an analysis of Mosers worm

This article is based off a previous writeup on the subject. To read that article, click here.

Introduction

#

Moser’s Worm is an important unsolved problem in geometry with a very amusing name. The name comes from an odd hypothetical. Suppose you have an inch-long pet worm, who’s a very restless sleeper. When the worm goes to bed, there’s no way of knowing how it might be twisted or curved. Nonetheless, you want to cover your worm with a little blanket. What’s the smallest blanket you could use, regardless of how the worm is positioned?

Although many unsolved problems in mathematics are known for their incomprehensibility, the standard statement of Moser’s worm only requires requires one definition:

From here on out, I might be a little bit looser. I’ll be certain to eliminate ambiguity whenever necesary, but the point of these notes is mostly to ensure that these concepts are written down somewhere they’re easy to locate later on, without necessarily being up to the hyper-rigorous standards of mathematical publication. This is my website, after all. Without further ado, here’s the statement of the problem:

It’s a shockingly simple question for something that has remained unsolved for so long! Although my background is in mathematics, and I studied differential geometry in my undergrad, this question in particular is one I don’t know a lot about. It’s a bit refreshing to work on something unrelated to what you normally do, so I’m looking forward to it! I should mention that I’m deliberately not looking into the already existing body of work on this problem. This is partly for my own satisfaction of discovering things for myself, and partly to avoid bias. Although I don’t suppose that I could fully crack an infamous problem outside my own specialty, it can still be very invigorating for a field to get a fresh pair of eyes and a fresh perspective.

There’s also the second benefit of writing an article on a subject from an outsider’s perspective, which is that it should be very approachable for the reader. Because I don’t know any of the jargon, techniques, or theorems in this field, I can’t possibly confuse a reader by using them. That said, I will use some basic calculus at a couple points.

Type 1 and Type 2 Curves

#

My goal here is to see what curves we don’t need to worry about. There’s two prongs to this:

1. Show that curve \(A\) is always accommodated when curve \(B\) is accommodated.
2. Show that when curves \(A\) and \(B\) are accommodated, curve \(C\) is also accommodated.

To that end, we give these definitions:

This entire problem exists in euclidean space. If \(d\) is the euclidean distance metric, and \(a\) and \(b\) are points, then \(\ell\langle a,b\rangle\) will be used as a synonym for \(d(a,b)\)

Immediately there are some lemmas we can form regarding taut curves without any further context. For example, this is the lemma that inspired the choice of the word “taut”:

proof:
Let \(X\) be a unit-length curve, and let \(x\) be a contiguous section of \(X\) that does not touch the border of \([X]\), except at two endpoints, \(a\) and \(b\). Then let \(X’\) be identical to \(X\), except that \(x\) is replaced with a straight line from \(a\) to \(b\). The length of \(X’\) is necessarily less than 1, although \(X’\) has the same convex hull. Finally, let \(X’’\) be the unit-length curve that includes all of \(X’\), but with an added length at the end. \(X’’\) must have a convex hull that includes everything in \([X]\) and more.

Hopefully, this drawing explains why I chose the word "taut" to describe this property. The curve on the left is "taut", in the sense that it could be stretched to coincide with the border of its convex hull, and that extra length could be used to create a larger hull. The light grey area is the same between the two shapes, but the dark grey area is added by using length more efficiently

Here’s another brief and straightforward observation about taut curves.

Let’s start our effort to characterize taut curves with yet another definition

An example of a Type 1 and Type 2 curve.

Type 1 and Type 2 curves can also be thought of in terms of path direction. A path along a Type 1 curve moves entirely either clockwise or counterclockwise across the edge of its convex hull. A path along a Type 2 curve starts moving either clockwise or counterclockwise along the border, but then moves in the opposite direction for a length.

Here’s a brief lemma about Type \(n\) curves that will be useful in several results coming up

proof:
Suppose, for the sake of contradiction, that \(X_a^b\) does not have zero curvature. Then \(\ell X_a^b>\ell\langle a,b\rangle\). Let \(X’\) be formed by replacing \(X_a^b\) in \(X\) with \(\langle a,b\rangle\). It is clear that \(\ell X’ < X\), and so \(X\) is accommodated by \(X’\), and is not taut.

I’ve only talked about Type 1 and Type 2 curves so far. Why hasn’t any special attention been given to curves of other types? As it turns out, these curves are never taut.

proof:
To force a contradiction, let \(X\) be such a curve. Let \(a\), \(b\), \(c\), \(d\), \(e\), \(f\), and \(g\) be points on \(X\), appearing in that order, with the added requirements that:

1. \(a\), \(b\), \(c\), \(d\), \(e\), \(f\), and \(g\) are all on the boundary of \([X]\).
2. \(a\) and \(g\) are the endpoints of \(X\).
3. \(d\) is the point furthest from \(\langle a,g \rangle\), on the opposite side from \(b\) and \(f\)
4. \(\langle b,c \rangle\) is internal to the convex hull of \(X\).
5. \(\langle e,f \rangle\) is internal to the convex hull of \(X\).

Here’s a drawing to explain.

For ease of upcoming calculations, let \(\langle a,g\rangle\) lie on the \(x\) axis, with \(a\) existing at \(0,0\).

Let \(\phi\) be the minimal distance from \(d\) to \(\langle a,g \rangle\). It immediately follows that:

$$ \frac{\ell X_b^f}{2}\leq\phi $$

In this case, define \(a’\) and \(g’\) to be these two points:

$$ a’=\left( 0, -\frac{\ell X_b^f}{2}\right) $$

$$ g’=\left( \ell\langle a,g\rangle, -\frac{\ell X_b^f}{2}\right) $$

Then we can define \(X’\) to be the curve

$$ X’ = \langle a’,a\rangle \cup X_a^b \cup \langle b,f\rangle \cup X_f^g \cup \langle g,g’\rangle $$

This is given in the drawing below:

Because \(a’\) and \(g’\) are both below \(d\), the entirety of \(X\) must be inside \(X’\).

Let’s limit a bit further which curves can be Type 1.

proof:
Let \(X\) be a curve as described above, and let \(c\) be the point on \(X\) that lies furthest from any line perpendicular to \(\langle a, b\rangle\). without loss of generality, assume \(c\) is closer to \(a\) than it is to \(b\).

Let \(\alpha\) be the line that includes the line segment \(\langle a, b\rangle\), and let \(c’\) be the point on \(\alpha\) closest to \(c\). Let \(X’\) be a curve identical to \(X\), except that the section between \(c\) and \(a\) is replaced with a straight line between \(c\) and \(c’\).

This situation is illustrated below:

It is clear that

$$ \ell\langle c,c’\rangle < \ell C_c^a $$

and therefore

$$ \ell X’<\ell X $$

Therefore, \(X’\) has a length less than \(X\), and so is a unit curve. The convex hull of \(X’\) can accommodate \(X\), but is larger. By definition, \(X\) is not taut.

Finally, these past few lemmas and a few other ideas are compiled in a theorem that helps to characterize Type 1 curves:

proof:
The fist of these requirements is implied by the definition of a Type 1 curve, and the second is implied by Lemma 3. Therefore, this proof focuses on the third requirement

First, suppose that a set of points as described exists. Without loss of generality, let \(f=a\). As before, I’ll include a diagram to clarify what the situation described in the lemma statement is.

