how Richard Feynman would integrate 1/(1+x^2)^2

We can use trig substitution (letting x=tanθ ) to do a typical calculus 2 integral, the integral of 1/(1+x^2)^2. However, we will use Feynman's technique of integration (aka Leibniz's integral rule, aka differentiation under the integral sign) to come up with a very nice integral formula.

10% off with the code "WELCOME10"
----------------------------------------
***Thanks to ALL my lovely patrons for supporting my channel and believing in what I do***
AP-IP Ben Delo Marcelo Silva Ehud Ezra 3blue1brown Joseph DeStefano
Mark Mann Philippe Zivan Sussholz AlkanKondo89 Adam Quentin Colley
Gary Tugan Stephen Stofka Alex Dodge Gary Huntress Alison Hansel
Delton Ding Klemens Christopher Ursich buda Vincent Poirier Toma Kolev
Tibees Bob Maxell A.B.C Cristian Navarro Jan Bormans Galios Theorist
Robert Sundling Stuart Wurtman Nick S William O'Corrigan Ron Jensen
Patapom Daniel Kahn Lea Denise James Steven Ridgway Jason Bucata
Mirko Schultz xeioex Jean-Manuel Izaret Jason Clement robert huff
Julian Moik Hiu Fung Lam Ronald Bryant Jan Řehák Robert Toltowicz
Angel Marchev, Jr. Antonio Luiz Brandao SquadriWilliam Laderer Natasha Caron Yevonnael Andrew Angel Marchev Sam Padilla ScienceBro Ryan Bingham
Papa Fassi Hoang Nguyen Arun Iyengar Michael Miller Sandun Panthangi
Skorj Olafsen Riley Faison Rolf Waefler Andrew Jack Ingham P Dwag Jason Kevin Davis Franco Tejero Klasseh Khornate Richard Payne Witek Mozga Brandon Smith Jan Lukas Kiermeyer Ralph Sato Kischel Nair Carsten Milkau Keith Kevelson Christoph Hipp Witness Forest Roberts Abd-alijaleel Laraki Anthony Bruent-Bessette Samuel Gronwold Tyler Bennett christopher careta Troy R Katy Lap C Niltiac, Stealer of Souls Jon Daivd R meh Tom Noa Overloop Jude Khine R3factor. Jasmine Soni L wan na Marcelo Silva
----------------------------------------

[AndrewDotsonvideos]

He has another famous technique (at least for people who calculate Feynman diagrams) called using Feynman parameters. It’s a way of re-casting an integral you’re solving into a form with temporarily more integrals that make the original integral easier to evaluate. Of course this is only helpful if the remaining Feynman parameter integrals can be solved analytically or are at least less expensive to solve numerically (it’s usually the latter). Not sure if you’ve ever made a video on it, but in the same spirit of Feynman integration tricks!

[plainbreadmike]

Can't get enough of these integrals with Feynman's technique videos, they're just so satisfying!

[flowingafterglow]

That answer is really slick, because if you look at it, the first term just looks like the integral of 1/u^2, where u = 1+x^2 and then the second term is just some version of the integral of 1/(1+x^2)

Random factors of 2, I agree, but the form is pretty cool

[davidalexander]

For definite integrals, I now see that there is actually no differentiating under an integral sign (requiring something like dominated convergence theorem) it's actually much prettier, we can write it as follows:

d/dx (1/a arctan(x/a))
= 1/(a^2+x^2)
Hence,
d/da d/dx (1/a arctan(x/a))
= d/da 1/(a^2+x^2).
By commutativity of partial derivatives,
d/dx d/da (1/a arctan(x/a))
= d/da 1/(a^2+x^2).
Thus, an anti derivative for
d/da (1/a arctan(x/a))
is
d/da 1/(a^2+x^2). (then work out these partial derivatives)

[NarutoSSj]

Rip Chen Leu. Although maybe never uttered by name again, you have a special place in all our hearts.

[MathNerd]

I recall seeing you do this in October 2018. Still a very neat video! :)

[chanduudarapu]

Thank you so much, lots of love from India 🇮🇳

[jellowz]

Thank po sir!
I hope you will also teach this topic "definition of exp z for imaginary z" under the linear equations with constant coefficient

[rjc]

Bro, you're so talented, I have my undergrad as a mech E and still come to your page for fun! Please don't stop the videos lol

[maciejkubera]

Great video as usual! On 0:49 You forgot to square the constant c ;) ;) ;) ;)

[SlipperyTeeth]

"Why do we add the +C at the end?" It depends on what you consider integration to be. Normally we just think of integration as the opposite of differentiation. But then, what is differentiation?

If you think of differentiation as a function from functions to functions, then integration should be its inverse function. But there isn't in general a left inverse for differentiation, because it's not one-to-one - and there are multiple right inverses. So you might consider "integration" to be the entire set/class of right inverses of differentiation - such that whenever you compose "integration"/differentiation, you pull back this abstract layer of set/class and compose them with every instance of an integration function. So differentiation after "integration" is just the set/class of differentiation after right inverses of differentiation - which all collapse to the identity. And there's the added bonus that with just a little more information (such as a single point on the curve) you'll be able to choose one of those integration functions to "act" as a left inverse for a specific input - so the whole set/class of integration functions can act as a left/right inverse for differentiation.

For single variable calculus, that's about all you need to consider, and this is a perfectly fine way to define the integration notation. For multivariable calculus, there's a new wrinkle. You can have a function that's constant in one variable, but not another (Let f(x, y) = y, then d/dx (f(x, y)) = 0). So if you integrate a function in the variable x, then you pick up a constant in the variable x. And then if you differentiate that by the variable a, it doesn't always become 0, because "constant in the variable x" doesn't imply "constant in the variable a". Sure, some functions are constant in both x and a, but not all. So if we compose differentiation with "integration", some of those compositions will collapse the constant, but not all. We didn't add +C at the end, it never should have been removed.

[Francesco-bfcb]

I've tried a differents (but much longer) method
You know when you differentiate f/g you get (f'g-g'f)/g², so know it becomes a differential equation

[adamlopez]

wow, what a nice way to solve this integral. Thank you for the video

[haaansolo]

I learned about Feynman's trick when it came up in Howard and Sheldon's fight on tbbt, and have been stunned by it ever since.

[vishalmishra]

This video has an innovative new method of solving such integrals. Here is the old boring way for the same -
set x = tanT which changes the problem to INT { cos^2 T = (1+cos2T)/2 } dT = T/2 + sin2T/4 = T/2 + 2tanT / 4sec^2 T = [ arctan x + x / (1+x^2) ] / 2

[jirisykora]

I don't know if anyone wrote it before but if you plug in -1 for a it's going to be the same as for a=1 because of the nature of the formula, where a^3 and tan^-1 cancel the negative sign of each other.

[frencyii]

You can also solve it by substituting x=tan(u), it allows u to simplify until coming to intregral(cos^2(x)), which is easily solvable with some goniometric formulas.

[neutron]

Didn't thought bout that amazing technique!

[pirnessa]

One mistake: C is a constant in terms of x not in a. Hence, the partial derivative d/da C is not zero in general, it is just another constant in terms of x (which you added back in the end). Nice video! (PLEASE SEE EDITS BELLOW BEFORE YOU COMMENT)

Edit: C may depend on parameter a however it wants. Thereby, C can be any function of parameter/variable a, and so, it might not be differentiable with respect to a. So in the end, the best way is just to find one antiderivative of ∫1/(a^2+x^2)^2dx. Then when we set a=1, we now know one antiderivative of ∫1/(1+x^2)^2dx. But we know that all the other antiderivatives of ∫1/(1+x^2)^2dx are obtained by adding a real constant to the antiderivative we already know, since g'(x)=0 for all real x if and only if g is a constant function. This is basically what @blackpenredpen did, but the reasoning that C is a constant with respect to a is not right although it is irrelevant mistake for the main point of the video.

Edit 2: First of all my claim is that where C:R->R is any function and R is the set of real numbers. If you think that I'm wrong and that ∫1/(a^2+x^2)dx=(1/a)arctan(x/a)+c, only when c is just any real number, i.e. constant with respect to both a and x. Then your claim against my claim is that if C:R->R is not a constant function, then
(1/a)arctan(x/a)+C(a) is not an antiderivative of 1/(a^2+x^2) with respect to x. To help you, I will go through what you are trying to prove and why it is not true.

So you need to take a function C:R->R which is not constant, for example you can think C(a)=a. Then you need to prove that (1/a)arctan(x/a)+C(a) is not an antiderivative of 1/(a^2+x^2) with respect to x. But you know the definition of antiderivative for multi variable functions, so you know that by the definition you need to prove that the partial derivative d/dx ( (1/a)arctan(x/a)+C(a) ) is not equal to 1/(a^2+x^2). But we know the following partial derivatives: d/dx (1/a)arctan(x/a)=1/(a^2+x^2) and d/dx C(a) = 0. So by the linearity of partial derivative you have d/dx ( (1/a)arctan(x/a)+C(a) )=1/(a^2+x^2). Thus, your claim is wrong and we have ended up proving that (1/a)arctan(x/a)+C(a) is an antiderivative of 1/(a^2+x^2) with respect to x if C:R->R is any function. In the comments you can find also different reasonings and how other people realized this.

If you still disagree, please read the 50+ other comments in detail, read my arguments, read others arguments, read why in the end they realized that C can be a function of a. Our comments are not the best source so I also recommend studying or recalling multivariable calculus and before that one variable calculus. Even better is to go to talk people in some university's math department. If after this you still feel that I'm wrong, then G I M M E A V A L I D P R O O F of the direction you are claiming and cite to my previous comments and show where I went wrong so the conversation is easier and faster.

[joykukreja]

Cool method. I did it by putting x = tan theta in 5 steps.