---
title: Dancer–Yan Spikes in Critical Elliptic Problems
url: https://www.emergentmind.com/topics/dancer-yan-spikes
type: topic
---

# Dancer–Yan Spikes in Critical Elliptic Problems

Dancer–Yan spikes are spike solutions associated with elliptic problems in which a compact or bounded direction alters the concentration mechanism available in the fully Euclidean setting. In the energy-critical Lane–Emden equation on the waveguide manifold \(\mathbb R_x^d\times \mathbb T_y\), they are positive-frequency solutions that decay in the noncompact \(x\)-directions and are periodic in \(y\), arising from the bifurcation of Aubin–Talenti bubbles after one direction is compactified [2606.01692]. In a distinct two-dimensional singularly perturbed free-boundary problem from plasma physics, the same terminology designates a precise Type I blow-up profile characterized by a refined inner–outer scaling and an Emden–log limiting profile [2507.20725].

## 1. Waveguide Lane–Emden setting

Let \(D=d+1\ge 3\) and
\[
2^*=\frac{2D}{D-2}=2+\frac{4}{d-1}.
\]
On the waveguide
\[
\mathcal X=\mathbb R_x^d\times \mathbb T_y,\qquad \mathbb T=\mathbb R/(2\pi\mathbb Z),
\]
the relevant positive-frequency standing waves have the form \(e^{i\sqrt\beta\,t}u(x,y)\) and satisfy
\[
-\Delta_{x,y}u+\beta u=u^{2^*-1},\qquad u>0,\ \beta>0.
\]
The variational formulation fixes the mass
\[
M(u)=\int_{\mathcal X}|u|^2,\qquad M(u)=c>0,
\]
and uses the energy
\[
E(u)=\frac12\int_{\mathcal X}|\nabla_{x,y}u|^2-\frac{D-2}{2D}\int_{\mathcal X}|u|^{2^*}.
\]
To isolate functions that “just fail to scatter” in the Euclidean directions, Luo introduces the semivirial functional
\[
Q(u)=\int_{\mathcal X}|\nabla_xu|^2-\frac{d}{D}\int_{\mathcal X}|u|^{2^*},
\]
and studies the constrained minimization problem
\[
m_c=\inf\{\,E(u):u\in H^1(\mathcal X),\ M(u)=c,\ Q(u)=0\}.
\]
A Lagrange-multiplier argument then shows that any optimizer solves the elliptic equation for some \(\beta>0\) [2606.01692].

This formulation is specific to the partially periodic geometry. The constraint \(Q(u)=0\) is not a generic Pohozaev identity imposed after the fact; it is built into the minimization problem to encode the Euclidean-direction criticality while keeping the periodic direction explicit.

## 2. Aubin–Talenti bubbles and the Dancer bifurcation picture

In the fully Euclidean space \(\mathbb R^D\), the zero-frequency equation
\[
-\Delta U=U^{2^*-1},\qquad U>0,
\]
has the Aubin–Talenti family of optimizers. These satisfy
\[
\int |\nabla W|^2=\int W^{2^*}=\mathcal S^{D/2},
\]
where
\[
\mathcal S=\inf_{0\neq f\in \dot H^1(\mathbb R^D)}\frac{\|\nabla f\|_2^2}{\|f\|_{2^*}^2}
\]
is the sharp Sobolev constant. They are the unique positive finite-energy solutions of the zero-frequency problem on \(\mathbb R^D\), and on \(\mathbb R^d\times \mathbb T\) they extend trivially by being constant in \(y\) [2606.01692].

Dancer’s bifurcation perspective begins precisely at this rigid Euclidean family. Once one breaks full Euclidean invariance by compactifying one direction to a torus, the bubble manifold bifurcates at small positive frequency \(\beta>0\) into genuinely \((x,y)\)-dependent solutions. The schematic Lyapunov–Schmidt expansion is
\[
u(x,y)=W(x)+\beta\,\psi(x,y)+O(\beta^2),
\]
with decay in \(x\in\mathbb R^d\) and \(2\pi\)-periodicity in \(y\). In Luo’s terminology, these are the Dancer–Yan spikes [2606.01692].

The bifurcation viewpoint is significant because it identifies the objects as perturbations of the sharp Sobolev optimizers. A plausible implication is that the waveguide problem should be analyzed relative to the Euclidean bubble threshold rather than by perturbative compactness alone.

