On area-minimizing subgraphs in integer lattices
Abstract.
We introduce area-minimizing subgraphs in an infinite graph via the formulation of functions of bounded variations initiated by De Giorgi. We classify area-minimizing subgraphs in the two-dimensional integer lattice up to isomorphisms, and prove general geometric properties for those in high-dimensional cases.
Mathematics Subject Classification 2010: 05C10, 28A75, 52C99.
1. Introduction
Area-minimizing submanifolds in the Euclidean space are important concepts in geometric measure theory. For the co-dimensional one case in , De Giorgi [DG61] initiated a formulation using functions of bounded variations, called an area-minimizing hypersurface or a boundary of a subset of least perimeter, see [Giu84]. A celebrated result is as follows.
Theorem 1.1 (Simons [Sim68]).
Every area minimizing hypersurface in for is flat, i.e. a boundary of a half space.
The following equation is called the minimal surface equation
(1) |
Bernstein first proved that any entire solution of the minimal surface equation on is affine, see [Ber27]. Such a statement on the triviality of entire solutions is now called the Bernstein theorem. By the observations of Fleming [Fle62] and De Giorgi [DG65], the Bernstein theorem of the minimal surface equation is reduced to the classification of area-minimizing hypersurfaces or area-minimizing cones in the Euclidean space. This leads to the following Bernstein theorem: any entire solution of the minimal surface equation on is affine if and only if for which the sharpness follows from the construction of a non-affine solution on by Bombieri, De Giorgi, and Giusti [BDGG69].
We recall the basic setting in A function is called of locally bounded variations in if and whose distributional derivative is a Radon measure in We denote by the space of functions of locally bounded variations in A Lebesgue measurable set is called a Caccioppoli set if where is the indicator function on A Caccioppoli set is called area-minimizing in if for any open set any Caccioppoli set with
where , is the Lebesgue measure, and is the BV seminorm. This in fact means that is an area-minimizing hypersurface. This is equivalent to that for any open set and with a.e. on
Such a function is called of least gradient in which has been extensively studied in the literature, see e.g. [BDGG69, Mir67, SWZ92, Juu05, MRSdL14, Mor17, JMN18, G1́8, Mor18, FM19, Zun19].
Let be a simple undirected graph with the set of vertices and the set of edges Two vertices are called neighbors, denoted by if there is an edge connecting and i.e. For any subset , denote by the complement of , by the vertex boundary of and by the exterior vertex boundary of . We write Given any we set
We say is the edge boundary of and set . We introduce discrete analogs of a subset of least perimeter and an area-minimizing subset.
Definition 1.2.
For a finite subset a subset is called of least perimeter in if for any with we have
where denotes the cardinality of a set.
Definition 1.3.
A proper nonempty subset is called area-minimizing in if for any finite , is of least perimeter in In this case, we identify the subset with its induced subgraph on and call a minimal subgraph in for convenience.
Remark 1.4.
The subset is called minimal if and only if for any finite with , then
For any we define
For finite the 1-Dirichlet energy on is given by, for any ,
For a subset any antisymmetric function on i.e. for any with is called a current on For a function we call the current is a current associated with on if
where
Let be the set of currents associated with on The 1-Laplacian is defined as a set-valued mapping
(2) |
A current is called minimal in if for all The following is well-known using the results in convex analysis [Roc70, RW98], see e.g. Chang [Cha16] and Hein-Bühler [HB10].
Proposition 1.5.
For finite and a given function the subdifferential of the functional at in is given by
The discrete analog of functions of least gradient was introduced in metric random walk spaces including graphs by Mazón, Pérez-Llanos, Rossi, and Toledo [MPLRT16], see also [GM21, MT23, MSDTM23].
Definition 1.6.
For finite a function is called of least gradient in if for any with , then
The following are the characterizations of minimal subgraphs.
Theorem 1.7.
For the following are equivalent:
-
(1)
is a minimal subgraph in
-
(2)
For any finite is of least gradient in
-
(3)
Now we turn to minimal subgraphs in the integer lattices. We denote by the graph of the -dimensional integer lattice consisting of the set of vertices and the set of edges
The first main result is the following statement.
Theorem 1.8.
Any minimal subgraph in is exactly one of the following subgraphs up to isomorphisms:
- (1)
- (2)
-
(3)
19 families of minimal subgraphs with geodesic simple boundary. More precisely, there are 7 families of connected minimal subgraphs: 5 fimilies all have exactly one connected component of boundaries; see Fig.15-Fig.15; 2 families both have exactly two connected of boundaries; see Fig.17, Fig.17. There are 12 families of disconnected minimal subgraphs: they are all complementary subgraphs of connected minimal subgraphs.
The classification of minimal subgraphs is quite complicated. The subtleties mainly lie in two aspects: 1. The boundaries of minimal subgraphs may not be geodesic. 2. The boundary of minimal subgraphs may not be simple. For the first aspect, we find that all boundaries of minimal subgraphs consist of geodesic rays instead of geodesic lines; for example see Claim 3.21. For the second aspect, the boundary of any minimal subgraph contains at most one unit square and distributes diagonally in some way; see Corollary 3.20 and Lemma 3.12. The arguments are based on topological, combinatorial and coarsely geometric methods.
