betaFeedback

WorksMathematics

Uniform restriction lower bounds and nongeodesic common nodal curves in integrable systems

Abstract

We construct a family of real analytic metrics on the two-torus, for which a fixed embedded nongeodesic circle is a nodal component of infinitely many real Laplace eigenfunctions. The examples address a question asked by Bourgain-Rudnick \cite{BR11}, and give a negative answer to the question asked by Toth-Zelditch \cite{TZ21} and Han \cite{Han}: whether a curve shared by infinitely many nodal sets must be geodesic. To explain the restriction obstruction, we prove uniform L2L^2 lower bounds for joint eigenfunctions of quantum completely integrable (QCI) systems under transverse crossing and isolation of a tangential characteristic branch, then extend to multi-branch setting. The results are used to characterize the possible joint nodal curves in the Liouville tori, narrowing down the choices. Finally, the counterexample realizes one possibility for which cancellation happens exactly on a nongeodesic curve.

Introduction and main results

A fixed curve can occur in the nodal sets of eigenfunctions at arbitrarily large eigenvalues. On a flat two-torus, a theorem of Bourgain–Rudnick says that a real analytic curve has this property precisely when it is a segment of a closed geodesic [5 Theorem 1.1]. Does the same geometric rigidity hold on a general analytic surface? Toth–Zelditch raise this question in their discussion of nodal curves [16 §1.3 of the arXiv version]; it is also formulated in Han’s problem notes [10 Problem 4.9].

We give an explicit negative answer. One fixed analytic metric, with no nonzero Killing field, admits a fixed embedded nongeodesic circle which is a regular nodal component of an infinite sequence of real joint eigenfunctions. The construction persists within an explicit family converging to a flat metric. Thus analyticity or the absence of continuous symmetries do not guarantee that the common nodal curves is a geodesic segment.

The nongeodesic counterexample

A fixed curve can occur in the nodal sets of eigenfunctions at arbitrarily large eigenvalues. A fixed curve can occur in the nodal sets of eigenfunctions at arbitrarily large eigenvalues. Fix a metric and a curve. Call a fixed curve persistent if it is contained in the nodal sets of eigenfunctions with eigenvalues tending to infinity.

Theorem 1.1 (A nongeodesic common nodal circle). On M=(R/2πZ)2M = (\mathbb{R}/2\pi\mathbb{Z})^2, let 0<ε<20 < \varepsilon< 2 and set

gε=Fε(x,y)(dx2+dy2),Fε=10+εcos⁡3x+4εcos⁡6y.(1)g_\varepsilon= F_\varepsilon(x,y)(dx^2+dy^2), \qquad F_\varepsilon= 10+\varepsilon\cos3x+4\varepsilon\cos6y. \tag*{(1)}

There are distinct eigenvalues λj→∞\lambda_j \to\infty and real L2(M,gε)L^2(M,g_\varepsilon)-normalized eigenfunctions uju_j such that

γ(t)=(2t,t)(mod2π),t∈R/2πZ,(2)\gamma(t)=(2t,t)\pmod{2\pi}, \qquad t\in\mathbb{R}/2\pi\mathbb{Z}, \tag*{(2)}

is a regular connected component of every nodal set {uj=0}\{u_j=0\}. If

a(x)=2+εcos⁡3x,b(y)=8+4εcos⁡6y,Qε=Fε−1(−b∂x2+a∂y2),a(x)=2+\varepsilon\cos3x,\qquad b(y)=8+4\varepsilon\cos6y,\qquad Q_\varepsilon=F_\varepsilon^{-1}(-b\partial_x^2+a\partial_y^2),

then −Δgεuj=λjuj-\Delta_{g_\varepsilon}u_j=\lambda_j u_j and Qεuj=0Q_\varepsilon u_j=0. The metric has no nonzero Killing vector field, and the signed geodesic curvature of the oriented circle (2)

kgε(t)=9εsin⁡6t10(2+εcos⁡6t)3/2.(3)k_{g_\varepsilon}(t)=\frac{9\varepsilon\sin6t}{10(2+\varepsilon\cos6t)^{3/2}}. \tag*{(3)}

In particular, the common nodal circle is not a geodesic. Moreover, gε→10(dx2+dy2)g_\varepsilon\to10(dx^2+dy^2) in C∞C^\infty as ε↓0\varepsilon\downarrow0.

The analytic part of the paper identifies the relevant obstruction. Restriction records the tangential component of a characteristic covector. If a branch with positive joint Liouville mass is isolated by this tangential projection and crosses the hypersurface transversely, a microlocal Cauchy estimate gives a uniform restriction lower bound. Distinct branches with the same tangential covector can instead cancel. Our example realizes this cancellation exactly on a nongeodesic curve. The general restriction theorem, its extension to disconnected fibers, and the explicit construction therefore give complementary information about persistent nodal sets.

The winding numbers of γ\gamma are (2,1)(2,1). Formula (3) distinguishes a coordinate straight line from a geodesic of the metric. Its curvature is nonzero on open arcs, so even such arcs can be shared by infinitely many nodal sets.

The proof in Section 7 uses nonreal Floquet multipliers for a one-dimensional weighted eigenvalue problem. The same scalar solutions, inserted at two different coordinate scales, give exact joint eigenfunctions with a common phase front. The restriction theorem below supplies necessary geometric conditions for persistence, while the construction shows that the nongeodesic alternative can occur.

Restriction on a regular joint torus

Let MM be a compact smooth manifold without boundary, of dimension nn, equipped with a smooth positive density. Let

P(h)=(P1(h),…,Pn(h)),Pj(h)∈Ψh1(M),P(h) = (P_1(h),\ldots,P_n(h)), \qquad P_j(h) \in\Psi_h^1(M),

be commuting self-adjoint operators whose full symbols have the form pj+hpj,1(h)p_j + hp_{j,1}(h), with pjp_j real and the lower-order symbols uniformly bounded in the appropriate symbol classes. Assume that ∑jPj(h)2\sum_j P_j(h)^2 is uniformly semiclassically elliptic of order two. Write

p=(p1,…,pn):T∗M⟶Rn,ΛE=p−1(E).p = (p_1,\ldots,p_n):T^*M \longrightarrow\mathbb{R}^n, \qquad\Lambda_E = p^{-1}(E).

We assume that EE is a regular value, so dp1∧⋯∧dpn≠0dp_1 \wedge\cdots\wedge dp_n \ne0 on the compact joint fiber. Each connected component is a Lagrangian torus by the Liouville–Arnold theorem. Consider normalized exact joint eigenfunctions

Pj(h)uh=Ej(h)uh,E(h)⟶E,∥uh∥L2(M)=1.(4)P_j(h)u_h = E_j(h)u_h, \qquad E(h) \longrightarrow E, \qquad\lVert u_h\rVert_{L^2(M)} = 1. \tag*{(4)}

The customary assumption E(h)=E+O(h)E(h) = E + O(h) is included. Allowing convergence without a specified rate is justified by the parameter-dependent local construction in Section 2.

Let H⊂MH \subset M be a compact smooth embedded hypersurface, possibly with boundary; all patches below lie in its interior. For a local defining function ρ\rho, set

TH∗M={(x,ξ):x∈H},ΣE=ΛE∩TH∗M,rH(x,ξ)=(x,ξ∣TxH).T_H^*M = \{(x,\xi):x \in H\}, \qquad\Sigma_E = \Lambda_E \cap T_H^*M, \qquad r_H(x,\xi) = (x,\xi|_{T_xH}).

An isolated transverse point z0∈ΣEz_0 \in\Sigma_E means the following. There is an index jj with Hpjρ(z0)≠0H_{p_j}\rho(z_0) \ne0, a sufficiently small neighborhood UU of z0z_0, and an open set W⊂T∗HW \subset T^*H containing rH(z0)r_H(z_0) such that

ΣE∩rH−1(W)⊂U.(5)\Sigma_E \cap r_H^{-1}(W) \subset U. \tag*{(5)}

Here UU is chosen so that, in coordinates H={x1=0}H = \{x_1 = 0\}, ∂ξ1pj≠0\partial_{\xi_1}p_j \ne0 and pj=Ejp_j = E_j is a single graph ξ1=a0(x,ξ′)\xi_1 = a_0(x,\xi') in UU. Thus (5) excludes competing points of the entire joint fiber over the selected tangential window. This definition is independent of the choice of defining function and coordinates.

Theorem 1.2 (Restriction on a connected regular joint fiber). Assume that ΛE\Lambda_E is connected and that HH has an isolated transverse point on ΛE\Lambda_E. Then the joint eigenfunctions (4) satisfy

∥uh∣H∥L2(H)≥c>0(6)\lVert u_h|_H\rVert_{L^2(H)} \ge c > 0 \tag*{(6)}

for all sufficiently small hh.

In particular, the hypotheses hold if rHr_H is injective on ΣE\Sigma_E and some Hamiltonian flow crosses HH transversely there. The local formulation (5) is sufficient because a single open flow box has positive normalized Liouville measure on a connected joint torus.

Finite fibers and the geometry of persistence

The rotating joint fiber of a Liouville torus has four connected components. To apply the restriction argument to this model, each component must have an observable patch.

Theorem 1.3 (Restriction on finitely many joint tori). Suppose that ΛE=⋃ν=1NΛE,ν\Lambda_E = \bigcup_{\nu=1}^{N} \Lambda_{E,\nu} is a compact regular joint fiber. Assume that every component ΛE,ν\Lambda_{E,\nu} contains an isolated transverse point over HH, with isolation always understood relative to the full fiber as in (5). Then (6) holds for every normalized sequence (4).

The components may carry unknown and sequence-dependent proportions of the mass. The proof in Section 6 uses the decomposition of a defect measure into normalized Liouville measures on the finitely many components.

For a Liouville metric g=(a(x)+b(y))(dx2+dy2)g=(a(x)+b(y))(dx^2+dy^2) and a rotating joint energy κ\kappa, write Aκ=a+κA_\kappa=\sqrt{a+\kappa} and Bκ=b−κB_\kappa=\sqrt{b-\kappa}. The four tangential momenta at a nonzero curve tangent vv are

σAκvx+τBκvy,σ,τ∈{+1,−1}.\sigma A_\kappa v_x+\tau B_\kappa v_y,\qquad\sigma,\tau\in\{+1,-1\}.

They are distinct precisely when

vxvy≠0,Aκ2vx2≠Bκ2vy2.(7)v_xv_y\ne0,\qquad A_\kappa^2v_x^2\ne B_\kappa^2v_y^2. \tag*{(7)}

Theorem 1.3 therefore gives a restriction lower bound near any such point. Its contrapositive forces every connected persistent joint nodal curve to follow one of the four line fields

ker⁡dx,ker⁡dy,ker⁡(Aκdx+Bκdy),ker⁡(Aκdx−Bκdy).\ker dx,\qquad\ker dy,\qquad\ker(A_\kappa dx+B_\kappa dy),\qquad\ker(A_\kappa dx-B_\kappa dy).

Section 6 proves this rigidity statement and its consequences: a fixed contractible closed curve cannot persist in a compact rotating window, a closed noncoordinate persistent curve requires an action resonance, and restriction on a curve with a separating point determines an exact joint eigenfunction stably.

The circle of Theorem 1.1 follows one of the permitted phase fronts. Its action resonance is Iy(0)=2Ix(0)I_y(0)=2I_x(0). Along the circle, two distinct transverse characteristic branches have the same tangential covector and their traces cancel. This realizes the obstruction to Theorem 1.3. The same local Cauchy estimate determines the quantum limit of the constructed real sequence: it is the equal mixture of Liouville measures on the two cancelling components (Proposition 7.3). Because its full joint fiber has four components, we retain a separate spherical example in Section 5 which demonstrates the necessity of branch isolation within the connected-fiber setting of Theorem 1.2.

Related results and organization

Upper restriction bounds for Laplace eigenfunctions were established by Burq–Gérard–Tzvetkov [3]; for a hypersurface the general bound is O(h−1/4)O(h^{-1/4}). In the quantum integrable surface setting, Toth obtained logarithmic bounds for suitable curves [13]. Bourgain–Rudnick proved uniform toral restriction estimates [4, 5]. Quantum ergodic restriction provides lower bounds along suitable subsequences under dynamical and asymmetry assumptions [15, 6], and Galkowski–Zélditch studied lower bounds for Cauchy data on negatively curved surfaces [9]. Here the joint equations provide invariance under all commuting Hamiltonian flows, and isolation of a tangential branch prevents cancellation.