## 3. Semivirial-vanishing geometry

Luo’s approach replaces the bifurcation construction with an energy-based variational method built around semivirial-vanishing geometry. The auxiliary functional is
\[
I(u)=E(u)-\tfrac12 Q(u)
=\frac12\int |\partial_yu|^2+\frac1{2D}\int |u|^{2^*}.
\]
On the constraint set \(Q=0\), one has \(E=I\). The geometry is organized by the \(x\)-scaling
\[
u_t(x,y)=t^{d/2}u(tx,y),
\]
for which
\[
\frac{d}{dt}E(u_t)=t^{-1}Q(u_t).
\]

Two structural facts are central. First, if \(Q(u)<0\), there is a unique \(t\in(0,1)\) such that \(Q(u_t)=0\), and moreover \(I(u_t)<I(u)\). Second,
\[
m_c=\inf\{\,I(u):M(u)=c,\ Q(u)\le 0\}.
\]
These statements convert the problem into one where minimizing sequences can be projected onto the semivirial manifold without losing control of the energy [2606.01692].

The paper explicitly contrasts this mechanism with the usual Brezis–Nirenberg scenario. No lower-order focusing perturbation is available in the model. Instead, the energy drop is produced by the bounded periodic direction, while the focusing takes place through scaling in the Euclidean directions alone. This suggests that the compact geometry acts as an effective variational defect relative to the Euclidean critical problem.

## 4. Strict sub-bubbling and normalized ground states

The Euclidean bubble threshold is
\[
m_{\rm bub}=\mathcal S^{D/2}/D.
\]
The main variational input is the strict sub-bubbling estimate
\[
\forall\,c>0:\qquad m_c<m_{\rm bub}.
\]
To prove it, Luo truncates a rescaled Aubin–Talenti bubble to a single period slab in the periodic direction. The key asymptotics are that the lost gradient tail satisfies
\[
a_R=\int_{|Y|>R}|\nabla_XW|^2\sim c_aR^{2-D},
\]
whereas the lost nonlinear tail is only \(O(R^{-D})\). Correspondingly,
\[
Q(U_R)=-a_R+O(R^{-D})<0
\]
for \(R\gg 1\), and
\[
E(U_R)=\frac{\mathcal S^{D/2}}{D}-\tfrac12 a_R+o(R^{2-D}).
\]
After projection back to \(\{Q=0\}\) by the unique scaling \(t_R<1\), the energy changes only by \(o(R^{2-D})\), which produces a test function strictly below the bubble level [2606.01692].

The leading-order decomposition makes the mechanism more explicit. Writing
\[
A_R=\int_{\Sigma_R}|\nabla_X(\eta_RW)|^2,\qquad
B_R=\int_{\Sigma_R}\eta_R^2|\partial_YW|^2,\qquad
C_R=\int_{\Sigma_R}\eta_R^{2^*}W^{2^*},
\]
one has
\[
A_R=\tfrac dD\mathcal S^{D/2}-a_R+o(R^{2-D}),\qquad
B_R=\tfrac1D\mathcal S^{D/2}-b_R+o(R^{2-D}),\qquad
C_R=\mathcal S^{D/2}-\gamma_R+o(R^{2-D}),
\]
with
\[
a_R\sim c_aR^{2-D},\qquad b_R\sim c_bR^{2-D},\qquad \gamma_R=O(R^{-D}).
\]
Thus the leading-order tail removal creates a negative contribution of order \(R^{2-D}\) in the gradient part, while the nonlinear loss is lower order. This geometric energy drop is what restores compactness below the sharp Sobolev threshold.

Once \(m_c<m_{\rm bub}\) is known, concentration of an Euclidean bubble can be excluded by the standard profile decomposition on \(\mathbb R^d\times \mathbb T\). Weak vanishing is also ruled out, since \(Q(u_n)=0\) forces a uniform lower bound on the critical \(L^{2^*}\)-norm. After compensating by at most an \(\mathbb R^d\)-translation, a minimizing sequence has a nonzero weak limit, and Brezis–Lieb splitting of mass, energy, and semivirial implies that the limit itself satisfies
\[
Q(u)=0,\qquad M(u)=c,\qquad E(u)=m_c.
\]
A secondary scaling argument yields \(\beta>0\), and positivity follows by the maximum principle. Therefore, for every \(c>0\), the infimum \(m_c\) is attained by a positive solution
\[
u_c>0\in H^1(\mathbb R^d\times\mathbb T),\qquad
-\Delta_{x,y}u_c+\beta_cu_c=u_c^{2^*-1},\qquad \beta_c>0.
\]
The same framework also records that there exists \(c_*>0\) such that if \(c<c_*\), any optimizer \(u_c\) has \(\partial_yu_c\not\equiv 0\) [2606.01692].

