14

This seems to be a curious problem:

Four points $A,B,C,D$ are chosen on a sphere of radius $1$ such that the centre $O$ of the sphere lies inside the tetrahedron $ABCD$. Prove that $|\vec{OA}+\vec{OB}+\vec{OC}+\vec{OD}|<2$.

An analogous problem in the plane can be formulated like this:

If three points $A,B,C$ are chosen on a circle of radius $1$ such that the centre $O$ of the circle lies inside triangle $ABC$, then $|\vec{OA}+\vec{OB}+\vec{OC}|<1$.

I don't know how the first problem is solved, but I also don't have a good solution to the second problem, just some thoughts:

We can assume that the angles $AOB$, $BOC$, $AOC$ satisfy the inequalities $$ 180^\circ\geq\ \angle AOB\ \geq\ \angle BOC\ \geq\ \angle AOC. $$ Then $AOB\geq120^\circ$ and point $C$ lies inside arc $A'B'$ (see figure). If $\vec{OP}=\vec{OA}+\vec{OB}$ and $\vec{OQ}=\vec{OP}+\vec{OC}$, then $|OP|<1$ and it is easy to see that $\angle POQ\>\angle OPQ$. (I wrote easy to see, but to be honest I could not strictly justify this point. These angles are equal if $C=A'$ or $C=B'$; if $C$ lies on arc $A'BAB'$ then the opposite inequality holds; if $C$ lies on arc $A'B'$ (as in the figure) then the above inequality holds.) Consider $\triangle OCQ$, $\angle CQO=\angle POQ\geq90^\circ$ $\Longrightarrow$ $OQ<OC=1$.

![enter image description here

Added 2024/01/07.
Patrick suggested another solution to the problem, but unfortunately his explanations are very vague.

I will try to present his solution here, it seems to be the best one that is presented here.

So, let $\vec{OS}=\vec{OA}+\vec{OB}+\vec{OC}+\vec{OD}$. Let us assume that $\vec{OS}\neq0$, otherwise our statement is obvious. Consider an axis $L$ passing through $O$ and parallel to $\vec{OS}$. We can look at $L$ as a number line, $1$ corresponds to the point of intersection of the ray $OS$ and the sphere. Let $\pi$ be the orthogonal projection of $\mathbb{R}^3$ onto the $L$. If $\pi(A)=a,\pi(B)=b,\pi(C)=c,\pi(D)=d,\pi(S)=s$, then $s=|\vec{OS}|$ and $a+b+c+d=s$. We can assume that $$ -1\leq a\leq b\leq c\leq d\leq1. $$ It is clear, if $d=1$, then $c<d=1$. (In the picture below, the named projections are marked in red.)

  1. If $a>0$, then the tetrahedron $ABCD$ lies on one side of the plane perpendicular to $\vec{OS}$ and passing through the center $O$ and so $O$ is outside the tetrahedron.

  2. If $b>-a$ and $a\neq-1$ and $P$ is the plane passing through the three points $A$, $O$, $S$ (in this plane is the figure below), then the tetrahedron $ABCD$ is located on one side of the plane perpendicular to $P$ and passing through $O$ and $A$.

  3. If $a=-1$, then $s=a+b+c+d=(a+d)+b+c=0+b+c<2$.

  4. If $b\leq-a$, then $s=a+b+c+d=(a+b)+c+d\leq0+c+d<2$.

The proof is now complete.

But that still leaves the question. Are there any simpler considerations?