The construction in Theorem 1.1 concerns a single fixed curve at infinitely many eigenvalues of a single metric. Prescribed-nodal-set results for individual eigenfunctions, such as [7], address an interesting yet different question. Common curved hypersurfaces in higher-dimensional flat tori already arise from products [5 §1.2]; our result concerns a closed surface. The underlying Liouville framework is standard [8], and the exact construction uses elementary Floquet theory. The geometric conclusion is the failure of geodesic rigidity even for analytic metrics without Killing fields arbitrarily close to a flat metric.

Sections 2 and 3 prove the local Cauchy estimate and Theorem 1.2. Sections 4 and 5 retain the nodal-intersection application and spherical examples. Sections 6 and 7 contain the finite-component extension and rigidity consequences, and the exact nongeodesic counterexample and its interpretation.

Disclaimer on AI usage

This paper uses generative AI models. Precisely, we used ChatGPT 6 Astra, Deepseek V4.1-Flash for the mathematical part and the writing, checking and graphing. Some other AI models are used in a nonessential way during the writing. Of course, the authors are responsible for the correctness of the whole content of this paper.

Acknowledgements

The first author is supported by the research project funded by Wuhan University of Technology (Grant No. 202601hx0314).

Both authors would like to thank Prof. John Toth and Prof. Long Jin for helpful discussions relating to this work.

A microlocal Cauchy estimate

We first prove a local estimate which does not require the joint fiber to be connected. Its constants are uniform for the joint energy in a sufficiently small fixed neighborhood of EE. In particular, convergence E(h)→EE(h) \to E is sufficient; no estimate on its rate is needed.

We use two consequences of joint ellipticity. For joint eigenfunctions with energy ee in a fixed bounded neighborhood of EE, elliptic regularity gives uniform bounds in every semiclassical Sobolev space and a fixed compact microlocalization:

u=Πhu+O(h∞)∥u∥L2(M)in every Hhm(M),Πh∈Ψhcomp(M).(8)u = \Pi_h u + O(h^\infty)\|u\|_{L^2(M)} \qquad\text{in every } H_h^m(M), \qquad\Pi_h \in\Psi_h^{\mathrm{comp}}(M). \tag*{(8)}

Moreover, a compact microlocal cutoff supported where ∣p−e∣≥c>0|p-e| \ge c > 0 annihilates uu to O(h∞)O(h^\infty), uniformly when the separation constant is uniform. These assertions follow by applying an elliptic parametrix to ∑k(Pk−ek)2\sum_k(P_k-e_k)^2; see, for example, [19]. Here and below an O(h∞)O(h^\infty) remainder has all the Sobolev bounds required for restriction. For a compactly microlocalized operator GhG_h we will use

∥RHGhw∥L2(H)≤Ch−1/2∥w∥L2(M),RHw=w∣H.(9)\|R_HG_hw\|_{L^2(H)} \le Ch^{-1/2}\|w\|_{L^2(M)}, \qquad R_Hw = w|_H. \tag*{(9)}

In local coordinates this follows from Cauchy–Schwarz in the bounded semiclassical normal-frequency variable.

Choose an isolated transverse point z0∈ΛE∩TH∗Mz_0 \in\Lambda_E \cap T_H^*M as in the hypothesis of Theorem 1.2. Write x=(t,x′)x=(t,x'), with H={t=0}H=\{t=0\}, and choose jj so that ∂ξtpj≠0\partial_{\xi_t}p_j \ne0 near z0z_0. Shrinking the tangential window WW and the selected neighborhood if necessary, we may arrange that their relevant closures are compact. The isolation hypothesis then persists for all ee sufficiently close to EE: every point of

p−1(e)∩TH∗M∩rH−1(W0‾),W0⋐W,(10)p^{-1}(e) \cap T_H^*M \cap r_H^{-1}(\overline{W_0}), \qquad W_0 \Subset W, \tag*{(10)}

lies in the selected noncharacteristic neighborhood. Indeed, a contrary sequence has a convergent subsequence by joint ellipticity and would produce a competing point of the full fiber at EE. This argument uses the isolation of the entire joint fiber, not just one of its connected components.

All coordinate calculations below are made in Lebesgue measure. If the ambient density is m(t,x′) dt dx′m(t,x')\,dt\,dx', conjugate locally by m1/2m^{1/2}. The conjugated operators have the same principal symbols and uniformly bounded subprincipal symbols. The restriction of m1/2um^{1/2}u differs from the natural density realization of u∣Hu|_H by a smooth positive multiplication operator. Thus the coordinate estimates give the stated estimates in the original norms. Spatial cutoffs are taken equal to one on a larger neighborhood of the chosen flow box; their errors are microlocally negligible in the smaller neighborhoods used below.

Factorization at the actual energy

The following noncharacteristic reduction is the usual first-order evolution factorization; compare [2, 12, 19]. We include its subprincipal construction because it is important for the size of the evolution error.

Lemma 2.1 (Uniform parameter-dependent factorization). For ee in a sufficiently small neighborhood of EE, there exist Qe,h∈Ψh0Q_{e,h} \in\Psi_h^0, elliptic in the selected neighborhood, and a tangential family Ae,h(t)∈Ψh1A_{e,h}(t) \in\Psi_h^1 such that

Qe,h(Pj−ej)=hDt−Ae,h(t)+h2Se,h(11)Q_{e,h}(P_j-e_j)=hD_t-A_{e,h}(t)+h^2S_{e,h} \tag*{(11)}

microlocally there. Here Dt=−i∂tD_t=-i\partial_t and, after the compact microlocal cutoffs, Se,hS_{e,h} is uniformly bounded on L2L^2; it may be taken in Ψh−1\Psi_h^{-1}. The tangential symbol is

ae,h=ae+ha1,e,h,ae real,a1,e,h∈S0 uniformly,a_{e,h}=a_e+ha_{1,e,h},\qquad a_e\ \text{real},\qquad a_{1,e,h}\in S^0\ \text{uniformly},

and

Ae,h−Ae,h∗∈hΨh0(12)A_{e,h}-A_{e,h}^*\in h\Psi_h^0 \tag*{(12)}

uniformly in ee and hh.

Proof. The implicit function theorem, applied with the parameter eje_j, gives

re≔pj−ej=e0,e(t,x′,ξt,ξ′)(ξt−ae(t,x′,ξ′)),∣e0,e∣≥c>0.r_e\coloneqq p_j-e_j=e_{0,e}(t,x',\xi_t,\xi')(\xi_t-a_e(t,x',\xi')),\qquad\lvert e_{0,e}\rvert\ge c>0.

All symbols and their derivatives are uniform on fixed smaller neighborhoods. Write the full Weyl symbol of Pj−ejP_j-e_j as re+hpj,1,hr_e+hp_{j,1,h} and set q0,e=e0,e−1q_{0,e}=e_{0,e}^{-1}. With the convention Hab={a,b}H_{ab}=\{a,b\}, the first Weyl correction is

de,h=q0,epj,1,h+12i{q0,e,re}.d_{e,h}=q_{0,e}p_{j,1,h}+\frac{1}{2i}\{q_{0,e},r_e\}.

Define

a1,e,h(t,x′,ξ′)=−de,h(t,x′,ae(t,x′,ξ′),ξ′).a_{1,e,h}(t,x',\xi')=-d_{e,h}(t,x',a_e(t,x',\xi'),\xi').

The function de,h+a1,e,hd_{e,h}+a_{1,e,h} vanishes on ξt=ae\xi_t=a_e, so smooth division gives

de,h+a1,e,h=(ξt−ae)ce,h.d_{e,h}+a_{1,e,h}=(\xi_t-a_e)c_{e,h}.

Taking q1,e,h=−e0,e−1ce,hq_{1,e,h}=-e_{0,e}^{-1}c_{e,h}, the Weyl product satisfies

(q0,e+hq1,e,h)#(re+hpj,1,h)=ξt−ae−ha1,e,h+O(h2).(q_{0,e}+hq_{1,e,h})\mathbin{\#}(r_e+hp_{j,1,h})=\xi_t-a_e-ha_{1,e,h}+O(h^2).

This proves (11). Extend aea_e as a real tangential symbol, using fixed cutoffs outside the working neighborhood. The real principal symbol and the bounded subprincipal term give (12). In particular, the construction factors pj−ejp_j-e_j itself: the quantity (ej−Ej)/h(e_j-E_j)/h never occurs. □

Propagation and comparison with Dirichlet data

Proposition 2.2 (Local Cauchy estimate). At an isolated transverse point z0z_0 there are a fixed χ∈C∗∞(T∗M)\chi\in C^\infty_*(T^*M), a uniformly bounded tangential operator BH∈Ψh0(H)B_H\in\Psi_h^0(H), a neighborhood V\mathcal V of EE, and constants C,h0>0C,h_0>0 such that

∥Op⁡hw(χ)u∥L2(M)≤C∥BH(u∣H)∥L2(H)+Ch1/2∥u∥L2(M)≤C′∥u∣H∥L2(H)+Ch1/2∥u∥L2(M)(13)\|\operatorname{Op}_h^w(\chi)u\|_{L^2(M)}\le C\|B_H(u|_H)\|_{L^2(H)}+Ch^{1/2}\|u\|_{L^2(M)}\le C'\|u|_H\|_{L^2(H)}+Ch^{1/2}\|u\|_{L^2(M)} \tag*{(13)}

for every exact joint eigenfunction Pku=ekuP_ku=e_ku, e∈Ve\in\mathcal V, and 0<h<h00<h<h_0. The principal symbol β\beta of BHB_H is supported in W0W_0 and equals one near rH(z0)r_H(z_0). The symbol χ\chi equals one on a nonempty relatively open flow box in ΛE\Lambda_E through z0z_0.

Proof. Write P=PjP=P_j and p=pjp=p_j in this proof, and initially take ∥u∥L2(M)=1\|u\|_{L^2(M)}=1. Choose β∈Cc∞(W0)\beta\in C^\infty_c(W_0), equal to one near rH(z0)r_H(z_0). On the transversal {t=0}\{t=0\} choose a compact cutoff

α(x′,ξt,ξ′)=β(x′,ξ′)ϑ(x′,ξt,ξ′)\alpha(x',\xi_t,\xi')=\beta(x',\xi')\vartheta(x',\xi_t,\xi')

in the selected neighborhood, where ϑ=1\vartheta=1 near the graph ξt=ae(0,x′,ξ′)\xi_t=a_e(0,x',\xi') over supp⁡β\operatorname{supp}\beta for all e∈Ve\in\mathcal V. Extend α\alpha along the HpH_p flow. Multiplying by a longitudinal cutoff outside a larger fixed slab produces b∈Cc∞(T∗M)b\in C^\infty_c(T^*M) such that

Hpb=0for ∣t∣<3δ,b=1near z0.(14)H_pb=0\quad\text{for }|t|<3\delta,\qquad b=1\quad\text{near }z_0. \tag*{(14)}

All supports lie inside the factorization neighborhood in this slab. The number δ>0\delta>0 is fixed independently of hh and ee.

Set Bh=Op⁡hw(b)B_h=\operatorname{Op}_h^w(b) and v=Bhuv=B_hu. The commutator formula and (14) give

(P−ej)v=[P,Bh]u=O(h2)in the working slab.(P-e_j)v=[P,B_h]u=O(h^2)\quad\text{in the working slab}.

This is a localized statement: the principal commutator vanishes in the slab, whereas the derivatives of the longitudinal cutoff are outside it. Insert nested cutoffs in the normal variable

before applying Lemma 2.1. Pseudolocality makes the separated cutoff errors O(h∞)O(h^{\infty}) in the inner slab, and hence

(hDt−Ae,h(t))v=fh,∥fh∥L2((−2δ,2δ)×Rn−1)≤Ch2.(15)(hD_t-A_{e,h}(t))v=f_h,\qquad\|f_h\|_{L^2((-2\delta,2\delta)\times\mathbb{R}^{n-1})}\le Ch^2. \tag*{(15)}

The real tangential principal symbol and (12) give the energy estimate

∥v(t)∥Lx′2≤C∥v(0)∥Lx′2+Ch∫min⁡(0,t)max⁡(0,t)∥fh(s)∥Lx′2 ds,∣t∣≤δ.\|v(t)\|_{L^2_{x'}}\le C\|v(0)\|_{L^2_{x'}}+\frac{C}{h}\int_{\min(0,t)}^{\max(0,t)}\|f_h(s)\|_{L^2_{x'}}\,ds,\qquad|t|\le\delta.