Let \(X’ = \langle c,a,b\rangle \cup X_b^e \cup \langle e,d\rangle\). As evident from the diagram, the convex hull of \(X’\) accommodates the convex hull of \(X\). Next, consider the inequality assumed earlier:

$$\begin{align*} \ell \langle a,b\rangle + \ell \langle c,a\rangle + \ell\langle e,d\rangle &\leq \ell X_a^e \\ \ell \langle c,a,b\rangle + \ell\langle e,d\rangle &\leq \ell X_a^e \\ \ell \langle c,a,b\rangle + \ell X_b^e + \ell\langle e,d\rangle &\leq \ell X_a^e + \ell X_b^e \\ \ell X’ &\leq \ell X \end{align*}$$

Therefore, \([X’]\) both accommodates \([X]\) and includes other area as well, but \(X’\) is shorter than \(X\). By definition, \(X\) is not taught.

An “if and only if” proof requires that the statement is proven in both directions, so Theorem 2 will now be proved in the opposite direction. Suppose that such a set of points as \({c, d, e, f}\) does not exist, but that \(X_1\) is a Type 1 curve such that \([X_1] \sqsupset X\). Let \(c\) and \(d\) be the endpoints of \(X_1\). An example of what this might look like is shown below:

Next, let \(\alpha\) be the line passing through \(c\) and \(d\). let \(X_2\) be the shortest curve with endpoints on \(\alpha\) that accommodates \(X\). \(X_2\) must consist of two lines parallel to \(\alpha\), as well as a section of the boundary of \([X]\), which may or may not include \(\langle a,b\rangle\). Let \(c’\) and \(d’\) be the endpoints of \(X_2\), and let \(e\) and \(f\) be the points where \(X_2\) meets the convex hull of \(X\), as shown below:

Finally, the length of \(X_2\) can be reduced even further. Let \(d’’\) and \(c’’\) be points on \(\langle d’,e\rangle\) and \(\langle c’,e \rangle\) that are both a distance \(\phi\) away from \(d’\) and \(c’\), with \(\phi\) having the largest possible value without \(\langle d’’,c’’\rangle\) intersecting the interior of \([X]\).

Let \(X_3\) be \(X_2\), without the lines from \(d’\) to \(d’’\) and from \(c’\) to \(c’’\). If \(f\) and \(e\) are both endpoints of \(X\), then \(X_3\) must contain the entirety of \(X\). This is impossible, as \(\ell X = 1\), so either \(e\) or \(f\) is not an endpoint of \(X\).

Instead, assume (for the sake of contradiction), that neither \(f\) nor \(e\) are endpoints of \(X\). This would require that at least one of the two points is not on a line parallel to \(\langle a,b \rangle\).

Therefore, \(\lbrace c’’, f, e, d’’\rbrace\) satisfies the requirements of the lemma.

The reason I set out this lemma is because I want to demonstrate that there is a limit on the “height” of Type 1 curves. If \(\langle a,b \rangle\) is very short, and \(X\) has a point very distant from \(\langle a,b \rangle\), it would be easy to find a way to maximize \(\ell X_a^e\) and minimize \(\ell \langle a,c \rangle\)

This theorem serves to characterize type 1 curves fairly precisely. One of the broader goal of this project is to characterize Type 2 curves equally as well.

proof:
The locations of \(a\) though \(d\) can be clarified by this diagram:

Suppose for the sake of contradiction that \(\ell\langle a.c\rangle > \ell\langle b,c\rangle\). Define \(X’\) such that

$$ X’ = X_b^a \cup \langle a,c\rangle \cup X_c^d $$

Because \(\ell\langle a,c\rangle > \ell\langle b,c\rangle\), \(\ell X’ < \ell X = 1\). Then let \(X’’\) be identical to \(X’\), except that an extra bit is added to increase the area of the convex hull while keeping \(\ell X’’\leq 1\). By definition, \(X\) is not taut.

By symmetry, \(X\) is also not taut if \(\ell \langle b.d\rangle > \ell \langle b,c\rangle\).

The non-internal sections of a Type 2 curve behave something like Type 1 curves themselves. For example, consider the following Lemma:

proof:
First, for clarity, a diagram of these lines is shown:

First, if \(p\) is a point on \(X_c^d\) that is on the “wrong” side of \(L_1\), then \(c\) is an internal point of \([X]\). This is not true by assumption.

The point \(p\) must be on the exterior of \([X]\), by assumption, and so cannot be on the wrong side of \(\langle c,d\rangle\).

Finally, suppose for the sake of contradiction that \(p\) lies on the wrong side of \(L_2\). Without loss of generality, let \(p\) specifically be the point furthest from \(L_2\). Then let \(p’\) be the point on the line containing \(b\) and \(d\) that is closest to \(p\). If \(X’ = X_a^p \bigcup \langle p, p’\rangle\), then \(X’\) is shorter than \(X\). Because \(\langle p,p’\rangle\) is perpendicular to \(\langle p’,d\rangle\), \(p’\) must also be further from \(L_2\) than any other point in \(X\). Therefore, \(X’\) accommodates \(X\), giving a contradiction.

This lemma comes with a few charming corollaries:

Finally, we can make a little theorem effectively characterizing taut Type 2 curves.

Defining Several Important Shapes

#

Now that the framework for which curves are worth considering has been well established, let’s build up a vocabulary of specific curves to consider. In some sense, the simplest unit-length curve is the single line segment. I don’t want to use the letter L or I for this purpose, so I’ll call this \(G\), in reference to the French spelling of the word line, or ligne.

There’s a bit of an opposite to \(G\). While \(G\) has a maximum distance from end to end (or diameter), it has no width whatsoever. A correlating question would be to search for the maximum radius of a circle accomodated by a single curve’s convex hull.

proof:
Consider the circle \(c\) of radius \(r\) defined with \((y-r)^2+x^2=(r)^2\). Suppose \(m\) is a curve with endpoints \(a\) and \(b\) that accomodates \(c\) with minimal length. In order to minimize the length of \(m\), \(\langle a,b \rangle\) must be tangent to \(c\). Without loss of generality, let \(a\) and \(b\) lie on the \(x\) axis. The locations of \(a\) and \(b\) that minimize the length of \(m\) are directly beneath the ends of \(c\), at \((\pm r,0)\)

To minimize the size of \(m\), assume that \(m\) is “taut” against the circle \(c\), as shown in the drawing below:

The total length of \(m\) is therefore \(2r+\pi r\). If \(m\) is a unit length curve, then $$\begin{align*} 1 =2r+\pi r &=r(2+\pi)\\ \frac{1}{2+\pi} &= r \end{align*}$$

Now that that’s done, we can use this as the basis of the definition of our second useful curve!

That’s the largest possible circle that can be accomodated with a unit length curve, but what about the largest possible square? We might as well define it as simply as possible:

Of course, \(S\) was chosen as a name because that is the first letter of the word “square”. This shape is shown below:

What other simple curves are important enough to be given brief names? There’s a fair argument to be made that the simplest curves would be those of constant curvature, also known as arcs.