enter image description here

kabenyuk
  • 10,712
  • 2
    For the circle case, if we use inner product, the problem boils down to a trigonometry inequality that is very easy to prove. – Reza Rajaei Dec 30 '23 at 08:58
  • @RezaRajaei, Thank you for your comment. It was very helpful. – kabenyuk Dec 30 '23 at 14:38
  • You're welcome. What's the source of the problem? – Reza Rajaei Dec 30 '23 at 14:57
  • 2
    It is easy to prove that $\vec{OA}+\vec{OB}+\vec{OC} = \vec{OH}$ where H is the orthocentre, which lies inside the triangle since it is acute – D S Dec 30 '23 at 15:14
  • @RezaRajaei This is a reformulation of an exercise from a book in Russian How to solve non-standard problems. – kabenyuk Dec 30 '23 at 15:49
  • @DS, Thank you, you are right, I have seen this formula, but to apply it to our problem we need another formula of length $|OH|$. But I will try to supplement my answer with this solution. – kabenyuk Dec 30 '23 at 16:02
  • Note that $OH^2 = 9R^2 - (a^2+b^2+c^2) = 9-(a^2+b^2+c^2)$, where R is the circumradius, and a,b,c are side lengths, however, this formula is not of much use – D S Dec 30 '23 at 16:23
  • Let the coordinates be $(a_i,b_i,c_i$ for $i = 1,2,3,4$. WLOG assume $c_1 = c_2 = c_3 \le 0$. Then intuitively it seems that the origin lies inside the tetrahedron iff $c_4 \ge -c_1$. – D S Dec 31 '23 at 15:43
  • this may allow us to get a bound.... – D S Dec 31 '23 at 15:46

4 Answers4

4

Not a bad idea given by Reza Rajaei in his comment. Here is another solution to the problem for the circle based on the dot product.

Let $\angle AOB=2\alpha$, $\angle AOC=2\beta$, and $\angle BOC=2\gamma$. First we prove a well-known trigonometric identity $$ \cos2\alpha+\cos2\beta+\cos2\gamma=-1-4\cos\alpha\cos\beta\cos\gamma. $$ Indeed, since $\alpha+\beta+\gamma=180^\circ$ we get \begin{align*} \cos2\alpha+\cos2\beta+\cos2\gamma & \\ = &\ 2\cos(\alpha+\beta)\cos(\alpha-\beta)+\cos2\gamma \\ = & -2\cos\gamma\cos(\alpha-\beta)+2\cos^2\gamma-1\\ = & -1-2\cos\gamma(\cos(\alpha-\beta)-\cos(\alpha+\beta))\\ = & -1 -4\cos\gamma\cos\alpha\cos\beta. \end{align*} Then since $0\leq\alpha,\beta,\gamma<90^\circ$ we have $$ \cos2\alpha+\cos2\beta+\cos2\gamma<-1.\tag1 $$

Now consider the dot product \begin{align*} (\vec{OA}+\vec{OB}+\vec{OC})\cdot(\vec{OA}+\vec{OB}+\vec{OC})&\\ = &\ 3+2\vec{OA}\cdot\vec{OB}+2\vec{OA}\cdot\vec{OC}+2\vec{OB}\cdot\vec{OC}\\ = &\ 3+2\cos2\alpha+2\cos2\beta+2\cos2\gamma. \end{align*} Hence and from $(1)$ it follows that $$ |\vec{OA}+\vec{OB}+\vec{OC}|=3+2(\cos2\alpha+\cos2\beta+\cos2\gamma)<3-2=1. $$

PS. Although this is not a solution to the main problem, but maybe someone will find a similar solution for the 3D problem. On the other hand I am sure that there must be a purely geometric solution for both circle and sphere.

Addition. (2023/12/31)

Here is another proof, of this assertion, which D S says in his commentary.

Let $O$ be the circumcenter of a triangle $\triangle ABC$ and let $Q$ denote the orthocenter of $\triangle ABC$, then $$ \vec{OQ}=\vec{OA}+\vec{OB}+\vec{OC}.\tag2 $$ This vector equality is sometimes called Sylvester's triangle problem.

As we know, the orthocentre is inside an acute triangle. In our case, the triangle $\triangle ABC$ is an acute triangle, as follows from the condition. Hence $|OQ|<1$ and according to $(2)$ we get $\vec{OA}+\vec{OB}+\vec{OC}<1$.

For the sake of completeness, I will give a short proof of $(2)$ (see the figure from my question). Let $O$ be the circumcenter of a triangle $\triangle ABC$, and let $M$ be the midpoint of side $AB$, $S$ the median intersection, and $\vec{SQ}=2\vec{OS}$. So $MS=2CS$ and the triangles $MOS$ and $CQS$ are similar.
Hence $OM\|CQ$, $CQ$ is an altitude of the triangle, and $Q$ is the orthocenter of the triangle. Now we get \begin{align*} &\vec{OA}+\vec{OB}=2\vec{OM}=\vec{CQ} \text{ and } \vec{OC}+\vec{CQ}=\vec{OQ}\\ \Longrightarrow\ & \vec{OA}+\vec{OB}+\vec{OC}=\vec{OQ}. \end{align*}

