![Arxiv:2102.11621V2 [Math.MG] 24 Feb 2021 Is No Known Isoperimetric-Type Result for Parallelohedra](http://data.docslib.org/img/9efd92f93954d834f095c82c1d29ffd0-1.webp)
ISOPERIMETRIC PROBLEMS FOR THREE-DIMENSIONAL PARALLELOHEDRA AND TRANSLATIVE, CONVEX MOSAICS
ZSOLT LANGI´
Abstract. The aim of this note is to investigate isoperimetric-type problems for 3-dimensional parallelohedra; that is, for convex polyhedra whose translates tile the 3-dimensional Euclidean space. Our main result states that among 3- dimensional parallelohedra with unit volume the one with minimal mean width is the regular truncated octahedron. In addition, we establish a connection between the edge lengths of 3-dimensional parallelohedra and the edge densities of the translative mosaics generated by them, and use our method to prove that among translative, convex mosaics generated by a parallelohedron with a given volume, the one with minimal edge density is the face-to-face mosaic generated by cubes.
1. Introduction Among the convex polyhedra in Euclidean 3-space R3 most known both inside and outside mathematics, we find the so-called parallelohedra; that is, the convex polyhedra whose translates tile space. All parallelohedra in 3-space can be defined as the Minkowski sums of at most six segments with prescribed linear dependen- cies between the generating segments, and thus, they are a subclass of zonotopes. Parallelohedra in R3 are also related to the Voronoi cells of lattices via Voronoi’s conjecture, stating (and proved in R3, see [6, 10, 12]) that any parallelohedron is an affine image of such a cell. Well-known examples of parallelohedra are the cube, the regular rhombic dodecahedron, and the regular truncated octahedron, which are the Voronoi cells generated by the integer, the face-centered cubic and the body-centered cubic lattice, respectively. Many papers investigate isoperimetric problems for zonotopes (see, e.g. [1, 2, 21]), with special attention on the relation between mean width and volume (see [5, 9]). On the other hand, apart from the celebrated proof of Kepler’s Conjecture by Hales [16], which, as a byproduct, implies that among parallelohedra with a given inradius, the one with minimum volume is the regular rhombic dodecahedron, there arXiv:2102.11621v2 [math.MG] 24 Feb 2021 is no known isoperimetric-type result for parallelohedra. In our paper we offer a new representation of 3-dimensional parallelohedra that might be useful to investigate their geometric properties. We use this representation to prove Theorem 1.
2010 Mathematics Subject Classification. 52B60, 52A40, 52C22. Key words and phrases. parallelohedra, zonotopes, discrete isoperimetric problems, density, tiling, Kepler’s conjecture, honeycomb conjecture, Kelvin’s conjecture. The author is supported by the National Research, Development and Innovation Office, NKFI, K-134199, the J´anosBolyai Research Scholarship of the Hungarian Academy of Sciences, and grants BME IE-VIZ TKP2020 and UNKP-20-5´ New National Excellence Program by the Ministry of Innovation and Technology. 1 2 Z. LANGI´
Theorem 1. Among unit volume 3-dimensional parallelohedra, regular truncated octahedra have minimal mean width. By their definition, parallelohedra are closely related to translative, convex mo- saics in R3 [29], which are the topic of the second part of the paper. During our investigation, we denote the family of translative, convex mosaics whose cells have unit volume by M. To state our results, we define the edge density and the surface density of a mosaic in the usual way: if M is a mosaic and i ∈ {1, 2}, and the limit 3 voli(skeli(M) ∩ (rB )) ρi(M) = lim sup 3 r→∞ vol3(rB ) 3 exists, where B is the Euclidean unit ball centered at the origin o, voli(·) denotes i-dimensional volume, and skeli(M) denotes the i-dimensional skeleton of M, then ρ1(M) is called the (upper) edge density, and ρ2(M) is called the (upper) sur- face density of M. As a special case of the relation between parallelohedra and translative, convex mosaics, one may expect that for a parallelohedron P and the associated face-to-face, translative, convex mosaic M, the edge density of M and the mean width of P , and similarly the surface density of M and the surface of P , are related in some way. In the second part of the paper we explore this connection, and, using a suitable modification of the proof of Theorem 1, we prove Theorem 2. Theorem 2. A translative, convex mosaic M ∈ M has minimal edge density if and only if M is a face-to-face mosaic with cubes as cells. The structure of the paper is as follows. In Section 2 we introduce and pa- rametrize 3-dimensional parallelohedra, and describe the tools and the notation necessary to prove our results. In Section 3 we prove Theorem 1. In Section 4 we investigate the properties of translative, convex mosaics, and show how to modify the arguments in Section 3 to prove Theorem 2. In Section 5, we collect additional remarks, and recall some old, and raise some new problems. Finally, in our Ap- pendix we collect the results of our computations for the minima of the quantities investigated in this paper for the different types of parallelohedra and translative, convex mosaics.
