ON GEOMETRIC ESTIMATES FOR SOME PROBLEMS ARISING

FROM MODELING PULL-IN VOLTAGE IN MEMS

DURVUDKHAN SURAGAN AND DONGMING WEI

Abstract. In this paper for all p > 1 we prove that the pull-in voltage of the
p-MEMS (micro-electro mechanical systems) problems in a smooth bounded do-
main of Rd , d ≥ 1, is minimized by symmetrizing the domain and the permittivity
arXiv:1801.08199v1 [math.AP] 23 Jan 2018

profile. The singular nonlinearities belong to a very general class and the obtained
results are new even for p = 2. Main ingredient of our proof is based on Talenti’s
comparison principle. Although there is no physical applications for the cases when
d ≥ 3, we nevertheless demonstrate our method to the multidimensional MEMS
problems in the whole space Rd , d ≥ 3, and the Dirichlet boundary value problems
of second order uniformly elliptic differential operators.

1. Introduction
Let us recall the second order differential equation with the singular nonlinearity
modeling stationary MEMS (micro-electro mechanical systems):
 f (x) d
−∆u(x) = λ (1−u(x)) 2 , 0 ≤ u(x) < 1, x ∈ Ω ⊂ R , d ≥ 1
(1.1)
u(x) = 0, x ∈ ∂Ω.
Here f describes the varying permittivity profile of the elastic membrane with
f ∈ C α (Ω̄) for some α ∈ (0, 1], 0 ≤ f ≤ 1, and f 6≡ 0. The Dirichlet pull-in voltage
is defined as
λ∗ (Ω, f ) = sup{λ > 0 | (1.1) possesses at least one classical solution}.
For the Dirichlet pull-in voltage the following inequality holds:
Theorem 1.1 (Proposition 2.2.1 [8]). A ball B (with the symmetrized permittivity
profile) is a minimizer of the Dirichlet pull-in voltage among all domains of given
volume, i.e.
λ∗ (B, f ∗ ) ≤ λ∗ (Ω, f )
for an arbitrary domain Ω ⊂ Rd with |Ω| = |B|, where | · | is the Lebesgue measure
in Rd , d ≥ 1. Here f ∗ is the symmetric decreasing rearrangement of f .
In this paper we consider a generalized stationary MEMS problem, so called p-
MEMS equation (see [3]):

1991 Mathematics Subject Classification. 47J10, 35J60.

Key words and phrases. p-MEMS problem, pull-in voltage, Talenti’s comparison principle, geo-
metric estimate.
The first author was supported by the MESRF (the Agreement number 02.a03.21.0008) and the
MESRK target program BR05236656. No new data was collected or generated during the course of
this research.

1
2 DURVUDKHAN SURAGAN AND DONGMING WEI


−∆p u(x) = −div(|∇u|p−2∇u) = λf (x)g(u), 0 ≤ u(x) < 1, x ∈ Ω ⊂ Rd ,
(1.2)
u(x) = 0, x ∈ ∂Ω.
where λ > 0 and f is a smooth positive function. The non-linearity g(u) is a non-
decreasing, positive function defined on [0, 1) with a singularity at u = 1:
(1.3) lim g(u) = +∞.
u→1−

We also assume that (λf (x)g(u))∗ = λf ∗ (x)g(u∗ ), where by ∗ we denote the sym-
metric decreasing rearrangement. For example, g(u) = (1 − u)−α, α > 0, satisfies the
all above assumptions (see (2.3)). Note that in the case d = 2, (1.2) is applicable for
some nonlinear material pull-in applications. This is motivated by considering the
membrane equation for nonlinear materials with constitutive stress-strain equation
in the form σ = |ǫ|p−2 ǫ subject to pull-in, where σ is the stress and ǫ is the strain.
Similarly, the p-MEMS pull-in voltage is defined as
λ∗p (Ω, f ) = sup{λ > 0 | (1.2) possesses at least one classical solution}.
In [4] Castorina, Esposito and Sciunzi proved that λ∗p < ∞ and for every λ ∈ (0, λ∗p )
there is a minimal (and semi-stable) solution uλ (i.e. uλ is the smallest positive
solution of (1.2) in a pointwise sense).
The main result of the present paper is: A ball B (with the symmetrized permittiv-
ity profile) is a minimizer of the Dirichlet p-MEMS pull-in voltage among all domains
of given volume, i.e.
λ∗p (B, f ∗ ) ≤ λ∗p (Ω, f ), 1 < p < ∞,
for an arbitrary smooth domain Ω ⊂ Rd , d ≥ 1, with |Ω| = |B|, where | · | is the
Lebesgue measure in Rd . Here f ∗ is the symmetric decreasing rearrangement of f .
To the best of our knowledge, the result is new even for the case p = 2 since we have
the general singular nonlinearity g(u), moreover, it gives a new proof for Theorem
1.1. Also we consider a similar stationary MEMS problem, but in infinity domain,
that is, on the whole Rd :
f (x)

