7

I'm a total newb and bottom rung hobbyist mathematician, just fair warning.

Say points a,b,c,d,e are in the plane (general position) and triangle abc contains point d, triangle ade contains point c. It seems intuitive that triangle bcd cannot contian point e, but I don't know how to prove it, without trying actual values for the coordinates. My first thought was to set up a system of inequalities, one for each statement of containment, similar to barrycentric coordinates (but using cross products), then somehow deduce a contradiction.

The inequality list looks like this. I gather one would have to check a number of different sign conditions, and perhaps all possible orderings of x and y. I'm ok to assume a specific ordering, like $a_x < e_x < c_x < d_x < b_x$ and $a_y < b_y < d_y < c_y < e_y$, if that helps. I already checked for simple sign conflicts (some orderings are easy to prove immediately via sign conflicts):

ABC contains D $$(d_x - a_x) (c_y - a_y) - (c_x - a_x) (d_y - a_y)>0$$ $$(d_x - b_x) (a_y - b_y) - (a_x - b_x) (d_y - b_y)>0$$ $$(d_x - c_x) (b_y - c_y) - (b_x - c_x) (d_y - c_y)>0$$

ADE contains C $$(c_x - a_x) (e_y - a_y) - (e_x - a_x) (c_y - a_y)>0$$ $$(c_x - d_x) (a_y - d_y) - (a_x - d_x) (c_y - d_y)>0$$ $$(c_x - e_x) (d_y - e_y) - (d_x - e_x) (c_y - e_y)>0$$

BCD contains E $$(e_x - b_x) (d_y - b_y) - (d_x - b_x) (e_y - b_y)>0$$ $$(e_x - c_x) (b_y - c_y) - (b_x - c_x) (e_y - c_y)>0$$ $$(e_x - d_x) (c_y - d_y) - (c_x - d_x) (e_y - d_y)>0$$

Can it be shown, just with the above statements, that the three "containment" statements cannot co-exist? I guess we also need to say that no three points are colinear and no two points are the same. Here's an example drawing, but of course I'm looking to prove this without assuming any particular configuration of the points. While a geometric reasoning proof would be interesting, my goal is to be able to programatically prove these sorts of statements, so analytical is best. I'm happy to improve the question with any feedback. thanks!

ABC contains D, ADE contains E, BCD cannot contain E

4 Answers4

2

Suppose $A,B,C,D,E$ are points in the plane, no three of which are collinear, such that

  • Triangle $ABC$ contains point $D$.$\\[4pt]$
  • Triangle $ADE$ contains point $C$.

Claim:$\;$Triangle $BCD$ does not contain point $E$.

Proof:

Since $A,B,C$ are not collinear, there are unique real numbers $r,s,t$ with $r+s+t=1$ such that $$ D=rA+sB+tC\qquad(\text{eq}1) $$ Since $A,D,E$ are not collinear, there are unique real numbers $u,v,w$ with $u+v+w=1$ such that $$ C=uA+vD+wE\qquad(\text{eq}2) $$ Since $B,C,D$ are not collinear, there are unique real numbers $x,y,z$ with $x+y+z=1$ such that $$ E=xB+yC+zD\qquad(\text{eq}3) $$ Since $D$ is strictly contained in triangle $ABC$, it follows that $r,s,t > 0$, and since $C$ is strictly contained in triangle $ADE$, it follows that $u,v,w > 0$.

Computing $u{*}(\text{eq}1)-r{*}(\text{eq}2)$, and then solving the resulting equation for $E$, we get $$ E = \Bigl(\frac{us}{wr}\Bigr) B + \Bigl(\frac{ut}{wr}+\frac{1}{w}\Bigr) C + \Bigl(-\frac{u}{wr}-\frac{v}{w}\Bigr) D $$ Summing the coefficients of $B,C,D$, we get \begin{align*} & \Bigl(\frac{us}{wr}\Bigr) + \Bigl(\frac{ut}{wr}+\frac{1}{w}\Bigr) + \Bigl(-\frac{u}{wr}-\frac{v}{w}\Bigr) \\[4pt] =\;& \frac{r+us+ut-u-vr}{wr} \\[4pt] =\;& \frac{r-u(1-s-t)-vr}{wr} \\[4pt] =\;& \frac{r-ur-vr}{wr} \\[4pt] =\;& \frac{r(1-u-v)}{wr} \\[4pt] =\;& \frac{rw}{wr} \\[4pt] =\;& 1 \\[4pt] \end{align*} hence, from the uniqueness of $x,y,z$, it follows that $$ z=-\frac{u}{wr}-\frac{v}{w} $$ so $z < 0$, and therefore $E$ is not contained in triangle $BCD$.