2. Preliminaries 2.1. Properties of three-dimensional parallelohedra. Minkowski [24] and Ven- kov [33] proved that P is a convex n-dimensional polytope for which there is a face-to-face tiling of Rn with translates of P if and only if (i) P and all its facets are centrally symmetric, and (ii) the projection of P along any of its (n − 2)-dimensional faces is a parallel- ogram or a centrally symmetric hexagon. Dolbilin [7] showed that these properties are already necessary if we require only the existence of a not necessarily face-to-face tiling with translates of P . The polytopes satisfying properties (i-ii) are called n-dimensional parallelohedra. The combinatorial classes of 3-dimensional parallelohedra are well known (cf. Figure 1). More specifically, any 3-dimensional parallelohedron is combinatorially isomorphic to one of the following: (1) a cube, (2) a hexagonal prism, ISOPERIMETRIC PROBLEMS 3
(3) Kepler’s rhombic dodecahedron (which we call a regular rhombic dodeca- hedron), (4) an elongated rhombic dodecahedron, (5) a regular truncated octahedron. We call a parallelohedron combinatorially isomorphic to the polyhedron in (i) a type (i) parallelohedron. It is also known that every parallelohedron P in R3 is a zonotope. More specifically, a type (1)-(5) parallelohedron can be attained as the Minkowski sum of 3, 4, 4, 5, 6 segments, respectively. A typical parallelohedron is of type (5), as every other parallelohedron can be obtained by removing some of the generating vectors of a type (5) parallelohedron. Since every 3-dimensional parallelohedron P ⊂ R3 is a zonotope, every edge of P is a translate of one of the segments generating P . Furthermore, for any generating segment S, the faces of P that contain a translate of S form a zone, i.e. they can be arranged in a sequence F1,F2,...,Fk,Fk+1 = F1 of faces of P such that for all values of i, Fi ∩Fi+1 is a translate of S. By property (ii) in the list in the beginning of Section 2, this sequence contains 4 or 6 faces. In these cases we say that S generates a 4-belt or a 6-belt in P , respectively. The numbers of 4- and 6-belts of a type (i) parallelohedron are 3 and 0, 3 and 1, 0 and 4, 1 and 4, and 0 and 6 for i = 1, 2, 3, 4, 5, respectively.
Type (1) Type (2) Type (3)
Type (4) Type (5)
Figure 1. The five combinatorial types of 3-dimensional paral- lelohedra. The type (5) parallelohedron in the picture is the regu- lar truncated octahedron generated by the six segments connecting the midpoints of opposite edges of a cube. The type (3) polyhe- dron is the regular rhombic dodecahedron generated by the four diagonals of the same cube. The rest of the polyhedra in the pic- ture are obtained by removing some generating segments from the type (5) polyhedron. 4 Z. LANGI´
Pm 3 Remark 1. Let P = i=1[o, vi] ⊂ R be a parallelohedron, and set I = {1, 2, . . . , m}. Then the volume, the surface area and the mean width of P is X (1) vol3(P ) = |Vijk|, {i,j,k}⊂I X (2) surf(P ) = 2 |vi × vj|, and {i,j}⊂I
m X (3) w(P ) = |vi|, i=1 respectively, where Vijk is the determinant of the matrix with column vectors vi, vj, vk. Proof. The formula in (1) is the well-known volume formula for zonotopes (cf. e.g. [31]). The one in (2) can be proved directly for type (5) parallelohedra, which implies its validity for all parallelohedra. Finally, Steiner’s formula [28] yields that 1 P the second quermassintegral W2(P ) of P is 3 ij(π − αij)lij, where αij and lij is the dihedral angle and the length of the edge between the ith and jth faces of P . 2π Pm Thus, from the zone property it follows that W2(P ) = 3 i=1 |vi|. On the other 3 3 hand, for any convex body K ⊂ R we have w(K) = 2π W2(K), which implies (3).