−∆u(x) = λ (1−u(x)) 2, 0 ≤ u(x) < 1, x ∈ Rd , d ≥ 3,
(1.4)
u(x) −→ 0, |x| −→ ∞,
where λ > 0 and f satisfies the assumptions: supp f ⊂ Ω ⊂ Rd , f ∈ C α (Ω̄) for some
α ∈ (0, 1], 0 ≤ f ≤ 1, and f 6≡ 0. Although problem (1.4) has no physical meaning
for d ≥ 3, it appears to be a very rich source of interesting mathematical phenomena.
Note that one and two dimensional cases there is no solution of the Poison equation
which vanishes at infinity. Therefore, there is no direct extension of above problem
in the low dimensional cases.
To analyse the main difference of the problems (1.1) and (1.4) first we shall discuss
linear analogues of these problems. A linear analogue of Theorem 1.1 is so called
a Rayleigh-Faber-Krahn inequality. To recall it let us consider the minimization
problem of the first eigenvalue of the Laplacian with the Dirichlet boundary condition
(among domains of given volume):

−∆u(x) = λD u(x), x ∈ Ω ⊂ Rd ,
(1.5)
u(x) = 0, x ∈ ∂Ω.
 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 3

The famous Rayleigh-Faber-Krahn inequality asserts that

λD D
1 (B) ≤ λ1 (Ω),

for any Ω with |Ω| = |B|, where B ⊂ Rd is a ball and | · | is the Lebesgue measure
in Rd . In two-dimensional case, the Dirichlet eigenvalues describe the fundamental
modes of vibration of an idealized drum with a given shape. In general, beyond the
intrinsic interest of geometric extremum problems, these type of results produce a
priori bounds for spectral invariants of operators on arbitrary domains. For example,
in other words Rayleigh-Faber-Krahn inequality says that the operator norm of the
inverse operator to the Dirichlet Laplacian is maximised in a ball among all Euclidean
domains of a given volume. An analogue of the Rayleigh-Faber-Krahn inequality for
general convolution type integral operators were given in [17] (see also [16]).
Similarly, we can consider a linear version of the problem (1.4):
(1.6) − ∆u(x) = µu(x), x ∈ Ω ⊂ Rd ,
with the nonlocal integral boundary condition
1 ∂εd (x − y)
Z Z
∂u(y)
(1.7) − u(x) + u(y)dSy − εd (x − y) dSy = 0, x ∈ ∂Ω,
2 ∂Ω ∂ny ∂Ω ∂ny
where εd is the fundamental solution of the Laplacian and ∂n∂ y denotes the outer
normal derivative at a point y on the boundary ∂Ω. The spectral problem (1.6)-(1.7)
is equivalent (see [11]) to
Z
(1.8) u(x) = µ εd (x − y)u(y)dy, x ∈ Ω ⊂ Rd .
Ω
This also means
Proposition 1.2. The problem (1.4) is equivalent to the nonlinear integral problem
f (y)
Z
(1.9) u(x) = λ εd (x − y) dy, 0 ≤ u(x) < 1.
Ω (1 − u(y))2
These discussions hint that the main difference between the problems (1.1) and
(1.4) is non locality of (1.4) as well as it can be analyzed separately. Note that
the boundary condition (1.7) describes the long-known in theoretical physics effect
of a transparent boundary condition skipping the outgoing waves and reflecting the
incoming waves. Moreover, the presence of such boundary conditions of the volume
potential permits reducing of the problem with conditions in the infinite domain to
a problem in a bounded domain and effectively apply numerical methods. We refer
[15] for further discussions and for spectral theory of (1.8).
As another consequence of our method, in this paper we present similar geometric
estimate for the pull-in voltage (upper bound of the spectrum) of the (nonlinear)
Dirichlet boundary value problem for a second order uniformly elliptic differential
operator
d  
X ∂ ∂
L= ajk (x)
j,k=1
∂x j ∂xk
with ajk (x) = akj (x).
4 DURVUDKHAN SURAGAN AND DONGMING WEI

In Section 2 we briefly discuss some preliminary results, in particular, we recall

the celebrated Talenti comparison principle [18], which states that the symmetric
decreasing rearrangement (Schwarz rearrangement) of the Newtonian potential of a
charge distribution is pointwise smaller than the potential resulting from symmetriz-
ing the charge distribution itself. Talenti’s comparison principle can be also extended
to the Dirichlet p-Laplacian and the Dirichlet uniformly elliptic boundary value prob-
lems. Main results of this paper and their proofs will be given in Section 3. Talenti’s
comparison principle plays a key role in the proofs.