The proof strategies of Theorem 1.8 are as follows. There are three key observations for the proof: Corollary 3.5, Lemma 3.7, and Lemma 3.9. Corollary 3.5 yields some geodesic convexity of minimal subgraphs, and implies that the boundary of any minimal subgraph determines this minimal subgraph itself in some sense. So the crucial point is to characterize the boundary of any minimal subgraph. Lemma 3.7 is a useful tool to describe the geometry of oriented boundary of any minimal subgraph, which suggests that the boundary behaves like a simple path in general. Lemma 3.9 gives a strong geometric restriction of the boundary of any minimal subgraph, which applies to deduce a strong property, Corollary 3.19, that the number of connected components of a minimal subgraph and its boundary is at most two. By the results in Lemma 3.1 and Lemma 3.19, it suffices to consider connected minimal subgraphs with at most two connected boundary components. There are two steps: First, we use several important properties to describe the geometry and topology of the boundary of any minimal subgraph (whether the boundary is geodesic or simple). Second, we use the arguments of currents to confirm the precise structure of boundary and the minimal subgraph; see Corollary 2.3 and Lemma 3.15.
On the other hand, the classification of high-dimensional minimal subgraphs(i.e. in ) seems much more complicated. The reasons are as follows: The result as in Lemma 3.7 fails in higher dimension, see Fig.18; the analogs of Corollary 3.5 and Lemma 3.9 in higher dimension don’t provide enough information about the local geometry.
Nevertheless, we can decompose any three dimensional minimal subgraph into a minimal sub-subgraph (union of three skeleton and two skeleton) and its one skeleton.
Theorem 1.9.
For a minimal subgraph , is minimal.
Recall that is called a rough isometry if
for any in and some positive constant . Note that this is stronger than usual definitions in [Woe00, BBI01]. For a minimal subgraph , we know does not always inherit the minimality of . But it shares the coarse geometry of and its boundary. We prove the following result.
Theorem 1.11.
For a minimal subgraph , the natural embedding (with the induced metric) is a rough isometry between metric spaces. Moreover, there exists a positive constant depending only on such that for sufficiently large and any
(3) | |||
(4) |
where denotes the -normed ball centered at of radius
We expect that determines the asymptotic geometry of and its boundary, so that the inequalities (3), (4) in Theorem 1.11 could be possibly improved, see Problem 5.1.
Moreover, we prove some restriction on the geometry of dimensional minimal subgraphs.
Theorem 1.12.
If is minimal, then there exist no two parallel hyperplanes bounding .
We remark that the conclusions in Section 4 are independent of those of Section 3, and note that the maximum principle holds for minimal subgraphs in in some sense; see Proposition 4.8.
The organization of this paper is as follows. In Section 2, we introduce some basic concepts and properties of minimal subgraphs. In Section 3, we focus on two dimensional minimal subgraphs and prove Theorem 1.8. In Section 4, we study the geometry of high-dimensional minimal subgraphs and prove Theorem 1.9. In Section 5, we list some open problems on geometry and topology of high-dimensional minimal subgraphs.
2. preliminaries
Let be a simple, undirected graph. For each edge we write and for associated directed edges. We say is connected if for any , there is a path connecting and for some For we denote by the vertex degree of in Usually, we consider the subgraph induced on a subset for which the degree of a vertex refers to that in The combinatorial distance or simply unless specially stated on the graph is defined as, for any and
For any function we consider the functional with Dirichlet boundary condition given by
where such that
Note that is of least gradient if and only if is the minimizer of One readily sees that if is of least gradient in then is of least gradient in A similar result holds for sets of least perimeter.
The following is the discrete co-area formula, see [Bar17].
Proposition 2.1.
For any function
where
Now we prove the following proposition.
Proposition 2.2.
For finite and the following are equivalent:
-
(1)
is of least perimeter in
-
(2)
is of least gradient in
-
(3)
Proof.
(1)(2): Consider any with For any Since is of least perimeter in
By the co-area formula,
(2)(1): For any with the result follows from the least gradient property of by choosing
(2)(3): is of least gradient in if and only if is the minimizer of Since is a convex function on is the minimizer if and only if is in the subdifferential of at which is by Proposition 1.5.
∎
We write for the ball of radius centered at
Proof of Theorem 1.7.
(1)(2): This follows from Proposition 2.2.
(2)(3): Consider a sequence of balls where for some For it follows from Proposition 2.2,
That is, for each there is a minimal current associated with on Since there are countable edges in and
there is a subsequence and a current on such that
One easily sees that and is minimal. Hence
(3)(2): For any minimal one easily verifies that for any finite which is also minimal in
This proves the theorem. ∎
Now we introduce some notions on . For , we say these vertices are horizontal (vertical,resp.) or is horizontal (vertical,resp.) to if they are in a horizontal (vertical,resp.) line. We say is left-horizontal (right-horizontal,up-vertical,down-vertical,resp.) to or a left-horizontal (right-horizontal,up-vertical,down-vertical,resp.) neighbor of if is horizontal (vertical,resp.) to and is on the left(right,up,down,resp.) of .
Given any two paths , is called flat if its vertices are all horizontal or vertical. The vertex is called a corner of if the subpath is not flat. We say a path is simple if all vertices have one or two neighbors in the path.
Given any path containing , is called a projective horizontal (vertical, resp.) interior point if there is a horizontal (vertical,resp.) line such that the distance projection (with respect to ) image of lies in the interior of that of ; see Fig. 21. Assume the path , we call is an isolated path in if for for the induced subgraph ; see Fig. 21.
By Theorem 1.7, we have the following.
Corollary 2.3.
is minimal if and only if .
Remark 2.4.
Given a minimal graph , we associate it with a geometric space . To be precise, we identify the graph structure with the natural corresponding 1-skeleton in , i.e. identify with , and associate any loop with a unit square enclosed by the loop. The boundary is called simple if for any , there are at most two neighbors of in . is called geodesic if holds for all
For the edge is called a boundary edge of if there is at most one unit square in containing . A vertex is called a boundary vertex if it is contained by a boundary edge. A boundary path is called oriented in oriented in short, if and contains one of the following subgraphs for any subpath ; see Fig. 23. If we equip with an orientation, is called right (left,resp.) oriented if these subgraphs for the subpaths lie on the right (left,resp.) hand side of ; see Fig. 23.