If we choose the generating segments in the form [o, pi] for some vectors pi ∈ 3 R , then in case of a type (1) or type (4) parallelohedron, the vectors vi are in general position, i.e. any three of them are linearly independent, whereas in case of the other types there are triples of vectors that are restricted to be co-planar. In particular, if P is a type (5) parallelohedron, then each of the four pairs of hexagon faces defines a linear dependence relation on the generating segments. More specifically, if v1, v2, v3, v4 are normal vectors of the four pairs of hexagon faces of the parallelohedron, then the directions of the six generating segments are the ones defined by the six cross-products vi × vj, i 6= j. Note that all triples of the vis are linearly independent, as otherwise some of the generating segments of P are collinear. Thus, we have X (4) P = [o, βij(vi × vj)] 1≤i • βij > 0 for all i 6= j, or • exactly one of the βijs is zero, or • exactly two of the βijs is zero: βi1j1 = βi2,j2 = 0, and they satisfy {i1, j1} ∩ {i2, j2} = ∅, or • exactly two of the βijs is zero: βi1j1 = βi2,j2 = 0, and they satisfy {i1, j1} ∩ {i2, j2}= 6 ∅, or • exactly three of the βijs are zero. and there is no s ∈ {1, 2, 3, 4} such that the indices of all nonzero βijs contain s, respectively. ISOPERIMETRIC PROBLEMS 5 It is easy to check that in the remaining cases P is planar. Thus, we can use this representation for all parallelohedra appearing in the paper with the assumption that βij ≥ 0 for all 1 ≤ i < j ≤ 4. Since no three of the vectors vi are linearly independent, but clearly all four of them are, up to multiplying by a constant they have a unique nontrivial linear combination λ1vi + λ2v2 + λ3v3 + λ4v4 = o. Since in the representation of P we considered only the directions of the vis, we may clearly assume that v1 + v2 + v3 + v4 = o. This implies that the tetrahedron conv{v1, v2, v3, v4} is centered, that is, its center of gravity is o. From the equation v1 + v2 + v3 + v4 = o it follows that the absolute values of the determinants of any three of the vis are equal. We choose this 2 common value 1, which yields that the volume of the tetrahedron is 3 . Throughout the paper for any i, j, k ∈ {1, 2, 3, 4}, we denote by Vijk the value of the determinant with vi, vj, vk as column vectors. We choose the indices in such a way that V123 = 1. Since v1 + v2 + v3 + v4 = o, we have that for any {i, j, s, t} = {1, 2, 3, 4}, the plane containing o, vi, vj strictly separates vs and vt, which implies that Vijs = −Vijt. In the proof we often use the function f : R6 → R, (5) f(τ12, τ13, τ14, τ23, τ24, τ34) = τ12τ13τ23 + τ12τ14τ24 + τ13τ14τ34 + τ23τ24τ34+ +(τ12 +τ34)(τ13τ24 +τ14τ23)+(τ13 +τ24)(τ12τ34 +τ14τ23)+(τ14 +τ23)(τ12τ34 +τ13τ24). We remark that f is not the third elementary symmetric function on six variables, 6 since after expansion it has only 16 < 20 = 3 members. An elementary computation yields that for any i, j, k, l, s, t ∈ {1, 2, 3, 4}, we have |vi × vj, vk × vl, vs × vt| = VijlVstk − VijkVstl. Using this formula and the properties of cross-products, we may express the volume, surface area and mean width of P in our representation. P Remark 2. For the parallelohedron P = i6=j βij[o, vi × vj] satisfying the condi- tions above, we have (6) vol3(P ) = f(β12, β13, β14, β23, β24, β34), (7) surf(P ) = 2 (β12β13 +β12β14 +β13β14)|v1|+... (β14β24 +β14β34 +β24β34)|v4|+ +β12β34(|v1 + v2|) + ... + β14β23|v1 + v4| , X (8) w(P ) = βij|vi × vj|, 1≤i Lemma 1. Let T = conv{p1, p2, p3, p4} be an arbitrary centered tetrahedron with volume V > 0 where the vertices are labelled in such a way that the determinant with columns p1, p2, p3 is positive. For any {i, j, s, t} = {1, 2, 3, 4}, let γij = −hps, pti 2 and ζij = γij|pi × pj| . Then 9 2 (1) f(γ12, γ13, γ14, γ23, γ24, γ34) = 4 V , P 27 2 (2) 1≤i Proof. Let χij = hpi, pji for all i, j. Consider the Gram matrix G defined by p1, p2, p3, and observe that as T is centered, the volume of the parallelepiped 3 spanned by them is 2 V . Since the determinant of the Gram matrix of d linearly independent vectors in Rn is the square of the volume of the parallelotope spanned 9 2 by the vectors, it follows that det(G) = 4 V . Furthermore, since T is centered, we have χ14 = −χ11 − χ12 − χ13, and we may obtain similar formulas for χ24 and χ34. Now, substituting these formulas into f(γ12, . . . , γ34) = f(−χ34,..., −χ12), we obtain that 2 2 2 f(γ12, . . . , γ34) = χ11χ22χ33 + 2χ12χ13χ23 − χ11χ2 − χ22χ13 − χ33χ12 = det(G), which implies the first identity. To obtain the second identity, observe that for any {i, j, s, t} = {1, 2, 3, 4}, we 2 2 have ζij = −hps, pti|pi × pj| = −χst χiiχjj − χij . This observation and an P argument similar to the one in the previous paragraph yields that 1≤i (∂τ12 f)(τ) = τ13τ23 + τ14τ24 + τ13τ24 + τ14τ23 + (τ13 + τ24 + τ14 + τ23)τ34, and we obtain similar expressions for all partial derivatives. Solving the system of equations (∂τ1i f)(τ) = 3f(τ), i = 2, 3, 4 for τ12, τ13, τ14, and substituting back the 4 solutions into the definition of f, we obtain that f(τ) = 27 (τ23 + τ24 + τ34). By a 4 4 4 similar argument, f(τ) = 27 (τ12 +τ13 +τ23) = 27 (τ12 +τ14 +τ24) = 27 (τ13 +τ14 +τ34) 8 P 8 also follows. Adding up, we obtain that in this case 4f(τ) = 27 i6=j τij = 27 , 2 implying that f(τ) = 27 . We show that in case of equality all coordinates of τ are equal. The equations 1 in the previous paragraph yield that τ12 + τ13 + τ23 = ... = τ23 + τ24 + τ34 = 2 , and 1 also that τ12 +τ13 +τ14 = ... = τ14 +τ24 +τ34 = 2 . Comparing suitable equations it ISOPERIMETRIC PROBLEMS 7 readily follows that τ12 = τ34, τ13 = τ24 and τ14 = τ23. Replacing τ34, τ24 and τ23 by τ12, τ13 and τ14, respectively, in the equations (∂τ12 f)(τ) = (∂τ13 f)(τ) = (∂τ14 f)(τ), 2 2 2 we obtain τ12 = τ13 = τ14, which characterizes equality in this case. Case 2, if exactly one coordinate of τ is zero. Without loss of generality, we 0 5 may assume that τ34 = 0, and let τ ∈ R be the vector obtained by removing the last coordinate of τ. Let us define the function g(τ12, . . . , τ24) = f(τ12, . . . , τ24, 0). Similarly like in Case 1, if f(τ) is a local maximum, we have that all five partial derivatives of g are equal at τ 0. Again, from this it follows that this common value is 0 equal to 3g(τ ). Factorizing the left-hand sides of the equations ∂τ13 − ∂τ14 = 0 and ∂τ23 − ∂τ24 = 0, we obtain that τ13 = τ14, τ23 = τ24, respectively. After substituting these equalities into the equation ∂τ13 − ∂τ23 = 0 and factorizing the left-hand side, we obtain that τ13 = τ23. Again, substituting back into ∂τ12 − ∂τ13 = 0 and 1 factorizing its left-hand side yields τ13 = 2τ12. From this, we have τ12 = 9 and 2 0 16 τij = 9 for any i ∈ {1, 2} and j ∈ {3, 4}, which implies that g(τ ) = 243 . Case 3, if at least two coordinates of τ are zero. Then, using an argument 1 similar to that in the previous cases, we obtain that the maximum of f(τ) is 16 , 1 27 and 0 if exactly two, exactly three or more than three coordinates of τ are zero, respectively. 3. Proof of Theorem 1 Recall from Subsection 2.1 that we represent P in the form X P = [o, βijvi × vj] 1≤i 3 P4 for some v1, v2, v3, v4 ∈ R satisfying i=1 vi = o, and for some βij ≥ 0. By our assumptions, |Vijk| = 1 for any {i, j, k} ⊂ {1, 2, 3, 4}, where Vijk denotes the deter- minant with vi, vj, vk as columns, and we assumed that V123 = 1, which implies, in particular, that V124 = −1. Then T = conv{v1, v2, v3, v4} is a centered tetrahedron, 2 with volume vol3(T ) = . 3 P By Remark 2, we need to find the minimum value of w(P ) = 1≤i n Lemma 3. Let P ⊂ R be a convex polytope with outer unit facet normals u1, . . . , uk. Let Fi denote the (n − 1)-dimensional volume of the ith facet of P . Then, up to congruence, there is a unique affine transformation L such that surf(L(P )) is max- imal in the affine class of P . Furthermore, P satisfies this property if and only if its surface area measure is isotropic, that is, if k X nFi (9) u ⊗ u = Id surf(P ) i i i=1 where Id denotes the identity matrix. Any convex polytope satisfying the conditions in (9) is said to be in surface isotropic position. We note that the volume of the projection body of any convex polyhedron is invariant under volume preserving linear transformations (cf. [27]). On the other hand, from Cauchy’s projection formula and the additivity of the support function (see. [13]) it follows that the projection body of the polytope in Lemma 3 is the zonotope Pk [o, F u ]. Note that the solution to Minkowski’s