2. Preliminaries
Let Ω be a measurable bounded domain of Rd . An open ball (with origin 0) B is
called a symmetric rearrangement of Ω if |B| = |Ω| and
B = x ∈ Rd | σd |x|d < |Ω| ,


d
2π 2
where σd = Γ( d
)
is the surface area of the unit ball in Rd . Let u be a nonnegative
2
measurable function vanishing at infinity in the sense that all of its positive level sets
have a finite measure, i.e.
Vol ({x|u(x) > t}) < ∞, ∀t > 0.
To define a symmetric decreasing rearrangement of u one uses (see, for example [14])
the layer-cake decomposition, which expresses a nonnegative function u in terms of
its level sets in the following way
Z ∞
u(x) = χ{u(x)>t} dt,
0
where χ is the characteristic function. Let u be a nonnegative measurable function
vanishing at infinity. Then
Z ∞
∗
(2.1) u (x) = χ{u(x)>t}∗ dt
0

is called a symmetric decreasing rearrangement of the function u. Note that the

symmetric decreasing rearrangement is also sometimes called the Schwarz rearrange-
ment. The simple definition (2.1) can be useful in many proofs, for example, if
0 ≤ v(x) − u(x), ∀x ∈ Rd , then we directly get
Z ∞ Z ∞
∗
u (x) = χ{u(x)>t}∗ dt ≤ χ{v(x)>t}∗ dt = v ∗ (x), ∀x ∈ Rd .
0 0

That is, if
0 ≤ u(x) ≤ v(x), ∀x ∈ Rd ,
then
0 ≤ v ∗ (x) − u∗ (x), ∀x ∈ Rd .
Moreover, if 0 ≤ u < 1, 0 ≤ f , 0 < α and 0 < λ, then by using the Taylor expansion
and the property of symmetric decreasing rearrangements we have
 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 5

 ∗ ∞ Y
n
!∗
λf X α−k+1
(2.2) F ∗ (u, f ) := = λf un
(1 − u)α n=0 k=1
k
∞ Y
n
X α−k+1 λf ∗
= λf ∗ (u∗ )n = = F (u∗ , f ∗)
n=0 k=1
k (1 − u∗ )α
λf
That is, for the function F (u, f ) := (1−u)α
, 0 ≤ u < 1, we have the property

(2.3) F ∗ (u, f ) = F (u∗ , f ∗ ).

Note that when n = 0, the product in (2.2) is an empty product and has value 1.
Theorem 2.1. [Talenti’s comparison principle for the Laplacian ] Consider a (smooth)
nonnegative function f with suppf ⊂ Ω ⊂ Rd , d ≥ 3, for a bounded set Ω, and its
symmetric decreasing rearrangement f ∗ . If solutions u and v of
−∆u = f, −∆v = f ∗ ,
vanish at infinity, then
u∗ (x) ≤ v(x), ∀x ∈ Rd .
Note that u and v exist, and are uniquely determined by the equation, i.e.
Z
u(x) = εd (x − y)f (y)dy
Ω
and Z
v(x) = εd (x − y)f ∗(y)dy,
B
where εd (·) is the fundamental solution of the Laplacian, that is,
1
εd (x − y) = , d ≥ 3,
(d − 2)σd |x − y|d−2
and σd is the surface area of d-dimensional unit ball. They are nonnegative since
the fundamental solution is nonnegative. The inequality also holds for nonnegative
measurable functions f vanishing when |x| → ∞. It is also known that the fundamen-
tal solution σd , d ≥ 3, does not change its formula under the symmetric decreasing
rearrangement, see e.g. Lieb and Loss [14]. Talenti’s comparison principle can be
extended to the Dirichlet boundary value problems of uniformly elliptic second order
differential operators and p-Laplacian (see, e.g. Section 4.3 of [2] as well as [18] and
[19]).
Theorem 2.2. [Talenti’s comparison principle for the Dirichlet p-Laplacian] Let
1 < p < ∞. Consider a (smooth) nonnegative function f in a smooth bounded domain
Ω ⊂ Rd , d ≥ 1, and its symmetric decreasing rearrangement f ∗ . Then solutions u
and v of
−∆p u = f in Ω, u|∂Ω = 0,
and
−∆p v = f ∗ in B, v|∂B = 0,
6 DURVUDKHAN SURAGAN AND DONGMING WEI

satisfy
u∗(x) ≤ v(x), ∀x ∈ B.
Here B is a ball centered at the origin with |B| = |Ω|, where | · | is the Lebesgue
measure in Rd .