3. geometry of two dimensional minimal subgraphs and proof of Theorem 1.8
In this section, we write for a minimal subgraph in
Lemma 3.1.
If is minimal, then so is the complementary subgraph .
Proof.
This is direct by definition and Corollary 2.3. ∎
Lemma 3.2.
can not contain an isolated geodesic path with length .
Proof.
If not, we may assume is an isolated geodesic path in with . Then one can remove this path except , and obtain the new subgraph . Taking , one easily sees that
This is impossible since is minimal.
∎
Lemma 3.3.
can not contain an isolated point(i.e. it has only one neighbor in ). As a consequence, is locally one of the three subgraphs and any boundary vertex has at least two boundary neighbors in ; see Fig. 24.
Proof.
If not, we get another subgraph by deleting the isolated point . It is clear that , where . This contradicts the minimality of ∎
Lemma 3.4.
If there is a finite non-closed simple path in lying on one side of a horizontal/vertical line and the line contains two endpoints of the path, then the domain enclosed by the line and the path is contained in .
Proof.
We argue by contradiction.
We may assume the path lies above on the horizontal line such that . Denote by the domain enclosed by the line and the path.
If is not in , then one can find such that the domain enclosed by the horizontal line through (except the line ) is contained in , and One can get another subgraph by adding to , then one easily deduces that
This contradicts the minimality of .
∎
Corollary 3.5.
If lie in a horizontal/vertical line and in the same connected component of , then the horizontal/vertical segment between and is contained in
Proof.
Corollary 3.6.
Any horizontal/vertical line cannot intersect two boundary edges and one other edge consecutively in a connected component of i.e. there cannot exist two boundary edges and one other edge satisfying that and lie in two horizontal/vertical lines in the given order.
Proof.
If not, one can deduce that the rectangle is contained in by Corollary 3.5. This is impossible, since is a boundary edge.
∎
Lemma 3.7.
If there is a finite simple right (left,resp.) oriented boundary path in , and is a projective horizontal (vertical,resp.) interior point in with . Then there is no vertex in on the left(right,resp.) hand side of such that it is adjacent and vertical (horizontal,resp.) to in .
Proof.
Otherwise, we may assume there is a vertex on the left side of with . By Lemma 3.3, there is a vertex .
Case 1. If are horizontal/vertical, then it is clear are vertical/horizontal. Furthermore, are also vertical/horizontal. If not, we can assume are vertical/horizontal. Combining Corollary 3.5 and the condition that is left oriented, one can deduce that is contained in two unit squares in and it is not a boundary edge of . It is a contradiction. Applying Lemma 3.3 again, there is a vertex . By the same argument, we have that are vertical/horizontal. Continuing the process, we finally get an isolated vertical/horizontal ray in . This contradicts the minimality of by Lemma 3.2.
Case 2. If are not horizontal/vertical, we may assume that are horizontal and are vertical. Recall that is projective interior vertex in , so that we may assume are vertical and are horizontal for some . If are horizontal, then and are both vertical. Combining Corollary 3.5 with the condition that is left oriented, one can deduce that is not a boundary edge of . It is a contradiction. Hence are vertical. Using same arguments in Case 1, one can get a vertical ray in . This contradicts the minimality of by Lemma 3.2. ∎
Remark 3.8.
The proof of Lemma 3.7 is valid for more general case and it is used frequently throughout this section.
Lemma 3.9.
can not contain two parallel horizontal or vertical rays.
Proof.
If not, we may assume that are two horizontal rays originating from two vertical vertices respectively with . Denote by the strip bounded by . So one can find subjected to , where is some positive integer. There are two cases.
Case 1. are in the same connected component of . Then contains by Corollary 3.5. One can obtain another graph via removing the rectangle . Note that have edges not in by Lemma 3.7. Therefore, this yields that
where . This is a contradiction by the minimality of
Case 2. are in different connected components of . We may assume that is above . For each with integer , there exist with
(5) |
Let . Then we get another graph by filling a rectangle . Note that these are boundary vertices of . Therefore, by using (5) one can show that
where . This contradicts the minimality of ∎
Corollary 3.10.
Any two disjoint infinite simple boundary paths in can not be contained in some unbounded infinite strip isometric to , where is some positive constant.
Proof.
The proof is similar to that of Lemma 3.9. Assume that is the line through . We may assume are two disjoint infinite boundary simple paths bounded by two horizontal rays and are vertical. Let be the vertical line through , for some integer .
We only prove the case that are in the same connected component of and the others are similar. It follows that the strip bounded by is contained in by Corollary 3.5. Note that there is at least one edge not in for with by Corollary 3.7. Therefore, we have
where and is obtained by removing from . This contradicts the minimality of ∎
Proof.
By Corollary 3.5, we have that for the first subgraph in Fig. 25 the rectangle and then is not a boundary path. It is a contradiction. For the second one in Fig. 25, applying Lemma 3.7 to the projective vertical interior point in the oriented path , this yields that or is not a boundary path. It is a contradiction.
For the third one in Fig. 25, we have
Claim 3.12.
The connected component containing of and the interior of the domain (or ) are disjoint; see Fig. 26.
Proof of Claim 3.12.
By Corollary 3.5 and Corollary 3.7, we deduce that and the interior of the domain are disjoint. By symmetry of the domian and , one can assume that there is a vertex lying in the interior of the domain . By definition of , it is easy to obtain that there is a vertex (distinct to or respectively) vertical or horizontal to or respectively. Then using Corollary 3.5 again, we get that or and therefore contains a unit square containing . This is impossible since is a boundary edge. ∎
Assume that is the connected component of in the domain . Now we consider the boundary .
Case 1. is finite. Using Lemma 3.3, we can find a closed simple path to enclose . By Corollary 3.5, we have and that is the domain enclosed by . Taking , by comparing the edge boundary of with that of , we get that
where is obtained by removing except from . This is impossible by the minimality of
Case 2. is infinite. Applying Corollary 3.10, we have that contains at most one infinite half simple path. If contains one infinite half simple path , then by Corollary 3.6 one deduces that only have finite backtracks in the direction of the line containing . Hence one can get that the subpath which is sufficiently far away from of is a geodesic ray. This contradicts the minimality of by Lemma 3.2. ∎
Lemma 3.13.
If there is a simple loop in , then the loop is a unit square.
Proof.
Note that the domain enclosed by any closed simple loop in is contained in by Corollary 3.5. Let be a simple loop in and be the domain enclosed by We may assume that is the smallest rectangle containing . Using Corollary 3.5, we have , where are the highest, the leftmost, the lowest, the rightmost subpath of
Since is a simple loop and for the smallest rectangle containing we can assume that are arranged counterclockwise on the loop . Let be the boundary subpaths connecting and , and , and , and in , respectively.
We claim that are geodesic. It suffices to prove the result for . Suppose it is not true, we may assume that and are horizontal, and are vertical. If is on the left of , then by the fact that is the leftmost, there is a boundary edge for some integer such that and are vertical. This is impossible by Corollary 3.6. If is on the right of , then the square is contained in by Corollary 3.5. This implies is not a boundary edge and it is a contradiction. Thus we prove the claim.
It follows from the above claim that if and only if are pairwise horizontal or vertical. Note that have other neighbors in exactly adjacent to the subset by Lemma 3.7.
Case 1. At least three of hold. It is clear that by Lemma 3.7. This obviously yields that
where and the new subgraph is obtained by removing from except . This contradicts the minimality of
Case 2. Two or one of hold. We only need to prove the result for the subcase that one of holds, since the other case is similar to those of Case 1 and the previous subcase. Now we may assume and . Applying Lemma 3.7 and Lemma 3.11, we have and if . Therefore we get
where and the new subgraph is obtained by removing from except . This contradicts the minimality of
Case 3. None of holds, i.e. . Then is a rectangle with . By Lemma 3.7, Lemma 3.11, and or . Observing that is minimal, we deduce that
where and the new subgraph is obtained by removing from except . This implies and the result follows.
∎
Corollary 3.14.
Given any connected component of , if contains some loops, then these loops are exactly one unit square.
Proof.
Otherwise, we may assume that there are two unit distinct squares , in by Lemma 3.13. Suppose that and are arranged anticlockwise on respectively, and are the leftmost (rightmost, lowest, highest, resp.) in , respectively.
Recall that , are in the connected component . Applying Corollary 3.6, we may assume that and there is a simple boundary path realizing in the domain ; see Fig. 27.
Observe that both have exactly two neighbors in , by the proof of Lemma 3.7 for the simple boundary path with right oriented edges . Similar arguments yield that both have exactly two neighbors in .
Let be the rectangle containing and be the subgraph obtained by removing the part of in except . Using Lemma 3.6 and recalling that all have exactly two neighbors in , we have
where . This contradicts the minimality of
∎
Lemma 3.15.
Assume that is an oriented boundary geodesic line in , then there are two geodesics (denoted by if the geodesic does not exist) such that minimal currents in the region enclosed by and is determined by ; see Fig. 28-31. Moreover, the region is also minimal and there is an oriented boundary geodesic line if are not both .
Proof.
The first statement can be deduced by the equation (2) and Corollary 2.3. The currents in yileds that . Using Corollary 2.3 again, we have that is minimal. Applying Lemma 3.7, one can obtain that is an oriented boundary geodesic line (ray, line, resp.) in Fig. 28 (29, 30, resp.), and is an oriented boundary geodesic ray in Fig. 29. Thus, we get that ( resp.) is an oriented boundary geodesic line in Fig. 28 (Fig. 29, Fig. 30, resp.). ∎
Corollary 3.16.
Proof.
We argue by contradiction. Note that .
If is of the form in Fig. 28 or Fig. 30, and we assume that is a vertical line. Then it is easy to deduce that , which is impossible. If is also a horizontal line, then we can get a new minimal subgraph containing two boundary horizontal lines by Lemma 3.15. This is impossible by Lemma 3.9.
If is of the form in Fig. 29, then we can get a new minimal subgraph containing two boundary horizontal rays by Lemma 3.15. This is impossible by Lemma 3.9.
If is of the form in Fig. 31, then it always holds that . This is impossible.
Case 2. are both of the form in Fig. 31. Since , divide the plane into three domains , where is bounded by and ; see Fig. 32.
Thus the currents determined by are inconsistent with these determined by by Lemma 3.15, i.e. .
∎
Lemma 3.17.
Assume contains no loops. Let be a boundary geodesic line in with exactly one corner . Then the connected component of containing in a quadrant region enclosed by is or .
Proof.
If not, then one can find a vertex . We may assume that , where are flat geodesic rays from ; see Fig. 33. By Corollary 3.5, two flat geodesic rays from which don’t intersect with ; see Fig. 33.
Then is bounded by and .
Since , there is another bi-infinite simple boundary path with . Since has no loops, one can find an infinite simple subpath such that Thus we can find two disjoint infinite simple boundary paths of in the unbounded strip enclosed by and or enclosed by and . This is impossible by Corollary 3.10.
∎
Corollary 3.18.
Assume contains no loops. Let be a boundary geodesic line in with finitely many corners. Then the connected component of containing in a quadrant region enclosed by is or .
Proof.
The argument of Lemma 3.17 still works with minor modifications. ∎
Lemma 3.19.
Assume that is geodesic and contains no loops. Then both and have at most two connected components.
Proof.
If has more than two connected components, then there are three oriented disjoint boundary geodesic lines by Lemma 3.3 and Lemma 4.3. Using Corollary 3.16, one easily sees that there are two disjoint parallel boundary geodesic rays in a bounded strip. It is a contradiction by Lemma 3.9.
∎
Now we can prove Theorem 1.8.
Proof of Theorem 1.8.
We divide it into three cases.
Case 1. is non-geodesic.
We may assume that contains a path such that is not a boundary path with , and are horizontal; see Fig. 34. Then the rectangle determined by is contained in by Corollary 3.5.
If and , then this contradicts Lemma 3.11 by symmetry of and
If and , then we get the new subgraph by removing the rectangle except . By Lemma 3.7, we have
where . This contradicts the minimality of
Thus, either or occurs. So that we can assume and . Denote by the connected component of containing and the part of in the domain ; see Fig. 35.
Step 1 of Case 1. is enclosed by two geodesic rays from in the domain .
Claim 3.20.
contains no loops.
Proof of Claim 3.20.
Otherwise, contains exactly one unit square by Corollary 3.14. We can assume that are arranged anticlockwise, are horizontal, and is on the left of . By Lemma 3.6, there is a boundary simple path connecting in the domain . Applying Lemma 3.7 to the boundary simple paths and , we have
Consider the new subgraph by removing the part in the rectangle determined by from except Since , we have
where . This contradicts the minimality of ∎
Using Claim 3.20 and Lemma 3.2, we may assume that is the leftmost(rightmost,resp.) simple infinite path in such that is a path connecting and . Let be a vertical(horizontal,resp.) line through ; see Fig. 36. Applying Corollary 3.10, the distance projection of on is an infinite ray.
Claim 3.21.
are geodesic, and the length of is at most one.
Proof of Claim 3.21.
If is not geodesic, then there are four cases for some non-geodesic boundary subpath If is right-horizontal to , is up-vertical to , then there is a boundary edge such that and are vertical. This contradicts Corollary 3.6.
If is down-vertical to , is left-horizontal to , then by that fact that the distance projection of on is an infinite ray, there is an edge such that and are horizontal. This contradicts Corollary 3.6.
If is left-horizontal to , is up-vertical to , then we have by Corollary 3.5. This contradicts that is leftmost.
If is up-vertical to , is left-horizontal to , then we have by Corollary 3.5. This contradicts that is leftmost.
Hence we have that is geodesic. The same argument yields that is geodesic.
Claim 3.22.
The domain enclosed by in domain is contained in As a consequence, we have
Proof of Claim 3.22.
By same arguments in Lemma 3.17, one can show that . By definition, implies that is an isolated path. It is impossible by Lemma 3.2. Thus we prove the result.
∎
Similar arguments yield that is the domain enclosed by two geodesic rays and from , where the subpath , the subpath and .
Claim 3.23.
At most one of and has corners. If (,resp.) has corners, then (,resp.) contains exactly two corners and the two corners are adjacent.
Proof of Claim 3.23.
We argue via Corollary 2.3 and the equation (2). We give only some arguments and the remaining are similar. If both and have corners, then the current flowing out of ; see the first picture in Fig. 37 for . This is a contradiction by Corollary 2.3. Note that if has corners, then it has at least two corners by Lemma 3.9. On the other hand, has at most two corners by Corollary 2.3; see Fig. 38. The remaining claims are easy to verify by Corollary 2.3; see the last two pictures in Fig. 37 and Fig. 39.
∎
Claim 3.24.
For and , denote by the rectangles containing and , and respectively, and let and , and represent sides of respectively. Let be the length of respectively; see the graph in Fig. 40. Then we have that In particular, (,resp.) has at most four (two,resp.) corners. If has two corners, then two corners in and are both adjacent.
Proof of Claim 3.24.
Claim 3.25.
is connected.
Proof of Claim 3.25.
If not, let be another connected component of and Then is geodesic. Otherwise, contains two quadrants located diagonally such that by Claim 3.23 and Claim 3.24, which is impossible. So that one can find a geodesic line in by Lemma 3.3 and Lemma 4.3. Recalling Corollary 3.16 for and or and , we have two disjoint geodesic rays in bounded infinite strip, which contradicts Corollary 3.10. ∎
Thus, must be one of the forms in (1) of Theorem 1.8 by Claim 3.21, Claim 3.23, Claim 3.24 and Claim 3.25. On the other hand, one can use the argument of currents to show that these forms in (1) of Theorem 1.8 are indeed minimal by Corollary 2.3.
Case 2. is geodesic and simple.
Subcase 2-1 The number of connected components of is one, i.e. is connected.
Subcase 2-1-1 If is connected, then one easily deduces that is one of the forms in Fig. 15-Fig. 15 by Corollary 2.3 and the condition that is geodesic and simple.
Subcase 2-1-2 If is not connected, then recalling the condition of Case 2, we obtain that has exactly two geodesic and simple connected components by Lemma 3.19. These yield that are oriented and of the form in Fig. 29 by Corollary 3.16.
Subcase 2-1-2-1 If is of the form in Fig. 17, then we have the following claim.
Claim 3.26.
In this case, is minimal if and only if
Proof of Claim 3.26.
Since is of the form in Fig. 17, is minimal if and only if the new subgraph in Fig. 41 is also minimal by Lemma 3.15. If we remove the rectangle to get another new graph , then we have
where . This implies that
On the other hand, we shall prove that is minimal if We argue it by induction on . If , then we get the result by Corollary 2.3; see Fig. 42. To be precise,
for , the currents of can be arbitrarily given (for example, zero currents) whenever they are of sum zero along respective lines; see in Fig. 42. For the currents of can be arbitrarily given whenever they are of sum zero along respective lines. The currents of are indicated in Fig. 42; see in Fig. 42. For one can see in Fig. 42 to find that the minimal currents of are indicated.
Now let the current from to be as in Fig. 41. Then we obtain the currents in Fig. 41 by Corollary 2.3 and the currents in the interior of the upper half plane bounded by the horizontal line through and , which are indicated by crossed arrows in Fig. 41. Denote by the vertical ray from in , and the horizontal ray from through . Then one gets a new connected graph bounded by and ; see Fig. 41. Note that , and one can reverse the currents of the geodesic line . One can show that this preserves that . Thus, by applying induction assumption we finish the proof.
∎
Subcase 2-1-2-2 If is of the form in Fig. 17, then is minimal if and only if the new subgraph in Fig. 43 is minimal by Lemma 3.15. Now we show that in Fig. 43 is indeed minimal by Corollary 2.3. More precisely, divide the plane into three regions and ; see Fig. 43. The currents in the interiors of are indicated as the crossed arrows in Fig. 43, and the currents for the horizontal lines or vertical lines through the rectangle in Fig. 43 only need to be of sum zero along these lines. Hence, this yields that .
Subcase 2-2 The number of connected components of is two.
One can show that has exactly two connected components by Lemma 3.19. Recalling that is geodesic and simple, one obtains that the complementary subgraph of is connected. Lemma 3.1 yields that is also minimal.
Case 3. is geodesic and non-simple.
Subcase 3-1 contains some loops. Then one of connected components of contains exactly one unit square by Corollary 3.14.
We may assume that there is a square where is the connected component of containing . Using Lemma 3.11 and symmetry of and , one can assume that is in the following shadow region; see Fig. 44.
One easily gets that the minimal currents for the square ; see Fig. 44. Then by we deduce that there are two neighbors (distinct from ) of and two neighbors (distinct from ) of . Applying Lemma 3.3 and Corollary 3.14 and recalling that is geodesic, we have two flat boundary geodesic rays from through respectively and two flat boundary geodesic rays from through respectively. Using Lemma 3.17, it implies that is isometric to the subgraph in Fig. 15. Similar arguments as in the proof of Claim 3.25 yield that .
Subcase 3-2 contains no loops.
Denote by the number of neighbors of in for any , and by the connected component of containing for some . Recalling that is non-simple, we may assume that
Subcase 3-2-1 Assume and are neighbors of .
By Lemma 3.3, Lemma 4.3 and the fact that is geodesic, there are four boundary geodesic rays from through respectively. By Corollary 3.5 and recalling that is geodesic and contains no loops, we obtain that these geodesic rays are all flat or have no corners. The four geodesic rays divide the plane into four quadrant regions. Applying Lemma 3.17 to these four quadrant regions, we have that is isometric to the subgraph with in Fig. 15. By similar arguments to Claim 3.25, we get .
Subcase 3-2-2 Assume and for any . Let be neighbors of in , and (, resp.) be up-vertical(left-horizontal, right-horizontal, resp.) to .
Subcase 3-2-2-1 Assume for some .
We may assume with boundary neighbors , and is in the geodesic subpath between and . By similar arguments of Subcase 3-2-1, we get two flat boundary geodesic rays from through respectively, two flat boundary geodesic rays from through respectively, and (,resp.) is right-horizontal(down-vertical,resp.) to . Using Corollary 3.18 for geodesic boundary line , we have where is the quadrant domain enclosed by and is the quadrant domain enclosed by . This yields that is of length at most two by Lemma 3.2. By similar arguments in Claim 3.25, we have and it is isometric to the subgraph in Fig. 15 with , or Fig. 15.
Subcase 3-2-2-2 Assume for any .
By similar arguments in Subcase 3-2-1, we get three boundary geodesic rays from through respectively and is flat or vertical. If has corners, then is horizontal by Corollary 3.5 and the facts that is geodesic and contains no loops. So that we can assume is horizontal by symmetry of . Let be regions enclosed by . We get or by Lemma 3.17.
We claim or . Otherwise, recalling for any and contains no loops, we have another boundary geodesic line with . By Lemma 3.9, there is a vertical edge in such that is up-vertical to . Then one can find one geodesic ray in from which doesn’t contain is above the horizontal line through . Then we get a contradiction by Corollary 3.10 for two disjoint geodesic rays and , which proves the claim.
Similarly, we have or . Note that Thus one of them is isolated and it contradicts Lemma 3.2.
Finally, we prove the result. ∎
4. Geometry of high-dimensional minimal subgraphs
For and denote by the -normed sphere centered at of radius . We say is a unit -cube if it is a dimensional cube with all edges of length one. Similar to the setting for , we identify the set with some -cube if are exactly the vertices of , where . So that any subgraph in can be identified with a dimensional cell complex or a geometric space in , called the geometric realization of which includes all cells whose vertices are contained in Given any subgraph , let be the -skeleton of (i.e. the union of -cubes of ) for .
Proposition 4.1.
For any minimal subgraph in , every connected component of is also minimal.
Proof.
Assume is one of connected components of . One can restrict the minimal recurrents of to , which yields the result by Corollary 2.3. ∎
Lemma 4.2.
Given any minimal subgraph and any , then . Moreover, it never occurs that for any in .
Proof.
If , then the currents flowing out of are at least , but the currents flowing into are at most This is impossible by Lemma 2.3.
If , then applying Lemma 2.3 to one can get that the current from to is one. By symmetry of and , one get the current from to is one. This is impossible. ∎
Lemma 4.3.
If is minimal, then both and are infinite.
Proof.
If not, then with some . Hence or . One can get a new graph by removing from or adding to . Comparing with , one easily sees that is not minimal. ∎
Proposition 4.4.
