 − π ζ ( 4 ) A h − 4 . \mathcal{E}(L)-\mathcal{E}(\Gamma)\ge-2\zeta(6)h^{-6}-\frac{3\pi\zeta(5)}{4}(1/a+a/A)h^{-5}-\frac{\pi\zeta(4)}{A}h^{-4}. E ( L ) − E ( Γ ) ≥ − 2 ζ ( 6 ) h − 6 − 4 3 π ζ ( 5 ) ( 1/ a + a / A ) h − 5 − A π ζ ( 4 ) h − 4 .
The integrals are 3 π / ( 8 z 5 ) 3\pi/(8z^{5}) 3 π / ( 8 z 5 ) and π / ( 2 z 4 ) \pi/(2z^{4}) π / ( 2 z 4 ) ; the omitted repulsive terms are nonnegative. With 2 ζ ( 6 ) < 21 / 10 2\zeta(6)<21/10 2 ζ ( 6 ) < 21/10 , 3 π ζ ( 5 ) / 4 < 99 / 40 3\pi\zeta(5)/4<99/40 3 π ζ ( 5 ) /4 < 99/40 , π ζ ( 4 ) < 341 / 100 \pi\zeta(4)<341/100 π ζ ( 4 ) < 341/100 and ζ ( 12 ) > 1 \zeta(12)>1 ζ ( 12 ) > 1 and V ≥ 21 / 20 V\ge21/20 V ≥ 21/20 , we obtain E ( L ) > P ( a , A ) \mathcal{E}(L)>P(a,A) E ( L ) > P ( a , A ) , where
P ( a , A ) = a − 12 − 21 10 a − 6 − 21 a 6 10 A 6 − 99 a 4 40 A 5 − 21 A 6 10 ( 21 / 20 ) 6 − 99 A 5 40 a ( 21 / 20 ) 5 − 99 a A 4 40 ( 21 / 20 ) 5 − 341 A 3 100 ( 21 / 20 ) 4 . \begin{aligned}
P(a,A)={}&a^{-12}-\frac{21}{10}a^{-6}-\frac{21a^{6}}{10A^{6}}-\frac{99a^{4}}{40A^{5}}-\frac{21A^{6}}{10(21/20)^{6}}\\
&-\frac{99A^{5}}{40a(21/20)^{5}}-\frac{99aA^{4}}{40(21/20)^{5}}-\frac{341A^{3}}{100(21/20)^{4}}.
\end{aligned} P ( a , A ) = a − 12 − 10 21 a − 6 − 10 A 6 21 a 6 − 40 A 5 99 a 4 − 10 ( 21/20 ) 6 21 A 6 − 40 a ( 21/20 ) 5 99 A 5 − 40 ( 21/20 ) 5 99 a A 4 − 100 ( 21/20 ) 4 341 A 3 .
On this rectangle,
∂ a P ≤ − 12 ( 5 / 4 ) 13 + 63 5 ( 5 / 4 ) 7 + 99 40 ( 101 / 100 ) 5 ( 7 / 10 ) 2 ( 21 / 20 ) 5 < − 150. \partial_{a}P\le-12(5/4)^{13}+\frac{63}{5}(5/4)^{7}+\frac{99}{40}\frac{(101/100)^{5}}{(7/10)^{2}(21/20)^{5}}<-150. ∂ a P ≤ − 12 ( 5/4 ) 13 + 5 63 ( 5/4 ) 7 + 40 99 ( 7/10 ) 2 ( 21/20 ) 5 ( 101/100 ) 5 < − 150.
For fixed a a a , P P P is strictly concave in A A A . At the endpoints,
P ( 4 / 5 , 7 / 9 ) > − 5 2 , P ( 4 / 5 , 101 / 100 ) > − 37 10 . P(4/5,7/9)>-\frac{5}{2},\qquad P(4/5,101/100)>-\frac{37}{10}. P ( 4/5 , 7/9 ) > − 2 5 , P ( 4/5 , 101/100 ) > − 10 37 .
Thus E ( L ) > − 37 / 10 \mathcal{E}(L)>-37/10 E ( L ) > − 37/10 , contradicting the hypothesis; hence a > 4 / 5 a>4/5 a > 4/5 . For h ≥ 13 / 10 h\ge13/10 h ≥ 13/10 , the second endpoint gives
E ( L ) > − 124 25 . \mathcal{E}(L)>-\frac{124}{25}. E ( L ) > − 25 124 .
Therefore h < 13 / 10 h<13/10 h < 13/10 for every primitive layer, equivalently λ 1 ( L ∗ ) > 10 / 13 \lambda_1(L^*)>10/13 λ 1 ( L ∗ ) > 10/13 . This completes the proof of Lemma 4.4 . □
Proposition 4.5 (A uniform Korkine–Zolotarev domain). Under the hypotheses of Lemma 4.4 , the lattice has a Korkine–Zolotarev-reduced basis of the form
B = ( a a x a y 0 δ δ z 0 0 h ) , a δ h = V , 0 ≤ x , z ≤ 1 2 , − 1 2 ≤ y ≤ 1 2 , B=\begin{pmatrix}a&ax&ay\\0&\delta&\delta z\\0&0&h\end{pmatrix},\qquad a\delta h=V,\quad0\le x,z\le\frac{1}{2},\quad-\frac{1}{2}\le y\le\frac{1}{2}, B = a 0 0 a x δ 0 a y δ z h , a δ h = V , 0 ≤ x , z ≤ 2 1 , − 2 1 ≤ y ≤ 2 1 , with
4 5 < a < 29 25 , 69 100 < δ < 127 100 , 22 25 < h < 13 10 . (41) \frac{4}{5}<a<\frac{29}{25},\qquad\frac{69}{100}<\delta<\frac{127}{100},\qquad\frac{22}{25}<h<\frac{13}{10}. \tag*{(41)} 5 4 < a < 25 29 , 100 69 < δ < 100 127 , 25 22 < h < 10 13 . (41) Moreover 21 / 26 < a δ < 5 / 4 21/26<a\delta<5/4 21/26 < a δ < 5/4 . For the unit-covolume shape Λ = V − 1 / 3 L \Lambda=V^{-1/3}L Λ = V − 1/3 L , write ( a 0 , δ 0 , h 0 ) = V − 1 / 3 ( a , δ , h ) (a_0,\delta_0,h_0)=V^{-1/3}(a,\delta,h) ( a 0 , δ 0 , h 0 ) = V − 1/3 ( a , δ , h ) . Then
77 100 < a 0 ≤ 2 1 / 6 , 69 100 < δ 0 < 123 100 , 7 8 < h 0 < 32 25 . (42) \frac{77}{100}<a_0\le2^{1/6},\qquad\frac{69}{100}<\delta_0<\frac{123}{100},\qquad\frac{7}{8}<h_0<\frac{32}{25}. \tag*{(42)} 100 77 < a 0 ≤ 2 1/6 , 100 69 < δ 0 < 100 123 , 8 7 < h 0 < 25 32 . (42) and
25 32 < a 0 δ 0 < 8 7 , λ 1 ( Λ ∗ ) > 39 50 . \frac{25}{32}<a_0\delta_0<\frac{8}{7},\qquad\lambda_1(\Lambda^*)>\frac{39}{50}. 32 25 < a 0 δ 0 < 7 8 , λ 1 ( Λ ∗ ) > 50 39 . Proof. Choose a reduced basis B B B with
a = min n ∈ Z 3 ∖ { 0 } ∣ B n ∣ , δ 2 = min ( j , k ) ∈ Z 2 ∖ { 0 } { δ 2 ( j + z k ) 2 + h 2 k 2 } . a=\min_{n\in\mathbb{Z}^3\setminus\{0\}}|Bn|,\qquad\delta^2=\min_{(j,k)\in\mathbb{Z}^2\setminus\{0\}}\{\delta^2(j+zk)^2+h^2k^2\}. a = n ∈ Z 3 ∖ { 0 } min ∣ B n ∣ , δ 2 = ( j , k ) ∈ Z 2 ∖ { 0 } min { δ 2 ( j + z k ) 2 + h 2 k 2 } . Size-reduced lifts give the stated shear ranges and
δ 2 ≥ a 2 ( 1 − x 2 ) , h 2 ≥ δ 2 ( 1 − z 2 ) ≥ 3 4 δ 2 . \delta^2\ge a^2(1-x^2),\qquad h^2\ge\delta^2(1-z^2)\ge\frac{3}{4}\delta^2. δ 2 ≥ a 2 ( 1 − x 2 ) , h 2 ≥ δ 2 ( 1 − z 2 ) ≥ 4 3 δ 2 . The Hermite bound a 6 ≤ 2 V 2 a^6\le2V^2 a 6 ≤ 2 V 2 and the preceding lemma give
a < 29 25 , δ = V a h > 69 100 , δ 2 ≤ 2 V 3 a < ( 127 100 ) 2 . a<\frac{29}{25},\qquad\delta=\frac{V}{ah}>\frac{69}{100},\qquad\delta^2\le\frac{2V}{\sqrt{3}a}<\left(\frac{127}{100}\right)^2. a < 25 29 , δ = ah V > 100 69 , δ 2 ≤ 3 a 2 V < ( 100 127 ) 2 . Also h ≥ 3 δ / 2 h\ge\sqrt{3}\delta/2 h ≥ 3 δ /2 and a δ h = V a\delta h=V a δ h = V imply
h 2 ≥ 3 V 2 a > ( 22 25 ) 2 . h^2\ge\frac{\sqrt{3}V}{2a}>\left(\frac{22}{25}\right)^2. h 2 ≥ 2 a 3 V > ( 25 22 ) 2 . Now a δ = V / h a\delta=V/h a δ = V / h gives the product bounds. Normalization yields a 0 > 77 / 100 a_0>77/100 a 0 > 77/100 , h 0 < 32 / 25 h_0<32/25 h 0 < 32/25 , a 0 ≤ 2 1 / 6 < 1123 / 1000 a_0\le2^{1/6}<1123/1000 a 0 ≤ 2 1/6 < 1123/1000 ; a 0 δ 0 h 0 = 1 a_0\delta_0h_0=1 a 0 δ 0 h 0 = 1 then gives δ 0 > 69 / 100 \delta_0>69/100 δ 0 > 69/100 and
δ 0 2 ≤ 2 3 a 0 < ( 123 100 ) 2 , h 0 2 ≥ 3 2 a 0 > ( 7 8 ) 2 . \delta_0^2\le\frac{2}{\sqrt{3}a_0}<\left(\frac{123}{100}\right)^2,\qquad h_0^2\ge\frac{\sqrt{3}}{2a_0}>\left(\frac{7}{8}\right)^2. δ 0 2 ≤ 3 a 0 2 < ( 100 123 ) 2 , h 0 2 ≥ 2 a 0 3 > ( 8 7 ) 2 . Inverting h 0 h_0 h 0 gives the product bounds. Since Λ ∗ = V 1 / 3 L ∗ \Lambda^*=V^{1/3}L^* Λ ∗ = V 1/3 L ∗ ,
λ 1 ( Λ ∗ ) > ( 21 / 20 ) 1 / 3 ( 10 / 13 ) > 39 / 50. \lambda_1(\Lambda^*)>(21/20)^{1/3}(10/13)>39/50. λ 1 ( Λ ∗ ) > ( 21/20 ) 1/3 ( 10/13 ) > 39/50. This completes the proof of Proposition 4.5 . □
Proposition 4.6 (A common endpoint comparison domain). Suppose 21 / 20 ≤ V = ∣ L ∣ ≤ 11 / 10 21/20\le V=|L|\le11/10 21/20 ≤ V = ∣ L ∣ ≤ 11/10 and E ( L ) ≤ − 124 / 25 \mathcal{E}(L)\le-124/25 E ( L ) ≤ − 124/25 . Then L ≅ U ( p ) Z 3 L\cong U(p)\mathbb{Z}^3 L ≅ U ( p ) Z 3 , where
U ( p ) = ( a a x a y 0 δ δ z 0 0 h ) , h = V a δ , p = ( a , δ , x , y , z ) , U(p)=\begin{pmatrix}a&ax&ay\\0&\delta&\delta z\\0&0&h\end{pmatrix},\qquad h=\frac{V}{a\delta},\qquad p=(a,\delta,x,y,z), U ( p ) = a 0 0 a x δ 0 a y δ z h , h = a δ V , p = ( a , δ , x , y , z ) ,
and p p p belongs to the compact domain D ( V ) D(V) D ( V ) defined by
4 5 ≤ a ≤ 29 25 , 69 100 ≤ δ ≤ 127 100 , 0 ≤ x , z ≤ 1 2 , − 1 2 ≤ y ≤ 1 2 , (43) \frac{4}{5}\leq a\leq\frac{29}{25},\quad\frac{69}{100}\leq\delta\leq\frac{127}{100},\quad0\leq x,z\leq\frac{1}{2},\quad-\frac{1}{2}\leq y\leq\frac{1}{2}, \tag*{(43)} 5 4 ≤ a ≤ 25 29 , 100 69 ≤ δ ≤ 100 127 , 0 ≤ x , z ≤ 2 1 , − 2 1 ≤ y ≤ 2 1 , (43)
22 25 ≤ h ≤ 13 10 , 10 V 13 ≤ a δ ≤ 25 V 22 , \frac{22}{25}\leq h\leq\frac{13}{10},\quad\frac{10V}{13}\leq a\delta\leq\frac{25V}{22}, 25 22 ≤ h ≤ 10 13 , 13 10 V ≤ a δ ≤ 22 25 V ,
δ 2 + a 2 x 2 ≥ a 2 , h 2 + δ 2 z 2 ≥ δ 2 , \delta^{2}+a^{2}x^{2}\geq a^{2},\qquad h^{2}+\delta^{2}z^{2}\geq\delta^{2}, δ 2 + a 2 x 2 ≥ a 2 , h 2 + δ 2 z 2 ≥ δ 2 ,
a 2 ( i + x j + y k ) 2 + δ 2 ( j + z k ) 2 + h 2 k 2 ≥ a 2 , a^{2}(i+xj+yk)^{2}+\delta^{2}(j+zk)^{2}+h^{2}k^{2}\geq a^{2}, a 2 ( i + x j + y k ) 2 + δ 2 ( j + z k ) 2 + h 2 k 2 ≥ a 2 ,
a − 2 i 2 + δ − 2 ( j − x i ) 2 + h − 2 { k − z j + ( x z − y ) i } 2 ≥ 100 169 , a^{-2}i^{2}+\delta^{-2}(j-xi)^{2}+h^{-2}\{k-zj+(xz-y)i\}^{2}\geq\frac{100}{169}, a − 2 i 2 + δ − 2 ( j − x i ) 2 + h − 2 { k − z j + ( x z − y ) i } 2 ≥ 169 100 ,
( i , j , k ) ∈ { − 1 , 0 , 1 } 3 ∖ { 0 } . (44) (i,j,k)\in\{-1,0,1\}^{3}\setminus\{0\}. \tag*{(44)} ( i , j , k ) ∈ { − 1 , 0 , 1 } 3 ∖ { 0 } . (44)
In particular, put T F = − 26537251 / 5000000 T_F=-26537251/5000000 T F = − 26537251/5000000 and D F = D ( 21 / 20 ) D_F=D(21/20) D F = D ( 21/20 ) .
Proof. Lemma 4.4 and Proposition 4.5 give the box, the Gauss inequalities and
λ 1 ( L ) > 4 / 5 , λ 1 ( L ∗ ) > 10 / 13 , 10 V 13 ≤ a δ = V / h ≤ 25 V 22 . \lambda_1(L)>4/5,\qquad\lambda_1(L^*)>10/13,\qquad\frac{10V}{13}\leq a\delta=V/h\leq\frac{25V}{22}. λ 1 ( L ) > 4/5 , λ 1 ( L ∗ ) > 10/13 , 13 10 V ≤ a δ = V / h ≤ 22 25 V .
For the remaining primal indices,
∣ k ∣ ≥ 2 ⟹ h 2 k 2 ≥ 4 ( 22 / 25 ) 2 > ( 29 / 25 ) 2 , |k|\geq2\Longrightarrow h^{2}k^{2}\geq4(22/25)^{2}>(29/25)^{2}, ∣ k ∣ ≥ 2 ⟹ h 2 k 2 ≥ 4 ( 22/25 ) 2 > ( 29/25 ) 2 ,
∣ k ∣ ≤ 1 , ∣ j ∣ ≥ 2 ⟹ δ 2 ( j + z k ) 2 ≥ 9 δ 2 / 4 > a 2 , |k|\leq1,\ |j|\geq2\Longrightarrow\delta^{2}(j+zk)^{2}\geq9\delta^{2}/4>a^{2}, ∣ k ∣ ≤ 1 , ∣ j ∣ ≥ 2 ⟹ δ 2 ( j + z k ) 2 ≥ 9 δ 2 /4 > a 2 ,
∣ j ∣ , ∣ k ∣ ≤ 1 , ∣ i ∣ ≥ 2 ⟹ a 2 ( i + x j + y k ) 2 ≥ a 2 . |j|,|k|\leq1,\ |i|\geq2\Longrightarrow a^{2}(i+xj+yk)^{2}\geq a^{2}. ∣ j ∣ , ∣ k ∣ ≤ 1 , ∣ i ∣ ≥ 2 ⟹ a 2 ( i + x j + y k ) 2 ≥ a 2 .
For the dual indices outside this cube,
∣ U − T n ∣ 2 ≥ 4 ∥ U ∥ 1 ∥ U ∥ ∞ ≥ 4 ( 503 / 200 ) ( 58 / 25 ) > 100 169 . |U^{-T}n|^{2}\geq\frac{4}{\lVert U\rVert_{1}\lVert U\rVert_{\infty}}\geq\frac{4}{(503/200)(58/25)}>\frac{100}{169}. ∣ U − T n ∣ 2 ≥ ∥ U ∥ 1 ∥ U ∥ ∞ 4 ≥ ( 503/200 ) ( 58/25 ) 4 > 169 100 .
The positive diagonal floors give compactness; a δ ∈ [ 21 / 26 , 105 / 88 ] a\delta\in[21/26,105/88] a δ ∈ [ 21/26 , 105/88 ] at V = 21 / 20 V=21/20 V = 21/20 and a δ ∈ [ 11 / 13 , 5 / 4 ] a\delta\in[11/13,5/4] a δ ∈ [ 11/13 , 5/4 ] at V = 11 / 10 V=11/10 V = 11/10 . This completes the proof of Proposition 4.6 . □
Lemma 4.7 (A uniform FCC stationary basin). Let
t = ( 441 1600 ) 1 / 6 , L F = t D 3 , D 3 = { ( i , j , k ) ∈ Z 3 : i + j + k ∈ 2 Z } , t=\left(\frac{441}{1600}\right)^{1/6},\qquad L_F=tD_3,\qquad D_3=\{(i,j,k)\in\mathbb{Z}^{3}:i+j+k\in2\mathbb{Z}\}, t = ( 1600 441 ) 1/6 , L F = t D 3 , D 3 = {( i , j , k ) ∈ Z 3 : i + j + k ∈ 2 Z } , so that ∣ L F ∣ = 21 / 20 |L_F|=21/20 ∣ L F ∣ = 21/20 . In the chart (2.5), put
L ( S ) = det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 L F , ∣ L ( S ) ∣ = 21 / 20 , L(S)=\det(I+S)^{-1/6}(I+S)^{1/2}L_F,\qquad|L(S)|=21/20, L ( S ) = det ( I + S ) − 1/6 ( I + S ) 1/2 L F , ∣ L ( S ) ∣ = 21/20 , Ψ ( S ) = 1 2 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) , ∇ s Ψ = ( ∂ u Ψ , ∂ v Ψ , ∂ x Ψ , ∂ y Ψ , ∂ z Ψ ) T . \Psi(S)=\frac{1}{2}E(L(S),6)-E(L(S),3),\qquad\nabla_s\Psi=(\partial_u\Psi,\partial_v\Psi,\partial_x\Psi,\partial_y\Psi,\partial_z\Psi)^{T}. Ψ ( S ) = 2 1 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) , ∇ s Ψ = ( ∂ u Ψ , ∂ v Ψ , ∂ x Ψ , ∂ y Ψ , ∂ z Ψ ) T . Then
∂ ϵ 2 Ψ ( S + ϵ T ) ∣ ϵ = 0 > 1 1000 ∥ T ∥ F 2 ( ∥ S ∥ F ≤ 1 50 , T ≠ 0 , T = T T , tr T = 0 ) , \left.\partial_{\epsilon}^{2}\Psi(S+\epsilon T)\right|_{\epsilon=0}>\frac{1}{1000}\lVert T\rVert_{F}^{2}\qquad\left(\lVert S\rVert_{F}\leq\frac{1}{50},\ T\ne0,\ T=T^{T},\ \operatorname{tr}T=0\right), ∂ ϵ 2 Ψ ( S + ϵ T ) ϵ = 0 > 1000 1 ∥ T ∥ F 2 ( ∥ S ∥ F ≤ 50 1 , T = 0 , T = T T , tr T = 0 ) , and
S T ∇ s Ψ ( S ) = ∂ ϵ Ψ ( ( 1 + ϵ ) S ) ∣ ϵ = 0 > 1 1000 ∥ S ∥ F 2 ( 0 < ∥ S ∥ F ≤ 43 1000 ) . S^{T}\nabla_s\Psi(S)=\left.\partial_{\epsilon}\Psi((1+\epsilon)S)\right|_{\epsilon=0}>\frac{1}{1000}\lVert S\rVert_{F}^{2}\qquad\left(0<\lVert S\rVert_{F}\leq\frac{43}{1000}\right). S T ∇ s Ψ ( S ) = ∂ ϵ Ψ (( 1 + ϵ ) S ) ∣ ϵ = 0 > 1000 1 ∥ S ∥ F 2 ( 0 < ∥ S ∥ F ≤ 1000 43 ) . Consequently
∇ s Ψ ( S ) = 0 ⟺ S = 0 ( ∥ S ∥ F ≤ 43 / 1000 ) , \nabla_s\Psi(S)=0\Longleftrightarrow S=0\qquad(\lVert S\rVert_{F}\leq43/1000), ∇ s Ψ ( S ) = 0 ⟺ S = 0 (∥ S ∥ F ≤ 43/1000 ) , and
Ψ ( S ) ≥ Ψ ( 0 ) + 1 2000 ∥ S ∥ F 2 ( ∥ S ∥ F ≤ 43 1000 ) . \Psi(S)\geq\Psi(0)+\frac{1}{2000}\lVert S\rVert_{F}^{2}\qquad\left(\lVert S\rVert_{F}\leq\frac{43}{1000}\right). Ψ ( S ) ≥ Ψ ( 0 ) + 2000 1 ∥ S ∥ F 2 ( ∥ S ∥ F ≤ 1000 43 ) . Convexity is asserted only on ∥ S ∥ F ≤ 1 / 50 \lVert S\rVert_{F}\leq1/50 ∥ S ∥ F ≤ 1/50 .
Proof. Cubic symmetry gives ∇ s Ψ ( 0 ) = 0 \nabla_s\Psi(0)=0 ∇ s Ψ ( 0 ) = 0 . Use the chart and polynomial construction of Lemma 2.9 , with its D , p 3 , p 6 D,p_3,p_6 D , p 3 , p 6 . For w = t ( i , j , k ) w=t(i,j,k) w = t ( i , j , k ) , write
q w = t 2 ( i 2 + j 2 + k 2 ) , ℓ w ( S ) = u ( 2 i 2 − j 2 − k 2 ) + v ( j 2 − k 2 ) + 2 x i j + 2 y i k + 2 z j k i 2 + j 2 + k 2 . q_w=t^2(i^2+j^2+k^2),\qquad\ell_w(S)=\frac{u(2i^2-j^2-k^2)+v(j^2-k^2)+2xij+2yik+2zjk}{i^2+j^2+k^2}. q w = t 2 ( i 2 + j 2 + k 2 ) , ℓ w ( S ) = i 2 + j 2 + k 2 u ( 2 i 2 − j 2 − k 2 ) + v ( j 2 − k 2 ) + 2 x ij + 2 y ik + 2 z j k .
Sum, with both signs, over the exact vector set
F = { t ( i , j , k ) : ( i , j , k ) ∈ [ − 40 , 40 ] 3 ∩ Z 3 , i + j + k ∈ 2 Z , 0 < i 2 + j 2 + k 2 ≤ 1600 } , \mathcal{F}=\left\{t(i,j,k):(i,j,k)\in[-40,40]^3\cap\mathbb{Z}^3,\quad i+j+k\in2\mathbb{Z},\quad0<i^2+j^2+k^2\le1600\right\}, F = { t ( i , j , k ) : ( i , j , k ) ∈ [ − 40 , 40 ] 3 ∩ Z 3 , i + j + k ∈ 2 Z , 0 < i 2 + j 2 + k 2 ≤ 1600 } ,
and form P P P by Lemma 2.9 . Since t − 6 = 1600 / 441 t^{-6}=1600/441 t − 6 = 1600/441 , its coefficients are rational, deg P ≤ 12 \deg P\le12 deg P ≤ 12 , and cubic symmetry gives ∇ s P ( 0 ) = 0 \nabla_sP(0)=0 ∇ s P ( 0 ) = 0 .
Since
λ 1 ( L F ) > 1 , min w ∈ L F ∖ ( F ∪ { 0 } ) ∣ w ∣ > 32 , \lambda_1(L_F)>1,\qquad\min_{w\in L_F\setminus(F\cup\{0\})}|w|>32, λ 1 ( L F ) > 1 , w ∈ L F ∖ ( F ∪ { 0 }) min ∣ w ∣ > 32 ,
(18) gives
∑ w ∈ L F ∖ ( F ∪ { 0 } ) q w − s ≤ T s ( 1 , 32 ) , s = 3 , 6. \sum_{w\in L_F\setminus(F\cup\{0\})}q_w^{-s}\le T_s(1,32),\qquad s=3,6. w ∈ L F ∖ ( F ∪ { 0 }) ∑ q w − s ≤ T s ( 1 , 32 ) , s = 3 , 6.
The finite rational sums give
∑ w ∈ F q w − 3 + T 3 ( 1 , 32 ) < 164 25 , ∑ w ∈ F q w − 6 + T 6 ( 1 , 32 ) < 5 2 . \sum_{w\in\mathcal{F}}q_w^{-3}+T_3(1,32)<\frac{164}{25},\qquad\sum_{w\in\mathcal{F}}q_w^{-6}+T_6(1,32)<\frac{5}{2}. w ∈ F ∑ q w − 3 + T 3 ( 1 , 32 ) < 25 164 , w ∈ F ∑ q w − 6 + T 6 ( 1 , 32 ) < 2 5 .
With E + ( L F , 3 ) = 164 / 25 E^+(L_F,3)=164/25 E + ( L F , 3 ) = 164/25 , E + ( L F , 6 ) = 5 / 2 E^+(L_F,6)=5/2 E + ( L F , 6 ) = 5/2 and tails T s ( 1 , 32 ) T_s(1,32) T s ( 1 , 32 ) , (20) gives
ε ( 1 / 50 ) < 3 / 1000 , ε ( 43 / 1000 ) < 551 / 100000. \varepsilon(1/50)<3/1000,\qquad\varepsilon(43/1000)<551/100000. ε ( 1/50 ) < 3/1000 , ε ( 43/1000 ) < 551/100000.
In coordinates ( u , v , x , y , z ) (u,v,x,y,z) ( u , v , x , y , z ) ,
∂ ϵ 2 P ( S + ϵ T ) ∣ ϵ = 0 ≥ ∣ η ∣ T C H ∣ η ∣ ( ∥ S ∥ F ≤ 1 / 50 ) , \left.\partial_\epsilon^2P(S+\epsilon T)\right|_{\epsilon=0}\ge|\eta|^{\mathsf T}C_H|\eta|\qquad(\|S\|_F\le1/50), ∂ ϵ 2 P ( S + ϵ T ) ϵ = 0 ≥ ∣ η ∣ T C H ∣ η ∣ ( ∥ S ∥ F ≤ 1/50 ) ,
where η = ( T 11 / 2 , ( T 22 − T 33 ) / 2 , T 12 , T 13 , T 23 ) T \eta=(T_{11}/2,(T_{22}-T_{33})/2,T_{12},T_{13},T_{23})^{\mathsf T} η = ( T 11 /2 , ( T 22 − T 33 ) /2 , T 12 , T 13 , T 23 ) T and
C H = 1 10 6 ( 4078438 − 123613 − 296771 − 296771 − 585394 − 123613 1356119 − 298719 − 298719 − 3375 − 296771 − 298719 455411 − 55420 − 55466 − 296771 − 298719 − 55420 455411 − 55466 − 585394 − 3375 − 55466 − 55466 578379 ) . C_H=\frac{1}{10^6}
\begin{pmatrix}
4078438 & -123613 & -296771 & -296771 & -585394\\
-123613 & 1356119 & -298719 & -298719 & -3375\\
-296771 & -298719 & 455411 & -55420 & -55466\\
-296771 & -298719 & -55420 & 455411 & -55466\\
-585394 & -3375 & -55466 & -55466 & 578379
\end{pmatrix}. C H = 1 0 6 1 4078438 − 123613 − 296771 − 296771 − 585394 − 123613 1356119 − 298719 − 298719 − 3375 − 296771 − 298719 455411 − 55420 − 55466 − 296771 − 298719 − 55420 455411 − 55466 − 585394 − 3375 − 55466 − 55466 578379 .
The L D L T LDL^{\mathsf T} L D L T pivots of C H − ( 4 / 1000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_H-(4/1000)\operatorname{diag}(6,2,2,2,2) C H − ( 4/1000 ) diag ( 6 , 2 , 2 , 2 , 2 ) exceed
4 , 1 , 3 10 , 1 4 , 1 3 . 4,\qquad1,\qquad\frac{3}{10},\qquad\frac{1}{4},\qquad\frac{1}{3}. 4 , 1 , 10 3 , 4 1 , 3 1 .
Together with the remainder, this gives the inner Hessian bound.
For the radial polynomial B B B of Lemma 2.9 , put s = ( u , v , x , y , z ) T \mathbf{s}=(u,v,x,y,z)^{\mathsf T} s = ( u , v , x , y , z ) T . Since ∇ s P ( 0 ) = 0 \nabla_sP(0)=0 ∇ s P ( 0 ) = 0 , Euler’s identity gives ∂ ϵ P ( ( 1 + ϵ ) S ) ∣ ϵ = 0 = s T B ( S ) s \left.\partial_\epsilon P((1+\epsilon)S)\right|_{\epsilon=0}=\mathbf{s}^{\mathsf T}B(S)\mathbf{s} ∂ ϵ P (( 1 + ϵ ) S ) ∣ ϵ = 0 = s T B ( S ) s . On ∥ S ∥ F ≤ 43 / 1000 \|S\|_F\le43/1000 ∥ S ∥ F ≤ 43/1000 ,
η T B ( S ) η ≥ ∣ η ∣ T C R ∣ η ∣ , \eta^{\mathsf T}B(S)\eta\ge|\eta|^{\mathsf T}C_R|\eta|, η T B ( S ) η ≥ ∣ η ∣ T C R ∣ η ∣ ,
where
C R = 1 10 6 ( 4057854 − 142444 − 331103 − 331103 − 648836 − 142444 1347120 − 333824 − 333824 − 5546 − 331103 − 333824 407641 − 62250 − 62365 − 331103 − 333824 − 62250 407641 − 62365 − 648836 − 5546 − 62365 − 62365 543475 ) . C_R=\frac{1}{10^6}
\begin{pmatrix}
4057854 & -142444 & -331103 & -331103 & -648836\\
-142444 & 1347120 & -333824 & -333824 & -5546\\
-331103 & -333824 & 407641 & -62250 & -62365\\
-331103 & -333824 & -62250 & 407641 & -62365\\
-648836 & -5546 & -62365 & -62365 & 543475
\end{pmatrix}. C R = 1 0 6 1 4057854 − 142444 − 331103 − 331103 − 648836 − 142444 1347120 − 333824 − 333824 − 5546 − 331103 − 333824 407641 − 62250 − 62365 − 331103 − 333824 − 62250 407641 − 62365 − 648836 − 5546 − 62365 − 62365 543475 .
These are the rational monomial bounds of Lemma 2.9 , with 77 raw positions and 17711 nonnegative triples after symmetry folding.
The matrix C R − ( 651 / 100000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_R-(651/100000)\operatorname{diag}(6,2,2,2,2) C R − ( 651/100000 ) diag ( 6 , 2 , 2 , 2 , 2 ) is positive definite. Its exact L D L T LDL^{T} L D L T pivots are respectively greater than
4 , 13 10 , 27 100 , 3 20 , 11 100 . 4,\qquad\frac{13}{10},\qquad\frac{27}{100},\qquad\frac{3}{20},\qquad\frac{11}{100}. 4 , 10 13 , 100 27 , 20 3 , 100 11 .
For 0 < ∥ S ∥ F ≤ 43 / 1000 0<\lVert S\rVert_F\le43/1000 0 < ∥ S ∥ F ≤ 43/1000 , ∇ s Ψ ( 0 ) = ∇ s P ( 0 ) = 0 \nabla_s\Psi(0)=\nabla_sP(0)=0 ∇ s Ψ ( 0 ) = ∇ s P ( 0 ) = 0 and the remainder give
∂ ϵ Ψ ( ( 1 + ϵ ) S ) ∣ ϵ = 0 ≥ ∣ s ∣ T C R ∣ s ∣ − ϵ ( 43 / 1000 ) ∥ S ∥ F 2 > 1 1000 ∥ S ∥ F 2 . \left.\partial_\epsilon\Psi((1+\epsilon)S)\right|_{\epsilon=0}\ge|s|^{T}C_R|s|-\epsilon(43/1000)\lVert S\rVert_F^2>\frac{1}{1000}\lVert S\rVert_F^2. ∂ ϵ Ψ (( 1 + ϵ ) S ) ∣ ϵ = 0 ≥ ∣ s ∣ T C R ∣ s ∣ − ϵ ( 43/1000 ) ∥ S ∥ F 2 > 1000 1 ∥ S ∥ F 2 .
Hence ∇ s Ψ ( S ) ≠ 0 \nabla_s\Psi(S)\ne0 ∇ s Ψ ( S ) = 0 for S ≠ 0 S\ne0 S = 0 ; integration along t S tS tS gives Ψ ( S ) − Ψ ( 0 ) ≥ ∥ S ∥ F 2 / 2000 \Psi(S)-\Psi(0)\ge\lVert S\rVert_F^2/2000 Ψ ( S ) − Ψ ( 0 ) ≥ ∥ S ∥ F 2 /2000 . This completes the proof of Lemma 4.7 . □
Lemma 4.8 (An explicit FCC reference chart). Let a F , δ F , h F > 0 a_F,\delta_F,h_F>0 a F , δ F , h F > 0 be given by
a F 6 = 441 200 , δ F 6 = 11907 12800 , h F 6 = 49 75 , a_F^6=\frac{441}{200},\qquad\delta_F^6=\frac{11907}{12800},\qquad h_F^6=\frac{49}{75}, a F 6 = 200 441 , δ F 6 = 12800 11907 , h F 6 = 75 49 , and define the two reference matrices
U F , σ = ( a F a F / 2 σ a F / 2 0 δ F δ F / 3 0 0 h F ) , σ ∈ { − 1 , 1 } . U_{F,\sigma}=
\begin{pmatrix}
a_F & a_F/2 & \sigma a_F/2\\
0 & \delta_F & \delta_F/3\\
0 & 0 & h_F
\end{pmatrix},\qquad\sigma\in\{-1,1\}. U F , σ = a F 0 0 a F /2 δ F 0 σ a F /2 δ F /3 h F , σ ∈ { − 1 , 1 } . Then U F , σ Z 3 ≅ ( 21 / 20 ) 1 / 3 F U_{F,\sigma}\mathbb{Z}^3\cong(21/20)^{1/3}F U F , σ Z 3 ≅ ( 21/20 ) 1/3 F and det U F , σ = 21 / 20 \det U_{F,\sigma}=21/20 det U F , σ = 21/20 . Exact rational enclosures are
1140866336564584 10 15 < a F < 1140866336564585 10 15 , \frac{1140866336564584}{10^{15}}<a_F<\frac{1140866336564585}{10^{15}}, 1 0 15 1140866336564584 < a F < 1 0 15 1140866336564585 , 988019229787417 10 15 < δ F < 988019229787418 10 15 , \frac{988019229787417}{10^{15}}<\delta_F<\frac{988019229787418}{10^{15}}, 1 0 15 988019229787417 < δ F < 1 0 15 988019229787418 , 931513463100523 10 15 < h F < 931513463100524 10 15 . \frac{931513463100523}{10^{15}}<h_F<\frac{931513463100524}{10^{15}}. 1 0 15 931513463100523 < h F < 1 0 15 931513463100524 . For a Korkine–Zolotarev matrix U ( p ) U(p) U ( p ) of determinant 21 / 20 21/20 21/20 , the relative matrix M σ = U ( p ) U F , σ − 1 M_\sigma=U(p)U_{F,\sigma}^{-1} M σ = U ( p ) U F , σ − 1 is upper triangular, with
( M σ ) 11 = a / a F , ( M σ ) 22 = δ / δ F , ( M σ ) 33 = h / h F , (M_\sigma)_{11}=a/a_F,\qquad(M_\sigma)_{22}=\delta/\delta_F,\qquad(M_\sigma)_{33}=h/h_F, ( M σ ) 11 = a / a F , ( M σ ) 22 = δ / δ F , ( M σ ) 33 = h / h F , ( M σ ) 12 = a ( x − 1 / 2 ) / δ F , ( M σ ) 23 = δ ( z − 1 / 3 ) / h F , (M_\sigma)_{12}=a(x-1/2)/\delta_F,\qquad(M_\sigma)_{23}=\delta(z-1/3)/h_F, ( M σ ) 12 = a ( x − 1/2 ) / δ F , ( M σ ) 23 = δ ( z − 1/3 ) / h F , ( M σ ) 13 = a { y − σ / 2 − ( x − 1 / 2 ) / 3 } / h F . (M_\sigma)_{13}=a\{y-\sigma/2-(x-1/2)/3\}/h_F. ( M σ ) 13 = a { y − σ /2 − ( x − 1/2 ) /3 } / h F . Then
S σ = 3 M σ T M σ tr ( M σ T M σ ) − I S_\sigma=\frac{3M_\sigma^{T}M_\sigma}{\operatorname{tr}(M_\sigma^{T}M_\sigma)}-I S σ = tr ( M σ T M σ ) 3 M σ T M σ − I gives the lattice in (19) , up to an isometry. Thus
∥ S σ ∥ F ≤ 43 / 1000 , ∇ p [ 1 2 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ] = 0 \lVert S_\sigma\rVert_F\le43/1000,\qquad\nabla_p\left[\frac{1}{2}E(U(p)\mathbb{Z}^3,6)-E(U(p)\mathbb{Z}^3,3)\right]=0 ∥ S σ ∥ F ≤ 43/1000 , ∇ p [ 2 1 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ] = 0 ⟹ U ( p ) Z 3 ≅ ( 21 / 20 ) 1 / 3 F . \Longrightarrow\quad U(p)\mathbb{Z}^3\cong(21/20)^{1/3}F. ⟹ U ( p ) Z 3 ≅ ( 21/20 ) 1/3 F . Proof. Taking sixth powers verifies the enclosures, and
δ F 2 = 3 4 a F 2 , h F 2 = 2 3 a F 2 , a F δ F h F = 21 20 . \delta_F^2=\frac{3}{4}a_F^2,\qquad h_F^2=\frac{2}{3}a_F^2,\qquad a_F\delta_Fh_F=\frac{21}{20}. δ F 2 = 4 3 a F 2 , h F 2 = 3 2 a F 2 , a F δ F h F = 20 21 . The normalized Gram matrix of U F , + U_{F,+} U F , + is the FCC form ( 1 i = j + 1 2 1 i ≠ j ) i , j = 1 3 (1_{i=j}+\frac{1}{2}1_{i\ne j})_{i,j=1}^3 ( 1 i = j + 2 1 1 i = j ) i , j = 1 3 . The column operation v 3 ← v 3 − v 1 v_3\leftarrow v_3-v_1 v 3 ← v 3 − v 1 gives U F , − U_{F,-} U F , − ; inversion gives the relative entries. Since det M σ = 1 \det M_\sigma=1 det M σ = 1 , right polar decomposition M σ = O P M_\sigma=OP M σ = O P yields
det ( I + S σ ) − 1 / 6 ( I + S σ ) 1 / 2 = P . \det(I+S_\sigma)^{-1/6}(I+S_\sigma)^{1/2}=P. det ( I + S σ ) − 1/6 ( I + S σ ) 1/2 = P . Orthogonal conjugation preserves the Frobenius norm and stationary basin. This completes the proof of Lemma 4.8. □
Lemma 4.9 (A full stationary basin for the simple-hexagonal lattice). Put
Y 0 = 110131033 10 8 , t 0 = ( 4 ( 11 / 10 ) 2 3 Y 0 ) 1 / 6 , Y_0=\frac{110131033}{10^8},\qquad t_0=\left(\frac{4(11/10)^2}{3Y_0}\right)^{1/6}, Y 0 = 1 0 8 110131033 , t 0 = ( 3 Y 0 4 ( 11/10 ) 2 ) 1/6 , and let
L 0 = t 0 span Z { ( Y 0 , 0 , 0 ) , ( 0 , 1 , 0 ) , ( 0 , 1 / 2 , 3 / 2 ) } . L_0=t_0\operatorname{span}_{\mathbb{Z}}\left\{(\sqrt{Y_0},0,0),(0,1,0),(0,1/2,\sqrt{3}/2)\right\}. L 0 = t 0 span Z { ( Y 0 , 0 , 0 ) , ( 0 , 1 , 0 ) , ( 0 , 1/2 , 3 /2 ) } . In the chart (19) , put
L ( S ) = det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 L 0 , ∣ L ( S ) ∣ = 11 / 10 , L(S)=\det(I+S)^{-1/6}(I+S)^{1/2}L_0,\qquad|L(S)|=11/10, L ( S ) = det ( I + S ) − 1/6 ( I + S ) 1/2 L 0 , ∣ L ( S ) ∣ = 11/10 , Ψ ( S ) = 1 2 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) , ∇ s Ψ = ( ∂ u Ψ , ∂ v Ψ , ∂ x Ψ , ∂ y Ψ , ∂ z Ψ ) T . \Psi(S)=\tfrac{1}{2}E(L(S),6)-E(L(S),3),\qquad\nabla_s\Psi=(\partial_u\Psi,\partial_v\Psi,\partial_x\Psi,\partial_y\Psi,\partial_z\Psi)^{\mathsf T}. Ψ ( S ) = 2 1 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) , ∇ s Ψ = ( ∂ u Ψ , ∂ v Ψ , ∂ x Ψ , ∂ y Ψ , ∂ z Ψ ) T . Then
∂ ϵ 2 Ψ ( S + ϵ T ) ∣ ϵ = 0 > 1 1000 ∥ T ∥ F 2 ( ∥ S ∥ F ≤ 3 / 100 , T ≠ 0 , T = T T , tr T = 0 ) , \left.\partial_\epsilon^2\Psi(S+\epsilon T)\right|_{\epsilon=0}>\frac{1}{1000}\lVert T\rVert_F^2\qquad(\lVert S\rVert_F\leq3/100,\ T\ne0,\ T=T^{\mathsf T},\ \operatorname{tr}T=0), ∂ ϵ 2 Ψ ( S + ϵ T ) ϵ = 0 > 1000 1 ∥ T ∥ F 2 (∥ S ∥ F ≤ 3/100 , T = 0 , T = T T , tr T = 0 ) , S T ∇ s Ψ ( S ) = ∂ ϵ Ψ ( ( 1 + ϵ ) S ) ∣ ϵ = 0 > 1 1000 ∥ S ∥ F 2 ( 3 / 100 ≤ ∥ S ∥ F ≤ 7 / 100 ) . \mathsf{S}^{\mathsf T}\nabla_s\Psi(S)=\left.\partial_\epsilon\Psi((1+\epsilon)S)\right|_{\epsilon=0}>\frac{1}{1000}\lVert S\rVert_F^2\qquad(3/100\leq\lVert S\rVert_F\leq7/100). S T ∇ s Ψ ( S ) = ∂ ϵ Ψ (( 1 + ϵ ) S ) ∣ ϵ = 0 > 1000 1 ∥ S ∥ F 2 ( 3/100 ≤ ∥ S ∥ F ≤ 7/100 ) . The root Y ∗ Y_* Y ∗ is determined by
d d Y { E ( H Y , 6 ) − 2 ( 11 / 10 ) 2 E ( H Y , 3 ) } ∣ Y = Y ∗ = 0 , \left.\frac{\mathrm d}{\mathrm dY}\left\{E(H_Y,6)-2(11/10)^2E(H_Y,3)\right\}\right|_{Y=Y_*}=0, d Y d { E ( H Y , 6 ) − 2 ( 11/10 ) 2 E ( H Y , 3 ) } Y = Y ∗ = 0 , 1.1013103262 < Y ∗ < 1.1013103264. 1.1013103262<Y_*<1.1013103264. 1.1013103262 < Y ∗ < 1.1013103264. Put
S ∗ = Y ∗ − Y 0 Y ∗ + 2 Y 0 diag ( 2 , − 1 , − 1 ) , ∥ S ∗ ∥ F < 3 ⋅ 10 − 9 . S_*=\frac{Y_*-Y_0}{Y_*+2Y_0}\operatorname{diag}(2,-1,-1),\qquad\lVert S_*\rVert_F<3\cdot10^{-9}. S ∗ = Y ∗ + 2 Y 0 Y ∗ − Y 0 diag ( 2 , − 1 , − 1 ) , ∥ S ∗ ∥ F < 3 ⋅ 1 0 − 9 . Then
∇ s Ψ ( S ) = 0 ⟺ S = S ∗ ( ∥ S ∥ F ≤ 7 / 100 ) , \nabla_s\Psi(S)=0\Longleftrightarrow S=S_*\qquad(\lVert S\rVert_F\leq7/100), ∇ s Ψ ( S ) = 0 ⟺ S = S ∗ (∥ S ∥ F ≤ 7/100 ) , argmin ∥ S ∥ F ≤ 7 / 100 Ψ ( S ) = { S ∗ } . \underset{\lVert S\rVert_F\leq7/100}{\operatorname{argmin}}\ \Psi(S)=\{S_*\}. ∥ S ∥ F ≤ 7/100 argmin Ψ ( S ) = { S ∗ } .
Proof. Lemma A.6 and hexagonal symmetry give
( ∂ v , ∂ x , ∂ y , ∂ z ) Ψ ( diag ( 2 u , − u , − u ) ) = 0 ( 6 ∣ u ∣ ≤ 7 / 100 ) , (\partial_v,\partial_x,\partial_y,\partial_z)\Psi(\operatorname{diag}(2u,-u,-u))=0\qquad(\sqrt{6}|u|\leq7/100), ( ∂ v , ∂ x , ∂ y , ∂ z ) Ψ ( diag ( 2 u , − u , − u )) = 0 ( 6 ∣ u ∣ ≤ 7/100 ) ,
∂ u Ψ ( S ∗ ) = 0 ⟹ ∇ s Ψ ( S ∗ ) = 0. \partial_u\Psi(S_*)=0\quad\Longrightarrow\quad\nabla_s\Psi(S_*)=0. ∂ u Ψ ( S ∗ ) = 0 ⟹ ∇ s Ψ ( S ∗ ) = 0.
For every w ∈ L 0 ∖ { 0 } w\in L_0\setminus\{0\} w ∈ L 0 ∖ { 0 } ,
w = t 0 ( Y 0 k , p / 2 , 3 j / 2 ) , p , j , k ∈ Z , ( p , j , k ) ≠ ( 0 , 0 , 0 ) , p + j ∈ 2 Z . w=t_0\left(\sqrt{Y_0}k,p/2,\sqrt{3}j/2\right),\qquad p,j,k\in\mathbb{Z},\qquad(p,j,k)\ne(0,0,0),\qquad p+j\in2\mathbb{Z}. w = t 0 ( Y 0 k , p /2 , 3 j /2 ) , p , j , k ∈ Z , ( p , j , k ) = ( 0 , 0 , 0 ) , p + j ∈ 2 Z .
The exact finite vector set F ⊂ L 0 \mathcal{F}\subset L_0 F ⊂ L 0 is specified by
∣ p ∣ ≤ 69 , ∣ j ∣ ≤ 40 , ∣ k ∣ ≤ 33 , p + j ∈ 2 Z , 0 < p 2 + 3 j 2 + 4 Y 0 k 2 ≤ 4800. |p|\leq69,\qquad|j|\leq40,\qquad|k|\leq33,\qquad p+j\in2\mathbb{Z},\qquad0<p^2+3j^2+4Y_0k^2\leq4800. ∣ p ∣ ≤ 69 , ∣ j ∣ ≤ 40 , ∣ k ∣ ≤ 33 , p + j ∈ 2 Z , 0 < p 2 + 3 j 2 + 4 Y 0 k 2 ≤ 4800.
Clear the positive denominator of Y 0 Y_0 Y 0 in the cutoff; the set is invariant under planar hexagonal symmetries and coordinate sign changes. Put
q w = t 0 2 4 ( p 2 + 3 j 2 + 4 Y 0 k 2 ) , q_w=\frac{t_0^2}{4}\left(p^2+3j^2+4Y_0k^2\right), q w = 4 t 0 2 ( p 2 + 3 j 2 + 4 Y 0 k 2 ) ,
ℓ w ( S ) = u ( 8 Y 0 k 2 − p 2 − 3 j 2 ) + v ( p 2 − 3 j 2 ) + 4 x Y 0 k p + 4 y 3 Y 0 k j + 2 z 3 p j p 2 + 3 j 2 + 4 Y 0 k 2 . \ell_w(S)=\frac{u(8Y_0k^2-p^2-3j^2)+v(p^2-3j^2)+4x\sqrt{Y_0}kp+4y\sqrt{3Y_0}kj+2z\sqrt{3}pj}{p^2+3j^2+4Y_0k^2}. ℓ w ( S ) = p 2 + 3 j 2 + 4 Y 0 k 2 u ( 8 Y 0 k 2 − p 2 − 3 j 2 ) + v ( p 2 − 3 j 2 ) + 4 x Y 0 k p + 4 y 3 Y 0 k j + 2 z 3 p j .
Lemma 2.9 gives P P P , deg P ≤ 12 \deg P\leq12 deg P ≤ 12 . Exact determinant multiplication and sign summation give rational coefficients; the exponents of x , y , z x,y,z x , y , z are all even or all odd. Retain the 138 raw positions through degree six, including odd powers of v v v , and all 25193 nonnegative triples. Use (18) with
λ 1 ( L 0 ) > 1 , min w ∈ L 0 ∖ ( F ∪ { 0 } ) ∣ w ∣ > 32 , a = 1 , R = 32. \lambda_1(L_0)>1,\qquad\min_{w\in L_0\setminus(\mathcal{F}\cup\{0\})}|w|>32,\qquad a=1,\quad R=32. λ 1 ( L 0 ) > 1 , w ∈ L 0 ∖ ( F ∪ { 0 }) min ∣ w ∣ > 32 , a = 1 , R = 32.
Finite rational summation gives
∑ w ∈ F q w − 3 + T 3 ( 1 , 32 ) < 3332 500 , ∑ w ∈ F q w − 6 + T 6 ( 1 , 32 ) < 339 100 . \sum_{w\in\mathcal{F}}q_w^{-3}+T_3(1,32)<\frac{3332}{500},\qquad\sum_{w\in\mathcal{F}}q_w^{-6}+T_6(1,32)<\frac{339}{100}. w ∈ F ∑ q w − 3 + T 3 ( 1 , 32 ) < 500 3332 , w ∈ F ∑ q w − 6 + T 6 ( 1 , 32 ) < 100 339 .
Lemma 2.9 , with these moments, gives the complete binomial and spatial remainder
∂ ϵ 2 { Ψ ( S + ϵ T ) − P ( S + ϵ T ) } ∣ ϵ = 0 ≤ ε ( R ) ∥ T ∥ F 2 , \left.\partial_{\epsilon}^{2}\{\Psi(S+\epsilon T)-P(S+\epsilon T)\}\right|_{\epsilon=0}\leq\varepsilon(R)\lVert T\rVert_F^2, ∂ ϵ 2 { Ψ ( S + ϵ T ) − P ( S + ϵ T )} ϵ = 0 ≤ ε ( R ) ∥ T ∥ F 2 ,
ε ( 3 / 100 ) < 7 2000 , ε ( 7 / 100 ) < 393 10000 . \varepsilon(3/100)<\frac{7}{2000},\qquad\varepsilon(7/100)<\frac{393}{10000}. ε ( 3/100 ) < 2000 7 , ε ( 7/100 ) < 10000 393 .
The weighted monomial maxima give
∂ ϵ 2 P ( S + ϵ T ) ∣ ϵ = 0 ≥ ∣ η ∣ T C H ∣ η ∣ ( ∥ S ∥ F ≤ 3 / 100 ) , \left.\partial_{\epsilon}^{2}P(S+\epsilon T)\right|_{\epsilon=0}\geq\lvert\eta\rvert^{\mathsf{T}}C_H\lvert\eta\rvert\qquad(\lVert S\rVert_F\leq3/100), ∂ ϵ 2 P ( S + ϵ T ) ϵ = 0 ≥ ∣ η ∣ T C H ∣ η ∣ (∥ S ∥ F ≤ 3/100 ) ,
where η = ( T 11 / 2 , ( T 22 − T 33 ) / 2 , T 12 , T 13 , T 23 ) T \eta=(T_{11}/2,(T_{22}-T_{33})/2,T_{12},T_{13},T_{23})^{\mathsf{T}} η = ( T 11 /2 , ( T 22 − T 33 ) /2 , T 12 , T 13 , T 23 ) T and
C H = 1 10 6 ( 13105271 − 2625768 − 102166 − 102139 − 2618185 − 2625768 5071238 − 37497 − 34804 − 1356023 − 102166 − 37497 237405 − 21289 − 34557 − 102139 − 34804 − 21289 237764 − 37032 − 2618185 − 1356023 − 34557 − 37032 5074732 ) . C_H=\frac{1}{10^{6}}
\begin{pmatrix}
13105271 & -2625768 & -102166 & -102139 & -2618185\\
-2625768 & 5071238 & -37497 & -34804 & -1356023\\
-102166 & -37497 & 237405 & -21289 & -34557\\
-102139 & -34804 & -21289 & 237764 & -37032\\
-2618185 & -1356023 & -34557 & -37032 & 5074732
\end{pmatrix}. C H = 1 0 6 1 13105271 − 2625768 − 102166 − 102139 − 2618185 − 2625768 5071238 − 37497 − 34804 − 1356023 − 102166 − 37497 237405 − 21289 − 34557 − 102139 − 34804 − 21289 237764 − 37032 − 2618185 − 1356023 − 34557 − 37032 5074732 .
The exact L D L T LDL^{\mathsf{T}} L D L T pivots of C H − ( 9 / 2000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_H-(9/2000)\operatorname{diag}(6,2,2,2,2) C H − ( 9/2000 ) diag ( 6 , 2 , 2 , 2 , 2 ) are greater than
13 , 4 , 1 5 , 1 5 , 3. 13,\qquad4,\qquad\frac{1}{5},\qquad\frac{1}{5},\qquad3. 13 , 4 , 5 1 , 5 1 , 3.
Thus ∇ s 2 Ψ ( S ) ≻ 0 \nabla_s^2\Psi(S)\succ0 ∇ s 2 Ψ ( S ) ≻ 0 for ∥ S ∥ F ≤ 3 / 100 \lVert S\rVert_F\leq3/100 ∥ S ∥ F ≤ 3/100 .
For the radial polynomial B B B of Lemma 2.9 , on ∥ S ∥ F ≤ 7 / 100 \lVert S\rVert_F\leq7/100 ∥ S ∥ F ≤ 7/100 ,
η T B ( S ) η ≥ ∣ η ∣ T C R ∣ η ∣ , \eta^{\mathsf{T}}B(S)\eta\geq\lvert\eta\rvert^{\mathsf{T}}C_R\lvert\eta\rvert, η T B ( S ) η ≥ ∣ η ∣ T C R ∣ η ∣ ,
where
C R = 1 10 6 ( 12644187 − 3333878 − 142310 − 142138 − 3309727 − 3333878 4476221 − 60016 − 54472 − 1713026 − 142310 − 60016 185878 − 27728 − 53659 − 142138 − 54472 − 27728 187091 − 58498 − 3309727 − 1713026 − 53659 − 58498 4487900 ) . C_R=\frac{1}{10^{6}}
\begin{pmatrix}
12644187 & -3333878 & -142310 & -142138 & -3309727\\
-3333878 & 4476221 & -60016 & -54472 & -1713026\\
-142310 & -60016 & 185878 & -27728 & -53659\\
-142138 & -54472 & -27728 & 187091 & -58498\\
-3309727 & -1713026 & -53659 & -58498 & 4487900
\end{pmatrix}. C R = 1 0 6 1 12644187 − 3333878 − 142310 − 142138 − 3309727 − 3333878 4476221 − 60016 − 54472 − 1713026 − 142310 − 60016 185878 − 27728 − 53659 − 142138 − 54472 − 27728 187091 − 58498 − 3309727 − 1713026 − 53659 − 58498 4487900 .
The matrix C R − ( 4031 / 100000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_R-(4031/100000)\operatorname{diag}(6,2,2,2,2) C R − ( 4031/100000 ) diag ( 6 , 2 , 2 , 2 , 2 ) is positive definite: its exact L D L T LDL^{\mathsf{T}} L D L T pivots exceed
12 , 3 , 9 100 , 8 100 , 7 10 . 12,\qquad3,\qquad\frac{9}{100},\qquad\frac{8}{100},\qquad\frac{7}{10}. 12 , 3 , 100 9 , 100 8 , 10 7 .
Since ∣ ∂ u Ψ ( 0 ) ∣ < 10 − 7 \lvert\partial_u\Psi(0)\rvert<10^{-7} ∣ ∂ u Ψ ( 0 )∣ < 1 0 − 7 and ( ∂ v , ∂ x , ∂ y , ∂ z ) Ψ ( 0 ) = 0 (\partial_v,\partial_x,\partial_y,\partial_z)\Psi(0)=0 ( ∂ v , ∂ x , ∂ y , ∂ z ) Ψ ( 0 ) = 0 ,
∣ ∂ ϵ Ψ ( ϵ S ) ∣ ∣ ϵ = 0 < 1 100000 ∥ S ∥ F 2 ( ∥ S ∥ F ≥ 3 / 100 ) . \left.\lvert\partial_{\epsilon}\Psi(\epsilon S)\rvert\right|_{\epsilon=0}<\frac{1}{100000}\lVert S\rVert_F^2\qquad(\lVert S\rVert_F\geq3/100). ∣ ∂ ϵ Ψ ( ϵ S )∣ ∣ ϵ = 0 < 100000 1 ∥ S ∥ F 2 (∥ S ∥ F ≥ 3/100 ) .
Euler’s identity and the integrated complete Hessian remainder give
∂ ϵ Ψ ( ( 1 + ϵ ) S ) ∣ ϵ = 0 > 1 1000 ∥ S ∥ F 2 \left.\partial_{\epsilon}\Psi((1+\epsilon)S)\right|_{\epsilon=0}>\frac{1}{1000}\lVert S\rVert_F^2 ∂ ϵ Ψ (( 1 + ϵ ) S ) ∣ ϵ = 0 > 1000 1 ∥ S ∥ F 2
on 3 / 100 ≤ ∥ S ∥ F ≤ 7 / 100 3/100\leq\lVert S\rVert_F\leq7/100 3/100 ≤ ∥ S ∥ F ≤ 7/100 . Hence
3 / 100 ≤ ∥ S ∥ F ≤ 7 / 100 ⟹ ∇ s Ψ ( S ) ≠ 0 , 3/100\leq\lVert S\rVert_F\leq7/100\quad\Longrightarrow\quad\nabla_s\Psi(S)\ne0, 3/100 ≤ ∥ S ∥ F ≤ 7/100 ⟹ ∇ s Ψ ( S ) = 0 ,
{ S : ∥ S ∥ F ≤ 3 / 100 , ∇ s Ψ ( S ) = 0 } = { S ∗ } , \{S:\lVert S\rVert_F\leq3/100,\ \nabla_s\Psi(S)=0\}=\{S_*\}, { S : ∥ S ∥ F ≤ 3/100 , ∇ s Ψ ( S ) = 0 } = { S ∗ } ,
by inner strict convexity and ∇ s Ψ ( S ∗ ) = 0 \nabla_s\Psi(S_*)=0 ∇ s Ψ ( S ∗ ) = 0 . This completes the proof of Lemma 4.9 . □
Lemma 4.10 (The simple-hexagonal reference chart). Put Y 0 = 110131033 / 10 8 Y_0 = 110131033/10^8 Y 0 = 110131033/1 0 8 . Define positive reference lengths by
a 0 6 = 484000000 330393099 , δ 0 6 = 27 64 a 0 6 , h 0 6 = Y 0 3 a 0 6 . a_0^6 = \frac{484000000}{330393099}, \qquad\delta_0^6 = \frac{27}{64}a_0^6, \qquad h_0^6 = Y_0^3 a_0^6. a 0 6 = 330393099 484000000 , δ 0 6 = 64 27 a 0 6 , h 0 6 = Y 0 3 a 0 6 . Then
U 0 = ( a 0 a 0 / 2 0 0 δ 0 0 0 0 h 0 ) U_0 = \begin{pmatrix} a_0 & a_0/2 & 0 \\ 0 & \delta_0 & 0 \\ 0 & 0 & h_0 \end{pmatrix} U 0 = a 0 0 0 a 0 /2 δ 0 0 0 0 h 0 generates ( 11 / 10 ) 1 / 3 H Y 0 (11/10)^{1/3}H_{Y_0} ( 11/10 ) 1/3 H Y 0 , with triangular plane first and vertical axis third. Exact rational bounds are
1065701881072846 10 15 < a 0 < 1065701881072847 10 15 , \frac{1065701881072846}{10^{15}} < a_0 < \frac{1065701881072847}{10^{15}}, 1 0 15 1065701881072846 < a 0 < 1 0 15 1065701881072847 , 922924901869947 10 15 < δ 0 < 922924901869948 10 15 , \frac{922924901869947}{10^{15}} < \delta_0 < \frac{922924901869948}{10^{15}}, 1 0 15 922924901869947 < δ 0 < 1 0 15 922924901869948 , 1118383081907223 10 15 < h 0 < 1118383081907224 10 15 . \frac{1118383081907223}{10^{15}} < h_0 < \frac{1118383081907224}{10^{15}}. 1 0 15 1118383081907223 < h 0 < 1 0 15 1118383081907224 . For M = U ( p ) U 0 − 1 M = U(p)U_0^{-1} M = U ( p ) U 0 − 1 ,
M = ( a / a 0 a ( x − 1 / 2 ) / δ 0 a y / h 0 0 δ / δ 0 δ z / h 0 0 0 h / h 0 ) . M = \begin{pmatrix} a/a_0 & a(x-1/2)/\delta_0 & ay/h_0 \\ 0 & \delta/\delta_0 & \delta z/h_0 \\ 0 & 0 & h/h_0 \end{pmatrix}. M = a / a 0 0 0 a ( x − 1/2 ) / δ 0 δ / δ 0 0 a y / h 0 δ z / h 0 h / h 0 . Consequently
S = 3 M T M tr ( M T M ) − I . S = \frac{3M^{\mathsf T}M}{\operatorname{tr}(M^{\mathsf T}M)}-I. S = tr ( M T M ) 3 M T M − I . Use (19) with L 0 = U 0 Z 3 L_0 = U_0\mathbb{Z}^3 L 0 = U 0 Z 3 , and let Y ∗ Y_\ast Y ∗ be the root in Lemma A.6 . For ∥ S ∥ F ≤ 7 / 100 \lVert S\rVert_F \le7/100 ∥ S ∥ F ≤ 7/100 ,
∇ S [ 1 2 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) ] = 0 ⟺ S = Y ∗ − Y 0 Y ∗ + 2 Y 0 diag ( − 1 , − 1 , 2 ) , \nabla_S \left[\frac{1}{2}E(L(S),6)-E(L(S),3)\right] = 0 \Longleftrightarrow S = \frac{Y_\ast-Y_0}{Y_\ast+2Y_0}\operatorname{diag}(-1,-1,2), ∇ S [ 2 1 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) ] = 0 ⟺ S = Y ∗ + 2 Y 0 Y ∗ − Y 0 diag ( − 1 , − 1 , 2 ) , and
∇ S 2 [ 1 2 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) ] ≻ 0 ( ∥ S ∥ F ≤ 3 / 100 ) . \nabla_S^2 \left[\frac{1}{2}E(L(S),6)-E(L(S),3)\right] \succ0 \qquad(\lVert S\rVert_F \le3/100). ∇ S 2 [ 2 1 E ( L ( S ) , 6 ) − E ( L ( S ) , 3 ) ] ≻ 0 (∥ S ∥ F ≤ 3/100 ) . Proof.
δ 0 2 = 3 4 a 0 2 , h 0 2 = Y 0 a 0 2 , a 0 δ 0 h 0 = 11 10 , U 0 Z 3 = ( 11 / 10 ) 1 / 3 H Y 0 . \delta_0^2 = \frac{3}{4}a_0^2, \qquad h_0^2 = Y_0a_0^2, \qquad a_0\delta_0h_0 = \frac{11}{10}, \qquad U_0\mathbb{Z}^3 = (11/10)^{1/3}H_{Y_0}. δ 0 2 = 4 3 a 0 2 , h 0 2 = Y 0 a 0 2 , a 0 δ 0 h 0 = 10 11 , U 0 Z 3 = ( 11/10 ) 1/3 H Y 0 . Inversion gives M M M ; the polar argument of Lemma 4.8 gives S S S . Lemma 4.9 applies after the cyclic permutation of the coordinate axes. This completes the proof of Lemma 4.10 .
Theorem 4.11 (FCC optimality at 21/20). At ∣ L ∣ = 21 / 20 \lvert L\rvert= 21/20 ∣ L ∣ = 21/20 ,
E ( L ) = m ( 21 / 20 ) ⟺ L ≅ ( 21 / 20 ) 1 / 3 F , \mathcal{E}(L) = m(21/20) \Longleftrightarrow L \cong(21/20)^{1/3}F, E ( L ) = m ( 21/20 ) ⟺ L ≅ ( 21/20 ) 1/3 F , and
− 5.30745025029370902112 < m ( 21 / 20 ) < − 5.30745025029370902111. -5.30745025029370902112 < m(21/20) < -5.30745025029370902111. − 5.30745025029370902112 < m ( 21/20 ) < − 5.30745025029370902111. Proof. The minimum is attained. By the FCC reference bound, Proposition 4.6 and Lemmas A.5 and 4.7 ,
E ( L ) = m ( 21 / 20 ) < T F ⟹ L ≅ U ( p ) Z 3 , p ∈ D F \mathcal{E}(L) = m(21/20) < T_F \Longrightarrow L \cong U(p)\mathbb{Z}^3,\qquad p \in D_F E ( L ) = m ( 21/20 ) < T F ⟹ L ≅ U ( p ) Z 3 , p ∈ D F ⟹ L ≅ ( 21 / 20 ) 1 / 3 F . \Longrightarrow L \cong(21/20)^{1/3}F. ⟹ L ≅ ( 21/20 ) 1/3 F . The FCC reference evaluation gives the stated energy interval. This completes the proof of Theorem 4.11 .
Theorem 4.12 (Simple-hexagonal optimality at 11/10). At ∣ L ∣ = 11 / 10 \lvert L\rvert= 11/10 ∣ L ∣ = 11/10 ,
E ( L ) = m ( 11 / 10 ) ⟺ L ≅ ( 11 / 10 ) 1 / 3 H Y ∗ , \mathcal{E}(L) = m(11/10) \Longleftrightarrow L \cong(11/10)^{1/3}H_{Y_\ast}, E ( L ) = m ( 11/10 ) ⟺ L ≅ ( 11/10 ) 1/3 H Y ∗ , where
1.1013103262 < Y ∗ < 1.1013103264. 1.1013103262 < Y_\ast< 1.1013103264. 1.1013103262 < Y ∗ < 1.1013103264.
Proof. By Proposition 4.6 and Lemmas A.6 , A.7 and 4.9 ,
E ( L ) = m ( 11 / 10 ) < − 24842327 5000000 ⟹ L ≅ ( 11 / 10 ) 1 / 3 H Y ∗ . \mathcal{E}(L)=m(11/10)<-\frac{24842327}{5000000}\quad\Longrightarrow\quad L\cong(11/10)^{1/3}H_{Y_*}. E ( L ) = m ( 11/10 ) < − 5000000 24842327 ⟹ L ≅ ( 11/10 ) 1/3 H Y ∗ .
The root bracket gives the planar area bound
983 1000 < ( 3 ( 11 / 10 ) 2 2 Y ∗ ) 1 / 3 < 123 125 . \frac{983}{1000}<\left(\frac{\sqrt{3}(11/10)^2}{2Y_*}\right)^{1/3}<\frac{123}{125}. 1000 983 < ( 2 Y ∗ 3 ( 11/10 ) 2 ) 1/3 < 125 123 .
This completes the proof of Theorem 4.12 . □
The transition interval
For 21 / 20 ≤ V ≤ 11 / 10 21/20\le V\le11/10 21/20 ≤ V ≤ 11/10 , the profile below reduces minimization to five-coordinate stationary points in a compact domain. Write
Φ V ( Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) , ∣ Λ ∣ = 1 , \Phi_V(\Lambda)=E(\Lambda,6)-2V^2E(\Lambda,3),\qquad|\Lambda|=1, Φ V ( Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) , ∣Λ∣ = 1 ,
E ( V 1 / 3 Λ ) = E ( Λ , 6 ) 2 V 4 − E ( Λ , 3 ) V 2 = Φ V ( Λ ) 2 V 4 , \mathcal{E}(V^{1/3}\Lambda)=\frac{E(\Lambda,6)}{2V^4}-\frac{E(\Lambda,3)}{V^2}=\frac{\Phi_V(\Lambda)}{2V^4}, E ( V 1/3 Λ ) = 2 V 4 E ( Λ , 6 ) − V 2 E ( Λ , 3 ) = 2 V 4 Φ V ( Λ ) ,
λ 1 ( Λ ) = min v ∈ Λ ∖ { 0 } ∣ v ∣ . \lambda_1(\Lambda)=\min_{v\in\Lambda\setminus\{0\}}|v|. λ 1 ( Λ ) = v ∈ Λ ∖ { 0 } min ∣ v ∣.
Proposition 5.1 (One profile comparison on a covolume interval). Let f : [ V 0 , V 1 ] → R f:[V_0,V_1]\to\mathbb{R} f : [ V 0 , V 1 ] → R be continuous and concave as a function of V 2 V^2 V 2 , where 0 < V 0 < V 1 0<V_0<V_1 0 < V 0 < V 1 . Suppose
E ( Λ , 6 ) − 2 V i 2 E ( Λ , 3 ) ≥ f ( V i ) ( ∣ Λ ∣ = 1 , i = 0 , 1 ) . (45) E(\Lambda,6)-2V_i^2E(\Lambda,3)\ge f(V_i)\qquad(|\Lambda|=1,\ i=0,1). \tag*{(45)} E ( Λ , 6 ) − 2 V i 2 E ( Λ , 3 ) ≥ f ( V i ) ( ∣Λ∣ = 1 , i = 0 , 1 ) . (45) Set
b 0 = − f + ′ ( V 0 ) 4 V 0 , b 1 = − f − ′ ( V 1 ) 4 V 1 , A 1 = f ( V 1 ) + 2 V 1 2 b 1 , b_0=-\frac{f'_+(V_0)}{4V_0},\qquad b_1=-\frac{f'_-(V_1)}{4V_1},\qquad A_1=f(V_1)+2V_1^2b_1, b 0 = − 4 V 0 f + ′ ( V 0 ) , b 1 = − 4 V 1 f − ′ ( V 1 ) , A 1 = f ( V 1 ) + 2 V 1 2 b 1 , with 0 < b 0 ≤ b 1 < ∞ 0<b_0\le b_1<\infty 0 < b 0 ≤ b 1 < ∞ and A 1 > 0 A_1>0 A 1 > 0 .
J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 ) ) , J(\Lambda)=E(\Lambda,6)-g(E(\Lambda,3)), J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 )) , g ( t ) = max V 0 2 ≤ V 2 ≤ V 1 2 { f ( V ) + 2 V 2 t } . g(t)=\max_{V_0^2\le V^2\le V_1^2}\{f(V)+2V^2t\}. g ( t ) = V 0 2 ≤ V 2 ≤ V 1 2 max { f ( V ) + 2 V 2 t } . Then
J ( Λ ) = min V 0 ≤ V ≤ V 1 { Φ V ( Λ ) − f ( V ) } , (46) J(\Lambda)=\min_{V_0\le V\le V_1}\{\Phi_V(\Lambda)-f(V)\}, \tag*{(46)} J ( Λ ) = V 0 ≤ V ≤ V 1 min { Φ V ( Λ ) − f ( V )} , (46) and, for ∣ Λ ∣ = 1 |\Lambda|=1 ∣Λ∣ = 1 ,
J ( Λ ) ≥ 0 ⟺ Φ V ( Λ ) ≥ f ( V ) ( V 0 ≤ V ≤ V 1 ) . J(\Lambda)\ge0\quad\Longleftrightarrow\quad\Phi_V(\Lambda)\ge f(V)\quad(V_0\le V\le V_1). J ( Λ ) ≥ 0 ⟺ Φ V ( Λ ) ≥ f ( V ) ( V 0 ≤ V ≤ V 1 ) . Moreover,
J ( Λ ) < 0 ⟹ b 0 < E ( Λ , 3 ) < b 1 , E ( Λ , 6 ) < A 1 , λ 1 ( Λ ) > ( 2 / A 1 ) 1 / 12 ; J(\Lambda)<0\quad\Longrightarrow\quad b_0<E(\Lambda,3)<b_1,\quad E(\Lambda,6)<A_1,\quad\lambda_1(\Lambda)>(2/A_1)^{1/12}; J ( Λ ) < 0 ⟹ b 0 < E ( Λ , 3 ) < b 1 , E ( Λ , 6 ) < A 1 , λ 1 ( Λ ) > ( 2/ A 1 ) 1/12 ; and inf J < 0 ⟹ arg min J ≠ ∅ \inf J<0\Longrightarrow\arg\min J\ne\varnothing inf J < 0 ⟹ arg min J = ∅ . For every
V ∈ arg max V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ ∗ , 3 ) } , V\in\underset{V_0\le v\le V_1}{\arg\max}\{f(v)+2v^2E(\Lambda_*,3)\}, V ∈ V 0 ≤ v ≤ V 1 arg max { f ( v ) + 2 v 2 E ( Λ ∗ , 3 )} , Λ ∗ ∈ arg min l o c J ⟹ Λ ∗ ∈ arg min l o c Φ V , \Lambda_*\in\underset{\mathrm{loc}}{\arg\min}\,J\quad\Longrightarrow\quad\Lambda_*\in\underset{\mathrm{loc}}{\arg\min}\,\Phi_V, Λ ∗ ∈ loc arg min J ⟹ Λ ∗ ∈ loc arg min Φ V , Λ ∗ ∈ arg min J ⟹ Λ ∗ ∈ arg min Φ V . \Lambda_*\in\arg\min J\quad\Longrightarrow\quad\Lambda_*\in\arg\min\Phi_V. Λ ∗ ∈ arg min J ⟹ Λ ∗ ∈ arg min Φ V . For V 0 < V < V 1 V_0<V<V_1 V 0 < V < V 1 ,
− f − ′ ( V ) 4 V ≤ E ( Λ ∗ , 3 ) ≤ − f + ′ ( V ) 4 V . -\frac{f'_-(V)}{4V}\le E(\Lambda_*,3)\le-\frac{f'_+(V)}{4V}. − 4 V f − ′ ( V ) ≤ E ( Λ ∗ , 3 ) ≤ − 4 V f + ′ ( V ) . No differentiability at a contact is required. In particular,
J ≥ 0 , J ( Λ ∗ ) = 0 ⟹ Λ ∗ ∈ arg min J . J\ge0,\quad J(\Lambda_*)=0\quad\Longrightarrow\quad\Lambda_*\in\arg\min J. J ≥ 0 , J ( Λ ∗ ) = 0 ⟹ Λ ∗ ∈ arg min J . Endpoint equality uses (5.1); shape variations remain unrestricted at reduction boundaries.
Proof. Concavity in V 2 V^2 V 2 gives
g ( t ) = f ( V 0 ) + 2 V 0 2 t ( t ≤ b 0 ) , g ( t ) = f ( V 1 ) + 2 V 1 2 t ( t ≥ b 1 ) . g(t)=f(V_0)+2V_0^2t\quad(t\le b_0),\qquad g(t)=f(V_1)+2V_1^2t\quad(t\ge b_1). g ( t ) = f ( V 0 ) + 2 V 0 2 t ( t ≤ b 0 ) , g ( t ) = f ( V 1 ) + 2 V 1 2 t ( t ≥ b 1 ) .
Together with (45) and monotonicity of g g g ,
J ( Λ ) < 0 ⟹ { b 0 < E ( Λ , 3 ) < b 1 , 2 λ 1 ( Λ ) − 12 ≤ E ( Λ , 6 ) < g ( b 1 ) = A 1 . J(\Lambda)<0\ \Longrightarrow\ \begin{cases} b_0<E(\Lambda,3)<b_1,\\ 2\lambda_1(\Lambda)^{-12}\le E(\Lambda,6)<g(b_1)=A_1. \end{cases} J ( Λ ) < 0 ⟹ { b 0 < E ( Λ , 3 ) < b 1 , 2 λ 1 ( Λ ) − 12 ≤ E ( Λ , 6 ) < g ( b 1 ) = A 1 .