One may obtain this directly by differentiating ∥v(t)∥2\|v(t)\|^2; the skew-adjoint part contributes at most C∥v(t)∥2C\|v(t)\|^2. Consequently,

∥v∥L2({∣t∣<δ})≤C∥RHBhu∥L2(H)+Ch.(16)\|v\|_{L^2(\{|t|<\delta\})}\le C\|R_HB_hu\|_{L^2(H)}+Ch. \tag*{(16)}

We next compare the localized trace with the original Dirichlet data. Define the tangential symbol

βe(t,x′,ξ′)=b(t,x′,ae(t,x′,ξ′),ξ′),βe(0,x′,ξ′)=β(x′,ξ′),\beta_e(t,x',\xi')=b(t,x',a_e(t,x',\xi'),\xi'),\qquad\beta_e(0,x',\xi')=\beta(x',\xi'),

and let Be,h(t)=Op⁡h,x′w(βe(t))B_{e,h}(t)=\operatorname{Op}^{w}_{h,x'}(\beta_e(t)). Smooth division in the selected neighborhood gives

b−βe=ce(ξt−ae).(17)b-\beta_e=c_e(\xi_t-a_e). \tag*{(17)}

Choose a compact microlocal cutoff Θh\Theta_h supported in that neighborhood and equal to the identity near the relevant support of bb at HH and near every point of (10) over supp⁡β\operatorname{supp}\beta. Joint ellipticity and the absence of competing branches give

RHBhu=RHBhΘhu+O(h∞),Op⁡h,x′w(β)RHu=RHBe,hΘhu+O(h∞).(18)R_HB_hu=R_HB_h\Theta_hu+O(h^{\infty}),\qquad\operatorname{Op}^{w}_{h,x'}(\beta)R_Hu=R_HB_{e,h}\Theta_hu+O(h^{\infty}). \tag*{(18)}

For the second equality, only the tangentially localized restriction is being discarded outside the selected neighborhood. To justify it, first use (8); the restriction relation then retains precisely the ambient covectors over supp⁡β\operatorname{supp}\beta. Those on the joint fiber lie where Θh\Theta_h is the identity, and the remaining compact set is uniformly joint elliptic. This also proves the asserted O(h∞)O(h^{\infty}) after taking the trace.

Quantizing (17), with nested compact microlocal cutoffs, gives

(Bh−Be,h)Θh=Ce,h(hDt−Ae,h)Θh+hFe,h+O(h∞),(19)(B_h-B_{e,h})\Theta_h=C_{e,h}(hD_t-A_{e,h})\Theta_h+hF_{e,h}+O(h^{\infty}), \tag*{(19)}

where Ce,hC_{e,h} and Fe,hF_{e,h} are uniformly bounded and compactly microlocalized. Importantly, the cutoff state need not solve the homogeneous equation to order h2h^2. Instead,

(hDt−Ae,h)Θhu=Qe,h[P,Θh]u+O(h2)=O(h)(hD_t-A_{e,h})\Theta_hu=Q_{e,h}[P,\Theta_h]u+O(h^2)=O(h)

on the support needed in (19). The commutator is O(h)O(h), which is sufficient. Applying (9) to (19) and then (18) yields

RHBhu=Op⁡h,x′w(β)RHu+O(h1/2)L2(H).(20)R_HB_hu=\operatorname{Op}^{w}_{h,x'}(\beta)R_Hu+O(h^{1/2})_{L^2(H)}. \tag*{(20)}

After undoing the density conjugation this is the same assertion with a uniformly bounded tangential operator BHB_H of principal symbol β\beta.

Finally choose η∈Cc∞((−δ,δ))\eta\in C^\infty_c((-\delta,\delta)) equal to one near zero and set χ(t,x′,ξ)=η(t)b(t,x′,ξ)\chi(t,x',\xi)=\eta(t)b(t,x',\xi). Then

Op⁡hw(χ)=ηBh+O(h)L2→L2.\operatorname{Op}^{w}_h(\chi)=\eta B_h+O(h)_{L^2\to L^2}.

Equations (16) and (20) prove (13). This last cutoff makes the estimate global on MM. The construction makes χ=1\chi=1 on a small flow box through z0z_0. Homogeneity restores the factor ∥u∥L2(M)\|u\|_{L^2(M)} in the error term.

□\square

The connected-fiber restriction theorem

We recall explicitly the invariant-measure argument, so that its use of connectedness and its lack of a nonresonance assumption are clear.

Lemma 3.1 (The joint semiclassical measure). Suppose p−1(E)=ΛEp^{-1}(E)=\Lambda_E is compact, connected, and regular. Every semiclassical measure of normalized joint eigenfunctions with E(h)→EE(h)\to E is the normalized Liouville measure μE\mu_E on ΛE\Lambda_E. In particular, for each fixed compactly supported symbol aa,

∥Op⁡hw(a)uh∥L2(M)2⟶∫ΛE∣a∣2 dμE.(21)\left\|\operatorname{Op}_{h}^{w}(a)u_h\right\|_{L^2(M)}^2 \longrightarrow\int_{\Lambda_E}|a|^2\,d\mu_E. \tag*{(21)}

Proof. Compact microlocalization (8) ensures that every subsequence has a further subsequence with a semiclassical measure of total mass one. The joint equations and E(h)→EE(h)\to E put its support in ΛE\Lambda_E. Self-adjointness and the commutator identity give

∫Hpja dμ=0,j=1,…,n.\int H_{p_j}a\,d\mu=0,\qquad j=1,\ldots,n.

Thus μ\mu is invariant under all commuting Hamiltonian flows. At a regular point their vector fields are independent and span TΛET\Lambda_E. In Liouville–Arnold angle coordinates they generate all translations of the connected torus. Their only invariant probability measure is normalized Haar measure, namely μE\mu_E. This is the connected-fiber form of the standard joint-measure description in [14]. Every convergent subsequence has this same limit, proving (21) for the original family. □

Proof of Theorem 1.2. Apply Proposition 2.2 at the isolated transverse point. Transversality implies that a small relatively open patch Γ0⊂ΛE∩TH∗M\Gamma_0\subset\Lambda_E\cap T_H^*M through that point has a flow tube

T={exp⁡(tHpj)(z):z∈Γ0, ∣t∣<t0}T=\{\exp(tH_{p_j})(z):z\in\Gamma_0,\ |t|<t_0\}

which is open in ΛE\Lambda_E. The cutoff in the proposition can be chosen equal to one on a smaller such tube. Consequently,

c0:=∫ΛE∣χ∣2 dμE>0.c_0:=\int_{\Lambda_E}|\chi|^2\,d\mu_E>0.

Lemma 3.1 gives ∥Op⁡hw(χ)uh∥L2(M)≥c0/2\left\|\operatorname{Op}_{h}^{w}(\chi)u_h\right\|_{L^2(M)}\ge\sqrt{c_0/2} for sufficiently small hh. On the other hand, (13) bounds this norm by C∥uh∣H∥L2(H)+Ch1/2C\|u_h|_H\|_{L^2(H)}+Ch^{1/2}. Absorbing the last term proves the asserted positive lower bound. □

Corollary 3.2 (Conditional LpL^p restriction bounds). In the setting of Theorem 1.2, assume in addition that HH is compact and an independent estimate gives ∥uh∣H∥L∞(H)≤C\|u_h|_H\|_{L^\infty(H)}\le C for the normalized family. Then, for every 1≤p≤∞1\le p\le\infty,

cp≤∥uh∣H∥Lp(H)≤Cp.c_p\le\|u_h|_H\|_{L^p(H)}\le C_p.

Proof. The upper bounds follow from the assumed sup-norm estimate. For 1≤p≤21\le p\le2 use

∥uh∣H∥L2(H)2≤∥uh∣H∥L∞(H)2−p∥uh∣H∥Lp(H)p.\|u_h|_H\|_{L^2(H)}^2\le\|u_h|_H\|_{L^\infty(H)}^{2-p}\|u_h|_H\|_{L^p(H)}^p.

For 2≤p≤∞2\le p\le\infty, Hölder’s inequality and the L2L^2 lower bound give the result. □

Remark 3.3 (Connectedness and nonresonance). No arithmetic nonresonance assumption is used: invariance under all nn flows already forces Haar measure on a connected regular fiber. For a disconnected regular fiber, invariant probability measures can distribute their mass among its torus components. The local estimate above remains valid on an isolated sheet; the additional issue is whether every possible component carrying mass is detected by the restriction.

Remark 3.4 (Turning boundaries). The transverse hypothesis cannot hold at a smooth one-sided boundary of the projected torus. Indeed, suppose locally π(ΛE)⊂{ρ≥0}\pi(\Lambda_E)\subset\{\rho\ge0\}, with H={ρ=0}H=\{\rho=0\} and dρ≠0d\rho\ne0. At z∈ΛE∩TH∗Mz\in\Lambda_E\cap T_H^*M, every HpjH_{p_j} flow stays on ΛE\Lambda_E. The function t↦ρ(π(exp⁡(tHpj)z))t\mapsto\rho(\pi(\exp(tH_{p_j})z)) has a local minimum at zero, and hence Hpjρ(z)=0H_{p_j}\rho(z)=0 for every jj. Thus a lower-bound problem at such a turning hypersurface requires an argument other than the noncharacteristic Cauchy estimate used here.

Sign changes from restriction lower bounds

The restriction estimate has a nodal consequence when combined with decay of period integrals. For a complex-valued eigenfunction, it is preferable to count sign changes of its real and imaginary parts separately: the union of their zero sets is the entire curve if either part vanishes identically.

For a continuous real-valued function ff on a parametrized compact arc II, let Nsc(f;I)N_{\mathrm{sc}}(f;I) be the supremum of the integers N≥0N \ge0 for which there are parameters s0<⋯<sNs_0 < \cdots< s_N in II satisfying f(si)f(si+1)<0f(s_i)f(s_{i+1}) < 0 for 0≤i<N0 \le i < N. We set Nsc(0;I)=0N_{\mathrm{sc}}(0;I) = 0 and allow the value +∞+\infty. For functions with isolated zeros this is the usual number of sign changes.

Whenever a square root of the Laplacian is used below, its symbol is understood near the positive joint energy. A smooth spectral modification at low energy puts it in the global semiclassical calculus without changing the eigenfunctions under consideration.

Proposition 4.1 (A sign-change criterion). Let γ\gamma be a fixed compact smooth arc, and let vhv_h be continuous complex-valued functions on γ\gamma. Suppose that

sup⁡0<h<h0∥vh∥L∞(γ)<∞,(22)\sup_{0<h<h_0}\|v_h\|_{L^\infty(\gamma)} < \infty, \tag*{(22)}

and that every nontrivial compact subarc II in the interior of γ\gamma satisfies

lim inf⁡h→0∫I∣vh∣2 ds>0,∫Ivh ds⟶0.(23)\liminf_{h\to0}\int_I |v_h|^2\,ds > 0,\qquad\int_I v_h\,ds \longrightarrow0. \tag*{(23)}

Then

Nsc(Re⁡vh;γ)+Nsc(Im⁡vh;γ)⟶∞.(24)N_{\mathrm{sc}}(\operatorname{Re}v_h;\gamma)+N_{\mathrm{sc}}(\operatorname{Im}v_h;\gamma)\longrightarrow\infty. \tag*{(24)}

In particular, if the functions vhv_h are real-valued, their number of sign changes tends to infinity.

Proof. Fix an integer N≥1N \ge1 and choose NN pairwise disjoint nontrivial compact subarcs I1,…,INI_1,\ldots,I_N in the interior of γ\gamma. The first condition in (4.2) and (4.1) imply, for each jj, that

∫Ij(∣Re⁡vh∣+∣Im⁡vh∣) ds≥∫Ij∣vh∣ ds≥∫Ij∣vh∣2 ds∥vh∥L∞(γ)≥cj>0(25)\int_{I_j}(|\operatorname{Re}v_h|+|\operatorname{Im}v_h|)\,ds\ge\int_{I_j}|v_h|\,ds\ge\frac{\int_{I_j}|v_h|^2\,ds}{\|v_h\|_{L^\infty(\gamma)}}\ge c_j>0 \tag*{(25)}

for all sufficiently small hh. If neither real nor imaginary part takes both positive and negative values on IjI_j, then each has a constant weak sign there. Consequently,

∫Ij(∣Re⁡vh∣+∣Im⁡vh∣) ds=∣Re⁡∫Ijvh ds∣+∣Im⁡∫Ijvh ds∣≤2∣∫Ijvh ds∣=o(1),\int_{I_j}(|\operatorname{Re}v_h|+|\operatorname{Im}v_h|)\,ds=\left|\operatorname{Re}\int_{I_j}v_h\,ds\right|+\left|\operatorname{Im}\int_{I_j}v_h\,ds\right|\le\sqrt{2}\left|\int_{I_j}v_h\,ds\right|=o(1),

contradicting (4.4). Thus each IjI_j contributes a sign transition of at least one of the two parts. Since the arcs are disjoint, the sum in (4.3) is at least NN for all sufficiently small hh. The integer NN was arbitrary. □

We next give hypotheses under which the three estimates in the proposition follow from results on joint eigenfunctions. In this paragraph MM is a surface, P1(h)=−h2ΔgP_1(h)=\sqrt{-h^2\Delta_g}, and the limiting first energy is E1=1E_1=1. For a geodesic segment γ\gamma parametrized by arclength τ\tau, set

Cγ={(x,ξ):x∈γ, ∣ξ∣g2=1}.C_\gamma=\{(x,\xi):x\in\gamma,\ |\xi|_g^2=1\}.

The admissibility condition of Wang–Wu [17 Definition 1], applied to the moment map (p12,p2)(p_1^2,p_2), is

∣∂τp2(x,ξ)∣≥cϵ>0on Cγ∩{∣p2−E2∣<ϵ}(26)|\partial_\tau p_2(x,\xi)|\ge c_\epsilon>0\quad\text{on }C_\gamma\cap\{|p_2-E_2|<\epsilon\} \tag*{(26)}

for some ϵ>0\epsilon>0. Here ∂τ\partial_\tau differentiates at fixed cotangent components in Fermi coordinates along γ\gamma, with τ\tau the arclength coordinate. This condition is a separate assumption; it does not follow from the noncharacteristic crossing condition of Theorem 1.2.

Corollary 4.2 (Nodal intersections on admissible geodesic segments). Assume the connected regular joint-fiber and joint-eigenvalue hypotheses of Theorem 1.2, with MM a surface, P1(h)=−h2ΔgP_1(h)=\sqrt{-h^2\Delta_g}, E1=1E_1=1, and E(h)=E+O(h)E(h)=E+O(h). Let γ\gamma be a compact embedded geodesic segment. Suppose that

  1. every point of γ\gamma belongs to π(ΛE)\pi(\Lambda_E);

  2. dξp1∧dξp2≠0d_\xi p_1\wedge d_\xi p_2\ne0 on ΛE∩Tγ∗M\Lambda_E\cap T_\gamma^*M;

  3. rγ:ΛE∩Tγ∗M→Tγ∗r_\gamma:\Lambda_E\cap T_\gamma^*M\to T_\gamma^* is injective;

  4. γ\gamma satisfies (26).

Then every normalized joint eigenfunction family in that theorem satisfies

Nsc(Re⁡uh∣γ;γ)+Nsc(Im⁡uh∣γ;γ)⟶∞.N_{\mathrm{sc}}(\operatorname{Re}u_h|_\gamma;\gamma)+N_{\mathrm{sc}}(\operatorname{Im}u_h|_\gamma;\gamma)\longrightarrow\infty.

For real-valued joint eigenfunctions this is a conclusion about the sign changes of uh∣γu_h|_\gamma itself.

Proof. At any point (x,ξ)∈ΛE∩Tγ∗M(x,\xi)\in\Lambda_E\cap T_\gamma^*M, the two vectors ∂ξp1\partial_\xi p_1 and ∂ξp2\partial_\xi p_2 span TxMT_xM by the fiber-rank assumption. Therefore at least one of the two Hamiltonian vector fields crosses γ\gamma at this point. The first and third hypotheses allow us to apply Theorem 1.2 to every nontrivial compact subarc I⊂γI\subset\gamma. We obtain

lim inf⁡h→0∥uh∥L2(I)>0.\liminf_{h\to0}\|u_h\|_{L^2(I)}>0.

The fiber-rank hypothesis also gives

∥uh∥L∞(γ)≤C.(27)\|u_h\|_{L^\infty(\gamma)}\le C. \tag*{(27)}

Indeed, after joint elliptic localization to a neighborhood of the joint fiber, the rank-one pointwise estimate on the energy circle gives this bound on a neighborhood of each point of γ\gamma; see [18 Theorem 1 and the second assertion of Theorem 2]. Compactness of γ\gamma makes the constant uniform. The rank condition here is imposed on all characteristic lifts over γ\gamma, rather than merely on one of its branches.

Every fixed subsegment II inherits (26). Since P1(h)2=−h2ΔgP_1(h)^2=-h^2\Delta_g and E1(h)2=1+O(h)E_1(h)^2=1+O(h), [17 Theorem 2] applies and yields

∣∫Iuh ds∣≤CIh1/2.(28)\left|\int_I u_h\,ds\right|\le C_Ih^{1/2}. \tag*{(28)}

Proposition 4.1 now applies to uh∣γu_h|_\gamma. □

Remark 4.3. This proof uses unweighted periods over finitely many fixed subsegments, as stated in [17]; it does not require a weighted extension of that theorem. The constants CIC_I in (28) are allowed to depend on II. Consequently the argument gives qualitative divergence and does not, by itself, give a uniform power-law lower bound for the number of sign changes. This is the same comparison between absolute mass and signed periods that underlies nodal arguments such as [11].

Spherical harmonics and the necessity of branch isolation

The sphere illustrates the connected-fiber restriction theorem and shows why crossing alone is insufficient, even when the full joint fiber is connected. Equip S2S^2 with the round metric

g=dθ2+sin⁡2θ dϕ2,0<θ<π,g=d\theta^2+\sin^2\theta\,d\phi^2,\qquad0<\theta<\pi,

and consider

P1(h)=−h2ΔS2,P2(h)=−ih∂ϕ.P_1(h)=\sqrt{-h^2\Delta_{S^2}},\qquad P_2(h)=-ih\partial_\phi.

These operators commute, and their principal symbols are

p1=(ξθ2+ξϕ2sin⁡2θ)1/2,p2=ξϕ.p_1=\left(\xi_\theta^2+\frac{\xi_\phi^2}{\sin^2\theta}\right)^{1/2},\qquad p_2=\xi_\phi.

The normalized complex spherical harmonics Yℓm(θ,ϕ)=Fℓ,m(θ)eimϕY_\ell^m(\theta,\phi)=F_{\ell,m}(\theta)e^{im\phi} are joint eigenfunctions, where Fℓ,mF_{\ell,m} is a real multiple of Pℓm(cos⁡θ)P_\ell^m(\cos\theta).

Fix 0<μ<10 < \mu< 1, put hℓ=(ℓ+12)−1h_\ell= \left(\ell+ \frac{1}{2}\right)^{-1}, and choose an integer mℓm_\ell nearest to μ/hℓ\mu/h_\ell. Then, for all sufficiently large ℓ\ell, 0≤mℓ≤ℓ0 \le m_\ell\le\ell and

E1,ℓ=hℓℓ(ℓ+1)=1−hℓ2/4=1+O(hℓ2),E2,ℓ=hℓmℓ=μ+O(hℓ).(29)E_{1,\ell} = h_\ell\sqrt{\ell(\ell+1)} = \sqrt{1-h_\ell^2/4} = 1+O(h_\ell^2), \qquad E_{2,\ell} = h_\ell m_\ell= \mu+ O(h_\ell). \tag*{(29)}

Thus the joint eigenvalues obey the O(h)O(h) energy-window condition. The limiting joint fiber is

Λμ={∣ξ∣g=1, ξφ=μ},ξθ2+μ2sin⁡2θ=1.(30)\Lambda_\mu= \{|\xi|_g=1,\ \xi_\varphi=\mu\}, \qquad\xi_\theta^2+\frac{\mu^2}{\sin^2\theta}=1. \tag*{(30)}

It is a compact connected regular Lagrangian torus: the two signs of ξθ\xi_\theta join at the two turning latitudes. Its projection is

π(Λμ)={arcsin⁡μ≤θ≤π−arcsin⁡μ}.\pi(\Lambda_\mu)=\{\arcsin\mu\le\theta\le\pi-\arcsin\mu\}.

Meridian restriction

Let II be a nontrivial compact interval strictly inside (arcsin⁡μ,π−arcsin⁡μ)(\arcsin\mu,\pi-\arcsin\mu), and let HI={(θ,0):θ∈I}H_I=\{(\theta,0):\theta\in I\}. For the local defining function ρ=φ\rho=\varphi, one has

Hp2ρ=1,rHI(θ,0,ξθ,μ)=(θ,ξθ).H_{p_2}\rho=1,\qquad r_{H_I}(\theta,0,\xi_\theta,\mu)=(\theta,\xi_\theta).

The latter map is injective on the joint fiber. Therefore Theorem 1.2 gives

∥Yℓmℓ∣HI∥L2(HI)≥cI>0(31)\left\|Y_\ell^{m_\ell}\big|_{H_I}\right\|_{L^2(H_I)}\ge c_I>0 \tag*{(31)}

for all sufficiently large ℓ\ell. The fiber-rank condition also holds over HIH_I, so the local pointwise bound is uniform there. These are nonturning-region estimates; the different behavior at the turning latitudes is described, for example, in [1].

For completeness, the real meridian restriction has order hℓ−1h_\ell^{-1} sign changes on II. Set wℓ(θ)=sin⁡θ Fℓ,mℓ(θ)w_\ell(\theta)=\sqrt{\sin\theta}\,F_{\ell,m_\ell}(\theta). The associated Legendre equation gives the exact equation

wℓ′′+1hℓ2(1−(hℓmℓ)2−hℓ2/4sin⁡2θ)wℓ=0.(32)w_\ell''+\frac{1}{h_\ell^2}\left(1-\frac{(h_\ell m_\ell)^2-h_\ell^2/4}{\sin^2\theta}\right)w_\ell=0. \tag*{(32)}

The coefficient in parentheses is bounded above and below by positive constants on II for all large ℓ\ell. Sturm comparison with constant positive coefficients therefore bounds the number of zeros in II above and below by constant multiples of hℓ−1h_\ell^{-1}, up to additive constants. Every such zero is simple, by uniqueness for the second-order equation, and is consequently a sign change. This conclusion follows directly from the separated equation and does not require admissibility in the sense of (26).

Cancellation on the equator

Consider instead Heq={θ=π/2}H_{\mathrm{eq}}=\{\theta=\pi/2\} and ρ=θ−π/2\rho=\theta-\pi/2. On the same connected regular joint torus,

Hp1ρ=ξθ∣ξ∣g=±1−μ2≠0.H_{p_1}\rho=\frac{\xi_\theta}{|\xi|_g}=\pm\sqrt{1-\mu^2}\ne0.

Thus a commuting Hamiltonian already crosses the equator at every characteristic lift. If crossing by the second operator is desired, one can equally use

P~2(h)=P2(h)+2P1(h),Hp~2ρ=±21−μ2≠0.\widetilde P_2(h)=P_2(h)+2P_1(h),\qquad H_{\widetilde p_2}\rho=\pm2\sqrt{1-\mu^2}\ne0.

This replacement preserves the joint fiber, with limiting energy (1,μ+2)(1,\mu+2).

Now choose mℓm_\ell to be an integer nearest to μ/hℓ\mu/h_\ell among the integers satisfying ℓ+mℓ\ell+m_\ell odd. The distance to that parity class is at most one, so (29) remains valid. The identity

Pℓm(−t)=(−1)ℓ+mPℓm(t)P_\ell^{m}(-t)=(-1)^{\ell+m}P_\ell^m(t)

implies

Yℓmℓ(π/2,φ)=0,∥Yℓmℓ∣Heq∥L2(Heq)=0.(33)Y_\ell^{m_\ell}(\pi/2,\varphi)=0,\qquad\left\|Y_\ell^{m_\ell}\big|_{H_{\mathrm{eq}}}\right\|_{L^2(H_{\mathrm{eq}})}=0. \tag*{(33)}

The failure of branch isolation is explicit:

rHeq(π2,φ,±1−μ2,μ)=(φ,μ).r_{H_{\mathrm{eq}}}\left(\frac{\pi}{2},\varphi,\pm\sqrt{1-\mu^2},\mu\right)=(\varphi,\mu).

Both normal branches have the same tangential covector, and their restrictions cancel exactly. Hence a noncharacteristic crossing cannot replace branch isolation in Theorem 1.2. This obstruction occurs on a single connected regular torus, independently of the additional issue of distributing mass among several components of a joint fiber.

Finite joint fibers and rigidity of persistent nodal curves

The local Cauchy estimate also applies when the full regular joint fiber has several connected components. This extension is useful for Liouville metrics, whose rotating joint fibers consist of four tori. We first prove the extension stated in Theorem 1.3, and then use it to classify the curves on which exact joint eigenfunctions can vanish persistently. The lower bounds in this section follow from the microlocal argument of the paper; separation of variables will be used only for a uniform upper bound.

Detecting every component of a regular fiber

Proof of Theorem 1.3. Write the compact regular fiber as

ΛE=⨆ν=1NΛν.\Lambda_E = \bigsqcup_{\nu=1}^{N} \Lambda_\nu.

Each component is a compact Liouville torus. Let μν\mu_\nu be its normalized Liouville measure. Joint ellipticity and the normalization of uhu_h imply that every semiclassical defect measure is a probability measure supported on ΛE\Lambda_E. Invariance under all the commuting Hamiltonian flows, which span the tangent space of each component, gives

μ=∑ν=1Ncνμν,cν≥0,∑ν=1Ncν=1.(34)\mu= \sum_{\nu=1}^{N} c_\nu\mu_\nu,\qquad c_\nu\ge0,\qquad\sum_{\nu=1}^{N} c_\nu= 1. \tag*{(34)}

The weights need not be known and may depend on the subsequence.

For each ν\nu, choose the transverse point and tangential window supplied by the theorem. Shrink the window and the transverse patch so that their closures lie in the original neighborhoods. The isolation hypothesis concerns preimages in the entire fiber: no other sheet, whether in Λν\Lambda_\nu or in another component, has a tangential covector in this smaller window. Applying Proposition 2.2, with the transverse Hamiltonian chosen for this component, gives a cutoff χν\chi_\nu in a small flow tube such that

∥Op⁡hw(χν)uh∥L2(M)≤Cν∥uh∣H∥L2(H)+o(1).(35)\left\|\operatorname{Op}^{w}_{h}(\chi_\nu)u_h\right\|_{L^2(M)} \leq C_\nu\left\|u_h|_H\right\|_{L^2(H)} + o(1). \tag*{(35)}

We choose the support disjoint from the other components and the cutoff nonzero on an open subset of Λν\Lambda_\nu. Consequently,

dν:=∫Λν∣χν∣2 dμν>0.d_\nu:= \int_{\Lambda_\nu} |\chi_\nu|^2\,d\mu_\nu> 0.

All cutoffs and constants can be chosen for nearby joint energies, as in the local estimate. In particular, convergence E(h)→EE(h) \to E suffices here.

If the asserted restriction lower bound failed, a sequence with h→0h \to0 and ∥uh∣H∥L2(H)→0\left\|u_h|_H\right\|_{L^2(H)} \to0 would have a subsequence with defect measure (6.1). Equation (35) would then imply, for every ν\nu,

0=lim⁡h→0∥Op⁡hw(χν)uh∥L2(M)2=cνdν.0 = \lim_{h\to0}\left\|\operatorname{Op}^{w}_{h}(\chi_\nu)u_h\right\|_{L^2(M)}^2 = c_\nu d_\nu.

Since dν>0d_\nu> 0, every cνc_\nu would vanish, contradicting their sum being one. This proves the theorem without a condition selecting a component or prescribing its mass. □\square

Corollary 6.1 (Uniformity in the joint energy). Suppose that the geometric hypotheses of Theorem 1.3 hold at EE. There are a neighborhood UU of EE and constants c,h0>0c,h_0 > 0 such that every exact joint eigenfunction satisfying

Pj(h)u=Ej′u(1≤j≤n),E′∈U,0<h<h0,P_j(h)u = E'_j u \quad(1 \leq j \leq n),\qquad E' \in U,\qquad0 < h < h_0,

obeys

∥u∣H∥L2(H)≥c∥u∥L2(M).(36)\lVert u|_{H}\rVert_{L^{2}(H)} \ge c\lVert u\rVert_{L^{2}(M)}. \tag*{(36)}

The same conclusion holds with HH replaced by a fixed small patch containing the transverse points used in the hypothesis.

Proof. Otherwise, choosing successively smaller energy neighborhoods and smaller upper bounds on hh, one could find normalized exact joint eigenfunctions with

hk→0,Ek′→E,∥uk∣H∥L2(H)→0.h_k \to0,\qquad E'_k \to E,\qquad\lVert u_k|_{H}\rVert_{L^{2}(H)} \to0.

The proof just given applies to this sequence and yields a contradiction. For a patch, choose all the local Cauchy cutoffs over that patch. Uniformity of their estimates follows by factoring at the actual nearby energy: the transverse derivative remains bounded away from zero, and the tangential windows continue to exclude the other sheets after shrinking. No estimate of Ek′−EE'_k-E by O(hk)O(h_k) is required. □

Remark 6.2. A transverse isolated patch on only one component does not suffice for a lower bound on arbitrary normalized joint eigenfunctions: that component can have weight zero in (34). Requiring such a patch on every component is what removes the need to determine the weights. The example in Theorem 1.1 will fail precisely at this point: the two components carrying the cancelling waves have the same tangential covectors along the common nodal curve.

The four branches of a Liouville metric

Let M=T2=(R/2πZ)2M=\mathbb{T}^{2}=(\mathbb{R}/2\pi\mathbb{Z})^{2} and let

g=F(x,y)(dx2+dy2),F(x,y)=a(x)+b(y)>0,(37)g=F(x,y)(dx^{2}+dy^{2}),\qquad F(x,y)=a(x)+b(y)>0, \tag*{(37)}

where a,ba,b are real smooth periodic functions. The commuting operators

L=−Δg=−F−1(∂x2+∂y2),Q=F−1(−b(y)∂x2+a(x)∂y2)(38)L=-\Delta_{g}=-F^{-1}(\partial_{x}^{2}+\partial_{y}^{2}),\qquad Q=F^{-1}(-b(y)\partial_{x}^{2}+a(x)\partial_{y}^{2}) \tag*{(38)}

are self-adjoint on their natural domains in L2(M,F dx dy)L^{2}(M,F\,dx\,dy). Their semiclassical principal symbols are

ℓ=ξx2+ξy2F,q=bξx2−aξy2F.\ell=\frac{\xi_{x}^{2}+\xi_{y}^{2}}{F},\qquad q=\frac{b\xi_{x}^{2}-a\xi_{y}^{2}}{F}.

For h>0h>0 and κ∈R\kappa\in\mathbb{R}, define the exact joint eigenspace

Eh,κ={u∈C∞(M;C):h2Lu=u, h2Qu=κu}.(39)\mathcal{E}_{h,\kappa}=\{u\in C^{\infty}(M;\mathbb{C}):h^{2}Lu=u,\ h^{2}Qu=\kappa u\}. \tag*{(39)}

We work in the rotating interval

R=(−min⁡a,min⁡b),(40)\mathcal{R}=(-\min a,\min b), \tag*{(40)}

which is nonempty because min⁡a+min⁡b>0\min a+\min b>0. For κ∈R\kappa\in\mathcal{R}, put

Aκ(x)=a(x)+κ,Bκ(y)=b(y)−κ.(41)A_{\kappa}(x)=\sqrt{a(x)+\kappa},\qquad B_{\kappa}(y)=\sqrt{b(y)-\kappa}. \tag*{(41)}

The full joint fiber {ℓ=1,q=κ}\{\ell=1,q=\kappa\} is the disjoint union of the four graph tori

Λστ(κ)={(x,y;σAκ(x),τBκ(y))},σ,τ∈{+1,−1}.(42)\Lambda_{\sigma\tau}(\kappa)=\{(x,y;\sigma A_{\kappa}(x),\tau B_{\kappa}(y))\},\qquad\sigma,\tau\in\{+1,-1\}. \tag*{(42)}

They are regular, since

det⁡∂(ℓ,q)∂(ξx,ξy)=−4ξxξyF≠0.\det\frac{\partial(\ell,q)}{\partial(\xi_x,\xi_y)}=-\frac{4\xi_x\xi_y}{F}\ne0.

For completeness, these second-order differential operators fit the first-order framework of the main theorem by setting

P1(h)=(1+h2L)1/2,P2(h)=h2Q(1+h2L)−1/2.(43)P_1(h)=(1+h^{2}L)^{1/2},\qquad P_2(h)=h^{2}Q(1+h^{2}L)^{-1/2}. \tag*{(43)}

They commute and are self-adjoint, with real principal symbols

p1=1+ℓ,p2=q1+ℓ.p_1=\sqrt{1+\ell},\qquad p_2=\frac{q}{\sqrt{1+\ell}}.

The symbol map (ℓ,q)↦(p1,p2)(\ell,q)\mapsto(p_1,p_2) has invertible differential. The space (39) is therefore exactly the joint eigenspace at energy (2,κ/2)(\sqrt{2},\kappa/\sqrt{2}) for (43), and its classical fiber is (42).

Let γ\gamma be a smooth regular curve. A point of γ\gamma with nonzero tangent v=(vx,vy)v=(v_x,v_y) is called separating at energy κ\kappa if

vxvy≠0,Aκ(x)2vx2≠Bκ(y)2vy2.(44)v_xv_y\ne0,\qquad A_\kappa(x)^2v_x^2\ne B_\kappa(y)^2v_y^2. \tag*{(44)}

This is independent of the regular parametrization. Indeed, the four tangential covectors in (42) are represented by

σAκvx+τBκvy,σ,τ∈{+1,−1},(45)\sigma A_\kappa v_x+\tau B_\kappa v_y,\qquad\sigma,\tau\in\{+1,-1\}, \tag*{(45)}

and they are pairwise distinct exactly when (44) holds.

Lemma 6.3 (Restriction near a separating point). Suppose that (44) holds at a point of γ\gamma at energy κ0∈R\kappa_0\in\mathbb{R}. There are a fixed compact subarc γ0⊂γ\gamma_0\subset\gamma, an open interval J⋐RJ\Subset\mathbb{R} containing κ0\kappa_0, and constants c,h0>0c,h_0>0 such that

∥u∣γ0∥L2(γ0)≥c∥u∥L2(M),u∈Eh,κ,κ∈J,0<h<h0.(46)\|u|_{\gamma_0}\|_{L^2(\gamma_0)}\ge c\|u\|_{L^2(M)},\qquad u\in E_{h,\kappa},\quad\kappa\in J,\quad0<h<h_0. \tag*{(46)}

Curve norms are taken with respect to metric arclength.

Proof. At the specified point, choose four disjoint neighborhoods of the tangential covectors (45). Shrinking the base arc makes each neighborhood isolate its sheet from the entire four-component fiber. On each branch, the vectors ∂ξp1,∂ξp2\partial_{\xi}p_1,\partial_{\xi}p_2 span the base tangent space. If ρ\rho is a local defining function for the curve, at least one of Hp1ρ,Hp2ρH_{p_1}\rho,H_{p_2}\rho is therefore nonzero. The transverse Hamiltonian may depend on the branch, which is permitted in Theorem 1.3. Thus every component has a transverse isolated patch over the same short arc. Apply Corollary 6.1 to the system (43). This proves the estimate for all nearby κ\kappa. □

We will also use the following elementary upper bound. Its proof is included to keep the subsequent stability assertion independent of additional pointwise estimates.

Lemma 6.4 (Uniform upper bound in a rotating window). For every compact K⋐RK\Subset\mathbb{R}, there are CK,h0>0C_K,h_0>0 such that

∥u∥L∞(M)≤CK∥u∥L2(M),u∈Eh,κ,κ∈K,0<h<h0.(47)\|u\|_{L^\infty(M)}\le C_K\|u\|_{L^2(M)},\qquad u\in E_{h,\kappa},\quad\kappa\in K,\quad0<h<h_0. \tag*{(47)}

Proof. The identities −∂x2=aL+Q-\partial_x^2=aL+Q and −∂y2=bL−Q-\partial_y^2=bL-Q give

−h2uxx=Aκ2u,−h2uyy=Bκ2u.(48)-h^2u_{xx}=A_\kappa^2u,\qquad-h^2u_{yy}=B_\kappa^2u. \tag*{(48)}

On a fixed coordinate period, each scalar equation has exact independent solutions with uniform expansions

fσ(x)=Aκ(x)−1/2eiσSx(x)/h+O(h),gτ(y)=Bκ(y)−1/2eiτSy(y)/h+O(h),f_\sigma(x)=A_\kappa(x)^{-1/2}e^{i\sigma S_x(x)/h}+O(h),\qquad g_\tau(y)=B_\kappa(y)^{-1/2}e^{i\tau S_y(y)/h}+O(h),

where Sx′=AκS_x'=A_\kappa, Sy′=BκS_y'=B_\kappa, and the errors are in C0C^0, uniformly for κ∈K\kappa\in K. To verify this elementary WKB estimate, the substitution t=Sx(x)t=S_x(x), f=Aκ−1/2wf=A_\kappa^{-1/2}w gives

wtt+h−2w=rκ(t)w,rκ=Aκ′′2Aκ3−3(Aκ′)24Aκ4.w_{tt}+h^{-2}w=r_\kappa(t)w,\qquad r_\kappa=\frac{A_\kappa''}{2A_\kappa^3}-\frac{3(A_\kappa')^2}{4A_\kappa^4}.

The solution with initial data w(0)=1w(0)=1, wt(0)=iσ/hw_t(0)=i\sigma/h satisfies

w(t)=eiσt/h+h∫0tsin⁡((t−s)/h)rκ(s)w(s) ds.w(t)=e^{i\sigma t/h}+h\int_0^t\sin((t-s)/h)r_\kappa(s)w(s)\,ds.

Boundedness of rκr_\kappa and Gronwall’s inequality give the asserted expansion; the yy equation is identical in form.

Every solution of both equations in (48) has an exact representation

u(x,y)=∑σ,τ=±1cστfσ(x)gτ(y).u(x,y)=\sum_{\sigma,\tau=\pm1}c_{\sigma\tau}f_\sigma(x)g_\tau(y).

Indeed, expand first in the xx basis; its two coefficient functions must solve the yy equation. The scalar basis functions themselves need not be periodic. Their Gram matrices on a full coordinate period have positive diagonal entries bounded away from zero and off-diagonal entries O(h)O(h): integration by parts applies to the leading off-diagonal phases because Aκ,BκA_\kappa,B_\kappa are bounded away

from zero. Hence the tensor-product Gram matrix, together with the positive bounds on FF, gives

∑σ,τ∣cστ∣2≤CK∥u∥L2(M)2.\sum_{\sigma,\tau} |c_{\sigma\tau}|^2 \le C_K \|u\|_{L^2(M)}^2.

The basis functions are uniformly bounded, so this coefficient estimate proves (47). □

Persistent curves and stable restriction

Define the closed nowhere-zero one-forms

ακ±=Aκ(x) dx±Bκ(y) dy.(49)\alpha_\kappa^\pm= A_\kappa(x)\,dx \pm B_\kappa(y)\,dy. \tag*{(49)}

A phase front is a leaf of ker⁡ακ+\ker\alpha_\kappa^+ or ker⁡ακ−\ker\alpha_\kappa^-. Locally, it is a level curve of Sx±SyS_x \pm S_y; these primitives need not be single-valued on the torus. All curves below are fixed independently of hh.

Proposition 6.5 (Rigidity of persistent nodal curves). Let γ\gamma be a connected smooth embedded regular curve. Suppose that normalized exact joint eigenfunctions uj∈Ehj,κju_j \in\mathcal{E}_{h_j,\kappa_j} satisfy

hj→0,κj→κ∈R,uj∣γ≡0.h_j \to0,\qquad\kappa_j \to\kappa\in\mathbb{R},\qquad u_j|_\gamma\equiv0.

Then γ\gamma is contained in a leaf of exactly one of the four foliations

ker⁡dx,ker⁡dy,ker⁡ακ+,ker⁡ακ−.(50)\ker dx,\qquad\ker dy,\qquad\ker\alpha_\kappa^+,\qquad\ker\alpha_\kappa^-. \tag*{(50)}

In particular, it is a coordinate curve or a phase front of the limiting joint energy. If γ\gamma is closed, it is a closed leaf.

Proof. A separating point at the limiting energy would, by Lemma 6.3, give a fixed subarc on which the restriction norms of uju_j are bounded below for all large jj. This contradicts their vanishing. Thus, at every point of a regular parametrization γ(t)=(x(t),y(t))\gamma(t)=(x(t),y(t)),

x′y′=0orAκ(x)2x′2=Bκ(y)2y′2.x'y'=0\quad\text{or}\quad A_\kappa(x)^2x'^2=B_\kappa(y)^2y'^2.

These are precisely the four tangent lines in (50). Since Aκ,Bκ>0A_\kappa,B_\kappa>0, the four lines are distinct at every point. The continuous tangent line of γ\gamma therefore selects a locally constant label, which is constant because γ\gamma is connected. Uniqueness for integral curves of the resulting smooth line field places γ\gamma in one leaf. A closed embedded curve tangent to that field is itself a closed leaf. □

Corollary 6.6 (A contractible closed curve cannot persist). Let γ\gamma be a fixed smooth embedded contractible closed curve and let K⋐RK\Subset\mathbb{R} be compact. There are c,h0>0c,h_0>0 such that

∥u∣γ∥L2(γ)≥c∥u∥L2(M),u∈Eh,κ,κ∈K,0<h<h0.(51)\|u|_\gamma\|_{L^2(\gamma)} \ge c\|u\|_{L^2(M)},\qquad u\in\mathcal{E}_{h,\kappa},\quad\kappa\in K,\quad0<h<h_0. \tag*{(51)}

In particular, γ\gamma cannot be a persistent nodal component in this rotating window.

Proof. Fix κ∈K\kappa\in K. If γ\gamma had no separating point, the line-field argument in the preceding proof would make it a closed leaf of one of (50). Coordinate leaves are noncontractible. On a phase-front leaf,

Aκ(x)x′±Bκ(y)y′=0.A_\kappa(x)x' \pm B_\kappa(y)y'=0.

Regularity implies that neither x′x' nor y′y' vanishes; both have constant sign. For a parametrization by R/Z\mathbb{R}/\mathbb{Z} and a lift (x~,y~)(\widetilde{x},\widetilde{y}), its winding numbers therefore satisfy

2πm=∫01x~′(t) dt≠0,2πn=∫01y~′(t) dt≠0.2\pi m=\int_0^1 \widetilde{x}'(t)\,dt\ne0,\qquad2\pi n=\int_0^1 \widetilde{y}'(t)\,dt\ne0.

Such a leaf is also noncontractible, a contradiction. Hence a separating point exists at every κ∈K\kappa\in K. Apply Lemma 6.3 at these points and take a finite subcover of their energy neighborhoods. The minimum of the finitely many lower-bound constants and frequency thresholds gives (51). □

Corollary 6.7 (Action resonance for closed noncoordinate curves). Under the hypotheses of Proposition 6.5, suppose that γ\gamma is closed and noncoordinate. Define the positive actions

Ix(κ)=∫02πAκ(x) dx,Iy(κ)=∫02πBκ(y) dy.(52)I_x(\kappa)=\int_0^{2\pi} A_\kappa(x)\,dx,\qquad I_y(\kappa)=\int_0^{2\pi} B_\kappa(y)\,dy. \tag*{(52)}

The winding numbers (m,n)(m,n) of γ\gamma satisfy mn≠0mn \ne0 and, for one fixed sign,

mIx(κ)±nIy(κ)=0.(53)mI_x(\kappa) \pm nI_y(\kappa) = 0. \tag*{(53)}

Consequently Ix(κ)/Iy(κ)=∣n/m∣∈QI_x(\kappa)/I_y(\kappa) = \lvert n/m\rvert\in\mathbb{Q}. If this action ratio is irrational, every smooth embedded closed persistent nodal curve is a coordinate circle.

Proof. By Proposition 6.5, the curve is a leaf of ker⁡ακ±\ker\alpha_\kappa^\pm for one sign. The preceding proof shows that mn≠0mn \ne0. Integrating its annihilating one-form and using periodicity gives

0=∫γακ±=mIx(κ)±nIy(κ),0 = \int_\gamma\alpha_\kappa^\pm= mI_x(\kappa) \pm nI_y(\kappa),

which proves the assertions. □

Remark 6.8. The phase coordinates

X(x)=∫0xAκ(t) dt,Y(y)=∫0yBκ(t) dtX(x) = \int_0^x A_\kappa(t)\,dt,\qquad Y(y) = \int_0^y B_\kappa(t)\,dt

identify the torus with a rectangular torus of side lengths Ix(κ)I_x(\kappa) and Iy(κ)I_y(\kappa). Its phase fronts become X±Y=constantX \pm Y = \text{constant}. Thus the resonance is exactly the condition for these leaves to close. It is necessary for persistence of a closed noncoordinate curve, but does not itself construct an exact eigenfunction vanishing on a chosen leaf.

Proposition 6.9 (Stable uniqueness from restriction). Let γ\gamma be a fixed compact smooth embedded regular curve, possibly with boundary, and let K⊂RK \subset\mathbb{R}. Suppose that γ\gamma has a separating point for every κ∈K\kappa\in K. There are c,C,h0>0c,C,h_0 > 0 such that

c∥u∥L2(M)≤∥u∣γ∥L2(γ)≤C∥u∥L2(M),u∈Eh,κ,κ∈K,0<h<h0.(54)c\lVert u\rVert_{L^2(M)} \le\lVert u|_\gamma\rVert_{L^2(\gamma)} \le C\lVert u\rVert_{L^2(M)},\qquad u \in E_{h,\kappa},\quad\kappa\in K,\quad0 < h < h_0. \tag*{(54)}

Restriction is therefore injective on each such joint eigenspace, with a uniformly bounded inverse on its range. In particular, for u,vu,v in the same Eh,κE_{h,\kappa},

∥u−v∥L2(M)≤c−1∥(u−v)∣γ∥L2(γ).(55)\lVert u-v\rVert_{L^2(M)} \le c^{-1}\lVert(u-v)|_\gamma\rVert_{L^2(\gamma)}. \tag*{(55)}

Proof. The finite-cover argument using Lemma 6.3 proves the lower estimate uniformly on KK. Lemma 6.4 gives the upper estimate because

∥u∣γ∥L2(γ)≤length⁡g(γ)1/2∥u∥L∞(M).\lVert u|_\gamma\rVert_{L^2(\gamma)} \le\operatorname{length}_g(\gamma)^{1/2}\lVert u\rVert_{L^\infty(M)}.

Finally, u−vu-v belongs to the same exact joint eigenspace, so the lower estimate applied to this difference gives (55). □

These statements allow every vector in a joint eigenspace, including linear combinations when the joint eigenvalue has multiplicity. They concern exact joint eigenfunctions in a regular rotating window; no assertion is made for arbitrary Laplace eigenfunctions, or for differences of functions with different joint eigenvalues. The construction in the next section realizes one of the noncoordinate phase fronts permitted by Proposition 6.5, and shows that this exceptional foliation need not consist of geodesics.

A nongeodesic common nodal circle

We now construct the exceptional noncoordinate phase front allowed by § 6. The existence argument is exact and independent of the restriction lower bound. The final calculation identifies the transverse branches whose traces cannot be separated.

The Floquet construction

Let f>0f>0 be a nonconstant smooth function with period ℓ=2π/N\ell=2\pi/N, where N≥3N\ge3 is an integer. Let r,sr,s be relatively prime positive integers with r≠sr\ne s. On T2\mathbb{T}^{2}, set

a(x)=r2f(rx),b(y)=s2f(sy),F=a+b,g=F(dx2+dy2),(56)a(x)=r^{2}f(rx),\qquad b(y)=s^{2}f(sy),\qquad F=a+b,\qquad g=F(dx^{2}+dy^{2}), \tag*{(56)}

and define

L=−Δg=−F−1(∂x2+∂y2),Q=F−1(−b∂x2+a∂y2).(57)L=-\Delta_{g}=-F^{-1}(\partial_{x}^{2}+\partial_{y}^{2}),\qquad Q=F^{-1}(-b\partial_{x}^{2}+a\partial_{y}^{2}). \tag*{(57)}

The operators commute and are formally self-adjoint for dVg=F dx dydV_{g}=F\,dx\,dy.

Proposition 7.1 (Matching separated equations). For the metric (56), there exist real normalized joint eigenfunctions uju_{j} with

Luj=λjuj,Quj=0,λj→∞,Lu_{j}=\lambda_{j}u_{j},\qquad Qu_{j}=0,\qquad\lambda_{j}\to\infty,

at distinct eigenvalues, for which

γr,s(t)=(st,rt)(mod2π)(58)\gamma_{r,s}(t)=(st,rt)\pmod{2\pi} \tag*{(58)}

is a regular connected nodal component. This embedded circle has winding (s,r)(s,r) and signed geodesic curvature

kg(t)=(r2−s2)f′(rst)2(r2+s2)f(rst)3/2.(59)k_{g}(t)=\frac{(r^{2}-s^{2})f'(rst)}{2(r^{2}+s^{2})f(rst)^{3/2}}. \tag*{(59)}

It is not a geodesic, and gg has no nonzero Killing field. If ff is real analytic, so are the metric and the eigenfunctions.

Proof. The twisted one-dimensional problem. Put ω=e2πi/N\omega=e^{2\pi i/N} and consider the operator

Tϕ=−f−1ϕ′′,D(T)={ϕ∈H2([0,ℓ]):ϕ(ℓ)=ωϕ(0), ϕ′(ℓ)=ωϕ′(0)}T\phi=-f^{-1}\phi'',\qquad D(T)=\{\phi\in H^{2}([0,\ell]):\phi(\ell)=\omega\phi(0),\ \phi'(\ell)=\omega\phi'(0)\}

on L2([0,ℓ],f(t) dt)L^{2}([0,\ell],f(t)\,dt). It is associated with the closed quadratic form ∫0ℓ∣ϕ′∣2 dt\int_{0}^{\ell}|\phi'|^{2}\,dt on the twisted H1H^{1} domain. Since ∣ω∣=1|\omega|=1, the boundary terms cancel; the operator is self-adjoint. Compactness of the form-domain embedding gives compact resolvent. The only constant function obeying the boundary condition is zero, so TT is strictly positive. Choose eigenpairs at distinct eigenvalues

−ϕj′′=λjfϕj,λj→∞.(60)-\phi_{j}''=\lambda_{j}f\phi_{j},\qquad\lambda_{j}\to\infty. \tag*{(60)}

Extend by ϕj(t+ℓ)=ωϕj(t)\phi_{j}(t+\ell)=\omega\phi_{j}(t). The matching values and derivatives, together with the differential equation, give a smooth global solution. It is 2π2\pi-periodic because ωN=1\omega^{N}=1.

The multiplier is nonreal, so ϕj\phi_{j} and ϕj‾\overline{\phi_{j}} are independent solutions: otherwise they would have the same multiplier, whereas ω≠ω‾\omega\ne\overline{\omega}. Their nonzero Wronskian gives a constant nonzero current

Jj=Im⁡(ϕj′(t)ϕj(t)‾)≠0.(61)J_{j}=\operatorname{Im}(\phi_{j}'(t)\overline{\phi_{j}(t)})\ne0. \tag*{(61)}

In particular, ϕj\phi_{j} never vanishes.

Exact joint eigenfunctions and the common component. Define the real, doubly periodic function

Uj(x,y)=Im⁡(ϕj(rx)ϕj(sy)‾).(62)U_{j}(x,y)=\operatorname{Im}(\phi_{j}(rx)\overline{\phi_{j}(sy)}). \tag*{(62)}

Equation (60) yields

−∂x2Uj=λjr2f(rx)Uj=λjaUj,−∂y2Uj=λjs2f(sy)Uj=λjbUj.-\partial_{x}^{2}U_{j}=\lambda_{j}r^{2}f(rx)U_{j}=\lambda_{j}aU_{j},\qquad-\partial_{y}^{2}U_{j}=\lambda_{j}s^{2}f(sy)U_{j}=\lambda_{j}bU_{j}.

Consequently,

LUj=λjUj,QUj=0.(63)LU_{j}=\lambda_{j}U_{j},\qquad QU_{j}=0. \tag*{(63)}

Along (58), both arguments in (62) equal rstrst, and hence

Uj∣γr,s=0,dUj∣γr,s=Jj(r dx−s dy)≠0.(64)U_{j}|_{\gamma_{r,s}}=0,\qquad dU_{j}|_{\gamma_{r,s}}=J_{j}(r\,dx-s\,dy)\ne0. \tag*{(64)}

Coprimality of r,sr,s makes (58) embedded. The implicit function theorem and compactness give a tubular neighborhood in which its nodal set is exactly this circle. Thus it is a regular connected component of the full nodal set. Normalize UjU_{j} in L2(M,g)L^{2}(M,g) to obtain uju_{j}. Distinct eigenvalues make the normalized functions orthogonal. Analyticity follows from the analytic differential equation when ff is analytic.

Geodesic curvature. Write D=r2+s2D=r^2+s^2 and orient the curve by increasing tt. Its Euclidean unit tangent and left normal are

T0=(s,r)D,n0=(−r,s)D.T_0=\frac{(s,r)}{\sqrt{D}},\qquad n_0=\frac{(-r,s)}{\sqrt{D}}.

For a conformal metric g=F(dx2+dy2)g=F(dx^2+dy^2), the signed geodesic curvature is

kg=F−1/2(k0−12∂n0log⁡F).k_g=F^{-1/2}\left(k_0-\frac{1}{2}\partial_{n_0}\log F\right).

Here k0=0k_0=0, while on the curve

F=Df(rst),∂n0F=(s2−r2)Df′(rst).F=Df(rst),\qquad\partial_{n_0}F=(s^2-r^2)\sqrt{D}f'(rst).

Substitution proves (59). Since r≠sr\ne s and ff is nonconstant, this curvature is not identically zero.

Absence of Killing fields. If X=P∂x+R∂yX=P\partial_x+R\partial_y is a Killing field, the trace-free parts of LXg=0\mathcal{L}_Xg=0 give

Px=Ry,Py=−Rx.P_x=R_y,\qquad P_y=-R_x.

Thus P,RP,R are harmonic periodic functions and hence constants. The remaining equation becomes

Pr3f′(rx)+Rs3f′(sy)=0for all (x,y).Pr^3f'(rx)+Rs^3f'(sy)=0\qquad\text{for all }(x,y).

Integrating first in yy and then in xx, and using that f′f' is not identically zero, gives P=R=0P=R=0. □

Proof of Theorem 1.1. Apply Proposition 7.1 with N=3N=3, r=1r=1, s=2s=2, and f(t)=2+εcos⁡3tf(t)=2+\varepsilon\cos3t. Positivity follows from 0<ε<20<\varepsilon<2. Formula (59) becomes (3). □

Why the choices matter. The small period makes a nonreal Floquet multiplier into a periodic solution on the original circle. The factors r2,s2r^2,s^2 exactly compensate for differentiating ϕj(rx)\phi_j(rx) and ϕj(sy)\phi_j(sy). Matching their arguments along rx=syrx=sy produces cancellation for every jj, while r≠sr\ne s removes geodesicity. Finite translation symmetries remain, but there is no nonzero Killing field; the persistence mechanism is the matching of scalar equations. Since the common curve is not geodesic, it cannot be a curve fixed pointwise by a nontrivial isometric reflection.

Remark 7.2 (A simpler example with a Killing field). The broad existence phenomenon also has a one-variable version. For g0=f(x)(dx2+dy2)g_0=f(x)(dx^2+dy^2) and any x0x_0 with f′(x0)≠0f'(x_0)\ne0, rephase each of the same Floquet solutions so that ϕj(x0)>0\phi_j(x_0)>0 is real. Then vj(x,y)=Im⁡ϕj(x)v_j(x,y)=\operatorname{Im}\phi_j(x) satisfies −Δg0vj=λjvj-\Delta_{g_0}v_j=\lambda_jv_j, and x=x0x=x_0 is a regular nodal circle because vj′(x0)=Jj/ϕj(x0)≠0v_j'(x_0)=J_j/\phi_j(x_0)\ne0. Its geodesic curvature has magnitude ∣f′(x0)∣/(2f(x0)3/2)\lvert f'(x_0)\rvert/(2f(x_0)^{3/2}). This direct comparison explains why the additional features of Proposition 7.1 matter: it removes the Killing field, makes the curve noncoordinate, and places the joint sequence in a rotating energy window.

Action resonance and transverse branch cancellation

For (56), the energy κ=0\kappa=0 lies in the rotating interval (−r2min⁡f,s2min⁡f)(-r^2\min f,s^2\min f). With hj=λj−1/2h_j=\lambda_j^{-1/2}, the constructed sequence lies in Ehj,0E_{h_j,0}. Its full joint fiber has the four graph components (42), so the finite-component formulation of Section 6 is the appropriate one.

Along γr,s(t)=(st,rt)\gamma_{r,s}(t)=(st,rt), the one-form

α0−=rf(rx) dx−sf(sy) dy\alpha_0^-=r\sqrt{f(rx)}\,dx-s\sqrt{f(sy)}\,dy

annihilates the tangent. Thus the common nodal circle is a phase front of the kind allowed by Proposition 6.5. If

I=∫02πf(t) dt,I=\int_0^{2\pi}\sqrt{f(t)}\,dt,

periodicity and change of variables give

Ix(0)=rI,Iy(0)=sI,sIx(0)−rIy(0)=0.I_x(0)=rI,\qquad I_y(0)=sI,\qquad sI_x(0)-rI_y(0)=0.

The winding (s,r)(s,r) therefore satisfies the resonance of Corollary 6.7. Resonance closes the phase front; the Floquet construction supplies exact eigenfunctions vanishing on it.

The failure of the restriction hypothesis is particularly transparent in Theorem 1.1. On γ(t)=(2t,t)\gamma(t)=(2t,t), put A=2+εcos⁡6tA=\sqrt{2}+\varepsilon\cos6t, so B=2AB=2A. The four covectors restricted to the tangent (2,1)(2,1) give

(σA,τB)(2,1)=2A(σ+τ)∈{4A,0,0,−4A}.(65)(\sigma A,\tau B)(2,1)=2A(\sigma+\tau)\in\{4A,0,0,-4A\}. \tag*{(65)}

The mixed branches are distinct in T∗MT^*M, but both have zero tangential covector. They cannot be separated by a pseudodifferential cutoff on the curve. For the local defining function ρ=x−2y\rho=x-2y and the Laplace symbol pL=(ξx2+ξy2)/Fεp_L=(\xi_x^2+\xi_y^2)/F_\varepsilon, these same branches satisfy

HpLρ∣Λ+−(0)=2A,HpLρ∣Λ−+(0)=−2A.(66)H_{p_L}\rho\big|_{\Lambda_{+-}(0)}=\frac{2}{A},\qquad H_{p_L}\rho\big|_{\Lambda_{-+}(0)}=-\frac{2}{A}. \tag*{(66)}

Both flows cross the curve transversely. The obstruction is therefore the coincidence of tangential covectors, despite nonzero geodesic curvature and transverse Hamiltonian crossing. The two same-sign branches remain individually observable; the next proposition uses the Cauchy estimate to show that they carry no limiting mass in this sequence.

The distinction from § 5 should be kept explicit. The spherical example violates branch isolation on one connected torus. Here the cancellation occurs between two components of a four-component regular fiber. It exhibits failure of the hypothesis that every component be detectable in Theorem 1.3.

The quantum limit selected by the common nodal curve

The local Cauchy estimate determines the semiclassical distribution of the constructed eigenfunctions, although the full joint fiber is disconnected. This gives a direct application of the restriction method to the example, without requiring asymptotic information about its Floquet eigenfunctions.

Proposition 7.3 (A unique limit on two components). Let uju_j be the real normalized eigenfunctions of Proposition 7.1, and put hj=λj−1/2h_j=\lambda_j^{-1/2}. Denote by μστ\mu_{\sigma\tau} the normalized Liouville measure on the graph torus Λστ(0)\Lambda_{\sigma\tau}(0) in (6.9). Then the full sequence has the unique semiclassical measure

μ=12μ+−+12μ−+.(67)\mu=\frac{1}{2}\mu_{+-}+\frac{1}{2}\mu_{-+}. \tag*{(67)}

In particular, for every c∈C∞(T∗M)c\in C^\infty(T^*M),

lim⁡j→∞⟨Op⁡hjw(c)uj,uj⟩L2(M)=12∫Λ+−(0)c dμ+−+12∫Λ−+(0)c dμ−+.\lim_{j\to\infty}\langle\operatorname{Op}_{h_j}^{w}(c)u_j,u_j\rangle_{L^2(M)} =\frac{1}{2}\int_{\Lambda_{+-}(0)}c\,d\mu_{+-}+\frac{1}{2}\int_{\Lambda_{-+}(0)}c\,d\mu_{-+}.

The same conclusion holds for any sequence of real normalized functions in Ehj,0E_{h_j,0}, with hj→0h_j\to0, that vanishes identically on γr,s\gamma_{r,s}.

Proof. Write pL=(ξx2+ξy2)/Fp_L=(\xi_x^2+\xi_y^2)/F for the principal symbol of h2Lh^2L. The joint equations and ellipticity imply that any subsequential defect measure is supported on {pL=1,q=0}\{p_L=1,q=0\}. Invariance under both commuting Hamiltonian flows gives

μ=∑σ,τ=±1cστμστ,cστ≥0,∑σ,τcστ=1,\mu=\sum_{\sigma,\tau=\pm1}c_{\sigma\tau}\mu_{\sigma\tau},\qquad c_{\sigma\tau}\geq0,\qquad\sum_{\sigma,\tau}c_{\sigma\tau}=1,

as in (6.1). We show that the common zero curve determines these weights.

Along γr,s(t)=(st,rt)\gamma_{r,s}(t)=(st,rt) one has

A0=rf(rst),B0=sf(rst),γ˙r,s=(s,r).A_0=r\sqrt{f(rst)},\qquad B_0=s\sqrt{f(rst)},\qquad\dot{\gamma}_{r,s}=(s,r).

The four tangential momenta are therefore

(σA0,τB0)(γ˙r,s)=rsf(rst)(σ+τ).(\sigma A_0,\tau B_0)(\dot{\gamma}_{r,s})=rs\sqrt{f(rst)}(\sigma+\tau).

The ++++ and −−-- branches each have a unique, nonzero tangential covector, whereas the two mixed branches both restrict to zero. Thus a short arc has tangential windows isolating the ++++ and −−-- sheets from the entire fiber. In a coordinate neighborhood of this arc, use the defining function ρ=rx−sy\rho= rx - sy, up to an additive constant. On the two isolated branches,

HpLρ∣Λ++(0)=2(r2−s2)f(rst)F,HpLρ∣Λ−−(0)=−2(r2−s2)f(rst)F.H_{p_L}\rho\big|_{\Lambda_{++}(0)} = \frac{2(r^2-s^2)\sqrt{f(rst)}}{F}, \qquad H_{p_L}\rho\big|_{\Lambda_{--}(0)} = -\frac{2(r^2-s^2)\sqrt{f(rst)}}{F}.

Both are nonzero because r≠sr \ne s. The first-order symbol p1=1+pLp_1 = \sqrt{1+p_L} has the same transversality. Proposition 2.2 therefore supplies cutoffs χ++\chi_{++} and χ−−\chi_{--} in flow tubes on the corresponding components such that

∥Op⁡hj(χσσ)uj∥L2(M)≤C∥uj∣γr,s∥L2(γr,s)+o(1)=o(1),σ=±1.\left\|\operatorname{Op}_{h_j}(\chi_{\sigma\sigma})u_j\right\|_{L^2(M)} \leq C\left\|u_j\big|_{\gamma_{r,s}}\right\|_{L^2(\gamma_{r,s})} + o(1) = o(1), \qquad\sigma= \pm1.

Choose each cutoff nonzero on an open part of its torus and supported away from the other components. Passing to the defect measure gives

cσσ∫Λσσ(0)∣χσσ∣2 dμσσ=0.c_{\sigma\sigma}\int_{\Lambda_{\sigma\sigma}(0)} |\chi_{\sigma\sigma}|^2\,d\mu_{\sigma\sigma}=0.

The integrals are positive, hence c++=c−−=0c_{++}=c_{--}=0.

Since the functions uju_j are real, their defect measure is invariant under time reversal ι(x,ξ)=(x,−ξ)\iota(x,\xi)=(x,-\xi). This follows by conjugating a pseudodifferential observable; its principal symbol is changed from c(x,ξ)c(x,\xi) to c(x,−ξ)c(x,-\xi). The map ι\iota exchanges Λ+−(0)\Lambda_{+-}(0) and Λ−+(0)\Lambda_{-+}(0) and sends their normalized Liouville measures to one another. Thus c+−=c−+c_{+-}=c_{-+}. Their sum is one, proving (67). Every subsequential defect measure has been identified with the same probability measure, so the entire sequence converges. The proof uses only reality, the joint equations, and the zero trace, which also proves the final assertion. □

Remark 7.4 (The spatial limit). The limit has positive spatial density everywhere, despite exact vanishing on the fixed circle. Set

Z=∫T2F(x,y)A0(x)B0(y) dx dy.Z=\int_{\mathbb{T}^2}\frac{F(x,y)}{A_0(x)B_0(y)}\,dx\,dy.

On each graph torus the absolute Jacobian is

∣det⁡∂(pL,q)∂(ξx,ξy)∣=4A0B0F.\left|\det\frac{\partial(p_L,q)}{\partial(\xi_x,\xi_y)}\right|=\frac{4A_0B_0}{F}.

Thus the unnormalized joint Liouville density is F dx dy/(4A0B0)F\,dx\,dy/(4A_0B_0), and the projection of its normalization is

π∗μστ=1ZF(x,y)A0(x)B0(y) dx dy.\pi_*\mu_{\sigma\tau}=\frac{1}{Z}\frac{F(x,y)}{A_0(x)B_0(y)}\,dx\,dy.

Consequently,

∣uj∣2 dVg⇀1ZFA0B0 dx dy.(68)|u_j|^2\,dV_g \rightharpoonup\frac{1}{Z}\frac{F}{A_0B_0}\,dx\,dy. \tag*{(68)}

This positive spatial limit is compatible with exact nodality because the two remaining branches have indistinguishable tangential covectors and their traces cancel.

Geometry of the explicit example

For ε=1\varepsilon=1, the conformal factor ranges from 5 to 15. The curvature in (3) vanishes only at t=kπ/6t=k\pi/6 and is nonzero on the intervening arcs. Figure 1 and Figure 2 display the geometry and one eigenfunction. The three-dimensional drawing records the topology; the metric and curvature are specified by their formulas, not by an embedding in Euclidean space.

Figure 1. The analytic metric g=(10+cos⁡3x+4cos⁡6y)(dx2+dy2)g=(10+\cos3x+4\cos6y)(dx^2+dy^2) and its common nodal circle γ(t)=(2t,t)\gamma(t)=(2t,t). Top left: a coordinate fundamental square; the two red segments join to one circle. Top right: the winding (2,1)(2,1) on a topological torus. Bottom: the signed geodesic curvature.

Figure 2. A numerical member of the exact family Uj(x,y)=Im⁡(ϕj(x)ϕj(2y)‾)U_j(x,y)=\operatorname{Im}(\phi_j(x)\overline{\phi_j(2y)}) for ε=1\varepsilon=1, rescaled for display. The scalar eigenfunction is computed in the twisted Fourier basis ei(3k+1)te^{i(3k+1)t}, with eigenvalue approximately 12.9925664812.99256648.

References

School of Mathematics and Statistics, Wuhan University of Technology, Wuhan, Hubei, China

Email address: [email protected]

Department of Mathematics and Statistics, McGill University, Montréal, Quebec, Canada

Email address: [email protected]

References

  1. [1][BDWZ12] N. Burq, S. Dyatlov, R. Ward, and M. Zworski, Weighted eigenfunction estimates with applications to compressed sensing, SIAM J. Math. Anal. 44 (2012), no. 5, 3481–3501.
  2. [2][BGT04] N. Burq, P. Gérard, and N. Tzvetkov, The Cauchy problem for the nonlinear Schrödinger equation on compact manifolds, in Phase space analysis of partial differential equations, Vol. I, Publ. Cent. Ric. Mat. Ennio Giorgi, Scuola Norm. Sup., Pisa, 2004, 21–52.DOI
  3. [3][BGT07] N. Burq, P. Gérard, and N. Tzvetkov, Restrictions of the Laplace–Beltrami eigenfunctions to submanifolds, Duke Math. J. 138 (2007), no. 3, 445–486.
  4. [4][BR09] J. Bourgain and Z. Rudnick, Restriction of toral eigenfunctions to hypersurfaces, C. R. Math. Acad. Sci. Paris 347 (2009), no. 21–22, 1249–1253.arxiv.org/abs/0907.4824
  5. [5][BR11] J. Bourgain and Z. Rudnick, On the nodal sets of toral eigenfunctions, Invent. Math. 185 (2011), no. 1, 199–237. doi:10.1007/s00222-010-0307-2; arXiv:1003.1743.DOI
  6. [6][DZ13] S. Dyatlov and M. Zworski, Quantum ergodicity for restrictions to hypersurfaces, Nonlinearity 26 (2013), no. 1, 35–52.DOI
  7. [7][EPS15] A. Enciso and D. Peralta-Salas, Eigenfunctions with prescribed nodal sets, J. Differential Geom. 101 (2015), no. 2, 197–211. arXiv:1404.1039.arxiv.org/abs/1404.1039
  8. [8][GT20] J. Galkowski and J. A. Toth, Pointwise bounds for joint eigenfunctions of quantum completely integrable systems, Comm. Math. Phys. 375 (2020), 915–947. doi:10.1007/s00222-020-03730-3.DOI
  9. [9][GZ21] J. Galkowski and S. Zelditch, Lower bounds for Cauchy data on curves in a negatively curved surface, Israel J. Math. 244 (2021), no. 2, 971–1000.arxiv.org/abs/2002.09456
  10. [10][Han] X. Han, Problems, online seminar notes, §4.1.5, Problem 4.9, p. 61, undated. https://www.csun.edu/~xiaolong/Han_Pblm.pdf (accessed 16 October 2026).
  11. [11][JZ16] J. Jung and S. Zelditch, Number of nodal domains and singular points of eigenfunctions of negatively curved surfaces with an isometric involution, J. Differential Geom. 102 (2016), no. 1, 37–66.arxiv.org/abs/1310.2919
  12. [12][KTZ07] H. Koch, D. Tataru, and M. Zworski, Semiclassical Lᵖ estimates, Ann. Henri Poincaré 8 (2007), no. 5, 885–916.
  13. [13][Tot09] J. A. Toth, L²-restriction bounds for eigenfunctions along curves in the quantum completely integrable case, Comm. Math. Phys. 288 (2009), no. 1, 379–401.
  14. [14][TZ03] J. A. Toth and S. Zelditch, Norms of modes and quasi-modes revisited, in Harmonic analysis at Mount Holyoke (South Hadley, MA, 2001), Contemp. Math., vol. 320, Amer. Math. Soc., Providence, RI, 2003, 435–458.
  15. [15][TZ13] J. A. Toth and S. Zelditch, Quantum ergodic restriction theorems: manifolds without boundary, Geom. Funct. Anal. 23 (2013), no. 2, 715–775.DOI
  16. [16][TZ21] J. A. Toth and S. Zelditch, Nodal intersections and geometric control, J. Differential Geom. 117 (2021), no. 2, 345–393. doi:10.4310/jdg/1612975018. The geodesicity question cited here is in §1.3, p. 12, of arXiv:1708.05754v1 (2017).DOI
  17. [17][WW] W. Wang and X. Wu, Average estimate for eigenfunctions along geodesics in the quantum completely integrable case, arXiv:2506.15992v2 (2026).arxiv.org/abs/2506.15992
  18. [18][WXb] X. Wu and X. Xiao, Trace estimates and improved pointwise bounds for joint eigenfunctions, arXiv:2604.22197 (2026).arxiv.org/abs/2604.22197
  19. [19][Zwo12] M. Zworski, Semiclassical analysis, Graduate Studies in Mathematics, vol. 138, Amer. Math. Soc., Providence, RI, 2012.

Paper details

vlnc:W_kepue63v
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
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