Normalized solutions for a Sobolev critical quasilinear

Schrödinger equation

Yuxin Li*, Meijie Yang† and Xiaojun Chang ‡

arXiv:2506.10870v1 [math.AP] 12 Jun 2025

School of Mathematics and Statistics & Center for Mathematics and Interdisciplinary Sciences, Northeast Normal
University, Changchun 130024, Jilin, PR China

Abstract
In this paper, we study the existence of normalized solutions for the following quasilinear Schrödinger
equation with critical exponent:
∗
−2
−∆u − u∆(u2 ) + λu = τ|u|q−2 u + |u|2·2 u, x ∈ RN ,

under the mass constraint RN |u|2 dx = c for some prescribed c > 0. Here τ ∈ R is a parameter, λ ∈ R
R

appears as a Lagrange multiplier, N ≥ 3, 2∗ := N−22N

and 2 < q < 2 · 2∗ . By deriving precise energy level
estimates and establishing new convergence theorems, we apply the perturbation method to establish
several existence results for τ > 0 in the Sobolev critical regime:
(a) For the case of 2 < q < 2 + N4 , we obtain the existence of two solutions, one of which is a local
minimizer, and the other is a mountain pass type solution, under explicit conditions on c > 0;
(b) For the case of 2 + N4 ≤ q < 4 + N4 , we obtain the existence of normalized solutions of mountain
pass type under different conditions on c > 0;
(c) For the case of 4 + N4 ≤ q < 2 · 2∗ , we obtain the existence of a ground state normalized solution
under different conditions on c > 0.
Moreover, when τ ≤ 0, we derive the non-existence result for 2 < q < 2 · 2∗ and all c > 0. Our research
provides a comprehensive analysis across the entire range q ∈ (2, 2 · 2∗ ) and for all N ≥ 3. The methods
we have developed are flexible and can be extended to a broader class of nonlinearities.

Keywords: Normalized solutions; Quasilinear Schrödinger equations; Combined nonlinearities; Sobolev

critical growth; Perturbation method.

1 Introduction
In this paper, we investigate the quasilinear Schrödinger equation with a Sobolev critical exponent,
which takes the following form:
∗
(
i∂t ψ + ∆ψ + ψ∆(|ψ|2 ) + τ|ψ|q−2 ψ + |ψ|2·2 −2 ψ = 0, (t, x) ∈ R+ × RN ,
(1.1)
ψ(0, x) = ψ0 (x), x ∈ RN ,
* [email protected]
† [email protected]
‡ [email protected]

1
where N ≥ 3, 2 < q < 2 · 2∗ , τ > 0 is a parameter, i denotes the imaginary unit, and ψ : R+ × RN → C rep-
resents the time-dependent wave function with the initial datum ψ0 . This class of quasilinear equations has
received widespread attention in recent years due to its rich applications in mathematical physics, for exam-
ple, in dissipative quantum mechanics, plasma physics and fluid mechanics. For more physical motivations
and applications, we refer the interested readers to [8, 9, 27, 29, 30, 44] and the references therein.
We are interested in the standing wave solutions to (1.1), especially those of the form ψ(t, x) = e−iλt u(x),
where λ ∈ R. In this case, the function u(x) satisfies the following stationary equation
∗ −2
−∆u − u∆(u2 ) + λu = τ|u|q−2 u + |u|2·2 u, x ∈ RN . (1.2)

When looking for solutions to (1.2), a typical strategy is to treat λ ∈ R as a fixed parameter. The solutions
of (1.2) correspond to the critical points of the energy functional given by
1 1
Z Z Z Z Z
λ τ ∗
Jλ (u) := |∇u|2 dx + |u|2 dx + |u|2 |∇u|2 dx − |u|q dx − |u|2·2 dx,
2 RN 2 RN RN q RN 2 · 2∗ RN

which is defined on the natural space

Z
Xe := {u ∈ H 1 (RN ) : |u|2 |∇u|2 dx < +∞}.
RN

Owing to the presence of the quasilinear term RN |u|2 |∇u|2 dx, the functional Jλ is non-differentiable in
R

Xe for N ≥ 2. To address this difficulty, several variational techniques have been developed, including
minimization methods [40, 44, 45], the Nehari manifold approach [2, 39], the dual approach [13, 38], and
perturbation approaches [34, 37].
It is noteworthy that the number 2 · 2∗ acts as a critical exponent for (1.2), as first observed in [39]. In
that work, Liu, Wang and Wang considered the following quasilinear Schrödinger equation

−∆u +V (x)u − u∆(u2 ) = f (u), x ∈ RN , (1.3)

where f (u) = |u|q−2 u and V (x) is a given potential. By using a Pohozaev type identity, they proved that
(1.3) admits no positive solutions in Xe if q ≥ 2 · 2∗ and x · ∇V (x) ≥ 0 for all x ∈ RN . In [6], Bezerra,
Miyagaki and Soares demonstrated the existence of positive solution for (1.3) with combined nonlinearities
∗
f (u) = |u|q−2 u+|u|2·2 −2 u for 4 < q < 2·2∗ . They utilized a change of variables, similar to those in [13, 38],
to transform (1.3) into a semilinear elliptic equation whose associated functionals are well-defined in the
Sobolev space H 1 (RN ). Subsequently, Liu, Liu and Wang [36] established the existence of both positive
and nodal ground states of (1.3) by analyzing the behavior of solutions in subcritical cases and passing to
the limit as the exponent tends to 2 · 2∗ . In [35], they adopted a perturbation method to investigate a more
general quasilinear elliptic equations and achieved new existence results. For further details on critical
quasilinear problems, the reader is referred to [16, 47, 53] and their references.
Alternatively, it is intriguing to investigate solutions to (1.2) with a prescribed L2 -norm, motivated by
the physical principle of mass conservation. Specifically, we seek pairs (u, λ) ∈ Xe × R satisfying
∗
(
−∆u − u∆(u2 ) + λu = τ|u|q−2 u + |u|2·2 −2 u, x ∈ RN ,
R 2
(1.4)
RN |u| dx = c,

where c > 0 is a given constant, and the parameter λ ∈ R arises as a Lagrange multiplier. Solutions of
this form are commonly termed normalized solutions, and the present paper is dedicated to exploring this
topic. Formally, a natural approach to find solutions of (1.4) is to search for critical points of the associated
functional
1 1
Z Z Z Z
τ ∗
I(u) := |∇u|2 dx + |u|2 |∇u|2 dx − |u|q dx − |u|2·2 dx (1.5)
2 RN RN q RN 2 · 2∗ RN

2
on the constraint set Z
e := {u ∈ Xe :
S(c) |u|2 dx = c}.
RN

The L2 -constraint gives rise to a critical exponent q̄ := 4 + N4 , which is pivotal for analyzing dynamical
properties like global existence, blow-up, and stability of ground states. Furthermore, this threshold signif-
icantly influences the geometric structure of the functional I. Specifically, for the pure nonlinearity |u|q−2 u,
the case q < q̄ is known as L2 -subcritical. In this scenario, the functional I is bounded from below on S(c),
e
and global minimizers always exist. Conversely, when q > q̄, the case is designated as L2 -supercritical.
Here, the functional I is unbounded from below on S(c), e necessitating additional technical arguments, as
one can not expect to directly locate a minimum of I on S(c).
e
The L2 -subcritical case, i.e., 2 < p < q̄, has been widely studied. In [14], Colin, Jeanjean and Squassina
considered the following equation
(
−∆u − u∆(u2 ) + λu = |u| p−2 u, x ∈ RN ,
R 2
(1.6)
RN |u| dx = c.

They employed a minimization approach:

e := inf E(u),
m(c) (1.7)
u∈S(c)
e

where
1 1
Z Z Z
E(u) := |∇u|2 dx + |u|2 |∇u|2 dx − |u| p dx.
2 RN RN p RN

They proved that the minimization problem (1.7) has a minimizer for p ∈ (2, 2 + N4 ) across all c > 0, and for
p ∈ [2 + N4 , 4 + N4 ) when c > 0 is sufficiently large. Notably, the non-differentiability of E did not impact
their proof. Jeanjean and Luo [24] later extended these results and provided a sharp conclusion. Using a
perturbation method, Jeanjean, Luo, and Wang [25] investigated the case p ∈ (2 + N4 , 4 + N4 ) and established
the existence of a mountain pass type normalized solution, as well as another solution that is either a local
minimizer or a global minimizer, depending on different conditions on the mass c > 0. In [56], Zhang,
Li and Wang presented a change of variables technique to construct multiple normalized solutions of (1.6)
with energies tending to 0 for N ≥ 2 and p ∈ (2, 4 + N4 ). By demonstrating the equivalence between the
quasilinear and semilinear equations under the L2 -constraint, they established the existence of infinitely
many pairs of normalized solutions in this dual framework.
In the L2 -critical/supercritical case, i.e., p ≥ q̄, Ye and Yu [54] studied the L2 -critical problem and
identified a threshold value of c > 0 that distinguishes between the existence and nonexistence of critical
points of the functional E restricted on S(c).
e Successively, Li and Zou [31] employed a perturbation method
to establish the existence of ground state normalized solutions and further obtained multiplicity results
through index theory. Zhang, Chen, and Wang [55] worked directly with the functional E in spite of
its non-smoothness in low-dimensional space. They proved the existence of ground states to (1.6) for
p ∈ [4 + N4 , N−2
2N
] when N = 3 and p ∈ [4 + N4 , ∞) when N = 1, 2. More recently, Mao and Lu [42] studied
the existence of normalized solutions for quasilinear Schrödinger equations with combined nonlinearities
of the form τ|u|q−2 u + |u| p−2 u, where 2 < q < 2 + N4 < p ≤ 2∗ . In a more general setting, Deng, Squassina,
Zhang and Zhong [15] utilized a dual approach, fixed point index theory, and global bifurcation techniques
to explore the existence of normalized solutions for quasilinear Schrödinger equations.
We observe that the aforementioned studies were all conducted within the H 1 -subcritical and critical
framework, i.e., p ≤ 2∗ . In a recent result, Jeanjean, Zhang and Zhong [26] obtained the existence of
ground state normalized solutions for (1.6), including the case 2∗ < p < 2 · 2∗ . They also analyzed the
asymptotic behavior of these ground states as the mass c > 0 varies. For normalized solution of quasilinear

3
Schrödinger equations involving potentials with power nonlinearity in the range 4 + N4 < p < 2 · 2∗ , we refer
the reader to [17, 18].
In this paper, we focus on the study of (1.4), which involves the critical exponent 2 · 2∗ . This equation
serves as an analogue to the Brezis-Nirenberg problem within the framework of normalized solutions for
quasilinear Schrödinger equations. If we ignore the fourth term −u∆(u2 ), the equation (1.4) reduces to the
classical nonlinear Schrödinger equation, and the associated Sobolev critical problem becomes
∗
(
−∆u + λu = τ|u|q−2 u + |u|2 −2 u, x ∈ RN ,
R 2
(1.8)
RN |u| dx = c,

where 2 < q < 2∗ . In [48], Soave employed Nehari-Pohozaev manifold decomposition to establish the
existence and asymptotic behavior of ground state solutions to (1.8) in the cases of L2 -subcritical, L2 -
critical, and L2 -supercritical perturbations, respectively. Subsequently, for the L2 -subcritical scenario where
q ∈ (2, 2 + N4 ), Jeanjean and Le [23] demonstrated the existence of a second solution at the mountain pass
level for N ≥ 4. Concurrently, Wei and Wu [50] established the existence of mountain pass type solutions
for N ≥ 3. For more results on this topic, the reader may refer to [11, 12, 22, 32].
The aim of this paper is to conduct a thorough analysis of the existence of normalized solutions for (1.4)
across the entire range q ∈ (2, 2 · 2∗ ) and for all N ≥ 3. To achieve this, we will investigate in detail the
∗
interaction between the lower-order term τ|u|q−2 u and the Sobolev critical term |u|2·2 −2 u, and analyze how
this interaction influences the structure of the constrained functional.
Compared to the classical Schrödinger equation (1.8), the study of (1.4) is significantly more intricate.
We must not only address the non-smoothness caused by the quasilinear term −u∆(u2 ), but also resolve the
∗
loss of compactness resulting from the Sobolev critical term |u|2·2 −2 u. The main challenges are outlined
as follows:
• In the process of compactness analysis, we encounter two main obstacles arising from the quasilin-
ear term −u∆(u2 ). On the one hand, it is widely recognized that Brezis-Lieb type lemma plays a
crucial role in studying L2 -constraint problems. However, for our problem, such a splitting lemma
is not readily available. The primary difficulty here lies in the emergence of undesirable cross-term
interactions:
Z Z Z Z
|v1 + v2 |2 |∇(v1 + v2 )|2 dx = |v1 |2 |∇v1 |2 dx + 2 |v1 |2 ∇v1 ∇v2 dx + |v1 |2 |∇v2 |2 dx
RN RN RN RN
Z Z Z
+2 v1 v2 |∇v1 |2 dx + 4 v1 v2 ∇v1 ∇v2 dx + 2 v1 v2 |∇v2 |2 dx
RN RN RN
Z Z Z
+ |v2 |2 |∇v1 |2 dx + 2 ∇v1 ∇v2 |v2 |2 dx + |v2 |2 |∇v2 |2 dx.
RN RN RN

On the other hand, for the bounded Palais-Smale sequence {un }, the weak convergence un ⇀ u in
H 1 (RN ) does not imply the desired convergence
Z Z
un φ|∇un |2 + |un |2 ∇un ∇φ dx → uφ|∇u|2 + |u|2 ∇u∇φ dx as n → +∞,



RN RN
∗
where φ ∈ C0∞ (RN ). In our arguments, the critical term |u|2·2 −2 u introduces additional complexity,
preventing us from bypassing the discussion on such splitting property as was done in [25, 31] for
addressing the subcritcal problems.
• When investigating the problem (1.8), the parameter range of q in the L2 -subcritcal perturbation term
τ|u|q−2 u typically yields two distinct types of solutions: local minimizers and mountain pass solu-
tions. Since problem (1.4) exhibits a similar variational structure, it is reasonable to seek these solu-
tions when q lies within the L2 -subcritcal range, i.e., q < 4 + N4 . However, for quasilinear problems
with Sobolev critical growth, we face the following challenges:

4
 (i) How can we establish the strict subadditivity of the local minimization energy, and then derive
the compactness of minimizing sequences in Xe from the local minimization geometry?
(ii) How can we establish a precise upper bound estimate for the corresponding mountain pass
energy levels, a commonly used strategy for recovering compactness in such problems? This
step is crucial in the Sobolev critical setting, ensuring that the minimax levels remain below the
threshold needed for compactness recovery. For our problem (1.4), by the Sobolev inequality
as follows,
Z  2∗ Z
2
2·2∗
S |u| dx ≤4 |u|2 |∇u|2 dx,
RN RN
∗
we observe that the terms −u∆(u2 ) and |u|2·2 −2 u in (1.4) are comparable. This necessitates
careful handling of the additional perturbation −∆u when performing subtle estimates and anal-
yses. Therefore, it is essential to separately determine the compactness threshold for our prob-
lem when 2 < q < 2 + N4 and when q ≥ 2 + N4 , and then conduct a meticulous analysis to obtain
the relevant energy estimates.

• Beyond the aforementioned obstacles, the non-vanishing of the Lagrange multiplier is also instru-
mental in controlling potential losses of compactness. The conventional approach to proving λ ̸= 0
involves combining the Nehari-Pohozaev type identity, as seen in [31]. Within this framework, the
result that λ > 0 could only be established for (1.6) when p ≤ 2∗ , which leads to existence results
that are valid only in low-dimensional spaces, specifically for N ≤ 3 (see also [42, 55]). However,
since we are actually dealing with a H 1 -supercritical problem, the standard method for determining
the sign of λ ∈ R is no longer applicable. Therefore, alternative techniques are needed to ascertain the
sign of λ ∈ R, which will remove additional conditions on N, p, and allow for extension to a broader
class of problems.

1.1 Main results

It is well known that for N ≥ 2, the functional I defined by (1.5) typically lacks smoothness in X,
e which
precludes the direct application of classical critical point theory. To handle the quasilinear term −u∆(u2 ),
it is necessary to work in a smaller space to achieve smoothness. To overcome this challenge, we adopt
a perturbation method as used in [25] and [31]. To be more precise, we consider the following perturbed
functional:
µ
Z
Iµ (u) := |∇u|θ dx + I(u), µ ∈ (0, 1] (1.9)
θ RN
on the reflexive Banach space X := W 1,θ (RN ) ∩ H 1 (RN ), where
 
4N 4N + 4
< θ < min ,N .
N +2 N +2

Then Iµ is a well-defined and of class C 1 on X, as shown in [31, Lemma A.1]. At this point, we study the
perturbed problem by searching for critical points of Iµ restricted on
Z
S(c) := {u ∈ X : |u|2 dx = c}.
RN

We introduce the related Pohozaev type mainfold defined by

Λµ (c) := {u ∈ S(c) : Qµ (u) = 0}, (1.10)

5
where
Z Z Z
2
Qµ (u) := µ(1 + γθ ) θ
|∇u| dx + |∇u| dx + (N + 2) |u|2 |∇u|2 dx
RN RN RN
Z Z
∗
− τγq |u|q dx − γ2·2∗ |u|2·2 dx, (1.11)
RN RN

with γq := N(q−2)
2q . Our main strategy is to first identify the critical point uµ of the perturbed functional Iµ
for any fixed µ ∈ (0, 1]. Subsequently, we take the limit as µ → 0+ , which yields the normalized solutions
to the original problem (1.4).
To state our main results, we introduce the following constants and notations. Throughout this paper,
we define
∥∇u∥22
S := inf 2
,
u∈D 1,2 (RN ) ∥u∥2∗
u̸=0
where
∗
D 1,2 (RN ) := {u ∈ L2 (RN ) : |∇u| ∈ L2 (RN )}
is the completion of C0∞ (RN ) with the norm ∥u∥D 1,2 (RN ) = ∥∇u∥2 .
Set
 N
1 2
c0 := > 0, (1.12)
2K
where
" 2∗
# α α−α
0
  2∗ " 2∗
# α α−α
2
τC2 (q, N) α0 τC2 (q, N)2∗ S 2 2 0
1 4 2 α0 τC2 (q, N)2∗ S 2 2 0
K := − ∗ + − ∗ > 0,
q α2 22 −1 q 2 · 2∗ S α2 22 −1 q
where α0 and α2 are given by (3.2).
 N
2
N +2  .
c̄1 :=  (1.13)
2+ N4
2 + N4 , N


2τN C1
 4N+4−Nq  (N+2)[N(q−2)−4]
(N 2 + 4)q
 
1 N(q−2) D1 (N) 2q(N + 2) 2N(q−2)
c̄2 := q , (1.14)
τC1 (q, N) N[4N − q(N − 2)] τC2 (q, N) 4N − q(N − 2)
where
N2 + 4
 
2
D1 (N) := min , .
2N(N + 2) N
!N
2
4N + 4
c̄3 := . (1.15)
τN(N + 2)C2 4 + N4 , N

Here C1 (q, N) and C2 (q, N) denote the best constants for the Gagliardo-Nirenberg type inequality in RN , as
given in (2.1) and (2.4), respectively.
We define the set
e : ∥v∇v∥22 < ρ̃0 },
V0 (c) := {v ∈ S(c) (1.16)
where
 2(N+2) 4N−q(N−2)
τN C2 (q, N) 1 (N − 2)γq
 
4N+4−Nq
ρ̃0 := − c04N+4−Nq .
2 q N(N + 2)
Now, we present our main results on the existence and multiplicity of normalized solutions to equation
(1.4). We first consider the case where q ∈ (2, 2 + N4 ).

6
 ˜ 0) ∈
Theorem 1.1. Let N ≥ 3, q ∈ (2, 2 + N4 ), τ > 0 and c ∈ (0, c0 ). Then (1.4) admits a solution pair (ṽ0 , λ
N +
e ∩ L (R )) × R satisfying
(S(c) ∞

ṽ0 ∈ V0 (c), ṽ0 > 0, I(ṽ0 ) = inf I(v) < 0.

v∈V0 (c)

Theorem 1.2. Let N ≥ 3, q ∈ (2, 2 + N4 ) and c ∈ (0, c0 ). Then there exists a sufficiently large τ∗ = τ∗ (c) > 0
¯ ∈ (S(c)
such that for any τ > τ∗ , (1.4) admits a solution pair (ū, λ) e ∩ H 1 (RN ) ∩ L∞ (RN )) × R+ satisfying
rad
N
S2
0 < I(ū) < I(ṽ0 ) + .
2N
Remark 1.3. In Theorem 1.1, we develop new analytical techniques to address the obstacles caused by the
lack of compactness due to the critical exponent and the non-radial space setting. We begin by establishing
the strict subadditial property of the minimizing energy associated with the perturbed functional Iµ restricted
on Vµ (c), as defined in (3.3). Subsequently, by employing the concentration-compactness principle, we
obtain local minimizers vµ ∈ Vµ (c) for any given µ > 0, satisfying 0 > Iµ (vµ ) = mµ (c) := infu∈Vµ (c) Iµ (u).
These minimizers do not necessarily exhibit radial symmetry.
To prove a convergence result, we need to establish a Brezis-Lieb type lemma. However, the quasilinear
term −u∆(u2 ) poses significant challenges in achieving this. To overcome this issue, in Lemma 3.7, for
the minimizing sequence {un } of Iµ on Vµ (c), we decomposing un into two disjoint families of functions,
thereby effectively neutralizing the impact of the cross terms. This leads to the following result: weak
convergence un ⇀ vµ in X implies that

un → vµ strongly in X and ∥un ∇un ∥22 → ∥vµ ∇vµ ∥22 .

Next, we establish a new profile decomposition theorem and apply it to derive the convergence of the
sequence {vµn } ⊂ Vµn (c) within the non radially symmetry space framework, as stated in Proposition 7.1,
to obtain a local minimizer for the original problem (1.4) in V0 (c) ⊂ X. e
While in Theorem 1.2, we work in the radially symmetric space Xr , a setting crucial for establishing
the compactness of Palais-Smale sequences at the mountain pass level Mµ (c). However, in the Sobolev
critical case, deriving sharp energy estimates is a vital step towards recovering compactness, and this will
be achieved by choosing appropriate test functions. To address this, we slightly modify the Aubin-Talenti
bubbles associated with the Sobolev inequality, centering them at the origin and cutting them off (see
(3.21)). This allows us to devise innovative approaches to establish the rigorous inequality concerning the
energy level Mµ (c). More precisely, for small µ > 0, starting from vµ ∈ Vµ (c), we construct the following
family of functions:
Weε,t := vµ + tUε , Wε,t (x) := η N−2
4 Weε,t (ηx) ∈ Sr (c),
  2
eε,t ∥2 N+2
∥W 2
where ε > 0, t > 0 and η := c . Due to the quasilinear term
Z Z
2 2
|∇W
eε,t | dx = |∇(vµ + tUε )2 |2 dx,
RN RN

we must handle several cross terms. Using the regularity results from Remark A.2, we derive the necessary
estimates for these interaction terms, see (3.36)-(3.40). Since vµ is a solution to the perturbed problem, we
develop refined techniques to carefully control the rate at which the parameter µ tends to 0. This ultimately
establish the following relationship between mµ (c) and Mµ (c):
1 N 4
Mµ (c) ≤ mµ (c) + S 2 − δ, for 2 < q < 2 + and µ > 0 small enough.
2N N
We then apply this vital inequality to show the strong convergence in X,
e as detailed in Proposition 7.2.

7
 In comparison with (1.8), the presence of both −∆u and −u∆(u2 ) in (1.4) gives rise to two distinct
exponents, namely 2 + N4 and 4 + N4 . These exponents correspond to the L2 -critical indices for −∆u and
−u∆(u2 ), respectively. This dual nature prevents us from classifying the problem as either L2 -subcritical or
L2 -supercritical in the conventional sense. Instead, an intermediate gap emerges, i.e., 2 + N4 < q < 4 + N4 ,
which complicates the structure of the associated functional restricted on S(c). Our result in this case can
be stated as follows.

