Result 1
TFX3Area-minimizing representatives of integral homology classes can be recovered from those of real homology and mod n homologies for n\in \mathbb{Z}_{\ge 2}.
The Hasse principle in number theory states that information about integral solutions to Diophantine equations can be pieced together from real solutions and solutions modulo prime powers. We show that an analogous Hasse principle holds for area-minimizing submanifolds: information about area-minimizing submanifolds in integral homology can be recovered from those in real homology and mod n homology for all n\in \mathbb{Z}_{\ge 2}. As a consequence we answer several questions of Almgren, Morgan and White and prove: area-minimizing submanifolds are not generically calibrated, products of area-minimizing submanifolds are not generically area-minimizing, and classification of area-minimizing pairs of planes mod n for n\ge 4.
Area-minimizing representatives of integral homology classes can be recovered from those of real homology and mod n homologies for n\in \mathbb{Z}_{\ge 2}.
Every area-minimizing integral current on a compact Riemannian manifold is area-minimizing mod n for infinitely many n.
An area-minimizing integral cone with subanalytic link is area-minimizing mod n for n sufficiently large.
A pair of oriented planes is area-minimizing mod n for n\ge 4 if and only if the pair is an area-minimizing integral current.
With smooth or subanalytic boundary, the solution to Plateau problem mod n is obtained by integral currents for n large enough, thus enjoying an improvement of regularity.
On orientable manifolds, area-minimizing integral currents of codimension at least two are not generically calibrated.
On a product of two closed oriented manifolds with non-vanishing Betti numbers, products of area-minimizing representatives of some homology classes are not generically area-minimizing.
(Compact setting) Assume that
M is a (d+c)-dimensional compact closed not necessarily orientable smooth manifold, with d,c\in\mathbb{Z}_{\ge 1},
[\Sigma]\in H_d(M,\mathbb{Z}) is a d-dimensional integral homology class,
g is a smooth Riemannian metric on M.
Use G_d(M,\mathbb{Z}) to denote the torsion subgroup of H_d(M,\mathbb{Z}), i.e.,
Define the d-th torsion number \tau_d of M to be
where \operatorname{lcm} means the least common multiple. Set \tau_d=1 if H_d(M,\mathbb{Z}) is torsion-free.
We classify the integral homology class [\Sigma] according to its algebraic properties.
We say [\Sigma] is a torsion class if there is an integer v such that v[\Sigma]=0, i.e., belonging to G_d(M,\mathbb{Z}). Otherwise we say [\Sigma] is a non-torsion class,
If [\Sigma] is a non-torsion class, M is orientable, and d\ge1 ,c\ge 2, then the representable multiplicity \mu_{[\Sigma]} of [\Sigma] is defined to be the smallest integer \mu_{[\Sigma]}\in\mathbb{Z}_{\ge 2} such that \mu_{[\Sigma]}[\Sigma] can be represented by a smoothly embedded connected submanifold.
The R-reduction [\Sigma]^R of the integral homology class [\Sigma] is defined to be
where \iota is the natural injection in (?) and 1 is the unit element in R.
We say a map \left\lVert\cdot \right\rVert^R:R\to \mathbb{R}_{\ge0} is a norm on R if \left\lVert\cdot\right\rVert^R satisfies the following three properties:
(reversibility) for all e\in R, we have
(triangle inequality) for all e,f\in R, we have
(positivity) we have the equivalence of propositions, for all e\in R,
The R-flat chain \mathcal{F}_g(M,R) in the metric g is the \mathcal{F}^R closure of the set of R-coefficient polyhedron chains.
When the canonical homomorphism \mathbb{Z}\to R is norm non-increasing, define the R-reduction T^R of an integral current T\in[\Sigma] to be
Define the R-area \left\lVert[\Sigma]\right\rVert^R_g of [\Sigma]^R to be
We say an R-flat chain T\in[\Sigma]^R is R-area-minimizing in [\Sigma]^R with respect to metric g, if T achieves the R-area of [\Sigma]^R defined in (2.1.7), i.e.,
The following inequality holds
Define
Set R=\mathbb{R} or \mathbb{Z} or \mathbb{Z}/n\mathbb{Z} for n\in\mathbb{Z}_{\ge 2} and use Definition 2.1.8 for \left\lVert\cdot\right\rVert^R.
The following inequality always holds
For a subset S\subset H_d(M,\mathbb{Z}), define the set of R-area-minimizing representatives \mathbf{MIN}_g^R\left(S\right) of S^R to be
For all [\Sigma]\in H_d(M,\mathbb{Z}), there exists a positive integer N_{[\Sigma],g}, and an open set \Omega_{[\Sigma],g} containing g in the space of Riemannian metrics, both depending only on [\Sigma] and g, such that for all metrics h\in\Omega_{[\Sigma],g}, all n\ge N_{[\Sigma],g}, with \tau_d|n
equality holds in (2.1.2), i.e., \left\lVert[\Sigma]\right\rVert_h^{\mathbb{Z}/n\mathbb{Z}}=\left\lVert[\Sigma]\right\rVert_h^\mathbb{Z},
the \mathbb{Z}/n\mathbb{Z}-reduction restricted to \mathbf{MIN}_h^{\mathbb{Z}}\left([\Sigma]\right) is a one to one map from \mathbf{MIN}_h^{\mathbb{Z}}\left([\Sigma]\right) to \mathbf{MIN}_h^{\mathbb{Z}/n\mathbb{Z}}\left([\Sigma]\right).
For all metrics h\in\Omega_{[\Sigma],g}, all n\ge N_{[\Sigma],g}, with \tau_d\not|n, we have
and the \mathbb{Z}/n\mathbb{Z}-reduction is a bijection from \mathbf{MIN}_h^{\mathbb{Z}}\big([\Sigma]+nG_d(M,\mathbb{Z})\big) to \mathbf{MIN}_h^{\mathbb{Z}/n\mathbb{Z}}([\Sigma]).
Area-minimizing representatives of [\Sigma]^{\mathbb{Z}/n\mathbb{Z}} are \mathbb{Z}/n\mathbb{Z}-reductions of area-minimizing integral currents for n large enough, thus enjoying much better regularity than general \mathbb{Z}/n\mathbb{Z} area-minimizing representatives.
If d\ge 1,c\ge 2, M is orientable, and [\Sigma] a non-torsion class, then for any real number \rho>1 and all integers n\ge 2 with \operatorname{gcd}\left(n,\mu_{[\Sigma]}\right)=1, there exists a non-empty open set \Omega'_{[\Sigma],n,\rho} in the space of Riemannian metrics, depending on [\Sigma],n, and \rho, such that for all metrics h\in\Omega'_{[\Sigma],n,\rho} we have
If c\ge 2 and M is orientable, then for any [\Sigma]\not=0, there exists a non-empty open set \Omega_{[\Sigma],\mathbb{R}} in the space of Riemannian metrics, such that for any metric h\in \Omega_{[\Sigma],\mathbb{R}} and any T\in\mathbf{MIN}_h^\mathbb{Z}([\Sigma]), there is no weakly closed measurable d-form calibrating T.
Assume that c,d\in\mathbb{Z}_{\ge1} and
W is either a (d-1)-dimensional compact smooth submanifold or a compact (d-1)-dimensional subanalytic set of M=\mathbb{R}^{d+c},
[\Sigma]\in H_d(\mathbb{R}^{d+c},W,\mathbb{Z}) is a d-dimensional relative integral homology class with d\ge 2.
Assume d\ge 2. For d-dimensional R-flat chains T with \partial T supported in W, we define
For d\ge 2,c\ge1 define the R-area \left\lVert[\Sigma]\right\rVert^R_\delta of [\Sigma]^R to be
We say an R-flat chain T\in[\Sigma]^R is R-area-minimizing in [\Sigma]^R with respect to metric g, if T achieves the R-area of [\Sigma]^R defined in (2.1.7), i.e.,
The following inequality always holds
For a subset S\subset H_d(\mathbb{R}^{d+c},W,\mathbb{Z}), define the set of R-area-minimizing representatives \mathbf{MIN}_\delta^R\left(S\right) of S^R to be
For all [\Sigma]\in H_d(\mathbb{R}^{d+c},W,\mathbb{Z}), there exists a positive integer N_{[\Sigma],\delta}, depending only on [\Sigma], such that for all n\ge N_{[\Sigma],\delta},
we have \left\lVert[\Sigma]\right\rVert_\delta^{\mathbb{Z}/n\mathbb{Z}}=\left\lVert[\Sigma]\right\rVert_\delta^\mathbb{Z},
the \mathbb{Z}/n\mathbb{Z}-reduction restricted to \mathbf{MIN}_\delta^{\mathbb{Z}}\left([\Sigma]\right) is a one to one map from \mathbf{MIN}_h^{\mathbb{Z}}\left([\Sigma]\right) to \mathbf{MIN}_\delta^{\mathbb{Z}/n\mathbb{Z}}\left([\Sigma]\right).
Assume that d\ge 1,c\ge 1, and B is an integral current cycle, i.e., \partial B=0, supported in W, then exists N_{B}\ge 0 so that for all integers n\ge N_B, the \mathbb{Z}/n\mathbb{Z}-reduction of an \mathbb{Z}-area-minimizing integral current T with boundary B is \mathbb{Z}/n\mathbb{Z}-area-minimizing.
Assume that P,Q are oriented d-dimensional subspaces of \mathbb{R}^{d+c}, with d\ge 2,c\ge 1,n\ge4. Then P^{\mathbb{Z}/n\mathbb{Z}}+Q^{\mathbb{Z}/n\mathbb{Z}} is \mathbb{Z}/n\mathbb{Z} area-minimizing if and only if P+Q is \mathbb{Z}-area-minimizing.
For c=d\ge 2, assume that P,Q are two oriented d-dimensional planes intersecting only at the origin in \mathbb{R}^{d+c}.
For n\ge 4, the \mathbb{Z}/n\mathbb{Z}-flat chain P^{\mathbb{Z}/n\mathbb{Z}}+Q^{\mathbb{Z}/n\mathbb{Z}} is \mathbb{Z}/n\mathbb{Z}-area-minimizing if and only if P+Q is \mathbb{Z}-area-minimizing.
For any natural homomorphism among homology groups, quantify the effect of the homomorphism on area-minimizing representatives.
Assumption 2.1.1 holds with another set of objects M',d',c',[\Sigma'],g' in place of M,d,c,[\Sigma],g.
Define
The following inequality holds,
If d\le 2, or c\le 2, or both [\Sigma],[\Sigma'] are of dimension 3 and codimension 3, then we have \left\lVert[\Sigma]\right\rVert^{\mathbb{R}}_{g}\times \left\lVert[\Sigma']\right\rVert^{\mathbb{R}}_{g'}= \left\lVert[\Sigma]\times [\Sigma']\right\rVert^{\mathbb{R}}_{g\times g'},
When both [\Sigma],[\Sigma'] are of codimension 1 and both M,M' are orientable,
If d\ge 1,c\ge 2, [\Sigma] is a non-torsion class on an orientable M, then for any real number \kappa>1, any metric g' on M', there exists a non-empty open set \Omega_{[\Sigma],g',\kappa} in the space of Riemannian metrics on M, a non-empty open set \Omega_{[\Sigma],g',\kappa}'\ni g' in the space of Riemannian metrics on M', and an integer \nu_{[\Sigma],g',\kappa}\ge 2, such that for any metrics h\in \Omega_{[\Sigma],g',\kappa}, h'\in \Omega_{[\Sigma],g',\kappa}' and any non-torsion homology class [\sigma']\in \nu H_m(M',\mathbb{Z}) with any integers \nu\ge \nu_{[\Sigma],g',\kappa}, 0\le m\le d'+c', we have,
Use G_d(M,W,\mathbb{Z}) to denote the torsion subgroup of H_d(M,W,\mathbb{Z}), i.e.,
Define the d-th torsion number \tau_d of (M,W) to be
where \operatorname{lcm} means the least common multiple. Set \tau_d=1 if H_d(M,W,\mathbb{Z}) is torsion-free.
Let \{f_1,\cdots,f_{b_d},v_1,\cdots,v_{|I|}\} be a basis of H_d(M,W,\mathbb{Z}) so that
\{f_1,\cdots,f_{b_d}\} is a \mathbb{Z}-linear independent basis of the factor \mathbb{Z}^{b_d} in (?),
for all j\in I, the element v_j generates the \mathbb{Z}/p_j^{\nu_j}\mathbb{Z} factor in (?).
If X,Y\in H_d(M,W,\mathbb{Z}) are two integral homology classes with the same \mathbb{Z}/n\mathbb{Z}-reduction, i.e., X^{\mathbb{Z}/n\mathbb{Z}}=Y^{\mathbb{Z}/n\mathbb{Z}}, then
Furthermore, when \tau_d|n, then X-Y is n times a non-torsion class.
It suffices to determine the kernel of \mathbb{Z}/n\mathbb{Z}-reduction. By [AH, Section 3.2 p. 215 property (5)], the kernel of \otimes \mathbb{Z}/n\mathbb{Z} of an abelian group is n times the group. Thus, if X,Y has the same \mathbb{Z}/n\mathbb{Z}-reduction, we have
If \tau_d|n, then nH_d(M,W,\mathbb{Z}) contains no nonzero element of the torsion subgroup G_d(M,W,\mathbb{Z}). We are done.
With the same assumptions as in Theorem 3, there exists an infinite subset K\subset \mathbb{Z}_{\ge 2}, such that for all k\in K, the integral homology class k[\Sigma] can be represented by a smoothly embedded connected compact submanifold,
First, le us prove the lemma without the connectedness assumption. Let us argue by contradiction. Suppose there exists only a finite subset F\subset \mathbb{Z}_{\ge 2} so that for k\in F, the integral homology class has a smoothly embedded representative. By [RT, Theorem II.29] some multiple of \left(\prod_{k\in F}k\right)[\Sigma] admits a smoothly embedded representative, a contradiction. Here we regard \prod_{k\in F}k as 1 in case F is empty.
Next, let us show that when the codimension of [\Sigma] with respect to the ambient manifold M is at least two, any smoothly embedded representative of k[\Sigma] with finitely many connected components can be made into a connected submanifold by connected sum of submanifolds. The case of d\ge 2 is classical [CL] and a detailed write up can be found in [ZLns, Lemma 2.8.2]. For d=1, we utilize a direct construction. For two disjoint smoothly embedded circles \gamma_1,\gamma_3, we take another curve \gamma_2 that starts from a point on \gamma_1 and end at a point on \gamma_3. Let \gamma_4 be \gamma_2 reversely traversed. Then \gamma_1+\gamma_2+\gamma_3+\gamma_4 is homologically equivalent to \gamma_1+\gamma_3. As embeddings are generic maps when d=1<2\le c [MH, 2.13 Theorem, 2.6 Theorem], a generic perturbation of \gamma_1+\gamma_2+\gamma_3+\gamma_4 gives a closed embedded circle \gamma_1\#\gamma_3\in [\gamma_1]+[\gamma_3]\in H_1(M,\mathbb{Z}).