i1 i i n problem [28] states that some unit vectors u1, . . . , uk ∈ R and positive numbers F1,...,Fk are the outer unit normals and volumes of the facets of a convex poly- n Pk tope if and only if the uis span R , and i=1 Fiui = o. On the other hand, a translate of the parallelohedron P in our investigation can be written in the form 1 P 2 1≤i X 3βij (10) (vi × vj) ⊗ (vi × vj) = Id, w(P )|vi × vj| 1≤i Step 2. Consider the function f(ζ12, . . . , ζ34), where ζij = −hvs, vti|vi × vj| for {i, j, s, t} = {1, 2, 3, 4}, 2 and v1, v2, v3, v4 are the vertices of a centered tetrahedron T with volume 3 . Set 2 γij = −hvs, vti and τij = γij|vi × vj| for all {i, j, s, t} = {1, 2, 3, 4}. To give an upper bound on the value of f, we apply the Cauchy-Schwartz Inequality, which states that for any nonnegative real numbers xi, yi, i = 1, 2, . . . , k, we q q Pk Pk 2 Pk 2 have i=1 xiyi ≤ i=1 xi i=1 yi , with equality if and only if the xis and the y s are proportional. To do this, we write each member of f as the product i √ √ ζijζklζmn = γijγklγmn τijτklτmn. Thus, we obtain that p p p f(ζ12, . . . , ζ34) ≤ f(γ12, . . . , γ34) f(τ12, . . . , τ34) = f(τ12, . . . , τ34), where we used the fact that by Lemma 1, f(γ12, . . . , γ34) = 1 for all centered tetrahe- dra with volume 2 . Furthermore, observe that by Lemma 1 we have P τ = 3 1≤i≤√j≤4 ij 3 for all such tetrahedra. Hence, by Lemma√ 2, we have f(ζ12, . . . , ζ34) ≤ 2. 1 Now, assume that f(ζ12, . . . , ζ34) = 2. Then, by Lemma 2, τij = 2 for all values i 6= j. This, by the Cauchy-Schwartz Inequality, implies that for some t ∈ R, γij = t for all values i 6= j. Since T is centered, this implies that γii = 3t for all is. In other words, the Gram matrix of the vectors v1, . . . , v4 is a scalar multiple of the matrix 4 Id −E, where E is the matrix with all entries equal to 1. Since the Gram matrix of a vector system determines the vectors up to orthogonal transformations, and it is easy to check that 4 Id −E is the Gram√ matrix of the vertex set of a centered regular tetrahedron of circumradius 3, the equality part in Theorem 1 follows. 4. Translative, convex mosaics and the proof of Theorem 2 Our next lemma establishes a relationship between the edge density of a transla- tive, convex mosaic M and the edge lengths of the generating parallelohedron P . Pk 3 Lemma 4. Let P = i=1[o, pi] be a unit volume parallelohedron in R , and let M be a tiling of R3 with translates of P . Set 1, if the projection of P along pi is a parallelogram, wi = 2, if the projection of P along pi is a centrally symmetric hexagon. Then 6 X (11) ρ1(M) ≥ wi|pi| i=1 with equality if M is face-to-face. Proof. First, we assume that M is face-to-face and P is a cell of M, and choose an edge E of P . Let E contain the edges E0 of the tiling for which there is a sequence of cells P1,...,Pk in which consecutive members have a common face and this face 0 0 contains an edge parallel to E, and E ⊂ P1 and E ⊂ Pk. Furthermore, let P be the family of cells containing edges from E. Then the projection of the members of P in the direction of E onto a plane is a translative, convex, edge-to-edge tiling of the plane. By the second property of parallelohedra in the list in the beginning of Section 2, the projections of the elements of P are either parallelograms, in which 10 Z. LANGI´ every vertex belongs to exactly four parallelograms, or they are centrally symmetric hexagons, in which every vertex belongs to three hexagons. Thus, a cell contains 4 or 6 translates of any given edge, and in the first case every edge belongs to exactly 4 cells, and in the second case every edge belongs to exactly 3 cells. 