3. Main results
3.1. The pull-in voltage for the p-MEMS problem. We consider the pull-in
voltage for the p-MEMS problem with 1 < p < ∞:

−∆p u(x) = λf (x)g(u), 0 ≤ u(x) < 1, x ∈ Ω ⊂ Rd , d ≥ 1,
(3.1)
u(x) = 0, x ∈ ∂Ω.
where λ > 0 (is the applied voltage) and (the permittivity profile) f is a smooth
positive function. The non-linearity g(u) is a non-decreasing, positive function defined
on [0, 1) with a singularity at u = 1:
(3.2) lim g(u) = +∞.
u→1−

We also assume that (λf (x)g(u))∗ = λf ∗ (x)g(u∗), where ∗ is for the symmetric
decreasing rearrangement. For instance, g(u) = (1 − u)−α , α > 0, satisfies the all
above assumptions (see (2.3)). As usual, the p-MEMS pull-in voltage is defined as
λ∗p := λ∗p (Ω, f ) = sup{λ > 0 | (3.1) possesses at least one classical solution}.
The main result of this paper is
Theorem 3.1. A ball B (with f ∗ ) is a minimizer of the Dirichlet p-MEMS pull-in
voltage among all domains of given volume, i.e.
λ∗p (B, f ∗ ) ≤ λ∗p (Ω, f ), 1 < p < ∞,
for an arbitrary smooth domain Ω ⊂ Rd , d ≥ 1, with |Ω| = |B|, where | · | is the
Lebesgue measure in Rd .
Proof of Theorem 3.1. In [4] it was proved that λ∗p < ∞ and for every λ ∈ (0, λ∗p ) there
is a minimal (and semi-stable) solution uλ (i.e. uλ is the smallest positive solution of
(3.1) in a pointwise sense). Consider the following Picard iteration scheme
(3.3) − ∆p um (x) = λf (x)g(um−1), u0 (x) ≡ 0, m = 1, 2, . . . ,
with um = 0 on the boundary ∂Ω of the smooth bounded domain Ω. The sequence
converges uniformly to a positive solution uλ satisfying u ≥ uλ in Ω (see [4]), where
u is a positive solution of (3.1). Consider the following two sequences
(3.4) − ∆p un (x) = λf (x)g(un−1) in Ω, un (x)|∂Ω = 0, n = 1, 2, . . . ,
and
(3.5) − ∆p vn (x) = λf ∗ (|x|)g(vn−1 ) in B, vn (x)|∂B = 0, n = 1, 2, . . . ,
with u0 ≡ 0 and v0 ≡ 0. We have
−∆p u1 (x) = λf (x)g(0) in Ω, u1 (x)|∂Ω = 0,
and
−∆p v1 (x) = λf ∗ (|x|)g(0) in B, v1 (x)|∂B = 0.
 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 7

Here f ∗ is the symmetric decreasing rearrangement of the positive function f . There-

fore, by Talenti’s comparison principle for the Dirichlet p-Laplacian for 1 < p < ∞
(see Theorem 2.2) we obtain
(3.6) u∗1 (x) ≤ v1 (x), ∀x ∈ B,
where B is the ball centered at the origin with |B| = |Ω|. We also have
(3.7) − ∆p u2 = λf g(u1) in Ω, u2 |∂Ω = 0.
In addition, let us consider
−∆p ṽ2 (x) = λ(f g(u1))∗ in B, ṽ2 |∂B = 0,
that is,
(3.8) − ∆p ṽ2 (x) = λf ∗ g(u∗1) in B, ṽ2 |∂B = 0.
By Talenti’s comparison principle for (3.7) and (3.8) we obtain
u∗2 (x) ≤ ṽ2 (x), ∀x ∈ B.
By using (3.6) we get
−∆p ṽ2 (x) = λf ∗ g(u∗1) ≤ λf ∗ g(v1 ) = −∆p v2 (x),
in B, that is, by the comparison principle for the operator −∆p (see, e.g. [9]) we get
ṽ2 (x) ≤ v2 (x), ∀x ∈ B.
This gives
u∗2 (x) ≤ v2 (x), ∀x ∈ B.
Further, by repeating this process we arrive at u∗n ≤ vn in B for all n ≥ 0, that
is, max u∗n ≤ max vn for all n ≥ 0. Since max u∗n = max un , it means that for a
B B B Ω
given λ if {vn } converges, then the sequence {un } is also converges. This fact proves
λ∗p (B, f ∗ ) ≤ λ∗p (Ω, f ). 
Since the singular nonlinearity g belongs to a very general class Theorem 3.1 implies
new results even for the case p = 2. For example, it gives a new geometric pull-in
voltage estimate for the electrostatic MEMS problem with effects of Casimir force
(see [13]):
(  
1 σ
−∆u(x) = λ (1−u(x)) 2 + (1−u(x))4
, 0 ≤ u(x) < 1, x ∈ Ω ⊂ Rd ,
(3.9)
u(x) = 0, x ∈ ∂Ω.
Here σ > 0 and the second term on the right-hand side describes the Casimir force.
Moreover, with a general nonlinearity g the result is new for the problem