quasi
  • 58,772
  • I'm not quite grasping part where r,s,t>0 implies D is contained in ABC. So far I've found three ways to show point inside triangle: Barrycentric coordinates, cross products, and the area sum method (used by David Raveh below). It seems like here we're saying ABC contains D iff D = rA +sB +tC and r,s,t>0. Is that the idea or what am I missing? Obviously there are r,s,t>0 where rA+sB+tC is not inside ABC, right? – Jay Hurley Jan 08 '23 at 18:49
  • @Jay Hurley: I had inadvertently omitted the condition $r+s+t=1$. Now fixed. – quasi Jan 09 '23 at 03:02
  • To help complete my understanding, is it right that the interior points of ABC are rA+sB+tC where 0<r,s,t<1 and r+s+t=1? We have r,s,t>0, but do we also need r,s,t<1, or is that extraneous already having r+s+t=1 and r,s,t>0? – Jay Hurley Jan 09 '23 at 05:54
  • @Jay Hurley: Yes, a point $D$ lies in the interior of triangle $ABC$ if and only if $D$ is a strict convex combination of $A,B,C$, or equivalently, $D=rA+sB+tC$, where $r,s,t$ are real numbers such that $r+s+t=1$ and $r,s,t > 0$ (hence also $r,s,t < 1$). – quasi Jan 09 '23 at 06:53
  • are we sure about the "only if" part? Triangle with points A(1,1), B(2,1), C(1,2), and coefficients r=1.1, s = 0, t = 0. Then rA +sB + tC = (1.1, 1.1) which is within the triangle, but r+s+t = 1.1 which is greater than 1. Seems like the 'if" works but not the "only if", or am I missing something? – Jay Hurley Mar 02 '23 at 17:42
  • @Jay Hurley: Since the three points $A(1,1),;B(2,1),;C(1,2)$ are not collinear, it follows that for every point $D$ of the plane, there are unique real numbers $r,s,t$ with $r+s+t=1$ such that $D=rA+sB+tC$. It's those values of $r,s,t$ for which the "if and only if" criterion applies. In particular, for the point $D(1.1,1.1)$, the unique values of $r,s,t$ with $r+s+t=1$ such that $D=rA+sB+tC$ are the values $r=.8,;s=.1,;t=.1$. – quasi Mar 02 '23 at 21:03
2

One can give a geometrical proof by contradiction.

Suppose point $E$ lies inside triangle $BDC$. As all vertices of $BCD$ belong to $\angle BAC$, then all interior points of $BCD$ are also interior points of $\angle BAC$, and $E$ lies then inside $\angle BAC$, as $D$ does.

Hence $\angle AED$ is all inside $\angle BAC$ (with the exception of point $A$) and cannot contain point $C$, which contradicts a given hypothesis. It follows that $E$ doesn't lie inside $BDC$.

Intelligenti pauca
  • 50,470
  • 4
  • 42
  • 77
0

My method was by area: a point is inside a triangle if the $3$ triangle areas it creates (found via cross-product) add up to the total triangle areas: thus, $D$ and $C$ inside $ABC$ and $ADE$ respectively, imply

  1. $$|AD\times AC|+|AB\times AD|+|BD\times CD|=|AB\times AC|$$
  2. $$|AD\times AC|+|AC\times AE|+|DE\times CD|=|AD\times AE|$$ $$|AB\times AC|+|AC\times AE|+|DE\times CD|=|BD\times CD|+|AB\times AD|+|AD\times AE|$$

and is also clear from the construction of the triangles that both $C$ and $D$ lie in the triangle $AEB$. Edit: I fully justify this by considering the extension of the line $DE$ to where it intersects $AB$, and call this point $F$ [this point exists because it must cross two lines of the $ABC$ triangle and it doesn't cross $AC$ (note that $D$ can't lie on the line $AB$)]. So $AED$ is a proper subset of $AEF$. Making similar arguments, $AEF$ is a proper subset of $AEB$. So $C$ lies in $AEB$, and D as well. This implies that

  1. $$|AB\times AD|+|AD\times AE|+|BD\times DE|=|AB\times AE|$$
  2. $$|AB\times AC|+|AC\times AE|+|BC\times BE|=|AB\times AE|$$ $$|AB\times AD|+|AD\times AE|+|BD\times DE|=|AB\times AC|+|AC\times AE|+|BC\times BE|$$

hold as well. Suppose $E$ lay in $DBC$. Then

  1. $$|BD\times DE|+|DE\times CD|+|BC\times BE|=|BD\times CD|$$

holds. But (1), (2), and (5) imply that

$$|AB\times AD|+|AD\times AE|+|BD\times DE|+|BC\times BE|=|AB\times AC|+|AC\times AE|$$

and (3) and (4) thus imply $$2|BC\times BE|=0$$ so we have achieved our contradiction.

David Raveh
  • 1,795
  • I'm not sure about this: "and is also clear from the construction of the triangles that both $C$ and $D$ lie in the triangle $ABE$". It seems to me that this is an essential part of what's to be proved. Why is it clear? – joriki Jan 08 '23 at 11:01
  • I don't agree that this statement is "essentially" the same as what is to be proved, as it can be shown that ABC and ADE are subsets of ABE. I will try to prove this fully tomorrow; I need to sleep :) – David Raveh Jan 08 '23 at 11:07
  • I didn't write that it's essentially the same, just that it's an essential part. Sleep well! :-) – joriki Jan 08 '23 at 11:07
  • @joriki I guess no sleep for me. Let me know if my justification remains unsatisfactory. Thanks and good night (morning??) – David Raveh Jan 08 '23 at 12:25
  • In "$AED$ is a proper subset of $AEF$", you seem to be assuming that $E$, $D$ and $F$ occur in this order – it's not obvious to me why they couldn't occur in the order $DEF$ or even $DFE$. Good day (from Germany)! – joriki Jan 08 '23 at 12:37
  • $ED$ must cross $CB$ and must end inside $ABC$. So if I extend $ED$ it must intersect one other line in the triangle, and I call this point $F$. Note that extending the line fully forces it to cross 2 lines. Guten Morgan aus America. I am now fully expecting you to ask how I know it crosses $CB$ – David Raveh Jan 08 '23 at 12:44
  • No, you did explain why it crosses $CB$ – because it doesn't cross $AC$ (since $C$ is inside $ADE$) and $D$ is inside $ABC$, so any line through $D$ has to cross both $AB$ and $CB$. What I don't see is why $ED$ crossing $CB$ and "ending" inside $ABC$ (I assume by this you mean that its one endpoint $D$ lies in $ABC$) implies that $E$, $D$ and $F$ occur in this order. Why can't $ED$ cross $CB$ in $F$ such that the order is $DEF$ or $DFE$? You seem to be assuming something about the position of $E$ there that's not obvious to me. – joriki Jan 08 '23 at 12:53
  • Thanks David. I'm interested as to whether it is a requirement to first prove that ABE contains C and D, then use those additional premises to show E cannot be inside BCD. My program that proves these sorts of statements indeed results in "not possible" if I include the extra two implied containments. Do you think this is impossible to prove with just ABC contains D and ADEC contains C? Or will it be necessary for my program to go through and first prove that C and D are forced to be in ABE? – Jay Hurley Jan 08 '23 at 19:03
  • @JayHurley I can't comment on how your program will run, but the statement can be proved without showing the 2 implied containments (see Intelligenti pauca answer). Using an area method, it is useful to make use of the implied containments to obtain useful relationships, but I doubt they must be obtained by this method. Great question, kept me up at night. – David Raveh Jan 08 '23 at 19:20
0

I started working on this answer before the others were posted. It doesn’t add anything geometrically to their proofs, but it does relate them to your system of equations, and since you wrote that that’s your main focus, perhaps it will be relevant for you.

Each of your inequalities is a statement that three points form a clockwise triangle in a particular order. To simplify them, assume $a_x=a_y=0$ without loss of generality and let $PQ$ denote $|\vec P\times\vec Q|=p_xq_y-q_xp_y$, which is twice the oriented area of the triangle $PQA$ (so $QP=-PQ$). Then the system simplifies to

\begin{eqnarray} DC & > & 0 \;, \tag1 \\ BD & > & 0 \;, \tag2 \\ CD + DB + BC & > & 0 \;, \tag3 \\ \\ CE & > & 0 \;, \tag4 \\ DC & > & 0 \;, \tag5 \\ EC + CD + DE & > & 0 \;, \tag6 \\ \\ ED + DB + BE & > & 0 \;, \tag7 \\ EB + BC + CE & > & 0 \;, \tag8 \\ EC + CD + DE & > & 0 \;. \tag9 \end{eqnarray}

The first thing to note is that your choice of orientations is justified: You can choose the orientation of $ABC$ arbitrarily without loss of generality; then the consistency of $(1)$ and $(5)$ fixes the orientation of $ADE$, and then the consistency of $(6)$ and $(9)$ fixes the orientation of $BCD$.

Note also that for each containment, adding the three equations yields the equation for the orientation of the containing triangle (after cancellation of opposite terms), which corresponds to the addition of triangle areas in David Raveh’s answer.

Adding $(4)$ and $(5)$ to $(9)$ yields $DE>0$, and adding that and $(2)$ to $(7)$ yields $BE>0$. Together with $(1)$, $(2)$, $(4)$ and $(5)$, we thus have all $6$ equations that all pairs of points other than $A$ lie in the circular order around $A$ that they do in your diagram. We can simplify a bit further by introducing the corresponding six positive variables $p=BD$, $q=DC$, $r=CE$, $s=BC$, $t=DE$ and $u=BE$. The four equations that state more than the positivity of these variables then read

\begin{eqnarray} s & > & p + q \;, \tag{3'}\\ t & > & r + q \;, \tag{6'}\\ u & > & p + t \;, \tag{7'}\\ r + s & > & u \;. \tag{8'} \end{eqnarray}

I was hoping that this in itself would yield a contradiction, but it doesn’t; these inequalities can be satisfied with positive values of the variables, so we do need to use their definitions in terms of the $PQ$ to derive a contradiction.

At this point, having read the other answers, I realized that the circular order of the other points around $A$ derived above is actually all we need – if $B$, $D$, $C$ and $E$ lie around $A$ in that circular order, then $E$ can't lie in $BCD$.

One more thing you might want to take a look at: There are criteria for the planarity of a graph, that is, for when it’s possible to embed it in the plane such that none of its edges cross. One such criterion is that a plane graph cannot contain (a subdivision of) the complete graph on $5$ vertices, $K_5$. So if you can prove that your premises imply that none of the segments between pairs of points cross, that would yield a contradiction, since the diagram would then be an embedding of $K_5$ in the plane without crossings. This is easy for most of the potential crossings, since they’re excluded either by the triangle containments or by the circular order around $A$; but it’s not as easy for the potential crossing of $BC$ and $DE$ (the one actually realized in the diagram).

joriki
  • 238,052
  • 1
    Thanks joriki, you've come to the same conclusion I had, except a lot faster (took me weeks). There are cases where the "product terms" exhibit contradiction on their own. In fact, in my example, if the x-order was Ax<Cx<Ex<Dx<Bx, which is just nudging E to the right past C, the original equations have a sign conflict. The second to last (middle equation in BCD contains E) becomes ->+. The ordering in the OP is the challenging case, where I think you're right: we have to "get inside" the product terms to deduce the contradiction, if possible. – Jay Hurley Jan 08 '23 at 18:56
  • Secondly, on the circular order around A. I see in the equations that D is left of AB, C is left of AD, and E is left of AC. And it makes intuitive sense that for E to be inside BCD, then E would have to be between AB and AC, but that still seems like a leap of intuition to me and I don't get how to make that clear with the x,y values of each point. I'm interested if it's possible or impossible with just these statements, or if we have to prove something else first and add that to the mix before the contradiction can be shown. – Jay Hurley Jan 08 '23 at 19:09
  • @JayHurley: I think you meant "$E$ is left of $AC$"? To your point: All of $ABC$ is on the same side of $AB$ as $C$, and all of $ABC$ is on the same side of $AC$ as $B$. So if $E$ were in $ABC$, it would have to be left of $AB$ like $C$ and right of $AC$ like $B$. (Note that this is not implied in just the relationships that you mentioned – $D$ being left of $AB$ and $C$ being left of $AD$ doesn't by itself imply that $C$ is left of $AB$ (the angles $\angle BAD$ and $\angle DAC$ could add up to more than $\pi$), but we know independently that it is.) – joriki Jan 08 '23 at 19:19