Theorem 1.4. Assume that one of the following conditions holds:

(i) N ≥ 3, q = 2 + N4 , c ∈ (0, c̄1 );

(ii) N = 3, q ∈ (2 + N4 , 4 + N4 ), c ∈ (0, c̄2 );

(iii) N ≥ 4, q ∈ (2 + N4 , 2∗ ], c ∈ (0, c̄2 );

(iv) N ≥ 4, 2∗ < q < 4 + N4 , c ∈ (0, c∗1 ) for some sufficiently small c∗1 > 0.
Then there exists τ∗ = τ∗ (c) > 0 sufficiently large such that for any τ > τ∗ , (1.4) admits a solution pair
¯ ∈ (S(c)
(ū, λ) e ∩ H 1 (RN ) ∩ L∞ (RN )) × R+ satisfying
rad

N
S2
0 < I(ū) < .
2N
Remark 1.5. In Theorem 1.4, it is essential to choose τ > 0 large enough to control the sign of the Lagrange
multiplier. However, this leads to an additional difficulty:
• In the process of analyzing the convergence uµn → ū as µn → 0, it is necessary to prove I(ū) ≥ 0 in
order to derive a contradiction to
1 N
lim M̌µn (c) < S2,
n→+∞ 2N
where M̌µn (c) is given in Lemma 4.2.

To handle this issue, we first select an appropriate Gagliardo-Nirenberg inequality by comparing the values
of 2∗ and 4 + N4 ( see (2.1) and (2.4), respectively). We then introduce an innovative classification into two
cases:
(i) ∥∇ū∥22 + ∥ū∇ū∥22 ≤ ρ∗ ,
(ii) ∥∇ū∥22 + ∥ū∇ū∥22 ≥ ρ∗ ,

where ρ∗ is defined by (7.19). By employing a contradiction argument, we exclude the first case in our
convergence result Proposition 7.2, and combine this with the fact that λ¯ ̸= 0 to establish the strong conver-
gence for the approximating solutions {uµn }. Moreover, this analysis clarifies the necessity of the condition
on c ∈ (0, c̄2 ) for q ≤ 2∗ , see (7.20) and (7.21). The argument is delicate, primarily due to the interplay
between I(ū) = M̌0 (c) > 0 and the L2 -subcritical perturbation q < 4 + N4 of the problem.
However, when N ≥ 4, 2∗ < q < 4+ N4 , no such explicit expression on the mass c > 0 is available. This is
mainly because, in the H 1 -supercritical framework, additional techniques such as interpolation inequalities
and the Sobolev inequality must be employed, making the process more intricate.

Next, we turn our attention to the L2 -supercritical case q ∈ [4 + N4 , 2 · 2∗ ). Our results are as follows.
Theorem 1.6. Assume that one of the following conditions holds:
(i) N = 3 or 4, q = 4 + N4 , c ∈ (0, c̄3 );

8
  i
(ii) N = 3 or 4, q ∈ 4 + N4 , 2(N+2)
N−2 , c > 0.

Then there exists τ∗ = τ∗ (c) > 0 sufficiently large such that for any τ > τ∗ , (1.4) admits a solution pair
¯ 0 ) ∈ (S(c)
(ū0 , λ e ∩ H 1 (RN ) ∩ L∞ (RN )) × R+ satisfying
rad

N
S2
0 < I(ū0 ) < .
2N
Moreover, ū0 is a ground state normalized solution in the sense that

e (u) = 0, u ∈ S(c)}
I(ū0 ) = inf{I(u) : dI|S(c) e

Theorem 1.7. Assume that one of the following conditions holds:

 
(i) N = 3 or 4, q ∈ 2(N+2)
N−2 , 2 · 2∗ , τ > 0;

(ii) N ≥ 5, q ∈ 4 + N4 , 2 · 2∗ , τ > 0.


¯ 0) ∈
Then there exists c∗2 > 0 sufficiently small such that for any c ∈ (0, c∗2 ), (1.4) admits solution pair (ū0 , λ
e 1 N ∞ N +
(S(c) ∩ Hrad (R ) ∩ L (R )) × R satisfying
N
S2
0 < I(ū0 ) < .
2N
Moreover, ū0 is a ground state normalized solution in the sense that

e (u) = 0, u ∈ S(c)}
I(ū0 ) = inf{I(u) : dI|S(c) e

Remark 1.8. In Theorem 1.6, when q = 4 + N4 , the limitation on c arises solely from the construction of the
mountain pass geometry. When q > 4 + N4 , although no restriction is imposed on the mass c, it is necessary
to assume that τ > 0 is large enough for the case N = 3, 4, q ≤ q∗N := 2(N+2)
N−2 . This condition stems from
the lack of a precise energy estimate and the difficulty in determining the sign of the Lagrange multiplier in
such cases. For quasilinear problems, the energy threshold for recovering compactness differs from that of
the classical problems. The energy estimate obtained in Lemma 4.4 is influenced by this special value q∗N .
The primary reason lies in the fact that, when selecting test functions for estimation, comparisons are made
based on the term −∆u, while the quasilinear term −u∆(u2 ) plays a less significant role, as shown in (1.19).
In Theorem 1.7, we can overcome these obstacles to obtain existence results for all τ > 0. Indeed, for
both ranges of q, we provide the precise energy estimates. Moreover, to prove that λ ̸= 0, we first investigate
the perturbed functional Iµ , establish some properties of Λµ (c), and then define m b µ (c) := infu∈Λµ (c) Iµ (u),
which offers an alternative minimax characterization of the mountain pass level M bµ (c) defined in Lemma
5.7. By constructing a family of functions with disjoint compact supports, we show that m b µ (c) is a non-
increasing function with respect to c > 0, see Lemma 5.5. Using the monotonicity of m b µ (c) and the implicit
function theorem, we conclude that λ ¯ 0 > 0 in a small neighborhood of the mass c > 0 for any τ > 0, leading
to the existence of a mountain pass ground state solution for the original problem.

Finally, we establish a nonexistence result.

Theorem 1.9. Let 2 < q < 2 · 2∗ , τ ≤ 0 and c ∈ (0, +∞). Then (1.4) has no solutions in S(c)
e × R+ .

9
1.2 Highlights of this paper
In contrast to the pure nonlinearity case (see [14, 25, 31]), the functional I consists of four distinct terms
N
that exhibit different scaling behaviors with respect to the dilation t 2 u(t·), which enriches significantly the
geometric structures of the constrained functional. In this subsection, we provide a thorough exposition
of the challenges we have overcome and the methods we have developed in addressing our problem. We
consider these contributions to be central to our work and among its most innovative aspects.
First of all, since we are dealing with problems involving the Sobolev critical exponent, establishing
a rigorous upper bound for the energy level is a crucial step in proving our main results. Through precise
calculations and innovative analytical techniques, we derive two key energy estimates: the first is presented
in Remark 1.3, and the second is as follows:
( N
1
M̌µ (c) ≤ 2N S 2 − δ, for 2(N+2)
N−2 < q < 2 · 2 ,
∗
1 N 4N
(1.17)
Mbµ (c) ≤ S 2 − δ, for qN < q < N−2 ,
2N

where µ > 0 and δ > 0 are sufficiently small. Here M̌µ (c), M
bµ (c) are the mountain pass levels given in
Lemma 4.2 and Lemma 5.7, respectively, and qN is defined by
(
4 + N4 , for N ≥ 5,
qN := 2(N+2) (1.18)
N−2 , for N = 3, 4.

To achieve these estimates, a pivotal issue lies in how to choose appropriate test functions, and we are
confronted with two main challenges:
• The test function defined by (3.21) is not suitable for use in the calculation of the mountain pass level.
The reason is that the truncation function in (3.21) lacks a clear and precise definition in the annular
region B2 \B1 . As a result, the function Uε does not belong to S(c), as shown in (3.26).
• Since we are employing the perturbation method, the Nehari-Pohozaev structure for the original
functional I is not well understood. This prevents us from projecting the function Uε onto S(c) first
and then onto Λµ (c) to accurately compute the mountain pass level, as was typically done in [48] for
the L2 -supercritical case. Moreover, for the L2 -subcritical case, such a projection onto the Nehari-
Pohozaev manifold is not unique, which introduces additional complications.
To address these challenges, we first construct a family of test functions as shown in equation (4.2). By
carefully selecting appropriate values of α > 0 and β > 0, we are able to show lim ∥U bε ∥2 = c, which
2 ε→0 √
c b
ensures that the function U
bε lies near the constraint S(c), see (4.3); Next, we define Vbε :=
bε ∥2 Uε to project
∥U
bε (x) onto Sr (c). Then for all q ∈ (2, 2 · 2∗ ), the following inequality holds.
U
1 N t 2  (4α+1)(N−2)  N+2
 
N (4α+1)(N 2 −4)
I0 (t 2 Vbε (tx)) ≤ S 2 + O ε 2N +t O ε 4N
2N 2
N(N+2)   N(q−2)
t N−2 (4α+1)(N+2) τt 2  N (N−2)q 
− ∗
O ε 4 − O ε2− 8 , ∀ t > 0. (1.19)
2·2 q
This inequality implies (1.17) and clarifies the necessity of the condition q > qN . For more details, see
Lemma 4.4.
Secondly, we develop two distinct convergence frameworks: one for the non-radial case through profile
decomposition, and another for radial solutions using concentration compactness. Based on the perturbed
method, for any given µ ∈ (0, 1], we study the existence of solutions to the following perturbed problem:
∗
(
−µ∆θ u − ∆u − u∆(u2 ) + λu = τ|u|q−2 u + |u|2·2 −2 u in RN ,
R 2
(1.20)
RN |u| dx = c.

10
We address the local minimizers {vµn } and the mountain pass solutions {uµn } separately, providing two
different analytical versions of concentration compactness type results in Section 7 to establish strong con-
e As previously mentioned, as µ → 0+ , these two convergence issues fundamentally relies on
vergence in X.
the Brezis-Lieb lemma for the quasilinear term as follows:
Z Z Z
|wµn |2 |∇wµn |2 dx = |w̃µn |2 |∇w̃µn |2 dx + |w̄|2 |∇w̄|2 dx + on (1), (1.21)
RN RN RN
where w̃µn := wµn − w̄. To demonstrate (1.21), we first derive the L∞ -estimate for the approximating solu-
tions by using Moser’s iteration, a process that simultaneously proves the existence of weak solutions to the
original problem, see Lemma A.3 for further details. We note that this splitting property relies essentially
on the relationship between solutions to the perturbed equations and those of the limiting equation. This
not only enables us to establish a profile decomposition theorem in X,
e but also serves as the crucial mecha-
nism for our compactness analysis within the Sobolev critical regime. Let us now briefly outline the proof
strategies for both convergence schemes.
• To prove Proposition 7.1, we employ the profile decomposition theorem developed in Theorem A.4
for the approximating solutions {vµn } to separately decompose each term in the functional Iµn . By
utilizing a novel scaling transformation as in (7.4), we project each decomposed component function
onto the set Vµn (c). Using the definition of mµn (c), we show that the perturbation tends to zero, i.e.,
µn ∥∇vµn ∥θθ → 0, and we establish strong convergence in the non-radial symmetric space X. e

• In the proof of Proposition 7.2, we discuss in the radially symmetric space Xr . For any fixed µ ∈ (0, 1],
we establish the existence of normalized solutions uµ to (1.20). This proof is standard since, in
this case, 2 · 2∗ is no longer a critical exponent for the perturbation problem. Nevertheless, we still
confront a loss of compactness in the convergence issues as µ → 0+ , due to the Sobolev critical term
∗
|u|2·2 −2 u. By utilizing the splitting property (1.21), we perform a detailed compactness analysis on
the Nehari-Pohozaev manifold Λµ (c) to obtain an alternative conclusion.
We believe that these new convergence results constitute one of the central contributions of our work, offer-
ing fresh insights for further developments in the field, particularly for quasilinear problems with Sobolev
critical growth.
Finally, the rigorous treatment of the Lagrange multiplier’s non-vanishing property serves as a key
element in our analysis, as it permits the relaxation of dimensional and exponent-range restrictions encoun-
tered in earlier studies. Unlike the classical Schrödinger equation, the Nehari-Pohozaev identity we derive
in (7.17) includes additional gradient terms, which complicates the verification of λ ̸= 0. Consequently, our
analysis distinguishes between two regimes.
• In cases where precise energy estimates are unavailable: We consider sufficiently large τ > 0 to
effectively control the mountain pass energy. This facilitates the application of our convergence
theorem, as established in Proposition 7.2, and simultaneously ensure λ > 0 through (7.17).
• In cases where precise energy estimates are available: We innovatively exploit the monotonicity of
the mountain pass level with respect to the mass c > 0. This approach enables the establishment of
λ > 0 for arbitrary τ > 0 and sufficiently small c > 0.
Previous works [31, 42, 55] focused on results for specific space dimensions. In contrast, this paper pro-
vides a comprehensive study of the existence and multiplicity of normalized solutions to equation (1.4),
covering the entire spectrum of subcritical perturbations in the interval 2 < q < 2 · 2∗ and for N ≥ 3. The
main innovation of this work is the novel analytical approaches we employ to resolve convergence issues,
which empower us to rigorously establish the non-triviality of the corresponding Lagrange multipliers.
This breakthrough unifies the analysis across the full range of q and N. We believe that this methodological
advancement provides a valuable framework for addressing related problems in quasilinear Schrödinger
equations.

11
1.3 Paper outline
This paper is organized as follows.
• Section 1 presents our main results and highlights of this paper.
• Section 2 provides some preliminaries that will be frequently used throughout the paper.

• Section 3 is devoted to the case where q ∈ (2, 2 + N4 ). We obtain a local minimizer and study the
mountain pass structure for Iµ |S(c) in Subsection 3.1 and Subsection 3.2, respectively.

• Section 4 explores the case where q ∈ [2 + N4 , 4 + N4 ), offering a precise energy estimate.

• Section 5 addresses the case where q ∈ [4 + N4 , 2 · 2∗ ). We explore properties of the associated Nehari-
Pohozaev manifold and establish an equivalent minimax characterization of the mountain pass level.
• Section 6 demonstrates the compactness of the Palais-Smale sequences obtained in Sections 3-5 and
proves the existence of mountain pass type critical points for Iµ |S(c) .
• Section 7 investigates convergence issues as µ → 0+ and completes the proof of Theorems 1.1-1.2,
Theorem 1.4, Theorems 1.6-1.7 and Theorem 1.9.
• The Appendix provides an L∞ estimate, a splitting lemma and the profile decomposition of approxi-
mating solutions, all of which play a fundamental role in proving our main results.

Throughout this paper, we make use of the following notations:

• Ls (RN ), 1 ≤ s ≤ +∞, denotes the usual Lebesgue space endowed with the norm
Z 1
s
∥u∥s := |u|s dx , 1 ≤ s < +∞, ∥u∥∞ := inf{M > 0 : |u(x)| ≤ M almost everywhere in RN };
RN

1
• W 1,s (RN ), 1 ≤ s < +∞, denotes the usual Sobolev space endowed with the norm ∥u∥W 1,s (RN ) := (∥∇u∥ss + ∥u∥ss ) s .
In particular, W 1,2 (RN ) is denoted by H 1 (RN );
1,s
• Wrad (RN ) := {u ∈ W 1,s (RN ) | u(x) = u(|x|) a.e. in RN };
• C0∞ (RN ) denotes the space of infinitely differentiable functions with compact support in RN ;
• For any x ∈ RN and r > 0, Br (x) := {y ∈ RN : |y − x| < r} denotes the open ball centered at x with radius r, and
Br := Br (0) is the open ball centered at the origin with radius r;
• The symbols ” → ” and ” ⇀ ” represent strong convergence and weak convergence, respectively;
• O (h) and o(h) mean that |O (h)| ≤ C|h| and o(h)/|h| → 0, respectively;
• C,C1 ,C2 , · · · denote positive constants whose value may change from line to line.

2 Preliminary
In this section, we present some preliminary results that will be utilized later. We begin with several
useful inequalities. The following Gagliardo-Nirenberg inequality can be found in [51]:
 
γp 1−γ p 2N
∥u∥ p ≤ C1 (p, N)∥∇u∥2 ∥u∥2 , where N ≥ 3, p ∈ 2, . (2.1)
N −2

12
 Next, we introduce the following sharp Gagliardo-Nirenberg type inequality, which is detailed in [1]:
Z  4N−p(N−2) Z  N(p−2)
C(p, N)
Z 2(N+2) 2(N+2)
p
2
|u| dx ≤
2
p−2 |u|dx |∇u| , ∀u ∈ E 1 , (2.2)
RN N+2 RN RN
∥Q p ∥1

where 2 < p < 2 · 2∗ , the constant C(p, N) is explicitly given by

p(N + 2)
C(p, N) = 4−N(p−2) N(p−2)
,
[4N − (N − 2)p] 2(N+2) [2N(p − 2)] 2(N+2)

and the Banach space E q , q ≥ 1, is defined as

E q := {u ∈ Lq (RN ) : |∇u| ∈ L2 (RN )},

equipped with the norm ∥u∥E q := ∥∇u∥2 + ∥u∥q . For more details, see [28]. Moreover, Q p is an extremal
function for the inequality (2.2) in RN and is the unique radially symmetric positive solution to the following
equation (see [46]):
p
−∆u + 1 = u 2 −1 in RN .
By substituting u with u2 in (2.2), we have the following Gagliardo-Nirenberg type inequality:
Z  4N−p(N−2)  Z  N(p−2)
C(p, N)
Z 2(N+2) 2(N+2)
|u| p dx ≤ p−2 |u|2 dx 4 |u|2 |∇u|2 . (2.3)
RN N+2 RN RN
∥Q p ∥1

For simplicity, we rewrite (2.3) as:

Z Z  4N−p(N−2) Z  N(p−2)
2(N+2) 2(N+2)
p 2 2 2
|u| dx ≤ C2 (p, N) |u| dx |u| |∇u| . (2.4)
RN RN RN

We need the Lions’ lemma as follows.

Nr
Lemma 2.1. [33, Lemma I.1] Let 1 < r ≤ ∞, 1 ≤ s < ∞ with s ̸= N−r if r < N. Assume that {un } is bounded
in Ls (RN ), {∇un } is bounded in Lr (RN ), and
Z
sup |un |s dx → 0, for some R > 0.
y∈RN B(y,R)

Nr
Then un → 0 as n → +∞ in Lβ (RN ) for any s < β < N−r .

Adapting the argument from [31, Lemma 2.1], we obtain the following result.
Lemma 2.2. Any critical point u of Iµ restricted on S(c) is contained in Λµ (c).
Define
1,θ 1
Sr (c) := S(c) ∩ Xr , Xr := Wrad (RN ) ∩ Hrad (RN ).
Let u ∈ Sr (c) be arbitrary but fixed. For any given µ ∈ (0, 1] and s ∈ R, we introduce an auxiliary functional,
similar to that in [21], to construct a Palais-Smale sequence:

I˜µ : R × Sr (c) → R,
(s, u) 7→ Iµ (H (s, u)),

13
where Ns
H (s, u(x)) := e 2 u(es x), x ∈ RN .
By a direct computation, we obtain
N(q−2)s ∗
µ e2s τe 2 eN(2 −1)s ∗
I˜µ (s, u) = eθ(1+γθ )s ∥∇u∥θθ + ∥∇u∥22 + e(N+2)s ∥u∇u∥22 − ∥u∥qq − ∗
∥u∥2·2
2·2∗ . (2.5)
θ 2 q 2·2

Note that I˜µ is of class C 1 , and a Palais-Smale sequence for I˜µ |R×Sr (c) is also a Palais-Smale sequence for
I˜µ |R×S(c) , as shown in [52, Theorem 1.28]. We also recall that the tangent space at a point u ∈ S(c) ⊂ X is
given by Z
Tu S(c) = {v ∈ X : uvdx = 0}.
RN
Following standard arguments as in [31], we have
Lemma 2.3. For u ∈ S(c) and s ∈ R, the map ϕ 7→ H (s, ϕ) from Tu S(c) to TH (s,ϕ) S(c) is a linear isomor-
phism with inverse ψ 7→ H (−s, ϕ).

3 The case 2 < q < 2 + N4

3.1 Local minimizers for the perturbed problem
Observe that for any u ∈ S(c), using (2.4) and the Sobolev inequality, we get

µ 1
Z Z Z
Iµ (u) ≥ |∇u|θ dx + |∇u|2 dx + |u|2 |∇u|2 dx
θ RN 2 RN RN
4N−q(N−2)   N(q−2)   2∗ Z  N
τC2 (q, N)c 2(N+2) 1 4 2 N−2
Z 2(N+2)
2 2 2 2
− |u| |∇u| dx − ∗
|u| |∇u| dx
q R N 2·2 S R N

2∗
 4N−q(N−2)

1 τC2 (q, N)c
 
2(N+2) N(q−2)
−1 1 4 2 2
≥ ξµ (u)  − (ξµ (u)) 2(N+2) − (ξµ (u)) N−2  , (3.1)
2 q 2 · 2∗ S

where
µ
Z Z Z
ξµ (u) := |∇u|θ dx + |∇u|2 dx + |u|2 |∇u|2 dx.
θ RN RN RN
Set
N(q − 2) 4N − q(N − 2) 2
α0 := − 1, α1 := , α2 := . (3.2)
2(N + 2) 2(N + 2) N −2
Since q < 2 + N4 , it is easy to verify that
 
2
α0 ∈ (−1, 0), α1 ∈ , 1 , α2 ∈ (0, 2].
N

We define
  2∗
1 τC2 (q, N) α1 α0 1 4 2
f (c, ρ) := − c ρ − ρ α2 .
2 q 2 · 2∗ S
Similar to the proof in [22, Lemma 2.1], we obtain the following result.

14
Lemma 3.1. For each c > 0, the function gc (ρ) := f (c, ρ) admits a unique global maximum and the maxi-
mum value satisfies

 max gc (ρ) > 0 if c < c0 ,
 ρ>0


max gc (ρ) = 0 if c = c0 ,
 ρ>0

max g (ρ) < 0 if c > c ,

c 0
ρ>0

where c0 is defined by (1.12).

Remark 3.2. Taking into account the ranges of α0 and α2 , a standard argument establishes that the function
gc (ρ) admits a unique maximum point at ρc given by
1
2∗
" #α
2 −α0
α0 τC2 (q, N)2∗ S 2 α1
ρc := − ∗ c α2 −α0 .
α2 22 −1 q

Obviously, ρ0 := ρc0 > 0 depends only on τ, q, N and is independent of µ ∈ (0, 1]. Moreover, it is straightfor-
ward to verify that Lemma 3.1 holds for both the µ = 0 case and the L2 -subcritical case, i.e., 2 < q < 4 + N4 .
Lemma 3.3. Let (c1 , ρ1 ) ∈ (0, +∞) × (0, +∞) be such that f (c1 , ρ1 ) ≥ 0. Then for any c2 ∈ (0, c1 ], we have
that  
c2
f (c2 , ρ2 ) ≥ 0 if ρ2 ∈ ρ1 , ρ1 .
c1
Proof. The proof can be derived following the arguments in [22], and is therefore omitted here.
Remark 3.4. For further reference, we note that by Lemma 3.3, it is not difficult to obtain that f (c0 , ρ0 ) = 0
and f (c, ρ0 ) > 0 for all c ∈ (0, c0 ). Moreover, for any c1 ∈ (0, c], we have
hc i
1
f (c1 , ρ) ≥ 0 for any ρ ∈ ρ0 , ρ0 .
c
Define Bρ0 := {u ∈ X : ξµ (u) < ρ0 }. Consider the set Vµ (c) given by

Vµ (c) := S(c) ∩ Bρ0 , (3.3)

and the minimization problem:

mµ (c) := inf Iµ (u) for any c ∈ (0, c0 ).

u∈Vµ (c)

The properties of mµ (c) are stated as follows.

Lemma 3.5. For any µ ∈ (0, 1] and c ∈ (0, c0 ), the following three assertions hold.
(i) mµ (c) < 0 < inf Iµ (u).
u∈∂Vµ (c)

(ii) The function c 7→ mµ (c) is a continuous mapping.

(iii) For all α ∈ (0, c), we have mµ (c) ≤ mµ (α) + mµ (c − α). Moreover, if mµ (α) or mµ (c − α) is attained,
then the inequality is strict.

15
Proof. (i) For any u ∈ ∂Vµ (c), using (3.1) and Remark 3.4, we get that

Iµ (u) ≥ ξµ (u) f (∥u∥22 , ξµ (u)) = ρ0 f (c, ρ0 ) > 0.

N
Now for any u ∈ S(c), define us (x) := s 2 u(sx), ∀s ∈ (0, +∞). Then we have

µ s2
Z Z Z
ψu (s) := Iµ (us ) = sθ(1+γθ ) |∇u|θ dx + |∇u|2 dx + sN+2 |u|2 |∇u|2 dx
θ RN 2 RN RN
1 N(2∗ −1)
Z Z
τ N(q−2) ∗
− s 2 |u|q dx − ∗
s |u|2·2 dx. (3.4)
q RN 2·2 R N

Due to the fact that N(q−2)2 < 2 and N(2∗ − 1) > N + 2, there exists s0 > 0 small enough such that
ξµ (us0 ) < ρ0 and Iµ (us0 ) = ψu (s0 ) < 0. This implies that mµ (c) < 0.

(ii) Let c ∈ (0, c0 ) be arbitrary and {cn } ⊂ (0, c0 ) be such that cn → c. From the definition of mµ (cn ) and
the fact that mµ (cn ) < 0, for any sufficiently small ε > 0, there exists un ∈ Vµ (cn ) such that

Iµ (un ) ≤ mµ (cn ) + ε and Iµ (un ) < 0. (3.5)

1 1
− N+2 c
We set vn (x) := tnN+2 un (tn x) for tn = cn . Consequently, vn ∈ S(c). Moreover, if cn ≥ c, which
means cn → c+ , we have
µ µ
∥∇vn ∥θθ + ∥∇vn ∥22 + ∥vn ∇vn ∥22 ≤ ∥∇un ∥θθ + ∥∇un ∥22 + ∥un ∇un ∥22 < ρ0 ,
θ θ
which implies that vn ∈ Vµ (c). On the other hand, −
cn
 taking cn < c such that cn → c , by Lemma 3.3, we
can see that f (cn , ρ) ≥ 0 for any ρ ∈ c ρ0 , ρ0 . Using Remark 3.4 again, we get that f (cn , ξµ (un )) <
0, which implies that ξµ (un ) < ccn ρ0 and
µ µ
∥∇vn ∥θθ + ∥∇vn ∥22 + ∥vn ∇vn ∥22 ≤ tn ∥∇un ∥θθ + tn ∥∇un ∥22 + tn ∥un ∇un ∥22 < ρ0 .
θ θ
This implies that vn ∈ Vµ (c). Therefore, we deduce that

mµ (c) ≤ Iµ (vn ) = Iµ (un ) + (Iµ (vn ) − Iµ (un )) = Iµ (un ) + on (1),

which together with (3.5) implies that

mµ (c) ≤ mµ (cn ) + ε + on (1).

Reversing the argument we can deduce similarly that mµ (cn ) ≤ mµ (c) + ε + on (1). Therefore, (ii)
holds.

(iii) It is suffices to prove that for any fixed α ∈ (0, c), the following holds
 ci
mµ (tα) ≤ tmµ (α), ∀ t ∈ 1, . (3.6)
α
To prove (3.6), we first note that by point (i), for any ε > 0 sufficiently small, there exists u ∈ Vµ (α)
such that

Iµ (u) ≤ mµ (α) + ε and Iµ (u) < 0. (3.7)

16
 From Lemma 3.3, we know that f (α, ρ) ≥ 0 for any ρ ∈ [ αc ρ0 , ρ0 ]. Consequently, using (3.7), we
1 1
have ξµ (u) < αc ρ0 . Next, we define a new function v(x) := t N+2 u(t − N+2 x). Then

ξµ (v) < tξµ (u) < ρ0 .

This implies that v ∈ Vµ (tα). Therefore, we obtain

mµ (tα) ≤ Iµ (v) < tIµ (u) ≤ t(mµ (α) + ε).

Since ε > 0 is arbitrary, we conclude that mµ (tα) ≤ tmµ (α) If mµ (α) is reached, then there exists a
function u0 ∈ Vµ (α) such that Iµ (u0 ) = mµ (α) and Iµ (u0 ) < 0. In this case, we can take ε = 0 in (3.7),
and the strict inequality follows directly.

The following result is a direct consequence of applying Lemma 2.1.

Lemma 3.6. For any µ ∈ (0, 1] and c ∈ (0, c0 ), let {un } ⊂ Bρ0 be such that ∥un ∥22 → c and Iµ (un ) → mµ (c).
Then, there exist δ > 0 and a sequence {yn } ⊂ RN such that
Z
|un |2 dx ≥ δ > 0, for some R > 0.
BR (yn )

Lemma 3.7. For any µ ∈ (0, 1] and c ∈ (0, c0 ), if a sequence {un } ⊂ Bρ0 satisfies ∥un ∥22 → c and Iµ (un ) →
mµ (c), then there exists vµ ∈ Vµ (c) such that, up to translation, un → vµ in X.
Proof. We employ the concentration-compactness principle to analyze the sequences {un }. Since {un } ⊂
Bρ0 , it follows that {un } is bounded in X and {un ∇un } is bounded in L2 (RN )N . Define ωn := un (· + yn ).
Then, by Lemma 3.6 and the Rellich compactness theorem, there exists 0 ̸= vµ ∈ X such that, up to a
subsequence,
in W 1,θ (RN );

ωn ⇀ vµ

in H 1 (RN );

ω ⇀ v
n µ
(3.8)

 ωn ∇ωn ⇀ ω̃ in L2 (RN )N ;
 β Nθ
ωn → vµ in Lloc (RN ), β ∈ [1, N−θ

).
Next, we claim that ω̃ = vµ ∇vµ . To see this, observe that since ∥∇ω2n ∥22 = 4 RN |ωn |2 |∇ωn |2 dx is bounded,
R

we have ω2n ⇀ ω̄ in H 1 (RN ). By the uniqueness of weak limits, it follows that ω̄ = v2µ . Consequently, we
conclude that ω̃ = vµ ∇vµ , and the claim holds.
Denote c1 = ∥vµ ∥22 > 0. Clearly, c1 ≤ c. In what follows, we are going to prove that the dichotomy does
not occur, i.e., c1 = c. Arguing by contradiction, we may assume that 0 < c1 < c. First of all, for any ε > 0,
there exists R1 = R1 (ε) > 0 such that
Z
c1 − ε < |vµ |2 dx ≤ c1 .
BR1

Thus, for n sufficiently large, we have

Z Z
c1 − ε < |ωn |2 dx ≤ c1 and |ωn |2 dx ≤ ε. (3.9)
BR1 2BR1 \BR1

Let R̃ = max{R, R1 }, where R is given in Lemma 3.6. Choose a smooth function ηR̃ defined by
(
1, 0 ≤ |t| ≤ R̃,
ηR̃ (t) :=
0, |t| ≥ 2R̃,

17
with |∇ηR̃ | ≤ R̃2 . Set zn (x) := ηR̃ (|x − yn |)un (x) and wn (x) := (1 − ηR̃ (|x − yn |)) un (x). Clearly, zn and wn
belong to X and un = zn + wn . Moreover, we have
Z Z
2
lim inf |zn | dx = lim inf |un |2 dx ≥ δ > 0.
n→+∞ BR̃ (yn ) n→+∞ BR̃ (yn )

From this, we derive that

µ
Z  
Iµ (un ) = Iµ (zn ) + Iµ (wn ) + |∇un |θ − |∇zn |θ − |∇wn |θ dx
θ B2R̃ (yn )\BR̃ (yn )
1
Z
|∇un |2 − |∇zn |2 − |∇wn |2 dx


+
2 B2R̃ (yn )\BR̃ (yn )
Z
|un |2 |∇un |2 − |zn |2 |∇zn |2 − |wn |2 |∇wn |2 dx


+
B2R̃ (yn )\BR̃ (yn )
Z
τ
− (|un |q − |zn |q − |wn |q ) dx
q B2R̃ (yn )\BR̃ (yn )
1
Z  
∗ ∗ ∗
− |un |2·2 − |zn |2·2 − |wn |2·2 dx. (3.10)
2 · 2∗ B2R̃ (yn )\BR̃ (yn )

We can now verify that

Z   Z  
|∇un |θ − |∇zn |θ − |∇wn |θ dx ≥ 1 − ηθR̃ − (1 − ηR̃ )θ |∇un |θ dx − oR̃ (1),
B2R̃ (yn )\BR̃ (yn ) B2R̃ (yn )\BR̃ (yn )
Z Z (3.11)
2 2 2
1 − η2R̃ − (1 − ηR̃ )2 2



|∇un | − |∇zn | − |∇wn | dx ≥ |∇un | dx − oR̃ (1),
B2R̃ (yn )\BR̃ (yn ) B2R̃ (yn )\BR̃ (yn )
(3.12)
Z
|un |2 |∇un |2 − |zn |2 |∇zn |2 − |wn |2 |∇wn |2 dx


B2R̃ (yn )\BR̃ (yn )
Z
1 − η4R̃ − (1 − ηR̃ )4 |un |2 |∇un |2 dx − oR̃ (1).


≥ (3.13)
B2R̃ (yn )\BR̃ (yn )

Moreover, using (3.9) and the interpolation inequality, we obtain

Z
(|un |q − |zn |q − |wn |q ) dx ≤ 3ε, (3.14)
B2R̃ (yn )\BR̃ (yn )

and Z  
∗ ∗ ∗
|un |2·2 − |zn |2·2 − |wn |2·2 dx ≤ 3ε, (3.15)
B2R̃ (yn )\BR̃ (yn )

for n large enough. Substituting (3.11)-(3.15) into (3.10), we deduce that

Iµ (un ) ≥ Iµ (zn ) + Iµ (wn ) − on (1) − oR̃ (1). (3.16)

By a direct computation, we have

∥wn ∥22 = ⟨un (x) − ηR̃ (|x − yn |)un (x), un (x) − ηR̃ (|x − yn |)un (x)⟩
= ⟨un , un ⟩ − 2 ⟨un , ηR̃ (|x − yn |)un ⟩ + ⟨ηR̃ (|x − yn |)un , ηR̃ (|x − yn |)un ⟩ .
2 (RN ) that as n → +∞,
It follows from ωn → vµ in Lloc
Z Z
∥wn ∥22 → c − 2 ηR̃ v2µ dx + η2R̃ v2µ dx := δR̃ > 0.
RN RN

18
Moreover, observe that
µ
Z Z Z
ξµ (wn ) ≤ oR̃ (1) + |∇ωn |θ dx + |∇ωn |2 dx + |ωn |2 |∇ωn |2 dx,
θ RN \BR̃ RN \BR̃ RN \BR̃

which implies that {wn } ⊂ Bρ0 for R̃ sufficiently large. Hence, in view of Lemma 3.5 (ii) we get
lim inf Iµ (wn ) ≥ lim inf mµ (∥wn ∥22 ) = mµ (δR̃ ).
n→+∞ n→+∞

Taking the limit as n → +∞ in (3.16) and using (3.8), we infer that

mµ (c) ≥ mµ (δR̃ ) + Iµ (ηR̃ vµ ) − oR̃ (1).
Then, letting R̃ → +∞, we deduce that
mµ (c) ≥ mµ (c − c1 ) + Iµ (vµ ). (3.17)
By the lower semi-continuity we have vµ ∈ Vµ (c1 ), which implies that Iµ (vµ ) ≥ mµ (c1 ). Suppose Iµ (vµ ) >
mµ (c1 ). Then, combining (3.17) and Lemma 3.5 (iii), we obtain
mµ (c) > mµ (c − c1 ) + mµ (c1 ) ≥ mµ (c − c1 + c1 ) = mµ (c),
which is impossible. Therefore, we must have Iµ (vµ ) = mµ (c1 ), meaning that vµ is a local minimizer of Iµ
on Vµ (c1 ). Applying Lemma 3.5 (iii) with the strict inequality, we deduce from (3.17) that
mµ (c) ≥ mµ (c − c1 ) + Iµ (vµ ) = mµ (c − c1 ) + mµ (c1 ) > mµ (c − c1 + c1 ) = mµ (c),
which is again a contradiction. Hence, it follows that ∥vµ ∥22 = c. Consequently, ωn → vµ in L2 (RN ). Using
the lower semi-continuity and interpolation, we infer that
mµ (c) = lim Iµ (un ) = lim Iµ (ωn ) ≥ Iµ (vµ ) ≥ mµ (c).
n→+∞ n→+∞

This implies that ∥∇ωn ∥22 → ∥∇vµ ∥22 , ∥∇ωn ∥θθ → ∥∇vµ ∥θθ and ∥ωn ∇ωn ∥22 → ∥vµ ∇vµ ∥22 . Therefore, un → vµ
in X.
Building on the preceding preliminary results, we conclude that
Lemma 3.8. For any fixed µ ∈ (0, 1] and c ∈ (0, c0 ), the functional Iµ has a critical point vµ ∈ Vµ (c)
satisfying
˜ µ vµ = 0,
Iµ (vµ ) = mµ (c), Iµ′ (vµ ) + λ
˜ µ ∈ R.
for some λ
Proof. For any sequence {un } ⊂ Vµ (c) satisfying Iµ (un ) → mµ (c), we deduce that, up to translation, un →
vµ ∈ Vµ (c) in X, as shown by Lemma 3.7. In particular, using the fact that mµ (c) < 0 and the argument
presented in Lemma 3.7, we conclude that vµ ∈ Vµ (c) is a local minimizer for Iµ on Vµ (c). This completes
our proof.
Remark 3.9. For further reference, we note that vµ ∈ Vµ (c) obtained in the above lemma is positive and
radially symmetric non-increasing. In fact, let v∗µ denote the Schwartz rearrangement of |vµ |. Consequently,
v∗µ is a positive, radially symmetric, and non-increasing function. Moreover, it is straightforward to verify
that Lemma 4.3 in [14] also holds for the Sobolev critical case, that is,
∥∇v∗µ ∥θθ ≤ ∥∇vµ ∥θθ , ∥∇v∗µ ∥22 + ∥v∗µ ∇v∗µ ∥22 ≤ ∥∇vµ ∥22 + ∥vµ ∇vµ ∥22 , ∥v∗µ ∥22 = ∥vµ ∥22 .
Thus, v∗µ ∈ Vµ (c) and Iµ (v∗µ ) = Iµ (vµ ), which implies that mµ (c) is reached by v∗µ satisfying
∗
−µ∆θ v∗µ − ∆v∗µ − v∗µ ∆(v∗2 ˜ ∗ ∗ q−2 ∗
vµ + |v∗µ |2·2 −2 v∗µ in RN ,
µ ) + λµ vµ = τ|vµ |

˜ µ ∈ R. Hereafter, without loss of generality, we still denote v∗ by vµ .

for some λ µ

19
3.2 The mountain pass structure and Palais-Smale sequences
Starting from the previous local minimizer uµ,c := vµ obtained in Lemma 3.8, we can derive the follow-
ing mountain pass structure for Iµ restricted on Sr (c).
Lemma 3.10. For any c ∈ (0, c0 ) and fixed µ ∈ (0, 1], there exists ζ0 (c) > 0 independent of µ such that
Mµ (c) := inf max Iµ (γ(t)) ≥ ζ0 (c) > sup max{Iµ (γ(0)), Iµ (γ(1))},
γ∈Γµ (c) t∈[0,1] γ∈Γµ (c)

where
Γµ (c) := {γ ∈ C ([0, 1], Sr (c)) : γ(0) = uµ,c , Iµ (γ(1)) < 2mµ (c)}.
Proof. Define
V0 (c) := {u ∈ S(c) : ∥∇u∥22 + ∥u∇u∥22 < ρ0 }.
Then for any u ∈ ∂V0 (c), we have
4N−q(N−2)

1 τC2 (q, N)c 2(N+2)


N(q−2) −1
Iµ (u) ≥ ∥∇u∥22 + ∥u∇u∥22 − ∥∇u∥22 + ∥u∇u∥22 2(N+2)
2 q
  2∗ 
1 4 2 2

2
2 N−2
− ∥∇u∥2 + ∥u∇u∥2
2 · 2∗ S
= ρ0 f (c, ρ0 ) := ζ0 (c) > 0.
For any γ ∈ Γµ (c), we have γ(0) = uµ,c ∈ Vµ (c) ⊂ V0 (c) and Iµ (γ(1)) < 2mµ (c), which implies that γ(1) ∈
/
Vµ (c). Moreover, by the continuity of γ(t) on [0, 1], there exists t0 ∈ (0, 1) such that
γ(t0 ) ∈ ∂V0 (c) and max Iµ (γ(t)) ≥ Iµ (γ(t0 )) ≥ ζ0 (c),
t∈[0,1]

which implies that the conclusion is valid.

We introduce the family of paths
Γ̃µ (c) := {γ̃ ∈ C ([0, 1], R × Sr (c)) : γ̃(0) = (0, uµ,c ), I˜µ (γ̃(1)) < 2mµ (c)},
and the associated minimax value
M̃µ (c) := inf max I˜µ (γ̃(t)),
γ̃∈Γ̃µ (c) t∈[0,1]

where I˜µ is given by (2.5). Then we have the following result:

Lemma 3.11. Let c ∈ (0, c0 ). Then Mµ (c) = M̃µ (c).
Proof. On the one hand, for any γ ∈ Γµ (c), let γ̃(t) := (0, γ(t)). It is easy to observe that
γ̃ ∈ Γ̃µ (c), and I˜µ (γ̃(t)) = I˜µ (0, γ(t)) = Iµ (γ(t)),
which implies that M̃µ (c) ≤ Mµ (c). On the other hand, for any γ̃ ∈ Γ̃µ (c), we have H ◦ γ̃ ∈ Γµ (c). By Lemma
3.10, there exists ζ˜ µ (c) > 0 such that

inf max I˜µ (γ̃(t)) ≥ ζµ (c) > ζ˜ µ (c) ≥ sup max{I˜µ (γ̃(0)), I˜µ (γ̃(1))},
γ̃∈Γ̃µ (c) t∈[0,1] γ̃∈Γ̃µ (c)

where ζµ (c) := inf Iµ (u) > 0. This implies that Mµ (c) ≤ M̃µ (c). Combining these two results, we
u∈∂Vµ (c)
conclude that Mµ (c) = M̃µ (c). This completes our proof.

20
 Following the Ghoussoub’s minimax approach introduced in [19], we establish a result showing the
existence of a Palais-Smale sequence of I˜µ at level M̃µ (c) with an additional property.
Lemma 3.12. For any µ ∈ (0, 1] and c ∈ (0, c0 ), there exists a sequence {un } ⊂ Sr (c) satisfying

Iµ (un ) → Mµ (c) > 0, ∥Iµ′ |Sr (c) (un )∥X ∗ → 0, Qµ (un ) → 0.

Proof. We will adopt the notation from [19, Theorem 5.2]. Define the following sets:

Λ−
µ (c) := {u ∈ Λr (c) : Iµ (u) < 0}, Eµc := {u ∈ Sr (c) : Iµ (u) < 2mµ (c)}.
Let
A = γ̃([0, 1]), X = R × Sr (c), F = {γ̃([0, 1]) : γ̃ ∈ Γ̃µ (c)},
B = ({0} × Λ− c ˜
µ (c)) ∪ ({0} × Eµ ), F = {(s, u) ∈ R × Sr (c) : Iµ (s, u) ≥ M̃µ (c)}.

It is straightforward to verify that F is a homotopy stable family of compact subsets of X with an extended
closed boundary B. Moreover, using the definition of M̃µ (c) and I˜µ (s, u), we can conclude that F is a dual
set for F with φ = I˜µ and d˜ = M̃µ (c). Consequently, by [19, Theorem 5.2], for the minimizing sequence
{yn = (0, βn )} ⊂ Γ̃µ (c) with βn ≥ 0 a.e. in RN , there exists a Palais-Smale sequence {(sn , wn )} ⊂ R × Sr (c)
for I˜µ |R×Sr (c) at the level M̃µ (c) > 0. That is, as n → +∞,

∂s I˜µ (sn , wn ) → 0 and ∥∂u I˜µ (sn , wn )∥(Twn Sr (c))∗ → 0, (3.18)

with the additional property that

|sn | + distX (wn , βn ([0, 1])) → 0. (3.19)
By (3.19), we know that {sn } is bounded. Moreover, it follows from (3.18) that Qµ (H (sn , wn )) → 0 and
that
Z Z Z
µeθ(1+γθ )sn |∇wn |θ−2 ∇wn ∇φdx + e2sn ∇wn ∇φdx + 2e(N+2)sn |∇wn |2 wn φdx
RN RN RN
Z Z Z
N(q−2)sn ∗ −1)s ∗ −2
+ 2e(N+2)sn |wn |2 ∇wn ∇φdx − τ e 2 |wn |q−2 wn φdx − eN(2 n
|wn |2·2 wn φdx
RN RN RN
= o(1)∥φ∥X , ∀φ ∈ Twn Sr (c),

which implies that as n → +∞,

⟨Iµ′ (H (sn , wn )), H (sn , φ)⟩ = o(1)∥φ∥X = o(1)∥H (sn , φ)∥X , ∀φ ∈ Twn Sr (c). (3.20)

Let un := H (sn , wn ) ∈ Sr (c). Then taking into account (3.20) and using Lemma 2.3, we deduce that {un } ⊂
Sr (c) is a Palais-Smale sequence for Iµ restricted on Sr (c) at the level Mµ (c), with Qµ (un ) → 0.
At this stage, since we are dealing with a critical exponent regime, it is necessary to derive a refined
energy estimate, which is crucial for obtaining the compactness of the Palais-Smale sequence as µ → 0+ .
Following the approach of Brezis-Nirenberg [7], we define a function uε by
N−2
(N(N − 2)ε) 8
uε (x) := N−2 , ε > 0.
(ε + |x|2 ) 4

N+2
Clearly, the function vε := u2ε solves the equation −∆vε = vεN−2 . Let ξ ∈ C0∞ (RN ) be a radially non-increasing
cut-off function such that ξ ≡ 1 in B1 and ξ ≡ 0 in RN \B2 . We then define

Uε (x) := ξ(x)uε (x). (3.21)

21
As ε → 0, the function Uε satisfies the following estimates:
Z Z
N N−2
4 |Uε |2 |∇Uε |2 dx = |∇(Uε2 )|2 dx = S 2 + O (ε 2 ), (3.22)
RN RN
Z
∗ N N
|Uε |2·2 dx = S 2 + O (ε 2 ). (3.23)
RN
Z
N−2
|∇Uε |2 dx = O (ε 4 | ln ε|). (3.24)
RN

N − (N−2)r
Z O (ε 2

 8 ), 2∗ < r < 2 · 2∗ ;
r N
|Uε | dx = O (ε 4 | ln ε|), r = 2∗ ; (3.25)
RN
O (ε (N−2)r

), 1 ≤ r < 2∗ .

8

In particular, we have Z
N−2
|Uε |2 dx = O (ε 4 ). (3.26)
RN
In the remainder of this subsection, we will prove the following lemma.

Lemma 3.13. Let q ∈ (2, 2 + N4 ), τ > 0 and c ∈ (0, c0 ). Then there exist 0 < µ0 < 1 and δ0 > 0 such that
N
S2
Mµ (c) ≤ mµ (c) + − δ′ for all µ ∈ (0, µ0 ).
2N
Proof. Let uµ,c be a local minimizer of Iµ on Vµ (c). Then, by Lemma 3.7, we have

∥uµ,c ∥22 = c, Iµ (uµ,c ) = mµ (c), uµ,c (x) > 0, ∀ x ∈ RN , (3.27)

and
⟨Iµ′ (uµ,c ) + λc uµ,c , φ⟩ = 0, for any φ ∈ X. (3.28)
Using φ = uµ,c as a test function in (3.28), we obtain
Z Z Z Z
µ |∇uµ,c |θ dx + |∇uµ,c |2 dx + 4 |uµ,c |2 |∇uµ,c |2 dx + λc |uµ,c |2 dx
RN RN RN RN
Z Z
∗
−τ |uµ,c |q dx − |uµ,c |2·2 dx = 0. (3.29)
RN RN

4N
Combining (3.29) and Qµ (uµ,c ) = 0, it follows from θ > N+2 that

τ(4N + 2q − Nq) N −2
∥uµ,c ∥qq ≤ λc ∥uµ,c ∥22 + ∥∇uµ,c ∥22 . (3.30)
q(N + 2) N +2

Moreover, using φ = Uε as a test function in (3.28), we have

Z Z Z Z
2
µ |∇uµ,c |θ−2
∇uµ,c ∇Uε dx + ∇uµ,c ∇Uε dx + 2 |uµ,c | ∇uµ,c ∇Uε dx + 2 |∇uµ,c |2 uµ,cUε dx
RN RN RN RN
Z Z Z
2·2 ∗ −1
+ λc uµ,cUε dx − τ uq−1
µ,c Uε dx − uµ,c Uε dx = 0. (3.31)
RN RN RN

For any ε > 0 and t > 0, we define

N−2
eε,t := uµ,c + tUε ,
W Wε,t (x) := η 4 eε,t (ηx),
W

22
   2
eε,t ∥2 N+2
∥W 2
where η = c . Then we immediately get that

2−N (N−2)q−4N
∥∇Wε,t ∥22 = η 2 eε,t ∥22 , ∥∇Wε,t
∥∇W 2 2
∥2 = ∥∇W 2 2
eε,t ∥2 , ∥Wε,t ∥qq = η 4 eε,t ∥qq ,
∥W (3.32)
∗ ∗ − N+2
∥Wε,t ∥2·2 e 2·2 2
2·2∗ = ∥Wε,t ∥2·2∗ , ∥Wε,t ∥2 = η
2 eε,t ∥22 = c.
∥W (3.33)
In view of (3.32)-(3.33), we have

1  2−N Z t 2 2−N
Z Z
2−N
I0 (Wε,t ) ≤ I0 (uµ,c ) + η 2 −1 |∇uµ,c |2 dx + tη 2 ∇uµ,c ∇Uε dx + η 2 |∇Uε |2 dx
2 RN R3 2 RN
t4 t2
Z Z Z
+ |∇Uε2 |2 dx + t 2 |∇(uµ,cUε )|2 dx + ∇u2 ∇U 2 dx
4 RN RN 2 RN µ,c ε
Z Z
+t ∇u2µ,c ∇(uµ,cUε )dx + t 3 ∇Uε2 ∇(uµ,cUε )dx
RN RN
Z 1 2·2∗
Z Z
τ (N−2)q−4N τ (N−2)q−4N q q ∗
+ 1−η 4 uqµ,c dx − η 4 t Uε dx − ∗
t Uε2·2 dx
q RN q RN 2·2 R N

:= L1 (t).

It is not difficult to verify that, uniformly for small ε > 0, L1 (t) → −∞ as t → +∞ and L1 (t) → I0 (uµ,c )
as t → 0 due to η → 1. Hence, there exist ε0 > 0 and 0 < t1 < t2 < +∞ such that for any t ∈ (0,t1 ) or
t ∈ (t2 , +∞) and ε ∈ (0, ε0 ], we have
N
S2
I0 (Wε,t ) < I0 (uµ,c ) + .
2N
Next, we address the case t1 ≤ t ≤ t2 . By (3.26), we have

∥uµ,c + tUε ∥22 2t

Z  N−2 
N+2
η 2 = = 1+ uµ,cUε dx + t 2 O ε 4 , as ε → 0.
c c RN

Utilizing the Taylor expansion, we deduce that, as ε → 0

N+2 (N−2)q−4Nt 4N − (N − 2)q
Z
(N−2)q−4N
 N−2 
1−η 4 = 1−η 2 2(N+2) = uµ,cUε dx + t 2 O ε 4 . (3.34)
c N +2 RN
t 2(2 − N)
Z  N−2 
2−N N+2 2−N
η 2 −1 = η 2 N+2 −1 = uµ,cUε dx − t 2 O ε 4 . (3.35)
c N + 2 RN
Moreover, since the bounded function uµ,c is a radial critical point of Iµ , by (3.24), (3.25), (3.26) and the
regularity of uµ,c (see Remark A.2), as ε → 0, we have the following estimates:
Z  N−2 
u2µ,c |∇Uε |2 dx = O ε 4 | ln ε| . (3.36)
N
ZR  N−2 
|∇uµ,c |2Uε2 dx = O ε 4 . (3.37)
RN

Z  N−2 p 
∇u2µ,c ∇Uε2 dx = O ε 4 | ln ε| . (3.38)
RN

23
 "Z #
2 3ξ2 (r)ξ′ (r)rN−1 ξ3 (r)rN
Z 2
3(2 − N)
Z
3(N−2)
Uε2 ∇uµ,c ∇Uε dx = CB3N ωN ε 8
3(N−2)
dr + 3N−2 dr
R N 0 (ε + r2 ) 4 2 0 (ε + r2 ) 4
 N+2
O (ε 8 ),
 if N > 4;
3
:= ζε (N) = O (ε 4 | ln ε|), if N = 4; (3.39)
 3
O (ε 8 ), if N = 3.


!
√2
sN+1
Z Z
3(N−2) 2−N
|∇Uε |2 uµ,cUε dx ≤ CB3N ωN ε
ε
8 O (1) + ε 4
3N+2 ds
RN 0 (1 + s2 ) 4
 N−2 
=O ε 8 . (3.40)

 !1 
√2 θ
sθ+N−1
Z Z
(N−2)θ N(2−θ) ε
|∇uµ,c |θ−2 ∇uµ,c ∇Uε dx ≤ CBθN ωN ε 8θ O (1) + ε 4 ds 
(N+2)θ
RN 0 (1 + s2 ) 4
 4N−θ(N+2) 
= O ε 8θ . (3.41)

Z  N−2 
∗ −1
uµ,cUε2·2 dx = O ε 8 . (3.42)
RN

From (3.22)-(3.26), (3.27) and (3.30)-(3.42), we derive that

1 1 1
Z Z Z Z
τ ∗
I0 (Wε,t ) ≤ |∇uµ,c |2 dx + |∇u2µ,c |2 dx − uqµ,c dx − 2·2
uµ,c dx
2 RN 4 RN q RN 2 · 2∗ RN
Z Z Z Z
+ t2 u2µ,c |∇Uε |2 dx + t 2 |∇uµ,c |2Uε2 dx + t 2 ∇u2µ,c ∇Uε2 dx + 2t 3 Uε2 ∇uµ,c ∇Uε dx
RN RN RN RN
Z Z Z
+ 2t 3 |∇Uε |2 uµ,cUε dx − λct uµ,cUε dx − µt |∇uµ,c |θ−2 ∇uµ,c ∇Uε dx
RN RN RN
1  2−N Z  2−N  Z
+ η 2 −1 |∇uµ,c |2 dx + η 2 − 1 t ∇uµ,c ∇Uε dx
2 RN R3
Z  Z
τ (N−2)q−4N (N−2)q−4N
 
+ 1−η 4 uqµ,c dx + τ 1 − η 4 t uq−1
µ,c Uε dx
q RN RN
∗
t 2 2−N t4 t 2·2
Z Z Z Z
∗ ∗ ∗
+ η 2 |∇Uε | dx + 2
|∇Uε2 |2 dx − ∗
Uε2·2 dx − t 2·2 −1 uµ,cUε2·2 −1 dx
2 RN 4 N 2 · 2 N N
 N−2  R  N−2   N−2R p  R
 N−2 
≤ I0 (uµ,c ) + t 2 O ε 4 | ln ε| + t 2 O ε 4 + t 2 O ε 4 | ln ε| + 2t 3 ζε (N) + 2t 3 O ε 8
 4N−θ(N+2)   N−2   N−2 p 
− µt O ε 8θ + t 2O ε 4 − t 2O ε 4 | ln ε|
 N−2  t4 N  N−2  t 2·2∗ N ∗
 N  N−2 
2 4 2 − t 2·2 O ε 2 2·2∗ −1
+ t O ε | ln ε| + S + t O ε 2 −
4 2 S − t O ε 8 .
4 2 · 2∗
(3.43)

It is straightforward to observe that

∗
t 4 t 2·2 1
− ≤ , ∀t > 0.
4 2 · 2∗ 2N

24
 α
 
θ(N+2)−4N θ(N+2)−3N−2
Moreover, by choosing µ = ε θ for some α > 0 satisfying α ∈ 8 , 8 , it follows from
(3.43) that there exist small ε0 > 0 and δ′ > 0 such that
N
S2
sup I0 (Wε0 ,t ) ≤ I0 (uµ,c ) + − 2δ′ .
t>0 2N

Clearly, Wε0 ,t ∈ Sr (c) for all t ≥ 0, Wε0 ,0 = uµ,c and I(Wε0 ,t ) ≤ Iµ (Wε0 ,t ) < 2mµ (c) for sufficiently large t > 0.
Therefore, there exists a sufficiently large t0 > 0 such that

Iµ (Wε0 ,t0 ) < 2mµ (c).

Let γ(t) := Wε0 ,tt0 . Then γ(t) ∈ Γµ (c). Moreover, observe that there exist 0 < T1 < T2 such that

sup Iµ (Wε,t ) = sup Iµ (Wε,t ), sup I(Wε,t ) = sup I(Wε,t ).

t>0 t∈[T1 ,T2 ] t>0 t∈[T1 ,T2 ]

From this, we deduce that

N N N
S2 ′ S2 ′ S2
Mµ (c) ≤ sup Iµ (Wε0 ,t ) ≤ C(ε0 )µ + I0 (uµ,c ) + − 2δ ≤ Iµ (uµ,c ) + − δ = mµ (c) + − δ′ .
t>0 2N 2N 2N

This completes the proof.

4 The case 2 + N4 ≤ q < 4 + N4

4.1 Mountain pass structure
In this section, we investigate the case 2 + N4 ≤ q < 4 + N4 . We first show that the functional Iµ exhibits
the mountain pass geometry.
Lemma 4.1. Let µ ∈ (0, 1] and τ > 0. Assume that one of the following conditions holds:
(i) N ≥ 3, q = 2 + N4 , c ∈ (0, c̄1 ).

(ii) N = 3, 2 + N4 < q < 4 + N4 , c > 0.

(iii) N ≥ 4, 2 + N4 < q ≤ 2∗ , c > 0.

(iv) N ≥ 4, 2∗ < q < 4 + N4 , c > 0.

Then there exists Kµ = Kµ (c, τ, q, N) > 0 small enough such that

0< sup Iµ (u) < inf Iµ (u), (4.1)

u∈Aµ (Kµ ,c) u∈∂Aµ (2Kµ ,c)

where
µ 1
Z
Aµ (Kµ , c) := {u ∈ S(c) : ∥∇u∥θθ + ∥∇u∥22 + |u|2 |∇u|2 dx < Kµ },
θ 2 RN
µ 1
Z
2
∂Aµ (2Kµ , c) := {u ∈ S(c) : ∥∇u∥θ + ∥∇u∥2 +
θ
|u|2 |∇u|2 dx = 2Kµ }.
θ 2 RN

25
Proof. For any u ∈ Aµ (Kµ , c), v ∈ ∂Aµ (2Kµ , c), we have

µ 1 µ 1
Z Z
∥∇u∥θθ + ∥∇u∥22 + |u|2 |∇u|2 dx < Kµ , ∥∇v∥θθ + ∥∇v∥22 + |v|2 |∇v|2 dx = 2Kµ .
θ 2 RN θ 2 RN

Case 1: N = 3, 2 + N4 ≤ q < 4 + N4 or N ≥ 4, 2 + N4 ≤ q ≤ 2∗ . Using (2.1) and the Sobolev inequality, we

get that
1
Z Z
τ ∗
Iµ (v) − Iµ (u) ≥ Kµ − |v|q dx − |v|2·2 dx
q RN 2 · 2∗ RN
q 2q−N(q−2)   2∗ Z
 N(q−2)  N
τC (q, N)c
Z
4 1
2 4 2 4
2 2
N−2
≥ Kµ − 1 |∇v| dx − ∗
|v| |∇v| dx
q RN 2·2 S RN
N(q−2) q 2q−N(q−2) 2   2∗
2 2 τC1 (q, N)c 4 N(q−2)
2 N−2 4 2 N
≥ Kµ − Kµ 4 − ∗ KµN−2 .
q 2 S

It follows from 0 < c < c̄1 when q = 2 + N4 and N(q−2)

4 > 1 when q > 2 + N4 that there exists a sufficiently
small Kµ > 0 such that Iµ (v) − Iµ (u) > 0, which gives (4.1).
Case 2: N ≥ 4, 2∗ < q < 4 + N4 . Using the interpolation inequality and the Sobolev inequality, we have

Z  2·2∗∗−q Z  q−2∗ ∗   2∗ Z  N
τ 2∗ 2
2·2∗ 2 1 4 2 2 2
N−2
Iµ (v) − Iµ (u) ≥ Kµ − |v| dx |v| dx − |v| |∇v| dx
q RN RN 2 · 2∗ S RN
2·2∗ −q q−2 ∗ 2 ∗
∗ Z  N
2q−2 τ
 Z    Z
2
2
2 2
2 1 4 2 2 2
N−2
≥ Kµ − 2∗ |∇v| dx |v| |∇v| dx − ∗
|v| |∇v| dx
S2q RN RN 2·2 S RN
2∗ +q 2   2 ∗
2 2 τ N 2 N−2 4 2 N
≥ Kµ − 2∗ KµN−2 − ∗ KµN−2 .
S2q 2 S

This establishes equation (4.1), thereby completing the proof.

Lemma 4.2. Under the assumptions of Lemma 4.1, the following properties hold:

Γ̌µ (c) := {γ ∈ C ([0, 1], Sr (c)) : γ(0) ∈ Aµ (Kµ , c), Iµ (γ(1)) < 0} ̸= ∅

and
M̌µ (c) := inf max Iµ (γ(t)) ≥ ζˇ 0 (c) > sup max{Iµ (γ(0)), Iµ (γ(1))},
γ∈Γ̌µ (c) t∈[0,1] γ∈Γ̌µ (c)

where ζˇ 0 (c) > 0 is independent of µ > 0.

Proof. The conclusion can be readily derived by employing Lemma 4.1 and invoking classical arguments.
Therefore, we omit the details here.

Lemma 4.3. Under the assumptions of Lemma 4.1, there exists a sequence {un } ⊂ Sr (c) satisfying

Iµ (un ) → M̌µ (c) > 0, ∥Iµ′ |Sr (c) (un )∥X ∗ → 0, Qµ (un ) → 0.

Proof. Let I˜µ be defined by (2.5). As argued before, we set

Γµ (c) := {γ ∈ C ([0, 1], R × Sr (c)) : γ(0) = (0, γ1 (0)), γ1 (0) ∈ Aµ (Kµ , c), I˜µ (γ(1)) < 0},

26
and
M µ (c) := inf max I˜µ (γ(t)).
γ∈Γµ (c) t∈[0,1]

Similar to Lemma 3.11, we can see that M̌µ (c) = M µ (c). Moreover, define

Eµ0 := {u ∈ Sr (c) : Iµ (u) < 0}.

Let
A = γ([0, 1]), X = R × Sr (c), F = {γ([0, 1]) : γ ∈ Γµ (c)},
B = ({0} × Aµ (Kµ , c)) ∪ ({0} × Eµ0 ), F = {(s, u) ∈ R × Sr (c) : I˜µ (s, u) ≥ M µ (c)}.
By employing the same argument as in Lemma 3.12, we complete the proof.
We now present a precise estimate of the energy level M̌µ (c) as defined in Lemma 4.2. In order to
achieve the desired result, we need to modify the test function Uε given by (3.21) appropriately. Define the
radial symmetry nonnegative functions as follows:
 √  N−2
4
 ε
 ε+|x|2

 , 0 ≤ |x| < ε−α ,
Ubε (x) := BN  ε2α+1/2  N−2
4 1−εβ |x| (4.2)

 1+ε2α+1 1−εβ−α
, ε−α ≤ |x| < ε−β ,

|x| ≥ ε−β ,

0,
N−2
where BN := (N(N − 2)) 8 and α > 0, β > 0 are parameters to be determined later. Then we derive
Z Z +∞
bε ∥22 =
∥U bε |2 dx = ωN
|U bε (r)|2 rN−1 dr
|U
RN 0
N−2
! N−2
ε−α rN−1
Z ε−β
ε2α+1/2
2
(1 − εβ r)2 rN−1
Z 
ε 4
= ωN B2N N−2 dr + dr
0 (ε + r2 ) 2 ε−α 1 + ε2α+1 (1 − εβ−α )2
1
ε−α− 2 sN−1
 Z
N+2
= ωN B2N ε 4 N−2 ds
0 (1 + s2 ) 2

! N−2
2
2ε−β(N+2) − N(N + 1)ε−2α − 2N(N + 2)ε−β−α + (N + 1)(N + 2)ε−2β ε−Nα
 
ε2α+1/2

+
1 + ε2α+1 N(N + 1)(N + 2)(ε−β − ε−α )2

N+2
= ωN B2N ε 4 I(ε,
˜ α)
N−2 N−2 
2ε−β(N+2)− − N(N + 1)ε−2α − 2N(N + 2)ε−β−α + (N + 1)(N + 2)ε−2β ε−Nα− 4
 
4
+ N−2 , (4.3)
N(N + 1)(N + 2)(ε−β − ε−α )2 (1 + ε−2α−1 ) 2
where
Z ε−α− 12
sN−1
˜ α) :=
I(ε, N−2 ds = O (ε−2α−1 ), as ε → 0.
0 (1 + s2 ) 2

2 −4N+8−N(N−2)q
Choose max{ 4N 16(N−2) , 0} < α < N−2
8
bε ∥2 . From this, we derive
and β > α such that c = lim ∥U 2 ε→0
(4α+1)(N−2)
that εβ = O (ε 4N ), and more precisely,
(4α+1)(N−2)  1
ε 4N N(N + 1)(N + 2)c N
lim = . (4.4)
ε→0 εβ 2ωN B2N

27
Moreover, we have
Z Z +∞
bε ∥22 =
∥∇U bε |2 dx = ωN
|∇U bε′ (r)|2 rN−1 dr
|U
RN 0
Z −α− 1
! N−2
N − 2 2 N−2 ε 2
2
ε2β−Nβ 1 − εNβ−Nα


sN+1 ε2α+1/2
 
= ωN B2N ε 4 N+2 ds +
2
2 0 (1 + s2 ) 2 1 + ε2α+1 N 1 − εβ−α
 (4α+1)(N−2) 
=O ε 2N . (4.5)

Z Z +∞
∗ ∗ ∗
bε ∥2·2
∥U 2·2∗ =
bε |2·2 dx = ωN
|U bε (r)|2·2 rN−1 dr
|U
RN 0
−α− 21
!N Z 4N
sN−1 ε2α+1/2 1−εβ−α ε−βN s N−2 (1 − s)N−1
Z ε

∗
= ωN B2·2
N ds + 4N ds
0 (1 + s2 )N 1 + ε2α+1 0 (1 − εβ−α ) N−2
!N Z 4N
sN−1 ε2α+1/2 1−εβ−α s N−2 (1 − s)N−1
 Z +∞ 
N ∗
=S 2 + ωN B2·2
N − 1 ds + ε−βN 4N ds
ε−α− 2 (1 + s2 )N 1 + ε2α+1 0 (1 − εβ−α ) N−2
 
N (4α+1)(N+2)
= S +O ε
2 4 . (4.6)

Z +∞ 2
d
Z Z
4 bε |2 |∇U
|U bε |2 dx = bε2 )|2 dx = ωN
|∇(U bε2 (r) rN−1 dr
U
RN RN 0 dr
−α− 12
!N−2 Z
sN+1 ε2α+1/2 εβ(2−N) s2 (1 − s)N−1
1−εβ−α
 Z ε

4
= ωN BN (N − 2)2 ds + 4 ds
0 (1 + s2 )N 1 + ε2α+1 0 (1 − εβ−α )4
!N−2 Z
sN+1 ε2α+1/2 1−εβ−α s2 (1 − s)N−1
 Z +∞ 
N
= S 2 + ωN B4N − (N − 2)2 1 ds + 4εβ(2−N) ds
ε−α− 2 (1 + s2 )N 1 + ε2α+1 0 (1 − εβ−α )4
 
N (4α+1)(N 2 −4)
= S 2 +O ε 4N . (4.7)

sN−1
Z +∞ Z 1
(N−2)q
N
 N (N−2)q 
bε (r)|q rN−1 dr ≥ ωN B ε 2 − q
bε ∥qq = ωN
∥U |U N
8
(N−2)q
ds = O ε2− 8 . (4.8)
0 0 (1 + s2 ) 4

 
2(N+2) ∗
Lemma 4.4. Let q ∈ N−2 , 2 · 2 , τ > 0 and c > 0. Then there exist 0 < µ1 < 1 and δ′ > 0 such that

N
S2
M̌µ (c) ≤ − δ′ for all µ ∈ (0, µ1 ).
2N
Moreover, the same conclusion holds provided that q ∈ (2, 2 · 2∗ ) for N ≥ 3 and τ > 0 is suffciently large.
√ √
c b 1 c 3
Proof. Define Vbε := bε ∥2 Uε ∈ Sr (c). Then we immediately get that 2 ≤ ≤ 2 for sufficiently small
∥U ∥U
bε ∥2

28
ε > 0. It follows from (1.9) and (4.3)-(4.8) that
 √ 2 Z  √ 4
t2 c c
Z
N
I0 (t 2 Vbε (tx)) = |∇Ubε |2 dx + t N+2 bε |2 |∇U
|U bε |2 dx
2 ∥U bε ∥2 RN ∥U
bε ∥2 RN
 √ q qγq Z  √ 2·2∗ N(N+2) Z
c τt q c t N−2 bε |2·2∗ dx
− |Uε | dx −
b
∗
|U
∥Uε ∥2
b q R N ∥Uε ∥2
b 2 · 2 RN
" N #
2 S2
 2
t  (4α+1)(N−2)
 (4α+1)(N −4)
≤ O ε 2N + t N+2 +O ε 4N
2 4
N(N+2) 
τt qγq  N − (N−2)q  t N−2
 
N (4α+1)(N+2)
− O ε 2 8 − S +O ε
2 4
q 2 · 2∗
1 N t 2  (4α+1)(N−2)  N+2
 
(4α+1)(N 2 −4)
≤ S + O ε
2 2N +t O ε 4N
2N 2
N(N+2)
τt qγq  N − (N−2)q 
 
t N−2 (4α+1)(N+2)
− O ε 4 − O ε2 8 , ∀t > 0.
2 · 2∗ q
4N −4N+8−N(N−2)q 2
If q > 2(N+2)
N−2 , we can choose max{ 16(N−2) , 0} < α < N−2
8 , such that there exist small enough
′
constants ε1 > 0 and δ > 0 satisfying the inequality
N
N S2
bε (tx)) ≤
sup I0 (t 2 U 1 − 2δ′ .
t>0 2N
2
If 2 < q ≤ 2(N+2) −1 and max{ 4N −12N+8−N(N−2)q , 0} < α < N−2
N−2 , we can select τ = ε 16(N−2) 8 , such that the above
inequality still holds.
Note that there exist constants 0 < T3 < T4 such that
N
bε (tx)) = sup Iµ (t N2 U
sup Iµ (t 2 U bε (tx)), N
bε (tx)) = sup I0 (t N2 U
sup I0 (t 2 U bε (tx)).
t>0 t∈[T3 ,T4 ] t>0 t∈[T3 ,T4 ]

From this, we conclude that

N N
bε (tx)) ≤ C(ε1 )µ + S − 2δ′ ≤ S − δ′ ,
N 2 2
bε (tx)) ≤ C(ε1 )µ + sup I0 (t N2 U
sup Iµ (t 2 U 1 1
t>0 t∈[T3 ,T4 ] 2N 2N
where µ > 0 is sufficiently small. Moreover, we have

bε (tx)) = µ t θ(1+γθ ) t2
Z Z Z
N
Iµ (t 2 U bε |θ dx +
|∇U bε |2 dx + t N+2
|∇U bε |2 |∇U
|U bε |2 dx
1 1 1 1 1
θ RN 2 RN RN
N(q−2) ∗
t N(2 −1)
Z Z
τt 2 ∗
− bε |q dx −
|U bε |22 dx,
|U
q RN
1
22∗ RN
1

which implies that there exist constants t1 > 0 small enough and t2 > 0 large enough such that
N N
bε (t1 x) ∈ Aµ (Kµ , c) and Iµ (t 2 U
t12 U 2 ε1 (t2 x)) < 0.
b
1
N
Let γε1 (t) := (t1 + (t2 − t1 )t) 2 U
bε ((t1 + (t2 − t1 )t)x). Then γε ∈ Γ̌µ (c) and we can deduce that
1 1
N
bε (tx)) ≤ S − δ′ , for some small µ > 0.
N 2
M̌µ (c) ≤ sup Iµ (t 2 U 1
t>0 2N
This completes our proof.

29
5 The case 4 + N4 ≤ q < 2 · 2∗
5.1 Properties of Λµ (c)
We first develop several properties of Λµ (c) as defined in (1.10). By adapting the approach from [3]
(see also [31]), we obtain
Lemma 5.1. Let 0 < µ ≤ 1 and c > 0. Then Λµ (c) is a C 1 -manifold of codimension 1 in S(c), and therefore
it is a C 1 -manifold of codimension 2 in X.
Lemma 5.2. For any µ ∈ (0, 1] and any u ∈ S(c), there exists a unique number su > 0 such that usu ∈ Λµ (c).
Proof. Using (1.11) and (3.4), we perform a direct computation to obtain ψ′u (s) = ds
d
Iµ (us ) = 1s Q(us ), which
implies that
ψ′u (s) = 0 ⇐⇒ us ∈ Λµ (c).
It is straightforward to observe that ψu (s) → 0+ as s → 0 and ψu (s) → −∞ as s → +∞. Consequently, the
maximum value max ψu (s) is attained at some su > 0, such that ψ′u (su ) = 0 and usu ∈ Λµ (c). Moreover, it
s∈(0,+∞)
follows from qγq ≥ N + 2 and qγq ≥ θ(1 + γθ ) that the critical point su > 0 is unique for any u ∈ X\{0}.
From Lemma 5.2, we immediately get the following results.
Corollary 5.3. Let µ ∈ (0, 1], q ∈ [4 + N4 , 2 · 2∗ ), τ > 0 and c > 0. Then for any u ∈ Λµ (c), there holds
Iµ (u) = max Iµ (us ).
s>0

Corollary 5.4. Let µ ∈ (0, 1], q ∈ [4 + 4 ∗

N , 2 · 2 ), τ > 0 and c > 0. Then there holds
inf Iµ (u) = m
b µ (c) = inf max Iµ (us ).
u∈Λµ (c) u∈S(c) s>0

b µ (c).
Inspired by [4, Lemma 5.3], we establish the following monotonicity property for m
Lemma 5.5. Let µ ∈ (0, 1], q ∈ [4 + N4 , 2 · 2∗ ), τ > 0 and c > 0. Then the function c 7→ m
b µ (c) is non-
increasing on (0, +∞).
Proof. For any 0 < ĉ1 < ĉ2 < +∞, it suffices to show that m b µ (ĉ2 ) ≤ m
b µ (ĉ1 ). By the definition of m
b µ (ĉ1 ),
there exists u ∈ Λµ (ĉ1 ) such that
Iµ (u) < m
b µ (ĉ1 ) + ε,
for any ε > 0. Let η ∈ C0∞ (RN ) a cut-off function satisfying η ≡ 1 in B1 , η ≡ 0 in RN \B2 and 0 ≤ η ≤ 1 for
1 ≤ |x| ≤ 2. Then, for δ ∈ (0, 1] small, define
uδ (x) := η(δx)u(x).
It is easy to verify that uδ → u in X as δ → 0. By continuity, we have
ε
b µ (ĉ1 ) + , Qµ (uδ ) → Qµ (u) = 0.
Iµ (uδ ) → Iµ (u) < m (5.1)
4
By Lemma 5.2, for any δ > 0, there exists some sδ > 0 such that uδsδ ∈ Λµ (c). We claim that {sδ } is bounded.
Indeed, if sδ → +∞ as δ → 0, recalling that uδ → u ̸= 0 in X as δ → 0, we have
Iµ (uδsδ ) µ θ(1+γθ )−N−2
Z
1
Z Z
0 = lim = sδ |∇uδ |θ dx + |∇uδ |2 dx + |uδ |2 |∇uδ |2 dx
δ→0 sN+2
δ
θ RN 2sNδ RN RN

τ N(q−2) −N−2 1 N(2∗ −1)−N−2

Z Z
∗
− sδ 2 |uδ |q dx − s |uδ |2·2 dx.
q RN 2 · 2∗ δ RN
= −∞,

30
which yields a contradiction. Thus, we may assume that, up to a subsequence, sδ → ŝ as δ → 0. Then
obviously we can check that Qµ (uδsδ ) → Qµ (uŝ ). By the uniqueness and the fact that Qµ (u) = 0, we get
ŝ = 1. Moreover, we have

1 − sN+2 θ(1 + γθ )(1 − sN+2 )

 Z
µ
Iµ (u) − Iµ (us ) = Qµ (u) + 1 − sθ(1+γθ ) − |∇u|θ dx
N +2 θ N +2 RN
1 − s2 1 − sN+2 N(q − 2)(1 − sN+2 ) 1 − sqγq
 Z  Z
2
+ − |∇u| dx + τ − |u|q dx
2 N +2 RN 2(N + 2)q q RN
∗
!Z
1 − sN+2 (N − 2)(1 − sN(2 −1) ) ∗
+ − |u|2·2 dx
4 4N RN

1 − sN+2 1 − s2 1 − sN+2
 Z
≥ Qµ (u) + − |∇u|2 dx
N +2 2 N +2 RN
∗
!Z
1 − sN+2 (N − 2)(1 − sN(2 −1) ) ∗
+ − |u|2·2 dx. (5.2)
4 4N R N

Hence,
!Z
1 − sN+2 1 − s2δ 1 − sN+2
Iµ (uδsδ ) ≤ Iµ (uδ ) − δ
Qµ (uδ ) − − δ
|∇uδ |2 dx
N +2 2 N +2 RN
N(2∗ −1)
!
1 − sN+2 (N − 2)(1 − sδ ) Z ∗
− δ
− |uδ |2·2 dx,
4 4N RN

which, in combination with (5.1), shows that there exists δ0 ∈ (0, 1) sufficiently small such that
ε ε ε
Iµ (uδsδ0 ) ≤ Iµ (uδ0 ) + ≤ Iµ (u) + < m
b µ (ĉ1 ) + . (5.3)
0 8 4 2
 1
2
ĉ2 −∥uδ0 ∥22
For the above δ0 > 0, we set v ∈ C0∞ (RN ) such that suppv ⊂ B4/δ0 \B2/δ0 and define v δ0 := ∥v∥22
v.
δ δ
Then, we have ∥vδ0 ∥22 = ĉ2 − ∥uδ0 ∥22 . For any κ ∈ (0, 1), we set wκ := uδ0 + vκ0 with ∥vκ0 ∥22 = ∥vδ0 ∥22 and it
is easy to verify that
δ 2 2
dist{suppuδ0 , suppvκ0 } ≥ − > 0.
δ0 κ δ0
Furthermore, we have

∥∇wκ ∥22 = ∥∇uδ0 ∥22 + κ2 ∥∇vδ0 ∥22 , ∥∇wκ ∥θθ = ∥∇uδ0 ∥θθ + κθ(1+γθ ) ∥∇vδ0 ∥θθ , (5.4)

∥wκ ∇wκ ∥22 = ∥uδ0 ∇uδ0 ∥22 + κN+2 ∥vδ0 ∇vδ0 ∥22 , (5.5)
N(q−2) ∗ ∗ ∗ −1) ∗
2·2 δ0 2·2 2·2
∥wκ ∥qq = ∥uδ0 ∥qq + κ 2 ∥vδ0 ∥qq , ∥wκ ∥2·2∗ = ∥u ∥2·2∗ + κ
N(2
∥vδ0 ∥2·2∗. (5.6)
By (5.4)-(5.6), we deduce that as κ → 0,

Iµ (wκ ) → Iµ (uδ0 ) and Qµ (wκ ) → Qµ (uδ0 ).

Moreover, it follows from ∥wκ ∥22 = ∥uδ0 ∥22 + ∥vδ0 ∥22 = ĉ2 that wκ ∈ S(ĉ2 ) for any κ ∈ (0, 1). Consequently,
by Lemma 5.2, there exists sκ > 0 such that wκsκ ∈ Λµ (ĉ2 ). Using a similar argument as before, we infer that

31
{sκ } is bounded. Thus we may assume that up to a subsequence, sκ → ŝ0 as κ → 0. In view of (5.4)-(5.6)
again, as κ → 0, we get
θ(1+γθ )
∥∇wκsκ ∥22 → ŝ20 ∥∇uδ0 ∥22 , ∥∇wκsκ ∥θθ → ŝ0 ∥∇uδ0 ∥θθ , ∥wκsκ ∇wκsκ ∥22 → ŝN+2
0 ∥uδ0 ∇uδ0 ∥22 ,
N(q−2)
2·2 ∗ N(2∗ −1) 2·2 ∗
∥wκsκ ∥qq → ŝ0 2
∥uδ0 ∥qq , ∥wκsκ ∥2·2∗ → ŝ0 ∥uδ0 ∥2·2∗,

which implies that there exists κ0 ∈ (0, 1) sufficiently small such that
δ ε
Iµ (wκsκ ) ≤ Iµ (uŝ00 ) + .
2
Therefore, together with Corollary 5.3 and (5.3) we deduce
ε ε ε
b µ (ĉ2 ) ≤ Iµ (wκsκ ) ≤ Iµ (uδŝ00 ) +
m ≤ max Iµ (uδs 0 ) + = Iµ (uδsδ0 ) + ≤ m
b µ (ĉ1 ) + ε.
2 s>0 2 0 2
This completes our proof, owing to the arbitrariness of ε > 0.

5.2 The existence of Palais-Smale sequence

b µ (c). To start with, we show
In this subsection, we will give an additional minimax characterization of m
the functional Iµ possesses a mountain pass geometry on the constraint S(c).
Lemma 5.6. Let µ ∈ (0, 1] and τ > 0. Assume that one of the following conditions holds:
(i) q = 4 + N4 , c ∈ (0, c̄3 ).

(ii) 4 + N4 < q < 2 · 2∗ , c > 0.

Then there exists Kµ = Kµ (c, τ, q, N) > 0 small enough such that

0< sup Iµ (u) < inf Iµ (u). (5.7)

u∈Aµ (Kµ ,c) u∈∂Aµ (2Kµ ,c)

Proof. Using (2.4) and the Sobolev inequality, for any u ∈ Aµ (Kµ , c), v ∈ ∂Aµ (2Kµ , c), we get that
4N−q(N−2)  N(q−2)   2∗ Z N
τC2 (q, N)c
Z 
2(N+2)
2 2
2(N+2) 1 4 2 2 2
N−2
Iµ (v) − Iµ (u) ≥ Kµ − |v| |∇v| dx − |v| |∇v| dx
q RN 2 · 2∗ S RN
N(q−2) 4N−q(N−2)
N(q−2) 2   2∗
2 2(N+2) τC2 (q, N)c 2(N+2) 2 N−2 4 2 N
≥ Kµ − Kµ2(N+2) − KµN−2 .
q 2∗ S
If q = 4 + N4 , we have

2! 2   2∗
2τN C2 4 + N4 , N c N 2 N−2 4 2 N
Iµ (v) − Iµ (u) ≥ 1− Kµ − ∗ KµN−2 .
4N + 4 2 S

From 0 < c < c̄3 , where c̄3 is defined by (1.15), it follows that there exists sufficiently small Kµ > 0 such
that Iµ (v) − Iµ (u) > 0, which implies (5.7). If 4 + N4 < q < 2 · 2∗ , choosing Kµ > 0 suffciently small, we also
N(q−2)
have the same conclusion since 2(N+2) > 1.

Next, we present the following conclusions without proof, as the arguments are analogous to those in
Lemma 4.2 and Lemma 4.3.

32
Lemma 5.7. Under the assumptions of Lemma 5.6, there holds:
bµ (c) := {γ ∈ C ([0, 1], Sr (c)) : γ(0) ∈ Aµ (Kµ , c), Iµ (γ(1)) < 0} ̸= ∅.
Γ

and
bµ (c) := inf max Iµ (γ(t)) ≥ b
M ζ0 (c) > sup max{Iµ (γ(0)), Iµ (γ(1))},
bµ (c) t∈[0,1]
γ∈Γ bµ (c)
γ∈Γ

ζ0 (c) > 0 is independent of µ > 0.

where b
Lemma 5.8. Under the assumptions of Lemma 5.6, there exists a sequence {un } ⊂ Sr (c) satisfying

Iµ (un ) → M
bµ (c) > 0, ∥Iµ′ |Sr (c) (un )∥X ∗ → 0, Qµ (un ) → 0.

bµ (c) = m
Lemma 5.9. Under the assumptions of Lemma 5.6, we have M b µ (c).

Proof. On the one hand, for any u ∈ Λµ (c), there exist s1 < 0 small and s2 > 0 large such that

us1 ∈ Aµ (Kµ , c) and Iµ (us2 ) < 0.

Let bγ(t) := u(1−t)s1 +ts2 , ∀t ∈ [0, 1]. Clearly, bγ ∈ Γ

bµ (c). By Corollary 5.3, we have

bµ (c) ≤ inf Iµ (u) = m

M b µ (c).
u∈Λµ (c)

On the other hand, letting s → 0 in (5.2), we get

1 N 1 1
Z Z
∗
Iµ (u) ≥ Qµ (u) + |∇u|2 dx + |u|2·2 dx ≥ Qµ (u), ∀u ∈ S(c).
N +2 2(N + 2) RN 2N RN N +2

Thus for any γ ∈ Γ bµ (c), we have Qµ (γ(1)) ≤ (N + 2)Iµ (γ(1)) < 0. By employing a similar argument as in
Lemma 5.2, it can be readily verified that there exists u1 ∈ Aµ (Kµ , c) such that Qµ (u1 ) > 0. This implies
that there exists some t0 such that γ(t0 ) ∈ Λµ (c). Consequently, we deduce that

max Iµ (γ(t)) ≥ inf Iµ (u) = m

b µ (c),
t∈[0,1] u∈Λµ (c)

bµ (c) ≥ m
which in turn implies that M b µ (c). This completes the proof.

Remark 5.10. By Lemma 2.2, any critical point of Iµ |S(c) lies in the set Λµ (c). Therefore, if m
b µ (c) attained,
the minimizer must be a ground state critical point of Iµ |S(c) .
Recalling the definition of qN given by (1.18), we now present the following result for further conve-
nience.
4N
, τ > 0 and c > 0. Then there exist 0 < µ2 < 1 and δ′′ > 0 such that


Lemma 5.11. Let q ∈ qN , N−2
N
bµ (c) ≤ S − δ′′ , for all µ ∈ (0, µ2 ).
2
M
2N

Moreover, the same conclusion holds if q ∈ (4 + N4 , 2 · 2∗ ) for N ≥ 3 and τ > 0 is sufficiently large.

33
6 The compactness of the Palais-Smale sequences
In this section, we establish the compactness of the Palais-Smale sequences obtained in Lemma 3.12,
Lemma 4.3 and Lemma 5.8. For simplicity, we denote these three different mountain pass values, i.e.,
Mµ (c), M̌µ (c) and Mbµ (c), uniformly by Mµ∗ (c).

Lemma 6.1. Assume that µ ∈ (0, 1]. Let {un } ⊂ Sr (c) be a sequence obtained in Lemma 3.12, Lemma 4.3
or Lemma 5.8. Then, there exists uµ ̸≡ 0 such that, up to a subsequence, un ⇀ uµ in Xr as n → +∞.
Proof. We first establish the boundedness of {un } in Xr , which we will discuss in the following two cases.
Case 1: 2 < q ≤ 4 + N4 . It follows from (2.4), Iµ (un ) → Mµ∗ (c), and Qµ (un ) → 0 that
 Z  Z
µ N −2 1 N −2
Mµ∗ (c) + on (1) ≥ 1 − θ(1 + γθ ) |∇un |θ dx + − |∇un |2 dx
θ N(N + 2) RN 2 N(N + 2) RN
  Z  N(q−2)
2 γq (N − 2) 1 4N−q(N−2)
Z 2(N+2)
2 2 2 2
+ |un | |∇un | dx + τ − C2 (q, N)c 2(N+2) |un | |∇un | dx
N RN N(N + 2) q R N

Z Z  Z  N(q−2)
2(N+2)
θ 2 2 2 2
≥ C1 |∇un | dx + (1 + |un | )|∇un | dx −C2 (1 + |un | )|∇un | dx
RN RN RN

 N
2
4 4 2N+2
for some C1 ,C2 > 0. For q < 4 + or q < 4 + with 0 < c < c̄3 = , we deduce that
N N τN C2 (4+ N4 ,N )
2 2
R R
{ RN (1 + |un | )|∇un | dx} and { RN |∇un | dx} are both bounded.
θ

Case 2: 4 + N < q < 2 · 2 . Using Iµ (un ) → Mµ∗ (c) and Qµ (un ) → 0 again, together with (2.4), we obtain
4 ∗

that {∥un ∥θθ } is bounded and thus {un } is bounded in Xr . Therefore, up to translation, un ⇀ uµ weakly in
Xr .
Next, we proceed by contradiction, assuming that uµ = 0. Then ∥un ∥rr → 0 in Lr (RN ) for r ∈ (2, 2∗ ). By
Nθ
the interpolation inequality, we further obtain ∥un ∥rr → 0 in Lr (RN ) for r ∈ (2, N−θ ). In particular,
Z Z
4N
|un |q dx → 0 and |un | N−2 dx → 0.
RN RN

Therefore, by Qµ (un ) → 0, we have

Z Z Z
µ(1 + γθ ) |∇un |θ dx + |∇un |2 dx + (N + 2) |un |2 |∇un |2 dx → 0.
RN RN RN

Thus, we deduce that Iµ (un ) → 0, which contradicts the fact that Mµ∗ (c) > 0.
Lemma 6.2. Assume that µ ∈ (0, 1]. Let {un } ⊂ Sr (c) is a sequence obtained in Lemma 3.12, Lemma 4.3
¯ µ ∈ R, such that, up to a subsequence,
or Lemma 5.8. Then, there exist uµ ∈ Xr \{0} and λ

¯ µ uµ = 0.
Iµ (uµ ) = Mµ∗ (c) and Iµ′ (uµ ) + λ (6.1)

¯ µ ̸= 0, we have
Moreover, if λ
lim ∥un − uµ ∥X = 0.
n→+∞

Proof. Using Lemma 6.1, we have

 
Nθ
un ⇀ uµ in Xr , un → uµ in Lr (RN ), r ∈ 2, . (6.2)
N −θ

34
By standard arguments (see [5] or [26]), for any w ∈ X, we get
1
⟨Iµ′ (un ) + λn un , w⟩ = on (1) with λn = − ⟨Iµ′ (un ), un ⟩. (6.3)
c
¯µ ∈ R
Since ⟨Iµ′ (un ), un ⟩ is bounded, it follows that {|λn |} is also bounded. Consequently, there exists λ
¯
such that, up to a subsequence, λn → λµ as n → +∞. By the boundedness of {un } ⊂ X, we get that
¯ µ un → 0.
Iµ′ (un ) + λ
To prove (6.1), it suffices to show that for any w ∈ X,
¯ µ uµ , w⟩ as n → +∞.
⟨Iµ′ (un ) + λn un , w⟩ → ⟨Iµ′ (uµ ) + λ (6.4)
Since un ⇀ uµ in Xr , we readily obtain the following convergences:
R R

 RN ∇un ∇wdx → RN ∇uµ ∇wdx,

λ R
¯ R
n RN un wdx → λµ RN uµ wdx,
q−2 u wdx → q−2 u wdx,
R R
RN |un | n RN |uµ | µ


∗ −2 2·2∗ −2 u wdx.

2·2
R R
|u | → |u |

RN n un wdx R N µ µ

Taking into account (6.3), it remains to prove that

Z Z
|∇un |θ−2 ∇un ∇wdx → |∇uµ |θ−2 ∇uµ ∇wdx, (6.5)
RN RN
and Z Z
un w|∇un |2 + |un |2 ∇un ∇w dx → uµ w|∇uµ |2 + |uµ |2 ∇uµ ∇w dx.



(6.6)
RN RN
θ
Since {|∇un |} is bounded in Lθ (RN ), it follows that {|∇un |θ−2 ∇un } is bounded in L θ−1 (RN )N . Conse-
quently,
θ
|∇un |θ−2 ∇un ⇀ |∇uµ |θ−2 ∇uµ in L θ−1 (RN )N .
Combining this with the weak convergence for any |∇w| ∈ Lθ (RN ), we conclude that (6.5) holds. Similarly,
using the Young’s inequality, we have that
4 |un |4 2|∇un |4 4 2|un |4 |∇un |4
(|un ||∇un |2 ) 3 ≤ + and (|un |2 |∇un |) 3 ≤ + ,
3 3 3 3
4
which implies that {|un ||∇un |2 } and {|un |2 |∇un |} are bounded in L 3 (RN ) since {un } is bounded in X. Thus,
we also have
4
un |∇un |2 ⇀ uµ |∇uµ |2 in L 3 (RN ),
4
|un |2 ∇un ⇀ |uµ |2 ∇uµ in L 3 (RN )N .
By a similar argument, (6.6) is thereby established.
Finally, in view of (6.2), we select the test functions wn = un and w = uµ in (6.4), respectively. By
invoking the weak lower semi-continuity property (see [14, Lemma 4.3]), we infer that
Z Z Z Z Z Z
µ |∇un |θ → µ |∇uµ |θ , |∇un |2 dx → |∇uµ |2 dx, |un |2 |∇un |2 dx → |uµ |2 |∇uµ |2 dx,
RN RN RN RN RN RN
These convergences imply that
Z Z
λn ¯µ
|un |2 dx → λ |uµ |2 dx and Iµ (uµ ) = Mµ∗ (c).
RN RN
¯ µ ̸= 0, it follows that
Thus, if λ
R
|un |2 dx →
R
|uµ |2 dx. Consequently, un → uµ strongly in Xr , thereby
RN RN
completing the proof.

35
7 Convergence issues
In this section, we address the convergence behavior as µ → 0+ and demonstrate that the sequences of
critical points of Iµ , obtained in Lemma 3.8 and Lemma 6.2, converge to critical points of I = I0 restricted
on S(c).
e More precisely, for any fixed µn ∈ (0, 1], we identify the critical points vµn (or uµn , respectively) of
Iµn |S(cn ) with 0 < cn ≤ c. These critical points solve the following problem:
∗
(
−µn ∆θ vµn − ∆vµn − vµn ∆(v2µn ) + λµn vµn = τ|vµn |q−2 vµn + |vµn |2·2 −2 vµn in RN ,
R 2
(7.1)
RN |vµn | dx = cn .

We then analyze the convergence process as µn → 0+ , n → +∞, considering the different types of critical
points involved. This analysis is crucial for proving our main results and is presented in detail in the
following two subsections.