−∆u(x) = λf (x)g(u), 0 ≤ u(x) < 1, x ∈ Ω ⊂ Rd ,
(3.10)
u(x) = 0, x ∈ ∂Ω.
1
Note that in the well-known case when g(u) = (1−u(x)) 2 the above proof seems new

and simple.
Now we demonstrate our method to the multidimensional MEMS problems in
the whole Euclidean space Rd and nonlinear Dirichlet boundary value problems of
uniformly elliptic differential operators. Note that, moreover, following the above
Dirichlet p-MEMS problem analysis (see [8] and [7]) one can obtain results on, for
8 DURVUDKHAN SURAGAN AND DONGMING WEI

example, the pull-in distance of the p-MEMS problem (using the bifurcation results of
Crandall and Rabinowitz [5] and [6]) and/or analogues of the bi-Laplacian nonlinear
singular MEMS problems (see, e.g. [7]).

3.2. The pull-in voltage for the Newtonian potential. We consider the pull-in
voltage for the stationary deflection of an infinity elastic membrane satisfying

 f (x)
−∆u(x) = λ (1−u(x)) 2, 0 ≤ u(x) < 1, x ∈ Rd , d ≥ 3,
(3.11)
u(x) −→ 0, |x| −→ ∞,
where λ > 0 is the applied voltage and the permittivity profile f satisfies the assump-
tions supp f ⊂ Ω ⊂ Rd , f ∈ C α (Ω̄) for some α ∈ (0, 1], 0 ≤ f ≤ 1, and f 6≡ 0.
As usual, the pull-in voltage is defined as
λ∗ (Ω, f ) = sup{λ > 0 | (3.1) possesses at least one classical solution}.
Theorem 3.2. There exists a positive pull-in voltage λ∗ < ∞ such that
a) For any λ < λ∗ , there exists at least one solution of (3.11).
b) For any λ > λ∗ , there is no solution of (3.11).
Proof of Theorem 3.2. By Proposition 1.2 the problem (3.11) is equivalent to the
nonlinear integral problem (1.9). Thus, since (1.9) has the trivial solution u = 0 with
λ = 0, by the implicit function theorem (1.9) has a solution. In addition, since the
fundamental solution εd is positive the integral on the right hand sight of (1.9) is
positive. This means that λ must be positive, that is, 0 < λ < λ∗ . Now we need to
show that λ∗ < ∞. Let 0 ≤ u(x) < 1 be a solution of (1.9). We also use the following
known fact (see [15]): The first eigenvalue µ1 of the spectral problem
Z
(3.12) φ1 (x) = µ1 εd (x − y)φ1(y)dy
Ω

is simple and positive as well as the corresponding eigenfunction φ1 can be chosen

positive. Thus, let us multiply (1.9) by φ1 and integrate over Ω, then we have

f (y)
Z Z Z
u(x)φ1 (x)dx = λ εd (x − y) dyφ1(x)dx, 0 ≤ u(x) < 1.
Ω Ω Ω (1 − u(y))2
By (3.12) we obtain
φ1 (y)f (y)
Z Z
λ
u(x)φ1 (x)dx = dy, 0 ≤ u(x) < 1,
Ω µ1 Ω (1 − u(y))2
that is,
R R
µ1 Ω
u(x)φ1 (x)dx µ1 Ω φ1 (x)dx
λ= φ1 (y)f (y)
≤R .
φ (y)f (y)dy
R
Ω (1−u(y))2
dy Ω 1

This means
R
µ 1 φ1 (x)dx
(3.13) λ∗ ≤ R Ω < ∞,
Ω 1
φ (y)f (y)dy
 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 9

for all x ∈ Rd . It also shows that there is no solution of (1.9) for any λ > λ∗ . On
the other hand, by the definition of λ∗ for any λ ∈ (0, λ∗) there exists λ̃ ∈ (λ, λ∗ ) for
which (1.9) has a solution uλ̃ , that is,
f (y) f (y)
Z Z
(3.14) uλ̃ (x) = λ̃ εd (x − y) 2
dy ≥ λ εd (x − y) dy,
Ω (1 − uλ̃ (y)) Ω (1 − uλ̃ (y))2
This also means that uλ̃ is a supsolution of (3.14) for the parameter λ. On the
other hand, since
Z
(3.15) 0 ≤ λ εd (x − y)f (y)dy,
Ω
u ≡ 0 is a subsolution of
f (y)
Z
(3.16) uλ̃ (x) ≤ λ εd (x − y) dy.
Ω (1 − uλ̃ (y))2
Therefore, by the method of sub- and supsolutions (see the proof of [8, Theorem
2.1.1]) we prove existence of a solution uλ of (1.9) for any λ ∈ (0, λ∗ ). 
Now we are ready to prove the following result.
Theorem 3.3. We have
λ∗ (f ∗ ) ≤ λ∗ (f )
for the permittivity profile f satisfying the assumptions supp f ⊂ Ω ⊂ Rd , d ≥ 3,
f ∈ C α (Ω̄) for some α ∈ (0, 1], 0 ≤ f ≤ 1, and f 6≡ 0. Here f ∗ is the symmetric
decreasing rearrangement of f .
Proof of Theorem 3.3. Let u be any positive solution of (1.9). Define the sequence
(the Picard iteration scheme)
f (y)
Z
(3.17) um (x) = λ εd (x − y) dy, u0 (x) ≡ 0, m = 1, 2, . . . ,
Rd (1 − um−1 (y))2
with supp f ⊂ Ω. We have u > u0 ≡ 0 and whenever u ≥ um−1 , then
 
1 1
Z
u(x) − um (x) = λ εd (x − y)f (y) − dy ≥ 0,
Rd (1 − u(y))2 (1 − um−1 (y))2
for all x ∈ Rd , that is, 1 > u ≥ um in Rd for each m ≥ 0. Moreover, from (3.17)
it is straightforward to see that the sequence {um } is monotone increasing. Thus, it
converges uniformly to a positive solution uλ satisfying u ≥ uλ in Rd . Consider the
following two sequences
f (y)
Z
(3.18) un (x) = λ εd (x − y) dy, u0 (x) ≡ 0, n = 1, 2, . . . ,
Rd (1 − un−1(y))2
with supp f ⊂ Ω, and
f ∗ (|y|)
Z
(3.19) vn (x) = λ εd (x − y) dy, v0 (x) ≡ 0, n = 1, 2, . . . .
Rd (1 − vn−1 (y))2
We have
−∆u1 = λf, −∆v1 = λf ∗ ,
10 DURVUDKHAN SURAGAN AND DONGMING WEI

therefore, by Talenti’s comparison principle for the Laplacian (see Theorem 2.1) we
obtain
(3.20) u∗1 (x) ≤ v1 (x), ∀x ∈ Rd .
We also have
f
(3.21) − ∆u2 = λ .
(1 − u1 )2
In addition, let us consider
f ∗ (|y|)
Z
ṽ2 (x) = λ εd (x − y) dy,
Rd (1 − u∗1 (y))2
that is,
f∗
(3.22) − ∆ṽ2 = λ .
(1 − u∗1 )2
Thus, by Theorem 2.1 for (3.21) and (3.22) we obtain
u∗2 (x) ≤ ṽ2 (x), ∀x ∈ Rd .
By using (3.20) we get
f ∗ (|y|) f ∗ (|y|)
Z Z
ṽ2 (x) = λ εd (x − y) dy ≤ λ εd (x − y) dy = v2 (x),
Rd (1 − u∗1 (y))2 Rd (1 − v1 (y))2
in Rd , that is,
u∗2 (x) ≤ v2 (x), ∀x ∈ Rd .
Further, by continuing this process we obtain that u∗n ≤ vn for all n ≥ 0, that is,
max u∗n ≤ max vn for all n ≥ 0. Since max u∗n = max un , it means that for a given
B B B Ω
λ if {vn } converges, then the sequence {un } is also converges. Thus, we arrive at
λ∗ (f ∗ ) ≤ λ∗ (f ). 
We have the following upper bound for the pull-in voltage:
Proposition 3.4. Let µ1 (Ω) be the first eigenvalue of the Newtonian potential (3.12)
in Ω. Then
4µ1 (Ω)
(3.23) λ∗ (f ) ≤ (inf f )−1 .
27 Ω

Proof of Proposition 3.23. As in the proof of Theorem 3.2, for any λ ∈ (0, λ∗ ) we
have
φ1 (y)f (y)
Z Z
λ
u(x)φ1 (x)dx = dy, 0 ≤ u < 1.
Ω µ1 Ω (1 − u(y))2
4
Since u(1 − u)2 ≤ 27 we obtain
Z
λ
Z
u(y)φ1(y)f (y) 27λ inf f Z
Ω
u(x)φ1 (x)dx = 2
dy ≥ u(y)φ1(y)dy
Ω µ1 Ω u(y)(1 − u(y)) 4µ1 Ω

proving the inequality (3.23). 

