betaFeedback

WorksMathematics

Rigidity and classification of complete translating solitons and self-expanders with constant curvature in R4\mathbb{R}^{4}

Abstract

For complete immersed surfaces in Euclidean four-space whose mean curvature vector has constant length, translators are affine planes parallel to their fixed nonzero velocity, and self-expanders are planes through their centre. This rigidity requires no Lagrangian assumption, properness, volume growth control, or bound on the second fundamental form. Complete Lagrangian translators and self-expanders in C2\mathbb{C}^{2} with constant Gaussian curvature are flat and, up to unitary equivalence, are properly embedded products of two complete planar curves. Translating factors are lines or grim reapers. Expanding factors are lines through the origin or nonstraight self-expanders. The proofs use moving frames, a transnormal function, a cubic tensor identity, and analytic continuation.

Introduction

Mean curvature flow evolves a family of immersed submanifolds xt=x(⋅,t):Mn→Rn+px_t=x(\cdot,t):M^n\to\mathbb{R}^{n+p} by

ddtx(p,t)=H⃗(p,t),x(p,0)=x(p),p∈Mn,\frac{d}{dt}x(p,t)=\vec{H}(p,t),\qquad x(p,0)=x(p),\qquad p\in M^n,

where H⃗(p,t)\vec{H}(p,t) is the mean curvature vector of MtM_t at p∈Mnp\in M^n. Its self-similar solutions play a central role both in the local analysis of singularities and in the long-time behaviour of the flow. Two basic examples are translating solitons (or, simply, translators), characterised by

H⃗=V⊥,0≠V∈Rn+p constant,(1)\vec{H}=V^\perp,\qquad0\ne V\in\mathbb{R}^{n+p}\ \text{constant}, \tag*{(1)}

and self-expanders, satisfying

H⃗=x⊥.(2)\vec{H}=x^\perp. \tag*{(2)}

Under our convention Δx=H⃗\Delta x=\vec{H}, equation (1.1) generates translation by tVtV, while (1.2) generates the homothetic flow xt=2txx_t=\sqrt{2t}x for t>0t>0, up to tangential reparametrisation. The normalisation in (1.2) is essential for the numerical coefficients below. Basic examples and the theory of translators and expanders may be found in [1, 2, 9, 10, 12, 13, 25].

The interaction between intrinsic and extrinsic geometry becomes particularly interesting for Lagrangian mean curvature flow in Cn\mathbb{C}^n. The Lagrangian condition is preserved along smooth mean curvature flow, and the parallel complex structure identifies the tangent and normal bundles. This produces additional symmetry in the second fundamental form, a feature absent for general submanifolds. Foundational constructions and rigidity results include [3, 4, 6, 8, 14, 15, 24].

From view points of submanifolds theory, it is very natural to study rigidity and classification theorems for the Lagrangian self-shrinkers, the Lagrangian translators and the Lagrangian self-expanders. Two curvature assumptions that often arise in classification problems must be distinguished. The extrinsic condition constant ∣H⃗∣2|\vec{H}|^2 fixes the mean curvature length. The intrinsic condition constant KK fixes the Gaussian curvature. For an immersed surface in R4\mathbb{R}^4, the Gauss equation gives 2K=∣H⃗∣2−∣B∣22K = |\vec{H}|^2 - |B|^2. Thus neither of these two constancy assumptions implies the other without further information. Recent results on Lagrangian translators and Lagrangian self-expanders under hypotheses involving ∣H⃗∣2|\vec{H}|^2 or ∣B∣2|B|^2 appear in [16, 19, 20]; complete Lagrangian self-shrinkers are investigated in [8, 18, 21]. In contrast, our second classification problem below concerns constant Gaussian curvature.

We first address the mean curvature length problem without imposing the Lagrangian condition.

Theorem 1.1. Let x:M2→R4x : M^2 \to\mathbb{R}^4 be a smooth, complete, connected translator without boundary. If ∣H⃗∣2|\vec{H}|^2 is constant, then H⃗=0\vec{H} = 0 and B=0B = 0. Consequently, xx is a global isometry onto an affine two-plane p0+Pp_0 + P with V∈PV \in P.

Theorem 1.2. Let x:M2→R4x : M^2 \to\mathbb{R}^4 be a smooth, complete, connected self-expander without boundary. If ∣H⃗∣2|\vec{H}|^2 is constant, then H⃗=B=0\vec{H} = B = 0, and xx is a global isometry onto a linear two-plane through the origin.

Remark 1.1. The above conclusion, without conditions the squared norm of the second fundamental form being bounded from above and Lagrangian condition, proves that the conclusion of [20] still holds true. The same transnormal function occurs for self-shrinkers, with a different Laplacian, in the preprint of Li, Wei and Xing [21].

Remark 1.2. The mechanisms are different. For translators, the nonzero mean curvature case is ruled out locally using a moving frame, the Gauss equation, and Codazzi compatibility. For expanders, completeness and a transnormal identity force the constant mean curvature to vanish. In particular, the translator’s vanishing conclusion does not itself require completeness; completeness is used only to identify the entire plane.

Our second condition prescribes the intrinsic Gaussian curvature KK. In particular, constant KK fixes the difference of two extrinsic curvature functions. The constant ∣B∣2|B|^2 problem for complete Lagrangian surfaces was treated by Li, Liu and Qiao for translators [16] and by Li and Wei for expanders [19]. Chen [7] classified general constant-curvature Lagrangian surfaces in C2\mathbb{C}^2 locally on an open dense set. For the intrinsic problem, let JJ be the standard complex structure on C2\mathbb{C}^2. An immersion is Lagrangian when J(TM)=NMJ(TM) = NM. The stronger cubic-tensor symmetry leads to global classifications.

Theorem 1.3. Let x:M2→C2x:M^2\to\mathbb{C}^2 be a smooth, complete, connected Lagrangian translator without boundary. If the Gaussian curvature is constant, then the surface is flat and isometric to R2\mathbb{R}^2, and there exist a unitary transformation UU and complete unit-speed plane curves γi:R→C\gamma_i:\mathbb{R}\to\mathbb{C} such that

x(s,t)=U(γ1(s),γ2(t)),γi′′=(Vi)⊥γi,(V1,V2)=U−1V.x(s,t)=U\bigl(\gamma_1(s),\gamma_2(t)\bigr),\qquad\gamma_i^{\prime\prime}=(V_i)^{\perp_{\gamma_i}},\qquad(V_1,V_2)=U^{-1}V.

Here ⊥γi\perp_{\gamma_i} denotes projection onto the normal line of the curve. Each factor is a straight line or a grim reaper curve; the product is a proper embedding. Conversely, every such product is a complete flat Lagrangian translator, provided the factor equations hold for the prescribed VV.

Theorem 1.4. Let x:M2→C2x:M^2\to\mathbb{C}^2 be a smooth, complete, connected Lagrangian self-expander without boundary. If its gaussian curvature KK is constant, then K=0K=0, and MM is isometric to R2\mathbb{R}^2, and

x(s,t)=U(γ1(s),γ2(t)),γi′′=(γi)⊥γi,x(s,t)=U\bigl(\gamma_1(s),\gamma_2(t)\bigr),\qquad\gamma_i^{\prime\prime}=(\gamma_i)^{\perp_{\gamma_i}},

for a unitary transformation UU and complete unit speed planar self-expanders γi\gamma_i. Each factor is a line through the origin or a nonstraight complete expanding curve. The product is properly embedded. Conversely, each such product is a complete flat Lagrangian self-expander.

Remark 1.3. In fact, the authors of the present paper [22] proved that the conclusion remains valid for self-shrinkers under the same hypotheses as those in Theorems 1.3 and 1.4.

Joyce, Lee and Tsui [14] constructed families of Lagrangian translators and self-expanders. The translating products of a line and a grim reaper, or of two grim reapers, already appear in Castro and Lerma [5]. For self-shrinkers, Li and Wang [17] obtained a product decomposition for complete flat Lagrangian surfaces. Here the exclusion of negative constant KK uses an identity for the covariant derivative of the Lagrangian cubic tensor. After proving flatness, we extend the local product decomposition to the universal cover. The known classification of planar soliton curves [12] then gives injectivity and properness on the original surface.

The exclusion of negative constant Gaussian curvature rests on an explicit identity for the covariant derivative of the Lagrangian cubic tensor, combined with an approximate extremal sequence. Flatness is then promoted to a global Cartesian splitting by commutation of shape operators and real-analytic elliptic regularity. Finally, planar curve classification gives embeddedness and eliminates nontrivial deck transformations. We include the calculations and the completeness arguments to make precise where each hypothesis enters.