We have
For R=\mathbb{Z} or \mathbb{Z}/n\mathbb{Z} for n\in\mathbb{Z}_{\ge 2}, finite mass R-flat chains form the space of R-rectifiable currents.
Every d-dimensional \mathbb{Z}-rectifiable current T belongs in the dual space of smooth differential forms, with actions on a form \phi defined by
where \operatorname{Rect} T is the Hausdorff d-dimensional (countably) underlying rectifiable set of T defined in [HF, Section 4.1.28(4)], \theta_T is the density of the mass measure of T and \eta is the simple unit d-vector representing the oriented tangent space to \operatorname{Rect} T Hausdorff d-dimensional a.e.
Every d-dimensional \mathbb{Z}-flat chain decomposes as a sum of a d-dimensional \mathbb{Z}-rectifiable current and the boundary of a (d+1)-dimensional \mathbb{Z}-rectifiable current and vice versa.
For every \mathbb{Z}/n\mathbb{Z}-rectifiable current T, there exists at least one \mathbb{Z}-rectifiable current V, called Federer’s “representative" modulo n of T such that the \mathbb{Z}/n\mathbb{Z}-mass measure of T coincides with the \mathbb{Z}-mass measure of V and T=V^{\mathbb{Z}/n\mathbb{Z}}.
In general, Federer’s “representative" modulo n
is not unique and many times there are infinitely many,
of a \mathbb{Z}/n\mathbb{Z}-cycle may not even be a \mathbb{Z}-cycle at all.
We say an R-rectifiable current T is smooth at a point p in the support of T if there exists an open set U containing p on M, such that T restricted to U equals \alpha times a smooth submanifold N, with \alpha\in R. The definitions of regular sets and singular sets of T are as follows:
The singular set of T, \operatorname{Sing} T, is defined as set of points in the support of T where T is not smooth,
The regular set of T, \operatorname{Reg}T, is defined as the set of points in the support of T where T is smooth.
We call \mathcal{S}_jT\setminus\mathcal{S}_{j-1}T the j-th stratum of the Almgren stratification, with S_{-1}T=\emptyset.
The j-th Almgren stratification of T has Hausdorff dimension at most j.
We define the Almgren stratification of any subset S of \operatorname{Supp} T by setting
Suppose T is a d-dimensional \mathbb{Z}/n\mathbb{Z}-flat chain on a not necessarily compact (d+c)-dimensional Riemannian manifold M^{d+c}. Let S be a rectifiable current that is a Federer’s “representative" modulo n of T. Then
where \operatorname{Reg}(T) denotes the regular part of T.
Note that regular set of T can have density at most \frac{n}{2}. We will prove that if p\in\operatorname{Supp} \partial S\cap \textnormal{Reg}(T), then \theta_T(p)= \frac{n}{2}.
Since p is in the regular set of T, there exists a neighborhood \mathbf{B}_r(p) in which T restricted to \mathbf{B}_r(p) equals \theta_T(p) N for some smooth submanifold N of \mathbf{B}_r(p), so that in some coordinate system (x_1,\cdots,x_{d+c}), N is a smooth ball contained in the x_1\cdots x_d-plane. By [RY, Corollary 1.4], there exists a \mathbb{Z}-flat chain R in \mathbf{B}_r(p) so that
as a \mathbb{Z}-rectifiable current. Here S\mathop{\hbox{\vrule height 7pt width .3pt depth 0pt \vrule height .3pt width 5pt depth 0pt}}\nolimits \mathbf{B}_r(p) means S restricted to \mathbf{B}_r(p). Since S,N are both of finite mass, we deduce that R is also of finite mass, thus a \mathbb{Z}-rectifiable current by [HF, Section 4.2.16].
If R=0, then \partial S restricted to \mathbf{B}_r(p) is empty, as S coincides with \theta_T(p)N. Thus, we only have to deal with the case of R\not=0.
Recall Fact 3.3.2. There exist a d-dimensional rectifiable set B, an \mathcal{H}^d measurable simple unit d-vector field \eta , so that the action of the current R on differential forms is
with \phi any smooth d-dimensional form.
By [HF, page 430], the support of S restricted to \mathbf{B}_r(p) is contained in N, thus by (Proof) the support of R is also contained in N, i.e., B\subset N. This implies that \eta=\pm \mathbf{T}_x N Hausdorff d-dimensional a.e. Here \mathbf{T}_x N means the unique simple d-vector field representing the tangent space at x to N.
Thus, we have
By [HF, Section 4.1.28], we deduce that
However, since we have R\not=0, i.e., \theta_R(x) being a positive integer on B Hausdorff d-dimensional a.e. , and \theta_T(p)\le \frac{n}{2} by [HF, page 430], we have
The inequality
holds where equality is achieved if and only if \theta_T(p)=\frac{n}{2},\theta_R(x)=1.
Thus, we have
By Fact 3.3.4, we have
In other words, equality holds in (Proof) Hausdorff d-dimensional a.e. on B. By Fact Proof, we deduce that
on B Hausdorff d-dimensional a.e. Thus \partial S intersect \operatorname{Reg} T at \theta_T=\frac{n}{2} and we are done.
Suppose T is a d-dimensional \mathbb{Z}/n\mathbb{Z}-area-minimizing \mathbb{Z}/n\mathbb{Z}- flat chain on a not necessarily compact (d+c)-dimensional Riemannian manifold (M^{d+c},h) with possibly non-zero boundary \partial T. Then
has Hausdorff dimension at most (d-2).
Recall Section ?. Since \theta_T(p,h) is an upper-semicontinuous function of p, the set \{\theta_T(p,h)<\frac{n}{2}\} is open. By [DHMS, Theorem 1.7], the d-th stratum of the Almgren stratification of \operatorname{Sing} T\cap (\operatorname{Supp}\partial T)^\complement\cap \{\theta_T(p,h)<\frac{n}{2}\} is of Hausdorff dimension at most (d-2). On the other hand, by [DHMS, Lemma 8.6], for any point p in the (d-1)-th stratum of the Almgren stratification of \operatorname{Sing} T\cap (\operatorname{Supp}\partial T)^\complement\cap \{\theta_T(p,h)<\frac{n}{2}\}, we have \theta_T(p,h)\ge\frac{n}{2}, so the (d-1)-th stratum of the Almgren stratification of \operatorname{Sing} T\cap W\cap \{\theta_T(p,h)<\frac{n}{2}\} is empty. By Fact 3.5.1, we are done.
Assume that S is a d-dimensional \mathbb{Z}-rectifiable current on a not necessarily compact (d+c)-dimensional Riemannian manifold M^{d+c}. If
with W and W\cup O closed, O disjoint from W and O having Hausdorff dimension at most (d-2), then we have
By Fact 3.3.4, \partial S is a \mathbb{Z}-rectifiable current. By the definition of support [HF, Section 4.1.1 p.344], it suffices to prove that for every point p\in M\setminus W, there is a neighborhood U(p) in which T(\phi)=0 with \phi any smooth d-form vanishing outside of U(p). Recall Fact 3.3.2. We have
where at the last step we used O having Hausdorff dimension at most (d-2) and \phi vanishing on W. We are done.
For any R-homology class \omega\in H_d(M,R), define
For any subset W of H_d(M,R), define the set of R-area-minimizing representatives \mathbf{MIN}_g^R(W) of W to be
\left\lVert\cdot\right\rVert^R_g depends continuously on g.
