Better Integration by Parts

From “Integration by Parts Done Right“, we learned

\int u' \cdot v \;dx = u\cdot v -\int u\cdot v' \; dx.\quad\quad\quad(1)

It is equally true that

\int u' \cdot v \;dx = \underline{(u + C)}\cdot v -\int \underline{(u+C)}\cdot v' \; dx\quad\quad\quad(2)

for any constant C ( see Exercise-1).

Notice (1) is a special case of (2) when C=0.

I have observed in many cases when proper choice of C is made, \int (u+C)\cdot v'\;dx can be tremendously simplified.

For example, to evaluate \int \log(x+k)\; dx, we let C=k so

\int \log(x+k)\;dx = \int x'\cdot \log(x+k)\;dx

\overset{(2)}{=} (x+k) \log(x+k)-\int (x+k)\cdot (\log(x+k))'\;dx

= (x+k)\log(x+k) -\int (x+k)\cdot \frac{1}{x+k}\;dx

= (x+k)\log(x+k) - \int \;dx

= (x+k)\log(x+k)-x

Had we chosen C=0, we would have

\int \log(x+k)\;dx = \int x'\cdot \log(x+k)\;dx

\overset{(2)}{=} x \log(x+k)-\int x\cdot (\log(x+k))'\;dx

= x\log(x+k) -\int x\cdot \frac{1}{x+k}\;dx

= x\log(x+k) - \int \frac{x+k-k}{x+k}\;dx

= x\log(x+k)-\int 1-\frac{k}{x+k}\;dx

= x\log(x+k)-x+k\int \frac{1}{x+k}\;dx

= x\log(x+k)-x+k\log(x+k)

= (x+k)\log(x+k)-x

Now let’s look at \int \arctan(\sqrt{x+k})\;dx.

\int \arctan(\sqrt{x+k})\;dx = \int x' \arctan(\sqrt{x+k})\;dx

\overset{(2)}{=} (x+\underbrace{k+1}_{C})\arctan(\sqrt{x+k})-\int (x+\underbrace{k+1}_{C})\cdot(\arctan(\sqrt{x+k}))'\;dx

= (x+k+1)\arctan(\sqrt{x+k})-\int (x+k+1)\cdot \frac{1}{1+(\sqrt{x+k})^2}\cdot\frac{1}{2\sqrt{x+k}}\;dx

= (x+k+1)\arctan(\sqrt{x+k})-\int (x+k+1)\cdot \frac{1}{1+x+k}\cdot\frac{1}{2\sqrt{x+k}}\;dx

= (x+k+1)\arctan(\sqrt{x+k})-\int \frac{1}{2\sqrt{x+k}}\;dx

= (x+k+1)\arctan(\sqrt{x+k})-\sqrt{x+k}

If we choose C=k,

\int \arctan(\sqrt{x+k})\;dx = \int x' \arctan(\sqrt{x+k})\; dx

\overset{(2)}{=} (x+k)\arctan(\sqrt{x+k})-\int (x+k)\frac{1}{1+x+k}\cdot\frac{1}{2\sqrt{x+k}}\;dx

= (x+k)\arctan(\sqrt{x+k})-\frac{1}{2}\int\frac{\sqrt{x+k}}{x+k+1}\;dx

u = \sqrt{x+k} \implies x=u^2-k, \frac{dx}{du} = 2u

= (x+k)\arctan(\sqrt{x+k})-\frac{1}{2}\int\frac{2u^2}{u^2+1}\;du

= (x+k)\arctan(\sqrt{x+k})-\frac{1}{2}\int 2(\frac{u^2+1-1}{u^2+1})\;du

= (x+k)\arctan(\sqrt{x+k})-\int (1-\frac{1}{u^2+1})\;du

= (x+k)\arctan(\sqrt{x+k})-(u-\arctan(u))

= (x+k)\arctan(\sqrt{x+k})-(\sqrt{x+k}-\arctan(\sqrt{x+k}))

= (x+k+1)\arctan(\sqrt{x+k})-\sqrt{x+k}.

Fig. 1 Omega‘s choice: C=k

It takes longer manually if C=0 (see Exercise-3)

To illustrate Just-in-Time determination of C, we evaluate \int x \arctan(x)\; dx:

\int x \arctan(x) \;dx = \int(\frac{x^2}{2})'\arctan(x)\;dx

= (\frac{x^2}{2} + C)\arctan(x) - \int (\frac{x^2}{2} + C)(\arctan(x))'\; dx

= (\frac{x^2}{2} + C)\arctan(x) - \int (\frac{x^2}{2}+C)\cdot \frac{1}{1+x^2}\; dx

While the last integral is not difficult we see by choosing C=\frac{1}{2}, it is greatly simplified.