7.1 Local minimizers for the original problem

In this subsection, we primarily consider the convergence behavior of the critical points obtained in
˜ n := λ
Lemma 3.8. For simplicity, we set vn := vµn , λ ˜ µ and m∗ (c) := lim mµ (c).
n 0 n
n→+∞

˜ ∈ S(c)
Proposition 7.1. Let 2 < q < 2 + N4 and c ∈ (0, c0 ). Then there exists (v̌, λ) e × R+ satisfying
∗ −2
−∆u − u∆(u2 ) + λu = τ|u|q−2 u + |u|2·2 u, x ∈ RN . (7.2)
Proof. We first claim that m∗0 (c) < 0. In fact, observe that for any 0 < µ2 < µ1 ≤ 1 and u ∈ X, one has
Iµ1 (u) ≥ Iµ2 (u) and Vµ1 (c) ⊂ Vµ2 (c). Hence, we have
mµ1 (c) = inf Iµ1 (u) ≥ inf Iµ2 (u) ≥ inf Iµ2 (u) = mµ2 (c),
u∈Vµ1 (c) u∈Vµ1 (c) u∈Vµ2 (c)

which implies that the energy mµ (c) is non-decreasing with respect to µ ∈ (0, 1]. Therefore, by Lemma 3.5
(i), we have m∗0 (c) ≤ m1 (c) < 0, thus confirming the claim. Moreover, we infer that {vn } is bounded in
H 1 (RN ) and {vn ∇vn } is bounded in L2 (RN )N due to the fact that vn ∈ Vµn (c). Define
Z
δ¯ := lim sup sup |vn |2 dx.
n→+∞ y∈RN B1 (y)

We now prove that δ¯ > 0. Suppose, for the sake of contradiction, that δ¯ = 0. It follows from Lemma 2.1
that vn → 0 in Lβ (RN ) for all 2 < β < 2 · 2∗ . Using the Sobolev inequality, we obtain
µn 1 1
Z Z Z Z
∗
Iµn (vn ) = |∇vn |θ dx + |∇vn |2 dx + |vn |2 |∇vn |2 dx − |vn |2·2 dx + on (1)
θRN 2 RN RN 2 · 2∗ RN
∗
 
 2
1 1 4 2 α2 
≥ ξµn (vn )  − ∗
ρ0 + on (1).
2 2·2 S

Taking into account that f (c0 , ρ0 ) = 0, see Remark 3.4, we arrive at

m∗0 (c) = lim mµn (c) = lim Iµn (vn ) ≥ 0,
n→+∞ n→+∞

which is a contradiction. Thus, δ¯ > 0, i.e., there exist {yn } ⊂ RN and δ¯ > 0 such that lim inf B1 (yn ) |vn |2 dx ≥
R
n→+∞
δ¯ > 0. Let v̌n (·) := vn (· + yn ). Then
Z
Iµn (v̌n ) → m∗0 (c), ˜ n v̌n = 0,
Iµ′ n (v̌n ) + λ |v̌n |2 dx ≥ δ¯ > 0,
B1 (0)

36
which implies that there exists v̌ ∈ X\{0}
e such that, up to a subsequence,

v̌n ⇀ v̌ in H 1 (RN ), v̌n ∇v̌n ⇀ v̌∇v̌ in L2 (RN )N , r

v̌n → v̌ in Lloc (RN ) for 2 < r < 2 · 2∗ .
˜ n } is bounded, and we may assume that
Moreover, by testing (7.1) with vn , it is standard to show that {λ
˜λn → λ
˜ in R. By Lemma A.1 and Remark A.2, we can prove that v̌ satisfies

˜ v̌, φ⟩ = 0 for any φ ∈ H 1 (RN ) ∩ L∞ (RN ),

⟨I ′ (v̌) + λ (7.3)

which implies that v̌ is a nontrivial solution of (7.2). Thus, with the aid of Remark A.2 and Theorem A.4,
we can find a profile decomposition of {v̌n } satisfying

v̌n (· + yin ) ⇀ ṽi in H 1 (RN ), v̌n (· + yin )∇v̌n (· + yin ) ⇀ ṽi ∇ṽi in L2 (RN )N as n → +∞,
i i
∥∇v̌n ∥22 = ∑ ∥∇ṽ j ∥22 + ∥∇vin ∥22 + on (1), ∥v̌n ∇v̌n ∥22 = ∑ ∥ṽ j ∇ṽ j ∥22 + ∥vin ∇vin ∥22 + on (1),
j=0 j=0

i ∞ i
∗ ∗ ∗
∥v̌n ∥22 = ∑ ∥ṽ j ∥22 + ∥vin ∥22 + on (1), lim sup ∥v̌n ∥qq = ∑ ∥ṽ j ∥qq , 2·2
∥v̌n ∥2·2∗ =
2·2
∑ ∥ṽ j ∥2·2 i 2·2
∗ + ∥vn ∥2·2∗ + on (1),
j=0 n→+∞ j=0 j=0
j
where vin (·) := v̌n − ∑ij=0 ṽ j (· − yn ). Thus, we obtain

i
µn
Z
Iµn (v̌n ) = |∇v̌n |θ dx + ∑ I(ṽ j ) + I(vin ) + on (1).
θ RN j=0

For j = 0, 1, · · · , i, set
 2N  2N
∥v̌n ∥2 ∥v̌n ∥2
ω j := , ωin := .
∥ṽ j ∥2 ∥vin ∥2
Clearly, ω j ≥ 1 and ωin ≥ 1. Moreover, from the convergence of ∑ij=0 ∥ṽ j ∥22 , there exists some j0 ≥ 0 such
that
∥v̌n ∥2 2N
 
inf ω j = ω j0 = .
j≥0 ∥ṽ j0 ∥2
For j = 0, 1, · · · , i, define the scaled functions
1−N
 
− N1 1−N
 1

2N
Ṽ j (x) := ω j ṽ j ω j x , Vni (x) := (ωin ) 2N vin (ωin )− N x . (7.4)

We then obtain the following relations:

−(N+1)θ
+1 −1
∥Ṽ j ∥22 = ∥Vni ∥22 = c, ∥∇Ṽ j ∥θθ = ω j 2N
∥∇ṽ j ∥θθ , ∥∇Ṽ j ∥22 = ω j N ∥∇ṽ j ∥22 , ∥Ṽ j ∇Ṽ j ∥22 = ω−1 2
j ∥ṽ j ∇ṽ j ∥2 .

This implies that Ṽ j ∈ Vµ (c) for any µ ∈ (0, 1], and similarly, Vni ∈ Vµ (c). Moreover, using the Young’s
inequality and the Sobolev inequality, for any j = 0, 1, · · · , i, there exist ε j > 0 and Cε j > 0 such that

  2∗
4 2 ∗
∥ṽ j ∥qq ≤ ε j ∥ṽ j ∥22 +Cε j ∥ṽ j ∇ṽ j ∥22 .
S

37
 1
−1− N
(1−ω j )q
Therefore, choosing ε j = 3τ∥ṽ j ∥22
∥ṽ j ∇ṽ j ∥22 and picking c > 0 smaller if necessary, we have

− N2
I(Ṽ j ) 1−ωj −1− N1
I(ṽ j ) = 1 + ∥∇ṽ j ∥22 + (1 − ω j )∥ṽ j ∇ṽ j ∥22
N 2
ωj
   
τ 1−N (q−2) N −2 1−N (2·2∗ −2)
2·2∗
− 1 − ω j 2N ∥ṽ j ∥qq − 1 − ω j 2N ∥ṽ j ∥2·2 ∗,
q 4N
    2∗
I(Ṽ j ) −1− N1 τ τ N − 2 4 2 ∗
≥ 1 + (1 − ω j )∥ṽ j ∇ṽ j ∥22 − ε j ∥ṽ j ∥22 − Cε j + ∥ṽ j ∇ṽ j ∥22
ω jN
q q 4N S
−1− N1
I(Ṽ j ) 1−ωj
≥ 1 + ∥ṽ j ∇ṽ j ∥22 .
N 3
ωj

Similarly, we also get

1
I(Vni ) 1 − (ωin )−1− N i i 2
I(vin ) ≥ 1 + ∥vn ∇vn ∥2 .
(ωin ) N 3

Thus, as n → +∞ and i → +∞, we deduce that

−1− N1
µn
Z i
Iµn (Ṽ j ) i 1−ωj
Iµn (v̌n ) ≥ θ
|∇v̌n | dx + ∑ 1 +∑ ∥ṽ j ∇ṽ j ∥22
θ RN j=0 N j=0 3
ωj
Iµn (Vni ) −1− N1
1 − (ωin )
+ 1 + ∥vin ∇vin ∥22 + on (1)
i
(ωn ) N 3
1
1 − (ωin )−1− N i i 2
  i
1 −1− N1 2
≥ mµn (c) + inf 1 − ω j ∑ ∥ṽ j ∇ṽ j ∥2 + ∥vn ∇vn ∥2 + on (1).
3 j≥0 j=0 3

Let δ˜ 1 := min{ω j0 , ωin }. Then we have

−1− 1
1 − δ˜ 1 N
mµn (c) ≥ mµn (c) + ∥v̌n ∇v̌n ∥22 + on (1). (7.5)
3

Now, we claim that there exists some δ˜ 2 > 0 independent of n such that

∥v̌n ∇v̌n ∥22 ≥ δ˜ 2 > 0. (7.6)

Indeed, since m∗0 (c) < 0, for sufficiently large n, we can take a positive number σ̃ such that

Iµn (v̌n ) ≤ m∗0 (c) + σ̃ < 0.

Thus, for sufficiently large n, we get that

τ 1 2·2∗ ∗
∥v̌n ∥qq + ∗
∥v̌n ∥2·2∗ ≥ −(m0 (c) + σ̃) > 0,
q 2·2

38
which, combined with the Gagliardo-Nirenberg inequality and Sobolev inequality, implies (7.6). Thus it
follows from (7.5) that as n → +∞,
−1− 1
1 − δ˜ 1 N ˜
m∗0 (c) ≥ m∗0 (c) + δ2 ,
3

which implies that δ˜ 1 ≤ 1. Therefore, we deduce that

∥v̌n ∥2 ≤ ∥ṽ j0 ∥2 or ∥v̌n ∥2 ≤ ∥vin ∥2 .

If ∥v̌n ∥2 ≤ ∥vin ∥2 , then

lim ∥v̌n ∥qq = 0.
n→+∞

Similar to the argument as before, we can easily conclude that this is impossible since m∗0 (c) < 0. Therefore,
it is necessary that
∥v̌n ∥2 ≤ ∥ṽ j0 ∥2 ,
which implies that

∥v̌n ∥22 = ∥ṽ j0 ∥22 , ∥∇v̌n ∥22 = ∥∇ṽ j0 ∥22 , ∥v̌n ∇v̌n ∥22 = ∥ṽ j0 ∇ṽ j0 ∥22 .

Moreover, by Theorem A.4, there exists a fixed point y j0 ∈ RN such that

v̌n (x) = ṽ j0 (x − ynj0 ) ⇀ ṽ j0 (x − y j0 ) as n → +∞.

Thus, by the uniqueness of the weak limit, there holds v̌ = ṽ j0 (· − y j0 ). Consequently,

v̌n → v̌ in H 1 (RN ) and v̌n ∇v̌n → v̌∇v̌ in L2 (RN )N .

Since
2·2 ∗ 2·2 ∗
Qµn (v̌n ) + τγq ∥v̌n ∥qq + γ2·2∗ ∥v̌n ∥2·2 q
∗ → Q0 (v̌) + τγq ∥v̌∥q + γ2·2∗ ∥v̌∥2·2∗ ,

it follows that
µn ∥∇v̌n ∥θθ → 0 and I(v̌) = lim mµn (c) = m∗0 (c).
n→+∞

˜ > 0. In fact, from (7.3), v̌ satisfies the following Pohozaev

To proceed, it suffices to demonstrate that λ
identity

2−N
Z Z ˜ Z
λN τN
Z
N
Z
∗
|∇v̌|2 dx + (2 − N) |v̌|2 |∇v̌|2 dx − |v̌|2 dx + |v̌|q dx + |v̌|2·2 dx = 0.
2 RN RN 2 RN q RN 2 · 2∗ RN
Thus, we get

1
Z
2
Z ˜ Z
λ
0 ≥ m∗0 (c) = lim Iµn (v̌n ) = I(v̌) = |∇v̌|2 dx + |v̌|2 |∇v̌|2 dx − |v̌|2 dx,
n→+∞ N RN N RN 2 RN
which gives that
˜ v̌∥2 ≥ 2 4
Z Z
λ∥ 2 |∇v̌|2 dx + |v̌|2 |∇v̌|2 dx > 0.
N RN N RN
˜ > 0, and the proof is complete.
Therefore, λ
Building on the above result, we are now ready to complete the proof of Theorem 1.1.

39
Proof of Theorem 1.1 . Let µn → 0+ . By Lemma 3.8, there exists a sequence {vµn } ⊂ Vµn (c) with 0 < c <
c0 such that
˜ µ vµ = 0, Iµ (vµ ) = mµ (c) → m∗ (c).
Iµ′ n (vµn ) + λ n n n n n 0

˜ > 0 such that

e ∩ L∞ (RN ) and λ
Then, Proposition 7.1 implies that there exist 0 ̸= v̌ ∈ S(c)
˜ v̌ = 0,
I ′ (v̌) + λ I(v̌) = m∗0 (c).

Define
m0 (c) := inf{I(v) : v ∈ S(c),
e dI| e (v) = 0}.
S(c)

Clearly, m0 (c) ≤ m∗0 (c) < 0. Taking a sequence vn ∈ S(c)

e with dI| e (vn ) = 0 and I(vn ) → m0 (c), and
S(c)
e ∩ L∞ (RN )\{0} and a
applying a similar argument as in Proposition 7.1, we obtain a function ṽ0 ∈ S(c)
˜
constant λ0 > 0 such that, up to a subsequence, there holds
˜ 0 ṽ0 = 0,
I ′ (ṽ0 ) + λ I(ṽ0 ) = m0 (c).

Moreover, since Q0 (ṽ0 ) = 0, we have

 Z  Z
1 N −2 2 1 (N − 2)γq
Z
I(ṽ0 ) = − |∇ṽ0 |2 dx + |ṽ0 |2 |∇ṽ0 |2 dx − τ − |ṽ0 |q dx.
2 N(N + 2) RN N RN q N(N + 2) RN

Combining (2.4) and I(ṽ0 ) = m0 (c) < 0, we get that

  Z  N(q−2)
2 1 (N − 2)γq 4N−q(N−2)
Z 2(N+2)
|ṽ0 |2 |∇ṽ0 |2 dx < τ − C2 (q, N)c 2(N+2) |ṽ0 |2 |∇ṽ0 |2 dx . (7.7)
N RN q N(N + 2) RN

Since q < 4 + N4 , it follows from (7.7) that the ground state ṽ0 lies within V0 (c), as defined by (1.16). This
completes the proof.

7.2 Mountain pass solutions for the original problem

This subsection addresses the convergence issue of the critical points obtained in Lemma 6.2. To begin,
we recall the definition of Mµ∗ (c) from Section 6 and define M0∗ (c) := lim Mµ∗n (cn ). By applying Lemma
n→+∞
3.10, Lemma 4.2 and Lemma 5.7, we establish that M0∗ (c) > 0 for all relevant values of c. Building on this,
we establish the following proposition:

Proposition 7.2. Suppose N ≥ 3, 2 < q < 2 · 2∗ , c > 0 is fixed. Let µn → 0+ as n → +∞, and consider a
sequence {w̄n } ⊂ Sr (cn ) with 0 < cn ≤ c and {λ ¯ n } ⊂ R satisfying
 
1 N
Iµn (w̄n ) → M0∗ (c) ∈ 0, S 2 , Iµ′ n (w̄n ) + λ¯ n w̄n = 0, (7.8)
2N
¯ ∈ R such that, up to a subsequence, one of the following
Then there exist ū ∈ Hr1 (RN ) ∩ L∞ (RN )\{0} and λ
holds:
(i) either w̄n ⇀ ū weakly in X, ¯ ∈ R, and
e where ū solves (1.2) for some λ

1 N
I(ū) ≤ M0∗ (c) − S2;
2N

40
 (ii) or as n → +∞, we have that

µn ∥∇w̄n ∥θθ → 0, ∥∇w̄n ∥22 → ∥∇ū∥22 , ∥w̄n ∇w̄n ∥22 → ∥ū∇ū∥22 ,

¯ ̸= 0, then w̄n → ū in L2 (RN ). In particular, ū is a critical point of
with I(ū) = M0∗ (c). Moreover, if λ
I on S(c), where c = lim cn .
e
n→+∞

Proof. For the sequence {w̄n } ⊂ Sr (cn ) satisfying (7.8), we first show that
 Z Z Z 
θ 2 2 2
sup max µn |∇w̄n | dx, |w̄n | |∇w̄n | dx, |∇w̄n | dx < +∞. (7.9)
n∈N RN RN RN

¯ n w̄n = 0 and Lemma 2.2, we know that Qµ (w̄n ) = 0 for each n ∈ N.

In fact, from the equation Iµ′ n (w̄n ) + λ n
Considering that Iµn (w̄n ) → M0∗ (c) as n → +∞, we analyze the boundedness of {w̄n } by examining the
following two cases.
Case 1: 2 < q < 4 + N4 . By (2.4) we get that

N −2
M0∗ (c) + 1 ≥ Iµn (w̄n ) = Iµn (w̄n ) − Qµ (w̄n )
N(N + 2) n
 Z  Z
µn N −2 1 N −2
≥ 1 − θ(1 + γθ ) |∇w̄n |θ dx + − |∇w̄n |2 dx
θ N(N + 2) RN 2 N(N + 2) RN
  Z  N(q−2)
2 γq (N − 2) 1 4N−q(N−2)
Z 2(N+2)
2 2 2 2
+ |w̄n | |∇w̄n | dx + τ − C2 (q, N)c 2(N+2) |w̄n | |∇w̄n | dx ,
N RN N(N + 2) q RN

which implies that the sequences { RN (1 + |w̄n |2 )|∇w̄n |2 dx} and {µn ∥∇w̄n ∥θθ } are bounded. Moreover, by
R
∗
the Sobolev inequality, {w̄n } is bounded in H 1 (RN ) and L2·2 (RN ).
4 ∗
Case 2: 4 + N ≤ q < 2 · 2 . By Qµn (w̄n ) = 0, we get

1
M0∗ (c) + 1 ≥ Iµn (w̄n ) = Iµn (w̄n ) − Qµ (w̄n ).
qγq n

By similar arguments as in Case 1, we deduce that (7.9) also holds in this case.
Consequently, there exists ū ∈ Hr1 (RN ) with ∥ū∇ū∥22 < +∞ such that, up to a subsequence, w̄n ⇀ ū in
Hr (RN ) and w̄n ∇w̄n ⇀ ū∇ū in L2 (RN )N .
1

We next show that ū ̸≡ 0. Suppose, by contradiction, that ū ≡ 0. Then w̄n → 0 in Lr (RN ) for r ∈ (2, 2∗ ).
Using (7.9) and the interpolation inequality, we further infer that w̄n → 0 in Lr (RN ) for all r ∈ (2, 2 · 2∗ ).
From Qµn (w̄n ) = 0, we obtain
Z Z Z Z
∗
µn (1 + γθ ) |∇w̄n |θ dx + |∇w̄n |2 dx + (N + 2) |w̄n |2 |∇w̄n |2 dx − γ2·2∗ |w̄n |2·2 dx = on (1).
RN RN RN RN

Thus, we have
 N−2
(N + 2)S
Z
N +2 N
Z Z
2·2∗ 2 2 2·2∗
|w̄n | dx ≥ (N + 2) |w̄n | |∇w̄n | dx + on (1) ≥ |w̄n | dx + on (1),
4 RN R N 4 RN

∗ ∗ N
|w̄n |2·2 dx → 0 or |w̄n |2·2 dx ≥ S 2 + on (1). The former case implies
R R
which implies that either RN RN
Z Z Z
µn (1 + γθ ) |∇w̄n |θ dx + |∇w̄n |2 dx + (N + 2) |w̄n |2 |∇w̄n |2 dx → 0,
RN RN RN

41
which leads to Iµn (w̄n ) → 0. This contradicts M0∗ (c) ̸= 0. The latter case gives

M0∗ (c) + on (1) = Iµn (w̄n )

 Z  Z
µn θ(1 + γθ ) 1 1 1
Z
2 ∗
= 1− θ
|∇w̄n | dx + − |∇w̄n | dx + |w̄n |2·2 dx + on (1)
θ N +2 R N 2 N + 2 R N 2N RN

1 1 N
Z
2·2∗
≥ |w̄n | dx + on (1) ≥ S 2 + on (1),
2N RN 2N
N
which is also a contradiction since M0∗ (c) < 2N1
S 2 . Thus, ū ̸≡ 0.
We now claim that
¯ n } ⊂ R is bounded.
{λ (7.10)
¯ n = − 1 ⟨I ′ (w̄n ), w̄n ⟩. By (7.9), it suffices to show that
Indeed, by (7.8), for any µn ∈ (0, 1], we have λ cn µn

lim inf cn > 0.

n→+∞

q
To see this, suppose for contradiction that lim cn = 0. Then by (2.4), we have ∥w̄n ∥q → 0. Since Qµn (w̄n ) =
n→+∞
∗ ∗ N
0, arguing as before, we obtain either RN |w̄n |2·2 dx → 0 or RN |w̄n |2·2 dx ≥ S 2 + on (1). In both cases,
R R

we reach a contradiction, thus proving the claim (7.10). Therefore, we may assume that λ ¯n → λ¯ in R.
¯
Moreover, as in Proposition 7.1, (ū, λ) satisfies (7.2) and Q0 (ū) = 0.
To establish the alternatives, set w̃n := w̄n − ū, ∀n ∈ N. Then we have

w̃n ⇀ 0 in Hr1 (RN ), w̃n → 0 in Lγ (RN ), γ ∈ (2, 2 · 2∗ ), w̃n → 0 a.e. in RN .

Using the Brezis-Lieb type lemma (see [52]), we obtain




∥∇w̄n ∥θθ = ∥∇w̃n ∥θθ + ∥∇ū∥θθ + on (1),
2 2 2
∥∇w̄n ∥2 = ∥∇w̃n ∥2 + ∥∇ū∥2 + on (1),



∥w̄n ∥22 = ∥w̃n ∥22 + ∥ū∥22 + on (1), (7.11)
 q q q
∥w̄n ∥q = ∥w̃n ∥q + ∥ū∥q + on (1),




∥w̄ ∥2·2 = ∥w̃ ∥2·2 + ∥ū∥2·2∗ + o (1).
 ∗ ∗
n 2·2∗ n 2·2∗ 2·2∗ n

Moreover, by Remark A.2 and Lemma A.3, we have

Z Z Z
|w̄n |2 |∇w̄n |2 dx = |w̃n |2 |∇w̃n |2 dx + |ū|2 |∇ū|2 dx + on (1). (7.12)
RN RN RN

Therefore, by Qµn (w̄n ) = Q0 (ū) = 0, we deduce that

Z
∗
µn (1 + γθ )∥∇w̄n ∥θθ + ∥∇w̃n ∥22 + (N + 2) |w̃n |2 |∇w̃n |2 dx − γ2·2∗ ∥w̃n ∥2·2
2·2
∗ + on (1) = 0. (7.13)
RN

For simplicity, up to a subsequence, considering µn → 0, we may assume

Z
∗
µn ∥∇w̃n ∥θθ → ι1 , µn ∥∇w̄n ∥θθ → ι1 , ∥∇w̃n ∥22 → ι2 , |w̃n |2 |∇w̃n |2 dx → ι3 , ∥w̃n ∥2·2
2·2
∗ → ι4 , (7.14)
RN

where ι1 , ι2 , ι3 , ι4 ≥ 0. Passing to the limit as n → +∞ in (7.13), we infer that

N +2
(1 + γθ )ι1 + ι2 + (N + 2)ι3 = ι4 . (7.15)
4

42
Using the Sobolev inequality, we have
  2∗ Z  2∗
4 2 2
Z
2 2 2·2∗ 2 2
(N + 2) |w̃n | |∇w̃n | dx ≤ γ2·2∗ ∥w̃n ∥2·2∗ + on (1) ≤ γ2·2∗ |w̃n | |∇w̃n | dx + on (1).
RN S RN

Letting n → +∞, we deduce that

  2∗
4 2 2∗
(N + 2)ι3 ≤ γ2·2∗ ι32 ,
S
N N
which implies that either ι3 = 0 or ι3 ≥ S4 . If ι3 ≥ S4 , then by (7.11) and (7.12), we have
2 2

µn 1 1
Z
2·2∗
M0∗ (c) + on (1) = Iµn (ū) + ∥∇w̃n ∥θθ + ∥∇w̃n ∥22 + |w̃n |2 |∇w̃n |2 dx − ∥w̃n ∥2·2 ∗ + on (1).
θ 2 RN 2 · 2∗
Taking the limit in above equation, and using (7.14) and (7.15), we obtain
1 1 N −2
M0∗ (c) = lim Iµn (ū) + ι1 + ι2 + ι3 − ι4
n→+∞ θ 2 4N
N +2 1 1 N −2
= I(ū) + ι4 + (1 − θ(1 + γθ ))ι1 − ι2 − (N + 1)ι3 − ι4
4 θ 2 4N
 2   2 
(N + N + 2)(1 + γθ ) 1 N +N +2 1 2
= I(ū) + + − (1 + γθ ) ι1 + − ι2 + ι3
N(N + 2) θ N(N + 2) 2 N
1 N
≥ I(ū) + S2,
2N
which completes the proof of alternative (i) in Proposition 7.2. If instead ι3 = 0, then combining the Sobolev
inequality and (7.15), we get that
ι1 = ι2 = ι3 = ι4 = 0.
This implies that as n → +∞
Z Z
∗ ∗
µn ∥∇w̄n ∥θθ → 0, ∥∇w̄n ∥22 → ∥∇ū∥22 , |w̄n |2 |∇w̄n |2 dx → |ū|2 |∇ū|2 dx, ∥w̄n ∥2·2
2·2 2·2
∗ → ∥ū∥2·2∗ ,
RN RN

From these limits, we deduce that I0 (ū) = lim Iµn (w̄n ) = M0∗ (c). Moreover, if λ ¯ ̸= 0, without loss of
n→+∞
¯ > 0. Since λ
generality, we may assume that λ ¯ n → λ,
¯ we can assume λ ¯ n > 0 for sufficiently large n. We can
then choose cn = c, and since
¯ n w̄n , w̄n ⟩ → ⟨I ′ (ū) + λ
⟨Iµ′ n (w̄n ) + λ ¯ ū, ū⟩,

we conclude that w̄n → ū strongly in L2 (RN ), which completes the proof.

With the above results in hand, we are now in a position to complete the proofs of the remaining main
theorems, namely Theorem 1.2 through Theorem 1.9.
Proof of Theorem 1.2 . By Lemma 3.12 and Lemma 6.2, there exists a sequence {uµn } ⊂ Sr (cn ) with
0 < cn ≤ c satisfying
¯ µ uµ = 0 in X ∗ ,
Iµ′ n (uµn ) + λ Iµn (uµn ) → M0 (c), uµn ≥ 0.
n n r

Moreover, using Lemma 3.13, we have

N N N
S2 S2 S2
0 < M0 (c) = lim Mµn (c) ≤ lim mµn (c) + − δ′ = m∗0 (c) + − δ′ < .
n→+∞ n→+∞ 2N 2N 2N

43
Hence, one of the alternatives in Proposition 7.2 holds. Suppose that alternative (i) of Proposition 7.2
¯ ∈R
occurs, that is, up to a subsequence, uµn ⇀ ū ̸= 0 in H 1 (RN ), where ū solves problem (1.2) for some λ
and
1 N
I(ū) ≤ M0 (c) − S2,
2N
which gives that
1 N 1 N 1 N
lim Mµn (cn ) ≥ lim Iµn (ū) + S 2 ≥ lim mµn (∥ū∥22 ) + S 2 ≥ lim mµn (cn ) + S 2 . (7.16)
n→+∞ n→+∞ 2N n→+∞ 2N n→+∞ 2N
However, by Lemma 3.13, for any c ∈ (0, c0 ), we have
N
S2
Mµn (c) ≤ mµn (c) + − δ′ , for δ′ > 0,
2N
which yields a contradiction with (7.16). Consequently, this shows that necessarily (ii) of Proposition 7.2
¯ ū = 0 and I(ū) = M0 (c).
occurs, namely I ′ (ū) + λ
It now remains to show that there exists τ∗1 > 0 independent of n ∈ N such that if τ > τ∗1 , the Lagrange
¯ > 0. In fact, since uµ ∈ Sr (cn ) is bounded in H 1 (RN ) and L2·2∗ (RN ), by (2.4), there exist
multiplier λ n
constants K1 , K2 , K3 > 0 independent of n ∈ N such that

Z 4N−q(N−2)
Z  N(q−2)
2(N+2)
q 2 2
K1 ≤ |uµn | dx ≤ C2 (q, N)c 2(N+2) |uµn | |∇uµn | dx ≤ K2
RN RN

and Z
|∇uµn |2 dx ≤ K3 .
RN
We define the constant
(N − 2)K3
τ∗1 := .
(N + 2 − 4γq )K1
Then, we have

(N − 2)∥∇uµn ∥22 (N − 2)∥∇ū∥22

τ∗1 > lim q = q > 0.
n→+∞ (N + 2 − 4γq )∥uµ ∥q (N + 2 − 4γq )∥ū∥q
n

¯ ū = 0 and Q0 (ū) = 0, we obtain

From the equations I ′ (ū) + λ
 
¯λ∥ū∥2 = τ 1 − 4γq ∥ū∥q − N − 2 ∥∇ū∥2 . (7.17)
2 q 2
N +2 N +2
¯n =λ
Therefore, if τ > τ∗1 , it follows from (7.17) that lim λ ¯ > 0. This completes the proof.
n→+∞

Proof of Theorem 1.4 . By Lemma 4.3 and Lemma 6.2, there exists a sequence {uµn } ⊂ Sr (cn ) with 0 <
cn ≤ c satisfying
¯ µ uµ = 0 in X ∗ , Iµ (uµ ) → M̌0 (c), uµ ≥ 0.
Iµ′ n (uµn ) + λ n n r n n n

Moreover, using Lemma 4.4, we have

N N
S2 S2
0 < M̌0 (c) = lim M̌µn (c) ≤ − δ′ < .
n→+∞ 2N 2N

44
Hence, one of the alternatives in Proposition 7.2 holds. Now we suppose that (i) of Proposition 7.2 takes
¯ ∈R
place, that is, up to a subsequence, uµn ⇀ ū ̸= 0 in H 1 (RN ), where ū solves problem (1.2) for some λ
and
1 N
I(ū) ≤ M̌0 (c) − S 2 < 0. (7.18)
2N
We now distinguish the following three cases.
Case 1: q = 2 + N4 . From (1.13), (2.1) and the fact that Q0 (ū) = 0, we deduce that
N2 + 4 2 τ(N 2 + 4) 2+ N4 4
Z Z Z
2
I(ū) ≥ |∇ū|2 dx + |ū|2 |∇ū|2 dx − 2
C1 (2 + , N)c N |∇ū|2 dx ≥ 0,
2N(N + 2) RN N RN 2(N + 2) N RN
which is a contradiction with (7.18).
Case 2: N = 3, 2 + N4 < q < 4 + N4 or N ≥ 4, 2 + N4 < q ≤ 2∗ . To begin, we define
4
(N 2 + 4)q
  N(q−2)−4
1 q(N−2)−2N
ρ∗ := ρ∗ (c, τ, q, N) := q c N(q−2)−4 . (7.19)
τC1 (q, N) N[4N − q(N − 2)]
We then proceed by distinguishing between the following two subcases.
Subcase 1: ∥∇ū∥22 + ∥ū∇ū∥22 ≤ ρ∗ . From (2.1), (7.19) and Q0 (ū) = 0, we obtain
 Z
1 N −2
I(ū) ≥ − |∇ū|2 dx
2 N(N + 2) RN
  Z  N(q−2)
2 γq (N − 2) 1 4
Z
q 2N−q(N−2)
2 2 2
+ |ū| |∇ū| dx + τ − C1 (q, N)c 4 |∇ū| dx
N RN N(N + 2) q RN
≥ 0. (7.20)
Subcase 2: ∥∇ū∥22 + ∥ū∇ū∥22 ≥ ρ∗ . From (1.14), (2.4) and Q0 (ū) = 0, we have
 Z
1 N −2
I(ū) ≥ − |∇ū|2 dx
2 N(N + 2) RN
  Z  N(q−2)
2 γq (N − 2) 1 4N−q(N−2)
Z 2(N+2)
2 2 2 2
+ |ū| |∇ū| dx + τ − C2 (q, N)c 2(N+2) |ū| |∇ū| dx
N RN N(N + 2) q RN
 
4N − q(N − 2) 4N−q(N−2)
N(q−2)
≥ D1 (N) ∥∇ū∥22 + ∥ū∇ū∥22 − τ C2 (q, N)c 2(N+2) ∥∇ū∥22 + ∥ū∇ū∥22 2(N+2)


2q(N + 2)
   
2
N(q−2)
2 2(N+2)

2

4N+4−Nq
2 2(N+2) 4N − q(N − 2) 4N−q(N−2)
≥ ∥∇ū∥2 + ∥ū∇ū∥2 D1 (N) ∥∇ū∥2 + ∥ū∇ū∥2 −τ C2 (q, N)c 2(N+2)
2q(N + 2)
N(q−2)  4N+4−Nq   
2(N+2) 2(N+2) 4N − q(N − 2) 4N−q(N−2)
≥ ρ∗ D1 (N)ρ∗ −τ C2 (q, N)c 2(N+2)
2q(N + 2)
≥ 0. (7.21)
Thus, both (7.20) and (7.21) contradict to (7.18).
Case 3: N ≥ 4, 2∗ < q < 4 + N4 . By Q0 (ū) = 0, the interpolation inequality and the Sobolev inequality,
we have
 Z  Z
1 N −2 2 γq (N − 2) 1
Z
2 2 2
I(ū) = − |∇ū| dx + |ū| |∇ū| dx + τ − |ū|q dx
2 N(N + 2) RN N RN N(N + 2) q RN
 Z
1 N −2 2
Z
≥ − |∇ū|2 dx + |ū|2 |∇ū|2 dx
2 N(N + 2) RN N RN
∗ Z  2·2∗ −q Z  q−2∗
γq (N − 2) 1 2q−2
 
2 2
2 2 2
+τ − 2∗
|∇ū| dx |ū| |∇ū| dx .
N(N + 2) q S 2 RN RN

45
 1,θ 1 (RN )
Then, there exists a sufficiently small ρ̃∗ > 0 small enough such that for any ū ∈ Wrad (RN ) ∩ Hrad
2 2 2 2
satisfying ∥∇ū∥2 + ∥ū∇ū∥2 ≤ ρ̃∗ , we get I(ū) ≥ 0. Otherwise, if ∥∇ū∥2 + ∥ū∇ū∥2 ≥ ρ̃∗ , by choosing the
mass c > 0 sufficiently small and arguing similarly as in (7.21), we obtain a contradiction with (7.18). Thus,
conclusion (ii) of Proposition 7.2 holds, namely there exist ū ∈ Hr1 (RN ) ∩ L∞ (RN )\{0}, ū ≥ 0 and λ ¯ ∈R
such that
¯ ū = 0, 0 < ∥ū∥2 ≤ c, I(ū) = M̌0 (c).
I ′ (ū) + λ 2
¯ > 0. Therefore,
Following the same reasoning as in the proof of Theorem 1.2, for τ > τ∗1 , we obtain λ
1 N N
ū ∈ H (R ) ∩ L (R ) is a critical point of I on S(c).
∞ e

Proof of Theorem 1.6 . By Lemma 5.8 and Lemma 6.2, there exists a sequence {uµn } ⊂ Sr (cn ) with 0 <
cn ≤ c satisfying
¯ µ uµ = 0 in X ∗ , Iµ (uµ ) → M
Iµ′ n (uµn ) + λ b0 (c), uµn ≥ 0.
n n r n n

Moreover, by Lemma 5.11, there exists τ∗2 > 0 large enough such that
N N
bµn (c) ≤ S − δ′′ < S .
2 2
0<M
b0 (c) = lim M
n→+∞ 2N 2N
Using Q0 (ū) = 0, we deduce that
 Z  Z  Z
1 1 N +2 N +2 N −2 ∗
I(ū) = − |∇ū|2 dx + 1 − |ū|2 |∇ū|2 dx + − |ū|2·2 dx ≥ 0,
2 qγq RN qγq RN 4qγq 4N RN

which implies that (i) of Proposition 7.2 cannot take place. Hence, conclusion (ii) of Proposition 7.2 holds,
¯ ∈ R such that
namely there exist ū ∈ Hr1 (RN ) ∩ L∞ (RN )\{0}, ū ≥ 0 and λ
¯ ū = 0,
I ′ (ū) + λ 0 < ∥ū∥22 ≤ c, I(ū) = M
b0 (c).

¯ > 0. Let us set τ∗ = max{τ∗ , τ∗ } >

Moreover, proceeding as in the proof of Theorem 1.2, for τ > τ∗1 , we get λ 1 2
∗
0. Consequently, for any τ > τ , it follows from Proposition 7.2 that ∥ū∥22 = c and ū ∈ H 1 (RN ) ∩ L∞ (RN ) is
a critical point of I on S(c).
e Finally, to obtain a ground state normalized solution, we define

b 0 (c) := inf{I(v) : v ∈ S(c),

m e dI| e (v) = 0}.
S(c)

Clearly, m b 0 (c) ≤ I(v) = M

b0 (c). We then take a sequence ūn ∈ S(c)
e such that dI| e (ūn ) = 0 and I(ūn ) →
S(c)
m e ∩ L∞ (RN )\{0} and
b 0 (c). By employing a similar argument as in Proposition 7.2, we obtain ū0 ∈ S(c)
¯λ0 ∈ R such that, up to a subsequence, there holds

¯ 0 ū0 = 0,
I ′ (ū0 ) + λ I(ū0 ) = m
b 0 (c).
¯ 0 > 0, and hence ∥ū0 ∥2 = c. This
Using a similar argument as in the previous proof, we can show that λ 2
b 0 (c).
implies that ū0 is a minimizer of m

Proof of Theorem 1.7 . It suffices to show that λ¯ > 0 for any τ > 0. We first show λ
¯ ≥ 0 . To this end, we
¯ < 0. Then for sufficiently large n, we have λ
argue by contradiction that λ ¯ µ < 0. In view of Lemma 6.2,
n
we have
bµn (c) and ∥uµn ∥22 = c.
Iµn (uµn ) = M (7.22)

46
 N √
For simplicity, for any fixed n sufficiently large, define w̄ := uµn and let w̄t,s (x) := s 2 t w̄(sx) with t, s > 0.
We also define
µn θ(1+γθ ) θ s2
Z Z Z
hI (t, s) := Iµn (w̄t,s ) = s t2 |∇w̄|θ dx + t |∇w̄|2 dx + sN+2t 2 |w̄|2 |∇w̄|2 dx
θ RN 2 RN RN
1 N(2∗ −1) 2∗
Z Z
τ N(q−2) q ∗
− s 2 t2 |w̄|q dx − s t |w̄|2·2 dx,
q RN 2 · 2∗ RN
Z Z
θ
hQ (t, s) := Qµn (w̄t,s ) = µn (1 + γθ )sθ(1+γθ )t 2 |∇w̄|θ dx + s2t |∇w̄|2 dx
RN RN
Z Z
N(q−2) q
+ (N + 2)sN+2t 2 |w̄|2 |∇w̄|2 dx − τγq s 2 t2 |w̄|q dx
RN RN
Z
∗ −1) ∗ ∗
− γ2·2∗ sN(2 t2 |w̄|2·2 dx.
RN

By a direct calculation, we have

µn 1 1
Z Z Z Z Z
∂hI τ ∗
(1, 1) = |∇w̄|θ dx + |∇w̄|2 dx + 2 |w̄|2 |∇w̄|2 dx − |w̄|q dx − |w̄|2·2 dx
∂t 2 RN 2 RN R N 2 R N 2 RN
¯λµ Z
=− n |w̄|2 dx > 0,
2 RN
Z Z Z
∂hI
(1, 1) = µn (1 + γθ ) |∇w̄|θ dx + |∇w̄|2 dx + (N + 2) |w̄|2 |∇w̄|2 dx
∂s RN RN RN
Z Z
∗
− τγq |w̄|q dx − γ2·2∗ |w̄|2·2 dx = 0,
RN RN
∂2 h I
Z Z
(1, 1) = µn (1 + γθ )(θ(1 + γθ ) − N − 2) |∇w̄|θ dx − N |∇w̄|2 dx
∂s2 RN RN
N2 + N + 2
 Z  Z
N(q − 2) − 2 N +2 ∗
+ τγq N + 1 − q
|w̄| dx + N +1− |w̄|2·2 dx
2 R N 4 N − 2 RN

< 0,

which imply that for |δs | small enough and δt > 0, there holds

hI (1 − δt , 1 + δs ) < hI (1, 1). (7.23)

∂hQ
Clearly, hQ (1, 1) = 0. Now, we show that ∂s (1, 1) ̸= 0. In fact, if this is not true, by a direct computation,
we get
Z Z Z
∂hQ
(1, 1) = µn θ(1 + γθ )2 |∇w̄|θ dx + 2 |∇w̄|2 dx + (N + 2)2 |w̄|2 |∇w̄|2 dx
∂s RN RN RN
Z Z
∗
− τqγ2q |w̄|q dx − γ2·2∗ N(2∗ − 1) |w̄|2·2 dx = 0.
RN RN

Together with Qµn (w̄) = 0, this yields

Z Z
µn (1 + γθ )(N + 2 − θ(1 + γθ )) |∇w̄|θ dx + N |∇w̄|2 dx
RN RN
Z Z
∗
= τγq (N + 2 − qγq ) |w̄|q dx + γ2·2∗ (N + 2 − N(2∗ − 1)) |w̄|2·2 dx.
RN RN

4N
< θ < 4N+4 ∗
However, this is impossible due to the fact that N+2 N+2 , qγq ≥ N + 2 and N + 2 < N(2 − 1). Thus,
by the implicit function theorem, we infer that there exist ε > 0 and a continuous function ĥ : [1 − ε, 1 + ε] 7→

47
R such that ĥ(1) = 1 and hQ (t, ĥ(t)) = 0 for t ∈ [1 − ε, 1 + ε]. Combining (7.22) and (7.23), we deduce that
bµn ((1 − ε)c) = m
M b µn ((1 − ε)c) = inf Iµn (u) ≤ Iµn (w̄1−ε,ĥ(1−ε) ) < Iµn (w̄) = M
bµn (c),
u∈Λµn ((1−ε)c)

which contradicts Lemma 5.5, which states that mb µ (c) = M

bµ (c) is non-increasing on (0, +∞) for any µ > 0.
¯
Therefore, we infer that λ ≥ 0. Taking into account (7.17), we derive that
    Z  N(q−2)
N −2 2 4γq q 4γq 4N−q(N−2)
2 2
2(N+2)
∥∇ū∥2 ≤ τ 1 − ∥ū∥q ≤ τ 1 − C2 (q, N)c 2(N+2) |ū| |∇ū| ,
N +2 N +2 N +2 RN

which implies that

∥∇ū∥22 → 0 as c → 0.
Thus, it follows from (7.17) that there exists a sufficiently small c∗2 > 0 such that for any c ∈ (0, c∗2 ), we have
¯ > 0. From this, using Proposition 7.2 we conclude that ∥ū∥2 = c and ū ∈ H 1 (RN ) ∩ L∞ (RN ) is a critical
λ 2
point of I on S(c).
e The ground state solution can be obtained using the same approach.
e × R+ is a solution to (1.4). Then, by (1.4) and the
Proof of Theorem 1.9 . Suppose that (u, λ) ∈ S(c)
Pohozaev identity, we obtain
Z Z Z Z Z
∗
|∇u|2 dx + 4 |u|2 |∇u|2 dx + λ |u|2 dx = τ |u|q dx + |u|2·2 dx,
RN RN RN RN RN
and
2−N N
Z Z Z Z Z
λN τN ∗
|∇u|2 dx + (2 − N) |u|2 |∇u|2 dx − |u|2 dx + |u|q dx + ∗
|u|2·2 dx = 0.
2 RN R N 2 RN q RN 2 · 2 R N

Combining these two equations, we get

4N − (N − 2)q N −2
0 < λ∥u∥22 = τ ∥u∥qq − ∥∇u∥22 ≤ 0,
(N + 2)q N +2
which is a contradiction. The proof is complete.

A Appendix
This section is devoted to providing an L∞ estimate, a Brezis-Lieb type result, and a profile decomposi-
tions for the approximating solutions {un } ⊂ X.
e
To be more specific, assume that (µn , un , λn ) ∈ (0, 1] × Xe × R satisfies the following equation with a
general nonlinearity:
−µ∆θ u − ∆u − u∆(u2 ) + λu = L ′ (u) in RN , (A.1)
where the function L : R → [0, ∞) is of class C 1 and satisfies
∗ −1
L ′ (s) ≤ C(|s| + |s|2·2 ) for any s ∈ R and some constant C > 0. (A.2)
In the following, we always assume that
n o
sup max µn ∥∇un ∥θθ , ∥un ∇un ∥22 , ∥∇un ∥22 , |λn | < +∞.
n∈N

Thus, there exists (u0 , λ0 ) such that as n → +∞,

un ⇀ u0 in H 1 (RN ), un ∇un ⇀ u0 ∇u0 in L2 (RN )N , un → u0 a.e. in RN , λn → λ0 .
To begin, we establish the following result by using Moser’s iteration.

48
Lemma A.1. There exists a constant C > 0 such that ∥un ∥∞ ≤ C and ∥u0 ∥∞ ≤ C.
Proof. Following the idea from [36, Proposition 3.1] with slight modifications, for any T > 0, we define
the truncated function uTn as (
T T, if un > T,
un =
un , if un ≤ T.
Taking φ = un |uTn |4(r−1) with r > 1 as the test function in (A.1), and using (A.2) and the boundedness of
{λn }, we obtain
Z Z
4(r−1)
(4r − 3) un |∇un |2 dx + T 4(r−1) |∇un |2 dx
{un ≤T } {un >T }
Z Z
4(r−1) 2
+ 2(4r − 3) un un |∇un |2 dx + 2 T 4(r−1) u2n |∇un |2 dx
{un ≤T } {un >T }
Z Z
∗
≤ −λn u2n |uTn |4(r−1) dx +C (|un |2 + |un |2·2 )|uTn |4(r−1) dx
RN RN
Z Z 
∗
≤C u2n (uTn )4(r−1) dx + un2·2 (uTn )4(r−1) dx . (A.3)
RN RN

On the other hand, by a direct computation, we deduce that

Z Z
|∇(un (uTn )2(r−1) )|2 dx + |∇(u2n (uTn )2(r−1) )|2 dx
RN RN
Z Z
4(r−1)
= (2r − 1)2 un |∇un |2 dx + T 4(r−1) |∇un |2 dx
{un ≤T } {un >T }
Z Z
2(2r−1)
+ 4r2 un |∇un |2 dx + 4 T 4(r−1) u2n |∇un |2 dx. (A.4)
{un ≤T } {un >T }

Combining (A.3), (A.4), and applying the Sobolev inequality, we have

Z  2∗ Z  2∗
∗ 2 ∗ 2
|un (uTn )2(r−1) |2 dx + |u2n (uTn )2(r−1) |2 dx
RN RN
Z Z  2 
≤ Cr2 u2n (uTn )4(r−1) dx + an u2n (uTn )2(r−1) dx , (A.5)
RN RN
8
where an := unN−2 . We claim that an ∈ Lt (RN ) for some t > N2 . In fact, by (A.5), for any K > 0 to be
determined later, we apply the Hölder inequality to obtain
Z  2∗ Z Z  2
2
2 T 2(r−1) 2∗ 8
|un (un ) | dx ≤ Cr2 u2n (uTn )4(r−1) dx +Cr2 K N−2 u2n (uTn )2(r−1) dx
RN RN RN
Z  2 Z  2N  N−2
N
∗ N 
2 2 T 2(r−1) N−2
+Cr un2·2 dx un (un ) dx . (A.6)
|un |≥K RN

Now, let 4r = 2 · 2∗ and choose K > 0 sufficiently large such that

Z 2
2 2·2∗
N 1
Cr un dx < .
|un |≥K 2
Then, letting T → +∞ in (A.6), we have
Z  2∗ Z Z
2
2·2∗ r 8
un dx ≤ 2Cr2 un4r−2 dx + 2Cr2 K N−2 u4r
n dx,
RN RN RN

49
 ∗
which implies that un ∈ L2·2 r (RN ), i.e., an ∈ Lt (RN ) with t = Nr N
2 > 2 . Thus the claim holds.
Back to (A.5), utilizing the Hölder inequality again, we derive that
Z  2∗ Z  2∗
∗ 2 ∗ 2
|un (uTn )2(r−1) |2 dx + |u2n (uTn )2(r−1) |2 dx
RN RN
! t ′ −1
 
Z  1′ Z
t 2∗ −2t ′ t′
Z  2t ′  t1′
2 T 4t ′ (r−1) 4t ′ −2∗ t ′ −1
≤ Cr  (un ) un dx un dx + ∥an ∥t u2n (uTn )2(r−1) dx 

N R N R RN

 1′ 2t ′  t1′
Z Z !
t

4t ′ (r−1) 4t ′ −2∗
≤ Cr2 (uTn ) un dx + 2 T 2(r−1)
un (un ) dx , (A.7)
RN RN

where t ′ = t
t−1 < N
N−2 . Assume that
Z Z
′ ∗ ′
u4t
n
r−2
dx + u4rt
n dx < +∞.
RN RN

Then, letting T → +∞ in (A.7), we get

Z  2∗ Z  2∗ Z  1′ Z  1′ !
2 2
2∗ (2r−1) ∗r ′ r−2∗ t ′ t
un dx + u2·2
n dx ≤ Cr 2
u4t
n dx + u4rt
n dx ,
RN R N R N R N

which implies
1 1 1 1 !
2·2∗ r 2·2∗ r
Z  Z  Z  Z 
2∗ (2r−1) ∗r 1 ′ r−2∗ 4rt ′ ′ 4rt ′
un dx + u2·2
n dx ≤ (Cr )2 4r
u4t
n dx + u4rt
n dx .
RN R N RN R N

2 ∗ ′ ∗
Set d := 2t ′ > 1, and define ri+1 = ri d for i ∈ N with 4r0 t = 2 · 2 . Clearly, the sequence {ri } is positive
and increasing since r0 > 0. Then we obtain
1 1
2·2∗ ri+1 2·2∗ ri+1
Z  Z 
2∗ (2ri+1 −1) 2·2∗ ri+1
un dx + un dx
RN RN
1 1 !
2·2∗ ri 2·2∗ ri
Z  Z 
1
2 2∗ (2r −1) ∗
≤ (Cri+1 ) 4ri+1 un i dx + un2·2 ri dx .
RN RN

By iterating this process, we arrive at

1 1
2·2∗ ri+1 2·2∗ ri+1
Z  Z 
2∗ (2ri+1 −1) 2·2∗ ri+1
un dx + un dx
RN RN
1 1 !
i+1 1
Z 
2·2∗
Z 
2·2∗
∗ ∗
≤ (Crk2 ) 4rk
∏ u2n dx + un2·2 dx .
k=0 RN RN

Letting i → +∞, we deduce that

 
1
∥un ∥L∞ (RN ) ≤ C ∥un ∥L22∗ (RN ) + ∥un ∥L2·2∗ (RN ) .

Thus, there exists a constant C > 0 such that ∥un ∥L∞ (RN ) ≤ C. Similarly, the bound ∥u0 ∥L∞ (RN ) ≤ C can be
derived in the same way.

50
Remark A.2. By Lemma A.1, we get that ∥un ∥∞ ≤ C, which together with the similar arguments as in [31,
Theorem 4.1], we can see that (u0 , λ0 ) satisfies the following equation:
−∆u − u∆(u2 ) + λu = L ′ (u) in RN . (A.8)
By Corollary 4.23 and Theorem 4.24 of [20], we have u0 ∈ C 1,β0 (RN ) for some β0 > 0. For the approxi-
1,β
mating solutions un , it follows from the regularity result in [49] that un ∈ Cloc 1 (RN ) for some β1 ∈ (0, 1).
Lemma A.3. Let ūn := un − u0 . Then we have
Z Z Z
|un |2 |∇un |2 dx = |ūn |2 |∇ūn |2 dx + |u0 |2 |∇u0 |2 dx + on (1).
RN RN RN

Proof. By a direct computation, we obtain

Z Z
|un |2 |∇un |2 dx = |ūn + u0 |2 |∇ūn + ∇u0 |2 dx
RN RN
Z Z Z
= |ūn |2 |∇ūn |2 dx + 2 |ūn |2 ∇ūn ∇u0 dx + |ūn |2 |∇u0 |2 dx
RN RN RN
Z Z Z
+2 ūn u0 |∇ūn |2 dx + 4 ūn u0 ∇ūn ∇u0 dx + 2 ūn u0 |∇u0 |2 dx
RN RN RN
Z Z Z
+ |u0 |2 |∇ūn |2 dx + 2 ∇ūn ∇u0 |u0 |2 dx + |u0 |2 |∇u0 |2 dx. (A.9)
RN RN RN

Thus, we need only prove that the right hand side of (A.9) tends to 0 as n → +∞, except for the first and
last terms. Following the idea in [41, Lemma 3.1], we first show that
1 2 2
µn |∇un |θ → 0 in Lloc (RN ), un ∇un → u0 ∇u0 in Lloc (RN )N , ∇un → ∇u0 in Lloc (RN ). (A.10)
To this aim, taking φ = un ϕ with ϕ ∈ C0∞ (RN ) as the test function in (A.1), we obtain
Z Z Z Z
µn |∇un |θ ϕdx + µn |∇un |θ−2 ∇un ∇ϕun dx + |∇un |2 ϕdx + ∇un ∇ϕun dx
RN RN N RN
Z Z Z R Z
+4 |un |2 |∇un |2 ϕdx + 2 u3n ∇un ∇ϕdx + λn u2n ϕdx = L ′ (un )un ϕdx. (A.11)
RN RN RN RN

By the boundedness of {µn ∥∇un ∥θθ }, we deduce that

Z  Z 1
 θ−1 Z 1
θ θ
θ−2 θ θ
µn |∇un | ∇un ∇ϕun dx ≤ Cµn µn θ
|∇un | dx |∇ϕ| dx → 0, as n → +∞.
RN RN RN

We now claim that

Z Z Z Z
∇un ∇ϕun dx + 2 u3n ∇un ∇ϕdx + λn u2n ϕdx − L ′ (un )un ϕdx
RN RN RN RN
Z Z Z Z
→ ∇u0 ∇ϕu0 dx + 2 u30 ∇u0 ∇ϕdx + λ0 u20 ϕdx − L ′ (u0 )u0 ϕdx.
RN RN RN RN

We only prove that

Z Z Z Z
lim ∇un ∇ϕun dx = ∇u0 ∇ϕu0 dx, lim L ′ (un )un ϕdx = L ′ (u0 )u0 ϕdx,
n→+∞ RN RN n→+∞ RN RN

since the convergence of the other terms can be established using similar arguments. For any ε > 0, there
exists a bounded set Ω such that suppϕ ⊂ Ω satisfying
Z Z
∇un ∇ϕun dx < ε and L ′ (un )un ϕdx < ε, for any n ∈ N.
RN \Ω RN \Ω

51
And it follows from (A.2) and ∥un ∥L∞ (RN ) ≤ C that for any measurable subset E ⊂ RN with meas(E)
sufficiently small, Z
1
∇un ∇ϕun dx ≤ ∥un ∥∞ ∥∇ϕ∥∞ ∥∇un ∥L2 (E) (meas(E)) 2 < ε,
E
and Z Z
∗
L ′ (un )un ϕdx ≤ C (|un |2 + |un |2·2 )ϕdx ≤ Cmeas(E) < ε, for any n ∈ N,
E E
where meas(E) denotes the Lebesgue measure of the subset E. Therefore, the claim holds by using the
Vitali’s convergence theorem (see [10, Theorem 3]). Taking φ = u0 ϕ as the test function in (A.8), we have
Z Z Z Z Z
|∇u0 |2 ϕdx + ∇u0 ∇ϕu0 dx + 4 |u0 |2 |∇u0 |2 ϕdx + 2 u30 ∇u0 ∇ϕdx + λ0 u20 ϕdx
RN RN RN RN RN
Z
= L ′ (u0 )u0 ϕdx. (A.12)
RN

Combining (A.11) and (A.12), we obtain (A.10). Since u0 ∈ C 1,β0 (RN ), using the similar arguments as in
[17, Proposition A.1], we get that for any BR ⊂ RN ,
Z
|ūn |2 |∇ūn |2 dx → 0. (A.13)
BR

Notice that u0 ∈ H 1 (RN ), and by the regularity of u0 , we deduce that u0 (x) → 0 as |x| → +∞. Thus back to
(A.9), for suffciently large R > 0, it follows from Lemma A.1 and (A.13) that
Z Z Z
ūn u0 |∇ūn |2 dx = ūn u0 |∇ūn |2 dx + ūn u0 |∇ūn |2 dx → 0 as n → +∞.
RN BR RN \BR

The arguments for the other terms in (A.9), except for the first and last ones, follow in an analogous manner.

Inspired by [43, Theorem 1.4], we establish the following profile decompositions for the approximating
solutions {un } ⊂ X.
e

Theorem A.4. There are sequences {ũi }∞ 1 N i ∞ N 0 i

i=0 ⊂ H (R ), {yn }i=0 ⊂ R for any n ≥ 1, such that yn = 0, |yn −
j
yn | → +∞ for i ̸= j, and passing to a subsequence, the following conditions hold for any i ≥ 0:

un (· + yin ) ⇀ ũi in H 1 (RN ) as n → +∞,

un (· + yin )∇un (· + yin ) ⇀ ũi ∇ũi in L2 (RN )N as n → +∞,

Z i Z Z
lim |∇un |2 dx = ∑ |∇ũ j |2 dx + lim |∇uin |2 dx,
n→+∞ RN N n→+∞ RN
j=0 R
Z i Z Z
lim |un |2 |∇un |2 dx = ∑ |ũ j |2 |∇ũ j |2 dx + lim |uin |2 |∇uin |2 dx,
n→+∞ RN N n→+∞ RN
j=0 R
j
where uin (·) := un − ∑ij=0 ũ j (· − yn ) and
Z i Z Z
lim sup L (un )dx = ∑ L (ũ j )dx + lim sup L (uin )dx. (A.14)
n→+∞ RN j=0 R
N n→+∞ RN

Moreover, if in addition L satisfies

L (s) L (s)
lim = lim ∗ = 0, (A.15)
s→0 |s|2 s→+∞ |s|2·2

52
then  Z 
lim lim sup L (uin )dx = 0. (A.16)
i→+∞ n→+∞ RN

Remark A.5. We note that Theorem A.4 relies heavily on the Brezis-Lieb type property developed in
Lemma A.3. This property requires that the weak limit ũi for every i ≥ 0 actually weakly solves the limiting
equation (A.8). Compared with the decomposition results in [17], our approach only requires the regularity
and uniformly boundedness of the solutions, without depending on the sign of λ ∈ R. Furthermore, we
perform a more refined decomposition of the approximating solution sequence {un }, and our method allows
for iteration up to an infinite number of times. We believe that Theorem A.4 possesses significant potential
for applications in this field.
Before proceeding to Theorem A.4, we present the following new variants of Lions’ lemma.
Lemma A.6. Suppose that {wn } ⊂ H 1 (RN ) and {wn ∇wn } ⊂ L2 (RN )N are bounded, and that for some
r > 0, we have Z
sup |wn |2 dx → 0 as n → +∞.
y∈RN B(y,r)

Then, for any continuous function L : R → [0, ∞) satisfying (A.15), it holds

Z
L (wn )dx → 0 as n → +∞.
RN

Proof. Let ε > 0 and 2 < β < 2 · 2∗ be given, and suppose that L satisfies (A.15). Then, there exist constants
0 < δ < M and cε > 0 such that
∗
L (s) ≤ ε|s|2 if |s| ∈ [0, δ]; L (s) ≤ ε|s|2·2 if |s| ∈ (M, +∞); L (s) ≤ cε |s|β if |s| ∈ (δ, M].
Therefore, by using Lemma 2.1, we deduce that
Z Z
∗
lim sup L (wn )dx ≤ ε lim sup (|wn |2 + |wn |2·2 )dx.
n→+∞ RN n→+∞ RN
∗
Since the Sobolev inequality ensures that {wn } is bounded in L2 (RN ) and L2·2 (RN ), we conclude by letting
ε → 0.
Proof of Theorem A.4 . First we claim that: Up to a subsequence, there exist sequences k ∈ N ∪ {∞},
{ũi }ki=0 ⊂ X,
e for 0 ≤ i < k + 1 (if k = ∞, then k + 1 = ∞ as well), {uin } ⊂ X, e {yin } ⊂ RN and positive
k k 0
numbers {ci }i=0 , {ri }i=0 such that yn = 0, r0 = 0 and for any 0 ≤ i < k + 1 one has
1. un (· + yin ) ⇀ ũi in H 1 (RN ), un (· + yin )∇un (· + yin ) ⇀ ũi ∇ũi in L2 (RN )N and un (· + yin )χB(0,n) → ũi in
L2 (RN ) as n → +∞, ũi ̸= 0 if i ≥ 1;
j
2. |yin − yn | ≥ n − ri − r j for 0 ≤ j ̸= i < k + 1 and sufficiently large n;
3. u−1 i i−1 i
n = un and un := un − ũi (· − yn ) for n ≥ 1;
i−1 2 ≥ 21 supy∈RN i−1 2
R R
4. B(yin ,ri ) |un | dx ≥ ci B(y,ri ) |un | dx for sufficiently large n, ri ≥ max{i, ri−1 }, if i ≥ 1,
and
3
Z
ci = lim lim sup sup |uni−1 |2 dx > 0.
4 r→+∞ n→+∞ y∈RN B(y,r)

5.
i i
lim ∥∇un ∥22 = ∑ ∥∇ũ j ∥22 + lim ∥∇uin ∥22 , lim ∥un ∇un ∥22 = ∑ ∥ũ j ∇ũ j ∥22 + n→+∞
lim ∥uin ∇uin ∥22 .
n→+∞ n→+∞ n→+∞
j=0 j=0

53
By the boundedness of {un } in X,
e we obtain

un ⇀ u0 in H 1 (RN ), un ∇un ⇀ u0 ∇u0 in L2 (RN )N , un χBn → u0 in L2 (RN ),

where χBn denotes the characteristic function on the ball Bn . Let ũ0 := u0 , u0n := un − ũ0 . If
Z
lim sup |u0n |2 dx = 0
n→+∞ B(y,r)
y∈RN

for every r ≥ 1, then the proof of this claim is completed with K = 0. Otherwise, following an argument
analogous to [43, Theorem 1.4], we can deduce that the first four conclusions hold. Noting that |y1n | ≥ n − r1
and that {un (· + y1n )} is bounded in X,
e up to a subsequence, we deduce that there exists ũ1 ̸= 0 such that

un (· + y1n ) ⇀ ũ1 in H 1 (RN ), un (· + y1n )∇un (· + y1n ) ⇀ ũ1 ∇ũ1 in L2 (RN )N , un (· + y1n )χBn → ũ1 in L2 (RN ).

Since un (· + y1n ) weakly solves (A.1), by the standard argument as in Lemma A.1, we deduce that ũ1 weakly
solves (A.8). Thus, by Remark A.2 and Lemma A.3, we obtain
Z Z Z
|∇u0n (· + y1n )|2 dx = |∇ũ1 |2 dx + |∇u1n (· + y1n )|2 dx + on (1),
RN RN RN
Z Z Z
|u0n (· + y1n )|2 |∇u0n (· + y1n )|2 dx = |ũ1 |2 |∇ũ1 |2 dx + |u1n (· + y1n )|2 |∇u1n (· + y1n )|2 dx + on (1),
RN RN RN

where u1n := u0n − ũ1 (· − y1n ). Applying Remark A.2 and Lemma A.3 again, we have
Z Z Z Z
|∇un |2 dx = |∇ũ0 |2 dx + |∇ũ1 |2 dx + |∇u1n |2 dx + on (1),
RN RN RN RN
Z Z Z Z
|un |2 |∇un |2 dx = |ũ0 |2 |∇ũ0 |2 dx + |ũ1 |2 |∇ũ1 |2 dx + |u1n |2 |∇u1n |2 dx + on (1).
RN RN RN RN

Proceeding similarly as in [43] and iterating this procedure, we can prove this claim. Moreover, using
Lemma A.6, we can see that (A.14) and (A.16) also hold, thus completing the proof.

References
[1] M. Agueh, Sharp Gagliardo-Nirenberg inequalities via p-Laplacian type equations, NoDEA Nonlinear Differential
Equations Appl., 15 (2008) 457-472.

[2] A. Ambrosetti and Z.-Q. Wang, Positive solutions to a class of quasilinear elliptic equations on R, Discrete Contin.
Dyn. Syst., 9 (2003) 55-68.

[3] T. Bartsch and N. Soave, A natural constraint approach to normalized solutions of nonlinear Schrödinger equations
and systems, J. Funct. Anal., 272 (2017) 4998-5037.

[4] J. Bellazzini, L. Jeanjean and T. J. Luo, Existence and instability of standing waves with prescribed norm for a
class of Schrödinger-Poisson equations, Proc. Lond. Math. Soc., (3) 107 (2013) 303-339.

[5] H. Berestycki and P. L. Lions, Nonlinear scalar field equations, II: Existence of infinitely many solutions, Arch.
Ration. Mech. Anal., 82 (1983) 347-375.

[6] J. M. Bezerra do Ó, O. H. Miyagaki and S. H. M. Soares, Soliton solutions for quasilinear Schrödinger equations
with critical growth, J. Differential Equations, 248 (2010) 722-744.

[7] H. Brézis and L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical Sobolev exponents,
Commun. Pure Appl. Math., 36 (1983) 437-477.

54
 [8] L. Brüll, H. Lange and E. de Jager, Stationary, oscillatory and solitary waves type solutions of singular nonlinear
Schrödinger equations, Math. Methods Appl. Sci., 8 (1986) 559-575.

[9] L. Brüll and H. Lange, Solitary waves for quasilinear Schrödinger equations, Expo. Math., 4 (1986) 278-288.

[10] M. C. Chen, H. L. Lin and H. C. Wang, Vitali convergence theorem and Palais Smale conditions, Differential
Integral Equations, 15 (2002) 641-656.

[11] S. T. Chen and X. H. Tang, Another look at Schrödinger equations with prescribed mass, J. Differential Equations,
386 (2024) 435-479.

[12] S. T. Chen and X. H. Tang, Normalized solutions for Kirchhoff equations with Sobolev critical exponent and
mixed nonlinearities, Math. Ann., 391 (2024) 2783-2836.

[13] M. Colin and L. Jeanjean, Solutions for a quasilinear Schrödinger equation: a dual approach, Nonlinear Anal., 56
(2004) 213-226.

[14] M. Colin, L. Jeanjean and M. Squassina, Stability and instability results for standing waves of quasi-linear
Schrödinger equations, Nonlinearity, 23 (2010) 1353-1385.

[15] T. Deng, M. Squassina, J. J. Zhang and X. X. Zhong, Normalized solutions of quasilinear Schrödinger equations
with a general nonlinearity, Asymptot. Anal., 140 (2024) 5-24.

[16] Y. B. Deng, Y. X. Guo and S. S. Yan, Multiple solutions for critical quasilinear elliptic equations, Calc. Var. Partial
Differential Equations, 58 (2019) Paper No. 2, 26 pp.

[17] F. S. Gao and Y. X. Guo, Existence of normalized solutions for mass super-critical quasilinear Schrödinger equa-
tion with potentials, J. Geom. Anal., 34 (2024), Paper No. 329, 39 pp.

[18] F. S. Gao and Y. X. Guo, Normalized solution for a quasilinear Schrödinger equation with potentials and general
nonlinearities, Commun. Pure Appl. Anal., 24 (2025) 507-534.

[19] N. Ghoussoub, Duality and perturbation methods in critical point theory, Cambridge Tracts in Mathematics,
Cambridge University Press, vol. 107, Cambridge, (1993). With appendices by David Robinson.

[20] Q. Han and F. H. Lin, Elliptic partial differential equations, second ed., Courant Lecture Notes in Mathematics,
vol. 1, Courant Institute of Mathematical Sciences, New York; American Mathematical Society, Providence, RI
(2011).

[21] L. Jeanjean, Existence of solutions with prescribed norm for semilinear elliptic equations, Nonlinear Anal., 28
(1997) 1633-1659.

[22] L. Jeanjean, J. Jendrej, T. T. Le and N. Visciglia, Orbital stability of ground states for a Sobolev critical
Schrödinger equation, J. Math. Pures Appl., (9) 164 (2022) 158-179.

[23] L. Jeanjean and T. T. Le, Multiple normalized solutions for a Sobolev critical Schrödinger equations, Math. Ann.,
384 (2022) 101-134.

[24] L. Jeanjean and T. J. Luo, Sharp nonexistence results of prescribed L2 -norm solutions for some class of
Schrödinger-Poisson and quasi-linear equations, Z. Angew. Math. Phys., 64 (2013) 937-954.

[25] L. Jeanjean, T. J. Luo and Z.-Q.Wang, Multiple normalized solutions for quasi-linear Schrödinger equations, J.
Differential Equations, 259 (2015) 3894-3928.

[26] L. Jeanjean, J. J. Zhang and X. X. Zhong, Existence and limiting profile of energy ground states for a quasi-linear
Schrödinger equations: Mass super-critical case, arXiv:2501.03845, (2025).

[27] S. Kurihara, Large-amplitude quasi-solitons in superfluid films, J. Phys. Soc. Japan, 50 (1981) 3262-3267.

55
[28] I. Kuzin and S. Pohozaev, Entire solutions of semilinear elliptic equations, Progress in Nonlinear Differential
Equations and their Applications, vol. 33, Birkhäuser, Berlin (1997).

[29] E. W. Laedke, K. H. Spatschek and L. Stenflo, Evolution theorem for a class of perturbed envelope soliton solu-
tions, J. Math. Phys., 24 (1983) 2764-2769.

[30] H. Lange, M. Poppenberg and H. Teismann, Nash-Moser methods for the solution of quasilinear Schrödinger
equations, Comm. Partial Differential Equations, 24 (1999) 1399-1418.

[31] H. W. Li and W. M. Zou, Quasilinear Schrödinger equations: ground state and infinitely many normalized solu-
tions, Pacific J. Math., 322 (2023) 99-138.

[32] X. F. Li, Existence of normalized ground states for the Sobolev critical Schrödinger equation with combined
nonlinearities, Calc. Var. Partial Differential Equations, 60 (2021) Paper No. 169, 14 pp.

[33] P. L. Lions, The concentration-compactness principle in the calculus of variations. The locally compact case. II,
Ann. Inst. H. Poincaré Anal. Non Linéaire, 1 (1984) 223-283.

[34] J. Q. Liu, X. Q. Liu and Z.-Q. Wang, Multiple sign-changing solutions for quasilinear elliptic equations via
perturbation method, Comm. Partial Differential Equations, 39 (2014) 2216-2239.

[35] X. Q. Liu, J. Q. Liu and Z.-Q. Wang, Quasilinear elliptic equations with critical growth via perturbation method,
J. Differential Equations, 254 (2013) 102-124.

[36] X. Q. Liu, J. Q. Liu and Z.-Q. Wang, Ground states for quasilinear Schrödinger equations with critical growth,
Calc. Var. Partial Differential Equations, 46 (2013) 641-669.

[37] X. Q. Liu, J. Q. Liu and Z.-Q. Wang, Quasilinear elliptic equations via perturbation method, Proc. Amer. Math.
Soc., 141 (2013) 253-263.

[38] J. Q. Liu, Y. Q. Wang and Z.-Q. Wang, Soliton solutions for quasilinear Schrödinger equations, II, J. Differential
Equations, 187 (2003) 473-493.

[39] J. Q. Liu, Y. Q. Wang and Z.-Q. Wang, Solutions for quasilinear Schrödinger equations via the Nehari method,
Comm. Partial Differential Equations, 29 (2004) 879-901.

[40] J. Q. Liu and Z.-Q. Wang, Soliton solutions for quasilinear Schrödinger equations, I, Proc. Amer. Math. Soc., 131
(2003) 441-448.

[41] J. Q. Liu and Z.-Q. Wang, Multiple solutions for quasilinear elliptic equations with a finite potential well, J.
Differential Equations, 257 (2014) 2874-2899

[42] A. M. Mao and S. Y. Lu, Normalized solutions to the quasilinear Schrödinger equations with combined nonlin-
earities, Proc. Edinb. Math. Soc., 67 (2024) 349-387.

[43] J. Mederski, Nonradial solutions for nonlinear scalar field equations, Nonlinearity, 33 (2020) 6349-6381.

[44] M. Poppenberg, K. Schmitt and Z.-Q. Wang, On the existence of soliton solutions to quasilinear Schrödinger
equations, Calc. Var. Partial Differential Equations, 14 (2002) 329-344.

[45] D. Ruiz and G. Siciliano, Existence of ground states for a modified nonlinear Schrödinger equation, Nonlinearity,
23 (2010) 1221-1233.

[46] J. Serrin and M. Tang, Uniqueness of ground states for quasilinear elliptic equations, Indiana Univ. Math. J., 49
(2000) 897-923.

[47] E. A. B. Silva and G. F. Vieira, Quasilinear asymptotically periodic Schrödinger equations with critical growth,
Calc. Var. Partial Differential Equations, 39 (2010) 1-33.

56
[48] N. Soave, Normalized ground states for the NLS equation with combined nonlinearities: the Sobolev critical case,
J. Funct. Anal., 279 (2020) Paper No. 108610, 43 pp.

[49] P. Tolksdorf, Regularity for a more general class of quasilinear elliptic equations, J. Differential Equations, 51
(1984) 126-150.

[50] J. C. Wei and Y. Z. Wu, Normalized solutions for Schrödinger equations with critical sobolev exponent and mixed
nonlinearities, J. Funct. Anal., 283 (2022) Paper No. 109574, 46 pp.

[51] M. I. Weinstein, Nonlinear Schrödinger equations and sharp interpolation estimates, Comm. Math. Phys., 87
(1983) 567-576.

[52] M. Willem, Minimax theorems, Progress in Nonlinear Differential Equations and their Applications, vol. 24,
Birkhäuser Boston, Boston, (1996).

[53] M. B. Yang and Y. H. Ding, Existence of semiclassical states for a quasilinear Schrödinger equation with critical
exponent in RN , Ann. Mat. Pura Appl., 192 (2013) 783-804.

[54] H. Y. Ye and Y. Y. Yu, The existence of normalized solutions for L2 -critical quasilinear Schrödinger equations, J.
Math. Anal. Appl., 497 (2021) Paper No. 124839, 15 pp.

[55] L. Zhang, J. Q. Chen and Z.-Q. Wang, Ground states for a quasilinear Schrödinger equation: Mass critical and
supercritical cases, Appl. Math. Lett., 145 (2023) Paper No. 108763, 7 pp.

[56] L. Zhang, Y. Q. Li and Z.-Q. Wang, Multiple normalized solutions for a quasilinear Schrödinger equation via dual
approach, Topol. Methods Nonlinear Anal., 61 (2023) 465-489.

57