The paper is organised as follows. § 2 fixes conventions and gives auxiliary lemmas. § 3-§ 4 prove the general-surface rigidity statements. § 5-§ 7 establish the curvature identities and exclude nonzero constant KK. § 8-§ 9 prove the global product classification.

Preliminaries

Let x:M2→R4x:M^2\to\mathbb{R}^4 be a smooth immersion. Throughout, MM is connected and without boundary, and completeness refers to the induced metric. For an orthonormal tangent frame {e1,e2}\{e_1,e_2\}, write Bij=B(ei,ej)B_{ij}=B(e_i,e_j). Thus

H⃗=tr⁡gB=∑i=12Bii,Δx=H⃗,(3)\vec{H}=\operatorname{tr}_g B=\sum_{i=1}^{2}B_{ii},\qquad\Delta x=\vec{H}, \tag*{(3)}

where BB is the second fundamental form. Denote by DD, ∇\nabla and ∇⊥\nabla^\perp the Euclidean, Levi-Civita, and normal connections, respectively. For tangent vector fields X,YX,Y and a normal field η\eta, the Gauss and Weingarten formulas are

DXY=∇XY+B(X,Y),DXη=−AηX+∇X⊥η,(4)D_XY=\nabla_XY+B(X,Y),\qquad D_X\eta=-A_\eta X+\nabla^\perp_X\eta, \tag*{(4)}

where ⟨AηX,Y⟩=⟨B(X,Y),η⟩\langle A_\eta X,Y\rangle=\langle B(X,Y),\eta\rangle. Our curvature convention is

R(X,Y)=∇X∇Y−∇Y∇X−∇[X,Y].R(X,Y)=\nabla_X\nabla_Y-\nabla_Y\nabla_X-\nabla_{[X,Y]}.

With K=⟨R(e1,e2)e2,e1⟩K=\langle R(e_1,e_2)e_2,e_1\rangle, the Gauss equation gives

K=⟨B11,B22⟩−∣B12∣2,2K=∣H⃗∣2−∣B∣2(5)K=\langle B_{11},B_{22}\rangle-|B_{12}|^2,\qquad2K=|\vec{H}|^2-|B|^2 \tag*{(5)}

The covariant derivative of BB is

(∇XB)(Y,Z)=∇X⊥(B(Y,Z))−B(∇XY,Z)−B(Y,∇XZ),(\nabla_XB)(Y,Z)=\nabla^\perp_X(B(Y,Z))-B(\nabla_XY,Z)-B(Y,\nabla_XZ),

and the Euclidean Codazzi equation is

(∇XB)(Y,Z)=(∇YB)(X,Z).(6)(\nabla_XB)(Y,Z)=(\nabla_YB)(X,Z). \tag*{(6)}

Lemma 2.1. ([21]) Let (Nm,g)(N^m,g) be a nonempty complete Riemannian manifold. If uu is smooth and nonnegative and satisfies ∣∇u∣2=4u|\nabla u|^2=4u. Then uu vanishes somewhere.

Proof. If u≡0u\equiv0, the conclusion is immediate. On Ω={u>0}\Omega=\{u>0\} define r=ur=\sqrt{u}, then ∣∇r∣=1|\nabla r|=1. Fix p∈Ωp\in\Omega and take the maximal integral curve γ:[0,τ)→Ω\gamma:[0,\tau)\to\Omega of −∇r-\nabla r with γ(0)=p\gamma(0)=p. Its speed equals one and

ddtr(γ(t))=−1,r(γ(t))=r(p)−t,\frac{d}{dt}r(\gamma(t))=-1,\qquad r(\gamma(t))=r(p)-t,

so τ≤r(p)\tau\leq r(p). The curve is Cauchy as t↑τt\uparrow\tau because d(γ(t),γ(s))≤∣t−s∣d(\gamma(t),\gamma(s))\leq|t-s|. Completeness supplies q=lim⁡t↑τγ(t)q=\lim_{t\uparrow\tau}\gamma(t). If τ<r(p)\tau<r(p), then u(q)>0u(q)>0 and the smooth vector field −∇r-\nabla r extends the integral curve, contradicting maximality. Hence τ=r(p)\tau=r(p) and u(q)=0u(q)=0. □

Lemma 2.2. Under the hypotheses of Lemma 2.1, if u(q)=0u(q)=0, the self-adjoint Hessian endomorphism AA defined by ⟨AX,Y⟩=∇2u(X,Y)∣q\langle AX,Y\rangle=\nabla^2u(X,Y)|_q satisfies A2=2AA^2=2A. Consequently, its eigenvalues lie in {0,2}\{0,2\} and Δu(q)≤2m\Delta u(q)\leq2m.

Proof. At a zero of the nonnegative smooth function uu, ∇u(q)=0\nabla u(q)=0. Twice differentiating ∑k(uk)2=4u\sum_k(u_k)^2=4u in normal coordinates gives

2∑kuikujk+2∑kukukij=4uij.2\sum_k u_{ik}u_{jk}+2\sum_k u_k u_{kij}=4u_{ij}.

At qq the second term vanishes, so A2=2A\mathcal{A}^{2}=2\mathcal{A}, and the eigenvalue equation μ2=2μ\mu^{2}=2\mu proves the assertion. □

Lemma 2.3. Every complete connected totally geodesic isometric immersion x:Nm→RNx:N^{m}\to\mathbb{R}^{N} is a global isometry onto an affine mm-plane.

Proof. Fix p∈Np\in N and put P=dxp(TpN)P=dx_{p}(T_{p}N). For any v∈TpNv\in T_{p}N, completeness makes the geodesic γv(t)=exp⁡p(tv)\gamma_{v}(t)=\exp_{p}(tv) available for all tt. Since B=0B=0,

d2dt2x(γv(t))=B(γ˙v,γ˙v)=0,\frac{d^{2}}{dt^{2}}x(\gamma_{v}(t))=B(\dot{\gamma}_{v},\dot{\gamma}_{v})=0,

and hence

x(exp⁡pv)=x(p)+dxp(v).x(\exp_{p}v)=x(p)+dx_{p}(v).

By Hopf-Rinow, exp⁡p\exp_{p} is surjective. Therefore the image is precisely x(p)+Px(p)+P. If x(q1)=x(q2)x(q_{1})=x(q_{2}), choose viv_{i} with exp⁡pvi=qi\exp_{p}v_{i}=q_{i}; the displayed equation implies dxp(v1)=dxp(v2)dx_{p}(v_{1})=dx_{p}(v_{2}), so v1=v2v_{1}=v_{2} and q1=q2q_{1}=q_{2}. The bijective local isometry xx is a global isometry. □

Lemma 2.4. If (N,g)(N,g) is complete and f∈C1(N)f\in C^{1}(N) is bounded below, there are pj∈Np_{j}\in N such that

f(pj)→inf⁡Nf,∣∇f∣(pj)→0.f(p_{j})\to\inf_{N}f,\qquad|\nabla f|(p_{j})\to0.

For a function bounded above, the analogous statement holds at the supremum.

Proof. Choose εj↓0\varepsilon_{j}\downarrow0 and qjq_{j} with f(qj)≤inf⁡f+εj2f(q_{j})\leq\inf f+\varepsilon_{j}^{2}. Ekeland’s variational principle [11] on the complete metric space (N,d)(N,d) gives pjp_{j} satisfying f(pj)≤f(qj)f(p_{j})\leq f(q_{j}) and

f(q)≥f(pj)−εjd(pj,q)(∀q∈N).f(q)\geq f(p_{j})-\varepsilon_{j}d(p_{j},q)\qquad(\forall q\in N).

Apply this to q=exp⁡pj(tv)q=\exp_{p_{j}}(tv) for unit vv, first with t>0t>0 and then with t<0t<0, to find ∣dfpj(v)∣≤εj|df_{p_{j}}(v)|\leq\varepsilon_{j}. Taking the supremum over vv gives the required gradient estimate. Replace ff by −f-f for a maximum sequence. □

Rigidity of translators with constant ∣H⃗∣|\vec{H}|

Proposition 3.1. Let x:M2→R4x:M^{2}\to\mathbb{R}^{4} be a connected translator with nonzero fixed velocity VV. If ∣H⃗∣2|\vec{H}|^{2} is constant, then H⃗=B=0\vec{H}=B=0, even without completeness.

