Let $A = \{(x, y)\in \mathbb{R}^2 | x^2-xy+2y^2<1 \}$; define $f: \mathbb{R}^2 \to \mathbb{R}$ by $f(x, y) = xy$.

Express the integral

$$\int_A f$$

as an integral over the unit ball $B = \{(x, y)\in \mathbb{R}^2 | x^2+y^2<1 \}$.

Someone gave a solution here but it seems to be wrong.

My attempt:

Observe that $x^2-xy+2y^2=(x-\frac{1}{2}y)^2+(\frac{\sqrt{7}}{2}y)^2$.

Let $B^\ast = \{(r, \theta) | 0< r < 1, 0 < \theta < 2\pi\}$.

Let $A^\ast = \{(x, y)\in \mathbb{R}^2 | x^2-xy+2y^2<1$, and $x<0$ if $y=0\}$.

Then the function

$$g(r, \theta) = (r\cos\theta+\frac{1}{\sqrt{7}}r\sin\theta, \frac{2}{\sqrt{7}}r\sin\theta)$$

carries $B^\ast$ in a one-to-one fasion onto $A-\{(0, 0)\}$.

Now we obtain that

\begin{align*} \text{det} Dg & = \begin{vmatrix}\cos\theta+\frac{1}{\sqrt{7}}\sin\theta & -r\sin\theta+\frac{1}{\sqrt{7}}r\cos\theta \\ \frac{2}{\sqrt{7}}\sin\theta & \frac{2}{\sqrt{7}}r\cos\theta \\ \end{vmatrix} \\ & = \frac{2}{\sqrt{7}}r \\ & > 0 \end{align*}

if $(r, \theta)\in B^\ast$.

Since the non-negative $x$-axis has measure zero, using the formula of change of variables, we have that

\begin{align*} \int_A f & = \int_{A^\ast} f \\ & = \int_{B^\ast} f \circ g \cdot |\text{det} Dg| \\ & = \int_0^{2\pi} \int_0^1 \Big((r\cos\theta+\frac{1}{\sqrt{7}}r\sin\theta)\frac{2}{\sqrt{7}}r\sin\theta\frac{2}{\sqrt{7}}r \Big) drd\theta \\ & = \frac{2}{7}\int_0^{2\pi} \int_0^1 (2\cos\theta\sin\theta+\frac{2}{\sqrt{7}}\sin^2 \theta)r^3 drd\theta \\ & = \frac{2}{7}\int_0^{2\pi} \int_0^1 \Big( \sin2\theta+\frac{1}{\sqrt{7}}(1-\cos2\theta) \Big)r^3 drd\theta \\ & = \frac{2}{7}\int_0^{2\pi} \int_0^1 \Big( \sin2\theta-\frac{1}{\sqrt{7}}\cos2\theta+\frac{1}{\sqrt{7}} \Big)r^3 drd\theta \\ & = \frac{2}{7}\int_0^{2\pi} \Big[ \frac{1}{4}\Big( \sin2\theta-\frac{1}{\sqrt{7}}\cos2\theta+\frac{1}{\sqrt{7}} \Big)r^4 \Big]_{r=0}^{r=1} d\theta \\ & = \frac{2}{7}\int_0^{2\pi} \frac{1}{4}\Big( \sin2\theta-\frac{1}{\sqrt{7}}\cos2\theta+\frac{1}{\sqrt{7}} \Big) d\theta \\ & = \frac{1}{14} \Big[ -\frac{1}{2}\cos2\theta-\frac{1}{2\sqrt{7}}\sin2\theta+\frac{1}{\sqrt{7}}\theta \Big]_{\theta=0}^{\theta=2\pi} \\ & = \frac{\pi}{7\sqrt{7}} \end{align*}

Am I making any mistake? Any help or suggestion is appreciated.

  • $\begingroup$ I edited for proper use of \mid and \det. When you write 3\text{det} A you see $3\text{det} A$, without proper spacing. With 3\det A you see $3\det A$. The spacing is context-dependent; e.g. in $3\det(A)$ it doesn't put as much space on the right. ${}\qquad{}$ $\endgroup$ – Michael Hardy Jan 8 '16 at 20:01
  • $\begingroup$ @MichaelHardy Thanks :-) $\endgroup$ – Liebster Jugendtraum Jan 8 '16 at 20:03
  • $\begingroup$ There appears to be an arithmetic error in the original solution. $\endgroup$ – amd Jan 8 '16 at 21:16
  • $\begingroup$ @amd Did you mean the one given by Ron Gordon? I think what is incorrect in his solution is the statement "$A$ is an ellipse rotated at an angle of $\frac{\pi}{8}$". By the way, is there any mistake in my computation? $\endgroup$ – Liebster Jugendtraum Jan 8 '16 at 21:24
  • 1
    $\begingroup$ @LiebsterJugendtraum, no the rotation of $\frac\pi 8$ is correct, but the values he has for $a$ and $b$ aren’t. I’m not going to check your calculations in detail, but the determinant of the Jacobian looks right. Note that the area of the original ellipse is $\frac{\sqrt7}2\pi$, so you’d expect to see a factor of $\frac2{\sqrt7}$ when mapping to the unit disk. $\endgroup$ – amd Jan 8 '16 at 21:46

Looks reasonable to me. As a check on your computations, here’s a solution using an approach similar to Ron Gordon’s from the original question.

Examine the matrix $Q$ associated with the quadratic form $x^2-xy+2y^2$: $$ Q = \pmatrix{1&-\frac12\\-\frac12&2}. $$ $\det Q>0$ and $\operatorname{tr}Q>0$, so the eigenvalues are positive, which means that we have an ellipse centered on the origin and rotated through some angle $\phi$. Define the pullback $\beta:(\rho,\theta)\mapsto(x,y)$ by $$\begin{align} \beta^*x &= \rho\,(a\cos\theta\cos\phi-b\sin\theta\sin\phi) \\ \beta^*y &= \rho\,(a\cos\theta\sin\phi+b\sin\theta\cos\phi), \end{align}$$ where $a$ and $b$ are the half-axis lengths of the ellipse. (This is the polar equation of the ellipse multiplied by $\rho$, which can easily be derived by scaling and rotating the unit circle.) For this pullback, we also have $$\det{\partial(\beta^*x,\beta^*y)\over\partial(\rho,\theta)}=ab\rho.$$ So, if $C$ is the unit disk, $$\begin{align} \int_Af\,dA &= \int_C\beta^*(f\,dA) \\ &=ab\int_0^{2\pi}\int_0^1\rho^3(a\cos\theta\cos\phi-b\sin\theta\sin\phi)(a\cos\theta\sin\phi+b\sin\theta\cos\phi)\,d\rho\,d\theta \\ &=\frac{ab}4\int_0^{2\pi}(a\cos\theta\cos\phi-b\sin\theta\sin\phi)(a\cos\theta\sin\phi+b\sin\theta\cos\phi)\,d\theta \\ &=\frac{ab}4\int_0^{2\pi}a^2\cos^2\theta\cos\phi\sin\phi - b^2\sin^2\theta\cos\phi\sin\phi + \cdots\,d\theta \\ &=\frac{\pi ab}4(a^2-b^2)\cos\phi\sin\phi.\tag{*} \end{align}$$ Note that we can ignore terms that involve odd powers of $\cos\theta$ and $\sin\theta$ since we’re integrating from $0$ to $2\pi$.

To determine $a$, $b$ and $\phi$, go back to the matrix $Q$. If its eigenvalues are $\lambda_1 < \lambda_2$, then $a^2 = 1/\lambda_1$ and $b^2 = 1/\lambda_2$. It’s a bit more difficult to find an exact value for $\phi$ in general, but we don’t really need to since we can get its sine and cosine directly from the normalized eigenvectors (which is why I didn’t combine $\cos\phi\sin\phi$ into $\frac12\sin{2\phi}$). In this case, we have $$ a^2 = {2\over3-\sqrt2}, b^2 = {2\over3+\sqrt2} \\ \cos\phi = {\sqrt{2+\sqrt2}\over2}, \sin\phi = {\sqrt{2-\sqrt2}\over2}. $$ Plugging these values into (*) and simplifying gives ${\pi\over7\sqrt7}$, which agrees with the value computed in the question above.

  • $\begingroup$ Thank you! There is only one question remaining. So we can ignore $ab(\cos^2 \phi-\sin^2 \phi)\cos\theta \sin\theta$ because $\cos\theta \sin\theta$ is an odd function and hence $\int_0^{2\pi}\cos\theta \sin\theta=0$. Is that correct? :-) $\endgroup$ – Liebster Jugendtraum Jan 9 '16 at 15:27
  • 1
    $\begingroup$ @LiebsterJugendtraum Not so much because it’s odd (though we could use periodicity to shift the interval of integration) but because of the symmetry about the $x$-axis over an integral number of cycles. You could certainly carry that term along for a few more steps, but I already know that its integral is going to be $0$, so I prefer to reduce the clutter. $\endgroup$ – amd Jan 9 '16 at 21:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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