Mahler’s criterion [34 ] and uniform convergence imply inf J < 0 ⟹ argmin J ≠ ∅ \inf J<0\Longrightarrow\operatorname*{argmin}J\ne\varnothing inf J < 0 ⟹ argmin J = ∅ . For any maximizing V V V at E ( Λ ∗ , 3 ) E(\Lambda_\ast,3) E ( Λ ∗ , 3 ) ,
Φ V ( Λ ) − Φ V ( Λ ∗ ) = J ( Λ ) − J ( Λ ∗ ) + g ( E ( Λ , 3 ) ) − 2 V 2 E ( Λ , 3 ) − f ( V ) ≥ J ( Λ ) − J ( Λ ∗ ) . \begin{aligned}
\Phi_V(\Lambda)-\Phi_V(\Lambda_\ast)=J(\Lambda)-J(\Lambda_\ast)+g(E(\Lambda,3))-2V^2E(\Lambda,3)-f(V) \\
&\ge J(\Lambda)-J(\Lambda_\ast).
\end{aligned} Φ V ( Λ ) − Φ V ( Λ ∗ ) = J ( Λ ) − J ( Λ ∗ ) + g ( E ( Λ , 3 )) − 2 V 2 E ( Λ , 3 ) − f ( V ) ≥ J ( Λ ) − J ( Λ ∗ ) .
The interior one-sided condition is
− 4 V E ( Λ ∗ , 3 ) − f − ′ ( V ) ≤ 0 ≤ − 4 V E ( Λ ∗ , 3 ) − f + ′ ( V ) . -4VE(\Lambda_\ast,3)-f'_-(V)\le0\le-4VE(\Lambda_\ast,3)-f'_+(V). − 4 V E ( Λ ∗ , 3 ) − f − ′ ( V ) ≤ 0 ≤ − 4 V E ( Λ ∗ , 3 ) − f + ′ ( V ) .
This completes the proof of Proposition 5.1 . □ \square □
On a smooth branch attaining f ( V ) f(V) f ( V ) ,
∂ Y Φ V ( H Y ) ∣ Y = Y ( V ) = 0 , f ( V ) = Φ V ( H Y ( V ) ) , b ( V ) = E ( H Y ( V ) , 3 ) = − f ′ ( V ) 4 V . \left.\partial_Y\Phi_V(H_Y)\right|_{Y=Y(V)}=0,\qquad f(V)=\Phi_V(H_Y(V)),\qquad b(V)=E(H_Y(V),3)=-\frac{f'(V)}{4V}. ∂ Y Φ V ( H Y ) ∣ Y = Y ( V ) = 0 , f ( V ) = Φ V ( H Y ( V )) , b ( V ) = E ( H Y ( V ) , 3 ) = − 4 V f ′ ( V ) .
If ∂ Y E ( H Y , 3 ) ≠ 0 \partial_YE(H_Y,3)\ne0 ∂ Y E ( H Y , 3 ) = 0 and ∂ Y 2 Φ V ( H Y ) > 0 \partial_Y^2\Phi_V(H_Y)>0 ∂ Y 2 Φ V ( H Y ) > 0 at Y ( V ) Y(V) Y ( V ) , then
b ′ ( V ) = 4 V { ∂ Y E ( H Y , 3 ) } 2 ∂ Y 2 { E ( H Y , 6 ) − 2 V 2 E ( H Y , 3 ) } ∣ Y = Y ( V ) > 0. (47) b'(V)=\left.\frac{4V\{\partial_YE(H_Y,3)\}^2}{\partial_Y^2\{E(H_Y,6)-2V^2E(H_Y,3)\}}\right|_{Y=Y(V)}>0. \tag*{(47)} b ′ ( V ) = ∂ Y 2 { E ( H Y , 6 ) − 2 V 2 E ( H Y , 3 )} 4 V { ∂ Y E ( H Y , 3 ) } 2 Y = Y ( V ) > 0. (47)
For t t t in this branch’s moment range,
V = b − 1 ( t ) , g ( t ) = E ( H Y ( V ) , 6 ) . V=b^{-1}(t),\qquad g(t)=E(H_Y(V),6). V = b − 1 ( t ) , g ( t ) = E ( H Y ( V ) , 6 ) .
At a smooth profile minimum Λ ( q ) \Lambda(q) Λ ( q ) supported by this V V V ,
∇ q J = 0 , ∇ q 2 J = ∇ q 2 Φ V ( Λ ( q ) ) − 4 V b ′ ( V ) ∇ q E ( Λ ( q ) , 3 ) { ∇ q E ( Λ ( q ) , 3 ) } T ⪰ 0. \nabla_qJ=0,\qquad\nabla_q^2J=\nabla_q^2\Phi_V(\Lambda(q))-\frac{4V}{b'(V)}\nabla_qE(\Lambda(q),3)\{\nabla_qE(\Lambda(q),3)\}^{\mathsf T}\succeq0. ∇ q J = 0 , ∇ q 2 J = ∇ q 2 Φ V ( Λ ( q )) − b ′ ( V ) 4 V ∇ q E ( Λ ( q ) , 3 ) { ∇ q E ( Λ ( q ) , 3 ) } T ⪰ 0.
At contacts use the one-sided moment inequalities above.
Lemma 5.2 (Uniform tests over a density interval). In (21) –(22) , let
p = ( a , δ , x , y , z ) , h = ( a δ ) − 1 , B ⊂ { a > 0 , δ > 0 } , c = mid B , p=(a,\delta,x,y,z),\qquad h=(a\delta)^{-1},\qquad B\subset\{a>0,\delta>0\},\qquad c=\operatorname{mid}B, p = ( a , δ , x , y , z ) , h = ( a δ ) − 1 , B ⊂ { a > 0 , δ > 0 } , c = mid B , where B B B is a closed box. Put Q ( p ) = U ( p ) T U ( p ) Q(p)=U(p)^{\mathsf T}U(p) Q ( p ) = U ( p ) T U ( p ) . For [ V − , V + ] ⊂ [ 21 / 20 , 89 / 50 ] [V_-,V_+]\subset[21/20,89/50] [ V − , V + ] ⊂ [ 21/20 , 89/50 ] , set
Φ V ( p ) = E ( U ( p ) Z 3 , 6 ) − 2 V 2 E ( U ( p ) Z 3 , 3 ) . \Phi_V(p)=E(U(p)Z^3,6)-2V^2E(U(p)Z^3,3). Φ V ( p ) = E ( U ( p ) Z 3 , 6 ) − 2 V 2 E ( U ( p ) Z 3 , 3 ) . Let V 2 = ( 1 − t ) V − 2 + t V + 2 V^2=(1-t)V_-^2+tV_+^2 V 2 = ( 1 − t ) V − 2 + t V + 2 , 0 ≤ t ≤ 1 0\le t\le1 0 ≤ t ≤ 1 , and suppose
[ H e ] ⊃ { ∇ p 2 Φ V e ( p ) : p ∈ B } , [ g e ] ∋ ∇ p Φ V e ( c ) ( e = − , + ) , [H_e]\supset\{\nabla_p^2\Phi_{V_e}(p):p\in B\},\qquad[g_e]\ni\nabla_p\Phi_{V_e}(c)\quad(e=-,+), [ H e ] ⊃ { ∇ p 2 Φ V e ( p ) : p ∈ B } , [ g e ] ∋ ∇ p Φ V e ( c ) ( e = − , + ) , [ H ] = hull ( [ H − ] , [ H + ] ) , [ g c ] = hull ( [ g − ] , [ g + ] ) . [H]=\operatorname{hull}([H_-],[H_+]),\qquad[g_c]=\operatorname{hull}([g_-],[g_+]). [ H ] = hull ([ H − ] , [ H + ]) , [ g c ] = hull ([ g − ] , [ g + ]) . For every p ∈ B p\in B p ∈ B , P ∈ R 5 × 5 P\in\mathbb R^{5\times5} P ∈ R 5 × 5 , i ∈ { 1 , … , 5 } i\in\{1,\ldots,5\} i ∈ { 1 , … , 5 } , and fixed nonzero dyadic v ∈ R 5 v\in\mathbb R^5 v ∈ R 5 ,
∇ p Φ V ( p ) = 0 ⟹ p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ] ) ( B − c ) } , \nabla_p\Phi_V(p)=0\Longrightarrow p\in B\cap\{c-P[g_c]+(I-P[H])(B-c)\}, ∇ p Φ V ( p ) = 0 ⟹ p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ]) ( B − c )} , 0 ∉ hull e ∈ { − , + } [ ∂ i Φ V e ] ( B ) ⟹ ∇ p Φ V ( p ) ≠ 0 , 0\notin\operatorname{hull}_{e\in\{-,+\}}[\partial_i\Phi_{V_e}](B)\Longrightarrow\nabla_p\Phi_V(p)\ne0, 0 ∈ / hull e ∈ { − , + } [ ∂ i Φ V e ] ( B ) ⟹ ∇ p Φ V ( p ) = 0 , max e ∈ { − , + } H ∈ [ H e ] v T H v < 0 ⟹ v T ∇ p 2 Φ V ( p ) v < 0. \max_{\substack{e\in\{-,+\}\\ H\in[H_e]}}v^{\mathsf T}Hv<0\Longrightarrow v^{\mathsf T}\nabla_p^2\Phi_V(p)v<0. e ∈ { − , + } H ∈ [ H e ] max v T H v < 0 ⟹ v T ∇ p 2 Φ V ( p ) v < 0. The last condition excludes a local minimum in the full shape chart. For ∣ Λ 0 ∣ = 1 |\Lambda_0|=1 ∣ Λ 0 ∣ = 1 , write
Δ s = E ( U ( c ) Z 3 , s ) − E ( Λ 0 , s ) ( s = 3 , 6 ) , r i = sup p ∈ B ∣ p i − c i ∣ . \Delta_s=E(U(c)Z^3,s)-E(\Lambda_0,s)\quad(s=3,6),\qquad r_i=\sup_{p\in B}|p_i-c_i|. Δ s = E ( U ( c ) Z 3 , s ) − E ( Λ 0 , s ) ( s = 3 , 6 ) , r i = p ∈ B sup ∣ p i − c i ∣.
Let Q e ≥ 0 Q_e \ge0 Q e ≥ 0 be the 32-vertex bound (A.3); thus
sup p ∈ B ∣ ξ i ∣ ≤ r i ξ T ∇ p 2 Φ V ( p ) ξ ≤ Q e . \sup_{\substack{p\in B\\ |\xi_i|\le r_i}} \xi^{\mathsf T}\nabla_p^2\Phi_V(p)\xi\le Q_e. p ∈ B ∣ ξ i ∣ ≤ r i sup ξ T ∇ p 2 Φ V ( p ) ξ ≤ Q e .
For ∇ p Φ V ( p ) = 0 \nabla_p\Phi_V(p)=0 ∇ p Φ V ( p ) = 0 ,
Φ V ( p ) − Φ V ( Λ 0 ) ≥ ( 1 − t ) ( Δ 6 − 2 V − 2 Δ 3 − 1 2 Q − ) + t ( Δ 6 − 2 V + 2 Δ 3 − 1 2 Q + ) . \Phi_V(p)-\Phi_V(\Lambda_0)\ge(1-t)\left(\Delta_6-2V_-^2\Delta_3-\frac{1}{2}Q_-\right)+t\left(\Delta_6-2V_+^2\Delta_3-\frac{1}{2}Q_+\right). Φ V ( p ) − Φ V ( Λ 0 ) ≥ ( 1 − t ) ( Δ 6 − 2 V − 2 Δ 3 − 2 1 Q − ) + t ( Δ 6 − 2 V + 2 Δ 3 − 2 1 Q + ) .
In particular, with the same Λ 0 \Lambda_0 Λ 0 at both endpoints,
∇ p Φ V ( p ) = 0 , min e ∈ { − , + } ( Δ 6 − 2 V e 2 Δ 3 − 1 2 Q e ) > 0 \nabla_p\Phi_V(p)=0,\qquad\min_{e\in\{-,+\}}\left(\Delta_6-2V_e^2\Delta_3-\frac{1}{2}Q_e\right)>0 ∇ p Φ V ( p ) = 0 , e ∈ { − , + } min ( Δ 6 − 2 V e 2 Δ 3 − 2 1 Q e ) > 0
⟹ Φ V ( p ) > Φ V ( Λ 0 ) , U ( p ) Z 3 ∉ argmin ∣ Λ ∣ = 1 Φ V ( Λ ) . \Longrightarrow\quad\Phi_V(p)>\Phi_V(\Lambda_0),\qquad U(p)\mathbb{Z}^3\notin\operatorname*{argmin}_{|\Lambda|=1}\Phi_V(\Lambda). ⟹ Φ V ( p ) > Φ V ( Λ 0 ) , U ( p ) Z 3 ∈ / ∣Λ∣ = 1 argmin Φ V ( Λ ) .
For the scalar test assume
M ( p ) = U ( p ) U ( c ) − 1 , sup p ∈ B ∥ log ( M ( p ) T M ( p ) ) ∥ F ≤ R , M(p)=U(p)U(c)^{-1},\qquad\sup_{p\in B}\left\|\log\left(M(p)^{\mathsf T}M(p)\right)\right\|_F\le R, M ( p ) = U ( p ) U ( c ) − 1 , p ∈ B sup log ( M ( p ) T M ( p ) ) F ≤ R ,
and put d = e − 2 / 3 R d=e^{-\sqrt{2/3}R} d = e − 2/3 R , q c ( i , j , k ) = ( i , j , k ) Q ( c ) ( i , j , k ) T q_c(i,j,k)=(i,j,k)Q(c)(i,j,k)^{\mathsf T} q c ( i , j , k ) = ( i , j , k ) Q ( c ) ( i , j , k ) T . Set
K e = 4 V e 2 d − 3 E ( U ( c ) Z 3 , 3 ) − 4 d 6 E ( U ( c ) Z 3 , 6 ) + ∑ ( i , j , k ) ∈ Z 3 − 4 ≤ i , j , k ≤ 4 ( i , j , k ) ≠ ( 0 , 0 , 0 ) max { 0 , 28 [ d q c ( i , j , k ) ] − 6 − 16 V e 2 [ d q c ( i , j , k ) ] − 3 } . K_e=4V_e^2d^{-3}E(U(c)\mathbb{Z}^3,3)-4d^6E(U(c)\mathbb{Z}^3,6)
+\sum_{\substack{(i,j,k)\in\mathbb{Z}^3\\ -4\le i,j,k\le4\\ (i,j,k)\ne(0,0,0)}}\max\left\{0,28[dq_c(i,j,k)]^{-6}-16V_e^2[dq_c(i,j,k)]^{-3}\right\}. K e = 4 V e 2 d − 3 E ( U ( c ) Z 3 , 3 ) − 4 d 6 E ( U ( c ) Z 3 , 6 ) + ( i , j , k ) ∈ Z 3 − 4 ≤ i , j , k ≤ 4 ( i , j , k ) = ( 0 , 0 , 0 ) ∑ max { 0 , 28 [ d q c ( i , j , k ) ] − 6 − 16 V e 2 [ d q c ( i , j , k ) ] − 3 } .
In the value bound, replace each Q e Q_e Q e by R 2 max ( 0 , K e ) R^2\max(0,K_e) R 2 max ( 0 , K e ) , provided
( 25 d ∥ Q ( c ) − 1 ∥ ∞ − 1 ) 3 ≥ 7 4 V 2 . \left(25d\|Q(c)^{-1}\|_\infty^{-1}\right)^3\ge\frac{7}{4V^2}. ( 25 d ∥ Q ( c ) − 1 ∥ ∞ − 1 ) 3 ≥ 4 V 2 7 .
All intervals include the complete whole-box tails; neither ∇ p Φ V ( c ) = 0 \nabla_p\Phi_V(c)=0 ∇ p Φ V ( c ) = 0 nor endpoint stationarity is assumed.
Proof. Use the complete endpoint bounds of Lemma A.2 .
Normalized kernels. The unit-covolume constants are π 3 / 6 \pi^3/6 π 3 /6 , π 6 / 2160 \pi^6/2160 π 6 /2160 ; the primal kernel derivatives at either endpoint are
( − π ) r { π 6 120 G 6 + r ( π q ) − V 2 π 3 G 3 + r ( π q ) } , r = 0 , 1 , 2. (-\pi)^r\left\{\frac{\pi^6}{120}G_{6+r}(\pi q)-V^2\pi^3G_{3+r}(\pi q)\right\},\qquad r=0,1,2. ( − π ) r { 120 π 6 G 6 + r ( π q ) − V 2 π 3 G 3 + r ( π q ) } , r = 0 , 1 , 2.
For the reciprocal kernel use ( − 9 / 2 + r , − 3 / 2 + r ) (-9/2+r,-3/2+r) ( − 9/2 + r , − 3/2 + r ) instead of ( 6 + r , 3 + r ) (6+r,3+r) ( 6 + r , 3 + r ) , with volume factor one. The primal critical argument solves
∑ j = 0 6 + r ( π q ) j j ! − 2 V 2 b r q 3 ∑ j = 0 3 + r ( π q ) j j ! = 0 , ( b 0 , b 1 , b 2 ) = ( 1 2 , 5 7 , 2 5 ) . \sum_{j=0}^{6+r}\frac{(\pi q)^j}{j!}-2V^2b_rq^3\sum_{j=0}^{3+r}\frac{(\pi q)^j}{j!}=0,\qquad(b_0,b_1,b_2)=\left(\frac{1}{2},\frac{5}{7},\frac{2}{5}\right). j = 0 ∑ 6 + r j ! ( π q ) j − 2 V 2 b r q 3 j = 0 ∑ 3 + r j ! ( π q ) j = 0 , ( b 0 , b 1 , b 2 ) = ( 2 1 , 7 5 , 5 2 ) .
The decreasing integral ratio gives uniqueness; use at most 60 interval bisections in [ 1 / 4 , 8 ] [1/4,8] [ 1/4 , 8 ] , stopping at an inconclusive sign. For reciprocal derivatives the positive-magnitude ratio is < π 3 / 60 < 2 V 2 <\pi^3/60<2V^2 < π 3 /60 < 2 V 2 . Evaluate both terms at the same q q q before subtraction.
Use (25) on the whole box, with these coefficients and
Q ( p ) ⪰ e − 2 / 3 R ∥ Q ( c ) − 1 ∥ ∞ − 1 I , Q ( p ) − 1 ⪰ e − 2 / 3 R ∥ Q ( c ) ∥ ∞ − 1 I . Q(p)\succeq e^{-\sqrt{2/3}R}\|Q(c)^{-1}\|_\infty^{-1}I,\qquad Q(p)^{-1}\succeq e^{-\sqrt{2/3}R}\|Q(c)\|_\infty^{-1}I. Q ( p ) ⪰ e − 2/3 R ∥ Q ( c ) − 1 ∥ ∞ − 1 I , Q ( p ) − 1 ⪰ e − 2/3 R ∥ Q ( c ) ∥ ∞ − 1 I .
Round norm denominators upwards; shortest-vector bounds do not replace these spectral floors.
Stationary values and exclusions. For p ∈ B p \in B p ∈ B ,
∇ p Φ V ( p ) = 0 ⟹ Φ V ( p ) − Φ V ( c ) = − ∫ 0 1 τ ( p − c ) T ∇ p 2 Φ V ( c + τ ( p − c ) ) ( p − c ) d τ . \nabla_p\Phi_V(p)=0 \Longrightarrow\Phi_V(p)-\Phi_V(c)=-\int_0^1 \tau(p-c)^{\mathsf T}\nabla_p^2\Phi_V(c+\tau(p-c))(p-c)\,\mathrm{d}\tau. ∇ p Φ V ( p ) = 0 ⟹ Φ V ( p ) − Φ V ( c ) = − ∫ 0 1 τ ( p − c ) T ∇ p 2 Φ V ( c + τ ( p − c )) ( p − c ) d τ .
Since
∇ p r Φ V = ( 1 − t ) ∇ p r Φ V − + t ∇ p r Φ V + , r = 0 , 1 , 2 , \nabla_p^r\Phi_V=(1-t)\nabla_p^r\Phi_{V_-}+t\nabla_p^r\Phi_{V_+},\qquad r=0,1,2, ∇ p r Φ V = ( 1 − t ) ∇ p r Φ V − + t ∇ p r Φ V + , r = 0 , 1 , 2 ,
(A.3) gives the paired value bound. The scalar summands are convex in V 2 V^2 V 2 ; their endpoint values are upper bounds, not an affine identity for K e K_e K e . For a trace-free Frobenius-unit H H H ,
∂ ϵ 2 ( ∣ ϵ H / 2 v ∣ − 12 − 2 V 2 ∣ ϵ H / 2 v ∣ − 6 ) ∣ ϵ = 0 ≤ 4 V 2 ∣ v ∣ − 6 − 4 ∣ v ∣ − 12 + max ( 0 , 28 ∣ v ∣ − 12 − 16 V 2 ∣ v ∣ − 6 ) . \left.\partial_\epsilon^2\left(|\epsilon^{H/2}v|^{-12}-2V^2|\epsilon^{H/2}v|^{-6}\right)\right|_{\epsilon=0}
\leq4V^2|v|^{-6}-4|v|^{-12}+\max(0,28|v|^{-12}-16V^2|v|^{-6}). ∂ ϵ 2 ( ∣ ϵ H /2 v ∣ − 12 − 2 V 2 ∣ ϵ H /2 v ∣ − 6 ) ϵ = 0 ≤ 4 V 2 ∣ v ∣ − 6 − 4∣ v ∣ − 12 + max ( 0 , 28∣ v ∣ − 12 − 16 V 2 ∣ v ∣ − 6 ) .
in the metric convention of Lemma A.2. At V = 1 V=1 V = 1 , Φ 1 = E ( ⋅ , 6 ) − 2 E ( ⋅ , 3 ) = 2 E \Phi_1=E(\mathord\cdot,6)-2E(\mathord\cdot,3)=2\mathcal{E} Φ 1 = E ( ⋅ , 6 ) − 2 E ( ⋅ , 3 ) = 2 E . The guard removes every omitted correction; convex interpolation therefore gives the scalar bound. The endpoint hulls and the mean-value identity give (A.4) and the two exclusions. Select v v v at the midpoint and round to 2 − 20 Z 2^{-20}\mathbb{Z} 2 − 20 Z before both complete quadratic-form checks. Use the same witness Λ 0 \Lambda_0 Λ 0 , complete whole-box tails and outward arithmetic. This completes the proof of Lemma 5.2. □ \square □
Lemma 5.3 (A compact moment domain for the transition interval). Let f f f be continuous on [ 21 / 20 , 11 / 10 ] [21/20,11/10] [ 21/20 , 11/10 ] , concave in V 2 V^2 V 2 , and suppose
− 5 21 f + ′ ( 21 / 20 ) = E ( F , 3 ) , f ( 21 / 20 ) + 441 200 E ( F , 3 ) = E ( F , 6 ) , -\frac{5}{21}f_+'(21/20)=E(F,3),\qquad f(21/20)+\frac{441}{200}E(F,3)=E(F,6), − 21 5 f + ′ ( 21/20 ) = E ( F , 3 ) , f ( 21/20 ) + 200 441 E ( F , 3 ) = E ( F , 6 ) , − 5 22 f − ′ ( 11 / 10 ) = E ( H Y 1 , 3 ) , f ( 11 / 10 ) + 121 50 E ( H Y 1 , 3 ) = E ( H Y 1 , 6 ) , -\frac{5}{22}f_-'(11/10)=E(H_{Y_1},3),\qquad f(11/10)+\frac{121}{50}E(H_{Y_1},3)=E(H_{Y_1},6), − 22 5 f − ′ ( 11/10 ) = E ( H Y 1 , 3 ) , f ( 11/10 ) + 50 121 E ( H Y 1 , 3 ) = E ( H Y 1 , 6 ) , where H Y 1 H_{Y_1} H Y 1 is the unit-covolume S H SH S H minimizer shape at V = 11 / 10 V=11/10 V = 11/10 . Assume the endpoint comparisons
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) ( ∣ Λ ∣ = 1 , V ∈ { 21 / 20 , 11 / 10 } ) . E(\Lambda,6)-2V^2E(\Lambda,3)\geq f(V)\qquad(|\Lambda|=1,\ V\in\{21/20,11/10\}). E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) ( ∣Λ∣ = 1 , V ∈ { 21/20 , 11/10 }) . For ∣ Λ ∣ = 1 |\Lambda|=1 ∣Λ∣ = 1 ,
min 21 / 20 ≤ V ≤ 11 / 10 { E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) − f ( V ) } < 0 \min_{21/20\leq V\leq11/10}\{E(\Lambda,6)-2V^2E(\Lambda,3)-f(V)\}<0 21/20 ≤ V ≤ 11/10 min { E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) − f ( V )} < 0 ⟹ E ( F , 3 ) < E ( Λ , 3 ) < 81 10 , E ( Λ , 6 ) < 5 , λ 1 ( Λ ) > ( 2 5 ) 1 / 12 > 463 500 . \Longrightarrow E(F,3)<E(\Lambda,3)<\frac{81}{10},\qquad E(\Lambda,6)<5,\qquad\lambda_1(\Lambda)>\left(\frac{2}{5}\right)^{1/12}>\frac{463}{500}. ⟹ E ( F , 3 ) < E ( Λ , 3 ) < 10 81 , E ( Λ , 6 ) < 5 , λ 1 ( Λ ) > ( 5 2 ) 1/12 > 500 463 . If the comparison holds for all unit-covolume lattices and all 21 / 20 ≤ V ≤ 11 / 10 21/20\leq V\leq11/10 21/20 ≤ V ≤ 11/10 , then
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = f ( V ) , 21 / 20 < V < 11 / 10 E(\Lambda,6)-2V^2E(\Lambda,3)=f(V),\qquad21/20<V<11/10 E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = f ( V ) , 21/20 < V < 11/10 ⟹ E ( F , 3 ) ≤ E ( Λ , 3 ) < 81 / 10 , E ( Λ , 6 ) < 5 , λ 1 ( Λ ) > ( 2 / 5 ) 1 / 12 > 463 / 500. \Longrightarrow E(F,3)\leq E(\Lambda,3)<81/10,\qquad E(\Lambda,6)<5,\qquad\lambda_1(\Lambda)>(2/5)^{1/12}>463/500. ⟹ E ( F , 3 ) ≤ E ( Λ , 3 ) < 81/10 , E ( Λ , 6 ) < 5 , λ 1 ( Λ ) > ( 2/5 ) 1/12 > 463/500. Every lattice in either case has a Korkine–Zolotarev representative with a δ h = 1 a\delta h=1 a δ h = 1 and
463 500 < a ≤ 2 1 / 6 < 1122463 10 6 , 801 1000 < δ < 559 500 , 7 8 < h < 27 20 , \frac{463}{500}<a\leq2^{1/6}<\frac{1122463}{10^6},\qquad\frac{801}{1000}<\delta<\frac{559}{500},\qquad\frac{7}{8}<h<\frac{27}{20}, 500 463 < a ≤ 2 1/6 < 1 0 6 1122463 , 1000 801 < δ < 500 559 , 8 7 < h < 20 27 , 0 ≤ x , z ≤ 1 2 , − 1 2 ≤ y ≤ 1 2 . 0\leq x,z\leq\frac{1}{2},\qquad-\frac{1}{2}\leq y\leq\frac{1}{2}. 0 ≤ x , z ≤ 2 1 , − 2 1 ≤ y ≤ 2 1 .
Proof. Let L 0 L_0 L 0 be the covolume 11 / 10 11/10 11/10 reference in Lemma 4.9. In its normalized-linear chart, ∥ S ∗ ∥ F ≤ 3 / 10 9 \lVert S_*\rVert_F\leq3/10^9 ∥ S ∗ ∥ F ≤ 3/1 0 9 at the stationary lattice, and
E ( L 0 , 3 ) < 3332 500 , E ( L 0 , 6 ) < 339 100 . E(L_0,3)<\frac{3332}{500},\qquad E(L_0,6)<\frac{339}{100}. E ( L 0 , 3 ) < 500 3332 , E ( L 0 , 6 ) < 100 339 .
For ∥ S ∥ F ≤ 3 / 10 9 < 1 \lVert S\rVert_F\leq3/10^9<1 ∥ S ∥ F ≤ 3/1 0 9 < 1 and v ∈ L 0 v\in L_0 v ∈ L 0 ,
det ( I + S ) − 1 / 3 ∣ ( I + S ) v ∣ ≥ 1 − 3 / 10 9 1 + 3 / 10 9 ∣ v ∣ 2 . \det(I+S)^{-1/3}\lvert(I+S)v\rvert\geq\frac{1-3/10^9}{1+3/10^9}\lvert v\rvert^2. det ( I + S ) − 1/3 ∣( I + S ) v ∣ ≥ 1 + 3/1 0 9 1 − 3/1 0 9 ∣ v ∣ 2 .
Hence
E ( det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 L 0 , s ) ≤ ( 1 + 3 / 10 9 1 − 3 / 10 9 ) s E ( L 0 , s ) , s = 3 , 6. E\left(\det(I+S)^{-1/6}(I+S)^{1/2}L_{0,s}\right) \le\left(\frac{1+3/10^{9}}{1-3/10^{9}}\right)^{s}E(L_{0},s), \qquad s=3,6. E ( det ( I + S ) − 1/6 ( I + S ) 1/2 L 0 , s ) ≤ ( 1 − 3/1 0 9 1 + 3/1 0 9 ) s E ( L 0 , s ) , s = 3 , 6.
Since
E ( H Y 1 , s ) = ( 11 / 10 ) 2 s / 3 E ( det ( I + S ∗ ) − 1 / 6 ( I + S ∗ ) 1 / 2 L 0 , s ) , E(H_{Y_{1}},s)=(11/10)^{2s/3}E\left(\det(I+S_{*})^{-1/6}(I+S_{*})^{1/2}L_{0,s}\right), E ( H Y 1 , s ) = ( 11/10 ) 2 s /3 E ( det ( I + S ∗ ) − 1/6 ( I + S ∗ ) 1/2 L 0 , s ) ,
b 0 = E ( F , 3 ) , b 1 = E ( H Y 1 , 3 ) < 81 / 10 , A 1 = E ( H Y 1 , 6 ) < 5. b_{0}=E(F,3), \qquad b_{1}=E(H_{Y_{1}},3)<81/10, \qquad A_{1}=E(H_{Y_{1}},6)<5. b 0 = E ( F , 3 ) , b 1 = E ( H Y 1 , 3 ) < 81/10 , A 1 = E ( H Y 1 , 6 ) < 5.
Apply Proposition 5.1. After the global comparison, an interior equality is a zero minimum and satisfies
b 0 ≤ E ( Λ , 3 ) ≤ b 1 < 81 / 10 , E ( Λ , 6 ) ≤ A 1 < 5. b_{0}\le E(\Lambda,3)\le b_{1}<81/10,\qquad E(\Lambda,6)\le A_{1}<5. b 0 ≤ E ( Λ , 3 ) ≤ b 1 < 81/10 , E ( Λ , 6 ) ≤ A 1 < 5.
For v 0 ∈ Λ ∖ { 0 } v_{0}\in\Lambda\setminus\{0\} v 0 ∈ Λ ∖ { 0 } with ∣ v 0 ∣ = λ 1 ( Λ ) |v_{0}|=\lambda_{1}(\Lambda) ∣ v 0 ∣ = λ 1 ( Λ ) ,
∣ v 0 ∣ − 12 + ∣ − v 0 ∣ − 12 = 2 λ 1 ( Λ ) − 12 ≤ E ( Λ , 6 ) < 5 ⟹ λ 1 ( Λ ) > ( 2 / 5 ) 1 / 12 > 463 / 500. |v_{0}|^{-12}+|-v_{0}|^{-12}=2\lambda_{1}(\Lambda)^{-12}\le E(\Lambda,6)<5 \quad\Longrightarrow\quad\lambda_{1}(\Lambda)>(2/5)^{1/12}>463/500. ∣ v 0 ∣ − 12 + ∣ − v 0 ∣ − 12 = 2 λ 1 ( Λ ) − 12 ≤ E ( Λ , 6 ) < 5 ⟹ λ 1 ( Λ ) > ( 2/5 ) 1/12 > 463/500.
The Korkine–Zolotarev conditions give
δ ≥ 3 2 a > 801 1000 , δ 2 ≤ 2 3 a < ( 559 500 ) 2 , \delta\ge\frac{\sqrt{3}}{2a}>\frac{801}{1000},\qquad\delta^{2}\le\frac{2}{\sqrt{3}a}<\left(\frac{559}{500}\right)^{2}, δ ≥ 2 a 3 > 1000 801 , δ 2 ≤ 3 a 2 < ( 500 559 ) 2 ,
h 2 ≥ 3 2 a > ( 7 8 ) 2 , h ≤ 2 3 a 2 < 27 20 . h^{2}\ge\frac{\sqrt{3}}{2a}>\left(\frac{7}{8}\right)^{2},\qquad h\le\frac{2}{\sqrt{3}a^{2}}<\frac{27}{20}. h 2 ≥ 2 a 3 > ( 8 7 ) 2 , h ≤ 3 a 2 2 < 20 27 .
This completes the proof of Lemma 5.3. □
Lemma 5.4 (A uniform FCC basin). Let L F = ( 441 / 1600 ) 1 / 6 D 3 L_{F}=(441/1600)^{1/6}D_{3} L F = ( 441/1600 ) 1/6 D 3 , so ∣ L F ∣ = 21 / 20 |L_{F}|=21/20 ∣ L F ∣ = 21/20 . For trace-free symmetric S S S in (2.5), put
L F ( S ) = det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 L F , Λ F ( S ) = ( 20 / 21 ) 1 / 3 L F ( S ) , L_{F}(S)=\det(I+S)^{-1/6}(I+S)^{1/2}L_{F},\qquad\Lambda_{F}(S)=(20/21)^{1/3}L_{F}(S), L F ( S ) = det ( I + S ) − 1/6 ( I + S ) 1/2 L F , Λ F ( S ) = ( 20/21 ) 1/3 L F ( S ) , F V ( S ) = 1 2 E ( L F ( S ) , 6 ) − ( 20 V / 21 ) 2 E ( L F ( S ) , 3 ) . F_{V}(S)=\frac{1}{2}E(L_{F}(S),6)-(20V/21)^{2}E(L_{F}(S),3). F V ( S ) = 2 1 E ( L F ( S ) , 6 ) − ( 20 V /21 ) 2 E ( L F ( S ) , 3 ) . For every V > 0 V>0 V > 0 satisfying
220499 220500 ≤ ( 20 V / 21 ) 2 ≤ 458 441 , \frac{220499}{220500}\le(20V/21)^{2}\le\frac{458}{441}, 220500 220499 ≤ ( 20 V /21 ) 2 ≤ 441 458 , and every 0 < ∥ S ∥ F ≤ 1 / 40 0<\lVert S\rVert_{F}\le1/40 0 < ∥ S ∥ F ≤ 1/40 ,
S T ∇ S F V ( S ) > 1 1000 ∥ S ∥ F 2 , S^{\mathsf{T}}\nabla_{S}F_{V}(S)>\frac{1}{1000}\lVert S\rVert_{F}^{2}, S T ∇ S F V ( S ) > 1000 1 ∥ S ∥ F 2 , ∇ S F V ( 0 ) = 0 , F V ( S ) ≥ F V ( 0 ) + 1 2000 ∥ S ∥ F 2 ( ∥ S ∥ F ≤ 1 / 40 ) . \nabla_{S}F_{V}(0)=0,\qquad F_{V}(S)\ge F_{V}(0)+\frac{1}{2000}\lVert S\rVert_{F}^{2}\quad(\lVert S\rVert_{F}\le1/40). ∇ S F V ( 0 ) = 0 , F V ( S ) ≥ F V ( 0 ) + 2000 1 ∥ S ∥ F 2 (∥ S ∥ F ≤ 1/40 ) . On ∥ S ∥ F ≤ 1 / 40 \lVert S\rVert_{F}\le1/40 ∥ S ∥ F ≤ 1/40 ,
∇ S F V ( S ) = 0 ⟺ S = 0 , argmin ∥ S ∥ F ≤ 1 / 40 F V = { 0 } . \nabla_{S}F_{V}(S)=0\Longleftrightarrow S=0,\qquad\underset{\lVert S\rVert_{F}\le1/40}{\operatorname{argmin}}\ F_{V}=\{0\}. ∇ S F V ( S ) = 0 ⟺ S = 0 , ∥ S ∥ F ≤ 1/40 argmin F V = { 0 } . For the unit-covolume shape Λ F ( S ) \Lambda_{F}(S) Λ F ( S ) ,
Φ V ( Λ F ( S ) ) = E ( Λ F ( S ) , 6 ) − 2 V 2 E ( Λ F ( S ) , 3 ) \Phi_{V}(\Lambda_{F}(S))=E(\Lambda_{F}(S),6)-2V^{2}E(\Lambda_{F}(S),3) Φ V ( Λ F ( S )) = E ( Λ F ( S ) , 6 ) − 2 V 2 E ( Λ F ( S ) , 3 ) = 2 ( 21 / 20 ) 4 F V ( S ) , =2(21/20)^{4}F_{V}(S), = 2 ( 21/20 ) 4 F V ( S ) , E ( ( 20 V / 21 ) 1 / 3 L F ( S ) ) = E ( Λ F ( S ) , 6 ) 2 V 4 − E ( Λ F ( S ) , 3 ) V 2 \mathcal{E}((20V/21)^{1/3}L_{F}(S))=\frac{E(\Lambda_{F}(S),6)}{2V^{4}}-\frac{E(\Lambda_{F}(S),3)}{V^{2}} E (( 20 V /21 ) 1/3 L F ( S )) = 2 V 4 E ( Λ F ( S ) , 6 ) − V 2 E ( Λ F ( S ) , 3 ) = ( 21 / ( 20 V ) ) 4 F V ( S ) . =(21/(20V))^{4}F_{V}(S). = ( 21/ ( 20 V ) ) 4 F V ( S ) . ( 20 V / 21 ) 2 ∈ [ 220499 / 220500 , 458 / 441 ] ⟺ 2 V 2 ∈ [ 220499 / 100000 , 229 / 100 ] . (20V/21)^{2}\in[220499/220500,458/441]\Longleftrightarrow2V^{2}\in[220499/100000,229/100]. ( 20 V /21 ) 2 ∈ [ 220499/220500 , 458/441 ] ⟺ 2 V 2 ∈ [ 220499/100000 , 229/100 ] . Both factors are positive and independent of S S S .
Proof. Use the complete spherical sums, binomial polynomials, determinant, and moment tails in Lemma 4.7 at covolume 21 / 20 21/20 21/20 . Cubic symmetry gives ∇ s F V ( 0 ) = 0 \nabla_{\!s}F_V(0)=0 ∇ s F V ( 0 ) = 0 for every V V V . With the finite vector set F \mathcal{F} F of Lemma 4.7 and the same q w , ℓ w q_w,\ell_w q w , ℓ w , put
P V ( S ) = 1 2 det ( I + S ) 2 ∑ w ∈ F q w − 6 p 6 ( ℓ w ( S ) ) − ( 20 V 21 ) 2 det ( I + S ) ∑ w ∈ F q w − 3 p 3 ( ℓ w ( S ) ) . P_V(S)=\frac{1}{2}\det(I+S)^2\sum_{w\in\mathcal{F}}q_w^{-6}p_6(\ell_w(S))
-\left(\frac{20V}{21}\right)^2\det(I+S)\sum_{w\in\mathcal{F}}q_w^{-3}p_3(\ell_w(S)). P V ( S ) = 2 1 det ( I + S ) 2 w ∈ F ∑ q w − 6 p 6 ( ℓ w ( S )) − ( 21 20 V ) 2 det ( I + S ) w ∈ F ∑ q w − 3 p 3 ( ℓ w ( S )) .
Affinity in V 2 V^2 V 2 gives uniformity on 220499 / 220500 ≤ ( 20 V / 21 ) 2 ≤ 458 / 441 220499/220500\leq(20V/21)^2\leq458/441 220499/220500 ≤ ( 20 V /21 ) 2 ≤ 458/441 . If P V = ∑ d = 0 12 P V , d P_V=\sum_{d=0}^{12}P_{V,d} P V = ∑ d = 0 12 P V , d , define
( B V ( S ) ) i j = ∑ d = 2 12 1 d − 1 ∂ 2 P V , d ( S ) ∂ s i ∂ s j , 1 ≤ i , j ≤ 5 , s = ( u , v , x , y , z ) T . (B_V(S))_{ij}=\sum_{d=2}^{12}\frac{1}{d-1}\frac{\partial^2P_{V,d}(S)}{\partial s_i\partial s_j},\qquad1\leq i,j\leq5,\quad\mathbf{s}=(u,v,x,y,z)^{\mathsf{T}}. ( B V ( S ) ) ij = d = 2 ∑ 12 d − 1 1 ∂ s i ∂ s j ∂ 2 P V , d ( S ) , 1 ≤ i , j ≤ 5 , s = ( u , v , x , y , z ) T .
For all ∥ S ∥ F ≤ 1 / 40 \lVert S\rVert_F\leq1/40 ∥ S ∥ F ≤ 1/40 , the rational coefficient sums and the explicit weighted monomial maxima give
η T B V ( S ) η ≥ ∣ η ∣ T C ∣ η ∣ , \eta^{\mathsf{T}}B_V(S)\eta\geq\lvert\eta\rvert^{\mathsf{T}}C\lvert\eta\rvert, η T B V ( S ) η ≥ ∣ η ∣ T C ∣ η ∣ ,
where
C = 1 10 6 ( 3859306 − 75427 − 183283 − 183283 − 362275 − 75427 1284695 − 184271 − 184271 − 1747 − 183283 − 184271 188706 − 36951 − 36973 − 183283 − 184271 − 36951 188706 − 36973 − 362275 − 1747 − 36973 − 36973 264882 ) . C=\frac{1}{10^6}
\begin{pmatrix}
3859306 & -75427 & -183283 & -183283 & -362275\\
-75427 & 1284695 & -184271 & -184271 & -1747\\
-183283 & -184271 & 188706 & -36951 & -36973\\
-183283 & -184271 & -36951 & 188706 & -36973\\
-362275 & -1747 & -36973 & -36973 & 264882
\end{pmatrix}. C = 1 0 6 1 3859306 − 75427 − 183283 − 183283 − 362275 − 75427 1284695 − 184271 − 184271 − 1747 − 183283 − 184271 188706 − 36951 − 36973 − 183283 − 184271 − 36951 188706 − 36973 − 362275 − 1747 − 36973 − 36973 264882 .
The total Hessian remainder is at most ( 458 / 441 ) ε ( 1 / 40 ) < 33 / 10000 (458/441)\varepsilon(1/40)<33/10000 ( 458/441 ) ε ( 1/40 ) < 33/10000 , since 458 / 441 ≥ 1 458/441\geq1 458/441 ≥ 1 also majorizes the repulsive contribution.
The L D L T LDL^{\mathsf{T}} L D L T pivots of C − ( 43 / 10000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C-(43/10000)\operatorname{diag}(6,2,2,2,2) C − ( 43/10000 ) diag ( 6 , 2 , 2 , 2 , 2 ) exceed, respectively,
38 10 , 12 10 , 12 100 , 10 100 , 13 100 . \frac{38}{10},\quad\frac{12}{10},\quad\frac{12}{100},\quad\frac{10}{100},\quad\frac{13}{100}. 10 38 , 10 12 , 100 12 , 100 10 , 100 13 .
Euler’s identity and the complete remainder give, for 0 < ∥ S ∥ F ≤ 1 / 40 0<\lVert S\rVert_F\leq1/40 0 < ∥ S ∥ F ≤ 1/40 ,
s T ∇ s F V ( S ) > ∥ S ∥ F 2 1000 , F V ( S ) − F V ( 0 ) = ∫ 0 1 s T ∇ s F V ( t S ) d t ≥ ∥ S ∥ F 2 2000 . \mathbf{s}^{\mathsf{T}}\nabla_{\!s}F_V(S)>\frac{\lVert S\rVert_F^2}{1000},\qquad
F_V(S)-F_V(0)=\int_0^1\mathbf{s}^{\mathsf{T}}\nabla_{\!s}F_V(tS)\,dt\geq\frac{\lVert S\rVert_F^2}{2000}. s T ∇ s F V ( S ) > 1000 ∥ S ∥ F 2 , F V ( S ) − F V ( 0 ) = ∫ 0 1 s T ∇ s F V ( tS ) d t ≥ 2000 ∥ S ∥ F 2 .
Homogeneity gives the scaling identities. This completes the proof of Lemma 5.4. □
Lemma 5.5 (A tetragonal stationary basin across the transition window). In the common chart of Lemma 2.9 , set
Λ ( S ) = det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 L 149 / 500 . \Lambda(S)=\det(I+S)^{-1/6}(I+S)^{1/2}L_{149/500}. Λ ( S ) = det ( I + S ) − 1/6 ( I + S ) 1/2 L 149/500 . Set Φ V ( S ) = E ( Λ ( S ) , 6 ) − 2 V 2 E ( Λ ( S ) , 3 ) \Phi_V(S)=E(\Lambda(S),6)-2V^2E(\Lambda(S),3) Φ V ( S ) = E ( Λ ( S ) , 6 ) − 2 V 2 E ( Λ ( S ) , 3 ) . For every 2267 / 1000 ≤ 2 V 2 ≤ 579 / 250 2267/1000\leq2V^2\leq579/250 2267/1000 ≤ 2 V 2 ≤ 579/250 , the closed ball ∥ S ∥ F ≤ 19 / 1000 \lVert S\rVert_F\leq19/1000 ∥ S ∥ F ≤ 19/1000 is strictly convex for Φ V \Phi_V Φ V , with
∂ ϵ 2 Φ V ( S + ϵ T ) ∂ ϵ 2 ∣ ϵ = 0 > 1 400 ∥ T ∥ F 2 ( T ≠ 0 , T = T T , tr T = 0 ) . \left.\frac{\partial^2_\epsilon\Phi_V(S+\epsilon T)}{\partial\epsilon^2}\right|_{\epsilon=0}>\frac{1}{400}\lVert T\rVert_F^2\qquad(T\ne0,\ T=T^{\mathsf{T}},\ \operatorname{tr}T=0). ∂ ϵ 2 ∂ ϵ 2 Φ V ( S + ϵ T ) ϵ = 0 > 400 1 ∥ T ∥ F 2 ( T = 0 , T = T T , tr T = 0 ) . There is a real-analytic u V u_V u V , ∣ u V ∣ < 1 / 200 \lvert u_V\rvert<1/200 ∣ u V ∣ < 1/200 , such that, on this ball,
S V = diag ( 2 u V , − u V , − u V ) , ∇ s Φ V ( S ) = 0 ⟺ S = S V , S_V=\operatorname{diag}(2u_V,-u_V,-u_V),\qquad\nabla_{\!s}\Phi_V(S)=0\Longleftrightarrow S=S_V, S V = diag ( 2 u V , − u V , − u V ) , ∇ s Φ V ( S ) = 0 ⟺ S = S V , argmin ∥ S ∥ F ≤ 19 / 1000 Φ V = { S V } . \underset{\lVert S\rVert_F\leq19/1000}{\operatorname{argmin}}\ \Phi_V=\{S_V\}. ∥ S ∥ F ≤ 19/1000 argmin Φ V = { S V } .
Proof. Set L 0 = ( 43 / 40 ) 1 / 3 L 149 / 500 L_0=(43/40)^{1/3}L_{149/500} L 0 = ( 43/40 ) 1/3 L 149/500 in the polynomial and complete remainder construction of Lemma 2.9 . For i , j , k ∈ Z i,j,k\in\mathbb{Z} i , j , k ∈ Z , let F \mathcal{F} F consist of
w = ( ( 43 / 40 ) 2 4 ( 149 / 500 ) ) 1 / 6 ( 149 / 500 i , j , k ) , ∣ i ∣ ≤ 60 , ∣ j ∣ , ∣ k ∣ ≤ 32 , w=\left(\frac{(43/40)^2}{4(149/500)}\right)^{1/6}\left(\sqrt{149/500}\,i,j,k\right),\qquad\lvert i\rvert\leq60,\quad\lvert j\rvert,\lvert k\rvert\leq32, w = ( 4 ( 149/500 ) ( 43/40 ) 2 ) 1/6 ( 149/500 i , j , k ) , ∣ i ∣ ≤ 60 , ∣ j ∣ , ∣ k ∣ ≤ 32 ,
i + j + k ∈ 2 Z , 0 < 3 i 2 + 10 j 2 + 10 k 2 ≤ 10800. i+j+k\in2\mathbb{Z},\qquad0<3i^2+10j^2+10k^2\leq10800. i + j + k ∈ 2 Z , 0 < 3 i 2 + 10 j 2 + 10 k 2 ≤ 10800.
For this set,
q w = { ( ( 43 / 40 ) 2 4 ( 149 / 500 ) ) 1 / 6 } 2 ( ( 149 / 500 ) i 2 + j 2 + k 2 ) q_w=\left\{\left(\frac{(43/40)^2}{4(149/500)}\right)^{1/6}\right\}^2\left((149/500)i^2+j^2+k^2\right) q w = { ( 4 ( 149/500 ) ( 43/40 ) 2 ) 1/6 } 2 ( ( 149/500 ) i 2 + j 2 + k 2 )
ℓ w = u { 2 ( 149 / 500 ) i 2 − j 2 − k 2 } + v ( j 2 − k 2 ) ( 149 / 500 ) i 2 + j 2 + k 2 + 2 x 149 / 500 i j + 2 y 149 / 500 i k + 2 z j k ( 149 / 500 ) i 2 + j 2 + k 2 . \ell_w=\frac{u\{2(149/500)i^2-j^2-k^2\}+v(j^2-k^2)}{(149/500)i^2+j^2+k^2}
+\frac{2x\sqrt{149/500}\,ij+2y\sqrt{149/500}\,ik+2zjk}{(149/500)i^2+j^2+k^2}. ℓ w = ( 149/500 ) i 2 + j 2 + k 2 u { 2 ( 149/500 ) i 2 − j 2 − k 2 } + v ( j 2 − k 2 ) + ( 149/500 ) i 2 + j 2 + k 2 2 x 149/500 ij + 2 y 149/500 ik + 2 z j k .
λ 1 ( L 0 ) = min w ∈ L 0 ∖ { 0 } ∣ w ∣ > 1 , min w ∈ L 0 ∖ ( F ∪ { 0 } ) ∣ w ∣ > 32. \lambda_1(L_0)=\min_{w\in L_0\setminus\{0\}}|w|>1,\qquad\min_{w\in L_0\setminus(F\cup\{0\})}|w|>32. λ 1 ( L 0 ) = w ∈ L 0 ∖ { 0 } min ∣ w ∣ > 1 , w ∈ L 0 ∖ ( F ∪ { 0 }) min ∣ w ∣ > 32.
Use (18) with a = 1 a=1 a = 1 , R = 32 R=32 R = 32 and
E + ( L 0 , 3 ) < 643 / 100 , E + ( L 0 , 6 ) < 261 / 100. E^+(L_0,3)<643/100,\qquad E^+(L_0,6)<261/100. E + ( L 0 , 3 ) < 643/100 , E + ( L 0 , 6 ) < 261/100.
For L 0 ( S ) = ( 43 / 40 ) 1 / 3 Λ ( S ) L_0(S)=(43/40)^{1/3}\Lambda(S) L 0 ( S ) = ( 43/40 ) 1/3 Λ ( S ) ,
F V ( S ) = 1 2 E ( L 0 ( S ) , 6 ) − ( 40 V / 43 ) 2 E ( L 0 ( S ) , 3 ) , F_V(S)=\frac{1}{2}E(L_0(S),6)-(40V/43)^2E(L_0(S),3), F V ( S ) = 2 1 E ( L 0 ( S ) , 6 ) − ( 40 V /43 ) 2 E ( L 0 ( S ) , 3 ) ,
9068 / 9245 ≤ ( 40 V / 43 ) 2 ≤ 9264 / 9245 , Φ V ( S ) = 2 ( 43 / 40 ) 4 F V ( S ) . 9068/9245\leq(40V/43)^2\leq9264/9245,\qquad\Phi_V(S)=2(43/40)^4F_V(S). 9068/9245 ≤ ( 40 V /43 ) 2 ≤ 9264/9245 , Φ V ( S ) = 2 ( 43/40 ) 4 F V ( S ) .
The Hessian polynomial is affine in V 2 V^2 V 2 ; its complete remainder is < 3 / 1000 <3/1000 < 3/1000 on ∥ S ∥ F ≤ 19 / 1000 \lVert S\rVert_F\leq19/1000 ∥ S ∥ F ≤ 19/1000 . At the two endpoints its rational comparison matrices are
C − = 1 10 6 ( 10970754 − 388661 − 183919 − 183919 − 51920 − 388661 663661 − 369193 − 369193 − 2037 − 183919 − 369193 1176945 − 28014 − 25522 − 183919 − 369193 − 28014 1176945 − 25522 − 51920 − 2037 − 25522 − 25522 119928 ) , C_-=\frac{1}{10^6}
\begin{pmatrix}
10970754 & -388661 & -183919 & -183919 & -51920\\
-388661 & 663661 & -369193 & -369193 & -2037\\
-183919 & -369193 & 1176945 & -28014 & -25522\\
-183919 & -369193 & -28014 & 1176945 & -25522\\
-51920 & -2037 & -25522 & -25522 & 119928
\end{pmatrix}, C − = 1 0 6 1 10970754 − 388661 − 183919 − 183919 − 51920 − 388661 663661 − 369193 − 369193 − 2037 − 183919 − 369193 1176945 − 28014 − 25522 − 183919 − 369193 − 28014 1176945 − 25522 − 51920 − 2037 − 25522 − 25522 119928 ,
C + = 1 10 6 ( 10232230 − 380114 − 173000 − 173000 − 56122 − 380114 319921 − 353808 − 353808 − 1906 − 173000 − 353808 1057728 − 29813 − 27355 − 173000 − 353808 − 29813 1057728 − 27355 − 56122 − 1906 − 27355 − 27355 201534 ) . C_+=\frac{1}{10^6}
\begin{pmatrix}
10232230 & -380114 & -173000 & -173000 & -56122\\
-380114 & 319921 & -353808 & -353808 & -1906\\
-173000 & -353808 & 1057728 & -29813 & -27355\\
-173000 & -353808 & -29813 & 1057728 & -27355\\
-56122 & -1906 & -27355 & -27355 & 201534
\end{pmatrix}. C + = 1 0 6 1 10232230 − 380114 − 173000 − 173000 − 56122 − 380114 319921 − 353808 − 353808 − 1906 − 173000 − 353808 1057728 − 29813 − 27355 − 173000 − 353808 − 29813 1057728 − 27355 − 56122 − 1906 − 27355 − 27355 201534 .
The matrices C ± − ( 4 / 1000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_\pm-(4/1000)\operatorname{diag}(6,2,2,2,2) C ± − ( 4/1000 ) diag ( 6 , 2 , 2 , 2 , 2 ) are positive definite. Their exact L D L T LDL^{\mathsf T} L D L T pivots exceed, respectively,
( 10 , 3 / 5 , 9 / 10 , 4 / 5 , 1 / 10 ) , ( 10 , 29 / 100 , 3 / 5 , 1 / 4 , 17 / 100 ) . (10,3/5,9/10,4/5,1/10),\qquad(10,29/100,3/5,1/4,17/100). ( 10 , 3/5 , 9/10 , 4/5 , 1/10 ) , ( 10 , 29/100 , 3/5 , 1/4 , 17/100 ) .
The weighted monomial rule of Lemma 2.9 proves these finite bounds. Affinity in V 2 V^2 V 2 and the common remainder bound give
∂ ϵ 2 F V ( S + ϵ T ) ∣ ϵ = 0 > ( 1 / 1000 ) ∥ T ∥ F 2 \left.\partial_\epsilon^2F_V(S+\epsilon T)\right|_{\epsilon=0}>(1/1000)\lVert T\rVert_F^2 ∂ ϵ 2 F V ( S + ϵ T ) ϵ = 0 > ( 1/1000 ) ∥ T ∥ F 2
on the whole ball. Since 2 ( 43 / 40 ) 4 > 5 / 2 2(43/40)^4>5/2 2 ( 43/40 ) 4 > 5/2 , the displayed normalized bound follows.
The complete attraction derivative and Lemma A.3 give
∣ ∂ u E ( L 0 ( S ) , 3 ) ∣ S = 0 ∣ < 261 100 , \left|\left.\partial_uE(L_0(S),3)\right|_{S=0}\right|<\frac{261}{100}, ∣ ∂ u E ( L 0 ( S ) , 3 ) ∣ S = 0 ∣ < 100 261 ,
∣ ∂ u F 43 / 40 ( 0 ) ∣ < 8 ⋅ 10 − 6 , ∣ ∂ u F V ( 0 ) ∣ < 1 20 . |\partial_uF_{43/40}(0)|<8\cdot10^{-6},\qquad|\partial_uF_V(0)|<\frac{1}{20}. ∣ ∂ u F 43/40 ( 0 ) ∣ < 8 ⋅ 1 0 − 6 , ∣ ∂ u F V ( 0 ) ∣ < 20 1 .
On the ball, ∂ u u F V > 10 \partial_{uu}F_V>10 ∂ uu F V > 10 , and
∂ u F V ( − 1 / 200 , 0 , 0 , 0 , 0 ) < 0 < ∂ u F V ( 1 / 200 , 0 , 0 , 0 , 0 ) . \partial_uF_V(-1/200,0,0,0,0)<0<\partial_uF_V(1/200,0,0,0,0). ∂ u F V ( − 1/200 , 0 , 0 , 0 , 0 ) < 0 < ∂ u F V ( 1/200 , 0 , 0 , 0 , 0 ) .
Tetragonal symmetry gives
( ∂ v F V , ∂ x F V , ∂ y F V , ∂ z F V ) ( u , 0 , 0 , 0 , 0 ) = 0. (\partial_vF_V,\partial_xF_V,\partial_yF_V,\partial_zF_V)(u,0,0,0,0)=0. ( ∂ v F V , ∂ x F V , ∂ y F V , ∂ z F V ) ( u , 0 , 0 , 0 , 0 ) = 0.
Thus
∃ ! u V ∈ ( − 1 / 200 , 1 / 200 ) : ∂ u F V ( u V , 0 , 0 , 0 , 0 ) = 0 , ∇ s F V ( S V ) = 0. \exists! u_V\in(-1/200,1/200):\partial_uF_V(u_V,0,0,0,0)=0,\qquad\nabla_sF_V(S_V)=0. ∃ ! u V ∈ ( − 1/200 , 1/200 ) : ∂ u F V ( u V , 0 , 0 , 0 , 0 ) = 0 , ∇ s F V ( S V ) = 0.
Strict convexity gives uniqueness; ∂ u u F V ( S V ) > 0 \partial_{uu}F_V(S_V)>0 ∂ uu F V ( S V ) > 0 gives analyticity by the implicit-function theorem. This completes the proof of Lemma 5.5 . □
Lemma 5.6 (A uniform simple-hexagonal basin on a broad density interval). Put
Y 0 = 107818724 10 8 , Λ 0 = ( 4 3 Y 0 ) 1 / 6 span Z { ( Y 0 , 0 , 0 ) , ( 0 , 1 , 0 ) , ( 0 , 1 / 2 , 3 / 2 ) } . Y_0=\frac{107818724}{10^8},\qquad\Lambda_0=\left(\frac{4}{3Y_0}\right)^{1/6}\operatorname{span}_{\mathbb{Z}}\{(\sqrt{Y_0},0,0),(0,1,0),(0,1/2,\sqrt{3}/2)\}. Y 0 = 1 0 8 107818724 , Λ 0 = ( 3 Y 0 4 ) 1/6 span Z {( Y 0 , 0 , 0 ) , ( 0 , 1 , 0 ) , ( 0 , 1/2 , 3 /2 )} . Thus ∣ Λ 0 ∣ = 1 \lvert\Lambda_0\rvert=1 ∣ Λ 0 ∣ = 1 . In (19) , set
Λ ( S ) = det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 Λ 0 , \Lambda(S)=\det(I+S)^{-1/6}(I+S)^{1/2}\Lambda_0, Λ ( S ) = det ( I + S ) − 1/6 ( I + S ) 1/2 Λ 0 , Φ V ( S ) = E ( Λ ( S ) , 6 ) − 2 V 2 E ( Λ ( S ) , 3 ) . \Phi_V(S)=E(\Lambda(S),6)-2V^2E(\Lambda(S),3). Φ V ( S ) = E ( Λ ( S ) , 6 ) − 2 V 2 E ( Λ ( S ) , 3 ) . For every
231499 100000 ≤ 2 V 2 ≤ 242001 100000 , \frac{231499}{100000}\le2V^2\le\frac{242001}{100000}, 100000 231499 ≤ 2 V 2 ≤ 100000 242001 , there is a real-analytic u V u_V u V with
S V = diag ( 2 u V , − u V , − u V ) , ∣ u V ∣ < 1 125 , Y ( V ) = Y 0 1 + 2 u V 1 − u V . S_V=\operatorname{diag}(2u_V,-u_V,-u_V),\qquad\lvert u_V\rvert<\frac{1}{125},\qquad Y(V)=Y_0\frac{1+2u_V}{1-u_V}. S V = diag ( 2 u V , − u V , − u V ) , ∣ u V ∣ < 125 1 , Y ( V ) = Y 0 1 − u V 1 + 2 u V . For ∥ S ∥ F ≤ 1 / 50 \lVert S\rVert_F\le1/50 ∥ S ∥ F ≤ 1/50 ,
∇ S Φ V ( S ) = 0 ⟺ S = S V , argmin ∥ S ∥ F ≤ 1 / 50 Φ V = { S V } , \nabla_S\Phi_V(S)=0\Longleftrightarrow S=S_V,\qquad\underset{\lVert S\rVert_F\le1/50}{\operatorname{argmin}}\ \Phi_V=\{S_V\}, ∇ S Φ V ( S ) = 0 ⟺ S = S V , ∥ S ∥ F ≤ 1/50 argmin Φ V = { S V } , and
∂ ϵ 2 Φ V ( S + ϵ T ) ∣ ϵ = 0 > 1 400 ∥ T ∥ F 2 ( T ≠ 0 ) , \left.\partial_\epsilon^2\Phi_V(S+\epsilon T)\right|_{\epsilon=0}>\frac{1}{400}\lVert T\rVert_F^2\qquad(T\ne0), ∂ ϵ 2 Φ V ( S + ϵ T ) ϵ = 0 > 400 1 ∥ T ∥ F 2 ( T = 0 ) , Φ V ( S ) ≥ Φ V ( S V ) + 1 800 ∥ S − S V ∥ F 2 . \Phi_V(S)\ge\Phi_V(S_V)+\frac{1}{800}\lVert S-S_V\rVert_F^2. Φ V ( S ) ≥ Φ V ( S V ) + 800 1 ∥ S − S V ∥ F 2 . Proof. Use L 0 = ( 87 / 80 ) 1 / 3 Λ 0 L_0=(87/80)^{1/3}\Lambda_0 L 0 = ( 87/80 ) 1/3 Λ 0 . At the endpoint covolumes V − = 231499 / 200000 V_-=\sqrt{231499/200000} V − = 231499/200000 and V + = 242001 / 200000 V_+=\sqrt{242001/200000} V + = 242001/200000 . For L 0 ( S ) = ( 87 / 80 ) 1 / 3 Λ ( S ) L_0(S)=(87/80)^{1/3}\Lambda(S) L 0 ( S ) = ( 87/80 ) 1/3 Λ ( S ) ,
E ( L 0 ( S ) , s ) = det ( I + S ) s / 3 ∑ w ∈ L 0 ∖ { 0 } ( ∣ w ∣ 2 + w T S w ) − s . E(L_0(S),s)=\det(I+S)^{s/3}\sum_{w\in L_0\setminus\{0\}}(\lvert w\rvert^2+w^{\mathsf T}Sw)^{-s}. E ( L 0 ( S ) , s ) = det ( I + S ) s /3 w ∈ L 0 ∖ { 0 } ∑ (∣ w ∣ 2 + w T S w ) − s . Hence
F V ( S ) = 1 2 E ( L 0 ( S ) , 6 ) − ( 80 V 87 ) 2 E ( L 0 ( S ) , 3 ) , F_V(S)=\frac{1}{2}E(L_0(S),6)-\left(\frac{80V}{87}\right)^2E(L_0(S),3), F V ( S ) = 2 1 E ( L 0 ( S ) , 6 ) − ( 87 80 V ) 2 E ( L 0 ( S ) , 3 ) , Φ V ( S ) = 2 ( 87 / 80 ) 4 F V ( S ) , 2 ( 87 / 80 ) 4 > 5 2 . \Phi_V(S)=2(87/80)^4F_V(S),\qquad2(87/80)^4>\frac{5}{2}. Φ V ( S ) = 2 ( 87/80 ) 4 F V ( S ) , 2 ( 87/80 ) 4 > 2 5 . With Y 0 = 26954681 / 25000000 Y_0=26954681/25000000 Y 0 = 26954681/25000000 , let F ⊂ L 0 \mathcal{F}\subset L_0 F ⊂ L 0 consist of the vectors
w = ( 4 ( 87 / 80 ) 2 3 Y 0 ) 1 / 6 ( Y 0 k , p / 2 , 3 j / 2 ) , p , j , k ∈ Z , p + j ∈ 2 Z , w=\left(\frac{4(87/80)^2}{3Y_0}\right)^{1/6}(\sqrt{Y_0}k,p/2,\sqrt{3}j/2),\qquad p,j,k\in\mathbb{Z},\quad p+j\in2\mathbb{Z}, w = ( 3 Y 0 4 ( 87/80 ) 2 ) 1/6 ( Y 0 k , p /2 , 3 j /2 ) , p , j , k ∈ Z , p + j ∈ 2 Z , ∣ p ∣ ≤ 69 , ∣ j ∣ ≤ 40 , ∣ k ∣ ≤ 33 , \lvert p\rvert\le69,\qquad\lvert j\rvert\le40,\qquad\lvert k\rvert\le33, ∣ p ∣ ≤ 69 , ∣ j ∣ ≤ 40 , ∣ k ∣ ≤ 33 , 0 < 25000000 ( p 2 + 3 j 2 ) + 107818724 k 2 ≤ 120000000000. 0<25000000(p^2+3j^2)+107818724k^2\le120000000000. 0 < 25000000 ( p 2 + 3 j 2 ) + 107818724 k 2 ≤ 120000000000. # ( F ∩ [ 0 , ∞ ) 3 ) = 25471. \#(\mathcal{F}\cap[0,\infty)^3)=25471. # ( F ∩ [ 0 , ∞ ) 3 ) = 25471. The set retains coordinate signs and planar hexagonal symmetry. For each vector put
q w = { ( 4 ( 87 / 80 ) 2 3 Y 0 ) 1 / 6 } 2 4 ( p 2 + 3 j 2 + 4 Y 0 k 2 ) , q_w=\frac{\left\{\left(\frac{4(87/80)^2}{3Y_0}\right)^{1/6}\right\}^2}{4}(p^2+3j^2+4Y_0k^2), q w = 4 { ( 3 Y 0 4 ( 87/80 ) 2 ) 1/6 } 2 ( p 2 + 3 j 2 + 4 Y 0 k 2 ) , ℓ w ( S ) = u ( 8 Y 0 k 2 − p 2 − 3 j 2 ) + v ( p 2 − 3 j 2 ) + 4 x Y 0 k p + 4 y 3 Y 0 k j + 2 z 3 p j p 2 + 3 j 2 + 4 Y 0 k 2 . \ell_w(S)=\frac{u(8Y_0k^2-p^2-3j^2)+v(p^2-3j^2)+4x\sqrt{Y_0}kp+4y\sqrt{3Y_0}kj+2z\sqrt{3}pj}{p^2+3j^2+4Y_0k^2}. ℓ w ( S ) = p 2 + 3 j 2 + 4 Y 0 k 2 u ( 8 Y 0 k 2 − p 2 − 3 j 2 ) + v ( p 2 − 3 j 2 ) + 4 x Y 0 k p + 4 y 3 Y 0 k j + 2 z 3 p j . With the degree-six binomial polynomials p 3 , p 6 p_3,p_6 p 3 , p 6 of Lemma 2.9 , the polynomial to be checked is
P V ( S ) = 1 2 det ( I + S ) 2 ∑ w ∈ F q w − 6 p 6 ( ℓ w ( S ) ) P_V(S)=\frac{1}{2}\det(I+S)^2\sum_{w\in\mathcal{F}}q_w^{-6}p_6(\ell_w(S)) P V ( S ) = 2 1 det ( I + S ) 2 w ∈ F ∑ q w − 6 p 6 ( ℓ w ( S )) P V ( S ) − ( 80 V 87 ) 2 det ( I + S ) ∑ w ∈ F q w − 3 p 3 ( ℓ w ( S ) ) . \phantom{P_V(S)}-\left(\frac{80V}{87}\right)^2\det(I+S)\sum_{w\in\mathcal{F}}q_w^{-3}p_3(\ell_w(S)). P V ( S ) − ( 87 80 V ) 2 det ( I + S ) w ∈ F ∑ q w − 3 p 3 ( ℓ w ( S )) .
Retain the determinant factors and all 138 parity-allowed positions, including odd powers of v v v . Complete sign sums give
P V ∈ Q [ V 2 , u , v , x , y , z ] , degs P V ≤ 12 , degs ∇ S 2 P V ≤ 10. P_V \in\mathbb{Q}[V^2,u,v,x,y,z], \qquad\operatorname{degs} P_V \le12, \qquad\operatorname{degs} \nabla_S^2 P_V \le10. P V ∈ Q [ V 2 , u , v , x , y , z ] , degs P V ≤ 12 , degs ∇ S 2 P V ≤ 10.
Since
λ 1 ( L 0 ) = min w ∈ L 0 ∖ { 0 } ∣ w ∣ > 1 , min w ∈ L 0 ∖ ( F ∪ { 0 } ) ∣ w ∣ > 32 , \lambda_1(L_0)=\min_{w\in L_0\setminus\{0\}} |w|>1,\qquad\min_{w\in L_0\setminus(\mathcal{F}\cup\{0\})}|w|>32, λ 1 ( L 0 ) = w ∈ L 0 ∖ { 0 } min ∣ w ∣ > 1 , w ∈ L 0 ∖ ( F ∪ { 0 }) min ∣ w ∣ > 32 ,
(18) gives
∑ w ∈ L 0 ∖ ( F ∪ { 0 } ) q w − s ≤ T s ( 1 , 32 ) , s = 3 , 6. \sum_{w\in L_0\setminus(\mathcal{F}\cup\{0\})}q_w^{-s}\le T_s(1,32),\qquad s=3,6. w ∈ L 0 ∖ ( F ∪ { 0 }) ∑ q w − s ≤ T s ( 1 , 32 ) , s = 3 , 6.
Finite rational summation gives
∑ w ∈ F q w − 3 + T 3 ( 1 , 32 ) < 6783 1000 , ∑ w ∈ F q w − 6 + T 6 ( 1 , 32 ) < 3477 1000 . \sum_{w\in\mathcal{F}}q_w^{-3}+T_3(1,32)<\frac{6783}{1000},\qquad\sum_{w\in\mathcal{F}}q_w^{-6}+T_6(1,32)<\frac{3477}{1000}. w ∈ F ∑ q w − 3 + T 3 ( 1 , 32 ) < 1000 6783 , w ∈ F ∑ q w − 6 + T 6 ( 1 , 32 ) < 1000 3477 .
Apply the complete binomial and spatial Hessian remainder formulas of Lemma 2.9 , with these moment bounds and with the s = 3 s=3 s = 3 contributions multiplied by 107556 / 105125 107556/105125 107556/105125 . They give
∣ ∂ ϵ 2 { F V ( S + ϵ T ) − P V ( S + ϵ T ) } ∣ ϵ = 0 ∣ < 151 50000 ∥ T ∥ F 2 ( ∥ S ∥ F ≤ 1 / 50 ) . \left|\left.\partial_\epsilon^2\{F_V(S+\epsilon T)-P_V(S+\epsilon T)\}\right|_{\epsilon=0}\right|<\frac{151}{50000}\lVert T\rVert_F^2\qquad(\lVert S\rVert_F\le1/50). ∂ ϵ 2 { F V ( S + ϵ T ) − P V ( S + ϵ T )} ϵ = 0 < 50000 151 ∥ T ∥ F 2 (∥ S ∥ F ≤ 1/50 ) .
Use Lemma 2.9 with weights ( 6 , 2 , 2 , 2 , 2 ) (6,2,2,2,2) ( 6 , 2 , 2 , 2 , 2 ) and coordinate order ( u , v , x , y , z ) (u,v,x,y,z) ( u , v , x , y , z ) . Before addition, round every rational summand outwards to 2 − 80 Z 2^{-80}Z 2 − 80 Z and each monomial upper bound upwards to 2 − 80 Z 2^{-80}Z 2 − 80 Z by integer-square-root enclosure of its rational square. All-even monomials retain their nonnegative sign. Then
∂ ϵ 2 P V ± ( S + ϵ T ) ∣ ϵ = 0 ≥ ∣ η ∣ T C ± ∣ η ∣ , \left.\partial_\epsilon^2P_{V_\pm}(S+\epsilon T)\right|_{\epsilon=0}\geq|\eta|^{\mathsf T}C_\pm|\eta|, ∂ ϵ 2 P V ± ( S + ϵ T ) ϵ = 0 ≥ ∣ η ∣ T C ± ∣ η ∣ ,
where η \eta η is the coordinate vector of T T T and
C − = 1 10 6 ( 17656592 − 1723052 − 55398 − 55392 − 1720776 − 1723052 6304087 − 24382 − 23238 − 888593 − 55398 − 24382 32520 − 17272 − 23164 − 55392 − 23238 − 17272 32623 − 24247 − 1720776 − 888593 − 23164 − 24247 6305113 ) , C_-=\frac{1}{10^6}
\begin{pmatrix}
17656592 & -1723052 & -55398 & -55392 & -1720776\\
-1723052 & 6304087 & -24382 & -23238 & -888593\\
-55398 & -24382 & 32520 & -17272 & -23164\\
-55392 & -23238 & -17272 & 32623 & -24247\\
-1720776 & -888593 & -23164 & -24247 & 6305113
\end{pmatrix}, C − = 1 0 6 1 17656592 − 1723052 − 55398 − 55392 − 1720776 − 1723052 6304087 − 24382 − 23238 − 888593 − 55398 − 24382 32520 − 17272 − 23164 − 55392 − 23238 − 17272 32623 − 24247 − 1720776 − 888593 − 23164 − 24247 6305113 ,
C + = 1 10 6 ( 14612274 − 1662571 − 47365 − 47360 − 1660327 − 1662571 5713988 − 21547 − 20329 − 848744 − 47365 − 21547 325877 − 14687 − 20255 − 47360 − 20329 − 14687 325981 − 21409 − 1660327 − 848744 − 20255 − 21409 5715003 ) . C_+=\frac{1}{10^6}
\begin{pmatrix}
14612274 & -1662571 & -47365 & -47360 & -1660327\\
-1662571 & 5713988 & -21547 & -20329 & -848744\\
-47365 & -21547 & 325877 & -14687 & -20255\\
-47360 & -20329 & -14687 & 325981 & -21409\\
-1660327 & -848744 & -20255 & -21409 & 5715003
\end{pmatrix}. C + = 1 0 6 1 14612274 − 1662571 − 47365 − 47360 − 1660327 − 1662571 5713988 − 21547 − 20329 − 848744 − 47365 − 21547 325877 − 14687 − 20255 − 47360 − 20329 − 14687 325981 − 21409 − 1660327 − 848744 − 20255 − 21409 5715003 .
Exact rational elimination gives L D L T LDL^{\mathsf T} L D L T pivots of C ± − ( 201 / 50000 ) diag ( 6 , 2 , 2 , 2 , 2 ) C_\pm-(201/50000)\operatorname{diag}(6,2,2,2,2) C ± − ( 201/50000 ) diag ( 6 , 2 , 2 , 2 , 2 ) exceeding, respectively,
( 17 , 6 , 1 / 50 , 1 / 100 , 5 ) , ( 14 , 5 , 3 / 10 , 3 / 10 , 5 ) . (17,6,1/50,1/100,5),\qquad(14,5,3/10,3/10,5). ( 17 , 6 , 1/50 , 1/100 , 5 ) , ( 14 , 5 , 3/10 , 3/10 , 5 ) .
Affinity in V 2 V^2 V 2 extends
∂ ϵ 2 F V ± ( S + ϵ T ) ∣ ϵ = 0 > ∥ T ∥ F 2 / 1000 \left.\partial_\epsilon^2F_{V_\pm}(S+\epsilon T)\right|_{\epsilon=0}>\lVert T\rVert_F^2/1000 ∂ ϵ 2 F V ± ( S + ϵ T ) ϵ = 0 > ∥ T ∥ F 2 /1000
to V ∈ [ V − , V + ] V\in[V_-,V_+] V ∈ [ V − , V + ] , for T ≠ 0 T\ne0 T = 0 and ∥ S ∥ F ≤ 1 / 50 \lVert S\rVert_F\le1/50 ∥ S ∥ F ≤ 1/50 . The first diagonal entries give
∂ u 2 F V ( S ) > 14 ( ∥ S ∥ F ≤ 1 / 50 ) . \partial_u^2F_V(S)>14\qquad(\lVert S\rVert_F\le1/50). ∂ u 2 F V ( S ) > 14 (∥ S ∥ F ≤ 1/50 ) .
For the axial root, use the finite N = 6 N=6 N = 6 Ewald derivative sum in Lemma A.6 , substituting the explicit rational covolume 87 / 80 87/80 87/80 and Y 0 = 107818724 / 10 8 Y_0=107818724/10^8 Y 0 = 107818724/1 0 8 above. Its complete Gaussian derivative tails and finite interval evaluation give
11 10 9 < ∂ u F 87 / 80 ( 0 ) < 12 10 9 . \frac{11}{10^9}<\partial_uF_{87/80}(0)<\frac{12}{10^9}. 1 0 9 11 < ∂ u F 87/80 ( 0 ) < 1 0 9 12 .
The finite attraction derivative and its complete omitted-vector bound 6 T 3 ( 1 , 32 ) 6T_3(1,32) 6 T 3 ( 1 , 32 ) give
4739 1000 < ∂ u E ( L 0 ( S ) , 3 ) ∣ S = 0 < 4743 1000 < 19 4 . \frac{4739}{1000}<\left.\partial_uE(L_0(S),3)\right|_{S=0}<\frac{4743}{1000}<\frac{19}{4}. 1000 4739 < ∂ u E ( L 0 ( S ) , 3 ) ∣ S = 0 < 1000 4743 < 4 19 .
Hexagonal symmetry and the axial bounds give
( ∂ v F V , ∂ x F V , ∂ y F V , ∂ z F V ) ( u , 0 , 0 , 0 , 0 ) = 0 , ∣ ∂ u F V ( 0 ) ∣ < 11 100 , (\partial_v F_V,\partial_x F_V,\partial_y F_V,\partial_z F_V)(u,0,0,0,0)=0,\qquad|\partial_u F_V(0)|<\frac{11}{100}, ( ∂ v F V , ∂ x F V , ∂ y F V , ∂ z F V ) ( u , 0 , 0 , 0 , 0 ) = 0 , ∣ ∂ u F V ( 0 ) ∣ < 100 11 ,
and
∂ u F V ( 1 / 125 , 0 , 0 , 0 , 0 ) > 1 500 , ∂ u F V ( − 1 / 125 , 0 , 0 , 0 , 0 ) < − 1 500 , \partial_u F_V(1/125,0,0,0,0)>\frac{1}{500},\qquad\partial_u F_V(-1/125,0,0,0,0)<-\frac{1}{500}, ∂ u F V ( 1/125 , 0 , 0 , 0 , 0 ) > 500 1 , ∂ u F V ( − 1/125 , 0 , 0 , 0 , 0 ) < − 500 1 ,
at the two axial endpoints inside the ball. Hence
∃ ! u V ∈ ( − 1 / 125 , 1 / 125 ) : ∂ u F V ( u V , 0 , 0 , 0 , 0 ) = 0 , ∇ s F V ( S V ) = 0. \exists! \,u_V\in(-1/125,1/125):\partial_u F_V(u_V,0,0,0,0)=0,\qquad\nabla_s F_V(S_V)=0. ∃ ! u V ∈ ( − 1/125 , 1/125 ) : ∂ u F V ( u V , 0 , 0 , 0 , 0 ) = 0 , ∇ s F V ( S V ) = 0.
Strict convexity and ∂ u u F V ( S V ) > 14 \partial_{uu}F_V(S_V)>14 ∂ uu F V ( S V ) > 14 give uniqueness and analyticity. The factor 2 ( 87 / 80 ) 4 > 5 / 2 2(87/80)^4>5/2 2 ( 87/80 ) 4 > 5/2 gives the stated Hessian and value-gap bounds. This completes the proof of Lemma 5.6 .
Corollary 5.7 (Identification of the recognized stationary shapes). For 441 / 200 ≤ 2 V 2 ≤ 121 / 50 441/200\le2V^2\le121/50 441/200 ≤ 2 V 2 ≤ 121/50 , use the respective charts and hypotheses of Lemmas 5.4, 5.5 and 5.6 . Then
∇ s Φ V = 0 ⟹ Λ ≅ F , L X ( V ) , H Y ( V ) , \nabla_s\Phi_V=0\quad\Longrightarrow\quad\Lambda\cong F,\quad L_X(V),\quad H_Y(V), ∇ s Φ V = 0 ⟹ Λ ≅ F , L X ( V ) , H Y ( V ) , respectively. In the latter two cases,
149 500 66 67 < X < 149 500 202 199 , 107818724 10 8 41 42 < Y < 107818724 10 8 127 124 . \frac{149}{500}\frac{66}{67}<X<\frac{149}{500}\frac{202}{199},\qquad\frac{107818724}{10^8}\frac{41}{42}<Y<\frac{107818724}{10^8}\frac{127}{124}. 500 149 67 66 < X < 500 149 199 202 , 1 0 8 107818724 42 41 < Y < 1 0 8 107818724 124 127 . Proof. Since ∂ u { ( 1 + 2 u ) / ( 1 − u ) } = 3 / ( 1 − u ) 2 > 0 \partial_u\{(1+2u)/(1-u)\}=3/(1-u)^2>0 ∂ u {( 1 + 2 u ) / ( 1 − u )} = 3/ ( 1 − u ) 2 > 0 , the formulas X = X 0 ( 1 + 2 u ) / ( 1 − u ) X=X_0(1+2u)/(1-u) X = X 0 ( 1 + 2 u ) / ( 1 − u ) , Y = Y 0 ( 1 + 2 u ) / ( 1 − u ) Y=Y_0(1+2u)/(1-u) Y = Y 0 ( 1 + 2 u ) / ( 1 − u ) and ∣ u ∣ < 1 / 200 |u|<1/200 ∣ u ∣ < 1/200 , ∣ u ∣ < 1 / 125 |u|<1/125 ∣ u ∣ < 1/125 give the stated intervals, contained respectively in ( 23 / 100 , 307 / 1000 ) (23/100,307/1000) ( 23/100 , 307/1000 ) and ( 103 / 100 , 103 / 50 ) (103/100,103/50) ( 103/100 , 103/50 ) . Lemma A.8 gives exactly one stationary parameter there for each target V V V , identifying the classified branches. Cubic symmetry gives ∇ s Φ V ( Λ F ( S ) ) ∣ S = 0 = 0 \nabla_s\Phi_V(\Lambda_F(S))|_{S=0}=0 ∇ s Φ V ( Λ F ( S )) ∣ S = 0 = 0 for every V V V . This completes the proof of Corollary 5.7.
Theorem 5.8 (The complete transition strip). For 21 / 20 ≤ V ≤ 11 / 10 21/20\le V\le11/10 21/20 ≤ V ≤ 11/10 , every global minimizer is one of the three stationary candidates in Proposition 2.7 . Their ordering, contact covolumes and equality cases are therefore the ones stated there.
Proof. Let f ( V ) f(V) f ( V ) be the least candidate value in Proposition 2.7 . Branch stationarity and the moment ordering make f f f concave in V V V , including both contacts. The endpoint comparisons are Theorems 4.11 and 4.12 , and Lemma 5.3 gives the compact moment domain. Lemma A.13 yields
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) , ∣ Λ ∣ = 1 , 21 20 ≤ V ≤ 11 10 , E(\Lambda,6)-2V^2E(\Lambda,3)\ge f(V),\qquad|\Lambda|=1,\qquad\frac{21}{20}\le V\le\frac{11}{10}, E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) , ∣Λ∣ = 1 , 20 21 ≤ V ≤ 10 11 , with equality precisely for a stationary candidate attaining f f f . Since 2 V 4 E ( V 1 / 3 Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) 2V^4E(V^{1/3}\Lambda)=E(\Lambda,6)-2V^2E(\Lambda,3) 2 V 4 E ( V 1/3 Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) , Proposition 2.7 gives the classification, all equality cases and the two derivative jumps. This completes the proof of Theorem 5.8 .
The intervening simple-hexagonal phase
For 11 / 10 ≤ V ≤ 89 / 50 11/10\le V\le89/50 11/10 ≤ V ≤ 89/50 , moment contraction, Fourier convexity and layer alignment give
argmin ∣ Λ ∣ = 1 { E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) } = { Λ : Λ ≅ H Y ( V ) } . \underset{|\Lambda|=1}{\operatorname{argmin}}\{E(\Lambda,6)-2V^2E(\Lambda,3)\}=\{\Lambda:\Lambda\cong H_Y(V)\}. ∣Λ∣ = 1 argmin { E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 )} = { Λ : Λ ≅ H Y ( V )} .
The curve Y ( V ) Y(V) Y ( V ) is defined in Lemma 6.2 .
Lemma 6.1 (Area coordinates and complete derivative tails). For t , r > 0 t,r>0 t , r > 0 and x , y , z ∈ R x,y,z\in\mathbb{R} x , y , z ∈ R , set
η = ( t , r , x , y , z ) , p ( η ) = ( a , δ , x , y , z ) = ( t r , t / r , x , y , z ) , Y = 3 2 t 3 , \eta=(t,r,x,y,z),\qquad p(\eta)=(a,\delta,x,y,z)=\left(\sqrt{tr},\sqrt{t/r},x,y,z\right),\qquad Y=\frac{\sqrt{3}}{2t^3}, η = ( t , r , x , y , z ) , p ( η ) = ( a , δ , x , y , z ) = ( t r , t / r , x , y , z ) , Y = 2 t 3 3 , U ( η ) = U ( p ( η ) ) = ( t r t r x t r y 0 t / r t / r z 0 0 t − 1 ) , det U = 1. (48) U(\eta)=U(p(\eta))=
\begin{pmatrix}
\sqrt{tr} & \sqrt{tr}\,x & \sqrt{tr}\,y\\
0 & \sqrt{t/r} & \sqrt{t/r}\,z\\
0 & 0 & t^{-1}
\end{pmatrix},
\qquad\det U=1. \tag*{(48)} U ( η ) = U ( p ( η )) = t r 0 0 t r x t / r 0 t r y t / r z t − 1 , det U = 1. (48) :::`
For n = ( i , j , k ) ∈ Z 3 n=(i,j,k)\in\mathbb{Z}^{3} n = ( i , j , k ) ∈ Z 3 , put q n = ∣ U ( η ) n ∣ 2 q_n=|U(\eta)n|^2 q n = ∣ U ( η ) n ∣ 2 and q n ∗ = ∣ U ( η ) − T n ∣ 2 q_n^*=|U(\eta)^{-T}n|^2 q n ∗ = ∣ U ( η ) − T n ∣ 2 . Then
q n = tr ( i + x j + y k ) 2 + t r ( j + z k ) 2 + k 2 t 2 , q n ∗ = i 2 t r + r t ( j − x i ) 2 + t 2 { k − z ( j − x i ) − y i } 2 . (49) \begin{aligned}
q_n&=\operatorname{tr}(i+xj+yk)^2+\frac{t}{r}(j+zk)^2+\frac{k^2}{t^2},\\
q_n^*&=\frac{i^2}{tr}+\frac{r}{t}(j-xi)^2+t^2\{k-z(j-xi)-yi\}^2. \tag*{(49)}
\end{aligned} q n q n ∗ = tr ( i + x j + y k ) 2 + r t ( j + z k ) 2 + t 2 k 2 , = t r i 2 + t r ( j − x i ) 2 + t 2 { k − z ( j − x i ) − y i } 2 . (49)
Put
A = i + x j + y k , B = j + z k , C = k 2 t 2 , D = j − x i , F = k − z D − y i . A=i+xj+yk,\qquad B=j+zk,\qquad C=\frac{k^2}{t^2},\qquad D=j-xi,\qquad F=k-zD-yi. A = i + x j + y k , B = j + z k , C = t 2 k 2 , D = j − x i , F = k − z D − y i .
In coordinate order ( t , r , x , y , z ) (t,r,x,y,z) ( t , r , x , y , z ) the gradients are
∇ η q n = ( ( r A 2 + B 2 / r ) − 2 C / t , t ( A 2 − B 2 / r 2 ) , 2 t r A j , 2 t r A k , 2 t B k / r ) , \nabla_\eta q_n=\left((rA^2+B^2/r)-2C/t,\ t(A^2-B^2/r^2),\ 2trAj,\ 2trAk,\ 2tBk/r\right), ∇ η q n = ( ( r A 2 + B 2 / r ) − 2 C / t , t ( A 2 − B 2 / r 2 ) , 2 t r A j , 2 t r A k , 2 tB k / r ) ,
∇ η q n ∗ = ( − ( i 2 / r + r D 2 ) / t 2 + 2 t F 2 , ( D 2 − i 2 / r 2 ) / t , − 2 r D i / t + 2 t 2 F z i , − 2 t 2 F i , − 2 t 2 F D ) . \nabla_\eta q_n^*=\left(-(i^2/r+rD^2)/t^2+2tF^2,\ (D^2-i^2/r^2)/t,\ -2rDi/t+2t^2Fzi,\ -2t^2Fi,\ -2t^2FD\right). ∇ η q n ∗ = ( − ( i 2 / r + r D 2 ) / t 2 + 2 t F 2 , ( D 2 − i 2 / r 2 ) / t , − 2 r D i / t + 2 t 2 F z i , − 2 t 2 F i , − 2 t 2 F D ) .
The Hessians, in the order ( t , r , x , y , z ) (t,r,x,y,z) ( t , r , x , y , z ) , are given by
∇ η 2 q n = ( 6 C / t 2 A 2 − B 2 / r 2 2 r A j 2 r A k 2 B k / r A 2 − B 2 / r 2 2 t B 2 / r 3 2 t A j 2 t A k − 2 t B k / r 2 2 r A j 2 t A j 2 t r j 2 2 t r j k 0 2 r A k 2 t A k 2 t r j k 2 t r k 2 0 2 B k / r − 2 t B k / r 2 0 0 2 t k 2 / r ) , \nabla_\eta^2q_n=
\begin{pmatrix}
6C/t^2 & A^2-B^2/r^2 & 2rAj & 2rAk & 2Bk/r\\
A^2-B^2/r^2 & 2tB^2/r^3 & 2tAj & 2tAk & -2tBk/r^2\\
2rAj & 2tAj & 2trj^2 & 2trjk & 0\\
2rAk & 2tAk & 2trjk & 2trk^2 & 0\\
2Bk/r & -2tBk/r^2 & 0 & 0 & 2tk^2/r
\end{pmatrix}, ∇ η 2 q n = 6 C / t 2 A 2 − B 2 / r 2 2 r A j 2 r A k 2 B k / r A 2 − B 2 / r 2 2 t B 2 / r 3 2 t A j 2 t A k − 2 tB k / r 2 2 r A j 2 t A j 2 t r j 2 2 t r j k 0 2 r A k 2 t A k 2 t r j k 2 t r k 2 0 2 B k / r − 2 tB k / r 2 0 0 2 t k 2 / r ,
∂ t t q n ∗ = 2 ( i 2 / r + r D 2 ) / t 3 + 2 F 2 , ∂ r r q n ∗ = 2 i 2 / ( t r 3 ) , ∂ x x q n ∗ = 2 r i 2 / t + 2 t 2 z 2 i 2 , ∂ y y q n ∗ = 2 t 2 i 2 , ∂ z z q n ∗ = 2 t 2 D 2 , ∂ t r q n ∗ = ( i 2 / r 2 − D 2 ) / t 2 , ∂ t x q n ∗ = 2 r D i / t 2 + 4 t F z i , ∂ t y q n ∗ = − 4 t F i , ∂ t z q n ∗ = − 4 t F D , ∂ r x q n ∗ = − 2 D i / t , ∂ r y q n ∗ = ∂ r z q n ∗ = 0 , ∂ x y q n ∗ = − 2 t 2 z i 2 , ∂ x z q n ∗ = 2 t 2 i ( F − z D ) , ∂ y z q n ∗ = 2 t 2 i D . (50) \begin{aligned}
\partial_{tt}q_n^*&=2(i^2/r+rD^2)/t^3+2F^2,&
\partial_{rr}q_n^*&=2i^2/(tr^3),\\
\partial_{xx}q_n^*&=2ri^2/t+2t^2z^2i^2,&
\partial_{yy}q_n^*&=2t^2i^2,\\
\partial_{zz}q_n^*&=2t^2D^2,&
\partial_{tr}q_n^*&=(i^2/r^2-D^2)/t^2,\\
\partial_{tx}q_n^*&=2rDi/t^2+4tFzi,&
\partial_{ty}q_n^*&=-4tFi,\\
\partial_{tz}q_n^*&=-4tFD,&
\partial_{rx}q_n^*&=-2Di/t,\\
\partial_{ry}q_n^*&=\partial_{rz}q_n^*=0,&
\partial_{xy}q_n^*&=-2t^2zi^2,\\
\partial_{xz}q_n^*&=2t^2i(F-zD),&
\partial_{yz}q_n^*&=2t^2iD. \tag*{(50)}
\end{aligned} ∂ tt q n ∗ ∂ xx q n ∗ ∂ z z q n ∗ ∂ t x q n ∗ ∂ t z q n ∗ ∂ r y q n ∗ ∂ x z q n ∗ = 2 ( i 2 / r + r D 2 ) / t 3 + 2 F 2 , = 2 r i 2 / t + 2 t 2 z 2 i 2 , = 2 t 2 D 2 , = 2 r D i / t 2 + 4 tF z i , = − 4 tF D , = ∂ r z q n ∗ = 0 , = 2 t 2 i ( F − z D ) , ∂ r r q n ∗ ∂ y y q n ∗ ∂ t r q n ∗ ∂ t y q n ∗ ∂ r x q n ∗ ∂ x y q n ∗ ∂ y z q n ∗ = 2 i 2 / ( t r 3 ) , = 2 t 2 i 2 , = ( i 2 / r 2 − D 2 ) / t 2 , = − 4 tF i , = − 2 D i / t , = − 2 t 2 z i 2 , = 2 t 2 i D . (50)
Complete ∇ η 2 q n ∗ \nabla_\eta^2q_n^* ∇ η 2 q n ∗ by symmetry.
Let B η B_\eta B η be any closed area-coordinate rectangle with t , r > 0 t,r>0 t , r > 0 , and let B p B_p B p contain p ( B η ) p(B_\eta) p ( B η ) . Here ( η 0 , η 1 , η 2 , η 3 , η 4 ) = ( t , r , x , y , z ) (\eta_0,\eta_1,\eta_2,\eta_3,\eta_4)=(t,r,x,y,z) ( η 0 , η 1 , η 2 , η 3 , η 4 ) = ( t , r , x , y , z ) and ( p 0 , p 1 , p 2 , p 3 , p 4 ) = ( a , δ , x , y , z ) (p_0,p_1,p_2,p_3,p_4)=(a,\delta,x,y,z) ( p 0 , p 1 , p 2 , p 3 , p 4 ) = ( a , δ , x , y , z ) . The nontrivial embedding derivatives are
a t = a 2 t , a r = a 2 r , δ t = δ 2 t , δ r = − δ 2 r , a_t=\frac{a}{2t},\qquad a_r=\frac{a}{2r},\qquad\delta_t=\frac{\delta}{2t},\qquad\delta_r=-\frac{\delta}{2r}, a t = 2 t a , a r = 2 r a , δ t = 2 t δ , δ r = − 2 r δ ,
a t t = − a 4 t 2 , a t r = a 4 t r , a r r = − a 4 r 2 , a_{tt}=-\frac{a}{4t^2},\qquad a_{tr}=\frac{a}{4tr},\qquad a_{rr}=-\frac{a}{4r^2}, a tt = − 4 t 2 a , a t r = 4 t r a , a r r = − 4 r 2 a ,
δ t t = − δ 4 t 2 , δ t r = − δ 4 t r , δ r r = 3 δ 4 r 2 . \delta_{tt}=-\frac{\delta}{4t^2},\qquad\delta_{tr}=-\frac{\delta}{4tr},\qquad\delta_{rr}=\frac{3\delta}{4r^2}. δ tt = − 4 t 2 δ , δ t r = − 4 t r δ , δ r r = 4 r 2 3 δ .
For k = 2 , 3 , 4 k=2,3,4 k = 2 , 3 , 4 , ∂ η i η j p k = 0 \partial_{\eta_i\eta_j}p_k=0 ∂ η i η j p k = 0 ( 0 ≤ i , j ≤ 4 ) (0\leq i,j\leq4) ( 0 ≤ i , j ≤ 4 ) . For any C 2 C^2 C 2 remainder R ( p ) R(p) R ( p ) with complete derivative bounds
∣ ∂ p k R ∣ ≤ e k , ∣ ∂ p k p l R ∣ ≤ e k l on B p , |\partial_{p_k}R|\leq e_k,\qquad|\partial_{p_kp_l}R|\leq e_{kl}\qquad\text{on }B_p, ∣ ∂ p k R ∣ ≤ e k , ∣ ∂ p k p l R ∣ ≤ e k l on B p ,
for 0 ≤ i , j ≤ 4 0\leq i,j\leq4 0 ≤ i , j ≤ 4 , the area-coordinate tail errors obey
∣ ∂ η i ( R ∘ p ) ∣ ≤ ∑ k = 0 4 sup B η ∣ ∂ η i p k ∣ e k , |\partial_{\eta_i}(R\circ p)|\leq\sum_{k=0}^{4}\sup_{B_\eta}|\partial_{\eta_i}p_k|e_k, ∣ ∂ η i ( R ∘ p ) ∣ ≤ k = 0 ∑ 4 B η sup ∣ ∂ η i p k ∣ e k ,
∣ ∂ η i η j ( R ∘ p ) ∣ ≤ ∑ k = 0 4 ∑ l = 0 4 sup B η ∣ ∂ η i p k ∣ e k l sup B η ∣ ∂ η j p l ∣ + ∑ k = 0 1 sup B η ∣ ∂ η i η j p k ∣ e k . (51) |\partial_{\eta_i\eta_j}(R\circ p)|\leq\sum_{k=0}^{4}\sum_{l=0}^{4}\sup_{B_\eta}|\partial_{\eta_i}p_k|e_{kl}\sup_{B_\eta}|\partial_{\eta_j}p_l|+\sum_{k=0}^{1}\sup_{B_\eta}|\partial_{\eta_i\eta_j}p_k|e_k. \tag*{(51)} ∣ ∂ η i η j ( R ∘ p ) ∣ ≤ k = 0 ∑ 4 l = 0 ∑ 4 B η sup ∣ ∂ η i p k ∣ e k l B η sup ∣ ∂ η j p l ∣ + k = 0 ∑ 1 B η sup ∣ ∂ η i η j p k ∣ e k . (51)
All suprema are over the whole rectangle; no Korkine–Zolotarev inequality is required.
Proof. Direct differentiation gives (L9)–(L10); with ( D p ) i j = ∂ p i / ∂ η j (D p)_{ij} = \partial p_i/\partial\eta_j ( D p ) ij = ∂ p i / ∂ η j ( 0 ≤ i , j ≤ 4 ) (0 \le i,j \le4) ( 0 ≤ i , j ≤ 4 ) , the chain rule gives
∇ η ( R ∘ p ) = ( D p ) T ∇ p R , \nabla_{\eta}(R \circ p) = (D p)^{\mathsf T}\nabla_p R, ∇ η ( R ∘ p ) = ( D p ) T ∇ p R ,
∇ η 2 ( R ∘ p ) = ( D p ) T ( ∇ p 2 R ) ( D p ) + ∑ k = 0 4 ( ∂ p k R ) ∇ η 2 p k . \nabla_{\eta}^{2}(R \circ p) = (D p)^{\mathsf T}(\nabla_p^{2}R)(D p) + \sum_{k=0}^{4}(\partial_{p_k}R)\nabla_{\eta}^{2}p_k. ∇ η 2 ( R ∘ p ) = ( D p ) T ( ∇ p 2 R ) ( D p ) + k = 0 ∑ 4 ( ∂ p k R ) ∇ η 2 p k .
and hence (L12). This completes the proof of Lemma 6.1. □
Lemma 6.2 (A monotone stationary simple-hexagonal reference curve). For the unit-covolume simple-hexagonal lattice H Y H_Y H Y , put
B ( Y ) = E ( H Y , 3 ) , A ( Y ) = E ( H Y , 6 ) , Γ ( Y ) = A t ( Y ) B t ( Y ) . B(Y) = E(H_Y,3), \qquad A(Y) = E(H_Y,6), \qquad\Gamma(Y) = \frac{A_t(Y)}{B_t(Y)}. B ( Y ) = E ( H Y , 3 ) , A ( Y ) = E ( H Y , 6 ) , Γ ( Y ) = B t ( Y ) A t ( Y ) . Here the subscripts denote logarithmic derivatives:
f t ( Y ) = Y f ′ ( Y ) , f t t ( Y ) = Y 2 f ′ ′ ( Y ) + Y f ′ ( Y ) ( f = A , B , Γ , q , q ∗ ) . f_t(Y) = Yf'(Y), \qquad f_{tt}(Y) = Y^2f''(Y) + Yf'(Y) \qquad(f=A,B,\Gamma,q,q^*). f t ( Y ) = Y f ′ ( Y ) , f tt ( Y ) = Y 2 f ′′ ( Y ) + Y f ′ ( Y ) ( f = A , B , Γ , q , q ∗ ) . On the entire closed interval 11 / 10 ≤ Y ≤ 6 11/10 \le Y \le6 11/10 ≤ Y ≤ 6 ,
2 < B t ( Y ) < 30 , 3 5 < Γ t ( Y ) < 17 2 , 2 < B_t(Y) < 30, \qquad\frac{3}{5} < \Gamma_t(Y) < \frac{17}{2}, 2 < B t ( Y ) < 30 , 5 3 < Γ t ( Y ) < 2 17 , 2417273 10 6 < Γ ( 11 / 10 ) < 2417274 10 6 , 5744686 10 6 < Γ ( 39 / 10 ) < 5744687 10 6 , \frac{2417273}{10^6} < \Gamma(11/10) < \frac{2417274}{10^6}, \qquad\frac{5744686}{10^6} < \Gamma(39/10) < \frac{5744687}{10^6}, 1 0 6 2417273 < Γ ( 11/10 ) < 1 0 6 2417274 , 1 0 6 5744686 < Γ ( 39/10 ) < 1 0 6 5744687 , 8630510 10 6 < Γ ( 6 ) < 8630511 10 6 . \frac{8630510}{10^6} < \Gamma(6) < \frac{8630511}{10^6}. 1 0 6 8630510 < Γ ( 6 ) < 1 0 6 8630511 . The inverse branch Y ( V ) = Γ − 1 ( 2 V 2 ) Y(V)=\Gamma^{-1}(2V^2) Y ( V ) = Γ − 1 ( 2 V 2 ) is real analytic for
Γ ( 11 / 10 ) ≤ 2 V 2 ≤ Γ ( 6 ) \Gamma(11/10) \le2V^2 \le\Gamma(6) Γ ( 11/10 ) ≤ 2 V 2 ≤ Γ ( 6 ) and satisfies
A t ( Y ( V ) ) = 2 V 2 B t ( Y ( V ) ) , d d V B ( Y ( V ) ) = 4 V B t ( Y ( V ) ) Γ t ( Y ( V ) ) > 16 V 17 , A_t(Y(V)) = 2V^2B_t(Y(V)), \qquad\frac{\mathrm d}{\mathrm dV}B(Y(V)) = 4V\frac{B_t(Y(V))}{\Gamma_t(Y(V))} > \frac{16V}{17}, A t ( Y ( V )) = 2 V 2 B t ( Y ( V )) , d V d B ( Y ( V )) = 4 V Γ t ( Y ( V )) B t ( Y ( V )) > 17 16 V , A t t − 2 V 2 B t t = B t Γ t > 6 5 . A_{tt}-2V^2B_{tt}=B_t\Gamma_t>\frac{6}{5}. A tt − 2 V 2 B tt = B t Γ t > 5 6 . For f ( V ) = A ( Y ( V ) ) − 2 V 2 B ( Y ( V ) ) f(V)=A(Y(V))-2V^2B(Y(V)) f ( V ) = A ( Y ( V )) − 2 V 2 B ( Y ( V )) ,
f ′ ( V ) = − 4 V B ( Y ( V ) ) , f ′ ′ ( V ) − f ′ ( V ) V = − 4 V d d V B ( Y ( V ) ) < 0. f'(V)=-4VB(Y(V)), \qquad f''(V)-\frac{f'(V)}{V}=-4V\frac{\mathrm d}{\mathrm dV}B(Y(V))<0. f ′ ( V ) = − 4 V B ( Y ( V )) , f ′′ ( V ) − V f ′ ( V ) = − 4 V d V d B ( Y ( V )) < 0. Thus f f f is strictly concave as a function of V 2 V^2 V 2 . For 11 / 10 ≤ V ≤ 89 / 50 11/10 \le V \le89/50 11/10 ≤ V ≤ 89/50 ,
Y ( V ) < 217 50 , B ( Y ( V ) ) < 43 2 , A ( Y ( V ) ) < 64. Y(V)<\frac{217}{50}, \qquad B(Y(V))<\frac{43}{2}, \qquad A(Y(V))<64. Y ( V ) < 50 217 , B ( Y ( V )) < 2 43 , A ( Y ( V )) < 64. At V ∈ { 11 / 10 , 89 / 50 } V\in\{11/10,89/50\} V ∈ { 11/10 , 89/50 } , Theorem 4.12 and Proposition 3.20 give
argmin ∣ L ∣ = V E ( L ) = { L : L ≅ V 1 / 3 H Y ( V ) } . \underset{|L|=V}{\operatorname{argmin}} E(L)=\{L:L\cong V^{1/3}H_{Y(V)}\}. ∣ L ∣ = V argmin E ( L ) = { L : L ≅ V 1/3 H Y ( V ) } .
Proof. Put c = ( 4 / ( 3 Y ) ) 1 / 3 c=(4/(3Y))^{1/3} c = ( 4/ ( 3 Y ) ) 1/3 . The Gram matrix of H Y H_Y H Y is
Q Y = c ( 1 1 / 2 0 1 / 2 1 0 0 0 Y ) . Q_Y=c\begin{pmatrix}1&1/2&0\\1/2&1&0\\0&0&Y\end{pmatrix}. Q Y = c 1 1/2 0 1/2 1 0 0 0 Y .
For 1 ≤ Y ≤ 13 / 2 1\le Y\le13/2 1 ≤ Y ≤ 13/2 , σ ( Q Y ) = { c / 2 , 3 c / 2 , c Y } ⊂ ( 1 / 4 , 4 ) \sigma(Q_Y)=\{c/2,3c/2,cY\}\subset(1/4,4) σ ( Q Y ) = { c /2 , 3 c /2 , c Y } ⊂ ( 1/4 , 4 ) . The same bounds hold for Q Y − 1 Q_Y^{-1} Q Y − 1 and justify termwise differentiation.
Use the kernels
P s ( q ) = π s Γ ( s ) G s ( π q ) , D s ( q ) = π s Γ ( s ) G 3 / 2 − s ( π q ) , G a ( x ) = ∫ 1 ∞ e − x u u a − 1 d u . P_s(q)=\frac{\pi^s}{\Gamma(s)}G_s(\pi q), \qquad D_s(q)=\frac{\pi^s}{\Gamma(s)}G_{3/2-s}(\pi q), \qquad G_a(x)=\int_1^\infty e^{-xu}u^{a-1}\,\mathrm du. P s ( q ) = Γ ( s ) π s G s ( π q ) , D s ( q ) = Γ ( s ) π s G 3/2 − s ( π q ) , G a ( x ) = ∫ 1 ∞ e − xu u a − 1 d u .
The unit-covolume version of the Ewald identity in Lemma 2.8 is
E ( H Y , s ) = π s Γ ( s ) ( 1 s − 3 / 2 − 1 s ) + ∑ n ∈ Z 3 ∖ { 0 } { P s ( n T Q Y n ) + D s ( n T Q Y − 1 n ) } . E(H_Y,s)=\frac{\pi^s}{\Gamma(s)}\left(\frac{1}{s-3/2}-\frac{1}{s}\right)+\sum_{n\in\mathbb{Z}^3\setminus\{0\}}\left\{P_s(n^{\mathsf{T}}Q_Yn)+D_s(n^{\mathsf{T}}Q_Y^{-1}n)\right\}. E ( H Y , s ) = Γ ( s ) π s ( s − 3/2 1 − s 1 ) + n ∈ Z 3 ∖ { 0 } ∑ { P s ( n T Q Y n ) + D s ( n T Q Y − 1 n ) } .
The constant terms are π 3 / 6 \pi^3/6 π 3 /6 and π 6 / 2160 \pi^6/2160 π 6 /2160 . For 0 ≤ a ≤ 432 0\leq a\leq432 0 ≤ a ≤ 432 , 0 ≤ b ≤ 144 0\leq b\leq144 0 ≤ b ≤ 144 , group the complete finite cube by
m a b = # { ( i , j , k ) ∈ [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } : i 2 + i j + j 2 = a , k 2 = b } . m_{ab}=\#\{(i,j,k)\in[-12,12]^3\cap\mathbb{Z}^3\setminus\{0\}:i^2+ij+j^2=a,\ k^2=b\}. m ab = # {( i , j , k ) ∈ [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } : i 2 + ij + j 2 = a , k 2 = b } .
Omit m a b = 0 m_{ab}=0 m ab = 0 ; ∑ a = 0 432 ∑ b = 0 144 m a b = 25 3 − 1 \sum_{a=0}^{432}\sum_{b=0}^{144}m_{ab}=25^3-1 ∑ a = 0 432 ∑ b = 0 144 m ab = 2 5 3 − 1 . For q = c a + c Y b q=ca+cYb q = c a + c Y b , q ∗ = 4 a / ( 3 c ) + b / ( c Y ) q^*=4a/(3c)+b/(cY) q ∗ = 4 a / ( 3 c ) + b / ( c Y ) ,
E f i n ( H Y , s ) = π s Γ ( s ) ( 1 s − 3 / 2 − 1 s ) + ∑ a = 0 m a b > 0 432 ∑ b = 0 144 m a b { P s ( q ) + D s ( q ∗ ) } . E^{\mathrm{fin}}(H_Y,s)=\frac{\pi^s}{\Gamma(s)}\left(\frac{1}{s-3/2}-\frac{1}{s}\right)+\sum_{\substack{a=0\\m_{ab}>0}}^{432}\sum_{b=0}^{144}m_{ab}\{P_s(q)+D_s(q^*)\}. E fin ( H Y , s ) = Γ ( s ) π s ( s − 3/2 1 − s 1 ) + a = 0 m ab > 0 ∑ 432 b = 0 ∑ 144 m ab { P s ( q ) + D s ( q ∗ )} .
The substitution j ↦ − j j\mapsto-j j ↦ − j preserves the cube and converts the dual form i 2 − i j + j 2 i^2-ij+j^2 i 2 − ij + j 2 to the same multiplicities. Logarithmic differentiation gives
q t = − c a 3 + 2 c Y b 3 , q t t = c a 9 + 4 c Y b 9 , q_t=-\frac{ca}{3}+\frac{2cYb}{3},\qquad q_{tt}=\frac{ca}{9}+\frac{4cYb}{9}, q t = − 3 c a + 3 2 c Y b , q tt = 9 c a + 9 4 c Y b ,
q t ∗ = 4 a 9 c − 2 b 3 c Y , q t t ∗ = 4 a 27 c + 4 b 9 c Y . q_t^*=\frac{4a}{9c}-\frac{2b}{3cY},\qquad q_{tt}^*=\frac{4a}{27c}+\frac{4b}{9cY}. q t ∗ = 9 c 4 a − 3 c Y 2 b , q tt ∗ = 27 c 4 a + 9 c Y 4 b .
For K = P s , D s K=P_s,D_s K = P s , D s and 0 ≤ r ≤ 2 0\leq r\leq2 0 ≤ r ≤ 2 ,
Y d d Y K ( q ) = K ′ ( q ) q t , ( Y d d Y ) 2 K ( q ) = K ′ ′ ( q ) q t 2 + K ′ ( q ) q t t , Y\frac{\mathrm{d}}{\mathrm{d}Y}K(q)=K'(q)q_t,\qquad\left(Y\frac{\mathrm{d}}{\mathrm{d}Y}\right)^2K(q)=K''(q)q_t^2+K'(q)q_{tt}, Y d Y d K ( q ) = K ′ ( q ) q t , ( Y d Y d ) 2 K ( q ) = K ′′ ( q ) q t 2 + K ′ ( q ) q tt ,
P s ( r ) ( q ) = ( − π ) r π s Γ ( s ) G s + r ( π q ) , D s ( r ) ( q ) = ( − π ) r π s Γ ( s ) G 3 / 2 − s + r ( π q ) . P_s^{(r)}(q)=\frac{(-\pi)^r\pi^s}{\Gamma(s)}G_{s+r}(\pi q),\qquad D_s^{(r)}(q)=\frac{(-\pi)^r\pi^s}{\Gamma(s)}G_{3/2-s+r}(\pi q). P s ( r ) ( q ) = Γ ( s ) ( − π ) r π s G s + r ( π q ) , D s ( r ) ( q ) = Γ ( s ) ( − π ) r π s G 3/2 − s + r ( π q ) .
For q ≥ 20 q\geq20 q ≥ 20 , Lemma A.2 gives ∣ K ( r ) ( q ) ∣ < 5 e − 3 q |K^{(r)}(q)|<5e^{-3q} ∣ K ( r ) ( q ) ∣ < 5 e − 3 q , 0 ≤ r ≤ 2 0\leq r\leq2 0 ≤ r ≤ 2 . Since ∣ q t ∣ ≤ 2 q / 3 |q_t|\leq2q/3 ∣ q t ∣ ≤ 2 q /3 , ∣ q t t ∣ ≤ 4 q / 9 |q_{tt}|\leq4q/9 ∣ q tt ∣ ≤ 4 q /9 , q e − q ≤ 1 qe^{-q}\leq1 q e − q ≤ 1 and q 2 e − q ≤ 4 q^2e^{-q}\leq4 q 2 e − q ≤ 4 , every composed derivative through order two is bounded by 20 e − 2 q 20e^{-2q} 20 e − 2 q . For omitted indices, ∣ n ∣ ≥ 13 |n|\geq13 ∣ n ∣ ≥ 13 , q , q ∗ ≥ 169 / 4 > 20 q,q^*\geq169/4>20 q , q ∗ ≥ 169/4 > 20 , and 2 q , 2 q ∗ > π ∣ n ∣ 2 / 8 2q,2q^*>\pi|n|^2/8 2 q , 2 q ∗ > π ∣ n ∣ 2 /8 . The complete primal-plus-dual remainder is
T = 240 e − 169 π / 8 ( 1 + 8 ) 2 1 − e − 27 π / 8 < 6 10 26 . T=240\frac{e^{-169\pi/8}(1+\sqrt{8})^2}{1-e^{-27\pi/8}}<\frac{6}{10^{26}}. T = 240 1 − e − 27 π /8 e − 169 π /8 ( 1 + 8 ) 2 < 1 0 26 6 .
Add [ 0 , T ] [0,T] [ 0 , T ] to each moment and [ − T , T ] [-T,T] [ − T , T ] to each of its first two derivatives.
Use the outward arithmetic, checked radicals, rational exponentials and half-integer recurrences of Lemmas 2.8 and A.2 . Enclose c = ( 4 / ( 3 Y ) ) 1 / 3 c=(4/(3Y))^{1/3} c = ( 4/ ( 3 Y ) ) 1/3 by outward cubing, extending the positive endpoints until both checks hold. Monotonicity of G a G_a G a encloses interval arguments by endpoint evaluation.
For q − ≥ 20 q_-\geq20 q − ≥ 20 , let v = max { b ∈ { 20 , 40 , 80 , 160 , 320 } : b ≤ q − } v=\max\{b\in\{20,40,80,160,320\}:b\leq q_-\} v = max { b ∈ { 20 , 40 , 80 , 160 , 320 } : b ≤ q − } and use
K ∈ [ 0 , 5 e − 3 v ] , K ′ ∈ [ − 5 e − 3 v , 0 ] , K ′ ′ ∈ [ 0 , 5 e − 3 v ] . K\in[0,5e^{-3v}],\qquad K'\in[-5e^{-3v},0],\qquad K''\in[0,5e^{-3v}]. K ∈ [ 0 , 5 e − 3 v ] , K ′ ∈ [ − 5 e − 3 v , 0 ] , K ′′ ∈ [ 0 , 5 e − 3 v ] .
Evaluate nonzero multiplicities in increasing ( a , b ) (a,b) ( a , b ) order, with outward additions and no Taylor interpolation.
Now partition [ 11 / 10 , 6 ] [11/10,6] [ 11/10 , 6 ] into the 1024 closed rational cells
[ 11 10 + 49 k 10240 , 11 10 + 49 ( k + 1 ) 10240 ] , 0 ≤ k < 1024. \left[\frac{11}{10}+\frac{49k}{10240},\frac{11}{10}+\frac{49(k+1)}{10240}\right],\qquad0\leq k<1024. [ 10 11 + 10240 49 k , 10 11 + 10240 49 ( k + 1 ) ] , 0 ≤ k < 1024.
On each whole cell evaluate the six complete sums, check B t > 0 B_t>0 B t > 0 , and form
Γ = A t B t , Γ t = A t t − Γ B t t B t . \Gamma=\frac{A_t}{B_t},\qquad\Gamma_t=\frac{A_{tt}-\Gamma B_{tt}}{B_t}. Γ = B t A t , Γ t = B t A tt − Γ B tt .
The rational aggregate enclosures from these 1024 closed-cell evaluations are
20368 10 4 < B t < 282127 10 4 , 6184 10 4 < Γ t < 83996 10 4 . \frac{20368}{10^4}<B_t<\frac{282127}{10^4},\qquad\frac{6184}{10^4}<\Gamma_t<\frac{83996}{10^4}. 1 0 4 20368 < B t < 1 0 4 282127 , 1 0 4 6184 < Γ t < 1 0 4 83996 .
These imply the uniform bounds. The same evaluations give the three endpoint inequalities and
6342227 < 10 6 Γ ( 217 / 50 ) < 6342228 , 6342227 < 10^{6}\Gamma(217/50) < 6342228, 6342227 < 1 0 6 Γ ( 217/50 ) < 6342228 ,
and, at Y = 217 / 50 Y = 217/50 Y = 217/50 ,
B ( Y ) < 214321 10 4 < 43 2 , A ( Y ) < 636842 10 4 < 64. B(Y) < \frac{214321}{10^{4}} < \frac{43}{2}, \qquad A(Y) < \frac{636842}{10^{4}} < 64. B ( Y ) < 1 0 4 214321 < 2 43 , A ( Y ) < 1 0 4 636842 < 64.
Thus
Γ t > 0 , [ 121 / 50 , 7921 / 1250 ] ⊂ Γ ( [ 11 / 10 , 217 / 50 ] ) , \Gamma_t > 0, \qquad[121/50,7921/1250] \subset\Gamma([11/10,217/50]), Γ t > 0 , [ 121/50 , 7921/1250 ] ⊂ Γ ([ 11/10 , 217/50 ]) ,
and the inverse-function theorem gives a unique analytic Y ( V ) Y(V) Y ( V ) . At this root,
d B d V = 4 V B t Γ t > 0 , A t t − 2 V 2 B t t = B t Γ t , f ′ = − 4 V B , f ′ ′ − f ′ V = − 4 V d B d V < 0. \frac{\mathrm{d}B}{\mathrm{d}V} = 4V\frac{B_t}{\Gamma_t} > 0, \qquad A_{tt} - 2V^2B_{tt} = B_t\Gamma_t, \qquad f' = -4VB, \qquad f'' - \frac{f'}{V} = -4V\frac{\mathrm{d}B}{\mathrm{d}V} < 0. d V d B = 4 V Γ t B t > 0 , A tt − 2 V 2 B tt = B t Γ t , f ′ = − 4 V B , f ′′ − V f ′ = − 4 V d V d B < 0.
The derivative bounds and
Γ ( 217 / 50 ) > 7921 / 1250 ⟹ Y ( V ) < 217 / 50 , B t > 0 , A t = Γ B t > 0 \Gamma(217/50) > 7921/1250 \Longrightarrow Y(V) < 217/50, \qquad B_t > 0, \qquad A_t = \Gamma B_t > 0 Γ ( 217/50 ) > 7921/1250 ⟹ Y ( V ) < 217/50 , B t > 0 , A t = Γ B t > 0
give the moment caps from the values at 217 / 50 217/50 217/50 .
At V = 11 / 10 V = 11/10 V = 11/10 , Theorem 4.12 places the global SH parameter in the interval
( 1.1013103262 , 1.1013103264 ) ⊂ ( 11 / 10 , 6 ) . (1.1013103262,1.1013103264) \subset(11/10,6). ( 1.1013103262 , 1.1013103264 ) ⊂ ( 11/10 , 6 ) .
Its parameter satisfies 0 = Y ∂ Y Φ 11 / 10 ( H Y ) = A t ( Y ) − 2 ( 11 / 10 ) 2 B t ( Y ) 0 = Y\partial_Y\Phi_{11/10}(H_Y) = A_t(Y) - 2(11/10)^2B_t(Y) 0 = Y ∂ Y Φ 11/10 ( H Y ) = A t ( Y ) − 2 ( 11/10 ) 2 B t ( Y ) , so Γ ( Y ) = 2 ( 11 / 10 ) 2 \Gamma(Y) = 2(11/10)^2 Γ ( Y ) = 2 ( 11/10 ) 2 and Y = Y ( 11 / 10 ) Y = Y(11/10) Y = Y ( 11/10 ) . At V = 89 / 50 V = 89/50 V = 89/50 , 39 / 10 < Y ( V ) < 6 39/10 < Y(V) < 6 39/10 < Y ( V ) < 6 by the endpoint inequalities. The physical triangular planar area is
A p l a n e = ( 3 V 2 2 Y ) 1 / 3 . A_{\mathrm{plane}} = \left(\frac{\sqrt{3}V^2}{2Y}\right)^{1/3}. A plane = ( 2 Y 3 V 2 ) 1/3 .
Hence 77 / 100 < A p l a n e < 89 / 100 77/100 < A_{\mathrm{plane}} < 89/100 77/100 < A plane < 89/100 and d A p l a n e / d Y = − A p l a n e / ( 3 Y ) ≠ 0 \mathrm{d}A_{\mathrm{plane}}/\mathrm{d}Y = -A_{\mathrm{plane}}/(3Y) \ne0 d A plane / d Y = − A plane / ( 3 Y ) = 0 , so A t − 2 V 2 B t = 0 ⟺ F V ′ ( A p l a n e ) = 0 A_t - 2V^2B_t = 0 \Longleftrightarrow F'_V(A_{\mathrm{plane}}) = 0 A t − 2 V 2 B t = 0 ⟺ F V ′ ( A plane ) = 0 . Proposition 3.20 gives F V ′ ′ > 34 F''_V > 34 F V ′′ > 34 on [ 2 / 5 , 89 / 100 ] [2/5,89/100] [ 2/5 , 89/100 ] ; hence
{ A ∈ [ 2 / 5 , 89 / 100 ] : F 89 / 50 ′ ( A ) = 0 } = { ( 3 ( 89 / 50 ) 2 2 Y ( 89 / 50 ) ) 1 / 3 } , \{A \in[2/5,89/100] : F'_{89/50}(A) = 0\} = \left\{\left(\frac{\sqrt{3}(89/50)^2}{2Y(89/50)}\right)^{1/3}\right\}, { A ∈ [ 2/5 , 89/100 ] : F 89/50 ′ ( A ) = 0 } = ⎩ ⎨ ⎧ ( 2 Y ( 89/50 ) 3 ( 89/50 ) 2 ) 1/3 ⎭ ⎬ ⎫ ,
identifying the second endpoint. This completes the proof of Lemma 6.2 . □
Lemma 6.3 (Moment contraction of the long SH profile). Put V 0 = 11 / 10 V_0 = 11/10 V 0 = 11/10 , V 1 = 89 / 50 V_1 = 89/50 V 1 = 89/50 and
f ( V ) = A ( Y ( V ) ) − 2 V 2 B ( Y ( V ) ) , g ( t ) = max V 0 ≤ V ≤ V 1 { f ( V ) + 2 V 2 t } . f(V) = A(Y(V)) - 2V^2B(Y(V)), \qquad g(t) = \max_{V_0 \le V \le V_1}\{f(V) + 2V^2t\}. f ( V ) = A ( Y ( V )) − 2 V 2 B ( Y ( V )) , g ( t ) = V 0 ≤ V ≤ V 1 max { f ( V ) + 2 V 2 t } . For b i = B ( Y ( V i ) ) b_i = B(Y(V_i)) b i = B ( Y ( V i )) , the unique maximizing covolume is
V ∗ ( t ) = { V 0 , t ≤ b 0 , Γ ( B − 1 ( t ) ) / 2 , b 0 ≤ t ≤ b 1 , V 1 , t ≥ b 1 . (52) V_*(t) =
\begin{cases}
V_0, & t \le b_0,\\
\sqrt{\Gamma(B^{-1}(t))/2}, & b_0 \le t \le b_1,\\
V_1, & t \ge b_1.
\end{cases} \tag*{(52)} V ∗ ( t ) = ⎩ ⎨ ⎧ V 0 , Γ ( B − 1 ( t )) /2 , V 1 , t ≤ b 0 , b 0 ≤ t ≤ b 1 , t ≥ b 1 . (52) Here B − 1 B^{-1} B − 1 is restricted to [ Y ( V 0 ) , Y ( V 1 ) ] [Y(V_0),Y(V_1)] [ Y ( V 0 ) , Y ( V 1 )] ; the endpoint formulas agree, and
g ( t ) = { f ( V 0 ) + 2 V 0 2 t , t ≤ b 0 , A ( B − 1 ( t ) ) , b 0 ≤ t ≤ b 1 , f ( V 1 ) + 2 V 1 2 t , t ≥ b 1 . g(t) =
\begin{cases}
f(V_0) + 2V_0^2t, & t \le b_0,\\
A(B^{-1}(t)), & b_0 \le t \le b_1,\\
f(V_1) + 2V_1^2t, & t \ge b_1.
\end{cases} g ( t ) = ⎩ ⎨ ⎧ f ( V 0 ) + 2 V 0 2 t , A ( B − 1 ( t )) , f ( V 1 ) + 2 V 1 2 t , t ≤ b 0 , b 0 ≤ t ≤ b 1 , t ≥ b 1 . For every real interval B = [ β − , β + ] \mathcal{B} = [\beta_-,\beta_+] B = [ β − , β + ] , the following finite construction encloses { 2 V ∗ ( t ) 2 : t ∈ B } \{2V_*(t)^2 : t \in\mathcal{B}\} { 2 V ∗ ( t ) 2 : t ∈ B } . Use the endpoint moment enclosures [ b 0 − , b 0 + ] [b_0^-,b_0^+] [ b 0 − , b 0 + ] in (A.8), whose outer band lies strictly inside the nodal moment domain. Let G i G_i G i enclose the exact rational 2 V i 2 2V_i^2 2 V i 2 outwards. Retain G 0 G_0 G 0 if β − ≤ b 0 + \beta_- \le b_0^+ β − ≤ b 0 + and retain G 1 G_1 G 1 if β + ≥ b 1 − \beta_+ \ge b_1^- β + ≥ b 1 − . If B ∩ [ b 0 − , b 1 + ] = [ ℓ , h ] ≠ ∅ \mathcal{B} \cap[b_0^-,b_1^+] = [\ell,h] \ne\varnothing B ∩ [ b 0 − , b 1 + ] = [ ℓ , h ] = ∅ , define
i − = max ( 0 , min { j : B j + ≥ ℓ } − 1 ) , i_- = \max(0,\min\{j : B_j^+ \ge\ell\} - 1), i − = max ( 0 , min { j : B j + ≥ ℓ } − 1 ) , i + = min ( 4096 , min ( { j : B j − > h } ∪ { 4096 } ) ) , i_+ = \min(4096,\min(\{j : B_j^- > h\} \cup\{4096\})), i + = min ( 4096 , min ({ j : B j − > h } ∪ { 4096 })) ,
and retain the nonempty interval
[ max ( inf G 0 , Γ i − − ) , min ( sup G 1 , Γ i + + ) ] . (53) \left[\max(\inf G_0,\Gamma_{i-}^{-}),\min(\sup G_1,\Gamma_{i+}^{+})\right]. \tag*{(53)} [ max ( inf G 0 , Γ i − − ) , min ( sup G 1 , Γ i + + ) ] . (53)
The hull contains 2 V ∗ ( t ) 2 2V_*(t)^2 2 V ∗ ( t ) 2 for all t ∈ B t \in\mathcal{B} t ∈ B , endpoints included; both searches use the 4097 ordered nodes. Under the endpoint comparisons, profile minima lie in [ b 0 , b 1 ] [b_0,b_1] [ b 0 , b 1 ] . Intersect the enclosure with every inherited supporting interval before testing ∇ η Ψ V ( U ( η ) Z 3 ) = 0 \nabla_\eta\Psi V(U(\eta)Z^3)=0 ∇ η Ψ V ( U ( η ) Z 3 ) = 0 .
Proof. By Lemma 6.2 ,
1 4 V ∂ V { f ( V ) + 2 V 2 t } = t − B ( Y ( V ) ) , \frac{1}{4V}\partial_V\{f(V)+2V^2t\}=t-B(Y(V)), 4 V 1 ∂ V { f ( V ) + 2 V 2 t } = t − B ( Y ( V )) ,
∂ V [ 1 4 V ∂ V { f ( V ) + 2 V 2 t } ] = − d d V B ( Y ( V ) ) < 0. \partial_V\left[\frac{1}{4V}\partial_V\{f(V)+2V^2t\}\right]=-\frac{\mathrm{d}}{\mathrm{d}V}B(Y(V))<0. ∂ V [ 4 V 1 ∂ V { f ( V ) + 2 V 2 t } ] = − d V d B ( Y ( V )) < 0.
Hence (52) , with t = B ( Y ( V ∗ ) ) t=B(Y(V_*)) t = B ( Y ( V ∗ )) and g ( t ) = A ( Y ( V ∗ ) ) g(t)=A(Y(V_*)) g ( t ) = A ( Y ( V ∗ )) in its interior. The guards retain all endpoint cases; for t ∈ B ∩ [ b 0 , b 1 ] t\in\mathcal{B}\cap[b_0,b_1] t ∈ B ∩ [ b 0 , b 1 ] , the ordered searches, including their boundary conventions, give
y i − ≤ B − 1 ( t ) ≤ y i + , Γ i − − ≤ 2 V ∗ ( t ) 2 ≤ Γ i + + . y_{i-}\le B^{-1}(t)\le y_{i+},\qquad\Gamma_{i-}^{-}\le2V_*(t)^2\le\Gamma_{i+}^{+}. y i − ≤ B − 1 ( t ) ≤ y i + , Γ i − − ≤ 2 V ∗ ( t ) 2 ≤ Γ i + + .
At array boundaries use strict containment of both roots in the node domain; otherwise B i − + < ℓ B_{i-}^{+}<\ell B i − + < ℓ , B i + − > h B_{i+}^{-}>h B i + − > h . Intersection with [ inf G 0 , sup G 1 ] [\inf G_0,\sup G_1] [ inf G 0 , sup G 1 ] and the hull give (53) ; Proposition 5.1 gives the profile-minimum assertions. This completes the proof of Lemma 6.3 . □
Lemma 6.4 (Separate planar and layer moment bounds). Fix t > 0 t>0 t > 0 in the coordinates of Lemma 6.1 . Suppose that throughout a convex transverse rectangle containing ( 2 / 3 , 1 / 2 , 0 , 0 ) (2/\sqrt{3},1/2,0,0) ( 2/ 3 , 1/2 , 0 , 0 ) ,
∥ ∇ ( r , x , y , z ) 2 E ( U ( t , r , x , y , z ) Z 3 , 3 ) ∥ o p ≤ M . \left\|\nabla^2_{(r,x,y,z)}E(U(t,r,x,y,z)\mathbb{Z}^3,3)\right\|_{\mathrm{op}}\le M. ∇ ( r , x , y , z ) 2 E ( U ( t , r , x , y , z ) Z 3 , 3 ) op ≤ M . Then every point of that rectangle satisfies
− M 2 ( y 2 + z 2 ) ≤ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ≤ M 2 { ( r − 2 / 3 ) 2 + ( x − 1 / 2 ) 2 } . (54) -\frac{M}{2}(y^2+z^2)\le E(U\mathbb{Z}^3,3)-E(H_Y,3)\le\frac{M}{2}\left\{(r-2/\sqrt{3})^2+(x-1/2)^2\right\}. \tag*{(54)} − 2 M ( y 2 + z 2 ) ≤ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ≤ 2 M { ( r − 2/ 3 ) 2 + ( x − 1/2 ) 2 } . (54) If moreover
11 10 ≤ Y ≤ 217 50 , 11 10 ≤ V ≤ 89 50 , E ( U ( t , r , x , y , z ) Z 3 , 3 ) = B ( Y ( V ) ) , \frac{11}{10}\le Y\le\frac{217}{50},\qquad\frac{11}{10}\le V\le\frac{89}{50},\qquad E(U(t,r,x,y,z)\mathbb{Z}^3,3)=B(Y(V)), 10 11 ≤ Y ≤ 50 217 , 10 11 ≤ V ≤ 50 89 , E ( U ( t , r , x , y , z ) Z 3 , 3 ) = B ( Y ( V )) , then
− 17 M 8 ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 17 M 8 { ( r − 2 / 3 ) 2 + ( x − 1 / 2 ) 2 } . (55) -\frac{17M}{8}(y^2+z^2)\le2V^2-\Gamma(Y)\le\frac{17M}{8}\left\{(r-2/\sqrt{3})^2+(x-1/2)^2\right\}. \tag*{(55)} − 8 17 M ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 8 17 M { ( r − 2/ 3 ) 2 + ( x − 1/2 ) 2 } . (55)
Proof. Let Γ \Gamma Γ be the planar lattice of area t t t and u = ( t r y , t / r z ) u=(\sqrt{t}r\,y,\sqrt{t/r}\,z) u = ( t r y , t / r z ) . For s > 0 s>0 s > 0 , Montgomery’s theorem [24 Theorem 1] and Poisson summation give
∑ v ∈ Γ e − s ∣ v ∣ 2 ≥ ∑ v ∈ T t e − s ∣ v ∣ 2 , \sum_{v\in\Gamma}e^{-s|v|^2}\ge\sum_{v\in T_t}e^{-s|v|^2}, v ∈ Γ ∑ e − s ∣ v ∣ 2 ≥ v ∈ T t ∑ e − s ∣ v ∣ 2 ,
∑ v ∈ Γ e − s ∣ v + k u ∣ 2 = π t s ∑ w ∈ Γ ∗ e − π 2 ∣ w ∣ 2 / s cos ( 2 π k w ⋅ u ) ≤ ∑ v ∈ Γ e − s ∣ v ∣ 2 . \sum_{v\in\Gamma}e^{-s|v+ku|^2}=\frac{\pi}{ts}\sum_{w\in\Gamma^*}e^{-\pi^2|w|^2/s}\cos(2\pi kw\cdot u)\le\sum_{v\in\Gamma}e^{-s|v|^2}. v ∈ Γ ∑ e − s ∣ v + k u ∣ 2 = t s π w ∈ Γ ∗ ∑ e − π 2 ∣ w ∣ 2 / s cos ( 2 π k w ⋅ u ) ≤ v ∈ Γ ∑ e − s ∣ v ∣ 2 .
Multiply by e − s k 2 / t 2 e^{-sk^2/t^2} e − s k 2 / t 2 , sum over k ∈ Z k\in\mathbb{Z} k ∈ Z , subtract the common zero-vector term and use q − 3 = 1 2 ∫ 0 ∞ s 2 e − s q d s q^{-3}=\frac{1}{2}\int_0^\infty s^2e^{-sq}\,\mathrm{d}s q − 3 = 2 1 ∫ 0 ∞ s 2 e − s q d s . Thus
E ( Γ ⊕ t − 1 Z , 3 ) ≥ E ( T t ⊕ t − 1 Z , 3 ) = E ( H Y , 3 ) , (56) E(\Gamma\oplus t^{-1}\mathbb{Z},3)\ge E(T_t\oplus t^{-1}\mathbb{Z},3)=E(H_Y,3), \tag*{(56)} E ( Γ ⊕ t − 1 Z , 3 ) ≥ E ( T t ⊕ t − 1 Z , 3 ) = E ( H Y , 3 ) , (56)
E ( Γ + Z ( u , t − 1 ) , 3 ) ≤ E ( Γ ⊕ t − 1 Z , 3 ) . (57) E(\Gamma+\mathbb{Z}(u,t^{-1}),3)\le E(\Gamma\oplus t^{-1}\mathbb{Z},3). \tag*{(57)} E ( Γ + Z ( u , t − 1 ) , 3 ) ≤ E ( Γ ⊕ t − 1 Z , 3 ) . (57)
Layer reflection gives ∇ y , z E ( U ( t , r , x , y , z ) Z 3 , 3 ) = 0 \nabla_{y,z}E(U(t,r,x,y,z)\mathbb{Z}^3,3)=0 ∇ y , z E ( U ( t , r , x , y , z ) Z 3 , 3 ) = 0 at y = z = 0 y=z=0 y = z = 0 ; hexagonal symmetry gives ∇ r , x E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) = 0 \nabla_{r,x}E(U(t,r,x,0,0)\mathbb{Z}^3,3)=0 ∇ r , x E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) = 0 at ( r , x ) = ( 2 / 3 , 1 / 2 ) (r,x)=(2/\sqrt{3},1/2) ( r , x ) = ( 2/ 3 , 1/2 ) . The Hessian bound and Taylor’s formula on the two segments yield
E ( U ( t , r , x , y , z ) Z 3 , 3 ) ≥ E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) − M 2 ( y 2 + z 2 ) , E(U(t,r,x,y,z)\mathbb{Z}^3,3)\ge E(U(t,r,x,0,0)\mathbb{Z}^3,3)-\frac{M}{2}(y^2+z^2), E ( U ( t , r , x , y , z ) Z 3 , 3 ) ≥ E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) − 2 M ( y 2 + z 2 ) ,
E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) ≤ E ( H Y , 3 ) + M 2 { ( r − 2 / 3 ) 2 + ( x − 1 / 2 ) 2 } . E(U(t,r,x,0,0)\mathbb{Z}^3,3)\le E(H_Y,3)+\frac{M}{2}\left\{(r-2/\sqrt{3})^2+(x-1/2)^2\right\}. E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) ≤ E ( H Y , 3 ) + 2 M { ( r − 2/ 3 ) 2 + ( x − 1/2 ) 2 } .
The segments remain in the rectangle; (A3)–(A4) give (A1). For E ( U Z 3 , 3 ) = B ( Y ( V ) ) E(UZ^{3},3)=B(Y(V)) E ( U Z 3 , 3 ) = B ( Y ( V )) , Lemma 6.2 gives
E ( U Z 3 , 3 ) − E ( H Y , 3 ) = ∫ Γ ( Y ) V d d v B ( Y ( v ) ) d v , 1 4 v d d v B ( Y ( v ) ) > 4 17 . E(UZ^{3},3)-E(H_Y,3)=\int_{\sqrt{\Gamma(Y)}}^{V}\frac{\mathrm{d}}{\mathrm{d}v}B(Y(v))\,\mathrm{d}v,\qquad\frac{1}{4v}\frac{\mathrm{d}}{\mathrm{d}v}B(Y(v))>\frac{4}{17}. E ( U Z 3 , 3 ) − E ( H Y , 3 ) = ∫ Γ ( Y ) V d v d B ( Y ( v )) d v , 4 v 1 d v d B ( Y ( v )) > 17 4 .
Since Γ t > 0 \Gamma_t>0 Γ t > 0 and Y , Y ( V ) ∈ [ 11 / 10 , 217 / 50 ] Y,Y(V)\in[11/10,217/50] Y , Y ( V ) ∈ [ 11/10 , 217/50 ] , the whole integration segment remains in the inverse-branch domain. The two signs in (A1) yield (A2). This completes the proof of Lemma 6.4 . □ \square □
Lemma 6.5 (A Fourier transverse tube for the SH profile). Use the coordinates of Lemma 6.1 and the profile Φ V \Phi_V Φ V . For every 11 / 10 ≤ Y ≤ 217 / 50 11/10\le Y\le217/50 11/10 ≤ Y ≤ 217/50 , the conditions
∣ r − 2 3 ∣ ≤ 1 200 , 99 200 ≤ x ≤ 1 2 , ∣ y ∣ , ∣ z ∣ ≤ 1 40 , \left|r-\frac{2}{\sqrt{3}}\right|\le\frac{1}{200},\qquad\frac{99}{200}\le x\le\frac{1}{2},\qquad|y|,|z|\le\frac{1}{40}, r − 3 2 ≤ 200 1 , 200 99 ≤ x ≤ 2 1 , ∣ y ∣ , ∣ z ∣ ≤ 40 1 , − 17 640 ≤ 2 V 2 − Γ ( Y ) ≤ 51 3200 (58) -\frac{17}{640}\le2V^2-\Gamma(Y)\le\frac{51}{3200} \tag*{(58)} − 640 17 ≤ 2 V 2 − Γ ( Y ) ≤ 3200 51 (58) imply
∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) ≻ 0. \nabla^2_{(r,x,y,z)}\Phi_V(UZ^3)\succ0. ∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) ≻ 0. For 11 / 10 ≤ Y ≤ 217 / 50 11/10\le Y\le217/50 11/10 ≤ Y ≤ 217/50 , 11 / 10 ≤ V ≤ 89 / 50 11/10\le V\le89/50 11/10 ≤ V ≤ 89/50 and coordinates in (58) , at fixed t t t ,
E ( U ( t , r , x , y , z ) Z 3 , 3 ) = B ( Y ( V ) ) , ∇ r , x , y , z Φ V ( U ( t , r , x , y , z ) Z 3 ) = 0 E(U(t,r,x,y,z)\mathbb{Z}^3,3)=B(Y(V)),\qquad\nabla_{r,x,y,z}\Phi_V(U(t,r,x,y,z)\mathbb{Z}^3)=0 E ( U ( t , r , x , y , z ) Z 3 , 3 ) = B ( Y ( V )) , ∇ r , x , y , z Φ V ( U ( t , r , x , y , z ) Z 3 ) = 0 ⟹ ( r , x , y , z ) = ( 2 3 , 1 2 , 0 , 0 ) , Y = Y ( V ) . \Longrightarrow\quad(r,x,y,z)=\left(\frac{2}{\sqrt{3}},\frac{1}{2},0,0\right),\qquad Y=Y(V). ⟹ ( r , x , y , z ) = ( 3 2 , 2 1 , 0 , 0 ) , Y = Y ( V ) .
Proof. All finite decimals below denote exact rationals.
The layer decomposition. Let Λ 2 \Lambda_2 Λ 2 be the planar lattice of area t t t with basis ( t r , 0 ) (\sqrt{tr},0) ( t r , 0 ) and ( t r x , t / r ) (\sqrt{tr}x,\sqrt{t/r}) ( t r x , t / r ) , and set u = ( t r y , t / r z ) u=(\sqrt{tr}y,\sqrt{t/r}z) u = ( t r y , t / r z ) . For its dual vector indexed by ( m , n ) (m,n) ( m , n ) ,
R 2 = q = m 2 t r + r t ( n − x m ) 2 , w ⋅ u = m y + ( n − x m ) z . R^2=q=\frac{m^2}{tr}+\frac{r}{t}(n-xm)^2,\qquad w\cdot u=my+(n-xm)z. R 2 = q = t r m 2 + t r ( n − x m ) 2 , w ⋅ u = m y + ( n − x m ) z .
Equation (29) , Lemma 3.1 and planar Poisson summation give, for s = 3 , 6 s=3,6 s = 3 , 6 ,
E ( U Z 3 , s ) = E ( Λ 2 , s ) + 2 π ζ ( 2 s − 2 ) t ( s − 1 ) ( 1 / t ) 2 − 2 s + 2 t ∑ k = 1 ∞ ∑ ( m , n ) ∈ Z 2 ∖ { ( 0 , 0 ) } F s ( q , k ( 1 / t ) ) cos ψ , (59) \begin{aligned}
E(UZ^3,s)&=E(\Lambda_2,s)+\frac{2\pi\zeta(2s-2)}{t(s-1)}(1/t)^{2-2s}\\
&\quad+\frac{2}{t}\sum_{k=1}^{\infty}\sum_{(m,n)\in\mathbb{Z}^2\setminus\{(0,0)\}}F_s(q,k(1/t))\cos\psi, \tag*{(59)}
\end{aligned} E ( U Z 3 , s ) = E ( Λ 2 , s ) + t ( s − 1 ) 2 π ζ ( 2 s − 2 ) ( 1/ t ) 2 − 2 s + t 2 k = 1 ∑ ∞ ( m , n ) ∈ Z 2 ∖ {( 0 , 0 )} ∑ F s ( q , k ( 1/ t )) cos ψ , (59)
F s ( q , d ) = 2 π s Γ ( s ) ( R / d ) s − 1 K s − 1 ( 2 π d R ) , ψ = 2 π k { m y + ( n − x m ) z } . F_s(q,d)=\frac{2\pi^s}{\Gamma(s)}(R/d)^{s-1}K_{s-1}(2\pi dR),\qquad\psi=2\pi k\{my+(n-xm)z\}. F s ( q , d ) = Γ ( s ) 2 π s ( R / d ) s − 1 K s − 1 ( 2 π d R ) , ψ = 2 π k { m y + ( n − x m ) z } .
The zero mode has no transverse derivatives; for Φ V \Phi_V Φ V , the nonzero-layer coefficients are
C 0 ( q , d ) = π 6 60 ( R / d ) 5 K 5 ( 2 π d R ) − 2 V 2 π 3 ( R / d ) 2 K 2 ( 2 π d R ) , C_0(q,d)=\frac{\pi^6}{60}(R/d)^5K_5(2\pi dR)-2V^2\pi^3(R/d)^2K_2(2\pi dR), C 0 ( q , d ) = 60 π 6 ( R / d ) 5 K 5 ( 2 π d R ) − 2 V 2 π 3 ( R / d ) 2 K 2 ( 2 π d R ) ,
C 1 ( q , d ) = − π 7 60 ( R / d ) 4 K 4 ( 2 π d R ) + 2 V 2 π 4 ( R / d ) K 1 ( 2 π d R ) , C_1(q,d)=-\frac{\pi^7}{60}(R/d)^4K_4(2\pi dR)+2V^2\pi^4(R/d)K_1(2\pi dR), C 1 ( q , d ) = − 60 π 7 ( R / d ) 4 K 4 ( 2 π d R ) + 2 V 2 π 4 ( R / d ) K 1 ( 2 π d R ) ,
C 2 ( q , d ) = π 8 60 ( R / d ) 3 K 3 ( 2 π d R ) − 2 V 2 π 5 K 0 ( 2 π d R ) . (60) C_2(q,d)=\frac{\pi^8}{60}(R/d)^3K_3(2\pi dR)-2V^2\pi^5K_0(2\pi dR). \tag*{(60)} C 2 ( q , d ) = 60 π 8 ( R / d ) 3 K 3 ( 2 π d R ) − 2 V 2 π 5 K 0 ( 2 π d R ) . (60)
By Lemma 3.1, C 1 = ∂ q C 0 C_1=\partial_qC_0 C 1 = ∂ q C 0 and C 2 = ∂ q 2 C 0 C_2=\partial_q^2C_0 C 2 = ∂ q 2 C 0 . The factor 2 / t 2/t 2/ t in (59) includes the two signs of the layer index, and its dual sum includes both signs of every nonzero dual vector.
Finite Bessel enclosures. No Bessel approximation is assumed in the finite calculation. For ν = 0 , 1 \nu= 0,1 ν = 0 , 1 , use (26) . For p = ± 1 / 2 p = \pm1/2 p = ± 1/2 , a > 0 a > 0 a > 0 and N ∈ Z ≥ 0 N \in\mathbb{Z}_{\geq0} N ∈ Z ≥ 0 , Taylor’s theorem gives
sgn ( ( 1 + a ) p − ∑ j = 0 N ( p j ) a j ) = sgn ∏ j = 0 N ( p − j ) . \operatorname{sgn}\left((1+a)^p-\sum_{j=0}^{N}\binom{p}{j}a^j\right)=\operatorname{sgn}\prod_{j=0}^{N}(p-j). sgn ( ( 1 + a ) p − j = 0 ∑ N ( j p ) a j ) = sgn j = 0 ∏ N ( p − j ) .
Integrate this sign against the positive weight in (26) . Define
a ν , 0 = 1 , a ν , j = a ν , j − 1 4 ν 2 − ( 2 j − 1 ) 2 8 j ν . a_{\nu,0}=1,\qquad a_{\nu,j}=a_{\nu,j-1}\frac{4\nu^2-(2j-1)^2}{8j\nu}. a ν , 0 = 1 , a ν , j = a ν , j − 1 8 j ν 4 ν 2 − ( 2 j − 1 ) 2 .
The exact gamma moments give
∑ j = 0 15 a 0 , j ( v ) ≤ K 0 ( v ) π / ( 2 v ) e − v ≤ ∑ j = 0 14 a 0 , j ( v ) , \sum_{j=0}^{15}a_{0,j}(v)\leq\frac{K_0(v)}{\sqrt{\pi/(2v)}e^{-v}}\leq\sum_{j=0}^{14}a_{0,j}(v), j = 0 ∑ 15 a 0 , j ( v ) ≤ π / ( 2 v ) e − v K 0 ( v ) ≤ j = 0 ∑ 14 a 0 , j ( v ) ,
∑ j = 0 14 a 1 , j ( v ) ≤ K 1 ( v ) π / ( 2 v ) e − v ≤ ∑ j = 0 15 a 1 , j ( v ) . (61) \sum_{j=0}^{14}a_{1,j}(v)\leq\frac{K_1(v)}{\sqrt{\pi/(2v)}e^{-v}}\leq\sum_{j=0}^{15}a_{1,j}(v). \tag*{(61)} j = 0 ∑ 14 a 1 , j ( v ) ≤ π / ( 2 v ) e − v K 1 ( v ) ≤ j = 0 ∑ 15 a 1 , j ( v ) . (61)
For the retained arguments v > 7 v>7 v > 7 ,
∑ j = 1 15 ∣ a 0 , j ( 7 ) ∣ < 1 50 , ∑ j = 1 15 ∣ a 1 , j ( 7 ) ∣ < 1 10 . \sum_{j=1}^{15}|a_{0,j}(7)|<\frac{1}{50},\qquad\sum_{j=1}^{15}|a_{1,j}(7)|<\frac{1}{10}. j = 1 ∑ 15 ∣ a 0 , j ( 7 ) ∣ < 50 1 , j = 1 ∑ 15 ∣ a 1 , j ( 7 ) ∣ < 10 1 .
These sums decrease with v v v , so the bounds are positive. For ν = 1 , 2 , 3 , 4 \nu=1,2,3,4 ν = 1 , 2 , 3 , 4 , enclose the remaining orders by the positive recurrence in Lemma 3.1. Since K ν ′ < 0 K_\nu'<0 K ν ′ < 0 , on [ v − , v + ] [v_-,v_+] [ v − , v + ] use the lower bound at v + v_+ v + and the upper bound at v − v_- v − .
Use the outward arithmetic, checked radicals and scaled degree-21/20 exponential bounds of Lemma A.2 in every coefficient recurrence.
To preserve radial cancellation in (60) , first form the three direct interval enclosures I 0 , I 1 , I 2 I_0,I_1,I_2 I 0 , I 1 , I 2 . For the midpoint q c q_c q c of the squared-radius interval, evaluate the same expressions there with the full d , V d,V d , V intervals, obtaining J 0 , J 1 , J 2 J_0,J_1,J_2 J 0 , J 1 , J 2 . Intersect
I 0 ← I 0 ∩ ( J 0 + I 1 ( q − q c ) ) , I 1 ← I 1 ∩ ( J 1 + I 2 ( q − q c ) ) . (62) I_0\leftarrow I_0\cap\left(J_0+I_1(q-q_c)\right),\qquad I_1\leftarrow I_1\cap\left(J_1+I_2(q-q_c)\right). \tag*{(62)} I 0 ← I 0 ∩ ( J 0 + I 1 ( q − q c ) ) , I 1 ← I 1 ∩ ( J 1 + I 2 ( q − q c ) ) . (62)
The mean value theorem at each fixed d , V d,V d , V proves both inclusions; the first uses the unrefined I 1 I_1 I 1 .
The planar Hessian and its complete tail. Set
Q = r ( i + x j ) 2 + j 2 r , ∇ r , x Q = ( ( i + x j ) 2 − j 2 r 2 , 2 r j ( i + x j ) ) , Q=r(i+xj)^2+\frac{j^2}{r},\qquad\nabla_{r,x}Q=\left((i+xj)^2-\frac{j^2}{r^2},\,2rj(i+xj)\right), Q = r ( i + x j ) 2 + r j 2 , ∇ r , x Q = ( ( i + x j ) 2 − r 2 j 2 , 2 r j ( i + x j ) ) ,
∇ r , x 2 Q = ( 2 j 2 / r 3 2 j ( i + x j ) 2 j ( i + x j ) 2 r j 2 ) . \nabla^2_{r,x}Q=
\begin{pmatrix}
2j^2/r^3 & 2j(i+xj)\\
2j(i+xj) & 2rj^2
\end{pmatrix}. ∇ r , x 2 Q = ( 2 j 2 / r 3 2 j ( i + x j ) 2 j ( i + x j ) 2 r j 2 ) .
At area one, for s ∈ { 3 , 6 } s\in\{3,6\} s ∈ { 3 , 6 } ,
( P s ) a b = ∑ ( i , j ) ∈ Z 2 ∖ { ( 0 , 0 ) } { s ( s + 1 ) Q − s − 2 Q a Q b − s Q − s − 1 Q a b } , a , b ∈ { r , x } . (P_s)_{ab}=\sum_{(i,j)\in\mathbb{Z}^2\setminus\{(0,0)\}}\left\{s(s+1)Q^{-s-2}Q_aQ_b-sQ^{-s-1}Q_{ab}\right\},\qquad a,b\in\{r,x\}. ( P s ) ab = ( i , j ) ∈ Z 2 ∖ {( 0 , 0 )} ∑ { s ( s + 1 ) Q − s − 2 Q a Q b − s Q − s − 1 Q ab } , a , b ∈ { r , x } .
The finite sum uses
( i , j ) ∈ [ − 40 , 40 ] 2 ∩ Z 2 , i > 0 ∨ ( i = 0 , j > 0 ) , (i,j)\in[-40,40]^2\cap\mathbb{Z}^2,\qquad i>0\ \lor\ (i=0,j>0), ( i , j ) ∈ [ − 40 , 40 ] 2 ∩ Z 2 , i > 0 ∨ ( i = 0 , j > 0 ) ,
multiplicity 2, with diagonal interval squares Q a 2 Q_a^2 Q a 2 . On
1149 1000 ≤ r ≤ 29 25 , 49 100 ≤ x ≤ 51 100 \frac{1149}{1000}\leq r\leq\frac{29}{25},\qquad\frac{49}{100}\leq x\leq\frac{51}{100} 1000 1149 ≤ r ≤ 25 29 , 100 49 ≤ x ≤ 100 51
the two Gershgorin lower bounds of the planar Gram matrix exceed 1 / 2 1/2 1/2 . Thus Q ≥ ( i 2 + j 2 ) / 2 Q\geq(i^2+j^2)/2 Q ≥ ( i 2 + j 2 ) /2 .
Also
∣ Q a ∣ ≤ α a Q , ∣ Q a b ∣ ≤ β a b Q , α = ( 1 , 6 / 5 ) , β = ( 8 / 5 1 1 3 ) . |Q_a|\leq\alpha_aQ,\qquad|Q_{ab}|\leq\beta_{ab}Q,\qquad\alpha=(1,6/5),\qquad\beta=
\begin{pmatrix}
8/5 & 1\\
1 & 3
\end{pmatrix}. ∣ Q a ∣ ≤ α a Q , ∣ Q ab ∣ ≤ β ab Q , α = ( 1 , 6/5 ) , β = ( 8/5 1 1 3 ) .
The arithmetic–geometric mean inequality gives ∣ Q r ∣ ≤ Q / r |Q_r|\leq Q/r ∣ Q r ∣ ≤ Q / r , ∣ Q x ∣ ≤ r Q |Q_x|\leq rQ ∣ Q x ∣ ≤ r Q , ∣ Q r x ∣ ≤ Q |Q_{rx}|\leq Q ∣ Q r x ∣ ≤ Q ; also 2 / r 2 < 8 / 5 2/r^2<8/5 2/ r 2 < 8/5 and 2 r 2 < 3 2r^2<3 2 r 2 < 3 .
Since # { ( i , j ) ∈ Z 2 : max ( ∣ i ∣ , ∣ j ∣ ) = ℓ } = 8 ℓ \#\{(i,j) \in\mathbb{Z}^{2} : \max(|i|,|j|)=\ell\}=8\ell # {( i , j ) ∈ Z 2 : max ( ∣ i ∣ , ∣ j ∣ ) = ℓ } = 8 ℓ for ℓ ∈ Z ≥ 1 \ell\in\mathbb{Z}_{\geq1} ℓ ∈ Z ≥ 1 , the complete omitted planar moment beyond cutoff N N N is at most
8 2 s 2 s − 2 N 2 − 2 s , \frac{8\,2^{s}}{2s-2}N^{2-2s}, 2 s − 2 8 2 s N 2 − 2 s ,
and the omitted Hessian entry has absolute value at most
{ s ( s + 1 ) α a α b + s β a b } ( 8 2 s 2 s − 2 40 2 − 2 s ) . (63) \{s(s+1)\alpha_{a}\alpha_{b}+s\beta_{ab}\}\left(\frac{8\,2^{s}}{2s-2}40^{2-2s}\right). \tag*{(63)} { s ( s + 1 ) α a α b + s β ab } ( 2 s − 2 8 2 s 4 0 2 − 2 s ) . (63)
Add (63) symmetrically to the finite sums. At fixed t t t , the complete planar block is ( t − 6 P 6 − 2 V 2 t − 3 P 3 ) ⊕ 0 2 × 2 (t^{-6}P_{6}-2V^{2}t^{-3}P_{3})\oplus0_{2\times2} ( t − 6 P 6 − 2 V 2 t − 3 P 3 ) ⊕ 0 2 × 2 .
The finite nonzero-layer Hessian. Use the half-dual indices
k ∈ { 1 , 2 , 3 } , ( m , n ) ∈ [ − 4 , 4 ] 2 ∩ Z 2 , m > 0 ∨ ( m = 0 , n > 0 ) , k\in\{1,2,3\},\qquad(m,n)\in[-4,4]^{2}\cap\mathbb{Z}^{2},\qquad m>0\ \vee\ (m=0,n>0), k ∈ { 1 , 2 , 3 } , ( m , n ) ∈ [ − 4 , 4 ] 2 ∩ Z 2 , m > 0 ∨ ( m = 0 , n > 0 ) ,
with factor 4 / t 4/t 4/ t . The derivatives are
∇ r , x , y , z q = ( − m 2 t r 2 + ( n − x m ) 2 t , − 2 r m ( n − x m ) t , 0 , 0 ) , \nabla_{r,x,y,z}q=\left(-\frac{m^{2}}{tr^{2}}+\frac{(n-xm)^{2}}{t},-\frac{2rm(n-xm)}{t},0,0\right), ∇ r , x , y , z q = ( − t r 2 m 2 + t ( n − x m ) 2 , − t 2 r m ( n − x m ) , 0 , 0 ) ,
( q a b ) a , b ∈ { r , x } = ( 2 m 2 t r 3 − 2 m ( n − x m ) t − 2 m ( n − x m ) t 2 r m 2 t ) , (q_{ab})_{a,b\in\{r,x\}}=
\begin{pmatrix}
\frac{2m^{2}}{tr^{3}} & -\frac{2m(n-xm)}{t}\\
-\frac{2m(n-xm)}{t} & \frac{2rm^{2}}{t}
\end{pmatrix}, ( q ab ) a , b ∈ { r , x } = ( t r 3 2 m 2 − t 2 m ( n − x m ) − t 2 m ( n − x m ) t 2 r m 2 ) ,
∇ r , x , y , z ψ = ( 0 , − ( 2 π k ) m z , ( 2 π k ) m , ( 2 π k ) ( n − x m ) ) , ψ x z = ψ z x = − ( 2 π k ) m , \nabla_{r,x,y,z}\psi=(0,-(2\pi k)mz,(2\pi k)m,(2\pi k)(n-xm)),\qquad\psi_{xz}=\psi_{zx}=-(2\pi k)m, ∇ r , x , y , z ψ = ( 0 , − ( 2 π k ) m z , ( 2 π k ) m , ( 2 π k ) ( n − x m )) , ψ x z = ψ z x = − ( 2 π k ) m ,
with all other second phase derivatives zero. For each retained term the exact Hessian entry is
4 t [ ( C 2 q a q b + C 1 q a b ) cos ψ − C 1 ( q a ψ b + q b ψ a ) sin ψ − C 0 ψ a ψ b cos ψ − C 0 ψ a b sin ψ ] . (64) \frac{4}{t}\left[(C_{2}q_{a}q_{b}+C_{1}q_{ab})\cos\psi-C_{1}(q_{a}\psi_{b}+q_{b}\psi_{a})\sin\psi-C_{0}\psi_{a}\psi_{b}\cos\psi-C_{0}\psi_{ab}\sin\psi\right]. \tag*{(64)} t 4 [ ( C 2 q a q b + C 1 q ab ) cos ψ − C 1 ( q a ψ b + q b ψ a ) sin ψ − C 0 ψ a ψ b cos ψ − C 0 ψ ab sin ψ ] . (64)
For an interval phase use the globally valid enclosures
sin ψ ∈ [ − min ( 1 , max ∣ ψ ∣ ) , min ( 1 , max ∣ ψ ∣ ) ] , cos ψ ∈ [ max ( − 1 , 1 − ( max ∣ ψ ∣ ) 2 / 2 ) , 1 ] . \sin\psi\in[-\min(1,\max|\psi|),\min(1,\max|\psi|)],\qquad
\cos\psi\in[\max(-1,1-(\max|\psi|)^{2}/2),1]. sin ψ ∈ [ − min ( 1 , max ∣ ψ ∣ ) , min ( 1 , max ∣ ψ ∣ )] , cos ψ ∈ [ max ( − 1 , 1 − ( max ∣ ψ ∣ ) 2 /2 ) , 1 ] .
Complete nonzero-layer tails. Assume
7 12 ≤ t ≤ 37 40 , 1149 1000 ≤ r ≤ 29 25 , 49 100 ≤ x ≤ 51 100 , 2 V 2 ≤ 32 5 , ∣ z ∣ ≤ δ z . \frac{7}{12}\leq t\leq\frac{37}{40},\qquad\frac{1149}{1000}\leq r\leq\frac{29}{25},\qquad\frac{49}{100}\leq x\leq\frac{51}{100},\qquad2V^{2}\leq\frac{32}{5},\qquad|z|\leq\delta_{z}. 12 7 ≤ t ≤ 40 37 , 1000 1149 ≤ r ≤ 25 29 , 100 49 ≤ x ≤ 100 51 , 2 V 2 ≤ 5 32 , ∣ z ∣ ≤ δ z .
For input intervals I t , I r , I V I_{t},I_{r},I_{V} I t , I r , I V , put
t − = inf I t , t + = sup I t , r − = inf I r , r + = sup I r . t_{-}=\inf I_{t},\quad t_{+}=\sup I_{t},\quad r_{-}=\inf I_{r},\quad r_{+}=\sup I_{r}. t − = inf I t , t + = sup I t , r − = inf I r , r + = sup I r .
Evaluate all constants outwards and use upper endpoints as absolute bounds.
Let j = max ( ∣ m ∣ , ∣ n ∣ ) j=\max(|m|,|n|) j = max ( ∣ m ∣ , ∣ n ∣ ) . Completing the square in m m m gives
R 2 ≥ j 2 t min { 1 r , r 1 + r 2 x 2 } > ( 24 j 25 ) 2 . R^{2}\geq\frac{j^{2}}{t}\min\left\{\frac{1}{r},\frac{r}{1+r^{2}x^{2}}\right\}>\left(\frac{24j}{25}\right)^{2}. R 2 ≥ t j 2 min { r 1 , 1 + r 2 x 2 r } > ( 25 24 j ) 2 .
Here r / ( 1 + r 2 x 2 ) r/(1+r^{2}x^{2}) r / ( 1 + r 2 x 2 ) increases in r r r throughout the box. For ( m , n ) ∈ { − 1 , 0 , 1 } 2 ∖ { ( 0 , 0 ) } (m,n)\in\{-1,0,1\}^{2}\setminus\{(0,0)\} ( m , n ) ∈ { − 1 , 0 , 1 } 2 ∖ {( 0 , 0 )} , R > 111 / 100 R>111/100 R > 111/100 . Thus
R j = { 111 / 100 , j = 1 , 24 j / 25 , j ≥ 2. R_{j}=
\begin{cases}
111/100,&j=1,\\
24j/25,&j\geq2.
\end{cases} R j = { 111/100 , 24 j /25 , j = 1 , j ≥ 2.
Retain 1 ≤ k ≤ K 1\leq k\leq K 1 ≤ k ≤ K and ∣ m ∣ , ∣ n ∣ ≤ W |m|,|n|\leq W ∣ m ∣ , ∣ n ∣ ≤ W , where K ≥ 3 K\geq3 K ≥ 3 , W ≥ 4 W\geq4 W ≥ 4 . The radial floors give 2 π k R / t > 30 2\pi kR/t>30 2 π k R / t > 30 on every omitted index.
For integer ν ≥ 0 \nu\geq0 ν ≥ 0 , use the half-integer polynomials P ν P_{\nu} P ν of Lemma 3.1:
K ν ( u ) ≤ K ν + 1 / 2 ( u ) = π 2 u e − u P ν ( u ) ≤ π 2 u e − u P ν ( 30 ) ( u ≥ 30 ) . K_{\nu}(u)\leq K_{\nu+1/2}(u)=\sqrt{\frac{\pi}{2u}}e^{-u}P_{\nu}(u)\leq\sqrt{\frac{\pi}{2u}}e^{-u}P_{\nu}(30)\qquad(u\geq30). K ν ( u ) ≤ K ν + 1/2 ( u ) = 2 u π e − u P ν ( u ) ≤ 2 u π e − u P ν ( 30 ) ( u ≥ 30 ) .
Define the six coefficient bounds, already including the common 2 / t 2/t 2/ t :
B 0 = π 6 60 P 5 ( 30 ) t + 9 / 2 , B 1 = π 7 60 P 4 ( 30 ) t + 7 / 2 , B 2 = π 8 60 P 3 ( 30 ) t + 5 / 2 , B_{0}=\frac{\pi^{6}}{60}P_{5}(30)t_{+}^{9/2},\qquad
B_{1}=\frac{\pi^{7}}{60}P_{4}(30)t_{+}^{7/2},\qquad
B_{2}=\frac{\pi^{8}}{60}P_{3}(30)t_{+}^{5/2}, B 0 = 60 π 6 P 5 ( 30 ) t + 9/2 , B 1 = 60 π 7 P 4 ( 30 ) t + 7/2 , B 2 = 60 π 8 P 3 ( 30 ) t + 5/2 ,
D 0 = ( 2 sup I V 2 ) π 3 P 2 ( 30 ) t + 3 / 2 , D 1 = ( 2 sup I V 2 ) π 4 P 1 ( 30 ) t + 1 / 2 , D 2 = ( 2 sup I V 2 ) π 5 t − − 1 / 2 . D_{0}=(2\sup I_{V}^{2})\pi^{3}P_{2}(30)t_{+}^{3/2},\qquad
D_{1}=(2\sup I_{V}^{2})\pi^{4}P_{1}(30)t_{+}^{1/2},\qquad
D_{2}=(2\sup I_{V}^{2})\pi^{5}t_{-}^{-1/2}. D 0 = ( 2 sup I V 2 ) π 3 P 2 ( 30 ) t + 3/2 , D 1 = ( 2 sup I V 2 ) π 4 P 1 ( 30 ) t + 1/2 , D 2 = ( 2 sup I V 2 ) π 5 t − − 1/2 .
The corresponding layer powers are k − 7 / 2 k^{-7/2} k − 7/2 , k − 9 / 2 k^{-9/2} k − 9/2 (repulsion) and k − 1 / 2 k^{-1/2} k − 1/2 , k − 3 / 2 k^{-3/2} k − 3/2 (attraction).
Use coordinate order ( r , x , y , z ) (r,x,y,z) ( r , x , y , z ) and define
α = ( 1 / r − , r + , 0 , 0 ) , β = ( 0 , δ z ( 2 π t + r + ) , 2 π t + r + , 2 π t + / r − ) . \alpha=(1/r_-,r_+,0,0),\qquad\beta=(0,\delta_z(2\pi\sqrt{t_+r_+}),2\pi\sqrt{t_+r_+},2\pi\sqrt{t_+/r_-}). α = ( 1/ r − , r + , 0 , 0 ) , β = ( 0 , δ z ( 2 π t + r + ) , 2 π t + r + , 2 π t + / r − ) .
The nonzero entries of Q = ( Q a b ) Q=(Q_{ab}) Q = ( Q ab ) and E = ( E a b ) E=(E_{ab}) E = ( E ab ) are
Q r r = 2 / r − 2 , Q r x = Q x r = 1 , Q x x = 2 r + 2 , E x z = E z x = 2 π t + r + . Q_{rr}=2/r_-^2,\qquad Q_{rx}=Q_{xr}=1,\qquad Q_{xx}=2r_+^2,\qquad E_{xz}=E_{zx}=2\pi\sqrt{t_+r_+}. Q r r = 2/ r − 2 , Q r x = Q x r = 1 , Q xx = 2 r + 2 , E x z = E z x = 2 π t + r + .
Direct differentiation and the arithmetic–geometric mean inequality give
∣ q a ∣ ≤ α a R 2 , ∣ q a b ∣ ≤ Q a b R 2 , ∣ ψ a ∣ ≤ β a k R , ∣ ψ a b ∣ ≤ E a b k R . |q_a|\leq\alpha_aR^2,\qquad|q_{ab}|\leq Q_{ab}R^2,\qquad|\psi_a|\leq\beta_akR,\qquad|\psi_{ab}|\leq E_{ab}kR. ∣ q a ∣ ≤ α a R 2 , ∣ q ab ∣ ≤ Q ab R 2 , ∣ ψ a ∣ ≤ β a k R , ∣ ψ ab ∣ ≤ E ab k R .
For example ∣ q x ∣ ≤ r q |q_x|\leq rq ∣ q x ∣ ≤ r q , ∣ q r x ∣ ≤ q |q_{rx}|\leq q ∣ q r x ∣ ≤ q , ∣ ψ y ∣ ≤ 2 π k t / r R |\psi_y|\leq2\pi k\sqrt{t/r}\,R ∣ ψ y ∣ ≤ 2 π k t / r R , and ∣ ψ z ∣ ≤ 2 π k t / r R |\psi_z|\leq2\pi k\sqrt{t/r}\,R ∣ ψ z ∣ ≤ 2 π k t / r R .
Apply (64) with the full-dual factor 2 / t 2/t 2/ t instead of the half-dual factor 4 / t 4/t 4/ t . Set
A a b = α a α b B 2 + ( α a β b + α b β a ) B 1 + β a β b B 0 , B a b = Q a b B 1 + E a b B 0 , A_{ab}=\alpha_a\alpha_bB_2+(\alpha_a\beta_b+\alpha_b\beta_a)B_1+\beta_a\beta_bB_0,\qquad B_{ab}=Q_{ab}B_1+E_{ab}B_0, A ab = α a α b B 2 + ( α a β b + α b β a ) B 1 + β a β b B 0 , B ab = Q ab B 1 + E ab B 0 ,
C a b = α a α b D 2 + ( α a β b + α b β a ) D 1 + β a β b D 0 , D a b = Q a b D 1 + E a b D 0 . C_{ab}=\alpha_a\alpha_bD_2+(\alpha_a\beta_b+\alpha_b\beta_a)D_1+\beta_a\beta_bD_0,\qquad D_{ab}=Q_{ab}D_1+E_{ab}D_0. C ab = α a α b D 2 + ( α a β b + α b β a ) D 1 + β a β b D 0 , D ab = Q ab D 1 + E ab D 0 .
The absolute summand is bounded by
e − 2 π k R / t + [ A a b R 13 / 2 k − 7 / 2 + B a b R 11 / 2 k − 9 / 2 + C a b R 7 / 2 k − 1 / 2 + D a b R 5 / 2 k − 3 / 2 ] . e^{-2\pi kR/t_+}\left[A_{ab}R^{13/2}k^{-7/2}+B_{ab}R^{11/2}k^{-9/2}+C_{ab}R^{7/2}k^{-1/2}+D_{ab}R^{5/2}k^{-3/2}\right]. e − 2 π k R / t + [ A ab R 13/2 k − 7/2 + B ab R 11/2 k − 9/2 + C ab R 7/2 k − 1/2 + D ab R 5/2 k − 3/2 ] .
On the omitted region, for p ∈ { 13 / 2 , 11 / 2 , 7 / 2 , 5 / 2 } p\in\{13/2,11/2,7/2,5/2\} p ∈ { 13/2 , 11/2 , 7/2 , 5/2 } ,
∂ R log ( R p e − 2 π k R / t + ) = p − 2 π k R / t + R < 0 , \partial_R\log\left(R^pe^{-2\pi kR/t_+}\right)=\frac{p-2\pi kR/t_+}{R}<0, ∂ R log ( R p e − 2 π k R / t + ) = R p − 2 π k R / t + < 0 ,
so R ≥ R j R\geq R_j R ≥ R j permits substitution of R j R_j R j .
The omitted-index sums admit a common closed majorant. For a > 0 a>0 a > 0 , p ≥ 0 p\geq0 p ≥ 0 and shell bounds c , R 1 > 0 c,R_1>0 c , R 1 > 0 , R j = c j R_j=cj R j = c j ( j ≥ 2 ) (j\geq2) ( j ≥ 2 ) , put
J = W + 1 , L = K + 1 , b = 2 π c / t + , η p = ( 1 + 1 / J ) p + 1 e − b . J=W+1,\qquad L=K+1,\qquad b=2\pi c/t_+,\qquad\eta_p=(1+1/J)^{p+1}e^{-b}. J = W + 1 , L = K + 1 , b = 2 π c / t + , η p = ( 1 + 1/ J ) p + 1 e − b .
For K , W ∈ Z ≥ 1 K,W\in\mathbb{Z}_{\geq1} K , W ∈ Z ≥ 1 , η p < 1 \eta_p<1 η p < 1 , and 2 π k R j / t + ≥ p 2\pi kR_j/t_+\geq p 2 π k R j / t + ≥ p on the omitted indices, define
S ( p , a ; c , R 1 , K , W , t + ) = 8 J ( c J ) p e − b J ( 1 − e − b J ) ( 1 − η p ) + ∑ j = 1 W 8 j R j p L − a e − 2 π L R j / t + 1 − e − 2 π R j / t + . (65) S(p,a;c,R_1,K,W,t_+)=\frac{8J(cJ)^pe^{-bJ}}{(1-e^{-bJ})(1-\eta_p)}
+\sum_{j=1}^{W}\frac{8jR_j^pL^{-a}e^{-2\pi LR_j/t_+}}{1-e^{-2\pi R_j/t_+}}. \tag*{(65)} S ( p , a ; c , R 1 , K , W , t + ) = ( 1 − e − b J ) ( 1 − η p ) 8 J ( c J ) p e − b J + j = 1 ∑ W 1 − e − 2 π R j / t + 8 j R j p L − a e − 2 π L R j / t + . (65)
For j ∈ Z ≥ 1 j\in\mathbb{Z}_{\geq1} j ∈ Z ≥ 1 ,
# { ( m , n ) ∈ Z 2 : max ( ∣ m ∣ , ∣ n ∣ ) = j } = 8 j . \#\{(m,n)\in\mathbb{Z}^2:\max(|m|,|n|)=j\}=8j. # {( m , n ) ∈ Z 2 : max ( ∣ m ∣ , ∣ n ∣ ) = j } = 8 j .
The omitted shell/layer pairs satisfy
{ ( j , k ) ∈ Z ≥ 1 2 : j > W ∨ k > K } = { ( j , k ) ∈ Z ≥ 1 2 : j > W } ∪ { ( j , k ) ∈ Z ≥ 1 2 : j ≤ W , k > K } . \{(j,k)\in\mathbb{Z}_{\geq1}^2:j>W\lor k>K\}
=\{(j,k)\in\mathbb{Z}_{\geq1}^2:j>W\}\cup\{(j,k)\in\mathbb{Z}_{\geq1}^2:j\leq W,\ k>K\}. {( j , k ) ∈ Z ≥ 1 2 : j > W ∨ k > K } = {( j , k ) ∈ Z ≥ 1 2 : j > W } ∪ {( j , k ) ∈ Z ≥ 1 2 : j ≤ W , k > K } .
The two parts use k − a ≤ 1 k^{-a}\leq1 k − a ≤ 1 , shell ratio ≤ η p \leq\eta_p ≤ η p , and k − a ≤ L − a k^{-a}\leq L^{-a} k − a ≤ L − a , respectively; geometric summation gives (65) . Both applications have J ≥ 5 J\geq5 J ≥ 5 , p ≤ 13 / 2 p\leq13/2 p ≤ 13/2 , b > 6 b>6 b > 6 , hence η p < 1 \eta_p<1 η p < 1 . Set
S ( p , a ) = S ( p , a ; 24 / 25 , 111 / 100 , K , W , t + ) . S(p,a)=S(p,a;24/25,111/100,K,W,t_+). S ( p , a ) = S ( p , a ; 24/25 , 111/100 , K , W , t + ) .
The complete entrywise tail radius is therefore
E a b = A a b S ( 13 / 2 , 7 / 2 ) + B a b S ( 11 / 2 , 9 / 2 ) + C a b S ( 7 / 2 , 1 / 2 ) + D a b S ( 5 / 2 , 3 / 2 ) . \mathcal{E}_{ab}=A_{ab}S(13/2,7/2)+B_{ab}S(11/2,9/2)+C_{ab}S(7/2,1/2)+D_{ab}S(5/2,3/2). E ab = A ab S ( 13/2 , 7/2 ) + B ab S ( 11/2 , 9/2 ) + C ab S ( 7/2 , 1/2 ) + D ab S ( 5/2 , 3/2 ) .
These majorants and their lower-order counterparts converge uniformly on the whole rectangle, justifying both Fourier derivatives.
The finite closed-cell certificate. Use Lemma A.2 ’s outward binary64 operations and checked square/cube roots for all inputs, coefficients and remainders. The full box is
r ∈ [ 2 / 3 ] + [ − ( 1 / 200 ) + , ( 1 / 200 ) + ] , r \in[2/\sqrt{3}] + [-(1/200)^{+}, (1/200)^{+}], r ∈ [ 2/ 3 ] + [ − ( 1/200 ) + , ( 1/200 ) + ] ,
x ∈ [ ( 1 / 2 − 1 / 200 ) − , 1 / 2 ] , y , z ∈ [ − ( 1 / 40 ) + , ( 1 / 40 ) + ] , x \in[(1/2 - 1/200)^{-}, 1/2], \qquad y,z \in[-(1/40)^{+}, (1/40)^{+}], x ∈ [( 1/2 − 1/200 ) − , 1/2 ] , y , z ∈ [ − ( 1/40 ) + , ( 1/40 ) + ] ,
where [ 2 / 3 ] [2/\sqrt{3}] [ 2/ 3 ] is its outward enclosure. Compute P 3 , P 6 P_3,P_6 P 3 , P 6 once on this entire planar box, including (63) :
P 3 ∈ ( [ [ 8.2454 , 10.4117 ] [ − 1.3124 , .9107 ] [ − 1.3124 , .9107 ] [ 14.3371 , 18.8609 ] ] ) , P_3 \in\left(
\begin{bmatrix}
[8.2454,10.4117] & [-1.3124,.9107] \\
[-1.3124,.9107] & [14.3371,18.8609]
\end{bmatrix}
\right), P 3 ∈ ( [ [ 8.2454 , 10.4117 ] [ − 1.3124 , .9107 ] [ − 1.3124 , .9107 ] [ 14.3371 , 18.8609 ] ] ) ,
P 6 ∈ ( [ [ 25.7504 , 31.4001 ] [ − 3.7107 , 2.2614 ] [ − 3.7107 , 2.2614 ] [ 44.5971 , 57.1271 ] ] ) . P_6 \in\left(
\begin{bmatrix}
[25.7504,31.4001] & [-3.7107,2.2614] \\
[-3.7107,2.2614] & [44.5971,57.1271]
\end{bmatrix}
\right). P 6 ∈ ( [ [ 25.7504 , 31.4001 ] [ − 3.7107 , 2.2614 ] [ − 3.7107 , 2.2614 ] [ 44.5971 , 57.1271 ] ] ) .
Subsequent arithmetic retains the complete intervals underlying these outward rational summaries. The closed dyadic parameter cells are
I d , k = [ 11 10 + 81 k 25 2 d , 11 10 + 81 ( k + 1 ) 25 2 d ] , d , k ∈ Z ≥ 0 , 0 ≤ k < 2 d . I_{d,k}=\left[\frac{11}{10}+\frac{81k}{25\,2^d},\frac{11}{10}+\frac{81(k+1)}{25\,2^d}\right], \qquad d,k\in\mathbb{Z}_{\ge0}, \qquad0\le k<2^d. I d , k = [ 10 11 + 25 2 d 81 k , 10 11 + 25 2 d 81 ( k + 1 ) ] , d , k ∈ Z ≥ 0 , 0 ≤ k < 2 d .
On each whole cell [ Y − , Y + ] [Y_{-},Y_{+}] [ Y − , Y + ] , Lemma 6.2 gives
t ∈ [ ( 3 2 Y + ) 1 / 3 , ( 3 2 Y − ) 1 / 3 ] , t\in\left[\left(\frac{\sqrt{3}}{2Y_{+}}\right)^{1/3},\left(\frac{\sqrt{3}}{2Y_{-}}\right)^{1/3}\right], t ∈ ( 2 Y + 3 ) 1/3 , ( 2 Y − 3 ) 1/3 ,
2 V 2 ∈ [ Γ ( Y − ) − , Γ ( Y + ) + ] + [ − ( 17 / 640 ) + , ( 51 / 3200 ) + ] . 2V^2\in[\Gamma(Y_{-})^{-},\Gamma(Y_{+})^{+}]+[-(17/640)^{+},(51/3200)^{+}]. 2 V 2 ∈ [ Γ ( Y − ) − , Γ ( Y + ) + ] + [ − ( 17/640 ) + , ( 51/3200 ) + ] .
All operations are outward; these boxes satisfy the Fourier-tail guard above, including 2 V 2 < 32 / 5 2V^2<32/5 2 V 2 < 32/5 .
Sum retained integer indices increasingly: Fourier ( m , n ) (m,n) ( m , n ) lexicographically, then k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 , and upper Hessian entries in coordinate order. Use diagonal interval squares, increasing Bessel degree/order, and the radial intersections (62) before (64) . Assemble H H H from the complete planar block and retained Fourier Hessian (64) , in order ( r , x , y , z ) (r,x,y,z) ( r , x , y , z ) , then add
H a b ← H a b + [ − E ‾ a b , E ‾ a b ] , H b a ← H a b ( 1 ≤ a ≤ b ≤ 4 ) . H_{ab}\leftarrow H_{ab}+[-\overline{E}_{ab},\overline{E}_{ab}],\qquad H_{ba}\leftarrow H_{ab}\qquad(1\le a\le b\le4). H ab ← H ab + [ − E ab , E ab ] , H ba ← H ab ( 1 ≤ a ≤ b ≤ 4 ) .
For j = 1 , 2 , 3 , 4 j=1,2,3,4 j = 1 , 2 , 3 , 4 , require H j j > 0 H_{jj}>0 H j j > 0 and update, for j < a ≤ b ≤ 4 j<a\le b\le4 j < a ≤ b ≤ 4 ,
H a b ← { H a a − H a j 2 / H j j , a = b , H a b − H a j H j b / H j j , a < b , H b a ← H a b , H_{ab}\leftarrow
\begin{cases}
H_{aa}-H_{aj}^{2}/H_{jj}, & a=b,\\
H_{ab}-H_{aj}H_{jb}/H_{jj}, & a<b,
\end{cases}
\qquad H_{ba}\leftarrow H_{ab}, H ab ← { H aa − H aj 2 / H j j , H ab − H aj H j b / H j j , a = b , a < b , H ba ← H ab ,
H j j > 0 ( j = 1 , 2 , 3 , 4 ) ⟹ ∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) ≻ 0. H_{jj}>0\ (j=1,2,3,4)\quad\Longrightarrow\quad\nabla^{2}_{(r,x,y,z)}\Phi_{V}(UZ^{3})\succ0. H j j > 0 ( j = 1 , 2 , 3 , 4 ) ⟹ ∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) ≻ 0.
Here H a j 2 H_{aj}^{2} H aj 2 is the interval square; the implication is the Schur-complement criterion for every enclosed Hessian.
Start from [ 11 / 10 , 217 / 50 ] [11/10,217/50] [ 11/10 , 217/50 ] and bisect inconclusive cells, left child first:
# n o d e s = 81 , # a c c e p t e d l e a v e s = 41 , max d = 10. \#\mathrm{nodes}=81,\qquad\#\mathrm{accepted\ leaves}=41,\qquad\max d=10. # nodes = 81 , # accepted leaves = 41 , max d = 10.
The accepted indices are
( d , k ) ∈ { ( 3 , 7 ) } ∪ ( { 4 } × { 6 , … , 13 } ) ∪ ( { 5 } × { 5 , … , 11 } ) (d,k)\in\{(3,7)\}\cup\bigl(\{4\}\times\{6,\ldots,13\}\bigr)\cup\bigl(\{5\}\times\{5,\ldots,11\}\bigr) ( d , k ) ∈ {( 3 , 7 )} ∪ ( { 4 } × { 6 , … , 13 } ) ∪ ( { 5 } × { 5 , … , 11 } )
∪ ( { 6 } × { 4 , … , 9 } ) ∪ ( { 7 , 8 } × { 4 , 5 , 6 , 7 } ) {}\cup\bigl(\{6\}\times\{4,\ldots,9\}\bigr)\cup\bigl(\{7,8\}\times\{4,5,6,7\}\bigr) ∪ ( { 6 } × { 4 , … , 9 } ) ∪ ( { 7 , 8 } × { 4 , 5 , 6 , 7 } )
∪ ( { 9 } × { 3 , … , 7 } ) ∪ ( { 10 } × { 0 , … , 5 } ) . (66) {}\cup\bigl(\{9\}\times\{3,\ldots,7\}\bigr)\cup\bigl(\{10\}\times\{0,\ldots,5\}\bigr). \tag*{(66)} ∪ ( { 9 } × { 3 , … , 7 } ) ∪ ( { 10 } × { 0 , … , 5 } ) . (66)
On every listed closed cell,
min 1 ≤ j ≤ 4 H j j > 6 10000 , max 1 ≤ a ≤ b ≤ 4 E ‾ a b < 6 10 6 . \min_{1\le j\le4}H_{jj}>\frac{6}{10000},\qquad\max_{1\le a\le b\le4}\overline{E}_{ab}<\frac{6}{10^{6}}. 1 ≤ j ≤ 4 min H j j > 10000 6 , 1 ≤ a ≤ b ≤ 4 max E ab < 1 0 6 6 .
These bounds include planar, layer, dual, Bessel and rounding remainders. The coverage identity is
2 − 3 + 8 2 − 4 + 7 2 − 5 + 6 2 − 6 + 4 2 − 7 + 4 2 − 8 + 5 2 − 9 + 6 2 − 10 = 1. 2^{-3}+8\,2^{-4}+7\,2^{-5}+6\,2^{-6}+4\,2^{-7}+4\,2^{-8}+5\,2^{-9}+6\,2^{-10}=1. 2 − 3 + 8 2 − 4 + 7 2 − 5 + 6 2 − 6 + 4 2 − 7 + 4 2 − 8 + 5 2 − 9 + 6 2 − 10 = 1.
For ( d , k ) (d,k) ( d , k ) , ( e , ℓ ) (e,\ell) ( e , ℓ ) listed in (F1),
⋃ listed ( d , k ) I d , k = [ 11 / 10 , 217 / 50 ] , I d , k ∘ ∩ I e , ℓ ∘ = ∅ ( ( d , k ) ≠ ( e , ℓ ) ) . \bigcup_{\text{listed }(d,k)} I_{d,k} = [11/10,217/50],\qquad I^\circ_{d,k}\cap I^\circ_{e,\ell}=\varnothing\quad((d,k)\ne(e,\ell)). listed ( d , k ) ⋃ I d , k = [ 11/10 , 217/50 ] , I d , k ∘ ∩ I e , ℓ ∘ = ∅ (( d , k ) = ( e , ℓ )) .
Thus the complete transverse Hessian is positive throughout the rectangle.
For 11 / 10 ≤ V ≤ 89 / 50 11/10\le V\le89/50 11/10 ≤ V ≤ 89/50 and E ( U Z 3 , 3 ) = B ( Y ( V ) ) E(UZ^3,3)=B(Y(V)) E ( U Z 3 , 3 ) = B ( Y ( V )) , Lemma A.16 gives
− 85 4 ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 1275 4 { ( r − 2 / 3 ) 2 + ( x − 1 / 2 ) 2 } . -\frac{85}{4}(y^2+z^2)\le2V^2-\Gamma(Y)\le\frac{1275}{4}\{(r-2/\sqrt{3})^2+(x-1/2)^2\}. − 4 85 ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 4 1275 {( r − 2/ 3 ) 2 + ( x − 1/2 ) 2 } .
The radii (58) give both parameter bounds. At fixed t t t , put ξ = ( r , x , y , z ) \xi=(r,x,y,z) ξ = ( r , x , y , z ) , ξ 0 = ( 2 / 3 , 1 / 2 , 0 , 0 ) \xi_0=(2/\sqrt{3},1/2,0,0) ξ 0 = ( 2/ 3 , 1/2 , 0 , 0 ) and φ ( ξ ) = Φ V ( U ( t , ξ ) Z 3 ) \varphi(\xi)=\Phi_V(U(t,\xi)Z^3) φ ( ξ ) = Φ V ( U ( t , ξ ) Z 3 ) . Symmetry gives ∇ ξ φ ( ξ 0 ) = 0 \nabla_\xi\varphi(\xi_0)=0 ∇ ξ φ ( ξ 0 ) = 0 .
If ∇ ξ φ ( ξ ) = 0 \nabla_\xi\varphi(\xi)=0 ∇ ξ φ ( ξ ) = 0 , then
0 = ( ξ − ξ 0 ) ⋅ [ ∇ ξ φ ( ξ ) − ∇ ξ φ ( ξ 0 ) ] = ∫ 0 1 ( ξ − ξ 0 ) T ∇ ξ 2 φ ( ξ 0 + s ( ξ − ξ 0 ) ) ( ξ − ξ 0 ) d s > 0 \begin{aligned}
0=(\xi-\xi_0)\cdot[\nabla_\xi\varphi(\xi)-\nabla_\xi\varphi(\xi_0)] \\
&=\int_0^1(\xi-\xi_0)^T\nabla_\xi^2\varphi(\xi_0+s(\xi-\xi_0))(\xi-\xi_0)\,ds>0
\end{aligned} 0 = ( ξ − ξ 0 ) ⋅ [ ∇ ξ φ ( ξ ) − ∇ ξ φ ( ξ 0 )] = ∫ 0 1 ( ξ − ξ 0 ) T ∇ ξ 2 φ ( ξ 0 + s ( ξ − ξ 0 )) ( ξ − ξ 0 ) d s > 0
unless ξ = ξ 0 \xi=\xi_0 ξ = ξ 0 , since the rectangle is convex. Then B ( Y ) = B ( Y ( V ) ) B(Y)=B(Y(V)) B ( Y ) = B ( Y ( V )) and B t > 0 B_t>0 B t > 0 give Y = Y ( V ) Y=Y(V) Y = Y ( V ) . All gradients in this argument are in ξ = ( r , x , y , z ) \xi=(r,x,y,z) ξ = ( r , x , y , z ) at fixed t t t . This completes the proof of Lemma 6.5 . □ \square □
Lemma 6.6 (A height-two terminal region for the long SH profile). Use the proved endpoint comparisons and g g g , V ∗ V_* V ∗ from Lemma 6.3 . Put J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 ) ) J(\Lambda)=E(\Lambda,6)-g(E(\Lambda,3)) J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 )) . For Λ = U ( t , r , x , y , z ) Z 3 \Lambda=U(t,r,x,y,z)Z^3 Λ = U ( t , r , x , y , z ) Z 3 and V = V ∗ ( E ( Λ , 3 ) ) ∈ [ 11 / 10 , 89 / 50 ] V=V_*(E(\Lambda,3))\in[11/10,89/50] V = V ∗ ( E ( Λ , 3 )) ∈ [ 11/10 , 89/50 ] , assume
V 2 t 3 ≤ ( 9 / 10 ) 3 , V 2 / t 6 ≥ 64. V^2t^3\le(9/10)^3,\qquad V^2/t^6\ge64. V 2 t 3 ≤ ( 9/10 ) 3 , V 2 / t 6 ≥ 64. Then
Λ ∈ arg min ∣ Λ ′ ∣ = 1 J ( Λ ′ ) , J ( Λ ) < 0 ⟹ ⊥ , \Lambda\in\underset{|\Lambda'|=1}{\arg\min}\,J(\Lambda'),\qquad J(\Lambda)<0\quad\Longrightarrow\quad\perp, Λ ∈ ∣ Λ ′ ∣ = 1 arg min J ( Λ ′ ) , J ( Λ ) < 0 ⟹ ⊥ , J ≥ 0 , J ( Λ ) = 0 , B ( Y ( 11 / 10 ) ) < E ( Λ , 3 ) < B ( Y ( 89 / 50 ) ) ⟹ Λ ≅ H Y ( V ) . J\ge0,\qquad J(\Lambda)=0,\qquad B(Y(11/10))<E(\Lambda,3)<B(Y(89/50))
\quad\Longrightarrow\quad\Lambda\cong H_{Y(V)}. J ≥ 0 , J ( Λ ) = 0 , B ( Y ( 11/10 )) < E ( Λ , 3 ) < B ( Y ( 89/50 )) ⟹ Λ ≅ H Y ( V ) . Proof. Compact reduction and Proposition 5.1 give, at an interior moment,
Λ ∗ ∈ arg min J , V ∈ arg max v ∈ [ V 0 , V 1 ] [ f ( v ) + 2 v 2 E ( Λ ∗ , 3 ) ] ⟹ Λ ∗ ∈ arg min Φ V . \Lambda_*\in\arg\min J,\qquad V\in\underset{v\in[V_0,V_1]}{\arg\max}[f(v)+2v^2E(\Lambda_*,3)]
\quad\Longrightarrow\quad\Lambda_*\in\arg\min\Phi_V. Λ ∗ ∈ arg min J , V ∈ v ∈ [ V 0 , V 1 ] arg max [ f ( v ) + 2 v 2 E ( Λ ∗ , 3 )] ⟹ Λ ∗ ∈ arg min Φ V . The same implication holds at interior zeros after J ≥ 0 J\ge0 J ≥ 0 , while
inf J < 0 ⟹ arg min J ≠ ∅ . \inf J<0\quad\Longrightarrow\quad\arg\min J\ne\varnothing. inf J < 0 ⟹ arg min J = ∅ . For the physical lattice V 1 / 3 Λ ∗ V^{1/3}\Lambda_* V 1/3 Λ ∗ ,
A p h y s = V 2 / 3 t ≤ 9 / 10 , h p h y s = V 1 / 3 / t ≥ 2. A_{\mathrm{phys}}=V^{2/3}t\le9/10,\qquad h_{\mathrm{phys}}=V^{1/3}/t\ge2. A phys = V 2/3 t ≤ 9/10 , h phys = V 1/3 / t ≥ 2. The moment cap gives
2 ( min v ∈ Λ ∗ ∖ { 0 } ∣ v ∣ ) − 12 ≤ E ( Λ ∗ , 6 ) < 64 , min v ∈ Λ ∗ ∖ { 0 } ∣ v ∣ > 2 − 5 / 12 > 749 1000 . 2\left(\underset{v\in\Lambda_*\setminus\{0\}}{\min}|v|\right)^{-12}\le E(\Lambda_*,6)<64,\qquad
\underset{v\in\Lambda_*\setminus\{0\}}{\min}|v|>2^{-5/12}>\frac{749}{1000}. 2 ( v ∈ Λ ∗ ∖ { 0 } min ∣ v ∣ ) − 12 ≤ E ( Λ ∗ , 6 ) < 64 , v ∈ Λ ∗ ∖ { 0 } min ∣ v ∣ > 2 − 5/12 > 1000 749 . Since V > 1 V>1 V > 1 , min ( i , j ) ∈ Z 2 ∖ { ( 0 , 0 ) } V 1 / 3 ∣ U ( t , r , x , y , z ) ( i , j , 0 ) T ∣ > 7 / 10 \underset{(i,j)\in\mathbb Z^2\setminus\{(0,0)\}}{\min}V^{1/3}|U(t,r,x,y,z)(i,j,0)^T|>7/10 ( i , j ) ∈ Z 2 ∖ {( 0 , 0 )} min V 1/3 ∣ U ( t , r , x , y , z ) ( i , j , 0 ) T ∣ > 7/10 . Lemmas 3.15 and 3.17 give
Λ ∗ ≅ H Y . \Lambda_*\cong H_Y. Λ ∗ ≅ H Y . For this SH shape the parameter satisfies
Y = 3 h p h y s 2 2 A p h y s ≥ 20 3 9 > 11 10 . Y=\frac{\sqrt{3}h_{\mathrm{phys}}^2}{2A_{\mathrm{phys}}}\ge\frac{20\sqrt{3}}{9}>\frac{11}{10}. Y = 2 A phys 3 h phys 2 ≥ 9 20 3 > 10 11 .
The planar indices { ( i , j ) ∈ Z 2 : i 2 + i j + j 2 = 1 } \{(i,j) \in\mathbb{Z}^{2}: i^{2}+ij+j^{2}=1\} {( i , j ) ∈ Z 2 : i 2 + ij + j 2 = 1 } give
3 Y 4 ∑ ( i , j ) ∈ Z 2 i 2 + i j + j 2 = 1 1 = 9 Y 2 ≤ E ( H Y , 3 ) < 43 2 , 11 10 < Y < 43 9 < 6. \frac{3Y}{4}\sum_{\substack{(i,j)\in\mathbb{Z}^{2}\\i^{2}+ij+j^{2}=1}}1=\frac{9Y}{2}\le E(H_Y,3)<\frac{43}{2},\qquad\frac{11}{10}<Y<\frac{43}{9}<6. 4 3 Y ( i , j ) ∈ Z 2 i 2 + ij + j 2 = 1 ∑ 1 = 2 9 Y ≤ E ( H Y , 3 ) < 2 43 , 10 11 < Y < 9 43 < 6.
Moreover,
0 = Y ∂ Y Φ V ( H Y ) = A t ( Y ) − 2 V 2 B t ( Y ) ⟹ Γ ( Y ) = 2 V 2 , Y = Y ( V ) , J ( H Y ) = 0. 0=Y\partial_Y\Phi_V(H_Y)=A_t(Y)-2V^{2}B_t(Y)\implies\Gamma(Y)=2V^{2},\quad Y=Y(V),\quad J(H_Y)=0. 0 = Y ∂ Y Φ V ( H Y ) = A t ( Y ) − 2 V 2 B t ( Y ) ⟹ Γ ( Y ) = 2 V 2 , Y = Y ( V ) , J ( H Y ) = 0.
Under the height hypotheses and the strict interior moment condition,
Λ ∗ ∈ a r g m i n J , J ( Λ ∗ ) ≤ 0 ⟹ J ( Λ ∗ ) = 0 , Λ ∗ ≅ H Y ( V ) . \Lambda_*\in\mathop{\rm argmin}J,\quad J(\Lambda_*)\le0\implies J(\Lambda_*)=0,\quad\Lambda_*\cong H_Y(V). Λ ∗ ∈ argmin J , J ( Λ ∗ ) ≤ 0 ⟹ J ( Λ ∗ ) = 0 , Λ ∗ ≅ H Y ( V ) .
This excludes negative global minima in this region and identifies its eligible zero minima. This completes the proof of Lemma 6.6. □
Lemma 6.7 (Forward slope bounds and a large height-two tube). Use the area-aspect coordinates, Y ( V ) = Γ − 1 ( 2 V 2 ) Y(V)=\Gamma^{-1}(2V^{2}) Y ( V ) = Γ − 1 ( 2 V 2 ) , and Lemma A.15 . For Λ = U ( t , r , x , y , z ) Z 3 \Lambda=U(t,r,x,y,z)Z^{3} Λ = U ( t , r , x , y , z ) Z 3 , suppose
11 10 ≤ Y = 3 2 t 3 ≤ 217 50 , 11 10 ≤ V ≤ 89 50 , E ( Λ , 3 ) = B ( Y ( V ) ) , \frac{11}{10}\le Y=\frac{\sqrt{3}}{2t^{3}}\le\frac{217}{50},\qquad\frac{11}{10}\le V\le\frac{89}{50},\qquad E(\Lambda,3)=B(Y(V)), 10 11 ≤ Y = 2 t 3 3 ≤ 50 217 , 10 11 ≤ V ≤ 50 89 , E ( Λ , 3 ) = B ( Y ( V )) , max { ∣ r − 2 / 3 ∣ , ∣ x − 1 / 2 ∣ , ∣ y ∣ , ∣ z ∣ } ≤ 1 100 . \max\{|r-2/\sqrt{3}|,|x-1/2|,|y|,|z|\}\le\frac{1}{100}. max { ∣ r − 2/ 3 ∣ , ∣ x − 1/2∣ , ∣ y ∣ , ∣ z ∣ } ≤ 100 1 . If
∣ r − 2 / 3 ∣ ≤ d r , ∣ x − 1 / 2 ∣ ≤ d x , ∣ y ∣ ≤ d y , ∣ z ∣ ≤ d z , |r-2/\sqrt{3}|\le d_r,\qquad|x-1/2|\le d_x,\qquad|y|\le d_y,\qquad|z|\le d_z, ∣ r − 2/ 3 ∣ ≤ d r , ∣ x − 1/2∣ ≤ d x , ∣ y ∣ ≤ d y , ∣ z ∣ ≤ d z , then
Γ ( Y ) − 1275 4 ( d y 2 + d z 2 ) ≤ 2 V 2 ≤ Γ ( Y ) + 1275 4 ( d r 2 + d x 2 ) . (67) \Gamma(Y)-\frac{1275}{4}(d_y^{2}+d_z^{2})\le2V^{2}\le\Gamma(Y)+\frac{1275}{4}(d_r^{2}+d_x^{2}). \tag*{(67)} Γ ( Y ) − 4 1275 ( d y 2 + d z 2 ) ≤ 2 V 2 ≤ Γ ( Y ) + 4 1275 ( d r 2 + d x 2 ) . (67) For the whole outward area image [ Y − , Y + ] [Y_-,Y_+] [ Y − , Y + ] , use the node enclosures y j ∈ [ y ‾ j , y ‾ j ] y_j\in[\underline{y}_j,\overline{y}_j] y j ∈ [ y j , y j ] , Γ ( y j ) ∈ [ Γ ‾ j , Γ ‾ j ] \Gamma(y_j)\in[\underline{\Gamma}_j,\overline{\Gamma}_j] Γ ( y j ) ∈ [ Γ j , Γ j ] of Lemma A.14 . Put
i − = max { 0 } ∪ { j ∈ { 0 , … , 4096 } : y ‾ j ≤ Y − } , i_-=\max\{0\}\cup\{j\in\{0,\ldots,4096\}:\overline{y}_j\le Y_-\}, i − = max { 0 } ∪ { j ∈ { 0 , … , 4096 } : y j ≤ Y − } , i + = min { 4096 } ∪ { j ∈ { 0 , … , 4096 } : y ‾ j ≥ Y + } , i_+=\min\{4096\}\cup\{j\in\{0,\ldots,4096\}:\underline{y}_j\ge Y_+\}, i + = min { 4096 } ∪ { j ∈ { 0 , … , 4096 } : y j ≥ Y + } , ( g − , g + ) = ( Γ ‾ i − , Γ ‾ i + ) . (g_-,g_+)=(\underline{\Gamma}_{i_-},\overline{\Gamma}_{i_+}). ( g − , g + ) = ( Γ i − , Γ i + ) . Then g − ≤ Γ ( Y − ) ≤ Γ ( Y + ) ≤ g + g_-\le\Gamma(Y_-)\le\Gamma(Y_+)\le g_+ g − ≤ Γ ( Y − ) ≤ Γ ( Y + ) ≤ g + and every eligible 2 V 2 2V^{2} 2 V 2 belongs to
[ g − − 1275 4 ( d y 2 + d z 2 ) , g + + 1275 4 ( d r 2 + d x 2 ) ] . (68) \left[g_--\frac{1275}{4}(d_y^{2}+d_z^{2}),\ g_++\frac{1275}{4}(d_r^{2}+d_x^{2})\right]. \tag*{(68)} [ g − − 4 1275 ( d y 2 + d z 2 ) , g + + 4 1275 ( d r 2 + d x 2 ) ] . (68) Intersect with the inherited coefficient interval; exclude only an empty intersection under the exact moment equation.
For Λ ∈ a r g m i n ∣ Λ ′ ∣ = 1 J ( Λ ′ ) \Lambda\in\mathop{\rm argmin}_{|\Lambda'|=1}J(\Lambda') Λ ∈ argmin ∣ Λ ′ ∣ = 1 J ( Λ ′ ) with E ( Λ , 6 ) < 64 E(\Lambda,6)<64 E ( Λ , 6 ) < 64 and E ( Λ , 3 ) < 43 / 2 E(\Lambda,3)<43/2 E ( Λ , 3 ) < 43/2 , assume
u + 3 2 ≤ ( 9 10 ) 3 , ℓ 2 t + 6 ≥ 64 , (69) \frac{u_+^{3}}{2}\le\left(\frac{9}{10}\right)^{3},\qquad\frac{\ell}{2t_+^{6}}\ge64, \tag*{(69)} 2 u + 3 ≤ ( 10 9 ) 3 , 2 t + 6 ℓ ≥ 64 , (69) where t + t_+ t + is the upper area endpoint and 2 V 2 ∈ [ ℓ , u ] 2V^{2}\in[\ell,u] 2 V 2 ∈ [ ℓ , u ] is the nonempty contracted interval. Then Λ ≅ H Y ( V ) \Lambda\cong H_Y(V) Λ ≅ H Y ( V ) and J ( Λ ) = 0 J(\Lambda)=0 J ( Λ ) = 0 . The same conclusion holds throughout
41 10 ≤ Y ≤ 217 50 , ∣ r − 2 / 3 ∣ , ∣ x − 1 / 2 ∣ , ∣ y ∣ , ∣ z ∣ ≤ 1 100 . (70) \frac{41}{10}\le Y\le\frac{217}{50},\qquad|r-2/\sqrt{3}|,|x-1/2|,|y|,|z|\le\frac{1}{100}. \tag*{(70)} 10 41 ≤ Y ≤ 50 217 , ∣ r − 2/ 3 ∣ , ∣ x − 1/2∣ , ∣ y ∣ , ∣ z ∣ ≤ 100 1 . (70)
Proof. The separate planar and layer moment comparison of Lemma 6.4 , with M = 150 M=150 M = 150 , gives (G1). Monotonicity and the outward node bounds give (G2), including the one-sided boundary cases.
For Λ ∈ a r g m i n ∣ Λ ′ ∣ = 1 J ( Λ ′ ) \Lambda\in\mathop{\rm argmin}_{|\Lambda'|=1}J(\Lambda') Λ ∈ argmin ∣ Λ ′ ∣ = 1 J ( Λ ′ ) , the supporting-line identity and
A p h y s 3 = V 2 t 3 , h p h y s 6 = V 2 t 6 A_{\mathrm{phys}}^{3}=V^{2}t^{3},\qquad h_{\mathrm{phys}}^{6}=\frac{V^{2}}{t^{6}} A phys 3 = V 2 t 3 , h phys 6 = t 6 V 2
make (G3) exactly the height-two hypotheses of Lemma 6.6.
At Y = 41 / 10 Y = 41/10 Y = 41/10 , the complete expressions and tails of Lemma 6.2 give
6 < ∂ log Y E ( H 41 / 10 , 6 ) ∂ log Y E ( H 41 / 10 , 3 ) < 752 125 . 6 < \frac{\partial_{\log Y} E(H_{41/10},6)}{\partial_{\log Y} E(H_{41/10},3)} < \frac{752}{125}. 6 < ∂ l o g Y E ( H 41/10 , 3 ) ∂ l o g Y E ( H 41/10 , 6 ) < 125 752 .
The complete rational cube-and-tail evaluation has lower endpoint > 6.0155932456 > 6.0155932456 > 6.0155932456 . Lemma A.15 gives ∣ 2 V 2 − Γ ( Y ) ∣ < 51 / 400 \lvert2V^2-\Gamma(Y)\rvert< 51/400 ∣ 2 V 2 − Γ ( Y )∣ < 51/400 , hence 2 V 2 > 2349 / 400 2V^2 > 2349/400 2 V 2 > 2349/400 . Together with 2 V 2 ≤ 7921 / 1250 2V^2 \le7921/1250 2 V 2 ≤ 7921/1250 , this yields
A p h y s 3 = V 2 3 2 Y < ( 9 / 10 ) 3 , h p h y s 6 = 4 V 2 Y 2 3 > 64. A_{\mathrm{phys}}^3=\frac{V^2\sqrt{3}}{2Y}< (9/10)^3,\qquad h_{\mathrm{phys}}^6=\frac{4V^2Y^2}{3}>64. A phys 3 = 2 Y V 2 3 < ( 9/10 ) 3 , h phys 6 = 3 4 V 2 Y 2 > 64.
Thus the entire large tube (G4) satisfies the height-two recognition conditions. The argument uses global minimality; it does not assert recognition of arbitrary local minima. This completes the proof of Lemma 6.7. □
Lemma 6.8 (A uniform comparison for all layer shifts). For t , r > 0 t,r>0 t , r > 0 and x , y , z ∈ R x,y,z\in\mathbb{R} x , y , z ∈ R , put
U ( t , r , x , y , z ) = ( t r t r x t r y 0 t / r t / r z 0 0 t − 1 ) , det U = 1 , Y = 3 2 t 3 . U(t,r,x,y,z)=
\begin{pmatrix}
\sqrt{tr} & \sqrt{tr}x & \sqrt{tr}y\\
0 & \sqrt{t/r} & \sqrt{t/r}z\\
0 & 0 & t^{-1}
\end{pmatrix},
\qquad\det U=1,\qquad Y=\frac{\sqrt{3}}{2t^3}. U ( t , r , x , y , z ) = t r 0 0 t r x t / r 0 t r y t / r z t − 1 , det U = 1 , Y = 2 t 3 3 . Suppose V ≥ 11 / 10 V\ge11/10 V ≥ 11/10 and either
{ 11 / 10 ≤ Y ≤ 217 / 50 , 2 / 3 − 1 / 50 ≤ r ≤ 2 / 3 , 12 / 25 ≤ x ≤ 1 / 2 , or 6 / 5 ≤ Y ≤ 217 / 50 , 2 / 3 − 1 / 20 ≤ r ≤ 2 / 3 , 9 / 20 ≤ x ≤ 1 / 2. (71) \left\{
\begin{aligned}
11/10&\le Y\le217/50,\qquad2/\sqrt{3}-1/50\le r\le2/\sqrt{3},\qquad12/25\le x\le1/2, \quad\text{or}\\
6/5&\le Y\le217/50,\qquad2/\sqrt{3}-1/20\le r\le2/\sqrt{3},\qquad9/20\le x\le1/2.
\end{aligned}
\right. \tag*{(71)} { 11/10 6/5 ≤ Y ≤ 217/50 , 2/ 3 − 1/50 ≤ r ≤ 2/ 3 , 12/25 ≤ x ≤ 1/2 , or ≤ Y ≤ 217/50 , 2/ 3 − 1/20 ≤ r ≤ 2/ 3 , 9/20 ≤ x ≤ 1/2. (71) Then, for every y , z ∈ R y,z\in\mathbb{R} y , z ∈ R ,
Φ V ( U ( t , r , x , y , z ) Z 3 ) − Φ V ( U ( t , r , x , 0 , 0 ) Z 3 ) ≥ 4 10 t 7 { 2 − cos ( 2 π ( y − x z ) ) − cos ( 2 π z ) } . (72) \Phi_V(U(t,r,x,y,z)\mathbb{Z}^3)-\Phi_V(U(t,r,x,0,0)\mathbb{Z}^3)
\ge\frac{4}{10t^7}\{2-\cos(2\pi(y-xz))-\cos(2\pi z)\}. \tag*{(72)} Φ V ( U ( t , r , x , y , z ) Z 3 ) − Φ V ( U ( t , r , x , 0 , 0 ) Z 3 ) ≥ 10 t 7 4 { 2 − cos ( 2 π ( y − x z )) − cos ( 2 π z )} . (72) In particular,
argmin ( y , z ) ∈ R 2 Φ V ( U ( t , r , x , y , z ) Z 3 ) = { ( y , z ) : ( y − x z , z ) ∈ Z 2 } , \underset{(y,z)\in\mathbb{R}^2}{\operatorname{argmin}}\ \Phi_V(U(t,r,x,y,z)\mathbb{Z}^3)
=\{(y,z):(y-xz,z)\in\mathbb{Z}^2\}, ( y , z ) ∈ R 2 argmin Φ V ( U ( t , r , x , y , z ) Z 3 ) = {( y , z ) : ( y − x z , z ) ∈ Z 2 } , ( y − x z , z ) ∈ Z 2 , − 1 2 ≤ y ≤ 1 2 , 0 ≤ z ≤ 1 2 ⟹ ( y , z ) = ( 0 , 0 ) . (y-xz,z)\in\mathbb{Z}^2,\quad-\frac{1}{2}\le y\le\frac{1}{2},\quad0\le z\le\frac{1}{2}
\Longrightarrow(y,z)=(0,0). ( y − x z , z ) ∈ Z 2 , − 2 1 ≤ y ≤ 2 1 , 0 ≤ z ≤ 2 1 ⟹ ( y , z ) = ( 0 , 0 ) .
Proof. Write a = 2 π ( y − x z ) a=2\pi(y-xz) a = 2 π ( y − x z ) , b = 2 π z b=2\pi z b = 2 π z , and δ ( v ) = 1 − cos v \delta(v)=1-\cos v δ ( v ) = 1 − cos v . For m , n ∈ Z m,n\in\mathbb{Z} m , n ∈ Z let
d = d ( m , n ) = { min ( ∣ m ∣ , ∣ n ∣ ) , m n > 0 , 0 , m n ≤ 0. d=d(m,n)=
\begin{cases}
\min(\lvert m\rvert,\lvert n\rvert), & mn>0,\\
0, & mn\le0.
\end{cases} d = d ( m , n ) = { min (∣ m ∣ , ∣ n ∣) , 0 , mn > 0 , mn ≤ 0.
The telescoping identity for e i v − 1 e^{iv}-1 e i v − 1 , followed by Cauchy–Schwarz, gives the two inequalities
δ ( k ( m a + n b ) ) ≤ k 2 ( ∣ m ∣ + ∣ n ∣ ) { ∣ m ∣ δ ( a ) + ∣ n ∣ δ ( b ) } , \delta(k(ma+nb))\le k^2(\lvert m\rvert+\lvert n\rvert)\{\lvert m\rvert\delta(a)+\lvert n\rvert\delta(b)\}, δ ( k ( ma + nb )) ≤ k 2 (∣ m ∣ + ∣ n ∣) {∣ m ∣ δ ( a ) + ∣ n ∣ δ ( b )} ,
δ ( k ( m a + n b ) ) ≤ k 2 ( ∣ m ∣ + ∣ n ∣ − d ) { ( ∣ m ∣ − d ) δ ( a ) + ( ∣ n ∣ − d ) δ ( b ) + d δ ( a + b ) } . (73) \delta(k(ma+nb))\le k^2(\lvert m\rvert+\lvert n\rvert-d)\{(\lvert m\rvert-d)\delta(a)+(\lvert n\rvert-d)\delta(b)+d\delta(a+b)\}. \tag*{(73)} δ ( k ( ma + nb )) ≤ k 2 (∣ m ∣ + ∣ n ∣ − d ) {(∣ m ∣ − d ) δ ( a ) + (∣ n ∣ − d ) δ ( b ) + d δ ( a + b )} . (73)
They hold for all real a , b a,b a , b and every integer k ≥ 1 k\ge1 k ≥ 1 .
For V = 11 / 10 V=11/10 V = 11/10 , ( m , n ) ∈ Z 2 ∖ { ( 0 , 0 ) } (m,n)\in\mathbb{Z}^2\setminus\{(0,0)\} ( m , n ) ∈ Z 2 ∖ {( 0 , 0 )} and k ∈ Z ≥ 1 k\in\mathbb{Z}_{\ge1} k ∈ Z ≥ 1 , put
R m , n 2 = m 2 t r + r t ( n − x m ) 2 , R_{m,n}^2=\frac{m^2}{tr}+\frac{r}{t}(n-xm)^2, R m , n 2 = t r m 2 + t r ( n − x m ) 2 ,
C k , m , n = π 6 60 ( t R m , n k ) 5 K 5 ( 2 π k R m , n t ) − 121 50 π 3 ( t R m , n k ) 2 K 2 ( 2 π k R m , n t ) . C_{k,m,n}=\frac{\pi^6}{60}\left(\frac{tR_{m,n}}{k}\right)^5K_5\left(\frac{2\pi kR_{m,n}}{t}\right)
-\frac{121}{50}\pi^3\left(\frac{tR_{m,n}}{k}\right)^2K_2\left(\frac{2\pi kR_{m,n}}{t}\right). C k , m , n = 60 π 6 ( k t R m , n ) 5 K 5 ( t 2 π k R m , n ) − 50 121 π 3 ( k t R m , n ) 2 K 2 ( t 2 π k R m , n ) .
Poisson summation gives the exact shift difference
Φ 11 / 10 ( U ( t , r , x , y , z ) Z 3 ) − Φ 11 / 10 ( U ( t , r , x , 0 , 0 ) Z 3 ) = − 4 t ∑ k = 1 ∞ ∑ m , n ∈ Z m > 0 or m = 0 , n > 0 C k , m , n δ ( k ( m a + n b ) ) . (74) \Phi_{11/10}(U(t,r,x,y,z)\mathbb{Z}^{3})-\Phi_{11/10}(U(t,r,x,0,0)\mathbb{Z}^{3})
= -\frac{4}{t}\sum_{k=1}^{\infty}\sum_{\substack{m,n\in\mathbb{Z}\\m>0\ \text{or}\ m=0,n>0}} C_{k,m,n}\delta(k(ma+nb)). \tag*{(74)} Φ 11/10 ( U ( t , r , x , y , z ) Z 3 ) − Φ 11/10 ( U ( t , r , x , 0 , 0 ) Z 3 ) = − t 4 k = 1 ∑ ∞ m , n ∈ Z m > 0 or m = 0 , n > 0 ∑ C k , m , n δ ( k ( ma + nb )) . (74)
The three modes ( k , m , n ) = ( 1 , 1 , 0 ) , ( 1 , 0 , 1 ) , ( 1 , 1 , 1 ) (k,m,n)=(1,1,0),(1,0,1),(1,1,1) ( k , m , n ) = ( 1 , 1 , 0 ) , ( 1 , 0 , 1 ) , ( 1 , 1 , 1 ) provide the defects δ ( a ) , δ ( b ) , δ ( a + b ) \delta(a),\delta(b),\delta(a+b) δ ( a ) , δ ( b ) , δ ( a + b ) . Throughout both rectangles in (71) ,
7 12 < t < 37 40 , 11 10 < r < 29 25 , 449 1000 < x ≤ 1 2 . \frac{7}{12}<t<\frac{37}{40},\qquad\frac{11}{10}<r<\frac{29}{25},\qquad\frac{449}{1000}<x\leq\frac{1}{2}. 12 7 < t < 40 37 , 10 11 < r < 25 29 , 1000 449 < x ≤ 2 1 .
For j = max ( ∣ m ∣ , ∣ n ∣ ) j=\max(|m|,|n|) j = max ( ∣ m ∣ , ∣ n ∣ ) , completing the square gives
R m , n 2 ≥ j 2 t min { 1 r , r 1 + r 2 x 2 } > ( 19 j 20 ) 2 . R_{m,n}^{2}\geq\frac{j^{2}}{t}\min\left\{\frac{1}{r},\frac{r}{1+r^{2}x^{2}}\right\}>\left(\frac{19j}{20}\right)^{2}. R m , n 2 ≥ t j 2 min { r 1 , 1 + r 2 x 2 r } > ( 20 19 j ) 2 .
For ( m , n ) ∈ { − 1 , 0 , 1 } 2 ∖ { ( 0 , 0 ) } (m,n)\in\{-1,0,1\}^{2}\setminus\{(0,0)\} ( m , n ) ∈ { − 1 , 0 , 1 } 2 ∖ {( 0 , 0 )} , R m , n > 27 / 25 R_{m,n}>27/25 R m , n > 27/25 . Thus set
R 1 = 27 25 , R j = 19 j 20 ( j ≥ 2 ) . R_{1}=\frac{27}{25},\qquad R_{j}=\frac{19j}{20}\quad(j\geq2). R 1 = 25 27 , R j = 20 19 j ( j ≥ 2 ) .
Retain
I = { ( k , m , n ) ∈ Z 3 : 1 ≤ k ≤ 3 , ∣ m ∣ , ∣ n ∣ ≤ 4 , m > 0 or m = 0 < n } , # I = 120. \mathcal{I}=\{(k,m,n)\in\mathbb{Z}^{3}:1\leq k\leq3,\ |m|,|n|\leq4,\ m>0\ \text{or}\ m=0<n\},\qquad\#\mathcal{I}=120. I = {( k , m , n ) ∈ Z 3 : 1 ≤ k ≤ 3 , ∣ m ∣ , ∣ n ∣ ≤ 4 , m > 0 or m = 0 < n } , # I = 120.
Every omitted index has 2 π k R m , n / t > 28 2\pi kR_{m,n}/t>28 2 π k R m , n / t > 28 . The positive part of its coefficient is at most its repulsive part; Lemma 3.1 yields
( C k , m , n ) + ≤ π 6 120 P 5 ( 28 ) t + 11 / 2 R m , n 9 / 2 k − 11 / 2 e − 2 π k R m , n / t + , (C_{k,m,n})_{+}\leq\frac{\pi^{6}}{120}P_{5}(28)t_{+}^{11/2}R_{m,n}^{9/2}k^{-11/2}e^{-2\pi kR_{m,n}/t_{+}}, ( C k , m , n ) + ≤ 120 π 6 P 5 ( 28 ) t + 11/2 R m , n 9/2 k − 11/2 e − 2 π k R m , n / t + ,
Here t + = sup t t_{+}=\sup t t + = sup t , r − = inf r r_{-}=\inf r r − = inf r and r + = sup r r_{+}=\sup r r + = sup r on the current closed input box. Each of the three integer weights in (73) is at most
( ∣ m ∣ + ∣ n ∣ ) 2 ≤ 2 ( m 2 + n 2 ) ≤ 4 { m 2 + ( n − x m ) 2 } ≤ 4 t + ( r + + r − − 1 ) R m , n 2 . (|m|+|n|)^{2}\leq2(m^{2}+n^{2})\leq4\{m^{2}+(n-xm)^{2}\}\leq4t_{+}(r_{+}+r_{-}^{-1})R_{m,n}^{2}. ( ∣ m ∣ + ∣ n ∣ ) 2 ≤ 2 ( m 2 + n 2 ) ≤ 4 { m 2 + ( n − x m ) 2 } ≤ 4 t + ( r + + r − − 1 ) R m , n 2 .
All omitted contributions to each unscaled defect budget are therefore bounded above by
T = π 6 60 P 5 ( 28 ) t + 13 / 2 ( r + + r − − 1 ) S ( 13 / 2 , 7 / 2 ; 19 / 20 , 27 / 25 , 3 , 4 , t + ) , (75) T=\frac{\pi^{6}}{60}P_{5}(28)t_{+}^{13/2}(r_{+}+r_{-}^{-1})S(13/2,7/2;19/20,27/25,3,4,t_{+}), \tag*{(75)} T = 60 π 6 P 5 ( 28 ) t + 13/2 ( r + + r − − 1 ) S ( 13/2 , 7/2 ; 19/20 , 27/25 , 3 , 4 , t + ) , (75)
using the common complete shell sum (65) . Here J = 5 J=5 J = 5 , L = 4 L=4 L = 4 and
η 13 / 2 = ( 6 / 5 ) 15 / 2 e − 2 π ( 19 / 20 ) / t + < 1. \eta_{13/2}=(6/5)^{15/2}e^{-2\pi(19/20)/t_{+}}<1. η 13/2 = ( 6/5 ) 15/2 e − 2 π ( 19/20 ) / t + < 1.
On the omitted region, ∂ R log ( R 13 / 2 e − 2 π k R / t + ) = ( 13 / 2 − 2 π k R / t + ) / R < 0 \partial_{R}\log(R^{13/2}e^{-2\pi kR/t_{+}})=(13/2-2\pi kR/t_{+})/R<0 ∂ R log ( R 13/2 e − 2 π k R / t + ) = ( 13/2 − 2 π k R / t + ) / R < 0 . The half-lattice has half the signed shell sum, giving the coefficient π 6 / 60 \pi^{6}/60 π 6 /60 in (75) . These bounds are independent of the shift and its phase.
The two residual defect vectors from (73) are
b 2 = ( − C 1 , 1 , 0 , − C 1 , 0 , 1 ) − T ( 1 , 1 ) − ∑ ( k , m , n ) ∈ I k 2 ( C k , m , n ) + ( ∣ m ∣ + ∣ n ∣ ) ( ∣ m ∣ , ∣ n ∣ ) , \begin{aligned}
\mathbf{b}_{2}&=(-C_{1,1,0},-C_{1,0,1})-T(1,1)\\
&\quad-\sum_{(k,m,n)\in\mathcal{I}}k^{2}(C_{k,m,n})_{+}(|m|+|n|)(|m|,|n|),
\end{aligned} b 2 = ( − C 1 , 1 , 0 , − C 1 , 0 , 1 ) − T ( 1 , 1 ) − ( k , m , n ) ∈ I ∑ k 2 ( C k , m , n ) + ( ∣ m ∣ + ∣ n ∣ ) ( ∣ m ∣ , ∣ n ∣ ) ,
b 3 = ( − C 1 , 1 , 0 , − C 1 , 0 , 1 , − C 1 , 1 , 1 ) − T ( 1 , 1 , 1 ) − ∑ ( k , m , n ) ∈ I k 2 ( C k , m , n ) + ( ∣ m ∣ + ∣ n ∣ − d ) ( ∣ m ∣ − d , ∣ n ∣ − d , d ) . \begin{aligned}
\mathbf{b}_{3}&=(-C_{1,1,0},-C_{1,0,1},-C_{1,1,1})-T(1,1,1)\\
&\quad-\sum_{(k,m,n)\in\mathcal{I}}k^{2}(C_{k,m,n})_{+}(|m|+|n|-d)(|m|-d,|n|-d,d).
\end{aligned} b 3 = ( − C 1 , 1 , 0 , − C 1 , 0 , 1 , − C 1 , 1 , 1 ) − T ( 1 , 1 , 1 ) − ( k , m , n ) ∈ I ∑ k 2 ( C k , m , n ) + ( ∣ m ∣ + ∣ n ∣ − d ) ( ∣ m ∣ − d , ∣ n ∣ − d , d ) .
With underlines denoting complete interval lower endpoints, every accepted closed cell satisfies
max { min 1 ≤ j ≤ 2 ( b ‾ 2 ) j , min 1 ≤ j ≤ 3 ( b ‾ 3 ) j } > 10 − 7 . \max\left\{\min_{1\leq j\leq2}(\underline{b}_{2})_{j},\min_{1\leq j\leq3}(\underline{b}_{3})_{j}\right\}>10^{-7}. max { 1 ≤ j ≤ 2 min ( b 2 ) j , 1 ≤ j ≤ 3 min ( b 3 ) j } > 1 0 − 7 .
For the narrow rectangle use the closed cells
[ 11 10 + 81 k 25 2 d , 11 10 + 81 ( k + 1 ) 25 2 d ] \left[\frac{11}{10}+\frac{81k}{25\,2^d},\frac{11}{10}+\frac{81(k+1)}{25\,2^d}\right] [ 10 11 + 25 2 d 81 k , 10 11 + 25 2 d 81 ( k + 1 ) ]
with
( d , k ) = ( 3 , 7 ) , d = 4 , 5 ≤ k ≤ 13 , d = 5 , 4 ≤ k ≤ 9 , (d,k)=(3,7),\quad d=4,\ 5\le k\le13,\quad d=5,\ 4\le k\le9, ( d , k ) = ( 3 , 7 ) , d = 4 , 5 ≤ k ≤ 13 , d = 5 , 4 ≤ k ≤ 9 ,
d = 6 , 3 ≤ k ≤ 7 , d = 7 , 2 ≤ k ≤ 5 , d = 8 , 0 ≤ k ≤ 3. d=6,\ 3\le k\le7,\quad d=7,\ 2\le k\le5,\quad d=8,\ 0\le k\le3. d = 6 , 3 ≤ k ≤ 7 , d = 7 , 2 ≤ k ≤ 5 , d = 8 , 0 ≤ k ≤ 3.
For the broad rectangle take
[ 6 5 + 157 k 50 2 d , 6 5 + 157 ( k + 1 ) 50 2 d ] \left[\frac{6}{5}+\frac{157k}{50\,2^d},\frac{6}{5}+\frac{157(k+1)}{50\,2^d}\right] [ 5 6 + 50 2 d 157 k , 5 6 + 50 2 d 157 ( k + 1 ) ]
with
d = 4 , 14 ≤ k ≤ 15 , d = 5 , 7 ≤ k ≤ 27 , d = 6 , 6 ≤ k ≤ 13 , d=4,\ 14\le k\le15,\quad d=5,\ 7\le k\le27,\quad d=6,\ 6\le k\le13, d = 4 , 14 ≤ k ≤ 15 , d = 5 , 7 ≤ k ≤ 27 , d = 6 , 6 ≤ k ≤ 13 ,
d = 7 , 5 ≤ k ≤ 11 , d = 8 , 4 ≤ k ≤ 9 , d = 9 , 2 ≤ k ≤ 7 , d = 10 , 0 ≤ k ≤ 3. d=7,\ 5\le k\le11,\quad d=8,\ 4\le k\le9,\quad d=9,\ 2\le k\le7,\quad d=10,\ 0\le k\le3. d = 7 , 5 ≤ k ≤ 11 , d = 8 , 4 ≤ k ≤ 9 , d = 9 , 2 ≤ k ≤ 7 , d = 10 , 0 ≤ k ≤ 3.
On each whole cell enclose t = ( 3 / ( 2 Y ) ) 1 / 3 t=(\sqrt{3}/(2Y))^{1/3} t = ( 3 / ( 2 Y ) ) 1/3 and use the entire planar rectangle. Evaluate I \mathcal{I} I lexicographically in ( m , n ) (m,n) ( m , n ) , then k = 1 , 2 , 3 k=1,2,3 k = 1 , 2 , 3 , using the signed Bessel bounds, radial intersections, outward binary64 arithmetic and checked radicals of Lemma 6.5 . Include (75) in every component. The finite records are
narrow: ( # nodes , # leaves ) = ( 57 , 29 ) , accepted components > 5 ⋅ 10 − 7 , \text{narrow:}\quad(\#\text{nodes},\#\text{leaves})=(57,29),\qquad\text{accepted components}>5\cdot10^{-7}, narrow: ( # nodes , # leaves ) = ( 57 , 29 ) , accepted components > 5 ⋅ 1 0 − 7 ,
broad: ( # nodes , # leaves ) = ( 107 , 54 ) , accepted components > 4 ⋅ 10 − 7 . \text{broad:}\quad(\#\text{nodes},\#\text{leaves})=(107,54),\qquad\text{accepted components}>4\cdot10^{-7}. broad: ( # nodes , # leaves ) = ( 107 , 54 ) , accepted components > 4 ⋅ 1 0 − 7 .
The exact coverage identities are
2 − 3 + 9 2 − 4 + 6 2 − 5 + 5 2 − 6 + 4 2 − 7 + 4 2 − 8 = 1 , 2^{-3}+9\,2^{-4}+6\,2^{-5}+5\,2^{-6}+4\,2^{-7}+4\,2^{-8}=1, 2 − 3 + 9 2 − 4 + 6 2 − 5 + 5 2 − 6 + 4 2 − 7 + 4 2 − 8 = 1 ,
2 2 − 4 + 21 2 − 5 + 8 2 − 6 + 7 2 − 7 + 6 2 − 8 + 6 2 − 9 + 4 2 − 10 = 1. 2\,2^{-4}+21\,2^{-5}+8\,2^{-6}+7\,2^{-7}+6\,2^{-8}+6\,2^{-9}+4\,2^{-10}=1. 2 2 − 4 + 21 2 − 5 + 8 2 − 6 + 7 2 − 7 + 6 2 − 8 + 6 2 − 9 + 4 2 − 10 = 1.
The listed closed cells are consecutive with disjoint interiors. The budget inequalities and (73) give (72) at V = 11 / 10 V=11/10 V = 11/10 .
Poisson summation gives, for every shift,
E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) − E ( U ( t , r , x , y , z ) Z 3 , 3 ) ≥ 0. E(U(t,r,x,0,0)\mathbb{Z}^{3},3)-E(U(t,r,x,y,z)\mathbb{Z}^{3},3)\ge0. E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) − E ( U ( t , r , x , y , z ) Z 3 , 3 ) ≥ 0.
Since 2 { V 2 − ( 11 / 10 ) 2 } ≥ 0 2\{V^{2}-(11/10)^{2}\}\ge0 2 { V 2 − ( 11/10 ) 2 } ≥ 0 , the bound holds for V ≥ 11 / 10 V\ge11/10 V ≥ 11/10 . Moreover,
δ ( a ) + δ ( b ) = 0 ⟺ ( y − x z , z ) ∈ Z 2 ⟹ U ( t , r , x , y , z ) Z 3 = U ( t , r , x , 0 , 0 ) Z 3 . \delta(a)+\delta(b)=0\Longleftrightarrow(y-xz,z)\in\mathbb{Z}^{2}\Longrightarrow U(t,r,x,y,z)\mathbb{Z}^{3}=U(t,r,x,0,0)\mathbb{Z}^{3}. δ ( a ) + δ ( b ) = 0 ⟺ ( y − x z , z ) ∈ Z 2 ⟹ U ( t , r , x , y , z ) Z 3 = U ( t , r , x , 0 , 0 ) Z 3 .
Together with (72) , this gives the stated equality set. This completes the proof of Lemma 6.8 . □
Lemma 6.9 (A moment-based area–aspect root domain). Put V 0 = 11 / 10 V_0=11/10 V 0 = 11/10 , V 1 = 89 / 50 V_1=89/50 V 1 = 89/50 , and let f ( V ) = E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) f(V)=E(H_Y(V),6)-2V^2E(H_Y(V),3) f ( V ) = E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) be the reference value of Lemma 6.2 . Assume the proved global comparisons and uniqueness at these two endpoints. For the profile
J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 ) ) , g ( B ) = max V 0 ≤ V ≤ V 1 { f ( V ) + 2 V 2 B } , J(\Lambda)=E(\Lambda,6)-g(E(\Lambda,3)),\qquad
g(B)=\max_{V_0\le V\le V_1}\{f(V)+2V^2B\}, J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 )) , g ( B ) = V 0 ≤ V ≤ V 1 max { f ( V ) + 2 V 2 B } , the conditions ∣ Λ ∣ = 1 |\Lambda|=1 ∣Λ∣ = 1 and J ( Λ ) = min ∣ Λ ′ ∣ = 1 J ( Λ ′ ) < 0 J(\Lambda)=\min_{|\Lambda'|=1}J(\Lambda')<0 J ( Λ ) = min ∣ Λ ′ ∣ = 1 J ( Λ ′ ) < 0 imply Λ ≅ U ( η ) Z 3 \Lambda\cong U(\eta)\mathbb{Z}^{3} Λ ≅ U ( η ) Z 3 , with Korkine–Zolotarev coordinates from Lemma 6.1 in
12 25 ≤ t ≤ 8 7 , 3 5 ≤ r ≤ 7 6 , 0 ≤ x , z ≤ 1 2 , − 1 2 ≤ y ≤ 1 2 . \frac{12}{25}\le t\le\frac{8}{7},\qquad
\frac{3}{5}\le r\le\frac{7}{6},\qquad
0\le x,z\le\frac{1}{2},\qquad
-\frac{1}{2}\le y\le\frac{1}{2}. 25 12 ≤ t ≤ 7 8 , 5 3 ≤ r ≤ 6 7 , 0 ≤ x , z ≤ 2 1 , − 2 1 ≤ y ≤ 2 1 . It satisfies the additional necessary conditions
E ( Λ , 3 ) < 43 2 , E ( Λ , 6 ) < 64 , t r > ( 749 1000 ) 2 , ( t r ) 3 ≤ 2 , E(\Lambda,3)<\frac{43}{2},\qquad E(\Lambda,6)<64,\qquad
tr>\left(\frac{749}{1000}\right)^2,\qquad(tr)^3\le2, E ( Λ , 3 ) < 2 43 , E ( Λ , 6 ) < 64 , t r > ( 1000 749 ) 2 , ( t r ) 3 ≤ 2 , r 2 ( 1 − x 2 ) ≤ 1 , t 3 ( 1 − z 2 ) ≤ r , r^2(1-x^2)\le1,\qquad t^3(1-z^2)\le r, r 2 ( 1 − x 2 ) ≤ 1 , t 3 ( 1 − z 2 ) ≤ r , and
∣ U ( η ) n ∣ 2 ≥ t r ( n ∈ Z 3 ∖ { 0 } ) . |U(\eta)n|^2\ge tr\qquad(n\in\mathbb{Z}^{3}\setminus\{0\}). ∣ U ( η ) n ∣ 2 ≥ t r ( n ∈ Z 3 ∖ { 0 }) . If J ≥ 0 J\ge0 J ≥ 0 , the same bounds hold when ∣ Λ ∣ = 1 |\Lambda|=1 ∣Λ∣ = 1 , J ( Λ ) = 0 J(\Lambda)=0 J ( Λ ) = 0 and B ( Y ( V 0 ) ) < E ( Λ , 3 ) < B ( Y ( V 1 ) ) B(Y(V_0))<E(\Lambda,3)<B(Y(V_1)) B ( Y ( V 0 )) < E ( Λ , 3 ) < B ( Y ( V 1 )) ; the endpoint shapes are already identified.
Proof. The gap reduction and the moment bounds (A.8) give E ( Λ , 3 ) < 43 / 2 E(\Lambda,3)<43/2 E ( Λ , 3 ) < 43/2 and E ( Λ , 6 ) < 64 E(\Lambda,6)<64 E ( Λ , 6 ) < 64 . In unit Korkine–Zolotarev coordinates,
a = t r = min n ∈ Z 3 ∖ { 0 } ∣ U ( η ) n ∣ , δ = t / r , a=\sqrt{tr}=\min_{n\in\mathbb{Z}^{3}\setminus\{0\}}|U(\eta)n|,\qquad\delta=\sqrt{t/r}, a = t r = n ∈ Z 3 ∖ { 0 } min ∣ U ( η ) n ∣ , δ = t / r ,
2 a − 12 ≤ E ( Λ , 6 ) < 64 , a > 2 − 5 / 12 > 749 1000 . 2a^{-12}\leq E(\Lambda,6)<64,\qquad a>2^{-5/12}>\frac{749}{1000}. 2 a − 12 ≤ E ( Λ , 6 ) < 64 , a > 2 − 5/12 > 1000 749 .
The usual Korkine–Zolotarev and Hermite inequalities imply
δ ≥ 3 2 a , δ 2 ≤ 2 3 a , ( t − 1 ) 2 ≥ 3 2 a , a 6 ≤ 2. \delta\geq\frac{\sqrt{3}}{2}a,\qquad\delta^{2}\leq\frac{2}{\sqrt{3}a},\qquad(t^{-1})^{2}\geq\frac{\sqrt{3}}{2a},\qquad a^{6}\leq2. δ ≥ 2 3 a , δ 2 ≤ 3 a 2 , ( t − 1 ) 2 ≥ 2 a 3 , a 6 ≤ 2.
Consequently
t ≥ 3 2 ( 749 / 1000 ) 2 > 12 25 , r ≥ 3 2 ( 749 / 1000 ) 3 > 3 5 , t\geq\frac{\sqrt{3}}{2}(749/1000)^{2}>\frac{12}{25},\qquad r\geq\sqrt{\frac{\sqrt{3}}{2}(749/1000)^{3}}>\frac{3}{5}, t ≥ 2 3 ( 749/1000 ) 2 > 25 12 , r ≥ 2 3 ( 749/1000 ) 3 > 5 3 ,
r ≤ 2 3 < 7 6 , t 12 ≤ 128 27 < ( 8 / 7 ) 12 . r\leq\frac{2}{\sqrt{3}}<\frac{7}{6},\qquad t^{12}\leq\frac{128}{27}<(8/7)^{12}. r ≤ 3 2 < 6 7 , t 12 ≤ 27 128 < ( 8/7 ) 12 .
Since a 2 = t r a^{2}=tr a 2 = t r , the remaining projection bounds are
δ 2 + a 2 x 2 ≥ a 2 ⟺ r 2 ( 1 − x 2 ) ≤ 1 , \delta^{2}+a^{2}x^{2}\geq a^{2}\Longleftrightarrow r^{2}(1-x^{2})\leq1, δ 2 + a 2 x 2 ≥ a 2 ⟺ r 2 ( 1 − x 2 ) ≤ 1 ,
t − 2 + δ 2 z 2 ≥ δ 2 ⟺ t 3 ( 1 − z 2 ) ≤ r . t^{-2}+\delta^{2}z^{2}\geq\delta^{2}\Longleftrightarrow t^{3}(1-z^{2})\leq r. t − 2 + δ 2 z 2 ≥ δ 2 ⟺ t 3 ( 1 − z 2 ) ≤ r .
These give (L2)–(L3); closing the strict domain bounds retains every eligible lattice. This completes the proof of Lemma 6.9 . □
Lemma 6.10 (The complete long-profile comparison). With f f f as in Lemma 6.9 ,
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) ( ∣ Λ ∣ = 1 , V 0 ≤ V ≤ V 1 ) , E(\Lambda,6)-2V^{2}E(\Lambda,3)\geq f(V)\qquad(|\Lambda|=1,\ V_{0}\leq V\leq V_{1}), E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) ≥ f ( V ) ( ∣Λ∣ = 1 , V 0 ≤ V ≤ V 1 ) , and
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = f ( V ) ⟺ Λ ≅ H Y ( V ) . E(\Lambda,6)-2V^{2}E(\Lambda,3)=f(V)\Longleftrightarrow\Lambda\cong H_Y(V). E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = f ( V ) ⟺ Λ ≅ H Y ( V ) . Proof. By Proposition 5.1 and Lemma 6.9 ,
Λ ∗ ∈ argmin ∣ Λ ∣ = 1 J ( Λ ) , J ( Λ ∗ ) < 0 ⟹ Λ ∗ ≅ U ( η ) Z 3 , η ∈ ( L 1 ) . \Lambda_{*}\in\underset{|\Lambda|=1}{\operatorname{argmin}}\,J(\Lambda),\qquad J(\Lambda_{*})<0\Longrightarrow\Lambda_{*}\cong U(\eta)\mathbb{Z}^{3},\qquad\eta\in(\mathrm{L1}). Λ ∗ ∈ ∣Λ∣ = 1 argmin J ( Λ ) , J ( Λ ∗ ) < 0 ⟹ Λ ∗ ≅ U ( η ) Z 3 , η ∈ ( L1 ) . Numerical Lemma A.17 retains every such minimum; recognition forces J = 0 J=0 J = 0 . Its empty terminal sets give
J ( Λ ) ≥ 0 ( ∣ Λ ∣ = 1 ) , J(\Lambda)\geq0\qquad(|\Lambda|=1), J ( Λ ) ≥ 0 ( ∣Λ∣ = 1 ) , J ( Λ ) = 0 , V = V ∗ ( E ( Λ , 3 ) ) ⟹ Λ ≅ H Y ( V ) . J(\Lambda)=0,\quad V=V_{*}(E(\Lambda,3))\Longrightarrow\Lambda\cong H_Y(V). J ( Λ ) = 0 , V = V ∗ ( E ( Λ , 3 )) ⟹ Λ ≅ H Y ( V ) . Endpoint uniqueness handles the moment-band boundaries; the retained invariant handles interior zeros. By (46) ,
Φ V ( Λ ) − f ( V ) = J ( Λ ) + g ( E ( Λ , 3 ) ) − f ( V ) − 2 V 2 E ( Λ , 3 ) ≥ 0 , \Phi_V(\Lambda)-f(V)=J(\Lambda)+g(E(\Lambda,3))-f(V)-2V^{2}E(\Lambda,3)\geq0, Φ V ( Λ ) − f ( V ) = J ( Λ ) + g ( E ( Λ , 3 )) − f ( V ) − 2 V 2 E ( Λ , 3 ) ≥ 0 , Φ V ( Λ ) = f ( V ) ⟺ J ( Λ ) = 0 , V = V ∗ ( E ( Λ , 3 ) ) ⟺ Λ ≅ H Y ( V ) . \Phi_V(\Lambda)=f(V)\Longleftrightarrow J(\Lambda)=0,\quad V=V_{*}(E(\Lambda,3))\Longleftrightarrow\Lambda\cong H_Y(V). Φ V ( Λ ) = f ( V ) ⟺ J ( Λ ) = 0 , V = V ∗ ( E ( Λ , 3 )) ⟺ Λ ≅ H Y ( V ) . This completes the proof of Lemma 6.10 . □
Theorem 6.11 (The complete intervening simple-hexagonal phase). Let Y ( V ) = Γ − 1 ( 2 V 2 ) Y(V)=\Gamma^{-1}(2V^{2}) Y ( V ) = Γ − 1 ( 2 V 2 ) be the curve in Lemma 6.2 . For 11 / 10 ≤ V ≤ 89 / 50 11/10\leq V\leq89/50 11/10 ≤ V ≤ 89/50 ,
argmin ∣ L ∣ = V E ( L ) = { L : L ≅ V 1 / 3 H Y ( V ) } . \underset{|L|=V}{\operatorname{argmin}}\,E(L)=\{L:L\cong V^{1/3}H_Y(V)\}. ∣ L ∣ = V argmin E ( L ) = { L : L ≅ V 1/3 H Y ( V )} . Proof. The endpoint identifications in Theorem 4.12 and Proposition 3.20 , Lemma 6.10 , and
E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = 2 V 4 E ( V 1 / 3 Λ ) E(\Lambda,6)-2V^{2}E(\Lambda,3)=2V^{4}\mathcal{E}(V^{1/3}\Lambda) E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) = 2 V 4 E ( V 1/3 Λ ) give the result. This completes the proof of Theorem 6.11 . □
Proof of Theorem 1.1 . Existence was proved in § 2 . Theorem 5.8 , Theorem 6.11 and Proposition 3.20 give all minimizers on [ 21 / 20 , ∞ ) [21/20,\infty) [ 21/20 , ∞ ) . Uniqueness at the joining covolumes identifies the simple-hexagonal branches. Lemma 2.6 proves (i), and Lemma 6.2 proves (ii).
For 0 < V ≤ 21 / 20 0<V\le21/20 0 < V ≤ 21/20 and a non-FCC unit-covolume shape L L L ,
2 V 4 { E ( V 1 / 3 L ) − E ( V 1 / 3 F ) } = 2 ( 21 / 20 ) 4 { E ( ( 21 / 20 ) 1 / 3 L ) − E ( ( 21 / 20 ) 1 / 3 F ) } + 2 { ( 21 / 20 ) 2 − V 2 } { E ( L , 3 ) − E ( F , 3 ) } > 0. 2V^{4}\{\mathcal{E}(V^{1/3}L)-\mathcal{E}(V^{1/3}F)\}
=2(21/20)^{4}\{\mathcal{E}((21/20)^{1/3}L)-\mathcal{E}((21/20)^{1/3}F)\}
+2\{(21/20)^{2}-V^{2}\}\{\mathcal{E}(L,3)-\mathcal{E}(F,3)\}>0. 2 V 4 { E ( V 1/3 L ) − E ( V 1/3 F )} = 2 ( 21/20 ) 4 { E (( 21/20 ) 1/3 L ) − E (( 21/20 ) 1/3 F )} + 2 {( 21/20 ) 2 − V 2 } { E ( L , 3 ) − E ( F , 3 )} > 0.
Here Theorem 4.11 makes the first term positive and (1.3) makes the second nonnegative. This completes the classification of the minimizers; the derivative jumps follow from Proposition 2.7 .
For 0 < X < 307 / 1000 < 1 / 3 0<X<307/1000<1/3 0 < X < 307/1000 < 1/3 and Y > 1 Y>1 Y > 1 ,
argmin n ∈ D 3 ∖ { 0 } ( X n 1 2 + n 2 2 + n 3 2 ) = { ( 2 , 0 , 0 ) , ( − 2 , 0 , 0 ) } , \underset{n\in D_{3}\setminus\{0\}}{\operatorname{argmin}}\left(Xn_{1}^{2}+n_{2}^{2}+n_{3}^{2}\right)=\{(2,0,0),(-2,0,0)\}, n ∈ D 3 ∖ { 0 } argmin ( X n 1 2 + n 2 2 + n 3 2 ) = {( 2 , 0 , 0 ) , ( − 2 , 0 , 0 )} ,
argmin ( i , j , k ) ∈ Z 3 ∖ { 0 } ( i 2 + i j + j 2 + Y k 2 ) = { ( ± 1 , 0 , 0 ) , ( 0 , ± 1 , 0 ) , ( 1 , − 1 , 0 ) , ( − 1 , 1 , 0 ) } . \underset{(i,j,k)\in\mathbb{Z}^{3}\setminus\{0\}}{\operatorname{argmin}}\left(i^{2}+ij+j^{2}+Yk^{2}\right)=\{(\pm1,0,0),(0,\pm1,0),(1,-1,0),(-1,1,0)\}. ( i , j , k ) ∈ Z 3 ∖ { 0 } argmin ( i 2 + ij + j 2 + Y k 2 ) = {( ± 1 , 0 , 0 ) , ( 0 , ± 1 , 0 ) , ( 1 , − 1 , 0 ) , ( − 1 , 1 , 0 )} .
Thus the tetragonal and SH minimizers are neither simple cubic nor BCC.
For ∣ L ∣ = 1 |L|=1 ∣ L ∣ = 1 , L ≇ F L\not\cong F L ≅ F ,
E ( V 1 / 3 L ) ≥ E ( V 1 / 3 F ) ⟺ E ( L , 6 ) − E ( F , 6 ) ≥ 2 V 2 { E ( L , 3 ) − E ( F , 3 ) } . \mathcal{E}(V^{1/3}L)\ge\mathcal{E}(V^{1/3}F)\Longleftrightarrow E(L,6)-E(F,6)\ge2V^{2}\{E(L,3)-E(F,3)\}. E ( V 1/3 L ) ≥ E ( V 1/3 F ) ⟺ E ( L , 6 ) − E ( F , 6 ) ≥ 2 V 2 { E ( L , 3 ) − E ( F , 3 )} .
The denominator is positive by (1.3). Taking the infimum and using the just proved sharp FCC interval gives (1.9). Lemma 5.4, applied at V F T V_{\mathrm{FT}} V FT , gives positive FCC shape Hessian: divide its radial inequality along S = t T S=tT S = tT by t 2 t^{2} t 2 and let t → 0 t\to0 t → 0 .
By Proposition 4.3 and this classification,
m ( V ) > m ( 1 ) ( V > 1 ) , m ( V ) = E ( F , 6 ) 2 V 4 − E ( F , 3 ) V 2 ( 0 < V ≤ 1 ) . m(V)>m(1)\quad(V>1),\qquad m(V)=\frac{E(F,6)}{2V^{4}}-\frac{E(F,3)}{V^{2}}\quad(0<V\le1). m ( V ) > m ( 1 ) ( V > 1 ) , m ( V ) = 2 V 4 E ( F , 6 ) − V 2 E ( F , 3 ) ( 0 < V ≤ 1 ) .
Proposition 4.2 gives E ( F , 6 ) < E ( F , 3 ) E(F,6)<E(F,3) E ( F , 6 ) < E ( F , 3 ) , and
E ( F , 6 ) 2 V 4 − E ( F , 3 ) V 2 = E ( F , 6 ) 2 ( V − 2 − E ( F , 3 ) E ( F , 6 ) ) 2 − E ( F , 3 ) 2 2 E ( F , 6 ) . \frac{E(F,6)}{2V^{4}}-\frac{E(F,3)}{V^{2}}=\frac{E(F,6)}{2}\left(V^{-2}-\frac{E(F,3)}{E(F,6)}\right)^{2}-\frac{E(F,3)^{2}}{2E(F,6)}. 2 V 4 E ( F , 6 ) − V 2 E ( F , 3 ) = 2 E ( F , 6 ) ( V − 2 − E ( F , 6 ) E ( F , 3 ) ) 2 − 2 E ( F , 6 ) E ( F , 3 ) 2 .
Its unique minimum occurs at V = E ( F , 6 ) / E ( F , 3 ) < 1 V=\sqrt{E(F,6)/E(F,3)}<1 V = E ( F , 6 ) / E ( F , 3 ) < 1 . This gives the free-scale minimum. Finally, Proposition 3.20 and (3.8) prove (iii), including both expansions; Proposition 4.3 proves (iv). This completes the proof of Theorem 1.1 . □
In particular the sharp scale-invariant inequality is
E ( L , 3 ) 2 E ( L , 6 ) ≤ E ( F , 3 ) 2 E ( F , 6 ) , ∣ L ∣ = 1 , (76) \frac{E(L,3)^{2}}{E(L,6)}\le\frac{E(F,3)^{2}}{E(F,6)},\qquad|L|=1, \tag*{(76)} E ( L , 6 ) E ( L , 3 ) 2 ≤ E ( F , 6 ) E ( F , 3 ) 2 , ∣ L ∣ = 1 , (76)
with equality exactly for L ≅ F L\cong F L ≅ F . This proves the Bravais-lattice free-scale conjecture in [7 Conjecture 2] , using (1.3).
Appendix
Lemmas A.1, 2.8, 2.10 and (2.4), (2.11) enclose the finite sums and complete tails used for the moment bounds, transition brackets and exhausted covers; the decimal thresholds are outputs.
Lemma A.1 (Outward arithmetic). The outward operations below satisfy
x ∈ I , y ∈ J ⟹ x + y ∈ I + J , x − y ∈ I − J , x y ∈ I J , x\in I,\ y\in J\quad\Longrightarrow\quad x+y\in I+J,\quad x-y\in I-J,\quad xy\in IJ, x ∈ I , y ∈ J ⟹ x + y ∈ I + J , x − y ∈ I − J , x y ∈ I J , x / y ∈ I / J ( 0 ∉ J ) . x/y\in I/J\quad(0\notin J). x / y ∈ I / J ( 0 ∈ / J ) . Together with the displayed power, root, kernel and remainder bounds, they enclose all finite moment sums and derivatives below.
Proof. Let R N 53 \mathrm{RN}_{53} RN 53 denote binary64 rounding to nearest, ties-to-even, with gradual underflow, and define
x ↓ = pred ( R N 53 ( x ) ) , x ↑ = succ ( R N 53 ( x ) ) . x^{\downarrow} = \operatorname{pred}(\mathrm{RN}_{53}(x)), \qquad x^{\uparrow} = \operatorname{succ}(\mathrm{RN}_{53}(x)). x ↓ = pred ( RN 53 ( x )) , x ↑ = succ ( RN 53 ( x )) .
Here pred , succ \operatorname{pred},\operatorname{succ} pred , succ give adjacent binary64 numbers; round operations separately, without fusion or reassociation. For I = [ a , b ] I=[a,b] I = [ a , b ] , J = [ c , d ] J=[c,d] J = [ c , d ] , use
I + J = [ ( a + c ) ↓ , ( b + d ) ↑ ] , I − J = [ ( a − d ) ↓ , ( b − c ) ↑ ] , I+J=[(a+c)^{\downarrow},(b+d)^{\uparrow}], \qquad I-J=[(a-d)^{\downarrow},(b-c)^{\uparrow}], I + J = [( a + c ) ↓ , ( b + d ) ↑ ] , I − J = [( a − d ) ↓ , ( b − c ) ↑ ] ,
I J = [ min ( a c , a d , b c , b d ) ↓ , max ( a c , a d , b c , b d ) ↑ ] , IJ=[\min(ac,ad,bc,bd)^{\downarrow},\max(ac,ad,bc,bd)^{\uparrow}], I J = [ min ( a c , a d , b c , b d ) ↓ , max ( a c , a d , b c , b d ) ↑ ] ,
I − 1 = [ ( 1 / b ) ↓ , ( 1 / a ) ↑ ] ( 0 ∉ I ) , I / J = I J − 1 , − I = [ − b , − a ] . I^{-1}=[(1/b)^{\downarrow},(1/a)^{\uparrow}]\ (0\notin I), \qquad I/J=IJ^{-1}, \qquad-I=[-b,-a]. I − 1 = [( 1/ b ) ↓ , ( 1/ a ) ↑ ] ( 0 ∈ / I ) , I / J = I J − 1 , − I = [ − b , − a ] .
Use I [ 0 , 0 ] = [ 0 , 0 ] J = [ 0 , 0 ] I[0,0]=[0,0]J=[0,0] I [ 0 , 0 ] = [ 0 , 0 ] J = [ 0 , 0 ] . If I , J ⊂ [ 0 , ∞ ) I,J\subset[0,\infty) I , J ⊂ [ 0 , ∞ ) or I , J ⊂ ( − ∞ , 0 ] I,J\subset(-\infty,0] I , J ⊂ ( − ∞ , 0 ] , set I J ← max { 0 , I J } IJ\leftarrow\max\{0,IJ\} I J ← max { 0 , I J } . Directly,
I 2 = { [ 0 , max ( a 2 , b 2 ) ↑ ] , 0 ∈ I , [ max { 0 , min ( a 2 , b 2 ) ↓ } , max ( a 2 , b 2 ) ↑ ] , 0 ∉ I . I^2=
\begin{cases}
[0,\max(a^2,b^2)^{\uparrow}], & 0\in I,\\
[\max\{0,\min(a^2,b^2)^{\downarrow}\},\max(a^2,b^2)^{\uparrow}], & 0\notin I.
\end{cases} I 2 = { [ 0 , max ( a 2 , b 2 ) ↑ ] , [ max { 0 , min ( a 2 , b 2 ) ↓ } , max ( a 2 , b 2 ) ↑ ] , 0 ∈ I , 0 ∈ / I .
Integer powers use repeated squaring and multiplication. The binary midpoint is
R N 53 ( a + R N 53 ( R N 53 ( b − a ) / 2 ) ) . \mathrm{RN}_{53}(a+\mathrm{RN}_{53}(\mathrm{RN}_{53}(b-a)/2)). RN 53 ( a + RN 53 ( RN 53 ( b − a ) /2 )) .
Enclose rationals by these rules; representable constants use point intervals. For 0 ≤ a ≤ b 0\le a\le b 0 ≤ a ≤ b , m ∈ { 2 , 3 } m\in\{2,3\} m ∈ { 2 , 3 } ,
0 ≤ l ≤ u , [ l , l ] m ≤ a , [ u , u ] m ≥ b ⟹ [ a , b ] 1 / m ⊂ [ l , u ] . 0\le l\le u,\qquad[l,l]^m\le a,\qquad[u,u]^m\ge b\quad\Longrightarrow\quad[a,b]^{1/m}\subset[l,u]. 0 ≤ l ≤ u , [ l , l ] m ≤ a , [ u , u ] m ≥ b ⟹ [ a , b ] 1/ m ⊂ [ l , u ] .
Approximate roots select l , u l,u l , u ; the power inequalities certify them, with [ 0 , 0 ] 1 / m = [ 0 , 0 ] [0,0]^{1/m}=[0,0] [ 0 , 0 ] 1/ m = [ 0 , 0 ] .
Use the adjacent binary64 bounds for π \pi π certified by Lemma 2.8. For the exponential, scaling to 0 ≤ u ≤ 1 / 2 0\le u\le1/2 0 ≤ u ≤ 1/2 and using
∑ j = 0 21 ( − u ) j j ! ≤ e − u ≤ ∑ j = 0 20 ( − u ) j j ! \sum_{j=0}^{21}\frac{(-u)^j}{j!}\le e^{-u}\le\sum_{j=0}^{20}\frac{(-u)^j}{j!} j = 0 ∑ 21 j ! ( − u ) j ≤ e − u ≤ j = 0 ∑ 20 j ! ( − u ) j
followed by squaring encloses the exponential; for x ≥ 1024 x\ge1024 x ≥ 1024 , e − x ∈ [ 0 , 2 − 1024 ] e^{-x}\in[0,2^{-1024}] e − x ∈ [ 0 , 2 − 1024 ] . The Stieltjes Jacobi orders in Lemma 2.8 are
48 ( x < 4 ) , 32 ( 4 ≤ x < 8 ) , 24 ( 8 ≤ x < 16 ) , 16 ( x ≥ 16 ) . 48\ (x<4),\qquad32\ (4\le x<8),\qquad24\ (8\le x<16),\qquad16\ (x\ge16). 48 ( x < 4 ) , 32 ( 4 ≤ x < 8 ) , 24 ( 8 ≤ x < 16 ) , 16 ( x ≥ 16 ) .
For 0 < x ≤ 2 0<x\le2 0 < x ≤ 2 , use
G 1 / 2 ( x ) = π / x − 2 ∑ n = 0 ∞ ( − x ) n n ! ( 2 n + 1 ) \mathcal{G}_{1/2}(x)=\sqrt{\pi/x}-2\sum_{n=0}^{\infty}\frac{(-x)^n}{n!(2n+1)} G 1/2 ( x ) = π / x − 2 n = 0 ∑ ∞ n ! ( 2 n + 1 ) ( − x ) n
with partial sums of orders 33 , 32 33,32 33 , 32 , then x G a + 1 = a G a + e − x x\mathcal{G}_{a+1}=a\mathcal{G}_a+e^{-x} x G a + 1 = a G a + e − x . Round every sum and remainder outwards.
For r ∈ { 0 , 1 , 2 } r\in\{0,1,2\} r ∈ { 0 , 1 , 2 } , let m j ± ρ j m_j\pm\rho_j m j ± ρ j enclose the coefficient of ( q − q c ) j (q-q_c)^j ( q − q c ) j , 0 ≤ j ≤ 6 − r 0\le j\le6-r 0 ≤ j ≤ 6 − r . Use a Horner radius at least the outward value
( 1 + 2 − 40 ) [ ( 1 + 2 − 40 ) { ∑ j = 0 6 − r ρ j ∣ q − q c ∣ j + 2 − 46 ∑ j = 0 6 − r ∣ m j ∣ ∣ q − q c ∣ j } + ( 1 + 2 − 40 ) M 7 ∣ q − q c ∣ 7 − r ( 7 − r ) ! ] + 2 − 1067 . (1+2^{-40})\left[\left(1+2^{-40}\right)\left\{\sum_{j=0}^{6-r}\rho_j|q-q_c|^j+2^{-46}\sum_{j=0}^{6-r}|m_j||q-q_c|^j\right\}+\left(1+2^{-40}\right)\frac{M_7|q-q_c|^{7-r}}{(7-r)!}\right]+2^{-1067}. ( 1 + 2 − 40 ) [ ( 1 + 2 − 40 ) { j = 0 ∑ 6 − r ρ j ∣ q − q c ∣ j + 2 − 46 j = 0 ∑ 6 − r ∣ m j ∣∣ q − q c ∣ j } + ( 1 + 2 − 40 ) ( 7 − r )! M 7 ∣ q − q c ∣ 7 − r ] + 2 − 1067 .
Use the exactly representable point constant 1 + 2 − 40 1+2^{-40} 1 + 2 − 40 at each occurrence. For interval arguments, use endpoint monotonicity or add M 1 rad ( q ) M_1\operatorname{rad}(q) M 1 rad ( q ) before the outer factor 1 + 2 − 40 1+2^{-40} 1 + 2 − 40 . For each positive kernel F F F listed in Lemma A.2 and q ≥ 20 q\ge20 q ≥ 20 , use
∣ F ( r ) ( q ) ∣ ≤ 5 e − 3 b ( r = 0 , 1 , 2 ) , b = max { d ∈ { 20 , 40 , 80 , 160 , 320 } : d ≤ q } . |F^{(r)}(q)|\le5e^{-3b}\quad(r=0,1,2),\qquad b=\max\{d\in\{20,40,80,160,320\}:d\le q\}. ∣ F ( r ) ( q ) ∣ ≤ 5 e − 3 b ( r = 0 , 1 , 2 ) , b = max { d ∈ { 20 , 40 , 80 , 160 , 320 } : d ≤ q } .
Use G a ( x ) ≤ e − x / ( x − 7 ) \mathcal{G}_a(x)\le e^{-x}/(x-7) G a ( x ) ≤ e − x / ( x − 7 ) for a ≤ 8 a\le8 a ≤ 8 , x > 7 x>7 x > 7 in every terminal inequality.
For degree ≤ 6 \le6 ≤ 6 and ≤ 32 \le32 ≤ 32 operations, rounding contributes at most 2 − 46 ∑ j = 0 6 − r ∣ m j ∣ ∣ q − q c ∣ j 2^{-46}\sum_{j=0}^{6-r}|m_j||q-q_c|^j 2 − 46 ∑ j = 0 6 − r ∣ m j ∣∣ q − q c ∣ j and underflow < 2 − 1067 <2^{-1067} < 2 − 1067 . On the nearest-grid cells, ∣ q − q c ∣ ≤ 20 / 9999 |q-q_c|\le20/9999 ∣ q − q c ∣ ≤ 20/9999 . Coefficient radii, positive remainders, outward sums and ( 1 + 2 − 40 ) > 1 (1 + 2^{-40}) > 1 ( 1 + 2 − 40 ) > 1 prove enclosure by induction. This completes the proof of Lemma A.1 . □
Lemma A.2 (Validated comparison bounds). Fix V > 0 V > 0 V > 0 , T ∈ R T \in\mathbb{R} T ∈ R and a closed initial box B 0 ⊂ { a , δ > 0 } B_0 \subset\{a, \delta> 0\} B 0 ⊂ { a , δ > 0 } . Choose Korkine–Zolotarev representatives Λ ≅ U ( p ( Λ ) ) Z 3 \Lambda\cong U(p(\Lambda))\mathbb{Z}^3 Λ ≅ U ( p ( Λ )) Z 3 with p ( Λ ) ∈ B 0 p(\Lambda) \in B_0 p ( Λ ) ∈ B 0 , and put
M T = { p ( Λ ) ∈ B 0 : Λ ∈ argmin ∣ L ∣ = V E ( L ) , 1 2 E ( Λ , 6 ) − E ( Λ , 3 ) ≤ T } . \mathcal{M}_T = \left\{p(\Lambda) \in B_0 : \Lambda\in\underset{|L|=V}{\operatorname{argmin}}\, E(L),\quad\frac{1}{2}E(\Lambda,6)-E(\Lambda,3)\leq T\right\}. M T = { p ( Λ ) ∈ B 0 : Λ ∈ ∣ L ∣ = V argmin E ( L ) , 2 1 E ( Λ , 6 ) − E ( Λ , 3 ) ≤ T } . All Ewald, derivative, stationary-value and Newton bounds below hold on every positive-diagonal box B ⊂ B 0 B \subset B_0 B ⊂ B 0 . If R R R is the union of boxes contained in proved stationary basins, and Q , U Q,U Q , U are the queued and unresolved boxes, the strict tests and contractions preserve
M T ⊂ R ∪ ⋃ B ∈ Q ∪ U B . \mathcal{M}_T \subset R \cup\bigcup_{B\in Q\cup U} B. M T ⊂ R ∪ B ∈ Q ∪ U ⋃ B . For local profile minima, Proposition 5.1 supplies the corresponding local minima at every supporting parameter.
Proof. Use the coordinate formulas and complete tails of Lemma 2.10 on B B B , with the same V V V . Truncate to { n ∈ Z 3 : 0 < ∥ n ∥ ∞ ≤ N } \{n \in\mathbb{Z}^3 : 0 < \lVert n\rVert_\infty\leq N\} { n ∈ Z 3 : 0 < ∥ n ∥ ∞ ≤ N } , N = 2 , 3 , 4 N = 2,3,4 N = 2 , 3 , 4 ; pairing n , − n n,-n n , − n gives multiplicity two.
For r = 0 , 1 , 2 r = 0,1,2 r = 0 , 1 , 2 , g ( r ) g^{(r)} g ( r ) has one critical argument, the positive root of
∑ j = 0 6 + r ( π q ) j j ! − b r q 3 ∑ j = 0 3 + r ( π q ) j j ! = 0 , ( b 0 , b 1 , b 2 ) = ( 1 , 4 7 , 5 14 ) . \sum_{j=0}^{6+r}\frac{(\pi q)^j}{j!}-b_rq^3\sum_{j=0}^{3+r}\frac{(\pi q)^j}{j!}=0,\qquad(b_0,b_1,b_2)=\left(1,\frac{4}{7},\frac{5}{14}\right). j = 0 ∑ 6 + r j ! ( π q ) j − b r q 3 j = 0 ∑ 3 + r j ! ( π q ) j = 0 , ( b 0 , b 1 , b 2 ) = ( 1 , 7 4 , 14 5 ) .
For x > 0 x>0 x > 0 , covariance of t 3 , t t^3,t t 3 , t for the probability density e − x t t 3 + r / G 4 + r ( x ) e^{-xt}t^{3+r}/\mathcal{G}_{4+r}(x) e − x t t 3 + r / G 4 + r ( x ) on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) gives
( G 7 + r G 4 + r ) ′ ( x ) = − G 8 + r ( x ) G 4 + r ( x ) − G 7 + r ( x ) G 5 + r ( x ) G 4 + r ( x ) 2 < 0 , \left(\frac{\mathcal{G}_{7+r}}{\mathcal{G}_{4+r}}\right)'(x)=-\frac{\mathcal{G}_{8+r}(x)\mathcal{G}_{4+r}(x)-\mathcal{G}_{7+r}(x)\mathcal{G}_{5+r}(x)}{\mathcal{G}_{4+r}(x)^2}<0, ( G 4 + r G 7 + r ) ′ ( x ) = − G 4 + r ( x ) 2 G 8 + r ( x ) G 4 + r ( x ) − G 7 + r ( x ) G 5 + r ( x ) < 0 ,
lim x → 0 G 7 + r ( x ) G 4 + r ( x ) = ∞ , lim x → ∞ G 7 + r ( x ) G 4 + r ( x ) = 1. \lim_{x\to0}\frac{\mathcal{G}_{7+r}(x)}{\mathcal{G}_{4+r}(x)}=\infty,\qquad\lim_{x\to\infty}\frac{\mathcal{G}_{7+r}(x)}{\mathcal{G}_{4+r}(x)}=1. x → 0 lim G 4 + r ( x ) G 7 + r ( x ) = ∞ , x → ∞ lim G 4 + r ( x ) G 7 + r ( x ) = 1.
Hence the root is unique. Start with [ l , h ] = [ 1 , 4 ] [l,h]=[1,4] [ l , h ] = [ 1 , 4 ] . For at most 50 bisections, put m = ( l + h ) / 2 m=(l+h)/2 m = ( l + h ) /2 : l ← m l\leftarrow m l ← m for a strictly positive polynomial interval, h ← m h\leftarrow m h ← m for a strictly negative one; otherwise stop. The retained interval encloses the root; endpoint and applicable critical values enclose the signed range. For reciprocal kernels, π 6 240 < π 3 2 \frac{\pi^6}{240}<\frac{\pi^3}{2} 240 π 6 < 2 π 3 and G − 9 / 2 + r < G − 3 / 2 + r \mathcal{G}_{-9/2+r}<\mathcal{G}_{-3/2+r} G − 9/2 + r < G − 3/2 + r give ( − 1 ) r + 1 g ( r ) ( q ) > 0 (-1)^{r+1}g^{(r)}(q)>0 ( − 1 ) r + 1 g ( r ) ( q ) > 0 ; endpoints determine the orientation.
For the four positive kernels π 3 2 G 3 ( π q ) \frac{\pi^3}{2}\mathcal{G}_3(\pi q) 2 π 3 G 3 ( π q ) , 2 π 6 240 G 6 ( π q ) 2\frac{\pi^6}{240}\mathcal{G}_6(\pi q) 2 240 π 6 G 6 ( π q ) , π 3 2 G − 3 / 2 ( π q ) \frac{\pi^3}{2}\mathcal{G}_{-3/2}(\pi q) 2 π 3 G − 3/2 ( π q ) , 2 π 6 240 G − 9 / 2 ( π q ) 2\frac{\pi^6}{240}\mathcal{G}_{-9/2}(\pi q) 2 240 π 6 G − 9/2 ( π q ) , store interval derivatives through order seven on the binary64 grid
q 0 = RN 53 ( 1 / 1000 ) , q j + 1 = RN 53 ( q j RN 53 ( 10001 / 10000 ) ) . q_0=\operatorname{RN}_{53}(1/1000),\qquad q_{j+1}=\operatorname{RN}_{53}\left(q_j\operatorname{RN}_{53}(10001/10000)\right). q 0 = RN 53 ( 1/1000 ) , q j + 1 = RN 53 ( q j RN 53 ( 10001/10000 ) ) .
Stop at the first q j > 20 q_j>20 q j > 20 ; successive relative spacing is ≤ 1 / 9999 \leq1/9999 ≤ 1/9999 . Enclose coefficients by F ( k ) ( q j ) / k ! F^{(k)}(q_j)/k! F ( k ) ( q j ) / k ! and M 1 , M 7 M_1,M_7 M 1 , M 7 by the respective absolute derivative bounds at q max ( 0 , j − 1 ) q_{\max(0,j-1)} q m a x ( 0 , j − 1 ) . Complete monotonicity validates these on [ q max ( 0 , j − 1 ) , q j + 1 ] [q_{\max(0,j-1)},q_{j+1}] [ q m a x ( 0 , j − 1 ) , q j + 1 ] . Choose the nearest node, the lower one in a tie; outside the grid, use direct kernel bounds. For r = 0 , 1 , 2 r=0,1,2 r = 0 , 1 , 2 , use degree 6 − r 6-r 6 − r and remainder
sup ∣ F ( 7 ) ∣ ( 7 − r ) ! ∣ q − q c ∣ 7 − r . \frac{\sup\lvert F^{(7)}\rvert}{(7-r)!}\lvert q-q_c\rvert^{7-r}. ( 7 − r )! sup ∣ F ( 7 ) ∣ ∣ q − q c ∣ 7 − r .
Complete monotonicity bounds sup ∣ F ( 7 ) ∣ \sup\lvert F^{(7)}\rvert sup ∣ F ( 7 ) ∣ at the left endpoint and longer intervals by their endpoints. Retain every remainder.
Metric radii and a stationary-value bound. With c = midbin B c=\operatorname{midbin} B c = midbin B componentwise,
U c = U ( c ) , M ( p ) = U ( p ) U c − 1 , p ∈ B . U_c=U(c),\qquad M(p)=U(p)U_c^{-1},\qquad p\in B. U c = U ( c ) , M ( p ) = U ( p ) U c − 1 , p ∈ B .
Then
M 11 = a a c , M 22 = δ δ c , M 33 = a c δ c a δ , M_{11}=\frac{a}{a_c},\qquad M_{22}=\frac{\delta}{\delta_c},\qquad M_{33}=\frac{a_c\delta_c}{a\delta}, M 11 = a c a , M 22 = δ c δ , M 33 = a δ a c δ c ,
M 12 = a ( x − x c ) δ c , M 23 = δ ( z − z c ) h c , M 13 = a { y − y c − z c ( x − x c ) } h c . M_{12}=\frac{a(x-x_c)}{\delta_c},\qquad M_{23}=\frac{\delta(z-z_c)}{h_c},\qquad M_{13}=\frac{a\{y-y_c-z_c(x-x_c)\}}{h_c}. M 12 = δ c a ( x − x c ) , M 23 = h c δ ( z − z c ) , M 13 = h c a { y − y c − z c ( x − x c )} .
Polar decomposition gives
∥ ( M T M ) 1 / 2 − I ∥ F 2 = ∥ M ∥ F 2 − 2 tr ( ( M T M ) 1 / 2 ) + 3 ≤ ∥ M ∥ F 2 − 2 tr M + 3 = ∥ M − I ∥ F 2 . \begin{aligned}
\left\|(M^{T}M)^{1/2}-I\right\|_{F}^{2}=\|M\|_{F}^{2}-2\operatorname{tr}\left((M^{T}M)^{1/2}\right)+3 \\
&\leq\|M\|_{F}^{2}-2\operatorname{tr}M+3=\|M-I\|_{F}^{2}.
\end{aligned} ( M T M ) 1/2 − I F 2 = ∥ M ∥ F 2 − 2 tr ( ( M T M ) 1/2 ) + 3 ≤ ∥ M ∥ F 2 − 2 tr M + 3 = ∥ M − I ∥ F 2 .
Take outward bounds
e ≥ sup p ∈ B ∥ M ( p ) − I ∥ F , e\geq\sup_{p\in B}\|M(p)-I\|_{F}, e ≥ p ∈ B sup ∥ M ( p ) − I ∥ F ,
b ≥ sup p ∈ B 4 ∑ i = 1 3 ( M i i − 1 ) 2 + 2 ( M 12 2 + M 13 2 + M 23 2 ) + e 2 . b\geq\sup_{p\in B}\sqrt{4\sum_{i=1}^{3}(M_{ii}-1)^{2}+2(M_{12}^{2}+M_{13}^{2}+M_{23}^{2})+e^{2}}. b ≥ p ∈ B sup 4 i = 1 ∑ 3 ( M ii − 1 ) 2 + 2 ( M 12 2 + M 13 2 + M 23 2 ) + e 2 .
By the logarithmic series,
e < 1 ⟹ sup p ∈ B ∥ log ( M T M ) ∥ F ≤ 2 e + e 2 1 − e , e<1\Longrightarrow\sup_{p\in B}\|\log(M^{T}M)\|_{F}\leq2e+\frac{e^{2}}{1-e}, e < 1 ⟹ p ∈ B sup ∥ log ( M T M ) ∥ F ≤ 2 e + 1 − e e 2 ,
b < 1 ⟹ sup p ∈ B ∥ log ( M T M ) ∥ F ≤ b + b 2 2 ( 1 − b ) . b<1\Longrightarrow\sup_{p\in B}\|\log(M^{T}M)\|_{F}\leq b+\frac{b^{2}}{2(1-b)}. b < 1 ⟹ p ∈ B sup ∥ log ( M T M ) ∥ F ≤ b + 2 ( 1 − b ) b 2 .
As b ≥ e 2 b\geq e^{2} b ≥ e 2 , take
R = { min { 2 e + e 2 / ( 1 − e ) , b + b 2 / ( 2 ( 1 − b ) ) } , b < 1 , 2 e + e 2 / ( 1 − e ) , b ≥ 1 , e < 1 , + ∞ , e ≥ 1. R=
\begin{cases}
\min\{2e+e^{2}/(1-e),\ b+b^{2}/(2(1-b))\}, & b<1,\\
2e+e^{2}/(1-e), & b\geq1,\ e<1,\\
+\infty, & e\geq1.
\end{cases} R = ⎩ ⎨ ⎧ min { 2 e + e 2 / ( 1 − e ) , b + b 2 / ( 2 ( 1 − b ))} , 2 e + e 2 / ( 1 − e ) , + ∞ , b < 1 , b ≥ 1 , e < 1 , e ≥ 1.
R = + ∞ R=+\infty R = + ∞ disables radius tests. For every p ∈ B p\in B p ∈ B , det M = 1 \det M=1 det M = 1 and tr log ( M T M ) = 0 \operatorname{tr}\log(M^{T}M)=0 tr log ( M T M ) = 0 , hence
e − 2 / 3 R U c T U c ⪯ U T U , e − 2 / 3 R ( U c T U c ) − 1 ⪯ ( U T U ) − 1 . e^{-\sqrt{2/3}R}U_c^{T}U_c\preceq U^{T}U,\qquad e^{-\sqrt{2/3}R}(U_c^{T}U_c)^{-1}\preceq(U^{T}U)^{-1}. e − 2/3 R U c T U c ⪯ U T U , e − 2/3 R ( U c T U c ) − 1 ⪯ ( U T U ) − 1 .
Use metric floors for finite R R R ; otherwise keep the positive native floors. For H = H T H=H^{T} H = H T , tr H = 0 \operatorname{tr}H=0 tr H = 0 , ∥ H ∥ F = 1 \|H\|_{F}=1 ∥ H ∥ F = 1 ,
d 2 d t 2 ( 1 2 ∣ e t H / 2 v ∣ − 12 − ∣ e t H / 2 v ∣ − 6 ) ∣ t = 0 ≤ 2 ∣ v ∣ − 6 − 2 ∣ v ∣ − 12 + max { 0 , 14 ∣ v ∣ − 12 − 8 ∣ v ∣ − 6 } . \left.\frac{d^{2}}{dt^{2}}\left(\frac{1}{2}\lvert e^{tH/2}v\rvert^{-12}-\lvert e^{tH/2}v\rvert^{-6}\right)\right|_{t=0}
\leq2\lvert v\rvert^{-6}-2\lvert v\rvert^{-12}+\max\{0,14\lvert v\rvert^{-12}-8\lvert v\rvert^{-6}\}. d t 2 d 2 ( 2 1 ∣ e t H /2 v ∣ − 12 − ∣ e t H /2 v ∣ − 6 ) t = 0 ≤ 2 ∣ v ∣ − 6 − 2 ∣ v ∣ − 12 + max { 0 , 14 ∣ v ∣ − 12 − 8 ∣ v ∣ − 6 } .
Indeed, with A = ∣ v ∣ − 12 A=\lvert v\rvert^{-12} A = ∣ v ∣ − 12 , B = ∣ v ∣ − 6 B=\lvert v\rvert^{-6} B = ∣ v ∣ − 6 , s = ( v T H v ) 2 / ∣ v ∣ 4 s=(v^{T}Hv)^{2}/\lvert v\rvert^{4} s = ( v T H v ) 2 / ∣ v ∣ 4 and t = v T H 2 v / ∣ v ∣ 2 t=v^{T}H^{2}v/\lvert v\rvert^{2} t = v T H 2 v / ∣ v ∣ 2 , 0 ≤ s ≤ t ≤ 2 / 3 0\leq s\leq t\leq2/3 0 ≤ s ≤ t ≤ 2/3 , so
( 21 A − 12 B ) s + ( 3 B − 3 A ) t ≤ max { 0 , 2 B − 2 A , 12 A − 6 B } = 2 B − 2 A + max { 0 , 14 A − 8 B } . \begin{aligned}
(21A-12B)s+(3B-3A)t\leq\max\{0,2B-2A,12A-6B\} \\
&=2B-2A+\max\{0,14A-8B\}.
\end{aligned} ( 21 A − 12 B ) s + ( 3 B − 3 A ) t ≤ max { 0 , 2 B − 2 A , 12 A − 6 B } = 2 B − 2 A + max { 0 , 14 A − 8 B } .
On the metric R R R -ball about U c Z 3 U_c\mathbb{Z}^{3} U c Z 3 , this gives
K = 2 e 3 2 / 3 R E ( U c Z 3 , 3 ) − 2 e 6 2 / 3 R E ( U c Z 3 , 6 ) + ∑ ( i , j , k ) ∈ [ − 4 , 4 ] 3 ∩ Z 3 ( i , j , k ) ≠ ( 0 , 0 , 0 ) max { 0 , 14 { e − 2 / 3 R q c ( i , j , k ) } − 6 − 8 { e − 2 / 3 R q c ( i , j , k ) } − 3 } . (77) \begin{aligned}
K=2e^{3\sqrt{2}/3R}E(U_c\mathbb{Z}^{3},3)-2e^{6\sqrt{2}/3R}E(U_c\mathbb{Z}^{3},6) \\
&+\sum_{\substack{(i,j,k)\in[-4,4]^{3}\cap\mathbb{Z}^{3}\\(i,j,k)\neq(0,0,0)}}
\max\left\{0,14\left\{e^{-\sqrt{2}/3R}q_c(i,j,k)\right\}^{-6}-8\left\{e^{-\sqrt{2}/3R}q_c(i,j,k)\right\}^{-3}\right\}. \tag*{(77)}
\end{aligned} K = 2 e 3 2 /3 R E ( U c Z 3 , 3 ) − 2 e 6 2 /3 R E ( U c Z 3 , 6 ) + ( i , j , k ) ∈ [ − 4 , 4 ] 3 ∩ Z 3 ( i , j , k ) = ( 0 , 0 , 0 ) ∑ max { 0 , 14 { e − 2 /3 R q c ( i , j , k ) } − 6 − 8 { e − 2 /3 R q c ( i , j , k ) } − 3 } . (77)
If { 25 e − 2 / 3 R λ min ( U c T U c ) } 3 ≥ 7 / 4 \{25e^{-\sqrt{2}/3}R\lambda_{\min}(U_c^{\mathsf T}U_c)\}^{3} \ge7/4 { 25 e − 2 /3 R λ m i n ( U c T U c ) } 3 ≥ 7/4 , every omitted correction vanishes. Use Lemma 2.8 for the center moments. Integration along the center-to-p p p metric geodesic gives, for p ∈ B p \in B p ∈ B with ∇ p f ( p ) = 0 \nabla_p f(p) = 0 ∇ p f ( p ) = 0 ,
f ( p ) ≥ f ( c ) − 1 2 R 2 max { 0 , K } . (78) f(p) \ge f(c) - \frac{1}{2}R^2\max\{0,K\}. \tag*{(78)} f ( p ) ≥ f ( c ) − 2 1 R 2 max { 0 , K } . (78)
For a box of radii r i r_i r i , H i j ∈ [ h ‾ i j , h ‾ i j ] H_{ij} \in[\underline{h}_{ij},\overline{h}_{ij}] H ij ∈ [ h ij , h ij ] and σ ∈ { − 1 , 1 } 5 \sigma\in\{-1,1\}^{5} σ ∈ { − 1 , 1 } 5 , set
Q σ = ∑ i = 1 5 max ( 0 , h ‾ i i ) r i 2 + 2 ∑ 1 ≤ i < j ≤ 5 r i r j { h ‾ i j , σ i σ j = 1 , − h ‾ i j , σ i σ j = − 1. Q_\sigma= \sum_{i=1}^{5}\max(0,\overline{h}_{ii})r_i^2 + 2\sum_{1\le i<j\le5}r_ir_j
\begin{cases}
\overline{h}_{ij}, & \sigma_i\sigma_j=1,\\
-\underline{h}_{ij}, & \sigma_i\sigma_j=-1.
\end{cases} Q σ = i = 1 ∑ 5 max ( 0 , h ii ) r i 2 + 2 1 ≤ i < j ≤ 5 ∑ r i r j { h ij , − h ij , σ i σ j = 1 , σ i σ j = − 1.
For p ∈ B p \in B p ∈ B with ∇ p f ( p ) = 0 \nabla_p f(p) = 0 ∇ p f ( p ) = 0 ,
f ( p ) − f ( c ) = − ∫ 0 1 t ( p − c ) T ∇ p 2 f ( c + t ( p − c ) ) ( p − c ) d t ≥ − 1 2 max { 0 , max σ Q σ } . (79) \begin{aligned}
f(p)-f(c)=-\int_0^1 t(p-c)^{\mathsf T}\nabla_p^2f(c+t(p-c))(p-c)\,\mathrm{d}t \\
&\ge-\frac{1}{2}\max\{0,\max_\sigma Q_\sigma\}. \tag*{(79)}
\end{aligned} f ( p ) − f ( c ) = − ∫ 0 1 t ( p − c ) T ∇ p 2 f ( c + t ( p − c )) ( p − c ) d t ≥ − 2 1 max { 0 , σ max Q σ } . (79)
Only ∇ p f ( p ) = 0 \nabla_p f(p)=0 ∇ p f ( p ) = 0 is required: the nonnegative-diagonal quadratic majorant attains its box maximum at a signed vertex.
Stationarity and interval Newton. For a , δ > 0 a,\delta>0 a , δ > 0 ,
p ∈ argmin l o c f ⟹ ∇ p f ( p ) = 0 , ∇ p 2 f ( p ) ⪰ 0. p\in\operatorname*{argmin}_{\mathrm{loc}} f \quad\Longrightarrow\quad\nabla_p f(p)=0,\quad\nabla_p^2f(p)\succeq0. p ∈ loc argmin f ⟹ ∇ p f ( p ) = 0 , ∇ p 2 f ( p ) ⪰ 0.
Also on reduction boundaries. For p ∈ B p\in B p ∈ B , ∇ p f ( p ) = 0 \nabla_p f(p)=0 ∇ p f ( p ) = 0 , [ H ] ⊃ ∇ p 2 f ( B ) [H]\supset\nabla_p^2f(B) [ H ] ⊃ ∇ p 2 f ( B ) , [ g c ] ∋ ∇ p f ( c ) [g_c]\ni\nabla_p f(c) [ g c ] ∋ ∇ p f ( c ) and any fixed real matrix P P P ,
0 = ∇ p f ( c ) + ( ∫ 0 1 ∇ p 2 f ( c + t ( p − c ) ) d t ) ⏟ ∈ [ H ] ( p − c ) . 0=\nabla_p f(c)+\underbrace{\left(\int_0^1\nabla_p^2f(c+t(p-c))\,\mathrm{d}t\right)}_{\in[H]}(p-c). 0 = ∇ p f ( c ) + ∈ [ H ] ( ∫ 0 1 ∇ p 2 f ( c + t ( p − c )) d t ) ( p − c ) .
Thus
p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ] ) ( B − c ) } . (80) p\in B\cap\{c-P[g_c]+(I-P[H])(B-c)\}. \tag*{(80)} p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ]) ( B − c )} . (80)
No invertibility of P P P is assumed.
Finite exclusions. Enclose B 0 B_0 B 0 outwards. Rejection gives M T ∩ B = ∅ \mathcal{M}_T\cap B=\varnothing M T ∩ B = ∅ ; recognition gives R ← R ∪ B \mathcal{R}\leftarrow\mathcal{R}\cup B R ← R ∪ B .
A necessary reduction or moment inequality is contradicted on B B B .
B B B is contained in a proved stationary basin with ∣ det U r e f ∣ = V |\det U_{\mathrm{ref}}|=V ∣ det U ref ∣ = V . For every p ∈ B p\in B p ∈ B ,
M = U ( p ) U r e f − 1 , g i j = ( M T M ) i j , S = 3 M T M tr ( M T M ) − I , M=U(p)U_{\mathrm{ref}}^{-1},\qquad g_{ij}=(M^{\mathsf T}M)_{ij},\qquad S=\frac{3M^{\mathsf T}M}{\operatorname{tr}(M^{\mathsf T}M)}-I, M = U ( p ) U ref − 1 , g ij = ( M T M ) ij , S = tr ( M T M ) 3 M T M − I ,
{ tr ( M T M ) } 2 ∥ S ∥ F 2 = 3 ∑ 1 ≤ i < j ≤ 3 ( g i i − g j j ) 2 + 18 ∑ 1 ≤ i < j ≤ 3 g i j 2 , \{\operatorname{tr}(M^{\mathsf T}M)\}^{2}\lVert S\rVert_F^2
=3\sum_{1\le i<j\le3}(g_{ii}-g_{jj})^2+18\sum_{1\le i<j\le3}g_{ij}^2, { tr ( M T M ) } 2 ∥ S ∥ F 2 = 3 1 ≤ i < j ≤ 3 ∑ ( g ii − g j j ) 2 + 18 1 ≤ i < j ≤ 3 ∑ g ij 2 ,
U ( p ) Z 3 ≅ det ( I + S ) − 1 / 6 ( I + S ) 1 / 2 U r e f Z 3 . U(p)Z^3\cong\det(I+S)^{-1/6}(I+S)^{1/2}U_{\mathrm{ref}}Z^3. U ( p ) Z 3 ≅ det ( I + S ) − 1/6 ( I + S ) 1/2 U ref Z 3 .
The isometry is the right polar decomposition.
∃ i ∈ { 1 , … , 5 } : 0 ∉ [ ∂ i f ( B ) ] \exists i\in\{1,\ldots,5\}:0\notin[\partial_i f(B)] ∃ i ∈ { 1 , … , 5 } : 0 ∈ / [ ∂ i f ( B )] .
min 1 ≤ i ≤ 5 h ‾ i i < 0 \min_{1\le i\le5}\overline{h}_{ii}<0 min 1 ≤ i ≤ 5 h ii < 0 , or a fixed trace-free strain
H = ( 2 b 0 b 2 b 3 b 2 − b 0 + b 1 b 4 b 3 b 4 − b 0 − b 1 ) , b i ∈ 2 − 20 Z ∩ [ − 1 , 1 ] . H=\begin{pmatrix}
2b_0 & b_2 & b_3\\
b_2 & -b_0+b_1 & b_4\\
b_3 & b_4 & -b_0-b_1
\end{pmatrix},
\qquad b_i\in2^{-20}\mathbb{Z}\cap[-1,1]. H = 2 b 0 b 2 b 3 b 2 − b 0 + b 1 b 4 b 3 b 4 − b 0 − b 1 , b i ∈ 2 − 20 Z ∩ [ − 1 , 1 ] .
satisfies
sup p ∈ B d 2 d t 2 E ( e t H / 2 U ( p ) Z 3 ) ∣ t = 0 < 0. \sup_{p\in B}\left.\frac{\mathrm{d}^2}{\mathrm{d}t^2}\mathcal{E}\left(e^{tH/2}U(p)Z^3\right)\right|_{t=0}<0. p ∈ B sup d t 2 d 2 E ( e t H /2 U ( p ) Z 3 ) t = 0 < 0.
For v = U ( p ) n v=U(p)n v = U ( p ) n , v ∗ = U ( p ) − T n v^*=U(p)^{-\mathsf{T}}n v ∗ = U ( p ) − T n , q = ∣ v ∣ 2 q=|v|^2 q = ∣ v ∣ 2 , q ∗ = ∣ v ∗ ∣ 2 q^*=|v^*|^2 q ∗ = ∣ v ∗ ∣ 2 ,
d 2 d t 2 E ( e t H / 2 U ( p ) Z 3 ) ∣ t = 0 ≤ ∑ n ∈ Z 3 0 < ∥ n ∥ ∞ ≤ N { g ′ ′ ( q ) ( v T H v ) 2 + g ′ ( q ) v T H 2 v + g ~ ′ ′ ( q ∗ ) ( ( v ∗ ) T H v ∗ ) 2 + g ~ ′ ( q ∗ ) ( v ∗ ) T H 2 v ∗ } + 2 3 ∥ H ∥ F 2 ( P 1 + P 2 + D 1 + D 2 ) . \left.\frac{\mathrm{d}^2}{\mathrm{d}t^2}\mathcal{E}(e^{tH/2}U(p)\mathbb{Z}^3)\right|_{t=0}
\leq\sum_{\substack{n\in\mathbb{Z}^3\\0<\lVert n\rVert_\infty\leq N}}
\left\{g^{\prime\prime}(q)(v^{\mathsf{T}}Hv)^2+g^\prime(q)v^{\mathsf{T}}H^2v
+\widetilde{g}^{\prime\prime}(q^*)((v^*)^{\mathsf{T}}H v^*)^2+\widetilde{g}^\prime(q^*)(v^*)^{\mathsf{T}}H^2v^*\right\}
+\frac{2}{3}\lVert H\rVert_{\mathrm{F}}^2(P_1+P_2+D_1+D_2). d t 2 d 2 E ( e t H /2 U ( p ) Z 3 ) t = 0 ≤ n ∈ Z 3 0 < ∥ n ∥ ∞ ≤ N ∑ { g ′′ ( q ) ( v T H v ) 2 + g ′ ( q ) v T H 2 v + g ′′ ( q ∗ ) (( v ∗ ) T H v ∗ ) 2 + g ′ ( q ∗ ) ( v ∗ ) T H 2 v ∗ } + 3 2 ∥ H ∥ F 2 ( P 1 + P 2 + D 1 + D 2 ) .
The direction selector uses
F ^ 0 = diag ( 2 , − 1 , − 1 ) 6 , F ^ 1 = diag ( 0 , 1 , − 1 ) 2 , \widehat{F}_0=\frac{\operatorname{diag}(2,-1,-1)}{\sqrt{6}},\qquad
\widehat{F}_1=\frac{\operatorname{diag}(0,1,-1)}{\sqrt{2}}, F 0 = 6 diag ( 2 , − 1 , − 1 ) , F 1 = 2 diag ( 0 , 1 , − 1 ) ,
F ^ 2 = 1 2 ( 0 1 0 1 0 0 0 0 0 ) , F ^ 3 = 1 2 ( 0 0 1 0 0 0 1 0 0 ) , F ^ 4 = 1 2 ( 0 0 0 0 0 1 0 1 0 ) . \widehat{F}_2=\frac{1}{\sqrt{2}}
\begin{pmatrix}
0&1&0\\
1&0&0\\
0&0&0
\end{pmatrix},\qquad
\widehat{F}_3=\frac{1}{\sqrt{2}}
\begin{pmatrix}
0&0&1\\
0&0&0\\
1&0&0
\end{pmatrix},\qquad
\widehat{F}_4=\frac{1}{\sqrt{2}}
\begin{pmatrix}
0&0&0\\
0&0&1\\
0&1&0
\end{pmatrix}. F 2 = 2 1 0 1 0 1 0 0 0 0 0 , F 3 = 2 1 0 0 1 0 0 0 1 0 0 , F 4 = 2 1 0 0 0 0 0 1 0 1 0 .
At the midpoint, form the approximate symmetric matrix
A i j = ∑ n ∈ [ − 3 , 3 ] 3 ∩ Z 3 ∖ { ( 0 , 0 , 0 ) } { ( 21 q − 8 − 12 q − 5 ) ( v T F ^ i v ) ( v T F ^ j v ) + ( 3 q − 4 − 3 q − 7 ) ( F ^ i v ) T ( F ^ j v ) } , A_{ij}=\sum_{\substack{n\in[-3,3]^3\cap\mathbb{Z}^3\\\setminus\{(0,0,0)\}}
\left\{(21q^{-8}-12q^{-5})(v^{\mathsf{T}}\widehat{F}_i v)(v^{\mathsf{T}}\widehat{F}_jv)
+(3q^{-4}-3q^{-7})(\widehat{F}_iv)^{\mathsf{T}}(\widehat{F}_jv)\right\},} A ij = n ∈ [ − 3 , 3 ] 3 ∩ Z 3 ∖ {( 0 , 0 , 0 )} { ( 21 q − 8 − 12 q − 5 ) ( v T F i v ) ( v T F j v ) + ( 3 q − 4 − 3 q − 7 ) ( F i v ) T ( F j v ) } , ∑
where v = U ( c ) n v=U(c)n v = U ( c ) n , q = ∣ v ∣ 2 q=|v|^2 q = ∣ v ∣ 2 . Starting with P = I 5 P=I_5 P = I 5 , make 15 lexicographic sweeps over 0 ≤ i < j ≤ 4 0\leq i<j\leq4 0 ≤ i < j ≤ 4 . Skip ∣ A i j ∣ < 10 − 14 |A_{ij}|<10^{-14} ∣ A ij ∣ < 1 0 − 14 ; otherwise set
θ = 1 2 arctan 2 ( 2 A i j , A j j − A i i ) , J { i , j } = ( cos θ sin θ − sin θ cos θ ) , \theta=\frac{1}{2}\arctan2(2A_{ij},A_{jj}-A_{ii}),\qquad
J_{\{i,j\}}=
\begin{pmatrix}
\cos\theta&\sin\theta\\
-\sin\theta&\cos\theta
\end{pmatrix}, θ = 2 1 arctan 2 ( 2 A ij , A j j − A ii ) , J { i , j } = ( cos θ − sin θ sin θ cos θ ) ,
and update ( A , P ) ← ( J T A J , P J ) (A,P)\leftarrow(J^{\mathsf{T}}AJ,PJ) ( A , P ) ← ( J T A J , P J ) , A i j = A j i = 0 A_{ij}=A_{ji}=0 A ij = A j i = 0 . For j = min arg min 0 ≤ k ≤ 4 A k k j=\min\operatorname*{arg\,min}_{0\leq k\leq4}A_{kk} j = min arg min 0 ≤ k ≤ 4 A k k , if A j j < − 1 A_{jj}<-1 A j j < − 1 , form
b 0 = 2 − 20 RNZ ( 2 20 P 0 j / 6 ) , b i = 2 − 20 RNZ ( 2 20 P i j / 2 ) ( 1 ≤ i ≤ 4 ) , b_0=2^{-20}\operatorname{RNZ}(2^{20}P_{0j}/\sqrt{6}),\qquad
b_i=2^{-20}\operatorname{RNZ}(2^{20}P_{ij}/\sqrt{2})\quad(1\leq i\leq4), b 0 = 2 − 20 RNZ ( 2 20 P 0 j / 6 ) , b i = 2 − 20 RNZ ( 2 20 P ij / 2 ) ( 1 ≤ i ≤ 4 ) ,
where RNZ \operatorname{RNZ} RNZ is nearest-integer rounding, ties-to-even. For max i ∣ b i ∣ ≤ 1 \max_i|b_i|\leq1 max i ∣ b i ∣ ≤ 1 , test this exact H H H at N = 4 N=4 N = 4 . The N = 3 N=3 N = 3 matrix selects the direction; only the strict N = 4 N=4 N = 4 bound deletes a box.
(5) The complete lower bounds (A.2), (A.3) satisfy
f ( c ) − 1 2 R 2 max ( 0 , K ) > T or f ( c ) − 1 2 max { 0 , max σ Q σ } > T . f(c)-\frac{1}{2}R^2\max(0,K)>T\quad\text{or}\quad
f(c)-\frac{1}{2}\max\{0,\max_\sigma Q_\sigma\}>T. f ( c ) − 2 1 R 2 max ( 0 , K ) > T or f ( c ) − 2 1 max { 0 , σ max Q σ } > T .
(6) The complete Newton enclosure (A.4) satisfies B ∩ { c − P [ g c ] + ( I − P [ H ] ) ( B − c ) } = ∅ B\cap\{c-P[g_c]+(I-P[H])(B-c)\}=\varnothing B ∩ { c − P [ g c ] + ( I − P [ H ]) ( B − c )} = ∅ .
Intersect with the consequences of s − ≤ a δ ≤ s + s_-\leq a\delta\leq s_+ s − ≤ a δ ≤ s + and the two Gauss inequalities:
a ∈ [ s − / δ + , s + / δ − ] , δ ∈ [ s − / a + , s + / a − ] , a\in[s_-/\delta_+,s_+/\delta_-],\qquad
\delta\in[s_-/a_+,s_+/a_-], a ∈ [ s − / δ + , s + / δ − ] , δ ∈ [ s − / a + , s + / a − ] ,
δ ≥ a 1 − x + 2 , a ≤ δ + 1 − x − 2 , \delta\geq a\sqrt{1-x_+^2},\qquad
a\leq\frac{\delta_+}{\sqrt{1-x_-^2}}, δ ≥ a 1 − x + 2 , a ≤ 1 − x − 2 δ + ,
a ≤ V δ − 2 1 − z + 2 , δ ≤ V a − 1 − z + 2 . a\leq\frac{V}{\delta_-^2\sqrt{1-z_+^2}},\qquad
\delta\leq\sqrt{\frac{V}{a_-\sqrt{1-z_+^2}}}. a ≤ δ − 2 1 − z + 2 V , δ ≤ a − 1 − z + 2 V .
When the respective right-hand radicands are positive, also impose
x ≥ 1 − ( δ + / a − ) 2 , z ≥ 1 − V 2 a − 2 δ − 4 . x\geq\sqrt{1-(\delta_+/a_-)^2},\qquad
z\geq\sqrt{1-\frac{V^2}{a_-^2\delta_-^4}}. x ≥ 1 − ( δ + / a − ) 2 , z ≥ 1 − a − 2 δ − 4 V 2 .
These intersections, Newton and closed bisection retain every point of $\mathcal{M}_T\cap B. At each visit: ≤ 10 \le10 ≤ 10 contractions, basin recognition, midpoint. Tests run in order, advancing after failure; subscripts give cube cutoffs:
R ≤ 9 / 20 : [ f ( c ) ] N = 3 ⟶ f ( c ) − 1 2 R 2 max ( 0 , K ) > T R \le9/20:\quad[f(c)]_{N=3}\longrightarrow f(c)-\frac{1}{2}R^{2}\max(0,K)>T R ≤ 9/20 : [ f ( c ) ] N = 3 ⟶ f ( c ) − 2 1 R 2 max ( 0 , K ) > T
⟶ ∃ i : 0 ∉ [ ∂ i f ( B ) ] N = 4 , \longrightarrow\exists i:\ 0\notin[\partial_i f(B)]_{N=4}, ⟶ ∃ i : 0 ∈ / [ ∂ i f ( B ) ] N = 4 ,
all R : [ ∂ t 2 E ( e t H / 2 U ( p ) Z 3 ) ∣ t = 0 ] N = 4 , p ∈ B < 0 , \text{all }R:\quad\left[\left.\partial_t^2\mathcal{E}(e^{tH/2}U(p)Z^3)\right|_{t=0}\right]_{N=4,\ p\in B}<0, all R : [ ∂ t 2 E ( e t H /2 U ( p ) Z 3 ) t = 0 ] N = 4 , p ∈ B < 0 ,
R ≤ 3 / 20 : N = 2 , 3 , 4 : f ( c ) − 1 2 max { 0 , max σ Q σ } > T ⟶ min i h ‾ i i < 0 , R\le3/20:\quad N=2,3,4:\ f(c)-\frac{1}{2}\max\{0,\max_{\sigma}Q_{\sigma}\}>T\longrightarrow\min_i\overline{h}_{ii}<0, R ≤ 3/20 : N = 2 , 3 , 4 : f ( c ) − 2 1 max { 0 , σ max Q σ } > T ⟶ i min h ii < 0 ,
R ≤ 3 / 20 : ( [ f ( c ) ] , [ ∇ p f ( c ) ] , [ ∇ p 2 f ( c ) ] ) N = 4 R\le3/20:\quad([f(c)],[\nabla_p f(c)],[\nabla_p^2f(c)])_{N=4} R ≤ 3/20 : ([ f ( c )] , [ ∇ p f ( c )] , [ ∇ p 2 f ( c )] ) N = 4
⟶ f ( c ) − 1 2 max { 0 , max σ Q σ } > T ⟶ x r e f ‘ ( A .4 ) ‘ . \longrightarrow f(c)-\frac{1}{2}\max\{0,\max_{\sigma}Q_{\sigma}\}>T\longrightarrow{xref}`(A.4)`. ⟶ f ( c ) − 2 1 max { 0 , σ max Q σ } > T ⟶ x r e f ‘ ( A .4 ) ‘.
The final tests retain [ H ] N = 4 ( B ) [H]_{N=4}(B) [ H ] N = 4 ( B ) . Invalid N = 2 , 3 N=2,3 N = 2 , 3 Hessians advance; invalid [ H ] N = 4 ( B ) [H]_{N=4}(B) [ H ] N = 4 ( B ) disables final value/Newton. No invalid enclosure rejects B B B . Choose P P P by partial-pivot elimination of the midpoint-Hessian midpoints, taking the first row in a pivot tie. A pivot smaller than 10 − 14 10^{-14} 1 0 − 14 in absolute value, or a nonfinite entry, disables Newton. An empty Newton or necessary-domain intersection excludes the box. A nonempty contraction is retained only when
v n e w + < RN 53 ( RN 53 ( 995 / 1000 ) v o l d − ) . v_{\mathrm{new}}^{+}<\operatorname{RN}_{53}\left(\operatorname{RN}_{53}(995/1000)\,v_{\mathrm{old}}^{-}\right). v new + < RN 53 ( RN 53 ( 995/1000 ) v old − ) .
where v ± v^{\pm} v ± are the outward coordinate-volume endpoints and RN 53 \operatorname{RN}_{53} RN 53 is binary64 round-to-nearest, ties-to-even. On acceptance, B ← B n e w B\leftarrow B_{\mathrm{new}} B ← B new , c ← mid b i n B c\leftarrow\operatorname{mid}_{\mathrm{bin}}B c ← mid bin B ; recompute scores and bisect, without re-enqueueing.
For the prescribed closed root enclosure B r e f B_{\mathrm{ref}} B ref , heap priority is lexicographic in
( 1 { B ∩ B r e f ≠ ∅ } , v + ( B ) ) . \left(\mathbf{1}_{\{B\cap B_{\mathrm{ref}}\ne\varnothing\}},\ v^{+}(B)\right). ( 1 { B ∩ B ref = ∅ } , v + ( B ) ) .
New keys rise only past strictly smaller parents. Removal moves the larger child into the hole, choosing the right child in a tie; the last key then rises from that hole by the same strict comparison. Start with the original root alone; insert lower children first. This fixes counts, not coverage.
With binary midpoint c c c and coordinate radii r i r_i r i , put
i m e t ( B ) = min argmax 1 ≤ i ≤ 5 ( 2 r a a c , 2 r δ δ c , a + r x δ c − 2 + z c 2 h c − 2 , a + + r y h c , δ + + r z h c ) . i_{\mathrm{met}}(B)=\min\operatorname*{argmax}_{1\le i\le5}\left(\frac{2r_a}{a_c},\frac{2r_\delta}{\delta_c},a_+r_x\sqrt{\delta_c^{-2}+z_c^2h_c^{-2}},\frac{a_++r_y}{h_c},\frac{\delta_++r_z}{h_c}\right). i met ( B ) = min 1 ≤ i ≤ 5 argmax ( a c 2 r a , δ c 2 r δ , a + r x δ c − 2 + z c 2 h c − 2 , h c a + + r y , h c δ + + r z ) .
Let r ( B ) r(B) r ( B ) be the computed whole-box basin-norm upper bound for the prescribed references, and B i − , B i + B_i^{-},B_i^{+} B i − , B i + the closed binary halves. For B ∩ B r e f ≠ ∅ B\cap B_{\mathrm{ref}}\ne\varnothing B ∩ B ref = ∅ , define
r i ( B ) = max σ ∈ { − , + } B i σ ∩ B r e f ≠ ∅ r ( B i σ ) , i = 1 , … , 5. r_i(B)=\max_{\substack{\sigma\in\{-,+\}\\B_i^{\sigma}\cap B_{\mathrm{ref}}\ne\varnothing}}r(B_i^{\sigma}),\qquad i=1,\ldots,5. r i ( B ) = σ ∈ { − , + } B i σ ∩ B ref = ∅ max r ( B i σ ) , i = 1 , … , 5.
The generic selector is
i ( B ) = { min argmin 1 ≤ i ≤ 5 r i ( B ) , B ∩ B r e f ≠ ∅ , min i r i ( B ) < + ∞ , i m e t ( B ) , otherwise . i(B)=
\begin{cases}
\min\operatorname*{argmin}_{1\le i\le5}r_i(B), & B\cap B_{\mathrm{ref}}\ne\varnothing,\ \min_i r_i(B)<+\infty,\\
i_{\mathrm{met}}(B), & \text{otherwise}.
\end{cases} i ( B ) = { min argmin 1 ≤ i ≤ 5 r i ( B ) , i met ( B ) , B ∩ B ref = ∅ , min i r i ( B ) < + ∞ , otherwise .
The bound above only selects a coordinate; recognition requires the strict whole-box basin predicate. Before domain contraction,
B − ∪ B + = B , B − = B i ( B ) − , B + = B i ( B ) + . B^{-}\cup B^{+}=B,\qquad B^{-}=B^{-}_{i(B)},\qquad B^{+}=B^{+}_{i(B)}. B − ∪ B + = B , B − = B i ( B ) − , B + = B i ( B ) + .
Retain both halves unless a necessary-domain test excludes one; insert B − B^{-} B − first. At depth 70 or without an interior binary midpoint, retain B B B in U \mathcal{U} U . Work limits never imply completion.
Intersections and closed-child unions preserve
M T ⊂ R ∪ ⋃ B ∈ Q ∪ U B , Q = U = ∅ ⟹ M T ⊂ R . \mathcal{M}_{T}\subset\mathcal{R}\cup\bigcup_{B\in Q\cup\mathcal{U}}B,\qquad Q=\mathcal{U}=\varnothing\Longrightarrow\mathcal{M}_{T}\subset\mathcal{R}. M T ⊂ R ∪ B ∈ Q ∪ U ⋃ B , Q = U = ∅ ⟹ M T ⊂ R .
Depth and work limits retain boxes in U \mathcal{U} U . This completes the proof of Lemma A.2 . □
Lemma A.3 (A tetragonal reference derivative). Put V = 43 / 40 V=43/40 V = 43/40 and X 0 = 149 / 500 X_0=149/500 X 0 = 149/500 . Then
− 2428944942886 10 18 < d d log X E ( V 1 / 3 L X ) ∣ X = X 0 < − 2428944942885 10 18 . -\frac{2428944942886}{10^{18}} < \left.\frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(V^{1/3}L_X)\right|_{X=X_0} < -\frac{2428944942885}{10^{18}}. − 1 0 18 2428944942886 < d log X d E ( V 1/3 L X ) X = X 0 < − 1 0 18 2428944942885 . In (19) , take L 0 = V 1 / 3 L X 0 L_0=V^{1/3}L_{X_0} L 0 = V 1/3 L X 0 . At S = 0 S=0 S = 0 ,
∇ ( v , x , y , z ) E ( L ( S ) ) = 0 , ∣ ∂ u E ( L ( S ) ) ∣ < 8 ⋅ 10 − 6 . \nabla_{(v,x,y,z)}\mathcal{E}(L(S))=0,\qquad|\partial_u\mathcal{E}(L(S))|<8\cdot10^{-6}. ∇ ( v , x , y , z ) E ( L ( S )) = 0 , ∣ ∂ u E ( L ( S )) ∣ < 8 ⋅ 1 0 − 6 .
Proof. Use Lemma 2.8 with Stieltjes order 96 and ( i , j , k ) ∈ [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } (i,j,k)\in[-6,6]^3\cap\mathbb{Z}^3\setminus\{0\} ( i , j , k ) ∈ [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } . With V V V fixed, put α = ( V 2 / ( 4 X ) ) 1 / 3 \alpha=(V^2/(4X))^{1/3} α = ( V 2 / ( 4 X ) ) 1/3 and ( q ˙ , q ˙ ∗ ) = d ( q , q ∗ ) / d log X (\dot{q},\dot{q}^{*})=\mathrm{d}(q,q^{*})/\mathrm{d}\log X ( q ˙ , q ˙ ∗ ) = d ( q , q ∗ ) / d log X ; then
q = α { X ( i + j − k ) 2 + ( i − j ) 2 + k 2 } , q=\alpha\{X(i+j-k)^2+(i-j)^2+k^2\}, q = α { X ( i + j − k ) 2 + ( i − j ) 2 + k 2 } ,
q ˙ = α 3 { 2 X ( i + j − k ) 2 − ( i − j ) 2 − k 2 } , \dot{q}=\frac{\alpha}{3}\{2X(i+j-k)^2-(i-j)^2-k^2\}, q ˙ = 3 α { 2 X ( i + j − k ) 2 − ( i − j ) 2 − k 2 } ,
q ∗ = α − 1 { ( i 2 + j 2 ) / 2 + k 2 + k ( i + j ) + ( i + j ) 2 / ( 4 X ) } , q^{*}=\alpha^{-1}\{(i^2+j^2)/2+k^2+k(i+j)+(i+j)^2/(4X)\}, q ∗ = α − 1 {( i 2 + j 2 ) /2 + k 2 + k ( i + j ) + ( i + j ) 2 / ( 4 X )} ,
q ˙ ∗ = α − 1 3 { ( i 2 + j 2 ) / 2 + k 2 + k ( i + j ) − ( i + j ) 2 / ( 2 X ) } . \dot{q}^{*}=\frac{\alpha^{-1}}{3}\{(i^2+j^2)/2+k^2+k(i+j)-(i+j)^2/(2X)\}. q ˙ ∗ = 3 α − 1 {( i 2 + j 2 ) /2 + k 2 + k ( i + j ) − ( i + j ) 2 / ( 2 X )} .
At X 0 X_0 X 0 ,
1284733594796636444756239754849703 2 110 < α < 1284733594796636444756239754849704 2 110 . \frac{1284733594796636444756239754849703}{2^{110}}<\alpha<\frac{1284733594796636444756239754849704}{2^{110}}. 2 110 1284733594796636444756239754849703 < α < 2 110 1284733594796636444756239754849704 .
Principal minors give the exact rational bound
39 100 I ⪯ 1 500 ( 649 − 351 − 149 − 351 649 − 149 − 149 − 149 649 ) ⪯ 2 I . \frac{39}{100}I\preceq\frac{1}{500}
\begin{pmatrix}
649&-351&-149\\
-351&649&-149\\
-149&-149&649
\end{pmatrix}\preceq2I. 100 39 I ⪯ 500 1 649 − 351 − 149 − 351 649 − 149 − 149 − 149 649 ⪯ 2 I .
Use a = ( 39 / 100 ) α − a=(39/100)\alpha_{-} a = ( 39/100 ) α − , a ∗ = 1 / ( 2 α + ) a^{*}=1/(2\alpha_{+}) a ∗ = 1/ ( 2 α + ) and ∣ q ˙ ∣ ≤ 2 q / 3 |\dot{q}|\le2q/3 ∣ q ˙ ∣ ≤ 2 q /3 , ∣ q ˙ ∗ ∣ ≤ 2 q ∗ / 3 |\dot{q}^{*}|\le2q^{*}/3 ∣ q ˙ ∗ ∣ ≤ 2 q ∗ /3 in the complete cube tails. With x = 49 π a x=49\pi a x = 49 π a , the derivative tail is at most
2 3 [ ( π 6 240 ∑ j = 0 6 6 ! j ! x j − 6 + π 3 2 ∑ j = 0 3 3 ! j ! x j − 3 ) Θ 6 ( a ) + π 6 240 V + π 3 2 Θ 6 ( a ∗ ) ] < 8.855 ⋅ 10 − 24 . \frac{2}{3}\left[\left(\frac{\pi^6}{240}\sum_{j=0}^{6}\frac{6!}{j!}x^{j-6}+\frac{\pi^3}{2}\sum_{j=0}^{3}\frac{3!}{j!}x^{j-3}\right)\Theta_6(a)+\frac{\pi^6}{240V}+\frac{\pi^3}{2}\Theta_6(a^{*})\right]<8.855\cdot10^{-24}. 3 2 [ ( 240 π 6 j = 0 ∑ 6 j ! 6 ! x j − 6 + 2 π 3 j = 0 ∑ 3 j ! 3 ! x j − 3 ) Θ 6 ( a ) + 240 V π 6 + 2 π 3 Θ 6 ( a ∗ ) ] < 8.855 ⋅ 1 0 − 24 .
Θ 6 \Theta_6 Θ 6 is the Gaussian cube tail in Lemma A.2. Using the kernels of Lemma A.2, the complete enclosure is
d d log X E ( V 1 / 3 L X ) ∣ X = X 0 ∈ ∑ n ∈ Z 3 0 < ∥ n ∥ ∞ ≤ 6 { g ′ ( q n ) q ˙ n + g ~ ′ ( q n ∗ ) q ˙ n ∗ } ∣ X = X 0 + [ − 8.855 , 8.855 ] 10 − 24 . \left.\frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(V^{1/3}L_X)\right|_{X=X_0}\in
\left.\sum_{\substack{n\in\mathbb{Z}^3\\0<\|n\|_\infty\le6}}\{g'(q_n)\dot{q}_n+\widetilde{g}'(q_n^{*})\dot{q}_n^{*}\}\right|_{X=X_0}
+[-8.855,8.855]10^{-24}. d log X d E ( V 1/3 L X ) X = X 0 ∈ n ∈ Z 3 0 < ∥ n ∥ ∞ ≤ 6 ∑ { g ′ ( q n ) q ˙ n + g ′ ( q n ∗ ) q ˙ n ∗ } X = X 0 + [ − 8.855 , 8.855 ] 1 0 − 24 .
Rational evaluation gives the stated interval; symmetry gives
∂ u E ( L ( S ) ) ∣ S = 0 = 3 d d log X E ( V 1 / 3 L X ) ∣ X = X 0 , \left.\partial_u\mathcal{E}(L(S))\right|_{S=0}
=3\left.\frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(V^{1/3}L_X)\right|_{X=X_0}, ∂ u E ( L ( S )) ∣ S = 0 = 3 d log X d E ( V 1/3 L X ) X = X 0 ,
( ∂ v , ∂ x , ∂ y , ∂ z ) E ( L ( S ) ) ∣ S = 0 = 0. \left.(\partial_v,\partial_x,\partial_y,\partial_z)\mathcal{E}(L(S))\right|_{S=0}=0. ( ∂ v , ∂ x , ∂ y , ∂ z ) E ( L ( S )) ∣ S = 0 = 0.
This completes the proof of Lemma A.3. □
Lemma A.4 (An explicit FCC comparison value). At covolume V = 21 / 20 V=21/20 V = 21/20 , the FCC energy satisfies
− 5.30745025029370902112 < E ( ( 21 / 20 ) 1 / 3 F ) < − 5.30745025029370902111. -5.30745025029370902112<\mathcal{E}((21/20)^{1/3}F)<-5.30745025029370902111. − 5.30745025029370902112 < E (( 21/20 ) 1/3 F ) < − 5.30745025029370902111. Thus E ( ( 21 / 20 ) 1 / 3 F ) < − 26537251 / 5000000 \mathcal{E}((21/20)^{1/3}F)<-26537251/5000000 E (( 21/20 ) 1/3 F ) < − 26537251/5000000 .
Proof. In Lemma 2.8, take X = 1 X=1 X = 1 , V = 21 / 20 V=21/20 V = 21/20 and ( i , j , k ) ∈ [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } (i,j,k)\in[-6,6]^3\cap\mathbb{Z}^3\setminus\{0\} ( i , j , k ) ∈ [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } , with
q ( i , j , k ) = α { ( i + j − k ) 2 + ( i − j ) 2 + k 2 } , q(i,j,k)=\alpha\{(i+j-k)^2+(i-j)^2+k^2\}, q ( i , j , k ) = α {( i + j − k ) 2 + ( i − j ) 2 + k 2 } ,
q ∗ ( i , j , k ) = 3 ( i 2 + j 2 ) + 4 k 2 + 2 i j + 4 k ( i + j ) 4 α , α = ( 441 / 1600 ) 1 / 3 . q^*(i,j,k)=\frac{3(i^2+j^2)+4k^2+2ij+4k(i+j)}{4\alpha},\qquad\alpha=(441/1600)^{1/3}. q ∗ ( i , j , k ) = 4 α 3 ( i 2 + j 2 ) + 4 k 2 + 2 ij + 4 k ( i + j ) , α = ( 441/1600 ) 1/3 .
The rational enclosure is
α − : = 844771120634149438190083412102055 2 110 < α , \alpha_-:=\frac{844771120634149438190083412102055}{2^{110}}<\alpha, α − := 2 110 844771120634149438190083412102055 < α ,
α + : = 844771120634149438190083412102056 2 110 > α . \alpha_+:=\frac{844771120634149438190083412102056}{2^{110}}>\alpha. α + := 2 110 844771120634149438190083412102056 > α .
The raw Gram spectrum lies in [ 1 / 2 , 7 / 2 ] [1/2,7/2] [ 1/2 , 7/2 ] , giving
q ≥ α 2 ∣ n ∣ 2 , q ∗ ≥ 2 7 α ∣ n ∣ 2 . q\geq\frac{\alpha}{2}|n|^2,\qquad q^*\geq\frac{2}{7\alpha}|n|^2. q ≥ 2 α ∣ n ∣ 2 , q ∗ ≥ 7 α 2 ∣ n ∣ 2 .
With Stieltjes order 96, Lemma 2.8 gives attraction < 2581 / 10 24 <2581/10^{24} < 2581/1 0 24 and repulsion < 711 / 10 24 <711/10^{24} < 711/1 0 24 ; adding these tails to the rational cube sum proves the bracket. This completes the proof of Lemma A.4 . □
Numerical lemma A.5 (The finite FCC comparison). For p ∈ D F p\in D_F p ∈ D F ,
U ( p ) Z 3 ∈ argmin ∣ L ∣ = 21 / 20 l o c E ( L ) , U(p)\mathbb{Z}^3\in\underset{|L|=21/20}{\operatorname{argmin}}^{\mathrm{loc}}\ \mathcal{E}(L), U ( p ) Z 3 ∈ ∣ L ∣ = 21/20 argmin loc E ( L ) , 1 2 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ≤ − 26537251 5000000 \frac{1}{2}\mathcal{E}(U(p)\mathbb{Z}^3,6)-\mathcal{E}(U(p)\mathbb{Z}^3,3)\leq-\frac{26537251}{5000000} 2 1 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ≤ − 5000000 26537251 ⟹ min σ ∈ { − 1 , 1 } ∥ 3 M σ T M σ tr ( M σ T M σ ) − I ∥ F ≤ 43 1000 , M σ = U ( p ) U F , σ − 1 . \Longrightarrow\min_{\sigma\in\{-1,1\}}\left\|\frac{3M_\sigma^{\mathsf T}M_\sigma}{\operatorname{tr}(M_\sigma^{\mathsf T}M_\sigma)}-I\right\|_F\leq\frac{43}{1000},\qquad M_\sigma=U(p)U_{F,\sigma}^{-1}. ⟹ σ ∈ { − 1 , 1 } min tr ( M σ T M σ ) 3 M σ T M σ − I F ≤ 1000 43 , M σ = U ( p ) U F , σ − 1 . Here U F , σ U_{F,\sigma} U F , σ is the explicit FCC reference in Lemma 4.8. Thus the recognized lattice is in the unique-stationary basin of Lemma 4.7 .
Proof. Apply Lemma A.2 on D F D_F D F with
V = 21 20 , T = − 26537251 5000000 . V=\frac{21}{20},\qquad T=-\frac{26537251}{5000000}. V = 20 21 , T = − 5000000 26537251 . Use (43) with outward rational endpoints, 21 / 26 ≤ a δ ≤ 105 / 88 21/26\leq a_\delta\leq105/88 21/26 ≤ a δ ≤ 105/88 , 22 / 25 ≤ h ≤ 13 / 10 22/25\leq h\leq13/10 22/25 ≤ h ≤ 13/10 and dual-square threshold 100 / 169 100/169 100/169 . In the basin test use U F , σ U_{F,\sigma} U F , σ and radius 43 / 1000 43/1000 43/1000 . Use the relative entries and outward rational enclosures [ a F ] , [ δ F ] [a_F],[\delta_F] [ a F ] , [ δ F ] of Lemma 4.8. Reference priority and directed splitting use
[ a F ] × [ δ F ] × { 1 / 2 } × { σ / 2 } × { 1 / 3 } , σ ∈ { − 1 , 1 } , [a_F]\times[\delta_F]\times\{1/2\}\times\{\sigma/2\}\times\{1/3\},\qquad\sigma\in\{-1,1\}, [ a F ] × [ δ F ] × { 1/2 } × { σ /2 } × { 1/3 } , σ ∈ { − 1 , 1 } , with Lemma A.1 ’s outward arithmetic; all other splitting scores are unchanged.
With h = 21 / ( 20 a δ ) h=21/(20a_\delta) h = 21/ ( 20 a δ ) , retain all remaining arithmetic, cubes, tails, metric guards, scheduling, selection, Newton and retention rules of Lemma A.2 , hence its original-box invariant.
The single original box, without resumed input, yields 59952 queued visits and 1685 geometric child prunes before insertion. In the order geometry, stationary value, gradient, FCC basin, empty Newton, negative curvature, the terminal counts are
( 1685 , 25602 , 3142 , 52 , 0 , 338 ) . (1685,25602,3142,52,0,338). ( 1685 , 25602 , 3142 , 52 , 0 , 338 ) . There are 8 Newton contractions and Q = U = ∅ Q=\mathcal{U}=\varnothing Q = U = ∅ , with no depth-70 rectangle. Lemma A.2 therefore places every eligible minimum in the unique-stationary FCC basin. This completes the proof of Numerical Lemma A.5. □
Lemma A.6 (A simple-hexagonal stationary parameter at covolume 11 / 10 11/10 11/10 ). Let Ψ ( Y ) = E ( ( 11 / 10 ) 1 / 3 H Y ) \Psi(Y)=\mathcal{E}((11/10)^{1/3}H_Y) Ψ ( Y ) = E (( 11/10 ) 1/3 H Y ) , where
H Y = ( 4 / ( 3 Y ) ) 1 / 6 span Z { ( 1 , 0 , 0 ) , ( 1 / 2 , 3 / 2 , 0 ) , ( 0 , 0 , Y ) } . (81) H_Y=(4/(3Y))^{1/6}\operatorname{span}_{\mathbb{Z}}\{(1,0,0),(1/2,\sqrt{3}/2,0),(0,0,\sqrt{Y})\}. \tag*{(81)} H Y = ( 4/ ( 3 Y ) ) 1/6 span Z {( 1 , 0 , 0 ) , ( 1/2 , 3 /2 , 0 ) , ( 0 , 0 , Y )} . (81)
Its branch critical set is
{ Y ∈ [ 103 / 100 , 103 / 50 ] : Y Ψ ′ ( Y ) = 0 } = { Y ∗ } , \{Y \in[103/100,103/50] : Y\Psi'(Y)=0\}=\{Y_*\}, { Y ∈ [ 103/100 , 103/50 ] : Y Ψ ′ ( Y ) = 0 } = { Y ∗ } ,
1.1013103262 < Y ∗ < 1.1013103264. 1.1013103262<Y_*<1.1013103264. 1.1013103262 < Y ∗ < 1.1013103264.
At the rational reference Y 0 = 110131033 / 10 8 Y_0=110131033/10^8 Y 0 = 110131033/1 0 8 ,
1.63271502522320 ⋅ 10 − 8 < 3 d Ψ d log Y ( Y 0 ) < 1.63271502522322 ⋅ 10 − 8 , 1.63271502522320\cdot10^{-8}<3\frac{d\Psi}{d\log Y}(Y_0)<1.63271502522322\cdot10^{-8}, 1.63271502522320 ⋅ 1 0 − 8 < 3 d log Y d Ψ ( Y 0 ) < 1.63271502522322 ⋅ 1 0 − 8 ,
− 4.968465474380345119512 < Ψ ( Y 0 ) < − 4.968465474380345119511. -4.968465474380345119512<\Psi(Y_0)<-4.968465474380345119511. − 4.968465474380345119512 < Ψ ( Y 0 ) < − 4.968465474380345119511.
At L 0 = ( 11 / 10 ) 1 / 3 H Y 0 L_0=(11/10)^{1/3}HY_0 L 0 = ( 11/10 ) 1/3 H Y 0 ,
E ( L 0 , 3 ) < 3332 500 , E ( L 0 , 6 ) < 339 100 , min v ∈ L 0 ∖ { 0 } ∣ v ∣ > 1. E(L_0,3)<\frac{3332}{500},\qquad E(L_0,6)<\frac{339}{100},\qquad\min_{v\in L_0\setminus\{0\}}|v|>1. E ( L 0 , 3 ) < 500 3332 , E ( L 0 , 6 ) < 100 339 , v ∈ L 0 ∖ { 0 } min ∣ v ∣ > 1.
Use (19) with this L 0 L_0 L 0 and its vertical axis first. Then ∂ u E ( L ( S ) ) ∣ S = 0 = 3 d Ψ / d log Y ( Y 0 ) \partial_u\mathcal{E}(L(S))|_{S=0}=3\,d\Psi/d\log Y(Y_0) ∂ u E ( L ( S )) ∣ S = 0 = 3 d Ψ/ d log Y ( Y 0 ) , and the stationary root lies in this chart with
S ∗ = diag ( 2 d , − d , − d ) , d = Y ∗ − Y 0 Y ∗ + 2 Y 0 , ∥ S ∗ ∥ F < 3 ⋅ 10 − 9 . S_*=\operatorname{diag}(2d,-d,-d),\qquad d=\frac{Y_*-Y_0}{Y_*+2Y_0},\qquad\lVert S_*\rVert_F<3\cdot10^{-9}. S ∗ = diag ( 2 d , − d , − d ) , d = Y ∗ + 2 Y 0 Y ∗ − Y 0 , ∥ S ∗ ∥ F < 3 ⋅ 1 0 − 9 .
Proof. With V = 11 / 10 V=11/10 V = 11/10 fixed, put α = ( 4 V 2 / ( 3 Y ) ) 1 / 3 \alpha=(4V^2/(3Y))^{1/3} α = ( 4 V 2 / ( 3 Y ) ) 1/3 and ( q ˙ , q ˙ ∗ ) = d ( q , q ∗ ) / d log Y (\dot q,\dot q^*)=d(q,q^*)/d\log Y ( q ˙ , q ˙ ∗ ) = d ( q , q ∗ ) / d log Y . For every ( i , j , k ) ∈ Z 3 ∖ { ( 0 , 0 , 0 ) } (i,j,k)\in\mathbb{Z}^3\setminus\{(0,0,0)\} ( i , j , k ) ∈ Z 3 ∖ {( 0 , 0 , 0 )} ,
q = α ( i 2 + i j + j 2 + Y k 2 ) , q ˙ = α 3 { − i 2 − i j − j 2 + 2 Y k 2 } , q=\alpha(i^2+ij+j^2+Yk^2),\qquad\dot q=\frac{\alpha}{3}\{-i^2-ij-j^2+2Yk^2\}, q = α ( i 2 + ij + j 2 + Y k 2 ) , q ˙ = 3 α { − i 2 − ij − j 2 + 2 Y k 2 } ,
q ∗ = α − 1 { 4 3 ( i 2 − i j + j 2 ) + k 2 Y } , q ˙ ∗ = 1 3 α { 4 3 ( i 2 − i j + j 2 ) − 2 k 2 Y } . q^*=\alpha^{-1}\left\{\frac{4}{3}(i^2-ij+j^2)+\frac{k^2}{Y}\right\},\qquad\dot q^*=\frac{1}{3\alpha}\left\{\frac{4}{3}(i^2-ij+j^2)-\frac{2k^2}{Y}\right\}. q ∗ = α − 1 { 3 4 ( i 2 − ij + j 2 ) + Y k 2 } , q ˙ ∗ = 3 α 1 { 3 4 ( i 2 − ij + j 2 ) − Y 2 k 2 } .
Thus ∣ q ˙ ∣ ≤ 2 q / 3 |\dot q|\le2q/3 ∣ q ˙ ∣ ≤ 2 q /3 , ∣ q ˙ ∗ ∣ ≤ 2 q ∗ / 3 |\dot q^*|\le2q^*/3 ∣ q ˙ ∗ ∣ ≤ 2 q ∗ /3 . Differentiate Lemma 2.8 on [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } [-6,6]^3\cap\mathbb{Z}^3\setminus\{0\} [ − 6 , 6 ] 3 ∩ Z 3 ∖ { 0 } , using its rational kernels and Stieltjes order 96. The raw Gram eigenvalues are 1 / 2 , 3 / 2 , Y 1/2,3/2,Y 1/2 , 3/2 , Y . Cubing on the displayed parameter interval, including Y 0 Y_0 Y 0 , gives
α − = 1.13572049 < α < 1.13572051 = α + , q ≥ α − 2 ∣ η ∣ 2 , q ∗ ≥ 2 3 α + ∣ η ∣ 2 . \alpha_-=1.13572049<\alpha<1.13572051=\alpha_+,\qquad q\ge\frac{\alpha_-}{2}|\eta|^2,\qquad q^*\ge\frac{2}{3\alpha_+}|\eta|^2. α − = 1.13572049 < α < 1.13572051 = α + , q ≥ 2 α − ∣ η ∣ 2 , q ∗ ≥ 3 α + 2 ∣ η ∣ 2 .
For the derivative tail, set
a = α − / 2 , b = 2 / ( 3 α + ) , x = 49 π a . a=\alpha_-/2,\qquad b=2/(3\alpha_+),\qquad x=49\pi a. a = α − /2 , b = 2/ ( 3 α + ) , x = 49 π a .
With the complete Gaussian cube tail Θ 6 \Theta_6 Θ 6 of Lemma 2.8, the omitted logarithmic derivative has absolute value at most
2 3 [ ( π 6 240 ∑ j = 0 6 6 ! j ! x j − 6 + π 3 2 ∑ j = 0 3 3 ! j ! x j − 3 ) Θ 6 ( a ) + π 6 240 V + π 3 2 Θ 6 ( b ) ] < 6 ⋅ 10 − 36 . \frac{2}{3}\left[\left(\frac{\pi^6}{240}\sum_{j=0}^{6}\frac{6!}{j!}x^{j-6}+\frac{\pi^3}{2}\sum_{j=0}^{3}\frac{3!}{j!}x^{j-3}\right)\Theta_6(a)+\frac{\pi^6}{240V}+\frac{\pi^3}{2}\Theta_6(b)\right]<6\cdot10^{-36}. 3 2 [ ( 240 π 6 j = 0 ∑ 6 j ! 6 ! x j − 6 + 2 π 3 j = 0 ∑ 3 j ! 3 ! x j − 3 ) Θ 6 ( a ) + 240 V π 6 + 2 π 3 Θ 6 ( b ) ] < 6 ⋅ 1 0 − 36 .
Including this tail, rational endpoint evaluation gives
− 1.32632707729240 ⋅ 10 − 10 < d Ψ d log Y ( 1.1013103262 ) < − 1.32632707729238 ⋅ 10 − 10 , -1.32632707729240\cdot10^{-10}<\frac{d\Psi}{d\log Y}(1.1013103262)<-1.32632707729238\cdot10^{-10}, − 1.32632707729240 ⋅ 1 0 − 10 < d log Y d Ψ ( 1.1013103262 ) < − 1.32632707729238 ⋅ 1 0 − 10 ,
1.60789194743864 ⋅ 10 − 10 < d Ψ d log Y ( 1.1013103264 ) < 1.60789194743866 ⋅ 10 − 10 . 1.60789194743864\cdot10^{-10}<\frac{d\Psi}{d\log Y}(1.1013103264)<1.60789194743866\cdot10^{-10}. 1.60789194743864 ⋅ 1 0 − 10 < d log Y d Ψ ( 1.1013103264 ) < 1.60789194743866 ⋅ 1 0 − 10 .
Continuity and branch uniqueness give the bracket.
At Y 0 Y_0 Y 0 , the rational cube-root endpoints are
1474249495201070240401860238111195 2 110 < α < 1474249495201070240401860238111196 2 110 . \frac{1474249495201070240401860238111195}{2^{110}}<\alpha<\frac{1474249495201070240401860238111196}{2^{110}}. 2 110 1474249495201070240401860238111195 < α < 2 110 1474249495201070240401860238111196 .
Add the complete ( E ( L 0 , 3 ) , E ( L 0 , 6 ) ) (E(L_0,3),E(L_0,6)) ( E ( L 0 , 3 ) , E ( L 0 , 6 )) tails < ( 7 , 4 ) ⋅ 10 − 38 <(7,4)\cdot10^{-38} < ( 7 , 4 ) ⋅ 1 0 − 38 componentwise to obtain the energy and moment bounds. Moreover,
Y 0 > 1 ⟹ min n ∈ Z 3 ∖ { 0 } q n = α > 1. Y_0>1\Longrightarrow\min_{n\in\mathbb{Z}^3\setminus\{0\}}q_n=\alpha>1. Y 0 > 1 ⟹ n ∈ Z 3 ∖ { 0 } min q n = α > 1.
Since ∂ u log Y 0 ( 1 + 2 u ) 1 − u ∣ u = 0 = 3 \partial_u \log\frac{Y_0(1+2u)}{1-u}\big|_{u=0}=3 ∂ u log 1 − u Y 0 ( 1 + 2 u ) u = 0 = 3 , the axial derivative corresponds to S = diag ( 2 u , − u , − u ) S=\operatorname{diag}(2u,-u,-u) S = diag ( 2 u , − u , − u ) , and ∥ S ∗ ∥ F 2 = 6 d 2 < ( 3 ⋅ 10 − 9 ) 2 \lVert S_*\rVert_F^2=6d^2<(3\cdot10^{-9})^2 ∥ S ∗ ∥ F 2 = 6 d 2 < ( 3 ⋅ 1 0 − 9 ) 2 . This completes the proof of Lemma A.6 . □
Numerical lemma A.7 (The finite simple-hexagonal comparison). With D ( 11 / 10 ) \mathcal{D}(11/10) D ( 11/10 ) from Proposition 4.6 and U 0 U_0 U 0 from Lemma 4.10 ,
p ∈ D ( 11 / 10 ) , U ( p ) Z 3 ∈ argmin ∣ L ∣ = 11 / 10 E ( L ) , p\in\mathcal{D}(11/10),\qquad U(p)\mathbb{Z}^3\in\operatorname*{argmin}_{\substack{|L|=11/10}}\mathcal{E}(L), p ∈ D ( 11/10 ) , U ( p ) Z 3 ∈ ∣ L ∣ = 11/10 argmin E ( L ) , 1 2 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ≤ − 24842327 5000000 \frac{1}{2}\mathcal{E}(U(p)\mathbb{Z}^3,6)-\mathcal{E}(U(p)\mathbb{Z}^3,3)\leq-\frac{24842327}{5000000} 2 1 E ( U ( p ) Z 3 , 6 ) − E ( U ( p ) Z 3 , 3 ) ≤ − 5000000 24842327 ⟹ ∥ 3 M T M tr ( M T M ) − I ∥ F ≤ 7 100 , M = U ( p ) U 0 − 1 . \Longrightarrow\qquad\left\lVert\frac{3M^{\mathsf T}M}{\operatorname{tr}(M^{\mathsf T}M)}-I\right\rVert_F\leq\frac{7}{100},\qquad M=U(p)U_0^{-1}. ⟹ tr ( M T M ) 3 M T M − I F ≤ 100 7 , M = U ( p ) U 0 − 1 . Proof. Use Lemma A.2 on D ( 11 / 10 ) \mathcal{D}(11/10) D ( 11/10 ) with
V = 11 10 , T = − 24842327 5000000 , h = 11 10 a δ , 11 13 ≤ a δ ≤ 5 4 , 22 25 ≤ h ≤ 13 10 , V=\frac{11}{10},\qquad T=-\frac{24842327}{5000000},\qquad h=\frac{11}{10a\delta},\qquad\frac{11}{13}\leq a\delta\leq\frac{5}{4},\qquad\frac{22}{25}\leq h\leq\frac{13}{10}, V = 10 11 , T = − 5000000 24842327 , h = 10 a δ 11 , 13 11 ≤ a δ ≤ 4 5 , 25 22 ≤ h ≤ 10 13 , and dual-square threshold 100 / 169 100/169 100/169 . The basin predicate is the strict whole-box S S S -norm test of Lemma 4.10 , with radius 7 / 100 7/100 7/100 . For the closed rational enclosures [ a 0 ] [a_0] [ a 0 ] , [ δ 0 ] [\delta_0] [ δ 0 ] in that lemma, set
B r e f = [ a 0 ] × [ δ 0 ] × { 1 / 2 } × { 0 } 2 . B_{\mathrm{ref}}=[a_0]\times[\delta_0]\times\{1/2\}\times\{0\}^2. B ref = [ a 0 ] × [ δ 0 ] × { 1/2 } × { 0 } 2 . This enclosure governs reference priority and directed splitting.
Use r ( B ) r(B) r ( B ) , r i ( B ) r_i(B) r i ( B ) and i m e t ( B ) i_{\mathrm{met}}(B) i met ( B ) from Lemma A.2 , with this B r e f B_{\mathrm{ref}} B ref . Retain the common rules, with selector:
i ( B ) = { min argmin 1 ≤ i ≤ 5 r i ( B ) , B ∩ B r e f ≠ ∅ , min i r i ( B ) < 999 1000 r ( B ) , i m e t ( B ) , otherwise . i(B)=\begin{cases}
\min\operatorname*{argmin}_{1\leq i\leq5}r_i(B), & B\cap B_{\mathrm{ref}}\ne\varnothing,\quad\min_i r_i(B)<\frac{999}{1000}r(B),\\
i_{\mathrm{met}}(B), & \text{otherwise}.
\end{cases} i ( B ) = { min argmin 1 ≤ i ≤ 5 r i ( B ) , i met ( B ) , B ∩ B ref = ∅ , min i r i ( B ) < 1000 999 r ( B ) , otherwise . Use B = B i ( B ) − ∪ B i ( B ) + B=B^-_{i(B)}\cup B^+_{i(B)} B = B i ( B ) − ∪ B i ( B ) + , retaining both closed children unless a certified interval predicate excludes one. All other rules, including the strict whole-box guards and original-box invariant, remain those of Lemma A.2 . From one closed root: 49389 queued visits, 776 geometric prunes before insertion, 604 retained Newton contractions, no depth-70 box, and Q = U = ∅ Q=U=\varnothing Q = U = ∅ . Terminal counts (geometry, value, gradient, SH recognition, empty Newton, curvature) are ( 776 , 20229 , 3703 , 66 , 59 , 250 ) (776,20229,3703,66,59,250) ( 776 , 20229 , 3703 , 66 , 59 , 250 ) . This completes the proof of Numerical Lemma A.7. □
Numerical lemma A.8 (Moment bounds along two shape intervals). For X ∈ [ 23 / 100 , 307 / 1000 ] X\in[23/100,307/1000] X ∈ [ 23/100 , 307/1000 ] and Y ∈ [ 103 / 100 , 103 / 50 ] Y\in[103/100,103/50] Y ∈ [ 103/100 , 103/50 ] ,
7.22696052 < E ( F , 3 ) < 7.22696053 , 7.22696052<\mathcal{E}(F,3)<7.22696053, 7.22696052 < E ( F , 3 ) < 7.22696053 , 7.39056 < E ( L X , 3 ) < 7.88582 , 23 100 ≤ X ≤ 307 1000 , 7.39056<\mathcal{E}(L_X,3)<7.88582,\qquad\frac{23}{100}\leq X\leq\frac{307}{1000}, 7.39056 < E ( L X , 3 ) < 7.88582 , 100 23 ≤ X ≤ 1000 307 , 7.94245 < E ( H Y , 3 ) < 11.28693 , 103 100 ≤ Y ≤ 103 50 , 7.94245<\mathcal{E}(H_Y,3)<11.28693,\qquad\frac{103}{100}\leq Y\leq\frac{103}{50}, 7.94245 < E ( H Y , 3 ) < 11.28693 , 100 103 ≤ Y ≤ 50 103 , d d log X E ( L X , 3 ) < 0 , \frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(L_X,3)<0, d log X d E ( L X , 3 ) < 0 , − 5.411 < d d log X ( d E ( L X , 6 ) / d log X d E ( L X , 3 ) / d log X ) < − 3.485 , -5.411<\frac{\mathrm{d}}{\mathrm{d}\log X}\left(\frac{\mathrm{d}\mathcal{E}(L_X,6)/\mathrm{d}\log X}{\mathrm{d}\mathcal{E}(L_X,3)/\mathrm{d}\log X}\right)<-3.485, − 5.411 < d log X d ( d E ( L X , 3 ) / d log X d E ( L X , 6 ) / d log X ) < − 3.485 , d d log Y E ( H Y , 3 ) > 0 , \frac{\mathrm{d}}{\mathrm{d}\log Y}\mathcal{E}(H_Y,3)>0, d log Y d E ( H Y , 3 ) > 0 , .969 < d d log Y ( d E ( H Y , 6 ) / d log Y d E ( H Y , 3 ) / d log Y ) < 6.256. .969<\frac{\mathrm{d}}{\mathrm{d}\log Y}\left(\frac{\mathrm{d}\mathcal{E}(H_Y,6)/\mathrm{d}\log Y}{\mathrm{d}\mathcal{E}(H_Y,3)/\mathrm{d}\log Y}\right)<6.256. .969 < d log Y d ( d E ( H Y , 3 ) / d log Y d E ( H Y , 6 ) / d log Y ) < 6.256.
At the endpoints the derivative ratios satisfy
3.41933 < d E ( L X , 6 ) / d log X d E ( L X , 3 ) / d log X ∣ X = 23 / 100 < 3.41935 , 3.41933 < \left.\frac{\mathrm{d}E(L_X,6)/\mathrm{d}\log X}{\mathrm{d}E(L_X,3)/\mathrm{d}\log X}\right|_{X=23/100} < 3.41935, 3.41933 < d E ( L X , 3 ) / d log X d E ( L X , 6 ) / d log X X = 23/100 < 3.41935 ,
2.19169 < d E ( L X , 6 ) / d log X d E ( L X , 3 ) / d log X ∣ X = 307 / 1000 < 2.19170 , 2.19169 < \left.\frac{\mathrm{d}E(L_X,6)/\mathrm{d}\log X}{\mathrm{d}E(L_X,3)/\mathrm{d}\log X}\right|_{X=307/1000} < 2.19170, 2.19169 < d E ( L X , 3 ) / d log X d E ( L X , 6 ) / d log X X = 307/1000 < 2.19170 ,
2.17722 < d E ( H Y , 6 ) / d log Y d E ( H Y , 3 ) / d log Y ∣ Y = 103 / 100 < 2.17724 , 2.17722 < \left.\frac{\mathrm{d}E(H_Y,6)/\mathrm{d}\log Y}{\mathrm{d}E(H_Y,3)/\mathrm{d}\log Y}\right|_{Y=103/100} < 2.17724, 2.17722 < d E ( H Y , 3 ) / d log Y d E ( H Y , 6 ) / d log Y Y = 103/100 < 2.17724 ,
3.38200 < d E ( H Y , 6 ) / d log Y d E ( H Y , 3 ) / d log Y ∣ Y = 103 / 50 < 3.38202. 3.38200 < \left.\frac{\mathrm{d}E(H_Y,6)/\mathrm{d}\log Y}{\mathrm{d}E(H_Y,3)/\mathrm{d}\log Y}\right|_{Y=103/50} < 3.38202. 3.38200 < d E ( H Y , 3 ) / d log Y d E ( H Y , 6 ) / d log Y Y = 103/50 < 3.38202.
Numerical lemma A.9 (Signs at the transition brackets). The transition covolumes satisfy
1.0648235667 < V F T < 1.0648235669 , 1.0648235667 < V_{\mathrm{FT}} < 1.0648235669, 1.0648235667 < V FT < 1.0648235669 , 1.0759906200 < V T H < 1.0759906201. (82) 1.0759906200 < V_{\mathrm{TH}} < 1.0759906201. \tag*{(82)} 1.0759906200 < V TH < 1.0759906201. (82) For the normalized energy differences D T F , D T H D_{\mathrm{TF}},D_{\mathrm{TH}} D TF , D TH of Lemma 2.6 ,
8.2 ⋅ 10 − 11 < D T F ( 1.0648235667 ) < 12.3 ⋅ 10 − 11 , 8.2 \cdot10^{-11} < D_{\mathrm{TF}}(1.0648235667) < 12.3 \cdot10^{-11}, 8.2 ⋅ 1 0 − 11 < D TF ( 1.0648235667 ) < 12.3 ⋅ 1 0 − 11 , − 7.5 ⋅ 10 − 11 < D T F ( 1.0648235669 ) < − 3.4 ⋅ 10 − 11 , -7.5 \cdot10^{-11} < D_{\mathrm{TF}}(1.0648235669) < -3.4 \cdot10^{-11}, − 7.5 ⋅ 1 0 − 11 < D TF ( 1.0648235669 ) < − 3.4 ⋅ 1 0 − 11 , − 11.1 ⋅ 10 − 11 < D T H ( 1.0759906200 ) < − 7.5 ⋅ 10 − 11 , -11.1 \cdot10^{-11} < D_{\mathrm{TH}}(1.0759906200) < -7.5 \cdot10^{-11}, − 11.1 ⋅ 1 0 − 11 < D TH ( 1.0759906200 ) < − 7.5 ⋅ 1 0 − 11 , 13.5 ⋅ 10 − 11 < D T H ( 1.0759906201 ) < 17.1 ⋅ 10 − 11 . 13.5 \cdot10^{-11} < D_{\mathrm{TH}}(1.0759906201) < 17.1 \cdot10^{-11}. 13.5 ⋅ 1 0 − 11 < D TH ( 1.0759906201 ) < 17.1 ⋅ 1 0 − 11 .
Verification of Lemmas A.8 and A.9 . For s = 3 , 6 s=3,6 s = 3 , 6 and a unit-covolume Gram matrix Q Q Q , use the kernels of Lemma 2.8 and the Poisson identity
E ( Q , s ) = π s Γ ( s ) [ 1 s − 3 2 − 1 s + ∑ ( i , j , k ) ∈ Z 3 ∖ { ( 0 , 0 , 0 ) } G s ( π q ( i , j , k ) ) + ∑ ( i , j , k ) ∈ Z 3 ∖ { ( 0 , 0 , 0 ) } G 3 2 − s ( π q ∗ ( i , j , k ) ) ] . E(Q,s)=\frac{\pi^s}{\Gamma(s)}\left[\frac{1}{s-\frac{3}{2}}-\frac{1}{s}+\sum_{(i,j,k)\in\mathbb{Z}^3\setminus\{(0,0,0)\}}G_s(\pi q(i,j,k))+\sum_{(i,j,k)\in\mathbb{Z}^3\setminus\{(0,0,0)\}}G_{\frac{3}{2}-s}(\pi q^*(i,j,k))\right]. E ( Q , s ) = Γ ( s ) π s s − 2 3 1 − s 1 + ( i , j , k ) ∈ Z 3 ∖ {( 0 , 0 , 0 )} ∑ G s ( π q ( i , j , k )) + ( i , j , k ) ∈ Z 3 ∖ {( 0 , 0 , 0 )} ∑ G 2 3 − s ( π q ∗ ( i , j , k )) .
where q = n T Q n q=n^TQn q = n T Q n and q ∗ = n T Q − 1 n q^*=n^TQ^{-1}n q ∗ = n T Q − 1 n for n = ( i , j , k ) T n=(i,j,k)^T n = ( i , j , k ) T . For L X L_X L X , put c = ( 4 X ) − 1 / 3 c=(4X)^{-1/3} c = ( 4 X ) − 1/3 and use
q = c [ X ( i − j − k ) 2 + ( i + j ) 2 + k 2 ] , q=c\left[X(i-j-k)^2+(i+j)^2+k^2\right], q = c [ X ( i − j − k ) 2 + ( i + j ) 2 + k 2 ] ,
q ∗ = 1 4 c [ X − 1 ( i − j ) 2 + ( i + j ) 2 + ( i − j + 2 k ) 2 ] . q^*=\frac{1}{4c}\left[X^{-1}(i-j)^2+(i+j)^2+(i-j+2k)^2\right]. q ∗ = 4 c 1 [ X − 1 ( i − j ) 2 + ( i + j ) 2 + ( i − j + 2 k ) 2 ] .
For H Y H_Y H Y , put c = ( 4 / ( 3 Y ) ) 1 / 3 c=(4/(3Y))^{1/3} c = ( 4/ ( 3 Y ) ) 1/3 and use
q = c ( i 2 + i j + j 2 + Y k 2 ) , q ∗ = c − 1 [ 4 3 ( i 2 − i j + j 2 ) + Y − 1 k 2 ] . q=c(i^2+ij+j^2+Yk^2),\qquad q^*=c^{-1}\left[\frac{4}{3}(i^2-ij+j^2)+Y^{-1}k^2\right]. q = c ( i 2 + ij + j 2 + Y k 2 ) , q ∗ = c − 1 [ 3 4 ( i 2 − ij + j 2 ) + Y − 1 k 2 ] .
The change j ↦ − j j\mapsto-j j ↦ − j shows that i 2 − i j + j 2 i^2-ij+j^2 i 2 − ij + j 2 may be replaced by i 2 + i j + j 2 i^2+ij+j^2 i 2 + ij + j 2 when the symmetric index cube is summed. For q 1 , q 2 ≥ 0 q_1,q_2\geq0 q 1 , q 2 ≥ 0 and logarithmic exponents a 1 , a 2 ∈ { − 2 / 3 , − 1 / 3 , 1 / 3 , 2 / 3 } a_1,a_2\in\{-2/3,-1/3,1/3,2/3\} a 1 , a 2 ∈ { − 2/3 , − 1/3 , 1/3 , 2/3 } , use
q = q 1 + q 2 , q t = a 1 q 1 + a 2 q 2 , q t t = a 1 2 q 1 + a 2 2 q 2 , q=q_1+q_2,\qquad q_t=a_1q_1+a_2q_2,\qquad q_{tt}=a_1^2q_1+a_2^2q_2, q = q 1 + q 2 , q t = a 1 q 1 + a 2 q 2 , q tt = a 1 2 q 1 + a 2 2 q 2 ,
d K ( q ) d t = K ′ ( q ) q t , d 2 K ( q ) d t 2 = K ′ ′ ( q ) q t 2 + K ′ ( q ) q t t . \frac{\mathrm{d}K(q)}{\mathrm{d}t}=K'(q)q_t,\qquad\frac{\mathrm{d}^2K(q)}{\mathrm{d}t^2}=K''(q)q_t^2+K'(q)q_{tt}. d t d K ( q ) = K ′ ( q ) q t , d t 2 d 2 K ( q ) = K ′′ ( q ) q t 2 + K ′ ( q ) q tt .
For both forms, sum on [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } [-12,12]^3\cap\mathbb{Z}^3\setminus\{0\} [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } . On every interval below, Sylvester’s criterion gives
Q > 1 5 I , tr Q < 5 , det Q = 1 ⟹ Q − 1 > 1 5 I , q , q ∗ ≥ 169 5 > 20 Q>\frac{1}{5}I,\qquad\operatorname{tr}Q<5,\qquad\det Q=1\quad\Longrightarrow\quad Q^{-1}>\frac{1}{5}I,\qquad q,q^*\geq\frac{169}{5}>20 Q > 5 1 I , tr Q < 5 , det Q = 1 ⟹ Q − 1 > 5 1 I , q , q ∗ ≥ 5 169 > 20
for omitted indices. For the four kernels K ( q ) = π s G s ( π q ) / Γ ( s ) K(q)=\pi^sG_s(\pi q)/\Gamma(s) K ( q ) = π s G s ( π q ) /Γ ( s ) and K ( q ) = π s G 3 / 2 − s ( π q ) / Γ ( s ) K(q)=\pi^sG_{3/2-s}(\pi q)/\Gamma(s) K ( q ) = π s G 3/2 − s ( π q ) /Γ ( s ) ,
∣ K ( j ) ( q ) ∣ ≤ 5 e − 3 q ( 0 ≤ j ≤ 2 , q ≥ 20 ) , ∣ q t ∣ ≤ 2 3 q , ∣ q t t ∣ ≤ 4 9 q . \lvert K^{(j)}(q)\rvert\leq5e^{-3q}\quad(0\leq j\leq2,\ q\geq20),\qquad\lvert q_t\rvert\leq\frac{2}{3}q,\qquad\lvert q_{tt}\rvert\leq\frac{4}{9}q. ∣ K ( j ) ( q )∣ ≤ 5 e − 3 q ( 0 ≤ j ≤ 2 , q ≥ 20 ) , ∣ q t ∣ ≤ 3 2 q , ∣ q tt ∣ ≤ 9 4 q .
Thus G a ( z ) ≤ e − z 2 / ( z − 7 ) G_a(z) \le e^{-z^2/(z-7)} G a ( z ) ≤ e − z 2 / ( z − 7 ) , a ≤ 8 a \le8 a ≤ 8 , z > 7 z > 7 z > 7 , bounds each omitted summand and its first two logarithmic derivatives by 10 e − 2 q 10e^{-2q} 10 e − 2 q , with complete remainder
τ : = 120 e − 169 π / 8 ( 1 + 8 ) 2 1 − e − 27 π / 8 . \tau:= 120\frac{e^{-169\pi/8}(1+\sqrt{8})^2}{1-e^{-27\pi/8}}. τ := 120 1 − e − 27 π /8 e − 169 π /8 ( 1 + 8 ) 2 .
Indeed 2 q ≥ 2 ∣ n ∣ 2 / 5 > ( π / 8 ) ∣ n ∣ 2 2q \ge2|n|^{2/5} > (\pi/8)|n|^2 2 q ≥ 2∣ n ∣ 2/5 > ( π /8 ) ∣ n ∣ 2 , and the Gaussian cube-tail estimate applies to each of the primal and dual sums.
For Lemma A.8, the closed cells are
23 100 + 77 j 256000 ≤ X ≤ 23 100 + 77 ( j + 1 ) 256000 , j ∈ Z , 0 ≤ j < 256 , \frac{23}{100}+\frac{77j}{256000}\le X\le\frac{23}{100}+\frac{77(j+1)}{256000},\qquad j\in\mathbb{Z},\quad0\le j<256, 100 23 + 256000 77 j ≤ X ≤ 100 23 + 256000 77 ( j + 1 ) , j ∈ Z , 0 ≤ j < 256 ,
103 100 + 103 j 102400 ≤ Y ≤ 103 100 + 103 ( j + 1 ) 102400 , j ∈ Z , 0 ≤ j < 1024. \frac{103}{100}+\frac{103j}{102400}\le Y\le\frac{103}{100}+\frac{103(j+1)}{102400},\qquad j\in\mathbb{Z},\quad0\le j<1024. 100 103 + 102400 103 j ≤ Y ≤ 100 103 + 102400 103 ( j + 1 ) , j ∈ Z , 0 ≤ j < 1024.
Outward evaluation at rational endpoints uses G a + 1 ( z ) = [ a G a ( z ) + e − z ] / z G_{a+1}(z)=[aG_a(z)+e^{-z}]/z G a + 1 ( z ) = [ a G a ( z ) + e − z ] / z , the half-integer alternating series (z ≤ 2 z\le2 z ≤ 2 ) and gamma Stieltjes bounds (z > 2 z>2 z > 2 ), giving
∂ log X ( ∂ log X E ( L X , 6 ) ∂ log X E ( L X , 3 ) ) ∈ ( − 5.410694 , − 3.485156 ) , \partial_{\log X}\left(\frac{\partial_{\log X}E(L_X,6)}{\partial_{\log X}E(L_X,3)}\right)\in(-5.410694,-3.485156), ∂ l o g X ( ∂ l o g X E ( L X , 3 ) ∂ l o g X E ( L X , 6 ) ) ∈ ( − 5.410694 , − 3.485156 ) ,
∂ log Y ( ∂ log Y E ( H Y , 6 ) ∂ log Y E ( H Y , 3 ) ) ∈ ( .969883 , 6.255668 ) . \partial_{\log Y}\left(\frac{\partial_{\log Y}E(H_Y,6)}{\partial_{\log Y}E(H_Y,3)}\right)\in(.969883,6.255668). ∂ l o g Y ( ∂ l o g Y E ( H Y , 3 ) ∂ l o g Y E ( H Y , 6 ) ) ∈ ( .969883 , 6.255668 ) .
For the endpoint evaluations in Lemma A.9 , use
V ∈ { 1.0648235667 , 1.0648235669 } : ( X 0 , Y 0 ) = ( .3012583513613 , 1.0488884438723 ) , V\in\{1.0648235667,1.0648235669\}:\quad(X_0,Y_0)=(.3012583513613,1.0488884438723), V ∈ { 1.0648235667 , 1.0648235669 } : ( X 0 , Y 0 ) = ( .3012583513613 , 1.0488884438723 ) ,
V ∈ { 1.0759906200 , 1.0759906201 } : ( X 0 , Y 0 ) = ( .2976831541610 , 1.0617382352094 ) . V\in\{1.0759906200,1.0759906201\}:\quad(X_0,Y_0)=(.2976831541610,1.0617382352094). V ∈ { 1.0759906200 , 1.0759906201 } : ( X 0 , Y 0 ) = ( .2976831541610 , 1.0617382352094 ) .
For the logarithmic objectives
f T ( t ) = E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) , f_T(t)=E(L e^t,6)-2V^2E(L e^t,3), f T ( t ) = E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) ,
f H ( t ) = E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) , f_H(t)=E(H e^t,6)-2V^2E(H e^t,3), f H ( t ) = E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) ,
( f , z 0 , m ) ∈ { ( f T , X 0 , 3.6 ) , ( f H , Y 0 , 6.1 ) } , t 0 = log z 0 , (f,z_0,m)\in\{(f_T,X_0,3.6),(f_H,Y_0,6.1)\},\qquad t_0=\log z_0, ( f , z 0 , m ) ∈ {( f T , X 0 , 3.6 ) , ( f H , Y 0 , 6.1 )} , t 0 = log z 0 ,
z ∈ [ z 0 − 10 − 4 , z 0 + 10 − 4 ] , t = log z ∈ [ log ( z 0 − 10 − 4 ) , log ( z 0 + 10 − 4 ) ] , z\in[z_0-10^{-4},z_0+10^{-4}],\qquad t=\log z\in[\log(z_0-10^{-4}),\log(z_0+10^{-4})], z ∈ [ z 0 − 1 0 − 4 , z 0 + 1 0 − 4 ] , t = log z ∈ [ log ( z 0 − 1 0 − 4 ) , log ( z 0 + 1 0 − 4 )] ,
f ′ ( log ( z 0 − 10 − 4 ) ) < 0 < f ′ ( log ( z 0 + 10 − 4 ) ) , f ′ ′ ( t ) > m . f'(\log(z_0-10^{-4}))<0<f'(\log(z_0+10^{-4})),\qquad f''(t)>m. f ′ ( log ( z 0 − 1 0 − 4 )) < 0 < f ′ ( log ( z 0 + 1 0 − 4 )) , f ′′ ( t ) > m .
For t ∗ t_* t ∗ in this interval with f ′ ( t ∗ ) = 0 f'(t_*)=0 f ′ ( t ∗ ) = 0 ,
f ( t 0 ) − ∣ f ′ ( t 0 ) ∣ 2 2 m ≤ f ( t ∗ ) ≤ f ( t 0 ) . f(t_0)-\frac{|f'(t_0)|^2}{2m}\le f(t_*)\le f(t_0). f ( t 0 ) − 2 m ∣ f ′ ( t 0 ) ∣ 2 ≤ f ( t ∗ ) ≤ f ( t 0 ) .
With f T ′ ( t T ∗ ) = f H ′ ( t H ∗ ) = 0 f'_T(t^*_T)=f'_H(t^*_H)=0 f T ′ ( t T ∗ ) = f H ′ ( t H ∗ ) = 0 ,
D T F ( V ) = f T ( t T ∗ ) − E ( F , 6 ) + 2 V 2 E ( F , 3 ) , D_{TF}(V)=f_T(t^*_T)-E(F,6)+2V^2E(F,3), D T F ( V ) = f T ( t T ∗ ) − E ( F , 6 ) + 2 V 2 E ( F , 3 ) ,
D T H ( V ) = f T ( t T ∗ ) − f H ( t H ∗ ) . D_{TH}(V)=f_T(t^*_T)-f_H(t^*_H). D T H ( V ) = f T ( t T ∗ ) − f H ( t H ∗ ) .
This completes the proof of Numerical Lemmas A.8 and A.9. □
Numerical lemma A.10 (Ten-decimal threshold locations). The two transition covolumes satisfy
1.06482356682 < V F T < 1.06482356684 , 1.07599062003 < V T H < 1.07599062004. 1.06482356682<V_{FT}<1.06482356684,\qquad1.07599062003<V_{TH}<1.07599062004. 1.06482356682 < V F T < 1.06482356684 , 1.07599062003 < V T H < 1.07599062004. More precisely, with D T F , D T H D_{TF},D_{TH} D T F , D T H as in Lemma A.9 ,
7.4 < 10 12 D T F ( 1.06482356682 ) < 9.3 , 7.4<10^{12}D_{TF}(1.06482356682)<9.3, 7.4 < 1 0 12 D T F ( 1.06482356682 ) < 9.3 , − 8.3 < 10 12 D T F ( 1.06482356684 ) < − 6.4 , -8.3<10^{12}D_{TF}(1.06482356684)<-6.4, − 8.3 < 1 0 12 D T F ( 1.06482356684 ) < − 6.4 , − 20.1 < 10 12 D T H ( 1.07599062003 ) < − 17.9 , -20.1<10^{12}D_{TH}(1.07599062003)<-17.9, − 20.1 < 1 0 12 D T H ( 1.07599062003 ) < − 17.9 , 4.5 < 10 12 D T H ( 1.07599062004 ) < 6.7. 4.5<10^{12}D_{TH}(1.07599062004)<6.7. 4.5 < 1 0 12 D T H ( 1.07599062004 ) < 6.7.
Proof. Use the same [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } [-12,12]^3 \cap\mathbb{Z}^3 \setminus\{0\} [ − 12 , 12 ] 3 ∩ Z 3 ∖ { 0 } primal and dual sums, rational kernels and complete remainder τ \tau τ through logarithmic order two. The exact grouped pairs and multiplicities of Lemma A.12 are ordered lexicographically, primal before dual:
1963 ⏟ L X , p r i m a l , 1979 ⏟ L X , d u a l , 1130 ⏟ H Y , p r i m a l / d u a l , \underbrace{1963}_{L_X,\mathrm{primal}},\qquad\underbrace{1979}_{L_X,\mathrm{dual}},\qquad\underbrace{1130}_{H_Y,\mathrm{primal/dual}}, L X , primal 1963 , L X , dual 1979 , H Y , primal/dual 1130 ,
total multiplicity per set = 25 3 − 1 = 15624. \text{total multiplicity per set}=25^3-1=15624. total multiplicity per set = 2 5 3 − 1 = 15624.
Moment lists prepend π 3 / 6 \pi^3/6 π 3 /6 or π 6 / 2160 \pi^6/2160 π 6 /2160 ; derivative lists prepend no constant. Append every row contribution. For a current list ( b 0 , … , b n − 1 ) (b_0,\ldots,b_{n-1}) ( b 0 , … , b n − 1 ) , balanced addition is
b j n e w = { b 2 j + b 2 j + 1 , 2 j + 1 < n , b 2 j , 2 j + 1 = n , 0 ≤ j < ⌈ n / 2 ⌉ , n n e w = ⌈ n / 2 ⌉ . b_j^{\mathrm{new}}=
\begin{cases}
b_{2j}+b_{2j+1}, & 2j+1<n,\\
b_{2j}, & 2j+1=n,
\end{cases}
\qquad0\leq j<\lceil n/2\rceil,\qquad n^{\mathrm{new}}=\lceil n/2\rceil. b j new = { b 2 j + b 2 j + 1 , b 2 j , 2 j + 1 < n , 2 j + 1 = n , 0 ≤ j < ⌈ n /2 ⌉ , n new = ⌈ n /2 ⌉ .
Iterate to n = 1 n=1 n = 1 , with outward operations as in Lemma A.2 and final tail interval
{ [ 0 , τ ] , r = 0 , [ − τ , τ ] , r = 1 , 2 , r = logarithmic derivative order . \begin{cases}
[0,\tau], & r=0,\\
[-\tau,\tau], & r=1,2,
\end{cases}
\qquad r=\text{logarithmic derivative order}. { [ 0 , τ ] , [ − τ , τ ] , r = 0 , r = 1 , 2 , r = logarithmic derivative order .
The kernels, indices and remainders are unchanged.
At each listed covolume V V V , the stationary objectives are
f T ( t ) = E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) , f H ( t ) = E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) . f_T(t)=E(L_{e^t},6)-2V^2E(L_{e^t},3),\qquad f_H(t)=E(H_{e^t},6)-2V^2E(H_{e^t},3). f T ( t ) = E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) , f H ( t ) = E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) .
The complete evaluations are
V ∈ { 1.06482356682 , 1.06482356684 } : X c = 0.3012583513613 , V\in\{1.06482356682,1.06482356684\}:\qquad X_c=0.3012583513613, V ∈ { 1.06482356682 , 1.06482356684 } : X c = 0.3012583513613 ,
( f , z c , m ) = ( f T , X c , 3.6 ) ; (f,z_c,m)=(f_T,X_c,3.6); ( f , z c , m ) = ( f T , X c , 3.6 ) ;
V ∈ { 1.07599062003 , 1.07599062004 } : ( X c , Y c ) = ( 0.2976831541610 , 1.0617382352094 ) , V\in\{1.07599062003,1.07599062004\}:\qquad(X_c,Y_c)=(0.2976831541610,1.0617382352094), V ∈ { 1.07599062003 , 1.07599062004 } : ( X c , Y c ) = ( 0.2976831541610 , 1.0617382352094 ) ,
( f , z c , m ) ∈ { ( f T , X c , 3.6 ) , ( f H , Y c , 6.1 ) } . (f,z_c,m)\in\{(f_T,X_c,3.6),(f_H,Y_c,6.1)\}. ( f , z c , m ) ∈ {( f T , X c , 3.6 ) , ( f H , Y c , 6.1 )} .
For every indicated triple,
z ∈ [ z c − 10 − 4 , z c + 10 − 4 ] , t 0 = log z c , z\in[z_c-10^{-4},z_c+10^{-4}],\qquad t_0=\log z_c, z ∈ [ z c − 1 0 − 4 , z c + 1 0 − 4 ] , t 0 = log z c ,
t = log z ∈ [ log ( z c − 10 − 4 ) , log ( z c + 10 − 4 ) ] , t=\log z\in[\log(z_c-10^{-4}),\log(z_c+10^{-4})], t = log z ∈ [ log ( z c − 1 0 − 4 ) , log ( z c + 1 0 − 4 )] ,
f ′ ( log ( z c − 10 − 4 ) ) < 0 < f ′ ( log ( z c + 10 − 4 ) ) , f ′ ′ ( t ) > m , f'(\log(z_c-10^{-4}))<0<f'(\log(z_c+10^{-4})),\qquad f''(t)>m, f ′ ( log ( z c − 1 0 − 4 )) < 0 < f ′ ( log ( z c + 1 0 − 4 )) , f ′′ ( t ) > m ,
∃ ! t ∗ ∈ [ log ( z c − 10 − 4 ) , log ( z c + 10 − 4 ) ] : f ′ ( t ∗ ) = 0 , \exists!\,t_*\in[\log(z_c-10^{-4}),\log(z_c+10^{-4})]:\qquad f'(t_*)=0, ∃ ! t ∗ ∈ [ log ( z c − 1 0 − 4 ) , log ( z c + 1 0 − 4 )] : f ′ ( t ∗ ) = 0 ,
f ( t 0 ) − ∣ f ′ ( t 0 ) ∣ 2 2 m ≤ f ( t ∗ ) ≤ f ( t 0 ) . f(t_0)-\frac{\lvert f'(t_0)\rvert^2}{2m}\leq f(t_*)\leq f(t_0). f ( t 0 ) − 2 m ∣ f ′ ( t 0 ) ∣ 2 ≤ f ( t ∗ ) ≤ f ( t 0 ) .
No bisection is used. With f T ′ ( t T ∗ ) = f H ′ ( t H ∗ ) = 0 f_T'(t_T^*)=f_H'(t_H^*)=0 f T ′ ( t T ∗ ) = f H ′ ( t H ∗ ) = 0 ,
D T F ( V ) = E ( L e t T ∗ , 6 ) − E ( F , 6 ) − 2 V 2 { E ( L e t T ∗ , 3 ) − E ( F , 3 ) } , D_{TF}(V)=E(L_{e^{t_T^*}},6)-E(F,6)-2V^2\{E(L_{e^{t_T^*}},3)-E(F,3)\}, D T F ( V ) = E ( L e t T ∗ , 6 ) − E ( F , 6 ) − 2 V 2 { E ( L e t T ∗ , 3 ) − E ( F , 3 )} ,
D T H ( V ) = E ( L e t H ∗ , 6 ) − E ( H e t H ∗ , 6 ) D_{TH}(V)=E(L_{e^{t_H^*}},6)-E(H_{e^{t_H^*}},6) D T H ( V ) = E ( L e t H ∗ , 6 ) − E ( H e t H ∗ , 6 )
− 2 V 2 { E ( L e t H ∗ , 3 ) − E ( H e t H ∗ , 3 ) } . \qquad-2V^2\{E(L_{e^{t_H^*}},3)-E(H_{e^{t_H^*}},3)\}. − 2 V 2 { E ( L e t H ∗ , 3 ) − E ( H e t H ∗ , 3 )} .
This completes the proof of Numerical Lemma A.10.
Numerical lemma A.11 (Derivative signs on the full comparison intervals). Let 21 / 20 ≤ V ≤ 11 / 10 21/20 \le V \le11/10 21/20 ≤ V ≤ 11/10 . The following inequalities hold throughout the displayed intervals:
d d log X E ( V 1 / 3 L X ) < 0 , X ∈ [ 77 / 1000 , 23 / 100 ] ∪ [ 13 / 25 , 49 / 50 ] , \frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(V^{1/3}L_X)<0,\qquad X\in[77/1000,23/100]\cup[13/25,49/50], d log X d E ( V 1/3 L X ) < 0 , X ∈ [ 77/1000 , 23/100 ] ∪ [ 13/25 , 49/50 ] , d d log X E ( V 1 / 3 L X ) > 0 , X ∈ [ 307 / 1000 , 12 / 25 ] ∪ [ 51 / 50 , 21 ] , \frac{\mathrm{d}}{\mathrm{d}\log X}\mathcal{E}(V^{1/3}L_X)>0,\qquad X\in[307/1000,12/25]\cup[51/50,21], d log X d E ( V 1/3 L X ) > 0 , X ∈ [ 307/1000 , 12/25 ] ∪ [ 51/50 , 21 ] , d 2 d ( log X ) 2 [ 2 V 4 E ( V 1 / 3 L X ) ] < 87 100 , X ∈ [ 12 / 25 , 13 / 25 ] , \frac{\mathrm{d}^{2}}{\mathrm{d}(\log X)^{2}}\left[2V^{4}\mathcal{E}(V^{1/3}L_X)\right]<\frac{87}{100},\qquad X\in[12/25,13/25], d ( log X ) 2 d 2 [ 2 V 4 E ( V 1/3 L X ) ] < 100 87 , X ∈ [ 12/25 , 13/25 ] , d 2 d ( log X ) 2 [ 2 V 4 E ( V 1 / 3 L X ) ] > 86 100 , X ∈ [ 49 / 50 , 51 / 50 ] , \frac{\mathrm{d}^{2}}{\mathrm{d}(\log X)^{2}}\left[2V^{4}\mathcal{E}(V^{1/3}L_X)\right]>\frac{86}{100},\qquad X\in[49/50,51/50], d ( log X ) 2 d 2 [ 2 V 4 E ( V 1/3 L X ) ] > 100 86 , X ∈ [ 49/50 , 51/50 ] , d d log Y E ( V 1 / 3 H Y ) < 0 , Y ∈ [ 27 / 100 , 103 / 100 ] , \frac{\mathrm{d}}{\mathrm{d}\log Y}\mathcal{E}(V^{1/3}H_Y)<0,\qquad Y\in[27/100,103/100], d log Y d E ( V 1/3 H Y ) < 0 , Y ∈ [ 27/100 , 103/100 ] , d d log Y E ( V 1 / 3 H Y ) > 0 , Y ∈ [ 103 / 50 , 14 ] . \frac{\mathrm{d}}{\mathrm{d}\log Y}\mathcal{E}(V^{1/3}H_Y)>0,\qquad Y\in[103/50,14]. d log Y d E ( V 1/3 H Y ) > 0 , Y ∈ [ 103/50 , 14 ] . In addition E ( H 1 , 3 ) > 15 / 2 E(H_1,3)>15/2 E ( H 1 , 3 ) > 15/2 and E ( H 1 , 6 ) < 5 E(H_1,6)<5 E ( H 1 , 6 ) < 5 .
Proof. Use the finite Ewald expressions in the proof of Lemma A.8, now with the cube − 32 ≤ i , j , k ≤ 32 -32\le i,j,k\le32 − 32 ≤ i , j , k ≤ 32 . For the tetragonal Gram matrix,
Q = c D T diag ( X , 1 , 1 ) D , D = ( 1 − 1 − 1 1 1 0 0 0 1 ) , c = ( 4 X ) − 1 / 3 . Q=cD^{T}\operatorname{diag}(X,1,1)D,\qquad D=\begin{pmatrix}1&-1&-1\\1&1&0\\0&0&1\end{pmatrix},\qquad c=(4X)^{-1/3}. Q = c D T diag ( X , 1 , 1 ) D , D = 1 1 0 − 1 1 0 − 1 0 1 , c = ( 4 X ) − 1/3 . Since spec ( D T D ) = { 2 − 2 , 2 , 2 + 2 } \operatorname{spec}(D^{T}D)=\{2-\sqrt{2},2,2+\sqrt{2}\} spec ( D T D ) = { 2 − 2 , 2 , 2 + 2 } ,
Q ≥ 117 200 c min ( X , 1 ) I > 1 25 I , tr Q = 3 c ( X + 1 ) < 25 Q\ge\frac{117}{200}c\min(X,1)I>\frac{1}{25}I,\qquad\operatorname{tr}Q=3c(X+1)<25 Q ≥ 200 117 c min ( X , 1 ) I > 25 1 I , tr Q = 3 c ( X + 1 ) < 25 on 77 / 1000 ≤ X ≤ 21 77/1000\le X\le21 77/1000 ≤ X ≤ 21 . For the simple-hexagonal Gram matrix,
Q = c ( 1 1 / 2 0 1 / 2 1 0 0 0 Y ) , c = ( 4 / ( 3 Y ) ) 1 / 3 , Q=c\begin{pmatrix}1&1/2&0\\1/2&1&0\\0&0&Y\end{pmatrix},\qquad c=(4/(3Y))^{1/3}, Q = c 1 1/2 0 1/2 1 0 0 0 Y , c = ( 4/ ( 3 Y ) ) 1/3 , one similarly has Q > ( 1 / 25 ) I Q>(1/25)I Q > ( 1/25 ) I and tr Q = c ( 2 + Y ) < 25 \operatorname{tr}Q=c(2+Y)<25 tr Q = c ( 2 + Y ) < 25 on 27 / 100 ≤ Y ≤ 14 27/100\le Y\le14 27/100 ≤ Y ≤ 14 . In both cases Q − 1 > ( 1 / 25 ) I Q^{-1}>(1/25)I Q − 1 > ( 1/25 ) I ; each moment and its first two logarithmic derivatives have error at most
120 e − 3267 π / 128 ( 1 + 128 / 3 ) 2 1 − e − 201 π / 128 . 120\frac{e^{-3267\pi/128}(1+\sqrt{128/3})^{2}}{1-e^{-201\pi/128}}. 120 1 − e − 201 π /128 e − 3267 π /128 ( 1 + 128/3 ) 2 . Here an omitted index has squared norm at least 1089 1089 1089 , q , q ∗ ≥ 1089 / 25 > 20 q,q^{*}\ge1089/25>20 q , q ∗ ≥ 1089/25 > 20 , and 2 q > ( 3 π / 128 ) ∣ n ∣ 2 2q>(3\pi/128)|n|^{2} 2 q > ( 3 π /128 ) ∣ n ∣ 2 . Thus the same composed-kernel bound 10 e − 2 q 10e^{-2q} 10 e − 2 q applies to each primal and dual summand.
For each ( a , b , M ) (a,b,M) ( a , b , M ) below, use the rational cells
X ∈ [ a + ( b − a ) j M , a + ( b − a ) ( j + 1 ) M ] , j ∈ Z , 0 ≤ j < M , X\in\left[a+\frac{(b-a)j}{M},a+\frac{(b-a)(j+1)}{M}\right],\qquad j\in\mathbb{Z},\quad0\le j<M, X ∈ [ a + M ( b − a ) j , a + M ( b − a ) ( j + 1 ) ] , j ∈ Z , 0 ≤ j < M , and likewise for Y Y Y . Insert 2 V 2 ∈ [ 441 / 200 , 121 / 50 ] 2V^{2}\in[441/200,121/50] 2 V 2 ∈ [ 441/200 , 121/50 ] directly in
d r d t r [ E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) ] , \frac{\mathrm{d}^{r}}{\mathrm{d}t^{r}}\left[E(L_{e^{t}},6)-2V^{2}E(L_{e^{t}},3)\right], d t r d r [ E ( L e t , 6 ) − 2 V 2 E ( L e t , 3 ) ] , d r d t r [ E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) ] ( r = 1 , 2 ) . \frac{\mathrm{d}^{r}}{\mathrm{d}t^{r}}\left[E(H_{e^{t}},6)-2V^{2}E(H_{e^{t}},3)\right]\qquad(r=1,2). d t r d r [ E ( H e t , 6 ) − 2 V 2 E ( H e t , 3 ) ] ( r = 1 , 2 ) . The tetragonal first-derivative calculations use
( a , b , M ) = ( 77 / 1000 , 23 / 100 , 256 ) , ( 307 / 1000 , 8 / 25 , 256 ) , ( 8 / 25 , 12 / 25 , 512 ) , (a,b,M)=(77/1000,23/100,256),\quad(307/1000,8/25,256),\quad(8/25,12/25,512), ( a , b , M ) = ( 77/1000 , 23/100 , 256 ) , ( 307/1000 , 8/25 , 256 ) , ( 8/25 , 12/25 , 512 ) , ( 13 / 25 , 49 / 50 , 1024 ) , ( 51 / 50 , 6 / 5 , 512 ) , ( 6 / 5 , 3 , 512 ) , ( 3 , 21 , 1024 ) . (13/25,49/50,1024),\quad(51/50,6/5,512),\quad(6/5,3,512),\quad(3,21,1024). ( 13/25 , 49/50 , 1024 ) , ( 51/50 , 6/5 , 512 ) , ( 6/5 , 3 , 512 ) , ( 3 , 21 , 1024 ) .
In that order the resulting bounds for the normalized derivatives are
( − 824 , − 2.49 ) , ( .00450 , .28887 ) , ( .03182 , .37258 ) , ( − .21241 , − .00785 ) , (-824,-2.49),\quad(.00450,.28887),\quad(.03182,.37258),\quad(-.21241,-.00785), ( − 824 , − 2.49 ) , ( .00450 , .28887 ) , ( .03182 , .37258 ) , ( − .21241 , − .00785 ) ,
( .01080 , .27165 ) , ( .12762 , 6.67785 ) , ( 5.30962 , 789.34618 ) . (.01080,.27165),\quad(.12762,6.67785),\quad(5.30962,789.34618). ( .01080 , .27165 ) , ( .12762 , 6.67785 ) , ( 5.30962 , 789.34618 ) .
The second-derivative calculations use
( a , b , M ) = ( 12 / 25 , 13 / 25 , 64 ) , ( 49 / 50 , 51 / 50 , 64 ) , (a,b,M)=(12/25,13/25,64),\qquad(49/50,51/50,64), ( a , b , M ) = ( 12/25 , 13/25 , 64 ) , ( 49/50 , 51/50 , 64 ) ,
and give respectively ( − 1.22492 , − .87808 ) (-1.22492,-.87808) ( − 1.22492 , − .87808 ) and ( .86503 , 1.26027 ) (.86503,1.26027) ( .86503 , 1.26027 ) . The simple-hexagonal first-derivative calculations use
( a , b , M ) = ( 27 / 100 , 51 / 50 , 1024 ) , ( 51 / 50 , 103 / 100 , 128 ) , ( 103 / 50 , 14 , 1024 ) , (a,b,M)=(27/100,51/50,1024),\quad(51/50,103/100,128),\quad(103/50,14,1024), ( a , b , M ) = ( 27/100 , 51/50 , 1024 ) , ( 51/50 , 103/100 , 128 ) , ( 103/50 , 14 , 1024 ) ,
and give ( − 766.26542 , − .07665 ) (-766.26542,-.07665) ( − 766.26542 , − .07665 ) , ( − .40269 , − .03494 ) (-.40269,-.03494) ( − .40269 , − .03494 ) , and ( 7.75627 , 1178.55666 ) (7.75627,1178.55666) ( 7.75627 , 1178.55666 ) . Including complete tails and outward rounding gives these signs and
7.90860 < E ( H 1 , 3 ) < 7.90861 , 4.61394 < E ( H 1 , 6 ) < 4.61395. 7.90860 < E(H_1,3) < 7.90861,\qquad4.61394 < E(H_1,6) < 4.61395. 7.90860 < E ( H 1 , 3 ) < 7.90861 , 4.61394 < E ( H 1 , 6 ) < 4.61395.
This completes the proof of Numerical Lemma A.11. □
Numerical lemma A.12 (A finite moment-to-parameter enclosure). Let [ V 0 , V 1 ] = [ 21 / 20 , 11 / 10 ] [V_0,V_1]=[21/20,11/10] [ V 0 , V 1 ] = [ 21/20 , 11/10 ] and
f ( V ) = min { E ( F , 6 ) − 2 V 2 E ( F , 3 ) , E ( L X ( V ) , 6 ) − 2 V 2 E ( L X ( V ) , 3 ) , E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) } . f(V)=\min\left\{
\begin{array}{l}
E(F,6)-2V^2E(F,3),\\
E(L_X(V),6)-2V^2E(L_X(V),3),\\
E(H_Y(V),6)-2V^2E(H_Y(V),3)
\end{array}
\right\}. f ( V ) = min ⎩ ⎨ ⎧ E ( F , 6 ) − 2 V 2 E ( F , 3 ) , E ( L X ( V ) , 6 ) − 2 V 2 E ( L X ( V ) , 3 ) , E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) ⎭ ⎬ ⎫ . For every finite interval T T T ,
{ 2 V 2 : b ∈ T , V ∈ argmax V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 b } } ⊂ C ( T ) . \left\{2V^2:\ b\in T,\quad V\in\underset{V_0\le v\le V_1}{\operatorname{argmax}}\{f(v)+2v^2b\}\right\}\subset C(T). { 2 V 2 : b ∈ T , V ∈ V 0 ≤ v ≤ V 1 argmax { f ( v ) + 2 v 2 b } } ⊂ C ( T ) . All maximizing endpoints are retained, including outside the strict moment band.
Proof. For Λ = L X , H Y \Lambda=L_X,H_Y Λ = L X , H Y , put
A ( Λ ) = E ( Λ , 6 ) , B ( Λ ) = E ( Λ , 3 ) , Γ ( Λ ) = A t ( Λ ) B t ( Λ ) , t = log X , log Y , A(\Lambda)=E(\Lambda,6),\qquad B(\Lambda)=E(\Lambda,3),\qquad\Gamma(\Lambda)=\frac{A_t(\Lambda)}{B_t(\Lambda)},\qquad t=\log X,\log Y, A ( Λ ) = E ( Λ , 6 ) , B ( Λ ) = E ( Λ , 3 ) , Γ ( Λ ) = B t ( Λ ) A t ( Λ ) , t = log X , log Y ,
respectively. The unit-covolume Ewald formulas and their first two logarithmic derivatives use Lemma A.8 and
K s , + ( q ) = π s G s ( π q ) Γ ( s ) , K s , − ( q ) = π s G 3 / 2 − s ( π q ) Γ ( s ) , K_{s,+}(q)=\frac{\pi^sG_s(\pi q)}{\Gamma(s)},\qquad K_{s,-}(q)=\frac{\pi^sG_{3/2-s}(\pi q)}{\Gamma(s)}, K s , + ( q ) = Γ ( s ) π s G s ( π q ) , K s , − ( q ) = Γ ( s ) π s G 3/2 − s ( π q ) ,
( K s , + ( r ) ( q ) , K s , − ( r ) ( q ) ) = ( − π ) r π s Γ ( s ) ( G s + r ( π q ) , G 3 / 2 − s + r ( π q ) ) , r = 0 , 1 , 2 , \left(K_{s,+}^{(r)}(q),K_{s,-}^{(r)}(q)\right)
=\frac{(-\pi)^r\pi^s}{\Gamma(s)}
\left(G_{s+r}(\pi q),G_{3/2-s+r}(\pi q)\right),\qquad r=0,1,2, ( K s , + ( r ) ( q ) , K s , − ( r ) ( q ) ) = Γ ( s ) ( − π ) r π s ( G s + r ( π q ) , G 3/2 − s + r ( π q ) ) , r = 0 , 1 , 2 ,
with constant terms π 3 / 6 \pi^3/6 π 3 /6 , π 6 / 2160 \pi^6/2160 π 6 /2160 . For ( i , j , k ) ∈ ( [ − 12 , 12 ] 3 ∩ Z 3 ) ∖ { 0 } (i,j,k)\in([-12,12]^3\cap\mathbb{Z}^3)\setminus\{0\} ( i , j , k ) ∈ ([ − 12 , 12 ] 3 ∩ Z 3 ) ∖ { 0 } , use
( a , b ) = ( ( i − j − k ) 2 , ( i + j ) 2 + k 2 ) ( L X , p r i m a l ) , (a,b)=\left((i-j-k)^2,(i+j)^2+k^2\right)\qquad(L_X,\ \mathrm{primal}), ( a , b ) = ( ( i − j − k ) 2 , ( i + j ) 2 + k 2 ) ( L X , primal ) ,
( a , b ) = ( ( i − j ) 2 , ( i + j ) 2 + ( i − j + 2 k ) 2 ) ( L X , d u a l ) , (a,b)=\left((i-j)^2,(i+j)^2+(i-j+2k)^2\right)\qquad(L_X,\ \mathrm{dual}), ( a , b ) = ( ( i − j ) 2 , ( i + j ) 2 + ( i − j + 2 k ) 2 ) ( L X , dual ) ,
( a , b ) = ( i 2 + i j + j 2 , k 2 ) ( H Y , p r i m a l / d u a l ) . (a,b)=\left(i^2+ij+j^2,k^2\right)\qquad(H_Y,\ \mathrm{primal/dual}). ( a , b ) = ( i 2 + ij + j 2 , k 2 ) ( H Y , primal/dual ) .
Group equal pairs with their full positive integer multiplicities; sum lexicographically, primal before dual. With c = ( 4 X ) − 1 / 3 c=(4X)^{-1/3} c = ( 4 X ) − 1/3 for L X L_X L X and c = ( 4 / ( 3 Y ) ) 1 / 3 c=(4/(3Y))^{1/3} c = ( 4/ ( 3 Y ) ) 1/3 for H Y H_Y H Y ,
( q 1 , q 2 , e 1 , e 2 ) = ( c X a , c b , 2 / 3 , − 1 / 3 ) ( L X , p r i m a l ) , (q_1,q_2,e_1,e_2)=(cX^a,cb,2/3,-1/3)\qquad(L_X,\ \mathrm{primal}), ( q 1 , q 2 , e 1 , e 2 ) = ( c X a , c b , 2/3 , − 1/3 ) ( L X , primal ) ,
( q 1 , q 2 , e 1 , e 2 ) = ( a / ( 4 c X ) , b / ( 4 c ) , − 2 / 3 , 1 / 3 ) ( L X , d u a l ) , (q_1,q_2,e_1,e_2)=(a/(4cX),b/(4c),-2/3,1/3)\qquad(L_X,\ \mathrm{dual}), ( q 1 , q 2 , e 1 , e 2 ) = ( a / ( 4 c X ) , b / ( 4 c ) , − 2/3 , 1/3 ) ( L X , dual ) ,
( q 1 , q 2 , e 1 , e 2 ) = ( c a , c Y b , − 1 / 3 , 2 / 3 ) ( H Y , p r i m a l ) , (q_1,q_2,e_1,e_2)=(ca,cY^b,-1/3,2/3)\qquad(H_Y,\ \mathrm{primal}), ( q 1 , q 2 , e 1 , e 2 ) = ( c a , c Y b , − 1/3 , 2/3 ) ( H Y , primal ) ,
( q 1 , q 2 , e 1 , e 2 ) = ( 4 a / ( 3 c ) , b / ( c Y ) , 1 / 3 , − 2 / 3 ) ( H Y , d u a l ) , (q_1,q_2,e_1,e_2)=(4a/(3c),b/(cY),1/3,-2/3)\qquad(H_Y,\ \mathrm{dual}), ( q 1 , q 2 , e 1 , e 2 ) = ( 4 a / ( 3 c ) , b / ( c Y ) , 1/3 , − 2/3 ) ( H Y , dual ) ,
q = q 1 + q 2 , q t = e 1 q 1 + e 2 q 2 , q t t = e 1 2 q 1 + e 2 2 q 2 , q=q_1+q_2,\qquad q_t=e_1q_1+e_2q_2,\qquad q_{tt}=e_1^2q_1+e_2^2q_2, q = q 1 + q 2 , q t = e 1 q 1 + e 2 q 2 , q tt = e 1 2 q 1 + e 2 2 q 2 ,
K t = K ′ q t , K t t = K ′ ′ q t 2 + K ′ q t t . K_t=K' q_t,\qquad K_{tt}=K''q_t^2+K'q_{tt}. K t = K ′ q t , K tt = K ′′ q t 2 + K ′ q tt .
Extend cube-root endpoints until their cubes enclose the input. Use the complete outward bounds of Lemmas 2.8 and A.2 ; on every closed cell,
det Q = 1 , Q > 1 5 I , tr Q < 5 ⟹ Q − 1 > 1 5 I ⟹ q , q ∗ ≥ 169 5 > 20 \det Q = 1,\qquad Q > \frac{1}{5}I,\qquad\operatorname{tr} Q < 5 \mathrel{\Longrightarrow} Q^{-1} > \frac{1}{5}I \mathrel{\Longrightarrow} q,q^{*} \ge\frac{169}{5} > 20 det Q = 1 , Q > 5 1 I , tr Q < 5 ⟹ Q − 1 > 5 1 I ⟹ q , q ∗ ≥ 5 169 > 20
for omitted indices. Numerical Lemma A.8 gives τ < 2648 / 10 29 \tau< 2648/10^{29} τ < 2648/1 0 29 . Enlarge the finite expressions, for s = 3 , 6 s=3,6 s = 3 , 6 , by
[ E ( Λ , s ) ] ← [ E ( Λ , s ) ] + [ 0 , τ ] , [ ∂ t r E ( Λ , s ) ] ← [ ∂ t r E ( Λ , s ) ] + [ − τ , τ ] ( r = 1 , 2 ) . [E(\Lambda,s)] \mathrel{\leftarrow} [E(\Lambda,s)] + [0,\tau],\qquad[\partial_t^r E(\Lambda,s)] \mathrel{\leftarrow} [\partial_t^r E(\Lambda,s)] + [-\tau,\tau]\qquad(r=1,2). [ E ( Λ , s )] ← [ E ( Λ , s )] + [ 0 , τ ] , [ ∂ t r E ( Λ , s )] ← [ ∂ t r E ( Λ , s )] + [ − τ , τ ] ( r = 1 , 2 ) .
Evaluate the complete intervals B j , A j , Γ j = A t , j / B t , j B_j,A_j,\Gamma_j=A_{t,j}/B_{t,j} B j , A j , Γ j = A t , j / B t , j at the rational nodes and on every intervening closed cell:
X j = 303104 + 7 j 1024000 ( 0 ≤ j ≤ 1024 ) , Y j = 421888 + 9 j 409600 ( 0 ≤ j ≤ 4096 ) . X_j=\frac{303104+7j}{1024000}\qquad(0\le j\le1024),\qquad Y_j=\frac{421888+9j}{409600}\qquad(0\le j\le4096). X j = 1024000 303104 + 7 j ( 0 ≤ j ≤ 1024 ) , Y j = 409600 421888 + 9 j ( 0 ≤ j ≤ 4096 ) .
The outward cell bounds are
B t ( L X ) ∈ [ − 1033173 / 10 6 , − 927064 / 10 6 ] , B_t(L_X)\in[-1033173/10^6,-927064/10^6], B t ( L X ) ∈ [ − 1033173/1 0 6 , − 927064/1 0 6 ] ,
Γ t ( L X ) ∈ [ − 4031155 / 10 6 , − 3984355 / 10 6 ] , \Gamma_t(L_X)\in[-4031155/10^6,-3984355/10^6], Γ t ( L X ) ∈ [ − 4031155/1 0 6 , − 3984355/1 0 6 ] ,
B t ( H Y ) ∈ [ 1408367 / 10 6 , 2245006 / 10 6 ] , B_t(H_Y)\in[1408367/10^6,2245006/10^6], B t ( H Y ) ∈ [ 1408367/1 0 6 , 2245006/1 0 6 ] ,
Γ t ( H Y ) ∈ [ 1912446 / 10 6 , 5777144 / 10 6 ] , Γ t = A t t − Γ B t t B t . \Gamma_t(H_Y)\in[1912446/10^6,5777144/10^6],\qquad\Gamma_t=\frac{A_{tt}-\Gamma B_{tt}}{B_t}. Γ t ( H Y ) ∈ [ 1912446/1 0 6 , 5777144/1 0 6 ] , Γ t = B t A tt − Γ B tt .
The X X X intervals decrease; reverse that list. Both lists then satisfy
D j + < D j + 1 − ( D = A , B , Γ ) . D_j^{+}<D_{j+1}^{-}\qquad(D=A,B,\Gamma). D j + < D j + 1 − ( D = A , B , Γ ) .
The consecutive upper endpoint-to-endpoint Γ \Gamma Γ differences are < 9229 / 10 8 <9229/10^8 < 9229/1 0 8 for L X L_X L X and < 123 / 10 6 <123/10^6 < 123/1 0 6 for H Y H_Y H Y , on whole cells. Lemma A.9 gives
2 V F T 2 ∈ I F T = 2 [ ( 10648235667 / 10 10 ) 2 , ( 10648235669 / 10 10 ) 2 ] , 2V_{\mathrm{FT}}^2\in I_{\mathrm{FT}}=2\left[(10648235667/10^{10})^2,(10648235669/10^{10})^2\right], 2 V FT 2 ∈ I FT = 2 [ ( 10648235667/1 0 10 ) 2 , ( 10648235669/1 0 10 ) 2 ] ,
2 V T H 2 ∈ I T H = 2 [ ( 10759906200 / 10 10 ) 2 , ( 10759906201 / 10 10 ) 2 ] . 2V_{\mathrm{TH}}^2\in I_{\mathrm{TH}}=2\left[(10759906200/10^{10})^2,(10759906201/10^{10})^2\right]. 2 V TH 2 ∈ I TH = 2 [ ( 10759906200/1 0 10 ) 2 , ( 10759906201/1 0 10 ) 2 ] .
The roots of A t − 2 V 2 B t = 0 A_t-2V^2B_t=0 A t − 2 V 2 B t = 0 satisfy
X F T ∈ [ 3012583413613 , 3012583613613 ] / 10 13 ( 2 V 2 ∈ I F T ) , X_{\mathrm{FT}}\in[3012583413613,3012583613613]/10^{13}\qquad(2V^2\in I_{\mathrm{FT}}), X FT ∈ [ 3012583413613 , 3012583613613 ] /1 0 13 ( 2 V 2 ∈ I FT ) ,
X T H ∈ [ 2976831441610 , 2976831641610 ] / 10 13 ( 2 V 2 ∈ I T H ) , X_{\mathrm{TH}}\in[2976831441610,2976831641610]/10^{13}\qquad(2V^2\in I_{\mathrm{TH}}), X TH ∈ [ 2976831441610 , 2976831641610 ] /1 0 13 ( 2 V 2 ∈ I TH ) ,
Y T H ∈ [ 10617382252094 , 10617382452094 ] / 10 13 ( 2 V 2 ∈ I T H ) , Y_{\mathrm{TH}}\in[10617382252094,10617382452094]/10^{13}\qquad(2V^2\in I_{\mathrm{TH}}), Y TH ∈ [ 10617382252094 , 10617382452094 ] /1 0 13 ( 2 V 2 ∈ I TH ) ,
Y 1 ∈ [ 11013103163 , 11013103363 ] / 10 10 ( V = 11 / 10 ) . Y_1\in[11013103163,11013103363]/10^{10}\qquad(V=11/10). Y 1 ∈ [ 11013103163 , 11013103363 ] /1 0 10 ( V = 11/10 ) .
For each displayed shape bracket [ z − , z + ] [z_-,z_+] [ z − , z + ] and every V V V in its corresponding bracket, the complete intervals satisfy
sup [ Γ ( z + ) ] < 2 V 2 < inf [ Γ ( z − ) ] ( L X ) , \sup[\Gamma(z_+)]<2V^2<\inf[\Gamma(z_-)]\qquad(L_X), sup [ Γ ( z + )] < 2 V 2 < inf [ Γ ( z − )] ( L X ) ,
sup [ Γ ( z − ) ] < 2 V 2 < inf [ Γ ( z + ) ] ( H Y ) . \sup[\Gamma(z_-)]<2V^2<\inf[\Gamma(z_+)]\qquad(H_Y). sup [ Γ ( z − )] < 2 V 2 < inf [ Γ ( z + )] ( H Y ) .
Closed-cell monotonicity encloses each root. The corresponding complete B B B intervals are B F , B T , F , B T , H , B H , H , B H , 1 B_F,B_{T,F},B_{T,H},B_{H,H},B_{H,1} B F , B T , F , B T , H , B H , H , B H , 1 , with B F = [ E ( F , 3 ) ] B_F=[E(F,3)] B F = [ E ( F , 3 )] . Their outward rational summaries are
E ( F , 3 ) ∈ [ 7.22696052186863 , 7.22696052187584 ] , E(F,3)\in[7.22696052186863,7.22696052187584], E ( F , 3 ) ∈ [ 7.22696052186863 , 7.22696052187584 ] ,
E ( L X ( V F T ) , 3 ) ∈ [ 7.41125402927957 , 7.41125409252949 ] , E(L_X(V_{\mathrm{FT}}),3)\in[7.41125402927957,7.41125409252949], E ( L X ( V FT ) , 3 ) ∈ [ 7.41125402927957 , 7.41125409252949 ] ,
E ( L X ( V T H ) , 3 ) ∈ [ 7.42294739425582 , 7.42294746189465 ] , E(L_X(V_{\mathrm{TH}}),3)\in[7.42294739425582,7.42294746189465], E ( L X ( V TH ) , 3 ) ∈ [ 7.42294739425582 , 7.42294746189465 ] ,
E ( H Y ( V T H ) , 3 ) ∈ [ 7.99317817792011 , 7.99317821023425 ] , E(H_Y(V_{\mathrm{TH}}),3)\in[7.99317817792011,7.99317821023425], E ( H Y ( V TH ) , 3 ) ∈ [ 7.99317817792011 , 7.99317821023425 ] ,
E ( H Y ( V 1 ) , 3 ) ∈ [ 8.06262195018044 , 8.06262198794635 ] , E(H_Y(V_1),3)\in[8.06262195018044,8.06262198794635], E ( H Y ( V 1 ) , 3 ) ∈ [ 8.06262195018044 , 8.06262198794635 ] ,
E ( H Y ( V 1 ) , 6 ) ∈ ( 4.96288451736152 , 4.96288460874427 ) ⊂ ( − ∞ , 5 ) . E(H_Y(V_1),6)\in(4.96288451736152,4.96288460874427)\subset(-\infty,5). E ( H Y ( V 1 ) , 6 ) ∈ ( 4.96288451736152 , 4.96288460874427 ) ⊂ ( − ∞ , 5 ) .
Use the complete finite-expression intervals within these strictly ordered summaries. Each shape bracket is interior to its node domain; each full node 2 V 2 2V^2 2 V 2 range strictly contains its active range.
For either increasing list, with final index m m m , put
l = max { inf T , B 0 − } , h = min { sup T , B m + } , l=\max\{\inf T,B_0^-\},\qquad h=\min\{\sup T,B_m^+\}, l = max { inf T , B 0 − } , h = min { sup T , B m + } ,
j = min { ν ∈ { 0 , … , m } : B ν + ≥ l } , j=\min\{\nu\in\{0,\ldots,m\}:B_\nu^+\ge l\}, j = min { ν ∈ { 0 , … , m } : B ν + ≥ l } ,
k = min { ν ∈ { 0 , … , m } : B ν − > h } ∪ { m } ( l ≤ h ) , k=\min\{\nu\in\{0,\ldots,m\}:B_\nu^->h\}\cup\{m\}\qquad(l\le h), k = min { ν ∈ { 0 , … , m } : B ν − > h } ∪ { m } ( l ≤ h ) ,
I ( T ) = { ∅ , l > h , [ B max ( 0 , j − 1 ) − , B k + ] ∩ I , l ≤ h , \mathcal{I}(T)=
\begin{cases}
\varnothing, & l>h,\\
[B_{\max(0,j-1)}^-,B_k^+]\cap I, & l\le h,
\end{cases} I ( T ) = { ∅ , [ B m a x ( 0 , j − 1 ) − , B k + ] ∩ I , l > h , l ≤ h ,
( I , I ) = { ( [ I F T − , I T H + ] , I L ) , L X , ( [ I T H − , 2 V 1 2 ] , I H ) , H Y . (I,\mathcal{I})=
\begin{cases}
([I_{F T}^-,I_{T H}^+],I_L), & L_X,\\
([I_{T H}^-,2V_1^2],I_H), & H_Y.
\end{cases} ( I , I ) = { ([ I F T − , I T H + ] , I L ) , ([ I T H − , 2 V 1 2 ] , I H ) , L X , H Y .
For every shape in the full tabulated interval,
B ( Λ ) ∈ T , Γ ( Λ ) ∈ I ⟹ Γ ( Λ ) ∈ I ( T ) . B(\Lambda)\in T,\quad\Gamma(\Lambda)\in I\quad\Longrightarrow\quad\Gamma(\Lambda)\in\mathcal{I}(T). B ( Λ ) ∈ T , Γ ( Λ ) ∈ I ⟹ Γ ( Λ ) ∈ I ( T ) .
Take C ( T ) C(T) C ( T ) to be the hull of the applicable contributions:
{ 2 V 0 2 } ⊂ C ( T ) ( inf T ≤ B F + ) , \{2V_0^2\}\subset C(T)\qquad(\inf T\le B_F^+), { 2 V 0 2 } ⊂ C ( T ) ( inf T ≤ B F + ) ,
{ 2 V 1 2 } ⊂ C ( T ) ( sup T ≥ B H , 1 − ) , \{2V_1^2\}\subset C(T)\qquad(\sup T\ge B_{H,1}^-), { 2 V 1 2 } ⊂ C ( T ) ( sup T ≥ B H , 1 − ) ,
[ 2 V 0 2 , I F T + ] ⊂ C ( T ) ( T ∩ B F ≠ ∅ ) , [2V_0^2,I_{F T}^+]\subset C(T)\qquad(T\cap B_F\ne\varnothing), [ 2 V 0 2 , I F T + ] ⊂ C ( T ) ( T ∩ B F = ∅ ) ,
I F T ⊂ C ( T ) ( T ∩ [ B T , F − , B T , F + ] ≠ ∅ ) , I_{F T}\subset C(T)\qquad(T\cap[B_{T,F}^-,B_{T,F}^+]\ne\varnothing), I F T ⊂ C ( T ) ( T ∩ [ B T , F − , B T , F + ] = ∅ ) ,
I L ( T ) ⊂ C ( T ) ( T ∩ [ B T , F − , B T , H + ] ≠ ∅ ) , I_L(T)\subset C(T)\qquad(T\cap[B_{T,F}^-,B_{T,H}^+]\ne\varnothing), I L ( T ) ⊂ C ( T ) ( T ∩ [ B T , F − , B T , H + ] = ∅ ) ,
I T H ⊂ C ( T ) ( T ∩ [ B T , H − , B H , H + ] ≠ ∅ ) , I_{T H}\subset C(T)\qquad(T\cap[B_{T,H}^-,B_{H,H}^+]\ne\varnothing), I T H ⊂ C ( T ) ( T ∩ [ B T , H − , B H , H + ] = ∅ ) ,
I H ( T ) ⊂ C ( T ) ( T ∩ [ B H , H − , B H , 1 + ] ≠ ∅ ) . I_H(T)\subset C(T)\qquad(T\cap[B_{H,H}^-,B_{H,1}^+]\ne\varnothing). I H ( T ) ⊂ C ( T ) ( T ∩ [ B H , H − , B H , 1 + ] = ∅ ) .
Concavity in V 2 V^2 V 2 gives all affine/contact/branch/endpoint cases. An inconclusive test retains
C ( T ) = [ 2 V 0 2 , 2 V 1 2 ] = [ 441 / 200 , 121 / 50 ] . C(T)=[2V_0^2,2V_1^2]=[441/200,121/50]. C ( T ) = [ 2 V 0 2 , 2 V 1 2 ] = [ 441/200 , 121/50 ] .
This completes the proof of Numerical Lemma A.12. □
Numerical lemma A.13 (The finite transition-profile comparison). Let [ V 0 , V 1 ] = [ 21 / 20 , 11 / 10 ] [V_0,V_1]=[21/20,11/10] [ V 0 , V 1 ] = [ 21/20 , 11/10 ] and
Φ V ( Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) , \Phi_V(\Lambda)=E(\Lambda,6)-2V^2E(\Lambda,3), Φ V ( Λ ) = E ( Λ , 6 ) − 2 V 2 E ( Λ , 3 ) , f ( V ) = min { E ( F , 6 ) − 2 V 2 E ( F , 3 ) , E ( L X ( V ) , 6 ) − 2 V 2 E ( L X ( V ) , 3 ) , E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) } , f(V)=\min\left\{
\begin{aligned}
&E(F,6)-2V^2E(F,3),\\
&E(L_X(V),6)-2V^2E(L_X(V),3),\\
&E(H_Y(V),6)-2V^2E(H_Y(V),3)
\end{aligned}
\right\}, f ( V ) = min ⎩ ⎨ ⎧ E ( F , 6 ) − 2 V 2 E ( F , 3 ) , E ( L X ( V ) , 6 ) − 2 V 2 E ( L X ( V ) , 3 ) , E ( H Y ( V ) , 6 ) − 2 V 2 E ( H Y ( V ) , 3 ) ⎭ ⎬ ⎫ , g ( t ) = max V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 t } , J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 ) ) . g(t)=\max_{V_0\le v\le V_1}\{f(v)+2v^2t\},\qquad J(\Lambda)=E(\Lambda,6)-g(E(\Lambda,3)). g ( t ) = V 0 ≤ v ≤ V 1 max { f ( v ) + 2 v 2 t } , J ( Λ ) = E ( Λ , 6 ) − g ( E ( Λ , 3 )) . Then
J ( Λ ) ≥ 0 ( ∣ Λ ∣ = 1 ) , J(\Lambda)\ge0\qquad(|\Lambda|=1), J ( Λ ) ≥ 0 ( ∣Λ∣ = 1 ) , ∣ Λ ∣ = 1 , V ∈ ( V 0 , V 1 ) ∩ argmax V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ , 3 ) } |\Lambda|=1,\quad V\in(V_0,V_1)\cap\operatorname*{argmax}_{V_0\le v\le V_1}\{f(v)+2v^2E(\Lambda,3)\} ∣Λ∣ = 1 , V ∈ ( V 0 , V 1 ) ∩ V 0 ≤ v ≤ V 1 argmax { f ( v ) + 2 v 2 E ( Λ , 3 )} ⟹ J ( Λ ) = 0 ⟺ ∃ Ξ ∈ { F , L X ( V ) , H Y ( V ) } : Λ ≅ Ξ , Φ V ( Ξ ) = f ( V ) . \Longrightarrow\quad J(\Lambda)=0\Longleftrightarrow\exists\Xi\in\{F,L_X(V),H_Y(V)\}:\ \Lambda\cong\Xi,\ \Phi_V(\Xi)=f(V). ⟹ J ( Λ ) = 0 ⟺ ∃Ξ ∈ { F , L X ( V ) , H Y ( V )} : Λ ≅ Ξ , Φ V ( Ξ ) = f ( V ) . The three norm conditions (T13) are those of Lemmas 5.4, 5.5 and 5.6 .
Proof. Use the five-coordinate bounds of Lemmas A.2 and 5.2 ; 2 V 2 2V^2 2 V 2 is contracted, never bisected.
Admissible minima. In (21) –(22) , put p = ( a , δ , x , y , z ) p=(a,\delta,x,y,z) p = ( a , δ , x , y , z ) , a , δ > 0 a,\delta>0 a , δ > 0 , h = ( a δ ) − 1 h=(a\delta)^{-1} h = ( a δ ) − 1 ; all derivatives below are shape derivatives of Φ V ( U ( p ) Z 3 ) \Phi_V(U(p)Z^3) Φ V ( U ( p ) Z 3 ) . Proposition 5.1 gives
inf ∣ Λ ∣ = 1 J ( Λ ) < 0 ⟹ { Λ ∗ ∈ arg min ∣ Λ ∣ = 1 J : J ( Λ ∗ ) < 0 } ≠ ∅ , \inf_{|\Lambda|=1}J(\Lambda)<0\quad\Longrightarrow\quad\left\{\Lambda_* \in\underset{|\Lambda|=1}{\arg\min}\,J:J(\Lambda_*)<0\right\}\ne\varnothing, ∣Λ∣ = 1 inf J ( Λ ) < 0 ⟹ { Λ ∗ ∈ ∣Λ∣ = 1 arg min J : J ( Λ ∗ ) < 0 } = ∅ ,
Λ ∗ ∈ arg min ∣ Λ ∣ = 1 J , J ( Λ ∗ ) ≤ 0 , V ∈ arg max V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ ∗ , 3 ) } \Lambda_* \in\underset{|\Lambda|=1}{\arg\min}\,J,\quad J(\Lambda_*)\le0,\quad V\in\underset{V_0\le v\le V_1}{\arg\max}\left\{f(v)+2v^2E(\Lambda_*,3)\right\} Λ ∗ ∈ ∣Λ∣ = 1 arg min J , J ( Λ ∗ ) ≤ 0 , V ∈ V 0 ≤ v ≤ V 1 arg max { f ( v ) + 2 v 2 E ( Λ ∗ , 3 ) }
⟹ Λ ∗ ∈ arg min ∣ Λ ∣ = 1 Φ V , ∇ p Φ V ( Λ ∗ ) = 0 , ∇ p 2 Φ V ( Λ ∗ ) ⪰ 0. \Longrightarrow\quad\Lambda_*\in\underset{|\Lambda|=1}{\arg\min}\,\Phi_V,\quad\nabla_p\Phi_V(\Lambda_*)=0,\quad\nabla_p^2\Phi_V(\Lambda_*)\succeq0. ⟹ Λ ∗ ∈ ∣Λ∣ = 1 arg min Φ V , ∇ p Φ V ( Λ ∗ ) = 0 , ∇ p 2 Φ V ( Λ ∗ ) ⪰ 0.
No differentiability at a phase contact is used; after J ≥ 0 J\ge0 J ≥ 0 ,
J ( Λ ∗ ) = 0 ⟹ Λ ∗ ∈ arg min J . J(\Lambda_*)=0\Longrightarrow\Lambda_*\in\arg\min J. J ( Λ ∗ ) = 0 ⟹ Λ ∗ ∈ arg min J .
Lemma 5.3 gives
E ( U ( p ) Z 3 , 3 ) < 81 / 10 , E ( U ( p ) Z 3 , 6 ) < 5 , E(U(p)Z^3,3)<81/10,\qquad E(U(p)Z^3,6)<5, E ( U ( p ) Z 3 , 3 ) < 81/10 , E ( U ( p ) Z 3 , 6 ) < 5 ,
and the closed Korkine–Zolotarev root rectangle
463 500 ≤ a ≤ 1122463 10 6 , 801 1000 ≤ δ ≤ 559 500 , 0 ≤ x , z ≤ 1 2 , − 1 2 ≤ y ≤ 1 2 . (83) \frac{463}{500}\le a\le\frac{1122463}{10^6},\qquad\frac{801}{1000}\le\delta\le\frac{559}{500},\qquad0\le x,z\le\frac{1}{2},\qquad-\frac{1}{2}\le y\le\frac{1}{2}. \tag*{(83)} 500 463 ≤ a ≤ 1 0 6 1122463 , 1000 801 ≤ δ ≤ 500 559 , 0 ≤ x , z ≤ 2 1 , − 2 1 ≤ y ≤ 2 1 . (83)
Enlarge strict bounds to closed intervals. Complete N = 6 N=6 N = 6 unit-covolume Ewald bounds and homogeneity give
2 V 2 E ( H 11 / 10 , 3 ) − 2 E ( H 11 / 10 , 6 ) > 0 , 2V^2E(H_{11/10},3)-2E(H_{11/10},6)>0, 2 V 2 E ( H 11/10 , 3 ) − 2 E ( H 11/10 , 6 ) > 0 ,
− 4968464328760 10 12 < E ( ( 11 / 10 ) 1 / 3 H 11 / 10 ) < − 4968464328755 10 12 < − 124 25 . (84) -\frac{4968464328760}{10^{12}}<\mathcal{E}\left((11/10)^{1/3}H_{11/10}\right)<-\frac{4968464328755}{10^{12}}<-\frac{124}{25}. \tag*{(84)} − 1 0 12 4968464328760 < E ( ( 11/10 ) 1/3 H 11/10 ) < − 1 0 12 4968464328755 < − 25 124 . (84)
For every eligible minimizer and supporting V V V ,
E ( Λ , 6 ) 2 V 4 − E ( Λ , 3 ) V 2 ≤ E ( H 11 / 10 , 6 ) 2 V 4 − E ( H 11 / 10 , 3 ) V 2 ≤ E ( ( 11 / 10 ) 1 / 3 H 11 / 10 ) < − 124 25 . \begin{aligned}
\frac{E(\Lambda,6)}{2V^4}-\frac{E(\Lambda,3)}{V^2}\le\frac{E(H_{11/10},6)}{2V^4}-\frac{E(H_{11/10},3)}{V^2} \\
&\le\mathcal{E}((11/10)^{1/3}H_{11/10})<-\frac{124}{25}.
\end{aligned} 2 V 4 E ( Λ , 6 ) − V 2 E ( Λ , 3 ) ≤ 2 V 4 E ( H 11/10 , 6 ) − V 2 E ( H 11/10 , 3 ) ≤ E (( 11/10 ) 1/3 H 11/10 ) < − 25 124 .
Proposition 4.5 gives
7 8 ≤ h ≤ 32 25 , 25 32 ≤ a δ ≤ 8 7 , δ 2 + a 2 x 2 ≥ a 2 , h 2 + δ 2 z 2 ≥ δ 2 . (85) \frac{7}{8}\le h\le\frac{32}{25},\qquad\frac{25}{32}\le a\delta\le\frac{8}{7},\qquad\delta^2+a^2x^2\ge a^2,\qquad h^2+\delta^2z^2\ge\delta^2. \tag*{(85)} 8 7 ≤ h ≤ 25 32 , 32 25 ≤ a δ ≤ 7 8 , δ 2 + a 2 x 2 ≥ a 2 , h 2 + δ 2 z 2 ≥ δ 2 . (85)
q p ( i , j , k ) ≥ a 2 , q p ∗ ( i , j , k ) ≥ 1521 2500 ( ( i , j , k ) ∈ { − 1 , 0 , 1 } 3 ∖ { 0 } ) . (86) q_p(i,j,k)\ge a^2,\qquad q_p^*(i,j,k)\ge\frac{1521}{2500}\quad\left((i,j,k)\in\{-1,0,1\}^3\setminus\{0\}\right). \tag*{(86)} q p ( i , j , k ) ≥ a 2 , q p ∗ ( i , j , k ) ≥ 2500 1521 ( ( i , j , k ) ∈ { − 1 , 0 , 1 } 3 ∖ { 0 } ) . (86)
Reject certified violations of (T1)–(T4); (T4) does not replace full-box spectral floors in Ewald tails.
Choose a Korkine–Zolotarev representative p ( Λ ) p(\Lambda) p ( Λ ) satisfying (T1)–(T4) for each eligible minimum, and put
K = { ( p ( Λ ) , 2 V 2 ) : Λ ∈ arg min ∣ L ∣ = 1 J ( L ) , J ( Λ ) ≤ 0 , \mathcal{K}=\left\{(p(\Lambda),2V^2):\Lambda\in\underset{|L|=1}{\arg\min}\,J(L),\ J(\Lambda)\le0,\right. K = { ( p ( Λ ) , 2 V 2 ) : Λ ∈ ∣ L ∣ = 1 arg min J ( L ) , J ( Λ ) ≤ 0 ,
E ( Λ , 3 ) ∈ [ E ( F , 3 ) , E ( H Y ( V 1 ) , 3 ) ] , V ∈ arg max V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ , 3 ) } } . \left.E(\Lambda,3)\in[E(F,3),E(H_Y(V_1),3)],\quad V\in\underset{V_0\le v\le V_1}{\arg\max}\{f(v)+2v^2E(\Lambda,3)\}\right\}. E ( Λ , 3 ) ∈ [ E ( F , 3 ) , E ( H Y ( V 1 ) , 3 )] , V ∈ V 0 ≤ v ≤ V 1 arg max { f ( v ) + 2 v 2 E ( Λ , 3 )} } .
Complete moment bounds. Use outward binary64 arithmetic, checked radicals, and complete kernel/Taylor bounds of Lemma A.2 ; enclose all rationals and grids, retaining every inconclusive box. The constant terms are π 3 / 6 \pi^3/6 π 3 /6 , π 6 / 2160 \pi^6/2160 π 6 /2160 , and the primal/dual argument derivatives are
( − π ) r π s Γ ( s ) ( G s + r ( π q ) , G 3 / 2 − s + r ( π q ∗ ) ) , s = 3 , 6 , r ∈ Z ≥ 0 . \frac{(-\pi)^r\pi^s}{\Gamma(s)}\left(\mathcal{G}_{s+r}(\pi q),\mathcal{G}_{3/2-s+r}(\pi q^*)\right),\qquad s=3,6,\quad r\in\mathbb{Z}_{\ge0}. Γ ( s ) ( − π ) r π s ( G s + r ( π q ) , G 3/2 − s + r ( π q ∗ ) ) , s = 3 , 6 , r ∈ Z ≥ 0 .
For the complete-cube E f i n E^{\mathrm{fin}} E fin , including its constant, assume
Q ( p ) ⪰ α I , Q ( p ) − 1 ⪰ β I ( p ∈ B ) . Q(p)\succeq\alpha I,\qquad Q(p)^{-1}\succeq\beta I\qquad(p\in B). Q ( p ) ⪰ α I , Q ( p ) − 1 ⪰ β I ( p ∈ B ) .
Then
0 ≤ E ( U ( p ) Z 3 , s ) − E f i n ( U ( p ) Z 3 , s ) ≤ π s Γ ( s ) { A s , 0 ( α ) + Θ N ( β ) π β ( N + 1 ) 2 } . (87) 0 \le E(U(p)Z^3,s)-E^{\mathrm{fin}}(U(p)Z^3,s)\le\frac{\pi^s}{\Gamma(s)}\left\{A_{s,0}(\alpha)+\frac{\Theta_N(\beta)}{\pi^\beta(N+1)^2}\right\}. \tag*{(87)} 0 ≤ E ( U ( p ) Z 3 , s ) − E fin ( U ( p ) Z 3 , s ) ≤ Γ ( s ) π s { A s , 0 ( α ) + π β ( N + 1 ) 2 Θ N ( β ) } . (87)
and, with ρ i \rho_i ρ i from Lemma A.2 ,
∣ ∂ i ( E − E f i n ) ∣ ≤ ρ i π s Γ ( s ) { A s , 1 ( α ) + Θ N ( β ) } . (88) \left|\partial_i(E-E^{\mathrm{fin}})\right|\le\rho_i\frac{\pi^s}{\Gamma(s)}\left\{A_{s,1}(\alpha)+\Theta_N(\beta)\right\}. \tag*{(88)} ∂ i ( E − E fin ) ≤ ρ i Γ ( s ) π s { A s , 1 ( α ) + Θ N ( β ) } . (88)
Bound each positive moment before subtraction; opposite indices may be paired.
For coordinate midpoint c c c , radii r i r_i r i and metric radius R R R , use the larger of the native full-box spectral floors and
e − 2 / 3 R ∥ Q c − 1 ∥ ∞ − 1 , e − 2 / 3 R ∥ Q c ∥ ∞ − 1 , e^{-\sqrt{2/3}R}\lVert Q_c^{-1}\rVert_\infty^{-1},\qquad e^{-\sqrt{2/3}R}\lVert Q_{c}\rVert_\infty^{-1}, e − 2/3 R ∥ Q c − 1 ∥ ∞ − 1 , e − 2/3 R ∥ Q c ∥ ∞ − 1 ,
rounding all norm upper endpoints upwards. At R ≤ 9 / 20 R\le9/20 R ≤ 9/20 , compute the complete center moments [ M s ] [M_s] [ M s ] and whole-box gradients [ G s , i ] [G_{s,i}] [ G s , i ] at N = 3 N=3 N = 3 , including (T5)–(T6), and set
[ M s ] = ( [ M s ] + [ − ∑ i = 1 5 r i max ∣ [ G s , i ] ∣ , ∑ i = 1 5 r i max ∣ [ G s , i ] ∣ ] ) ∩ [ ( e − 2 / 3 R ) s M s ‾ , ( e − 2 / 3 R ) − s M s ‾ ] ∩ [ 0 , ∞ ) . (89) [M_s]=\left([M_s]+\left[-\sum_{i=1}^{5}r_i\max\left|[G_{s,i}]\right|,\sum_{i=1}^{5}r_i\max\left|[G_{s,i}]\right|\right]\right)
\cap\left[(e^{-\sqrt{2/3}R})^s\overline{M_s},(e^{-\sqrt{2/3}R})^{-s}\overline{M_s}\right]\cap[0,\infty). \tag*{(89)} [ M s ] = ( [ M s ] + [ − i = 1 ∑ 5 r i max ∣ [ G s , i ] ∣ , i = 1 ∑ 5 r i max ∣ [ G s , i ] ∣ ] ) ∩ [ ( e − 2/3 R ) s M s , ( e − 2/3 R ) − s M s ] ∩ [ 0 , ∞ ) . (89)
Retain the displayed powers and outward evaluation order. Then
{ E ( U ( p ) Z 3 , s ) : p ∈ B } ⊂ [ M s ] . \{E(U(p)Z^3,s):p\in B\}\subset[M_s]. { E ( U ( p ) Z 3 , s ) : p ∈ B } ⊂ [ M s ] .
M 3 > 81 / 10 or M 6 > 5 ⟹ K ∩ ( B × G ) = ∅ . \mathcal{M}_3>81/10\ \text{or}\ \mathcal{M}_6>5\quad\Longrightarrow\quad\mathcal{K}\cap(B\times G)=\varnothing. M 3 > 81/10 or M 6 > 5 ⟹ K ∩ ( B × G ) = ∅ .
Supporting-parameter contraction. Initially G = [ 2 V 0 2 , 2 V 1 2 ] G=[2V_0^2,2V_1^2] G = [ 2 V 0 2 , 2 V 1 2 ] . Lemma A.12 gives
G ∩ C ( [ M 3 ] ) = ∅ ⟹ K ∩ ( B × G ) = ∅ . G\cap C([\mathcal{M}_3])=\varnothing\quad\Longrightarrow\quad\mathcal{K}\cap(B\times G)=\varnothing. G ∩ C ([ M 3 ]) = ∅ ⟹ K ∩ ( B × G ) = ∅ .
Otherwise, with [ l , u ] = G ∩ C ( [ M 3 ] ) [l,u]=G\cap C([\mathcal{M}_3]) [ l , u ] = G ∩ C ([ M 3 ]) , update
G ⟵ G ∩ [ 2 − 13 ⌈ 2 13 l ⌉ , 2 − 13 ⌈ 2 13 u ⌉ ] . (90) G\longleftarrow G\cap\left[2^{-13}\lceil2^{13}l\rceil,2^{-13}\lceil2^{13}u\rceil\right]. \tag*{(90)} G ⟵ G ∩ [ 2 − 13 ⌈ 2 13 l ⌉ , 2 − 13 ⌈ 2 13 u ⌉ ] . (90)
The affine FCC piece, both contacts and exterior endpoints remain; branch inversions stay inside their active intervals. Every child and Newton contraction inherits G G G .
Witness and stationary tests. Use N = 6 N=6 N = 6 , (T5), and the ordered witnesses
F , L n / 500 ( n = 147 , 148 , 149 , 150 , 151 ) , H n / 100 ( n = 103 , 104 , … , 112 ) . (91) F,\qquad L_{n/500}\ (n=147,148,149,150,151),\qquad H_{n/100}\ (n=103,104,\ldots,112). \tag*{(91)} F , L n /500 ( n = 147 , 148 , 149 , 150 , 151 ) , H n /100 ( n = 103 , 104 , … , 112 ) . (91)
For a supporting V V V , Proposition 2.7 gives
J ( Λ ) = Φ V ( Λ ) − f ( V ) , Φ V ( W ) ≥ f ( V ) , Φ V ( Λ ) > Φ V ( W ) ⟹ J ( Λ ) > 0. J(\Lambda)=\Phi_V(\Lambda)-f(V),\qquad\Phi_V(W)\ge f(V),\qquad\Phi_V(\Lambda)>\Phi_V(W)\quad\Longrightarrow\quad J(\Lambda)>0. J ( Λ ) = Φ V ( Λ ) − f ( V ) , Φ V ( W ) ≥ f ( V ) , Φ V ( Λ ) > Φ V ( W ) ⟹ J ( Λ ) > 0.
For G = [ ℓ , u ] G=[\ell,u] G = [ ℓ , u ] , evaluate the endpoint signed kernels of Lemma 5.2 at V = e / 2 V=\sqrt{e}/2 V = e /2 , e = ℓ , u e=\ell,u e = ℓ , u , using the exact coefficient e e e , without evaluating a square root, and reciprocal volume one:
π 6 240 ⟼ π 6 120 , π 3 2 ⟼ V 2 π 3 . \frac{\pi^6}{240}\longmapsto\frac{\pi^6}{120},\qquad\frac{\pi^3}{2}\longmapsto V^2\pi^3. 240 π 6 ⟼ 120 π 6 , 2 π 3 ⟼ V 2 π 3 .
Use (2.10) and the complete errors (2.11) with these coefficients. At both endpoints, check the critical-root signs on [ 1 / 4 , 8 ] [1/4,8] [ 1/4 , 8 ] ; uniqueness and at most 60 bisections enclose each critical argument. Joint and separate endpoint evaluations use the same bounds. Affinity in V 2 V^2 V 2 makes the endpoint hull an enclosure of the objective and all shape derivatives.
For the 32-sign Hessian bound Q e Q_e Q e of (A.3), require one common witness W W W with
( [ M 6 ] − E ( W , 6 ) ) − e ( [ M 3 ] − E ( W , 3 ) ) − 1 2 Q e > 0 , e = ℓ , u . (92) ([M_6]-E(W,6))-e([M_3]-E(W,3))-\frac{1}{2}Q_e>0,\qquad e=\ell,u. \tag*{(92)} ([ M 6 ] − E ( W , 6 )) − e ([ M 3 ] − E ( W , 3 )) − 2 1 Q e > 0 , e = ℓ , u . (92)
The scalar metric test replaces Q e Q_e Q e by R 2 max ( 0 , K e ) R^2\max(0,K_e) R 2 max ( 0 , K e ) , where
K e = 2 e ( e − 2 / 3 R ) − 3 [ M 3 ] − 4 ( e − 2 / 3 R ) 6 [ M 6 ] + 2 ∑ n ∈ C 4 + max { 0 , 28 b n 2 − 8 e b n } , K_e=2e\left(e^{-\sqrt{2/3}R}\right)^{-3}[M_3]-4\left(e^{-\sqrt{2/3}R}\right)^6[M_6]+2\sum_{n\in C_4^+}\max\{0,28b_n^2-8eb_n\}, K e = 2 e ( e − 2/3 R ) − 3 [ M 3 ] − 4 ( e − 2/3 R ) 6 [ M 6 ] + 2 n ∈ C 4 + ∑ max { 0 , 28 b n 2 − 8 e b n } ,
C N + = { n ∈ [ − N , N ] 3 ∩ Z 3 : n 1 > 0 ∨ n 1 = 0 < n 2 ∨ n 1 = n 2 = 0 < n 3 } , N ∈ N . C_N^+ = \left\{ n \in[-N,N]^3 \cap\mathbb{Z}^3 : n_1 > 0 \mathbin{\vee} n_1 = 0 < n_2 \mathbin{\vee} n_1 = n_2 = 0 < n_3 \right\}, \qquad N \in\mathbb{N}. C N + = { n ∈ [ − N , N ] 3 ∩ Z 3 : n 1 > 0 ∨ n 1 = 0 < n 2 ∨ n 1 = n 2 = 0 < n 3 } , N ∈ N .
Evaluate [ b n ] = 1 / ( e − 2 / 3 R q c ( n ) ) 3 [b_n]=1/(e^{-\sqrt{2/3}R_{q_c}(n)})^3 [ b n ] = 1/ ( e − 2/3 R q c ( n ) ) 3 outwards; evaluate both polynomial terms on this same interval, then take the upper endpoint and its maximum with zero. Convexity gives
( e − 2 / 3 R q c ( n ) ) 3 ≥ 7 2 ℓ ⟹ 28 b 2 − 8 e b ≤ 0 ( 0 ≤ b ≤ ( e − 2 / 3 R q c ( n ) ) − 3 , e ∈ [ ℓ , u ] ) . \left(e^{-\sqrt{2/3}R_{q_c}(n)}\right)^3 \ge\frac{7}{2\ell} \implies28b^2-8eb\le0 \qquad\left(0\le b\le\left(e^{-\sqrt{2/3}R_{q_c}(n)}\right)^{-3},\ e\in[\ell,u]\right). ( e − 2/3 R q c ( n ) ) 3 ≥ 2 ℓ 7 ⟹ 28 b 2 − 8 e b ≤ 0 ( 0 ≤ b ≤ ( e − 2/3 R q c ( n ) ) − 3 , e ∈ [ ℓ , u ] ) .
Enable this correction only when
( 25 e − 2 / 3 R ∥ Q c − 1 ∥ ∞ ) 3 ≥ 7 2 ℓ ; \left(25e^{-\sqrt{2/3}R}\lVert Q_c^{-1}\rVert_{\infty}\right)^3 \ge\frac{7}{2\ell}; ( 25 e − 2/3 R ∥ Q c − 1 ∥ ∞ ) 3 ≥ 2 ℓ 7 ;
all omitted-index corrections then vanish. In K e K_e K e , use M 3 M_3 M 3 and M 6 M_6 M 6 . Lemma 5.2 gives (T10) for every 2 V 2 ∈ G 2V^2\in G 2 V 2 ∈ G ; neither ∇ p Φ V − ( c ) = 0 \nabla_p\Phi_{V^-}(c)=0 ∇ p Φ V − ( c ) = 0 nor ∇ p Φ V + ( c ) = 0 \nabla_p\Phi_{V^+}(c)=0 ∇ p Φ V + ( c ) = 0 is assumed.
The remaining exclusions are
∃ i : 0 ∉ hull 2 V 2 ∈ G [ ∂ i Φ V ] ( B ) , or max e ∈ { ℓ , u } sup p ∈ B ∇ p T ∇ p 2 Φ e 2 ( U ( p ) Z 3 ) v < 0 \exists i:\ 0\notin\operatorname{hull}_{2V^2\in G}[\partial_i\Phi_V](B), \qquad\text{or}\qquad\max_{e\in\{\ell,u\}}\sup_{p\in B}\nabla_p^{\mathsf T}\nabla_p^2\Phi\,\frac{\sqrt e}{2}(U(p)Z^3)_v<0 ∃ i : 0 ∈ / hull 2 V 2 ∈ G [ ∂ i Φ V ] ( B ) , or e ∈ { ℓ , u } max p ∈ B sup ∇ p T ∇ p 2 Φ 2 e ( U ( p ) Z 3 ) v < 0
for a fixed dyadic v v v . Affinity and Lemma 5.2 exclude every intermediate-parameter local minimum.
For endpoint hulls [ H ] [H] [ H ] , [ g c ] [g_c] [ g c ] , use
B ⟵ B ∩ { c − P [ g c ] + ( I − P [ H ] ) ( B − c ) } . (93) B\longleftarrow B\cap\{c-P[g_c]+(I-P[H])(B-c)\}. \tag*{(93)} B ⟵ B ∩ { c − P [ g c ] + ( I − P [ H ]) ( B − c )} . (93)
The rounded midpoint inverse P P P needs neither accuracy nor invertibility for (A.4); failed construction disables Newton.
On the midpoint endpoint-Hessian hull, use at most 40 Jacobi rotations with angle
1 2 atan2 ( 2 A i j , A j j − A i i ) , \frac{1}{2}\operatorname{atan2}(2A_{ij},A_{jj}-A_{ii}), 2 1 atan2 ( 2 A ij , A j j − A ii ) ,
choosing the lexicographically first maximal ∣ A i j ∣ |A_{ij}| ∣ A ij ∣ , i ≠ j i\ne j i = j , and stopping at max i ≠ j ∣ A i j ∣ < 10 − 12 \max_{i\ne j}|A_{ij}|<10^{-12} max i = j ∣ A ij ∣ < 1 0 − 12 . For accumulated rotation matrix O O O ,
k ∈ argmin i A i i , A k k < 0 ⟹ v = round 2 − 20 Z ( O e k ) . k\in\operatorname*{argmin}_i A_{ii},\qquad A_{kk}<0\implies v=\operatorname{round}_{2^{-20}\mathbb{Z}}(Oe_k). k ∈ i argmin A ii , A k k < 0 ⟹ v = round 2 − 20 Z ( O e k ) .
Round ties away from zero; deletion requires both complete strict quadratic-form bounds. Construct P P P by partial-pivot elimination of the midpoint Hessian; a pivot of magnitude < 10 − 14 <10^{-14} < 1 0 − 14 disables Newton.
Basin recognition. The unit reference Korkine–Zolotarev parameters are
p F = ( 2 1 / 6 , 3 2 − 5 / 6 , 1 2 , 1 2 , 1 3 ) , p T = ( ( 149 125 ) 1 / 3 , ( 125 149 ) 1 / 6 , 1 2 , 1 2 , 0 ) , p_F=\left(2^{1/6},\sqrt{3}\,2^{-5/6},\frac{1}{2},\frac{1}{2},\frac{1}{3}\right),\qquad
p_T=\left(\left(\frac{149}{125}\right)^{1/3},\left(\frac{125}{149}\right)^{1/6},\frac{1}{2},\frac{1}{2},0\right), p F = ( 2 1/6 , 3 2 − 5/6 , 2 1 , 2 1 , 3 1 ) , p T = ( ( 125 149 ) 1/3 , ( 149 125 ) 1/6 , 2 1 , 2 1 , 0 ) ,
p H = ( ( 10 8 80864043 ) 1 / 6 , 3 2 ( 10 8 80864043 ) 1 / 6 , 1 2 , 0 , 0 ) . (94) p_H=\left(\left(\frac{10^8}{80864043}\right)^{1/6},\frac{\sqrt{3}}{2}\left(\frac{10^8}{80864043}\right)^{1/6},\frac{1}{2},0,0\right). \tag*{(94)} p H = ( ( 80864043 1 0 8 ) 1/6 , 2 3 ( 80864043 1 0 8 ) 1/6 , 2 1 , 0 , 0 ) . (94)
For FCC and FCT also use the isometric reference y 0 = − 1 / 2 y_0=-1/2 y 0 = − 1/2 . For each reference, Lemma A.2 with c = p 0 c=p_0 c = p 0 gives
M = U ( p ) U ( p 0 ) − 1 , S = 3 M T M tr ( M T M ) − I , g i j = ( M T M ) i j , M=U(p)U(p_0)^{-1},\qquad S=\frac{3M^{\mathsf T}M}{\operatorname{tr}(M^{\mathsf T}M)}-I,\qquad g_{ij}=(M^{\mathsf T}M)_{ij}, M = U ( p ) U ( p 0 ) − 1 , S = tr ( M T M ) 3 M T M − I , g ij = ( M T M ) ij ,
{ tr ( M T M ) } 2 ∥ S ∥ F 2 = 3 ∑ 1 ≤ i < j ≤ 3 ( g i i − g j j ) 2 + 18 ( g 12 2 + g 13 2 + g 23 2 ) . \left\{\operatorname{tr}(M^{\mathsf T}M)\right\}^2\lVert S\rVert_F^2
=3\sum_{1\le i<j\le3}(g_{ii}-g_{jj})^2+18(g_{12}^2+g_{13}^2+g_{23}^2). { tr ( M T M ) } 2 ∥ S ∥ F 2 = 3 1 ≤ i < j ≤ 3 ∑ ( g ii − g j j ) 2 + 18 ( g 12 2 + g 13 2 + g 23 2 ) .
Accept by Lemmas 5.4, 5.5 and 5.6 only under the corresponding whole-box condition:
G ⊂ [ 220499 / 100000 , 229 / 100 ] , sup p ∈ B ∥ S ∥ F ≤ 1 / 40 ( F ) , G\subset[220499/100000,229/100],\qquad\sup_{p\in B}\lVert S\rVert_F\le1/40\qquad(F), G ⊂ [ 220499/100000 , 229/100 ] , p ∈ B sup ∥ S ∥ F ≤ 1/40 ( F ) ,
G ⊂ [ 2267 / 1000 , 579 / 250 ] , sup p ∈ B ∥ S ∥ F ≤ 19 / 1000 ( T ) , G\subset[2267/1000,579/250],\qquad\sup_{p\in B}\lVert S\rVert_F\le19/1000\qquad(T), G ⊂ [ 2267/1000 , 579/250 ] , p ∈ B sup ∥ S ∥ F ≤ 19/1000 ( T ) ,
G ⊂ [ 231499 / 100000 , 242001 / 100000 ] , sup p ∈ B ∥ S ∥ F ≤ 1 / 50 ( H ) . (95) G\subset[231499/100000,242001/100000],\qquad\sup_{p\in B}\lVert S\rVert_F\le1/50\qquad(H). \tag*{(95)} G ⊂ [ 231499/100000 , 242001/100000 ] , p ∈ B sup ∥ S ∥ F ≤ 1/50 ( H ) . (95)
Use checked radicals and right polar decomposition. For a supporting V V V , Corollary 5.7 gives
∇ p Φ V ( U ( p ) Z 3 ) = 0 , (96) \nabla_p\Phi_V(U(p)Z^3)=0, \tag*{(96)} ∇ p Φ V ( U ( p ) Z 3 ) = 0 , (96)
⟹ U ( p ) Z 3 ≅ Ξ ∈ { F , L X ( V ) , H Y ( V ) } , J ( U ( p ) Z 3 ) = Φ V ( Ξ ) − f ( V ) ≥ 0. \Longrightarrow U(p)Z^3\cong\Xi\in\{F,L_X(V),H_Y(V)\},\qquad J(U(p)Z^3)=\Phi_V(\Xi)-f(V)\ge0. ⟹ U ( p ) Z 3 ≅ Ξ ∈ { F , L X ( V ) , H Y ( V )} , J ( U ( p ) Z 3 ) = Φ V ( Ξ ) − f ( V ) ≥ 0.
Subdivision. Let D ( B ) D(B) D ( B ) denote the outward intersections (T1)–(T4), the unit-covolume product/Gauss contractions of Lemma A.2 , and the guarded shear bounds
a δ ∈ [ 25 / 32 , 8 / 7 ] , x ≥ 1 − ( δ + / a − ) 2 , z ≥ 1 − ( a 2 δ 4 ) − 1 . a\delta\in[25/32,8/7],\qquad x\ge\sqrt{1-(\delta_+/a_-)^2},\qquad z\ge\sqrt{1-(a^2\delta^4)^{-1}}. a δ ∈ [ 25/32 , 8/7 ] , x ≥ 1 − ( δ + / a − ) 2 , z ≥ 1 − ( a 2 δ 4 ) − 1 .
Each shear bound requires a positive radicand. Use at most 10 inner and 10 outer passes, stopping at unchanged binary endpoints; if needed, finish with one final call to the same inner loop. Intersect the a , δ a,\delta a , δ ranges with (T1) and repeat these contractions. Apply D D D on enqueueing and removal, and before Newton volume comparison; enqueueing applies it again. Empty intersections reject.
Use the binary max-heap and insertion/removal/child-order ties of Lemma A.2 , with key v + ( B ) v^+(B) v + ( B ) and a single root. For the ordered witnesses ( W 1 , … , W m ) (W_1,\ldots,W_m) ( W 1 , … , W m ) in (T9), set
j ∗ = min a r g m i n 1 ≤ j ≤ m E ‾ ( W j , 6 ) − ( mid G ) E ‾ ( W j , 3 ) . j_*=\min\mathop{\rm argmin}_{1\le j\le m}\overline{E}(W_j,6)-(\operatorname{mid}G)\overline{E}(W_j,3). j ∗ = min argmin 1 ≤ j ≤ m E ( W j , 6 ) − ( mid G ) E ( W j , 3 ) .
The scalar test uses W j ∗ W_{j_*} W j ∗ ; rectangular tests cycle through j ∗ , … , m , 1 , … , j ∗ − 1 j_*,\ldots,m,1,\ldots,j_*-1 j ∗ , … , m , 1 , … , j ∗ − 1 , using one W j W_j W j at both endpoints of (T10). Write g N , c N g_N,c_N g N , c N for the complete gradient and fixed-direction exclusions above at cutoff N N N ; the same subscript on moments or Hessians specifies their complete cutoff. At each arrow continue only if no terminal test succeeds:
D ⟶ (T13) , D\longrightarrow\text{(T13)}, D ⟶ (T13) ,
R ≤ 9 / 20 : ( [ M s ] , [ G s , i ] ) 3 ⟶ (T7) ⟶ (T8) ⟶ (T13) R\le9/20:\quad([M_s],[G_{s,i}])_3\longrightarrow\text{(T7)}\longrightarrow\text{(T8)}\longrightarrow\text{(T13)} R ≤ 9/20 : ([ M s ] , [ G s , i ] ) 3 ⟶ (T7) ⟶ (T8) ⟶ (T13)
⟶ (T10) Q e = R 2 max ( 0 , K e ) ⟶ g 4 , \longrightarrow\text{(T10)}_{Q_e=R^2\max(0,K_e)}\longrightarrow g_4, ⟶ (T10) Q e = R 2 m a x ( 0 , K e ) ⟶ g 4 ,
R ≤ 3 / 20 : N = 2 , 3 , 4 : ( [ H − ] , [ H + ] ) N ⟶ (T10) ⟶ g N ⟶ c N ; R\le3/20:\quad N=2,3,4:\quad([H_-],[H_+])_N\longrightarrow\text{(T10)}\longrightarrow g_N\longrightarrow c_N; R ≤ 3/20 : N = 2 , 3 , 4 : ([ H − ] , [ H + ] ) N ⟶ (T10) ⟶ g N ⟶ c N ;
R ≤ 3 / 20 : [ H − ] 4 , [ H + ] 4 valid : [ M s ] 4 ⟶ (T10) R\le3/20:\quad[H_-]_4,[H_+]_4\ \text{valid}:\quad[M_s]_4\longrightarrow\text{(T10)} R ≤ 3/20 : [ H − ] 4 , [ H + ] 4 valid : [ M s ] 4 ⟶ (T10)
⟶ [ ∇ p r Φ V e ( c ) ] N = 4 ; r = 1 , 2 ; e = − , + ⟶ (T11) . \longrightarrow[\nabla_p^r\Phi_{V_e}(c)]_{N=4;\ r=1,2;\ e=-,+}\longrightarrow\text{(T11)}. ⟶ [ ∇ p r Φ V e ( c ) ] N = 4 ; r = 1 , 2 ; e = − , + ⟶ (T11) .
Here s = 3 , 6 s=3,6 s = 3 , 6 and i = 1 , … , 5 i=1,\ldots,5 i = 1 , … , 5 . Invalid endpoint Hessians stop the Hessian sequence and disable its final Newton attempt; no invalid bound deletes a box.
When (T11) is formed, B N = ∅ B_N=\varnothing B N = ∅ rejects. Otherwise set B ~ = D ( B N ) \widetilde{B}=D(B_N) B = D ( B N ) for the Newton child B N B_N B N . If B ~ = ∅ \widetilde{B}=\varnothing B = ∅ , reject; otherwise retain it only under
v + ( B ~ ) < RN 53 ( RN 53 ( 995 / 1000 ) v − ( B ) ) . v^+(\widetilde{B})<\operatorname{RN}_{53}\left(\operatorname{RN}_{53}(995/1000)\,v^-(B)\right). v + ( B ) < RN 53 ( RN 53 ( 995/1000 ) v − ( B ) ) .
This selects a contraction, not an exclusion. Reinsertion applies D D D again and leaves the pathwise bisection count k k k unchanged.
Otherwise, with the ordinary scores of Lemma A.2 ,
i ∗ = min a r g m a x i s i ( B ) , b = mid b i n ( B i ∗ ) , i_*=\min\mathop{\rm argmax}_i s_i(B),\qquad b=\operatorname{mid}_{\mathrm{bin}}(B_{i_*}), i ∗ = min argmax i s i ( B ) , b = mid bin ( B i ∗ ) ,
where mid b i n \operatorname{mid}_{\mathrm{bin}} mid bin is the binary midpoint. Set
k ≥ 120 or b ∉ ( B ‾ i ∗ , B ‾ i ∗ ) : U ← U ∪ { ( B , G ) } , k\ge120\ \text{or}\ b\notin(\underline{B}_{i_*},\overline{B}_{i_*}):\qquad U\leftarrow U\cup\{(B,G)\}, k ≥ 120 or b ∈ / ( B i ∗ , B i ∗ ) : U ← U ∪ {( B , G )} ,
k < 120 , B ‾ i ∗ < b < B ‾ i ∗ : B − = B ∩ { p i ∗ ≤ b } , B + = B ∩ { p i ∗ ≥ b } . k<120,\qquad\underline{B}_{i_*}<b<\overline{B}_{i_*}:\qquad B^-=B\cap\{p_{i_*}\le b\},\qquad B^+=B\cap\{p_{i_*}\ge b\}. k < 120 , B i ∗ < b < B i ∗ : B − = B ∩ { p i ∗ ≤ b } , B + = B ∩ { p i ∗ ≥ b } .
Before domain contraction,
B − ∪ B + = B , G − = G + = G , k − = k + = k + 1. B^-\cup B^+=B,\qquad G^-=G^+=G,\qquad k^-=k^+=k+1. B − ∪ B + = B , G − = G + = G , k − = k + = k + 1.
Enqueue D ( B − ) D(B^-) D ( B − ) before D ( B + ) D(B^+) D ( B + ) ; the scalar interval G G G is never bisected.
Invariant and finite outcome. Let R \mathcal{R} R be the union of accepted products B × G B \times G B × G , and Q , U \mathcal{Q},\mathcal{U} Q , U the queued and unresolved pairs. For ( p , 2 V 2 ) ∈ K ∩ ( B × G ) (p,2V^2) \in K \cap(B \times G) ( p , 2 V 2 ) ∈ K ∩ ( B × G ) ,
∇ p Φ V = 0 , ∇ p 2 Φ V ⪰ 0. \nabla_p\Phi_V=0,\qquad\nabla_p^2\Phi_V \succeq0. ∇ p Φ V = 0 , ∇ p 2 Φ V ⪰ 0.
Whenever (T5)–(T8) are evaluated, and whenever (T11) is formed, respectively,
E ( U ( p ) Z s , s ) ∈ [ M s ] , 2 V 2 ∈ G ∩ C ( [ M 3 ] ) , E(U(p)\mathcal{Z}^s,s)\in[\mathcal{M}_s],\qquad2V^2\in G\cap C([\mathcal{M}_3]), E ( U ( p ) Z s , s ) ∈ [ M s ] , 2 V 2 ∈ G ∩ C ([ M 3 ]) ,
p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ] ) ( B − c ) } . p\in B\cap\{c-P[g_c]+(I-P[H])(B-c)\}. p ∈ B ∩ { c − P [ g c ] + ( I − P [ H ]) ( B − c )} .
Thus the stationary exclusions and (T11) retain every such pair. Closed bisection gives
B × G = ( B − × G ) ∪ ( B + × G ) , K ⊂ R ∪ ⋃ ( B , G ) ∈ Q ∪ U ( B × G ) . B\times G=(B^-\times G)\cup(B^+\times G),\qquad K\subset\mathcal{R}\cup\bigcup_{(B,G)\in\mathcal{Q}\cup\mathcal{U}}(B\times G). B × G = ( B − × G ) ∪ ( B + × G ) , K ⊂ R ∪ ( B , G ) ∈ Q ∪ U ⋃ ( B × G ) .
Boxes and supporting intervals are closed; the inclusion holds on reduction boundaries and both contact bands with their one-sided parameter conditions.
The finite record is
( # nodes , # domain-excluded children ) = ( 127320 , 2137 ) , (\#\text{nodes},\#\text{domain-excluded children})=(127320,2137), ( # nodes , # domain-excluded children ) = ( 127320 , 2137 ) ,
( # value , # gradient , # curvature ) = ( 50642 , 3587 , 0 ) , (\#\text{value},\#\text{gradient},\#\text{curvature})=(50642,3587,0), ( # value , # gradient , # curvature ) = ( 50642 , 3587 , 0 ) ,
( # empty Newton , # moment/support ) = ( 364 , 1954 ) , (\#\text{empty Newton},\#\text{moment/support})=(364,1954), ( # empty Newton , # moment/support ) = ( 364 , 1954 ) ,
( # F C C , # tetragonal , # S H ) = ( 31 , 11 , 30 ) , (\#\mathrm{FCC},\#\text{tetragonal},\#\mathrm{SH})=(31,11,30), ( # FCC , # tetragonal , # SH ) = ( 31 , 11 , 30 ) ,
( # retained Newton , # bisections ) = ( 11946 , 58755 ) , (\#\text{retained Newton},\#\text{bisections})=(11946,58755), ( # retained Newton , # bisections ) = ( 11946 , 58755 ) ,
( # moment evaluations , # support contractions ) = ( 126921 , 45445 ) , (\#\text{moment evaluations},\#\text{support contractions})=(126921,45445), ( # moment evaluations , # support contractions ) = ( 126921 , 45445 ) ,
( # Hessian levels , # endpoint pairs ) = ( 265719 , 3899 ) . (\#\text{Hessian levels},\#\text{endpoint pairs})=(265719,3899). ( # Hessian levels , # endpoint pairs ) = ( 265719 , 3899 ) .
Hence
Q = U = ∅ , K ⊂ R , { ( p ( Λ ) , 2 V 2 ) ∈ K : J ( Λ ) < 0 } = ∅ . \mathcal{Q}=\mathcal{U}=\varnothing,\qquad K\subset\mathcal{R},\qquad\{(p(\Lambda),2V^2)\in K:J(\Lambda)<0\}=\varnothing. Q = U = ∅ , K ⊂ R , {( p ( Λ ) , 2 V 2 ) ∈ K : J ( Λ ) < 0 } = ∅ .
No depth-120 box remains. Endpoint comparisons outside the moment band and the exhausted cover inside it give
J ( Λ ) ≥ 0 , J ( Λ ) = 0 ⟹ Φ V ( Λ ) = f ( V ) , J(\Lambda)\geq0,\qquad J(\Lambda)=0\Longrightarrow\Phi_V(\Lambda)=f(V), J ( Λ ) ≥ 0 , J ( Λ ) = 0 ⟹ Φ V ( Λ ) = f ( V ) ,
for every supporting V V V . Basin identification and Proposition 2.7 give exactly the stated phase/contact equality cases; endpoint uniqueness was proved separately. This completes the proof of Numerical Lemma A.13. □
Lemma A.14. (The long SH reference moment graph.) For the unit-covolume family H Y H_Y H Y ,
A ( Y ) = E ( H Y , 6 ) , B ( Y ) = E ( H Y , 3 ) , Γ ( Y ) = A t ( Y ) B t ( Y ) , t = log Y . A(Y)=E(H_Y,6),\qquad B(Y)=E(H_Y,3),\qquad\Gamma(Y)=\frac{A_t(Y)}{B_t(Y)},\qquad t=\log Y. A ( Y ) = E ( H Y , 6 ) , B ( Y ) = E ( H Y , 3 ) , Γ ( Y ) = B t ( Y ) A t ( Y ) , t = log Y . For 11 / 10 ≤ Y ≤ 217 / 50 11/10\leq Y\leq217/50 11/10 ≤ Y ≤ 217/50 ,
2062436 10 6 < B t < 20076780 10 6 , 987405 10 6 < Γ t < 5931991 10 6 . (97) \frac{2062436}{10^6}<B_t<\frac{20076780}{10^6},\qquad\frac{987405}{10^6}<\Gamma_t<\frac{5931991}{10^6}. \tag*{(97)} 1 0 6 2062436 < B t < 1 0 6 20076780 , 1 0 6 987405 < Γ t < 1 0 6 5931991 . (97) For ( V 0 , V 1 ) = ( 11 / 10 , 89 / 50 ) (V_0,V_1)=(11/10,89/50) ( V 0 , V 1 ) = ( 11/10 , 89/50 ) , Γ ( Y ( V i ) ) = 2 V i 2 \Gamma(Y(V_i))=2V_i^2 Γ ( Y ( V i )) = 2 V i 2 and
1101310326229 10 12 < Y ( V 0 ) < 1101310326324 10 12 , \frac{1101310326229}{10^{12}}<Y(V_0)<\frac{1101310326324}{10^{12}}, 1 0 12 1101310326229 < Y ( V 0 ) < 1 0 12 1101310326324 , 4336021407082 10 12 < Y ( V 1 ) < 4336021407176 10 12 . (98) \frac{4336021407082}{10^{12}}<Y(V_1)<\frac{4336021407176}{10^{12}}. \tag*{(98)} 1 0 12 4336021407082 < Y ( V 1 ) < 1 0 12 4336021407176 . (98) For b i = B ( Y ( V i ) ) b_i=B(Y(V_i)) b i = B ( Y ( V i )) and a i = A ( Y ( V i ) ) a_i=A(Y(V_i)) a i = A ( Y ( V i )) ,
8.06262196892621 < b 0 < 8.06262196911185 , 8.06262196892621<b_0<8.06262196911185, 8.06262196892621 < b 0 < 8.06262196911185 , 4.96288456272628 < a 0 < 4.96288456316477 , 4.96288456272628<a_0<4.96288456316477, 4.96288456272628 < a 0 < 4.96288456316477 , 21.41363138319924 < b 1 < 21.41363138365243 , 21.41363138319924<b_1<21.41363138365243, 21.41363138319924 < b 1 < 21.41363138365243 , 63.56748489078682 < a 1 < 63.56748489357881. (99) 63.56748489078682<a_1<63.56748489357881. \tag*{(99)} 63.56748489078682 < a 1 < 63.56748489357881. (99)
All finite decimals are rational; b 1 < 43 / 2 b_1 < 43/2 b 1 < 43/2 , a 1 < 64 a_1 < 64 a 1 < 64 .
For the 4097 nodes indexed by j ∈ Z j \in\mathbb{Z} j ∈ Z ,
y j = 11 10 + 81 j 25 ⋅ 4096 = 225280 + 162 j 204800 , 0 ≤ j ≤ 4096 , (100) y_j = \frac{11}{10} + \frac{81j}{25 \cdot4096} = \frac{225280 + 162j}{204800}, \qquad0 \le j \le4096, \tag*{(100)} y j = 10 11 + 25 ⋅ 4096 81 j = 204800 225280 + 162 j , 0 ≤ j ≤ 4096 , (100)
for D ∈ { A , B , Γ } D \in\{A, B, \Gamma\} D ∈ { A , B , Γ } ,
D ( y j ) ∈ [ D j − , D j + ] , D j + − D j − < 10 − 9 ( 0 ≤ j ≤ 4096 ) , D(y_j) \in[D_j^-,D_j^+], \qquad D_j^+ - D_j^- < 10^{-9} \qquad(0 \le j \le4096), D ( y j ) ∈ [ D j − , D j + ] , D j + − D j − < 1 0 − 9 ( 0 ≤ j ≤ 4096 ) ,
D j + < D j + 1 − ( 0 ≤ j < 4096 ) . D_j^+ < D_{j+1}^- \qquad(0 \le j < 4096). D j + < D j + 1 − ( 0 ≤ j < 4096 ) .
Moreover,
Γ j + 1 + − Γ j − < 1651 10 6 ( 0 ≤ j < 4096 ) . (101) \Gamma_{j+1}^+ - \Gamma_j^- < \frac{1651}{10^6} \qquad(0 \le j < 4096). \tag*{(101)} Γ j + 1 + − Γ j − < 1 0 6 1651 ( 0 ≤ j < 4096 ) . (101)
[ y i − , y i + ] ⊂ ( y 0 , y 4096 ) ( i = 0 , 1 ) , and [y_i^-,y_i^+] \subset(y_0,y_{4096}) \quad(i=0,1), \qquad\text{and} [ y i − , y i + ] ⊂ ( y 0 , y 4096 ) ( i = 0 , 1 ) , and
Γ ( y 0 ) < 2 V 0 2 < 2 V 1 2 < Γ ( y 4096 ) . \Gamma(y_0) < 2V_0^2 < 2V_1^2 < \Gamma(y_{4096}). Γ ( y 0 ) < 2 V 0 2 < 2 V 1 2 < Γ ( y 4096 ) .
Proof. Use Lemma 6.2 , with its multiplicities, on
n ∈ ( [ − 12 , 12 ] 3 ∩ Z 3 ) ∖ { 0 } , Q Y , Q Y − 1 ⪰ 1 4 I . n \in([-12,12]^3 \cap\mathbb{Z}^3) \setminus\{0\}, \qquad Q_Y,Q_Y^{-1} \succeq\frac{1}{4}I. n ∈ ([ − 12 , 12 ] 3 ∩ Z 3 ) ∖ { 0 } , Q Y , Q Y − 1 ⪰ 4 1 I .
The complete tail is
T = 240 e − 169 π / 8 ( 1 + 8 ) 2 1 − e − 27 π / 8 < 5295 10 29 . (102) T = 240\frac{e^{-169\pi/8}(1+\sqrt{8})^2}{1-e^{-27\pi/8}} < \frac{5295}{10^{29}}. \tag*{(102)} T = 240 1 − e − 27 π /8 e − 169 π /8 ( 1 + 8 ) 2 < 1 0 29 5295 . (102)
For F ∈ { A , B } F \in\{A,B\} F ∈ { A , B } and derivative orders r = 0 , 1 , 2 r=0,1,2 r = 0 , 1 , 2 , use
[ F ] ← [ F ] + [ 0 , T ] , [ F t ] ← [ F t ] + [ − T , T ] , [ F t t ] ← [ F t t ] + [ − T , T ] . [F] \leftarrow[F]+[0,T], \qquad[F_t] \leftarrow[F_t]+[-T,T], \qquad[F_{tt}] \leftarrow[F_{tt}]+[-T,T]. [ F ] ← [ F ] + [ 0 , T ] , [ F t ] ← [ F t ] + [ − T , T ] , [ F tt ] ← [ F tt ] + [ − T , T ] .
For [ y j , y j + 1 ] [y_j,y_{j+1}] [ y j , y j + 1 ] , j ∈ Z j \in\mathbb{Z} j ∈ Z , 0 ≤ j < 4096 0 \le j < 4096 0 ≤ j < 4096 , the complete enclosures give
Γ = A t B t , Γ t = A t t − Γ B t t B t . \Gamma= \frac{A_t}{B_t}, \qquad\Gamma_t = \frac{A_{tt}-\Gamma B_{tt}}{B_t}. Γ = B t A t , Γ t = B t A tt − Γ B tt .
Together with (A.6)–(A.10),
B t > 0 , Γ t > 0 , A t = Γ B t > 4 , B_t > 0, \qquad\Gamma_t > 0, \qquad A_t = \Gamma B_t > 4, B t > 0 , Γ t > 0 , A t = Γ B t > 4 ,
B ( y j + 1 ) − B ( y j ) > 1 / 3000 , Γ ( y j + 1 ) − Γ ( y j ) > 1 / 10000 , B(y_{j+1})-B(y_j) > 1/3000, \qquad\Gamma(y_{j+1})-\Gamma(y_j) > 1/10000, B ( y j + 1 ) − B ( y j ) > 1/3000 , Γ ( y j + 1 ) − Γ ( y j ) > 1/10000 ,
D j + < D j + 1 − ( D ∈ { A , B , Γ } , 0 ≤ j < 4096 ) , D_j^+ < D_{j+1}^- \qquad(D \in\{A,B,\Gamma\},\ 0 \le j < 4096), D j + < D j + 1 − ( D ∈ { A , B , Γ } , 0 ≤ j < 4096 ) ,
Γ ( y 0 ) < 2 V 0 2 < 2 V 1 2 < Γ ( y 4096 ) . \Gamma(y_0) < 2V_0^2 < 2V_1^2 < \Gamma(y_{4096}). Γ ( y 0 ) < 2 V 0 2 < 2 V 1 2 < Γ ( y 4096 ) .
The same finite sums and complete tails give
8.06015362123 < B ( 11 / 10 ) < 8.06015362124 , 8.06015362123 < B(11/10) < 8.06015362124, 8.06015362123 < B ( 11/10 ) < 8.06015362124 ,
21.43203378605 < B ( 217 / 50 ) < 21.43203378608. 21.43203378605 < B(217/50) < 21.43203378608. 21.43203378605 < B ( 217/50 ) < 21.43203378608.
Hence
B ( y 0 ) < b 0 − < b 0 + < b 1 − < b 1 + < B ( y 4096 ) . B(y_0) < b_0^- < b_0^+ < b_1^- < b_1^+ < B(y_{4096}). B ( y 0 ) < b 0 − < b 0 + < b 1 − < b 1 + < B ( y 4096 ) .
For i ∈ { 0 , 1 } i \in\{0,1\} i ∈ { 0 , 1 } and y i ± y_i^\pm y i ± in (A.7),
2.4199999998707 < Γ ( y 0 − ) < 2.4199999998756 , 2.4199999998707 < \Gamma(y_0^-) < 2.4199999998756, 2.4199999998707 < Γ ( y 0 − ) < 2.4199999998756 ,
2.4200000000669 < Γ ( y 0 + ) < 2.4200000000719 , 2.4200000000669 < \Gamma(y_0^+) < 2.4200000000719, 2.4200000000669 < Γ ( y 0 + ) < 2.4200000000719 ,
6.3367999999742 < Γ ( y 1 − ) < 6.3367999999839 , 6.3367999999742 < \Gamma(y_1^-) < 6.3367999999839, 6.3367999999742 < Γ ( y 1 − ) < 6.3367999999839 ,
6.3368000001024 < Γ ( y 1 + ) < 6.3368000001121. 6.3368000001024 < \Gamma(y_1^+) < 6.3368000001121. 6.3368000001024 < Γ ( y 1 + ) < 6.3368000001121.
Since B t > 0 B_t>0 B t > 0 and A t = Γ B t > 0 A_t=\Gamma B_t>0 A t = Γ B t > 0 ,
Γ ( y i − ) < 2 V i 2 < Γ ( y i + ) ⟹ Y ( V i ) ∈ ( y i − , y i + ) , \Gamma(y_i^-) < 2V_i^2 < \Gamma(y_i^+) \Longrightarrow Y(V_i) \in(y_i^-,y_i^+), Γ ( y i − ) < 2 V i 2 < Γ ( y i + ) ⟹ Y ( V i ) ∈ ( y i − , y i + ) ,
E ( H Y ( V i ) , s ) ∈ [ E ( H y i − , s ) , E ( H y i + , s ) ] ( i ∈ { 0 , 1 } , s ∈ { 3 , 6 } ) . E(H_{Y(V_i)},s) \in[E(H_{y_i^-},s),E(H_{y_i^+},s)] \qquad(i \in\{0,1\},\ s \in\{3,6\}). E ( H Y ( V i ) , s ) ∈ [ E ( H y i − , s ) , E ( H y i + , s )] ( i ∈ { 0 , 1 } , s ∈ { 3 , 6 }) .
These are (A.7)–(A.8). This completes the proof of Lemma A.14. □
Lemma A.15 (A transverse moment Hessian bound on the SH tube). In Lemma 6.1 , fix t t t and put
Y = 3 2 t 3 , w = ( r − 2 / 3 , x − 1 / 2 , y , z ) . Y=\frac{\sqrt{3}}{2t^3},\qquad w=(r-2/\sqrt{3},x-1/2,y,z). Y = 2 t 3 3 , w = ( r − 2/ 3 , x − 1/2 , y , z ) . For 11 / 10 ≤ Y ≤ 217 / 50 11/10\leq Y\leq217/50 11/10 ≤ Y ≤ 217/50 and ∥ w ∥ ∞ ≤ 1 / 100 \lVert w\rVert_{\infty}\leq1/100 ∥ w ∥ ∞ ≤ 1/100 ,
∥ ∇ w 2 E ( U ( t , r , x , y , z ) Z 3 , 3 ) ∥ o p < 150. \lVert\nabla_w^2 E(U(t,r,x,y,z)\mathbb{Z}^3,3)\rVert_{\mathrm{op}}<150. ∥ ∇ w 2 E ( U ( t , r , x , y , z ) Z 3 , 3 ) ∥ op < 150. 0 < δ ≤ 1 / 100 0<\delta\leq1/100 0 < δ ≤ 1/100 , ∥ w ∥ ∞ ≤ δ \lVert w\rVert_{\infty}\leq\delta ∥ w ∥ ∞ ≤ δ imply
∣ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ∣ < 300 δ 2 . \left|E(U\mathbb{Z}^3,3)-E(H_Y,3)\right|<300\delta^2. E ( U Z 3 , 3 ) − E ( H Y , 3 ) < 300 δ 2 . Moreover,
E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) , 1 4 V d d v E ( H Y ( v ) , 3 ) > 4 17 E(U\mathbb{Z}^3,3)=E(H_Y(V),3),\qquad\frac{1}{4V}\frac{\mathrm{d}}{\mathrm{d}v}E(H_Y(v),3)>\frac{4}{17} E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) , 4 V 1 d v d E ( H Y ( v ) , 3 ) > 17 4 ( min { V , Γ ( Y ) / 2 } ≤ v ≤ max { V , Γ ( Y ) / 2 } ) \left(\min\{V,\sqrt{\Gamma(Y)/2}\}\leq v\leq\max\{V,\sqrt{\Gamma(Y)/2}\}\right) ( min { V , Γ ( Y ) /2 } ≤ v ≤ max { V , Γ ( Y ) /2 } ) ⟹ ∣ 2 V 2 − Γ ( Y ) ∣ < 1275 δ 2 . \Longrightarrow\quad\left|2V^2-\Gamma(Y)\right|<1275\delta^2. ⟹ 2 V 2 − Γ ( Y ) < 1275 δ 2 .
Proof. Use Lemma 6.1 , N = 4 N=4 N = 4 , and the complete tails (25) , with coefficients ( π 3 / 2 , 0 ) (\pi^3/2,0) ( π 3 /2 , 0 ) for ( E ( ⋅ , 3 ) , E ( ⋅ , 6 ) ) (E(\cdot,3),E(\cdot,6)) ( E ( ⋅ , 3 ) , E ( ⋅ , 6 )) . For every primal or dual index n ∈ C 4 + n\in C_4^+ n ∈ C 4 + , the Hessian contribution is
2 { K 3 ′ ′ ( q ) ∂ i q ∂ j q + K 3 ′ ( q ) ∂ i j q } , i , j ∈ { r , x , y , z } . 2\{K_3^{\prime\prime}(q)\partial_iq\partial_jq+K_3^{\prime}(q)\partial_{ij}q\},\qquad i,j\in\{r,x,y,z\}. 2 { K 3 ′′ ( q ) ∂ i q ∂ j q + K 3 ′ ( q ) ∂ ij q } , i , j ∈ { r , x , y , z } .
Use full old-coordinate boxes and spectral floors, with both terms of (L12). On the closed cells
I d , k = [ 11 10 + 81 k 25 ⋅ 2 d , 11 10 + 81 ( k + 1 ) 25 ⋅ 2 d ] I_{d,k}=\left[\frac{11}{10}+\frac{81k}{25\cdot2^d},\frac{11}{10}+\frac{81(k+1)}{25\cdot2^d}\right] I d , k = [ 10 11 + 25 ⋅ 2 d 81 k , 10 11 + 25 ⋅ 2 d 81 ( k + 1 ) ]
and ∥ w ∥ ∞ ≤ 1 / 100 \lVert w\rVert_{\infty}\leq1/100 ∥ w ∥ ∞ ≤ 1/100 , the complete bounds are
sup Y ∈ I d , k ∥ w ∥ ∞ ≤ 1 / 100 max 1 ≤ i ≤ 4 ∑ j = 1 4 ∣ ∂ w i ∂ w j E ( U Z 3 , 3 ) ∣ ≤ c , \sup_{\substack{Y\in I_{d,k}\\\lVert w\rVert_{\infty}\leq1/100}}\max_{1\leq i\leq4}\sum_{j=1}^{4}\left|\partial_{w_i}\partial_{w_j}E(U\mathbb{Z}^3,3)\right|\leq c, Y ∈ I d , k ∥ w ∥ ∞ ≤ 1/100 sup 1 ≤ i ≤ 4 max j = 1 ∑ 4 ∂ w i ∂ w j E ( U Z 3 , 3 ) ≤ c ,
for
( d , k , c ) ∈ { ( 2 , 0 , 129 ) , ( 2 , 1 , 145 ) , ( 3 , 4 , 122 ) , ( 3 , 5 , 134 ) , ( 3 , 6 , 146 ) , ( 4 , 14 , 138 ) , ( 4 , 15 , 144 ) } . (d,k,c)\in\{(2,0,129),(2,1,145),(3,4,122),(3,5,134),(3,6,146),(4,14,138),(4,15,144)\}. ( d , k , c ) ∈ {( 2 , 0 , 129 ) , ( 2 , 1 , 145 ) , ( 3 , 4 , 122 ) , ( 3 , 5 , 134 ) , ( 3 , 6 , 146 ) , ( 4 , 14 , 138 ) , ( 4 , 15 , 144 )} .
Thus
⋃ ( d , k ) ∈ { ( 2 , 0 ) , ( 2 , 1 ) , ( 3 , 4 ) , ( 3 , 5 ) , ( 3 , 6 ) , ( 4 , 14 ) , ( 4 , 15 ) } I d , k = [ 11 / 10 , 217 / 50 ] , \bigcup_{(d,k)\in\{(2,0),(2,1),(3,4),(3,5),(3,6),(4,14),(4,15)\}}I_{d,k}=[11/10,217/50], ( d , k ) ∈ {( 2 , 0 ) , ( 2 , 1 ) , ( 3 , 4 ) , ( 3 , 5 ) , ( 3 , 6 ) , ( 4 , 14 ) , ( 4 , 15 )} ⋃ I d , k = [ 11/10 , 217/50 ] ,
( # n o d e s , # l e a v e s ) = ( 13 , 7 ) , Q = U = ∅ . (\#\mathrm{nodes},\#\mathrm{leaves})=(13,7),\qquad Q=U=\varnothing. ( # nodes , # leaves ) = ( 13 , 7 ) , Q = U = ∅ .
For
H = ∇ ( w 1 , w 2 , w 3 , w 4 ) 2 E ( U ( t , 2 3 + w 1 , 1 2 + w 2 , w 3 , w 4 ) Z 3 , 3 ) = H T , H=\nabla^2_{(w_1,w_2,w_3,w_4)}E\left(U\left(t,\frac{2}{\sqrt{3}}+w_1,\frac{1}{2}+w_2,w_3,w_4\right)\mathbb{Z}^3,3\right)=H^{\mathsf{T}}, H = ∇ ( w 1 , w 2 , w 3 , w 4 ) 2 E ( U ( t , 3 2 + w 1 , 2 1 + w 2 , w 3 , w 4 ) Z 3 , 3 ) = H T ,
∥ H ∥ o p ≤ ∥ H ∥ ∞ = max 1 ≤ i ≤ 4 ∑ j = 1 4 ∣ H i j ∣ < 145.911368 < 150. \lVert H\rVert_{\mathrm{op}}\leq\lVert H\rVert_{\infty}=\max_{1\leq i\leq4}\sum_{j=1}^{4}|H_{ij}|<145.911368<150. ∥ H ∥ op ≤ ∥ H ∥ ∞ = 1 ≤ i ≤ 4 max j = 1 ∑ 4 ∣ H ij ∣ < 145.911368 < 150.
By symmetry and Lemma 6.2 ,
∇ w E ( U Z 3 , 3 ) ∣ w = 0 = 0 , 1 4 V d d V E ( H Y ( V ) , 3 ) > 4 17 . \left.\nabla_wE(U\mathbb{Z}^3,3)\right|_{w=0}=0,\qquad\frac{1}{4V}\frac{\mathrm{d}}{\mathrm{d}V}E(H_Y(V),3)>\frac{4}{17}. ∇ w E ( U Z 3 , 3 ) w = 0 = 0 , 4 V 1 d V d E ( H Y ( V ) , 3 ) > 17 4 .
Taylor’s formula and E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) E(U\mathbb{Z}^3,3)=E(H_Y(V),3) E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) give
∣ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ∣ ≤ 73 ∣ w ∣ 2 < 300 δ 2 , |E(U\mathbb{Z}^3,3)-E(H_Y,3)|\leq73|w|^2<300\delta^2, ∣ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ∣ ≤ 73∣ w ∣ 2 < 300 δ 2 ,
∣ 2 V 2 − Γ ( Y ) ∣ ≤ 17 4 ∣ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ∣ < 1275 δ 2 . |2V^2-\Gamma(Y)|\leq\frac{17}{4}|E(U\mathbb{Z}^3,3)-E(H_Y,3)|<1275\delta^2. ∣2 V 2 − Γ ( Y ) ∣ ≤ 4 17 ∣ E ( U Z 3 , 3 ) − E ( H Y , 3 ) ∣ < 1275 δ 2 .
This completes the proof of Lemma A.15. □
:::
Lemma A.16 (A separate layer bound on a larger transverse rectangle). Put Y = 3 / ( 2 t 3 ) Y=\sqrt{3}/(2t^{3}) Y = 3 / ( 2 t 3 ) and B = E ( U ( t , r , x , y , z ) Z 3 , 3 ) B=E(U(t,r,x,y,z)Z^{3},3) B = E ( U ( t , r , x , y , z ) Z 3 , 3 ) . On the full rectangle
11 10 ≤ Y ≤ 217 50 , ∣ r − 2 3 ∣ ≤ 1 200 , 99 200 ≤ x ≤ 1 2 , ∣ y ∣ , ∣ z ∣ ≤ 1 40 , \frac{11}{10}\leq Y\leq\frac{217}{50},\qquad\left|r-\frac{2}{\sqrt{3}}\right|\leq\frac{1}{200},\qquad\frac{99}{200}\leq x\leq\frac{1}{2},\qquad|y|,|z|\leq\frac{1}{40}, 10 11 ≤ Y ≤ 50 217 , r − 3 2 ≤ 200 1 , 200 99 ≤ x ≤ 2 1 , ∣ y ∣ , ∣ z ∣ ≤ 40 1 , one has
∥ ∇ ( y , z ) 2 B ∥ op < 10 , − 5 ( y 2 + z 2 ) ≤ B − E ( H Y , 3 ) ≤ 75 { ( r − 2 3 ) 2 + ( x − 1 2 ) 2 } . \lVert\nabla^{2}_{(y,z)}B\rVert_{\operatorname{op}}<10,\qquad-5(y^{2}+z^{2})\leq B-E(H_Y,3)\leq75\left\{\left(r-\frac{2}{\sqrt{3}}\right)^{2}+\left(x-\frac{1}{2}\right)^{2}\right\}. ∥ ∇ ( y , z ) 2 B ∥ op < 10 , − 5 ( y 2 + z 2 ) ≤ B − E ( H Y , 3 ) ≤ 75 { ( r − 3 2 ) 2 + ( x − 2 1 ) 2 } . If B = E ( H Y ( V ) , 3 ) B=E(H_Y(V),3) B = E ( H Y ( V ) , 3 ) for 11 / 10 ≤ V ≤ 89 / 50 11/10\leq V\leq89/50 11/10 ≤ V ≤ 89/50 , then
− 85 4 ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 1275 4 { ( r − 2 3 ) 2 + ( x − 1 2 ) 2 } . (103) -\frac{85}{4}(y^{2}+z^{2})\leq2V^{2}-\Gamma(Y)\leq\frac{1275}{4}\left\{\left(r-\frac{2}{\sqrt{3}}\right)^{2}+\left(x-\frac{1}{2}\right)^{2}\right\}. \tag*{(103)} − 4 85 ( y 2 + z 2 ) ≤ 2 V 2 − Γ ( Y ) ≤ 4 1275 { ( r − 3 2 ) 2 + ( x − 2 1 ) 2 } . (103) In particular,
− 17 640 ≤ 2 V 2 − Γ ( Y ) ≤ 51 3200 . -\frac{17}{640}\leq2V^{2}-\Gamma(Y)\leq\frac{51}{3200}. − 640 17 ≤ 2 V 2 − Γ ( Y ) ≤ 3200 51 .
Proof. On all 41 closed cells (F1) and the full transverse rectangle, the complete N = 4 N=4 N = 4 Hessians and transformed tails of Lemma A.15 give
∥ ∇ ( y , z ) 2 B ∥ op ≤ max a ∈ { y , z } ∑ b ∈ { y , z } ∣ B a b ∣ < 43519 10000 < 10 , B a b = ∂ a ∂ b B . \lVert\nabla^{2}_{(y,z)}B\rVert_{\operatorname{op}}\leq\max_{a\in\{y,z\}}\sum_{b\in\{y,z\}}|B_{ab}|<\frac{43519}{10000}<10,\qquad B_{ab}=\partial_{a}\partial_{b}B. ∥ ∇ ( y , z ) 2 B ∥ op ≤ a ∈ { y , z } max b ∈ { y , z } ∑ ∣ B ab ∣ < 10000 43519 < 10 , B ab = ∂ a ∂ b B .
Lemma A.15 gives
∥ ∇ ( r , x ) 2 E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) ∥ op < 150. \left\lVert\nabla^{2}_{(r,x)}E(U(t,r,x,0,0)Z^{3},3)\right\rVert_{\operatorname{op}}<150. ∇ ( r , x ) 2 E ( U ( t , r , x , 0 , 0 ) Z 3 , 3 ) op < 150.
Apply Lemma 6.4 with ( 10 , 150 ) (10,150) ( 10 , 150 ) and ( 4 V ) − 1 d d V E ( H Y ( V ) , 3 ) > 4 / 17 (4V)^{-1}\frac{\mathrm{d}}{\mathrm{d}V}E(H_Y(V),3)>4/17 ( 4 V ) − 1 d V d E ( H Y ( V ) , 3 ) > 4/17 . This completes the proof of Lemma A.16 . □
Numerical lemma A.17 (Finite exhaustion of the long-profile domain). Use Lemma 6.9 and
( V 0 , V 1 ) = ( 11 / 10 , 89 / 50 ) , 2 V 2 ∈ G = [ 2 V − 2 , 2 V + 2 ] ⊂ [ 2 V 0 2 , 2 V 1 2 ] . (V_{0},V_{1})=(11/10,89/50),\qquad2V^{2}\in G=[2V_{-}^{2},2V_{+}^{2}]\subset[2V_{0}^{2},2V_{1}^{2}]. ( V 0 , V 1 ) = ( 11/10 , 89/50 ) , 2 V 2 ∈ G = [ 2 V − 2 , 2 V + 2 ] ⊂ [ 2 V 0 2 , 2 V 1 2 ] . From the single closed root (L1),
# processed rectangles = 107013 , Q = U = ∅ . \#\text{processed rectangles}=107013,\qquad Q=\mathcal{U}=\varnothing. # processed rectangles = 107013 , Q = U = ∅ . For b i = E ( H Y ( V i ) , 3 ) b_i=E(H_Y(V_i),3) b i = E ( H Y ( V i ) , 3 ) and a representative η ( Λ ) \eta(\Lambda) η ( Λ ) in (L1), at every step,
Λ ∈ argmin ∣ L ∣ = 1 J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1 , \Lambda\in\underset{|L|=1}{\operatorname{argmin}}\,J(L),\qquad J(\Lambda)\leq0,\qquad b_0<E(\Lambda,3)<b_1, Λ ∈ ∣ L ∣ = 1 argmin J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1 , V ∈ argmax V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ , 3 ) } ⟹ Λ ≅ H Y ( V ) or ( η ( Λ ) , 2 V 2 ) ∈ ⋃ ( Q , G ) ∈ Q ∪ U ( Q × G ) . V\in\underset{V_0\leq v\leq V_1}{\operatorname{argmax}}\{f(v)+2v^{2}E(\Lambda,3)\}
\Longrightarrow\Lambda\cong H_Y(V)\quad\text{or}\quad(\eta(\Lambda),2V^{2})\in\bigcup_{(Q,G)\in\mathcal{Q}\cup\mathcal{U}}(Q\times G). V ∈ V 0 ≤ v ≤ V 1 argmax { f ( v ) + 2 v 2 E ( Λ , 3 )} ⟹ Λ ≅ H Y ( V ) or ( η ( Λ ) , 2 V 2 ) ∈ ( Q , G ) ∈ Q ∪ U ⋃ ( Q × G ) .
Proof. In the common interval estimates, use
η = ( t , r , x , y , z ) , Φ V ( η ) = E ( U ( η ) Z 3 , 6 ) − 2 V 2 E ( U ( η ) Z 3 , 3 ) . \eta=(t,r,x,y,z),\qquad\Phi_V(\eta)=E(U(\eta)Z^{3},6)-2V^{2}E(U(\eta)Z^{3},3). η = ( t , r , x , y , z ) , Φ V ( η ) = E ( U ( η ) Z 3 , 6 ) − 2 V 2 E ( U ( η ) Z 3 , 3 ) .
Eligible minima and the two endpoint boundaries. For b i = E ( H Y ( V i ) , 3 ) b_i=E(H_Y(V_i),3) b i = E ( H Y ( V i ) , 3 ) , endpoint comparison and uniqueness give
E ( Λ , 3 ) ≤ b 0 ⟹ J ( Λ ) = Φ V 0 ( Λ ) − f ( V 0 ) ≥ 0 , E(\Lambda,3)\leq b_0\Longrightarrow J(\Lambda)=\Phi_{V_0}(\Lambda)-f(V_0)\geq0, E ( Λ , 3 ) ≤ b 0 ⟹ J ( Λ ) = Φ V 0 ( Λ ) − f ( V 0 ) ≥ 0 ,
E ( Λ , 3 ) ≥ b 1 ⟹ J ( Λ ) = Φ V 1 ( Λ ) − f ( V 1 ) ≥ 0 , E(\Lambda,3)\geq b_1\Longrightarrow J(\Lambda)=\Phi_{V_1}(\Lambda)-f(V_1)\geq0, E ( Λ , 3 ) ≥ b 1 ⟹ J ( Λ ) = Φ V 1 ( Λ ) − f ( V 1 ) ≥ 0 ,
Φ V i ( Λ ) = f ( V i ) ⟹ Λ ≅ H Y ( V i ) , E ( Λ , 3 ) = b i ( i = 0 , 1 ) . \Phi_{V_i}(\Lambda)=f(V_i)\Longrightarrow\Lambda\cong H_Y(V_i),\qquad E(\Lambda,3)=b_i\quad(i=0,1). Φ V i ( Λ ) = f ( V i ) ⟹ Λ ≅ H Y ( V i ) , E ( Λ , 3 ) = b i ( i = 0 , 1 ) .
Lemma 6.2 gives d d V E ( H Y ( V ) , 3 ) > 0 \frac{\mathrm{d}}{\mathrm{d}V}E(H_Y(V),3)>0 d V d E ( H Y ( V ) , 3 ) > 0 , hence
Λ ∈ argmin ∣ L ∣ = 1 J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1 \Lambda\in\underset{|L|=1}{\operatorname{argmin}}\,J(L),\qquad J(\Lambda)\leq0,\qquad b_0<E(\Lambda,3)<b_1 Λ ∈ ∣ L ∣ = 1 argmin J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1
⟹ ∃ ! V ∈ ( V 0 , V 1 ) : E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) . \Longrightarrow\exists!V\in(V_0,V_1):E(\Lambda,3)=E(H_Y(V),3). ⟹ ∃ ! V ∈ ( V 0 , V 1 ) : E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) .
For this V V V ,
b 0 < E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) < b 1 , b_0 < E(\Lambda,3)=E(H_Y(V),3)<b_1, b 0 < E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) < b 1 ,
Λ ∈ argmin ∣ L ∣ = 1 Φ V ( L ) , ∇ η Φ V = 0 , ∇ η 2 Φ V ⪰ 0. \Lambda\in\underset{|L|=1}{\operatorname{argmin}}\ \Phi_V(L),\qquad\nabla_\eta\Phi_V=0,\qquad\nabla_\eta^2\Phi_V\succeq0. Λ ∈ ∣ L ∣ = 1 argmin Φ V ( L ) , ∇ η Φ V = 0 , ∇ η 2 Φ V ⪰ 0.
Proposition 5.1 uses all variations in ( 0 , ∞ ) 2 × R 3 (0,\infty)^2\times\mathbb{R}^3 ( 0 , ∞ ) 2 × R 3 , including at reduction boundaries. Choose a Korkine–Zolotarev representative η ( Λ ) \eta(\Lambda) η ( Λ ) in (L1), with Λ ≅ U ( η ( Λ ) ) Z 3 \Lambda\cong U(\eta(\Lambda))\mathbb{Z}^3 Λ ≅ U ( η ( Λ )) Z 3 , and set
K = { ( η ( Λ ) , 2 V 2 ) : Λ ∈ argmin ∣ L ∣ = 1 J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1 , V ∈ argmax V 0 ≤ v ≤ V 1 { f ( v ) + 2 v 2 E ( Λ , 3 ) } } . \mathcal{K}=\left\{(\eta(\Lambda),2V^2):
\begin{aligned}
&\Lambda\in\underset{|L|=1}{\operatorname{argmin}}\ J(L),\quad J(\Lambda)\leq0,\quad b_0<E(\Lambda,3)<b_1,\\
&V\in\underset{V_0\leq v\leq V_1}{\operatorname{argmax}}\{f(v)+2v^2E(\Lambda,3)\}
\end{aligned}
\right\}. K = ⎩ ⎨ ⎧ ( η ( Λ ) , 2 V 2 ) : Λ ∈ ∣ L ∣ = 1 argmin J ( L ) , J ( Λ ) ≤ 0 , b 0 < E ( Λ , 3 ) < b 1 , V ∈ V 0 ≤ v ≤ V 1 argmax { f ( v ) + 2 v 2 E ( Λ , 3 )} ⎭ ⎬ ⎫ .
Necessary domain contractions and direct lower sums. With box endpoints − , + -,+ − , + , intersect (L1), then iterate at most ten times:
t ∈ [ 749 1000 2 r + , 2 1 / 3 r − ] , r ∈ [ 749 1000 2 t + , 2 1 / 3 t − ] , t\in\left[\frac{749}{1000}\frac{2}{r_+},\frac{2^{1/3}}{r_-}\right],\qquad
r\in\left[\frac{749}{1000}\frac{2}{t_+},\frac{2^{1/3}}{t_-}\right], t ∈ [ 1000 749 r + 2 , r − 2 1/3 ] , r ∈ [ 1000 749 t + 2 , t − 2 1/3 ] ,
r ≤ ( 1 − x + 2 ) − 1 / 2 , t ≤ ( r + 1 − z + 2 ) 1 / 3 , r ≥ t − 3 ( 1 − z + 2 ) . (104) r\leq(1-x_+^2)^{-1/2},\qquad
t\leq\left(\frac{r_+}{1-z_+^2}\right)^{1/3},\qquad
r\geq t_-^3(1-z_+^2). \tag*{(104)} r ≤ ( 1 − x + 2 ) − 1/2 , t ≤ ( 1 − z + 2 r + ) 1/3 , r ≥ t − 3 ( 1 − z + 2 ) . (104)
After the corresponding strict preliminary comparisons, also intersect
x ≥ 1 − r − − 2 ( r − > 1 ) , z ≥ 1 − r + / t − 3 ( r + / t − 3 < 1 ) . x\geq\sqrt{1-r_-^{-2}}\quad(r_->1),\qquad
z\geq\sqrt{1-r_+/t_-^3}\quad(r_+/t_-^3<1). x ≥ 1 − r − − 2 ( r − > 1 ) , z ≥ 1 − r + / t − 3 ( r + / t − 3 < 1 ) .
Stop at unchanged binary endpoints. The bounds follow from (L2)–(L3); Q = ∅ Q=\varnothing Q = ∅ excludes. Use the coordinate hull
P = hull p ( Q ) , p ( η ) = ( a , δ , x , y , z ) = ( t r , t / r , x , y , z ) . (105) P=\operatorname{hull}p(Q),\qquad p(\eta)=(a,\delta,x,y,z)=(\sqrt{tr},\sqrt{t/r},x,y,z). \tag*{(105)} P = hull p ( Q ) , p ( η ) = ( a , δ , x , y , z ) = ( t r , t / r , x , y , z ) . (105)
For n ∈ C 1 + n\in C_1^+ n ∈ C 1 + ,
q n ( P ‾ ) < t r ⟹ K ∩ ( Q × G ) = ∅ . q_n(\overline{P})<tr\Longrightarrow\mathcal{K}\cap(Q\times G)=\varnothing. q n ( P ) < t r ⟹ K ∩ ( Q × G ) = ∅ .
In the outward reciprocal-then-square order, accumulate
D s ( Q ) = 2 ∑ n ∈ C 1 + { q n ( P ‾ ) } − s , s = 3 , 6 , D_s(Q)=2\sum_{n\in C_1^+}\{q_n(\overline{P})\}^{-s},\qquad s=3,6, D s ( Q ) = 2 n ∈ C 1 + ∑ { q n ( P ) } − s , s = 3 , 6 ,
( D 3 , D 6 ) ← ( 0 , 0 ) , D 3 ← D 3 + 2 ( 1 / q n ( P ‾ ) 3 ) , (D_3,D_6)\leftarrow(0,0),\qquad D_3\leftarrow D_3+2(1/q_n(\overline{P})^3), ( D 3 , D 6 ) ← ( 0 , 0 ) , D 3 ← D 3 + 2 ( 1/ q n ( P ) 3 ) ,
D 6 ← D 6 + 2 ( 1 / q n ( P ‾ ) 3 ) 2 . (106) D_6\leftarrow D_6+2(1/q_n(\overline{P})^3)^2. \tag*{(106)} D 6 ← D 6 + 2 ( 1/ q n ( P ) 3 ) 2 . (106)
Then
E ( U ( p ) Z 3 , s ) ≥ D s , D 3 ≥ 43 / 2 ∨ D 6 ≥ 64 ⟹ K ∩ ( Q × G ) = ∅ . E(U(p)\mathbb{Z}^3,s)\geq D_s,\qquad D_3\geq43/2\ \lor\ D_6\geq64\Longrightarrow\mathcal{K}\cap(Q\times G)=\varnothing. E ( U ( p ) Z 3 , s ) ≥ D s , D 3 ≥ 43/2 ∨ D 6 ≥ 64 ⟹ K ∩ ( Q × G ) = ∅ .
Apply these lower-endpoint tests on enqueueing and removal, and to Newton children before volume comparison.
Uniform elimination of all layer shifts. After the domain pass, if either whole-box guard (71) holds, use outward bounds for 2 / 3 2/\sqrt{3} 2/ 3 , Y = 3 / ( 2 t 3 ) Y=\sqrt{3}/(2t^3) Y = 3 / ( 2 t 3 ) and r ≤ 2 / 3 r\leq2/\sqrt{3} r ≤ 2/ 3 . Lemma 6.8 gives
Λ ∈ argmin ∣ L ∣ = 1 Φ V ( L ) , 2 V 2 ≥ 121 / 50 ⟹ y = z = 0. \Lambda\in\underset{|L|=1}{\operatorname{argmin}}\ \Phi_V(L),\qquad2V^2\geq121/50\Longrightarrow y=z=0. Λ ∈ ∣ L ∣ = 1 argmin Φ V ( L ) , 2 V 2 ≥ 121/50 ⟹ y = z = 0.
Thus
Q ∩ { y = z = 0 } = ∅ ⟹ K ∩ ( Q × G ) = ∅ , Q\cap\{y=z=0\}=\varnothing\Longrightarrow\mathcal{K}\cap(Q\times G)=\varnothing, Q ∩ { y = z = 0 } = ∅ ⟹ K ∩ ( Q × G ) = ∅ ,
Q ∩ { y = z = 0 } ≠ ∅ ⟹ Q ← Q ∩ ( R 3 × [ − 2 − 20 , 2 − 20 ] 2 ) . Q\cap\{y=z=0\}\neq\varnothing\Longrightarrow Q\leftarrow Q\cap\left(\mathbb{R}^3\times[-2^{-20},2^{-20}]^2\right). Q ∩ { y = z = 0 } = ∅ ⟹ Q ← Q ∩ ( R 3 × [ − 2 − 20 , 2 − 20 ] 2 ) .
Repeat the domain pass; K \mathcal{K} K and the coordinate-volume ordering are retained. No parameter subdivision or Fourier evaluation is used.
Moment bounds. Use the complete tails (25) and full-box spectral floors of Lemmas A.13 and A.2, with outward binary64 radicals and Taylor remainders, for t , r > 0 t,r>0 t , r > 0 . The evaluation data are
old-coordinate value/scalar/gradient: mid P P P , rad P P P , R P R_P R P , whole P P P ,
area value/gradient/Hessian/Newton: p ( mid Q ) p(\operatorname{mid} Q) p ( mid Q ) , ( ρ i ) i = 1 5 (\rho_i)_{i=1}^{5} ( ρ i ) i = 1 5 , whole Q Q Q ,
where mid P ≠ p ( mid Q ) \operatorname{mid} P \ne p(\operatorname{mid} Q) mid P = p ( mid Q ) in general. The floors
e − 2 / 3 R P ∥ Q ( mid P ) − 1 ∥ ∞ − 1 , e − 2 / 3 R P ∥ Q ( mid P ) ∥ − 1 e^{-\sqrt{2/3}R_P}\lVert Q(\operatorname{mid} P)^{-1}\rVert_{\infty}^{-1}, \qquad e^{-\sqrt{2/3}R_P}\lVert Q(\operatorname{mid} P)\rVert^{-1} e − 2/3 R P ∥ Q ( mid P ) − 1 ∥ ∞ − 1 , e − 2/3 R P ∥ Q ( mid P ) ∥ − 1
combine by maximum with the native full-P P P norm floors; norm upper bounds round upwards. For R P ≤ 9 / 20 R_P \le9/20 R P ≤ 9/20 , take N = 3 N=3 N = 3 and intersect (T7) of Lemma A.13 with
[ ( e − 2 / 3 R P ) s M ‾ s , ( e − 2 / 3 R P ) − s M ‾ s ] \left[(e^{-\sqrt{2/3}R_P})^{s}\overline{M}_s,(e^{-\sqrt{2/3}R_P})^{-s}\overline{M}_s\right] [ ( e − 2/3 R P ) s M s , ( e − 2/3 R P ) − s M s ]
to obtain [ M s ] [M_s] [ M s ] ; evaluate the displayed powers outwards.
Moment band and supporting-parameter contraction. The whole-box exclusions are
M ‾ 3 > 43 / 2 ∨ M ‾ 6 > 64 ∨ M ‾ 3 < b 0 − ∨ M ‾ 3 > b 1 + ⟹ K ∩ ( Q × G ) = ∅ . (107) \underline{M}_3>43/2\lor\underline{M}_6>64\lor\overline{M}_3<b_0^-\lor\underline{M}_3>b_1^+\Longrightarrow K\cap(Q\times G)=\varnothing. \tag*{(107)} M 3 > 43/2 ∨ M 6 > 64 ∨ M 3 < b 0 − ∨ M 3 > b 1 + ⟹ K ∩ ( Q × G ) = ∅ . (107)
With the complete endpoint intervals (A.8), intersect G G G with Lemma 6.3’s enclosure for [ M 3 ] [M_3] [ M 3 ] ; exclude an empty intersection. Otherwise, for [ l , u ] [l,u] [ l , u ] and w = RN 53 ( u − l ) w=\operatorname{RN}_{53}(u-l) w = RN 53 ( u − l ) ,
S ← 1 , S ← 2 S while S < 8192 , w S < 16 , S\gets1,\qquad S\gets2S\quad\text{while }S<8192,\ wS<16, S ← 1 , S ← 2 S while S < 8192 , w S < 16 ,
G ← G ∩ [ S − 1 ⌊ S l ⌋ , S − 1 ⌈ S u ⌉ ] . (108) G\gets G\cap[S^{-1}\lfloor Sl\rfloor,S^{-1}\lceil Su\rceil]. \tag*{(108)} G ← G ∩ [ S − 1 ⌊ S l ⌋ , S − 1 ⌈ S u ⌉] . (108)
Thus 2 V ∗ 2 ∈ G 2V_*^2\in G 2 V ∗ 2 ∈ G survives also for V ∗ = V 0 , V 1 V_*=V_0,V_1 V ∗ = V 0 , V 1 ; never bisect G G G . Only for b 0 < E ( Λ , 3 ) < b 1 b_0<E(\Lambda,3)<b_1 b 0 < E ( Λ , 3 ) < b 1 impose
∂ V { f ( V ) + 2 V 2 E ( Λ , 3 ) } = 0 ⟺ E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) . \partial_V\{f(V)+2V^2E(\Lambda,3)\}=0\Longleftrightarrow E(\Lambda,3)=E(H_Y(V),3). ∂ V { f ( V ) + 2 V 2 E ( Λ , 3 )} = 0 ⟺ E ( Λ , 3 ) = E ( H Y ( V ) , 3 ) .
Outside this band, no parameter-derivative equation is imposed.
Additional conditional geometry contraction. After the domain pass and before terminal tests, apply Lemma 6.7 whenever
11 / 10 ≤ 3 2 t 3 ≤ 217 / 50 , ∣ r − 2 / 3 ∣ , ∣ x − 1 / 2 ∣ , ∣ y ∣ , ∣ z ∣ ≤ 1 / 100. 11/10\le\frac{\sqrt{3}}{2t^3}\le217/50,\qquad|r-2/\sqrt{3}|,\ |x-1/2|,\ |y|,\ |z|\le1/100. 11/10 ≤ 2 t 3 3 ≤ 217/50 , ∣ r − 2/ 3 ∣ , ∣ x − 1/2∣ , ∣ y ∣ , ∣ z ∣ ≤ 1/100.
For the outward image [ Y − , Y + ] [Y_-,Y_+] [ Y − , Y + ] and transverse bounds ( d r , d x , d y , d z ) (d_r,d_x,d_y,d_z) ( d r , d x , d y , d z ) in (G2), use y j ∈ [ y ‾ j , y ‾ j ] y_j\in[\underline{y}_j,\overline{y}_j] y j ∈ [ y j , y j ] and Γ ( y j ) ∈ [ Γ j − , Γ j + ] \Gamma(y_j)\in[\Gamma_j^-,\Gamma_j^+] Γ ( y j ) ∈ [ Γ j − , Γ j + ] . Binary search gives
i − = max ( { 0 } ∪ { j ∈ { 0 , … , 4096 } : y ‾ j ≤ Y − } ) , i_-=\max\left(\{0\}\cup\{j\in\{0,\ldots,4096\}:\overline{y}_j\le Y_-\}\right), i − = max ( { 0 } ∪ { j ∈ { 0 , … , 4096 } : y j ≤ Y − } ) ,
i + = min ( { 4096 } ∪ { j ∈ { 0 , … , 4096 } : y ‾ j ≥ Y + } ) . i_+=\min\left(\{4096\}\cup\{j\in\{0,\ldots,4096\}:\underline{y}_j\ge Y_+\}\right). i + = min ( { 4096 } ∪ { j ∈ { 0 , … , 4096 } : y j ≥ Y + } ) .
Under E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) E(UZ^3,3)=E(H_Y(V),3) E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) , (G2) gives
G ← G ∩ [ Γ ‾ i − − 1275 4 ( d y 2 + d z 2 ) , Γ ‾ i + + 1275 4 ( d r 2 + d x 2 ) ] . G\gets G\cap\left[\underline{\Gamma}_{i_-}-\frac{1275}{4}(d_y^2+d_z^2),\overline{\Gamma}_{i_+}+\frac{1275}{4}(d_r^2+d_x^2)\right]. G ← G ∩ [ Γ i − − 4 1275 ( d y 2 + d z 2 ) , Γ i + + 4 1275 ( d r 2 + d x 2 ) ] .
No interpolation is used; only G = ∅ G=\varnothing G = ∅ excludes. Retain these uncoarsened bounds for (L15) and (G4); then apply (L8).
Direct rational area-coordinate jets. Use (L9)–(L10) and Lemma 5.2 at V = e / 2 V=\sqrt{e}/2 V = e /2 , e ∈ { inf G , sup G } e\in\{\inf G,\sup G\} e ∈ { inf G , sup G } , with covolume one. In Lemma 6.1, use ∂ i = ∂ / ∂ η i \partial_i=\partial/\partial\eta_i ∂ i = ∂ / ∂ η i , ∂ i j = ∂ 2 / ( ∂ η i ∂ η j ) \partial_{ij}=\partial^2/(\partial\eta_i\partial\eta_j) ∂ ij = ∂ 2 / ( ∂ η i ∂ η j ) and K ′ = d K / d q K'=dK/dq K ′ = d K / d q , K ′ ′ = d 2 K / d q 2 K''=d^2K/dq^2 K ′′ = d 2 K / d q 2 . Sum
2 K ′ ( q n ) ∂ i q n , 2 { K ′ ′ ( q n ) ∂ i q n ∂ j q n + K ′ ( q n ) ∂ i j q n } , 1 ≤ i , j ≤ 5. (109) 2K'(q_n)\partial_iq_n,\qquad2\{K''(q_n)\partial_iq_n\partial_jq_n+K'(q_n)\partial_{ij}q_n\},\qquad1\le i,j\le5. \tag*{(109)} 2 K ′ ( q n ) ∂ i q n , 2 { K ′′ ( q n ) ∂ i q n ∂ j q n + K ′ ( q n ) ∂ ij q n } , 1 ≤ i , j ≤ 5. (109)
and reciprocal terms over C N + C_N^+ C N + , using interval squares on diagonals and common squared-length arguments before subtraction. For each unique critical argument in [ 1 / 4 , 8 ] [1/4,8] [ 1/4 , 8 ] , full-parameter endpoint signs and at most 60 rational-kernel bisections retain its zero. Add (25) on all of P P P , at the same N N N and full-box floors, transformed by (L12). Since p ( Q ) ⊂ P p(Q)\subset P p ( Q ) ⊂ P , these tails cover all of Q Q Q , including non-Korkine–Zolotarev points; finite terms use (L9)–(L10) and (L11).
Comparisons and stationary exclusions. Use the witnesses
H n / 10 , n = 11 , 12 , … , 43. (110) H_{n/10}, \qquad n=11,12,\ldots,43. \tag*{(110)} H n /10 , n = 11 , 12 , … , 43. (110)
With complete N = 6 N=6 N = 6 positive moments and e = mid G e=\operatorname{mid} G e = mid G , start at
min n ∈ { 11 , … , 43 } argmin n ∈ { 11 , … , 43 } E ( H n / 10 , 6 ) − e E ( H n / 10 , 3 ) . \min_{n\in\{11,\ldots,43\}}\operatorname*{argmin}\limits_{n\in\{11,\ldots,43\}} E(H_{n/10},6)-eE(H_{n/10},3). n ∈ { 11 , … , 43 } min n ∈ { 11 , … , 43 } argmin E ( H n /10 , 6 ) − e E ( H n /10 , 3 ) .
Use it for the scalar test; rectangular tests visit all 33 cyclically, with one witness for both endpoints.
For 11 / 10 ≤ Y ≤ 6 11/10\leq Y\leq6 11/10 ≤ Y ≤ 6 , B t , Γ t > 0 B_t,\Gamma_t>0 B t , Γ t > 0 and
∂ log Y Φ V ( H Y ) = ( Γ ( Y ) − 2 V 2 ) ∂ log Y E ( H Y , 3 ) , \partial_{\log Y}\Phi_V(H_Y)=(\Gamma(Y)-2V^2)\partial_{\log Y}E(H_Y,3), ∂ l o g Y Φ V ( H Y ) = ( Γ ( Y ) − 2 V 2 ) ∂ l o g Y E ( H Y , 3 ) ,
Φ V ( H Y ( V ) ) ≤ Φ V ( H n / 10 ) . \Phi_V(H_Y(V))\leq\Phi_V(H_{n/10}). Φ V ( H Y ( V )) ≤ Φ V ( H n /10 ) .
For the same supporting V V V ,
Φ V ( U Z 3 ) > Φ V ( H n / 10 ) ⟹ J ( U Z 3 ) > 0. \Phi_V(UZ^3)>\Phi_V(H_{n/10})\quad\Longrightarrow\quad J(UZ^3)>0. Φ V ( U Z 3 ) > Φ V ( H n /10 ) ⟹ J ( U Z 3 ) > 0.
For the scalar metric test, use Lemma A.13 on P P P , mid P \operatorname{mid} P mid P , R P R_P R P : both occurrences of max ( 0 , 28 b n 2 / ( 8 e b n ) ) \max(0,28b_n^2/(8eb_n)) max ( 0 , 28 b n 2 / ( 8 e b n )) use the same lower squared-length endpoint, with its zero-skip and complete omitted-correction guard. The initial coordinate gradient uses all of P P P . For t , r > 0 t,r>0 t , r > 0 , (L5) gives
det ( ∂ p i ∂ η j ) 1 ≤ i , j ≤ 5 = − 1 2 r ≠ 0 , ∇ η Φ V = ( ∂ p i ∂ η j ) 1 ≤ i , j ≤ 5 T ∇ p Φ V , \det\left(\frac{\partial p_i}{\partial\eta_j}\right)_{1\leq i,j\leq5}=-\frac{1}{2r}\ne0,\qquad
\nabla_\eta\Phi_V=\left(\frac{\partial p_i}{\partial\eta_j}\right)_{1\leq i,j\leq5}^{\mathsf T}\nabla_p\Phi_V, det ( ∂ η j ∂ p i ) 1 ≤ i , j ≤ 5 = − 2 r 1 = 0 , ∇ η Φ V = ( ∂ η j ∂ p i ) 1 ≤ i , j ≤ 5 T ∇ p Φ V ,
so ∇ η Φ V = 0 ⟺ ∇ p Φ V = 0 \nabla_\eta\Phi_V=0\Longleftrightarrow\nabla_p\Phi_V=0 ∇ η Φ V = 0 ⟺ ∇ p Φ V = 0 .
For one fixed witness W W W , compute at p ( mid Q ) p(\operatorname{mid} Q) p ( mid Q ) the complete differences
Δ s = E ( U ( p ( mid Q ) ) Z 3 , s ) − E ( W , s ) , s = 3 , 6. \Delta_s=E(U(p(\operatorname{mid} Q))Z^3,s)-E(W,s),\qquad s=3,6. Δ s = E ( U ( p ( mid Q )) Z 3 , s ) − E ( W , s ) , s = 3 , 6.
With the area radii and 32-sign Hessian bounds Q e Q_e Q e of (A.3), from (L11)–(L12),
min e ∈ { inf G , sup G } Δ 6 − e Δ 3 − 1 2 Q e > 0 ⟹ K ∩ ( Q × G ) = ∅ . (111) \min_{e\in\{\inf G,\sup G\}}\frac{\Delta_6-e\Delta_3-\frac{1}{2}Q_e}{ } >0\quad\Longrightarrow\quad K\cap(Q\times G)=\varnothing. \tag*{(111)} e ∈ { i n f G , s u p G } min Δ 6 − e Δ 3 − 2 1 Q e > 0 ⟹ K ∩ ( Q × G ) = ∅ . (111)
The complete endpoint hulls also give
0 ∉ hull e ∈ { inf G , sup G } [ ∂ i Φ e / 2 ( Q ) ] ⟹ K ∩ ( Q × G ) = ∅ , 0\notin\operatorname{hull}_{e\in\{\inf G,\sup G\}}\left[\partial_i\Phi_{\sqrt{e/2}}(Q)\right]\quad\Longrightarrow\quad K\cap(Q\times G)=\varnothing, 0 ∈ / hull e ∈ { i n f G , s u p G } [ ∂ i Φ e /2 ( Q ) ] ⟹ K ∩ ( Q × G ) = ∅ ,
max e ∈ { inf G , sup G } sup η ∈ Q v T ∇ η 2 Φ e / 2 v < 0 ⟹ K ∩ ( Q × G ) = ∅ \max_{e\in\{\inf G,\sup G\}}\sup_{\eta\in Q}v^{\mathsf T}\nabla_\eta^2\Phi_{\sqrt{e/2}}v<0\quad\Longrightarrow\quad K\cap(Q\times G)=\varnothing e ∈ { i n f G , s u p G } max η ∈ Q sup v T ∇ η 2 Φ e /2 v < 0 ⟹ K ∩ ( Q × G ) = ∅
for 1 ≤ i ≤ 5 1\leq i\leq5 1 ≤ i ≤ 5 or one fixed dyadic v v v at both endpoints, respectively. For Newton, use (A.4) at mid Q \operatorname{mid} Q mid Q with full area-gradient/Hessian hulls; exclude empty intersections. Lemma A.13 and
∇ η j Φ V = ( 1 − θ ) ∇ η j Φ V − + θ ∇ η j Φ V + , j = 0 , 1 , 2 , V 2 = ( 1 − θ ) V − 2 + θ V + 2 , 0 ≤ θ ≤ 1 \nabla_\eta^j\Phi_V=(1-\theta)\nabla_\eta^j\Phi_{V_-}+\theta\nabla_\eta^j\Phi_{V_+},\qquad j=0,1,2,\qquad
V^2=(1-\theta)V_-^2+\theta V_+^2,\qquad0\leq\theta\leq1 ∇ η j Φ V = ( 1 − θ ) ∇ η j Φ V − + θ ∇ η j Φ V + , j = 0 , 1 , 2 , V 2 = ( 1 − θ ) V − 2 + θ V + 2 , 0 ≤ θ ≤ 1
give the complete enclosures for every V ∈ [ V − , V + ] V\in[V_-,V_+] V ∈ [ V − , V + ] .
Three terminal recognitions. On the complete rectangle and retained interval, test
V 2 t 3 ‾ ≤ ( 9 / 10 ) 3 , V 2 t 6 ≥ 64 , V − > 1. (112) \overline{V^2t^3}\leq(9/10)^3,\qquad\frac{V^2}{t^6}\geq64,\qquad V_->1. \tag*{(112)} V 2 t 3 ≤ ( 9/10 ) 3 , t 6 V 2 ≥ 64 , V − > 1. (112)
For ( η ( Λ ) , 2 V 2 ) ∈ K (\eta(\Lambda),2V^2)\in K ( η ( Λ ) , 2 V 2 ) ∈ K , Lemma 6.6 gives Λ ≅ H Y ( V ) \Lambda\cong H_Y(V) Λ ≅ H Y ( V ) and J ( Λ ) = 0 J(\Lambda)=0 J ( Λ ) = 0 . Use the uncoarsened conditional geometry bounds of Lemma 6.7.
Static Fourier recognition requires
11 10 ≤ 3 2 t 3 ≤ 217 50 , ∣ r − 2 3 ∣ ≤ 1 200 , 99 200 ≤ x ≤ 1 2 , ∣ y ∣ , ∣ z ∣ ≤ 1 40 . (113) \frac{11}{10}\leq\frac{\sqrt{3}}{2t^3}\leq\frac{217}{50},\qquad
\left|r-\frac{2}{\sqrt{3}}\right|\leq\frac{1}{200},\qquad
\frac{99}{200}\leq x\leq\frac{1}{2},\qquad
|y|,|z|\leq\frac{1}{40}. \tag*{(113)} 10 11 ≤ 2 t 3 3 ≤ 50 217 , r − 3 2 ≤ 200 1 , 200 99 ≤ x ≤ 2 1 , ∣ y ∣ , ∣ z ∣ ≤ 40 1 . (113)
The outward guards and E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) E(UZ^3,3)=E(H_Y(V),3) E ( U Z 3 , 3 ) = E ( H Y ( V ) , 3 ) give
− 17 640 ≤ 2 V 2 − Γ ( 3 2 t 3 ) ≤ 51 3200 . -\frac{17}{640}\leq2V^2-\Gamma\left(\frac{\sqrt{3}}{2t^3}\right)\leq\frac{51}{3200}. − 640 17 ≤ 2 V 2 − Γ ( 2 t 3 3 ) ≤ 3200 51 .
Lemma 6.5 and ∇ η Φ V = 0 \nabla_{\eta}\Phi_{V}=0 ∇ η Φ V = 0 give
( r , x , y , z ) = ( 2 / 3 , 1 / 2 , 0 , 0 ) , (r,x,y,z)=\left(2/\sqrt{3},1/2,0,0\right), ( r , x , y , z ) = ( 2/ 3 , 1/2 , 0 , 0 ) ,
∂ log Y Φ V ( H Y ) = 0 ⟹ Γ ( Y ) = 2 V 2 ⟹ Y = Y ( V ) , J = 0. \partial_{\log Y}\Phi_{V}(H_{Y})=0\Longrightarrow\Gamma(Y)=2V^{2}\Longrightarrow Y=Y(V),\qquad J=0. ∂ l o g Y Φ V ( H Y ) = 0 ⟹ Γ ( Y ) = 2 V 2 ⟹ Y = Y ( V ) , J = 0.
For the four-dimensional Hessian test, after moment contraction require
11 / 10 ≤ 3 2 t 3 ≤ 217 / 50 , ∣ r − 2 / 3 ∣ , ∣ x − 1 / 2 ∣ , ∣ y ∣ , ∣ z ∣ ≤ 1 / 100 , 11/10\leq\frac{\sqrt{3}}{2t^{3}}\leq217/50,\qquad\lvert r-2/\sqrt{3}\rvert,\ \lvert x-1/2\rvert,\ \lvert y\rvert,\ \lvert z\rvert\leq1/100, 11/10 ≤ 2 t 3 3 ≤ 217/50 , ∣ r − 2/ 3 ∣ , ∣ x − 1/2 ∣ , ∣ y ∣ , ∣ z ∣ ≤ 1/100 ,
and binary area/density widths ≤ RN 53 ( 1 / 200 ) , RN 53 ( 1 / 5 ) \leq\operatorname{RN}_{53}(1/200),\operatorname{RN}_{53}(1/5) ≤ RN 53 ( 1/200 ) , RN 53 ( 1/5 ) . The widths only schedule the test. Set
W = hull c o o r d ( proj r , x , y , z Q ∪ { ( 2 / 3 , 1 / 2 , 0 , 0 ) } ) . W=\operatorname{hull}_{\mathrm{coord}}\left(\operatorname{proj}_{r,x,y,z}Q\cup\{(2/\sqrt{3},1/2,0,0)\}\right). W = hull coord ( proj r , x , y , z Q ∪ {( 2/ 3 , 1/2 , 0 , 0 )} ) .
Use (L9)–(L11) and the complete N = 4 N=4 N = 4 tail on Q t × W Q_{t}\times W Q t × W .
Throughout Q t × W Q_{t}\times W Q t × W , for the primal and dual Gram matrices,
Q 0 , Q 1 ≻ 0 , Q 1 = Q 0 − 1 , ∥ Q 0 ∥ ∞ , ∥ Q 1 ∥ ∞ ≤ 25 / 8 ⟹ Q 0 , Q 1 ⪰ ( 8 / 25 ) I . Q_{0},Q_{1}\succ0,\qquad Q_{1}=Q_{0}^{-1},\qquad\lVert Q_{0}\rVert_{\infty},\lVert Q_{1}\rVert_{\infty}\leq25/8\Longrightarrow Q_{0},Q_{1}\succeq(8/25)I. Q 0 , Q 1 ≻ 0 , Q 1 = Q 0 − 1 , ∥ Q 0 ∥ ∞ , ∥ Q 1 ∥ ∞ ≤ 25/8 ⟹ Q 0 , Q 1 ⪰ ( 8/25 ) I .
Differentiate (L9), with outward row-norm bounds
D σ a ≥ sup Q t × W ∥ ∂ a Q σ ∥ ∞ , E σ a b ≥ sup Q t × W ∥ ∂ a b Q σ ∥ ∞ , σ = 0 , 1. D_{\sigma a}\geq\sup_{Q_{t}\times W}\lVert\partial_{a}Q_{\sigma}\rVert_{\infty},\qquad E_{\sigma ab}\geq\sup_{Q_{t}\times W}\lVert\partial_{ab}Q_{\sigma}\rVert_{\infty},\qquad\sigma=0,1. D σ a ≥ Q t × W sup ∥ ∂ a Q σ ∥ ∞ , E σ ab ≥ Q t × W sup ∥ ∂ ab Q σ ∥ ∞ , σ = 0 , 1.
For transverse a , b a,b a , b and every index,
∣ ∂ a q σ ∣ ≤ D σ a ∣ n ∣ 2 , ∣ ∂ a b q σ ∣ ≤ E σ a b ∣ n ∣ 2 . \lvert\partial_{a}q_{\sigma}\rvert\leq D_{\sigma a}\lvert n\rvert^{2},\qquad\lvert\partial_{ab}q_{\sigma}\rvert\leq E_{\sigma ab}\lvert n\rvert^{2}. ∣ ∂ a q σ ∣ ≤ D σ a ∣ n ∣ 2 , ∣ ∂ ab q σ ∣ ≤ E σ ab ∣ n ∣ 2 .
For n ∈ Z 3 ∖ [ − 4 , 4 ] 3 n\in\mathbb{Z}^{3}\setminus[-4,4]^{3} n ∈ Z 3 ∖ [ − 4 , 4 ] 3 , q σ ( n ) ≥ 8 q_{\sigma}(n)\geq8 q σ ( n ) ≥ 8 . For the unit-covolume primal/dual moment kernels K s K_{s} K s , s = 3 , 6 s=3,6 s = 3 , 6 ,
K s ( j ) = d j K s / d q j . K_{s}^{(j)}=\mathrm{d}^{j}K_{s}/\mathrm{d}q^{j}. K s ( j ) = d j K s / d q j .
Their integral representations and G a ( u ) ≤ e − u / ( u − 7 ) G_{a}(u)\leq e^{-u}/(u-7) G a ( u ) ≤ e − u / ( u − 7 ) , a ≤ 8 a\leq8 a ≤ 8 , give
∣ K s ( j ) ( q ) ∣ < 100 e − π q q , j = 1 , 2 , q ≥ 8. \lvert K_{s}^{(j)}(q)\rvert<\frac{100e^{-\pi q}}{q},\qquad j=1,2,\quad q\geq8. ∣ K s ( j ) ( q )∣ < q 100 e − π q , j = 1 , 2 , q ≥ 8.
For e ∈ { 2 V − 2 , 2 V + 2 } e\in\{2V_{-}^{2},2V_{+}^{2}\} e ∈ { 2 V − 2 , 2 V + 2 } , each omitted Hessian entry of E ( U ( η ) Z 3 , 6 ) − e E ( U ( η ) Z 3 , 3 ) E(U(\eta)Z^{3},6)-eE(U(\eta)Z^{3},3) E ( U ( η ) Z 3 , 6 ) − e E ( U ( η ) Z 3 , 3 ) has absolute bound
T a b = 100 25 8 ( 1 + ∣ e ∣ ) ∑ σ = 0 1 { D σ a D σ b S 2 + E σ a b S 0 } . (114) T_{ab}=100\frac{25}{8}(1+\lvert e\rvert)\sum_{\sigma=0}^{1}\{D_{\sigma a}D_{\sigma b}S_{2}+E_{\sigma ab}S_{0}\}. \tag*{(114)} T ab = 100 8 25 ( 1 + ∣ e ∣) σ = 0 ∑ 1 { D σ a D σ b S 2 + E σ ab S 0 } . (114)
with the complete omitted cube and no extra opposite-pair factor. For j = 0 , 2 j=0,2 j = 0 , 2 , Lemma A.2 gives
S j ≥ ∑ n ∈ Z 3 ∖ [ − 4 , 4 ] 3 ∣ n ∣ j e − 8 π ∣ n ∣ 2 / 25 , S_{j}\geq\sum_{n\in\mathbb{Z}^{3}\setminus[-4,4]^{3}}\lvert n\rvert^{j}e^{-8\pi\lvert n\rvert^{2}/25}, S j ≥ n ∈ Z 3 ∖ [ − 4 , 4 ] 3 ∑ ∣ n ∣ j e − 8 π ∣ n ∣ 2 /25 ,
S 0 ≤ Θ 4 ( 8 / 25 ) , S 2 ≤ 25 e − 4 π Θ 4 ( 4 / 25 ) . S_{0}\leq\Theta_{4}(8/25),\qquad S_{2}\leq25e^{-4\pi}\Theta_{4}(4/25). S 0 ≤ Θ 4 ( 8/25 ) , S 2 ≤ 25 e − 4 π Θ 4 ( 4/25 ) .
Indeed ( u e − 4 π u / 25 ) ′ < 0 (ue^{-4\pi u/25})'<0 ( u e − 4 π u /25 ) ′ < 0 for u ≥ 25 u\geq25 u ≥ 25 . Outward evaluation and diagonal interval squares make (L17) cover every omitted primal/dual index.
Positive outward Schur pivots throughout Q t × W Q_{t}\times W Q t × W , including (L17), give
H V − , H V + ≻ 0 , H V : = ∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) , H_{V_{-}},H_{V_{+}}\succ0,\qquad H_{V}:=\nabla^{2}_{(r,x,y,z)}\Phi_{V}(UZ^{3}), H V − , H V + ≻ 0 , H V := ∇ ( r , x , y , z ) 2 Φ V ( U Z 3 ) ,
H V = ( 1 − θ ) H V − + θ H V + ≻ 0 , V 2 = ( 1 − θ ) V − 2 + θ V + 2 , 0 ≤ θ ≤ 1. H_{V}=(1-\theta)H_{V_{-}}+\theta H_{V_{+}}\succ0,\qquad V^{2}=(1-\theta)V_{-}^{2}+\theta V_{+}^{2},\qquad0\leq\theta\leq1. H V = ( 1 − θ ) H V − + θ H V + ≻ 0 , V 2 = ( 1 − θ ) V − 2 + θ V + 2 , 0 ≤ θ ≤ 1.
For ( η ( Λ ) , 2 V 2 ) ∈ K (\eta(\Lambda),2V^{2})\in\mathcal{K} ( η ( Λ ) , 2 V 2 ) ∈ K , SH symmetry and the positive transverse Hessian give ( r , x , y , z ) = ( 2 / 3 , 1 / 2 , 0 , 0 ) (r,x,y,z)=(2/\sqrt{3},1/2,0,0) ( r , x , y , z ) = ( 2/ 3 , 1/2 , 0 , 0 ) . Then
∇ η Φ V = 0 ⟹ Γ ( Y ) = 2 V 2 ⟹ Y = Y ( V ) , J ( Λ ) = 0. \nabla_{\eta}\Phi_{V}=0\Longrightarrow\Gamma(Y)=2V^{2}\Longrightarrow Y=Y(V),\qquad J(\Lambda)=0. ∇ η Φ V = 0 ⟹ Γ ( Y ) = 2 V 2 ⟹ Y = Y ( V ) , J ( Λ ) = 0.
Schedule, subdivision and retention rules. Initialize the contracted root (L1) with G = [ 2 V 0 2 , 2 V 1 2 ] G=[2V_0^2,2V_1^2] G = [ 2 V 0 2 , 2 V 1 2 ] . Remove a box of greatest outward coordinate-volume upper bound, using the binary max-heap and insertion/removal tie rules of Lemma A.13 ; insert lower-coordinate children first. On removal, apply in order
domain ⟶ guarded shift restriction ⟶ domain ⟶ conditional geometry contraction ⟶ ( L 15 ) , ( L 16 ) . \text{domain}\longrightarrow\text{guarded shift restriction}\longrightarrow\text{domain}
\longrightarrow\text{conditional geometry contraction}\longrightarrow(L15),(L16). domain ⟶ guarded shift restriction ⟶ domain ⟶ conditional geometry contraction ⟶ ( L 15 ) , ( L 16 ) .
The ordered tests stop at the first terminal success; (L17) requires a valid moment enclosure and supporting-parameter contraction:
R P ≤ 9 / 20 : ( N = 3 , mid P , whole P ) ⟶ ( L 7 ) , ( L 8 ) R_P\le9/20:\quad(N=3,\text{ mid }P,\text{ whole }P)\longrightarrow(L7),(L8) R P ≤ 9/20 : ( N = 3 , mid P , whole P ) ⟶ ( L 7 ) , ( L 8 )
⟶ ( L 15 ) , ( L 16 ) ⟶ scheduled ( L 17 ) \longrightarrow(L15),(L16)\longrightarrow\text{scheduled }(L17) ⟶ ( L 15 ) , ( L 16 ) ⟶ scheduled ( L 17 )
⟶ scalar test ⟶ ( N = 4 , gradient on P ) . \longrightarrow\text{scalar test}\longrightarrow(N=4,\text{ gradient on }P). ⟶ scalar test ⟶ ( N = 4 , gradient on P ) .
For the nested area tests, use the center p ( mid Q ) p(\operatorname{mid}Q) p ( mid Q ) and radii ( ρ i ) i = 1 5 (\rho_i)_{i=1}^{5} ( ρ i ) i = 1 5 :
R P ≤ 3 / 20 : ( N = 3 , area-center moments ) R_P\le3/20:\quad(N=3,\text{ area-center moments}) R P ≤ 3/20 : ( N = 3 , area-center moments )
⟶ [ ( L 14 ) ⟶ area-gradient hull ⟶ dyadic curvature ] N = 2 , 3 , 4 \longrightarrow[(L14)\longrightarrow\text{ area-gradient hull}\longrightarrow\text{ dyadic curvature}]_{N=2,3,4} ⟶ [( L 14 ) ⟶ area-gradient hull ⟶ dyadic curvature ] N = 2 , 3 , 4
⟶ ( N = 4 , area-center moments ) ⟶ ( L 14 ) \longrightarrow(N=4,\text{ area-center moments})\longrightarrow(L14) ⟶ ( N = 4 , area-center moments ) ⟶ ( L 14 )
⟶ ( N = 4 , endpoint jets ) ⟶ Newton . \longrightarrow(N=4,\text{ endpoint jets})\longrightarrow\text{Newton}. ⟶ ( N = 4 , endpoint jets ) ⟶ Newton .
Invalid endpoint Hessians stop the N = 2 , 3 , 4 N=2,3,4 N = 2 , 3 , 4 sequence and disable its final Newton attempt. Use Lemma A.13 ’s fixed dyadic direction, rounded inverse and complete interval checks; failed enclosures disable the dependent tests. Retain nonempty Newton intersections by Lemma A.2 ’s outward-volume rule, never as an exclusion.
For [ r H ] ∋ 2 / 3 [r_H]\ni2/\sqrt{3} [ r H ] ∋ 2/ 3 , the SH-line intersection test is
Q r ∩ [ r H ] ≠ ∅ , 1 / 2 ∈ Q x , 0 ∈ Q y , 0 ∈ Q z . Q_r\cap[r_H]\ne\varnothing,\qquad1/2\in Q_x,\qquad0\in Q_y,\qquad0\in Q_z. Q r ∩ [ r H ] = ∅ , 1/2 ∈ Q x , 0 ∈ Q y , 0 ∈ Q z .
For an SH-line box, the t t t -priority rule is
[ 3 / ( 2 t 3 ) ] ∩ [ 11 / 10 , 217 / 50 ] ≠ ∅ , [ 3 / ( 2 t 3 ) ] ⊄ [ 11 / 10 , 217 / 50 ] , [\sqrt{3}/(2t^3)]\cap[11/10,217/50]\ne\varnothing,\qquad[\sqrt{3}/(2t^3)]\not\subset[11/10,217/50], [ 3 / ( 2 t 3 )] ∩ [ 11/10 , 217/50 ] = ∅ , [ 3 / ( 2 t 3 )] ⊂ [ 11/10 , 217/50 ] ,
RN 53 ( sup Q t − inf Q t ) > [ 1 ] / [ 10000 ] , inf Q t < ( mid Q ) t < sup Q t \operatorname{RN}_{53}(\sup Q_t-\inf Q_t)>[1]/[10000],\qquad\inf Q_t<(\operatorname{mid}Q)_t<\sup Q_t RN 53 ( sup Q t − inf Q t ) > [ 1 ] / [ 10000 ] , inf Q t < ( mid Q ) t < sup Q t
⟹ bisect t first . \Longrightarrow\quad\text{bisect }t\text{ first}. ⟹ bisect t first .
Here [ 1 ] / [ 10000 ] [1]/[10000] [ 1 ] / [ 10000 ] is the upper endpoint of outward interval division. For an SH-line box, if [ 3 / ( 2 t 3 ) ] ⊂ [ 11 / 10 , 217 / 50 ] [\sqrt{3}/(2t^3)]\subset[11/10,217/50] [ 3 / ( 2 t 3 )] ⊂ [ 11/10 , 217/50 ] , use
( max ∣ Q r − [ r H ] ∣ 1 / 200 , max ∣ Q x − 1 / 2 ∣ 1 / 200 , max ∣ Q y ∣ 1 / 40 , max ∣ Q z ∣ 1 / 40 ) ; \left(
\frac{\max|Q_r-[r_H]|}{1/200},
\frac{\max|Q_x-1/2|}{1/200},
\frac{\max|Q_y|}{1/40},
\frac{\max|Q_z|}{1/40}
\right); ( 1/200 max ∣ Q r − [ r H ] ∣ , 1/200 max ∣ Q x − 1/2∣ , 1/40 max ∣ Q y ∣ , 1/40 max ∣ Q z ∣ ) ;
otherwise use
( max ∣ Q r − [ r H ] ∣ , max ∣ Q x − 1 / 2 ∣ , max ∣ Q y ∣ , max ∣ Q z ∣ ) 1 / 10000 . \frac{\left(\max|Q_r-[r_H]|,\max|Q_x-1/2|,\max|Q_y|,\max|Q_z|\right)}{1/10000}. 1/10000 ( max ∣ Q r − [ r H ] ∣ , max ∣ Q x − 1/2∣ , max ∣ Q y ∣ , max ∣ Q z ∣ ) .
Round maxima upwards, retaining fractional divisions. If the maximum > 1 >1 > 1 , split the first maximizer in ( r , x , y , z ) (r,x,y,z) ( r , x , y , z ) ; these fallback scores never authorize recognition. Otherwise use
2 ρ t ( mid Q ) t , 2 ρ r ( mid Q ) r , s x ( P ) , s y ( P ) , s z ( P ) , \frac{2\rho_t}{(\operatorname{mid}Q)_t},\qquad
\frac{2\rho_r}{(\operatorname{mid}Q)_r},\qquad
s_x(P),\ s_y(P),\ s_z(P), ( mid Q ) t 2 ρ t , ( mid Q ) r 2 ρ r , s x ( P ) , s y ( P ) , s z ( P ) ,
with the last three scores from Lemma A.13 . Bisect the first maximizer in ( t , r , x , y , z ) (t,r,x,y,z) ( t , r , x , y , z ) at an interior binary midpoint; both closed children inherit G G G . Retain as unresolved if no such midpoint exists or depth is 120. Newton preserves depth; never subdivide G G G .
The comparison invariant and finite outcome. Let R \mathcal{R} R be the union of recognized products Q × G Q \times G Q × G . Retain
K ⊂ R ∪ ⋃ ( Q , G ) ∈ Q ∪ U ( Q × G ) . \mathcal{K} \subset\mathcal{R} \cup\bigcup_{(Q,G)\in\mathcal{Q}\cup\mathcal{U}} (Q \times G). K ⊂ R ∪ ( Q , G ) ∈ Q ∪ U ⋃ ( Q × G ) .
Initial inclusion is Lemma 6.9 ; Lemma A.13 , (L4)–(L17) and (A.4) preserve it, including
Q − ∪ Q + = Q , G − = G + = G Q^{-}\cup Q^{+}=Q,\qquad G^{-}=G^{+}=G Q − ∪ Q + = Q , G − = G + = G
under closed bisection. Recognition gives
( η ( Λ ) , 2 V 2 ) ∈ K ∩ R ⟹ Λ ≅ H Y ( V ) , J ( Λ ) = 0. (\eta(\Lambda),2V^{2})\in\mathcal{K}\cap\mathcal{R}\Longrightarrow\Lambda\cong H_Y(V),\qquad J(\Lambda)=0. ( η ( Λ ) , 2 V 2 ) ∈ K ∩ R ⟹ Λ ≅ H Y ( V ) , J ( Λ ) = 0.
The endpoint half-line identities give all exterior equality cases. The finite record is
( # r e m o v a l s , # b i s e c t i o n s , # r e t a i n e d N e w t o n ) = ( 107013 , 52350 , 2621 ) , (\#\mathrm{removals},\#\mathrm{bisections},\#\mathrm{retained\ Newton})=(107013,52350,2621), ( # removals , # bisections , # retained Newton ) = ( 107013 , 52350 , 2621 ) ,
# d o m a i n - e x c l u d e d c h i l d r e n b e f o r e i n s e r t i o n = 309 , \#\mathrm{domain\text{-}excluded\ children\ before\ insertion}=309, # domain - excluded children before insertion = 309 ,
( # v a l u e , # g r a d i e n t ) = ( 27970 , 17423 ) , (\#\mathrm{value},\#\mathrm{gradient})=(27970,17423), ( # value , # gradient ) = ( 27970 , 17423 ) ,
( # N e w t o n e x c l u s i o n s , # m o m e n t ) = ( 456 , 4046 ) , (\#\mathrm{Newton\ exclusions},\#\mathrm{moment})=(456,4046), ( # Newton exclusions , # moment ) = ( 456 , 4046 ) ,
( # h e i g h t , # d y n a m i c t r a n s v e r s e , # s t a t i c F o u r i e r ) = ( 232 , 5 , 23 ) , (\#\mathrm{height},\#\mathrm{dynamic\ transverse},\#\mathrm{static\ Fourier})=(232,5,23), ( # height , # dynamic transverse , # static Fourier ) = ( 232 , 5 , 23 ) ,
( # a l i g n m e n t e x c l u s i o n s , # a l i g n m e n t c o n t r a c t i o n s ) = ( 1887 , 61 ) . (\#\mathrm{alignment\ exclusions},\#\mathrm{alignment\ contractions})=(1887,61). ( # alignment exclusions , # alignment contractions ) = ( 1887 , 61 ) .
The 61 alignment contractions retain y = z = 0 y=z=0 y = z = 0 and create no children. Thus
Q = U = ∅ ⟹ { J ( Λ ) ≥ 0 ( ∣ Λ ∣ = 1 ) , J ( Λ ) = 0 ⟹ ∃ V ∈ [ V 0 , V 1 ] : Λ ≅ H Y ( V ) . \mathcal{Q}=\mathcal{U}=\varnothing\Longrightarrow
\begin{cases}
J(\Lambda)\ge0 & (|\Lambda|=1),\\
J(\Lambda)=0\Longrightarrow\exists V\in[V_0,V_1]:\ \Lambda\cong H_Y(V).
\end{cases} Q = U = ∅ ⟹ { J ( Λ ) ≥ 0 J ( Λ ) = 0 ⟹ ∃ V ∈ [ V 0 , V 1 ] : Λ ≅ H Y ( V ) . ( ∣Λ∣ = 1 ) ,
Proposition 5.1 gives the comparison for every V V V . This completes the proof of Numerical Lemma A.17 . □
Acknowledgements. The research of S. Luo is partially supported by the National Natural Science Foundation of China (NSFC) under Grant Nos. 12261045 and 12001253, and by the Jiangxi Jieqing Fund under Grant No. 20242BAB23001. The research of J. Wei is partially supported by the General Research Fund (GRF) of Hong Kong ”New frontiers in singularity formation of nonlinear partial differential equations”. The authors used AI to assist the long but finite computations.
Statements and Declarations: there is no conflict of interest.
Data availability: the manuscript has no associated data.
References
(S. Luo) School of Mathematics and Statistics, Jiangxi Normal University, Nanchang, 330022, China
(J. Wei) Department of Mathematics, Chinese University of Hong Kong, Shatin, NT, Hong Kong
Email address , S. Luo: luosp1989@163.com
Email address , J. Wei: wei@math.cuhk.edu.hk