Proof. Set T=V⊤T=V^{\top}, f=⟨V,x⟩f=\langle V,x\rangle and c=∣H⃗∣2c=|\vec{H}|^{2}. Because V=T+H⃗V=T+\vec{H} is a fixed ambient vector, (4) yields, for all tangent XX,

∇XT=AH⃗X,∇X⊥H⃗=−B(X,T).(7)\nabla_{X}T=A_{\vec{H}}X,\qquad\nabla_{X}^{\perp}\vec{H}=-B(X,T). \tag*{(7)}

Taking the gradient and Laplacian of ff gives

∇f=T,∣T∣2=∣V∣2−c=a2,Δf=⟨V,H⃗⟩=c,(8)\nabla f=T,\qquad|T|^{2}=|V|^{2}-c=a^{2},\qquad\Delta f=\langle V,\vec{H}\rangle=c, \tag*{(8)}

where a≥0a\geq0. We will prove by contradiction that c=0c=0. Assume c>0c>0. If a=0a=0, then ff is constant, contrary to Δf=c>0\Delta f=c>0. Thus a>0a>0. The constant length of T=∇fT=\nabla f implies

0=12∇∣T∣2=∇2f(T,⋅)=⟨AH⃗T,⋅⟩,0=\frac{1}{2}\nabla|T|^{2}=\nabla^{2}f(T,\cdot)=\langle A_{\vec{H}}T,\cdot\rangle,

so AH⃗T=0A_{\vec H}T=0. On a local oriented chart let e1=Tae_1=\frac{T}{a} and choose e2e_2 perpendicular to it. Since tr⁡AH⃗=⟨H⃗,H⃗⟩=c\operatorname{tr} A_{\vec H}=\langle\vec H,\vec H\rangle=c, we obtain

AH⃗e1=0,AH⃗e2=ce2.(9)A_{\vec H}e_1=0,\qquad A_{\vec H}e_2=ce_2. \tag*{(9)}

Put k=ca>0k=\frac{c}{a}>0. The first identity of (7) and metric compatibility give

∇e1e1=∇e1e2=0,∇e2e1=ke2,∇e2e2=−ke1.(10)\nabla_{e_1}e_1=\nabla_{e_1}e_2=0,\qquad\nabla_{e_2}e_1=ke_2,\qquad\nabla_{e_2}e_2=-ke_1. \tag*{(10)}

Since kk is constant, these connection coefficients give

[e1,e2]=−ke2,R(e1,e2)e2=−k2e1,K=−k2.(11)[e_1,e_2]=-ke_2,\qquad R(e_1,e_2)e_2=-k^2e_1,\qquad K=-k^2. \tag*{(11)}

The negative value is not yet a contradiction; we must impose the normal Codazzi equations.

Let h=ch=\sqrt{c} and choose a local orthonormal normal frame ν1=H⃗h,ν2\nu_1=\frac{\vec H}{h},\nu_2. Equation (9) yields ⟨B11,ν1⟩=⟨B12,ν1⟩=0\langle B_{11},\nu_1\rangle=\langle B_{12},\nu_1\rangle=0 and ⟨B22,ν1⟩=h\langle B_{22},\nu_1\rangle=h. Since B11+B22=hν1B_{11}+B_{22}=h\nu_1, for smooth functions b,db,d we have

B11=bν2,B12=dν2,B22=hν1−bν2.(12)B_{11}=b\nu_2,\qquad B_{12}=d\nu_2,\qquad B_{22}=h\nu_1-b\nu_2. \tag*{(12)}

The Gauss equation gives

K=⟨bν2,hν1−bν2⟩−∣dν2∣2=−b2−d2.K=\langle b\nu_2,h\nu_1-b\nu_2\rangle-\lvert d\nu_2\rvert^2=-b^2-d^2.

Together with (11), this gives the fixed radius relation

b2+d2=k2.(13)b^2+d^2=k^2. \tag*{(13)}

Define normal connection functions by ∇ei⊥ν1=ωiν2\nabla^\perp_{e_i}\nu_1=\omega_i\nu_2 and ∇ei⊥ν2=−ωiν1\nabla^\perp_{e_i}\nu_2=-\omega_i\nu_1. The second identity of (7) with T=ae1T=ae_1 and (12) yield

hω1=−ab,hω2=−ad.(14)h\omega_1=-ab,\qquad h\omega_2=-ad. \tag*{(14)}

Step 3. Use (10) to expand the first Codazzi equation

(∇e1B)(e2,e2)=(∇e2B)(e1,e2).(\nabla_{e_1}B)(e_2,e_2)=(\nabla_{e_2}B)(e_1,e_2).

The ν2\nu_2 component of the left-hand side is hω1−e1(b)h\omega_1-e_1(b). The right-hand side has ν2\nu_2 component e2(d)+2kbe_2(d)+2kb: indeed,

(∇e2B)(e1,e2)=∇e2⊥B12−kB22+kB11.(\nabla_{e_2}B)(e_1,e_2)=\nabla^\perp_{e_2}B_{12}-kB_{22}+kB_{11}.

Thus

hω1−e1(b)=e2(d)+2kb.(15)h\omega_1-e_1(b)=e_2(d)+2kb. \tag*{(15)}

Similarly, the ν2\nu_2 component of (∇e1B)(e1,e2)=(∇e2B)(e1,e1)(\nabla_{e_1}B)(e_1,e_2)=(\nabla_{e_2}B)(e_1,e_1) gives

e1(d)=e2(b)−2kd.(16)e_1(d)=e_2(b)-2kd. \tag*{(16)}

Substituting (14) gives the closed system

e1(b)+e2(d)=−(a+2k)b,e1(d)−e2(b)=−2kd.(17)\begin{aligned} e_1(b)+e_2(d)&=-(a+2k)b,\\ e_1(d)-e_2(b)&=-2kd. \tag*{(17)} \end{aligned}

Since k>0k>0, equation (13) allows a smooth local angle θ\theta with b=kcos⁡θb=k\cos\theta, d=ksin⁡θd=k\sin\theta. Substituting into (17) and inverting the rotation matrix gives

e1(θ)=asin⁡θcos⁡θ,e2(θ)=−2k−acos⁡2θ.(18)e_1(\theta)=a\sin\theta\cos\theta,\qquad e_2(\theta)=-2k-a\cos^2\theta. \tag*{(18)}

From (3.5), [e1,e2](θ)=−ke2(θ)[e_1,e_2](\theta)=-ke_2(\theta). Differentiating (3.12) gives

e1(e2(θ))=2a2sin⁡2θcos⁡2θ,e_1(e_2(\theta))=2a^2\sin^2\theta\cos^2\theta,
e2(e1(θ))=a(cos⁡2θ−sin⁡2θ)(−2k−acos⁡2θ).e_2(e_1(\theta))=a(\cos^2\theta-\sin^2\theta)(-2k-a\cos^2\theta).

The commutator equality therefore reduces, after collecting terms, to

a(a+3k)cos⁡2θ=2k(a+k).(19)a(a+3k)\cos^2\theta=2k(a+k). \tag*{(19)}

All coefficients are positive constants, so cos⁡2θ\cos^2\theta must be constant. Locally, smoothness then forces θ\theta to be constant, because each level set of cos⁡2\cos^2 consists of isolated points. But (3.12) gives e2(θ)=−2k−acos⁡2θ<0e_2(\theta)=-2k-a\cos^2\theta<0. This is a contradiction. Hence c>0c>0 is impossible.

Now H⃗=0\vec{H}=0 and V⊥=0V^\perp=0, so V=TV=T is a nowhere-zero tangent vector, and (3.1) gives B(X,V)=0B(X,V)=0 for every XX. Choosing e1=V/∣V∣e_1=V/\lvert V\rvert yields B11=B12=0B_{11}=B_{12}=0. Since H⃗=B11+B22=0\vec{H}=B_{11}+B_{22}=0, we also have B22=0B_{22}=0. Thus B≡0B\equiv0. □

Proof of Theorem 1.1. Proposition 3.1 gives B=H⃗=0B=\vec{H}=0, and Lemma 2.3 identifies the complete immersion with an entire affine plane p0+Pp_0+P. Since VV is tangent to the plane at every point, it lies in PP. □

Rigidity of self-expanders with constant ∣H⃗∣\lvert\vec{H}\rvert

Proof of Theorem 1.2. Write c=∣H⃗∣2≥0c=\lvert\vec{H}\rvert^2\ge0 and T=x⊤T=x^\top. For the constant cc, the self-expander equation gives

u=∣x∣2−c=∣T∣2≥0.(20)u=\lvert x\rvert^2-c=\lvert T\rvert^2\ge0. \tag*{(20)}

Directly from (2.1),

∇u=2T,∣∇u∣2=4u,Δu=2⟨x,H⃗⟩+4=4+2c.(21)\nabla u=2T,\qquad\lvert\nabla u\rvert^2=4u,\qquad\Delta u=2\langle x,\vec{H}\rangle+4=4+2c. \tag*{(21)}

