3

Given a multivariate polynomial equation like $$xy - bx - ay - ab = 0$$ how do I simplify it, make sense of it, or characterize it?

My attempts so far are below.


Attempt 1

Viewing $x,y$ as variables and $a,b$ as constants, it's obviously of degree 2. Thus, I won't be able to find any linear solutions in $x$ or $y$. But I should be able to somehow make some sense of it, more than in the current form.

One approach I've tried is finding roots. As a quadratic, it should be expressable as a product of monomials. I've looked up on this site and elsewhere, and there seems to be no standard method to do that for multivariate polynomials, but we can treat $x,y$ individually (viewing the other as a constant), giving something like $$x = \frac {a(y-b)}{y-b} \\ = a$$ for $y \neq b$. But this same argument also gives $y = b$ (!), violating our condition.

A form like $$(x-p)(x-r)(y-q)(y-s) = 0$$ would be ideal. The above suggests it would simply be $$(x - a)(y - b) = 0$$ which seems to meet the conditions: $[x = a \land y \neq b] \lor [x \neq a \land y = b]$.

Multiplying out, we get $$(x - a)(y - b) = xy - bx - ay + ab$$ and so we've (in some ways) solved the problem.

Still remaining question: What is the locus of $x,y$ that meet that equation?


The above seems to have found a simplification, but through a confused and roundabout path (and still incomplete). Is there a more direct or disciplined approach? Or a way to generalize this? When I see a polynomial in one variable, I immediately know what to do (for some definition of do). What about multivariate?

How can I "solve" these types of equations? I request that answers don't just solve this question, but show me the method (or point me to a good reference). Which courses is this normally covered in?

(Note: I've searched math.SE and other sites and found posts like this, this, this, and this none of which helped me.)


Attempt 2

As pointed out in the comments, there's an error in the above. Rather, it should be: $$ x = a \frac {y+b}{y-b}, \quad y \neq b \\ y = b \frac {x+a}{x-a}, \quad x \neq a $$ so, if I simplified correctly, $$ x = a \left ( \frac {1 + \frac{x+a}{x-a}} {-1 + \frac{x+a}{x-a}} \right ) $$ which leaves me just as confused! What next?


Update

Let's start a fresh:

$$xy−bx−ay−ab=0$$

Trying to isolate and solve for $x$ or $y$ is a mess, as the above shows. Lesson: We can't always solve for $x$! Instead, we can notice that:

  1. This is a second degree polynomial in two variables
  2. and therefore either a conic or degenerate conic
  3. which are often best characterized not by $x=...$ but by $(\frac x p - h)^2...$ or something similar.

How do we do that? Let's try to factor our polynomial (or at least the higher degree terms of it) using an analog of completing the square: To get an $xy$ term, our solution will need to look like $$(x - p)(y - q) = r.$$ Then we must have $q =b$, because, when multiplied out, the only coefficient for $x^1$ is $q$. Likewise, $p = a$, giving $$(x-a)(y-b) = r.$$

Finally, since $$(x-a)(y-b) = xy -bx -ay + ab$$ then $r = 2ab$, giving $$(x-a)(y-b) = 2ab.$$ From knowledge of conics, this is a hyperbola with center $(a, b)$ and "scale factor" (I don't know the proper term) $2ab$. QED.

SRobertJames
  • 4,278
  • 1
  • 11
  • 27
  • The basic technique here seems to be exactly what's used throughout multivariate mathematics: Analyze each variable individually, treating the others as constants. This makes it clear that the locus is two perpendicular lines! – SRobertJames Jun 19 '23 at 19:07
  • 1
    you have locus $xy - bx - ay - ab = 0$
    and calculation: $(x - a)(y - b) = xy - bx - ay + ab$ . Look carefully at the $\pm $ signs. Also, please write out the $x,y$ version I recommended on your previous question.
    – Will Jagy Jun 19 '23 at 19:19
  • The other thing useful for anyone is to draw graphs, and get good at sketching graphs by hand. For this problem: compare, what is the locus of $xy=0$ but then what is the locus of $xy=1?$ – Will Jagy Jun 19 '23 at 19:44
  • @WillJagy From $(x-a)(y-b)=0$, I conclude that either $x = a$ or $y = b$. These both fulfill the original equation. Can you explain what direction beyond that you're pointing me in? Re your individual points: 1. I've checked the $\pm$ signs and they seem correct. Nor do they seem to show any pattern of interest 2. I'm not sure change you recommend: I've used scalars here, no vectors, and there's no $x^2$ to complete the square with 3. $xy = k$ is a hyperbola, but it's not clear how that helps me here (degenerate hyperbola is two intersecting lines?). Can you please elaborate? – SRobertJames Jun 19 '23 at 19:52
  • $xy-bx-ay - ab = (x-a)(y-b) - 2ab$ so the product is not zero – Will Jagy Jun 19 '23 at 19:54
  • 2
    Don't "solve" for anything here.... Part 1: what is the locus $xy=1?$ Part 1 and a half: what is the locus $(x-a)(y-b) = 1?$ Part 2: what is the locus $(x-a)(y-b) = 2ab?$ – Will Jagy Jun 19 '23 at 20:31
  • Let me put in an advertisement for pure translations. If I first give you $x^2 + y^2 = 25$ you tell me it is a circle of radius $5$ around $(0,0)$ If I then, as in the other problem, give you $(x-3)^2 + (y-4)^2 = 25,$ you tell mi it is a circle of radius $5$ around center $(3,4)$ Similar: first I give you $xy=1.$ This is a hyperbola, one arc in the first quadrant and one in the third. It does have a "center," it is symmetric (in more than one way) around the origin. Next: what is the locus $(x-3)(y-4)= 1?$ Around what point is it symmetric? – Will Jagy Jun 19 '23 at 20:42
  • @WillJagy Can you take a look at the update? In particular, did I extract the general method properly: If you can't solve for $x$, factor the polynomial, or at least some of it, using a method similar to completing the square. – SRobertJames Jun 19 '23 at 20:55
  • well, you are right that we are discussing conic sections as long as thre are constants, take capital letters $A,B,C,D,E,F$ and the locus of this set of $x,y$ points: $Ax^2 + Bxy +Cy^2 + Dx + Ey + F=0$ When $B=0$ you can find this as a translation by completing squares, $A(x-a)^2 + C(y-b)^2 = W$ for some new constant $W.$ Instead, if $B$ is nonzero but $A=C=0$ you have a translated hyperbola, as in the current problem. If all of $A,B,C \neq 0$ we also need rotations. Off to get groceries – Will Jagy Jun 19 '23 at 21:04
  • anyway, if you know enough about the locus of $g(x,y) = C,$ then the locus of $g(x-a,y-b)= C $ is just the original translated – Will Jagy Jun 19 '23 at 21:12
  • 1
    there seems to be no standard method to do that for multivariate polynomials --- True, but the school algebra method factoring by grouping is the obvious method to try in this case. See also the answers and comments to How to simpify this? – Dave L. Renfro Jun 19 '23 at 21:54

0 Answers0