kabenyuk
  • 10,712
  • 1
    I checked the inequalities I referred to in the comments above but it seems that the final inequality is not easy to work with, considering the fact that $2$ is kind of a limit point. However, many special cases can be simply verified. As you suggested, there must be an elegant approach, maybe a physical notion about vectors or an invariant. – Reza Rajaei Dec 31 '23 at 08:37
  • 1
    Thank you for your comments and efforts. So far the 3D case is not working out. I'm trying to get more ways to prove for the 2D case in the hope that one of them will help for the 3D case as well. – kabenyuk Dec 31 '23 at 11:28
  • @RezaRajaei, thank you for your detailed comments. If you had posted this as a reply, we could have discussed some of your points of reasoning underneath this reply of yours. There is nothing wrong with this answer not solving the problem completely. Let it be a partial answer. – kabenyuk Jan 03 '24 at 06:46
  • @kebenyuk I just edited the idea and posted it. – Reza Rajaei Jan 03 '24 at 06:49
3

I think the idea mentioned by @D S in the comments might work. WLOG, we may assume:

$$B=(x_1,y_1,-z_1), \\ C=(x_2,y_2,-z_1), \\D=(x_3,y_3,-z_1),$$ where $z_1 \geq 0.$ Moreover, we assume $A=(x_4,y_4,z_4)$ with $z_4>0$ and $O=(0,0,0).$ Note that if $z_4 \leq z_1$, then $O$ does not lie inside the tetrahedron. Hence, we suppose $z_4>z_1.$

I think at this point we need to make use of the $2$D case. Let $H$ be the projection of $O$ onto the plane $z=-z_1.$ Since $O$ lies inside the tetrahedron,$|\vec{HB}+\vec{HC}+\vec{HD}|$ is exactly the $2$D case where radius $1$ is replaced with radius $\sqrt{1-z_1^2}$. To see that, just consider the circle shaped by the intersection of the sphere and the plane $z=-z_1.$ Thus, we will have:

$$|\vec{HB}+\vec{HC}+\vec{HD}|^2=(x_1+x_2+x_3)^2+(y_1+y_2+y_3)^2<1-z_1^2. \ \ \ \ \ \ (*)$$

Now, notice that our main goal is to prove:

$$(x_1+x_2+x_3+x_4)^2+(y_1+y_2+y_3+y_4)^2+(-3z_1+z_4)^2<4.$$

Considering $(*)$, it is enough to show:

\begin{align*} (1-z_1^2)+x_4^2+2x_4(x_1+x_2+x_3)+y_4^2+2y_4(y_1+y_2+y_3)+9z_1^2+z_4^2-6z_1z_4 <4\\ \iff 4z_1^2-3z_1z_4+x_4(x_1+x_2+x_3)+y_4(y_1+y_2+y_3)<1. \end{align*}

However, using $(*)$ once again together with $z_4>z_1 \geq 0$, we have:

\begin{align*} 4z_1^2-3z_1z_4+x_4(x_1+x_2+x_3)+y_4(y_1+y_2+y_3)&\\ \leq 4z_1^2-3z_1z_4+ \frac{x_4^2+(x_1+x_2+x_3)^2}{2}+\frac{y_4^2+(y_1+y_2+y_3)^2}{2} & \\ <4z_1^2-3z_1z_4+\frac{2-z_1^2-z_4^2}{2} &\\ <4z_1^2-3z_1^2-\frac{z_1^2+z_4^2}{2}+1 &\\ =\frac{z_1^2-z_4^2}{2}+1<1. \end{align*}

We are done.


Another proof for the $2$D case:

WLOG, we may assume $A=(x_1, y_1)$ and $B=(-x_1,y_1)$ where $y_1 > 0$. In order to have $O=(0,0) $ inside the triangle, we must have $C=(x_3,-y_3)$ such that $y_3>y_1> 0$.