If \mathbf{M}_g^R\le A\mathbf{M}_h^R for A>0, then \left\lVert\cdot\right\rVert_g^R\le A\left\lVert\cdot\right\rVert_h^R.
\left\lVert\cdot\right\rVert^R_g is a norm in the sense of Definition 2.1.4, i.e., satisfying reversibility, triangle inequality and positivity.
For any integer k\in \mathbb{Z}, \omega\in H_d(M,R), we have
We have
\left\lVert\cdot\right\rVert^{\mathbb{R}}_g is turns the the vector space H_d(M,\mathbb{R}) into a finite dimensional Banach space,
\left\lVert[\Sigma]\right\rVert^{\mathbb{R}}_g\not=0 if and only if [\Sigma] is a non-torsion class.
we have
For any torsion class \tau\in H_d(M,\mathbb{Z}), we have
If an integer k\in \left(-\frac{n}{2},\frac{n}{2}\right] invertible in \mathbb{Z}/n\mathbb{Z} with its unique inverse in \left(-\frac{n}{2},\frac{n}{2}\right] denoted by l, then for any \omega\in H_d(M,\mathbb{Z}/n\mathbb{Z}), we have
and thus
Fix a basis \{v_1,\cdots,v_b\} of \mathbb{R}^b. For any continuous Banach norm \left\lVert\cdot\right\rVert on \mathbb{R}^n, there exists a C>0 that depends continuously on \left\lVert\cdot\right\rVert, so that
with each a_j\in \mathbb{R}.
For v=\sum_{j=1}^b a_jv_j with each a_j\in\mathbb{R} define
It is straightforward to verify that \left\lVert\cdot\right\rVert^\infty is a continuous function Banach norm on \mathbb{R}^n. We need to prove that \frac{\left\lVert\cdot\right\rVert^\infty}{\left\lVert\cdot\right\rVert} is bounded. By homogeneity, it suffices to verify this on the norm unit sphere \{v\in \mathbb{R}^n|\left\lVertv\right\rVert=1\}. This follows from continuity of \left\lVert\cdot\right\rVert^\infty, and the the compactness of norm unit sphere.
There exists a continuous function \upsilon mapping the space of Riemannian metrics to \mathbb{R}_{\ge 0}, such that
Recall Assumption 3.1.1. Every non-torsion class [\Sigma]\in H_d(M,\mathbb{Z}) can be written as
with each \alpha_j\in \mathbb{Z},\beta_i\in\mathbb{Z}\cap[0,p_j^{\nu_j}], and not all a_i are 0. By Lemma 3.8.2 there is a continuous function C from the space of Riemannian metrics to \mathbb{R}_{> 0} so that
where at the first equality we have used Fact 3.8.6. By the mediant inequality and \max_j|\alpha_j|\ge 1, we have
By Fact 3.8.1, we are done.
(Definition of calibrations)
We say a closed d-dimensional Hausdorff measurable d-form \phi on a (possibly open) ambient manifold is a calibration if its comass is at most 1.
We say a d-dimensional integral current T is calibrated by \phi, if d-dimensional Hausdorff measure a.e. where \phi restricted to the tangent space of T equals the volume form of T.
If a d-dimensional integral current T on a Riemannian manifold is calibrated by a degree d smooth calibration form \phi, then T is both \mathbb{Z}-area-minimizing and \mathbb{R}-area-minimizing, and any integral current homologous to T and having the same area as T must also be calibrated by \phi.
We say an \mathbb{R}-flat cochain \alpha, is a calibration on (M,g), if
d\alpha=0,
\alpha(T)\le \mathbf{M}^{\mathbb{R}}_g(T), for all \mathbb{R}-flat chains T.
We say an \mathbb{R}-flat chain T is calibrated by \alpha, if
Every element of \mathbf{MIN}_g^{\mathbb{R}}\left([\Sigma]\right) is calibrated by a calibration \mathbb{R}-flat cochain.
A \mathbb{Z}-area-minimizing integral current representing [\Sigma] is calibrated by an \mathbb{R}-flat cochain if and only if \left\lVert[\Sigma]\right\rVert_g^{\mathbb{Z}}=\left\lVert[\Sigma]\right\rVert_g^{\mathbb{R}}.
[VDDMom, Definition on p.506 and discussion right after it] An \mathbb{R}_{\operatorname{an}}-definable set of \mathbb{R} is a finite union of intervals and points.
Compact subanalytic sets in Euclidean spaces are \mathbb{R}_{\operatorname{an}}-definable.
If W is a (d-1)-dimensional compact subanalytic set of \mathbb{R}^{d+c} and V is a (c+1)-dimensional subspace of \mathbb{R}^{d+c}, with c,d\ge 1, then for Haar measure almost every g\in O(d+c), W\cap g.V consists of at most \mu_W points. Furthermore, affine translates W+p of W satisfies
The set of points in g\in O(d+c) satisfying \dim g.V\cap W\ge 1 is \mathbb{R}_{\operatorname{an}}-definable and has dimension most \dim(O(d+c))-1
The case of d=1 is trivial as \dim W=0. From now on we assume that d\ge 2. By [MComin, Exercise 3.13] (which itself is just a successive application of [MComin, Theorem 3.12]), for every g\in O(d+c), every affine translation p+W, the intersection g.V\cap (p+W) has at most \mu_W connected components, which are \mathbb{R}_{\operatorname{an}}-definable sets. By Fact 3.10.3, Haar almost every g\in O(d+c), the intersection g.V\cap W consists only of 0-dimensional \mathbb{R}_{\operatorname{an}}-definable sets, i.e., discrete points. We are done.
Let V be a (c+1)-dimensional subspace of \mathbb{R}^{d+c} and v a unit length vector in V. Then the subset O(V,v) of O(d+c) defined by
is a smooth \mathbb{R}_{\operatorname{an}}-definable set of dimension
Choose a coordinate (x_1,\cdots,x_{d+c}) system on \mathbb{R}^{d+c} so that V is the coordinate plane x_1\cdots x_{c+1} and v is the coordinate vector \partial_{x_1}. Identify \mathbb{R}^{(d+c)^2} with the space of matrices via entries.
Consider the real algebraic map
defined by
where \mathbb{R}^{\frac{(d+c)(d+c+1)}{2}} is the the linear space of symmetric matrices, t denotes transpose and g_{i,j} means the ij-th entry of g.
Then O(V,v) is the real algebraic subvariety of O(d+c) defined by
Let us prove that dF has rank \frac{(d+c)(d+c+1)}{2}+(d-1) when g\in O(V,v). Then by the implicit function theorem [JL, Theorem 7.9], we deduce that O(V,v) is a smooth submanifold of dimension
And we are done.
First, we calculate the differential of F when g is invertible.
For any fixed g with \det g\not=0, do a linear change of variables as follows: left multiplication by g on \mathbb{R}^{(d+c)^2} and left multiplication by g^{-1} composed with right multiplication by (g^t)^{-1} on \mathbb{R}^{\frac{(d+c)(d+c+1)}{2}}, we can rewrite
where g_i.h^j denotes the inner product of the i-th row of g with the j-th column of h.
Now let us specialize to g\in O(V,v)=F^{-1}(I,0). Then g_{c+2},\cdots,g_{c+d} are orthonormal frames of a vector subspace dimension (d-1) and their first entries are all zero, i.e.,
Complete g_{c+2},\cdots,g_{c+d} to an orthnormal frame of \mathbb{R}^{d+c}:
Define \kappa:\mathbb{R}^{d+c}\to\mathbb{R}^{(d+c)^2} by setting
In other words, \kappa(w) is the (d+c)\times (d+c) matrix formed by regarding -w as the first row vector and w as the first column vector and add them to the zero matrix. By definition,
We claim that
dF in (Proof) sends the span of the set of symmetric matrices and the set \kappa(\mathbb{R}^{d+c}) onto \mathbb{R}^{\frac{(d+c)(d+c+1)}{2}}.
To prove the above fact, let e_i^j denote the (d+c)\times(d+c) matrix with all entries zero outside of the ij-th entry, which is set to 1. Then direct calculation shows that and we are done:
where 0_{d+c} denotes the (d+c)\times(d+c) 0 matrix.
Let N be a d-dimensional open submanifold in a complete (d+c)-dimensional manifold M. Then an d-dimensional R-flat chain T such that \operatorname{Supp} \partial T\cap N=\emptyset restricted to N equals \alpha N for some \alpha\in R.
Apply [MSfc, 6.1 Theorem]
The d-dimensional R-homology of W, H_d(W,R), satisfies the following properties
H_{d-1}(W,R) is finitely generated abelian group and is free R=\mathbb{Z} or \mathbb{R},
every element of H_{d-1}(W,R) can be represented by a unique (d-1)-dimensional simplicial R-cycle and vice versa,
every (d-1)-dimensional R-flat chain cycle supported in W can be represented by a simplicial cycle,
the (d-1)-dimensional homology of R-flat chains supported in W is isomorphic to H_{d-1}(W,R).
every element of H_{d-1}(W,R) can be represented by a unique (d-1)-dimensional R-flat chain cycle and vice versa.
First, W admits a finite triangulation ([BPhs, 7.4 Theorem] for subanalytic W and [SCtr] for smooth W). Next, there is no d-dimensional simplices, so there is non-nontrivial exact (d-1)-dimensional chains. Thus, any homology class of H_{d-1}(W,R) has a unique representative by a simplicial cycle. This means that the group H_{d-1}(W,R) is a subgroup of the finitely generated abelian group of R-simplicial d-dimensional chains in W. By [AKba, Proposition 4.52] H_{d-1}(W,R) is finitely generated. When R=\mathbb{Z} or \mathbb{R}, the group of R-simplicial chains is free, so its subgroup H_{d-1}(W,R) is also free [AKba, Theorem 4.55]. We are done with the first bullet and the second bullet.
For the third bullet, as \partial T=0, by Fact 3.11.1, every R-flat cycle is represented by an simplicial R-chain and then apply the second bullet.
The fourth bullet and the last bullet follows from the third bullet and the second bullet.
Use \mathcal{C}(\mathbb{R}^n,R) to denote the subspace of \mathbb{R}-flat chains on \mathbb{R}^n that are cycles, i.e.,
For any R-flat chain \beta\in \mathcal{C}(\mathbb{R}^n,R) define
For any subset S of \mathcal{C}_d(\mathbb{R}^{d+c},R), define the set of R-area-minimizing representatives \mathbf{MIN}^R_\delta(S) of S to be
\left\lVert\cdot\right\rVert^R_\delta is a norm in the sense of Definition 2.1.4 on \mathcal{C}(\mathbb{R}^n,R) i.e., satisfying reversibility, triangle inequality and positivity.
For any integer k\in \mathbb{Z}, \beta\in\mathcal{C}(\mathbb{R}^n,R), we have
We have
\left\lVert\cdot\right\rVert^{\mathbb{R}}_\delta turns the vector space H_d(\mathbb{R}^{d+c},W,\mathbb{R}) into a finite dimensional Banach space,
\left\lVertk\beta\right\rVert^\mathbb{R}_\delta= |k|\left\lVert\beta\right\rVert^\mathbb{R}_\delta for any k\in\mathbb{R},
and
We have
Assume N is a d-dimensional connected embedded compact closed submanifold of M. For any real number \lambda>0, there exists a smooth tubular neighborhood U(N) of N, a smooth Riemannian metric h on M, and a smooth deformation retract \pi_N of U(N) onto N, such that
any stationary varifold on (M,h) whose support is not contained in U(N) has area larger than \lambda \mathbf{M}_h^{\mathbb{Z}}(N),
\pi_N^\ast \operatorname{dvol}_{N} is a calibration form on (U(N),h),
N is the only integral current calibrated by \pi_N^\ast \operatorname{dvol}_{N} in (U(N),h),
if N represents a non-zero \mathbb{R}-homology class, then \pi_N^\ast \operatorname{dvol}_{N} can be extended to a smooth calibration form on (M,h),
N is \mathbb{Z}/n\mathbb{Z}-area-minimizing in U(N) for n\in \mathbb{Z}_{\ge2}.
The lemma combines several results of Yongsheng Zhang’s thesis [YZt] and published in [YZj, YZa]. Take a tubular neighborhood U(N) of N inside M. The first bullet is [YZj, Lemma 3.1] applied to U(N) and M. The second to the fourth bullet come from [YZa, Sections 3.3, 3.4]. The last bullet follows from [ZLa, Lemma 2.12]. For an alternative proof the last bullet, by [ZLns, Fact 7.0.6] \pi^\ast_N\operatorname{dvol}_N is a calibration form on U(N) implies that \pi is a d-dimensional area-non-increasing map from U(N) to N, i.e.,
for R=\mathbb{Z}, or \mathbb{Z}/n\mathbb{Z} with n\in\mathbb{Z}_{\ge2 } and T any R-flat chains supported in U(N). For any R-flat chain N' homologous to N in U(N), we necessarily have \pi_\ast N'=N since \pi is a deformation retract. By (Proof), we are done.
There exist an open set \Omega_g containing g in the space of smooth Riemannian metrics, positive real numbers C_g,c_g,r_g>0 depending on g, such that, for any metric h\in \Omega_g, any point p\in M and all radii r\in(0,r_g], we have
for any d-dimensional integral stationary varifold V on M in the metric h, the function,
is monotonically increasing in r, where V(\cdot) denotes the mass measure of V, and \mathbf{B}_r(p,h) is the radius r geodesic ball in metric h centered at p,
and the density \theta_V({p,h}) of V at p in h is well-defined and satisfies
The argument is the Riemannian adaptation of the first two sections of [Dar]. First by [PEcir, Section 8 Theorem], the injectivity radius \operatorname{Inj}_g(M) of M in g depends continuously on the metric g. Thus, there exists an open neighborhood \Omega_g containing g in the space of Riemannian metrics so that
For any metric h\in\Omega_g, and any point p\in M, adopt a normal coordinate (x_1,\cdots,x_{d+c}) in \mathbf{B}_{r_g}(p,h). Let \mathbf{r}_{p,h} denote the distance function to p on (M,h). By [AG, Chapter 2 ] we have
A straightforward calculation using Taylor expansion shows that for any unit length tangent vector u in \mathbf{B}_{r_g}(p,h), we have
for some c_{p,h}>0 which depends only on the C^3 norm of the ambient metric h, i.e.,
By compactness of M and shrinking \Omega_g if necessary, we can set
Arguing as in the first two sections of [Dar], we deduce from (Proof) that \exp(c_gr)V(\mathbf{B}_r(p,h))r^{-d} is monotonically increasing in r.
The second bullet follows from the first bullet and
Using the same notation as Lemma 4.1.1, for R=\mathbb{Z} or R=\mathbb{Z}/n\mathbb{Z} with n\ge 2 and all p\in M,h\in\Omega_g and any T\in\mathbf{MIN}_h^{R}\left([\Sigma]\right), we have
with C_{[\Sigma]}>0 depending only on [\Sigma] and \Omega_g, not on R.