It’s worth showing that the shape given in this definition does in fact have a length of 1. The two places where \(y=0\) are given with:

$$\begin{align*} 0 &= \sqrt{r^{2}-x^{2}}-r\sin\left(\frac{\pi r-1}{2r}\right)\\ r\sin\left(\frac{\pi r-1}{2r}\right) &= \sqrt{r^{2}-x^{2}}\\ r^2\sin^2\left(\frac{\pi r-1}{2r}\right) &= r^2-x^2\\ r^2 - r^2\sin^2\left(\frac{\pi r-1}{2r}\right) &= x^2\\ r^2\left(1-\sin^2\left(\frac{\pi r-1}{2r}\right)\right) &= x^2\\ r^2\cos^2\left(\frac{\pi r-1}{2r}\right) &= x^2\\ \pm r\cos\left(\frac{\pi r-1}{2r}\right) &= x \end{align*}$$

Note also that \(y’=\frac{x}{\sqrt{r^{2}-x^{2}}}\). Using the standard formula for length of a curve gives:

$$\begin{align*} \int_{-r\cos\left(\frac{\pi r-1}{2r}\right)}^{r\cos\left(\frac{\pi r-1}{2r}\right)}\sqrt{1+y’^2}dx &= \int_{-r\cos\left(\frac{\pi r-1}{2r}\right)}^{r\cos\left(\frac{\pi r-1}{2r}\right)}\sqrt{1+\frac{x^2}{r^2-x^2}}dx\\ &= \int_{-r\cos\left(\frac{\pi r-1}{2r}\right)}^{r\cos\left(\frac{\pi r-1}{2r}\right)}\sqrt{\frac{r^2}{r^2-x^2}}dx\\ &= \int_{-r\cos\left(\frac{\pi r-1}{2r}\right)}^{r\cos\left(\frac{\pi r-1}{2r}\right)}\frac{1}{\sqrt{1-\frac{x^2}{r^2}}}dx\\ &= \left.-r\cos^{-1}\left(\frac{x}{r}\right)\right|_{-r\cos\left(\frac{\pi r-1}{2r}\right)}^{r\cos\left(\frac{\pi r-1}{2r}\right)}\\ &= -r\cos^{-1}\left(\frac{r\cos\left(\frac{\pi r-1}{2r}\right)}{r}\right) + r\cos^{-1}\left(\frac{-r\cos\left(\frac{\pi r-1}{2r}\right)}{r}\right)\\ &= -r\cos^{-1}\left(\cos\left(\frac{\pi r-1}{2r}\right)\right) + r\cos^{-1}\left(-\cos\left(\frac{\pi r-1}{2r}\right)\right)\\ &= -r\left(\frac{\pi r-1}{2r}\right) + r\left(-\cos^{-1}\left(\cos\left(\frac{\pi r-1}{2r}\right)\right)+\pi\right)\\ &= -r\left(\frac{\pi r-1}{2r}\right) + r\left(-\left(\frac{\pi r-1}{2r}\right)+\pi\right)\\ &= -\frac{\pi r-1}{2} -\frac{\pi r-1}{2}+\pi r\\ &= -\pi r+1 +\pi r = 1 \end{align*}$$

One could also argue that the simplest shape to consider is two line segments meeting at a single vertex, also known as a 2-chain (no relation), with both line segments of equal length. The convex hulls of these shapes are simple triangles. From this perspective, it would only make sense to define

As you might have guessed, the name \(V\) was chosen due to its resemblance to an angle. The name \(E\) was chosen because of the similarity to an equilateral triangle.

All of these have been type 1 curves! In a sense, the simplest type 2 curve is one that can be formed with just four points.

Of course, this is also shown below:

A hopeful reader might wonder if this curve might in fact be accomodated by a type 1 curve. \(Z\) certainly doesn’t look like it could be accomodated by a type 1 curve, by any of the lemmas encountered earlier.

To see if it’s possible to accomodate this with a type 1 curve, let’s start by placing the \(Z\) curve on it’s side, so that the starting and ending points are on the \(y\)-axis. I will refer to this instance of \(Z\) as \(z\), and label its points with:

$$z_1 = \left(0,0\right)$$ $$z_2 = \left(\frac{2\sqrt{5}}{15},-\frac{\sqrt{5}}{15}\right)$$ $$z_3 = \left(\frac{3\sqrt{5}}{15},\frac{\sqrt{5}}{15}\right)$$ $$z_4 = \left(\frac{\sqrt{5}}{3},0\right)$$

I’m loath to stretch the length of this paper out any further with information the reader could easily find for themselves. That said, it is important to show that these points really do form an instance of \(Z\). First, note that the distances between each point really does come out to ⅓.

$$\ell\langle z_1,z_2\rangle = \sqrt{\left(\frac{2\sqrt{5}}{15} -0\right)^2 + \left(-\frac{\sqrt{5}}{15} -0\right)^2} = \sqrt{\frac{20}{15^2} + \frac{5}{15^2}} = \frac{5}{15}=\frac{1}{3}$$

$$\ell\langle z_2,z_3\rangle = \sqrt{\left(\frac{3\sqrt{5}}{15} -\frac{2\sqrt{5}}{15}\right)^2 + \left(\frac{\sqrt{5}}{15} +\frac{\sqrt{5}}{15} \right)^2} = \sqrt{\frac{5}{15^2} + \frac{20}{15^2}} = \frac{5}{15}=\frac{1}{3}$$

$$\ell\langle z_3,z_4\rangle = \sqrt{\left(\frac{5\sqrt{5}}{15} -\frac{3\sqrt{5}}{15}\right)^2 + \left(\frac{0-\sqrt{5}}{15}\right)^2} = \sqrt{\frac{20}{15^2} + \frac{5}{15^2}} = \frac{5}{15}=\frac{1}{3}$$

Additionally, the slope between \(z_1\) and \(z_2\) is \(\frac{-\frac{\sqrt{5}}{15}-0}{\frac{2\sqrt{5}}{15}-0} = \frac{-1}{2}\). The slope between \(z_2\) and \(z_3\) is \(\frac{\frac{\sqrt{5}}{15}+\frac{\sqrt{5}}{15}}{\frac{3\sqrt{5}}{15}-\frac{2\sqrt{5}}{15}} = 2\). Finally, the slope between \(z_4\) and \(z_3\) is \(\frac{0-\frac{\sqrt{5}}{15}}{\frac{\sqrt{5}}{3}-\frac{3\sqrt{5}}{15}} = \frac{-1}{2}\). This shows that each edge is orthogonal to the previous one, confirming that this is an instance of \(Z\).

I want to define a type 1 curve that gets very close to accomodating this. For a start, I’ll have this curve include \(z_1\), \(z_2\), and \(z_3\). This curve is shown in purple:

The length of \(\langle z_1, z_3, z_4\rangle\) is given by:

$$ \ell\langle z_1, z_3\rangle + \ell\langle z_3, z_4\rangle = \sqrt{\left(\frac{\sqrt{5}}{15}-0\right)^2+\left(\frac{3\sqrt{5}}{15}-0\right)^2} + \frac{1}{3} = \sqrt{\frac{5}{15^2}+ \frac{45}{15^2}} + \frac{5}{15} = \frac{5\sqrt{2}+5}{15}=\frac{\sqrt{2}+1}{3}$$

In order to get as close as possible to accomodating \(z_2\), we may place a single point at a distance of \(\frac{2-\sqrt{2}}{3}\) from \(z_1\). The optimal point \(p\) would be the one where \(\langle z_1,p \rangle\) and \(\langle p, z_4 \rangle\) are orthogonal. (trust me)