 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 11

One can prove other upper estimates of the pull-in voltage that depends on the
global properties of the permittivity profile (see (3.13)). For the Dirichlet case (see
[10]) we have the estimate
R D
D φ (x)dx
4µ (Ω)
(3.24) (λD )∗ (f ) ≤ 1 R ΩD 1 ,
3 φ (y)f (y)dy
Ω 1

where µD D
1 and φ1 are the first eigenvalue and the first eigenfunction of the Dirichlet
Laplacian, respectively.

3.3. The pull-in voltage for uniformly elliptic problems. We consider the pull-
in voltage problem for the second order uniformly elliptic differential operator:

−Lu(x) = λf (x)g(u), 0 < u, x ∈ Ω ⊂ Rd ,
(3.25)
u(x) = 0, x ∈ ∂Ω.
where
d  
X ∂ ∂
L= ajk (x)
j,k=1
∂x j ∂xk
with ajk (x) = akj (x) is a second order uniformly elliptic differential operator, that is,
there exists a positive constant c such that
d
X d
X
aj,k (x)ξj ξk ≥ c ξj2
j,k=1 j

for all ξ and x ∈ Ω. Let us assume that f and g are (smooth) positive functions such
that ∂g(t)
∂t
> 0 in R+ and (f (x)g(u(x)))∗ = f ∗ (x)g(u∗(x)),
As before, λ > 0 is the applied voltage and the pull-in is defined as
λ∗L := λ∗L (Ω, f ) = sup{λ > 0 | (3.23) possesses at least one classical solution}.
We obtain the following geometric estimate for λ∗L .
Theorem 3.5. A ball B (with the symmetrized permittivity profile) is a minimizer
of the Dirichlet pull-in voltage of the operator L among all domains of given volume,
i.e.
λ∗L (B, f ∗ ) ≤ λ∗L (Ω, f ),
for an arbitrary smooth domain Ω ⊂ Rd , d ≥ 1, with |Ω| = |B|, where | · | is the
Lebesgue measure in Rd .
Proof of Theorem 3.5. In [1, Chapter IV] it was proved that λ∗L < ∞ and for every
λ ∈ (0, λ∗L ) there is a minimal (and semi-stable) solution uλ (i.e. uλ is the smallest
positive solution of (3.25) in a pointwise sense). Define the sequence
(3.26) − Lum (x) = λf (x)g(um−1), u0 (x) ≡ 0, m = 1, 2, . . . ,
with um = 0 on the boundary ∂Ω of the smooth bounded domain Ω. Dirichlet
boundary value problem (3.25) is solvable if and only if the sequence {um } is uniformly
bounded. Moreover, the sequence {um } is uniformly bounded, then it converges
uniformly to a minimal solution uλ satisfying u ≥ uλ in Ω (see [12]), where u is a
12 DURVUDKHAN SURAGAN AND DONGMING WEI

positive solution of (3.25). As in the previuos proofs let us consider the following two
sequences
(3.27) − Lun (x) = λf (x)g(un−1) in Ω, un (x)|∂Ω = 0, n = 1, 2, . . . ,
and
(3.28) − Lvn (x) = λf ∗ (|x|)g(vn−1) in B, vn (x)|∂B = 0, n = 1, 2, . . . ,
with u0 ≡ 0 and v0 ≡ 0. We have
−Lu1 (x) = λf (x)g(0) in Ω, u1 |∂Ω = 0,
and
−Lv1 (x) = λf ∗ (|x|)g(0) in B, v1 (x)|∂B = 0.
Here f ∗ is the symmetric decreasing rearrangement of the positive function f . There-
fore, by Talenti’s comparison principle for the Dirichlet boundary value problem for
second order uniformly elliptic differential operators we obtain
(3.29) u∗1 (x) ≤ v1 (x), ∀x ∈ B,
where B is the ball centered at the origin with |B| = |Ω|. We also have
(3.30) − Lu2 = λf g(u1) in Ω, u2 |∂Ω = 0.
In addition, let us consider
−Lṽ2 (x) = λ(f g(u1))∗ in B, ṽ2 |∂B = 0,
that is,
(3.31) − Lṽ2 (x) = λf ∗ g(u∗1) in B, ṽ2 |∂B = 0.
By Talenti’s comparison principle for (3.30) and (3.31) we obtain
u∗2 (x) ≤ ṽ2 (x), ∀x ∈ B.
By using (3.29) we get
−Lṽ2 (x) = λf ∗ g(u∗1) ≤ λf ∗ g(v1 ) = −Lv2 (x),
in B, that is, by the (classical) comparison principle for −L we get
ṽ2 (x) ≤ v2 (x), ∀x ∈ B.
This gives
u∗2 (x) ≤ v2 (x), ∀x ∈ B.
By induction we arrive at u∗n ≤ vn in B for all n ≥ 0, that is, max u∗n ≤ max vn for
B B
all n ≥ 0. Since max u∗n = max un , it means that for a given λ if {vn } converges, then
B Ω
the sequence {un } is also converges. This fact proves λ∗L (B, f ∗ ) ≤ λ∗L (Ω, f ). 
 ON GEOMETRIC ESTIMATES FOR PULL-IN VOLTAGE 13

