Escolar Documentos
Profissional Documentos
Cultura Documentos
CARLOS C. ARANDA
2. A review on statistics
In this section, we recall some basics on mathematical statistics. Our main goal
is to join theories that usually are studied in a no related approach.
Definition 2.1. The characteristic S1 is subordinate to S2 if S1 is a measurable
function of S2 : S1 = ϕ(S2 ). In the σ-algebra representation the statistic S1 is
subordinate to S2 if σ(S1 ) ⊆ σ(S2 ), where σ(S) is the σ-algebra generated by the
measurable function S : (Ω, σ, Pθ ) → R. We assume that there exists a measure
P such that dPdP the Radon Nykodim derivative is well defined for all θ in the
θ
parameter space Θ.
Definition 2.2. If S1 is subordinate to S2 and S2 is subordinate to S1 , we denom-
inate this statistics equivalents.
Definition 2.3 (Fisher [8], Neyman [20] and Zacks [25]). The statistic S is suf-
ficient if for any measurable function T : (Ω, σ) → R, the conditional distribution
Pθ (T ∈ B|σ(S)) is independent of θ.
Definition 2.4. The sufficient statistic S0 is a minimal statistic if it is subordinate
to any S sufficient statistic.
fθ
Theorem 2.5 (Lehmann-Scheffe [14]). The σ-algebra U = σ(r(x, θ) ≡ fP (x) =
dPθ
dP (x), θ ∈ Θ) is a minimal sufficient σ-algebra.
Remark 2.6. If we define S(x) such that S(x) = S(x0 ) if and only if the relation
h(x, x0 ) = ffθθ(x
(x)
0)
no depend on θ. Then S is a sufficient minimal statistic.
R
Theorem 2.7. (Blackwell [3], Rao [21]) If we assume Eθ (T ) = T (x)fθ (x)dP = θ,
θ ∈ Θ. Then
Z Z
(E(T |S)(x) − θ)2 fθ (x)dP ≤ (T (x) − θ)2 fθ (x)dP,
for all θ ∈ Θ.
3. A priori estimates
This section is concerned with the study of a priori estimates for the equation
− ∆u = f (x, u, ∇u), u(x) ≥ 0, x ∈ Ω ⊂ RN , (3.1)
and the fully coupled systems
−∆u = f1 (x, u, v),
−∆v = f2 (x, u, v),
u(x), v(x) ≥ 0, x ∈ Ω ⊂ RN ,
where Ω is a open set in RN , N > 2. A priori estimate means: we obtain bounds
independent of any given solution and any given boundary condition. This is central
because in Schrödinger equations we have no boundary and therefore for a statistical
analysis we need this class of a priori estimates. Another central topic in this article
is the derivation of weakly coupled Harnack inequalities for fully coupled systems.
In [7] it is stated that
EJDE-2012/60 SHORT-TITLE?? 3
has not “black holes” solutions. Black holes solutions means that the gravitational
potential of the cluster behaves like 1r (r = |x|) near the center. In 1972 Peebles
gives for the first time a derivation of the steady state distribution of the star near
a massive collapsed object, the same year Peebles motivated the observer and theo-
retician with the title of his paper “Black holes are where you find them”([15]-[16].
The question of the existence of black hole in a globular cluster is still open (1995).
Core collapse does occur, for instance using Hubble Space Telescope, Bendinelli
documented the first detection of a collapsed cluster in M31 [2]
EJDE-2012/60 SHORT-TITLE?? 5
Tous ces corps devenus invisibles son à le même place où ils on été
observés, puisqu’ils n’ en ont point changé, durant leur apparation
il existe donc dans les spaces célestes, des corps obscurs aussi con-
sidérables et peut être en aussi grand nombre que les etoiles. Un
astre lumineux de meme densite que la terre, et dont le diametre
serait deux cents cinquante fois plus grand que celui du soleil, ne
laisserant en vertu de son attraction parvenr aucun de ses rayons
jusqu’à nous, il es done possible que les plus grands corps lumineux
de l’univers, soient par cela même invisibles. Une étoile qui sans
être de cette grandeur, surpaserait considerablement le soileil, af-
faiblirait sensiblemente la vı̂tesse de la lumiere et augmenterait ainsi
l’étendue de la lumière.