1 Now, we assign a weight wE to each edge E of M; we set w(E) = k , where k is the number of cells E belongs to. Let Nr denote the number of cells of M, and Er denote 3 3 2 the family of the edges of the cells contained in rB . As Nr = vol3(rB ) + O(r ), we have 6 3 X 2 X 2 vol1(skel1(M) ∩ (rB )) = vol1(E) + O(r ) = Nr wi|pi| + O(r ). E∈Er i=1 Finally, if M 0 is a not necessarily face-to-face tiling by translates of P , then we may divide each edge into finitely many pieces in such a way that relative interior points on each piece belong to the same cells. It is easy to see that any part of an edge in a 6-belt belongs to exactly 3 cells, and a part of an edge in a 4-belt belongs to either 4 or 2 cells. Thus, repeating the procedure, the weighted sum of the edge lengths 3 in rB , defined like in the previous case, yields a value for ρ1(M) asymptotically not less than that for a face-to-face mosaic generated by P . Now we prove Theorem 2. First, we consider only face-to-face mosaics. Let P be the parallelohedron generating M. It is easy to see that if P is a type (1) parallelohedron, then ρ1(M) is minimal if and only if P is a cube, and in this case ρ1(P ) = 3. Consider the case that P is a type (2) parallelohedron; that is, a centrally symmetric hexagon based prism. If the lengths of the edges of the hexagon are a1, a2, a3, and the length of its lateral edges is b, then ρ1(M) = a1 +a2 +a3 +2b. An elementary computation shows that if vol3(P ) = 1, then this quantity is minimal if and only if P is a right prism, its base is a regular hexagon and 4b = 3a1 = 3a2 = 7/6 3a3. Furthermore, in this case ρ1(M) = 3 ≈ 3.60. Next, consider the case that P is a type (3) or a type (5) parallelohedron. Then all edges of P belong to 6-belts, and hence, ρ1(M) = 2w(P ), where w(P ) denotes the mean width of P . Thus, by 5/6 Theorem 1, we have that in this case ρ1(M) ≥ 2 · 3 ≈ 5.35. We are left with the case of type (4) parallelohedra. To prove Theorem 2 in this case, we use the notation introduced in Subsection 2.1, and thus, we represent P in the form X P = [o, βijvi × vj] 1≤i vol3(P ) = f(β12, β13, . . . , β24, 0). ISOPERIMETRIC PROBLEMS 11 We intend to find the minimum of ρ1(M) under the condition that vol3(P ) = 1. We carry out the same two steps as in the proof of Theorem 1. In particular, we observe that there is an o-symmetric convex polytope in R3 with outer unit facet normals ± vi×vj where i 6= j and {i, j} 6= {3, 4}, and areas of the corresponding |vi×vj | β12 facets 2 |v1 × v2|, β13|v1 × v3|, . . . , β24|v2 × v4|. Similarly like in the previous section, we may assume that this polytope is in surface isotropic position, implying that β12 X 2βij ρ1(M) (v1 ×v2)⊗(v1 ×v2)+ (vi ×vj)⊗(vi ×vj) = Id . |v1 × v2| |vi × vj| 3 i∈{1,2},j∈{3,4} Multiplying this equation both from the left and from the right by some of the vis, we obtain that ρ (M) β = −hv , v i|v × v | · 1 , 12 3 4 1 2 3 and for any i ∈ {1, 2} and j ∈ {3, 4}, we have ρ (M) β = −hv , v i|v × v | · 1 . ij 3 4 1 2 6 Let ζij = −hvs, vti|vi × vj| be defined as in the previous section. Then our opti- mization problem can be rewritten as finding the maximum of 216 vol3(P ) 3 = f(2ζ12, ζ13, . . . , ζ24, 0) (ρ1(M)) 2 in the family of all centered tetrahedra T with vol3(T ) = and hv3, v4i = 0, under P 3 the condition that 1≤i 5. Remarks and open problems For a convex polyhedron P ⊂ R3, let E(P ) denote the total edge length of P . Melzak [22] in 1965 made the conjecture that for any unit volume convex polyhedron P ⊂ R3, the total edge length of P is E(P ) ≥ 22/3 · 311/6 ≈ 11.89, with equality 12 Z. LANGI´ for certain regular triangle based prisms. This conjecture is still open despite being the subject of scientific research even now [30]. A slight modification of the proof of Theorem 2 provides a proof of this conjecture for the special case of parallelohedra. Remark 3. For any unit volume parallelohedron P in R3, we have E(P ) ≥ 12, with equality if and only if P is a cube. The origin of the Honeycomb Conjecture, stating that in a decomposition of the Euclidean plane into regions of equal area, the regular hexagonal grid has the least perimeter, can be traced back to ancient times [32]. This problem has been in the focus of research throughout the second half of the 20th century [8, 23], and was finally settled by Hales [15]. The most famous analogous conjecture for mosaics in 3-dimensional Euclidean space is due to Lord Kelvin [19], who in 1887 conjectured that in a tiling of space with cells of unit volume, the mosaic with minimal surface area is composed of slightly modified truncated octahedra. Even though this conjecture was disproved by Weaire and Phelan in 1994 [34], who discovered a tiling of space with two slighly curved ‘polyhedra’ of equal volume and with less total surface area than in Lord Kelvin’s mosaic, the original problem of finding the mosaics with equal volume cells and minimal surface area has been extensively studied (see, e.g. [20, 11, 25]). On the other hand, in the author’s knowledge, there is no subfamily of mosaics for which Kelvin’s problem is solved. This is our motivation for the following conjecture. Before we state it, we observe that for any (not necessarily face-to-face) translative, convex mosaic M of R3, generated by translates of a parallelohedron P , we have 1 ρ2(M) = 2 surf(P ). Conjecture 1. Among translative, convex mosaics M in R3 generated by a unit volume parallelohedron P , ρ2(M) is minimal if and only if P is a regular truncated octahedron. The following conjecture, called Rhombic Dodecahedral Conjecture, can be found in [1]. We note that this conjecture is the lattice variant of the so-called Strong Dodecahedral Conjecture [3] proved by Hales in [17]. For other variants of the Dodecahedral Conjecture, see also [4] or [18]. 3 Conjecture 2 (Bezdek, 2000). The surface√ area of any parallelohedron in R with unit inradius is at least as large as 12 2 ≈ 16.97, which is the surface area of the regular rhombic dodecahedron of unit inradius. As it was mentioned in the introduction, Voronoi cells of lattice packings of con- gruent balls is an important subfamily of 3-dimensional parallelohedra. Regarding this subclass, Dar´oczy[5] gave an example of a packing whose Voronoi cells have a smaller mean width than that of the regular rhombic dodecahedron of the same inradius. This is the motivation behind our last question. Here we note that the mean widths of a cube, a regular truncated√ √ octahedron and a regular rhombic do- decahedron of unit inradius is 6 > 2 6 = 2 6, respectively, which in our opinion, makes the question indeed intriguing. Problem 1. Find the minimum of the mean widths of 3-dimensional parallelohedra of unit inradius. ISOPERIMETRIC PROBLEMS 13 6. Appendix In three tables we collected the results of our investigation for the minimal values of the mean width and the total edge length of the different types of parallelohedra, and the edge density of translative, convex mosaics defined by different types of parallelohedra. Recall (cf. Section 2) that the different types of parallelohedra correspond to the following polyhedra. • Type (1) parallelohedra: parallelopipeds. • Type (2) parallelohedra: hexagonal prisms. • Type (3) parallelohedra: rhombic dodecahedra. • Type (4) parallelohedra: elongated rhombic dodecahedra. • Type (5) parallelohedra: truncated octahedra. Finally, note that while the mean width w(·) is a continuous function of its variable with respect to Hausdorff distance, the quantities ρ1(·) and E(·) are not. Type Minimum of w(P ) Optimal parallelohedra (1) 3 cube regular hexagon based right prism, with 2/3 37/6 2 (2) base and lateral edges of lengths 5/6 21/3 ≈ 2.86 3 31/6 and 21/3 , respectively 2/3 1/2 regular rhombic dodecahedron√ with edge (3) 2 · 3 ≈ 2.75 3 length 24/3 (4) 34/3 not known ≥ 22/3 ≈ 2.73 3 regular truncated octahedron of edge (5) 21/6 ≈ 2.67 1 length 27/6 Table 1. Minima of the mean widths of different types of paral- lelohedra Type Infimum of ρL(P ) Optimal parallelohedra (1) 3 cube regular hexagon based right prism, with 7/6 2 (2) 3 ≈ 3.60 base and lateral edges of lengths 35/6 31/6 and 2 , respectively 5/3 1/2 regular rhombic dodecahedron√ with edge (3) 2 · 3 ≈ 5.50 3 length 24/3 (4) 34/3·51/3 not known ≥ 22/3 ≈ 4.66 5/6 regular truncated octahedron with edge (5) 2 · 3 ≈ 5.35 1 length 27/6 . Table 2. Infima of the edge densities of translative, convex mo- saics generated by different types of parallelohedra 14 Z. LANGI´ Type Infimum of E(P ) Optimal parallelohedra (1) 12 cube 3/2 4/3 regular hexagon based right prism,√ with (2) 3 2 ≈ 13.09 3 base and lateral edges of length √2 3 5/3 3/2 regular rhombic dodecahedron√ with edge (3) 2 · 3 ≈ 16.50 3 lengths 24/3 (4) ≥ 211/3 ≈ 12.70 not known 5 2 regular truncated octahedron with edge (5) 2 6 · 3 ≈ 16.04 1 length 27/6 Table 3. Infima of the total edge lengths of different types of parallelohedra Acknowledgements The author is indebted to A. Jo´osand G. Domokos for many fruitful discussions of this problem. References [1] K. Bezdek, E. Dar´oczy-Kissand K.J. Liu, Voronoi polyhedra in unit ball packings with small surface area, Period. Math. Hungar. 