Geometric evolution equations with triple junctions in higher dimensions
Harald Garcke
University of Regensburg
joint work with Daniel Depner (University of Regensburg) and Yoshihito Kohsaka (Muroran IT)
February 2014
1 Surface cluster
2 First variation
3 Mean curvature flow with junctions
4 Known results (curves and graphs)
5 An existence result
6 Numerical approximation of flows with triple junctions by a method introduced by Barrett, G., N¨urnberg
Harald Garcke Geometric evolution equations with triple junctions
Cluster of hypersurfaces
Given smooth hypersurfaces
Γ1, . . . , ΓN in Rd +1 with (piecewise) smooth boundaries
Boundaries ∂Γi at piecewise smooth junctions
Energy for such structures Γ = (Γ1, . . . , ΓN)
F (Γ) =
N
X
i =1
Z
Γi
γid Hd =
N
X
i =1
γiHd(Γi) Consider gradient flow of F
First variation of F
Given vector field ζ : Rd +1 → Rd +1 we set
Γi(ε) := {x + εζ(x ) | x ∈ Γi}
Remark: Triple junctions stay triple junctions under the variation
Harald Garcke Geometric evolution equations with triple junctions
First variation of F
Transport theorem gives d
d ε Z
Γi(ε)
1d Hd = − Z
Γi(ε)
ViHid Hd + Z
∂Γi(ε)
v∂Γ,id Hd −1
Quantities:
Vi : normal velocity of Γi(ε) Hi : mean curvature
v∂Γ,i : outer conormal velocity of boundary
First variation of F
d
d εF (Γ(ε)) =
N
X
i =1
d d ε
Z
Γi(ε)
γid Hd = X
i
Z
Γi(ε)
−γiViHid Hd +X
i
Z
∂Γi(ε)
γiv∂Γ,id Hd −1
Steepest descent w.r.t. L2-metric
Vi = γiHi on Γi mean curvature flow in bulk.
Harald Garcke Geometric evolution equations with triple junctions
Boundary conditions on triple junctions Boundary term on junction
v∂Γi = ζ(x ) · τi τi outer unit conormal on ∂Γi On triple junction with i = 1, 2, 3
3
X
i =1
γiτi
!
· ζ(x) = 0
ζ arbitrary ⇒
3
P
i =1
γiτi = 0
symmetric case (γ1 = γ2 = γ3) ⇒ 120◦ angle condition
Gradient flow (symmetric case)
Vi = Hi on Γi 120◦ angle condition on triple junction
Typical solution (Numerics by Barrett, G., N¨urnberg)
equal area non-equal area
Harald Garcke Geometric evolution equations with triple junctions
Lower dimensional singularities
E.g. Four triple junctions meet at quadruple point Equal surface energies:
cos θ = −13 (tetrahedron angles) θ : angle between junctions
Double bubble conjecture
Hutchings, Morgan, Ritor´e, Ros (2000) proved:
The standard double bubble provides the least-perimeter way to enclose two prescribed volumes in R3
standard double bubble use an instability argument involving
the second variation of area to rule this out
Harald Garcke Geometric evolution equations with triple junctions
Allen-Cahn systems
Mean curvature flow with triple junctions arises as singular limit of Allen-Cahn systems:
Search u : Ω → RN such that
ε∂tu = ε∆u − 1εDψ(u) , ε > 0
ψ : RN → R potential with at least three global minima (multi-well potential)
Well-posedness
Existence of varifold solutions : Brakke, 1978
In this setting no uniqueness, no satisfactory regularity results
Question:
Do classical solutions exist ? What is the best formulation ?
Harald Garcke Geometric evolution equations with triple junctions
Local well-posedness for curves
Bronsard Reitich, 1993 Idea:
Parametrize curves over fixed intervals
Xi : [0, 1] × [0, T ] → R2 , i = 1, 2, 3 (p, t) 7→ Xi(p, t)
Solve
Xti = Xppi
|Xp|2 de Turck like trick (instead of Xti = Xssi , Xs arc length derivative)
How to handle triple junctions Require
I
X1(p, t) = X2(p, t) = X3(p, t) at boundary points p = 0, 1 II
P3 i =1
Xpi
|Xpi| = 0 (⇔ angle condition)
Harald Garcke Geometric evolution equations with triple junctions
How to prove local existence Linearize
Check Lopatinski-Shapiro conditions Use fixed point argument
Fourth order curve flow with junctions V = −∆sH in the plane + junctions
Local existence: Garcke and Novick-Cohen
Long time behaviour
Mantegazzu, Novaga, Tortorelli
Situation with Dirichlet boundary conditions
Solution exists as long as lengths Li, i = 1, . . . , N are bounded from below.
Schn¨urer etal / Bellettini Novaga
Asymptotic behaviour of lens-shape geometries
Harald Garcke Geometric evolution equations with triple junctions
Higher dimensional case Additional difficulty:
junctions have dimension d − 1 > 0 tangential degree of freedom
No general existence result for smooth solutions known before Special case of graphs: local existence of smooth solutions (Freire, 2010)
Graph situation
MCF with triple junction is a free boundary problem
Harald Garcke Geometric evolution equations with triple junctions
Another graph situation
The projection of γ(t) on the plane is the free boundary
Another graph situation
The resulting PDEs and the boundary conditions are highly nonlinear
Harald Garcke Geometric evolution equations with triple junctions
Triple junction flow of graphs
Complete set of equations
∂tgi
p1 + |∇gi|2 = div ∇gi
p1 + |∇gi|2
!
in Di(t)
2
X
i =1
1
p|∇gi|2 + 1(−∇gi ,1) − 1
p|∇g3|2 + 1(−∇g3,1) = 0 on γ(t) g1 = g2 = g3 on γ(t)
Three second order PDEs Four boundary conditions
Local existence for graphs Theorem (Freire)
Given C3+α initial surface with triple junctions which initially fulfill - angle condition
- H1 + H2 + H3 = 0
Then there exists a unique local solution in C2+α,1+α2
Harald Garcke Geometric evolution equations with triple junctions
Generic initial data are not graphs
MCF with boundary conditions Related simpler problem:
Consider mean curvature flow with 90◦ boundary condition
V = H on Γ
^(Γ, ∂Ω) = 90◦ at outer boundary ∂Ω
Harald Garcke Geometric evolution equations with triple junctions
MCF with boundary conditions
Use curvilinear coordinate system Search for w (σ, t).
w solves second order para- bolic PDE + Neumann type boundary condition, see
Stahl 1996 Depner 2010
Evolving triple junctions are more difficult to describe in higher dimensions
E.g. tangential motion in general necessary
normal velocitites V1, V2 given
V3 fixed through V1 + V2 + V3 = 0 triple junction has to move tangentially
Harald Garcke Geometric evolution equations with triple junctions
Parametrizing evolving suface clusters
Γ1∗, Γ2∗, Γ3∗ initial hyper-surfaces such that γ := ∂Γ1∗ = ∂Γ2∗ = ∂Γ3∗
ν∗1, ν∗2, ν∗3 triple junction outer conormals on γ N∗1, N∗2, N∗3 normals to Γ1∗, Γ2∗, Γ3∗
Parametrize new surface over Γ∗ = (Γ1∗, Γ2∗, Γ3∗):
Ψi(σ, w , r ) := σ + wN∗i (σ) + r τ∗i(σ)
(σ, t) 7→ Ψi(σ, ρi(σ, t), µi(pri(σ), t)) =: Φi(σ, t) τ∗i : extension of conormal with support local to γ
pri : Γi∗ → ∂Γi∗ pr|∂Γi i
∗ = id
PDE formulation Degrees of freedom
(ρ1, ρ2, ρ3) change in normal direction on (Γ1∗, Γ2∗, Γ3∗)
(µ1, µ2, µ3) change in tangential direction on (∂Γ1∗, ∂Γ2∗, ∂Γ3∗) Equations:
MCF: V i = Hi on Γi∗
Persistence of triple junction:
σ + ρ1N∗1 + µ1ν∗1 = σ + ρ2N∗2 + µ2ν∗2 = σ + ρ3N∗3 + µ3ν∗3 on γ 4 conditions angle conditions on γ 2 conditions
Harald Garcke Geometric evolution equations with triple junctions
Persistence of triple junction gives:
ρ1 + ρ2 + ρ3 = 0 on γ µi = √1
3(ρk − ρj) on γ This implies
(µ1, µ2, µ3) on γ are given by (ρ1, ρ2, ρ3) on γ
Parametrization σ + ρi(σ, t)N∗i + µi(pri(σ), t)τ∗i implies non local dependence on boundary values
PDE formulation (NPDE)
ρit = Fi(ρi, ρ ◦ pr) + bi(ρi, ρ ◦ pr)n 3 P
j=1
Pij(ρ, ρ ◦ pr)Fj(ρj, ρ ◦ pr)o Fi second order non-local differential operator
(Pij)ij matrix leading to a coupling of the three equations via non-local boundary terms
(BC1) ρ1 + ρ2 + ρ3 = 0 on γ (BC2) angle conditions on γ
nonlinear conditions involving first derivatives of ρ and µ
Harald Garcke Geometric evolution equations with triple junctions
Strategy for local existence result 1. Linearize
2. Show existence/uniqueness of linear equation in energy spaces 3. Show C2+α,1+α2 regularity of solutions (Solonnikov theory) 4. Use a contraction argument (Baconneau, Lunardi)
Special feature: Non-local term includes derivatives of highest order
1. Linear problem
The linear problem (in angle condition I only show principal part)
∂tui = ∆Γi
∗ui + |Πi∗|2ui on Γi∗
∂ν1
∗u1 = ∂ν2
∗u2 = ∂ν3
∗u3 on γ
u1 + u2 + u3 = 0 on γ
Equations are formulated on each surface coupled through boundary conditions on triple junction
Harald Garcke Geometric evolution equations with triple junctions
Existence of weak solution (linearized problem) Testing procedure gives
d dt
P
i
R
Γi∗ 1
2(ui)2 + P
i
R
Γi∗
|∇Γ∗
i ui|2 = R
Γ∗i
P
i
(∂νi
∗ui)ui + l.o.t.
= R
Γ∗i
∂ν1
∗u1(u1 + u2 + u3) + l.o.t.
= 0 + l.o.t.
Solution in energy space exists
Regularity of solutions
Use results by Solonnikov (locally)
Situation in a neighborhood U of p ∈ γ
Write equation locally with the same parameter domain D = B1(0) ∩ Rd+
system on same domain of definition
Harald Garcke Geometric evolution equations with triple junctions
Are the complementarity (Lopatinski-Shapiro) conditions fulfilled ? Use suitable coordinates !
In a fixed point we get:
∂tui − ∆ui = 0 + b.c. u1 + u2 + u3 = 0, ∂du1 = ∂du2 = ∂du3 Lopatinski-Shapiro condition:
(ϕ1, ϕ2, ϕ3) = 0 is the only bounded solution of the ODE system λϕj − ϕ00j + |ξ|2ϕj = 0 for z ∈ R+ ,
ϕ1 + ϕ2 + ϕ3 = 0 for z = 0 , ϕ01 = ϕ02 = ϕ03 for z = 0
Lopatinski-Shapiro condition
λϕj − ϕ00j + |ξ|2ϕj = 0 for z ∈ R+ , ϕ1 + ϕ2 + ϕ3 = 0 for z = 0 ,
ϕ01 = ϕ02 = ϕ03 for z = 0 . Multiply by ϕj and integrate
λ
∞
R
0
|ϕj|2 + P
j
(−
∞
R
0
ϕ00j ϕj + |ξ|2|ϕj|2) = 0 We compute
−
∞
R
0
ϕ00j ϕj =
∞
R
0
|ϕ0j|2 − [ϕ0jϕj]∞0 P
j
(ϕ0jϕj)(0) = ϕ01(0)P
j
ϕj(0) = 0 Since Re λ > 0 we obtain (ϕ1, ϕ2, ϕ3) ≡ 0
Harald Garcke Geometric evolution equations with triple junctions
Solonnikov theory gives C2+α,1+α2 -solvability of linear problem Use contraction argument for the fully nonlinear, non-local
PDE on the surface cluster (similar as Baconneau, Lunardi) Remark: Here it is important that the non-local term vanishes in the linearization
Theorem (Depner, G., Kohsaka, 2014, ARMA)
Let (Γ10, Γ20, Γ03) be a C2+α surface cluster with a C2+α triple junction curve γ
We assume the compatibility conditions - (Γ10, Γ20, Γ30) fufill the angle conditions - H01 + H02 + H03 = 0
Then there exists a local C2+α,1+α2 solution of Vi = Hi
+angle conditions with initial data (Γ10, Γ20, Γ30)
Harald Garcke Geometric evolution equations with triple junctions
Remarks
1. Baconneau, Lunardi and Freire have a loss of regularity.
How to avoid this ?
Parametrize over a smooth cluster close by.
2. Generalizations to more than three surfaces are no problem
Remarks
3. Completely missing is (Huisken for MCF)
continuation criterium
(maybe possible in terms of curvature and surface area of cluster patches)
blow-up analysis at first singular time
Harald Garcke Geometric evolution equations with triple junctions
Quadruple junctions are more difficult
A triple bubble with a quadruple point
∂Γ is only piecewise smooth !
Higher order flows (surface diffusion) G.+Novick-Cohen
Vi = −∆ΓiHi on Γi angle condition
H1 + H2 + H3 = 0 continuity of chem.pot.
∇Γ1H1 · ν∂Γ1 = ∇Γ2H2 · ν∂Γ2 = ∇Γ3H3 · ν∂Γ3 = 0 flux balance Properties:
d dt
X
i
Z
Γi
1d Hn ≤ 0 surface area decreasing volume preserving
Harald Garcke Geometric evolution equations with triple junctions
Stationary solutions
0 = d dt
X
i
Z
Γi
1d Hn = −X
i
Z
Γi
|∇Hi|2
(Γ1, Γ2, Γ3) stationary ⇒ Hi constant
b.c. ⇒ H1 + H2 + H3 = 0 Stationary solutions: Double bubbles, triple bubles,...
Surface diffusion: no junctions
Escher Simonett showed for V = −∆sH
Spheres are (up to rescaling and translation) asymptotically stable under surface diffusion
Method: Center manifold theory
Needed since equilibria are not discrete
(indeed they build a (d + 2)-dimensional set translation + rescaling possible)
Harald Garcke Geometric evolution equations with triple junctions
Stability for surface diffusion with junctions
Problem: Fully nonlinear boundary conditions lead to nonlinear phase manifold
Usual stability theory in Banach-spaces not applicable
Try to use generalized principle of linearized stability for PDEs with nonlinear BC’s (Pr¨uss, Simonett, Zacher)
Some ingredients: ∂tu + A(u(t))u(t) = F (u(t)) Equilibria form a C1-manifold E
tangent space for E is given by N(A0) (A0 linearization) σ(A0) \ {0} ⊂ C+
0 is semi-simple eigenvalue
Work in progress (Arab, Abels, G.)
Mercedes star is stable under surface diffusion.
Double bubble is stable under surface diffusion
Harald Garcke Geometric evolution equations with triple junctions
Conclusions
Local well-posedness for mean curvature flow with triple junctions can be shown in higher dimensions also in the parametric case
Results for longer times are missing in higher dimensions (Continuation results, classification of singularities)
The situation with codimension 2 singularities is more difficult Semi-group theory for parabolic PDEs with fully nonlinear
BC’s can help to show stability results
The approach of Barrett, G., N¨urnberg can easily deal with junctions