If \(p = \left(p_x, p_y\right)\), then these two requirements may be expressed as:

$$\sqrt{\left(p_x-0\right)^2+\left(p_y-0\right)^2} = \frac{2-\sqrt{2}}{3}$$

$$\frac{p_y-0}{p_x-0}=-\frac{\frac{\sqrt{5}}{3}-p_x}{0-p_y}$$

If we take the first of these two equations and square both sides, this becomes:

$$p_x^2+p_y^2 = \frac{6-4\sqrt{2}}{9}$$

Meanwhile, we multiply both sides of the other equation by \(p_xp_y\) to get

$$p_y^2=\frac{\sqrt{5}}{3}p_x-p_x^2$$

Now we can plug this into the other equation for \(p_y^2\) and solve to get

$$\frac{\sqrt{5}}{3}p_x = \frac{6-4\sqrt{2}}{9} $$ $$p_x = \frac{6-4\sqrt{2}}{3\sqrt{5}} $$

Square this to get

$$p_x^2 = \frac{68-48\sqrt{2}}{45} $$

Plug this and \(\frac{\sqrt{5}}{3}p_x\) into our equation for \(p_y^2\) to get

$$\begin{align*} p_y^2 &= \frac{6-4\sqrt{2}}{9}-\frac{68-48\sqrt{2}}{45}\\ p_y^2 &= \frac{30-20\sqrt{2}}{45}-\frac{68-48\sqrt{2}}{45}\\ p_y^2 &= \frac{-38+28\sqrt{2}}{45}\\ p_y &= \pm\frac{\sqrt{-38+28\sqrt{2}}}{3\sqrt{5}} \end{align*}$$

Of course, this is pretty ugly, but it is accurate. We’ll use the negative version of \(p_y\) to get \(p=\left(\frac{6-4\sqrt{2}}{3\sqrt{5}},\frac{\sqrt{-38+28\sqrt{2}}}{3\sqrt{5}}\right)\), as shown:

Convex Hulls of Multiple Curves

#

The focus of this section is to expand our understanding of convex hulls of sets of taught curves. While discussing convex hulls of curves, it will be useful to clarify something that went mostly unstated in the previous section:

My plan is to spend this section establishing some useful lemmas about the convex hulls of multiple curves, as a bit of set dressing for the sections to come. The nature of this problem requires that we seek to minimize the size of a convex hull, so I’ll give a definition relating to this:

Here’s the first lemma for this section:

proof:
Let \(a’\) be the instance of \(A\) that is internal to \(X\) but does not touch its boundary, let \(\omega’=\lbrace a’, b, c, \dots\rbrace\), and let \(X’\) be the convex hull of the union of \(\omega’\).

It can be immediately seen by definition that \(\omega’\in\Omega\), and therefore that \(X’\in [\Omega]\). Because \(X\) is a minimal element of \([\Omega]\), \(X\lesssim X’\). \(X’\) contains no points outside the boundary of \(X\), and so \(X\nless X’\).

Therefore, \(X’\) has the same area as \(X\). This implies that \(X=X’\).

In order to produce the next lemma. we give a special definition:

This definition comes with an almost immediate lemma:

proof:

Let \(a,b\in\omega\). First, we know that \([\omega]\sqsupseteq[a,b]\). We also know that if \(a\) and \(b\) are not fully intersecting, they at least have a fully intersecting pair of linear transformations, \(a’\) and \(b’\), such that \([a,b]\sqsupseteq[a’,b’]\). By the transitive property, \([\omega]\sqsupseteq[a’,b’]\)

We can extend this lemma just a little bit further, finally reaching a theorem about the nature of minimal convex hulls.

proof:

Let \(\omega\) be such that \([\omega]\in[\Omega]^-\). In order to force a contradiction, assume that there exists \(a,b\in\omega\) such that \(a\) and \(b\) are not fully intersecting. Because of Lemma 6, we may assume that both \(a\) and \(b\) touch the boundary of \([\omega]\). In fact, they both extend the boundary of \([\omega]\), meaning that removing them would result in a smaller overall convex hull.

Let’s assume (WLOG) that \(a’\) is a linear transformation of \(a\), such that \(a’\sqsubset[a,b]\). Let \(\omega’\) be formed by replacing \(a\) in \(\omega\) with \(a’\). Note that, by the definition of fully intersecting, \([a,b]\sqsupseteq[a’b]\), and therefore, \([\omega]\sqsupseteq[\omega’]\). This means that if there is any point on the boundary of \([\omega]\) that is not present in \([\omega’]\), then \([\omega]\sqsupset[\omega’]\). But this cannot be the case if \([\omega]\in[\Omega]^-\), so it must be that \([\omega]=[\omega’]\).

We now know that there are some points present in \([a,b]\) that are absent in \([a’,b]\), but that none of these points are on the boundary of \([\omega]\). As mentioned, \(a\) lies on the boundary of \([\omega]\). If \(a’\) is also on the boundary of \([\omega]\), that means that \(a\) must not have actually expanded the boundary of \([\omega]\). This also violates the spirit of Lemma 6, giving a contradiction.

For the final lemma of this this section, it’ll be helpful to note one final definition. To be clear, this isn’t a definition of my own, but one that is commonly used.

I’ll introduce first the specific case of this lemma, then extend it out to a more general case.

Well, as is my predilection, I’ll try to prove this by contradiction, and I’ll try to prove a simpler version first.

Minimal Covers of Finite Sets

#

In a previous section, I limited the space of curves worth considering by defining taughtness. I showed that some curves are accomodated by other curves, and therefore not worth considering. In the next section, I’ll try to limit the space of curves even further by finding some curves that are accomodated by sets of other curves.

My initial goal was to hopefully find a finite set of taut curves that proxy accommodate all other taut curves, and then to find the smallest set that contains all curves in that finite set. This goal seems reasonable at first, as it is much easier to find \([\Xi]^-\) when \(\Xi\) is a finite set of curves as opposed to an infinite one. There are some severe problems with this as an objective, but there’s still much to be learned by exploring it!

In this section, to illustrate that it is in fact possible to find the minimal size of covering shapes for certain finite numbers of curves, we’ll find a few examples. These next two theorems are 5% insight and 95% bookkeeping, so feel free to skip their proofs if you’re in a hurry.

proof:
First, assume without loss of generality that the instance of \(S\) is located on the points \((0,0)\), \(\left(\frac{1}{3},0\right)\), \(\left(\frac{1}{3},\frac{1}{3}\right)\), and \(\left(0,\frac{1}{3}\right)\), as in the definition’s example. Call this instance \(s\). Let the intance of \(G\) have endpoints \(A\) and \(B\), so that \(\ell\langle A,B\rangle = 1\). Define this instance of the line so that \(\langle A,B\rangle = g\). Note that the line between \(A\) and \(B\) must pass through the square in order for these shapes to be fully intersecting.

