1 Introduction

Let \(C, C_1, C_2, C_3, \ldots \) be planar convex bodies. We say that \(\ C_1, C_2, \ldots \) can be packed into 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. If the area of C is equal to the sum of areas of the bodies, then the packing is perfect.

There are many results concerning packings. For example, Moon and Moser showed [12] that any collection of squares whose total area does not exceed 1/2 can be packed into a square of sidelength 1. Richardson [15] proved that any collection of triangles homothetic to T, whose total area does not exceed half the area of T, can be packed in T and made the conjecture that such a result is true also for the translative packing by positive homothetic copies. This has been confirmed in [4].

In this note, we will study perfect packing.

In 1966 Moser posed the following well-known problem (see problem LM6 in [13]): Find the smallest \(\varepsilon \ge 0\) such that the squares of harmonic sidelengths \(\ 1/2, 1/3, 1/4, \ldots \ \) can be packed into a rectangle of area \(\ \frac{1}{6}\pi ^2-1+\varepsilon \ \) (the sum of areas of the squares equals \(\ \frac{1}{6}\pi ^2-1\)).

The following upper bounds for \(\varepsilon \) were obtained sequentially: \(\ 1/20\) [11], 1/127 [8], 1/198 [1], and 1/1244918662 (see [14, 9], and [3]).

This packing problem can be extended. Let \(S_n^t\) be a square of sidelength \(n^{-t}\) for \(\ n=1,2,\ldots \). If \(\ t> 1/2\), then the total area of the squares is equal to \(\ \sum _{n=1}^{\infty } \frac{1}{n^{2t}} = \zeta (2t)\), where \( \zeta ( s)\) is the Riemann zeta function. The question is whether \(\ S_1^t, S_2^t, \ldots \) (for \(t>1/2\)) can be packed perfectly into a rectangle. Obviously, for \(t=1\), we get Moser’s original question.

Some results are known for \(t < 1\). Chalcraft [2] showed that \(\ S_1^t, S_2^t, S_3^t, \ldots \ \) can be packed perfectly into a square for all t in the range [0.5964, 0.6]. Joós [10] checked that these squares can be packed perfectly for all t in the range \([\log _3 2, 2/3]\) (\(\log _3 2 \approx 0.63\)). Wästlund [17] proved that \(\ S_1^t, S_2^t, S_3^t, \ldots \ \) can be packed into a finite collection of squares of the same area as the sum of areas of the squares, provided that \(\ 1/2< t < 2/3\). In [5] it is shown that for all t in the range \(\ (1/2, 2/3]\), the squares \(\ S_1^t, S_2^t, S_3^t, \ldots \ \) can be packed perfectly into a single square. Recently, Tao [16] proved that for any \(\ 1/2<t<1\), and any \(n_0\) that is sufficiently large depending on t, the squares \(\ S_{n_0}^t, S_{n_0+1}^t, \ldots \) can be packed perfectly into a square.

In this note, we will give an analog of this problem for the packing of triangles.

Let \(T_n^t\) be an equilateral triangle of sidelength \(n^{-t}\) for \(\ n=1,2,\ldots \). The question arises whether \(\ T_1^t, T_2^t, \ldots \) (for \(t>1/2\)) can be packed perfectly into an equilateral triangle. More precisely: whether \(\ T_1^t, T_2^t, \ldots \) can be packed perfectly, for \(\ 0.5 <t \le 0.761202... \) (now \(1+1/2^{t} \le \sqrt{\zeta (2t)}\)); whether \(\ T_2^t, T_3^t, \ldots \) can be packed perfectly, provided that \(\ 0.761202... < t \le \ 0.943674...\) (for such values of t the sum of sidelengths of \(T_2^t\) and \(T_3^t\) is smaller than \(\sqrt{\zeta (2t)-1}\)); whether \(\ T_3^t, T_4^t, \ldots \) can be packed perfectly, provided that \(\ 0.943674...<t\le 1.121936...\) (now \(1/3^t +1/4^t \le \sqrt{\zeta (2t)-1-1/2^{2t}}\)), etc.. It is only known [7] that \(T_3^1, T_4^1, \ldots \) can be packed into an equilateral triangle of side of length \((\pi ^2/6-5/4)^{1/2}+1/270\).

In Sect. 2, we will show that equilateral triangles of sidelengths 1, \(2^{-t}\), \(3^{-t}\), \(4^{-t},\ldots \ \) can be packed perfectly into an equilateral triangle \(T_t\), provided that \(\ 1/2<t \le 37/72\).

In Sect. 3, we will check that if \(\ 1/2<t \le 37/72\), then all packed triangles can be positive homothetic copies of \(T_t\) as well as all packed triangles can be negative homothetic copies of \(T_t\).

In addition, in Sect. 4, we will consider square-packing. We will prove that, for \(\ 1/2<t \le (154+3\sqrt{2})/306\), squares of sidelengths 1, \(2^{-t}\), \(3^{-t}\), \(4^{-t},\ldots \ \) can be packed perfectly into a square \(S_t\) in such a way that some squares have a side parallel to a diagonal of \(S_t\) and the remaining squares have a side parallel to a side of \(S_t\).

2 Perfect Packing

Let t be a fixed number from the interval \(\ (1/2, 37/72] \) and let \(T_t\) be an equilateral triangle of area \(\ \frac{\sqrt{3}}{4} \zeta (2t)\).

The outline of the packing method is as follows. For each \(n\ge 2\), the empty space in \(T_t\), i.e., the part of \(T_t\) not covered by packed triangles \(\ T_1^t, \ldots , T_{n-1}^t\), will be divided into at most \(3(n-1)\) trapezoids. Then, \(T_n^t\) will be packed into a corner of one of these trapezoids.

Let R(lu) be an isosceles trapezoid with legs of length l, with the measure of the base angles equal to \(60^{\circ }\) and with the shorter base of length u. Clearly, the length of the longer base of R(lu) is equal to \(l+u\). If \(u=0\), then R(l, 0) is an equilateral triangle.

A trapezoid R(lu) is x-big, provided that \(\ l\ge 2x\) (see Fig. 1, right).

A trapezoid R(lu) is standard, provided that \(\ u \le l\).

Obviously, each x-big trapezoid is also v-big for any \(v<x\). Moreover, each standard trapezoid is x-big for sufficiently small x.

Fig. 1
figure 1

Standard trapezoids R(lu)

Proposition 1

The area of any standard trapezoid that is not x-big is smaller than \( 3\sqrt{3}x^2\).

Proof

The area of any standard trapezoid that is not x-big is smaller than three times the area of an equilateral triangle of sidelength 2x (see Fig. 1, left), i.e., is smaller than \( 3\cdot \frac{\sqrt{3}}{4} \cdot (2x)^2=3\sqrt{3}x^2\). \(\square \)

Lemma 2

Let R(lu) be an x-big trapezoid. Then R(lu) can be divided into either four or five parts: an equilateral triangle of sidelength x and at most four trapezoids which are either standard or x-big.

Proof

Let \(T_i\) be the equilateral triangle of sidelength x. We divide R(lu) into \(T_i\) (denoted by“+”in Figs. 2 and 3) and four sets \(A_i\), \(B_i\), \(C_i\), \(D_i\) which are either standard trapezoids or x-big trapezoids (possibly some of them are triangles), or the empty set.

Case 1: \(l\ge 4x\). The trapezoid R(lu) is divided into: \(\ T_i \cup A_i \cup B_i \cup C_i \), where \(A_i=R(x,x)\), \(B_i=R(l-2x,u)\) and \(C_i=R(2x,l+u-3x)\). The trapezoid \(A_i\) is standard. Moreover, \(B_i\) and \(C_i\) are x-big (see Fig. 2, left, when \(l>4x\)). In this case \(D_i = \emptyset \).

Case 2: \(3x\le l<4x\).

Subcase 2a: \(u<x\). The trapezoid R(lu) is divided into: \(T_i\), two standard trapezoids \(A_i=R(x,x)\), \(B_i=R(l-2x,u)\) and one x-big trapezoid \(C_i=R(2x,l+u-3x)\) (as in Fig. 2, left, when \(3x<l\le 4x\)). We take \(D_i = \emptyset \).

Subcase 2b: \(u\ge x\). The trapezoid R(lu) is divided into: \(T_i\), three standard trapezoids \(A_i=R(x,x)\), \(B_i=R(x,0)\), \(C_i=R(x,l-3x)\) and one x-big trapezoid \(D_i=R(l,u-x)\) (as in Fig. 2, right).

Case 3: \(2x\le l<3x\).

Subcase 3a: \(u<x\) and \(l+u\ge 3x\). The trapezoid R(lu) is divided into: \(T_i\) and three standard trapezoids \(A_i=R(x,x)\), \(B_i=R(x,l+u-3x)\), and \(C_i=R(l-x,u)\) as in Fig. 3, left. We take \(D_i = \emptyset \).

Subcase 3b: \(u<x\) and \(l+u<3x\). The trapezoid R(lu) is divided into: \(T_i\) and three standard trapezoids \(A_i=R(x,l+u-2x)\), \(B_i=R(x,0)\), and \(C_i=R(l-x,u)\) as in Fig. 3, middle. We take \(D_i = \emptyset \).

Subcase 3c: \(u\ge x\). The trapezoid R(lu) is divided into: \(T_i\), three standard trapezoids \(A_i=B_i=R(x,0)\), \(C_i=R(x,l-2x)\) and one x-big trapezoid \(D_i=R(l,u-x)\) as in Fig. 3, right. \(\square \)

Fig. 2
figure 2

Divisions of the x-big trapezoid R(lu), when \(l\ge 3x\)

Fig. 3
figure 3

Divisions of the x-big trapezoid R(lu), when \(2x\le l< 3x\)

Since \(t\le 37/72\), it follows that the sidelength of \(T_t\) is greater or equal to \(\sqrt{ \zeta (37/36)} >6\).

2.1 Packing Method \(M_{\triangle }\)

[1]:

The first triangle is packed into the lower left vertex of \(T_t\). After packing \(T_1^t\), the uncovered part of \(T_t\) is divided into \(\ A_1 \cup B_1 \cup C_1\ \) as in the proof of Lemma 2 (comp. Fig. 2, left, for \(u=0\)). We take \(\ \mathcal {R}_1 = \{ A_1, B_1, C_1\}\).

[2]:

Assume that \(n>1\), that the triangles \(\ T_1^t, \ldots , T_{n-1}^t\) are packed into \(T_t\) and that the family \(\mathcal {R}_{n-1}\) is defined. We choose one of the \(n^{-t}\)-big trapezoids from \(\mathcal {R}_{n-1}\) in any way. Denote this trapezoid by R. We pack \(T_n^t\) into R at the vertex at the longer base of R. After packing \(T_n^t\), the uncovered part of R is divided into \(\ A_n \cup B_n \cup C_n \cup D_n\), as in the proof of Lemma 2, and we take \(\ \mathcal {R}_n = \bigl ( \mathcal {R}_{n-1}{\setminus } \{R\} \bigr ) \cup \{ A_n, B_n, C_n, D_n\}\). It is possible that \(D_n=\emptyset \).

Observe that \(\mathcal {R}_{n-1}\) contains at most \(3(n-1)\) trapezoids with mutually disjoint interiors, for any \(n\ge 2\). Each trapezoid from \(\mathcal {R}_{n-1}\) is either \(n^{-t}\)-big or standard. It is possible that a trapezoid from \(\mathcal {R}_{n-1}\) is both \(n^{-t}\)-big and standard.

Example 1

Figure 4 illustrates the initial stage of the packing process for \(t=0.51\). The area of \(T_{0.51}\) is equal to \(\zeta (1.02)\sqrt{3}/4 \approx 21.9\), thus the sidelength \(\sigma \) of \(T_{0.51}\) is greater than 7.11. After packing \(T_1^{0.51}\), the uncovered part of \(T_{0.51}\) is divided into three trapezoids: \(\ A_1 = R(1,1)\), \(B_1=R(\sigma -2,0)\) and \(C_1 = R(2, \sigma -3)\). Both trapezoids \(B_1\) and \(C_1\) are 2-big and therefore we have two possibilities to pack \(T_2^{0.51}\). For instance, we will pack each triangle into a trapezoid of maximum leg length. After packing \(T_2^{0.51}\) into \(B_1\), the uncovered part of \(T_{0.51}\) is partitioned into \(A_1\), \(C_1\), \(A_2\), \(B_2\) and \(C_2\) (\(D_2=\emptyset \)).

Fig. 4
figure 4

Packing method \(M_{\triangle }\) for \(t=0.51\)

Theorem 3

For each t in the range \(\ 1/2<t\le 37/72 \approx 0.5138\), the triangles \(T_n^t\) can be packed perfectly into the triangle \(T_t\) by the method \(M_{\triangle }\).

Proof

The proof is similar to that presented in [6]. Let t be a fixed number from the interval \(\ (1/2, 37/72] \). The area of \(T_t\) is equal to \(\ \frac{\sqrt{3}}{4} \sum _{i=1}^{\infty } \frac{1}{i^{2t}} = \frac{\sqrt{3}}{4} \zeta (2t) \ge \frac{\sqrt{3}}{4} \zeta (37/36) = \frac{\sqrt{3}}{4} \cdot 36.579\ldots \). Consequently, \(T_t\) is 1-big (its leg length is greater than 2). We pack \(T_1^t, T_2^t,... \) into \(T_t\) by the method \(M_{\triangle }\). To prove Theorem 3 it suffices to show that, for any n, there is at least one \(n^{-t}\)-big trapezoid in \(\mathcal {R}_{n-1}\) (into which \(T_n^t\) will be packed).

First, we estimate the sum of areas of trapezoids in \(\mathcal {R}_{n-1}\), i.e., the area of the uncovered part of \(T_t\) after packing \(T_{n-1}^t\). This value is equal to the sum of areas of unpacked triangles \(\ T_n^t, T_{n+1}^t, \ldots \), i.e., is equal to

$$\begin{aligned} \frac{\sqrt{3}}{4} \Bigl ( \frac{1}{n^{2t}} + \frac{1}{(n+1)^{2t}}+\ldots \Bigr )&> \frac{\sqrt{3}}{4} \cdot \int _{n}^{+\infty }\frac{1}{x^{2t}}\textrm{d}x = \frac{\sqrt{3}}{4} \cdot \frac{1}{2t-1}n^{1-2t} \\&\ge \frac{\sqrt{3}}{4} \cdot \frac{1}{2\cdot \frac{37}{72}-1}n^{1-2t}= 9\sqrt{3}n^{1-2t}. \end{aligned}$$

Assume that there is an integer n such that the triangle \(T_n^t\) cannot be packed into \(T_t\) by the method \(M_{\triangle }\), i.e., that there is no \(n^{-t}\)-big trapezoid in \(\mathcal {R}_{n-1}\). This means that all trapezoids in \(\mathcal {R}_{n-1}\) are standard and that the length of the leg of each such trapezoid is smaller than \(2n^{-t}\) (if \(l\ge 2n^{-t}\), then R(lu) is \(n^{-t}\)-big). By Proposition 1, the area of each such trapezoid is smaller than \(3\sqrt{3}n^{-2t}\). Since there are at most \(3(n-1)\) trapezoids in \(\mathcal {R}_{n-1}\), it follows that the total area of trapezoids in \(\mathcal {R}_{n-1}\) is smaller than \(\ (3n-3) \cdot 3\sqrt{3} n^{-2t}<9\sqrt{3}n^{1-2t}\), which is a contradiction.

Consequently, \(\ T_1^t, T_2^t, \ldots \ \) can be packed into \(T_t\). \(\square \)

3 Positive or Negative Copies

One side of \(T_r\) is called the bottom. In our drawings, the bottom is the horizontal side of the triangle. By \(T_+(l,u)\) we mean the trapezoid T(lu) lying such that its bases are parallel to the bottom of \(T_r\) and that the longer base is lower than the shortest one (see Fig. 5, left).

We call a trapezoid x-classic if it arises from rotating a trapezoid \(T_+(l,u)\) by a multiple of \(60^{\circ }\), where the respective T(lu) is x-big or standard (see Fig. 5, left and right).

The image of any positive homothetic copy of \(T_r\) in rotation of \(60^{\circ }\) is a negative homothetic copy of \(T_r\) as well as the image of any negative homothetic copy of \(T_r\) in rotation of \(60^{\circ }\) is a positive homothetic copy of \(T_r\). The image of a triangle T(x, 0) in rotation of \(180^{\circ }\) is denoted by \(-T(x,0)\).

Fig. 5
figure 5

Standard trapezoids T(lu)

Lemma 4

Let R be an x-big classic trapezoid. Then R can be divided into: \(T_+(x,0)\) and at most four x-classic trapezoids. Moreover, R can be divided into: \(-T_+(x,0)\) and at most four x-classic trapezoids.

Proof

Let R be the x-big classic trapezoid.

Since R is the image of \(T_+(l,u)\) in rotation by a multiple of \(60^\circ \), it follows that R can be divided into the same shapes as \(T_+(l,u)\), with possibly switching roles of \(T_+(x,0)\) and \(-T_+(x,0)\) (see Fig. 6). Consequently, to prove Lemma 4 it suffices to check that \(T_+(l,u)\) can be divided into \(T_+(x,0)\) (denoted by “+”) and at most four x-classic trapezoids as well as that \(T_+(l,u)\) can be divided into \(-T_+(x,0)\) (denoted by“-”in Figs. 6, 7 and 8) and at most four x-classic trapezoids.

The first option was discussed in the proof of Lemma 2. Now consider negative copies of the triangle.

Case 1: \(l\ge 4x\). The trapezoid R is divided into: \(-T_+(x,0)\), one standard trapezoid and two x-big trapezoids (as in Fig. 7, left; if \(l\ge 4x\), then the upper trapezoid is x-big).

Case 2: \(3x\le l<4x\). If \(u<x\), then \(T_+(l,u)\) is divided as in Fig. 7, left (now the upper trapezoid is standard). If \(u\ge x\), then \(T_+(l,u)\) is divided as in Fig. 7, right.

Case 3: \(2x\le l<3x\). If \(u<x\) and \(l+u\ge 3x\), then \(T_+(l,u)\) is divided as in Fig. 8, left. If \(u<x\) and \(l+u<3x\), then \(T_+(l,u)\) is divided as in Fig 8, middle. If \(u\ge x\), then \(T_+(l,u)\) is divided as in Fig. 8, right. \(\square \)

Fig. 6
figure 6

Divisions of the x-big classic trapezoid R, when \(l\ge 4x\)

Fig. 7
figure 7

Divisions of the x-big trapezoid \(T_+(l,u)\), when \(l \ge 3x\)

Fig. 8
figure 8

Divisions of the x-big trapezoid \(T_+(l,u)\), when \(2x\le l< 3x\)

3.1 Packing Method \(M_{\triangle ^{\pm }}\)

Let \(\mathbb {N} = \mathbb {A} \cup \mathbb {B}\), where \(\mathbb {A} \cap \mathbb {B} = \emptyset \). Since \(t\le 37/72\), it follows that the sidelength of \(T_t\) is greater or equal to \(\sqrt{ \zeta (37/36)} >6\).

[1]:

If \( 1 \in \mathbb {A}\), then the first triangle is packed into \(T_t\) in the place “+” described in the proof of Lemma 2 (see Fig. 2, left, if \(u=0\)). If \( 1 \in \mathbb {B}\), then the first triangle is packed in the place “-” described in the proof of Lemma 4 (see Fig. 7, left, if \(l>4x\) and \(u=0\)). After packing \(T_1^t\), the uncovered part of \(T_t\) is divided into three 1-classic trapezoids as in the proof of Lemma 4. We take as \(\ \mathcal {R}^{\pm }_1\) the family of these three trapezoids.

[2]:

Assume that \(n>1\), that the triangles \(\ T_1^t, \ldots , T_{n-1}^t\) are packed into \(T_t\) and that the family \(\mathcal {R^{\pm }}_{n-1}\) is defined. We choose one of the \(n^{-t}\)-big trapezoids from \(\mathcal {R^ {\pm }}_{n-1}\) in any way. Denote this trapezoid by R. If \( n \in \mathbb {A}\), then \(T_n^t\) is packed in the place “+” in R; if \( n \in \mathbb {B}\), then \(T_n^t\) is packed in the place “-” in R (see Figs. 2, 3, 6, 7 or 8). After packing \(T_n^t\), the uncovered part of R is divided into at most four \(n^{-t}\)-classic trapezoids as in the proof of Lemma 4. We take as \(\ \mathcal {R^{\pm }}_n \) the union of the family of these trapezoids and \( \mathcal {R^{\pm }}_{n-1}\setminus \{R\} \).

Fig. 9
figure 9

Packing method for \(t=0.51\) and \(\mathbb {B}=\emptyset \)

Clearly, \(\mathcal {R^{\pm }}_{n-1}\) contains at most \(3(n-1)\) trapezoids with mutually disjoint interiors, for any \(n\ge 2\). Each trapezoid from \(\mathcal {R^{\pm }}_{n-1}\) is either \(n^{-t}\)-big or standard.

Figure 9 illustrates the initial stage of the packing process in the case when \(\mathbb {A} = \mathbb {N}\) and \( \mathbb {B}=\emptyset \); note that the first twelve triangles are packed in the same places using the algorithms \(M_{\triangle }\) and \(M_{\triangle ^{\pm }}\). On the other hand, \(\mathbb {B} = \mathbb {N}\) and \( \mathbb {A}=\emptyset \) in Fig. 10.

Theorem 5

For each t in the range \(\ 1/2<t\le 37/72\), the triangles \(T_n^t\) can be packed perfectly into the triangle \(T_t\) so that each packed triangle \(T_n\) is a positive homothetic copy of \(T_n^t\), provided that \(n \in \mathbb {A}\) and that each packed triangle \(T_n^t\) is a negative homothetic copy of \(T_t\), provided that \(n \in \mathbb {B}\).

Proof

We pack \(T_1^t, T_2^t,... \) into \(T_t\) by the method \(M_{\triangle ^{\pm }}\). As in the proof of Theorem 3, the sum of areas of trapezoids in \(\mathcal {R^{\pm }}_{n-1}\) is greater than \(\ 9\sqrt{3}n^{1-2t}.\) If there is an integer n such that the triangle \(T_n^t\) cannot be packed into \(T_t\) by our method, then the total area of trapezoids in \(\mathcal {R^{\pm }}_{n-1}\) is smaller than \(\ (3n-3) \cdot 3\sqrt{3} n^{-2t}< 9\sqrt{3}n^{1-2t}\), which is a contradiction. \(\square \)

Fig. 10
figure 10

Packing method for \(t=0.51\) and \(\mathbb {A}=\emptyset \)

4 Squares

Denote by \(S_t\) the square of area \(\zeta (2t)\), where \(\ 1/2<t\le (154+3\sqrt{2})/306\approx 0.517\). Let P(ha) be a right trapezoid with height h and with bases of length a and \(a-h\). In this section, a trapezoid P(ha) is x-big, provided that \(\ h\ge \frac{3\sqrt{2}}{2}x\). A trapezoid P(ha) is standard, if \(\ a \le \frac{3\sqrt{2}}{2}h\). A trapezoid P(ha) is x-classic, provided that it is either standard or x-big and provided that its bases are parallel either to a side of \(S_t\) or to a diagonal of \(S_t\). Observe that the area of each standard trapezoid that is not x-big is smaller than \( \ \bigl ( \frac{3\sqrt{2}}{2}-\frac{1}{2} \bigr ) \bigl ( \frac{3\sqrt{2}}{2} x \bigr )^2 = \frac{27\sqrt{2}-9}{4} \cdot x^2\).

Lemma 6

Let P be an x-big classic trapezoid. Then P can be divided into: a square of sidelength x with a side parallel to the bases of P and at most four x-classic trapezoids. Moreover, P can be divided into: a square of sidelength x with a diagonal parallel to the bases of P and at most five x-classic trapezoids.

Proof

Observe that P(ha) can be divided (see Fig. 11) into a square of sidelength x with a side parallel to bases of P(ha) and into:

Case 1::

two standard trapezoids P(xx) and \(P(x, (3\sqrt{2}/2-1)x)\) and two big trapezoids: \(P(3\sqrt{2}x/2, a-x)\) and \(P(h-3\sqrt{2}x/2, a- 3\sqrt{2}x/2)\), provided that \(h\ge 3 \sqrt{2}x\);

Case 2::

two standard trapezoids P(xh/2) and one big trapezoid \(P(h, a-x)\), provided that \(h< 3 \sqrt{2}x\) and \(a-x \ge h\); clearly \(h/2< 3 \sqrt{2}x/2\);

Case 3::

four standard trapezoids: \(P(h-x,a-x)\), P(xx) and two trapezoids \(P(x, (a-x)/2)\), provided that \(a-x<h< 3 \sqrt{2}x\) and \(a>3x\); clearly \((a-x)/2< 3 \sqrt{2}x/2\) as well as \(a-x<h=2(h-x)-h+2x\le 2(h-x)-3 \sqrt{2}x/2+2x<2(h-x)\);

Case 4::

two standard trapezoids: \(P(h-x,a-x)\) and \(P(x, a-x)\), provided that \(a-x<h< 3 \sqrt{2}x\) and \(a\le 3x\); clearly \(a-x\le 3x-x=2x\) as well as \(a-x<h<2(h-x)\).

Moreover, P(ha) can be divided (see Fig. 12) into a square of sidelength x with a diagonal parallel to bases of P(ha) and into:

Case 5::

two standard trapezoids P(xx) and P(x, 2x) and two big trapezoids: \(P(3\sqrt{2}x/2, a-\sqrt{2}x)\) and \(P(h-3\sqrt{2}x/2, a-3\sqrt{2}x/2)\), provided that \(h\ge 3 \sqrt{2}x\);

Case 6::

five standard trapezoids: \(P(z,\sqrt{2}x)\), \(P(h-z, h-z)\), where \(z=\sqrt{2}x-(a-h)\), \(P(\sqrt{2}x, \sqrt{2}x)\) and two trapezoids \(P(x, (h-z)\sqrt{2}/2-x/2)\), provided that \(h< 3 \sqrt{2}x\) and \( z > a-h\); since \(z+a-h=\sqrt{2}x\) and \(z>a-h\), it follows that \(z>\sqrt{2}x/2\); clearly, \(\ \frac{\sqrt{2}}{2}(h-z)-\frac{1}{2}x < \frac{\sqrt{2}}{2}\Bigl (3\sqrt{2}x-\frac{\sqrt{2}}{2}x\Bigr ) - \frac{1}{2}x=2x\);

Case 7::

five standard trapezoids: \(P(h-z,h-z)\), \(P(a-h,\sqrt{2}x)\), \(P(\sqrt{2}x, \sqrt{2}x)\) and two trapezoids \(P(x, h\sqrt{2}/2-x)\), provided that \(5\sqrt{2}x/2 \le h< 3 \sqrt{2}x\) and \( z \le a-h< \sqrt{2}x\); clearly, \(\ h \frac{\sqrt{2}}{2}-x < 3 \sqrt{2}x\cdot \frac{\sqrt{2}}{2}-x =2x\) as well as \(\sqrt{2}x = z+a-h \le a-h+a-h=2(a-h)\);

Case 8::

four standard trapezoids: \(P(h-z,h-z)\), \(P(a-h,\sqrt{2}x)\) and two trapezoids \(P(x, h\sqrt{2}/2-x/2)\), provided that \(h<5\sqrt{2}x/2\) and \( z \le a-h<\sqrt{2}x\); clearly, \(\ h\sqrt{2}/2-x/2 < (5\sqrt{2}x/2) \cdot (\sqrt{2}/2)-x/2=2x\) as well as \(\sqrt{2}x = z+a-h \le a-h+a-h=2(a-h)\);

Case 9::

one big trapezoid \(P(h, a - \sqrt{2}x)\) and four standard trapezoids \(P(x, (h\sqrt{2}+x)/4)\), provided that \(h<3\sqrt{2}x\) and \(a-h\ge \sqrt{2}x\); clearly, \(\ (h\sqrt{2}+x)/4< (3\sqrt{2}x\cdot \sqrt{2}+x)/4 <\frac{3\sqrt{2}}{2}x\). \(\square \)

Fig. 11
figure 11

Divisions of the x-big classic trapezoid P(ha), when a side of the square is parallel to the bases of P(ha)

Fig. 12
figure 12

Divisions of the x-big classic trapezoid P(ha), when a diagonal of the square is parallel to the bases of P(ha)

Let \(\mathbb {N} = \mathbb {A} \cup \mathbb {B}\), where \(\mathbb {A} \cap \mathbb {B} = \emptyset \). Since \(t\le (154+3\sqrt{2})/306\), it follows that the sidelength of \(S_t\) is greater or equal to \(\sqrt{ \zeta ((154+3\sqrt{2})/153)} >5.4\).

4.1 Packing Method \(M_{\square }\)

[1]:

The square \(S_t\) is partitioned into two right isosceles triangles: \(A_1\) containing the lower left vertex of \(S_t\) and \(B_1\) containing the upper right vertex of \(S_t\). We choose one of them, say \(A_1\), and pack the first square into it in the following way. If \( 1 \in \mathbb {A}\), then the first square is packed in the place marked in Fig. 11 (the upper left picture). If \( 1 \in \mathbb {B}\), then the first square is packed in the place marked in Fig. 12 (the upper left picture). After packing \(S_1^t\), the uncovered part of \(A_1\) is divided into four 1-classic trapezoids as in Figs. 11 and 12. We take as \(\ \mathcal {P}_1\) the union of the family of these four trapezoids and \(\{B_1\}\).

[2]:

Assume that \(n>1\), that the squares \(\ S_1^t, \ldots , S_{n-1}^t\) are packed into \(S_t\) and that the family \(\mathcal {S}_{n-1}\) is defined. We choose one of the \(n^{-t}\)-big trapezoids from \(\mathcal {P}_{n-1}\) in any way. Denote this trapezoid by P. The square \(S_n^t\) is packed into P in the place marked in Figs. 11 and 12. After packing \(S_n^t\), the uncovered part of P is divided into at most five trapezoids. We take as \(\ \mathcal {P}_n \) the union of the family of these trapezoids and \( \mathcal {P}_{n-1}\setminus \{P\} \).

Figure 13 illustrates the initial stage of the square-packing process when \(\mathbb {A}\) is the set of even numbers and \(\mathbb {B}\) is the set of odd numbers.

Theorem 7

For each t in the range \(\ 1/2<t\le (154+3\sqrt{2})/306\approx 0.517\), the squares \(S_n^t\) can be packed perfectly into the square \(S_t\) so that a side of each packed square \(S_i\) is parallel to a side of \(S_t\) for \(i \in \mathbb {A}\) while a side of each packed square \(S_i\) is parallel to a diagonal of \(S_t\) for \(i \in \mathbb {B}\).

Proof

Let t be a fixed number from the interval \(\ (1/2, (154+3\sqrt{2})/306] \). We place \(S_1^t, S_2^t, \ldots \) by the method \(M_{\square }\). The sum of areas of trapezoids in \(\mathcal {P}_{n-1}\) is greater than

$$\begin{aligned} \int _{n}^{+\infty }\frac{1}{x^{2t}}\textrm{d}x \ge \frac{1}{2\cdot \frac{154+3\sqrt{2}}{306}-1} \cdot n^{1-2t} = (27\sqrt{2}-9)n^{1-2t}. \end{aligned}$$

Assume that there is an integer n such that the square \(S_n^t\) cannot be packed into \(S_t\) by our method, i.e., that there is no \(n^{-t}\)-big trapezoid in \(\mathcal {P}_{n-1}\). Since there are at most 4n trapezoids in \(\mathcal {P}_{n-1}\), it follows that the total area of trapezoids in \(\mathcal {P}_{n-1}\) is smaller than \(\ 4n \cdot \frac{27\sqrt{2}-9}{4} \cdot n^{-2t} = (27\sqrt{2}-9)n^{1-2t}\), which is a contradiction. \(\square \)

Fig. 13
figure 13

Square-packing method for \(t=0.51\)