Assume Assumption 2.2.1. Let T\in\mathbf{MIN}^R_\delta([\Sigma]) for R=\mathbb{Z} or R=\mathbb{Z}/n\mathbb{Z} with n\in\mathbb{Z}_{\ge 2}. There exists C_{[\Sigma]}>0 depending only on [\Sigma], not on R, so that for any point p\in\mathbb{R}^{d+c}, then \theta_T(p), the density of the R-mass measure of T at p, is at most \theta(B), i.e.,
Use \partial\Sigma to denote the unique \mathbb{Z}-flat cycle representing the class \partial[\Sigma]\in H_{d-1}(W,\mathbb{Z}).
Assume that c,d\in\mathbb{Z}_{\ge1}, and R=\mathbb{Z} or R=\mathbb{Z}/n\mathbb{Z} with n\ge 2. If W is a closed smooth (d-1)-dimensional submanifold of \mathbb{R}^{d+c} and B is a smooth (d-1)-dimensional integral cycle with \operatorname{Supp} B\subset W, and T is an R-area-minimizing flat chain with boundary
then for any point p\in\mathbb{R}^{d+c}, we have \theta_T(p)\le C_{B}, with C_{B} a positive real number depending only on B.
Let T\in\mathbf{MIN}^R_\delta([B]). We have
We will use both Allard’s interior and Allard’s boundary monotonicity formula for varifolds stationary outside W. The goal is to turn the uniform mass bound (Proof) into the uniform density bound (5.0.1)
First let us note that the mass measure V_T of T is a stationary integral varifold in W^\complement, since any diffeomorphism of \mathbb{R}^{d+c} which is identity on W must increase the mass. By Allard’s boundary monotonicity formula [WAb, (2) Theorem of Section 3.4], there exists s_{[B]}>0,c_{[B]}>0 depending only on B, so that for any 0<r\le s, and any p\in W,
is monotonically increasing. We want to remark that though Allard commented [WAb, Section 3.3] that he assumes the dimension of the varifold is at least 2, but the proof of [WAb, (2) Theorem of Section 3.4] did not use that assumption. Also, Allard did not assume that the boundary is connected. Thus, Allard’s boundary monotonicity formula does apply to all cases we consider in this proof.
This implies that
for any point p\in W.
On the other hand, for points p at least \frac{1}{2}s_{[B]} away from W, by Allard’s interior monotonicity formula in [WA, (2) Corollary of Section 5.1], we have
For p of distance at most \frac{1}{2}s_{[B]} from W, let q be a point on W minimizing |p-q|. Then \mathbf{B}_{|p-q|}(p)\subset \mathbf{B}_{2|p-q|}(q), by Allard’s interior monotonicity formula in [WA, (2) Corollary of Section 5.1], we have
On the other hand, by Allard’s boundary monotonicity formula [WAb, (2) Theorem of Section 3.4], we have
Bringing (Proof) into (Proof), we deduce that
Combining all three cases, (Proof) (Proof) and (Proof), we are done by setting
Assume that c,d\in\mathbb{Z}_{\ge1}, and R=\mathbb{Z} or R=\mathbb{Z}/n\mathbb{Z} with n\ge 2. If W is a compact (d-1)-dimensional subanalytic set of \mathbb{R}^{d+c} and B is a (d-1)-dimensional integral cycle with \operatorname{Supp} B\subset W, and T is an R-area-minimizing flat chain with boundary
then for any point p\in\mathbb{R}^{d+c}, we have \theta_T(p)\le C_{B}, with C_{B} a positive real number depending only on B.
For any point p\in\mathbb{R}^{d+c}, define \pi_p to be the orthogonal projection of \mathbb{R}^{d+c}\setminus p to the unit sphere \mathbf{S}_1(p) centered at p, i.e.,
For any point p\in \mathbb{R}^{d+c}, we have
Apply Gromov’s [MGfrm, Section 8.1-8.2] extended monotonicity formula proved in [EWW, Section 9.2, 9.3]. We deduce that [EWW, Theorem 1.3]
Here \operatorname{Cone}_p B^R denotes the cone over \partial \Sigma centered at p. Geometrically \operatorname{Cone}_p B^R is formed by unions of rays starting from p and passing through points in \operatorname{Supp} B^R, counted with multiplicities in R induced by \partial \Sigma. In other words, \operatorname{Cone}_p B^R defined is by the pushforward of B^R\times[0,\infty)\subset \mathbb{R}^{d+c}\times \mathbb{R} to \mathbb{R}^{d+c} under the map
By definition, \operatorname{Cone}_p B^R is homogeneous and the slice of \operatorname{Cone}_p B^R by the unit sphere centered at p is \pi_p\left( B^R\right), so we have
We are done
Use \left|B^R\right| to denote the maximum density of B^R, i.e.,
When we are in the second case of Assumption 2.2.1, use \left|[\Sigma]^R\right| to denote the maximum density of \partial \Sigma^R, i.e.,
\left|B^R\right| is finite and depends only on B^R.
Apply [GVvwl, Proposition 3.0.6], we deduce that \sup_{p\in\mathbb{R}^{d+c}}\theta_W(p) is finite. The proposed fact then follows from Lemma 3.11.1.
For any point p\in \mathbb{R}^{d+c}, we have
Let us first prove (5.2.1). Recall Fact 3.2.1. Fix a totally geodesic c-dimensional sphere S^c in the unit sphere \mathbf{S}^{d+c-1}(p) centered at p. Apply the spherical integral geometry formula [HF, 3.2.48 Theorem] with A= \pi_p(W), \alpha=\theta_{\pi_p\left( B^R\right)}, B=S^c, \beta=1, we deduce that
By definition of mass, we have
On the other hand, the other factors on the right hand side of (Proof) evaluates to
Thus, bringing (Proof) and (Proof) to (Proof), we deduce that (5.2.1) holds:
Now we go on to prove (5.2.1). By the area formula applied to mass, [LS, 27.2 Remarks (3)], we have
where \pi_p^{-1}(q)\cap W^+ means the intersection of \pi_p^{-1}(q) with points in W so that the Jacobian of \pi_p is nonzero, i.e., d\pi_p being full rank (d-1). We deduce that
Let P(g.S^c) denote the unique affine (c+1)-dimensional plane passing through p and g.S^c. We have
This implies that
Bringing (Proof) into (Proof), we deduce that
Apply Lemma 3.10.1 by translating p to the origin, we deduce that for Haar almost every g\in O(d+c), we have
For any point p\in \mathbb{R}^{d+c}, we have
When we are in the second case of Assumption 2.2.1, for any T\in\mathbf{MIN}^R_\delta([\Sigma]^R) we have
Define
Set B_{P,Q}=W_P+W_Q.
For p\in \mathbb{R}^{d+c}, and T\in\mathbf{MIN}^R_\delta([\Sigma_{P,Q}]) we have
and strict inequality holds for p\not=0.
By definition, we have
Recall Fact 3.10.1 with p translated to the origin. If an affine (c+1)-dimensional plane V intersects W_{P,Q} at finitely many points, then the cardinality \#|V\cap W_{P,Q}| is at most 4, i.e., V intersecting P,Q both at a line. Thus, we have
By Fact 5.2.4, we deduce that
To prove that strict inequality holds in (Proof) for p\not=0, we prove the contrapositive.
In other words, we will prove that equality in (Proof) implies p=0.
We need to recall the proof of Fact 5.2.4. Recall equality (Proof) Equality in (Proof) holds implies that for Haar a.e. g\in O(d+c), we have
If p\not\in P, then take a flat coordinate system (x_1,\cdots,x_{d+c}) so that P is spanned by x_1,\cdots,x_d-axes vectors and p=(p_1,\cdots,p_{d+c}) with p_{d+c}>0. Consider the affine (c+1)-dimensional plane P'_p parameterized by
Then P'_p\cap P=\emptyset. Thus,
and there exists an open set U(\pi_p(P'_p)) containing \pi_p(P'_p) so that
However, \pi_p(P'_p) is a totally geodesic c-dimensional sphere in \mathbf{S}_1(p), so there exists g_P so that g_P(S^c)=\pi_p(P'_p). For a neighborhood U(g_P) of g_P in O(d+c) we have h(\pi_p(P'_p))\subset U(\pi_p(P'_p)). In other words, for h\in U(g_P), we have \pi_P(W_P)\cap h(S^c)=\emptyset.
Thus, if (Proof) holds for Haar a.e. g\in O(d+c), we must have p\in V. The same argument works for Q and we deduce that if (Proof) holds then p\in Q. We are done.
Consider the quadruples of configures (M^{d+c},W,g,\Omega_g) with d,c\in\mathbb{Z}_{\ge 1}. Assume that one of the followings holds
(Assumption 2.1.1) M is a (d+c)-dimensional compact closed smooth manifold, W is the empty set, [\Sigma]\in H_d(M,W,\mathbb{Z}) is a d-dimensional integral homology class, and g is a smooth metric, \Omega_g defined in Lemma 4.1.1,
(Assumption 2.2.1) M=\mathbb{R}^{d+c} with d\ge 2, W is a (d-1)-dimensional compact closed smooth submanifold or a (d-1)-dimensional compact subanalytic set, [\Sigma]\in H_d(M,W,\mathbb{Z}) is a d-dimensional relative integral homology class, g=\delta is the standard flat metric and \Omega_g=\{g\}.
Assume that n\ge 3C_{[\Sigma]}. For any metric h\in \Omega_g, let
Then any \mathbb{Z}-rectifiable current S that is Federer’s “representative" modulo n of T satisfies
In other words, S is a relative cycle in H_d(M,W,\mathbb{Z}) and the singular set of S outside of W is of Hausdorff dimension at most (d-2).
Using the same notation as in Fact 6.0.1, there exists a positive real number N_{[\Sigma],g}, so that for n\ge N_{[\Sigma],g}, we have [S]\in[\Sigma]+nG_d(M,W,\mathbb{Z}). Furthermore, if \tau_d|n, then nG_d(M,W,\mathbb{Z})=0 and we have [S]=[\Sigma].
S\in\mathbf{MIN}_h^\mathbb{Z}([S]) and S\in \mathbf{MIN}_h^{\mathbb{Z}}([\Sigma]+nH_d(M,W,\mathbb{Z})).
Let Z be another integral current representing [S].
By definition of Federer’s “representative" modulo n (Fact 3.3.4), we have
On the other hand, by area-minimality of T and that the canonical homomorphism \mathbb{Z}\to \mathbb{Z}/n\mathbb{Z} is norm non-increasing, we deduce that
Combining (Proof) and (Proof), we deduce that S is \mathbb{Z}-area-minimizing in [S].
By Fact 3.3.4, we have
As [S]\in [\Sigma]+nH(M,W,\mathbb{Z}). We are done.
Set
Using the same notation as in Fact 3.3.4, S as a Federer’s “representative" modulo n is unique. In other words, if S' is another Federer’s “representative" modulo n of T, then we have S=S'.
First, S^{\mathbb{Z}/n\mathbb{Z}}=(S')^{\mathbb{Z}/n\mathbb{Z}}=T implies (S'-S)^{\mathbb{Z}/n\mathbb{Z}}=0. As S and S' has the same support and they both have boundaries supported in W, they must have the same regular set
In each connected component of \operatorname{Reg} S\cap \operatorname{Reg} S', S'-S equals a smooth submanifold counted with multiplicity divisible by n. by (?), S and S' has density strictly less than \frac{n}{2} Hausdorff d-dimensional a.e. . Thus, S-S' has density strictly less than n Hausdorff d-dimensional a.e. . This implies that S and S' coincides on their regular set. Since \operatorname{Sing} S\setminus W=\operatorname{Sing} S'\setminus W=\operatorname{Sing} has Hausdorff dimension at most (d-2), we deduce that S=S'.
Any \mathbb{Z}-rectifiable current S that is Federer’s “representative" modulo n of T determines a homology class \partial S\in H_{d-1}(W_{P,Q},\mathbb{Z}) provided n\ge 4.
a,b cannot be of of opposite signs.
The forms P^\ast and Q^\ast are calibration forms calibrating only P and Q, respectively.
Let T_n be a \mathbb{Z}/n\mathbb{Z}-area-minimizing flat chain with \partial T=B^{\mathbb{Z}/n\mathbb{Z}}.
There exists N_B depending only on B so that T_n is an integral current with \partial T_n=B for n\ge N_B.
A \mathbb{Z}-area-minimizing current in a non-zero torsion homology class can never be calibrated.
If d\ge 1,c\ge 2, [\Sigma] is a non-torsion class and M is an orientable, then for any \lambda>0, there exists a non-empty open set \Omega'_{[\Sigma],\lambda} in the space of Riemannian metrics, depending on [\Sigma] and \lambda, such that for any metric h\in \Omega'_{[\Sigma],\lambda}, we have
By Fact 3.8.1, it suffices to show the existence of one metric h in which inequality (10.1.1) holds.
Let N be a smoothly embedded connected submanifold representing \mu_{[\Sigma]}[\Sigma]. Apply Lemma 3.13.1 to N to obtain a metric h on M, tubular neighborhood U(N), and the deformation retract \pi_N.
By the last bullet of Lemma 3.13.1 and Fact 3.9.1, we have
By the first bullet of Fact 3.8.5, we have
Let T\in\mathbf{MIN}_h^{\mathbb{Z}}\left([\Sigma]\right). Since \pi is a deformation retract of U(N) onto N, by homotopy invariance of homology [AH], we deduce that H_d(U(N),\mathbb{Z})=\mathbb{Z}[N]=\mathbb{Z}[\mu_{[\Sigma]}\Sigma]. Thus, any integral current whose support is contained in U(N) is homologous to some integer multiple of \mu_{[\Sigma]}[\Sigma]. Since [\Sigma] is a non-torsion class, we deduce that
The support of T cannot be contained in U(N).
By the first bullet of Lemma 3.13.1, we deduce that
For any metric h on M, there exists a non-empty open set \Omega_{[\Sigma],h}\ni h in the space of Riemannian metrics on M, and an integer \nu_{[\Sigma],h} such that for any integer \nu\ge \nu_{[\Sigma],h} and any metric {\hat{h}}\in \Omega_{[\Sigma],h}, we have
By (3.8.5), there exists an integer \nu_{[\Sigma],h} such that for any \nu\ge \nu_{[\Sigma],h}, we have
There also exists an open set \Omega_{[\Sigma],h}\ni h in the space of Riemannian metrics on M, such that for any metric {\hat{h}}\ge \Omega_{[\Sigma],h}, we have
as quadratic forms, which by definition of mass [HF, p.358], implies that
for R=\mathbb{Z} or \mathbb{R}. Combining (Proof) and (Proof), by Fact 3.8.2, we have
Since \left(\frac{9}{8}\right)^3<\frac{3}{2}, we are done.
(The Hasse principle for variational problems) For a variational problem that can be formulated on chain spaces over \mathbb{R},\mathbb{Z} and \mathbb{Z}/n\mathbb{Z} for n\ge 2, the solutions in \mathbb{Z}-chains can be reconstructed from the corresponding solutions in \mathbb{R}-chains and \mathbb{Z}/n\mathbb{Z}-chains for all n \geq 2.
What are the obstructions to the Hasse principle for variational problems?
If d\ge 1,c\ge 2, [\Sigma] is a non-torsion class and M is an orientable, and K is the infinite subset of \mathbb{Z}_{\ge 2} defined in Lemma 3.1.2, then for any two integer k,l\in K with k<l and any real numbers \lambda>1, there exist a non-empty open set \Omega'_{[\Sigma],k,l,\lambda} in the space of Riemannian metrics, depending on [\Sigma], k, l, and \lambda, such that for any metric h\in \Omega'_{[\Sigma],k,l,\lambda}, we have
We have
The denominator of (12.4.1) equals
Using the notation of Lemma 4.1.1, Subsection ? and Fact 7.1.1 show that whenever
we have
Expand [\Sigma] with respect to the basis in Assumption 3.1.1, we have
for \bigcup_{j=1}^{b_d}\{a_j\}\subset \mathbb{N}, and \bigcup_{i\in I}\{b_i\}\subset \mathbb{Z}\cap (-\frac{n}{2},\frac{n}{2}]. Apply Lemma 3.8.3 and Fact 3.8.6, we obtain
This implies whenever
equation (Proof) holds. This implies that the numerator of (12.4.1) is at least
We conjecture that the value of the left-hand side of (12.4.1) is 1.
Define
\mathbf{LavGap}([\Sigma],R) is always a (possibly empty) open set.
\mathbf{LavGap}([\Sigma],\mathbb{R}) is the set of metrics g where no element of \mathbf{MIN}_{g}^{\mathbb{Z}}([\Sigma]) can be calibrated.
(Camillo De Lellis) The open set \mathbf{LavGap}([\Sigma],\mathbb{R}) is dense in the space of smooth Riemannian metrics.
If [\Sigma]\in H_d(\mathbb{CP}^{\frac{d+c}{2}},\mathbb{Z}), the Fubini-Study metric lies in the closure of \mathbf{LavGap}([\Sigma],\mathbb{R}).
The class 2[\Sigma] always admits a connected embedded smooth complex algebraic subvariety N as a representative. For instance, one can take Fermat hypersurfaces in an equator \mathbb{CP}^{\frac{d}{2}+1} of \mathbb{CP}^{\frac{d+c}{2}}. Then by [HL, page 70 Example I], N is \mathbb{Z}-area-minimizing in the Fubini-Study metric g_{FS} and calibrated by the form \frac{1}{(d/2)!}\omega^d, where \omega is the standard Kahler form in the Fubini-Study metric. Let \beta be a smooth function on \mathbb{CP}^{\frac{d}{2}+1} such that \beta\ge 1 and \beta^{-1}(1)=N. Since \beta g_{FS}\ge g_{FS} as quadratic forms, by the definition of comass in (?), in metric \beta g_{FS}, the form \frac{1}{(d/2)!}\omega^d still calibrates N. This implies that
Since \frac{1}{(d/2)!}\omega^d have comass 1 in \beta g_{FS} only on N, by the constancy theorem [FMgmt, Section 4.9], any integral current calibrated by \frac{1}{(d/2)!}\omega^d in \beta g_{FS} must be an integer multiple of N. As [\Sigma] is not a torsion class, we deduce that
N is the unique \mathbb{Z}-rectifiable current in 2[\Sigma] calibrated by \frac{1}{(d/2)!}\omega^d.
If \left\lVert[\Sigma]\right\rVert^{\mathbb{R}}_{\beta g_{FS}}=\left\lVert[\Sigma]\right\rVert^{\mathbb{Z}}_{\beta g_{FS}}, then by (Proof), any element T\in\mathbf{MIN}_{\beta g_{FS}}^{\mathbb{Z}}([\Sigma]) must satisfy \mathbf{M}_{\beta g_{FS}}^{\mathbb{Z}}(T)=\frac{1}{2}\mathbf{M}_{\beta g_{FS}}^{\mathbb{Z}}(N). Thus, we have 2T\in 2[\Sigma] and 2T has the same area as N. By Fact 3.9.1, this implies that 2T is also calibrated by \frac{1}{(d/2)!}\omega^d, which contradicts Fact Proof. Replacing \beta by 1+\epsilon(\beta-1) for positive \epsilon\to0. We are done.
[FMvbd, 3.3 Lemma] A d-dimensional \mathbb{Z}/n\mathbb{Z}-area-minimizing \mathbb{Z}/n\mathbb{Z}-flat chain in \mathbb{R}^{d+c} has Euclidean volume growth and has a tangent cone at infinity.
If a d-dimensional \mathbb{Z}-area-minimizing integral current T in \mathbb{R}^{d+c} has Euclidean volume growth, then the \mathbb{Z}/n\mathbb{Z}-flat chain T^{\mathbb{Z}/n\mathbb{Z}} is \mathbb{Z}/n\mathbb{Z}-area-minimizing for n large enough.
[FMwtc, Problem 16]A pairs of oriented planes is \mathbb{Z}/2\mathbb{Z}-area-minimizing if and only if the pair is \mathbb{Z}-area-minimizing with any orientations on P,Q.
(Morgan) A pairs of planes is \mathbb{Z}/3\mathbb{Z}-area-minimizing if and only if the pair is \mathbb{Z}-area-minimizing and the sum of characterizing angles is at least \frac{2}{3}\pi.
(The Hasse principle for p-adic variational problems) For a variational problem that can be formulated on chain spaces over \mathbb{Q}_p,\mathbb{Z}_p and \mathbb{Z}/p^n\mathbb{Z} for n\ge 1, all equipped with the p-adic norm, the solutions in \mathbb{Z}_p-chains can be reconstructed from the corresponding solutions in \mathbb{Q}_p-chains and \mathbb{Z}/p^n\mathbb{Z}-chains for all n\ge 1.
\mathbb{R}/\mathbb{Z}-area-minimizing flat chains represents the asymptotic behavior of \mathbb{Z}/n\mathbb{Z}-area-minimizing flat chains as n\to\infty, just as \mathbb{R}-area-minimizing flat chains represents the asymptotic behavior of \mathbb{Z}-area-minimizing flat chains.
A d-dimensional \mathbb{Z}/n\mathbb{Z}-size-minimizing flat chain T satisfies
([BWfrfc]) the d-th stratum in the Almgren stratification of T is a smooth open not necessarily connected d-dimensional submanifold,
([FMmcr, LScy]) near each point in the (d-1)-th stratum in the Almgren stratification of T, the support of T decomposes into three d-dimensional submanifolds with boundaries meeting along a (d-1)-dimensional submanifold at equal angles of \frac{2\pi}{3}, i.e., forming Y-shaped triple junctions.
(The Hasse principle for size-minimizing flat chains) \mathbb{Z}-size-minimizing representatives of an integral homology class [\Sigma] exist and can be reconstructed from \mathbb{R}-size-minimzing representatives of [\Sigma]^{\mathbb{R}} and \mathbb{Z}/n\mathbb{Z}-size-minimizing representatives of [\Sigma]^{\mathbb{Z}/n\mathbb{Z}} for n\in\mathbb{Z}_{\ge 2}.
The text on this page is from arXiv:2508.21045v6 by Zhenhua Liu, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2508/2508.21045.jsonl.