If is minimal, then we have , where is a constant depending only on .
Proof.
One can get a new graph by removing from for any and . Comparing with and using the minimality of , one can get
Thus we complete the proof. ∎
Now we are ready to prove Theorem 1.9.
Proof of Theorem 1.9.
It suffices to show the restriction of some minimal currents of to denoted by are well defined, and then they are minimal currents, i.e. the sum of the restrictive currents for any vertex of is zero; see Lemma 2.3. Note that may be nonempty. So that it suffices to show the currents on agree with those induced naturally by . Precisely speaking, the minimal current on any agrees with that induced by , where for and . This yields that by Lemma 4.2.
If , then If , then the result is true.
Thus we only need to consider the case Note that if , one easily deduces that there exists a square in containing the edge . Hence We have the following two cases.
Case 1. .
Subcase 1-1. The neighbors of and in are as in Fig. 46. Using Lemma 2.3 to and , we have that the currents on the oriented edges are all one, indicated by arrows in Fig. 46. Applying Lemma 2.3 to , we obtain that the current on is one, and hence . By symmetry, we get . Then the currents on the oriented edges are all one and the sum of the currents flowing into . This is a contradiction.
Subcase 1-2. The neighbors of and in are as in Fig. 46. Using Lemma 2.3 to and , we have that the currents on the oriented edges are both one. Since , and the sum of the currents flowing out of . This is a contradiction.
Subcase 1-3. The neighbors of and in are as in Fig. 47. Using Lemma 2.3 to and , we have that the currents on the oriented edges are all one, indicated by arrows in Fig. 47. Using Lemma 2.3 for , one can get the currents on the oriented edges are both one, and hence Then the currents on the oriented edges are all one. Thus the sum of the currents flowing in of . This is a contradiction.
Subcase 1-4. The neighbors of and in are as in Fig. 49 or Fig. 49. Using Lemma 2.3 to and , we deduce that the currents on oriented edges must be indicated by arrows in Fig. 49 or Fig. 49. Noting that , we have that the sum of currents flowing out of . This is a contradiction.
Case 2. .
Subcase 2-1. The neighbors of in are as in Fig. 52. Then it is obvious that which is impossible.
Subcase 2-2. The neighbors of and in are as in Fig. 52. If , then using Lemma 2.3 to , we have that the current on the oriented edge is one. So that we get the current on the oriented edge is one by applying Lemma 2.3 to , which agrees with that induced by , and we finish the proof in this setting. By symmetry of , the remaining setting in this subcase is that . In this case, by direct computation the currents flowing into , which is impossible.
Subcase 2-3. The neighbors of and in are as in Fig. 52. It follows that by . By Lemma 2.3 for , we have that the current on the oriented edge is one. Recalling that and using Lemma 2.3 to , we deduce that the current on the oriented edge is one, which agrees with that induced by . Hence we finish the proof in this subcase. ∎
Lemma 4.5.
For any minimal subgraph and any point , and there exists a positive constant depending only on such that
Proof.
For any , let be a positive integer with Then . We can get a new graph by removing from .
Comparing with , one can deduce that by the minimality of
This gives
Therefore, we obtain that
On the other hand, since
Hence we have
for some positive constant depending only on
∎
Corollary 4.6.
Given any minimal subgraph and any point , let be the set of connected components of . Then
In particular,
Proof.
The idea of the proof is similar to that of Lemma 4.5.
We argue by contradiction. Then this yields
for any , any and some positive constant depending only on . Hence we have
(6) |
We can get a new graph by removing from . Comparing with , using (6) and applying the minimality of , we have that
and hence
(7) |
Note that
which yields
for any . But recall that
This is impossible for sufficiently large . ∎
Lemma 4.7.
For any minimal subgraph and any point , we have
where
Proof.
Observe the new graph obtained by removing from and note that . Comparing with and recalling that is minimal, we get that
∎
Proof of Theorem 1.11.
The natural embedding is surely a rough isometry by Lemma 4.5. Recall that , , and one has for any given by Lemma 4.5. One can show the inequality (3) in Theorem 1.11 for sufficiently large .
It is obvious that .
Hence we have
By , direct computation yields that
∎
Proof of Theorem 1.12.
Suppose it is not true, let be one of two parallel hyperplanes and be the usual distance projection image from to . Assume the distance of two parallel hyperplanes is and is the new graph obtained by removing from for any given . Then one can show that
(8) | |||
(9) |
Before introducing a maximum principle property of minimal subgraphs, we adopt some notations.
Let be the -th projection for any integer , where
for any
For any subgraph and any , we define
and
Note that may be infinite.
Proposition 4.8.
Suppose that is minimal and given two finite subsets , such that the geometric realization of in is the closure of a bounded open domain. Then neither
(10) |
nor
(11) |
holds for any integer .
Proof.
It suffices to prove (10) fails for . We argue by contradiction.
Consider the nonempty set and the set . Note that and , by the assumption (10). Now one can obtain a new graph from by deleting the finite subset .
Using (10) again, one easily deduces that
where is any finite ball containing and is the boundary edge of in . This is impossible by the minimality of .
∎
5. Open problems
In this section, let always be a minimal subgraph in . We propose some open problems on geometry and topology of high-dimensional minimal subgraphs.
The first problem is the stronger version of Theorem 1.11.
Problem 5.1.
Are the following true:
The second problem is about the number of connected components and some “big” connected component of any minimal subgraph.
Problem 5.2.
Do and have only finitely many connected components? Does at least one of these connected components contains a subgraph isometric to
Motivated by positive density at infinity of any minimal hypersurface in , it is natural to ask the growth rates of any minimal subgraph and its boundary.
Problem 5.3.