Of course, \([s,g]\) is the convex hull of these points. A potential instance of \([s,g]\) is shown in the illustration below:

\([s,g]\) can be deconstructed into the central square and multiple triangles:

Let \(A=(a_x,a_y)\), and let \(B=(b_x,b_y)\). Additionally, let \(f(x)\) be the distance of \(x\) to the interval \(\left[0,\frac{1}{3}\right]\), so that

$$f(x) = \begin{cases} -x &\text{if } x\leq 0 \\ 0 &\text{if } x\in \left(0,\frac{1}{3}\right)\\ x-\frac{1}{3} &\text{if } x \geq \frac{1}{3} \end{cases}$$

To give an example of how this could be useful, consider the triangle formed by \(a_x\). If \(A\) is the lower point on the figure above, this would be the green triangle. Using the standard formula for area of a triangle, we get \(\frac{1}{2}\cdot\frac{1}{3}\cdot f(a_x)\)

Then, the full area of \([s,g]\) would be given by:

$$\frac{1}{9}+\sum_{i} \frac{f(i)}{6}$$

Where \(i\) is an element of \(\left\lbrace a_x,a_y,b_x,b_y\right\rbrace \). The presence of \(\frac{1}{9}\) is due to the area of \(s\). There is also the limitation from the length of \(g\) which implies that:

$$\begin{align*} 1 &= \ell \langle A,B\rangle\\ 1 &= \sqrt{\left(a_x-b_x\right)^2+\left(a_y-b_y\right)^2}\\ 1 &= \left(a_x-b_x\right)^2+\left(a_y-b_y\right)^2 \end{align*}$$

For the time being, let’s hold \(A\) constant, and consider the optimal placement of \(B\), assuming without loss of generality that \(b_x \geq 0\) and that \(b_y\geq \frac{1}{3}\). Solving the previous equation for \(b_y\) gives

$$\begin{align*} 1 &= \left(a_y-b_y\right)^2+\left(a_x-b_x\right)^2\\ 1- \left(a_y-b_y\right)^2 &= \left(a_x-b_x\right)^2\\ \sqrt{1- \left(a_y-b_y\right)^2} &= a_x-b_x\\ a_x - \sqrt{1- \left(a_y-b_y\right)^2} &= b_x \end{align*}$$

Therefore, the maximum value of \(b_x\) is:

$$b_x \leq a_x - \sqrt{1- \left(a_y-\frac{1}{3}\right)^2}$$

By symmetry, we also know that \(b_y= a_y - \sqrt{1- \left(a_x-b_x\right)^2}\). Therefore, the total area due to \(B\) can be expressed as:

$$A_B = f(b_x) + f(b_y) = f(b_x) + b_y = f(b_x) + a_y - \sqrt{1- \left(a_x-b_x\right)^2} $$

This function is non-differentiable or has boundaries when \(b_x=0\), when \(b_x=\frac{1}{3}\), and when \(b_x= a_x - \sqrt{1- \left(a_y-\frac{1}{3}\right)^2}\). If \(x\in\left[0,\frac{1}{3}\right]\), then

$$A_B = a_y - \sqrt{1- \left(a_x-b_x\right)^2}$$

The derivative of this is

$$\frac{dA_B}{db_x} = \frac{2a_x-2b_x}{2\sqrt{1- \left(a_x-b_x\right)^2}} = \frac{a_x-b_x}{\sqrt{1- \left(a_x-b_x\right)^2}}$$

This is equal to 0 only when \(b_x = a_x\). On the other hand, if \(x > \frac{1}{3}\), then

$$A_B = b_x-\frac{1}{3} + a_y - \sqrt{1- \left(a_x-b_x\right)^2} $$

The derivative of this is quite similar to the derivative from before:

$$\frac{dA_B}{db_x} = 1 + \frac{a_x-b_x}{\sqrt{1- \left(a_x-b_x\right)^2}}$$

For this derivative to be equal to 0, this comes to:

$$\begin{align*} 0 &= 1 + \frac{a_x-b_x}{\sqrt{1- \left(a_x-b_x\right)^2}}\\ 1 &= \frac{b_x-a_x}{\sqrt{1- \left(a_x-b_x\right)^2}}\\ \sqrt{1- \left(a_x-b_x\right)^2} &= b_x-a_x\\ 1- \left(a_x-b_x\right)^2 &= \left(b_x-a_x\right)^2\\ 1 &= 0 \end{align*}$$

This is obviously not possible, meaning that the critical points of this area function are when \(b_x=0\), when \(b_x=\frac{1}{3}\), when \(b_y=\frac{1}{3}\), or when \(0 < b_x=a_x < 0\).

If \(b_x,a_x \in \left[0,\frac{1}{3}\right]\), then the formula for area is given by \(a_y - \sqrt{1- \left(a_x-b_x\right)^2}\). It is clear from inspection that this function is larger when \(a_x=b_x\) as opposed to otherwise. Because of this, we conclude that \(B\) lies on either \(x=0\), \(x=\frac{1}{3}\), \(y=0\), or \(y=\frac{1}{3}\). By symmetry, \(A\) also lies on one of these lines. This leads to one of two possibilities:

In one possibility, assume that both \(A\) and \(B\) lie on \(x=0\) and \(y=0\). Without loss of generality, let \(b_x = 0\) while \(a_x = \frac{1}{3}\). Then the distance between \(A) and \(B\) gives the restriction that:

$$\begin{align*} 1 &= \sqrt{\left(a_x-b_x\right)^2+\left(a_y-b_y\right)^2}\\ 1 &= \sqrt{\left(\frac{1}{3}\right)^2+\left(a_y-b_y\right)^2}\\ 1 &= \frac{1}{9}+\left(a_y-b_y\right)^2\\ \frac{8}{9} &= \left(a_y-b_y\right)^2\\ \frac{2\sqrt{2}}{3} &= |a_y-b_y| \end{align*}$$

Assuming that the square lies between these two points, the area of the full shape is then given by:

$$A= \frac{1}{9}+\frac{1}{6}\cdot\frac{2\sqrt{2}}{3}=\cdot\frac{2+2\sqrt{2}}{18}\approx 0.268$$

In the other possibility, \(A\) lies on a vertical line, while \(B\) lies on a horizontal line, or vice versa. Without loss of generality, let \(a_x=0\), and let \(b_y = 0\).Then the distance between \(A) and \(B\) gives the restriction that:

$$\begin{align*} 1 &= \sqrt{\left(a_x-b_x\right)^2+\left(a_y-b_y\right)^2}\\ 1 &= \sqrt{b_x^2+a_y^2}\\ 1 &= b_x^2+a_y^2\\ 1 - a_y^2 &= b_x^2\\ \sqrt{1 - a_y^2} &= b_x \end{align*}$$

And the total area of the shape is given by:

$$A= \frac{1}{9}+\frac{1}{6}b_x + \frac{1}{6}a_y=\frac{1}{9}+\frac{\sqrt{1 - a_y^2}}{6} + \frac{a_y}{6}$$

Once again, we find the minimum of this using first-term calculus. The derivative of this area is:

$$\frac{dA}{da_y}= \frac{ -a_y}{6\sqrt{1 - a_y^2}}+\frac{1}{6}$$

This derivative evaluates to 0 when

$$\begin{align*} 0 &= \frac{ -a_y}{6\sqrt{1 - a_y^2}}+\frac{1}{6}\\ \frac{a_y}{6\sqrt{1 - a_y^2}} &= \frac{1}{6}\\ a_y &= \sqrt{1 - a_y^2}\\ a_y^2 &= 1 - a_y^2\\ 2a_y^2 &= 1\\ a_y &= \frac{1}{\sqrt{2}} \end{align*}$$

If \(a_y=\frac{1}{\sqrt{2}}\), then \(b_x=\frac{1}{\sqrt{2}}\) as well. This solution is invalid, however, as the instance of \(L\) in this case would not pass through the instance of \(S\). Rather than showing this mathematically, I’ll just show the results of a graphing utility.

The non-differentiable points in the second possibility are identical to the non-differentiable points in the first possibility, proving the theorem.

Admittedly, the statement of that theorem might have been highly intuitive to you. You might find it much less intuitive to consider elements of \(\left[G,V_{\frac{\pi}{6}}\right]^-\).

proof:

To start, let’s place an instance of \(E\) in the cartesian plane, and we’ll call this instance \(e\) (Euler’s number does not appear anywhere in this article). I’ll place this triangle with two points on the \(x\)-axis, and one point on the \(y\)-axis. Then the points in this triangle are:

$$e_1=\left(\frac{-1}{4},0\right)$$ $$e_2=\left(\frac{1}{4}\right)$$ $$e_3=\left(0,\frac{\sqrt{3}}{4}\right)$$

Note that it isn’t really important which two sides of this triangle make up \(e\), as we’ll be considering the convex hull, so all three sides are included in this graph. I will illustrate this triangle below:

Note that I have labeled the sections of this triangle with numbers 1 though 6. This is because the instance of \(G\) must pass through \(e\) by Theorem 4. Note that \(G\) cannot have endpoints in adjacent sections if it passes through \(e\). Note also that \(G\) cannot have both its endpoints in odd-numbered sections. Therefore, define the endpoints of \(G\) to be \(a=\left(a_x,a_y\right)\) and \(b=\left(b_x,b_y\right)\). Without loss of generality, assume \(a\) is in section 4, and assume \(b\) is in section 1 or 6.

What is the size of \(\left[e,\langle a,b\rangle\right]\)? First, Let’s find the area of \(\left[e,a\right]\). This consists of the area inside \(e\), as well as the area in the triangle defined by the points \(e_1\), \(e_2\), and \(a\).

The area of the grey region inside \(e\) is given by \(\frac{1}{2}\cdot\frac{1}{2}\cdot\frac{\sqrt{3}}{4}\). The area of the blue region created by \(a\) is given by \(\frac{1}{2}\cdot\frac{1}{2}\cdot a_y\). Therefore, the total area of these two regions is:

$$\frac{\sqrt{3}}{16}+\frac{a_y}{4}$$

Let’s start by assuming that \(b\) is located in region 6. In this case, the inclusion of \(b\) creates an additional triangle defined by the points \(e_3\), \(e_2\), and \(b\), as shown below:

The area of this green region is determined by the distance from \(b\) to the line from \(e_3\) to \(e_2\). There’s a useful formula that gives the area of a triangle given the locations of its three points:

$$\begin{align*} Area &=\left|{\frac{b_x\left(\frac{\sqrt{3}}{4}-0\right)+0\left(0-b_y\right)+\frac{1}{4}\left(b_y-\frac{\sqrt{3}}{4}\right)}{2}}\right|\\ &= \left|{\frac{\frac{b_x\sqrt{3}}{4}+\frac{1}{4}\left(b_y-\frac{\sqrt{3}}{4}\right)}{2}}\right|\\ &= \left|{\frac{b_x\sqrt{3}+b_y-\frac{\sqrt{3}}{4}}{8}}\right|\\ &= \frac{b_x\sqrt{3}}{8}+\frac{b_y}{8}-\frac{\sqrt{3}}{32} \end{align*}$$

Recall that \(b\) and \(a\) are exactly one unit apart. This gives the restriction that:

$$\begin{align*} 1 &= \left(b_y-a_y\right)^2 +\left(b_x-a_x\right)^2\\ 1-\left(b_x-a_x\right)^2 &= \left(b_y-a_y\right)^2 \\ \sqrt{1-\left(b_x-a_x\right)^2} &= b_y-a_y \\ \sqrt{1-\left(b_x-a_x\right)^2}+a_y &= b_y\\ \end{align*}$$

Substituting this into the equation of the area of the green region would give

$$Area = \frac{b_x\sqrt{3}}{8}+\frac{\sqrt{1-\left(b_x-a_x\right)^2}+a_y}{8}-\frac{\sqrt{3}}{32} $$

To find the minimum possible value of this area, we take the derivative with respect to \(b_x\) and set that derivative equal to 0.

$$\begin{align*} 0&= \frac{\sqrt{3}}{8}+\frac{-2\left(b_x-a_x\right)}{8\cdot 2\sqrt{1-\left(b_x-a_x\right)^2}}\\ 0&= \frac{\sqrt{3}}{8}-\frac{b_x-a_x}{8\sqrt{1-\left(b_x-a_x\right)^2}}\\ 0&= \sqrt{3}-\frac{b_x-a_x}{\sqrt{1-\left(b_x-a_x\right)^2}}\\ \sqrt{3} &= \frac{b_x-a_x}{\sqrt{1-\left(b_x-a_x\right)^2}}\\ \sqrt{3-3\left(b_x-a_x\right)^2} &= b_x-a_x\\ 3-3\left(b_x-a_x\right)^2 &= \left(b_x-a_x\right)^2\\ 3 &= 4\left(b_x-a_x\right)^2\\ \pm\frac{\sqrt{3}}{2} &= b_x-a_x\\ \end{align*}$$

Again, the distance between \(b\) and \(a\) must be exactly one unit. This means that:

$$\begin{align*} 1 &= \left(b_y-a_y\right)^2 +\left(\pm\frac{\sqrt{3}}{2}\right)^2\\ 1 &= \left(b_y-a_y\right)^2 + \frac{3}{4}\\ \frac{1}{4} &= \left(b_y-a_y\right)^2\\ \frac{1}{2} &= b_y-a_y\\ \end{align*}$$

Because \(b_x>0\), and \(b_y\) is so low above \(a_y\), we may assume that \(b= \left(a_x+\frac{\sqrt{3}}{2},a_y+\frac{1}{2}\right) \). This is signifigant, because we may observe that the slope from \(a\) to \(b\) is \(\frac{1}{\sqrt{3}}\). This is perpendicular to the slope from \(Z_3\) to \(Z_2\), meaning that this location of \(b\) maximizes the area of the green triangle.

The other posibility that would create a critical point for the area of the green triangle is that \(b\) lies on either the \(x\)-axis, or on the line \(y=\sqrt{3}x+\frac{\sqrt{3}}{4}\), as these are the bounds of \(b\)’s location. It is not possible for \(b\) to lie on the \(x\)-axis, as this would result in the line \(G\) lying outside \(e\). Therefore, the minimum area is achieved when \(b\) lies on \(y=\sqrt{3}x+\frac{\sqrt{3}}{4}\).

The other posibility is that \(b\) lies in section 1. If this is the case, we may ignore the area of \(e\), as the overall area can now be defined by a triangle between \(b\), \(e_1\), and \(e_2\).

In this case, the total area of \([e,\langle a,b\rangle]\) is \(\frac{1}{4}a_y+\frac{1}{4}b_y\). Holding the position of \(a\) constant, \(b_y\) is minimized by maximizing \(\left|\left(b_x-a_x\right)\right|\). This means that \(b\) must lie on the boundary of Section 1.

We have concluded that \(b\) must lie on the boundary of section 1, regardless of what other assumptions have been made. Let’s assume, without loss of generality, that \(b\) lies on the line \(y=\sqrt{3}x+\frac{\sqrt{3}}{4}\), while \(a\) is in section 4. From here, we’ll hold the location of \(b\) constant and find the optimal location of \(a\). As before, to minimize the area of the blue triangle formed by \(a\), we must minimize \(a_y\), which means maximizing \(\left|\left(b_x-a_x\right)\right|\). We conclude that \(a\) must lie on the boundary of section 1. If \(a\) lies on the line passing through \(e_3\) and \(e_2\), then \(\langle a,b\rangle\) does not intersect \(e\).

We have now shown that \([e, \langle a,b\rangle]\) is congruent to a triangle with base length 1 and height \(\frac{\sqrt{3}}{4}\). The area of such a triangle is \(\frac{\sqrt{3}}{4}\).

Proxy Accommodated Curves

#

As mentioned, one of the big hopes in this project is that we should be able to ignore certain curves because they are accomodated by the unions of other curves. For example, let’s look at what else might be covered by one of the elements of \([E,G]^-\).

proof

Let \(t\) be the triangle described in the theorem statement.

For any angle \(\theta\), define \(v_\theta\) to contain the points \(v_1=\left(0,\frac{1}{2}\cos{\frac{\theta}{2}}\right)\), \(v_2=\left(\frac{1}{2}\sin{\frac{\theta}{2}},0\right)\), and \(v_3=\left(-\frac{1}{2}\sin{\frac{\theta}{2}},0\right)\). Note first that the distance between \(v_1\) and \(v_2\) is:

$$\sqrt{\left(\frac{1}{2}\cos{\frac{\theta}{2}}\right)^2+\left(-\frac{1}{2}\sin{\frac{\theta}{2}}\right)^2}=\frac{1}{2}\sqrt{\cos^2{\frac{\theta}{2}}+\sin^2{\frac{\theta}{2}}}=1$$

By symmetry, the distance from The angle formed by this shape is \(\frac{1}{2}\). The angle formed at \(v_1\) is \(\theta\), confirming that this is an instance of \(V_\theta\). Note that \(\frac{1}{2}\sin{\frac{\theta}{2}} < \frac{1}{2}\) for all values of \(\theta\), meaning that \(v_2\) lies inside \(t\). By symmetry, \(v_2\) also lies inside \(t\).

\(v_1\) lies inside \(t\) whenever \(\frac{1}{2}\cos{\frac{\theta}{2}}<\frac{\sqrt{3}}{4}\). This is true when \(\theta<\frac{pi}{6}\). Since \(V_\theta\) is not taut for \(\theta < \frac{\pi}{6}\), we won’t consider these anyway.

For any radius \(r\), let \(a_r\) be the curve defined with \(y=\sqrt{r^{2}-x^{2}}-r\sin\left(\frac{\pi r-1}{2r}\right)\), where \(y>0\). The line that defines the right half of \(t\) is \(y=-\frac{\sqrt{3}}{2}x+\frac{\sqrt{3}}{4}\). These two lines intersect when

$$\begin{align*} \sqrt{r^{2}-x^{2}}-r\sin\left(\frac{\pi r-1}{2r}\right) &= -\frac{\sqrt{3}}{2}x+\frac{\sqrt{3}}{4}\\ \sqrt{r^{2}-x^{2}}-r\sin\left(\frac{\pi r-1}{2r}\right) &= \frac{\sqrt{3}}{4}(1-2x)\\ \sqrt{r^{2}-x^{2}} &= \frac{\sqrt{3}}{4}(1-2x)+r\sin\left(\frac{\pi r-1}{2r}\right)\\ r^{2}-x^{2} &= \frac{3}{16}(1-2x)^2+\frac{\sqrt{3}}{2}(1-2x)r\sin\left(\frac{\pi r-1}{2r}\right)+r^2\sin^2\left(\frac{\pi r-1}{2r}\right)\\ r^{2}-x^{2} &= \frac{3}{16}-\frac{3}{8}x+\frac{3}{4}x^2+\frac{\sqrt{3}r}{2}(1-2x)\sin\left(\frac{\pi r-1}{2r}\right)+r^2\sin^2\left(\frac{\pi r-1}{2r}\right)\\ r^2 &= \frac{3}{16}-\frac{3}{8}x+\frac{7}{4}x^2+\frac{\sqrt{3}r}{2}(1-2x)\sin\left(\frac{\pi r-1}{2r}\right)+r^2\frac{1-\cos\left(\frac{\pi r-1}{r}\right)}{2}\\ 0 &= \frac{3}{8}-\frac{3}{4}x+\frac{7}{2}x^2+\sqrt{3}r(1-2x)\sin\left(\frac{\pi r-1}{2r}\right)-r^2-r^2\cos\left(\frac{\pi r-1}{r}\right)\\ \end{align*}$$

Eliminating Type 2 Curves

#

So far, the previous sections have each had to do with results I have already proven. At the time that I write this, the subject of this section is not yet proven. Therefore, what follows is a hypothesis, rather than a lemma, theorem, or statement.

This would be a lovely result to find, although it’s easier stated than proven. Let’s start with a simpler hypothesis. Maybe we can use that as a stepping stone.

And in particular, I would sharpen this hypothesis even further:

To start exploring this, I’ll place an instance of \(Z’\) within the cartesian plane at the same location as in its definition, called \(z’\). As a reminder, the points used here are labeled as follows:

$$P = \left(p_x,p_y\right)=\left(\frac{6-4\sqrt{2}}{3\sqrt{5}},\frac{\sqrt{-38+28\sqrt{2}}}{3\sqrt{5}}\right)$$ $$Z_1=\left(0,0\right)$$ $$Z_3=\left(\frac{3\sqrt{5}}{15},\frac{\sqrt{5}}{15}\right)$$ $$Z_4=\left(\frac{\sqrt{5}}{3},0\right)$$

For convenience, we also recall that

$$Z_3=\left(\frac{2\sqrt{5}}{15},\frac{-\sqrt{5}}{15}\right)$$

It is possible, for the purposes of exploring Question 2.2, to expend the effective area of \(z’\)

If this shape is part of a union that includes certain other points, it would be possible for \(Z\) to be included, answering Question 2.2. I’m going to indicate in blue the points which would result in accomodating \(Z\). First, it should be clear that including \(\left(\frac{2\sqrt{5}}{15},\frac{-\sqrt{5}}{15}\right)\)would allow for \(Z\). In addition, there’s an infinite stretch of territory that would include that point by necessity.

It is important to give the precise equations for these lines, as they’ll be built upon moving forward. One boundary of the blue region is the line from \(P\) to \(Z_2\). This is defined with:

$$y_1\leq\frac{15p_y+\sqrt{5}}{15p_x-2\sqrt{5}}\left(x-p_x\right)+p_y$$

The other boundary of the blue region is the line from \(Z_2\) to \(Z_4\)

$$y_2\leq\frac{1}{3}\left(x-\frac{2\sqrt{5}}{15}\right)-\frac{\sqrt{5}}{15}$$

Recall that \(M\) includes a circle of radius \(\frac{1}{\pi+2}\). Because of this, the center of that circle cannot be located anywhere less than \(\frac{1}{\pi+2}\) units away from this blue region. So, what is the line consisting of points \(\frac{1}{\pi+2}\) away from \(y_1\)? The vector with a slope that is the opposite recipricoal of the slope of \(y_1\) is:

$$v_1=\left[15p_y+\sqrt{5},-15p_x+2\sqrt{5}\right]$$

The magnitude of this vector is:

$$\begin{align*} |v_1| &= \sqrt{\left(15p_y+\sqrt{5}\right)^2+\left(-15p_x+2\sqrt{5}\right)^2}\\ &= \sqrt{225p_y^2+30\sqrt{5}p_y+5+225p_x^2-60\sqrt{5}p_x+5}\\ &= \sqrt{225p_y^2+30\sqrt{5}p_y+225p_x^2-60\sqrt{5}p_x+10} \end{align*}$$

The line that lies \(\frac{1}{\pi+2}\) units away from \(y_1\) is therefore:

$$y\leq\frac{15p_y+\sqrt{5}}{15p_x-2\sqrt{5}}\left(x-p_x-\frac{15p_y+\sqrt{5}}{(\pi+2)|v_1|}\right)+p_y-\frac{15p_x-2\sqrt{5}}{(\pi+2)|v_1|}$$

Similarly, if we define

$$v_2=\left[-1,3\right]$$

Then the line that lies \(\frac{1}{\pi+2}\) units away from \(y_1\) is:

$$y\leq\frac{1}{3}\left(x-\frac{2\sqrt{5}}{15}+\frac{1}{(\pi+2)\sqrt{10}}\right) -\frac{\sqrt{5}}{15}+ \frac{3}{(\pi+2)\sqrt{10}}$$

There will also be a circle of radius \(\frac{1}{\pi+2}\) around \(Z_2\). I’ll illustrate this “forbidden zone” in black. Let \(f_1\) be the name of the boundary of this black region.

Next, I’ll notice that Wherever we place this instance of \(M\), it must be fully intersecting with \(z’\). If \(Z_4\) is in the interior of \(M\), then \(M\) is not fully intersecting with \(z’\). Similarly, if both \(Z_1\) and \(P\) are in the interior of \(M\), \(M\) is not fully intersecting with \(z’\). I’ll illustrate these regions in black as well.

What is the line with the greatest slope passing through both \(Z_1\) and \(M\)? There’s a whole host of ways to solve this question, but I’ve found that this is the quickest one. Let’s define \(p,q\) to be center of the circle inside \(M\), so that our diagram must include all points in the circle defined by

$$(y-q)^2+(x-p)^2=\frac{1}{\left(2+\pi\right)^2}$$

Any line passing through \(Z_1\) has the form \(y=mx\) where \(m\) is the slope. At what locations does this pass through the circle? Substituting this into the previous equation gives:

$$\begin{align*} (mx-q)^2+(x-p)^2 &=\frac{1}{\left(2+\pi\right)^2}\\ m^2x^2-2qmx+q^2+x^2-2px+p^2 &=\frac{1}{\left(2+\pi\right)^2}\\ (m^2+1)x^2+(-2qm-2p)x+q^2+p^2 -\frac{1}{\left(2+\pi\right)^2}&=0 \end{align*}$$

The line is tangent to the circle when this equation has only one solution. Using the discriminant gives:

$$\begin{align*} (-2qm-2p)^2-4(m^2+1)\left(q^2+p^2 -\frac{1}{\left(2+\pi\right)^2}\right) &= 0\\ 4q^2m^2+8qpm+4p^2-4(m^2+1)\left(\frac{1}{\left(2+\pi\right)^2}-q^2-p^2\right) &= 0\\ q^{2}m^{2}+2qpm+p^{2}-(m^2+1)\left(q^2+p^2-\frac{1}{\left(2+\pi\right)^{2}}\right) &= 0\\ \left(q^2+\frac{1}{\left(2+\pi\right)^2}-q^2-p^2\right)m^2+2qpm+p^2+\left(\frac{1}{\left(2+\pi\right)^2}-q^2-p^2\right) &= 0\\ \left(\frac{1}{\left(2+\pi\right)^2}-p^2\right)m^2+2qpm+\frac{1}{\left(2+\pi\right)^2}-q^2 &= 0 \end{align*}$$

Therefore, the possible slopes are given by

$$\begin{align*} m &=\frac{-2qp\pm\sqrt{\left(2qp\right)^2-4\left(\frac{1}{\left(2+\pi\right)^2}-p^2\right)\left(\frac{1}{\left(2+\pi\right)^2}-q^2\right)}}{2\left(\frac{1}{\left(2+\pi\right)^2}-p^2\right)}\\ m &=\frac{-qp\pm\sqrt{q^2p^2-\left(\frac{1}{\left(2+\pi\right)^2}-p^2\right)\left(\frac{1}{\left(2+\pi\right)^2}-q^2\right)}}{\frac{1}{\left(2+\pi\right)^2}-p^2}\\ m &=\frac{-qp\pm\sqrt{q^2p^2-\left(\frac{1}{\left(2+\pi\right)^4}-\frac{p^2+q^2}{\left(2+\pi\right)^2}+p^2q^2\right)}}{\frac{1}{\left(2+\pi\right)^2}-p^2}\\ m &=\frac{-qp\pm\sqrt{q^2p^2-\frac{1}{\left(2+\pi\right)^4}+\frac{p^2+q^2}{\left(2+\pi\right)^2}-p^2q^2}}{\frac{1}{\left(2+\pi\right)^2}-p^2}\\ m &=\frac{-qp\pm\sqrt{\frac{p^2+q^2}{\left(2+\pi\right)^2}-\frac{1}{\left(2+\pi\right)^4}}}{\frac{1}{\left(2+\pi\right)^2}-p^2}\\ m &=\frac{-qp\pm\frac{1}{2+\pi}\sqrt{p^2+q^2-\frac{1}{\left(2+\pi\right)^2}}}{\frac{1}{\left(2+\pi\right)^2}-p^2}\\ m &=\frac{qp\pm\frac{1}{2+\pi}\sqrt{p^2+q^2-\frac{1}{\left(2+\pi\right)^2}}}{p^2-\frac{1}{\left(2+\pi\right)^2}}\\ \end{align*}$$

Because of this, we can use the following equation to give an extra side to \(z’\)

$$y=\frac{qp+\frac{1}{2+\pi}\sqrt{p^2+q^2-\frac{1}{\left(2+\pi\right)^2}}}{p^2-\frac{1}{\left(2+\pi\right)^{2}}}x$$

This is shown here:

By the same token, we can draw a line from \(Z_3\) to a point on \(M\) that it is tangent to. We can take our previous equation, and replace \(p\) with \(p-\frac{\sqrt{5}}{5}\), while replacing \(q\) with \(q-\frac{\sqrt{5}}{15}\). This equation would be:

$$y-\frac{\sqrt{5}}{15}=\frac{\left(q-\frac{\sqrt{5}}{15}\right)\left(p-\frac{\sqrt{5}}{5}\right)-\frac{1}{2+\pi}\sqrt{\left(p-\frac{\sqrt{5}}{5}\right)^2+\left(q-\frac{\sqrt{5}}{15}\right)^2-\frac{1}{\left(2+\pi\right)^2}}}{\left(p-\frac{\sqrt{5}}{5}\right)^2-\frac{1}{\left(2+\pi\right)^2}}\left(x-\frac{\sqrt{5}}{5}\right)$$

This is shown here:

What is the lowest possible slope of the line tangent to \(M\) passing through \(Z_1\)? We may place the center of \(M\) at the point where \(f_1\) intersects the circle around \(Z_4\). This point is about \(\left(0.57391,0.0918346\right)\). Plugging these points in for \(p\) and \(q\) gives a particular slope for that line. We will define this with \(m_1\). Writing out the precise value of \(m_1\) would time a very long time, so sufice it to say that \(m_1\approx 0.5461546\).

Similarly, I would define \(m_2\) as the minimal slope of the line tangent to \(M\) passing through \(Z_3\). Placing the center of \(M\) at the point where \(f_1\) intersects the circle around \(P\) would mean setting \(p\approx 0.115614\), while \(q\approx -0.004593\). If \(M\) is placed at this point, then \(m_2\approx-0.127902662317\).

Now, graph the two lines

$$y=m_1x$$ $$y-\frac{\sqrt{5}}{15}=m_2\left(x-\frac{\sqrt{5}}{5}\right)$$

These two lines intersect at a point that is approximately \((0.306014,0.167131)\). Regardless of where \(M\) is located, this point must be included in the convex hull of \(M\) and \(z’\). Additionally, I’ve been rounding down, meaning that this point really is included, not just one close to it. This extended version of \(z’\) is shown below: