12

How can I find a fourth line $L$ that intersects three given lines $L_1$, $L_2$, $L_3$ in 3D space?

We can assume that $L_1$, $L_2$, $L_3$ are in "general position", so no two of them are coplanar, etc.

I'm not even sure that three lines is enough to uniquely define $L$, actually. If three lines is not enough, how many do I need?

The question is related to this one. Specifically, see idea #4 in my list of suggested approaches. It requires finding a line that intersects with a few given ones.

Edit:

Apparently, I need four lines, not three, to uniquely define $L$. So, how can I construct a fifth line that interesects four given ones?

I found this paper, and this one, but they are both difficult for me to read. Surely there must have been solutions before 2008, and, if so, I'm hoping that these are easier to understand.

bubba
  • 44,617
  • No, the $L$ will still not be unique, in the generic situation. – Ted Shifrin Dec 15 '13 at 03:58
  • I'm not going to buy Shifrin's book (which he mentions below),and I don't have access to a decent library that might have it. But, it turns out that the answer is in a book that I already own. Snyder and Sisam's "Analytic Geometry of Space". Art. 120. – bubba Jan 21 '16 at 12:15

5 Answers5

10

There is a one-parameter set of such lines, and their union is a ruled surface. The classic question from enumerative geometry is to ask (at most, over $\Bbb R$) how many lines meet four lines in general position. (These sorts of questions are best asked in projective space. But in Euclidean space by "general position" you rule out any sort of parallelism of planes and lines, etc.)

Here's a hint on how to get started: $L_1$ and an arbitrary point on $L_2$ determine a plane, and $L_3$ intersects that plane in a unique point.

Ted Shifrin
  • 125,228
  • Thanks very much. If the four-line question is classical, can you point me to an easy-to-read reference, please. Hopefully something that's not too abstract. I know nothing about projective geometry. I live in a Euclidean space :-) I'm interested in computing the common line, so a constructive process would be welcome. – bubba Dec 15 '13 at 03:26
  • I actually have a hands-on treatment of this in the abstract algebra book I wrote, but it is all set in projective space. I would suggest taking three general skew lines: say the $x$-axis, the line $(0,1,0) + t(1,0,1)$, and the line $(1,2,2)+s(1,0,2)$. Find the ruled surface explicitly (i.e., equation). Then take a general fourth line and see where it intersects that ruled surface. Now figure out, from the nature of the ruled surface, exactly what lines meet all four. If you get stuck in the middle, I'll help. :) BIG HINT: That ruled surface will be DOUBLY ruled. – Ted Shifrin Dec 15 '13 at 03:52
  • I did some numerical examples. I set up parametric equations for the lines, and wrote equations to express the intersections. I found that my equations often had only complex solutions. In any case (real or complex), there were always two distinct solutions. Is this what you would expect? – bubba Dec 15 '13 at 14:27
  • Well, that's the problem with the real number system. For the three lines I gave you, you get a familiar surface. Some lines in R3 do not intersect it (hence your complex roots), but lots do. And, yes, the surface is a quadric, so the general line will meet it in either 2 or 0 points. – Ted Shifrin Dec 15 '13 at 14:44
3

Three skew lines in ${\Bbb R}^3$ generate a hyperboloid of one sheet with equation given as follows:

$$ (x(D_1)\cdot (y(C_2)\cdot z(C_3) - z(C_2)\cdot y(C_3)) + x(D_2)\cdot (y(C_3)\cdot z(C_1) - z(C_3)\cdot y(C_1))+x(D_3)\cdot (y(C_1)\cdot z(C_2) - z(C_1)\cdot y(C_2)))\cdot x^2 + (y(D_1)\cdot (z(C_2)\cdot x(C_3) - x(C_2)\cdot z(C_3)) + y(D_2)\cdot (z(C_3)\cdot x(C_1) - x(C_3)\cdot z(C_1))+y(D_3)\cdot (z(C_1)\cdot x(C_2) - x(C_1)\cdot z(C_2)))\cdot y^2 + (z(D_1)\cdot (x(C_2)\cdot y(C_3) - y(C_2)\cdot x(C_3)) + z(D_2)\cdot (x(C_3)\cdot y(C_1) - y(C_3)\cdot x(C_1))+z(D_3)\cdot (x(C_1)\cdot y(C_2) - y(C_1)\cdot x(C_2)))\cdot z^2 + (x(C_1)\cdot (y(C_2)\cdot y(D_3) - y(D_2)\cdot y(C_3) + z(C_3)\cdot z(D_2) - z(D_3)\cdot z(C_2))+x(C_2)\cdot (y(C_3)\cdot y(D_1) - y(D_3)\cdot y(C_1) + z(C_1)\cdot z(D_3) - z(D_1)\cdot z(C_3))+x(C_3)\cdot (y(C_1)\cdot y(D_2) - y(D_1)\cdot y(C_2) + z(C_2)\cdot z(D_1) - z(D_2)\cdot z(C_1)))\cdot y\cdot z + (y(C_1)\cdot (z(C_2)\cdot z(D_3) - z(D_2)\cdot z(C_3) + x(C_3)\cdot x(D_2) - x(D_3)\cdot x(C_2))+y(C_2)\cdot (z(C_3)\cdot z(D_1) - z(D_3)\cdot z(C_1) + x(C_1)\cdot x(D_3) - x(D_1)\cdot x(C_3))+y(C_3)\cdot (z(C_1)\cdot z(D_2) - z(D_1)\cdot z(C_2) + x(C_2)\cdot x(D_1) - x(D_2)\cdot x(C_1)))\cdot z\cdot x + (z(C_1)\cdot (x(C_2)\cdot x(D_3) - x(D_2)\cdot x(C_3) + y(C_3)\cdot y(D_2) - y(D_3)\cdot y(C_2))+z(C_2)\cdot (x(C_3)\cdot x(D_1) - x(D_3)\cdot x(C_1) + y(C_1)\cdot y(D_3) - y(D_1)\cdot y(C_3))+z(C_3)\cdot (x(C_1)\cdot x(D_2) - x(D_1)\cdot x(C_2) + y(C_2)\cdot y(D_1) - y(D_2)\cdot y(C_1)))\cdot x\cdot y + (x(D_1)\cdot (y(C_2)\cdot y(D_3) - y(D_2)\cdot y(C_3) - z(C_3)\cdot z(D_2) + z(D_3)\cdot z(C_2))+x(D_2)\cdot (y(C_3)\cdot y(D_1) - y(D_3)\cdot y(C_1) - z(C_1)\cdot z(D_3) + z(D_1)\cdot z(C_3))+x(D_3)\cdot (y(C_1)\cdot y(D_2) - y(D_1)\cdot y(C_2) - z(C_2)\cdot z(D_1) + z(D_2)\cdot z(C_1)))\cdot x + (y(D_1)\cdot (z(C_2)\cdot z(D_3) - z(D_2)\cdot z(C_3) - x(C_3)\cdot x(D_2) + x(D_3)\cdot x(C_2))+y(D_2)\cdot (z(C_3)\cdot z(D_1) - z(D_3)\cdot z(C_1) - x(C_1)\cdot x(D_3) + x(D_1)\cdot x(C_3))+y(D_3)\cdot (z(C_1)\cdot z(D_2) - z(D_1)\cdot z(C_2) - x(C_2)\cdot x(D_1) + x(D_2)\cdot x(C_1)))\cdot y + (z(D_1)\cdot (x(C_2)\cdot x(D_3) - x(D_2)\cdot x(C_3) - y(C_3)\cdot y(D_2) + y(D_3)\cdot y(C_2))+z(D_2)\cdot (x(C_3)\cdot x(D_1) - x(D_3)\cdot x(C_1) - y(C_1)\cdot y(D_3) + y(D_1)\cdot y(C_3))+z(D_3)\cdot (x(C_1)\cdot x(D_2) - x(D_1)\cdot x(C_2) - y(C_2)\cdot y(D_1) + y(D_2)\cdot y(C_1)))\cdot z + (x(D_1)\cdot (y(D_2)\cdot z(D_3) - z(D_2)\cdot y(D_3)) + y(D_1)\cdot (z(D_2)\cdot x(D_3) - x(D_2)\cdot z(D_3))+z(D_1)\cdot (x(D_2)\cdot y(D_3) - y(D_2)\cdot x(D_3)))=0$$

where the three lines $L_1,L_2,L_3$ are given by two points on each $A_1,B_1,$ $A_2,B_2,$ and $A_3,B_3.$ And the plücker coordinates are $$(p_{01}(L_i),p_{02}(L_i),p_{03}(L_i),p_{23}(L_i),p_{31}(L_i),p_{12}(L_i))=(x(C_i),y(C_i),z(C_i),x(D_i),y(D_i),z(D_i))$$ where $C_i=B_i-A_i$ is the difference and $D_i=A_i \times B_i$ is the vector product.

Now in, say, geogebra pick the points and the lines on them, and using the equation, draw the hyperboloid of one sheet $Q_{123}$.

Then pick a fourth line, if this line intersects $Q_{123}$ in two points take the tangent planes to $Q_{123}$ on these two intersection points. Quadric with tangent plane For each intersection point the tangent plane will pick out two lines, one from each ruling. Now the two lines of the opposite ruling to the original three $(L_1,L_2,L_3)$ are the solution lines.

If this fourth line touches $Q_{123}$ similarly you get one solution line.

If this fourth line does not intersect $Q_{123}$ there are no lines meeting the four given lines.

See this geogebra project and this youtube video.

References

MECH 576

3264 and all that

  • bruh....how did this get so complicated – Clemens Bartholdy Jul 24 '22 at 01:25
  • Awesome! Thanks for this general solution in R3. It helped me to solve the puzzle at https://adventofcode.com/2023/day/24 analytically. Part 1 of the puzzle is quite easy, but in part 2 (which gets unlocked afterwards) a single straight trajectory intersecting 400 skewed trajectories needs to be found (which are of course specially crafted so that a solution exists). Your GeoGebra project helped me to better understand the structure of the problem and that 4 trajectories are sufficient to find the solution. – PiQuer Dec 27 '23 at 01:38
3

Note that lines are determined by 4 numbers (a point with three coordinates which can be nirmalized so that one coordinate is 0, and a direction with 2 coordinates up to scaling). Note that requiring a line to intersect another given line gives a single equation (e.g. for the x-axis, the requirement is that y=0 when z=0 or vice versa), cutting the number of free coordinates down by one. So to get down to a single line one would expect to require it to intersect four lines.

Brian Rushton
  • 13,375
2

To build up on Ted Shifrin's comment, assuming you have four skew lines $L_1$, $L_2$, $L_3$ and $L_4$, a costructive and an easy way to find a fifth line that intersects all the four lines would be as follows:

Construct two ruled surfaces (hyperboloids or hyperbolic paraboloids) uniquely defined by $L_1$, $L_2$, $L_3$ and $L_1$, $L_2$, $L_4$ that form the reguli of the quadrics $H_{123}$ and $H_{124}$. Find out the intersection between those two quadrics which definitely contains $L_1$ and $L_2$. Along with these two lines, the intersection must also contain two other lines (from the complementary reguli of the quadrics), say $M$ and $N$ that intersect all four given lines. Here's an example: $$L_1=(-3,-5,-2)+ t(1,0,0)$$ $$L_2=(-3,-5,2)+ t(0,1,0)$$ $$L_3=(3,5,0)+ t(0,0,1)$$ $$L_4=(3,10,0)+ t(1,0,1)$$ $$H_{123}=4\,xy-10\,xz+6\,yz-60$$ $$H_{124}=4\,xy-15\,xz+4\,yz+15\,{z}^{2}-10\,x+4\,y-25\,z-130$$ The intersection of these two quadrics yields 4 lines (I did it by finding the Gröbner basis of the ideal $\langle H_{123}, H_{124} \rangle$ using lexicographic ordering and factoring the first polynomial, there are other ways to do it and almost all computational environments have a Gröbner basis command if you want to check it out): $$L_1: z+2=y+5=0 $$ $$L_2: -z+2=x+3=0 $$ $$M: x+ \left( -\frac {1}{10}\,\sqrt {73}-{\frac {7}{10}} \right) y+\frac {1}{2}\,\sqrt {73}+ \frac {1}{2}= 5\,z\sqrt {73}+10\,\sqrt {73}+8\,y-55\,Z-70=0 $$ $$N: x+ \left( \frac {1}{10}\,\sqrt {73}-{\frac {7}{10}} \right) y-\frac {1}{2}\,\sqrt {73}+\frac {1}{2}= 5\,z\sqrt {73}+10\,\sqrt {73}-8\,y+55\,Z+70=0 $$ where $M$ and $N$ intersect $L_i, i=1,2,3,4$.

1

I occasionally need to actually write down the formulas for the lines in $\mathbb{R}^4$ which meet two given lines. Here is how I do it.

Make a projective change of coordinates so that $L_1$ is the $z$-axis and $L_2$ is the intersection of the plane at infinity with the $xy$ plane. Parametrize $L_3$ and $L_4$ as $(a z+ b, c z + d, z)$ and $(a' z'+ b', c' z' + d', z')$.

The line through $(a z+ b, c z + d, z)$ and $(a' z'+ b', c' z' + d', z')$ will meet $L_2$ if and only if $z=z'$, and will meet $L_1$ if and only if the vectors $\left[ \begin{smallmatrix} a z +b \\ c z + d \end{smallmatrix} \right]$ and $\left[ \begin{smallmatrix} a' z +b' \\ c' z + d' \end{smallmatrix} \right]$ are proportional. In other words, we want $$\det \begin{bmatrix} az+b & a' z + b' \\ cz+d & c' z+d' \\ \end{bmatrix} = 0.$$ This is a quadratic equation which can be solved for $z$.