References
[1] C. Bandle. Isoperimetric inequalities and applications. Monographs and Studies in Math-
ematics, Pitman, Boston, London, 1980.
[2] A. Burchard. A short course on rearrangement inequalities.
http://www.math.utoronto.ca/almut/rearrange.pdf, 2009.
[3] D. Castorina, P. Esposito and B. Sciunzi. p-MEMS equation on a ball. Methods and Ap-
plications of Analysis, 15:277–284, 2008.
[4] D. Castorina, P. Esposito and B. Sciunzi. Degenerate elliptic equations with singular
nonlinearities. Calc. Var., 34:279–306, 2009.
[5] M. G. Crandall and P. H. Rabinowitz. Bifurcation, perturbation of simple eigenvalues and
linearized stability. Arch. Rational Mech. Anal., 52:161–180, 1973.
[6] M. G. Crandall and P. H. Rabinowitz. Some continuation and variational methods for
positive solutions of nonlinear elliptic eigenvalue problems. Arch. Rational Mech. Anal.,
58:207–218, 1975.
[7] C. Cowan, P. Esposito, N. Ghoussoub, and A. Moradifam, The critical dimension for
a fourth order elliptic problem with singular nonlinearity. Arch. Rational Mech. Anal.,
198:763–787, 2010.
[8] P. Esposito, N. Ghoussoub and Y. Guo. Mathematical Analysis of Partial Differential
Equations Modeling Electrostatic MEMS, volume 20 of Courant Lecture Notes in Mathe-
matics. American Mathematical Society, Providence, RI, 2009.
[9] J. Garcia-Melian and J. Sabina de Lis, Maximum and comparison principles for operators
involving the p-Laplacian. J. Math. Anal. Appl., 218(1):49–65, 1998.
[10] Y. Guo, Z. Pan, and M. J. Ward, Touchdown and pull-in voltage behavior of a MEMS
device with varying dielectric properties. SIAM J. Appl. Math., 66:309–338, 2005.
[11] T. S. Kal’menov and D. Suragan. On spectral problems for the volume potential. Dokl.
Math., 428(1):16–19, 2009.
[12] H. B. Keller and D. S. Cohen. Some positone problems suggested by nonlinear heat gen-
eration. J. Math. Mech., 16:1361–1376, 1967.
[13] B. Lai. On the partial differential equations of electrostatic MEMS devices with effects of
Casimir force. Ann. Henri Poincaré, 16:239–253, 2015.
[14] E. H. Lieb and M. Loss. Analysis, volume 14 of Graduate Studies in Mathematics. American
Mathematical Society, Providence, RI, second edition, 2001.
[15] G. Rozenblum, M. Ruzhansky and D. Suragan. Isoperimetric inequalities for Schatten
norms of Riesz potentials. J. Funct. Anal., 271:224–239, 2016.
[16] M. Ruzhansky and D. Suragan. Isoperimetric inequalities for the logarithmic potential
operator. J. Math. Anal. Appl., 434:1676–1689, 2016.
[17] M. Ruzhansky and D. Suragan. Schatten’s norm for convolution type integral operator.
Russ. Math. Surv., 71:157–158, 2016.
[18] G. Talenti. Elliptic equations and rearrangements. Annali della Scuola Normale Superiore
di Pisa Classe di Scienze, 3:697–718, 1976.
[19] G. Talenti. Nonlinear elliptic equations, rearrangements of functions and Orlicz spaces.
Ann. Mat. Pura Appl., 120:159–184, 1977.

Durvudkhan Suragan:
Institute of Mathematics and Mathematical Modeling
125 Pushkin str., 050010 Almaty, Kazakhstan
and
RUDN University
6 Miklukho-Maklay St, Moscow, 117198, Russia
E-mail address suragan@list.ru

Dongming Wei:
Department of Mathematics
14 DURVUDKHAN SURAGAN AND DONGMING WEI

School of Science and Technology, Nazarbayev University

53 Kabanbay Batyr Ave, Astana 010000
Kazakhstan
E-mail address dongming.wei@nu.edu.kz