This quotation belongs to Pierre Simon de Laplace in Exposition du systeme du
Monde 1796, second volume p348.
Harnack inequality for weak solutions is a classical property in the study of linear
elliptic equations [19]. For nonlinear elliptic equations, we have
Theorem 3.8 (Schoen [23]). For N ≥ 3, let B3R be a ball of radius 3R in RN ,
and let u ∈ C 2 (B3R ) be a positive solution of
N +2
−∆u = N (N − 2)u N −2 .
Then
(max u)(min u) ≤ C(N )R2−N .
BR B2R
provided R ≤ R0 . If the terms up−1 and 1 are dropped in (3.9), then (3.12) holds
with C = C(N, p, r, Λ) and with no further restriction on R.
6 C. C. ARANDA EJDE-2012/60
The above theorem is valid for f (u) ≡ 0. This statement of classical Harnack
inequality is different from the usual ones where d ∈ L∞ (Ω) and generalizes the
usual Harnack inequality for homogeneous elliptic equations for the nonhomoge-
neous situation. The reader can see that in [11]. With the Moser iteration technique
developed in this paper, we obtain a priori estimates for a class of fully coupled
systems:
Theorem 3.12. If u,v are W 1,2 (Ω) non-negative solutions of
−∆u = d1 u + f1,1 (u) + f1,2 (v) + g1 in Ω ⊂ RN , (3.14)
N
−∆v = d2 v + f2,1 (u) + f2,2 (v) + g2 in Ω ⊂ R , , (3.15)
where:
N
(i) C1,i,j ur ≤ fi,j (u) ≤ C2,i,j ur for all i, j = 1, 2 and 1 < r < N −2 .
q
(ii) 0 ≤ gi , di ∈ L 2 (Ω), for all i = 1, 2 and q > N .
Then for any ball B3R (y) ⊂ Ω,
sup {u + v + 1} ≤ H(R), (3.16)
BR (y)
N
(i) C1,i,j ur ≤ fi,j (u) ≤ C2,i,j ur for all i, j = 1, 2 . . . n and 1 < r < N −2 .
q
(ii) 0 ≤ gi , di ∈ L 2 (Ω), for all i = 1, 2 and q > N .
Then for any ball B3R (y) ⊂ Ω
n
X
sup {1 + ui } ≤ H(R),
BR (y) i=1
To our knowledge, for fully coupled elliptic systems, one of the most representa-
tive results on Harnack inequality is as follows.
Theorem 3.14 (Busca and Sirakov [4]). Let us consider the problem
−∆u = f1 (u, v) in Ω, (3.18)
−∆v = f2 (u, v) in Ω. (3.19)
Assume f1 (u, v), f2 (u, v) are globally Lipschitz continuous functions, with Lipschitz
constant A, which satisfy the cooperativeness assumption:
∂f1 ∂f2
≥ 0, ≥ 0.
∂v ∂u
Let (u, v) be a nonnegative solution of (3.18), 3.19 in Ω. We suppose that the system
is fully coupled, in the sense that f1 (0, v) > 0 for all v > 0, and f2 (u, 0) > 0 for
u > 0. Then for any compact subset K of Ω there exists a function Φ(t) (depending
on A, K and Ω), continuous on [0, ∞), such that Φ(0) = 0 and
sup max{u, v} ≤ Φ( inf min{u, v}). (3.20)
x∈K
We call (3.23), (3.24) a weakly coupled Harnack inequalities because the coupling
is located in C1 (R) and C2 (R). This is a new phenomenon: the fully coupled
equation (3.21), 3.22 satisfies a weakly coupled Harnack inequalities. A nice survey
is [9] for more details on elliptics systems, critical hyperbola, hamiltonian elliptic
systems etc.
