$$f(a,b)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt{x^2+a^2}\sqrt{x^2+b^2}}$$

To use Landen's transformation

$$f(a,b)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt{x^2+(\frac{a+b}{2})^2}\sqrt{x^2+ab}}$$

$$f(a,b)=f\left(\frac{a+b}{2},\sqrt{ab}\right)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt{x^2+c^2}\sqrt{x^2+c^2}}=\int_0^{+\infty} \frac{\;\mathrm dx}{x^2+c^2}=\frac{\pi}{2c}$$


$\operatorname{AGM}$ is Arithmetic–geometric mean

I would like to find similar transform for:


$$g(a,b,c)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+a^3}\sqrt[3]{x^3+b^3}\sqrt[3]{x^3+c^3}}$$

Do you know similar kind of transform method for that integral?

Or which other methods can be used to evaluate that improper integral?

Thanks a lot for answers and helps.


I would like to share my results. Maybe someone can give some ideas to go forward.

$$g(a,b,c)=\frac{1}{F(a,b,c)}=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+a^3}\sqrt[3]{x^3+b^3}\sqrt[3]{x^3+c^3}}$$ $$\frac{1}{F(a.k,b.k,c.k)}=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+a^3k^3}\sqrt[3]{x^3+b^3k^3}\sqrt[3]{x^3+c^3k^3}}$$

$$\frac{1}{F(a.k,b.k,c.k)}=\int_0^{+\infty} \frac{\;\mathrm dx}{k^3\sqrt[3]{\frac{x^3}{k^3}+a^3}\sqrt[3]{\frac{x^3}{k^3}+b^3}\sqrt[3]{\frac{x^3}{k^3}+c^3}}$$ $ku=x$ $$\frac{1}{F(a.k,b.k,c.k)}=\int_0^{+\infty} \frac{\;k du}{k^3\sqrt[3]{u^3+a^3}\sqrt[3]{u^3+b^3}\sqrt[3]{u^3+c^3}}$$








$$\frac{1}{F(a,at,az)}=\frac{1}{a^2F(1,t,z)}=\frac{1}{a^2H(t,z)}=\frac{1}{a^2}\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+1}\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$

Now the problem is for 2 variables. $$\frac{1}{H(t,z)}=V(t,z)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+1}\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$

We can get a partial differential equation from here

$$\frac{\partial V(t,z)}{\partial t}=-\int_0^{+\infty} \frac{t^2\;\mathrm dx}{\sqrt[3]{x^3+1}(x^3+t^3)\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$

$$\frac{\partial V(t,z)}{\partial z}=-\int_0^{+\infty} \frac{z^2\;\mathrm dx}{\sqrt[3]{x^3+1}(x^3+z^3)\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$

$$\frac{\partial^2 V(t,z)}{\partial t \partial z }=\int_0^{+\infty} \frac{t^2z^2\;\mathrm dx}{\sqrt[3]{x^3+1}(x^3+t^3)(x^3+z^3)\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}=$$

$$=\frac{1 }{z^3-t^3}\int_0^{+\infty} \frac{t^2z^2\;\mathrm dx}{\sqrt[3]{x^3+1}(x^3+t^3)\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}+\frac{1 }{t^3-z^3}\int_0^{+\infty} \frac{t^2z^2\; \mathrm dx}{\sqrt[3]{x^3+1}(x^3+z^3)\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$

$$(t^3 -z^3 )\frac{\partial^2 V(t,z)}{\partial t \partial z }=z^2\frac{\partial V(t,z)}{\partial t }- t^2\frac{\partial V(t,z)}{\partial z }$$

Now I am looking for a transform that

$t=k(s,y)$ $z=l(s,y)$ To get same equation

$$(s^3 -y^3 )\frac{\partial^2 V(s,y)}{\partial s \partial y }=y^2\frac{\partial V(s,y)}{\partial s }- s^2\frac{\partial V(s,y)}{\partial y }$$

UPDATE 2: I tried to use a variable change .

$$\frac{1}{H(t,z)}=V(t,z)=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+1}\sqrt[3]{x^3+t^3}\sqrt[3]{x^3+z^3}}$$



$$\frac{1}{H(t,z)}=V(t,z)=\int_0^{1} \frac{\;du}{(1-u^3)\sqrt[3]{\frac{u^3}{1-u^3}+t^3}\sqrt[3]{\frac{u^3}{1-u^3}+z^3}}=$$

$$=\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{t^3+(1-t^3)u^3}\sqrt[3]{z^3+(1-z^3)u^3}}$$

$$=\frac{1}{\sqrt[3]{(1-t^3)}\sqrt[3]{(1-z^3)}}\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{(\cfrac{t}{\sqrt[3]{(1-t^3)}})^3+u^3}\sqrt[3]{(\cfrac{z}{\sqrt[3]{(1-z^3)}})^3+u^3}}$$

If we combine the results above, we can get

$$g(a,b,c)=\frac{1}{F(a,b,c)}=\int_0^{+\infty} \frac{\;\mathrm dx}{\sqrt[3]{x^3+a^3}\sqrt[3]{x^3+b^3}\sqrt[3]{x^3+c^3}}$$



$$g(a,b,c)=\frac{1}{F(a,b,c)}=\int_0^{1} \frac{\;du}{(1-u^3)\sqrt[3]{\frac{a^3u^3}{1-u^3}+b^3}\sqrt[3]{\frac{a^3u^3}{1-u^3}+c^3}}=\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{b^3-(b^3-a^3)u^3}\sqrt[3]{c^3-(c^3-a^3)u^3}}$$

$$g(a,b,c)=\frac{1}{F(a,b,c)}=\frac{1}{bc}\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{1-(1-(a/b)^3)u^3}\sqrt[3]{1-(1-(a/c)^3)u^3}}$$

$$\frac{1}{T(x,y)}=\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{1-x^3u^3}\sqrt[3]{1-y^3u^3}}$$

$$F(a,b,c)=bc .T(\sqrt[3]{1-(a/b)^3},\sqrt[3]{1-(a/c)^3})=ac.T(\sqrt[3]{1-(b/a)^3},\sqrt[3]{1-(b/c)^3})=ab.T(\sqrt[3]{1-(c/a)^3},\sqrt[3]{1-(c/b)^3})$$

It seems If we solve $\int_0^{1} \frac{\;du}{\sqrt[3]{1-u^3}\sqrt[3]{1-x^3u^3}\sqrt[3]{1-y^3u^3}}$, It will be enough.

Maybe It can be other group of integrals or can be found a relation with elliptic integrals such as $\int_0^{1} \frac{\;du}{\sqrt{1-u^2}\sqrt{1-x^2u^2}}$. I do not know yet.

  • $\begingroup$ For $b=c$ Mathematica evaluates it into some expression involging the Hypergeometric function with $\frac{b^3}{a^3}$ as the last argument. $\endgroup$
    – Nikolaj-K
    Jul 29, 2013 at 9:48
  • $\begingroup$ How did you get from the first line to the second? That doesn't seem right $\endgroup$
    – nbubis
    Jul 29, 2013 at 10:12
  • 3
    $\begingroup$ @nbubis Please see en.wikipedia.org/wiki/Landen's_transformation $\endgroup$
    – Mathlover
    Jul 29, 2013 at 10:15
  • $\begingroup$ Just wanted to tell what a great question this is. You've showed several general methods for finding Landen transformations, which can be used with different integrals $\endgroup$
    – Yuriy S
    Aug 21, 2016 at 17:01

1 Answer 1


There is a known AGM-like transformation for your integral $$ g(a,b,c) = \int_0^\infty \frac{dx} { \root 3 \of {x^3+a^3} \root 3 \of {x^3+b^3} \root 3 \of {x^3+c^3} }. $$ It was obtained only a few years ago, and is considerably more complicated and harder to prove than Landen's transformation; it's inspired by the familiar AGM, but does not reduce to it, and must be developed on its own terms.

Define $$ I(a,b,c) = \int_0^\infty \frac{dt}{\root 3 \of {t(t+a^3)(t+b^3)(t+c^3)}}. $$ The change of variables $x^3 = 1/t$, $dx = -t^{-4/3} dt/3$ gives $$ g(a,b,c) = \frac13\int_0^\infty \frac{dt} {\root 3 \of {t(1+a^3t)(1+b^3t)(1+c^3t)}} = \frac1{3abc} I(a^{-1},b^{-1},c^{-1}). $$

Now the AGM-like identity is $$ I(a,b,c) = I(a',b',c') $$ where the transformation $\psi: (a,b,c) \mapsto (a',b',c')$ is given by $$ a' = \frac13(a+b+c), \quad b', c' = \left[\frac12 \left( \frac{a^2 (b+c) + b^2(c+a) + c^2(a+b)}{3} \pm \frac{(a-b)(b-c)(c-a)}{\sqrt{-27}} \right)\right]^{1/3}. $$ Note the $\sqrt{-27}$ in the denominator: if $a,b,c$ are distinct positive reals then $b',c'$ are complex conjugates (chosen to be the principal cube roots), but then another application of $\psi$ returns positive $a'',b'',c''$ that are much closer than $a,b,c$. Repeated application of $\psi$ yields a sequence that converges cubically to $(M,M,M)$ for some $M=M(a,b,c)$ which is the analogue for these "Picard periods" of the AGM; and then $$ I(a,b,c) = I(M,M,M) = \frac{\pi \sqrt{4/3}}{M}. $$

For example, if we want to evaluate $g(3.1,4.1,5.9)$ then we write $$ g(3.1,4.1,5.9) = \frac1{3 \cdot 3.1 \cdot 4.1 \cdot 5.9} I\left(\frac1{3.1},\frac1{4.1},\frac1{5.9}\right). $$ Applying $\psi^3$ to $(1/3.1, 1/4.1, 1/5.9)$ yields $M = 0.2453101037492669116299747362\ldots$, so $I(1/3.1, 1/4.1, 1/5.9) = \pi \sqrt{4/3} / M = 14.78780805610937446473708974\ldots$, and $g(3.1,4.1,5.9) = 0.0657332322345471756512603615-$ which agrees with the numerical computation of $g(a,b,c)$ (try telling


to gp).

The reference where I found formulas equivalent to the above recipe for $I(a,b,c)$ is

Keiji Matsumoto and Hironori Shiga: A variant of Jacobi type formula for Picard curves, J. Math. Soc. Japan 62 #1 (2010), 305$-$319. (doi: 10.2969/jmsj/06210305)

This paper in turn attributes the formulas for $\psi$ and $M$ to Theorem 2.2 on page 134 of

Kenji Koike and Hironori Shiga: Isogeny formulas for the Picard modular form and a three terms arithmetic geometric mean, J. Number Theory 124 (2007), 123$-$141.

  • $\begingroup$ Very impressive. Thank you very much! $\endgroup$ Dec 26, 2013 at 7:24
  • $\begingroup$ @Noam D. Elkis I think that you mistyped $x^3 = 1/t$, $dx = -t^{-2/3} dt/3$ . I have updated it as $dx = -t^{-4/3} dt/3$ . Thanks a lot for your great answer. $\endgroup$
    – Mathlover
    Apr 3, 2015 at 7:08
  • $\begingroup$ This is just awesome. The method is similar to Carlson's integrals, but much more complex $\endgroup$
    – Yuriy S
    Jun 11, 2016 at 14:37
  • $\begingroup$ Nice answer (+1). $\endgroup$ Jun 11, 2017 at 6:28

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.