Now, we need to prove that:

$$|\vec{OA}+\vec{OB}+\vec{OC}|=|(x_3,-y_3+2y_1)|<1.$$

However, $$x_3^2+(-y_3+2y_1)^2=x_3^2+y_3^2-4y_3y_1+4y_1^2=1-4y_1(y_3-y_1)<1.$$

kabenyuk
  • 10,712
Reza Rajaei
  • 5,183
  • I have corrected the formulas a bit. I agree with you that $z_1<z_4$, but I would like to have a little bit more of an explanation for that. Other than that, it seems to be properly correct. Of course I wish there were less computation. – kabenyuk Jan 03 '24 at 11:48
  • 1
    @kabenyuk Yes. You are right. One may even argue why we are allowed to choose $B$, $C$, $D$ is that way having the same $-z_1$. That's easy because a (unique) plane passes through three points. For $z_1<z_4$, let's consider two cones connected at $O$ whose bases are the planes $z=-z_1$ and $z=z_1$. Now, pick any point $(x,y,z)$ on the sphere with $z<z_1$. It is at least intuitively obvious that the corresponding tetrahedron does not contain $O$. I just realized this argument gives us a very simple solution to the $2$D case. – Reza Rajaei Jan 03 '24 at 13:09
  • 1
    As for the $B,C,D$ coordinates, I get it. It's a great solution for the circle. – kabenyuk Jan 03 '24 at 13:25
  • 1
    I chose your answer because you took a very important step first. Good luck in all you do. – kabenyuk Jan 08 '24 at 14:22
  • @kabenyuk Thank you, Sir. – Reza Rajaei Jan 08 '24 at 15:37
3

Suppose that instead of maximizing the length of the sum of the four vectors we instead want to maximize the x-coordinate of the sum of the four vectors. If we can prove that there is a strict upper bound for this then this solves the stated problem. To see this note that we can choose the system of coordinates to that the sum of the vectors end up on the positive x-axis.

For the origin to be in the interior of the tetrahedron then one of the four points without loss of generality A has to have a negative x-coordinate. Otherwise the tetrahedron would not intersect the plane x=0 that contains the origin.

Now project onto the plane spanned by A and the x-axis. We have now 4-gon in the circle disk so that the origin is in the interior and one of the vertices is A.

One of the points obtained by projecting B,C,D has to have x-coordinate at most the same as the point -A for the origin to be in the interior (otherwise the line segments from A to these points would all be on the same side of the line trough A and the origin). The other points has x-coordinate at most the same as (1,0,0) as we are on a circle centered on the origin with radius 1.

Now the rather deranged tetrahedron spanned by the points A,-A,(1,0,0),(1,0,0) gives a strict upper bound of 2 for the x-coordinate of the sum.

Patrik
  • 141
  • This does not provide an answer to the question. To critique or request clarification from an author, leave a comment below their post. - From Review – Divide1918 Jan 05 '24 at 10:26
  • Hope the clarification makes the answer ok. I'm a bit curious what was difficult to understand before? – Patrik Jan 05 '24 at 11:12
  • The point B' would be my point A (it has to have negative x-coordinate). In your example all other points have x coordinates greater than 0.1 (the x-coordinate of your -B' and my -A). This tetrahedron will not contain the origin. – Patrik Jan 05 '24 at 14:06
  • How about this reformulation: Let A be the point furthest from the point P = A+B+C+D (sorry for sloppy notation). Let d be the distance from -A to P. Let S be the circle on the unit sphere consisting of all points that have distance d to P. Note that B, C, D can not all be on the side of S that are closer to P, as then the origin would not be in the tetrahedron (as the origin is on the line from A to -A). Now project on the axis parallel to P. We cant beat the length obtained by A and -A together two unit vectors in the same direction as P. – Patrik Jan 05 '24 at 15:46
  • In general no. But if we are free to choose coordinate systems then yes. Sorry for breaking the norms of this community, I felt that the reformulation had quite a different flavor than what I did in the original formulation so did not want to delete that. The trick with coordinates is not really needed but kind of nice in my mind so did not want to remove it. – Patrik Jan 05 '24 at 16:35
  • I think I'm beginning to understand your reasoning. I think it would be properly to supplement your answer with a good drawing. – kabenyuk Jan 05 '24 at 16:38
  • @Patrik, I am going to try to put your thoughts into a generally accepted language. In short, you're looking at the projection of $\mathbb{R}^3$ onto the axis $L$ passing through $O$ and parallel to $\vec{OS}$, where $\vec{OS}$ is our sum. We can identify $L$ with the number line. If $a,b,c,d,s$ are the projections of the points of $A,B,C,D,S$, then $$a+b+c+d=s \text{ and } |OS|=s. $$ It remains to realize that if $a<0$ and $b,c,d>-a$, then $O$ lies outside our tetrahedron. This is what I don't understand from you. – kabenyuk Jan 06 '24 at 08:58
  • If z is the distance from the point -A to S then look at all points on the unit sphere that have distance z to S, this is a circle. If b, c, d > -a then the points B, C, D are all on one side of the circle. This will not give a tetrahedron containing the origin as the origin is on the line from A to -A (and -A is on the circle). – Patrik Jan 06 '24 at 15:17
  • @Patrik It's still unclear, but I've apparently added your solution to my question. – kabenyuk Jan 07 '24 at 13:18
1