Lemma 2.1 gives q∈Mq\in M with u(q)=0u(q)=0. By Lemma 2.2, Δu(q)≤4\Delta u(q)\le4. Equation (4.2) therefore forces c=0c=0 and H⃗≡0\vec{H}\equiv0. Then equation (1.2) gives x⊥=0x^\perp=0, i.e. x=Tx=T is tangent everywhere. Taking normal parts of DXx=XD_Xx=X yields

B(X,T)=0.(22)B(X,T)=0. \tag*{(22)}

At any point where T≠0T\ne0, choose e1=T/∣T∣e_1=T/\lvert T\rvert. Equation (4.3) shows B11=B12=0B_{11}=B_{12}=0, and H⃗=0\vec{H}=0 gives B22=0B_{22}=0. By (4.2), Δu=4\Delta u=4, so uu cannot vanish on an open set and the set {T≠0}\{T\ne0\} is dense. Continuity gives B=0B=0 globally. By Lemma 2.3, xx parametrizes an entire affine plane. Finally u(q)=0u(q)=0 and c=0c=0 imply x(q)=0x(q)=0, so the plane passes through the origin. □

Lagrangian structure and drift identities

For the remainder of the paper assume that x:M2→C2x:M^2\to\mathbb{C}^2 is Lagrangian. Set

C(X,Y,W)=⟨B(X,Y),JW⟩(23)C(X,Y,W)=\langle B(X,Y),JW\rangle \tag*{(23)}

Because JJ is parallel and J(TM)=NMJ(TM)=NM, the cubic tensor CC is totally symmetric. The covariant derivative ∇C\nabla C is totally symmetric in its four arguments by the Codazzi equation. The normal curvature is identified with the tangent curvature through JJ. That is, R⊥(X,Y)JW=JR(X,Y)WR^\perp(X,Y)JW=JR(X,Y)W. Then K=0K=0 implies flat normal connection.

The equations (1) and (2) can be written as

H⃗=(λx+V)⊥,(λ,V)=(0,V), V≠0,or(1,0).(24)\vec{H}=(\lambda x+V)^\perp,\qquad(\lambda,V)=(0,V),\ V\ne0,\qquad\text{or}\qquad(1,0). \tag*{(24)}

Define

Z=−(λx+V)⊤,ρ=∣H⃗∣2,S=∣B∣2,L=Δ−⟨Z,∇⋅⟩.(25)Z=-(\lambda x+V)^\top,\qquad\rho=|\vec{H}|^2,\qquad S=|B|^2,\qquad\mathcal{L}=\Delta-\langle Z,\nabla\cdot\rangle. \tag*{(25)}

Thus L=Δ+⟨V⊤,∇⋅⟩\mathcal{L}=\Delta+\langle V^\top,\nabla\cdot\rangle for translators and L=Δ+⟨x⊤,∇⋅⟩\mathcal{L}=\Delta+\langle x^\top,\nabla\cdot\rangle for self-expanders. Direct differentiation of (24) gives

∇X⊥H⃗=B(X,Z),∇XZ=−λX−AH⃗X,∇ρ=2AH⃗Z.(26)\nabla^\perp_X\vec{H}=B(X,Z),\qquad\nabla_XZ=-\lambda X-A_{\vec{H}}X,\qquad\nabla\rho=2A_{\vec{H}}Z. \tag*{(26)}

The final formula follows from

X(ρ)=2⟨∇X⊥H⃗,H⃗⟩=2⟨AH⃗Z,X⟩.X(\rho)=2\langle\nabla^\perp_X\vec{H},\vec{H}\rangle=2\langle A_{\vec{H}}Z,X\rangle.

With the conventions in (24)–(25), we obtain the following drift identities.

Lemma 5.1. For a Lagrangian surface satisfying (5.2), we have

12Lρ=∣∇⊥H⃗∣2−λρ−∣AH⃗∣2,(27)\frac{1}{2}\mathcal{L}\rho=|\nabla^\perp\vec{H}|^2-\lambda\rho-|A_{\vec{H}}|^2, \tag*{(27)}

and

12LS=∣∇⊥B∣2−λS−32S2+2ρS−12ρ2−∣AH⃗∣2.(28)\frac{1}{2}\mathcal{L}S=|\nabla^\perp B|^2-\lambda S-\frac{3}{2}S^2+2\rho S-\frac{1}{2}\rho^2-|A_{\vec{H}}|^2. \tag*{(28)}

Consequently, as 2K=ρ−S2K=\rho-S,

LK=∣∇⊥H⃗∣2−∣∇⊥B∣2+(−2λ−2ρ+6K)K.(29)\mathcal{L}K=|\nabla^\perp\vec{H}|^2-|\nabla^\perp B|^2+(-2\lambda-2\rho+6K)K. \tag*{(29)}

Proof. Tracing the derivative of the first identity in (26) and using Codazzi gives

Δ⊥H⃗=∇Z⊥H⃗−λH⃗−∑iB(ei,AH⃗ei).\Delta^\perp\vec{H}=\nabla^\perp_Z\vec{H}-\lambda\vec{H}-\sum_i B(e_i,A_{\vec{H}}e_i).

Taking its inner product with H⃗\vec{H} and using 12Δρ=∣∇⊥H⃗∣2+⟨Δ⊥H⃗,H⃗⟩\frac{1}{2}\Delta\rho=|\nabla^\perp\vec{H}|^2+\langle\Delta^\perp\vec{H},\vec{H}\rangle yields (27). For clarity, here is the algebraic part of the Simons formula used in (28). With respect to a tangent orthonormal frame write

AJe1=[abbc],AJe2=[bccd].A_{Je_1}=\begin{bmatrix}a&b\\b&c\end{bmatrix},\qquad A_{Je_2}=\begin{bmatrix}b&c\\c&d\end{bmatrix}.

Then

S=a2+3b2+3c2+d2,ρ=(a+c)2+(b+d)2,AH⃗=(a+c)AJe1+(b+d)AJe2.(30)S=a^2+3b^2+3c^2+d^2,\qquad\rho=(a+c)^2+(b+d)^2,\qquad A_{\vec{H}}=(a+c)A_{Je_1}+(b+d)A_{Je_2}. \tag*{(30)}

A direct expansion of these two symmetric matrices verifies

∑i,j=12[(tr⁡(AJeiAJej))2+∣[AJei,AJej]∣2]=32S2−2ρS+12ρ2+∣AH⃗∣2,(31)\sum_{i,j=1}^{2}\left[(\operatorname{tr}(A_{Je_i}A_{Je_j}))^2+|[A_{Je_i},A_{Je_j}]|^2\right]=\frac{3}{2}S^2-2\rho S+\frac{1}{2}\rho^2+|A_{\vec{H}}|^2, \tag*{(31)}

The standard Euclidean Simons identity for the second fundamental form, combined with (5.4), is

12LS=∣∇⊥B∣2−λS−∑i,j=12[(tr⁡(AJeiAJej))2+∣[AJei,AJej]∣2].\frac{1}{2}\mathcal{L}S = \lvert\nabla^\perp B\rvert^2-\lambda S-\sum_{i,j=1}^{2}\left[\left(\operatorname{tr}(A_{Je_i}A_{Je_j})\right)^2+\lvert[A_{Je_i},A_{Je_j}]\rvert^2\right].

Substituting (5.9) into the Simons identity gives (5.6). Finally, 2K=ρ−S2K=\rho-S and subtraction of the first two equations give (5.7). □

The cubic-tensor identity at negative constant curvature

The next calculation is the key point in excluding K<0K<0. It is important that a frame identity valid only at one point is not differentiated as though it held on a neighbourhood.

Lemma 6.1. Suppose K=−κ<0K=-\kappa<0 is constant and (5.2) holds. At any point where ρ>0\rho>0, choose an orthonormal tangent frame with H⃗=ρ Je1\vec H=\sqrt{\rho}\,Je_1 and write Z=ξe1+ζe2Z=\xi e_1+\zeta e_2. Then

κ∣Z∣2=ρξ2+(ρ+8κ)(λ+ρ+3κ).(32)\kappa\lvert Z\rvert^2=\rho\xi^2+(\rho+8\kappa)(\lambda+\rho+3\kappa). \tag*{(32)}

Proof. The cubic tensor Cijk=⟨B(ei,ej),Jek⟩C_{ijk}=\langle B(e_i,e_j),Je_k\rangle and its covariant derivative are totally symmetric. At the point in question write

(C111,C112,C122,C222)=(a,b,c,−b),h0=a+c=ρ.(33)(C_{111},C_{112},C_{122},C_{222})=(a,b,c,-b),\qquad h_0=a+c=\sqrt{\rho}. \tag*{(33)}

The last component C222C_{222} is −b-b because the Je2Je_2 component of H⃗\vec H is zero at this point. Directly from the Gauss equation,

K=ac−c2−2b2,S=a2+3c2+4b2.(34)K=ac-c^2-2b^2,\qquad S=a^2+3c^2+4b^2. \tag*{(34)}

For clarity, the three normal-valued coefficients are

B11=aJe1+bJe2,B12=bJe1+cJe2,B22=cJe1−bJe2.B_{11}=aJe_1+bJe_2,\qquad B_{12}=bJe_1+cJe_2,\qquad B_{22}=cJe_1-bJe_2.

At the point, denote the independent components of the symmetric four-tensor ∇C\nabla C by

p=C111,1,q=C111,2,r=C112,2,s=C122,2,t=C222,2.p=C_{111,1},\qquad q=C_{111,2},\qquad r=C_{112,2},\qquad s=C_{122,2},\qquad t=C_{222,2}.

Because ∇X⊥H⃗=B(X,Z)\nabla^\perp_X\vec H=B(X,Z) by (5.4), we have

p+r=A=aξ+bζ,q+s=F=bξ+cζ,r+t=E=cξ−bζ.(35)p+r=A=a\xi+b\zeta,\qquad q+s=F=b\xi+c\zeta,\qquad r+t=E=c\xi-b\zeta. \tag*{(35)}

Define

P=r−h0ζ8,Q=q−F2.(36)P=r-\frac{h_0\zeta}{8},\qquad Q=q-\frac{F}{2}. \tag*{(36)}

The multiplicities of symmetric entries yield

∣∇⊥B∣2=p2+4q2+6r2+4s2+t2,∣∇⊥H⃗∣2=A2+2F2+E2.\lvert\nabla^\perp B\rvert^2=p^2+4q^2+6r^2+4s^2+t^2,\qquad\lvert\nabla^\perp\vec H\rvert^2=A^2+2F^2+E^2.

Using p=A−rp=A-r, s=F−qs=F-q, t=E−rt=E-r,

∣∇⊥H⃗∣2−∣∇⊥B∣2=2r(A+E)−8r2−8(q−F/2)2=ρξ28−8(P2+Q2),(37)\begin{aligned} \lvert\nabla^\perp\vec{H}\rvert^2-\lvert\nabla^\perp B\rvert^2 &=2r(A+E)-8r^2-8(q-F/2)^2\\ &=\frac{\rho\xi^2}{8}-8(P^2+Q^2), \tag*{(37)} \end{aligned}

where A+E=h0ξA+E=h_0\xi.

One must use the differentiated full cubic invariant, rather than differentiating C222=−bC_{222}=-b on a neighborhood. In an arbitrary frame, the intrinsic Gauss expression is

K=C111C122+C112C222−C1122−C1222.K=C_{111}C_{122}+C_{112}C_{222}-C_{112}^2-C_{122}^2.

Differentiate this invariant formula first, and only then evaluate in the adapted frame where C222=−bC_{222}=-b. Total symmetry of ∇C\nabla C gives

K1=cp+(a−2c)r−3bq+bs,(38)K_1=cp+(a-2c)r-3bq+bs, \tag*{(38)}
K2=cq+(a−2c)s−3br+bt.(39)K_2=cq+(a-2c)s-3br+bt. \tag*{(39)}

Inserting (6.4) and (6.5) simplifies these to

K1=(ρ8+K2)ξ+(h0−4c)P−4bQ,(40)K_1=\left(\frac{\rho}{8}+\frac{K}{2}\right)\xi+(h_0-4c)P-4bQ, \tag*{(40)}
K2=K2ζ−4bP−(h0−4c)Q.(41)K_2=\frac{K}{2}\zeta-4bP-(h_0-4c)Q. \tag*{(41)}

Let

B=[h0−4c−4b−4b−h0+4c],D=ρ−8K.\mathcal{B}= \begin{bmatrix} h_0-4c & -4b\\ -4b & -h_0+4c \end{bmatrix}, \qquad D=\rho-8K.

From h0=a+ch_0=a+c and (6.3),

(h0−4c)2+16b2=(a−3c)2+16b2=(a+c)2−8(ac−c2−2b2)=D,(h_0-4c)^2+16b^2=(a-3c)^2+16b^2=(a+c)^2-8(ac-c^2-2b^2)=D,

so B2=DI\mathcal{B}^2=DI. Since KK is constant and D=ρ+8κ>0D=\rho+8\kappa>0, equations (6.9)-(6.10) imply

B(PQ)=−((ρ/8+K/2)ξ(K/2)ζ).\mathcal{B} \begin{pmatrix} P\\ Q \end{pmatrix} = -\begin{pmatrix} (\rho/8+K/2)\xi\\ (K/2)\zeta \end{pmatrix}.

Taking norms and using B2=DI\mathcal{B}^2=DI yields

P2+Q2=1D[(ρ8+K2)2ξ2+K24ζ2].(42)P^2+Q^2=\frac{1}{D}\left[\left(\frac{\rho}{8}+\frac{K}{2}\right)^2\xi^2+\frac{K^2}{4}\zeta^2\right]. \tag*{(42)}

Combining (6.6) and (6.11), and simplifying the coefficient of ξ2\xi^2, gives the full identity

∣∇⊥H⃗∣2−∣∇⊥B∣2=−2K2D∣Z∣2−2KρDξ2.(43)\lvert\nabla^\perp\vec{H}\rvert^2-\lvert\nabla^\perp B\rvert^2 =-\frac{2K^2}{D}\lvert Z\rvert^2-\frac{2K\rho}{D}\xi^2. \tag*{(43)}

Indeed,

ρ(ρ−8K)−(ρ+4K)28D=−2K(ρ+K)D.\frac{\rho(\rho-8K)-(\rho+4K)^2}{8D} =-\frac{2K(\rho+K)}{D}.

Finally, since LK=0\mathcal{L}K=0, inserting (6.12) into (5.7), setting K=−κK=-\kappa, and multiplying by D/(2κ)>0D/(2\kappa)>0 gives

0=−κ∣Z∣2+ρξ2+(ρ+8κ)(λ+ρ+3κ),0=-\kappa\lvert Z\rvert^2+\rho\xi^2+(\rho+8\kappa)(\lambda+\rho+3\kappa),

which is (32). □

Exclusion of nonzero constant Gaussian curvature

Proposition 7.1. No complete immersed translator or self-expander in R4\mathbb{R}^{4} has positive constant Gaussian curvature.

Proof. For a surface, Ric⁡=Kg\operatorname{Ric}=Kg. A positive constant KK makes MM compact by the Bonnet-Myers theorem. For a translator, f=⟨V,x⟩f=\langle V,x\rangle satisfies Δf=⟨V,H⃗⟩=∣H⃗∣2\Delta f=\langle V,\vec{H}\rangle=|\vec{H}|^{2}. Integration gives H⃗=0\vec{H}=0, then Δ∣x∣2=4\Delta|x|^{2}=4, contradicting compactness. For a self-expander,

12Δ(∣x∣2)=2+⟨x,H⃗⟩=2+∣H⃗∣2>0,\frac{1}{2}\Delta(|x|^{2})=2+\langle x,\vec{H}\rangle=2+|\vec{H}|^{2}>0,

which is also impossible on a compact manifold. □

Proposition 7.2. A complete Lagrangian self-expander in C2\mathbb{C}^{2} cannot have negative constant Gaussian curvature.

Proof. Suppose K=−κ<0K=-\kappa<0. Let f=∣x∣2/2≥0f=|x|^{2}/2\ge0. For self-expanders, Z=−x⊤=−∇fZ=-x^{\top}=-\nabla f and λ=1\lambda=1. On Ω={ρ>0}\Omega=\{\rho>0\}, Lemma 6.1 implies

κ∣∇f∣2=ρξ2+(ρ+8κ)(1+ρ+3κ)≥8κ(1+3κ).(44)\kappa|\nabla f|^{2}=\rho\xi^{2}+(\rho+8\kappa)(1+\rho+3\kappa)\ge8\kappa(1+3\kappa). \tag*{(44)}

We show that Ω\Omega is dense. If H⃗=0\vec{H}=0 on a nonempty open set, the self-expander equation gives x⊥=0x^{\perp}=0 there. For tangent vectors X,YX,Y, the second equation of (5.4) yields

∇2f(X,Y)=⟨X,Y⟩+⟨x⊥,B(X,Y)⟩=⟨X,Y⟩.\nabla^{2}f(X,Y)=\langle X,Y\rangle+\langle x^{\perp},B(X,Y)\rangle=\langle X,Y\rangle.

Thus ∇X∇f=X\nabla_{X}\nabla f=X there. Commuting these derivatives yields R(X,Y)∇f=0R(X,Y)\nabla f=0. For a 2-dimensional Riemannian manifold, its curvature tensor is completely determined by the Gaussian curvature:

R(X,Y)W=K(⟨W,Y⟩X−⟨W,X⟩Y)R(X,Y)W=K\bigl(\langle W,Y\rangle X-\langle W,X\rangle Y\bigr)

where the corresponding curvature sign convention is adopted. Choose a local orthonormal frame {e1,e2}\{e_{1},e_{2}\}, and write ∇f=f1e1+f2e2\nabla f=f_{1}e_{1}+f_{2}e_{2}. It follows from R(X,Y)∇f=0R(X,Y)\nabla f=0 and K=−κ≠0K=-\kappa\ne0 that f1=f2=0f_{1}=f_{2}=0. Then on a surface of nonzero Gaussian curvature this implies ∇f=0\nabla f=0, contrary to ∇2f=g\nabla^{2}f=g. Consequently the continuous estimate (7.1) holds on all of MM. Since ff is bounded below, Lemma 2.4 supplies points pjp_{j} with ∣∇f∣(pj)→0|\nabla f|(p_{j})\to0. Inequality (7.1) gives instead ∣∇f∣2≥8(1+3κ)>0|\nabla f|^{2}\ge8(1+3\kappa)>0, a contradiction. □

Proposition 7.3. A complete Lagrangian translator in C2\mathbb{C}^{2} cannot have negative constant Gaussian curvature.

Proof. Suppose K=−κ<0K=-\kappa<0. The translator equation and Gauss identity yield

0≤ρ=∣V⊥∣2≤∣V∣2,∣Z∣=∣V⊤∣≤∣V∣,S=ρ+2κ≤∣V∣2+2κ.(45)0\le\rho=|V^{\perp}|^{2}\le|V|^{2},\qquad|Z|=|V^{\top}|\le|V|,\qquad S=\rho+2\kappa\le|V|^{2}+2\kappa. \tag*{(45)}

If H⃗=0\vec{H}=0 identically, then VV is tangent, and from DXV=0D_XV=0, it is a nonzero parallel tangent field which forces K=0K=0, contrary to hypothesis. Therefore ρ∗=sup⁡Mρ>0\rho_*=\sup_M\rho>0. Apply Lemma 2.4 to −ρ-\rho. We find pjp_j with

ρ(pj)→ρ∗,∣∇ρ∣(pj)→0.(46)\rho(p_j)\to\rho_*,\qquad|\nabla\rho|(p_j)\to0. \tag*{(46)}

For large jj, use the mean curvature frame of Lemma 6.1, so that H⃗=ρJe1\vec{H}=\sqrt{\rho}Je_1. By (7.2), the coefficients a,b,ca,b,c of (6.2) and ξ,ζ\xi,\zeta of ZZ are uniformly bounded. After taking a subsequence, these five real numbers converge to a,b,c,ξ,ζa,b,c,\xi,\zeta, with h0=a+c=ρ∗>0h_0=a+c=\sqrt{\rho_*}>0. Let

A=[abbc],z=(ξζ).\mathcal{A}=\begin{bmatrix}a&b\\b&c\end{bmatrix},\qquad z=\begin{pmatrix}\xi\\\zeta\end{pmatrix}.

At the points pjp_j, the gradient identity in (5.4) is ∇ρ=2AH⃗Z\nabla\rho=2A_{\vec{H}}Z. In the chosen frame AH⃗=h0AA_{\vec{H}}=h_0\mathcal{A}. Passing to the limit in (7.3) gives

Az=0.(47)\mathcal{A}z=0. \tag*{(47)}

Similarly, Lemma 6.1 with λ=0\lambda=0 passes to the limit as

κ∣z∣2=ρ∗ξ2+(ρ∗+8κ)(ρ∗+3κ).(48)\kappa|z|^2=\rho_*\xi^2+(\rho_*+8\kappa)(\rho_*+3\kappa). \tag*{(48)}

The right-hand side is positive, so z≠0z\ne0. Equation (47) therefore implies that det⁡A=ac−b2=0\det\mathcal{A}=ac-b^2=0. Combining this with the Gauss relation (6.3) gives

b2+c2=κ,h0c=(a+c)c=κ.(49)b^2+c^2=\kappa,\qquad h_0c=(a+c)c=\kappa. \tag*{(49)}

The eigenvalues of the symmetric 2×22\times2 matrix A\mathcal{A} are 00 and h0h_0. Hence ker⁡A\ker\mathcal{A} is one-dimensional. Direct calculation yields Av=0\mathcal{A}v=0 for v=(c,−b)Tv=(c,-b)^{\mathsf T}. Then v∈ker⁡Av\in\ker\mathcal{A}. Since z≠0z\ne0 and Az=0\mathcal{A}z=0, there necessarily exists a non-zero real number τ\tau such that z=τvz=\tau v. Thus,

ξ=τc,ζ=−τb,∣z∣2=τ2(c2+b2)=τ2κ\xi=\tau c,\qquad\zeta=-\tau b,\qquad|z|^2=\tau^2(c^2+b^2)=\tau^2\kappa

from (7.6). Therefore, inserting h0c=κh_0c=\kappa and ρ∗=h02\rho_*=h_0^2 yields

ρ∗ξ2=h02τ2c2=τ2(h0c)2=τ2κ2=κ∣z∣2.(50)\rho_*\xi^2=h_0^2\tau^2c^2=\tau^2(h_0c)^2=\tau^2\kappa^2=\kappa|z|^2. \tag*{(50)}

Substitution of (7.7) into (7.5) forces 0=(ρ∗+8κ)(ρ∗+3κ)>00=(\rho_*+8\kappa)(\rho_*+3\kappa)>0, the desired contradiction. □

From the above analysis, the following result can be derived.

Corollary 7.1. Every complete connected Lagrangian translator or self-expander in C2\mathbb{C}^2 with constant Gaussian curvature is flat.

Global splitting in the flat case

Proposition 8.1. Let x:M2→C2x:M^2\to\mathbb{C}^2 be a complete, connected, flat Lagrangian immersion satisfying (5.2). Up to a unitary transformation, its lift to the universal cover is a Cartesian product of two complete, unit-speed planar soliton curves in orthogonal complex lines.

Proof. As JJ identifies the tangent and normal connections, K=0K=0 implies the normal curvature R⊥R^{\perp} vanishes. By the Ricci equation all shape operators commute. At a point with B≠0B\ne0, at least one shape operator is nonscalar: otherwise Lagrangian symmetry would force B=0B=0. A nonscalar self-adjoint operator in dimension two has distinct eigenvalues, hence determines a smooth local orthonormal eigenframe {e1,e2}\{e_1,e_2\}; commutation diagonalizes the remaining shape operators simultaneously. In this common principal frame, for any normal vector ν∈NpM\nu\in N_pM, we have ⟨B(e1,e2),ν⟩=0\langle B(e_1,e_2),\nu\rangle=0. That is, B(e1,e2)=0B(e_1,e_2)=0. Total symmetry of CC then yields

B(e1,e1)=k1Je1,B(e2,e2)=k2Je2.(51)B(e_1,e_1)=k_1Je_1,\qquad B(e_2,e_2)=k_2Je_2. \tag*{(51)}

Write ∇eie1=ωie2\nabla_{e_i}e_1=\omega_i e_2, ∇eie2=−ωie1\nabla_{e_i}e_2=-\omega_i e_1. In the diagonal frame, total symmetry of ∇C\nabla C gives

e2(k1)=ω1k1,e1(k2)=−ω2k2,k1ω2+k2ω1=0.(52)e_2(k_1)=\omega_1k_1,\qquad e_1(k_2)=-\omega_2k_2,\qquad k_1\omega_2+k_2\omega_1=0. \tag*{(52)}

For completeness, these are obtained by equating C111,2C_{111,2} with C112,1C_{112,1}, C222,1C_{222,1} with C122,2C_{122,2}, and C112,2C_{112,2} with C122,1C_{122,1}, respectively. Moreover, the Je2Je_2 component of ∇e1⊥H⃗=B(e1,Z)\nabla^\perp_{e_1}\vec{H}=B(e_1,Z) equals zero. Since H⃗=k1Je1+k2Je2\vec{H}=k_1Je_1+k_2Je_2, this component is e1(k2)+k1ω1e_1(k_2)+k_1\omega_1. Together with (52),

k1ω1−k2ω2=0,k2ω1+k1ω2=0.k_1\omega_1-k_2\omega_2=0,\qquad k_2\omega_1+k_1\omega_2=0.

The determinant is k12+k22>0k_1^2+k_2^2>0 at every point of this neighbourhood. Thus

ω1=ω2=0(53)\omega_1=\omega_2=0 \tag*{(53)}

on this nonflat neighbourhood. In particular, the local diagonalising frame is parallel.

The universal Riemannian cover (M~,g~)(\widetilde{M},\widetilde{g}) is complete, simply connected, and flat; hence it is globally isometric to the Euclidean plane with coordinates (s,t)(s,t). Extend the local parallel eigenframe uniquely to global parallel fields ∂s,∂t\partial_s,\partial_t on the cover. As ∇∂s∂t=0\nabla_{\partial_s}\partial_t=0 and B(∂s,∂t)=0B(\partial_s,\partial_t)=0 on the initial neighbourhood,

xst=∇∂s∂t+B(∂s,∂t)=0(54)x_{st}=\nabla_{\partial_s}\partial_t+B(\partial_s,\partial_t)=0 \tag*{(54)}

To propagate this identity, use the soliton equation in the global Euclidean coordinates:

xss+xtt=λx+V−⟨λx+V,xs⟩xs−⟨λx+V,xt⟩xt.(55)x_{ss}+x_{tt}=\lambda x+V-\langle\lambda x+V,x_s\rangle x_s-\langle\lambda x+V,x_t\rangle x_t. \tag*{(55)}

Here ∣xs∣=∣xt∣=1|x_s|=|x_t|=1 and ⟨xs,xt⟩=0\langle x_s,x_t\rangle=0. Equation (55) is a semilinear elliptic system with analytic nonlinearities; standard analytic elliptic regularity [23] implies that xx is real analytic. Therefore the analytic tensor xstx_{st}, which vanishes on the open set in (54), vanishes throughout R2\mathbb{R}^2. If B≡0B\equiv0 instead, the same conclusion follows immediately. Thus

xst=0on R2.(56)x_{st}=0\quad\text{on }\mathbb{R}^2. \tag*{(56)}

Integrating gives x(s,t)=α(s)+β(t)x(s,t)=\alpha(s)+\beta(t), with ∣α′∣=∣β′∣=1|\alpha'|=|\beta'|=1. The orthogonality of xs,xtx_s,x_t and the Lagrangian condition yield

⟨α′,β′⟩=⟨Jα′,β′⟩=0.(57)\langle\alpha',\beta'\rangle=\langle J\alpha',\beta'\rangle=0. \tag*{(57)}

By (51) on the initial neighbourhood and analyticity,

α′′(s)∈span⁡{Jα′(s)},β′′(t)∈span⁡{Jβ′(t)}.\alpha^{\prime\prime}(s) \in\operatorname{span}\{J\alpha'(s)\}, \qquad\beta^{\prime\prime}(t) \in\operatorname{span}\{J\beta'(t)\}.

Therefore the real two-planes L1=span⁡{α′,Jα′}L_1 = \operatorname{span}\{\alpha',J\alpha'\} and L2=span⁡{β′,Jβ′}L_2 = \operatorname{span}\{\beta',J\beta'\} are each independent of their parameters. Equation (57) gives L1⊥L2L_1 \perp L_2. They are fixed orthogonal complex lines in C2\mathbb{C}^2. Choosing a unitary UU that sends the coordinate lines to L1,L2L_1,L_2 and absorbing the constant components of xx into the factors, we obtain

x(s,t)=U(γ1(s),γ2(t)).x(s,t) = U(\gamma_1(s),\gamma_2(t)).

The normal projection in (24) splits over these complex lines, giving

γi′′=(Vi)⊥γi for translators,γi′′=(γi)⊥γi for expanders,\gamma_i^{\prime\prime} = (V_i)^{\perp\gamma_i}\ \text{for translators}, \qquad\gamma_i^{\prime\prime} = (\gamma_i)^{\perp\gamma_i}\ \text{for expanders},

where U−1(λx+V)=(λγ1+V1,λγ2+V2)U^{-1}(\lambda x + V) = (\lambda\gamma_1 + V_1,\lambda\gamma_2 + V_2) and ⊥γi\perp\gamma_i denotes the projection onto the normal direction of the corresponding plane curve. Both parameters traverse all of R\mathbb{R}, so the curve factors are complete. □

Planar curve factors and proofs of classification

Translating plane curves

Let γ:R→C\gamma:\mathbb{R}\to\mathbb{C} be complete and unit speed with γ′′=v⊥γ\gamma^{\prime\prime}=v^{\perp\gamma} for fixed v∈Cv\in\mathbb{C}. If v=0v=0, the curve γ\gamma is a line. For v≠0v\ne0, rotate the plane so that v=(0,a)v=(0,a) with a=∣v∣>0a=|v|>0. Write T=γ′=(cos⁡θ,sin⁡θ)T=\gamma'=(\cos\theta,\sin\theta) and N=JT=(−sin⁡θ,cos⁡θ)N=JT=(-\sin\theta,\cos\theta). The curvature equation is

θ′=⟨v,N⟩=acos⁡θ.(58)\theta'=\langle v,N\rangle=a\cos\theta. \tag*{(58)}

An equilibrium solution is a straight line parallel to vv. Every nonequilibrium solution has cos⁡θ\cos\theta of a fixed sign; reversing orientation allows −π/2<θ<π/2-\pi/2<\theta<\pi/2. On that branch the curve is a graph y=y(u)y=y(u), and (58) becomes

y′′1+(y′)2=a.(59)\frac{y''}{1+(y')^2}=a. \tag*{(59)}

Integrating twice yields, after translations,

γa(u)=(u,−1alog⁡cos⁡(au)),−π2a<u<π2a.(60)\gamma_a(u)=\left(u,-\frac{1}{a}\log\cos(au)\right),\qquad-\frac{\pi}{2a}<u<\frac{\pi}{2a}. \tag*{(60)}

The arclength form ds=sec⁡(au)duds=\sec(au)du diverges at both endpoints, so the maximal arclength parametrisation has domain R\mathbb{R}. It is injective because the horizontal coordinate is strictly monotone, and proper because its height diverges at both ends. Uniqueness of the curve ODE shows that a complete curve must be one of these maximal solutions. For a prescribed velocity VV, each factor velocity Vi=(U−1V)iV_i=(U^{-1}V)_i is fixed: a line must be parallel to ViV_i when Vi≠0V_i\ne0, and a grim reaper has speed parameter ∣Vi∣|V_i|.

Expanding plane curves

Let γ:R→C\gamma:\mathbb{R}\to\mathbb{C} be a unit-speed solution of γ′′=γ′⊥\gamma''=\gamma'^\perp. Put

T=γ′,N=JT,p=⟨γ,T⟩,k=⟨γ,N⟩.T=\gamma',\qquad N=JT,\qquad p=\langle\gamma,T\rangle,\qquad k=\langle\gamma,N\rangle.

Since T′=kNT'=kN and N′=−kTN'=-kT, differentiating gives

p′=1+k2,k′=−pk,T′=kN.(61)p'=1+k^2,\qquad k'=-pk,\qquad T'=kN. \tag*{(61)}

As p′≥1p'\geq1, a solution defined for all s∈Rs\in\mathbb{R} has p(s)→+∞p(s)\to+\infty as s→+∞s\to+\infty and p(s)→−∞p(s)\to-\infty as s→−∞s\to-\infty, so it has a unique zero. Translate the arc-length parameter to make this zero occur at s=0s=0, reverse orientation if necessary, and rotate the ambient plane so that

p(0)=0,k(0)=a≥0,T(0)=(1,0),γ(0)=(0,a).(62)p(0)=0,\qquad k(0)=a\geq0,\qquad T(0)=(1,0),\qquad\gamma(0)=(0,a). \tag*{(62)}

If a=0a=0, uniqueness gives k=0k=0, p=sp=s and the line through the origin. Suppose a>0a>0. By uniqueness kk remains positive, and on s≥0s\geq0,

p(s)≥s,0<k(s)≤a,1≤p′(s)≤1+a2.p(s)\geq s,\qquad0<k(s)\leq a,\qquad1\leq p'(s)\leq1+a^2.

The analogous backward bounds hold for s≤0s\leq0. Therefore the solution of (9.4) exists for all s∈Rs\in\mathbb{R}. Conversely, solving (9.4) and then putting γ=pT+kN\gamma=pT+kN gives

γ′=(p′−k2)T+(pk+k′)N=T,\gamma'=(p'-k^2)T+(pk+k')N=T,

so the reconstructed curve satisfies the desired expander equation.

To prove embeddedness, write T=(cos⁡θ,sin⁡θ)T=(\cos\theta,\sin\theta) with θ(0)=0\theta(0)=0; thus θ′=k>0\theta'=k>0. For s>0s>0 set ϕ=arctan⁡(p/k)\phi=\arctan(p/k). Direct differentiation yields

ϕ′=kp′−pk′p2+k2=k+kp2+k2>k=θ′.(63)\phi'=\frac{kp'-pk'}{p^2+k^2}=k+\frac{k}{p^2+k^2}>k=\theta'. \tag*{(63)}

Here ϕ(0)=θ(0)=0\phi(0)=\theta(0)=0, hence 0<θ(s)<ϕ(s)<π/20<\theta(s)<\phi(s)<\pi/2 on s>0s>0. The symmetry of the initial-value system implies pp and θ\theta are odd and kk is even. Therefore −π/2<θ(s)<π/2-\pi/2<\theta(s)<\pi/2 for all s∈Rs\in\mathbb{R}. It follows that ddsγ1(s)=cos⁡θ(s)>0\frac{d}{ds}\gamma_1(s)=\cos\theta(s)>0, proving injectivity.

Also p(s)≥sp(s)\geq s for s≥0s\geq0 and p(s)≤sp(s)\leq s for s≤0s\leq0; since

∣γ(s)∣2=p(s)2+k(s)2≥s2,|\gamma(s)|^2=p(s)^2+k(s)^2\geq s^2,

the curve is proper. Denote the normalized curve by Γa\Gamma_a, with Γ0=R\Gamma_0=\mathbb{R}. Every complete planar expanding curve has this form up to rotation and reparametrization. The planar classification was established in [12].

Descent to the surface and proof of the main theorems

Proofs of Theorems 1.3 and 1.4. Corollary 7.1 gives K=0K=0. By Proposition 8.1, the universal cover M~≅R2\widetilde{M}\cong\mathbb{R}^2 has lifted immersion x~(s,t)=U(γ1(s),γ2(t))\widetilde{x}(s,t)=U(\gamma_1(s),\gamma_2(t)). The preceding subsections show that each complete factor is injective and proper, so the Cartesian product is injective and proper. If σ\sigma is a deck transformation, then x~∘σ=x~\widetilde{x}\circ\sigma=\widetilde{x}. Injectivity forces σ\sigma to be the identity. Hence the deck group is trivial and MM itself is isometric to R2\mathbb{R}^2. The original immersion is the proper embedded product stated in the theorems.

Conversely, for complete unit-speed curves in orthogonal complex lines, the product has metric ds2+dt2ds^2+dt^2 and is flat and complete. Its tangent planes are Lagrangian, and both its mean curvature vector and the normal projection of VV or xx split into the respective planar components. The factor curve equations imply the corresponding surface soliton equation. □

Acknowledgements

The first author was partially supported by NSFC grant No.12401060, Natural Science Foundation of Henan Province grant No.262300421232. The second author was partially supported by NSFC grant No. 12501063, Postdoctoral Fellowship Program of CPSF grant No. GZC20252036.

Declarations

Conflict of interest There are no conflicts of interest with third parties.

Data availability No datasets were generated or analysed in this study.

Zhi Li,

School of Mathematics and Statistics, Henan Normal University, 453007, Xinxiang, Henan, China.

Email address: [email protected]

Cheng Xing,

School of Mathematics and Statistics, Henan Normal University, 453007, Xinxiang, Henan, China.

Email address: [email protected]

References

  1. [1]S. J. Altschuler and L. F. Wu, Translating surfaces of the non-parametric mean curvature flow with prescribed contact angle, Calc. Var. Partial Differential Equations, 2 (1994), no. 1, 101-111.DOI
  2. [2]S. Ancari and X. Cheng, Volume properties and rigidity on self-expanders of mean curvature flow, Geom. Dedicata, 216 (2022), no. 2, Paper No. 24, 25 pp.arxiv.org/abs/2012.03364
  3. [3]H. Anciaux, Construction of Lagrangian self-similar solutions to the mean curvature flow in C^n, Geom. Dedicata, 120 (2006), 37-48.DOI
  4. [4]I. Castro and A. M. Lerma, Hamiltonian stationary self-similar solutions for Lagrangian mean curvature flow in the complex Euclidean plane, Proc. Amer. Math. Soc., 138 (2010), no. 5, 1821-1832.
  5. [5]I. Castro and A. M. Lerma, Translating solitons for Lagrangian mean curvature flow in complex Euclidean plane, Internat. J. Math., 23 (2012), no. 10, 1250101.arxiv.org/abs/1007.1886
  6. [6]I. Castro, A. M. Lerma, The Clifford torus as a self-shrinker for the Lagrangian mean curvature flow, Int. Math. Res. Not., 6 (2014), 1515-1527.
  7. [7]B.-Y. Chen, Classification of Lagrangian surfaces of constant curvature in the complex Euclidean plane, Proc. Edinb. Math. Soc., 48 (2005), no. 2, 337-364.DOI
  8. [8]Q. -M. Cheng, H. Hori and G. Wei, Complete Lagrangian self-shrinkers in ℝ⁴, Math. Z., 301 (2022), no. 4, 3417-3468.
  9. [9]X. Cheng and D. Zhou, Spectral properties and rigidity for self-expanding solutions of the mean curvature flows, Math. Ann., 371 (2018), no. 1-2, 371-389.arxiv.org/abs/1709.04086
  10. [10]J. Clutterbuck, O. Schnürer and F. Schulze, Stability of translating solutions to mean curvature flow, Calc. Var. Partial Differential Equations, 29 (2007), no. 3, 281-293.DOI
  11. [11]I. Ekeland, On the variational principle, J. Math. Anal. Appl., 47 (1974), no.2, 324-353.DOI
  12. [12]P. H. Halldorsson, Self-similar solutions to the curve shortening flow, Trans. Amer. Math. Soc., 364 (2012), no.10, 5285-5309.
  13. [13]H. P. Halldorsson, Helicoidal surfaces rotating/translating under the mean curvature flow, Geom. Dedicata, 162 (2013), 45-65.
  14. [14]D. Joyce, Y. -I. Lee and M. -T. Tsui, Self-similar solutions and translating solitons for Lagrangian mean curvature flow, J. Differential Geom., 84 (2010), 127-161.arxiv.org/abs/0801.3721
  15. [15]Y. -I. Lee and M. -T. Wang, Hamiltonian stationary shrinkers and expanders for Lagrangian mean curvature flows, J. Differential Geom., 83 (2009), 27-42.arxiv.org/abs/0707.0239
  16. [16]X. X. Li, Y. Y. Lu and R. N. Qiao, A uniqueness theorem of complete Lagrangian translator in C², Manuscripta Math., 164 (2021), no. 1-2, 251-265.DOI
  17. [17]H. Li and X. Wang, New characterizations of the Clifford torus as a Lagrangian self-shrinker, J. Geom. Anal., 27 (2017), no. 2, 1393-1412.arxiv.org/abs/1505.05593
  18. [18]Z. Li, R. X. Wang and G. Wei, The rigidity theorem for complete Lagrangian self-shrinkers, J. Geom. Anal., 35 (2025), no. 2, Paper No. 69, 17 pp.arxiv.org/abs/2412.01162
  19. [19]Z. Li and G. Wei, Complete Lagrangian self-expanders in C², J. Geom. Phys., 198 (2024), Paper No. 105107, 10 pp.DOI
  20. [20]Z. Li and G. Wei, Classification of Lagrangian translators and Lagrangian self-expanders in ℂ², Manuscripta Math., 176 (2025), no. 3, Paper No. 30, 17 pp.arxiv.org/abs/2405.14086
  21. [21]Z. Li, G. Wei and C. Xing, Rigidity theorems of complete self-shrinkers in the Euclidean space, 2026, arXiv:2610.29842v1.arxiv.org/abs/2610.29842
  22. [22]Z. Li, G. Wei and C. Xing, Complete lagrangian self-shrinkers with constant gaussian curvature, 2026, preprint.
  23. [23]C. B. Morrey, Jr., Multiple Integrals in the Calculus of Variations. Classics in Math., Springer-Verlag, Berlin, 2008; reprint of the 1966 edition.
  24. [24]A. Neves and G. Tian, Translating solutions to Lagrangian mean curvature flow, Trans. Amer. Math. Soc., 365 (2013), no. 11, 5655-5680.arxiv.org/abs/0711.4341
  25. [25]K. Smoczyk, Self-expanders of the mean curvature flow, Vietnam J. Math., 49 (2021), no. 2, 433-445.arxiv.org/abs/2005.05803

Paper details

vlnc:W_v722qbau
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents