Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

If I have two lines $$ \eqalign{ & L_1 \left( t \right):p_1 + td_1 \cr & L_2 \left( q \right):p_2 + qd_2 \cr} $$ living in $\mathbb{R}^n$, there exists a classical formula to find the distance between them involving dot and cross products. The question is: can I deduce that formula only using calculus? (In this case, 2 variables) i.e., find the values such that the function $$ f\left( {t,q} \right) = \left \| L_1 (t) - L_2(t) \right \| = \left \| p_1 + td_1 - p_2 - qd_2 \right\| $$ reaches its minimum value.

Oh sorry; for simplicity, to have the natural cross product, just take $\mathbb{R}^3$.

share|cite|improve this question

Consider the distance squared between two points on these lines are $d_2(t_1, t_2) = \vert\vert p_1 - p_2 + t_1 q_1 - t_2 q_2 \vert\vert^2$.

Simple algebra shows, here $(u,v)$ denotes dot product of two vectors:

$$ d_2(t_1,t_2) = \vert p_{12} \vert^2 + 2 t_1 (p_{12}, q_1)- 2 t_2 (p_{12},q_2) + t_1^2 \vert q_1 \vert^2 + t_2^2 \vert q_2 \vert^2 - 2 t_1 t_2 (q_1,q_2) $$

You now minimize it by requiring derivatives with respect to $t_1$ and $t_2$ to vanish. This yields

$$ t_1 = \frac{ (p_{12},q_1) \vert q_2\vert^2 - (p_{12},q_2) (q_1, q_2) }{(q_1, q_2)^2 - \vert q_1 \vert^2 \vert q_2 \vert^2} \; \;\; \text{and} \;\;\; t_2 = - \frac{ (p_{12},q_2) \vert q_1\vert^2 - (p_{12},q_1) (q_1, q_2) }{(q_1, q_2)^2 - \vert q_1 \vert^2 \vert q_2 \vert^2} $$

Upon substitution I obtain the minimal distance squared is

$$ \vert p_{12} \vert^2 + \frac{ \vert q_1\vert^2 (p_{12},q_2)^2 + \vert q_2\vert^2 (p_{12},q_1)^2 - 2 (q_1,q_2)(p_{12},q_1)(p_{12},q_2) }{ (q_1, q_2)^2 - \vert q_1 \vert^2 \vert q_2 \vert^2} $$

where $p_{12}=p_1-p_2$.

share|cite|improve this answer
What did you do? I did not understand you , sorry. – Daniel Aug 9 '11 at 20:11
@Daniel I constructed the function of distance squared between two points on two curve, i.e. point $p_1 + t_1 q_1$ and point $p_2 + t_2 q_2$. This gives bivariate polynomial in $t_1$ and $t_2$ with coefficients expressible as scalar products of appropriate vectors. Then I solved for minimal parameters, obtained from $\frac{d}{d{t_1}} d_2(t_1,t_2) = 0$ and $\frac{d}{d{t_2}} d_2(t_1,t_2) = 0$ and substituted them back. – Sasha Aug 9 '11 at 20:14
@Daniel I have updated my response filling in steps. Hope this helps. – Sasha Aug 9 '11 at 21:00
I assume that $(u,v)$ denotes the scalar (dot) product $u\cdot v$. – Grumpy Parsnip Aug 9 '11 at 21:11
By $d_2$, I assume you mean $d^2$. – robjohn Aug 17 '11 at 10:37

In the past, I have solved these types of problems without calculus just by requiring the minimizing line to be orthogonal to whatever it was supposed to minimize the distance between.

Since I am lazy, I will rewrite the equations as $ L_1 ( p):a + p b $ and $ L_2 ( q):c + q d $. We want to find values of $p$ and $q$ such that, if $P = L_1(p)$ and $Q = L_2(q)$, then $P-Q$ is orthogonal to both $b$ and $d$.

Using "$.$" for dot product (I don't see where cross product is needed), this becomes $$0 = ( a + p b - c - q d).b = ( a + p b - c - q d).d$$ or, letting $g = c-a$, $$\eqalign{p |b|^2 - q(d.b) = g.b \cr p(b.d)-q |d|^2 = g.d\cr}$$

Solving these for $p$ and $q$ and substituting back in $P$ and $Q$ should give the points (modulo and errors on my part). Note that, if $b$ and $d$ are not proportional, Cauchy-Schwarz shows this equation has a unique solution.

share|cite|improve this answer
"The Schwarz of inequality / And lemma too, he has no T. // The "Distribution" Schwartz you see / Is French, and so he has a T. -- Ralph Boas, Spelling Lesson. :D – J. M. Aug 17 '11 at 4:27
@J.M. that reminds me... :) – t.b. Aug 21 '11 at 9:38
...apparently you went on a Schwarz correction spree @Theo... :D – J. M. Aug 21 '11 at 10:53
@J.M. Yes I did :) (I only see your comment now that I'm back at it. Note that the system only removes one punctuation character after the notification, so a trailing ellipsis undermines notification.) Is it sooo hard to spell names correctly? I'll never understand how mathematicians can see $\ddot{u} = u$ as a mildly interesting equation (as opposed to $u = u$) while they fail to see the difference between Holder and Hölder, Schwarz and Schwartz, etc. – t.b. Aug 26 '11 at 23:33
@Theo: I could understand the "Holder" bit at least; not everybody a.) has an international keyboard; b.) can remember the ASCII code for those fancy letters; or c.) has the patience to bring up a character map application or search for that stuff in his/her favorite search engine. Oh well. We'll never run out of work... ;) – J. M. Aug 28 '11 at 20:29

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.