Using the integral relations method to the problem of local properties of weak
solutions (see [11, Theorem 8.17 section 8.6]), we have the following result.
Theorem 3.16 (Morrey [18], Stampacchia [24]). If u is a W 1,2 (Ω) solution of the
problem
− ∆u + cu = g in Ω, (3.25)
q
where c ∈ L∞ (Ω), g ∈ L 2 (Ω) for some q > N . Then we have, for any ball
B2R (y) ⊂ Ω and p > 1:
N 2N
sup u ≤ C(R− p ku+ kLp (B2R (y)) + R2− q kgkL q2 (Ω) ), (3.26)
BR (y)
Z 1/p R
sup {u + 1} ≤ |u + 1|p dx CR−N/(2p) (1 + )N/(2p) , (3.28)
BR (y) B 3R (y) 2
2
where
N
∞
X log max{Csq−N , Cs }
C = C(p, N ) exp ,
s=0
χs p
(χs p)2 16 1/2
Cs = 2 max{ max{k b̂ s k q , }, 1} ,
2|χs p − 1| L 2 (B(1+2−s−1 )R ) |χs p − 1|
4C 2
b̂s = |c| + |g| + s + C0 ,
|χ p|
and p ∈ R>0 − {χ−s }∞ N
s=0 , χ = N −2 . Moreover for p < 0, with the same constants,
Z 1/p R
inf {u + 1} ≥ |u + 1|p dx CR−N/(2p) (1 + )N/(2p) . (3.29)
BR B 3R (y) 2
2
Our main results - Theorems 3.5, 3.11, 3.12 and 3.15 - follow by combining
Theorem 3.17 and Lemma 3.4. In section 5, we collect some preliminary results:
we recall a lemma essentially proved in [1] and we state variants of classical Lemmas
by Morrey and Trudinger (see [11]).
EJDE-2012/60 SHORT-TITLE?? 9
4.1. Quantum mechanics and minimal statistics. We present here the con-
nection on quantum mechanics and minimal statistics. First at all, we note that
Z nZ uσ (x1 , x2 , x3 ) o
Eσ (X1 ) = x1 R dx 2 dx 3 dx1 = θ1 (σ),
R
T
Ω X1 =x1 u (x)dx
Ω σ
Z nZ uσ (x1 , x2 , x3 ) o
Eσ (X2 ) = x2 R dx 1 dx 3 dx2 = θ1 (σ),
R
T
Ω X2 =x2 u (x)dx
Ω σ
Z nZ uσ (x1 , x2 , x3 ) o
Eσ (X3 ) = x3 R dx1 dx2 dx3 = θ3 (σ).
R
T
Ω X3 =x3 u (x)dx
Ω σ
Therefore, we use Theorems 2.5, 2.7 and Remark 2.6 to obtain unbiased estimators
of minimum variance. This connection allow to use all the classical mathemati-
cal statistical procedure like the Bayesian approach on the parameter space. Of
fundamental importance is our a priori estimates without dependence on the dis-
tance to the boundary. In fact this provides a law of nature on the energy level on
non-relativistic quantum mechanics.
10 C. C. ARANDA EJDE-2012/60
Therefore,
Z Z Z
k β+r k β−1
C1 η u dx + |β| η u |∇u| dx ≤ k η k−1 uβ ∇η · ∇u dx.
2
(5.1)
Observe that
2k
uβ ∇η k · ∇u ≤ η k |∇u|uβ .
Rη
EJDE-2012/60 SHORT-TITLE?? 11
|β| −1 4k 2
Z Z Z
η k uβ−1 |∇u|2 dx + C1 η k uβ+r ≤ η k−2 uβ+1 dx. (5.2)
2 R2
2γ
With the assumption γ > r − 1, we fix β = γ − r and k = r−1 . Therefore
2γ γ−r+1
η k−2 uβ+1 = η r−1 −2 uγ−r+1 = η 2 r−1 uγ−r+1 .
1 q
γ
Using Young’s inequality ab ≤ 0 aq + 01−q b q−1 with q = γ−r+1 (and consequently
q γ
q−1 = r−1 ), we have
γ−r+1 2γ 1 2γ
R−2 η k−2 uβ+1 = η 2 r−1 uγ−r+1 ≤ 0 η r−1 uγ + 01−q R− r−1 .
It follows that
Z Z 1
Z
2γ 2γ
R−2 η k−2 uβ+1 dx ≤ 0 η r−1 uγ dx + 01−q R− r−1 dx
B2R
Z 1 2γ
= 0 η k uγ dx + 01−q ωN RN − r−1 ,
Choosing
2 2γ 2 C1
−1 4k 2 0 = 4( ) 0 = .
r−γ r−1 2
We find
Z Z
C1
uγ dx ≤ η k uγ dx
2 BR
q
2γ
≤ 201−q ωN RN − r−1
(r − γ)(r − 1)2 C1 1−r
γ 2γ
= 2( ) ωN RN − r−1 .
26 γ 2
This completes the proof.
q > N . Then, for any ball BR1 (y) ⊂ BR2 (y) ⊂ Ω and β 6= −1,
(β+1)/2
ku + 1k N (β+1)
L N −2 (BR1 )
(5.4)
n1 N
N −q
√ 1 o
(β+1)/2
≤ max{1 , 2}(1 + ) × (ku + 1kLβ+1 (BR2 ) ) ,
1 R2 − R1
where
1
p
2 C(N )
1 = r . (5.5)
(β+1)2 2
max 2(|β|) max{k|c| + |g| + 4C 16
|β| + C k q ,
0 L 2 (B ) |β| }, 1
R2
2
|∇w|2 dx for all
R
and C(N ) is the Sobolev embedding constant (C(N )kwk 2N ≤
N −2
w ∈ H01 (Ω)).
Proof. We define û = u + 1. We use a classical test function η 2 ûβ where η satisfies
1
for 0 < R1 < R2 , η ≡ 1 in BR1 , η ≡ 0 in Ω − BR2 with |∇η| ≤ R2 −R 1
2C 2
Z Z
+ η 2 ûβ+1 dx + C0 η 2 ûβ+1 .dx
Then we obtain
2C 2
Z Z
8
(|β| − ) |∇u|2 η 2 ûβ−1 dx ≤ + C0 )η 2 + |∇η|2 ûβ+1 dx. (5.9)
(|g| + |c| +
β+1
Now, if we set w = û 2 , we obtain
β + 1 β−1
∇w = û 2 ∇u,
2
(β + 1)2 β−1
|η∇w|2 = û |∇u|2 η 2 .
4
It follows from (5.9) that
(β + 1)2 2C 2
Z Z
8
|η∇w|2 dx ≤ + C0 )η 2 + |∇η|2 w2 dx.
(|g| + |c| + (5.10)
4(|β| − )
If we set
2C 2
b̂ ≡ |g| + |c| + + C0 ,
applying the Hölder inequality,
Z
b̂(ηw)2 ≤ kb̂kL q2 (B k(ηw)2 k q−2
q = kb̂kL q2 (B kηwk22q .
R2 ) R2 ) q−2
We arrive at
(β + 1)2
Z Z
8
|η∇w|2 dx ≤ kb̂kL q2 (B ) kηwk22q + |∇η|2 w2 dx.
4(|β| − ) R2 q−2
At this point we use the interpolation inequality for Lp norms: If p ≤ s ≤ r then
kuks ≤ 1 kukr + −µ
1 kukp , where
1 1
p − s
µ= 1 1 .
s − r
2q 2N
The condition q > N implies 2 < q−2 < N −2 , therefore
N
kηwk 2q ≤ 1 kηwk 2N + 1N −q kηwk2 .
q−2 N −2
So, we have
Z
|η∇w|2 dx
(5.11)
(β + 1)2
Z
N
8
≤ {kb̂kL q2 (B ) (1 kηwk 2N + 1N −q kηwk2 )2 + |∇η|2 w2 dx}.
4(|β| − ) R2 N −2
By the Sobolev inequality,
Z Z
2 2
C(N )kηwk 2N ≤ |η∇w| dx + |w∇η|2 dx.
N −2
14 C. C. ARANDA EJDE-2012/60
We deduce that
(β + 1)2
C(N )kηwk22N ≤ {kb̂kL q2 (B ) (1 kηwk 2N
N −2 4(|β| − ) R2 N −2
Z Z
N
8
+ 1N −q kηwk2 )2 + |∇η|2 w2 dx} + |w∇η|2
(5.12)
(β + 1)2 8
≤ max max{kb̂kL q2 (B ) , }, 1
4(|β| − ) R2
N
Z
× (1 kηwk 2N + 1 kηwk2 ) + 2 |w∇η|2 .
2
N −q
N −2
2N Z 1/2
≤ 1 kηwk 2N + 1N −q kηwk22 + 2 |w∇η|2 dx
N −2
q Z 1/2
2N
≤ 1 kηwk 2N + max(1N −q , 2) kηwk22 + |w∇η|2 dx
N −2
N √ Z 1/2
= 1 kηwk 2N + max(1 , 2) kηwk2 + |w∇η|2 dx
N −q 2
.
N −2
If p
1
2 C(N )
1 = r ,
(β+1)2 16)
max{ 2(|β|) max{kb̂k q
L2 (BR2 ) |β|
}, 1}
then
s
1p (β + 1)2 16
C(N )kηwk 2N ≤ max max{kb̂kL q2 (B ) , }, 1
2 N −2 2(|β|) R2 |β|
N √ Z 2 2 1/2
× max{1N −q , 2} w (η + |∇η|2 dx) .
We are led to
1 N √ 1
kwk 2N ≤ max{1N −q , 2} 1 + kwkL2 (BR2 ) .
L N −2 (BR1 ) 1 R2 − R1
EJDE-2012/60 SHORT-TITLE?? 15
Finally, using
Z β+1 2N
(N −2)/(2N )
kwk 2N = û 2 N −2 dx
L N −2 (BR1 ) BR1
n Z −2 o β+1
NN(β+1)
(β+1)N 2 β+1
= û N −2 dx = kûk 2
N (β+1) ,
BR1 L N −2 (BR1 )
and
β+1
kwkL2 (BR2 ) = kûkLβ+1
2
(BR )
.
2
Using (6.1),
C(χm−1 p)(1 + 2−1 R) 1
χm−1 p
Φ(χm p, R) ≤ Φ(χm−1 p, (1 + 2−m )R)
(1 + 2−m )R − R
C(χm−1 p)(1 + 2−1 R) m−1
1
n m−1
Y 1
o 3R
× (C(χs p)) χs p Φ(p, ).
s=0
2
Now
m−1 n m−1
X log C(χs p)) o
Y 1
s χs p
(C(χ p)) = exp .
s=0 s=0
χs p
Therefore, we study the convergence of
∞ ∞ log 1
√ N
X log C(χs p)) X 1 (χs p) max{ 2, (1 (χs p)) N −q }
=
s=0
χs p s=0
χs p
√ q
(6.2)
∞
log max{ 2(1 (χs p))−1 , ((1 (χs p))−1 ) q−N }
X
= ,
s=0
χs p
where
1
p
s 2 C(N )
1 (χ p) = r .
(χs p)2
max{ 2(χs p−1) max{kb̂kL q2 (B , 16 }, 1}
) χs p−1
(1+2−s−1 )R
r
max{max{kb̂kL q2 (B , 16 }, C(p, χ)}
) p−1
√ 3R
2
≤ C(p, χ) χs p 1
p .
