26

Let $x,y$ be real numbers such that $$\left(\sqrt{y^{2} - x\,\,}\, - x\right)\left(\sqrt{x^{2} + y\,\,}\, - y\right)=y$$ Show that $x+y=0$.

My try: Let $$\sqrt{y^2-x}-x=a,\sqrt{x^2+y}-y=b\Longrightarrow ab=y$$ and then $$\begin{cases} y^2=a^2+(2a+1)x+x^2\cdots\cdots (1)\\ x^2=b^2+(2b-1)y+y^2\cdots\cdots \end{cases}$$ $(1)+(2)$ then $$x=-\dfrac{a^2+b^2+(2b-1)ab}{2a+1}\cdots\cdots (3)$$ so $$x+y=ab-\dfrac{a^2+b^2+(2b-1)ab}{2a+1}=\dfrac{(a-b)(2ab-a+b)}{2a+1}$$ we take $(3)$ in $(2)$,we have $$b^2+(2b-1)y+y^2-x^2=\dfrac{(2ab-a+b)(2a^3b+a^3+3a^2b-2ab^3+ab^2+4ab-b^3+b)}{(2a+1)^2}=0$$

so $$(2ab-a+b)=0$$ or $$2a^3b+a^3+3a^2b-2ab^3+ab^2+4ab-b^3+b=0$$ if $$2ab-a+b=0\Longrightarrow x+y=\dfrac{(a-b)(2ab-a+b)}{2a+1}=0$$ and if $$2a^3b+a^3+3a^2b-2ab^3+ab^2+4ab-b^3+b=0$$ I don't prove $$x+y=\dfrac{(a-b)(2ab-a+b)}{2a+1}=0?$$

math110
  • 93,304
  • We can't apply inequalities here, as WA shows in (http://www.wolframalpha.com/input/?i=a%3D%28sqrt%28y%5E2%2Bx%29%2Bx%29%28sqrt%28x%5E2%2By%29-y%29-y%2C+x%2B1%3Dy%3D1) together with (http://www.wolframalpha.com/input/?i=a%3D%28sqrt%28y%5E2%2Bx%29%2Bx%29%28sqrt%28x%5E2%2By%29-y%29-y%2C+x-1%3Dy%3D1). (The variable $a$ takes a negative value as well as a positive one, so we don't have $LHS\le RHS$ or $RHS\le LHS$ consistently) – chubakueno Oct 29 '13 at 02:55
  • Yes,But This is different problem,Thank you – math110 Oct 29 '13 at 04:37
  • Maybe it helps to call $f(x,y) = x+y$, leave all that mess in terms of $f(x,y), x$ and $y$, and check that $\partial f/\partial x = \partial f / \partial y = 0$, and see that $f(x,y) = 0 $ at least once. – Ivo Terek Jul 15 '14 at 04:15
  • Why isn't it correct to prove by the conjecture $y=-x$? – rae306 Aug 26 '14 at 18:29
  • "Proof by conjecture" is not a valid method of proof. More precisely, just stating that $x=y$ is not a correct solution because you are assuming the conclusion. You want to suppose that a given pair of numbers $x$ and $y$ satisfying the first equation, and deduce from this alone that $x+y=0$ is also satisfied. Even if you find a solution of the form $y=-x$, there's no guarantee that all solutions are of this form. This is exactly what the OP is asking to be proven. – vociferous_rutabaga Aug 26 '14 at 19:06
  • By conjecture, x+y=0, so x = -y. Substitute -y for x in the equation. Solve left side, y=y. – Paul Magnussen Aug 26 '14 at 17:50
  • Now posted to, and closed on, MO, https://mathoverflow.net/questions/288126/amazing-problem-if-sqrty2-x-x-sqrtx2y-y-y-then-xy-0 – Gerry Myerson Dec 10 '17 at 11:41

7 Answers7

3

Assume that $x<0$ and $y>0$, your statement can be written (even if we change $x$ to $-x$),

If $x,y$ are strictly positive such that $(\sqrt{y^2+x}+x)(\sqrt{x^2+y}-y)=y$ then $x=y$.

This equality becomes, $$\overbrace{\dfrac{x+\sqrt{x+y^2}}{y+\sqrt{y+x^2}}}^{A}=\overbrace{\dfrac{y}{x^2+y-y^2}}^{B}.$$ Let us start with $x>y$, clearly $B<1$. Moreover $x\mapsto(x-\sqrt{y+x^2})$ is increasing and $x\mapsto \sqrt{y+x^2}$ strictly increasing, we conclude that $x\mapsto(x+\sqrt{x+y^2})-(y+\sqrt{y+x^2})$ is strictly increasing and thus $A>0$ when $x>y$. We deduces that $A>1$ and therefore $A$ cannot be equal to $B$. The case $x<y$ can be treated in the same way.

Krokop
  • 1,124
3

Multiplication:

$y=x y-x \sqrt{x^2+y}-y \sqrt{-x+y^2}+\sqrt{x^2+y} \sqrt{-x+y^2} \Rightarrow$

$x \sqrt{x^2+y}+y \sqrt{-x+y^2}=\sqrt{x^2+y} \sqrt{-x+y^2}+xy-y$

squaring both sides:

$2 x y \sqrt{x^2+y} \sqrt{y^2-x}+x^4+x^2 y-x y^2+y^4= -x^3-x y+y^2-2 x y^2+2 x^2 y^2+y^3+(2 x y-2 y )\sqrt{x^2+y} \sqrt{-x+y^2} \Rightarrow $

simplifying:

$x^4+x^3-2 x^2 y^2+x^2 y+x y^2+x y+y^4-y^3-y^2 = -2 y \sqrt{x^2+y} \sqrt{y^2-x}$

Squaring again:

$y^8 - 2 y^7 - 4 x^2 y^6 + 2 x y^6 - y^6 + 6 x^2 y^5 + 2 y^5 + 6 x^4 y^4 - 2 x^3 y^4 + 3 x^2 y^4 - 4 x y^4 + y^4 - 6 x^4 y^3 - 4 x^3 y^3 - 2 x y^3 - 4 x^6 y^2 - 2 x^5 y^2 + x^4 y^2 + x^2 y^2 + 2 x^6 y + 4 x^5 y + 2 x^4 y + x^8 + 2 x^7 + x^6= -4 x^3 y^2+4 x^2 y^4-4 x y^3+4 y^5$

The only thing we have to check is how it factors as $(y+x) (p(x,y))$. Which gives us:

$(x+y)^2(x^6-2 x^5 y+2 x^5-x^4 y^2-2 x^4 y+x^4+4 x^3 y^3+2 x^3 y-x^2 y^4-4 x^2 y^3-4 x^2 y^2+2 x^2 y-2 x y^5+6 x y^4+2 x y^3+y^6-2 y^5-y^4-2 y^3+y^2)=0$

Buddha
  • 1,256
  • 11
  • 16
2

We need to prove that $x=y$, where $$\left(\sqrt{y^{2}+x}+x\right)\left(\sqrt{x^{2} + y}- y\right)=y$$ or

$$\left(\sqrt{y^{2}+x}+x\right)\left(\sqrt{x^{2} + y}- y\right)=\left(\sqrt{x^{2}+y}+x\right)\left(\sqrt{x^{2} + y}- x\right)$$ or $$\left(\sqrt{y^{2}+x}+x\right)\left(\sqrt{x^{2} + y}- y\right)-\left(\sqrt{x^{2}+y}+x\right)\left(\sqrt{x^{2} + y}- y\right)+$$ $$+\left(\sqrt{x^{2}+y}+x\right)\left(\sqrt{x^{2} + y}- y\right)-\left(\sqrt{x^{2}+y}+x\right)\left(\sqrt{x^{2} + y}- x\right)=0$$ or $$\left(\sqrt{y^{2}+x}-\sqrt{x^2+y}\right)\left(\sqrt{x^{2} + y}- y\right)+\left(\sqrt{x^{2}+y}+x\right)(x-y)=0,$$ which gives $x=y$ or

$$\frac{(1-x-y)\left(\sqrt{x^{2} + y}- y\right)}{\sqrt{y^{2}+x}+\sqrt{x^2+y}}+\sqrt{x^{2}+y}+x=0,$$ which is $$\sqrt{x^2+y}-y\sqrt{x^2+y}+\sqrt{(x^2+y)(y^2+x)}+x\sqrt{y^2+x}+x^2+xy+y^2=0.$$ Now, we'll consider four cases.

  1. $x\geq0$, $y\geq 0$.

Since $$-y\sqrt{x^2+y}+\sqrt{(x^2+y)(y^2+x)}=\sqrt{x^2+y}\left(\sqrt{y^2+x}-y\right)\geq0,$$ we obtain $x=y=0.$

  1. $x\geq0,$ $y\leq0.$

It's obvious that this case gives $x=y=0$ again.

  1. $x\leq0$, $y\geq0.$

Since, $$\sqrt{(x^2+y)(y^2+x)}+x\sqrt{y^2+x}=\sqrt{y^2+x}\left(\sqrt{x^2+y}+x\right)\geq0,$$ it's enough to prove that $$x^2+xy+y^2\geq(y-1)\sqrt{x^2+y},$$ which is obvious for $y\leq1.$

But for $y\geq1$ by AM-GM we obtain: $$(y-1)\sqrt{x^2+y}\leq\frac{1}{2}((y-1)^2+x^2+y)$$ and it's enough to prove that $$x^2+xy+y^2\geq\frac{1}{2}((y-1)^2+x^2+y)$$ or $$\require{cancel} \cancel{(x+y)^2+y^2+y-1\geq0.}\\ (x+y)^2+y-1\geq0.$$ We see that for $y\geq1$ the equality does not occur and in the case $y<1$ the equality occurs for $$x^2+xy+y^2=(y-1)\sqrt{x^2+y}=0,$$ which gives $x=y=0$ again.

  1. $x\leq0$ and $y\leq0.$

In this case it's enough to prove that $$xy+x\sqrt{y^2+x}\geq0$$ or $$x\left(y+\sqrt{y^2+x}\right)\geq0,$$ which is obvious.

The equality occurs for $x^2+y^2=0$ and we got $x=y=0$ again.

Done!

2

In the following proof we divide the $(x,y)$ plane into regions (see diagram) and show that each region can contain no solutions except on the line $x + y = 0$.

enter image description here

Let $$ F(x,y) = U(x,y)V(x,y) – y $$ where $$ U(x,y) = \sqrt{y^2 – x} \,\, – x \\V(x,y) = \sqrt{x^2 + y} \,\, – y $$

Then solutions satisfy $$ F(x,y) = 0 $$

Substituting $y=-x$ shows that $x+y=0$ is a solution for all x.

U is nonreal where $x>y^2$ (regions A and C in the diagram, bounded by red lines) and V is nonreal in the region $y<-x^2$ (regions B and C, also bounded by red lines). In these regions F is nonreal except possibly in region C, where U and V are both unreal. But there the condition for F to be real reduces to $x+y=0$, a subset of the known solution.

The following statements and deductions relate to the other regions of the $(x,y)$ plane, where $U$ and $V$ are real.

$U<0 \Leftrightarrow x>\tfrac{1}{2}(-1 + \sqrt{1 + 4y^2})$ (regions I, J).

$V<0 \Leftrightarrow y>\tfrac{1}{2}(1 + \sqrt{1 + 4x^2})$ (all regions except F, G).

$U_{x} < 0$ (regions D-K)

$V_{x} < 0 \Leftrightarrow x < 0$ (regions D, E, F)

$U_{xx} < 0$ (regions D-K)

$V_{xx} < 0 \Leftrightarrow y < 0$ (regions D, J, K)

where a subscript x denotes partial differentiation with respect to x.

On the diagram the lines on which $U=0$ and $V=0$ are coloured green and blue, respectively. It is easily shown that $F$ is nonzero on all the coloured lines (with sign as indicated) except at $(0,0)$ and $(1,-1)$. These lines delimit, but are excluded from, the regions A-K.

From the results above we can make the following deductions.

In region D: $$ U>0, V>0, U_{x}<0, V_{x}<0 \\F_{x} = UV_{x} + VU_{x} < 0 $$ This region is bounded on the right by the line $y<-x^2$, on which $F>0$. So $F>0$ throughout region D and it can contain no solutions.

In region E: $$ U>0, V>0, U_{x}<0, V_{x}<0 \\F_{x} < 0 $$ so here there can be no solutions other than those known to exist on the line $y=-x$.

In region F: $$ U>0, V<0, U_{x}<0, V_{x}<0, U_{xx}<0, V_{xx}>0 \\F_{xx} = UV_{xx} + VU_{xx} + 2U_{x}V_{x} > 0 $$ This region is bounded on the left by the line $V=0$ and on the right by the line $x = 0$, and on both these lines $F<0$. So the positive second derivative $F_{xx}$ means there can be no solutions $F=0$ in this region.

In a similar way, solutions can be ruled out for the following regions:

In region G, bounded on right by the line $V=0$ on which $F<0$: $$ U>0, V<0, U_{x}<0, V_{x}>0 \\F_{x}>0 $$

In region I, bounded on left by the line $U=0$ on which $F<0$: $$ U<0, V>0, U_{x}<0, V_{x}>0 \\F_{x} < 0 $$

In region J, containing a segment of the known solution line $x+y=0$ on which $F=0$: $$ U<0, V>0, U_{x}<0, V_{x}>0 \\F_{x} < 0 $$

In region K, bounded on the left and right by lines on which $F>0$: $$ U>0, V>0, U_{x}<0, V_{x}>0, U_{xx}<0, V_{xx}<0 \\F_{xx} = UV_{xx} + VU_{xx} + 2U_{x}V_{x} < 0 $$

Finally, in region H: $$ U>0, V>0, U_{x}<0, V_{x}>0 $$ and we note that $U_{x}<0$ in region G also, so for a given $y$, $U<U_{max}$, where $U_{max} = U(0,y) = y$

For the same value of y, $V<V_{max}$, where $V_{max} = V(X,y)$, and X is the value of x on the right-hand boundary of the region. On this boundary, $y=\sqrt{X^2+X}$, so $$ V_{max} = V(X,y) = \sqrt{X^2+y} \, – y < \sqrt{X^2+X+y} \, - y = \sqrt{y^2 + y} \,\, – y < \tfrac{1}{2}. $$

Therefore $$ F = UV – y < U_{max} V_{max} – y < y \tfrac{1}{2} – y = \, –\tfrac{1}{2} y < 0 $$ which completes the proof that there are no solutions other than $x+y = 0$.

MartinG
  • 978
2

Assuming continuity in the area interval $ (0 < x < 1 )$ and $ (0 > y > -1 ) $ would create problems as $x$ and $y$ are not always real in these areas.

Nay, union of inside parabola areas of $ y_1 = - x^2 $ and $ y_2 = \sqrt{x} $ would violate $ x + y = 0, $ which is only the common chord of intersection of $ y_1,y_2$. So the shown line joining $(0,0)$ to $(1,-1)$ does not exist as real. E.g., $(\frac12, -\frac12)$ does not lie on the common line.

enter image description here

Narasimham
  • 40,495
1

This is not a solution, but brute force can be used to remove the radicals. Let $A=y^2-x$ and $B=x^2+y$. We have

$$\sqrt{AB}-y\sqrt{A}-x\sqrt{B}+xy=y$$

Isolating $\sqrt{AB}$ and squaring both sides:

$$\sqrt{AB}=y\sqrt{A}+x\sqrt{B}+y(1-x)\quad(1)$$ $$AB=y^2A+x^2B+y^2(1-x)^2+2xy\sqrt{AB}+2y^2(1-x)\sqrt{A}+2xy(1-x)\sqrt{B}$$

(1) allows us to remove $\sqrt{AB}$. We do this and also recall what $A$ and $B$ equal.

$$(y^2-x)(x^2+y)=y^2(y^2-x)+x^2(x^2+y)+y^2(1-x)^2+2xy\left(y\sqrt{A}+x\sqrt{B}+y(1-x)\right)+2y^2(1-x)\sqrt{A}+2xy(1-x)\sqrt{B}$$

Group $\sqrt{A}$ and $\sqrt{B}$ terms, then rearrange a bit:

$$(y^2-x)(x^2+y)=y^2(y^2-x)+x^2(x^2+y)+y^2(1-x)^2+2xy^2(1-x)+2y^2\sqrt{A}+2xy\sqrt{B}$$ $$x^2y^2+y^3-x^3-xy=y^4-xy^2+x^4+x^2y+y^2-2xy^2+x^2y^2+2xy^2-2x^2y^2+2y\left(y\sqrt{A}+x\sqrt{B}\right)$$ $$y^3-x^3-xy=y^4-xy^2+x^4+x^2y+y^2-2x^2y^2+2y\left(y\sqrt{A}+x\sqrt{B}\right)$$

(1) allows us to sub out the quantity in parentheses:

$$y^3-x^3-xy=y^4-xy^2+x^4+x^2y+y^2-2x^2y^2+2y\left(y(x-1)+\sqrt{AB}\right)$$ $$y^3-x^3-xy=y^4-xy^2+x^4+x^2y+y^2-2x^2y^2+2y^2(x-1)+2y\sqrt{AB}$$ $$y^3-x^3-xy=y^4+xy^2+x^4+x^2y-y^2-2x^2y^2+2y\sqrt{AB}$$ $$y^3-x^3-xy-y^4-xy^2-x^4-x^2y+y^2+2x^2y^2=2y\sqrt{AB}$$

Squaring both sides, we've reached a goal of no longer having radicals.

$$(y^3-x^3-xy-y^4-xy^2-x^4-x^2y+y^2+2x^2y^2)^2=4y^2(y^2-x)(x^2+y)$$

I had a CAS expand this, move it all to one side, and then, as expected, $(x+y)$ factors out of it (twice).

$$(x+y)^2 p(x,y)=0$$

where $$p(x,y)=x^6-2 x^5 y+2 x^5-x^4 y^2-2 x^4 y+x^4+4 x^3 y^3+2 x^3 y-x^2 y^4-4 x^2 y^3-4 x^2 y^2+2 x^2 y-2 x y^5+6 x y^4+2 x y^3+y^6-2 y^5-y^4-2 y^3+y^2$$

is a monster. It would be sufficient to show that $p(x,y)$ is never $0$ in the region of the plane where both $\sqrt{A}$ and $\sqrt{B}$ are defined aside from points along $x+y=0$ (like $(0,0)$). This is a pretty messy polynomial, but at least it's a polynomial.


EDIT: This approach seems to be useless; a CAS plot of the zero set of $p$ has several components, all of which are in the region where $\sqrt{A}$ and $\sqrt{B}$ are defined. They must be extraneous solutions from the squaring that was done twice.

2'5 9'2
  • 54,717
0

A more generalised approach over my earlier post. This is not intended to be an exhaustive proof but an experimental one. Constructive comments are most welcome.

Let

$$\sqrt{y^2-x}-x=Ay^n\qquad \cdots (1)\\ \sqrt{x^2+y}-y=\frac {y^{1-n}}A \qquad \cdots (2)\\$$ such that the original equation $$\left(\sqrt{y^2-x}-x\right)\left(\sqrt{x^2+y}-y\right)=y$$ is satisfied as required.

From $(1)$,

$$\begin{align} \sqrt{y^2-x}&=x+Ay^n\\ y^2-x&=x^2++2Axy^n+A^2y^{2n}\\ y^{2n}A^2+2xy^nA+(x^2-y^2+x)&=0\\ A^2+\frac {2x}{y^n}A+\frac{(x^2-y^2+x)}{y^{2n}}&=0\qquad \qquad \qquad \qquad \cdots (3) \end{align}$$

From $(2)$,

$$\begin{align} \sqrt{x^2+y}&=y+\frac {y^{1-n}}A\\ A\sqrt{x^2+y}&=Ay+y^{1-n}\\ A^2(x^2+y)&=A^2y^2+2Ay^{2-n}+y^{2(1-n)}\\ (x^2-y^2+y)A^2-2y^{2-n}A-y^{2(1-n)}&=0\\ A^2-\frac{2y^{2-n}}{x^2-y^2+y}A-\frac{y^{2(1-n)}}{x^2-y^2+y}&=0\qquad \qquad \qquad \cdots (4) \end{align}$$

Equating coefficients of $A^1$: $$\begin{align}\frac{2x}{y^n}&=-\frac{2y^{2-n}}{x^2-y^2+y}\\ y^2&=-x(x^2-y^2+y)\qquad \qquad \qquad \qquad \qquad \qquad \cdots (5)\end{align}$$

Equating coefficients of $A^0$:

$$\begin{align}\frac{x^2-y^2+x}{y^{2n}}&=-\frac{y^{2(1-n)}}{x^2-y^2+y}\\ y^2&=-(x^2-y^2+x)(x^2-y^2+y)\qquad \cdots (6)\end{align}$$

(5)=(6): $$\begin{align}x(x^2-y^2+y)&=(x^2-y^2+x)(x^2-y^2+y)\\ (x^2-y^2)(x^2+y^2-y)&=0\\ (x-y)(x+y)(x^2-y^2+y)&=0\\ \Rightarrow x-y=&0, x+y=0, x^2-y^2+y=0\end{align}$$

Checking by substitution into the original equation shows that only $$x+y=0$$ is valid.

This graph created on desmos.com might help illustrate the approach:

https://www.desmos.com/calculator/qrlbgbalix

  • It's an interesting approach, but I don't buy it. Equations (3) and (4) show us that $A$ is a solution to the two indicated quadratic equations. It does not follow that they must be the same equation! –  Aug 23 '14 at 06:12
  • @Bungo Thanks for your comments. $A$ is actually a parameter in this case, and defines a series of curves for each equation. However at the same value of $A$ both equations intersect at a point which satisfies the original equation. The locus of this point is the solution required. Setting both quadratic equations to be the same is essentially assigning the same value of $A$ for both equations. – Hypergeometricx Aug 23 '14 at 09:10
  • I agree that this is valid until you square the equations. But after squaring, each equation has a second root in addition to $A$. How do we know that the second root in (3) is the same as the second root in (4)? Sorry if I'm being dense. To take a simpler example: if $x = 1$ then for any $c$ I can express this as $x-c = 1-c$. Squaring both sides, I get $x^2 - 2xc + c^2 = 1 - 2c + c^2$, or $x^2 - 2xc - (1 - 2c) = 0$. If I carry out this procedure with two different values of $c$, I get two valid equations with $1$ as a root, but the coefficients are not the same. –  Aug 23 '14 at 16:40
  • That's true. By equating coefficients we arrive at one particular solution, but there may be others, e.g. where the two quadratics are not the same but have one common root...would it be correct to say that? Any suggestions on other possible approaches? – Hypergeometricx Aug 23 '14 at 16:58
  • Yes, that's what I was getting at. If two quadratic equations have BOTH roots in common then their coefficients are the same, up to a common factor. (Actually that's one more issue with your approach: $x^2 + 2x + 1 = 0$ and $2x^2 + 4x + 2 = 0$ have the same roots but the coefficients are not equal.) Sorry, I don't have any suggestions at the moment - the other answer is pretty thorny and I haven't read through it yet. :-) –  Aug 23 '14 at 17:03
  • P.S. We can see that $y=-x$ is a solution simply by plugging it into the original equation. The hard part is proving that there are no other solutions. –  Aug 23 '14 at 17:09