39 (1999), 107-118. [2] K. Bezdek, On rhombic dodecahedra, Beitr¨ageAlgebra Geom. 41 (2000), 411-416. [3] K. Bezdek, On a stronger form of Rogers’ lemma and the minimum surface area of Voronoi cells in unit ball packings, J. reine angew. Math. 518 (2000), 131–143. [4] K. Bezdek and Z. L´angi, Density bounds for outer parallel domains of unit ball packings, Proc. Steklov Inst. Math. 288 (2015), 209-225. [5] E. Dar´oczy-Kiss, On the minimum intrinsic 1-volume of Voronoi cells in lattice unit sphere packings, Period. Math. Hungar. 39 (1999), 119-123. [6] B. Delone, Sur la partition r ´eguli`ere de l’espace `a 4 dimensions, Bulletin de l’Acad´emiedes Sciences de l’URSS. VII s´erie,1929, no. 1, 79–110. [7] N.P. Dolbilin, Properties of faces of parallelohedra, Proc. Steklov Inst. Math. 266 (2009), 105–119. [8] L. Fejes T´oth, Uber¨ das k¨urzesteKurvennetz, das eine Kugeloberfl¨achein fl¨achengleiche konvexe Teil zerlegt, Math. Naturwiss. Anz. Ungar. Akad. Wiss. 62 (1943), 349–354. [9] P. Filliman, Extremum problems for zonotopes, Geom. Dedicata 27 (1988), 251–262. [10] E.S. Fedorov, Elements of the study of figures (in Russian), Zap. Mineralog. Obsch., 1885. [11] R. Gabbrielli, A new counter-example to Kelvin’s conjecture on minimal surfaces, Phil. Mag. Lett. 89(8) (2009), 483–491. [12] A. Garber and A. Magazinov, Voronoi conjecture for five-dimensional parallelohedra, arXiv:1906.05193 [math.CO], 12 June, 2019. [13] , R.J. Gardner, Geometric tomography, Encyclopedia of Mathematics and its Applications 58, Cambridge University Press, Cambridge, 2006. [14] A. Giannopoulos and M. Papadimitrakis, Isotropic surface area measures, Mathematika 46 (1999), 1-13. [15] T.C. Hales, The honeycomb conjecture, Discrete Comput. Geom. 25 (2001), 1–22. [16] T.C. Hales, A proof of the Kepler conjecture, Ann. Math. 162 (2005), 1065–1185. [17] T.C. Hales, Dense sphere packings: a blueprint for formal proofs, Cambridge University Press, Cambridge, 2012. [18] T.C. Hales and S. McLaughlin, The dodecahedral conjecture, J. Amer. Math. Soc. 23 (2010), 299-344. [19] Lord Kelvin (Sir William Thomson), LXIII. On the division of space with minimum par- titional area, The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science 24 (1887), 503-514. ISOPERIMETRIC PROBLEMS 15 [20] R. Kusner and J. Sullivan, Comparing the Weaire-Phelan equal-volume foam to Kelvin’s foam Forma 11(3) (1996), 233-242. [21] J. Linhart, Extremaleigenschaften der reguliren 3-Zonotope, Studia Sci. Math. Hungar. 21 (1986), 94-98. [22] Z.A. Melzak, Problems connected with convexity, Canad. Math. Bull. 8 (1965), 565-573. [23] F. Morgan, The hexagonal honeycomb conjecture, Trans. Amer. Math. Soc. 351 (1999), 1753–1763. [24] H. Minkowski, Allgemeine Lehrs¨atze ¨uber die convexen Polyeder, G¨ott. Nachr. (1897), 198–219. [25] E.´ Oudet, Approximation of partitions of least perimeter by Γ-convergence: around Kelvin’s conjecture, Experiment. Math. 20(3) (2011), 260–270. [26] M. Petty, Surface area of a convex body under affine transformations, Proc. Amer. Math. Soc, 12 (1961), 824-828. [27] M. Petty, Projection bodies, In: Proc. Colloq. Convexity, Copenhagen, 1965, (Københavns Univ. Mat. Inst., 1965) 234-241. [28] R. Schneider, Convex bodies: the Brunn-Minkowski theory, Encyclopedia of Mathematics and its Applications 151, Cambridge University Press, Cambridge, 2013. [29] E. Schulte, Tilings, Handbook of convex geometry, Vol. A, B, 899–932, North-Holland, Am- sterdam, 1993. [30] R.C. Scott, Minimizing the mass of the codimension-two skeleton of a convex, volume-one polyhedral region, Ph.D. thesis, Rice University, Houston TX, 2011. [31] G.C. Shephard, Combinatorial properties of associated zonotopes, Canad. J. Math. 24 (1974), 302-331. [32] M.T. Varro, On Agriculture, Loeb Classical Library, 1934. [33] B.A. Venkov, On a class of Euclidean polyhedra, Vestn. Leningr. Univ., Ser. Mat. Fiz., Khim. 9 (1954), 11–31. [34] D. Weaire and R. Phelan, A counter-example to Kelvin’s conjecture on minimal surfaces, Phil. Mag. Lett. 69(2) (1994), 107–110. Morphodynamics Research Group and Department of Geometry, Budapest University of Technology, Egry Jozsef´ utca 1., Budapest 1111, Hungary Email address: [email protected]