## 5. Two-dimensional free-boundary formulation and the classical spike ansatz

A second appearance of Dancer–Yan spikes occurs in the two-dimensional free-boundary problem studied by Bartolucci, Jevnikar, Wei, and Wu. Let \(\Omega\subset\mathbb R^2\) be a bounded \(C^{2,\beta}\)-domain, with \(p>1\) and \(I>0\). The plasma problem is
\[
-\Delta v=[v]_+^p\quad \text{in }\Omega,\qquad
v=\gamma\quad \text{on }\partial\Omega,\qquad
\int_\Omega [v]_+^p=I.
\]
The plasma region is \(\Omega_+=\{x\in\Omega:v(x)>0\}\), and \(\partial\Omega_+\subset \Omega\) is the free boundary. After introducing \(I^{1/q}=\lambda\), with \(q=p/(p-1)\), the paper rewrites the problem in the equivalent form
\[
-\Delta \psi=[\alpha+\lambda\psi]_+^p\quad \text{in }\Omega,\qquad
\psi=0\quad \text{on }\partial\Omega,\qquad
\int_\Omega [\alpha+\lambda\psi]_+^p=1.
\]
In the singular limit \(\alpha\to -\infty\), equivalently \(\lambda|\alpha|^{p-1}\to\infty\), one recovers
\[
-\varepsilon^2\Delta u=[u-1]_+^p,\qquad
u>0\ \text{in }\Omega,\qquad
u=0\ \text{on }\partial\Omega,\qquad
\int_\Omega [u-1]_+^p\to 0,\qquad \varepsilon\to 0,
\]
with
\[
\varepsilon^2=\frac{1}{|\alpha|^{p-1}\lambda}\to 0.
\]
Unlike the higher-dimensional case, the usual rescalings \(u(\varepsilon y)\) fail in dimension two because the planar limit
\[
-\Delta W=[W-1]_+^p,\qquad W\to 1 \text{ at infinity},
\]
is impossible [2507.20725].

Dancer and Yan’s remedy is a two-scale construction. Let \(\phi\) solve
\[
\begin{cases}
-\Delta \phi=\phi^p,& |y|<1,\\
\phi=0,& |y|=1,
\end{cases}
\]
radial and decreasing, and define
\[
I_{p-1}=\int_{B_1}\phi^{p-1},\qquad
I_p=\int_{B_1}\phi^p=-2\pi \phi'(1)>0.
\]
With \(R_2=1/\sqrt\pi\) so that \(|B_{R_2}|=1\), and with \(a-b>0\), \(s_\varepsilon>0\) is chosen by
\[
\bigl(\tfrac{\varepsilon}{s_\varepsilon}\bigr)^{2/(p-1)}\phi'(1)
=\frac{a-b}{\ln(\sqrt\pi\,s_\varepsilon)}.
\]
The Dancer–Yan trial function in \(B_{R_2}\) is
\[
U_{\varepsilon,a,b}(x)=
\begin{cases}
a+(\varepsilon/s_\varepsilon)^{2/(p-1)}\phi(x/s_\varepsilon),& |x|\le s_\varepsilon,\\[1ex]
a+(a-b)\dfrac{\ln(|x|/s_\varepsilon)}{\ln(\sqrt\pi\,s_\varepsilon)},& s_\varepsilon\le |x|\le R_2.
\end{cases}
\]
It satisfies
\[
-\Delta U_{\varepsilon,a,b}=\tfrac1{\varepsilon^2}[U_{\varepsilon,a,b}-a]_+^p
\quad \text{in } B_{R_2},\qquad
U_{\varepsilon,a,b}=b\quad \text{on }\partial B_{R_2}.
\]
In particular, when \(a=1\) and \(b=0\), it is an exact solution of
\[
-\varepsilon^2\Delta v=[v-1]_+^p,\qquad v|_{\partial B_{R_2}}=0.
\]

The asymptotic features of this spike are also explicit. One has \(s_\varepsilon\to 0\) and
\[
\varepsilon/s_\varepsilon\approx (\ln(1/s_\varepsilon))^{(p-1)/2}.
\]
The inner core width is \(s_\varepsilon\), the free-boundary layer is logarithmic in \(s_\varepsilon\), and the peak height obeys
\[
U_{\varepsilon,1,0}(0)=1+(\varepsilon/s_\varepsilon)^{2/(p-1)}\phi(0),
\]
so
\[
v(0)-1\sim (\varepsilon/s_\varepsilon)^{2/(p-1)}\phi(0)
\]
[2507.20725].