Reza Rajaei's solution to the sphere problem is wonderful, but requires some computation. After some thought about his solution, I have found that the computations required is not that much.

I think it is reasonable to post this reasoning as a new answer. enter image description here Circle case.

Let the three points of $A,B,C$ be placed on a circle with a radius of $1$. If the center $O$ of the circle lies inside $\triangle ABC$, then $$ |\vec{OA}+\vec{OB}+\vec{OC}|<1. $$

Proof. Let $OQ\perp AB$. We have $\vec{OA}+\vec{OB}=2\vec{OQ}$.

Let $A'$ and $B'$ be points centrally symmetric to points $A$ and $B$ respectively with respect to the center $O$. The center $O$ lies inside $\triangle ABC$ if and only if $C$ lies inside the arc $A'UB'$ (see figure on the left).

It is almost obvious that $$ |2\vec{OQ}+\vec{OC}|\leq|2\vec{OQ}+\vec{OA'}| $$ for any point $C$ on arc $A'UB'$. (You can for example compute $(\vec{OC}+2\vec{OQ})^2$ then use the geometric definition of dot product, the properties of the $\cos$, and the fact that $\angle A'OQ>90^\circ$.)

We have $2\vec{OQ}+\vec{OA'}=2\vec{OQ}-\vec{OA}=\vec{OB}$.

Therefore we get $$ |\vec{OA}+\vec{OB}+\vec{OC}|= |2\vec{OQ}+\vec{OC}|\leq |2\vec{OQ}+\vec{OA'}|= |\vec{OB}|\leq1. $$

Case sphere.

Let $A,B,C,D$ be four points on a sphere of radius $1$. If the centre $O$ of the sphere lies inside the tetrahedron $ABCD$, then
$$|\vec{OA}+\vec{OB}+\vec{OC}+\vec{OD}|<2.$$

Proof. Let $P$ be the plane passing through the three points $A,B,C$, and let $Q$ be the projection of the point $O$ onto the plane $P$. Consider a plane $P'$ that passes through the points $O$ and $D$ and is perpendicular to the plane $P$. The figure shows the circle lying in the intersection of the plane $P'$ and the sphere and the line $XY$ lying in the intersection of the planes $P$ and $P'$. Let $X'$ and $Y'$ be points centrally symmetric to points $X$ and $Y$ respectively with respect to the center $O$. We get exactly the same figure as in the case of a circle (in the figure on the right).

We know that $\vec{QA}+\vec{QB}+\vec{QC}=\vec{QH}$ and $|QH|<|QA|$ (|QA| is the radius of the circle lying in the intersection of the sphere with the plane $P$). Hence $$ |\vec{OH}| =|\vec{OQ}+\vec{QH}| \leq|\vec{OQ}+\vec{QX}| =|\vec{OX}| \leq1. $$

Just as in the case of a circle we have $$ |2\vec{OQ}+\vec{OD}|<1. $$

Consequently, \begin{align*} |\vec{OA}+\vec{OB}+\vec{OC}+\vec{OD}|\\ &=|\vec{QA}+\vec{QB}+\vec{QC}+3\vec{OQ}+\vec{OD}|\\ &=|\vec{QH}+3\vec{OQ}+\vec{OD}|\\ &=|\vec{OH}+2\vec{OQ}+\vec{OD}|\\ &<|\vec{OH}|+|2\vec{OQ}+\vec{OD}|=2. \end{align*}

That was what we had to prove.

PS. I hope that everything is all right here. But once again, all of the credit goes to Reza. It is too bad that he does not have a lot of votes for his answer.

kabenyuk
  • 10,712