= (\frac{x^2}{2} + \frac{1}{2})\arctan(x) - \frac{1}{2}\int\; dx

= (\frac{x^2}{2} + \frac{1}{2}) \arctan(x) - \frac{1}{2}x

Fig. 2 Omega‘s choice: C=0


Exercise-1 Show that \int u' \cdot v \;dx = \underline{(u + C)}\cdot v -\int \underline{(u+C)}\cdot v' \; dx where C is a constant.

Exercise-2 Evaluate \int \log(\sqrt{x+k})\;dx.

Exercise-3 Evaluate \int \frac{x}{(x+k+1)\sqrt{x+k}}\;dx.

More Extraordinary Sums

The summation derived in my previous post, namely

\frac{\pi^2}{6} = 1 + \frac{1}{4} + \frac{1}{9} + \frac{1}{16} + \dots + \frac{1}{k^2} + \dots,\quad\quad\quad(1)

is Leonhard Euler’s solution to the problem of finding the precise summation for the reciprocals of squared natural numbers.

We now proceed to find the precise summation for the reciprocals of fourth powered natural numbers:

1 + \frac{1}{2^4} + \frac{1}{3^4} + \frac{1}{4^4} + \dots + \frac{1}{k^4} + \dots

Recall Euler’s key equation in finding (1):

1+\frac{x^2}{3!}+\frac{x^4}{5!} + \frac{x^6}{7!} +\frac{x^7}{9!} +\cdots = (1-\frac{x^2}{\pi^2})(1-\frac{x^2}{4\pi^2})(1-\frac{x^2}{9\pi^2})(1-\frac{x^2}{16\pi^2})\cdots,\quad\quad\quad(2)

whose right hand side can be expressed as

\prod\limits_{i=1}^{n}(1-c_ix^2)\quad\quad\quad(3)

where n=1,2,3,..., c_i = \frac{1}{(i\pi)^2}. Computer algebra exploration (see Fig. 1) suggests that for finite value n, its coefficient of x^4 is

\sum\limits_{i=1}^{n}\left(c_i\sum\limits_{j=i+1}^{n}c_j\right).

Fig. 1

Fig. 2

Since

\sum\limits_{i=1}^{n}\left(c_i\sum\limits_{j=i+1}^{n}c_j\right)=\frac{1}{2}\left((c_1+c_2+c_3+\dots+c_n)^2-(c_1^2+c_2^2+c_3^2+\dots+c_n^2)\right)

holds for n=1,2,3,.... (see (*)), the coefficient of x^4 on the right hand side of (2) becomes

\frac{1}{2}\left((c_1+c_2+c_3+\dots)^2-(c_1^2+c_2^2+ c_3^2 + \dots)\right)

\overset{c_i=\frac{1}{(i\pi)^2}}{=}\frac{1}{2}\left(\frac{1}{\pi^2}  + \frac{1}{4\pi^2} + \frac{1}{9\pi^2} + \frac{1}{16\pi} + \dots)^2-(\frac{1}{\pi^4}+\frac{1}{16\pi^4} + \frac{1}{81\pi^4} + \frac{1}{256\pi^4} + \dots)\right)

= \frac{1}{2}\left(\frac{1}{\pi^4}\left(\boxed{1+\frac{1}{4}+\frac{1}{9}+\frac{1}{16}+\dots}\right)^2-\frac{1}{\pi^4}(1+\frac{1}{16}+\frac{1}{81}+\frac{1}{256}+\dots)\right)

\overset{(1)}{=} \frac{1}{2}\left(\frac{1}{\pi^4}\left(\boxed{\frac{\pi^2}{6}}\right)^2-\frac{1}{\pi^4}(1+\frac{1}{16}+\frac{1}{81}+\frac{1}{256}+\dots)\right)

= \frac{1}{2}\left(\frac{1}{\pi^4}\frac{\pi^4}{6^2}-\frac{1}{\pi^4}(1+\frac{1}{16}+\frac{1}{81}+\frac{1}{256}+\dots)\right)

=\frac{1}{2}\left(\frac{1}{36}-\frac{1}{\pi^4}(1+\frac{1}{16}+\frac{1}{81}+\frac{1}{256}+\dots)\right)

= \frac{1}{72}-\frac{1}{2\pi^4}\left(1+\frac{1}{16}++\frac{1}{81} + \frac{1}{256} + \cdots\right).

Equate the coefficients of x^4 on both sides of (2), we have

\frac{1}{5!} = \frac{1}{72} - \frac{1}{2\pi^4}\left(\underline{1+\frac{1}{16} + \frac{1}{81}+ \frac{1}{256}+ \dots}\right)

i.e.,

-\frac{1}{180} =  - \frac{1}{2\pi^4}\left(\underline{1+\frac{1}{16} + \frac{1}{81}+ \frac{1}{256}+ \dots}\right)

Cross multiply -2\pi^4 then yields

1 + \frac{1}{16} + \frac{1}{81} + \frac{1}{256} + \frac{1}{625} +  \frac{1}{1296} + \dots + \frac{1}{k^4}+ ...= \frac{\pi^4}{90}.