## 6. Singular-limit classification, vanishing level, and multi-spike structure

The 2025 classification theorem concerns nonnegative solutions \(v_n\) of
\[
-\varepsilon_n^2\Delta v_n=[v_n-1]_+^p\quad \text{in }\Omega,\qquad \varepsilon_n\to 0,
\]
under the uniform bounds
\[
\int_\Omega \frac{[v_n-1]_+^{p-1}}{\varepsilon_n^2}\le H_{p-1}<\infty,
\qquad
\int_\Omega \frac{\ln(1/s_n)\,[v_n-1]_+^{p}}{\varepsilon_n^2}\le H_p<\infty,
\]
where \(s_n\to 0\) is defined by
\[
(\varepsilon_n/s_n)^{2/(p-1)}\phi'(1)\ln(\sqrt\pi\,s_n)=1,
\]
and \(\theta_n\equiv \phi'(1)\ln(\sqrt\pi s_n)\to +\infty\). If \(x_n\in\Omega\) is a local maximizer with \(x_n\to x_*\in\Omega\), then exactly one of four behaviors occurs [2507.20725].

| Case | Limiting description | Mass behavior |
|---|---|---|
| (A) Vanishing | \(v_n\le 1\) on compact subsets eventually | no local spike |
| (B–i) Type I | \(u_n\to w^*\) on expanding balls | both \((p-1)\)- and \((p)\)-mass quantize |
| (B–ii) Type II | \(t_n\to\infty\), \(s_nt_n\to 0\) | \((p-1)\)-mass quantizes, \((p)\)-mass vanishes in the spike |
| (B–iii) Fading spike | \(s_nt_n\) bounded away from \(0\) | both masses decay too fast for quantization |

The refined scaling is
\[
\tilde v_n(y)=\theta_n\bigl[v_n(x_n+s_ny)-1\bigr],
\]
followed by
\[
t_n=\Bigl(\frac{\phi(0)}{\theta_n\,[v_n(x_n)-1]}\Bigr)^{(p-1)/2},
\qquad
u_n(z)=t_n^{2/(p-1)}\tilde v_n(t_nz).
\]
The classical Dancer–Yan spike is exactly case (B–i): \(t_n\to t_\infty\in (T_0,\infty)\), and on each ball \(\{|z|<R_n\}\) with \(R_n\to\infty\),
\[
u_n(z)\to w^*(z),
\]
where \(w^*\) equals the Emden solution \(\phi\) for \(|z|\le 1\) and \(\phi'(1)\ln |z|\) outside. In this regime,
\[
\varepsilon_n^{-2}\int_{|x-x_n|<R_ns_nt_n}[v_n-1]_+^{p-1}\to I_{p-1},
\]
and
\[
\theta_n\varepsilon_n^{-2}\int_{|x-x_n|<R_ns_nt_n}[v_n-1]_+^{p}\to I_p\,t_\infty^{-2/(p-1)}.
\]

The “vanishing level” is simply the threshold \(v=1\), and the geometry near a Type I or II spike is encoded by
\[
r_n\approx s_nt_n,
\qquad
v_n(x_n)-1=\frac{\phi(0)}{\theta_n\,t_n^{2/(p-1)}}.
\]
Hence the free boundary lies at distance \(O(s_nt_n)\) from \(x_n\), while the amplitude above \(1\) is
\[
O\bigl(1/(\theta_n t_n^{2/(p-1)})\bigr).
\]
Equivalently,
\[
\theta_n\bigl(v_n(x_n)-1\bigr)t_n^{2/(p-1)}=\phi(0).
\]

The global theory with Dirichlet boundary data \(v_n|_{\partial\Omega}=0\) and non-vanishing total \((p)\)-mass shows that boundary regions contain no spikes, that only finitely many spike sequences occur, and that these are of Type I or II. The centers are separated by distances much larger than the local core scales, and away from the union of \(B_{O(s_nt_n)}(x_{n,j})\) one has \(v_n\le 1\) by Green’s representation. Matching inner expansions to outer Green expansions quantizes the \((p-1)\)-mass to \(mI_{p-1}\), while a Pohozaev identity around each spike yields drift equations forcing the limiting centers to be critical points of the Kirchhoff–Routh Hamiltonian
\[
\mathcal H(q_1,\dots,q_m)
=\sum_{i=1}^m M_i^2\,H(q_i,q_i)+\sum_{i\ne j}M_iM_j\,G(q_i,q_j),
\]
with
\[
M_i=\sum_{j:x_{\infty,j}=q_i} t_{\infty,j}^{-2/(p-1)}.
\]

A central clarification provided by the classification is that, in dimension \(d=2\), it is not true that any solution in the singular limit is a Dancer–Yan spike. The spiking structure is more rich. By contrast, for \(d\ge 3\), simple \(\varepsilon_n\)-rescaling around a nondegenerate maximum leads to a unique radially decaying solution of
\[
-\Delta W=[W-1]_+^p\quad \text{on }\mathbb R^d,
\]
and that classification forces a single spike type [2507.20725].

Source: https://www.emergentmind.com/topics/dancer-yan-spikes