1 Introduction

We say that convex bodies \(\ C_1, C_2, \ldots \) can be packed into a convex body C if it is possible to apply translations and rotations to the sets \(C_n\) so that the resulting translated and rotated bodies are contained in C and have mutually disjoint interiors. It is known that if C is either a square [1], or a triangle (see [2] and [3]), or a disk [4], then any collection of homothetic copies of C, whose total area does not exceed half the area of C can be packed into C. The conjecture is that the same packing density is achievable for any planar convex body (see [5] and [6]). However, only the following upper bound was found [7]: any collection of homothetic copies of C, with total area not greater than a quarter of the area of C, can be packed into C.

Meir and Moser [8] proved that any collection of d-dimensional cubes, whose total volume does not exceed \(2^{1-d}\) times the volume of K, can be packed into the d-dimensional cube K. In the paper [2] concerning packing of triangles, Richardson writes: “The original paper of Kranakis and Meertens [9] conjectured that Theorem 1 may be extendible to families of similar tetrahedra (in three dimensions) or even to any family of similar d-dimensional simplexes.” We confirm this conjecture in three dimensions for a tetrahedron \(T_r\) of vertices (0, 0, 0), (0, 1, 0), (1, 1, 0) and (0, 0, 1). We will show that any collection of homothetic copies of \(T_r\), whose total volume does not exceed one quarter of the volume of \(T_r\), can be packed into \(T_r\).

2 Division of \(T_r\)

The unit cube can be divided into two congruent triangular prisms. Observe that such a prism can be partitioned into three tetrahedra congruent to \(T_r\) (see Fig. 1). The unit cube can be partitioned into six tetrahedra congruent to \(T_r\).

Fig. 1
figure 1

Partition of a prism into three congruent tetrahedra

Denote by \(\lambda T_r\) the image of \(T_r\) in the homothety with ratio \(\lambda \) and center (0, 0, 1). Moreover, let \(\ a_1\ge a_2 \ge \ldots \ \) be positive numbers, let \(S_i=a_i T_r\) for \(\ i=1,2,\ldots \ \) and let \(\ a_1^3+a_2^3+\ldots \le 1/4\). Obviously, \(\ a_1+a_2 \le 1\), otherwise \(\ a_1^3+a_2^3 > (1/2)^3+(1/2)^3=1/4\), which is a contradiction.

To describe the method of packing tetrahedra \(S_1, S_2, \ldots \) into \(T_r\), let’s introduce some notations. Denote by \(k_1\) the smallest integer such that

$$\begin{aligned} a_4+a_{10}+\ldots + a_{6k_1-2}+a_{6k_1+1} > 1-a_1 \end{aligned}$$

(\(k_1=3\) in Fig. 3); if \(a_4+a_7>1-a_1\), then \(k_1=1\). Moreover, let \(k_2\) be the smallest integer such that

$$\begin{aligned} a_{6k_1+4}+a_{6k_1+10}+\ldots + a_{6k_2-2}+a_{6k_2+1} > 1-a_1-a_{6k_1+1} \end{aligned}$$

(\(k_2= 7\) in Fig. 3). At the end of this section we will define layers and slices. According to the packing method described in Sect. 3, tetrahedra \(\ S_1, \ldots , S_{6k_1}\ \) will be packed into the first layer of the first slice; tetrahedra \(\ S_{6k_1+1}, \ldots , S_{6k_2}\ \) will be packed into the second layer of the first slice. If

$$\begin{aligned} a_1+a_{6k_1+1}+a_{6k_2+1}+\ldots +a_{6k_i+1}+a_{6k_i+2} \le 1 \end{aligned}$$

for \(i\ge 2\), then denote by \(k_{i+1}\) the smallest integer such that

$$\begin{aligned} a_{6k_i+4}+a_{6k_i+10}+\ldots + a_{6k_{i+1}-2}+a_{6k_{i+1}+1} > 1-a_1-a_{6k_1+1}- \ldots - a_{6k_i+1}. \end{aligned}$$

If j is the smallest integer such that

$$\begin{aligned} a_1+a_{6k_1+1}+\ldots +a_{6k_j+1}+a_{6k_j+2} > 1, \end{aligned}$$

then we take \(s_1=a_{6k_j+2}\) (\(j=5\) and \(k_j=13\) in Fig. 3, right).

Fig. 2
figure 2

Division of \(T_r\)

Tetrahedra will be packed into layers \(L_i\) contained in slices \(H_n\). In Sect. 3 we will compare the total volume of tetrahedra packed into \(L_i\) with the volume of some layer \(L_i^*\). Let

$$\begin{aligned} H_1 = T_r \setminus (1-a_1)T_r, \end{aligned}$$

and

$$\begin{aligned} H_1^* = (1-a_1)T_r \setminus (1-a_1-a_{s_1})T_r. \end{aligned}$$

(see Fig. 2). Moreover, let

$$\begin{aligned} L_1&= \{ (x,y,z)\in H_1 : \ 0\le x \le a_1 \}, \\ L_1^*&= \{ (x,y,z)\in H_1^* : \ 0\le x \le a_{6k_1+1} \}, \\ L_2&= \{ (x,y,z)\in H_1 : \ a_1\le x \le a_1+a_{6k_1+1} \}, \\ L_2^*&= \{ (x,y,z)\in H_1^* : \ a_{6k_1+1}\le x \le a_{6k_1+1}+a_{6k_2+1} \}, \\ L_3&= \{ (x,y,z)\in H_1 : \ a_1+a_{6k_1+1}\le x \le a_1+a_{6k_1+1}+a_{6k_2+1} \}, \\ L_3^*&= \{ (x,y,z)\in H_1^* : \ a_{6k_1+1}+a_{6k_2+1}\le x \le a_{6k_1+1}+a_{6k_2+1} +a_{6k_3+1}\}, \ \mathrm{etc.} \end{aligned}$$
Fig. 3
figure 3

Bottom of \(H_1\)

Fig. 4
figure 4

Covering \(L_1^*\) with \(\ S_2, S_3, \ldots , S_{s_{6k_1+1}}\)

3 Packing Method

Observe that \( S_1, S_2\) and \(S_3\) can be packed into the part of \(L_1\) with \(\ y\ge 1-a_1\) (see Figs. 1, 3, 4). Moreover, \(S_4, S_5, \ldots , S_{9}\) can be paced into the cube

$$\begin{aligned} \{ (x,y,z): \ 0\le x \le a_4, \ 1-a_1-a_4 \le y \le 1-a_1, \ 0\le z \le a_4 \}. \end{aligned}$$

Packing method, Part I.

Tetrahedra \(S_1, S_2, \ldots , S_{6k_1}\) are packed into \(L_1\) (see Figs. 3,  4, where \(k_1=3\)). Tetrahedra \(S_{6k_i+1}, S_{6k_i+2}, \ldots , S_{6k_{i+1}}\) are packed into \(L_{i+1}\), for \(i=1,2, \ldots , j-1\). Moreover \(S_{6k_j+1}=S_{s_1-1}\) is placed in \(L_{j+1}\). Clearly, all packed tetrahedra \(S_1, \ldots , S_{s_1-1}\) are contained in \(H_1\).

Lemma 1

The sum of volumes of \(S_2, \ldots , S_{s_1}\) is greater than the volume of \(H_1^*\).

Proof

It is easy to verify that \(\ c^3/2+d^3/2 \ge cd^2,\) provided that \(\ 0<d\le c <1\). Consequently,

$$\begin{aligned} \frac{1}{6}a_2^3+\frac{1}{6}a_3^3+\ldots + \frac{1}{6}a_7^3 \ge 3 \cdot \frac{1}{6}a_4^3 + 3\cdot \frac{1}{6}a_7^3 \ge a_4a_7^2 \ge a_4a_{6k_1+1}^2. \end{aligned}$$

Moreover, if \(a_1+a_4+a_7\le 1\), then

$$\begin{aligned} \frac{1}{6}a_8^3+\ldots + \frac{1}{6}a_{13}^3 \ge 3\cdot \frac{1}{6} a_{10}^3+3\cdot \frac{1}{6} a_{13}^3\ge a_{10}a_{6k_1+1}^2, \end{aligned}$$

etc. This means that that the total volume of tetrahedra \(S_2, S_3, \ldots , S_{6k_1+1}\) is greater than the volume of \(L_1^*\) (see Fig. 4). For the same reason, the total volume of tetrahedra \(S_{6k_i+2}, \ldots , S_{6k_{i+1}+1}\) is greater than the volume of \(L_{i+1}^*\), for \(i=1,2, \ldots , j-1\). This implies that the sum of volumes of \(S_2, \ldots , S_{s_1}\) is greater than the volume of \(H_1^*\) (see Fig. 5 ).

Fig. 5
figure 5

Covering \(H_1^*\) with \(L_1, L_2, \ldots , L_{k_j}\) and \(S_{s_1}\)

Fig. 6
figure 6

\(a_1+a_{6k_1+1}+\ldots +a_{6k_m+1}+a_{6k_m+2}\le 1\ \) and \(a_1+a_{6k_1+1}+\ldots +a_{6k_m+1}+a_{6k_m+1}> 1\ \)

An additional explanation is required if \(\ a_1+a_{6k_1+1}+ \ldots + a_{6k_m+1}+a_{6k_m+2} \le 1\), while \(\ a_1+a_{6k_1+1}+ \ldots + a_{6k_m+1}+a_{6k_m+1} > 1\ \) for some integer m. We can imagine that \(S_{80}\) can be packed into \(L_6\) in Fig. 3 (left). In this case there is enough space to pack \(\ S_{6k_m+1}, S_{6k_m+2}\) and \(\ S_{6k_m+3}\) into the part of \(L_{m+1}\) with \(\ y \ge 1-a_{6k_m+1}\) (see Fig. 6). Clearly, at least three next tetrahedra \(\ S_{6k_m+4}, S_{6k_m+5}\) and \(\ S_{6k_m+6}\) can be packed into this layer. Moreover, \(m<j\), i.e., \(S_{6k_{m+1}+1 }\) is placed in the next layer in \(H_1\). \(\square \)

Packing method, Part II.

Similar to Lemma 1, one can find an integer \(s_2\) such that tetrahedra \(S_{s_1}, \ldots , S_{s_2-1}\) can be packed into \(H_2\) and that the sum of volumes of \(S_{s_1+1}, \ldots S_{s_2}\) is greater than the volume of \(H_2^*\), where

$$\begin{aligned} H_2 = H_1^* = (1-a_1)T_r \setminus (1-a_1-a_{s_1})T_r \end{aligned}$$

and

$$\begin{aligned} H_2^* = (1-a_1-a_{s_1})T_r \setminus (1-a_1-a_{s_1}-a_{s_2})T_r. \end{aligned}$$

Assume that \(s_2, \ldots , s_k\) are defined for \(k\ge 2\) and that \(\ a_1+a_{s_1}+\ldots + a_{s_k} \le 1\). Then there exists an integer \(s_{k+1}\) such that tetrahedra \(S_{s_k}, \ldots , S_{s_{k+1}-1}\) can be packed into \(H_{k+1}\) and that the sum of volumes of \(S_{s_k+1}, \ldots , S_{s_{k+1}}\) is greater than the volume of \(H_{k+1}^*\), where \(\ H_{k+1} = H_k^* \ \) and

$$\begin{aligned} H_{k+1}^*= (1-a_1- a_{s_1} - \ldots -a_{s_{k}} )T_r \setminus (1-a_1-a_{s_1}- \ldots -a_{s_{k+1}})T_r, \end{aligned}$$

provided that \(\ a_1 +a_{s_1}+\ldots + a_{s_{k+1}} \le 1\ \) or

$$\begin{aligned} H_{k+1}^*= (1-a_1- a_{s_1} - \ldots -a_{s_{k}} )T_r, \end{aligned}$$

provided that \(\ a_1 +a_{s_1}+\ldots + a_{s_{k+1}} > 1\). In the latter case we stop the packing process; there is not enough space in \(T_r\) to create a new slice of height \(a_{s_{k+1}}\).

Theorem 2

Any collection of homothetic copies of \(T_r\), whose total volume does not exceed 1/24, can be packed into \(T_r\).

Proof

Let \(S_1, S_2, \ldots \) be a collection of homothetic copies of \(T_r\) with total volume not greater than 1/24. Without loss of generality we can assume that \(a_1\ge a_2 \ge \ldots \), where \(S_i=a_iT_r\), for \(i=1,2,\ldots \) Assume that the tetrahedra cannot be packed into \(T_r\) by the method described above. This implies that there is an integer l such that \(a_1+a_{s_1}+\ldots +a_{s_l}>1\). The sum of volumes of \(S_2, \ldots , S_{s_l}\) is greater than the sum of volumes of \(H_1^*, \ldots , H_{s_l}^*\), i.e., is greater than \(\frac{1}{6}(1-a_1)^3\). As a consequence, the total volume of all tetrahedra \(S_1, S_2, \ldots \) is greater than

$$\begin{aligned} \frac{1}{6}a_1^3 + \frac{1}{6} (1-a_1)^3 > \frac{1}{6} \cdot \left( \frac{1}{2} \right) ^3 + \frac{1}{6} \cdot \left( \frac{1}{2} \right) ^3 = \frac{1}{24}, \end{aligned}$$

which is a contradiction. \(\square \)