Whether do there exist positive constants depending only on such that
Note that the monotonicity formula of any minimal subgraph or its boundary does not hold in general; see Figure 15. Proposition 4.4 gives the upper bounds in Problem 5.3. So that it suffices to consider the lower bounds in Problem 5.3.
The last question is as follows.
Problem 5.5.
For a minimal subgraph , do bounded connected components of exist? How to characterize the bounded connected components of if they exist?
Acknowledgements. We thank Florentin Münch for helpful discussions and suggestions on the 1-Laplaican and the definition of area-minimizing subgraphs in B.H. is supported by NSFC, no. 12371056 and Shanghai Science and Technology Program [Project No. 22JC1400100]. Z.H. is supported by NSFC, no.12301094 and Guangzhou Basic and Applied Basic Research Foundation No. 2024A04J3483.
Data Available Statement. The data used to support the findings of this study are available from the corresponding author upon request.
References
- [Bar17] Martin T. Barlow. Random walks and heat kernels on graphs, volume 438 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2017.
- [BBI01] D. Burago, Yu. Burago, and S. Ivanov. A course in metric geometry, volume 33 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2001.
- [BDGG69] E. Bombieri, E. De Giorgi, and E. Giusti. Minimal cones and the Bernstein problem. Invent. Math., 7:243–268, 1969.
- [Ber27] Serge Bernstein. Über ein geometrisches Theorem und seine Anwendung auf die partiellen Differentialgleichungen vom elliptischen Typus. Math. Z., 26(1):551–558, 1927.
- [Cha16] K. C. Chang. Spectrum of the 1-Laplacian and Cheeger’s constant on graphs. J. Graph Theory, 81(2):167–207, 2016.
- [DG61] Ennio De Giorgi. Frontiere orientate di misura minima. Editrice Tecnico Scientifica, Pisa, 1961. Seminario di Matematica della Scuola Normale Superiore di Pisa, 1960-61.
- [DG65] Ennio De Giorgi. Una estensione del teorema di Bernstein. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (3), 19:79–85, 1965.
- [Fle62] Wendell H. Fleming. On the oriented Plateau problem. Rend. Circ. Mat. Palermo (2), 11:69–90, 1962.
- [FM19] Morteza Fotouhi and Amir Moradifam. General least gradient problems with obstacle. Calc. Var. Partial Differential Equations, 58(5):Paper No. 182, 19, 2019.
- [G1́8] Wojciech Górny. Planar least gradient problem: existence, regularity and anisotropic case. Calc. Var. Partial Differential Equations, 57(4):Paper No. 98, 27, 2018.
- [Giu84] Enrico Giusti. Minimal surfaces and functions of bounded variation, volume 80 of Monographs in Mathematics. Birkhäuser Verlag, Basel, 1984.
- [GM21] Wojciech Górny and José M. Mazón. Least gradient functions in metric random walk spaces. ESAIM Control Optim. Calc. Var., 27:Paper No. S28, 32, 2021.
- [HB10] M. Hein and T. Bühler. An inverse power method for nonlinear eigenproblems with applications in -spectral clustering and sparse pca. Advances in Neural Information Processing Systems (NIPS), pages 847–855, 2010. MIT Press, Cambridge, MA.
- [JMN18] Robert L. Jerrard, Amir Moradifam, and Adrian I. Nachman. Existence and uniqueness of minimizers of general least gradient problems. J. Reine Angew. Math., 734:71–97, 2018.
- [Juu05] Petri Juutinen. -Harmonic approximation of functions of least gradient. Indiana Univ. Math. J., 54(4):1015–1029, 2005.
- [Mir67] M. Miranda. Comportamento delle successioni convergenti di frontiere minimali. Rend. Sem. Mat. Univ. Padova, 38:238–257, 1967.
- [Mor17] Amir Moradifam. Least gradient problems with Neumann boundary condition. J. Differential Equations, 263(11):7900–7918, 2017.
- [Mor18] Amir Moradifam. Existence and structure of minimizers of least gradient problems. Indiana Univ. Math. J., 67(3):1025–1037, 2018.
- [MPLRT16] José M. Mazón, Mayte Pérez-Llanos, Julio D. Rossi, and Julián Toledo. A nonlocal 1-Laplacian problem and median values. Publ. Mat., 60(1):27–53, 2016.
- [MRSdL14] José M. Mazón, Julio D. Rossi, and Sergio Segura de León. Functions of least gradient and 1-harmonic functions. Indiana Univ. Math. J., 63(4):1067–1084, 2014.
- [MSDTM23] José M. Mazón, Marcos Solera-Diana, and J. Julián Toledo-Melero. Variational and diffusion problems in random walk spaces, volume 103 of Progress in Nonlinear Differential Equations and their Applications. Birkhäuser/Springer, Cham, 2023.
- [MT23] José M. Mazón and Julián Toledo. Cahn-Hilliard equations on random walk spaces. Anal. Appl. (Singap.), 21(4):959–1000, 2023.
- [Roc70] R. Tyrrell Rockafellar. Convex analysis. Princeton Mathematical Series, No. 28. Princeton University Press, Princeton, N.J., 1970.
- [RW98] R. Tyrrell Rockafellar and Roger J.-B. Wets. Variational analysis, volume 317 of Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, 1998.
- [Sim68] James Simons. Minimal varieties in riemannian manifolds. Ann. of Math. (2), 88:62–105, 1968.
- [SWZ92] Peter Sternberg, Graham Williams, and William P. Ziemer. Existence, uniqueness, and regularity for functions of least gradient. J. Reine Angew. Math., 430:35–60, 1992.
- [Woe00] W. Woess. Random walks on infinite graphs and groups. Number 138 in Cambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 2000.
- [Zun19] Andres Zuniga. Continuity of minimizers to weighted least gradient problems. Nonlinear Anal., 178:86–109, 2019.