Using a Computer Algebra System, we can obtain the precise summation for the reciprocals of even powered natural numbers endlessly.

Fig. 3

A question arises naturally at this point:

For odd integer i, what is 1 +\frac{1}{2^i} +\frac{1}{3^i} + \frac{1}{5^i} +  \dots ?

We know 1+\frac{1}{2^1}+\frac{1}{3^1} + \frac{1}{4^1} + ...= 1 + \frac{1}{2} + \frac{1}{3} + \frac{1}{4} + \dots is divergent (see “My shot at Harmonic Series“)

For all other odd values of i, it is convergent (see Exercise-1)

But can we evaluate them precisely?

Some have conjectured that the sum in question is of the form \left(\frac{p}{q}\right)\cdot\pi^i for some fraction \frac{p}{q} as well. But to this day, no one knows if this is true.

The past two hundred years of mathematical research have not advanced our knowledge on the odd powers.


Prove: \left(\sum\limits_{i=1}^{n}c_i\right)^2-\sum\limits_{i=1}^{n}c_i^2 = 2\sum\limits_{i=1}^{n-1}\left(c_i\sum\limits_{j=i+1}^{n}c_j\right)\quad\quad\quad(*)

For n = 2, (*) is true:

\left(\sum\limits_{i=1}^{2}c_i\right)^2-\sum\limits_{i=1}^{2}c_i^2 = (c_1+c_2)^2-(c_1^2+c_2^2) = c_1^2+2c_1c_2+c_2^2-c_1^2-c_2^2 = 2c_1c_2.

2\sum\limits_{i=1}^{1}\left(c_i\sum\limits_{j=i+1}^{2}c_j\right) =2c_1\sum\limits_{j=2}^{2}c_j = 2c_1c_2

Assume for n=k, (*) is true:

\left(\sum\limits_{i=1}^{k}c_i\right)^2-\sum\limits_{i=1}^{k}c_i^2 = 2\sum\limits_{i=1}^{k-1}\left(c_i\sum\limits_{j=i+1}^{k}c_j\right).\quad\quad\quad(A-1)

Let’s show (*) is true when n=k+1:

\left(\sum\limits_{i=1}^{k+1}c_i\right)^2-\sum\limits_{i=1}^{k+1}c_i^2

=  \left(\sum\limits_{i=1}^{k}c_i +c_{k+1}\right)^2-\sum\limits_{i=1}^{k+1}c_i^2

=  \left(\sum\limits_{i=1}^{k}c_i\right)^2 + 2 (\sum\limits_{i=1}^{k}c_i)c_{k+1} + c_{k+1}^2-\sum\limits_{i=1}^{k}c_i^2 - c_{k+1}^2

= \left(\sum\limits_{i=1}^{k}c_i\right)^2 + 2 \sum\limits_{i=1}^{k}(c_i c_{k+1}) -\sum\limits_{i=1}^{k}c_i^2

= \underline{\left(\sum\limits_{i=1}^{k}c_i\right)^2 -\sum\limits_{i=1}^{k}c_i^2} + 2 \sum\limits_{i=1}^{k}(c_i c_{k+1})

\overset{(A-1)}{=} 2\sum\limits_{i=1}^{k-1}\left(c_i\sum\limits_{j=i+1}^{k}c_j\right) + 2 \sum\limits_{i=1}^{k}(c_i c_{k+1})

= 2\left(\sum\limits_{i=1}^{k-1}\left(c_i\sum\limits_{j=i+1}^{k}c_j\right) + \sum\limits_{i=1}^{k}(c_i c_{k+1})\right)

= 2\biggl(c_1(c_2+c_3+\dots+c_k) +c_1c_{k+1} + c_2(c_3+c_4 + \dots + c_k) + c_2c_{k+1}+\dots+ c_{k-2}(c_{k-1} + c_k) + c_{k-2}c_{k+1}+ c_{k-1}c_k +c_{k-1}c_{k+1}+ c_kc_{k+1}\biggr)

= 2\biggl(c_1(c_2+c_3 + \dots + c_{k+1})+ c_2(c_3+c_4 + \dots + c_{k+1})+\dots+ c_{k-2}(c_{k-1} + c_k + c_{k+1}) + c_{k-1}(c_k+c_{k+1}) + c_k c_{k+1}\biggr)

= 2\sum\limits_{i=1}^{k}\left(c_i\sum\limits_{j=i+1}^{k+1}c_j\right).

= 2\sum\limits_{i=1}^{(k+1)-1}\left(c_i\sum\limits_{j=i+1}^{k+1}c_j\right).


Exercise-1 Show that for i \ge 2, \sum\limits_{k=1}^{\infty}\frac{1}{k^i} is convergent.

Exercise-2 Derive \sum\limits_{k=1}^{\infty}\frac{1}{k^i} for i = 6, 8, 10, ..., 30.

Deriving the Extraordinary Euler Sum

By definition,

\sin(x) = x - \frac{x^3}{3!} + \frac{x^5}{5!} - \frac{x^7}{7!} + \frac{x^9}{9!} + \dots

Dividing by x \ne 0 yields

\frac{\sin(x)}{x} = 1 - \frac{x^2}{3!} + \frac{x^4}{5!} - \frac{x^6}{7!} + \frac{x^8}{9!}+\dots\quad\quad\quad(1)

Assuming the properties of finite polynomials hold true for infinite series,

\sin(x) = 0 when x=\pm \pi, \pm 2\pi, \pm 3\pi, ...

means

1 - \frac{x^2}{3!} + \frac{x^4}{5!} - \frac{x^6}{7!} + \frac{x^8}{9!}+\dots

= (1-\frac{x}{\pi})(1-\frac{x}{-\pi})(1-\frac{x}{2\pi})(1-\frac{x}{-2\pi})(1-\frac{x}{3\pi})(1-\frac{x}{-3\pi})\dots

= (1-\frac{x}{\pi})(1+\frac{x}{\pi})(1-\frac{x}{2\pi})(1+\frac{x}{2\pi})(1-\frac{x}{3\pi})(1+\frac{x}{3\pi})\dots

=(1-\frac{x^2}{\pi^2})(1-\frac{x^2}{4\pi^2})(1-\frac{x^2}{9\pi^2})(1-\frac{x^2}{16\pi^2})\dots

i.e.,

1 - \frac{x^2}{3!} + \frac{x^4}{5!} - \frac{x^6}{7!} + \frac{x^8}{9!}+\dots=(1-\frac{x^2}{\pi^2})(1-\frac{x^2}{4\pi^2})(1-\frac{x^2}{9\pi^2})(1-\frac{x^2}{16})\dots\quad\quad\quad(2)

It can be shown (see Exercise-1) that the coefficient of x^2 on the right hand side of (2) is

-\frac{1}{\pi^2}\left(1+ \frac{1}{4}+ \frac{1}{9} + \frac{1}{16}+\dots\right).

Therefore, equate the like powers of x^2 on both sides of (2) gives

-\frac{1}{3!} = -\frac{1}{\pi^2}\left(1+ \frac{1}{4}+ \frac{1}{9} + \frac{1}{16} + \dots\right).

Multiplying -\pi^2 throughout, we obtain the Extraordinary Euler Sum

\frac{\pi^2}{6} = 1 + \frac{1}{4} + \frac{1}{9} + \frac{1}{16} + \dots


Omega knows Euler’s sum


Exercise-1 show the coefficient of x^2 on the right hand side of (2) is

-\frac{1}{\pi^2}\left(1+ \frac{1}{4}+ \frac{1}{9} + \frac{1}{16} + \dots\right).

Exercise-2 Show that \sum\limits_{i=1}^{n} \frac{1}{i^2} is convergent.

Solving Minimization and Maximization Problems without Calculus [1]

Problem:

If x and N are real numbers such that N=\sqrt{5x+6} + \sqrt{7x+8}, then what is the smallest possible value of N? The answer must be exact.


Solution:

For N to be valid, we must have

5x+6 \ge 0\quad\quad\quad(1)

and

7x+8 \ge 0.\quad\quad\quad(2)

It follows that

(1) \implies x \ge -\frac{6}{5}, \;\;  (2) \implies x \ge -\frac{8}{7}.

Since

\left(-\frac{6}{5}-(-\frac{8}{7}) = -\frac{2}{35} < 0\right) \implies -\frac{6}{5} <-\frac{8}{7},

it must be true that

x \ge  -\frac{8}{7}.

Moreover, N(x) = \sqrt{5x+6} + \sqrt{7x+8} is an increase function:

\forall x_1,  x_2, (x_1 < x_2) \implies N(x_1) < N(x_2).

Therefore,

at x = -\frac{8}{7}, N attains minimum value N(-\frac{8}{7}) = \sqrt{\frac{2}{7}}.

That is,

The smallest possible value of N is \sqrt{\frac{2}{7}}.

Using Omega CAS Exlorer, we visualize our solution in Fig. 1.

Fig. 1

How Poetry and Math Intersect

Some say the world will end in fire,

Some say in ice.

From what I’ve tasted of desire

I hold with those who favor fire.

– Robert Frost


– Michael Xue