2 C(N )
From this inequality, we deduce, using the integral test, the convergence of the
series (6.2). Finally,
sup{u + 1}
BR
= lim Φ(χm p, R)
m→∞
3R n m 1
Pm−2 s+2 −1 Pm−1 1
Pm−1 1
o
≤ Φ(p, ) lim 2 χm−1 p 2 p s=0 χs R p s=0 χs (1 + 2−1 R) j=0 χj p
2 m→∞
m−1
Y 1
× (C(χs p)) χs p
s=0
3R R
= Φ(p, )CR−N/(2p) (1 + )N/(2p) ,
2 2
where
P∞ ∞
1 s+2 Y s
C = C(|c|, |g|, C, C0 , N, p, q) = 2 p s=0 χs (C(χs p))1/χ p ,
s=0
In a similar manner, we can prove (3.29).
Proof of Theorem 3.5. If a non-negative function u ∈ W 1,2 (Ω) solves the equation
−∆u = du + f (u) + G(∇u) + g in Ω ⊂ RN ,
with the conditions in Theorem 3.5, then −∆u ≥ C1 ur . We set c = d + f (u)/u.
Now, we apply Theorem 3.17. From (3.28) and (3.3), we deduce
Z 1/p R
sup {u + 1} ≤ |u + 1|p dx CR−N/(2p) (1 + )N/(2p)
BR (y) B 3R (y) 2
2
1/p 3R N/p R
≤ kukLp (B 3R (y)) + ωN ( ) CR−N/(2p) (1 + )N/(2p)
2 2 2
N
− 2 1/p 3R N/p
−N/(2p) R
≤ C(N, p, r, C1 )R p r−1 + ωN ( ) CR (1 + )N/(2p) ,
2 2
where max{r − 1, 1} < p < r and C = C(|c|, |g|, C, C0 , N, p, q). Condition (iii) in
Theorem 3.5 implies
f (u)
k k q/2 ≤ C2 kur−1 kLq/2 (B(1+2−s−1 )R )
u L (B(1+2−s−1 )R )
1 (6.3)
= C2 kuk r−1
(r−1)q .
L 2 (B(1+2−s−1 )R )
To use Lemma 3.4 and Theorem 3.17, we need to satisfy the conditions r − 1 <
(r−1)q/2 < r, N/2 ≤ q/2. Moreover, to get simple statements, we set an additional
condition 1 < (r − 1)q/2. For 1 < r < N/(N − 2) there exists q(r) satisfying all the
required restrictions. We deduce
N 2
1
q(r)
− r−1 1
kuk r−1
(r−1)q(r) ≤ C(r, N, C1 )(R (r−1) 2 ) r−1 . (6.4)
L 2 (B(1+2−s−1 )R )
18 C. C. ARANDA EJDE-2012/60
For the demonstration of this affirmation, see for example the proof of [5, Lemma
1.36]. Thus
Z 1/p0 Z Z 1/p0
{u + 1}p0 dx = {u + 1}p0 dx {u + 1}−p0 dx
BR (y) BR (y) BR (y)
Z −1/p0
× {u + 1}−p0 dx
BR (y)
Z −1/p0
≥ C(R, p0 ) {u + 1}−p0 dx .
BR (y)
Collecting this inequality with (3.29) and (3.28) in Theorem 3.17, we derive the
conclusion of Theorem 3.11.
Proof of Theorem 3.12. From equations (3.14) (3.15), we compute
2
X fi,1 (u) fi,2 (v)
−∆(u + v) ≤ (d1 + d2 )(u + v) + + (u + v) + (g1 + g2 )
i=1
u v
≤ (d1 + d2 )(u + v) + C(ur−1 + v r−1 )(u + v) + (g1 + g2 ).
From −∆u ≥ C0,1,1 ur , −∆v ≥ C0,2,2 v r using Lemma 3.4 and with the same method
of proof of Theorem 3.5, we obtain the desired result.
Proof of Theorem 3.15. By Theorem 3.12, if the non-negative pair (u, v) solves
equations (3.21) and (3.22), then (u, v) ∈ L∞
loc (Ω), where there are not dependence
in the local bound on (u, v). Therefore the result follows from Theorem 3.11.
Acknowledgments. The author would like to express his gratitude with the Em-
bassy of France at Ottawa and Revenu Quebec for their kind support. The author
would like to thank the anonymous referees for their useful suggestions.
References
[1] M. F. Bidaut Veron, S. Pohozaev; Nonexistence results and estimates for some nonlinear
elliptic problems. J. Anal. Math., 84 (2001), 1-49.
[2] O. Bendinelli, C. Cacciari, S. Djorgorski, L. Federici, .. Ferraro, Fi Fusi Pecci, G. Parmeggiani,
N. Weir, E Zaratti; The first detection of a collapsed core globular cluster in M31. Astrophy.
I., 409 (1993), LI7-LI9.
[3] D. Blackwell; Conditional expectation and unbiased sequential estimation. Ann. Math.
Statist., 1947 18, 105-110.
[4] J. Busca, B. Sirakov; Harnack type estimates for nonlinear elliptic systems and applications.
Ann. Inst. H. Poincar Anal. Non Linaire 21 (2004), no. 5, 543–590.
[5] Y. Z. Chen, L. C. Wu; Second order elliptic equations. Translations of mathematical mono-
graphs, vol. 174 AMS.
[6] E. N. Dancer; Superlinear problems on domains with holes of asymptotic shape and exterior
problems. Math. Z., 229 (1998), 475-491.
[7] L. D. Landau, E. M. Lifshitz; Quantum mechanics non-relativistic theory. Volume 3 of Course
of Theoretical Physics, third edition Pergamon Press.
EJDE-2012/60 SHORT-TITLE?? 19
[8] R. A. Fisher; On the mathematical fundations of theoretical statistics. Phil. Trans. Roy. Soc.
A., 1922, 222, 309-368.
[9] Bernhard Ruf; Superlinear Elliptic Equations and Systems. Dipartimento de Matematica,
Universita degli Studi di Milano, May 23, 2007.
[10] B. Gidas, J. Spruck; Global and local behavior of positive solutions of nonlinear elliptic
equations. Comm. Pure Appl. Math., 34 (1981), 525-598.
[11] D. Gilbarg, N. Trudinger; Elliptic partial differential equations of second order. Second edi-
tion 1983-Springer-Verlag.
[12] M. Hénon; Numerical experiments on the stability of spherical stellar systems. Astronomy
and Astrophysics 24 (1973), 229-238.
[13] O. Ladyzhenskaya, N. Ural’tseva; Linear and quasilinear elliptic equations. 1968 Academic
Press.
[14] E. L. Lehmann, H. Scheffe; Completeness, similar regions and unbiased estimation. Pt. I.
Sankhya, 1950, 10 1950, 10, 305-340.
[15] P.I.E. Peebles; Black holes are where you find them. Gen. Rel. Grav., 3 (1972), 63-82.
[16] P.I.E. Peebles; Star distribution near a collapsed object. Ast. I., 178 (1972), 371-375.
[17] T. Matukuma; Sur la dynamique des amas globulaires stellaires. Proc. Imp. Acad. 6 (1930),
133-136.
[18] C. B. Morrey; Second order elliptic equations in several variables and Hölder continuity.
Math. Z. 72, 146-164 (1959).
[19] J. Moser; On Harnack Theorem for elliptic differential equations. Comm. Pure Appl. Math.
14 577-591 (1961).
[20] J. Neyman; Su un teorema concernente le cossidette statistiche sufficienti. Inst. Ital. Atti.
Giorn. (1935), 6 320-334.
[21] J. Rao; Sufficient statistics and minimum variances estimates. Proc. Cambr. Phill. Soc. 1949,
45, 213-218.
[22] J. Serrin, H. Zou; Cauchy-Liouville and universal boundedness theorems for quasilinear el-
liptic equations and inequalities. Acta Math.,189 (2002),79-142.
[23] R. Schoen; Courses at Stanford University, 1988, and New York University, 1989.
[24] G. Stampacchia; Problemi al contorno ellitici, con dati discontinui, dotati di soluzioni
hölderiane. Ann. Mat. Pura Appl. (4) 51, 1-37 (1960).
[25] S. Zacks; The theory of statistical inference. Wiley, N. Y., 1971.